ESPResSo User`s Guide
ESPResSo User’s Guide
for version 3.4-dev-466-g94ac079
February 10, 2015
Todo list
Better throw some out (e.g. switches)? . . . . . . . . . . . . . . . . . . . . . . . . . 86
Missing: lattice switch, dpd tgamma, n rigidbonds . . . . . . . . . . . . . . . . . . 86
Which commands can be used to set the read-only variables? . . . . . . . . . . . . 86
Docs missing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Docs missing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Someone who knows or wrote this command should check this. . . . . . . . . . . . 112
Document the usage! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
Document the usage! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
Document the usage and what it is! . . . . . . . . . . . . . . . . . . . . . . . . . . 114
I think there is still a bug in there (Hanjo) . . . . . . . . . . . . . . . . . . . . . . 118
Describe the different energies components returned by the different commands! . 118
Document arguments nb inter, nb intra, tot nb inter and tot nb intra . . . . . . . 119
Description of how electrostatic contribution to Pressure is calculated . . . . . . . 119
Check this! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Missing descriptions of parameters of several observables . . . . . . . . . . . . . . . 130
any suggestion for a more suitable name? . . . . . . . . . . . . . . . . . . . . . . . 131
Formatted printing is not fully supported yet. . . . . . . . . . . . . . . . . . . . . . 133
Does not work yet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
Processing data from Tcl input or from input files is not fully supported yet. . . . 135
Complex conjugate product must be defined. . . . . . . . . . . . . . . . . . . . . . 136
Maybe not all parameters are printed. . . . . . . . . . . . . . . . . . . . . . . . . . 138
I do not understand this. How does the error look? . . . . . . . . . . . . . . . . . . 156
Missing commands: Probably all from scripts/auxiliary.tcl? . . . . . . . . . . 157
Complete in broad strokes the applicability of the electrokinetics model. Also
mention the difference in temperatures between EK and LB species. . . . . . 175
At the moment this fails badly, if you try to parse incorrectly formatted files. This
will be fixed in the future. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
The list contains all features, but there are tons of docs missing! . . . . . . . . . . 235
Docs missing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
How to use it? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
Documentation! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
BOND ANGLEDIST and BOND ENDANGLEDIST are completely undocumented.239
Cleanup: References, mathematics . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
3
Contents
1 Introduction
1.1 Guiding principles . . . . . .
1.2 Available simulation methods
1.3 Basic program structure . . .
1.4 On units . . . . . . . . . . . .
1.5 Requirements . . . . . . . . .
1.6 Syntax description . . . . . .
2 First steps
2.1 Quick installation . . . . . .
2.2 Running ESPResSo . . . . .
2.3 Creating the first simulation
2.4 tutorial.tcl . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. . . .
. . . .
script
. . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
9
9
10
12
12
14
14
.
.
.
.
15
15
16
16
22
3 Getting, compiling and running ESPResSo
3.1 Running configure . . . . . . . . . . . . . . . . . .
3.2 make: Compiling, testing and installing ESPResSo . .
3.3 Running ESPResSo . . . . . . . . . . . . . . . . . . .
3.4 myconfig.hpp: Activating and deactivating features
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
23
23
25
27
27
4 Setting up particles
4.1 part: Creating single particles . . .
4.2 Creating groups of particle . . . . .
4.3 constraint: Setting up constraints .
4.4 Virtual sites . . . . . . . . . . . . . .
4.5 Grand canonical feature . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
29
29
35
42
46
49
.
.
.
.
.
.
.
50
50
57
60
63
67
68
69
5 Setting up interactions
5.1 Isotropic non-bonded interactions . .
5.2 Anisotropic non-bonded interactions
5.3 Bonded interactions . . . . . . . . .
5.4 Object-in-fluid interactions . . . . .
5.5 Bond-angle interactions . . . . . . .
5.6 Dihedral interactions . . . . . . . . .
5.7 Coulomb interaction . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5
5.8
5.9
Dipolar interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
Special interaction commands . . . . . . . . . . . . . . . . . . . . . . . . . 82
6 Setting up the system
6.1 setmd: Setting global variables. . . . . . . . . . . . . .
6.2 thermostat: Setting up the thermostat . . . . . . . .
6.3 nemd: Setting up non-equilibrium MD . . . . . . . . .
6.4 cellsystem: Setting up the cell system . . . . . . . .
6.5 CUDA . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.6 Creating bonds when particles collide . . . . . . . . .
6.7 Catalytic Reactions . . . . . . . . . . . . . . . . . . . .
6.8 Galilei Transform and Particle Velocity Manipulation .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
86
86
88
93
94
95
96
97
99
7 Running the simulation
7.1 integrate: Running the simulation . . . . . . . . . . . . . . . .
7.2 time_integration: Runtime of the integration loop . . . . . . .
7.3 minimize_energy: Run steepest descent minimization . . . . . .
7.4 change_volume: Changing the box volume . . . . . . . . . . . .
7.5 lees_edwards_offset: Applying shear between periodic images
7.6 Stopping particles . . . . . . . . . . . . . . . . . . . . . . . . . .
7.7 velocities: Setting the velocities . . . . . . . . . . . . . . . . .
7.8 Fixing the particle sorting . . . . . . . . . . . . . . . . . . . . . .
7.9 Parallel tempering . . . . . . . . . . . . . . . . . . . . . . . . . .
7.10 Metadynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
101
. 101
. 102
. 102
. 102
. 103
. 104
. 104
. 104
. 105
. 109
8 Analysis in Tcl
8.1 Available observables . . . . . . . . . . . . . . . .
8.2 Analyzing groups of particles (molecules) . . . .
8.3 Storing configurations . . . . . . . . . . . . . . .
8.4 uwerr: Computing statistical errors in time series
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
111
111
122
126
127
9 Analysis in the core
129
9.1 Observables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
9.2 Correlations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
10 Input / Output
10.1 No generic checkpointing! . . . . . . . . . . .
10.2 blockfile: Using the structured file format .
10.3 Writing and reading binary files . . . . . . . .
10.4 Writing VTF files . . . . . . . . . . . . . . . .
10.5 writevtk: Particle Visualization in paraview
10.6 Reading and Writing PDB/PSF files . . . . .
10.7 Online-visualisation with VMD . . . . . . . .
10.8 Error handling . . . . . . . . . . . . . . . . .
6
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
143
. 143
. 144
. 148
. 149
. 151
. 152
. 153
. 155
11 Auxilliary commands
11.1 Finding particles and bonds . . . . . . . . . . . . . . . . . . . . . . . . .
11.2 Additional Tcl math-functions . . . . . . . . . . . . . . . . . . . . . . .
11.3 Checking for features of ESPResSo . . . . . . . . . . . . . . . . . . . . .
157
. 157
. 158
. 164
12 Lattice-Boltzmann
12.1 Setting up a LB fluid . . . . . . . . . . . . . . . . . . . .
12.2 LB as a thermostat . . . . . . . . . . . . . . . . . . . . .
12.3 The Shan Chen bicomponent fluid . . . . . . . . . . . .
12.4 SC as a thermostat . . . . . . . . . . . . . . . . . . . . .
12.5 SC component-dependent interactions between particles
12.6 Reading and setting single lattice nodes . . . . . . . . .
12.7 Removing total fluid momentum . . . . . . . . . . . . .
12.8 Visualization . . . . . . . . . . . . . . . . . . . . . . . .
12.9 Setting up boundary conditions . . . . . . . . . . . . . .
12.10Choosing between the GPU and CPU implementations .
12.11Electrohydrodynamics . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
174
. 174
. 176
. 178
. 179
.
.
.
.
183
. 184
. 184
. 184
. 186
13 Electrokinetics
13.1 Electrokinetic Equations
13.2 Setup . . . . . . . . . .
13.3 Output . . . . . . . . .
13.4 Catalytic Reaction . . .
14 Object-in-fluid
14.1 Membranes . . . . .
14.2 Parameters . . . . .
14.3 Geometry . . . . . .
14.4 Available commands
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
15 Immersed Boundary Method for soft elastic objects
16 External package: mbtools
16.1 Introduction . . . . . . . . . .
16.2 Installing and getting started
16.3 The main.tcl script . . . . .
16.4 Analysis . . . . . . . . . . . .
16.5 System generation . . . . . .
16.6 Utils . . . . . . . . . . . . . .
16.7 mmsg . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
165
165
167
168
169
169
170
171
171
171
172
173
192
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
194
. 194
. 195
. 196
. 198
. 202
. 209
. 215
17 Under the hood
218
17.1 Internal particle organization . . . . . . . . . . . . . . . . . . . . . . . . . 218
18 Getting involved
220
7
18.1
18.2
18.3
18.4
Community support and mailing lists .
Contributing your own code . . . . . .
Developers’ guide . . . . . . . . . . . .
User’s guide . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
A ESPResSo quick reference
220
221
221
221
222
B Features
235
B.1 General features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
B.2 Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
B.3 Debug messages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
C Sample scripts
241
D Maxwell Equations Molecular Dynamics (MEMD)
D.1 Equations of motion . . . . . . . . . . . . . . .
D.2 Discretization . . . . . . . . . . . . . . . . . . .
D.3 Initialization of the algorithm . . . . . . . . . .
D.4 Time integrator . . . . . . . . . . . . . . . . . .
D.5 Self–energy . . . . . . . . . . . . . . . . . . . .
D.6 For which systems to use the algorithm . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
242
242
243
243
244
245
246
E The
E.1
E.2
E.3
E.4
E.5
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
248
248
250
252
253
254
MMM family of algorithms
Introduction . . . . . . . . .
MMM2D . . . . . . . . . .
MMM1D . . . . . . . . . .
ELC . . . . . . . . . . . . .
Errors . . . . . . . . . . . .
F Bibliography
8
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
256
1. Introduction
ESPResSo is a simulation package designed to perform Molecular Dynamics (MD) and
Monte Carlo (MC) simulations. It is meant to be a universal tool for simulations of a
variety of soft matter systems. ESPResSo features a broad range of interaction potentials which opens up possibilities for performing simulations using models with different
levels of coarse-graining. It also includes modern and efficient algorithms for treatment
of electrostatics (P3M, MMM-type algorithms, Maggs algorithm, . . . ), hydrodynamic
interactions (DPD, Lattice-Boltzmann), and magnetic interactions. It is designed to
exploit the capabilities of parallel computational environments. The program is being
continuously extended to keep the pace with current developments both in the algorithms
and software.
The kernel of ESPResSo is written in C with computational efficiency in mind. Interaction between the user and the simulation engine is provided via a Tcl scripting
interface. This enables setup of arbitrarily complex systems which users might want to
simulate in future, as well as modifying simulation parameters during runtime.
1.1. Guiding principles
ESPResSo is a tool for performing computer simulation and this user guide describes how
to use this tool. However, it should be borne in mind that being able to operate a tool
is not sufficient to obtain physically meaningful results. It is always the responsibility of
the user to understand the principles behind the model, simulation and analysis methods
he is using. ESPResSo will not do that for you!
It is expected that the users of ESPResSo and readers of this user guide have a thorough
understanding of simulation methods and algorithms they are planning to use. They
should have passed a basic course on molecular simulations or read one of the renown
textbooks, e.g. [23]. It is not necessary to understand everything that is contained in
ESPResSo, but it is inevitable to understand all methods that you want to use. Using
the program as a black box without proper understanding of the background will most
probably result in wasted user and computer time with no useful output.
To enable future extensions, the functionality of the program is kept as general as
possible. It is modularized, so that extensions to some parts of the program (e.g. implementing a new potential) can be done by modifying or adding only few files, leaving
most of the code untouched.
To facilitate the understanding and the extensibility, much emphasis is put on readability of the code. Hard-coded assembler loops are generally avoided in hope that the
overhead in computer time will be more than compensated for by saving much of the
user time while trying to understand what the code is supposed to do.
9
Hand-in-hand with the extensibility and readability of the code comes the flexibility of
the whole program. On the one hand, it is provided by the generalized functionality of
its parts, avoiding highly specialized functions. An example can be the implementation
of the Generic Lennard-Jones potential described in section 5.1.3 where the user can
change all available parameters. Where possible, default values are avoided, providing
the user with the possibility of choice. ESPResSo cannot be aware whether your particles
are representing atoms or billiard balls, so it cannot check if the chosen parameters make
sense and it is the user’s responsibility to make sure they do.
On the other hand, flexibility of ESPResSo stems from the employment of Tcl at the
steering level. Apart from the ability to modify the simulation and system parameters at
runtime, many simple tasks which are not computationally critical can be implemented
at this level, without even touching the C-kernel. For example, simple problem-specific
analysis routines can be implemented in this way and made interact with the simulation
core. Another example of the program’s flexibility is the possibility to integrate system
setup, simulation and analysis in one single control script. ESPResSo provides commands
to create particles and set up interactions between them. Capping of forces helps prevent
system blow-up when initially some particles are placed on top of each other. Using
the Tcl interface, one can simulate the randomly set-up system with capped forces,
interactively check whether it is safe to remove the cap and switch on the full interactions
and then perform the actual productive simulation.
1.2. Available simulation methods
ESPResSo provides a number of useful methods. The following table shows the various methods as well as their status. The table distinguishes between the state of the
development of a certain feature and the state of its use. We distinguish between five
levels:
Core means that the method is part of the core of ESPResSo, and that it is extensively
developed and used by many people.
Good means that the method is developed and used by independent people from different groups.
Group means that the method is developed and used in one group.
Single means that the method is developed and used by one person only.
None means that the method is developed and used by nobody.
If you believe that the status of a certain method is wrong, please report so to the
developers.
10
Feature
Development Status
Integrators, Thermostats, Barostats
Velocity-Verlet Integrator
Core
Langevin Thermostat
Core
GHMC Thermostat
Single
DPD Thermostat
None
Isotropic NPT
None
NEMD
None
Quarternion Integrator
None
Interactions
Short-range Interactions
Core
Directional Lennard-Jones
Single
Gay-Berne Interaction
None
Constraints
Core
Relative Virtual Sites
Good
Center-of-mass Virtual Sites
None
RATTLE Rigid Bonds
None
Coulomb Interaction
P3M
Core
P3M on GPU
Single
Dipolar P3M
Group
Ewald on GPU
Single
MMM1D
Single
MMM2D
Single
MMM1D on GPU
Single
ELC
Good
MEMD
Single
ICC*
Group
Hydrodynamic Interaction
Lattice-Boltzmann
Core
Lattice-Boltzmann on GPU
Group
DPD
None
Shan-Chen Multicomponent Fluid Group
Tunable Slip Boundary
Single
Analysis
uwerr
None
Input/Output
Blockfiles
Core
VTF output
Core
VTK output
Group
PDB output
Good
Online visulation with VMD
Good
Miscellaneous
Usage Status
Core
Core
Single
Good
Single
Group
Good
Core
Single
Single
Core
Good
Good
Group
Core
Single
Good
Single
Good
Good
Single
Good
Group
Group
Core
Core
Good
Group
Single
Good
Core
Core
Group
Good
Good
11
Grand canonical feature
Metadynamics
Parallel Tempering
Electrokinetics
Object-in-fluid
Collision Detection
Catalytic Reactions
mbtools package
Single
Single
Single
Group
Group
Group
Single
Group
Single
Single
Single
Group
Group
Group
Single
Group
1.3. Basic program structure
As already mentioned, ESPResSo consists of two components. The simulation engine is
written in C for the sake of computational efficiency. The steering or control level is
interfaced to the kernel via an interpreter of the Tcl scripting language.
The kernel performs all computationally demanding tasks. Before all, integration of
Newton’s equations of motion, including calculation of energies and forces. It also takes
care of internal organization of data, storing the data about particles, communication
between different processors or cells of the cell-system. The kernel is modularized so that
basic functions are accessed via a set of well-defined lean interfaces, hiding the details of
the complex numerical algorithms.
The Tcl interpreter contains several special commands as an extension to Tcl, which
provide the interaction with the simulation engine. Thus, the user has at hand the
full realm of Tcl commands and constructs plus a few commands to communicate with
the simulation engine. The interfacing commands are designed so that they can both
set properties of the system (set up particles, interactions, thermostat) and retrieve
information about the already set-up entities. The standard Tcl constructs allow for
a flexible decision-making in the course of the simulation. This can be for example
exploited to check whether a simulation has reached the desired state. With a certain
overhead in efficiency, it can also be used to reject/accept new configurations in combined
MD/MC schemes. In principle, any parameter which is accessible from the Tcl level
can be changed at any moment of runtime. In this way methods like thermodynamic
integration become readily accessible.
The focus of the user guide is documenting the Tcl commands, their behaviour and
use in the simulation. It only describes certain technical details of implementation which
are necessary for understanding how the commands work. Technical documentation of
the code and program structure is contained in the Developers’ guide (see section 18.3).
1.4. On units
What is probably one of the most confusing subjects for beginners of ESPResSo is, that
ESPResSo does not predefine any units. While most MD programs specify a set of units,
like, for example, that all lengths are measured in ˚
Angstr¨om or nanometers, times are
12
measured in nano- or picoseconds and energies are measured in kJ/mol, ESPResSo does
not do so.
Instead, the length-, time- and energy scales can be freely chosen by the user. Once
these three scales are fixed, all remaining units are derived from these three basic choices.
The probably most important choice is the length scale. A length of 1.0 can mean
a nanometer, an ˚
Angstr¨
om, or a kilometer - depending on the physical system, that
the user has in mind when he writes his ESPResSo-script. When creating particles
that are intended to represent a specific type of atoms, one will probably use a length
scale of ˚
Angstr¨
om. This would mean, that e.g. the parameter σ of the Lennard-Jones
interaction between two atoms would be set to twice the van-der-Waals radius of the
atom in ˚
Angstr¨
om. Alternatively, one could set σ to 2.0 and measure all lengths in
multiples of the van-der-Waals radius. When simulation colloidal particles, which are
usually of micrometer size, one will choose their diameter (or radius) as basic length
scale, which is much larger than the ˚
Angstr¨om scale used in atomistic simulations.
The second choice to be made is the energy scale. One can for example choose to
set the Lennard-Jones parameter to the energy in kJ/mol. Then all energies will
be measured in that unit. Alternatively, one can choose to set it to 1.0 and measure
everything in multiples of the van-der-Waals binding energy of the respective particles.
The final choice is the time (or mass) scale. By default, ESPResSo uses a reduced
mass of 1, so that the mass unit is simply the mass of all particles. Combined with the
energy and length scale, this is sufficient to derive the resulting time scale:
s
[mass]
[time] = [length]
.
[energy]
˚
This means, that if you measure lengths in
pAngstr¨om, energies in kB T at 300 K and
˚
masses in 39.95u, then your time scale is A 39.95u/kB T = 0.40 ps.
On the other hand, if you want a particular time scale, then the mass scale can be
derived from the time, energy and length scales as
[mass] = [energy]
[time]2
.
[length]2
By activating the feature MASSES, you can specify particle masses in the chosen unit
system.
A special note is due regarding the temperature, which is coupled to the energy scale
by Boltzmann’s constant. However, since ESPResSo does not enforce a particular unit
system, we also don’t know the numerical value of the Boltzmann constant in the current
unit system. Therefore, when specifying the temperature of a thermostat, you actually
do not define the temperature, but the value of the thermal energy kB T in the current
unit system. For example, if you measure energy in units of kJ/mol and your real
temperature should be 300 K, then you need to set the thermostat’s effective temperature
to kB 300 Kmol/kJ = 2.494.
As long as one remains within the same unit system throughout the whole ESPResSoscript, there should be no problems.
13
1.5. Requirements
The following libraries and tools are required to be able to compile and use ESPResSo:
Tcl/Tk ESPResSo requires the Toolkit Command Language Tcl/Tk 1 in the version 8.3
or later. Some example scripts will only work with Tcl 8.4. You do not only need
the interpreter, but also the header files and libraries. Depending on the operating
system, these may come in separate development packages. If you want to use a
graphical user interface (GUI) for your simulation scripts, you will also need Tk.
FFTW For some algorithms (e.g. P3 M), ESPResSo needs the FFTW library version 3
or later 2 for Fourier transforms. Again, the header files are required.
MPI Finally, if you want to use ESPResSo in parallel, you need a working MPI environment (that implements the MPI standard version 1.2).
1.6. Syntax description
Throughout the user’s guide, formal definitions of the syntax of several Tcl-commands
can be found. The following conventions are used in these descriptions:
• Different variants of a command are labeled (1), (2), . . .
• Keywords and literals of the command that have to be typed exactly as given are
written in typewriter font.
• If the command has variable arguments, they are set in italicfont. The description following the syntax definition should contain a detailed explanation of the
argument and its type.
• ( alt1 | alt2 ) specifies, that one of the alternatives alt1 or alt2 can be used.
• [argument] specifies, that the argument argument is optional, i.e. it can be omitted.
• When an optional argument or a whole command is marked by a superscript label
(1), this denotes that the argument can only be used, when the corresponding
feature (see appendix B on page 235) specified in “Required features” is activated.
Example
(1) constraint
(2) constraint
type id
(3) constraint
(4) constraint
Required features:
1
2
http://www.tcl.tk/
http://www.fftw.org/
14
wall normal nx ny nz dist d type id
sphere center cx cy cz radius rad direction direction
rod center cx cy lambda lambda 1
ext_magn_field fx fy fz 2,3
CONSTRAINTS
1 ELECTROSTATICS
2 ROTATION
3 DIPOLES
2. First steps
2.1. Quick installation
If you have installed the requirements (see section 1.5 on the preceding page) in standard
locations, to compile ESPResSo, it is usually enough to execute the following sequence
of two steps in the directory where you have unpacked the sources:
./configure
make
This will compile ESPResSo in a freshly created object path named according to your
CPU and operating system. As you have not yet specified a configuration, a standard
version will be built with the most often used features. Usually you will want to build
another version of ESPResSo with options better suited for your purpose.
In some cases, e.g. when ESPResSo needs to be compiled for several different platforms
or when different versions with different sets of features are required, it might be useful
to execute the commands not in the source directory itself, but to start configure
from another directory (see section 3.1.1 on page 24). Furthermore, many features
of ESPResSo can be selectively turned on or off in the local configuration header (see
section 3.4 on page 27) before starting the compilation with make.
The shell script configure prepares the source code for compilation. It will determine
how to use and where to find the different libraries and tools required by the compilation
process, and it will test what compiler flags are to be used. The script will find out most
of these things automatically. If something is missing, it will complain and give hints on
how to solve the problem. The configuration process can be controlled with the help of
a number of options that are explained in section 3.1 on page 23.
The command make will compile the source code. Depending on the options passed
to the program, make can also be used for a number of other things:
• It can install and uninstall the program to some other directories. However, normally it is not necessary to actually install ESPResSo to run it.
• It can test ESPResSo for correctness.
• It can build the documentation.
The details of the usage of make are described in section 3.2 on page 25.
When these steps have successfully completed, ESPResSo can be started with the
command (see section 3.3 on page 27)
Espresso script
15
where script is a Tcl script that tells ESPResSo what to do, and has to be written by the
user. You can find some examples in the samples folder of the source code directory.
If you want to run in parallel, you should have compiled ESPResSo to use MPI, and
need to tell MPI to run ESPResSo in parallel. The actual invocation is implementation
dependent, but in many cases, such as OpenMPI, you can use
mpirun -n n nodes Espresso script
where n nodes is the number of prcessors to be used.
2.2. Running ESPResSo
ESPResSo is implemented as an extension to the Tcl scripting language. This means
that you need to write a script for any task you want to perform with ESPResSo. To
learn about the Tcl script language and especially the ESPResSo extensions, this chapter
offers two tutorial scripts. The first will guide you step-by-step through creating your
first simulation script, while the second script is a well documented example simulation
script. Since the latter is slightly more complex and uses more advanced features of
ESPResSo, we recommend to work through both scripts in the presented order. If you
want to learn about the Tcl language in greater detail, there is an excellent tutorial 1 .
2.3. Creating the first simulation script
This section introduces some of the features of ESPResSo by constructing step by step
a simulation script for a simple salt crystal. We cannot give a full Tcl tutorial here;
however, most of the constructs should be self–explanatory. We also assume that the
reader is familiar with the basic concepts of a MD simulation here. The code pieces can
be copied step by step into a file, which then can be run using Espresso file from the
ESPResSo source directory.
Our script starts with setting up the initial configuration. Most conveniently,
one would like to specify the density and the number of particles of the system as
parameters:
set n_part 200; set density 0.7
set box_l [expr pow($n_part/$density,1./3.)]
These variables do not change anything in the simulation engine, but are just standard
Tcl variables; they are used to increase the readability and flexibility of the script. The
box length is not a parameter of this simulation; it is calculated from the number of
particles and the system density. This allows to change the parameters later easily, e.g.
to simulate a bigger system.
The parameters of the simulation engine are modified by the setmd command.
For example
1
http://www.tcl.tk/man/tcl8.5/tutorial/tcltutorial.html
16
setmd box_l $box_l $box_l $box_l
setmd periodic 1 1 1
defines a cubic simulation box of size box_l, and periodic boundary conditions in
all spatial dimensions. We now fill this simulation box with particles
set q 1; set type 0
for {set i 0} { $i < $n_part } {incr i} {
set posx [expr $box_l*[t_random]]
set posy [expr $box_l*[t_random]]
set posz [expr $box_l*[t_random]]
set q [expr -$q]; set type [expr 1-$type]
part $i pos $posx $posy $posz q $q type $type
}
This loop adds n_part particles at random positions, one by one. In this construct,
only two commands are not standard Tcl commands: the random number generator
t_random and the part command, which is used to specify particle properties, here the
position, the charge q and the type. In ESPResSo the particle type is just an integer
number which allows to group particles; it does not imply any physical parameters. Here
we use it to tag the charges: positive charges have type 0, negative charges have type 1.
Now we define the ensemble that we will be simulating. This is done using the
thermostat command. We also set some integration scheme parameters:
setmd time_step 0.01; setmd skin 0.4
set temp 1; set gamma 1
thermostat langevin $temp $gamma
This switches on the Langevin thermostat for the NVT ensemble, with temperature temp
and friction gamma. The skin depth skin is a parameter for the link–cell system which
tunes its performance, but cannot be discussed here.
Before we can really start the simulation, we have to specify the interactions
between our particles. We use a simple, purely repulsive Lennard-Jones interaction
to model the hard core repulsion [25], and the charges interact via the Coulomb
potential:
set sig 1.0; set cut
[expr 1.12246*$sig]
set eps 1.0; set shift [expr 0.25*$eps]
inter 0 0 lennard-jones $eps $sig $cut $shift 0
inter 1 0 lennard-jones $eps $sig $cut $shift 0
inter 1 1 lennard-jones $eps $sig $cut $shift 0
inter coulomb 10.0 p3m tunev2 accuracy 1e-3 mesh 32
The first three inter commands instruct ESPResSo to use the same purely repulsive
Lennard–Jones potential for the interaction between all combinations of the two particle types 0 and 1; by using different parameters for different combinations, one could
simulate differently sized particles. The last line sets the Bjerrum length to the value
10, and then instructs ESPResSo to use P3 M for the Coulombic interaction and to try
17
to find suitable parameters for an rms force error below 10−3 , with a fixed mesh size of
32. The mesh is fixed here to speed up the tuning; for a real simulation, one will also
tune this parameter.
If we want to calculate the temperature of our system from the kinetic energy, we need
to know the number of the degrees of freedom of the particles. In ESPResSo these are
usually 3 translational plus 3 rotational degrees of freedom (if the feature ROTATION
is activated). You can get this number in the following way 2 :
if { [regexp "ROTATION" [code_info]] } {
set deg_free 6
} else { set deg_free 3 }
Now we can integrate the system:
set integ_steps 200
for {set i 0} { $i < 20 } { incr i} {
set temp [expr [analyze energy kinetic]/(($deg_free/2.0)*$n_part)]
puts "t=[setmd time] E=[analyze energy total], T=$temp"
integrate $integ_steps
}
This code block is the primary simulation loop and runs 20×integ_steps MD steps.
Every integ_steps time steps, the potential, electrostatic and kinetic energies are
printed out (the latter one as temperature). However, the simulation will crash:
ESPResSo complains about particle coordinates being out of range. The reason for
this is simple: Due to the initial random setup, the overlap energy is around a
million kT, which we first have to remove from the system. In ESPResSo, this is
can be accelerated by capping the forces, i. e. modifying the Lennard–Jones force
such that it is constant below a certain distance. Before the integration loop, we
therefore insert this equilibration loop:
for {set cap 20} {$cap < 200} {incr cap 20} {
puts "t=[setmd time] E=[analyze energy total]"
inter forcecap $cap; integrate $integ_steps
}
inter forcecap 0
This loop integrates the system with a force cap of initially 20 and finally 200. The last
command switches the force cap off again. With this equilibration, the simulation script
runs fine.
However, it takes some time to simulate the system, and one will probably like to
write out simulation data to configuration files, for later analysis. For this purpose
ESPResSo has commands to write simulation data to a Tcl stream in an easily
parsable form. We add the following lines at end of integration loop to write the
configuration files “config 0” through “config 19”:
2
There also exists a Tcl function degrees_of_freedom which does the same.
18
set f [open "config_$i" "w"]
blockfile $f write tclvariable {box_l density}
blockfile $f write variable box_l
blockfile $f write particles {id pos type}
close $f
The created files “config ...” are human–readable and look like
{tclvariable
{box_l 10}
{density 0.7}
}
{variable {box_l 10.0 10.0 10.0} }
{particles {id pos type}
{0 3.51770181433 4.3208975936 5.30529948918 0}
{1 3.93145531704 6.58506447035 6.95045147034 1}
...
}
As you can see, such a blockfile consists of several Tcl lists, which are called blocks,
and can store any data available from the simulation. Reading a configuration is
done by the following simple script:
set f [open $filename "r"]
while { [blockfile $f read auto] != "eof" } {}
close $f
The blockfile read auto commands will set the Tcl variables box_l and density to
the values specified in the file when encountering the tclvariable block, and set the
box dimensions for the simulation when encountering the variable block. The particle
positions and types of all 216 particles are restored when the particles block is read.
Note that it is important to have the box dimensions set before reading the particles, to
avoid problems with the periodic boundary conditions.
With these configurations, we can now investigate the system. As an example, we
will create a second script which calculates the averaged radial distribution functions
g++ (r) and g+− (r). The radial distribution function for a the current configuration
can be obtained using the analyze command:
set rdf [analyze rdf 0 1 0.9 [expr $box_l/2] 100]
set rlist ""
set rdflist ""
foreach value [lindex $rdf 1] {
lappend rlist
[lindex $value 0]
lappend rdflist [lindex $value 1]
}
The shown analyze rdf command returns the distribution function of particles of type
1 around particles of type 0 (i. e. of opposite charges) for radii between 0.9 and half the
19
Figure 2.1.: VMD Snapshot of the salt system
box length, subdivided into 100 bins. Changing the first two parameters to either “0 0”
or “1 1” allows to determine the distribution for equal charges. The result is a list of
r and g(r) pairs, which the following foreach loop divides up onto two lists rlist and
rdflist.
To average over a set of configurations, we put the two last code snippets into a
loop like this:
set cnt 0
for {set i 0} {$i < 100} {incr i} { lappend avg_rdf 0}
foreach filename $argv {
set f [open $filename "r"]
while { [blockfile $f read auto] != "eof" } {}
close $f
set rdf [analyze rdf 0 1 0.9 [expr $box_l/2] 100]
set rlist ""
set rdflist ""
foreach value [lindex $rdf 1] {
lappend rlist
[lindex $value 0]
lappend rdflist [lindex $value 1] }
set avg_rdf [vecadd $avg_rdf $rdflist]
incr cnt
}
set avg_rdf [vecscale [expr 1.0/$cnt] $avg_rdf]
Initially, the sum of all g(r), which is stored in avg_rdf, is set to 0. Then the loops
over all configurations given by argv, calculates g(r) for each configuration and adds
up all the g(r) in avg_rdf. Finally, this sum is normalized by dividing by the number
of configurations. Note the “1.0/$cnt”; this is necessary, since “1/$cnt” is interpreted
20
4
3.5
3
g(r)
2.5
2
1.5
1
0.5
0
1
1.5
2
2.5
3
r
Figure 2.2.: Radial distribution functions g++ (r) between equal charges (rectangles) and
g+− (r) for opposite charges (circles). The plus symbols denote g(r) for an
uncharged system.
as an integer division, which results in 0 for cnt > 1. argv is a predefined variable: it
contains all the command line parameters. Therefore this script should be called like
Espresso script [config... ]
The printing of the calculated radial distribution functions is simple. Add to the
end of the previous snippet the following lines:
set plot [open "rdf.data" "w"]
puts $plot "\# r rdf(r)"
foreach r $rlist rdf $avg_rdf { puts $plot "$r $rdf" }
close $plot
This instructs the Tcl interpreter to write the avg_rdf to the file rdf.data in gnuplot–
compatible format. Fig. 2.2 shows the resulting radial distribution functions, averaged
over 100 configurations. In addition, the distribution for a neutral system is given,
which can be obtained from our simulation script by simply removing the command
inter coulomb ... and therefore not turning on P3 M.
The code example given before is still quite simple, and the reader is encouraged to
try to extend the example a little bit, e. g. by using differently sized particle, or changing
the interactions. If something does not work, ESPResSo will give comprehensive error
messages, which should make it easy to identify mistakes. For real simulations, the
simulation scripts can extend over thousands of lines of code and contain automated
adaption of parameters or online analysis, up to automatic generation of data plots.
Parameters can be changed arbitrarily during the simulation process, as needed for e. g.
21
simulated annealing. The possibility to perform non–standard simulations without the
need of modifications to the simulation core was one of the main reasons why we decided
to use a script language for controlling the simulation core.
2.4. tutorial.tcl
In the directory samples/ of the es sources, you will find a well documented simulation
script tutorial.tcl, which takes you step by step through a slightly more complicated
simulation of a polyelectrolyte system. The basic structure of the script is however
the same as in the previous example and probably the same as the structure of most
ESPResSo simulation scripts.
Initially, some parameters and global variables are set, the interactions are initialized,
and particles are added. For this, the script makes use of the polymer command, which
provides a faster way to set up chain molecules.
The actual simulation falls apart again into two loops, the warmup loop with increasing
force capping, and the final simulation loop. Note that the electrostatic interaction is
only activated after equilibrating the excluded volume interactions, which speeds up the
warmup phase. However, depending on the problem, this splitted warmup may not be
possible due to physical restrictions. ESPResSo cannot detect these mistakes and it is
your responsibility to find simulation procedure suitable to your specific problem.
22
3. Getting, compiling and running
ESPResSo
This chapter will describe how to get, compile and run the ESPResSo software.
ESPResSo releases are available as source code packages from the ESPResSo home
page1 . This is where new users should get the code. The code within release packages is
tested and known to run on a number of platforms. Alternatively, people that want to
use the newest features of ESPResSo or that want to start contributing to the software
can instead obtain the current development code via the version control system software
git2 from ESPResSo’s project page at Github 3 . This code might be not as well tested
and documented as the release code; it is recommended to use this code only if you have
already gained some experience in using ESPResSo.
Unlike most other software, no binary distributions of ESPResSo are available, and the
software is usually not installed globally for all users. Instead, users of ESPResSo should
compile the software themselves. The reason for this is that it is possible to activate
and deactivate various features before compiling the code. Some of these features are
not compatible with each other, and some of the features have a profound impact on
the performance of the code. Therefore it is not possible to build a single binary that
can satisfy all needs. A user should always activate only those features that are actually
needed. This means, however, that learning how to compile ESPResSo is a necessary
evil. The build system of ESPResSo uses the GNU autotools, which are developed since
more than 20 years and allow to compile software easily on a wide range of platforms.
3.1. Running configure
The first step of building ESPResSo is to run the shell script configure which is to be
found in the top level source directory. The script collects all the information required
by the compilation process. It will determine how to use and where to find the compiler,
as well as the different libraries and tools required by the compilation process, and it
will test what compiler flags are to be used. The script will find out about most of these
things automatically. If something is missing, it will complain and give hints how to
solve the problem. The generic syntax of calling the configure script is:
configure [options ...] [variable=value ...]
1
http://espressomd.org
http://git.org
3
https://github.com/espressomd/espresso
2
23
If you are using the development source code from the git repository, before you
can call configure, it is necessary to have the GNU autotools (autoconf and automake)
installed. Then you can call the script bootstrap.sh from the top level source directory,
which will generate the configure script.
3.1.1. Source and build directories
Usually, when a program is compiled, the resulting binary files are put into the same
directory as the sources of the program. In ESPResSo’s build system, the source directory
that contains all the source files can be completely separated from the build directory,
where the files created by the build process are put. The location of the build directory
is the current working directory at the time when configure is called. In this way, you
can build several variants of ESPResSo, each variant having different activated features,
and for as many platforms as you want. All further commands concerning compiling
and running ESPResSo have to be called from the build directory. None of the files in
the source directory is ever modified when by the build process.
Example When the source directory is $srcdir (i.e. the files where unpacked to this
directory), then the build directory can be set to $builddir by calling the configurescript from there:
cd $builddir
$srcdir/configure
make
Espresso
3.1.2. Options and Variables
The behaviour of configure can be controlled by the means of command line options
and variables. In the following, only important command line options and variables
ESPResSo will be explained. For a complete list of options, variables and explanations
thereof, call
configure --help
--with-mpi=( yes | no | guess )/ --without-mpi By default, configure will automatically determine whether an MPI compiler is available. If it is, it will use it. If
you specify --without-mpi or --with-mpi=no, then MPI will not be used, even if
it is available.
--with-efence / --without-efence Whether or not to use the “electric fence” memory debugging library. 4 Efence is not used by default.
4
http://freshmeat.net/projects/efence/
24
--with-tcl=TCL By default, configure will automatically determine which version of
Tcl is used. If the wrong version is chosen automatically, you can specify the name
of the library with this option, e.g. tcl8.4.
--with-tk=TK / --without-tk By default, the GUI toolkit Tk is not used by ESPResSo.
This option can be used to activate Tk and to specify which Tk version to use,
e.g. tk8.4. If you only specify --with-tk and do not give a version number,
configure will try to automatically deduce the right version.
--with-fftw / --without-fftw This can be used to specify whether the FFTW library is to be used, and which version. By default, version 3 will be used if it is
found, otherwise version 2 is used. Note that quite a number of central features of
ESPResSo require FFTW.
--with-cuda=path / --without-cuda This switch enables CUDA support. path should
be the path to the CUDA directory, which can be omitted if it is the NVIDIA default path, i.e. /usr/local/cuda. The variable NVCCFLAGS can be used to define
compiler flags for the NVIDIA CUDA-compiler nvcc. For example, NVCCFLAGS
= "-gencode arch=compute_20,code=sm_20" will compile code only for Fermi
cards. Default is to compile for compute model 2.0, i.e. everything with a Fermi
chip or newer. Note that we require at least compute model 1.1, that is G90.
However, to use G90 (e. g. Tesla C1060), you need to manually specificy compute
model 1.1.
LDFLAGS=linker-flags This variable can be used to change the flags that the linker will
get when linking the ESPResSo binaries. This variable can be used to modify the
path where the compiler finds library files when they are installed in non-standard
places, e.g. LDFLAGS="-L/home/juser/lib".
CPPFLAGS=preprocessor-flags This variable can be used to change the flags that the
preprocessor will see. This variable can be used to modify the path wherer the
compiler finds include files when they are installed in non-standard places, e.g.
CPPFLAGS="-I/home/juser/include".
CXXFLAGS=C++-compiler flags This variable can be used to modify the compilation
flags, e.g. to change the optimization level for debugging (CXXFLAGS="-g -O0").
3.2. make: Compiling, testing and installing ESPResSo
The command make is mainly used to compile the ESPResSo source code, but it can do
a number of other things. The generic syntax of the make command is:
make [options] [target...] [variable=value]
When no target is given, the target all is used. The following targets are available:
25
all Compiles the complete ESPResSo source code. The variable myconf can be used to
specify the name of the configuration header to be used.
check Runs the testsuite. By default, all available tests will be run on 1, 2, 3, 4, 6,
or 8 processors. Which tests are run can be controlled by means of the variable
tests, which processor numbers are to be used can be controlled via the variable
processors. Note that depending on your MPI installation, MPI jobs can only
be run in the queueing system, so that ESPResSo will not run from the command
line. In that case, you may not be able to run the testsuite, or you have to directly
submit the testsuite script testsuite/test.sh to the queueing system.
Example: make check tests="madelung.tcl" processors="1 2"
will run the test madlung.tcl on one and two processors.
clean Deletes all files that were created during the compilation.
mostlyclean Deletes most files that were created during the compilation. Will keep for
example the built doxygen documentation and the ESPResSo binary.
dist Creates a .tar.gz-file of the ESPResSo sources. This will include all source files
as they currently are in the source directory, i.e. it will include local changes. This
is useful to give your version of ESPResSo to other people. The variable extra can
be used to specify additional files and directories that are to be included in the
archive file.
Example: make dist extra="myconfig.hpp internal"
will create the archive file and include the file myconfig.hpp and the directory
internal with all files and subdirectories.
install Install ESPResSo. The variables prefix and exec-prefix can be used to
specify the installation directories, otherwise the defaults defined by the configure
script are used. prefix sets the prefix where all ESPResSo files are to be installed,
exec-prefix sets the prefix where the executable files are to be installed and is
required only when there is an architecture-specific directory.
Example: make install prefix=/usr/local
will install all files below /usr/local.
uninstall Uninstalls ESPResSo, i.e. removes all files that were installed during make
install. The variables are identical to the variables of the install-target.
ug
Creates the User guide in the doc/ug subdirectory (only when using the development sources).
dg
Creates the Developers’ guide in the doc/dg subdirectory (only when using the
development sources).
doxygen
tutorials
26
Creates the Doxygen code documentation in the doc/doxygen subdirectory.
Creates the ESPResSo tutorials in the doc/tutorials subdirectory.
doc
Creates all documentation in the doc subdirectory (only when using the development sources).
A number of options are available when calling make. The most interesting option
is probably -j num_jobs , which can be used for parallel compilation on computers
that have more than one CPU or core. num jobs specifies the maximal number of jobs
that will be run. Setting num jobs to the number of available processors speeds up the
compilation process significantly.
3.3. Running ESPResSo
When ESPResSo is found in your path, it can be run via
Espresso [tcl script [args]]
When ESPResSo is called without any arguments, it is started in the interactive mode,
where new commands can be entered on the command line. When the name of a tcl script is given, the script is executed. Any further arguments are passed to the script.
If you want to run ESPResSo in parallel using MPI, the actual invocation depends on
your MPI implementation. In many cases, e.g. OpenMPI, the command will be
mpiexec -n n nodes Espresso [tcl script [args]]
where n nodes denotes the number of MPI nodes to be used. However, note that depending on your MPI installation, MPI jobs can only be run in a queueing system, so
that ESPResSo will not run from the command line. Also, older installations sometimes
require “-np” instead of “-n” or “mpirun” instead of “mpiexec”.
3.4. myconfig.hpp: Activating and deactivating features
ESPResSo has a large number of features that can be compiled into the binary. However,
it is not recommended to actually compile in all possible features, as this will slow down
ESPResSo significantly. Instead, compile in only the features that are actually required.
A strong gain in speed can be achieved, by disabling all non-bonded interactions except
for a single one, e.g. LENNARD_JONES. For the developers, it is also possible to turn on or
off a number of debugging messages. The features and debug messages can be controlled
via a configuration header file that contains C-preprocessor declarations. Appendix B
on page 235 lists and describes all available features. The file myconfig-sample.hpp
that configure will generate in the build directory contains a list of all possible features
that can be copied into your own configuration file. When no configuration header is
provided by the user, a default header, found in src/core/myconfig-default.hpp, will
be used that turns on the default features.
When you distinguish between the build and the source directory, the configuration
header can be put in either of these. Note, however, that when a configuration header
is found in both directories, the one in the build directory will be used.
27
By default, the configuration header is called myconfig.hpp. The name of the configuration header can be changed either when the configure-script is called via the
variable MYCONFIG (see section 3.1 on page 23), or when make is called with the setting
myconfig=myconfig header (see section 3.2 on page 25).
The configuration header can be used to compile different binary versions of ESPResSo
with a different set of features from the same source directory. Suppose that you have
a source directory $srcdir and two build directories $builddir1 and $builddir2 that
contain different configuration headers:
• $builddir1/myconfig.hpp:
#define ELECTROSTATICS
#define LENNARD-JONES
• $builddir2/myconfig.hpp:
#define LJCOS
Then you can simply compile two different versions of ESPResSo via
cd $builddir1
$srcdir/configure
make
cd $builddir2
$srcdir/configure
make
28
4. Setting up particles
4.1. part: Creating single particles
4.1.1. Defining particle properties
Syntax
part pid [pos x y z ] [type typeid ] [v vx vy vz ] [f fx fy fz ]
[bond bondid pid2 ...] [q charge] 1 [quat q1 q2 q3 q4 ] 2
[omega_body/lab x y z ] 2 [torque_body/lab x y z ] 2
[rinertia x y z ] 2 [[un]fix x y z ] 3 [ext_force x y z ] 3
[ext_torque x y z ] 2,3 [exclude pid2 ...] 4 [exclude delete pid2 ...] 4
[mass mass] 5 [dipm moment] 6 [dip dx dy dz ] 6 [virtual v ] 7,8
[vs_relative pid distance] 8 [vs_auto_relate_to pid ] 8 [temp T ] 9
[gamma g] 9 [rotation rot] 10 [solvation lA kA lB kB ] 11
[swimming ( ( v_swim v swim | f_swim f swim ) | off )] 12
[swimming ( ( v_swim v swim | f_swim f swim ) ( pusher | puller )
dipole_length dipole length rotational_friction rotational friction |
off )] 12,13
Required features: 1 ELECTROSTATICS 2 ROTATION 3 EXTERNAL_FORCES 4 EXCLUSION
5 MASS 6 DIPOLES 7 VIRTUAL_SITES_COM 8 VIRTUAL_SITES_RELATIVE
9 LANGEVIN_PER_PARTICLE 10 ROTATION_PER_PARTICLE 11 SHANCHEN
12 ENGINE 13 LB or LB_GPU
Description
This command modifies particle data, namely position, type (monomer, ion, . . . ), charge,
velocity, force and bonds. Multiple properties can be changed at once. If you add a new
particle the position has to be set first because of the spatial decomposition.
Arguments
• pid
• [pos x y z ] Sets the position of this particle to (x, y, z).
• [type typeid ] Restrictions: typeid ≥ 0.
The typeid is used in the inter command (see section 5 on page 50) to define the
parameters of the non bonded interactions between different kinds of particles.
• [v vx vy vz ] Sets the velocity of this particle to (vx, vy, vz). The velocity remains
variable and will be changed during integration.
29
• [f fx fy fz ] Set the force acting on this particle to (f x, f y, f z). The force remains
variable and will be changed during integration. However, whereas the velocity
is modified with respect to the velocity you set upon integration, the force it
recomputed during the integration step and any force set in this way is lost
during the integration step.
• [bond bondid pid2 ...] Restrictions: bondid ≥ 0; pid2 must be an existing particle. The bondid is used for the inter command to define bonded interactions.
• bond delete Will delete all bonds attached to this particle.
• [q charge] Sets the charge of this particle to q.
• [quat q1 q2 q3 q4 ] Sets the quaternion representation of the rotational position
of this particle.
• [omega_body] x y z ( [ ])omega_body x y z The command [omega body] sets
the angular momentum of this particle in the particle’s co-rotating frame (or
body frame) and the command [omega lab] sets it for the particle in the fixed
frame (or laboratory frame). If you set the angular momentum of the particle in
the lab frame, the orientation of the particle ([quat]) must be set before invoking
[omega lab], otherwise the conversion from lab to body frame will not be handled
properly.
• [torque_body/lab x y z ] The command [torque body] sets the torque of this
particle in the particle’s co-rotating frame (or body frame) and the command
[torque lab] sets it for the particle in the fixed frame (or laboratory frame). If
you set the torque of the particle in the lab frame, the orientation of the particle
([quat]) must be set before invoking [torque lab], otherwise the conversion from
lab to body frame will not be handled properly.
• [rinertia x y z ] Sets the diagonal elements of this particles rotational inertia
tensor. These correspond with the inertial moments along the coordinate axes in
the particle’s co-rotating coordinate system. When the particle’s quaternions are
set to 1 0 0 0, the co-rotating and the fixed (lab) frame are co-aligned.
• [fix x y z ] Fixes the particle in space. By supplying a set of 3 integers as arguments it is possible to fix motion in x , y, or z coordinates independently. For
example fix 0 0 1 will fix motion only in z. Note that fix without arguments is
equivalent to fix 1 1 1.
• [unfix] Release any external influence from the particle.
• [ext_force x y z ] An additional external force is applied to the particle.
• [ext_torque x y z ] An additional external torque is applied to the particle. This
torque is specified in the laboratory frame!
• [exclude pid2 ...+] Restrictions: pid2 must be an existing particle. Between
the current particle an the exclusion partner(s), no nonbonded interactions are
30
calculated. Note that unlike bonds, exclusions are stored with both partners.
Therefore this command adds the defined exclusions to both partners.
• [exclude delete pid2 ...] Searches for the given exclusion and deletes it. Again
deletes the exclusion with both partners.
• [mass mass] Sets the mass of this particle to mass. If not set, all particles have
a mass of 1 in reduced units.
• [dipm moment] Sets the dipol moment of this particle to moment.
• [dip dx dy dz ] Sets the orientation of the dipole axis to (dx, dy, dz).
• [virtual v ] Declares the particles as virtual (1) or non-virtual (0, default). Please
read chapter 4.4 before using virtual sites.
• [vs_auto_relate_to pid ] Automatically relates a virtual site to a non-virtual
particle for the “relative” implementation of virtual sites. pid is the id of the
particle to which the virtual site should be related.
• [vs_relative pid distance] Allows for manual access to the attributes of virtual
sites in the “relative” implementation. pid denotes the id of the particle to which
this virtual site is related and distance the distance between non-virtual and
virtual particle.
• [temp T ] If used in combination with the Langevin thermostat (as documented in
section 6.2), sets the temperature T individually for the particle with id pid . This
allows to simulate systems containing particles of different temperatures. Caution:
this has no influence on any other thermostat then the Langevin thermostat.
• [gamma g] If used in combination with the Langevin thermostat (as documented
in section 6.2), sets the frictional coefficient T individually for the particle with id
pid . This allows to simulate systems containing particles with different diffusion
constants. Caution: this has no influence on any other thermostat then the
Langevin thermostat.
• [rotation rot] Specifies whether a particle’s rotational degrees of freedom are
integrated (value of 1) or not (0). If set to zero, the content of the torque and
omega variables are meaningless. The default is 1.
• [solvation lA kA lB kB ] Sets the four solvation coupling constants for the two
components of a Shan-Chen fluid, as documented in Section 12.4.
• [swimming ( ( v_swim v swim | f_swim f swim ) | off )] Enables the particle to be self-propelled in the direction determined by its quaternion. For setting
the quaternion of the particle see quat. The self-propulsion speed will relax to
a constant velocity, that is specified by v_swim. Alternatively it is possible to
achieve a constant velocity by imposing a constant force term f_swim that is balanced by friction of a (Langevin) thermostat. The way the velocity of the particle
decays to the constant terminal velocity in either of these methods is completely
31
determined by the friction coefficient. You may only set one of the possibilities
v_swim or f_swim as you cannot relax to constant force and constant velocity at
the same time. The option off (re)sets v swim and f swim both to 0.0 and thus
disables swimming. This option applies to all non-lattice-Boltzmann thermostats.
Note that there is no real difference between v_swim and f_swim, since the latter
may aways be chosen such that the same terminal velocity is achieved for a given
friction coefficient.
• [swimming ( ( v_swim v swim | f_swim f swim ) ( pusher | puller )
dipole_length dipole length rotational_friction rotational friction |
off )]
For an explanation of the parameters v_swim, f_swim and off see the previous
item. In lattice-Boltzmann self-propulsion is less trivial than for normal MD, because the self-propulsion is achieved by a force-free mechanism, which has strong
implications for the far-field hydrodynamic flow field induced by the self-propelled
particle. In ESPResSo only the dipolar component of the flow field of an active
particle is taken into account. This flow field can be generated by a pushing
or a pulling mechanism, leading to change in the sign of the dipolar flow field
with respect to the direction of motion. You can specify the nature of the particle’s flow field by using the keywords pusher or puller. You will also need to
specify a dipole length which determines the distance of the source of propulsion
from the particle’s center. Note that you should not put this distance to zero;
ESPResSo (currently) does not support mathematical dipole flow fields. The key
rotational_friction can be used to set the friction that causes the orientation
of the particle to change in shear flow. The torque on the particle is determined
by taking the cross product of the difference between the fluid velocity at the center of the particle and at the source point and the vector connecting the center
and source.
You may ask: “Why are there two methods v swim and f swim for the selfpropulsion using the lattice-Bolzmann algorithm?” The answer is straightforward.
When a particle is accelerating, it has a monopolar flow-field contribution which
vanishes when it reaches its terminal velocity (for which there will only be a
dipolar flow field). The major difference between the above two methods is that
with v swim the flow field only has a monopolar moment and only while the
particle is accelerating. As soon as the particle reaches a constant speed (given
by v swim) this monopolar moment is gone and the flow field is zero! In contrast,
f swim always, i.e., while accelerating and while swimming at constant force
possesses a dipolar flow field.
Please note that even though swimming is interoperable with the CPU version
of LB it is only supported on less than three Open MPI nodes, i.e. n_nodes ≤ 1.
Warning: The options [omega], [torque], and [tbf] are deprecated and will
be removed in some future version.
32
4.1.2. Getting particle properties
Syntax
(1) part pid print [( id | pos | type | folded_position | type | q |
v | f | torque_body | torque_lab | body_frame_velocity | fix |
ext_force | ext_torque | bond | exclusions connections [range] |
swimming )]...
(2) part
Description
Variant (1) will return a list of the specified properties of particle pid , or all properties,
if no keyword is specified. Variant (2) will return a list of all properties of all particles.
Note that there is a difference between the *_body and *_lab. The first prints the
variable in the co-rotating frame, whereas the second gives the variable in the stationary
frame, the body and laboratory frames, respectively. One would typically want to output
the variable in the laboratory frame, since it is the frame of interest. However for some
tests involving reading and writing the variable it may be desireable to know it in the
body frame as well. Be careful with reading and writing, if you write in the lab frame,
then read in the lab frame. If you are setting the variable in the lab frame, the orientation
of the particle’s quat must be set before, otherwise the conversion from lab to body frame
will not be handled properly. Also be careful about the order in which you write and
read in data from a blockfile, for instance if you output the variable in both frames!
The body_frame_velocity command is a print-only command that gives the velocity
in the body frame, which can be useful for determining the translational diffusion tensor
of an anisotropic particle via the velocity auto-correlation (Green-Kubo) method.
Example
part 40 print id pos q bonds
will return a list like
40 8.849 1.8172 1.4677 1.0 {}
This routine is primarily intended for effective use in Tcl scripts.
When the keyword connection is specified, it returns the connectivity of the
particle up to range (defaults to 1). For particle 5 in a linear chain the result up to
range = 3 would look like:
{ { 4 } { 6 } } { { 4 3 } { 6 7 } } { {4 3 2 } { 6 7 8 } }
The function is useful when you want to create bonded interactions to all other particles
a certain particle is connected to. Note that this output can not be used as input to the
part command. Check results if you use them in ring structures.
If none of the options is specified, it returns all properties of the particle, if it
exists, in the form
0 pos 2.1 6.4 3.1 type 0 q -1.0 v 0.0 0.0 0.0 f 0.0 0.0 0.0
bonds { {0 480} {0 368} ... }
33
which may be used as an input to this function later on. The first integer is the particle
number.
Variant (2) returns the properties of all stored particles in a tcl-list with the same
format as specified above:
{0 pos 2.1 6.4 3.1 type 0 q -1.0 v 0.0 0.0 0.0 f 0.0 0.0 0.0
bonds{{0 480}{0 368}...}}
{1 pos 1.0 2.0 3.0 type 0 q 1.0 v 0.0 0.0 0.0 f 0.0 0.0 0.0
bonds{{0 340}{0 83}...}}
{2...{{...}...}}
{3...{{...}...}}
...
When using pos, the particle position returned is unfolded, for convenience in diffusion calculations etc. Note that therefore blockfiles will contain imaged positions, but
un-imaged velocities, which should not be interpreted together. However, that is fine for
restoring the simulation, since the particled data is loaded the same way.
4.1.3. Deleting particles
Syntax
(1) part pid delete
(2) part deleteall
Description
In variant (1), the particle pid is deleted and all bonds referencing it. Variant (2)
will delete all particles currently present in the simulation. Variant (3) will delete all
currently defined exclusions.
4.1.4. Exclusions
Syntax
(1) part auto_exclusions [range]
(2) part delete_exclusions
Required features:
EXCLUSIONS
Description
Variant (1) will create exclusions for all particles pairs connected by not more than
range bonds (range defaults to 2). This is typically used in atomistic simulations, where
nearest and next nearest neighbour interactions along the chain have to be omitted since
they are included in the bonding potentials. For example, if the system contains particles
0 . . . 100, where particle n is bonded to particle n − 1 for 1 ≤ n ≤ 100, then it will result
in the exclusions:
• particle 1 does not interact with particles 2 and 3
34
• particle 2 does not interact with particles 1, 3 and 4
• particle 3 does not interact with particles 1, 2, 4 and 5
• ...
Variant (2) deletes all exclusions currently present in the system.
4.2. Creating groups of particle
4.2.1. polymer: Setting up polymer chains
Syntax
polymer num polymers monomers per chain bond length
[start pid ] [pos x y z ] [mode ( RW | SAW | PSAW ) [shield [trymax ]]]
[charge q] 1 [distance dcharged ] 1 [types typeidneutral [typeidcharged ]]
[bond bondid ] [angle φ [θ [x y z ]]] [constraints] 2
Required features:
1 ELECTROSTATICS
2 CONSTRAINTS
Description
This command will create num polymers polymer or polyelectrolyte chains with monomers per chain
monomers per chain. The length of the bond between two adjacent monomers will be
set up to be bond length.
Arguments
• num polymers Sets the number of polymer chains.
• monomers per chain Sets the number of monomers per chain.
• bond length Sets the initial distance between two adjacent monomers. The distance during the course of the simulation depends on the applied potentials. For
fixed bond length please refer to the Rattle Shake algorithm[2]. The algorithm is
based on Verlet algorithm and satisfy internal constraints for molecular models
with internal constrains, using Lagrange multipliers.
• [start pid ] Sets the particle number of the start monomer to be used with the
part command. This defaults to 0.
• [pos x y z ] Sets the position of the first monomer in the chain to x , y, z (defaults
to a randomly chosen value)
• [mode ( RW | PSAW | SAW ) [shield [trymax ]]] Selects the setup mode:
RW (Random walk) The monomers are randomly placed by a random walk with
a steps size of bondl ength.
PSAW (Pruned self-avoiding walk) The position of a monomer is randomly chosen in a distance of bond length to the previous monomer. If the position is
closer to another particle than shield , the attempt is repeated up to trymax
35
times. Note, that this is not a real self-avoiding random walk, as the particle
distribution is not the same. If you want a real self-avoiding walk, use the
SAW mode. However, PSAW is several orders of magnitude faster than SAW,
especially for long chains.
SAW (Self-avoiding random walk) The positions of the monomers are chosen as
in the plain random walk. However, if this results in a chain that has a
monomer that is closer to another particle than shield , a new attempt of
setting up the whole chain is done, up to trymax times.
The default for the mode is RW, the default for the shield is 1.0, and the default
for trymax is 30000, which is usually enough for PSAW. Depending on the length
of the chain, for the SAW mode, trymax has to be increased by several orders of
magnitude.
• [charge valency] Sets the valency of the charged monomers. If the valency of
the charged polymers valency is smaller than 10−10 , the charge is assumed to be
zero, and the types are set to typeidcharged = typeidneutral . If charge is not set, it
defaults to 0.0.
• [distance dcharged ] Sets the stride between the indices of two charged monomers.
This defaults defaults to 1, meaning that all monomers in the chain are charged.
• [types typeidneutral typeidcharged ] Sets the type ids of the neutral and charged
monomer types to be used with the part command. If only typeidneutral is defined,
typeidcharged defaults to 1. If the option is omitted, both monomer types default
to 0.
• [bond bondid ] Sets the type number of the bonded interaction to be set up between
the monomers. This defaults to 0. Any bonded interaction, no matter how many
bonding-partners needed, is stored with the second particle in this bond. See
chapter 5.3.
• [angle φ [θ [x y z ]]] Allows for setting up helices or planar polymers: φ and
theta are the angles between adjacent bonds. x , y and z set the position of the
second monomer of the first chain.
• [constraints] If this option is specified, the particle setup-up tries to obey previously defined constraints (see section 4.3 on page 42).
4.2.2. counterions: Setting up counterions
Syntax
counterions N [start pid ] [mode ( SAW | RW ) [shield [trymax ]]]
[charge val ] 1 [type typeid ]
Required features:
1 ELECTROSTATICS
Description
This command will create N counterions in the simulation box.
36
Arguments
• [start pid ] Sets the particle id of the first counterion. It defaults to the current number of particles, i.e. counterions are placed after all previously defined
particles.
• [mode ( SAW | RW ) [shield [trymax ]]] Specifies the setup method to place the
counterions. It defaults to SAW. See the polymer command for a detailed description.
• [charge val ] Specifies the charge of the counterions. If not set, it defaults to −1.0.
• [type typeid ] Specifies the particle type of the counterions. It defaults to 2.
4.2.3. salt: Setting up salt ions
Syntax
salt N+ N− [start pid ] [mode ( SAW | RW ) [shield [trymax ]]]
[charges val+ [val− ]] 1 [types typeid+ [typeid− ]] [rad r ]
Required features:
1 ELECTROSTATICS
Description
Create N+ positively and N− negatively charged salt ions of charge val+ and val− within
the simulation box.
Arguments
• [start pid ] Sets the particle id of the first (positively charged) salt ion. It defaults
to the current number of particles.
• [mode ( SAW | RW ) [shield [trymax ]]] Specifies the setup method to place the
counterions. It defaults to SAW. See the polymer command for a detailed description.
• [charge val+ [val− ]] Sets the charge of the positive salt ions to val+ and the one
of the negatively charged salt ions to val− . If not set, the values default to 1.0
and −1.0, respectively.
• [type typeid+ [typeid− ]] Specifies the particle type of the salt ions. It defaults to
3 respectively 4.
• [rad r ] The salt ions are only placed in a sphere with radius r around the origin.
4.2.4. diamond: Setting up diamond polymer networks
Syntax
diamond a bond length monomers per chain [counterions NCI ]
[charges valnode valmonomer valCI ] 1 [distance dcharged ] 1 [nonet]
Required features:
1 ELECTROSTATICS
37
Description
Creates a diamond-shaped polymer network with 8 tetra-functional nodes connected by
2 ∗ 8 polymer chains of length monomers per chain in a unit cell of length a. Chain
monomers are placed at a mutual distance bond length along the vector connecting
network nodes. The polymer is created starting from particle ID 0. Nodes are assigned
type 0, monomers (both charged and uncharged) are type 1 and counterions type 2. For
inter-particle bonds interaction 0 is taken which must be a two-particle bond.
Figure 4.1.: Diamond-like polymer network with monomers per chain=15.
Arguments
• a Determines the size of the of the unit cell.
• bond length Specifies the bond length of the polymer chains connecting the 8
tetra-functional nodes.
• monomers per chain Sets the number of chain monomers between the functional
nodes.
• [counterions NCI ] Adds NCI counterions to the system.
• [charges valnode valmonomer valCI ] Sets the charge of the nodes to valnode , the
charge of the connecting monomers to valmonomer , and the charge of the counterions to valCI .
• [distance dcharged ] Specifies the distance between charged monomers along the
interconnecting chains. If dcharged > 1 the remaining chain monomers are uncharged.
• [nonet] Do not create bonds between the chains.
38
4.2.5. icosaeder: Setting up an icosaeder
Syntax
icosaeder a monomers per chain [counterions NCI ]
[charges valmonomers valCI ] 1 [distance dcharged ] 1
Required features:
1 ELECTROSTATICS
Description
Creates a modified icosaeder to model a fullerene (or soccer ball). The edges are modeled
by polymer chains connected at the corners of the icosaeder. For inter-particle bonds
interaction 0 is taken which must be a two-particle bond. Two particle types are used
for the pentagons and the interconnecting links. For an example, see figure 4.2.
Figure 4.2.: Icosaeder with monomers per chain=15.
Arguments
• a Length of the links. Defines the size of the icosaeder.
• monomers per chain Specifies the number of chain monomers along one edge.
• [counterions NCI ] Specifies the number of counterions to be placed into the system.
• [charges valmonomers valCI ] Set the charges of the monomers to valmonomers and
the charges of the counterions to valCI .
• [distance dcharged ] Specifies the distance between two charged monomer along
the edge. If dcharged > 1 the remaining monomers are uncharged.
39
4.2.6. crosslink: Cross-linking polymers
Syntax
crosslink num polymer monomers per chain [start pid ] [catch rcatch ]
[distLink link dist] [distChain chain dist] [FENE bondid ]
[trials trymax ]
Description
Attempts to end-crosslink the current configuration of num polymer equally long polymers with monomers per chain monomers each, returning how many ends are successfully connected.
Arguments
• [start pid ] pid specifies the first monomer of the chains to be linked. It has to
be specified if the polymers do not start at id 0.
• [catch rc atch] Set the radius around each monomer which is searched for possible
new monomers to connect to. rcatch defaults to 1.9.
• [distLink link dist] The minimal distance of two interconnecting links. It defaults to 2.
• [distChain chain dist] The minimal distance for an interconnection along the
same chain. It defaults to 0. If set to monomers per chain, no interchain connections are created.
• [FENE bondid ] Sets the bond type for the connections to bondid .
• [trials trymax ] If not specified, trymax defaults to 30000.
4.2.7. copy_particles: copying a set of particles
Syntax
copy_particles [set id1 id2 ...| range from to ...] [shift s x s y s z ]
Description
Copy a group of particles including their bonds. Positions can be [shift]ed by an
offset (s x , s y, s z ), otherwise the copied set is at exactly the same position as
the original set. The particles can be given as a combination of [list]s or [range]s.
The new particles obtain in any case consecutive identities after the largest current
identity. The mapping of the particles is returned as a list of old-new pairs, which
can be conveniently read into an array:
array set newidentities [copy_particles ...]
puts "particle 42 is now at position $newidentities(42)"
Bonds within the defined particle set are copied with translated identities, but not
bonds with particles outside the list. That is, if the particle set corresponds to a molecule,
intramolecular bonds are preserved, but not intermolecular ones.
40
Examples of use:
copy_particles set {1 2 3 4} shift 0.0 0.0 0.0
copy_particles set {1 2} set {3 4}
copy_particles range 1 4
All these examples do the same - making exact copies of particles 1 through 4.
41
4.3. constraint: Setting up constraints
Syntax
(1) constraint wall normal nx ny nz dist d type id [penetrable flag]
[reflecting flag] [only_positive flag] [tunable_slip flag]
(2) constraint sphere center cx cy cz radius rad direction direction
type id [penetrable flag] [reflecting flag]
(3) constraint cylinder center cx cy cz axis nx ny nz radius rad
length length direction direction type id [penetrable flag]
[reflecting flag]
(4) constraint rhomboid corner px py pz a ax ay az b bx by bz
c cx cy cz direction direction type id [penetrable flag]
[reflecting flag]
(5) constraint maze nsphere n dim d sphrad rs cylrad rc type id
[penetrable flag]
(6) constraint pore center cx cy cz axis nx ny nz radius rad length
length type id
(7) constraint stomatocyte center x y z orientation ox oy oz
outer_radius Ro inner_radius Ri layer_width w direction
direction type id [penetrable flag] [reflecting flag]
(8) constraint slitpore pore_mouth z channel_width c
pore_width w pore_length l upper_smoothing_radius us
lower_smoothing_radius ls
(9) constraint rod center cx cy lambda lambda 1
(10) constraint plate height h sigma sigma 1
(11) constraint ext_magn_field fx fy fz 2,3
(12) constraint plane cell x y z type id
(13) constraint mindist_position x y z
(14) constraint hollow_cone center x y z orientation ox oy oz
outer_radius Ro inner_radius Ri width w opening_angle alpha
direction direction type id [penetrable flag] [reflecting flag]
(15) constraint spherocylinder center cx cy cz axis nx ny nz radius
rad length length direction direction type id [penetrable flag]
[reflecting flag]
(16) constraint mindist_position x y z
Required features:
CONSTRAINTS
1 ELECTROSTATICS
2 ROTATION
3 DIPOLES
Description
The constraint command offers a variety of surfaces that can be defined to interact with
desired particles. Variants (1) to (7) create interactions via a non-bonded interaction
potential, where the distance between the two particles is replaced by the distance of
the center of the particle to the surface. The constraints are identified like a particle via
its type for the non-bonded interaction. After a type is defined for each constraint one
has to define the interaction of all different particle types with the constraint using the
42
inter command. In variants (1) to (7), constraints are able to be penetrated if flag is set
to 1. Otherwise, when the penetrable option is ignored or flag is set to 0, the constraint
cannot be violated, i.e. no particle can go through the constraint surface. In variants
(1) to (4) and (7) it is also possible to specify a flag indicating if the constraints should
be reflecting. The flags can equal 1 or 2. The flag 1 corresponds to a reflection process
where the normal component of the velocity is reflected and the tangential component
remains unchanged. If the flag is 2, also the tangential component is turned around, so
that a bounce back motion is performed. The second variant is useful for boundaries
of DPD. The reflection property is only activated if an interaction is defined between a
particular particle and the constraint! This will usually be a lennard-jones interaction
with = 0, but finite interaction range.
In variant (1) if the only_positive flag is set to 1, interactions are only calculated if
the particle is on the side of the wall in which the normal vector is pointing. This has only
an effect for penetrable walls. If the tunable_slip flag is set to 1, then slip boundary
interactions apply that are essential for microchannel flows like the Plane Poiseuille or
Plane Couette Flow. You also need to use the tunable slip interaction (see 5.9.1) for this
too work.
Variants (9) and (10) create interactions based on electrostatic interactions. The
corresponding force acts in direction of the normal vector of the surface and applies to
all charged particles.
Variant (11) does not define a surface but is based on magnetic dipolar interaction
with an external magnetic field. It applies to all particles with a dipole moment.
The resulting surface in variant (1) is a plane defined by the normal vector nx ny nz
and the distance d from the origin. The force acts in direction of the normal. Note
that the d describes the distance from the origin in units of the normal vector so that
the product of d and n is a point on the surface. Therefore negative distances are quite
common!
The resulting surface in variant (2) is a sphere with center cx cy cz and radius rad .
The direction determines the force direction, -1 or [inside] for inward and +1 or [outside]
for outward.
The resulting surface in variant (3) is a cylinder with center cx cy cz and radius rad .
The length parameter is half of the cylinder length. The axis is a vector along the
cylinder axis, which is normalized in the program. The direction is defined the same
way as for the spherical constraint.
The resulting surface in variant (4) is a rhomboid, defined by one corner located at
px py pz and three adjacent edges, defined by the three vectors connecting the corner p
with it’s three neighboring corners, a (ax ay az ), b (bx by bz ) and c (cx cy cz ).
The resulting surface in variant (5) is n spheres of radius rs along each dimension,
connected by cylinders of radius rc . The spheres have simple cubic symmetry. The
spheres are distributed evenly by dividing the boxl by n. Dimension of the maze can be
controlled by d : 0 for one dimensional, 1 for two dimensional and 2 for three dimensional
maze.
Variant (6) sets up a cylindrical pore similar to variant (3) with a center cx cy cz and
radius rad . The length parameter is half of the cylinder length. The axis is a vector
43
along the cylinder axis, which is normalized in the program. The argument radius rad
can be replaced by the argument radii rad1 rad2 to obtain a pore with a conical
shape and corresponding opening radii. The first radius is in the direction opposite to
the axis vector.
Variant (7) creates a stomatocyte shaped boundary. This command should be used
with care. The position can be any point in the simulation box, and the orientation of the
(cylindrically symmetric) stomatocyte is given by a vector, which points in the direction
of the symmetry axis, it does not need to be normalized. The parameters: outer radius
Ro, inner radius Ri , and layer width w , specify the shape of the stomatocyte. Here
inappropriate choices of these parameters can yield undersired results. The width is
used as a scaling parameter. That is, a stomatocyte given by Ro:Ri :w = 7:3:1 is half the
size of the stomatocyte given by 7:3:2. Not all choices of the parameters give reasonable
values for the shape of the stomatocyte, but the combination 7:3:1 is a good point to
start from when trying to modify the shape.
In variant (8), a slit-shaped pore in a T-orientation to a flat channel is created. The
geometry is depicted in Fig. 4.3. It translationally invariant in y direction. The pore
(lower vertical part) extends in z-direction, and the channel (upper horizontal part).
The pore mouth is defined as the z-coordinate, where the lower plane of the channel
and the slit pore intersect. It is always centered in the x-direction. A corresponding
dielectric command decorates the surface with surface charges that can be calculated
with the ICC? algorithm.
channel_width
pore_mouth
upper
pore_width
_smoothing
_radius
lower_smoothing
_radius
Figure 4.3.: The slitpore created by the constraint slitpore.
Variant (8) specifies an electrostatic interaction between the charged particles in the
system to an infinitely long rod with a line charge of lambda which is alinge along the
z-axis and centered at cx and cy .
Variant (9) specifies the electrostatic interactinos between the charged particles in the
system and an inifinitely large plate in the x-y-plane at height h. The plate carries a
charge density of sigma.
44
Variant (10) specifies the dipolar coupling of particles with a dipolar moment to an
external field fx fy fz .
Variant (11) creates an infinite plane at a fixed position. For non-initializing a direction of the constraint values of the positions have to be negative. For the tunable-slip
boundary interactions you have to set two constraints.
Variant (14) creates a hollow-cone shaped boundary. The position can be any point
in the simulation box, and the orientation of the (cylindrically symmetric) cone is given
by a vector, which points in the direction of the symmetry axis, it does not need to be
normalized. The parameters: outer radius Ro, inner radius Ri , width w , and opening angle alpha, specify the shape of the object. The inner radius gives the narrow end
opening size, the outer radius the length of the shaft, and the width the layer width, i.e.,
the thickness of the cone. The opening angle (between 0 and π/2) specifies the angle of
the cone.
Variant (15) creates a spherocylinder, that is, a cylinder capped by a hemisphere on
either side. The parameter length length specifies the length of the shaft, excluding the
two hemispherical caps.
Variant (16) calculates the smallest distance to all non-penetrable constraints, that can
be repulsive (wall, cylinder, sphere, rhomboid, maze, pore, slitpore). Negative distances
mean that the position is “within” the area that particles should not access. Helpful to
find initial configurations.)
Example
To create an infinite plane in z-direction at z = 20.0 of type id 1, use:
constraint plane cell -10 -10 20 type 1
4.3.1. Deleting a constraint
Syntax
constraint delete [num]
Description
This command will delete constraints. If num is specified only this constraint will
deleted, otherwise all constraints will be removed from the system.
4.3.2. Getting the force on a constraint
Syntax
constraint force n
Description
Returns the force acting on the nth constraint. Note, however, that this are only forces
due to interactions with particles, not with other constraints. Also, these forces still
do not mean that the constraints move, they are just the negative of the sum of forces
45
acting on all particles due to this constraint. Similarly, the total energy does not containt
constraint-constraint contributions.
4.3.3. Getting the currently defined constraints
Syntax
constraint [num]
Description
Prints out all constraint information. If num is specified only this constraint is displayed,
otherwise all constraints will be printed.
4.3.4. harmonic_force: Creating a harmonic trap
Syntax
harmonic_force { x y z } k
Required features:
CUDA
Description
Calculates a spring force for all particles, where the equilibrium position of the spring
is at xyz and it’s force constant is k . A more flexible trap can be constructed with
constraints, but this one runs on the GPU.
4.4. Virtual sites
Virtual sites are particles, the positions and velocities of which are not obtained by
integrating an equation of motion. Rather, their coordinates are obtained from the
position (and orientation) of one or more other particles. In this way, rigid arrangements
of particles can be constructed and a particle can be placed in the center of mass of a set
of other particles. Virtual sites can interact with other particles in the system by means
of interactions. Forces are added to them according to their respective particle type.
Before the next integration step, the forces accumulated on a virtual site are distributed
back to those particles, from which the virtual site was derived.
There are two distinct types of virtual sites, described in the following.
4.4.1. Virtual sites in the center of mass of a molecule
To activate this implementation, enable the feature VIRTUAL_SITES_COM (sec. 3.4).
Virtual sites are then placed in the center of mass of a set of particles (as defined below).
Their velocity will also be that of the center of mass. Forces accumulating on the virtual
sites are distributed back to the particles which form the molecule. To place a virtual
site at the center of a molecule, perform the following steps in that order
46
1. Create a particle of the desired type for each molecule. It should be placed at least
roughly in the center of the molecule to make sure, it’s on the same node as the
other particles forming the molecule, in a simulation with more than one cpu.
2. Make it a virtual site using
part pid virtual 1
3. Declare the list of molecules and the particles they consist of:
eval analyze set {molid {listofparticleids..} ...}
The lists of particles in a molecule comprise the non-virtual particles and the virtual site.
4. Assign to all particles that belong to the same molecule a common molecule id
part pid mol molid
5. Update the position of all virtual particles (optional)
integrate 0
Please note that the use of virtual sites requires that the particles are numbered consecutively. I.e., the particle ids should go from zero to N − 1, where N is the number of
particles.
4.4.2. Rigid arrangements of particles
The “relative” implementation of virtual sites allows for the simulation of rigid arrangements of particles. It can be used, e.g., for extended dipoles and raspberry-particles, but
also for more complex configurations. Position and velocity of a virtual site are obtained
from the position and orientation of exactly one non-virtual particle, which has to be
placed in the center of mass of the rigid body. Several virtual sites can be related to one
and the same non-virtual particle. The position of the virtual site is given by
x~v = x~n + On (Ov E~z )d,
(4.1)
where x~n is the position of the non-virtual particle, On is the orientation of the nonvirtual particle, Ov denotes the orientation of the vector x~v − x~n with respect to the nonvirtual particle’s body fixed frame and d the distance between virtual and non-virtual
particle. In words: The virtual site is placed at a fixed distance from the non-virtual
particle. When the non-virtual particle rotates, the virtual sites rotates on an orbit
around the non-virtual particle’s center.
To use this implementation of virtual sites, activate the feature VIRTUAL_SITES_RELATIVE (see sec. 3.4). To set up a virtual site,
1. Place the particle to which the virtual site should be related. It needs to be in the
center of mass of the rigid arrangement of particles you create. Let its particle id
be n.
47
2. Place a particle at the desired relative position, make it virtual and relate it to the
first particle
part v pos pos virtual 1 vs_auto_relate n
3. Repeat the previous step with more virtual sites, if desired.
4. To update the positions of all virtual sites, call
integrate 0
Please note:
• The relative position of the virtual site is defined by its distance from the nonvirtual particle, the id of the non-virtual particle and a quaternion which defines
the vector from non-virtual particle to virtual site in the non-virtual particle’s
body-fixed frame. The first two are saved in the virtual site’s vs relative-attribute,
while the latter is saved in the quaternion attribute. Take care, not to overwrite
these after using vs auto relate.
• Virtual sites can not be placed relative to other virtual sites, as the order in which
the positions of virtual sites are updated is not guaranteed. Always relate a virtual
site to a non-virtual particle placed in the center of mass of the rigid arrangement
of particles.
• Don’t forget to declare the particle virtual in addition to calling vs auto relate
• In case you know the correct quaternions, you can also setup a virtual site using
part v virtual 1 quat q vs_relative n d
where n is the id of the non-virtual particle and d is its distance from the virtual
site.
• In a simulation on more than one CPU, the effective cell size needs to be larger
than the largest distance between a non-virtual particle and its associated virtual
sites. To this aim, you need to set the global variable min global cut to this
largest distance. ESPResSo issues a warning when creating a virtual site with
vs_auto_relate_to and the cutoff is insufficient.
• If the virtual sites represent actual particles carrying a mass, the inertia tensor of
the non-virtual particle in the center of mass needs to be adapted.
• The presence of rigid bodies constructed by means of virtual sites adds a contribution to the pressure and stress tensor.
• The use of virtual sites requires that the particles are numbered consecutively, i.e.,
the particle ids should go from zero to N − 1, where N is the number of particles.
48
4.4.3. Additional features
The behaviour of virtual sites can be fine-tuned with the following switches in myconfig.hpp
(sec. 3.4)
• VIRTUAL_SITES_NO_VELOCITY specifies that the velocity of virtual sites is not computed
• VIRTUAL_SITES_THERMOSTAT specifies that the Langevin thermostat should also
act on virtual sites
• THERMOSTAT_IGNORE_NON_VIRTUAL specifies that the thermostat does not act on
non-virtual particles
4.5. Grand canonical feature
For using ESPResSo conveniently for simulations in the grand canonical ensemble, or
other purposes, when particles of certain types are created and deleted frequently.
Syntax
part gc ( type | ( ( find | delete | status | number ) type ) )
Required features:
GRANDCANONICAL
Description
By giving only a particle type as an argument, ESPResSo will initialize arrays on the
c-level which will keep track of particles of the given type. When using the keyword
find and a particle type, the command will return a randomly chosen particle id, for
a particle of the given type. Similarly using the delete variant will delete a randomly
chosen particle of the given type. Notice however, that changing the type of an existing
particle will not change the arrays. If the change is permanent, then re-initializing the
arrays of the involved types will remove this inconsistency. The keyword status will
return a list with all particles with the given type, similarly giving number as argument
will return the number of particles which sharing the given type.
49
5. Setting up interactions
In ESPResSo, interactions are set up and investigated by the inter command. There
are mainly two types of interactions: non-bonded and bonded interactions. Non-bonded
interactions only depend on the type of the two involved particles. This also applies to
the electrostatic interaction; however, due to its long-ranged nature, it requires special
care and ESPResSo handles it separately with a number of state-of-the-art algorithms.
The particle type and the charge are both defined using the part command.
A bonded interaction defines an interaction between a number of specific particles;
it only applies to the set of particles for which it has been explicitly set. A bonded
interaction between a set of particles has to be specified explicitly by the part bond
command, while the inter command is used to define the interaction parameters.
Syntax
inter
Description
Without any arguments, inter returns a list of all defined interactions as a Tcllist. The format of each entry corresponds to the syntax for defining the interaction
as described below. Typically, this list looks like
{0 0 lennard-jones 1.0 2.0 1.1225 0.0 0.0} {0 FENE 7.0 2.0}
5.1. Isotropic non-bonded interactions
Syntax
inter type1 type2 [interaction] [parameters]
Description
This command defines an interaction of type interaction between all particles of type
type1 and type2 . The possible interaction types and their parameters are listed
below. If the interaction is omitted, the command returns the currently defined
interaction between the two types using the syntax to define the interaction, e.g.
0 0 lennard-jones 1.0 2.0 1.1225 0.0 0.0
For many non-bonded interactions, it is possible to artificially cap the forces, which
often allows to equilibrate the system much faster. See the subsection 5.9.5 for details.
50
5.1.1. Tabulated interaction
Syntax
inter type1 type2 tabulated filename
Required features:
TABULATED
Description
This defines an interaction between particles of the types type1 and type2 according
to an arbitrary tabulated pair potential. filename specifies a file which contains the
tabulated forces and energies as a function of the separation distance. The tabulated
potential allows capping the force using inter forcecap, see section 5.9.5.
At present the required file format is simply an ordered list separated by whitespace.
The data reader first looks for a # character and begins reading from that point in the
file. Anything before the # will be ignored.
The first three parameters after the # specify the number of data points Npoints and
the minimal and maximal tabulated separation distances rmin and rmax . The number
of data points obviously should be an integer, the two other can be arbitrary positive
doubles. Take care when choosing the number of points, since a copy of each lookup table
is kept on each node and must be referenced very frequently. The maximal tabulated
separation distance also acts as the effective cutoff value for the potential.
The remaining data in the file should consist of n data triples r, F (r) and V (r). r gives
the particle separation, V (r) specifies the interaction potential, and F (r) = −V 0 (r)/r
the force (note the factor 1/r!). The values of r are assumed to be equally distributed
between rmin and rmax with a fixed distance of (rmax − rmin )/(Npoints − 1); the distance
values r in the file are ignored and only included for human readability.
5.1.2. Lennard-Jones interaction
Syntax
inter type1 type2 lennard-jones σ rcut [( cshift |auto ) [roff [rcap [ rmin ]]]]
Required features:
LENNARD_JONES
Description
This command defines the traditional (12-6)-Lennard-Jones interaction between particles
of the types type1 and type2 . The potential is defined by
(
4(( r−rσ off )12 − ( r−rσ off )6 + cshift ) , if rmin + roff < r < rcut + roff
VLJ (r) =
. (5.1)
0
, otherwise
The traditional Lennard–Jones potential is the “work–horse” potential of particle–
particle interactions in coarse–grained simulations. It is a simple model of the van–der–
Waals interaction, and is attractive at large distance, but strongly repulsive at short
distances. roff + σ corresponds to the sum of the radii of the interaction particles; at this
1
radius, VLJ (r ) = 4cshift . The minimum of the potential is at r = roff + 2 6 σ. At this
51
value of r, VLJ (r) = − + 4cshift . The attractive part starts beyond this value of r . rcut
determines the radius where the potential is cut off.
If cshift is not set or it is set to the string auto, the shift will be automatically computed
such that the potential is continuous at the cutoff radius. If roff is not set, it is set to 0.
The total force on a particle can be capped by using the command inter forcecap,
see section 5.9.5, or on an individual level using the rcap variable. When rcap is set
and inter forcecap individual has been issued before, the maximal force that is
generated by this potential is the force at rcap . By default, force capping is off, i.e. the
cap radius is set to 0.
An optional additional parameter can be used to restrict the interaction from a minimal distance rmin . This is an optional parameter, set to 0 by default.
A special case of the Lennard–Jones potential is the Weeks–Chandler–Andersen (WCA)
potential, which one obtains by putting the cutoff into the minimum, i.e. choosing
1
rcut = 2 6 σ. The WCA potential is purely repulsive, and is often used to mimick hard
sphere repulsion.
When coupling particles to a Shan-Chen fluid, if the affinity interaction is set, the
Lennard-Jones potential is multiplied by the function
(
A(r) =
(1−α1 )
[1
2
1
+ tanh(2φ)] +
(1−α2 )
[1
2
1
+ tanh(−2φ)] , if r > rcut + 2 6 σ
,
, otherwise
(5.2)
where αi is the affinity to the i-th fluid component (see 5.2.3), and the order parameter
−ρ2
φ is calculated from the fluid component local density as φ = ρρ11 +ρ
. For example, if
2
the affinities are chosen so that the first component is a good solvent (α1 = 1) and the
second one is a bad solvent (α2 = 0), then, if the two particles are both in a region rich
1
in the first component, then φ ' 1, and A(r) ' 0 for r > rcut + 2 6 σ. Therefore, the
interaction potential will be very close to the WCA one. Conversely, if both particles
are in a region rich in the second component, then φ ' −1, and A(r) ' 1, so that the
potential will be very close to the full LJ one. If the cutoff has been set large enough,
the particle will experience the attractive part of the potential, mimiking the effective
attraction induced by the bad solvent.
5.1.3. Generic Lennard-Jones interaction
Syntax
inter type1 type2 lj-gen σ rcut cshift roff e1 e2 b1 b2 [( rcap |auto ) λ δ]
Required features:
52
LENNARD_JONES_GENERIC
Description
This command defines a generalized version of the Lennard-Jones interaction (see section
5.1.2) between particles of the types type1 and type2 . The potential is defined by
(
(b1 ( r−rσ off )e1 − b2 ( r−rσ off )e2 + cshift ) , if rmin + roff < r < rcut + roff
VLJ (r) =
.
0
, otherwise
(5.3)
Note that the prefactor 4 of the standard LJ potential is missing, so the normal LJ
potential is recovered for b1 = b2 = 4, e1 = 12 and e2 = 6.
The total force on a particle can be capped by using the command inter forcecap,
see section 5.9.5, or on an individual level using the rcap variable. When rcap is set
and inter forcecap individual has been issued before, the maximal force that is
generated by this potential is the force at rcap . By default, force capping is off, i.e. the
cap radius is set to 0.
The optional LJGEN_SOFTCORE feature activates a softcore version
of the potential,
p
where the following transformations apply: → λ and r−roff → (r − roff )2 − (1 − λ)δσ 2 .
λ allows to tune the strength of the interaction, while δ varies how smoothly the potential
goes to zero as λ → 0. Such a feature allows one to perform alchemical transformations,
where a group of atoms can be slowly turned on/off during a simulation.
5.1.4. Lennard-Jones cosine interaction
Syntax
(1) inter type1 type2 lj-cos σ rcut roff
(2) inter type1 type2 lj-cos2 σ roff ω
Required features:
(1) LJCOS
(2) LJCOS2
Description
specifies a Lennard-Jones interaction with cosine tail [54] between particles of the types
type1 and type2 . The first variant behaves as follows: Until the minimum of the Lennard1
Jones potential at rmin = roff + 2 6 σ, it behaves identical to the unshifted Lennard-Jones
potential (cshift = 0). Between rmin and rcut , a cosine is used to smoothly connect the
potential to 0, i.e.
1
V (r) = cos α(r − roff )2 + β − 1 ,
(5.4)
2
−1
where α = π (rcut − roff )2 − (rmin − roff )2
and β = π − (rmin − roff )2 α.
1
In the second variant, the cutoff radius is rcut = rmin + ω, where rmin = roff + 2 6 σ as
in the first variant. The potential between rmin and rcut is given by
hπ
i
V (r) = cos2
(r − rmin ) .
(5.5)
2ω
For r < rmin , V (r) is implemented as normal Lennard-Jones potential, see equation 5.1
with cshift = 0.
Only the second variant allows capping the force using inter forcecap, see section 5.9.5.
53
5.1.5. Smooth step interaction
Syntax
inter type1 type2 smooth-step σ1 n k0 σ2 rcut
Required features:
SMOOTH_STEP
Description
This defines a smooth step interaction between particles of the types type1 and type2 ,
for which the potential is
V (r) = (σ1 /d)n + /(1 + exp [2k0 (r − σ2 )])
(5.6)
for r < rcut , and V (r) = 0 elsewhere. With n around 10, the first term creates a short
range repulsion similar to the Lennard-Jones potential, while the second term provides
a much softer repulsion. This potential therefore introduces two length scales, the range
of the first term, σ1 , and the range of the second one, σ2 , where in general σ1 < σ2 .
5.1.6. BMHTF potential
Syntax
inter type1 type2 bmhtf-nacl A B C D σ rcut
Required features:
BMHTF_NACL
Description
This defines an interaction with the short-ranged part of the Born-Meyer-Huggins-TosiFumi potential between particles of the types type1 and type2 , which is often used to
simulate NaCl crystals. The potential is defined by:
V (r) = A exp [B (σ − r )] − Cr −6 − Dr −8 + shift ,
(5.7)
where shift is chosen such that V (rcut ) = 0. For r ≥ rcut , the V (r) = 0.
For NaCl, the parameters should be chosen as follows:
types A (kJ/mol) B (˚
A−1 ) C (˚
A6 kJ/mol) D ˚
A8 kJ/mol σ (˚
A)
Na-Na 25.4435
3.1546
101.1719
48.1771
2.34
Na-Cl 20.3548
3.1546
674.4793
837.0770
2.755
3.1546
6985.6786
14031.5785 3.170
Cl-Cl 15.2661
The cutoff can be chosen relatively freely because the potential decays fast; a value
around 10 seems reasonable.
In addition to this short ranged interaction, one needs to add a Coulombic, long–
ranged part. If one uses elementary charges, i.e. a charge of q = +1 for the Na–particles,
and q = −1 for the Cl–particles, the corresponding prefactor of the Coulomb interaction
is ≈ 1389.3549˚
A kJ/mol.
54
5.1.7. Morse interaction
Syntax
inter type1 type2 morse α rmin rcut
Required features:
MORSE
Description
This defines an interaction using the Morse potential between particles of the types
type1 and type2 . It serves similar purposes as the Lennard-Jones potential, but has a
deeper minimum, around which it is harmonic. This models the potential energy in a
diatomic molecule. This potential allows capping the force using inter forcecap, see
section 5.9.5.
For r < rcut , this potential is given by
V (r) = (exp [−2α (r − rmin )] − 2 exp [−α (r − rmin )]) − shift ,
(5.8)
where shift is again chosen such that V (rcut ) = 0. For r ≥ rcut , the V (r) = 0.
5.1.8. Buckingham interaction
Syntax
inter type1 type2 buckingham A B C D rcut rdiscont shift
Required features:
BUCKINGHAM
Description
This defines a Buckingham interaction between particles of the types type1 and type2 ,
for which the potential is given by
V (r) = A exp(−Br) − Cr−6 − Dr−4 + shift
(5.9)
for rdiscont < r < rcut . Below rdiscont , the potential is linearly continued towards r = 0,
similarly to force capping, see below. Above r = rcut , the potential is 0. This potential
allows capping the force using inter forcecap, see section 5.9.5.
5.1.9. Soft-sphere interaction
Syntax
inter type1 type2 soft-sphere a n rcut roffset
Required features:
SOFT_SPHERE
Description
This defines a soft sphere interaction between particles of the types type1 and type2 ,
which is defined by a single power law:
V (r) = a (r − roffset )−n
(5.10)
for r < rcut , and V (r) = 0 above. There is no shift implemented currently, which means
that the potential is discontinuous at r = rcut . Therefore energy calculations should be
used with great caution.
55
5.1.10. Hat interaction
Syntax
inter type1 type2 hat Fmax rc
Required features:
HAT
Description
This defines a simple force ramp between particles of the types type1 and type2 . The
maximal force Fmax acts at zero distance and zero force is applied at distances rc and
bigger. For distances smaller than rc , the force is given by
r
F (r) = Fmax · 1 −
,
(5.11)
rc
for distances exceeding rc , the force is zero.
The potential energy is given by
V (r) = Fmax · (r − rc ) ·
r + rc
−1 ,
2rc
(5.12)
which is zero for distances bigger than rc and continuous at distance rc .
This is the standard conservative DPD potential and can be used in combination with
inter DPD 5.9.2. The potential is also useful for live demonstrations, where a big time
step may be employed to obtain quick results on a weak machine, for which the physics
do not need to be entirely correct.
5.1.11. Hertzian interaction
Syntax
inter type1 type2 hertzian σ Required features:
HERTZIAN
Description
This defines an interaction according to the Hertzian potential between particles of the
types type1 and type2 . The Hertzian potential is defined by
(
5/2
1 − σr
r<σ
V (r) =
(5.13)
0
r ≥ σ.
The potential has no singularity and is defined everywhere; the potential has nondifferentiable maximum at r = 0, where the force is undefined.
5.1.12. Gaussian
Syntax
inter type1 type2 gaussian σ rcut
Required features:
56
GAUSSIAN
Description
This defines an interaction according to the Gaussian potential between particles of the
typers type1 and type2 . The Gaussian potential is defined by
(
1 r 2
e− 2 ( σ ) r < rcut
(5.14)
V (r) =
0
r ≥ rcut
The Gaussian potential is smooth except at the cutoff, and has a finite overlap energy
of . It can be used to model e.g. overlapping polymer coils.
Currently, there is no shift implemented, which means that the potential is discontinuous at r = rcut . Therefore use caution when performing energy calculations. However,
you can often choose the cutoff such that the energy difference at the cutoff is less than
a desired accuracy, since the potential decays very rapidly.
5.2. Anisotropic non-bonded interactions
5.2.1. Directional Lennard-Jones interaction
Syntax
inter type1 type2 lj-angle σ rcut b1a b1b b2a b2b [rcap z0 δz κ 0 ]
Required features:
LJ_ANGLE
Description
1111
0000
0000
1111
0000
1111
111
000
000
111
000
111
Specifies a 12-10 Lennard-Jones interaction with angular dependence between particles
of the types type1 and type2 . These two particles need two bonded partners oriented
in a symmetric way. They define an orientation for the central particle. The purpose of
using bonded partners is to avoid dealing with torques, therefore the interaction does not
need the ROTATION feature. The angular part of the potential minimizes the system
when the two central beads are oriented along the vector formed by these two particles.
The shaded beads on the image are virtual particles that are formed from the orientation
of the bonded partners, connected to the central beads. They are used to define angles.
The potential is of the form
σ 10 σ 12
U (rik , θjik , θikn ) = 5
−6
cos2 θjik cos2 θikn ,
(5.15)
r
r
where rik is the distance between the two central beads, and each angle defines the
orientation between the direction of a central bead (determined from the two bonded
57
partners) and the vector rik . Note that the potential is turned off if one of the angle
is more than π/2. This way we don’t end up creating a minimum for an anti-parallel
configuration.
Unfortunately, the bonded partners are not sought dynamically. One has to keep track
of the relative positions of the particle IDs. This can be done by setting the parameters
b1a , b1b , b2a , and b2b . Say the first bead type1 has particle ID n, then one should set
the simulation such as its two bonded partners have particle IDs n + b1a and n + b1b ,
respectively. On a linear chain, for example, one would typically have b1a = 1 and
b1b = −1 such that the central bead and its two bonded partners have position IDs n,
n + 1 , and n − 1 , respectively. This is surely not optimized, but once the simulation is
set correctly the algorithm is very fast.
The force can be capped using inter forcecap. It might turn out to be useful in
some cases to keep this capping during the whole simulation. This is due to the very
sharp angular dependence for small distance, compared to σ. Two beads might come
very close to each other while having unfavorable angles such that the interaction is
turned off. Then a change in the angle might suddenly turn on the interaction and the
system will blow up (the potential is so steep that one would need extremely small time
steps to deal with it, which is not very clever for such rare events).
For instance, when modeling hydrogen bonds (N-H...O=C), one can avoid simulating
hydrogens and oxygens by using this potential. This comes down to implementing a
HBond potential between N and C atoms.
The optional parameter rcap is the usual cap radius. The four other optional parameters (z0 , δz , κ, 0 ) describe a different interaction strength 0 for a subset of the
simulation box. The box is divided through the z plane in two different regions: region
1 which creates an interaction with strength , region 2 with interaction strength 0 . The
2nd region is defined by its z -midplane z0 , its total thickness δz , and the interface width
κ. Therefore, the interaction strength is everywhere except for the region of the box
z0 − δz/2 < z < z0 + δz/2. The interface width smoothly interpolates between the two
regions to avoid discontinuities. As an example, one can think of modeling hydrogen
bonds in two different environments: water, where the interaction is rather weak, and
in a lipid bilayer, where it is comparatively stronger.
5.2.2. Gay-Berne interaction
Syntax
inter type1 type2 gay-berne 0 σ0 rcutoff k1 k2 µ ν
Required features:
ROTATION
GAY_BERNE
Description
This defines a Gay-Berne potential for prolate and oblate particles between particles of
the types type1 and type2 . The Gay-Berne potential is an anisotropic version of the
classic Lennard-Jones potential, with orientational dependence of the range σ0 and the
well-depth 0 .
58
Assume two particles with orientations given by the unit vectors u
ˆ i and u
ˆ j and intermolecular vector r = rˆ
r. If r < rcut , then the interaction between these two particles is
given by
−12
−6
V (rij , u
ˆi , u
ˆ j ) = 4(ˆ
rij , u
ˆi , u
ˆ j ) r˜ij
− r˜ij
,
(5.16)
otherwise V (r) = 0. The reduced radius is
r˜ =
r − σ(ˆ
r, u
ˆi , u
ˆ j ) + σ0
,
σ0
#)− 1
"
2
r·u
ˆi − ˆ
r·u
ˆ j )2
(ˆ
r·u
ˆi + ˆ
r·u
ˆ j )2 (ˆ
1
1− χ
+
2
1 + χˆ
ui · u
ˆj
1 − χˆ
ui · u
ˆj
(5.17)
(
σ(ˆ
r, u
ˆi , u
ˆ j ) = σ0
(5.18)
and
(ˆ
r, u
ˆi , u
ˆj ) =
0
µ
− ν
r·u
ˆi + ˆ
r·u
ˆ j )2 (ˆ
r·u
ˆi − ˆ
r·u
ˆ j )2
χ0 (ˆ
2
1 − χ (ˆ
ui · u
ˆj )
+
. (5.19)
1−
2
1 + χ0 u
ˆi · u
ˆj
1 − χ0 u
ˆi · u
ˆj
2
1/µ
1/µ
2
0
The parameters χ =
− 1 / k1 + 1 and χ = k2 − 1 / k2 + 1 are responsible for the degree of anisotropy of the molecular properties. k1 is the molecular elongation, and k2 is the ratio of the potential well depths for the side-by-side and end-to-end
configurations. The exponents µ and ν are adjustable parameters of the potential. Several Gay-Berne parametrizations exist, the original one being k1 = 3, k2 = 5, µ = 2 and
ν = 1.
k12
5.2.3. Affinity interaction
Syntax
inter type1 type2 affinity α1 α2
Required features:
SHANCHEN
Description
Instead of defining a new interaction, this command acts as a modifier for existing
interactions, so that the conditions of good/bad solvent associated to the two components
of a Shan-Chen fluid. The two types must match those of the interaction that one wants
to modify, and the two affinity values α1 and α2 are values between 0 and 1. A value of 1
(of 0) indicates that the component acts as a good (bad) solvent. The specific functional
form depends on the interaction type and is listed in the interaction section. So far, only
the standard Lennard-Jones interaction is modified by the affinity interaction.
59
5.3. Bonded interactions
Syntax
inter bondid [interaction] [parameters]
Description
Bonded interactions are identified by their bonded interaction type identificator
bondid , which is a non-negative integer. The inter bondid command is used to
specify the type and parameters of a bonded interaction, which applies to all particles connected explicitly by this bond using the part command (see section 4.1
on page 29). Therefore, defining a bond between two particles always involves two
steps: defining the interaction and applying it. Assuming that two particles with
ids 42 and 43 already exist, one can create e.g. a FENE-bond between them using
inter 1 fene 10.0 2.0
part 42 bond 1 43
If a FENE-bond with the same interaction parameters is required between several
particles (e.g. in a simple chain molecule), one can use the same type id :
inter 1 fene 10.0 2.0
part 42 bond 1 43; part 43 bond 1 44
Bonds can have more than just two bond partners. For the inter command that does
not play a role as it only specifies the parameters, only when applying the bond using
the bond particle, the number of involved particles plays a role. The number of involved
particles and their order, if important, is nevertheless specified here for completeness.
5.3.1. FENE bond
Syntax
inter bondid fene K ∆rmax [r0 ]
Description
This creates a bond type with identificator bondid with a FENE (finite extension nonlinear expander) interaction. This is a rubber-band-like, symmetric interaction between
two particles with prefactor K , maximal stretching ∆rmax and equilibrium bond length
r0 . The bond potential diverges at a particle distance r = r0 −∆rmax and r = r0 +∆rmax .
It is given by
"
#
1
r − r0 2
2
V (r) = − K ∆rmax ln 1 −
.
(5.20)
2
∆rmax
5.3.2. Harmonic bond
Syntax
inter bondid harmonic K R [rcut ]
60
Description
This creates a bond type with identificator bondid with a classical harmonic potential.
It is a symmetric interaction between two particles. The potential is minimal at particle
distance r = R, and the prefactor is K. It is given by
1
V (r) = K (r − R)2
2
(5.21)
The third, optional parameter rcut defines a cutoff radius. Whenever a harmonic bond
gets longer than rcut , the bond will be reported as broken, and a background error will
be raised.
5.3.3. Quartic bond
Syntax
inter bondid quartic K0 K1 R [rcut ]
Description
This creates a bond type with identificator bondid with a quartic potential. The potential
is minimal at particle distance r = R. It is given by
1
1
V (r) = K0 (r − R)2 + K1 (r − R)4
2
4
(5.22)
The fourth, optional, parameter rcut defines a cutoff radius. Whenever a quartic bond
gets longer than rcut , the bond will be reported as broken, and a background error will
be raised.
5.3.4. Bonded coulomb
Syntax
inter bondid bonded_coulomb α
Description
This creates a bond type with identificator bondid with a coulomb pair potential. It is
given by
αq1 q2
V (r) =
,
(5.23)
r
where q1 and q2 are the charges of the bound particles. There is no cutoff, the bejerrum
length of other coulomb interactions is not taken into account.
5.3.5. Subtracted Lennard-Jones bond
Syntax
inter bondid subt_lj reserved R
61
Description
This creates a ”bond” type with identificator bondid , which acts between two particles
and actually subtracts the Lennard-Jones interaction between the involved particles. The
first parameter, reserved is a dummy just kept for compatibility reasons. The second
parameter, R, is used as a check: if any bond length in the system exceeds this value, the
program terminates. When using this interaction, it is worthwhile to consider capping
the Lennard-Jones potential appropriately so that round-off errors can be avoided.
This interaction is useful when using other bond potentials which already include
the short–ranged repulsion. This often the case for force fields or in general tabulated
potentials.
5.3.6. Rigid bonds
Syntax
inter bondid rigid_bond constrained bond distance positional tolerance
velocity tolerance
Description
To simulate rigid bonds, ESPResSo uses the Rattle Shake algorithm which satisfies internal constraints for molecular models with internal constraints, using Lagrange multipliers.[2]
5.3.7. Tabulated bond interactions
Syntax
(1) inter bondid tabulated bond filename
(2) inter bondid tabulated angle filename
(3) inter bondid tabulated dihedral filename
Description
This creates a bond type with identificator bondid with a two-body bond length (variant
(1)), three-body angle (variant (2)) or four-body dihedral (variant (3)) tabulated potential. The tabulated forces and energies have to be provided in a file filename, which is
formatted identically as the files for non-bonded tabulated potentials (see section 5.1.1).
The potential is calculated as follows:
• Variant (1) is a two body interaction depending on the distance of two particles.
The force acts in the direction of the connecting vector between the particles.
The bond breaks above the tabulated range, but for distances smaller than the
tabulated range, a linear extrapolation based on the first two tabulated force values
is used.
• Variant (2) is a three-body angle interaction similar to the angle potential (see
section 5.5). It is assumed that the potential is tabulated for all angles between
0 and π, where 0 corresponds to a stretched polymer, and just as for the tabulated pair potential, the forces are scaled with the inverse length of the connecting
62
vectors. The force on particles p1 and p3 (in the notation of section 5.5) acts perpendicular to the connecting vector between the particle and the center particle
p2 in the plane defined by the three particles. The force on the center particle p2
balances the other two forces.
• Variant (3) tabulates a torsional dihedral angle potential (see section 5.6). It is
assumed that the potential is tabulated for all angles between 0 and 2π. This
potential is not tested yet! Use on own risk, and please report your findings and
eventually necessary fixes.
5.3.8. Virtual bonds
Syntax
inter bondid virtual_bond
Description
This creates a virtual bond type with identificator bondid , i.e. a pair bond without
associated potential or force. It can used to specify topologies and for some analysis
that rely on bonds, or e.g. for bonds that should be displayed in VMD.
5.4. Object-in-fluid interactions
Please cite [12] (BibTeX-key cimrak in file doc/ug/citations.bib) when using the
interactions in this section in order to simulate extended objects embedded in a LB
fluid. For more details also see the documentation at
cell-in-fluid.fri.uniza.sk/oif-documentation
The following interactions are implemented in order to mimic the mechanics of elastic
or rigid objects immersed in the LB fluid flow. Their mathematical formulations were
inspired by [21]. Details on how the bonds can be used for modeling objects are described
in section 14.
5.4.1. Stretching force
Syntax
inter bondid stretching_force L0AB ks
Description
This type of interaction is available for closed 3D immersed objects as well as for 2D
sheet flowing in the 3D flow.
For each edge of the mesh, LAB is the current distance between point A and point B.
0
LAB is the distance between these points in the relaxed state, that is if the current edge
has the length exactly L0AB , then no forces are added. ∆LAB is the deviation from the
63
relaxed state, that is ∆LAB = LAB − L0AB . The stretching force between A and B is
calculated using
Fs (A, B ) = ks κ(λAB )∆LAB nAB .
(5.24)
Here, nAB is the unit vector pointing from A to B , ks is the stretching constant, λAB =
LAB /L0AB , and κ is a nonlinear function that resembles neo-Hookean behaviour
κ(λAB ) =
0.5 + λ−2.5
λAB
AB
.
λAB + λ−3
AB
(5.25)
The stretching force acts between two particles and is symmetric. Therefore, if an
interaction is defined by
inter 1 stretching_force 2.0 4.0
then the following two commands
part 42 bond 1 43
part 43 bond 1 42
are equivalent.
5.4.2. Bending force
Syntax
inter bondid bending_force θ0 kb
Description
The tendency of an elastic object to maintain the resting shape is achieved by prescribing the preferred angles between the neighbouring triangles of the mesh. This type of
interaction is available for closed 3D immersed objects as well as for 2D sheet flowing in
the 3D flow.
Denote the angle between two triangles in the resting shape by θ0 . For closed immersed
objects, one always has to set the inner angle. The deviation of this angle ∆θ = θ − θ0
defines two bending forces for two triangles A1 BC and A2 BC
Fbi (Ai BC ) = kb
∆θ
nA BC
θ0 i
(5.26)
Here, nAi BC is the unit normal vector to the triangle Ai BC . The force Fbi (Ai BC ) is
assigned to the vertex not belonging to the common edge. The opposite force divided
by two is assigned to the two vertices lying on the common edge. This procedure is done
twice, for i = 1 and for i = 2.
64
Unlike the stretching force, the bending force is strictly asymmetric. After creating
the interaction
inter 33 bending_force 0.7 4.0,
it is important how the bond is created. Particles need to be mentioned in the correct
order. Command
part 0 bond 33 1 2 3
creates a bond related to the angle between the triangles 012 and 123. The particle 0
corresponds to point A1, particle 1 to C, particle 2 to B and particle 3 to A2. There are
two rules that need to be fulfilled:
• there has to be an edge between particles 1 and 2
• orientation of the triangle 012, that is the normal vector defined as a vector product
01 × 02, must point to the inside of the immersed object.
Notice that also concave objects can be defined. If θ0 is larger than π, then the inner
angle is concave.
5.4.3. Local area conservation
Syntax
0
inter bondid area_force_local SABC
kal
Description
This interaction conserves the area of the triangles in the triangulation. This type of
interaction is available for closed 3D immersed objects as well as for 2D sheet flowing in
the 3D flow.
The deviation of the triangle surface SABC is computed from the triangle surface in
0
the resting shape ∆SABC = SABC − SABC
. The area constraint assigns the following
shrinking/expanding force to every vertex
∆SABC
wA
Fal (A) = −kal √
SABC
(5.27)
where kal is the area constraint coefficient, and wA is the unit vector pointing from the
centroid of triangle ABC to the vertex A. Similarly the analogical forces are assigned
to B and C . This interaction is symmetric, therefore after defining the interaction
65
inter 44 area_force_local 0.02 4.0
the following commands are equivalent
part 0 bond 44 1 2
part 0 bond 44 2 1
part 1 bond 44 0 2
5.4.4. Global area conservation
Syntax
inter bondid area_force_global S 0 kag
Description
This type of interaction is available solely for closed 3D immersed objects.
The conservation of local area is sometimes too restrictive. Denote by S the current
surface of the immersed object, by S0 the surface in the relaxed state and define ∆S =
S − S0 . The global area conservation force is defined as
Fag (A) = −kag
∆S
wA
S
(5.28)
Here, the above mentioned force divided by 3 is added to all three particles.
Again, this interaction is symmetric, as is the area force local.
5.4.5. Volume conservation
Syntax
inter bondid volume_force V 0 kv
Description
This type of interaction is available solely for closed 3D immersed objects.
The deviation of the objects volume V is computed from the volume in the resting
shape ∆V = V − V 0 . For each triangle the following force is computed
Fv (ABC ) = −kv
66
∆V
SABC nABC
V0
(5.29)
where SABC is the area of triangle ABC , nABC is the normal unit vector of plane ABC ,
and kv is the volume constraint coefficient. The volume of one immersed object is
computed from
X
V =
SABC nABC · hABC
(5.30)
ABC
where the sum is computed over all triangles of the mesh and hABC is the normal vector
from the centroid of triangle ABC to any plane which does not cross the cell. The force
Fv (ABC ) is equally distributed to all three vertices A, B , C .
This interaction is again symmetric. After the definition of the interaction by
inter 22 volume_force 65.3 3.0
the order of vertices is crucial. By the following command the bonds are defined
part 0 bond 22 1 2
Triangle 012 must have correct orientation, that is the normal vector defined by a vector
product 01 × 02. The orientation must point inside the immersed object.
5.5. Bond-angle interactions
Syntax
(1) inter bondid angle_harmonic K [φ0 ]
(2) inter bondid angle_cosine K [φ0 ]
(3) inter bondid angle_cossquare K [φ0 ]
Required features:
BOND_ANGLE
Description
This creates a bond type with identificator bondid with an angle dependent potential.
This potential is defined between three particles. The particle for which the bond is
created, is the central particle, and the angle φ between the vectors from this particle to
the two others determines the interaction. K is the bending constant, and the optional
parameter φ0 is the equilibirum bond angle in radian ranging from 0 to π. If this parameter is not given, it defaults to φ0 = π, which corresponds to a stretched configuration.
For example, for a bond defined by
part $p_2 bond 4 $p_1 $p_3
the minimal energy configurations are the following:
67
inter 4 angle type 1.0 [PI]
inter 4 angle type 1.0 [expr [PI]/2]
~ p3
~
p1
~
p2
~
~
p3
p1
~
p2
For the potential acting between the three particles three variants are possible
• Harmonic bond angle potential (1):
A classical harmonic potential,
V (φ) =
K
(φ − φ0 )2 .
2
(5.31)
Unlike the two following variants, this potential has a kink at φ = φ0 + π and
accordingly a discontinuity in the force, and should therefore be used with caution.
• Cosine bond angle potential (2):
V (α) = K [1 − cos(φ − φ0)]
(5.32)
Around φ0 , this potenial is close to a harmonic one (both are 1/2(φ − φ0 )2 in
leading order), but it is periodic and smooth for all angles φ.
• Cosine square bond angle potential (3):
K
[cos(φ) − cos(φ0 )]2
(5.33)
2
This form is used for example in the GROMOS96 force field. The potential is
1/8(φ − φ0 )4 around φ0 , and therefore much flatter than the two potentials before.
V (α) =
5.6. Dihedral interactions
Syntax
inter bondid dihedral n K p
Description
This creates a bond type with identificator bondid with a dihedral potential, i.e. a fourbody-potential. In the following, let the particle for which the bond is created be particle
p2 , and the other bond partners p1 , p3 , p4 , in this order, i.e. part p2 bond bondid p1 p3 p4 .
Then, the dihedral potential is given by
V (φ) = K [1 − cos(nφ − p)] ,
68
(5.34)
where n is the multiplicity of the potential (number of minimas) and can take any integer
value (typically from 1 to 6), p is a phase parameter and K is the bending constant of the
potential. φ is the dihedral angle between the particles defined by the particle quadrupel
p1 , p2 , p3 and p4 , i.e. the angle between the planes defined by the particle triples p1 , p2
and p3 and p2 , p3 and p4 :
P2
P1
φ
P3
P4
Together with appropriate Lennard-Jones interactions, this potential can mimic a large
number of atomic torsion potentials.
If you enable the feature OLD DIHEDRAL, then the old, less general form of the
potential is used:
V (φ) = K [1 + p cos(nφ)] ,
(5.35)
where p is rather a phase factor and can only take values p = ±1.
5.7. Coulomb interaction
Syntax
(1) inter coulomb 0.0
(2) inter coulomb
(3) inter coulomb parameters
Description
These commands allow to set up the calculation of the Coulomb interaction. The
Coulomb (or electrostatic) interaction is defined as follows. For a pair of particles at
distance r with charges q1 and q2 , the interaction is given by
U C (r) = lB kB T
q1 q2
.
r
(5.36)
where lB = e2o /(4πkB T ) denotes the Bjerrum length, which measures the strength of
the electrostatic interaction. As a special case, when the internal variable temperature
is set to zero, the value of bjerrum length you enter is treated as lB kB T rather than
lB . This occurs when the thermostat is switched off and ESPResSo performs an NVE
integration (see also Section 6.2).
Computing electrostatic interactions is computationally very expensive. ESPResSo
features some state-of-the-art algorithms to deal with these interactions as efficiently as
possible, but almost all of them require some knowledge to use them properly. Uneducated use can result in completely unphysical simulations.
69
Variant (1) disables Coulomb interactions. Variant (2) returns the current parameters of the coulomb interaction as a Tcl-list using the same syntax as used to
setup the method, e.g.
{coulomb 1.0 p3m 7.75 8 5 0.1138 0.0}
{coulomb epsilon 0.1 n_interpol 32768 mesh_off 0.5 0.5 0.5}
Variant (3) is the generic syntax to set up a specific method or its parameters,
the details of which are described in the following subsections. Note that using the
electrostatic interaction also requires assigning charges to the particles. This is done
using the part command to set the charge q, e.g.
inter coulomb 1.0 p3m tune accuracy 1e-4
part 0 q 1.0; part 1 q -1.0
5.7.1. Coulomb P3M
Syntax
inter coulomb lB p3m [gpu] rcut ( mesh | {meshx meshy meshz } ) cao
alpha
Required features:
ELECTROSTATICS
Description
For this feature to work, you need to have the fftw3 library installed on your system.
In ESPResSo, you can check if it is compiled in by checking for the feature FFTW.
This command activates the P3M method to compute the electrostatic interactions
between charged particles. The different parameters are described in more detail in [17].
[gpu] The optional flag gpu causes the far field portion of p3m to be calculated on the
GPU. It should be noted that this does not always provide significant increase in
performance. Furthermore it computes the far field interactions with only single
precision which limits the maximum precision. Furthermore the algorithm does
not work in combination with certain other methods implemented in ESPResSo
and only for the case of cubic boxes.
rcut The real space cutoff as a positive floating point number.
mesh The number of mesh points, as a single positive integer.
meshx ,y,z The number of mesh points in x, y and z direction. This is relevant for
noncubic boxes.
cao The charge-assignment order, an integer between 0 and 7.
alpha The Ewald parameter as a positive floating point number.
Make sure that you know the relevance of the P3M parameters before using P3M! If
you are not sure, read the following references [22, 27, 33, 16, 17, 18, 15, 10].
70
Tuning Coulomb P3M
Syntax
inter coulomb lB p3m ( tune | tunev2 ) [gpu] accuracy accuracy
[r_cut rcut ] [mesh mesh] [cao cao] [alpha α]
Required features:
ELECTROSTATICS
Description
It is not easy to calculate the various parameters of the P3M method such that the
method provides the desired accuracy at maximum speed. To simplify this, ESPResSo
provides a function to automatically tune the algorithm. Note that for this function to
work properly, your system should already contain an initial configuration of charges and
the correct initial box size. Also note that both provided tuning algorithms work very
well on homogenous charge distributions, but might not achieve the requested precision
for highly inhomogenous or symmetric systems. For example, because of the nature of
the P3M algorithm, systems are problematic where most charges are placed in one plane,
one small region, or on a regular grid.
The function employs the analytical expression of the error estimate for the P3M
method [27] and its real space error [33] to obtain sets of parameters that yield the
desired accuracy, then it measures how long it takes to compute the coulomb interaction
using these parameter sets and chooses the set with the shortest run time.
The function will only automatically tune those parameters that are not set to a
predetermined value using the optional parameters of the tuning command.
The two tuning methods follow different methods for determining the optimal parameters. While the tune version tests different values on a grid in the parameter space,
the tunev2 version uses a bisection to determine the optimal parameters. In general, for
small systems the tune version is faster, while for large systems tunev2 is faster. The
results of tunev2 are always at least as good as the parameters from the tune version,
and normally the obtained accuracy is much closer to the desired value.
During execution the tuning routines report the tested parameter sets, the corresponding k-space and real-space errors and the timings needed for force calculations (the setmd
variable timings controls the number of test force calculations). Since the error depends
on rcut /box l and αbox l the output is given in these units.
Note that the previous setting of rcut , cao and mesh will be remembered. If you want
to retune your electrostatics, e.g. after a major system change, you should use
inter coulomb lB p3m tune accuracy acc r_cut 0 mesh 0 cao 0
Additional P3M parameters
Syntax
inter coulomb [epsilon ( metallic | epsilon )] [n_interpol points]
[mesh_off xoff yoff zoff ]
71
Description
Once P3M algorithm has been set up, it is possible to set some additional P3M parameters with this command. The different parameters have the following meaning:
epsilon epsilon The dielectric constant of the surrounding medium, metallic (i.e. infinity) or some finite positive number. Defaults to metallic.
n_interpol ni nterpol Number of interpolation points for the charge assignment function. When this is set to 0, interpolation is turned off and the function is computed
directly. Defaults to 32768.
mesh_off mesho ff Offset of the first mesh point from the lower left corner of the simulation box in units of the mesh constant. Defaults to 0.5 0.5 0.5.
5.7.2. Coulomb Ewald GPU
Syntax
inter coulomb lB ewaldgpu rcut ( Kcut | {Kcut,x Kcut,y Kcut,x } ) alpha
Required features:
ELECTROSTATICS
Description
This command activates the Ewald method to compute the electrostatic interactions
between charged particles. The far field is computed by the GPU with single precision
and the near field by the CPU with double precision. It only works for the case of cubic
boxes.
lB Bjerrum length as positive floating point number
rcut Real space cutoff as positive floating point number
Kcut Reciprocal space cutoff as single positive integer
Kcut,xyz Reciprocal space cutoff in x, y and z direction (relevant for noncubic boxes)
alpha Ewald parameter as positive floating point number
Tuning Ewald GPU
Syntax
inter coulomb lB ewaldgpu tune accuracy accuracy precision precision
K_max Kmax
Required features:
ELECTROSTATICS
Description
The tuning algorithm first computes the optimal rcut and alpha for every Kcut between
one and Kmax as described in [33]. Then the performance for all those (Kcut , rcut , alpha)triplets will be measured via a short test simulation and the fastest will be chosen.
72
accuracy Maximal allowed root mean square error regarding the forces
precision Determines how precise alpha will be computed
Kmax Maximal reciprocal space cutoff Kcut to be tested in the tuning algorithm
Tuning Alpha Ewald GPU
Syntax
inter coulomb lB ewaldgpu tunealpha rcut ( Kcut | {Kcut,x Kcut,y Kcut,x } )
precision
Required features:
ELECTROSTATICS
Description
If Kcut and rcut are given by the user, then tunealpha computes the optimal alpha with
the chosen precision as described in [33]. But in general tune should be chosen for
tuning.
5.7.3. Debye-H¨
uckel potential
Syntax
inter coulomb lB dh κ rcut
Required features:
ELECTROSTATICS
Description
Defines the electrostatic potential by
U C−DH = lB kB T
q1 q2 exp(−κr)
r
for r < rcut
(5.37)
The Debye-H¨
uckel potential is an approximate method for calculating electrostatic
interactions, but technically it is treated as other short-ranged non-bonding potentials.
For r > rcut it is set to zero which introduces a step in energy. Therefore, it introduces
fluctuations in energy.
For κ = 0, this corresponds to the plain coulomb potential.
5.7.4. MMM2D
Please cite [5] (BibTeX-key mmm2d in file doc/ug/citations.bib) when using
MMM2D, and [60] (BibTeX-key icmmm2d in file doc/ug/citations.bib) when
using dielectric interfaces.
Syntax
inter coulomb lB mmm2d maximal pairwise error [fixed far cutoff ]
[dielectric t m b ] [dielectric-contrasts ∆t ∆b ] [capacitor U ]
Required features:
ELECTROSTATICS
73
Description
MMM2D coulomb method for systems with periodicity 1 1 0. Needs the layered cell
system. The performance of the method depends on the number of slices of the cell
system, which has to be tuned manually. It is automatically ensured that the maximal
pairwise error is smaller than the given bound. The far cutoff setting should only be
used for testing reasons, otherwise you are more safe with the automatical tuning. If
you even don’t know what it is, do not even think of touching the far cutoff. For details
on the MMM family of algorithms, refer to appendix E on page 248.
The last two, mutually exclusive arguments “dielectric” and “dielectric-constants”
allow to specify dielectric contrasts at the upper and lower boundaries of the simulation
box. The first form specifies the respective dielectric constants in the media, which
however is only used to calculate the contrasts. That is, specifying t = m = b = const
is always identical to t = m = b = 1. The second form specifies only the dielectric
−b
−t
contrasts at the boundaries, that is ∆t = m
and ∆b = m
. Using this form allows
m +t
m +b
to choose ∆t/b = −1, corresponding to metallic boundary conditions.
Using capacitor U allows to maintain a constant electric potential difference U
between the xy-plane at z = 0 and z = L, where L denotes the box length in z-direction.
This is done by countering the total dipol moment of the system with the electric field
Einduced and superposing a homogeneous electric field Eapplied = UL to retain U . This
mimics the induction of surface charges ±σ = Einduced · 0 for planar electrodes at z = 0
and z = L in a capacitor connected to a battery with voltage U . Using capacitor 0 is
equivalent to ∆t/b = −1.
Syntax
efield_caps ( total | induced | applied )
Required features:
ELECTROSTATICS
Description
The electric fields added by capacitor U can be obtained by calling the above command, where induced returns Einduced , applied returns Eapplied and total their sum.
5.7.5. MMM1D
Please cite [3] (BibTeX-key mmm1d in file doc/ug/citations.bib) when using
MMM1D.
Syntax
(1) inter coulomb lB mmm1d switch radius maximal pairwise error
(2) inter coulomb lB mmm1d tune maximal pairwise error
Required features:
ELECTROSTATICS
Description
MMM1D coulomb method for systems with periodicity 0 0 1. Needs the nsquared cell
system (see section 6.4 on page 94). The first form sets parameters manually. The switch
radius determines at which xy-distance the force calculation switches from the near to
the far formula. The Bessel cutoff does not need to be specified as it is automatically
74
determined from the particle distances and maximal pairwise error. The second tuning
form just takes the maximal pairwise error and tries out a lot of switching radii to find
out the fastest one. If this takes too long, you can change the value of the setmd variable
timings, which controls the number of test force calculations.
Syntax
(1) inter coulomb lB mmm1dgpu switch radius [bessel cutoff ]
maximal pairwise error
(2) inter coulomb lB mmm1dgpu tune maximal pairwise error
Required features:
CUDA
ELECTROSTATICS
MMM1D_GPU
Description
MMM1D is also available in a GPU implementation. Unlike its CPU counterpart, it
does not need the nsquared cell system. The first form sets parameters manually. The
switch radius determines at which xy-distance the force calculation switches from the
near to the far formula. If the Bessel cutoff is not explicitly given, it is determined from
the maximal pairwise error, otherwise this error only counts for the near formula. The
second tuning form just takes the maximal pairwise error and tries out a lot of switching
radii to find out the fastest one.
For details on the MMM family of algorithms, refer to appendix E on page 248.
5.7.6. Maxwell Equation Molecular Dynamics (MEMD)
Syntax
inter coulomb lB memd f mass mesh [epsilon ε∞ ]
Required features:
ELECTROSTATICS
Description
This is an implementation of the instantaneous 1/r Coulomb interaction
U = lB kB T
q1 q2
r
(5.38)
as the potential of mean force between charges which are dynamically coupled to a local
electromagnetic field.
The algorithm currently works with the following constraints:
• cellsystem has to be domain decomposition but without Verlet lists!
• system has to be periodic in three dimensions.
Arguments
• f mass is the mass of the field degree of freedom and equals to the square root of
the inverted speed of light.
• mesh is the number of mesh points for the interpolation of the electromagnetic
field in one dimension.
75
• ε∞ is the background dielectric permittivity at infinity. This defaults to metallic
boundary conditions, to match the results of P3M.
The arising self-interactions are treated with a modified version of the exact solution
of the lattice Green’s function for the problem.
Currently, forces have large errors for two particles within the same lattice cube. This
may be fixed in future development, but right now leads to the following rule of thumb
for the parameter choices:
• The lattice should be of the size of your particle size (i.e. the lennard jones epsilon).
That means: mesh ≈ box l/lj sigma
• The integration timestep should be in a range where no particle moves more than
one lattice box (i.e. lennard jones sigma) per timestep.
• The speed of light should satisfy the stability criterion c a/dt, where a is the
lattice spacing and dt is the timestep. For the second parameter, this means
f mass dt2 /a2 .
The main error of the MEMD algorithm stems from the lattice interpolation and is
proportional to the lattice size in three dimensions, which means ∆lattice ∝ a3 .
Without derivation here, the algorithmis error is proportional to 1/c2 , where c is the
adjustable speed of light. From the stability criterion, this yields
∆maggs = A · a3 + B · dt2 /a2
(5.39)
This means that increasing the lattice will help the algorithmic error, as we can tune
the speed of light to a higher value. At the same time, it increases the interpolation
error at an even higher rate. Therefore, momentarily it is advisable to choose the lattice
with a rather fine mesh of the size of the particles. As a rule of thumb, the error will
then be less than 10−5 for the particle force.
For a more detailed description of the algorithm, see appendix D on page 242 or the
publications [38, 43].
Spatially varying dielectrics with MEMD
Since MEMD is a purely local algorithm, one can apply local changes to some properties
and the propagation of the Coulomb force is still valid. In particular, it is possible to
arbitrarily select the dielectric permittivity on each site of the interpolating lattice.
Syntax
inter coulomb lB memd localeps node node x node y node z dir X /Y /Z
eps ε
Required features:
ELECTROSTATICS
Description
The keyword localeps after the inter coulomb command offers the possibility to assign
any value of ε to any lattice site.
76
Arguments
• lB is the bjerrum length of the background. It defines the reference value εbg via
the formula (5.40). This is a global variable.
• node x is the index of the node in x direction that should be changed
• node y is the index of the node in y direction that should be changed
• node z is the index of the node in z direction that should be changed
• X /Y /Z is the direction in which the lattice site to be changed is pointing. Has
to be one of the three (X, Y or Z).
• ε is the relative permittivity change in respect to the background permittivity set
by the parameter lB .
The permittivity on each lattice site is set relatively. By defining the (global) bjerrum
length of the system, the reference permittivity ε is fixed via the formula
lB = e2 /(4πεkB T )
(5.40)
The local changes of ε are in reference to this value and can be seen as a spatially
dependent prefactor to this epsilon. If left unchanged, this prefactor is 1.0 for every site
by default.
5.7.7. Electrostatic Layer Correction (ELC)
Please cite [6] (BibTeX-key elc in file doc/ug/citations.bib) when using ELC,
and in addition [61] (BibTeX-key icelc in file doc/ug/citations.bib) if you use
dielectric interfaces.
Syntax
inter coulomb elc maximal pairwise error gap size
[far cutoff ] [noneutralization] [dielectric t m b ]
[dielectric-contrasts ∆t ∆b ] [capacitor U ]
Required features:
ELECTROSTATICS
Description
This is a special procedure that converts a 3d method, to a 2d method, in computational order N. Currently, it only supports P3M. This means, that you will first have
to set up the P3M algorithm (via inter coulomb p3m params) before using ELC. The
algorithm is definitely faster than MMM2D for larger numbers of particles (> 400 at
reasonable accuracy requirements). The maximal pairwise error maximal pairwise error
sets the LUB error of the force between any two charges without prefactors (see the papers). The algorithm tries to find parameters to meet this LUB requirements or will
throw an error if there are none.
The gap size gap size gives the height of the empty region between the system box
and the neighboring artificial images (again, see the paper). ESPResSo does not make
77
sure that the gap is actually empty, this is the users responsibility. The method will
compute fine of the condition is not fulfilled, however, the error bound will not be
reached. Therefore you should really make sure that the gap region is empty (e. g. by
constraints).
The setting of the far cutoff far cutoff is only intended for testing and allows to directly
set the cutoff. In this case, the maximal pairwise error is ignored. The periodicity has
to be set to 1 1 1 still, and the 3d method has to be set to epsilon metallic, i.e. metallic
boundary conditions. For details, see appendix E on page 248.
By default, ELC just as P3M adds a homogeneous neutralizing background to the
system in case of a net charge. However, unlike in three dimensions, this background adds
a parabolic potential across the slab [8]. Therefore, under normal circumstance, you will
probably want to disable the neutralization using [noneutralization]. This corresponds
then to a formal regularization of the forces and energies [8]. Also, if you add neutralizing
walls explicitely as constraints, you have to disable the neutralization.
The dielectric contrast features work exactly the same as for MMM2D, see the documentation above. Same accounts for capacitor U , but the constant potential is maintained between the xy-plane at z = 0 and z = L − gap size. The command efield caps
to read out the electric fields added by capacitor U also applies for the capacitor-feature
of ELC.
Make sure that you read the papers on ELC ([6, 61]) before using it.
5.7.8. Dielectric interfaces with the ICC? algorithm
Syntax
iccp3m n induced charges convergence convergence criterion areas areas
normals normals sigmas sigmas epsilons epsilons [eps_out eps out ]
[relax relaxation parameter ] [max_iterations max iterations ]
[ext_field ext field ]
Required features:
ELECTROSTATICS
Description
The ICC? algorithm allows to take into account arbitrarily shaped dielectric interfaces.
This is done by iterating the charge on the particles with the ids 0 to n induced particles − 1
until the correctly represent the influence of the dielectric discontinuity. It relies on a
coulomb solver that is already initialized. This Coulomb solver can be P3M, P3M+ELC,
MMM2D or MMM1D. As most of the times, ICC? will be used with P3M the corresponding command is called iccp3m.
Please make sure to read the corresponding articles, mainly[7, 59, 32] before using it.
The particles with ids 0 to n induced particles − 1 are treated as iterated particles by
ICC?. The constitute the dielectric interface and should be fixed in space. The parameters areas and epsilons are Tcl lists containing one floating point number describing
each surface elements area and dielectric constant. sigmas allows to take into account a
(bare) charge density, thus a surface charge density in absence of any charge induction.
normals is a Tcl list of Tcl lists with three floating point numbers describing the outward
pointing normal vectors for every surface element. The parameter convergence criterion
78
allows to specify the accuracy of the iteration. It corresponds to the maximum relative
change of any of the interface particle’s charge. After max iterations the iteration stops
anyways. The dielectric constant in bulk, i. e. outside the dielectric walls is specified
by eps out. A homogenous electric field can be added to the calculation of dielectric
boundary forces by specifying it in the parameter ext field .
Quick setup of dielectric interfaces
Syntax
(1) dielectric sphere center cx cy cz radius r res res
(2) dielectric wall normal nx ny nz dist d res res
(3) dielectric cylinder center cx cy cz axis ax ay az radius r
direction d
(4) dielectric pore center cx cy cz axis ax ay az radius r length l
smoothing_radius rs res res
(5) dielectric slitpore pore_mouth z channel_width c
pore_width w pore_length l upper_smoothing_radius us
lower_smoothing_radius ls
Description
The command dielectric allows to conveniently create dielectric interfaces similar to
the constraint and the lbboundary command. Currently the creation of spherical, cylindrical and planar geometries as well as a pore and slitpore geometry is supported. Please
check the documentation of the corresponding constraint for the detailed geometry. It is
implemented in Tcl and places particles in the right positions and adds the correct values
to the global Tcl variables icc areas icc normals icc sigmas icc epsilons and increases
the global Tcl variable varn induced charges. Thus after setting up the shapes, it is still
necessary to register them by calling iccp3m, usually in the following way:
iccp3m $n_induced_charges epsilons $icc_epsilons normals
$icc_normals areas $icc_areas sigmas $icc_sigmas
5.8. Dipolar interaction
Syntax
(1) inter magnetic 0.0
(2) inter magnetic
(3) inter magnetic parameters
Description
These commands can be used to set up magnetostatic interactions, which is defined as
follows:
µi · ~r)(~
µj · ~r)
(~
µi · µ
~ j ) 3(~
D−P 3M
−
(5.41)
U
(~r) = lB kB T
r3
r5
79
where r = |~r|.
lB is a dimensionless parameter similar to the Bjerrum length in electrostatics which
helps to tune the effect of the medium on the magnetic interaction between two magnetic
dipoles.
Computing magnetostatic interactions is computationally very expensive. ESPResSo
features some state-of-the-art algorithms to deal with these interactions as efficiently as
possible, but almost all of them require some knowledge to use them properly. Uneducated use can result in completely unphysical simulations.
The commands above work as their couterparts for the electrostatic interactions
(see section 5.7.1 on page 70). Variant (1) disables dipolar interactions. Variant
(2) returns the current parameters of the dipolar interaction as a Tcl-list using the
same syntax as used to setup the method, e.g.
{coulomb 1.0 p3m 7.75 8 5 0.1138 0.0}
{coulomb epsilon 0.1 n_interpol 32768 mesh_off 0.5 0.5 0.5}
Variant (3) is the generic syntax to set up a specific method or its parameters,
the details of which are described in the following subsections. Note that using the
magnetostatic interaction also requires assigning dipole moments to the particles.
This is done using the part command to set the dipole moment dip, e.g.
inter coulomb 1.0 p3m tune accuracy 1e-4
part 0 dip 1 0 0; part 1 dip 0 0 1
5.8.1. Dipolar P3M
Syntax
inter magnetic lB p3m rcut mesh cao alpha
Required features:
DIPOLES
Description
This command activates the P3M method to compute the dipolar interactions between
charged particles. The different parameters are described in more detail in [10].
rcut The real space cutoff as a positive floating point number.
mesh The number of mesh points, as a single positive integer.
cao The charge-assignment order, an integer between 0 and 7.
alpha The Ewald parameter as a positive floating point number.
Make sure that you know the relevance of the P3M parameters before using P3M! If
you are not sure, read the following references [22, 27, 33, 16, 17, 18, 15].
Note that dipolar P3M does not work with non-cubic boxes.
80
Tuning dipolar P3M
Syntax
inter magnetic lB p3m ( tune | tunev2 ) accuracy accuracy
[r_cut rcut ] [mesh mesh] [cao cao] [alpha α]
Required features:
DIPOLES
Description
Tuning dipolar P3M works exactly as tuning Coulomb P3M. Therefore, for details on how
to tune the algorothm, refer to the documentation of Coulomb P3M (see section 5.7.1
on page 71).
For the magnetic case, the expressions of the error estimate are given in [10].
5.8.2. Dipolar Layer Correction (DLC)
Syntax
inter magnetic mdlc accuracy gap size [far cutoff ]
Required features:
DIPOLES
Description
Like ELC but applied to the case of magnetic dipoles, but here the accuracy is the one
you wish for computing the energy. farc utoff is set to a value that, assuming all dipoles
to be as larger as the largest of the dipoles in the system, the error for the energy would
be smaller thant the value given by accuracy. At this moment you cannot compute
the accuracy for the forces, or torques, nonetheless, usually you will have an error for
forces and torques smaller than for energies. Thus, the error for the energies is an upper
boundary to all errors in the calculations.
At present, the program assumes that the gap without particles is along the z-direction.
The gap-size is the length along the z-direction of the volume where particles are not
allowed to enter.
As a reference for the DLC method, see [9].
5.8.3. Dipolar all-with-all and no replicas (DAWAANR)
Syntax
inter magnetic lB dawaanr
Required features:
DIPOLES
Description
This interaction calculates energies and forces between dipoles by explicitly summing
over all pairs. For the directions in which the system is periodic (as defined by setmd
periodic), it applies the minimum image convention, i.e. the interaction is effectively
cut off at half a box length.
81
In periodic systems, this method should only be used if it is not possible to use dipolar
P3M or DLC, because those methods have a far better accuracy and are much faster.
In a non-periodic system, the DAWAANR-method gives the exact result.
5.8.4. Magnetic Dipolar Direct Sum (MDDS)
Syntax
inter magnetic lB mdds n_cut value n cut
Required features:
DIPOLES
MAGNETIC_DIPOLAR_DIRECT_SUM
Description
The command enables the “magnetic dipolar direct sum”. The dipole-dipole interaction
is computed by explicitly summing over all pairs. If the system is periodic in one or
more directions, the interactions with further value n cut replicas of the system in all
periodic directions is explicitly computed.
As it is very slow, this method is not intended to do simulations, but rather to check
the results you get from more efficient methods like P3M.
5.9. Special interaction commands
5.9.1. Tunable-slip boundary interaction
Syntax
inter type1 type2 tunable_slip T γL rcut δt vx vy vz
Required features:
TUNABLE_SLIP
Description
Simulating microchannel flow phenomena like the Plane Poiseuille and the Plane Couette
Flow require accurate boundary conditions. There are two main boundary conditions in
use:
1. slip boundary condition which means that the flow velocity at the the hydrodynamic boundaries is zero.
2. partial-slip boundary condition which means that the flow velocity at the hydrodynamic boundaries does not vanish.
In recent years, experiments have indicated that the no-slip boundary condition is
indeed usually not valid on the micrometer scale. Instead, it has to be replaced by the
partial-slip boundary condition
δB ∂n vk krB = vk krB ,
where vk denotes the tangential component of the velocity and ∂n vk its spatial derivative
normal to the surface, both evaluated at the position rB of the so-called hydrodynamic
82
boundary. This boundary condition is characterized by two effective parameters, namely
(i) the slip length δB and (ii) the hydrodynamic boundary rB .
Within the approach of the tunable-slip boundary interactions it is possible to tune the
slip length systematically from full-slip to no-slip. A coordinate-dependent Langevinequation describes a viscous layer in the vicinity of the channel walls which exerts an
additional friction on the fluid particles. T is the temperature, γL the friction coefficient
and rcut is the cut-off radius of this layer. δt is the timestep of the integration scheme.
With vx vy and vz it is possible to give the layer a reference velocity to create a Plane
Couette Flow. Make sure that the cutoff radius rcut is larger than the cutoff radius of
the constraint Lennard-Jones interactions. Otherwise there is no possibility that the
particles feel the viscous layer.
This method was tested for Dissipative Particle Dynamics but it is intended for mesoscopic simulation methods in general. Note, that to use tunable-slip boundary interactions you have to apply two wall constraints with Lennard-Jones in addition to the
tunable-slip interaction. Make sure that the cutoff radius rcut is larger than the cutoff
radius of the constraint Lennard-Jones interactions. Otherwise there is no possibility
that the particles feel the viscous layer. Please read reference [52] before using this
interaction.
5.9.2. DPD interaction
Syntax
inter type1 type2 inter_dpd gamma r cut wf tgamma tr cut twf
Required features:
INTER_DPD
Description
This is a special interaction that is to be used in conjunction with the Dissipative Particle
Dynamics algorithm 6.2.3 when the INTER_DPD implementation is used. The parameters
correspond to the parameters of the DPD thermostat 5.9.2, but can be set individually
for the different interactions.
5.9.3. Fixing the center of mass
Syntax
inter typeid1 typeid1 comfixed flag
Required features:
COMFIXED
Description
This interaction type applies a constraint on particles of type typeid1 such that during
the integration the center of mass of these particles is fixed. This is accomplished as
follows: The sum of all the forces acting on particles of type typeid1 are calculated. These
include all the forces due to other interaction types and also the thermostat. Next a
force equal in magnitude, but in the opposite direction is applied to all the particles.
This force is divided on the particles of type typeid1 relative to their respective mass.
83
Under periodic boundary conditions, this fixes the itinerant center of mass, that is, the
one obtained from the unfolded coordinates.
Note that the syntax of the declaration of comfixed interaction requires the same
particle type to be input twice. If different particle types are given in the input, the
program exits with an error message. flag can be set to 1 (which turns on the interaction)
or 0 (to turn off the interaction).
Since the necessary communication is lacking at present, this interaction only works
on a single node.
5.9.4. Pulling particles apart
Syntax
inter typeid1 typeid2 comforce flag dir force fratio
Required features:
COMFORCE
Description
The comforce interaction type enables one to pull away particle groups of two different
types. It is mainly designed for pulling experiments on bundles. Within a bundle of
molecules of type number typeid1 lets mark one molecule as of type typeid2 . Using
comforce one can apply a force such that t2 can be pulled away from the bundle. The
comforcef lag is set to 1 to turn on the interaction, and to 0 otherwise. The pulling
can be done in two different directions. Either parallel to the major axis of the bundle
(dir = 0) or perpendicular to the major axis of the bundle (dir = 1). force is used
to set the magnitude of the force. fratio is used to set the ratio of the force applied
on particles of typeid1 vs. typeid2 . This is useful if one has to keep the total applied
force on the bundle and on the target molecule the same. A force of magnitude force
is applied on typeid2 particles, and a force of magnitude (force * fratio) is applied on
typeid1 particles.
5.9.5. Capping the force during warmup
Syntax
inter forcecap ( Fmax | individual )
Description
Non-bonded interactions are often used to model the hard core repulsion between particles. Most of the potentials in the section are therefore singular at zero distance, and
forces usually become very large for distances below the particle size. This is not a problem during the simulation, as particles will simply avoid overlapping. However, creating
an initial dense random configuration without overlap is often difficult.
By artificially capping the forces, it is possible to simulate a system with overlaps.
By gradually raising the cap value Fmax , possible overlaps become unfavorable, and the
system equilibrates to a overlap free configuration.
84
This command will cap the force to Fmax , i.e. for particle distances which would
lead to larger forces than Fmax , the force remains at Fmax . Accordingly, the potential
is replaced by rFmax . Particles placed exactly on top of each other will be subject to a
force of magnitude Fmax along the first coordinate axis.
The force capping is switched off by setting Fmax = 0. Note that force capping always
applies to all Lennard-Jones, tabulated, Morse and Buckingham interactions regardless
of the particle types.
If instead of a force capping value, the string “individual” is given, the force capping
can be set individually for each interaction. The capping radius is in this case not
derived from the potential parameters, but is given by an additional signal floating
point parameter to the interaction.
85
6. Setting up the system
6.1. setmd: Setting global variables.
Syntax
(1) setmd variable
(2) setmd variable [value]+
Description
Variant (1) returns the value of the ESPResSo global variable variable, variant (2) can
be used to set the variable variable to value. The ’+’ in variant (2) means that for some
variables more than one value can be given (example: setmd boxl 5 5 5). The following
global variables can be set:
Better throw
some out (e.g.
switches)?
Missing: lattice switch, dpd tgamma, n rigidbonds
Which commands
can be used to
set the read-only
variables?
box_l (double[3]) Simulation box length. Note that if you change the box length
during the simulation, the folded particle coordinates will remain the same, i.e.,
the particle stay in the same image box, but at the same relative position in their
image box. If you want to scale the positions, use the change_volume command.
cell_grid (int[3], read-only) Dimension of the inner cell grid.
cell_size (double[3], read-only) Box-length of a cell.
dpd_gamma (double, read-only) Friction constant for the DPD thermostat.
dpd_r_cut (double, read-only) Cutoff for DPD thermostat.
gamma (double, read-only) Friction constant for the Langevin thermostat.
integ_switch (int, read-only) Internal switch which integrator to use.
lb_components (int, read-only) Number of fluid components.
local_box_l (int[3], read-only) Local simulation box length of the nodes.
max_cut (double, read-only) Maximal cutoff of real space interactions.
max_cut_nonbonded (double, read-only) Maximal cutoff of nonbonded real space
interactions.
max_cut_bonded (double, read-only) Maximal cutoff of bonded real space interactions.
86
max_num_cells (int) Maximal number of cells for the link cell algorithm. Reasonable values are between 125 and 1000, or for some problems (nt otalp articles /
nn odes).
max_part (int, read-only) Maximal identity of a particle. This is in general not
related to the number of particles!
max_range (double, read-only) Maximal range of real space interactions: maxc ut
+ skin.
max_skin (double, read-only) Maximal skin to be used for the link cell/verlet algorithm. This is the minimum of cell size - max range.
min_global_cut (double) Minimal total cutoff for real space. Effectively, this plus
the skin is the minimally possible cell size. Espresso typically determines this
value automatically, but some algorithms, e.g. virtual sites, require you to specify
it manually.
min_num_cells (int) Minimal number of cells for the link cell algorithm. Reasonable values range in 10−6 N 2 to 10−7 N 2 . In general just make sure that the Verlet
lists are not incredibly large. By default the minimum is 0, but for the automatic
P3M tuning it may be wise to set larger values for high particle numbers.
n_layers (int, read-only) Number of layers in cell structure LAYERED (see section 6.4 on page 94).
n_nodes (int, read-only) Number of nodes.
n_part (int, read-only) Total number of particles.
n_part_types (int, read-only) Number of particle types that were used so far in
the inter command (see chaptertcl:inter).
node_grid (int[3]) 3D node grid for real space domain decomposition (optional, if
unset an optimal set is chosen automatically).
nptiso_gamma0 (double, read-only)
Docs missing.
nptiso_gammav (double, read-only)
Docs missing.
npt_p_ext (double, read-only) Pressure for NPT simulations.
npt_p_inst (double) Pressure calculated during an NPT isotropic integration.
piston (double, read-only) Mass off the box when using NPT isotropic integrator.
periodicity (bool[3]) Specifies periodicity for the three directions. If the feature
PARTIAL PERIODIC is set, this variable can be set to (1,1,1) or (0,0,0) at the
moment. If not it is readonly and gives the default setting (1,1,1).
skin (double) Skin for the Verlet list.
temperature (double, read-only) Temperature of the simulation.
87
thermo_switch (double, read-only) Internal variable which thermostat to use.
time (double) The simulation time.
time_step (double) Time step for MD integration.
timings (int) Number of samples to (time-)average over.
transfer_rate (int, read-only) Transfer rate for VMD connection. You can use
this to transfer any integer value to the simulation from VMD.
verlet_flag (bool) Indicates whether the Verlet list will be rebuild. The program
decides this normally automatically based on your actions on the data.
verlet_reuse (double) Average number of integration steps the verlet list has been
re-used.
warnings (int) if non-zero (default), some warnings are printed out. Set this to
zero if you get annoyed by them.
6.2. thermostat: Setting up the thermostat
Syntax
(1) thermostat
(2) thermostat off
(3) thermostat parameters
Description
The thermostat command is used to change settings of the thermostat.
The different available thermostats will be described in the following subsections. Note
that for a simulation of the NPT ensemble, you need to use a standard thermostat for
the particle velocities (e.g. Langevin or DPD), and a thermostat for the box geometry
(e.g. the isotropic NPT thermostat).
You may combine different thermostats at your own risk by turning them on one by
one. Note that there is only one temperature for all thermostats, although for some
thermostats like the Langevin thermostat, particles can be assigned individual temperatures.
Since ESPResSo does not enforce a particular unit system, it cannot know about the
current value of the Boltzmann constant. Therefore, when specifying the temperature of
a thermostat, you actually do not define the temperature, but the value of the thermal
energy kB T in the current unit system (see the discussion on units, Section 1.4).
Variant (1) returns the thermostat parameters. A Tcl list is given containing all
the parameters needed to set the specific thermostat. (exactly the same as the input
command line, without the preceding thermostat).
Variant (2) turns off all thermostats and sets all thermostat variables to zero. Setting
temperature to zero also affects the way in which electrostatics are handled (see also
Section 5.7).
88
Variant (3) sets up one of the thermostats described below.
Note that their are three different types of noise which can be used in ESPResSo. The
one used typically in simulations is flat noise with the correct variance and it is the default
used in ESPResSo, though it can be explicitly specified using the feature FLATNOISE. You
can also employ Gaussian noise which is, in some sense, more realistic. Notably Gaussian
noise (activated using the feature GAUSSRANDOM) does a superior job of reproducing
higher order moments of the Maxwell-Boltzmann distribution. For typical generic coarsegrained polymers using FENE bonds the Gaussian noise tends to break the FENE bonds.
We thus offer a third type of noise, activate using the feature GAUSSRANDOMCUT, which
produces Gaussian random numbers but takes anything which is two standard deviations
(2σ) below or above zero and set it to −2σ or 2σ respectively. In all three cases the
distribution is made such that the second moment of the distribution is the same and
thus results in the same temperature.
6.2.1. Langevin thermostat
Syntax
thermostat langevin temperature gamma
Description
The Langevin thermostat consists of a friction and noise term coupled via the fluctuationdissipation theorem. The friction term is a function of the particle velocities. For a more
detailed explanation, refer to [25].
If the feature ROTATION is compiled in, the rotational degrees of freedom are also
coupled to the thermostat.
Using the Langevin thermostat, it is posible to set a temperature and a friction coefficient for every particle individually via the feature LANGEVIN_PER_PARTICLE. Consult
the reference of the part command (chapter 4) for information on how to achieve this.
6.2.2. GHMC thermostat
Syntax
thermostat ghmc temperature n md phi [-no_flip | -flip | -random_flip]
[-no_scale | -scale]
Description
ESPResSo implements Generalized Hybrid Monte Carlo (GHMC) as a thermostat. GHMC
is a simulation method for sampling the canonical ensemble [41]. The method consists of
MC cycles that combine a few constant energy MD steps, specified by n md , followed by
a Metropolis criterion for their acceptance. Prior to integration, the particles momenta
are mixed with momenta sampled from the appropriate Boltzmann distribution.
Given the particles momenta pj from the last j th GHMC cycle the new momenta are
generated by: pj+1 = cos(φ)pj + sin(φ)ξξ , where ξ is a noise vector of random Gaussian
89
variables with zero mean and variance 1/temperature (see [29] for more details). The
momenta mixing parameter cos(φ) corresponds to phi in the implementation.
In case the MD step is rejected, the particles momenta may be flipped. This is specified
by setting the -no_flip / -flip option, for the -random_flip option half of the rejected
MD steps randomly result in momenta flip. The default for momenta flip is -no_flip.
The ξ noise vector’s variance van be tuned to exactly 1/temperature by specifying the
-scale option. The default for temperature scaling is -no_scale.
6.2.3. Dissipative Particle Dynamics (DPD)
ESPResSo implements Dissipative Particle Dynamics (DPD) either via a global thermostat, or via a thermostat and a special DPD interaction between particle types. The
latter allows the user to specify friction coefficients on a per-interaction basis.
Thermostat DPD
Syntax
thermostat dpd temperature gamma r cut [ WF wf tgamma tr cut TWF twf ]
Required features:
DPD
or TRANS_DPD
Description
ESPResSo’s standard DPD thermostat implements the thermostat exactly as described in
[55]. We use the standard Velocity-Verlet integration scheme, e.g. DPD only influences
the calculation of the forces. No special measures have been taken to self-consistently
determine the velocities and the dissipative forces as it is for example described in [42].
DPD adds a velocity dependent dissipative force and a random force to the usual conservative pair forces (e.g. Lennard-Jones).
The dissipative force is calculated by
F~ijD = −ζwD (rij )(ˆ
rij · ~vij )ˆ
rij
(6.1)
F~ijR = σwR (rij )Θij rˆij
(6.2)
The random force by
where Θij ∈ [−0.5, 0.5[ is a uniformly distributed random number. The connection of σ
and ζ is given by the dissipation fluctuation theorem:
(σwR (rij )2 = ζwD (rij )kB T
(6.3)
The parameters gamma and rc ut define the strength of the friction ζ and the cutoff
radius.
According to the optional parameter WF (can be set to 0 or 1, default is 0) of the
thermostat command the functions wD and wR are chosen in the following way ( rij <
r cut ) :
r
(1 − rijc )2 , wf = 0
D
R
2
w (rij ) = (w (rij )) =
(6.4)
1
, wf = 1
90
For rij ≥ r cut wD and wR are identical to 0 in both cases.
The friction (dissipative) and noise (random) term are coupled via the fluctuationdissipation theorem. The friction term is a function of the relative velocity of particle
pairs. The DPD thermostat is better for dynamics than the Langevin thermostat, since
it mimics hydrodynamics in the system.
1
When using a Lennard-Jones interaction, r cut = 2 6 σ is a good value to choose, so
that the thermostat acts on the relative velocities between nearest neighbor particles.
Larger cutoffs including next nearest neighbors or even more are unphysical.
gamma is basically an inverse timescale on which the system thermally equilibrates.
Values between 0.1 and 1 are o.k, but you propably want to try this out yourself to get
a feeling for how fast temperature jumps during a simulation are. The dpd thermostat
does not act on the system center of mass motion. Therefore, before using dpd, you
have to stop the center of mass motion of your system, which you can achieve by using
the command galilei_transform 6.8. This may be repeated once in a while for long
runs due to round off errors (check this with the command system_CMS_velocity) 6.8.
Two restrictions apply for the dpd implementation of ESPResSo:
• As soon as at least one of the two interacting particles is fixed (see 4 on how to fix
a particle in space) the dissipative and the stochastic force part is set to zero for
both particles (you should only change this hardcoded behaviour if you are sure
not to violate the dissipation fluctuation theorem).
• DPD does not take into account any internal rotational degrees of freedom of the
particles if ROTATION is switched on. Up to the current version DPD only acts on
the translatorial degrees of freedom.
Transverse DPD thermostat This is an extension of the above standard DPD thermostat [31], which dampens the degrees of freedom perpendicular on the axis between
two particles. To switch it on, the feature TRANS_DPD is required instead of the feature
DPD.
The dissipative force is calculated by
F~ijD = −ζwD (rij )(I − rˆij ⊗ rˆij ) · ~vij
(6.5)
~ ij
F~ijR = σwR (rij )(I − rˆij ⊗ rˆij ) · Θ
(6.6)
The random force by
The parameters tgamma tr cut define the strength of the friction and the cutoff in
the same way as above. Note: This thermostat does not conserve angular momentum.
Interaction DPD
Syntax
thermostat inter_dpd temperature
Required features:
INTER_DPD
91
Description
Another way to use DPD is by using the interaction DPD. In this case, DPD is implemented via a thermostat and corresponding interactions. The above command will set
the global temperature of the system, while the friction and other parameters have to
be set via the command inter inter_dpd (see 5.9.2 on page 83). This allows to set the
friction on a per-interaction basis.
DPD interactions with fixed particles is switched off by default, because it is not clear
if the results obtained with that method are physically correct. If you want activate
inter_dpd with fixed particles please use:
Syntax
thermostat inter_dpd ignore_fixed_particles 0
Required features:
INTER_DPD
Description
By default the flag ignore_fixed_particles is switched ON.
Other DPD extensions
The features DPD_MASS_RED or DPD_MASS_LIN make the friction constant mass dependent:
ζ → ζMij
There are two implemented cases.
mm
i j
• DPD_MASS_RED uses the reduced mass: Mij = 2 mi +m
j
• DPD_MASS_LIN uses the real mass: Mij =
mi +mj
2
The prefactors are such that equal masses result in a factor 1.
6.2.4. Isotropic NPT thermostat
Syntax
thermostat npt_isotropic temperature gamma0 gammaV
Required features:
NPT
Description
This theormstat is based on the Anderson thermostat (see [1, 39]) and will thermalize
the box geometry. It will only do isotropic changes of the box.
Be aware that this feature is neither properly examined for all systems nor is it maintained regularly. If you use it and notice strange behaviour, please contribute to solving
the problem.
92
6.3. nemd: Setting up non-equilibrium MD
Syntax
(1)
(2)
(3)
(4)
(5)
(6)
nemd
nemd
nemd
nemd
nemd
nemd
exchange n slabs n exchange
shearrate n slabs shearrate
off
profile
viscosity
Required features:
NEMD
Description
Use NEMD (Non Equilibrium Molecular Dynamics) to simulate a system under shear
with help of an unphysical momentum change in two slabs in the system.
Variants (1) and (2) will initialise NEMD. Two distinct methods exist. Both methods
divide the simulation box into n slab slabs that lie parallel to the x-y-plane and apply a
shear in x direction. The shear is applied in the top and the middle slabs. Note, that the
methods should be used with a DPD thermostat or in an NVE ensemble. Furthermore,
you should not use other special features like part fix or constraints inside the top
and middle slabs. For further reference on how NEMD is implemented into ESPResSo
see [54].
Variant (1) chooses the momentum exchange method. In this method, in each step the
n exchange largest positive x-components of the velocity in the middle slab are selected
and exchanged with the n exchange largest negative x-components of the velocity in the
top slab.
Variant (2) chooses the shear-rate method. In this method, the targetted x-component
of the mean velocity in the top and middle slabs are given by
target velocity = ±shearrate
Lz
4
(6.7)
where Lz is the simulation box size in z-direction. During the integration, the xcomponent of the mean velocities of the top and middle slabs are measured. Then,
the difference between the mean x-velocities and the target x-velocities are added to the
x-component of the velocities of the particles in the respective slabs.
Variant (3) will turn off NEMD, variant (4) will print usage information of the parameters of NEMD. Variant (5) will return the velocity profile of the system in x-direction
(mean velocity per slab).
Variant (6) will return the viscosity of the system, that is computed via
η=
F
γL
˙ x Ly
(6.8)
where F is the mean force (momentum transfer per unit time) acting on the slab, Lx Ly
is the area of the slab and γ˙ is the shearrate.
93
NEMD as implemented generates a Pouseille flow, with shear flow rate varying over a
finite wavelength determined by the box. For a planar Couette flow (constant shear, infinite wavelength), consider using Lees-Edwards boundary conditions (see 7.5 on page 103)
to drive the shear.
6.4. cellsystem: Setting up the cell system
This section deals with the flexible particle data organization of ESPResSo. Due to
different needs of different algorithms, ESPResSo is able to change the organization of
the particles in the computer memory, according to the needs of the used algorithms.
For details on the internal organization, refer to section 17.1 on page 218.
6.4.1. Domain decomposition
Syntax
cellsystem domain_decomposition [-no_verlet_list]
Description
This selects the domain decomposition cell scheme, using Verlet lists for the calculation
of the interactions. If you specify -no_verlet_list, only the domain decomposition is
used, but not the Verlet lists.
The domain decomposition cellsystem is the default system and suits most applications with short ranged interactions. The particles are divided up spatially into small
compartments, the cells, such that the cell size is larger than the maximal interaction
range. In this case interactions only occur between particles in adjacent cells. Since
the interaction range should be much smaller than the total system size, leaving out
all interactions between non-adjacent cells can mean a tremendous speed-up. Moreover,
since for constant interaction range, the number of particles in a cell depends only on
the density. The number of interactions is therefore of the order N instead of order N 2
if one has to calculate all pair interactions.
6.4.2. N-squared
Syntax
cellsystem nsquare
Description
This selects the very primitive nsquared cellsystem, which calculates the interactions for
all particle pairs. Therefore it loops over all particles, giving an unfavorable computation
time scaling of N 2 . However, algorithms like MMM1D or the plain Coulomb interaction
in the cell model require the calculation of all pair interactions.
In a multiple processor environment, the nsquared cellsystem uses a simple particle
balancing scheme to have a nearly equal number of particles per CPU, i.e. n nodes have
94
m particles, and p − n nodes have m + 1 particles, such that n ∗ m + (p − n) ∗ (m + 1) = N ,
the total number of particles. Therefore the computational load should be balanced
fairly equal among the nodes, with one exception: This code always uses one CPU for
the interaction between two different nodes. For an odd number of nodes, this is fine,
because the total number of interactions to calculate is a multiple of the number of
nodes, but for an even number of nodes, for each of the p − 1 communication rounds,
one processor is idle.
E.g. for 2 processors, there are 3 interactions: 0-0, 1-1, 0-1. Naturally, 0-0 and 1-1 are
treated by processor 0 and 1, respectively. But the 0-1 interaction is treated by node 1
alone, so the workload for this node is twice as high. For 3 processors, the interactions
are 0-0, 1-1, 2-2, 0-1, 1-2, 0-2. Of these interactions, node 0 treats 0-0 and 0-2, node 1
treats 1-1 and 0-1, and node 2 treats 2-2 and 1-2.
Therefore it is highly recommended that you use nsquared only with an odd number
of nodes, if with multiple processors at all.
6.4.3. Layered cell system
Syntax
cellsystem layered n layers
Description
This selects the layered cell system, which is specifically designed for the needs of the
MMM2D algorithm. Basically it consists of a nsquared algorithm in x and y, but a
domain decomposition along z, i. e. the system is cut into equally sized layers along the
z axis. The current implementation allows for the cpus to align only along the z axis,
therefore the processor grid has to have the form 1x1xN. However, each processor may
be responsible for several layers, which is determined by n layers, i. e. the system is
split into N*n layers layers along the z axis. Since in x and y direction there are no
processor boundaries, the implementation is basically just a stripped down version of
the domain decomposition cellsystem.
6.5. CUDA
Syntax
(1) cuda list
(2) cuda setdevice id
(3) cuda getdevice
Description
This command can be used to choose the GPU for all subsequent GPU-computations.
Note that due to driver limitations, the GPU cannot be changed anymore after the first
GPU-using command has been issued, for example lbfluid. If you do not choose the
GPU manually before that, CUDA internally chooses one, which is normally the most
powerful GPU available, but load-independent.
95
Variant (1) lists the available devices by their ids and brand names. Variant (2) allows
to choose the device by its id, which can be determined using cuda list, or for example
the deviceQuery example code in the CUDA SDK. Variant (3) finally gives the id of
the currently active GPU.
6.6. Creating bonds when particles collide
Please cite [7] (BibTeX-key espresso2 in file doc/ug/citations.bib) when using
dynamic bonding.
Syntax
(1)
(2)
(4)
(5)
on_collision
on_collision off
on_collision [exception] bind_centers d bond1
on_collision [exception] bind_at_point_of_collision d bond1 bond2
type
Description
With the help of the feature COLLISION_DETECTION, bonds between particles can be
created automatically during the simulation, every time two particles collide. This is
useful for simulations of chemical reactions and irreversible adhesion processes.
Two methods of binding are available:
• bind_centers adds a bonded interaction between the colliding particles at the
first collision. This leads to the distance between the particles being fixed, the
particles can, however still slide around each other.
The parameters are as follows: d is the distance at which the bond is created.
bond1 denotes a pair bond and is the type of the bond created between the colliding
particles. Particles that are already bound by a bond of this type do not get a new
bond, in order to avoid creating multiple bonds.
• bind_at_point_of_collision prevents sliding of the particles at the contact.
This is achieved by creating two virtual sites at the point of collision. They are
rigidly connectd to the colliding particles, respectively. A bond is then created
between the virtual sites, or an angular bond between the two real particles and
the virtual particles. In the latter case, the virtual particles are the centers of
the angle potentials (particle 2 in the description of the angle potential, see 5.5).
Due to the rigid connection between each of the particles in the collision and its
respective virtual site, a sliding at the contact point is no longer possible. See
the documentation on rigid bodies for details. In addition to the bond between
the virtual sites, the bond between the colliding particles is also created. You
can either use a real bonded interaction to prevent wobbling around the point of
contact or you can use a virtual bond to prevent additional force contributions, at
the expense of RATTLE, see 5.3.6.
96
The parameters d and bond1 are the same as for the bind_centers method. bond2
determines the type of the bond created between the virtual sites (if applicable),
and can be either a pair or a triple (angle) bond. If it is a pair bond, it connects
the two virtual particles, otherwise it constraints the angle between the two real
particles around the virtual ones. type denotes the particle type of the virtual sites
created at the point of collision (if applicable). Be sure not to define a short-ranged
interaction for this particle type, as two particles will be generated in the same
place.
The code can throw an exception (background error) in case two particles collide
for the first time, if the exception keyword is added to the invocation. In conjunction with the catch command of Tcl, this can be used to intercept the collision:
if {[catch {integrate 0} err]} {
foreach exception [lrange $err 2 end] {
if {[lrange $exception 0 2] == "collision between particles"} {
set i [lindex $exception 3]
set j [lindex $exception 5]
puts "particles $i and $j collided"
}
}
}
The following limitations currently apply for the collision detection:
• The method is currently limited to simulations with a single cpu
• No distinction is currently made between different particle types
• The “bind at point of collision” approach requires the VIRTUAL_SITES_RELATIVE
feature
• The “bind at point of collision” approach cannot handle collisions between virtual
sites
6.7. Catalytic Reactions
With the help of the feature CATALYTIC_REACTIONS, one can define three particle types
to act as reactant, catalyzer, and product. Using these reaction categories, we model
the following chemical reaction system:
rt
*
)
pr;
(6.9)
rt −
→ pr,
(6.10)
ct
where the first line indicates that there is a reversible chemical reaction that converts
the reactant particles (rt) into product (pr ) particles, leading to an equilibrium state.
The second line indicates that in the presence of a catalyst (ct) the forward reaction
97
pathway is favored, i.e., conversion of reactants into products. The equilibrium reaction
is described by the equilibrium constant
Keq =
keq,+
[pr]
,
=
keq,[rt]
(6.11)
with [rt] and [pr] the reactant and product concentration and keq,± the forward and
backward reaction rate constants, respectively. The rate constants that specify the
change in concentration for the equilibrium and catalytic reaction are given by
d[rt]
dt
d[pr]
dt
d[rt]
d[pt]
−
=
dt
dt
= keq,- [pr] − keq,+ [rt];
(6.12)
= keq,+ [rt] − keq,- [pr];
(6.13)
= kct [rt],
(6.14)
respectively.
In the current ESPResSo implementation we assume keq,+ = keq,- ≡ k eq and therefore
Keq = 1. The user can specify k eq ≥ 0 and k ct ≡ kct > 0. The former rate constant is
applied to all reactant and product particles in the system, whereas the latter is applied
only to the reactant particles in the vicinity of a catalyst particle. Reactant particles
that have a distance of r or less to at least one catalyzer particle are therefore converted
into product particles with rate constant k eq + k ct. The conversion of particles is done
stochastically on the basis of the relevant rate constant (k ≥ 0):
Pcvt = 1 − exp (−k∆t) ,
(6.15)
with Pcvt the probability of the conversion and ∆t the integration time step. If the
equilibrium rate constant is not specified it is assumed that k eq = 0.
Syntax
(0) reaction reactant_type rt catalyzer_type ct product_type pt range
r ct_rate k ct [eq_rate k eq] [react_once on/off ]
(1) reaction off
(2) reaction print
Required features:
a
CATALYTIC_REACTIONSa
The current implementation also requires the use of verlet lists and domain
decomposition.
Description
• Variant (0) defines a reaction with particles of type number rt as reactant, type ct
as catalyzer and type pt as product1 . The catalytic reaction rate constant is given
1
Only one type of particle can be assigned to each of these three reaction species and no particle type
may be assigned to multiple species. That is, ESPResSo currently does not support particles of type 1
and 2 both to be reactants, nor can particles of type 1 be a reactant as well as a catalyst. Moreover,
98
by k ct 2 and to override the default rate constant for the equilibrium reaction
(k eq = 0), one can specify it by k eq. By default each reactant particle is checked
against each catalyst particle (react_once off ). However, when creating smooth
surfaces using many catalyst particles, it can be desirable to let the reaction rate
be independent of the surface density of these particles. That is, each particle has
a likelihood of reacting in the vicinity of the surface (distance is less than r) as
specified by the rate constant, i.e., not according to Pcvt = 1−exp (−nk∆t), with n
the number of local catalysts. To accomplish this, each reactant is considered only
once each time step by using the option react_once on. The reaction command
is set up such that the different properties may be influenced individually.
• Variant (1) disables the reaction. Note that at the moment, there can only be one
reaction in the simulation.
• Variant (2) returns the current reaction parameters.
In future versions of ESPResSo the capabilities of the CATALYTIC_REACTIONS feature
may be generalized to handle multiple reactant, catalyzer, and product types, as well
as more general reaction schemes. Other changes may involve merging the current
implementation with the COLLISION_DETECTION feature.
6.8. Galilei Transform and Particle Velocity Manipulation
The following commands may be useful in effecting the velocity of the system.
6.8.1. Particle motion and rotation
Syntax
kill_particle_motion [rotation] 1
Required features:
1 ROTATION
Description
This command halts all particles in the current simulation, setting their velocities to
zero, as well as their angular momentum if the option rotation is specified and the
feature ROTATION has been compiled in.
only one of these reactions can be implemented in a single Tcl script. If, for instance, there is a
reaction involving particle types 1, 2, and 4, there cannot be a second reaction involving particles
of type 5, 6, and 8. It is however possible to modify the reaction properties for a given set of types
during the simulation.
2
Currently only strictly positive values of the catalytic conversion rate constant are allowed. Setting
the value to zero is equivalent to reaction off.
99
6.8.2. Forces and torques acting on the particles
Syntax
kill_particle_forces [torques] 1
Required features:
1 ROTATION
Description
This command sets all forces on the particles to zero, as well as all torques if the option
torque is specified and the feature ROTATION has been compiled in.
6.8.3. The centre of mass of the system
Syntax
system_CMS
Description
Returns the center of mass of the whole system. It currently does not factor in the
density fluctuations of the Lattice-Boltzman fluid.
6.8.4. The centre-of-mass velocity
Syntax
system_CMS_velocity
Description
Returns the velocity of the center of mass of the whole system.
6.8.5. The Galilei transform
Syntax
galilei_transform
Description
Substracts the velocity of the center of mass of the whole system from every particle’s
velocity, thereby performing a Galilei transform into the reference frame of the center
of mass of the system. This transformation is useful for example in combination with
the DPD thermostat, since there, a drift in the velocity of the whole system leads to an
offset in the reported temperature.
100
7. Running the simulation
7.1. integrate: Running the simulation
Syntax
(1) integrate steps [recalc_forces] [reuse_forces]
(2) integrate set [nvt]
(3) integrate set npt_isotropic pext piston [x y z ] [-cubic_box]
Description
ESPResSo uses the Velocity Verlet algorithm for the integration of the equations of
motion. The command integrate with an integer steps as parameter integrates the
system for steps time steps.
Note that this implementation of the Velocity Verlet algorithm reuses forces, that is,
they are computed once in the middle of the time step, but used twice, at the beginning
and end. However, in the first time step after setting up, there are no forces present
yet. Therefore, ESPResSo has to compute them before the first time step. That has two
consequences: first, random forces are redrawn, resulting in a narrower distribution of
the random forces, which we compensate by stretching. Second, coupling forces of e. g.
the Lattice Boltzmann fluid cannot be computed and are therefore lacking in the first
half time step. In order to minimize these effects, ESPResSo has a quite conservative
heuristics to decide whether a change makes it necessary to recompute forces before
the first time step. Therefore, calling hundred times integrate 1 does the same as
integrate 100, apart from some small calling overhead.
However, for checkpointing, there is no way for ESPResSo to tell that the forces that
you read back in actually match the parameters that are set. Therefore, ESPResSo would
recompute the forces before the first time step, which makes it essentially impossible to
checkpoint LB simulations, where it is vital to keep the coupling forces. To work around
this, integrate has an additional parameter [reuse forces], which tells integrate to not
recalculate the forces for the first time step, but use that the values still stored with
the particles. Use this only if you are absolutely sure that the forces stored match your
current setup!
The opposite problem occurs when timing interactions: In this case, one would like
to recompute the forces, despite the fact that they are already correctly calculated. To
this aim, the option [recalc forces] can be used to enforce force recalculation.
Two methods for the integration can be set: For an NVT ensemble (thermostat) and
for an NPT isotropic ensemble (barostat). The current method can be detected with
the command integrate set without any parameters.
101
The NVT integrator is set without parameters (the temperature can be set with the
thermostat). For the NPT ensemble, the parameters that can be added are:
• pext The external pressure as float variable. This parameter is required.
• piston The mass of the applied piston as float variable. This parameter is required.
• x :y:z Three integers to set the box geometry for non-cubic boxes. This parameter
is optional.
• -cubic_box If this optional parameter is added, a cubic box is assumed.
7.2. time_integration: Runtime of the integration loop
Syntax
(1) time_integration
(2) time_integration steps
Description
This command runs the integration as would the integrate command and returns the
wall runtime in seconds.
7.3. minimize_energy: Run steepest descent minimization
Syntax
(1) minimize_energy fmax steps gamma maxdisplacement
Description
This command runs a steepest descent energy minimization on the system. Please note
that the behaviour is undefined if either a thermostat, Maggs electrostatics or LatticeBoltzmann is activated. It runs a simple steepest descent algorithm:
Iterate
pi = pi + min(gamma × Fi , maxdisplacement),
while the maximal force is bigger than fmax or for at most steps times. The energy is relaxed by gamma, while the change per coordinate per step is limited to maxdisplacement.
The combination of gamma and maxdisplacement can be used to get an poor man’s
adaptive update. Please be aware of the fact that this needs not to converge to a local
minimum in periodic boundary conditions.
7.4. change_volume: Changing the box volume
Syntax
(1) change_volume Vnew
(2) change_volume Lnew ( x | y | z | xyz )
102
Description
Changes the volume of either a cubic simulation box to the new volume Vnew or its
given x-/y-/z-/xyz-extension to the new box-length Lnew , and isotropically adjusts the
particles coordinates as well. The function returns the new volume of the deformed
simulation box.
7.5. lees_edwards_offset: Applying shear between periodic
images
Syntax
(1) lees_edwards_offset offsetnew
Required features:
LEES_EDWARDS
Description
Lees-Edwards Periodic Boundary Conditions are used to impose a shear flow of speed
γ˙ on the system relative to its periodic images by moving the PBC wrap such that:
v xunf olded = v xf olded + γy
˙ img (where v xunf olded is the x-component of the velocity
of an image particle outside the main simulation box, and yimg is the count of PBC
boundaries crossed in the y-direction). The absolute value of the shear offset is set using
this command; with the shear flow rate γ˙ then determined internally as the difference
between successive offsets. A typical usage would be to integrate by 1 MD timestep
and then to increase the offset to a new value using this command; this usage pattern
is intended to allow for arbitrary shear flow time profiles, such as an oscillatory shear.
A common calculation to make using Lees-Edwards boundary conditions is to find the
shear viscosity (or kinematic viscosity) by plotting shear stress (or shear stress/density)
against the applied strain for different values of constant γ.
˙
Lees-Edwards differs from the NEMD approach (see 6.3 on page 93) in that the shear
imposed is homogenous across the system (but only on average: symmetry breaking
effects are not ruled out) rather than reversing direction with a periodicity of the box
length. Accordingly the transport properties measured using Lees-Edwards are likely to
be different to (and arguably more physical than) those measured using NEMD or those
from equilibrium simulations by a Green-Kubo type approach.
When the shear flow rate γ˙ is non-zero, the Langevin thermostat will treat v xf olded
as being relative to a flow field which changes smoothly from −γ/2
˙
at the bottom of
the periodic box to γ/2
˙
at the top. This ‘laminar’ thermostatting is provided mostly
because it gives quite convenient equilibration of a flowing system. In order to correctly
observe transport properties, symmetry-breaking or entropy production in relation to
shear flow is probably better to use the DPD thermostat (see 6.2.3 on page 90) once
the initial heat-up has been carried out. The DPD thermostat removes kinetic energy
from the system based on a frictional term defined relative to a local reference frame
of a given particle-pair, without enforcing any specific flow pattern a priori. At high
rates of dissipation, this can however lead to an artefactual shear-banding type effect
at the periodic boundaries, such that the bulk fluid is nearly stationary. This effect is
103
removed using the modification proposed to the DPD thermostat by Chatterjee [11] to
allow treatment of systems with high dissipation rates, which is applied automatically
if LEES_EDWARDS is compiled in. Chatterjee’s modification is just to skip calculation of
DPD forces (both dissipative and random) for particle pairs which cross a boundary in
y.
The function returns the old value of the offset.
If LEES_EDWARDS is compiled in, then coordinates are folded into the primary simulation box as the integration progresses, to prevent a numerical overflow.
7.6. Stopping particles
Use the following functions, also see Section 6.8:
• kill_particle_motion: halts all particles in the current simulation, setting their
velocities to zero, as well as their angular momentum if the feature ROTATION
has been compiled in.
• kill_particle_forces: sets all forces on the particles to zero, as well as all
torques if the feature ROTATION has been compiled in.
7.7. velocities: Setting the velocities
Syntax
velocities vmax [start pid ] [count N ]
Description
Sets the velocities of the particles with particle IDs between pid and pid +N to a random
vector with a length less than vmax , and returns the absolute value of the total velocity
assigned. By default, all particles are affected.
7.8. Fixing the particle sorting
Syntax
sort_particles
Description
Resorts the particles, making sure that
• the domain decomposition is strictly fullfilled, i.e. each particle is on the processor
and in the cell that its position belongs to
• the particles within each cell are ordered with ascending identity.
104
Both conditions together form a unique particle ordering. This is important when doing
checkpointing, because this makes sure that random numbers are applied in a specific
order. Therefore, after writing or reading a checkpoint, you should call sort_particles.
7.9. Parallel tempering
Syntax
parallel_tempering::main -rounds N -swap swap -perform perform
[-init init] [-values {Ti }] [-connect master ] [-port port]
[-load jnode ] [-resrate Nreset ] [-info info]
Description
This command can be used to run a parallel tempering simulation. Since the simulation
routines and the calculation of the swap probabilities are provided by the user, the
method is not limited to sampling in the temperature space. However, we assume in
the following that the sampled values are temperatures, and call them accordingly. It is
possible to use multiple processors via TCP/IP networking, but the number of processors
can be smaller than the number of temperatures.
Arguments
• swap specifies the name of the routine calculating the swap probability for a system. The routine has to accept three parameters: the id of the system to evaluate,
and two temperatures T1 and T2 . The routine should return a list containing
the energy of the system at temperatures T1 and T2 , respectively.
• perform specifies the name of the routine performing the simulation between two
swap tries. The routine has to accept two parameters: the id of the system to
propagate and the temperature T at which to run it. Return values are ignored.
• init specifies the name of a routine initializing a system. This routine can for
example create the particles, perform some intial equilibration or open output
files. The routine has to accept two parameters: the id of the system to initialize
and its initial temperature T . Return values are ignored.
• R specifies the number of swap trial rounds; in each round, neighboring temperatures are tried for swapping alternatingly, i.e. with four temperatures, The first
swap trial round tries to swap 1 ↔ 2 and 3 ↔ 4, the second round 2 ↔ 3, and so
on.
• master the name of the host on which the parallel tempering master node is running.
• port the TCP/IP port on which the parallel tempering master should listen. This
defaults to 12000.
• jnode specifies how many systems to run per ESPResSo-instance. If this is more
than 1, it is the user’s responsibility to manage the storage of configurations, see
105
below for examples. This defaults to 1.
• Rreset specifies after how many swap trial rounds to reset the counters for the
acceptance rate statistics. This defaults to 10.
• info specifies which output the parallel tempering code should produce:
none parallel tempering will be totally quiet, except for fatal errors
comm information on client activities, such as connecting, is printed to stderr
all print lots of information on swap energies and probabilities to stdout. This
is useful for debugging and quickly checking the acceptance rates.
This defaults to all.
Introduction
The basic idea of parallel tempering is to run N simulations with configurations Ci in
parallel at different temperatures T1 < T2 < . . . < TN , and exchange configurations
between neighboring temperatures. This is done according to the Boltzmann rule, i.e.
the swap probability for two configurations A and B at two different parameters T1 and
T2 is given by
min (1, exp − [β(T2 )UA (T2 ) + β(T1 )UB (T1 ) − β(T1 )UA (T1 ) − β(T2 )UB (T2 )]) ,
(7.1)
where UC (T ) denotes the potential energy of configuration C at parameter T and β(T )
the corresponding inverse temperature. If T is the temperature, UC is indepedent of T ,
and β(T ) = 1/(kB T ). In this case, the swap probability reduces to the textbook result
min(1, exp − [(1/T2 − 1/T1 ) (UA − UB ) /kB ] .
(7.2)
However, T can also be chosen to be any other parameter, for example the Bjerrum
length, i.e. the the strength of the electrostatic interaction. In this case, β(T ) = β is a
constant, but the energy UC (T ) of a configuration C depends on T , and one needs the
full expression (7.1). ESPResSo always uses this expression.
In practice, one does not swap configurations, but temperatures, simply because exchanging temperatures requires much less communication than exchanging the properties
of all particles.
Th ESPResSo implementation of parallel tempering repeatedly propagates all configurations Ci and tries to swap neighboring temperatures. After the first propagation,
the routine attempts to swap temperatures T1 and T2 , T3 and T4 , and so on. After the
second propagation, swaps are attempted between temperatures T2 and T3 , T4 and T5 ,
and so on. For the propagation, parallel tempering relies on a user routine; typically,
one will simply propagate the configuration by a few 100 MD time steps.
Details on usage and an example
The parallel tempering code has to be loaded explicitely by source "scripts/parallel_tempering.tcl" from the Espresso directory. To make use of the parallel tempering
106
tool, one needs to implement three methods: the propagation, the energy calculation
and an initialization routine for a configuration. A typical initialization routine will look
roughly like this:
proc init {id temp} {
# create output files for temperature temp
set f [open "out-$temp.dat" w]; close $f
init_particle_positions
thermostat langevin $temp 1.0
equilibration_integration
global config
set config($id) "{[part]} [setmd time]"
}
The last two lines are only necessary if each instance of ESPResSo handles more than
one configuration, e.g. if you have 300 temperatures, but only 10 ESPResSo processes
(i.e.-load 30). In this case, all user provided routines need to save and restore the
configurations. Saving the time is not necessary because the simulation tine across swaps
is not meaningful anyways; it is however convenient for investigating the (temperature)history of individual configurations.
A typical propagation routine accordingly looks like this
proc perform {id temp} {
global config
particle delete
foreach p [lindex $config($id) 0] { eval part $p }
setmd time [lindex $config($id) 1]
thermostat langevin $temp 1.0
set f [open "out-$temp.dat" a];
integrate 1000
puts $f "[setmd time] [analyze energy]"
close $f
set config($id) "{[part]} [setmd time]"
}
Again, the saving and storing of the current particle properties in the config array are
only necessary if there is more than one configuration per process. In practice, one will
rescale the velocities at the beginning of perform to match the current temperature,
otherwise the thermostat needs a short time to equilibrate. The energies necessary to
determine the swap probablility are calculated like this:
proc swap {id temp1 temp2} {
global config
particle delete
foreach p $config($id) { eval part $p }
set epot [expr [analyze energy total] - [analyze energy kinetic]]
107
return "[expr $epot/$temp1] [expr $epot/$temp2]"
}
Note that only the potential energy is taken into account. The temperature enters only
indirectly through the inverse temperature prefactor, see Eqn. (7.1).
The simulation is then started as follows. One of the processes runs the command
for {set T 0} {$T < 3} {set T [expr $T + 0.01]} {
lappend temperatures $T }
parallel_tempering::main -load 30 -values $temperatures -rounds 1000 \
-init init -swap swap -perform perform
This command turns the ESPResSo instance executing it into the master part of the
parallel tempering simulation. It waits until a sufficient number of clients has connected.
This are additional ESPResSo instances, which are identical to the master script, except
that they execute
parallel_tempering::main -connect $host -load 30 \
-init init -swap swap -perform perform
Here, host is a variable containing the TCP/IP hostname of the computer running
the master process. Note that the master process waits until enough processes have
connected to start the simulation. In the example, there are 300 temperatures, and each
process, including the master process, will deal with 30 of them. Therefore, 1 master
and 9 slave processes are required. For a typical queueing system, a starting routine
could look like this:
master=
for h in $HOSTS; do
if [ "$master" == "" ]; then
ssh $h "cd run; ./pt_test.tcl"
master=$h;
else
ssh $h "cd run; ./pt_test.tcl -connect $host"
fi
done
where pt_test.tcl passes the -connect option on to parallel_tempering::main.
Sharing data
Syntax
parallel_tempering::set_shareddata data
Description
can be used at any time by the master process to specify additional data that is available
on all processes of the parallel tempering simulation. The data is accessible from all
processes as parallel_tempering::shareddata.
108
7.10. Metadynamics
Syntax
(1) metadynamics
(2) metadynamics set off
(3) metadynamics set distance pid1 pid2 dmin dmax bheight bwidth fbound
dbins
(4) metadynamics set relative_z pid1 pid2 zmin zmax bheight bwidth fbound
zbins
(5) metadynamics print_stat current_coord
(6) metadynamics print_stat coord_values
(7) metadynamics print_stat profile
(8) metadynamics print_stat force
(9) metadynamics load_stat profile list force list
Required features: METADYNAMICS
Description
Performs metadynamics sampling. Metadynamics is an efficient scheme to calculate the
potential of mean force of a system as a function of a given reaction coordinate from
a canonical simulation. The user first chooses a reaction coordinate (e.g. distance)
between two particles (pid1 and pid2 ). As the system samples values along this reaction
coordinate (here the distance between pid1 and pid2 ), an iterative biased force pulls the
system away from the values of the reaction coordinate most sampled. Ultimately, the
system is driven in such a way that it self-diffuses along the reaction coordinate between
the two boundaries (here dmin and dmax ). The potential of mean force (or free energy
profile) can be extracted by reading the profile.
Arguments
• pid1 ID of the first particle involved in the metadynamics scheme.
• pid2 ID of the second particle involved in the metadynamics scheme.
• dmin , zmin : minimum value of the reaction coordinate. While dmin must be
positive (it’s a distance), zmin can be negative since it’s the relative height of pid1
with respect to pid2 .
• dmax , zmax : maximum value of the reaction coordinate.
• bheight height of the bias function.
• bwidth width of the bias function.
• fbound strength of the ramping force at the boundaries of the reaction coordinate
interval.
• dbins , zbins : number of bins of the reaction coordinate.
• profile list Tcl list of a previous metadynamics profile.
109
• force list Tcl list of a previous metadynamics force.
Details on usage
Variant (1) returns the status of the metadynamics routine. Variant (2) turns metadynamics off (default value). Variant (3) sets a metadynamics scheme with the reaction
coordinate distance, which corresponds to the distance between any two particles of the
system (e.g. calculate the potential of mean force of the end-to-end distance of a polymer). Variant (4) sets a metadynamics scheme with the reaction coordinate relative_z:
relative height (i.e. z coordinate) of particle pid1 with respect to pid2 (e.g. calculate the
potential of mean force of inserting one particle pid1 through an interface with center of
mass pid2 ). Variant (5) prints the current value of the reaction coordinate. Variant (6)
prints a list of the binned values of the reaction coordinate (e.g. dbins values between
dmin and dmax ). Variant (7) prints the current potential of mean force for all values of
the reaction coordinate considered. Variant (8) prints the current force (norm rather
than vector) for all values of the reaction coordinate considered. Variant (9) loads a
previous metadynamics sampling by reading a Tcl list of the potential of mean force and
applied force. This is especially useful to restart a simulation.
Note that the metadynamics scheme works seamlessly with the VIRTUAL SITES
feature, allowing to define centers of mass of groups of particles as end points of the
reaction coordinate. One can therefore measure the potential of mean force of the
distance between a particle and a molecule or interface.
The metadynamics scheme has (as of now) only been implemented for one processor:
MPI usage is not supported. However, one can speed up sampling by communicating the
profile and force between independent simulations (denoted walkers). The print_stat and load_stat can be used to input/output metadynamics information between
walkers at regular intervals. Warning: the information extracted from print_stat contains the entire history of the simulation, while only the last increment of sampling
should be communicated between walkers in order to avoid counting the same samples
multiple times.
Details on implementation
As of now, only two reaction coordinates have been implemented: distance and relative_z. Many different reaction coordinates can be set up, and it is rather easy to implement
new ones. See the code in metadynamics.{h,c} for further details.
The bias functions that are applied to the potential of mean force and the biased
force are not gaussian function (as in many metadynamics codes) but so-called Lucy
functions. See [40] for more details. These avoid the calculation of exponentials.
110
8. Analysis in Tcl
ESPResSo has two fundamentally different classes of observables for analyzing the systems. On the one hand, some observables are computed from the Tcl level. In that
case, the observable is measured in the moment that the corresponding Tcl function is
called, and the results are returned to the Tcl script. In general, observables in this class
should only be computed after a large number of timesteps, as switching forth and back
between the C- and the Tcl-level is costly. This chapter describes all observables in this
class.
On the other hand, some observables are computed and stored in the C-core of
ESPResSo during a call to the function integrate, while they are set up and their
results are collected from the Tcl level. These observables are more complex to implement and offer less flexibility, while the are significantly faster and more memory
efficient, and they can be set up to be computed every few timesteps. The observables
in this class are described in chapter 9.
The class of Tcl-level analysis functions is mainly controlled via the analyze command.
It has two main uses: Calculation of observables (analyze observable) and definition and
analysis of topologies in the system (analyze topologycommand ). In addition, ESPResSo
offers the command uwerr (see section 8.4 for computing statistical errors in time series.
8.1. Available observables
The command analyze provides online-calculation of local and global observables.
8.1.1. Minimal distances between particles
Syntax
(1) analyze mindist [type list a type list b]
(2) analyze distto pid
(3) analyze distto x y z
Description
Variant (1) returns the minimal distance between two particles in the system. If the
type-lists are given, then the minimal distance between particles of only those types is
determined.
distto returns the minimal distance of all particles to particle pid (variant (2)), or
to the coordinates (x , y, z ) (Variant (3)).
111
8.1.2. Particles in the neighbourhood
Syntax
(1) analyze nbhood pid r catch
(2) analyze nbhood x y z rc atch
Description
Returns a Tcl-list of the particle ids of all particles within a given radius r catch around
the position of the particle with number pid in variant (1) or around the spatial coordinate (x , y, z ) in variant (2).
8.1.3. Particle distribution
Syntax
analyze distribution part type list a part type list b
[rmin [rmax [rbins [log flag [int flag]]]]]
Description
Returns its parameters and the distance distribution of particles with types specified in
part type list a around particles with types specified in part type list b with distances
between rmin and rmax , binned into rbins bins. The bins are either equidistant (if
log flag = 0) or logarithmically equidistant (if log flag ≥ 1). If an integrated distribution
is required, use int flag = 1. The distance is defined as the minimal distance between a
particle of one group to any of the other group.
Output format
The output corresponds to the blockfile format (see section 10.2 on page 144):
{ parameters }
{
{ r dist(r ) }
.
.
.
}
8.1.4. Radial density map
Syntax
analyze radial_density_map xbins ybins xrange yrange
[axisofrotation centerofrotation beadtypelist [thetabins]]
Description
Returns the radial density of particles around a given axis. Parameters are:
Someone who
knows or wrote
this command
should check this.
• xbins histogram bins in x direction.
112
• ybins histogram bins in y direction.
• xrange range for analysis in x direction.
• yrange range for analysis in y direction.
• axisofrotation rotate around given axis. (x, y, or z)
• centerofrotation rotate around given point.
• beadtypelist only analyze beads of given types.
• thetabins histogram bins in angle theta.
8.1.5. Modes
Syntax
analyze modes2d
Description
Analyzes the modes of a configuration. Requires that a grid is set and that the system
contains more than two particles. Output are four numbers in the order:
htRE
htIM
θRE
θIM
8.1.6. Lipid orientation
Syntax
(1) analyze get_lipid_orients
(2) analyze lipid_orient_order
Description
Document the
usage!
8.1.7. Bilayers
Syntax
(1) analyze bilayer_set
(2) analyze bilayer_density_profile
Description
Document the
usage!
113
8.1.8. GPB
Syntax
analyze cell_gpb Manningparameter outercellradius innercellradius
[accuracy [numberofinteractions]]
Description
Document the
usage and what it
is!
8.1.9. Get folded positions
Syntax
analyze get_folded_positions [-molecule] [shift x y z ]
Description
Outputs the folded positions of particles. Without any parameters, the positions of all
particles are given, folded to the box length. The optional parameter -molecule ensures
that molecules (particle groups) are kept intact. The optional shift parameters can be
used to shift the not separated molecules if needed.
8.1.10. Vkappa
Syntax
analyze Vkappa [(
reset | read | set Vκ,1 Vκ,2 avk
) ]
Description
Calculates the compressibility V × κT through the Volume fluctuations V × κT =
β hV 2 i − hV i2 [34]. Given no arguments this function calculates and returns the current value of the running average for the volume fluctuations. The argument reset
clears the currently stored values. With read the cumulative mean volume, cumulative
mean squared volume and how many samples were used can be retrieved. Likewise the
option set enables you to set those.
8.1.11. Radial distribution function
Syntax
analyze ( rdf | <rdf> ) part type list a part type list b [rmin rmax rbins]
Description
Returns its parameters and the radial distribution function (rdf) of particles with types
specified in part type list a around particles with types specified in part type list b. The
range is given by rmin and rmax and is divided into rbins equidistant bins.
Output format
The output corresponds to the blockfile format (see section 10.2 on page 144):
114
{ parameters }
{
{ r rdf (r ) }
.
.
.
}
8.1.12. Structure factor
Syntax
analyze structurefactor type order
Description
Returns the spherically averaged structure factor S(q) for particles of a given type type.
2π
The S(q) is calculated for all possible wave vectors, 2π
L <= q <= L order . Do not chose
parameter order too large, because the number of calculations grows as order 3 .
Output format
The output corresponds to the blockfile format (see section 10.2 on page 144):
{ q value S (q) value }
.
.
.
8.1.13. Van-Hove autocorrelation function G(r, t)
Syntax
analyze vanhove type rmin rmax rbins [tmax ]
Description
Returns the van Hove auto correlation function G(r, t) and the mean square displacement
msd(t) for particles of type ptype for the configurations stored in the array configs. This
tool assumes that the configurations stored with analyze append (see section 8.3 on
page 126) are stored at equidistant time intervals. G(r, t) is calculated for each multiple
of this time intervals. For each time t the distribution of particle displacements is
calculated according to the specification given by rmin, rmax and rbins. Optional
argument tmax defines the maximum value of t for which G(r, t) is calculated. If it is
omitted or set to zero, maximum possible value is used. If the particles perform a random
walk (i.e. a normal diffusion process) G(r, t)/r2 is a Gaussian distribution for all times.
Deviations of this behavior hint on another diffusion process or on the fact that your
system has not reached the diffusive regime. In this case it is also very questionable to
calculate a diffusion constant from the mean square displacement via the Stokes-Einstein
relation.
Output format
The output corresponds to the blockfile format (see section 10.2 on page 144):
115
{ msd { msd (0 ) msd (1 ) ... } }
{ vanhove { { G(0 , 0 ) G(1 , 0 ) ... }
{ G(0 , 1 ) G(1 , 1 ) ... }
.
.
.
}
}
The G(r, t) are normalized such that the integral over space always yields 1.
8.1.14. Center of mass
Syntax
analyze centermass part type
Description
Returns the center of mass of particles of the given type.
8.1.15. Moment of inertia matrix
Syntax
(1) analyze momentofinertiamatrix typeid
(2) analyze find_principal_axis typeid
Description
Variant (1) returns the moment of inertia matrix for particles of given type typeid . The
output is a list of all the elements of the 3x3 matrix. Variant (2) returns the eigenvalues
and eigenvectors of the matrix.
8.1.16. Gyration tensor
Syntax
analyze gyration_tensor [typeid ]
Description
Analyze the gyration tensor of particles of a given type typeid , or of all particles in the
system if no type is given. Returns a Tcl-list containing the squared radius of gyration,
three shape descriptors (asphericity, acylindricity, and relative shape anisotropy), eigenvalues of the gyration tensor and their corresponding eigenvectors. The eigenvalues are
sorted in descending order.
8.1.17. Aggregation
Syntax
analyze aggregation dist criteria s mol id f mol id
[min contact [charge criteria]]
116
Description
Returns the aggregate size distribution for the molecules in the molecule id range
s mol id to f mol id . If any monomers in two different molecules are closer than
dist criteria they are considered to be in the same aggregate. One can use the optional min contact parameter to specify a minimum number of contacts such that only
molecules having at least min contact contacts will be considered to be in the same
aggregate. The second optional parameter charge criteria enables one to consider aggregation state of only oppositely charged particles.
8.1.18. Identifying pearl-necklace structures
Syntax
analyze necklace pearl threshold back dist space dist first length
Description
Algorithm for identifying pearl necklace structures for polyelectrolytes in poor solvent
[36]. The first three parameters are tuning parameters for the algorithm: pearl threshold
is the minimal number of monomers in a pearl. back dist is the number of monomers
along the chain backbone which are excluded from the space distance criterion to form
clusters. space dist is the distance between two monomers up to which they are considered to belong to the same clusters. The three parameters may be connected by scaling
arguments. Make sure that your results are only weakly dependent on the exact choice
of your parameters. For the algorithm the coordinates stored in partCfg are used. The
chain itself is defined by the identity first of its first monomer and the chain length
length. Attention: This function is very specific to the problem and might not give
useful results for other cases with similar structures.
8.1.19. Finding holes
Syntax
analyze holes typeidprobe mesh size
Required features:
LENNARD_JONES
Description
Function for the calculation of the unoccupied volume (often also called free volume)
in a system. Details can be found in Schmitz and Muller-Plathe [49]. It identifies free
space in the simulation box via a mesh based cluster algorithm. Free space is defined
via a probe particle and its interactions with other particles which have to be defined
through LJ interactions with the other existing particle types via the inter command
before calling this routine. A point of the mesh is counted as free space if the distance of
the point is larger than LJ cut+LJ offset to any particle as defined by the LJ interaction
parameters between the probe particle type and other particle types. How to use this
function: Define interactions between all (or the ones you are interested in) particle
types in your system and a fictitious particle type. Practically one uses the van der
117
Waals radius of the particles plus the size of the probe you want to use as the Lennard
Jones cutoff. The mesh spacing is the box length divided by the meshs ize.
Output format
{ n holes mean hole size max hole size free volume fraction
{ sizes }
{ surfaces }
{ element lists }
}
I think there is
still a bug in there
(Hanjo)
A hole is defined as a continuous cluster of mesh elements that belong to the unoccupied volume. Since the function is quite rudimentary it gives back the whole information
suitable for further processing on the script level. sizes and surfaces are given in number
of mesh points, which means you have to calculate the actual size via the corresponding
volume or surface elements yourself. The complete information is given in the element lists for each hole. The element numbers give the position of a mesh point in the linear
representation of the 3D grid (coordinates are in the order x, y, z). Attention: the
algorithm assumes a cubic box. Surface results have not been tested. .
8.1.20. Temperature of the LB fluid
Syntax
analyze fluid temp 1 or 2 or 3
Required features:
1 LB
2 LB_GPU
3 ELECTROKINETICS
Description
This command returns the temperature of the lattice-Boltzmann (LB) fluid, see Chapter 12, by averaging over the fluid nodes. In case LB_BOUNDARIES or LB_BOUNDARIES_GPU
are compiled in and boundaries are defined, only the available fluid volume is taken into
account.
8.1.21. Energies
Syntax
(1)
(2)
(3)
(4)
Describe the different energies
components returned by the different commands!
analyze
analyze
analyze
analyze
energy
energy ( total | kinetic | coulomb | magnetic )
energy bonded bondid
energy nonbonded typeid1 typeid2
Description
Returns the energies of the system. Variant (1) returns all the contributions to the
total energy. Variant (2) returns the numerical value of the total energy or its kinetic
or Coulomb or magnetic contributions only. Variants (3) and (4) return the energy
contributions of the bonded resp. non-bonded interactions.
118
Output format (variant (1))
{ energy value } { kinetic value } { interaction value } ...
8.1.22. Pressure
Syntax
(1) analyze pressure
(2) analyze pressure total
(3) analyze pressure ( totals | ideal | coulomb |
tot_nonbonded_inter | tot_nonbonded_intra | vs_relative )
(4) analyze pressure bonded bondid
(5) analyze pressure nonbonded typeid1 typeid2
(6) analyze pressure nonbonded_intra [typeid ]
(7) analyze pressure nonbonded_inter [typeid ]
Description
Computes the pressure and its contributions in the system. Variant (1) returns all
the contributions to the total pressure. Variant (2) will return the total pressure only.
Variants (3), (4) and (5) return the corresponding contributions to the total pressure.
Warning: Pressure works only with certain interactions and features. Read
in detail before use!
The pressure is calculated (if there are no electrostatic interactions) by
P
2Ekinetic
j>i Fij rij
p=
+
Vf
3V
(8.1)
Document arguments nb inter,
nb intra, tot nb inter and tot nb intra
where f = 3 is the number of translational degrees of freedom of each particle, V is the
volume of the system, Ekinetic is the kinetic energy, Fij the force between particles i and
j, and rij is the distance between them. The kinetic energy divided by the degrees of
freedom is
2Ekinetic
1X
=
mi vi2 .
(8.2)
f
3
i
Note that Equation 8.1 can only be applied to pair potentials and central forces.
Description of how contributions from other interactions are calculated is beyond the
scope of this manual. Three body potentials are implemented following the procedure in
Ref. [58]. A different formula is used to calculate contribution from electrostatic interactions in P3M. For electrostatic interactions, the k-space contribution is not well tested,
so use with caution! Anything outside that is currently not implemented. Four-body
dihedral potentials are not included. In case of rigid body rotation, virial contribution
from torques is not included. The pressure contribution for rigid bodies constructed by
means of the VIRTUAL SITES RELATIVE mechanism is included. On the other hand,
the pressure contribution for rigid bonds is not included. All other constraints of any
119
Description of
how electrostatic
contribution to
Pressure is calculated
kind are not currently accounted for in the pressure calculations. The pressure is no
longer correct, e.g., when particles are confined to a plane.
The command is implemented in parallel.
Output format (variant (1))
{ { pressure total pressure }
{ ideal ideal gas pressure }
{ { bond type pressure }
.
.
.
}
{ { nonbonded type pressure }
.
.
.
}
{ coulomb pressure }
}
specifying the pressure, the ideal gas pressure, the contributions from bonded interactions, the contributions from non-bonded interactions and the electrostatic contributions.
8.1.23. Stress Tensor
Syntax
(1) analyze stress_tensor
(2) analyze stress_tensor total
(3) analyze stress_tensor ( totals | ideal | coulomb |
tot_nonbonded_inter | tot_nonbonded_intra )
(4) analyze stress_tensor bonded bondt ype
(5) analyze stress_tensor nonbonded typeid1 typeid2
(6) analyze stress_tensor nonbonded_intra [typeid ]
(7) analyze stress_tensor nonbonded_inter [typeid ]
Description
Computes the stress tensor of the system. The various options are equivalent to those
described by analyze pressure in 8.1.22 on the preceding page. It is called a stress
tensor but the sign convention follows that of a pressure tensor.
Warning: Stress tensor works only with certain interactions and features.
Same restrictions as in the case of Pressure are applicable (see section 8.1.22).
The stress tensor is calculated by
P
P
(k) (l)
(k) (l)
m
v
v
i
j>i Fij rij
(kl)
i
i
i
p
=
+
(8.3)
V
V
where the notation is the same as for analyze pressure in 8.1.22 on the previous page
and the superscripts k and l correspond to the components in the tensors and vectors.
120
Note that the angular velocities of the particles are not included in the calculation of
the stress tensor.
The command is implemented in parallel.
Output format (variant (1))
{ { pressure total pressure tensor }
{ ideal ideal gas pressure tensor }
{ { bond type pressure tensor }
.
.
.
}
{ { nonbonded type pressure tensor }
.
.
.
}
{ coulomb pressure tensor }
}
specifying the pressure tensor, the ideal gas pressure tensor, the contributions from
bonded interactions, the contributions from non-bonded interactions and the electrostatic contributions.
8.1.24. Local Stress Tensor
Syntax
analyze local_stress_tensor periodic x periodic y periodic z range start x
range start y range start z range x range y range z bins x bins y
bins z
Description
Computes local stress tensors in the system. A cuboid is defined starting at the coordinate (range start x ,range start y,range start z ) and going to the coordinate (range start x +range x ,
range start y+range y, range start z +range z ). This cuboid in divided into bins x bins
in the x direction, bins y bins in the y direction and bins z bins in the z direction such
that the total number of bins is bins x *bins y*bins z . For each of these bins a stress
tensor is calculated using the Irving Kirkwood method. That is, a given interaction
contributes towards the stress tensor in a bin proportional to the fraction of the line
connecting the two particles that is within the bin.
If the P3M and MMM1D electrostatic methods are used, these interactions are not
included in the local stress tensor. The DH and RF methods, in contrast, are included.
Concerning bonded interactions only two body interactions (FENE, Harmonic) are included (angular and dihedral are not). For all electrostatic interactions only the real
space part is included.
Care should be taken when using constraints of any kind, since these are not accounted
for in the local stress tensor calculations.
The command is implemented in parallel.
121
Output format (variant (1))
{ { LocalStressTensor }
{ { x bin y bin z bin } { pressure tensor } }
.
.
.
}
specifying the local pressure tensor in each bin.
8.2. Analyzing groups of particles (molecules)
Syntax
(1) analyze set chains [chain start n chains chain length]
(2) analyze set topo_part_sync
(3) analyze set
Description
The above set of functions is designed to facilitate analysis of molecules. Molecules are
expected to be a group of particles comprising a contiguous range of particle IDs. Each
molecule is a set of consecutively numbered particles and all molecules are supposed to
consist of the same number of particles. Some functions in this group require that the
particles constituting a molecule are connected into linear chains (particle n is connected
to n + 1 and so on) while others are applicable to molecules of whatever topology.
The analyze set command defines the structure of the current system to be used
with some of the analysis functions.
Variant (1) defines a set of n chains chains of equal length chain length which start
with the particle with particle number chain start and are consecutively numbered (i.e.
the last particle in that topology has number chain start + n chains ∗ chain length − 1).
Variant (2) synchronizes topology and particle data, assigning mol id values to particles.
Variant (3) will return the chains currently stored.
8.2.1. Chains
All analysis functions in this section require the topology of the chains to be set correctly.
The topology can be provided upon calling. This (re-)sets the structure info permanently,
i.e. it is only required once.
End-to-end distance
Syntax
analyze ( re | <re> ) [chain start n chains chain length]
122
Description
Returns the quadratic end-to-end-distance and its root averaged over all chains. If
<re> is used, the distance is averaged over all stored configurations (see section 8.3 on
page 126).
Output format
{ re error of re re2 error of re2 }
Radius of gyration
Syntax
analyze ( rg | <rg> ) [chain start n chains chain length]
Description
Returns the radius of gyration averaged over all chains. It is a radius of a sphere, which
would have the same moment of inertia as the molecule, defined as
2
RG
=
N
1 X
(~ri − ~rcm )2 ,
N
(8.4)
i=1
where ~ri are position vectors of individual particles constituting a molecule and ~rcm is
the position vector of its centre of mass. The sum runs over all N particles comprising
the molecule. For more information see any polymer science book, e.g. [47]. If <rg> is
used, the radius of gyration is averaged over all stored configurations (see section 8.3 on
page 126).
Output format
{ rg error of rg rg2 error of rg2 }
Hydrodynamic radius
Syntax
analyze ( rh | <rh> ) [chain start n chains chain length]
Description
Returns the hydrodynamic radius averaged over all chains. If <rh> is used, the hydrodynamic radius is averaged over all stored configurations (see section 8.3 on page 126).
The following formula is used for the computation:
N N
1
2 XX
1
= 2
,
RH
N
|~ri − ~rj |
(8.5)
i=1 j=i
The above-mentioned formula is only valid under certain assumptions. For more information, see Chapter 4 and equation 4.102 in [19].
Output format
{ rh error of rh }
123
Internal distances
Syntax
analyze ( internal_dist | <internal_dist> ) [chain start n chains chain length]
Description
Returns the averaged internal distances within the chains (over all pairs of particles).
If <internal_dist> is used, the values are averaged over all stored configurations (see
section 8.3 on page 126).
Output format
{ idf (0 ) idf (1 ) ... idf (chain length − 1 ) }
The index corresponds to the number of beads between the two monomers considered
(0 = next neighbours, 1 = one monomer in between, . . . ).
Internal distances II (specific monomer)
Syntax
analyze ( bond_dist | <bond_dist> ) [index index ]
[chain start n chains chain length]
Description
In contrast to analyze internal_dist, it does not average over the whole chain, but
rather takes the chain monomer at position index (default: 0, i.e. the first monomer
on the chain) to be the reference point to which all internal distances are calculated.
If <bond_dist> is used, the values will be averaged over all stored configurations (see
section 8.3 on page 126).
Output format
{ bdf (0 ) bdf (1 ) ... bdf (chain length − 1 − index ) }
Bond lengths
Syntax
analyze ( bond_l | <bond_l> ) [chain start n chains chain length]
Description
Analyzes the bond lengths of the chains in the system. Returns its average, the standard
deviation, the maximum and the minimum. If you want to look only at specific chains,
use the optional arguments, i.e. chain start = 2 ∗ MPC and n chains = 1 to only
include the third chain’s monomers. If <bond_l> is used, the value will be averaged
over all stored configurations (see section 8.3 on page 126). This function assumes linear
chain topology and does not check if the bonds really exist!
Output format
{ mean stddev max min }
124
Form factor
Syntax
analyze ( formfactor | <formfactor> ) qmin qmax qbins
[chain start n chains chain length]
Description
Computes the spherically averaged form factor of a single chain, which is defined by
S(q) =
1
chain length
length
chain
X
i,j=1
sin(qrij )
qrij
(8.6)
of a single chain, averaged over all chains for qbin + 1 logarithmically spaced q-vectors
qmin, . . . , qmax where qmin > 0 and qmax > qmin. If <formfactor> is used, the form
factor will be averaged over all stored configurations (see section 8.3 on the following
page).
Output format
{
{ q S (q) }
.
.
.
}
with q ∈ {qmin, . . . , qmax }.
Chain radial distribution function
Syntax
analyze rdfchain rmin rmax rbins [chain start n chains chain length]
Description
Returns three radial distribution functions (rdf) for the chains. The first rdf is calculated
for monomers belonging to different chains, the second rdf is for the centers of mass of
the chains and the third one is the distribution of the closest distances between the
chains (i.e. the shortest monomer-monomer distances). The distance range is given by
rmin and rmax and it is divided into rbins equidistant bins.
Output format
{
{r rdf1 (r ) rdf2 (r ) rdf3 (r ) }
.
.
.
}
125
Check this!
Mean square displacement of chains
Syntax
(1) analyze ( <g1>| <g2>| <g3> ) [chain start n chains chain length]
(2) analyze g123 [-init] [chain start n chains chain length]
Description
Variant (1) returns
• the mean-square displacement of the beads in the chain (<g1>)
• the mean-square displacement of the beads relative to the center of mass of the
chain (<g2>)
• or the motion of the center of mass (<g3>)
averaged over all stored configurations (see section 8.3). At short time scales, g1 and
g2 coincide, since the motion of the center of mass is much slower. At large timescales
g1 and g3 coincide and correspond to the center of mass motion, while g2 levels off. g2
and g3 together correspond to g1. For details, see Grest and Kremer [25].
Variant (2) returns all of these observables for the current configuration, as compared
to the reference configuration. The reference configuration is set, when the option -init
is used.
Output format (variant (1))
{ gi (0 ∗ dt) gi (1 ∗ dt) ... }
Output format (variant (2))
{ g1 (t) g2 (t) g3 (t) }
8.3. Storing configurations
Some observables (i.e. non-static ones) require knowledge of the particles’ positions
at more than one or two times. Therefore, it is possible to store configurations for
later analysis. Using this mechanism, the program is also able to work quasi-offline by
successively reading in previously saved configurations and storing them to perform any
analysis desired afterwards.
Note that the time at which configurations were taken is not stored. The most observables that work with the set of stored configurations do expect that the configurations
are taken at equidistant timesteps.
Note also, that the stored configurations can be written to a file and read from it via
the blockfile command (see section 10.2 on page 144).
126
8.3.1. Storing and removing configurations
Syntax
(1)
(2)
(3)
(4)
(5)
analyze
analyze
analyze
analyze
analyze
append
remove [index ]
replace index
push [size]
configs config
Description
Variant (1) appends the current configuration to the set of stored configurations. Variant
(2) removes the index th stored configuration, or all, if index is not specified. Variant
(3) will replace the index th configuration with the current configuration.
Variant (4) will append the current configuration to the set of stored configuration
and remove configurations from the beginning of the set until the number of stored
configurations is equal to size. If size is not specified, only the first configuration in the
set is removed.
Variants (1) to (4) return the number of currently stored configurations.
Variant (5) will append the configuration config to the set of stored configurations.
config has to define coordinates for all configurations in the format:
{x1 y1 z1 x2 y2 z2 ... }
8.3.2. Getting the stored configurations
Syntax
(1) analyze configs
(2) analyze stored
Description
Variant (1) returns all stored configurations, while variant (2) returns only the number
of stored configurations.
Output format (variant (1))
{
{x1 y1 z1 x2 y2 z2 ... }
.
.
.
}
8.4. uwerr: Computing statistical errors in time series
Syntax
(1) uwerr data nrep col [s tau] [plot]
(2) uwerr data nrep f [s tau [f args]] [plot]
127
Description
Calculates the mean value, the error and the error of the error for an arbitrary numerical
time series according to Wolff [62].
Arguments
• data is a matrix filled with the primary estimates ai,r
α from R replica with N1 , N2 , . . . , NR
measurements each.

a1,1
1
a2,1
1
..
.
a21,1
a22,1
..
.
a1,1
3
a2,1
3
..
.
···
···
..
.











 N1 ,1

N1 ,1
1 ,1
data =  a1
aN
a
·
·
·

2
3
 1,2

1,2
1,2
 a1
a2
a3
··· 


..
..
..
.. 


.
.
.
. 
R ,R
R ,R
R ,R
aN
aN
aN
···
1
2
3
• nrep is a vector whose elements specify the length of the individual replica.
nrep = (N1 , N2 , . . . , NR )
• f is a user defined Tcl function returning a double with first argument a vector
which has as many entries as data has columns. If f is given instead of the column,
the corresponding derived quantity is analyzed.
• f args are further arguments to f .
• s tau is the estimate S = τ /τint as explained in section (3.3) of [62]. The default
is 1.5 and it is never taken larger than minR
r=1 Nr /2.
• [plot] If plot is specified, you will get the plots of Γ/Γ(0) and τint vs. W . The
data and gnuplot script is written to the current directory.
Output format
mean error error of error act
error of act [Q]
where act denotes the integrated autocorrelation time, and Q denotes a quality measure, i.e. the probability to find a χ2 fit of the replica estimates.
The function returns an error message if the windowing failed or if the error in one of
the replica is to large.
128
9. Analysis in the core
Analysis in the core is a new concept introduced in ESPResSo since version 3.1. It was
motivated by the fact, that sometimes it is desirable that the analysis functions do more
than just return a value to the scripting interface. For some observables it is desirable
to be sampled every few integrations steps. In addition, it should be possible to pass
the observable values to other functions which compute history-dependent quantities,
such as correlation functions. All this should be done without the need to interrupt
the integration by passing the control to the script level and back, which produces a
significant overhead when performed too often.
Some observables in the core have their corresponding counterparts in the Tcl observables of the analyze command described in Chapter 8. However, only the coreobservables can be used on the fly with the toolbox of the correlator and on the fly
analysis of time series. Similarly, some special cases of using the correlator have their
redundant counterparts in the analysis in Tcl (Chapter 8), but the correlator provides a
general and versatile toolbox which can be used with any implemented core-observables.
The only trick to bridge the gap between Tcl based analysis and core analysis is the
tclcommand observable that allows use the return value of arbitrary Tcl functions (also
self-written) as input for the core analysis. See more below.
9.1. Observables
9.1.1. Introduction
The first step of the core analysis is to tell ESPResSo to create an observable. An
observable in the sense of the core analysis can be considered as a rule how to compute
a certain set of numbers from a given state of the system or a role how to collect data
from other observables. Any observable is represented as a single array of double values.
Any more complex shape (tensor, complex number, . . . ) must be compatible to this
prerequisite. Every observable however documents the storage order.
Creating an observable means just allocating the corresponding memory, assigning a
function to compute the observable value and reserving an id which will be used to refer
to the observable. In addition to the possibility to print the observable value (return
the observable value to the script interface), the id of a core-observable can be passed
to another analysis function. The observable value is computed from the current state
of the system at the moment when it is needed, i.e. when requested explicitly by the
user calling the observable print function or when requested automatically by some
other analysis function. Updating is an orthogonal concept: Observables that collect
129
data over time (e.g. the average observable) need to be updated regularly, even though
their current value is not of interest.
Not all observables are implemented in parallel. When performing a parallel computation, too frequent updates to observables which are not implemented in parallel may
produce a significant slowdown.
9.1.2. Creating an observable
To create a new observable, use
Syntax
observable new name [parameters+]
Description
Upon this call, ESPResSo allocates the necessary amount of memory and returns an
integer id which will be used later to refer to the observable. The parameter name and
further arguments have to correspond to one of the observables described below.
Available observables
Missing descriptions of parameters of several
observables
Currently the following observables are implemented. Particle specifications (see section 9.1.6 below) define a group of particles, from which the observable should be calculated. They are generic to all observables and are described after the list of observables.
Here are the observables, that only depend on the current state of the simulation
system:
• particle_positions particle specifications
Positions of the particles, in the format x1 , y1 , z1 , x2 , y2 , z2 , . . . xn , yn , zn .
The particles are ordered ascending according to their ids.
• particle_velocities particle specifications
Velocities of the particles, in the format
v1x , v1y , v1z , v2x , v2y , v2z , . . . vnx , vny , vnz . The particles are ordered ascending
according to their ids.
• particle_body_velocities particle specifications
Velocities of the particles in the body frame, in the format
v1x , v1y , v1z , v2x , v2y , v2z , . . . vnx , vny , vnz . The particles are ordered ascending
according to their ids. This command only produces a meaningful result when
ROTATIONS is compiled in.
• particle_forces particle specifications
Forces on the particles, in the format
f1x , f1y , f1z , f2x , f2y , f2z , . . . fnx , fny , fnz . The particles are ordered ascending
according to their ids.
130
• particle_angular_momentum particle specifications
Angular momenta (omega) of the particles, in the format
ω1x , ω1y , ω1z , ω2x , ω2y , ω2z , . . . ωnx , ωny , ωnz . The particles are ordered ascending
according to their ids and the angular velocity/momentum is specified in the laboratory frame.
• particle_body_angular_momentum particle specifications
Angular momenta (omega) of the particles, in the format
ω1x , ω1y , ω1z , ω2x , ω2y , ω2z , . . . ωnx , ωny , ωnz . The particles are ordered ascending
according to their ids and the angular velocity/momentum is specified in the body
(co-rotating) frame.
• com_position particle specifications [blocked size]
Position of the centre of mass. If blocked size is specified, the particles are subdivided into blocks of size size and the centre of mass position is calculated for each
block separately.
• com_velocity particle specifications [blocked size]
Velocity of the centre of mass. If blocked size is specified, the particles are subdivided into blocks of size size and the centre of mass velocity is calculated for each
block separately.
• com_force particle specifications [blocked size]
Total force on the specified particles. If blocked size is specified, the particles are
subdivided into blocks of size size and the total force is calculated for each block
separately.
• stress_tensor
The stress tensor. It only works with all particles. It is returned as a 9-dimensional
array:
{ σxx , σxy , σxz , σyx , σyy , σyz , σzx , σzy , σzz }
• stress_tensor_acf_obs
The observable for computation of the Stress tensor autocorrelation function. Similarly to the stress tensor, it only works with all particles. It is returned as a 6dimensional array:
{ σxy , σyz , σzx , (σxx − σyy ), (σxx − σzz ), (σyy − σzz ) }
where σij are the components of the stress tensor.
• particle_currents particle specifications
Electric currents due to individual particles. For a particle i: jix = qi vix /∆t where
∆t is the simulation time step. Required feature: ELECTROSTATICS
• currents particle specifications
P
x
Electric currents averaged over all particles: j x =
i qi vi /∆t where ∆t is the
simulation time step. Required feature: ELECTROSTATICS
131
any suggestion for
a more suitable
name?
• dipole_moment particle specifications
P
x
The dipole moment of the specified group of particles: µx =
i qi ri Required
feature: ELECTROSTATICS
• interacts_with particle specifications1 particle specifications2 cutoff
For each particle belonging to particle specifications1 the observable is unity if
a neighbour of a type from particle specifications2 is found within the distance
defined by the cutoff . If no such neighbour is found, the observable is zero. The
observable has one dimension per each particle of particle specifications1
• density_profile particle specifications profile specifications
Compute the density profile within the specified cube. For profile specifications,
see section 9.1.7.
• force_density_profile particle specifications profile specifications
Compute the force density profile within the specified cube. For profile specifications, see section 9.1.7.
• lb_velocity_profile particle specifications profile specifications
Compute the Lattice-Boltzmann velocity profile within the specified cube. For
profile specifications, see section 9.1.7.
• flux_density_profile particle specifications profile specifications
Compute the flux density within the specified cube. For profile specifications, see
section 9.1.7.
• radial_density_profile Compute the density profile in cylindrical coordinates.
For profile specifications, see section 9.1.7.
• radial_flux_density_profile
Compute the flux density profile in cylindrical coordinates. For profile specifications, see section 9.1.7.
• lb_radial_velocity_profile
Compute the Lattice-Boltzmann velocity profile in cylindrical coordinates. For
profile specifications, see section 9.1.7.
• rdf type listtype list[r min[r max [n bins]]]
Compute the radial distribution function, see 8.1.11.
• structure_factor order
Compute the structure factor. Remember it scales as order3 , see 8.1.12.
• radial_density_distribution type < type > minr < minr >\
maxr < maxr > rbins < rbins > (start point < X >< Y >< Z >\
end point < X >< Y >< Z > |id start point < id > id end point < id >)
Computes the radial density distribution for particles of the given type around the
axis given either as fixed positions or as particle ids. The binning is done between
minr and maxr with rbins.
132
• spatial_polymer_property (ids < id list > |type < type >)N < Npoly >
Calculates the mean charge along Npoly weak polyelectrolytes of the same length.
The particles can be specified as a list of particle ids or through the particle type.
• persistence_length ids < id list > max d < max d > cut off < cut off >
Calculates the persistence length of the given polymer specified through the id_list.
max_d is the maximum distance (in terms of particles) for which the correlation
is computed. With cut_off the number of particles at the polymer ends that are
ignored for the calculation, can be specified.
• polymer_pair_correlation ids < id list > maxr < maxr > minr < minr > k < k >\
rbins < rbins > N < Npoly > poly len < poly len >
Computes the pair correlation of particles on a polymer chain given through the
id_list, here k is the distance (in terms of particles) between the mononers on the
chain for which the pair correlation is computed. This is averaged over the whole
chain and all the given chains. The number of polymers for which this distribution
is computed has to be given as Npoly and their length through poly_len. The
distribution is calculated for distances between minr and maxr with rbins.
The tclcommand observable is a helpful tool, that allows to make the analysis framework
much more versatile, by allowing the evaluation of arbitrary tcl commands.
• tclcommand dimQ command
An arbitrary Tcl function that returns a list of floating point numbers of fixed size
dimQ can be specified. Although its execution might be slow, it allows to prototype
new observables without a lot of trouble. Many existing analysis commands can
be made to cooperate with the core analysis that way.
The following commands allow to collect data automatically over time once their autoupdate feature is enabled.
• average ref
The running average of the reference observable with id ref . It can be resetted by
observable no reset
9.1.3. Printing an observable
Syntax
observable id print [formatted]
Description
Prints the value of the observable with a given id. If the observable refers to the current
state of the system, its value is updated before printing.
9.1.4. Passing an observable to an analysis function
Currently the only analysis function which uses the core observables is the correlator
(section 9.2).
133
Formatted printing is not fully
supported yet.
9.1.5. Deleting an observable to an analysis function
Syntax
observable id delete
Does not work yet
Description
Deletes the observable, i.e. frees the allocated memory and makes the id free for a new
observable.
9.1.6. Particle specifications
You can specify from which particles the observable should be computed in one of the
following ways. In all cases, particle specifications refer to the current state of espresso.
Any later changes to particles (additions, deletions, changes of types) will not be automatically reflected in the observable.
• all
Requests observable calculation based on all particles in the system.
• types type list
Restricts observable calculation to a given particle type(s). The type list is a tcl
list of existing particle types.
• id id list
Restricts observable calculation to a given list of particle id(s). The id list is a tcl
list of existing particle ids.
9.1.7. Profile specifications
Profiles are specified by giving the spacial area that is to be profiled and the number of
bins in each spacial direction. The area to be analyzed is characterized by minx /maxx
miny/maxy and minz /maxz . The defaults correspond to the box size when the observable is created. The bin size in each direction defaults to 1, and can be change with
the parameter xbins/ybins/zbins. Changing one, two or three of them to a value > 1
with thus create a one-, two- or three-dimensional map of the desired quantity. The full
syntax thus reads as:
Syntax
observable new needs_profile_specs [other parameters] [ minx minx ]
[ maxx maxx ] [ miny miny ] [ maxy maxy ] [ minz minz ]
[ maxz maxz ] [ xbins xbins ] [ ybins ybins ] [ zbins zbins ]
Description
Radial profiles allow to do the same as usual profiles, except the coordinate system is a
cylindrical one and the binning is done in the cylindrical coordinates (defined with the
axis in z-direction). This is very helpful if the symmetry of the system is cylindrical.
134
The spacial are is characterized by a center (default to the center of the box) a maximum radial position maxr (defaults to the smaller value of the box lengths in x and
y directions) and a minimum and maximum value of z. It is possible to also resolve
different polar angles, thus using it as a full 3D mapping tool, but this will only rarely
be used. The full syntax is:
Syntax
observable new needs_radial_profile_specs [other parameters]
[ center <cx> <cy> <cx> ] [ maxr maxr ] [ minz minz ]
[ maxz maxz ] [ rbins rbins ] [ phibins phibins ] [ zbins zbins ]
Description
9.2. Correlations
9.2.1. Introduction
Time correlation functions are ubiquitous in statistical mechanics and molecular simulations when dynamical properties of many-body systems are concerned. A prominent
example is the velocity autocorrelation function, hv(t) · v(t + τ )i which is used in the
Green-Kubo relations. In general, time correlation functions are of the form
C(τ ) = hA (t) ⊗ B (t + τ )i ,
(9.1)
where t is time, τ is the lag time (time difference) between the measurements of (vector)
observables A and B, and ⊗ is an operator which produces the vector quantity C from
A and B. The ensemble average h·i is taken over all time origins t. Correlation functions
describing dynamics of large and complex molecules such as polymers span many orders
of magnitude, ranging from MD time step up to the total simulation time.
ESPResSo uses a fast correlation algorithm (see section 9.2.6) which enables efficient
computation of correlation functions spanning many orders of magnitude in the lag time.
The generic correlation interface of ESPResSo may process either observables defined
in the kernel, or data which it reads from an external file or values entered through the
scripting interface. Thus, apart from data processing on the fly, it can also be used as
an efficient correlator for stored data. In all cases it produces a matrix of n + 2 columns.
The first two columns are the values of lag times τ and the number of samples taken
for a particular value of τ . The remaining ones are the elements of the n-dimensional
vector C(τ ).
The uwerr command for computing averages and error estimates of a time series
of observables relies on estimates of autocorrelation functions and the respective autocorrelation times. The correlator provides the same functionality as a by-product of
computing the correlation function (see section 9.2.5.
An example of the usage of observables and correlations is provided in the script
correlation.tcl in the samples directory.
135
Processing data
from Tcl input or
from input files
is not fully supported yet.
9.2.2. Creating a correlation
Correlation first has to be defined by saying which observables are to be correlated,
what should be the correlation operation, sampling frequency, etc. When a correlation
is defined, its id is returned which is used further to do other operations with the correlation. The correlation can be either updated automatically on the fly without direct
user intervention, or by an explicit user call for an update.
Syntax
correlation new obs1 id1 [obs2 id2 ] corr_operation
operation dt dt tau_max tau max [tau_lin tau lin]
[compress1 name [compress2 name] ]
Description
Defines a new correlation and returns an integer id which has been assigned to it. Its
further arguments are described below.
Arguments
• obs1 and obs2
are ids of the observables A and B that are to correlated. The ids have to refer
to existing observables which have been previously defined by the observable
command. Some observables are already implemented, and others can be easily
added. This can be done with very limited ESPResSo knowledge just by following
the implementations that are already in. If obs2 is omitted, autocorrelation of
obs1 is calculated by default.
• corr_operation
The operation that is performed on A(t) and B(t + τ ) to obtain C(τ ). The
following operations are currently is available:
• scalar_product
P
Scalar product of A and B, i.e. C = Ai Bi
i
• componentwise_product
Comnponentwise product of A and B, i.e. Ci = Ai Bi
• square_distance_componentwise
Each component of the correlation vector is the square of the difference
between the corresponding components of the observables, i.e. Ci = (Ai −
Bi )2 . Example: when A is particle_positions, it produces the mean
square displacement (for each component separately).
• tensor_product
Tensor product of A and B, i.e. Ci·lB +j = Ai Bj , with lB the length of B.
• complex_conjugate_product
• fcs_acf wx wy wz
Complex conjugate product must
be defined.
136
Fluorescence Correlation Spectroscopy (FCS) autocorrelation function, i.e.
Gi (τ ) =
∆x2 (τ ) ∆y 2 (τ ) ∆z 2 (τ ) E
1D
i
i
i
exp −
−
−
,
N
wx2
wy2
wz2
(9.2)
2
where ∆x2i (τ ) = xi (0) − xi (τ ) is the square discplacement of particle i
in the x direction, and wx is the beam waist of the intensity profile of the
exciting laser beam,
2x2 2y 2 2z 2 W (x, y, z) = I0 exp − 2 − 2 − 2 .
wx
wy
wz
(9.3)
Equation 9.2 is a generalization of the formula presented by H¨ofling et
al. [28]. For more information, see references therein. Per each 3 dimensions of the observable, one dimension of the correlation output is produced.
If fcs_acf is used with other observables than particle_positions, the
physical meaning of the result is unclear.
• dt
The time interval of sampling data points. When autoupdate is used, dt has to be
a multiple of timestep. It is also used to produce time axis in real units. Warning:
if dt is close to the timestep, autoupdate is strongly recommended. Otherwise cpu
time is wasted on passing the control between the script and kernel.
• tau_max
This is the maximum value of τ for which the correlation should be computed.
Warning: Unless you are using the multiple tau correlator, choosing tau max of
more than 100dt will result in a huge computational overhead. In a multiple tau
correlator with reasonable parameters, tau max can span the entire simulation
without too much additional cpu time.
• tau_lin
The number of data-points for which the results are linearly spaced in tau. This
is a parameter of the multiple tau correlator. If you want to use it, make sure
that you know how it works. By default, it is set equal to tau max which results
in the trivial linear correlator. By setting tau lin ¡ tau max the multiple tau
correlator is switched on. In many cases, tau lin=16 is a good choice but this may
strongly depend on the observables you are correlating. For more information,
we recommend to read Ref. [45] or to perform your own tests.
• compress1 and compress2
Are functions used to compress the data when going to the next level of the
multiple tau correlator. Different compression functions for different observables
can be specified if desired, otherwise the same function is used for both. Default
is discard which takes one of the observable values and discards the other one.
This is safe for all observables but produces poor statistics in the tail. For some
observables, linear compression can be used which makes an average of two
137
neighbouring values but produces systematic errors. Depending on the observable,
the systematic error can be anything between harmless and disastrous. For more
information, we recommend to read Ref. [45] or to perform your own tests.
9.2.3. Inquiring about already existing correlations
Syntax
(1) correlation
(2) correlation n_corr
Maybe not all
parameters are
printed.
Description
Variant (1) returns a tcl list of the defined correlations including their parameters.
Variant (2) returns the number of currently defined correlations.
9.2.4. Collecting time series data for the correlation
Syntax
(1) correlation id autoupdate { start | stop}
(2) correlation id update
(3) correlation id finalize
Description
Variant (1) is the recommended way of updating the correlations. By specifying start
or stop it starts or stops automatically updating the correlation estimates. The automatic updates are done within the integration loop without further user intervention.
The update frequency is adjusted based on the value of dt provided when defining the
correlation. Note that autoupdate has to be started after setting the sim-time (e.g. after
setmdtime0 ).
Variant (2) is an explicit call for an instantaneous update of the correlation estimates,
using the current system state. It is only possible to use (2) if the correlation is not
being autoupdated. However, it is possible to use it after autoupdate has been stopped.
When updating by an explicit call, ESPResSo does not check if the lag time between two
updates corresponds the value of dt specified when creating the correlation.
Variant (3) correlates all data from history which are left in the buffers. Once this
has been done, the history is lost and no further updates are possible. When a new
observable value is passed to a correlation, level 0 of the compression buffers of the
multiple tau correlator (see section 9.2.6 for details) is updated immediately. Higher
levels are updated only when the lower level buffers are filled and there is a need to push
some values one level up. When the updating is stopped, a number of observable values
have not reached the higher level, especially when taum ax is comparable to the total
simulation time and if there are many compression levels. In such case, variant (3) is
very useful. If tau max is much shorter, it does not have a big effect.
138
9.2.5. Printing out the correlation and related quantities
Syntax
(1) correlation id write_to_file filename
(2) correlation id print
(3a) correlation id print [ average1 | variance1 | correlation_time ]
(3b) correlation id print [ average_errorbars ]
Description
Variant (1) writes the current status of the correlation estimate to the specified filename.
If the file exists, its contents will be overwritten.
Output format
The output looks as follows:
tau1 n_samples C1 C2 ... Cn
tau2 n_samples C1 C2 ... Cn
Where each line corresponds to a given value of tau, n_samples is the number of samples
which contributed to the correlation at this level and Ci are the individual components
of the correlation.
Variant (2) returns the current status of the correlation estimate as a Tcl variable.
Output format
The output looks as follows:
tau1 n_samples C1 C2 ... Cn
tau2 n_samples C1 C2 ... Cn
Variants (3a) and (3b) return the corresponding estimate of the statistical property
as a Tcl variable.
average1 prints the average of observable1.
variance1 prints the variance of observable1.
correlation_time prints the estimate of the correlation time.
average_errorbars prints the estimate of the error of the average based on the method
according to [62] (same as used by the uwerr command).
9.2.6. The correlation algorithm: multiple tau correlator
Here we briefly describe the multiple tau correlator which is implemented in ESPResSo.
For a more detailed description and discussion of its behaviour with respect to statistical
and systematic errors, please read the cited literature. This type of correlator has been
in use for years in the analysis of dynamic light scattering [48]. About a decade later
it found its way to the Fluorescence Correlation Spectroscopy (FCS) [37]. The book
of Frenkel and Smit [23] describes its application for the special case of the velocity
autocorrelation function.
139
Compression
level
0
τ =2
τ = p−1
τ = p−2
τ =1
i=0
i=1
i=2
i=3
.........
i=4 i=5
i = p−2
i = p−1
τ = 2(p−1)
τ =p
1
i=0
i=2
i=4
i=6
.....
i=p
τ = 2p
2
i=0
i=4
........
i = 2p
τ = p+2
i = p+2
.....
τ = 2(p+2)
i = 2(p+2)
.....
i = 2(p−1)
τ = 4(p−1)
i = 4(p−1)
Figure 9.1.: Schematic representation of buffers in the correlator.
Let us consider a set of N observable values as schematically shown in Figures 9.1,
where a value of index i was measured in time iδt. We are interested in computing the
correlation function according to Equation ?? for a range lag times τ = (i − j)δt between
the measurements i and j. To simplify the notation, we further drop δt when referring
to observables and lag times.
The trivial implementation takes all possible pairs of values corresponding to lag
times τ ∈ [τmin : τmax ]. Without loss of generality, let us further
consider τmin = 0. The
2
computational effort for such an algorithm scales as O τmax
. As a rule of thumb, this
3
is feasible if τmax < 10 . The multiple tau correlator provides a solution to compute
the correlation functions for arbitrary range of the lag times by coarse-graining the
high τ values. It applies the naive algorithm to a relatively small range of lag times
τ ∈ [0 : p − 1]. This we refer to as compression level 0. To compute the correlations for
lag times τ ∈ [p : 2(p − 1)], the original data are first coarse-grained, so that m values of
the original data are compressed to produce a single data point in the higher compression
level. Thus the lag time between the neighbouring values in the higher compression level
increases by a factor of m, while the number of stored values decreases by the same
factor and the number of correlation operations at this level reduces by a factor of m2 .
Correlations for lag times τ ∈ [2p : 4(p − 1)] are computed at compression level 2, which
is created in an analogous manner from level 1. This can continue hierarchically up to an
arbitrary level for which enough data is available.
Due to the hierarchical reduction of
the data, the algorithm scales as O p2 log(τmax ) . Thus an additional order of magnitude
in τmax costs just a constant extra effort.
The speedup is gained at the expense of statistical accuracy. The loss of accuracy
occurs at the compression step. In principle one can use any value of m and p to tune
the algorithm performance. However, it turns out that using a high m dilutes the data
140
at high τ . Therefore m = 2 is hard-coded in the ESPResSo correlator and cannot be
modified by user. The value of p remains an adjustable parameter which can be modified
by user by setting tau_lin when defining a correlation. In general, one should choose
p m to avoid loss of statistical accuracy. Choosing p = 16 seems to be safe but it may
depend on the properties of the analyzed correlation functions. A detailed analysis has
been performed in Ref. [45].
The choice of the compression function also influences the statistical accuracy and
can even lead to systematic errors. The default compression function is discard2 which
discards the second for the compressed values and pushes the first one to the higher
level. This is robust and can be applied universally to any combination of observables
and correlation operation. On the other hand, it reduces the statistical accuracy as
the compression level increases. In many cases, the average compression operation can
be applied, which averages the two neighbouring values and the average then enters
the higher level, preserving almost the full statistical accuracy of the original data. In
general, if averaging can be safely used or not, depends on the properties of the difference
1
1
1
(Ai ⊗ Bi+p + Ai+1 ⊗ Bi+p+1 ) − (Ai + Ai+1 ) ⊗ (Bi+p + Bi+p+1 )
(9.4)
2
2
2
For example in the case of velocity autocorrelation function, the above-mentioned difference has a small value and a random sign, i.e. different contributions cancel each other.
On the other hand, in the of the case of mean square displacement the difference is always positive, resulting in a non-negligible systematic error. A more general discussion
is presented in Ref. [45].
9.2.7. Checkpointing the correlator
It is possible to checkpoint the correlator. Therby the data is written directly to a file.
It is probably usefull to write to a binary file, as this is more precise. The accuracy of
the ascii files is not high.
Syntax
(1) correlation id write_checkpoint_binary filename
(2) correlation id write_checkpoint_ascii filename
Description
In order to load a checkpoint, the correlator has to be initialized. Therefore the observable(s) have to be created. Make sure that the correlator is exactly initilized as it
was when the checkpoint was created. If this is not fullfilled, and e.g. the size of an
observable has changed, loading the checkpoint failes.
Syntax
(1) correlation id read_checkpoint_binary filename
(2) correlation id read_checkpoint_ascii filename
Description
Depending on whether the checkpoint was written as binary or as text, the corresponding
variant for reading the checkpoint has to be used.
141
An simple example for checkpointing:
set pp [observable new particle_positions all]
set cor1 [correlation new obs1 $pp corr_operation square_distance_componentwise \
dt 0.01 tau_max 1000 tau_lin 16]
integrate 1000
correlation $cor1 write_checkpoint_binary "cor1.bin"
And then to continue the simulation:
set pp [observable new particle_positions all]
set cor1 [correlation new obs1 $pp corr_operation square_distance_componentwise \
dt 0.01 tau_max 1000 tau_lin 16]
correlation $cor1 read_checkpoint_binary "cor1.bin"
142
10. Input / Output
10.1. No generic checkpointing!
One of the most asked-for feature that seems to be missing in ESPResSo is checkpointing,
i.e. a simple way to tell ESPResSo to store and restore the current state of the simulation,
and to be able to write this state to or read it from a file. This would be most useful to
be able to restart a simulation from a specific point in time.
Unfortunately, it is impossible to provide a simple command (e.g. checkpoint), out of
two reasons. The main reason is that ESPResSo has no way to determine what information constitutes the actual state of the simulation. On the one hand, ESPResSo scripts
sometimes use Tcl-variables that contain essential information about a simulation, e.g.
the stored values of an observable that was computed in previous time steps, counters,
etc. These would have to be contained in a checkpoint. However, not all Tcl-variables
are of interest. For example, Tcl has a number of automatically set variables that contain
information about the hostname, the machine type, etc. These variables should most
probably not be included the simulation state. ESPResSo has no way to distinguish between these variables. On the other hand, the ESPResSo core has a number of internal
variables, e.g. the particle coordinates. While most of these are probably good candidates for being included into a checkpoint, this is not necessarily so. For example, when
you have particles in your system that have fixed coordinates, should these be stored in
a checkpoint, or not? If the system contains mostly fixed particles and only very few
moving particles, this would increase the memory size of a checkpoint needlessly. And
what about the interactions in the system, or the bonds? Should these be stored in a
checkpoint, or are they generated by the script?
Another problem with a generic checkpoint would be the control flow of the script.
In principle, the checkpoint would have to store where in the script the checkpointing
function was called to be able to return there. All this is even further complicated by
the fact that ESPResSo is running in parallel.
Instead, in ESPResSo, the user has to specify what information needs to be saved to a
file to be able to restore the simulation state. The blockfile and writemd commands
help you to do that. blockfile writes text files. When floating point numbers are stored
in such files (e.g. the particle positions), there is only a limited precision. Therefore, it
is not possible to bitwise reproduce a simulation state using this function. When you
need bitwise reproducibility, you will have to use the command writemd, which stores
positions, forces and velocities in binary format. Note that there is no command to write
other MD parameters like time step or interactions in binary format. You should restore
these using exactly the same Tcl command that you used to create them.
Finally, there is one more complication: random forces are computed in the order
143
the particles are stored in memory. This order usually differs after reading a blockfile
back, since the particles are stored in consecutive identity order. In memory, they are
usually not in a specific order. Therefore, you need to use sort_particles after writing
a blockfile that you want to use for checkpointing, so that the particles are resorted to
the same consecutive order. Note that this does not change physics, just the order the
random numbers are applied.
When using an LB fluid, you need to also write out the fluid nodes, see the lbfluid
command for further details.
10.2. blockfile: Using the structured file format
ESPResSo uses a standardized ASCII block format to write structured files for analysis or storage. Basically the file consists of blocks in curled braces, which have a
single word title and some data. The data itself may consist again of such blocks.
An example is:
{file {Demonstration of the block format}
{variable epsilon {_dval_ 1} }
{variable p3m_mesh_offset {_dval_ 5.0000000000e-01
5.0000000000e-01 5.0000000000e-01 } }
{variable node_grid {_ival_ 2 2 2 } }
{end}
Whitespace will be ignored within the format (space, tab and return).
The keyword variable should be used to indicate that a variable definition follows in
the form name data. data itself is a block with title _ival_ or _dval_ denoting integer
resp. double values, which then follow in a whitespace separated list.
Such blocks can be read in and written either from ESPResSo-scripts (see in the
following subsections), or from your own C-code using the C-Interface (see section ??).
10.2.1. Writing ESPResSo’s global variables
Syntax
(1) blockfile channel write variable {varname1 varname2 ...}
(2) blockfile channel write variable all
Description
Variant (1) writes the global variables varname1 varname2 . . . (which are known to the
setmd command (see section 6.1 on page 86) to channel . Variant (2) will write all known
global variables.
Note, that when the block is read, all variables with names listed in the Tcl variable
blockfile_variable_blacklist are ignored.
144
10.2.2. Writing Tcl variables
Syntax
(1) blockfile channel write tclvariable { varname1 varname2 ...}
(2) blockfile channel write tclvariable all
(3) blockfile channel write tclvariable reallyall
Description
These commands will write Tcl global variables to channel . Global variables are those
declared in the top scope of the Tcl script, or those that were explicitly declared
global. When reading the block, all variables with names listed in the Tcl variable
blockfile_tclvariable_blacklist are ignored.
Variant (1) writes the Tcl global variables varname1 , varname2 , . . . to channel .
Variant (2) will write all Tcl variables to the file, with the exception of the internally predefined globals from Tcl (tcl_version, argv, argv0, argc, tcl_interactive,
auto_oldpath, errorCode, auto_path, errorInfo, auto_index, env, tcl_pkgPath,
tcl_patchLevel, tcl_libPath, tcl_library and tcl_platform). Variant (3) will even
write those.
10.2.3. Writing particles, bonds and interactions
Syntax
(1) blockfile channel write particles what ( range | all )
(2) blockfile channel write bonds range
(3) blockfile channel write interactions
Description
Variant (1) writes particle information in a standardized format to channel . what can
be any list of parameters that can be specified in part parti d print, except for bonds.
Note that id and pos will automatically be added if missing. range is a Tcl list of
ranges which particles to write. A range is defined as start-stop, where start and stop
are particle identities. stop can also be the string “end”, denoting the highest used
particle identity. Thus " 0-5 10-end" are all particles with the exception of particles
6-9. The keyword all denotes all known particles, i.e. is eqivalent to "0-end").
Variant (2) writes the bond information in a standardized format to channel . The
involved particles and bond types must exist and be valid.
Variant (3) writes the interactions in a standardized format to channel .
10.2.4. Writing the random number generator states
Syntax
(1)
(2)
(3)
(4)
blockfile
blockfile
blockfile
blockfile
channel
channel
channel
channel
write
write
write
write
random
bit_random
seed
bitseed
145
Description
Variants (1) and (2) write the full information on the current states of the respecitive
random number generators (see sections 11.2.1 on page 162 and 11.2.2 on page 163) on
any node to channel . Using this information, it is possible to recover the exact states of
the generators.
Variants (3) and (4) write only the seed(s) which were used to initialize the random
number generators. Note that this information is not sufficient to restore the full state of
a random number generator, because the internal state might contain more information.
10.2.5. Writing all stored configurations
Syntax
blockfile channel write configs
Description
This command writes all configurations currently stored for off-line analysis (see section 8.3 on page 126) to channel .
10.2.6. Writing arbitrary blocks
Syntax
(1) blockfile channel write start tag
(2) blockfile channel write end
(3) blockfile channel write tag [arg]...
Description
channel has to be a Tcl channel. Variant (1) starts a block and gives it the title tag,
variant (2) ends the block. Between two calls to the command, arbitrary data can be
written to the channel. When variant (3) is used, the function blockfile_write_tag is
called with all of the commands arguments. This function should then write the data.
Example
set file [open "data.dat" w]
blockfile $file write start "mydata"
puts $file "{This is my data!}"
blockfile $file write end
will write
{mydata {This is my data!}}
to the file data.dat.
146
10.2.7. Reading blocks
Syntax
(1)
(2)
(3)
(4)
blockfile channel read start
blockfile channel read toend
blockfile channel read auto
blockfile channel read ( particles | interactions | bonds |
variable | seed | random | bitrandom | configs )
Description
Variants (1) and (2) are the low-level block-reading commands. Variant (1) reads the
start part of a block and returns the block title, while variant (2) reads the block data
and returns it.
Variants (3) and (4) read whole blocks. Variant (3) reads in one block and does the
following:
1. if a procedure blockfile_read_auto_tag exists, this procedure takes over (tag
is the first expression in the block). For most block types, at least all mentioned
above, i.e. particles, interactions, bonds, seed, random, bitrandom, configs,
and variable, the corresponding procedure will overwrite the current information
with the information from the block.
2. if the procedure does not exist, it returns
{ usertag rest of block }
3. if the file is at the end, it returns eof
Variant (4) checks for a block with tag block and then again executes the corresponding
blockfile_read_auto_tag, if it exists.
In the contrary that means that for a new blocktype you will normally implement two
procedures:
blockfile_write_tag channel write tag arg...
which writes the block including the header and enclosing braces and
blockfile_read_auto_tag channel read auto
which reads the block data and the closing brace. The parameters write, read , tag and
auto are regular parameters which will always have the specified value. They occur just
for technical reasons.
In a nutshell: The blockfile command is provided for saving and restoring the current
state of ESPResSo, e.g. for creating and using checkpoints. Hence you can transfer all
accessible information from files to ESPResSo and vice versa.
147
set out [open "|gzip
blockfile $out write
blockfile $out write
blockfile $out write
blockfile $out write
blockfile $out write
blockfile $out write
blockfile $out write
close $out
-c - > checkpoint.block.gz" "w"]
variable all
interactions
random
bitrandom
particles "id pos type q v f" all
bonds all
configs
This example writes all global variables, all interactions, the full current state of
the random number generator, all information (i.e. id, position, type-number, charge,
velocity, forces, bonds) of all particles, and all stored particle configurations to the file
checkpoint.block.gz which is compressed on-the-fly. If you want to be able to read in
the information using ESPResSo, note that interactions must be stored before particles
before bonding information, as for the bonds to be set all particles and all interactions
must already be known to ESPResSo.
set in [open "|gzip -cd checkpoint.block.gz" "r"]
while { [blockfile $in read auto] != "eof" } {}
close $in
This is basically all you need to restore the information in the blockfile, overwriting the
current settings in ESPResSo.
10.3. Writing and reading binary files
Binary files are written using the command
Syntax
writemd channel [posx|posy|posz|vx|vy|vz|fx|fy|fz]...
Description
This will write out particle data to the Tcl channel channel for all particles in binary
format. Apart from the mandatory particle id, only limited information can be stored.
The coordinates (posx, posy and posz), velocities (vx, vy and vz) and forces (fx, fy and
fz). Other information should be stored in a blockfile or reconstructed differently. Note
that since both blockfile and writemd are using a Tcl channel, it is actually possible to
mix them, so that you can write a single checkpoint file. However, the blockfile read
auto mechanism cannot handle the binary section, thus you need to read this section
manually. Reading of binary particle data happens through
Syntax
readmd channel
Description
For the exact format of the written binary sequence, see src/tcl/binary_file_tcl.cpp.
148
10.4. Writing VTF files
The formats VTF (VTF Trajectory Format), VSF (VTF Structure Format) and VCF
(VTF Coordinate Format) are formats for the visualization software VMD[30]1 . They
are intended to be human-readable and easy to produce automatically and modify.
The format distinguishes between structure blocks that contain the topological information of the system (i.e. the system size, particle names, types, radii and bonding
information, amongst others), while coordinate blocks (a.k.a. as timestep blocks) contain
the coordinates for the particles at a single timestep. For a visualization with VMD, one
structure block and at least one coordinate block is required.
Files in the VSF format contain a single structure block, files in the VCF format
contain at least one coordinate block, while files in the VTF format contain a single
structure block first and an arbitrary number of coordinate blocks afterwards, thus
allowing to store all information for a whole simulation in a single file. For more details
on the format, refer to the homepage of the format2 .
Creating files in these formats from within ESPResSo is supported by the commands
writevsf and writevcf, that write a structure respectively a coordinate block to the
given Tcl channel. To create a VTF file, first use writevsf at the beginning of the
simulation, and then writevcf after each timestep to generate a trajectory of the whole
simulation.
The structure definitions in the VTF/VSF formats are incremental, i.e. a user can
easily add further structure lines to the VTF/VSF file after a structure block has been
written to specify further particle properties for visualization.
Note that the ids of the particles in ESPResSo and VMD may differ. VMD requires
the particle ids to be enumerated continuously without any holes, while this is not
required in ESPResSo. When using writevsf and writevcf, the ESPResSo particle ids
are automatically translated into VMD particle ids. The function vtfpid allows the
user to get the VMD particle id for a given ESPResSo particle id.
Also note, that these formats can not be used to write trajectories where the number
of particles or their types varies between the timesteps. This is a restriction of VMD
itself, not of the format.
10.4.1. writevsf: Writing the topology
Syntax
writevsf channelId [( short | verbose )] [radius ( radii | auto )]
[typedesc typedesc]
Description
Writes a structure block describing the system’s structure to the channel given by
channelId . channelId must be an identifier for an open channel such as the return
1
2
http://www.ks.uiuc.edu/Research/vmd/
https://github.com/olenz/vtfplugin/wiki/VTF-format
149
value of an invocation of open. The output of this command can be used for a standalone VSF file, or at the beginning of a VTF file that contains a trajectory of a whole
simulation.
Arguments
• [( short | verbose )] Specify, whether the output is in a human-readable, but
somewhat longer format (verbose), or in a more compact form (short). The
default is verbose.
• [radius ( radii | auto )] Specify the VDW radii of the atoms. radii is either
auto, or a Tcl-list describing the radii of the different particle types. When the
keyword auto is used and a Lennard-Jones interaction between two particles of
the given type is defined, the radius is set to be σLJ
2 plus the LJ shift. Otherwise,
the radius 0.5 is substituted. The default is auto.
Example
writevsf $file radius {0 2.0 1 auto 2 1.0}
• [typedesc typedesc] typedesc is a Tcl-list giving additional VTF atom-keywords
to specify additional VMD characteristics of the atoms of the given type. If no
description is given for a certain particle type, it defaults to name name type
type , where name is an atom name and type is the type id.
Example
writevsf $file typedesc {0 "name colloid" 1 "name pe"}
10.4.2. writevcf: Writing the coordinates
Syntax
writevcf channelId [( short | verbose )] [( folded | absolute )]
[pids ( pids | all )] [userdata userdata]
Description
Writes a coordinate (or timestep) block that contains all coordinates of the system’s
particles to the channel given by channelId . channelId must be an identifier for an open
channel such as the return value of an invocation of open.
Arguments
• [( short | verbose )] Specify, whether the output is in a human-readable, but
somewhat longer format (verbose), or in a more compact form (short). The
default is verbose.
• [( folded | absolute )] Specify whether the particle positions are written in
absolute coordinates (absolute) or folded into the central image of a periodic
system (folded). The default is absolute.
• [pids ( pids | all )] Specify the coordinates of which particles should be written. If all is used, all coordinates will be written (in the ordered timestep
150
format). Otherwise, pids has to be a Tcl-list specifying the pids of the particles.
The default is all.
Example
pids {0 23 42}
• [userdata userdata] Specify arbitrary user data for the particles. userdata has
to be a Tcl list containing the user data for every particle. The user data is
appended to the coordinate line and can be read into VMD via the VMD plugin
VTFTools. The default is to provide no userdata.
Example
userdata {"red" "blue" "green"}
10.4.3. vtfpid: Translating ESPResSo particles ids to VMD particle ids
Syntax
vtfpid pid
Description
If pid is the id of a particle as used in ESPResSo, this command returns the atom id
used in the VTF, VSF or VCF formats.
10.5. writevtk: Particle Visualization in paraview
This feature allows to export the particle positions in a paraview 3 compatible VTK file.
Paraview is a powerful and easy to use open-source visualization program for scientific
data. Since ESPResSo can export the lattice-Boltzmann velocity field 12.8 in the VTK
format as well and paraview allows to visualize particles with glyphs and vector fields
with stream lines, glyphs, contourplots, etc., one can use it so completely visualize a
coupled lattice-Boltzmann MD simulation. It can also create videos without much effort
if one exports data of individual time steps into separate files with filenames including
a running index (data_0.vtk, data_1.vtk, ...).
Syntax
writevtk filename [( all | type )]
Description
Arguments
• filename Name of the file to export the particle positions into.
• [( all | type )] Specifies which particle type should be exportet. The default is
all. Alternatively, a type number can be given. Exporting the positions of all
3
http://www.paraview.org/
151
particles but in separate files might make sense if one wants to distinguish the
different particle types in the visualization (i.e. by color or size).
10.6. Reading and Writing PDB/PSF files
The PDB (Brookhaven Protein DataBase) format is a widely used format for describing
atomistic configurations. PSF is a format that is used to describe the topology of a PDB
file.
When visualizing your system with VMD, it is recommended to use the VTF format
instead (see section 10.4), as it was specifically designed for visualizations with VMD. In
contrast to the PDB/PSF formats, in VTF files it is possible to specify the VDW radii
of the particles, to have a varying simulation box size, etc.
10.6.1. writepsf: Writing the topology
Syntax
writepsf file [-molecule] NP MPC NC I Np S Nn S
Description
Writes the current topology to the file file (here, file is not a channel, since additional
information cannot be written anyway). NP , MPC and so on are parameters describing
a system consisting of equally long charged polymers, counterions and salt. This information is used to set the residue name and can be used to color the atoms in VMD.
If you specify -molecule, the residue name is taken from the molecule identity of the
particle. Of course different kinds of topologies can also be handled by modified versions
of writepsf.
10.6.2. writepdb: Writing the coordinates
Syntax
(1) writepdb file
(2) writepdbfoldchains file chain start n chains chain length box l
(3) writepdbfoldtopo file shift
Description
Variant (1) writes the corresponding particle data.
Variant (2) writes folded particle data where the folding is performed on chain centers
of mass rather than single particles. In order to fold in this way the chain topology
and box length must be specified. Note that this method is outdated. Use variant (3)
instead.
Variant (3) writes folded particle data where the folding is performed on chain centers
of mass rather than single particles. This method uses the internal box length and
topology information from espresso. If you wish to shift particles prior to folding then
supply the optional shift information. shift should be a three member tcl list consisting
152
of x, y, and z shifts respectively and each number should be a floating point (ie with
decimal point).
10.6.3. readpdb: Reading the coordinates and interactions
Syntax
readpdb pdb_file pdbfile type type first_id firstid
[ itp_file itpfile first_type fisttype]
[lj_with othertype epsilon sigma 1] [lj_rel_cutoff cutoff 1 ]
[fit_to_box]
Required features:
1 LENNARD_JONES
Description
Reads the positions and possibly charges, types and Lennard-Jones interactions from
the file pdbfile and a corresponding Gromacs topology file itpfile. The topology file must
contain the atoms and atomtypes sections, it may be necessary to use the Gromacs
preprocessor to obtain a complete file from a system configuration and a force field.
Any offset of the particle positions if removed, such that the lower left corner bounding
box of the particles is in the origin. If fit_to_box is given, the box size if increased to
hold the particles if necessary. If it is not set and the particles do not fit into the box,
the behavior is undefined.
type sets the particle type for the added particles. If there is a topology file give that
contains a types for the particles, the particles get types by the order in the topology
file plus firstype. If the corresponding type in the topology file has a charge, it is used,
otherwise the particle charge defaults to zero.
The particles get consecutive id’s in the order of the pdb file, starting at firstid . Please
be aware that existing particles get overwritten by values from the file.
The lj_with section produces Lennard-Jones interactions between the type othertype
and the types defined by the topology file. The interaction parameters are calculated
√
as othertype,j = othertype j and σothertype,j = 21 (σothertype + σj ), where j runs over the
atomtypes defined in the topology file. This corresponds to the combination rule 2 of
Gromacs. There may be multiple such sections. The cutoff is determined by cutoff as
cutoff × σij in a relative fashion. The potential is shifted so that it vanishes at the cutoff.
The command returns the number of particles that were successfully added.
Reading bonded interactions and dihedrals is currently not supported.
10.7. Online-visualisation with VMD
IMD (Interactive Molecular Dynamics) is the protocol that VMD uses to communicate
with a simulation. Tcl md implements this protocol to allow online visual analysis of
running simulations.
In IMD, the simulation acts as a data server. That means that a simulation can
provide the possibility of connecting VMD, but VMD need not be connected all the
time. You can watch the simulation just from time to time.
153
In the following the setup and usage of IMD is described.
10.7.1. imd: Using IMD in the script
Syntax
(1)
(2)
(3)
(4)
imd
imd
imd
imd
connect [port]
positions [( -unfolded |-fold_chains )]
listen seconds
disconnect
Description
In your simulation, the IMD connection is setup up using variant (1), where port is an
arbitrary port number (which usually has to be between 1024 and 65000). By default,
ESPResSo will try to open port 10000, but the port may be in use already by another
ESPResSo simulation. In that case it is a good idea to just try another port.
While the simulation is running, variant (2) can be used to transfer the current coordinates to VMD, if it is connected. If not, nothing happens and the command just
consumes a small amount of CPU time. Note, that before you can transfer coordinates
to VMD, VMD needs to be aware of the structure of the system. For that, you first
need to load a corresponding structure file (PSF or VSF) into VMD. Also note, that the
command prepare_vmd_connection (see section 10.7.3 on the next page) can be used
to automatically set up the VMD connection and transfer the structure file.
By specifying -unfolded, the unfolded coordinates of the particles will transferred,
while -fold_chains will fold chains according to their centers of mass and retains bonding connectivity. Note that this requires the chain structure to be specified first using
the analyze command.
Variant (3) can be used to let the simulation wait for seconds seconds or until IMD has
connected, before the script is continued. This is normally only useful in demo scripts,
if you want to see all frames of the simulation.
Variant (4) will terminate the IMD session. This is normally not only nice but also
the operating system will not free the port for some time, so that without disconnecting
for some 10 seconds you will not be able to reuse the port.
10.7.2. Using IMD in VMD
The PDB/PSF files created by ESPResSo via the command writepsf and writepdb can
be loaded into VMD. This should bring up an initial configuration.
Then you can use the VMD console to execute the command
imd connect host port
where host is the host running the simulation and port is the port it listens to. Note that
VMD crashes, if you do that without loading a structure file before. For more information
on how to use VMD to extract more information or hide parts of configuration, see the
VMD Quick Help.
154
10.7.3. Automatically setting up a VMD connection
Syntax
(1) prepare_vmd_connection filename [start] [wait wait] [localhost]
[constraints] ...
(2) prepare_vmd_connection [filename [wait [start [constraints]]]]
Description
To reduce the effort involved in setting up the IMD connection, starting VMD and loading the structure file, ESPResSo provides the command prepare_vmd_connection. It
writes out the required vsf structure description file to filename.vsf (default for filename
is vmd), doing some nice stuff such as coloring the molecules, bonds and counterions
appropriately, rotating your viewpoint, and connecting your system to the visualization
server.
If the option [constraints] is given, then the command will create graphics primitives
in VMD that represent some of the spatial constraints (sphere, rhomboid and cylinder
at present).
If [start] is given, the command will automatically try to start VMD and connect it to
the ESPResSo simulation. Otherwise it only writes the VMD setup script filename.vmd_start.script.
You can use this script later to connect to the ESPResSo simulation by running either
vmd -e vmd_start.script
or by running
source "vmd_start.script"
at VMD’s Tcl console. If you choose to not start VMD automatically, prepare_vmd_connection
puts the hostname into the VMD script, so that you can start it from any computer.
However, some more recent Linux distributions block any incoming transfer even from
the computer itself, if it does not come from localhost. If you encounter problems to
connect to VMD on the very same computer, try the [localhost] option, which will enforce to use the hostname localhost. Note that the [start] option implies the [localhost]
option, since VMD is necessarily started from the same computer.
If the option [wait] is provided, then the command waits for at most wait seconds for
VMD to connect. Since VMD usually takes a while to start, it is usually a good idea to
combine the [start] option with a waiting time of 100, so a bit less than a minute.
All remaining parameters are passed to the writevsf that is used to setup the system,
so that you can specify the sizes of particles etc.
prepare_vmd_connection also supports an older, deprecated syntax (variant 2) with
limited functionality. This syntax uses fixed position parameters and boolean values for
[start] and [constraints], as described above.
10.8. Error handling
Errors in the parameters are detected as early as possible, and hopefully selfexplanatory error messages returned without any changes to the data in the internal
155
data of ESPResSo. This include errors such as setting nonexistent properties of particles or simply misspelled commands. These errors are returned as standard Tcl
errors and can be caught on the Tcl level via
catch {script} err
I do not understand this. How
does the error
look?
When run noninteractively, Tcl will return a nice stack backtrace which allows to quickly
find the line causing the error.
However, some errors can only be detected after changing the internal structures, so
that ESPResSo is left in a state such that integration is not possible without massive
fixes by the users. Especially errors occuring on nodes other than the primary node fall
under this condition, for example a broken bond or illegal parameter combinations.
For error conditions such as the examples given above, a Tcl error message of the form
tcl error background 0 error a error b 1 error c
is returned. Following possibly a normal Tcl error message, after the background keyword all severe errors are listed node by node, preceeded by the node number. A special
error is <consent>, which means that one of the slave nodes found exactly the same errors as the master node. This happens mainly during the initialization of the integrate,
e.g. if the time step is not set. In this case the error message will be
background_errors 0 {time_step not set} 1 <consent>
In each case, the current action was not fulfilled, and possibly other parts of the internal
data also had to be changed to allow ESPResSo to continue, so you should really know
what you do if you try and catch these errors.
156
11. Auxilliary commands
Missing
commands:
Probably all from
scripts/auxiliary.tc
11.1. Finding particles and bonds
11.1.1. countBonds
Syntax
countBonds particlel ist
Description
Returns a Tcl-list of the complete topology described by particle list, which must have
the same format as the output of the command part (see section 4.1 on page 29).
The output list contains only the particle id and the corresponding bonding information, thus it looks like e.g.
{106 {0 107}} {107 {0 106} {0 108}} {108 {0 107} {0 109}} ...
{210 {0 209} {0 211}} {211 {0 210}} 212 213 ...
for a single chain of 106 monomers between particle 106 and 211, with additional loose
particles 212, 213, ... (e.g. counter-ions). Note, that the part command stores any
bonds only with the particle of lower particle number, which is why [part 109] would
only return ... bonds 0 110, therefore not revealing the bond between particle 109
and (the preceding) particle 108, while countBonds would return all bonds particle 109
participates in.
11.1.2. findPropPos
Syntax
findPropPos particlep ropertyl ist property
Description
Returns the index of property within particlep ropertyl ist, which is expected to have the
same format as [part particlei d ]. If property is not found, -1 is returned.
This function is useful to access certain properties of particles without hard-wiring
their index-position, which might change in future releases of part.
Example
[lindex [part $i] [findPropPos [part $i] type]]
157
This returns the particle type id of particle i without fixing where exactly that information has to be in the output of [part $i].
11.1.3. findBondPos
Syntax
findBondPos particlep ropertyl ist
Description
Returns the index of the bonds within particlep ropertyl ist, which is expected to have the
same format as [part particle number ]; hence its output is the same as [findPropPos
particlep ropertyl ist bonds]. If the particle does not have any bonds, -1 is returned.
11.1.4. timeStamp
Syntax
timeStamp path prefix postfix suffix
Description
Modifies the filename contained within path to be preceded by a prefix and having
postfix before the suffix ; e.g.
timeStamp ./scripts/config.gz DH863 001 gz
returns ./scripts/DH863_config001.gz. If postfix is −1, the current date is used in the
format %y%m%d. This would results in ./scripts/DH863_config021022.gz on October
22nd, 2002.
11.2. Additional Tcl math-functions
The following procedures are found in scripts/ABHmath.tcl.
• CONSTANTS
– PI
returns π with 16 digits precision.
– KBOLTZ
Returns Boltzmann constant in Joule/Kelvin
– ECHARGE
Returns elementary charge in Coulomb
– NAVOGADRO
Returns Avogadro number
– SPEEDOFLIGHT
158
Returns speed of light in meter/second
– EPSILON0
Returns dielectric constant of vaccum in Coulombˆ2/(Joule meter)
– ATOMICMASS
Returns the atomic mass unit u in kilogramms
• MATHEMATICAL FUNCTIONS
– sqr <arg>
returns the square of arg.
– min <arg1> <arg2>
returns the minimum of arg1 and arg2 .
– max <arg1> <arg2>
returns the maximum of arg1 and arg2 .
– sign <arg>
returns the signum-function of arg, namely +1 for arg > 0, -1 for < 0, and
=0 otherwise.
• RANDOM FUNCTIONS
– gauss_random
returns random numbers which have a Gaussian distribution
– dist_random <dist> [max]
returns random numbers in the interval [0, 1] which have a distribution according to the distribution function p(x) dist which has to be given as a tcl
list containing equally spaced values of p(x). If p(x) contains values larger
than 1 (default value of max) the maximum or any number larger than that
has to be given max . This routine basically takes the function p(x) and places
it into a rectangular area ([0,1],[0,max]). Then it uses to random numbers
to specify a point in this area and checks wether it resides in the area under
p(x). Attention: Since this is written in tcl it is probably not the fastest way
to do this!
– vec_random [len]
returns a random vector of length len (uniform distribution on a sphere) This
is done by chosing 3 uniformly distributed random numbers [−1, 1] If the
length of the resulting vector is <= 1.0 the vector is taken and normalized to
the desired length, otherwise the procedure is repeated until succes. On average the procedure needs 5.739 random numbers per vector. (This is probably
not the most efficient way, but it works!) Ask your favorit mathematician for
a proof!
159
– phivec_random <v> <phi> [len]
return a random vector at angle phi with v and length len
• PARTICLE OPERATIONS
Operations involving particle positions. The parameters pi can either denote the
particle identity (then the particle position is extracted with the The part command
command) or the particle position directly When the optional box parameter for
minimum image conventions is omited the functions use the the setmd box_l
command.
– bond_vec <p1> <p2>
Calculate bond vector pointing from particles p2 to p1 return = (p1 .pos p2 .pos)
– bond_vec_min <p1> <p2> [box]
Calculate bond vector pointing from particles p2 to p1 return = MinimumImage(p1 .pos
- p2 .pos)
– bond_length <p1> <p2>
Calculate bond length between particles p1 and p2
– bond_length_min <p1> <p2> [box]
Calculate minimum image bond length between particles p1 and p2
– bond_angle <p1> <p2> <p3> [type]
Calculate bond angle between particles p1 , p2 and p3 . If type is ”r” the
return value is in radiant. If it is ”d” the return value is in degree. The
default for type is ”r”.
– bond_dihedral <p1> <p2> <p3> <p4> [type]
Calculate bond dihedral between particles p1 , p2 , p3 and p4 If type is ”r”
the return value is in radiant. If it is ”d” the return value is in degree The
default for type is ”r”.
– part_at_dist <p> <dist>
return position of a new particle at distance dist from p with random orientation
– part_at_angle <p1> <p2> <phi> [len]
return position of a new particle at distance len (default=1.0) from p2 which
builds a bond angle phi for (p1 , p2 , p-new)
– part_at_dihedral <p1> <p2> <p3> <theta> [phi] [len]
return position of a new particle at distance len (default=1.0) from p3 which
builds a bond angle phi (default=random) for (p2 , p3 , p-new) and a dihedral
angle theta for (p1 , p2 , p3 , p-new)
160
• INTERACTION RELATED
Help functions related to interactions implemented in ESPResSo.
–
calc_lj_shift <lj_sigma> <lj_cutoff>
returns the value needed to shift the Lennard Jones potential to zero at the
cutoff.
• VECTOR OPERATIONS
A vector v is a tcl list of numbers with an arbitrary length Some functions are
provided only for three dimensional vectors. corresponding functions contain 3d
at the end of the name.
– veclen <v>
return the length of a vector
– veclensqr <v>
return the length of a vector squared
– vecadd <a> <b>
add vector a to vector b: return = (a+b)
– vecsub <a> <b>
subtract vector b from vector a: return = (a-b)
– vecscale <s> <v>
scale vector v with factor s: return = (s*v )
– vecdot_product <a> <b>
calculate dot product of vectors a and b: return = (a.b)
– veccross_product3d <a> <b>
calculate the cross product of vectors a and b: return = (a x b)
– vecnorm <v> [len]
normalize a vector to length len (default 1.0)
– unitvec <p1> <p2>
return unit vector pointing from position p1 to position p2
– orthovec3d <v> [len]
return orthogonal vector to v with length len (default 1.0) This vector does
not have a random orientation in the plane perpendicular to v
– create_dihedral_vec <v1> <v2> <theta> [phi] [len]
create last vector of a dihedral (v1 , v2 , res) with dihedral angle theta and
bond angle (v2 , res) phi and length len (default 1.0). If phi is ommited or
set to rnd then phi is assigned a random value between 0 and 2 Pi.
161
• TCL LIST OPERATIONS
– average <list>
Returns the avarage of the provided list
– list_add_value <list> <val>
Add val to each element of list
– flatten <list>
flattens a nested list
– list_contains <list> <val>
Checks wether list contains val . returns the number of occurences of val in
list.
• REGRESSION
– LinRegression <l>
where l is a listof pairs of points { {$x1 $y1} {$x2 $y2} ...} . LinRegression
returns the least-square linear fit ax + b and the standard errors σa and σb .
– LinRegressionWithSigma <l>
where l is a list of lists of points in the form {{$x1 $y1 $s1} {$x2 $y2 $s2} ...}
where s is the standard deviation of y. LinRegressionWithSigma returns the
least-square linear fit ax+b, the standard errors σa and σb , covariance cov(a, b)
and χ.
11.2.1. t_random
• Without further arguments,
t_random
returns a random double between 0 and 1 using the ’ran1’ random number generator from Numerical Recipes.
• t_random int <n>
returns a random integer between 0 and n-1.
• t_random seed
returns a tcl-list with the seeds of the random number generators on each of the
’n nodes’ nodes, while
t_random seed <seed(0)> ... <seed(n_nodes-1)>
sets those seeds to the new values respectively, re-initialising the random number generators on each node. Note that this is automatically done on invoking
Espresso, however due to that your simulation will always start with the same random sequence on any node unless you use this tcl-command to reset the sequences’
seeds.
162
• Since internally the random number generators’ random sequences are not based
on mere seeds but rather on whole random number tables, to recover the exact
state of the random number generators at a given time during the simulation run
(e. g. for saving a checkpoint) requires knowledge of all these values. They can be
accessed by
t_random stat
which returns a tcl-list with all status informations for any node (e. g. 8 nodes =>
approx. 350 parameters). To overwrite those internally in Espresso (e. g. upon
restoring a checkpoint) submit the whole list back using
t_random stat <status-list>
with status − list being the tcl-list mentioned above without any braces. Be careful!
A complete recovery of the current state of the simulation is only possible if you
make sure to include a call to the sort_particles command after you saved the
blockfile to make sure random numbers are applied in the same order.
The C implementation is t random
11.2.2. The bit_random command
• Without further arguments,
bit_random
returns a random double between 0 and 1 using the R250 generator XOR-ing a
table of 250 linear independent integers.
• bit_random seed
returns a tcl-list with the seeds of the random number generators on each of the
’n nodes’ nodes, while
bit_random seed <seed(0)> ... <seed(n_nodes-1)>
sets those seeds to the new values respectively, re-initialising the random number generators on each node. Note that this is automatically done on invoking
Espresso, however due to that your simulation will always start with the same random sequence on any node unless you use this tcl-command to reset the sequences’
seeds.
• Since internally the random number generators’ random sequences are not based
on mere seeds but an array of 250 linear independent integers whose bits are used
as matrix elements which are XOR-ed, to recover the exact state of the random
number generators at a given time during the simulation run (e. g. for saving a
checkpoint) requires knowledge of all these values. They can be accessed by
bit_random stat
163
which returns a tcl-list with all status informations for any node (e. g. 8 nodes
=> approx. 2016 parameters). To overwrite those internally in Espresso (e. g.
upon restoring a checkpoint) submit the whole list back using
bit_random stat <status-list>
with ¡status-list¿ being the tcl-list mentioned above without any braces. Be careful!
See 11.2.1 for more information on how to recover of the current state, include the
sequence the random numbers are applied.
• Note further that the bit-wise display of integers, as it is used by this random
number generator, is platform dependent. As long as you stay on the same architecture this doesn’t matter at all; however, it wouldn’t be wise to use a checkpoint
including the state of the R250 to restart the simulation on a different platform
- most likely, the integers will have a different bit-muster leading to a completely
different random matrix. So, if you’re using this random number generator, always
remain on the same platform!
11.3. Checking for features of ESPResSo
In an ESPResSo-Tcl-script, you can get information whether or not one or some of the
features are compiled into the current program with help of the following Tcl-commands:
• code_info
provides information on the version, compilation status and the debug status of the
used code. It is highly recommended to store this information with your simulation
data in order to maintain the reproducibility of your results. Exemplaric output:
ESPRESSO: v1.5.Beta (Neelix), Last Change: 23.01.2004
{ Compilation status { PARTIAL_PERIODIC } { ELECTROSTATICS }
{ EXTERNAL_FORCES } { CONSTRAINTS } { TABULATED }
{ LENNARD_JONES } { BOND_ANGLE_COSINE } }
{ Debug status { MPI_CORE FORCE_CORE } }
• has_feature <feature> ...
tests, if feature is compiled into the ESPResSo kernel. A list of possible features
and their names can be found here.
•
require_feature <feature> ...
tests, if feature is feature is compiled into the ESPResSo kernel, will exit the script
if it isn’t and return the error code 42. A list of possible features and their names
can be found here.
164
12. Lattice-Boltzmann
For an implicit treatment of a solvent, ESPResSo allows to couple the molecular dynamics simulation to a Lattice-Boltzmann fluid. The Lattice- Boltzmann-Method (LBM) is
a fast, lattice based method that, in its “pure” form, allows to calculate fluid flow in
different boundary conditions of arbitrarily complex geometries. Coupled to molecular
dynamics, it allows for the computationally efficient inclusion of hydrodynamic interactions into the simulation. The implementation of boundary conditions for the LBM is
a difficult task, a lot of research is still being conducted on this topic. The focus of the
ESPResSo implementation of the LBM is, of course, the coupling to MD and therefore
available geometries and boundary conditions are somewhat limited in comparison to
“pure” codes.
Here we restrict the documentation to the interface. For a more detailed description
of the method, please refer to the literature.
12.1. Setting up a LB fluid
Please cite [7] (BibTeX-key espresso2 in file doc/ug/citations.bib) if you use
the LB fluid and [46] (BibTeX-key lbgpu in file doc/ug/citations.bib) if you use
the GPU implementation.
Syntax
lbfluid [gpu] 2 [agrid agrid ] 1 or 2 [dens density ] 1 or 2 or 3
[visc viscosity] 1 or 2 or 3 [tau lb timestep] 1 or 2
[bulk_visc bulk viscosity] 1 or 2 or 3 [ext_force fx fy fz ] 1 or 2 or 3
[friction gamma ] 1 or 2 or 3 [couple 2pt/3pt ] 2
[gamma_odd gamma odd ] 1 or 2 or 3 [gamma_even gamma even] 1 or 2 or 3
[mobility] mobilities 3 [sc_coupling] coupling constants 3
Required features:
1 LB
2 LB_GPU
3 SHANCHEN
Description
The lbfluid command initializes the fluid with a given set of parameters. It is also
possible to change parameters on the fly, but this will only rarely be done in practice.
Before being able to use the LBM, it is necessary to set up a box of a desired size. The
parameter agrid is used to set the lattice constant of the fluid, so the size of the box in
every direction must be a multiple of agrid .
In ESPResSo the LB scheme and the MD scheme are not synchronized: In one LB time
step typically several MD steps are performed. This allows to speed up the simulations
165
and is adjusted with the parameter tau, the LB timestep. The parameters dens and
visc set up the density and (kinematic) viscosity of the LB fluid in (usual) MD units.
Internally the LB implementation works with a different set of units: all lengths are
expressed in agrid , all times in tau and so on. Therefore changing agrid and tau, might
change the behaviour of the LB fluid, e.g. at boundaries, due to characteristics of the
LBM itself. It should also be noted that the LB nodes are located at 0.5, 1.5, 2.5, etc
(in terms of agrid ). This has important implications for the location of hydrodynamic
boundaries which are generally considered to be halfway between two nodes to first order.
Currently it is not possible to precisely give a parameter set where reliable results are
expected, but we are currently performing a study on that. Therefore the LBM should
not be used as a black box, but only after a careful check of all parameters that were
applied.
The parameter ext_force allows to apply an external body force density that is
homogeneous over the fluid. It is again to be given in MD units. The parameter
bulk_viscosity allows to tune the bulk viscosity of the fluid and is given in MD units.
In the limit of low Mach (often also low Reynolds) number the results should be independent of the bulk viscosity up to a scaling factor. It is however known that the values
of the viscosity does affect the quality of the implemented link-bounce-back method.
gamma_odd and gamma_even are the relaxation parameters for the kinetic modes. Due
to their somewhat obscure nature they are to be given directly in LB units.
Before running a simulation at least the following parameters must be set up: agrid,
dens, visc, tau, friction. For the other parameters, the following are taken: bulk viscosity=0,
gamma odd =0, gamma even=0, fx = fy = fz = 0.
If the feature SHANCHEN is activated, the Lattice Boltzmann code (so far GPU version
only) is extended to a two-component Shan-Chen (SC) method. The lbfluid command
requires in this case to supply two values, for the respective fluid components, to each of
the options dens, visc, bulk_visc, friction, gamma_odd and gamma_even, when they
are used, otherwise they are set to the default values. The three elements of the coupling
matrix can be supplied with the option sc_coupling, and the mobility coefficient can
be specified with the option mobility. By default no copuling is activated, and the
relaxation parameter associated to the mobility is zero, corresponding to an infinite
value for mobility. Additional details are given in 12.3 and 12.4.
Syntax
lbfluid print_interpolated_velocity x y z
Description
This variant returns the velocity at point in countinous space. This can make it easier
to calculate flow profiles independent of the lattice constant.
Syntax
lbfluid save_ascii_checkpoint filename lbfluid save_binary_checkpoint
filename lbfluid load_ascii_checkpoint filename lbfluid
load_binary_checkpoint filename
166
Description
The first two save commands save all of the LB fluid nodes’ populations to filename
in ascii or binary format respectively. The two load commands load the populations
from filename. This is useful for restarting a simulation either on the same machine or
a different machine. Some care should be taken when using the binary format as the
format of doubles can depend on both the computer being used as well as the compiler.
One thing that one needs to be aware of is that loading the checkpoint also requires
the used to reuse the old forces. This is necessary since the coupling force between the
paricles and the fluid has already been applied to the fluid. Failing to reuse the old
forces breaks momentum conservation, which is in general a problem. It is particularly
problematic for bulk simulations as the system as a whole acquires a drift of the center
of mass, causing errors in the calculation of velocities and diffusion coefficients. The
correct way to restart an LB simulation is to first load in the particles with the correct
forces, and use “integrate steps reuse forces” upon the first call to integrate. This causes
the old forces to be reused and thus conserves momentum.
12.2. LB as a thermostat
Syntax
thermostat lb 1 or 2 or 3 T
Required features:
1 LB
2 LB_GPU
3 SHANCHEN
Description
The LBM implementation in ESPResSo uses Ahlrichs and D¨
unweg’s point coupling
method to couple MD particles the LB fluid. This coupling consists in a frictional
force and a random force:
F~ = −γ (~v − ~u) + F~R .
The momentum acquired by the particles is then transferred back to the fluid using a
linear interpolation scheme, to preserve total momentum. In the GPU implementation
the force can alternatively be interpolated using a three point scheme which couples
the particles to the nearest 27 LB nodes. This can be called using “lbfluid couple
3pt” and is described in D¨
unweg and Ladd by equation 301[20]. Note that the three
point coupling scheme is incompatible with the Shan Chen Lattice Boltmann. The
frictional force tends to decrease the relative velocity between the fluid and the particle
whereas the random forces are chosen so large that the average kinetic energy per particle
corresponds to the given temperature, according to a fluctuation dissipation theorem.
No other thermostatting mechanism is necessary then. Please switch off any other
thermostat before starting the LB thermostatting mechanism.
The LBM implementation provides a fully thermalized LB fluid, i.e. all nonconserved
modes, including the pressure tensor, fluctuate correctly according to the given temperature and the relaxation parameters. All fluctuations can be switched off by setting the
temperature to 0.
Regarind the unit of the temperature, please refer to Section 1.4.
167
12.3. The Shan Chen bicomponent fluid
Please cite [50] (BibTeX-key sega13c in file doc/ug/citations.bib) if you use the
Shan Chen implementation described below.
The Lattice Boltzmann variant of Shan and Chan[51] is widely used as it is simple
and yet very effective in reproducing the most important traits of multicomponent or
multiphase fluids. The version of the Shan-Chen method implemented in ESPResSo is an
extension to bi-component fluids of the multi-relaxation-times Lattice Boltzmann with
fluctuations applied to all modes, that is already present in ESPResSo. It features, in
addition, coupling with particles[50] and component-dependent particle interactions (see
sections 12.4 and12.5).
The Shan-Chen fluid is set up using the lbfluid command, supplying two values
(one per component) to the dens option. Optionally, two values can be set for each
of the usual transport coefficients (shear and bulk viscosity), and for the ghost modes.
It is possible to set a relaxation time also for the momentum modes, since they are
not conserved quantities in the Shan-Chen method, by using the option mobility. The
mobility transport coefficient expresses the propensity of the two components to mutually
diffuse, and, differently from other transport coefficients, only one value is needed, as
it carachterizes the mixture as a whole. When thermal fluctuations are switched on, a
random noise is added, in addition, also to the momentum modes. Differently from the
other modes, a correlated noise is added to the momentum ones, in order to preserve
the total momentum.
The fluctuating hydrodynamic equations that are simulated using the Shan-Chen approach are
X
∂
ˆ) +
~
~ +∇
~ · (Π
~ + ~σ
ρ
~u + (~u · ∇)~u = −∇p
~gζ ,
(12.1)
∂t
ζ
∂
~ · (ρζ ~u) = ∇
~ · (D
~ ζ + ξ~ˆζ ),
ρζ + ∇
∂t
(12.2)
~ · (ρ~u) = 0,
∂t ρ + ∇
(12.3)
wherePthe index ζ = 1, 2 specifies the component,
~uPis the fluid (baricentric) velocity,
P
ρ = ζ ρζ is the total density, and p = ζ pζ = ζ c2s ρζ is the internal pressure of
ˆ and ξ~ˆζ are associated,
the mixture (cs being the sound speed). Two fluctuating terms ~σ
~ ζ and to the viscous stress tensor Π.
~
respectivelu, to the diffusive current D
The coupling between the fluid components is realized by the force
XX
~gζ (~r) = −ρζ (~r)
gζζ 0 ρζ 0 (~r0 )(~r0 − ~r),
~
r0
(12.4)
ζ0
that acts on the component ζ at node position ~r, and depends on the densities on
the neighboring nodes located at ~r0 . The width of the interfacial regions between two
168
components, that can be obtained with the Shan-Chen method is usually 5-10 lattice
units. The coupling matrix gζζ 0 is in general symmetric, so in the present implementation
only three real values need to be specified with the option sc_coupling. The lbfluid
command sets the density of the two components to the values specified by the option
dens, and these can be modified with the lbnode command. Note that the number of
active fluid components can be accessed through the global variable lb_components.
12.4. SC as a thermostat
The coupling of particle dynamics to the Shan-Chen fluid has been conceived as an
extension of the Ahlrichs and D¨
unweg’s point coupling, with the force acting on a particle
given by
P
r)
ζ γζ ρζ (~
~
F =− P
(~v − ~u) + F~R + F~ ps ,
(12.5)
ρ
(~
r
)
ζ
ζ
ζ
where ζ identifies the component, ρζ (~r) is a linear interpolation of the component density
on the nodes surrounding the particle, γζ is the component-dependent friction coefficient,
F~R is the usual random force, and
X
F~ ps = −
κζ ∇ρζ (~r).
(12.6)
ζ
This is an effective solvation force, that can drive the particle towards density maxima
or minima of each component, depending on the sign of the constant κζ . Note that by
setting the coupling constant to the same negative value for both components will, in
absence of other forces, push the particle to the interfacial region.
In addition to the solvation force acting on particles, another one that acts on the
fluid components is present, representing the solvation force of particles on the fluid.
X X (~ri − ~r) (~r0 − ~r) ~r0 − ~r
F~ζfs (~r) = −λζ ρζ (~r)
Θ
·
,
(12.7)
|~ri − ~r| |~r0 − ~r| |~r0 − ~r|2
0
i
~
r
where Θ(x) = 1 if 0 < x < 1, and 0 otherwise, the sum over lattice nodes is performed on
the neighboring sites of ~r and the index i runs over all particles. Note that a dependence
on the particle index i is assumed for κζ and λζ . This force has the effect of raising or
lowering (depending on the sign of the coupling constant λζ ) the density in the eight
nodes around a particle. The particle property solvation (Chap. 4) sets the coupling
constants λA ,κA ,λB and κB , where A and B denote the first and second fluid component,
respectively. A complete description of the copuling scheme can be found in [50].
12.5. SC component-dependent interactions between particles
Often particle properties depend on the type of solvent in which they are. For example,
a polymer chain swells in a good solvent, and collapses in a bad one. One of the possible
169
ways to model the good or bad solvent condition in coarse-grained models is to employ
a WCA or a LJ (attractive) potential, respectively. If one wants to model the two
components of the SC fluid as good/bad solvent, it is possible to do it using the affinity
argument of the inter command. This non-bonded interaction type acts as a modifier
to other interactions. So far only the Lennard-Jones interaction is changed by the
affinity, so that it switches in a continuous way (after the potential minimum) from
the full interaction to the WCA one. For more information see 5.1.2 and 5.2.3.
12.6. Reading and setting single lattice nodes
Syntax
lbnode x y z ( print | set ) args
Required features:
1 LB
2 LB_GPU
3 SHANCHEN
Description
The lbnode command allows to inspect (print) and modify (set) single LB nodes. Note
that the indexing in every direction starts with 0. For both commands you have to specify
what quantity should be printed or modified. Print allows the following arguments:
rho
the density (one scalar1,2 or two scalars3 ).
u
the fluid velocity (three floats: ux , uy , uz )
pi
the fluid velocity (six floats: Πxx , Πxy , Πyy ,
Πxz , Πyz , Πzz )
pi_neq
the nonequilbrium part of the pressure tensor, components as above.
pop
the 19 population (check the order from the
source code please).
boundary
the flag indicating whether the node is a
fluid node (boundary = 0) or a boundary
node (boundary 6= 0). Does not support
set. Refer to the lbboundary command for
this functionality.
1 LB or LB_GPU;
2 SHANCHEN
Example: The line
puts [ lbnode 0 0 0 print u ]
prints the fluid velocity in node 0 0 0 to the screen. The command set allows to
change the density or fluid velocity in a single node. Setting the other quantities
can easily be implemented. Example:
puts [ lbnode 0 0 0 set u 0.01 0. 0.]
170
12.7. Removing total fluid momentum
Syntax
lbfluid remove_momentum
Required features:
1 LB
2 LB_GPU
3 SHANCHEN
Description
In some cases, such as free energy profile calculations, it might be useful to prevent
interface motion. This can be achieved using the command lbfluid remove_momentum,
that removes the total momentum of the fluid.
12.8. Visualization
Syntax
lbfluid print [vtk] property filename [filename]
Description
The print parameter of the lbfluid command is a feature to simplify visualization. It
allows for the export of the whole fluid field data into a file with name filename at once.
Currently supported values for the parameter property are boundary and velocity when
using LB or LB_GPU and density and velocity when using SHANCHEN. The additional option
vtk enables export in the vtk format which is readable by visualization software such as
paraview1 or mayavi22 . Otherwise gnuplot readable data will be exported. If you plan
to use paraview for visualization, note that also the particle positions can be exportet
in the VTK format 10.5. If the SHANCHEN bicomponent fluid is used, two filenames have
to be supplied when exporting the density field, to save both components.
12.9. Setting up boundary conditions
Syntax
(1) lbboundary shape shape args [velocity vx vy vz ]
(2) lbboundary force [nboundary ]
Required features:
LB_BOUNDARIES
Description
If nothing else is specified, periodic boundary conditions are assumed for the LB fluid.
Variant (1) allows to set up other (internal or external) boundaries.
The lbboundary command syntax is very close to the constraint syntax, as
usually one wants the hydrodynamic boundary conditions to be shaped similarily
to the MD boundaries. Currently the shapes mentioned above are available and
their syntax exactly follows the syntax of the constraint command. For example
1
2
http://www.paraview.org/
http://code.enthought.com/projects/mayavi/
171
lbboundary wall dist 1.5 normal 1. 0. 0.
creates a planar boundary condition at distance 1.5 from the origin of the coordinate
system where the half space x > 1.5 is treated as normal LB fluid, and the other half
space is filled with boundary nodes.
Intersecting boundaries are in principle possible but must be treated with care. In
the current, only partly satisfactory, all nodes that are within at least one boundary
are treated as boundary nodes. Improving this is nontrivial, and suggestions are very
welcome.
Currently, only the so called “link-bounce-back” algorithm for wall nodes is available.
This creates a boundary that is located approximately midway between the lattice nodes,
so in the above example this corresponds indeed to a boundary at x = 1.5. Note that the
location of the boundary is unfortunately not entirely independent of the viscosity. This
can e.g. be seen when using the sample script poiseuille.tcl with a high viscosity.
The bounce back boundary conditions allow to set velocity at a boundary to a nonzero
value. This allows to create shear flow and boundaries moving relative to each other.
This could be a fixed sphere in a channel moving at a finite speed – corresponding
to the galilei-transform of a moving sphere in a fixed channel. The velocity boundary
conditions are implemented according to [57] eq. 12.58. Using this implementation as a
blueprint for the boundary treatment an implementation of the Ladd-Coupling should
be relatively straightforward.
Variant (2) prints out the force on boundary number n_boundary.
12.10. Choosing between the GPU and CPU implementations
Syntax
(1) lbfluid cpu
(2) lbfluid gpu
Required features:
1 LB
2 LB_GPU
Description
A very recent development is an implementation of the LBM for NVIDIA GPUs using
the CUDA framework. On CUDA-supporting machines this can be activated by configuring with configure --with-cuda=/path/to/cuda and activating the feature LB_GPU.
Within the ESPResSo-Tcl-script, the lbfluid command can be used to choose between
the CPU and GPU implementations of the Lattice-Boltzmann algorithm, for further
information on CUDA support see section 6.5.
Variant (1) is the default and turns on the standard CPU implementation of the
Lattice-Boltzmann fluid, while variant (2) turns on the GPU implementation, implying
that all following LB-related commands are executed on the GPU.
Currently only a subset of the CPU commands are available for the GPU implementation. For boundary conditions analogous to the CPU implementation, the feature
LB_BOUNDARIES_GPU has to be activated.
172
12.11. Electrohydrodynamics
Syntax
setmd mu_E µEx µEy µEz
Required features:
LB
LB_ELECTROHYDRODYNAMICS
Description
If the feature LB_ELECTROHYDRODYNAMICS is activated, the (non-GPU) Lattice Boltzmann Code can be used to implicitely model surrounding salt ions in an external electric
field by having the charged particles create flow.
For that to work, you need to set the electrophoretic mobility (multiplied by the
external E-field) µE in all 3 dimensions for your system. The three given parameters
are float values and should, for a meaningful system, be less than 1.0.
For more information on this method and how it works, read the publication [26].
173
13. Electrokinetics
The electrokinetics setup in ESPResSo allows for the description of electro-hydrodynamic
systems on the level of ion density distributions coupled to a Lattice-Boltzmann (LB)
fluid. The ion density distributions may also interact with explicit charged particles,
which are interpolated on the LB grid. In the following paragraph we briefly explain the
electrokinetic model implemented in ESPResSo, before we come to the description of the
interface.
If you are interested in using the electrokinetic implementation in ESPResSo for scientific purposes, please contact G. Rempfer before you start your project.
13.1. Electrokinetic Equations
In the electrokinetics code we solve the following system of coupled continuity, diffusionadvection, Poisson, and Navier-Stokes equations:
∂nk
∂t
= − ∇ · ~jk ;
(13.1)
~jk = −Dk ∇nk − νk qk nk ∇Φ + nk~vfl ;
∆Φ = −4π lB kB T
X
qk n k ;
(13.2)
(13.3)
k
∂~vfl
~
+ ~vfl · ∇~vfl ρfl = −kB T ∇ρfl − qk nk ∇Φ
∂t
~ vfl + (η/3 + η b )∇(∇ · ~vfl );
+ η ∆~
∂ρfl
∂t
= − ∇ · (ρfl~vfl ) ,
which define relations between the following observables
nk
~jk
the number density of the particles of species k,
Φ
the electrostatic potential,
ρfl
the mass density of the fluid,
~vfl
the advective velocity of the fluid,
the number density flux of the particles of species k,
and input parameters
174
(13.4)
(13.5)
Dk
the diffusion constant of species k,
νk
the mobility of species k,
qk
the charge of a single particle of species k,
lB
the Bjerrum length,
kB T
the thermal energy given by the product of Boltzmann’s constant kB
and the temperature T ,
η
the dynamic viscosity of the fluid,
ηb
the bulk viscosity of the fluid.
The temperature T , and diffusion constants Dk and mobilities νk of individual species
are linked through the Einstein-Smoluchowski relation Dk /νk = kB T . The system of
equations described in Eqs. (13.1)-(13.5), combining diffusion-advection, electrostatics,
and hydrodynamics is conventionally referred to as the Electrokinetic Equations.
The electrokinetic equations have the following properties:
• On the coarse time and length scale of the model, the dynamics of the particle
species can be described in terms of smooth density distributions and potentials
as opposed to the microscale where highly localized densities cause singularities in
the potential.
In most situations, this restricts the application of the model to species of monovalent ions, since ions of higher valency typically show strong condensation and
correlation effects – the localization of individual ions in local potential minima
and the subsequent correlated motion with the charges causing this minima.
• Only the entropy of an ideal gas and electrostatic interactions are accounted for.
In particular, there is no excluded volume.
This restricts the application of the model to monovalent ions and moderate charge
densities. At higher valencies or densities, overcharging and layering effects can
occur, which lead to non-monotonic charge densities and potentials, that can not
be covered by a mean-field model such as Poisson-Boltzmann or this one.
Even in salt free systems containing only counter ions, the counter-ion densities
close to highly charged objects can be overestimated when neglecting excluded volume effects. Decades of the application of Poisson-Boltzmann theory to systems of
electrolytic solutions, however, show that those conditions are fulfilled for monovalent salt ions (such as sodium chloride or potassium chloride) at experimentally
realizable concentrations.
• Electrodynamic and magnetic effects play no role. Electrolytic solutions fulfill
those conditions as long as they don’t contain magnetic particles.
• The diffusion coefficient is a scalar, which means there can not be any crossdiffusion. Additionally, the diffusive behavior has been deduced using a formalism
175
Complete in broad
strokes the applicability of the
electrokinetics
model. Also mention the difference
in temperatures
between EK and
LB species.
relying on the notion of a local equilibrium. The resulting diffusion equation,
however, is known to be valid also far from equilibrium.
• The temperature is constant throughout the system.
• The density fluxes instantaneously relax to their local equilibrium values. Obviously one can not extract information about processes on length and time scales
not covered by the model, such as dielectric spectra at frequencies, high enough
that they correspond to times faster than the diffusive time scales of the charged
species.
13.2. Setup
13.2.1. Initialization
Syntax
electrokinetics 1 or 2 or 3 [agrid agrid ] [lb_density lb density]
[visc viscosity] [bulk_visc bulk viscosity] [friction gamma ]
[gamma_odd gamma odd ] [gamma_even gamma even] [T T ]
[bjerrum_length bjerrum length]
Required features:
1 ELECTROKINETICS
2 EK_BOUNDARIES
3 EK_REACTIONS
Description
The electrokinetics command initializes the LB fluid with a given set of parameters,
and it is very similar to the ESPResSo Lattice-Boltzmann lbfluid command in set-up.
We therefore refer the reader to Chapter 12 for details on the implementation of LB in
ESPResSo and describe only the major differences here.
The first major difference with the LB implementation is that the electrokinetics setup is a Graphics Processing Unit (GPU) only implementation. There is no Central
Processing Unit (CPU) version, and at this time there are no plans to make a CPU version available in the future. To use the electrokinetics features it is therefore imperative
that your computer contains a CUDA capable GPU which is sufficiently modern.
To set up a proper LB fluid using the electrokinetics command one has to specify
at least the following options: agrid , lb density, visc, friction, T , and bjerrum length.
The other options can be used to modify the behavior of the LB fluid. Note that the
electrokinetics command does not allow the user to set the time step parameter tau
as is the case for the lbfluid command, this parameter is instead taken directly from the
input of the setmd t_step command. The LB mass density is set independently from
the electrokinetic number densities, since the LB fluid serves only as a medium through
which hydrodynamic interactions are propagated, as will be explained further in the
next paragraph. If no lb density is specified, then our algorithm assumes lb density =
1.0. The two ‘new’ parameters are T the temperature at which the diffusive species
are simulated and bjerrum length the Bjerrum length associated with the electrostatic
properties of the medium. See the above description of the electrokinetic equations for
176
an explanation of the introduction of a temperature, which does not come in directly
via a thermostat that produces thermal fluctuations.
13.2.2. Diffusive Species
Syntax
electrokinetics 1 or 2 or 3 species number [density density] [D D]
[valency valency] [ext_force fx fy fz ]
Required features:
1 ELECTROKINETICS
2 EK_BOUNDARIES
3 EK_REACTIONS
Description
The electrokinetics command followed by an integer species number (in the range
0 to 10) and several options can be used to initialize the diffusive species. Here the
options specify: the number density density, the diffusion coefficient D, the valency of the
particles of that species valency, and an optional external (electric) force which is applied
to the diffusive species. As mentioned before, the LB density is completely decoupled
from the electrokinetic densities. This has the advantage that greater freedom can be
achieved in matching the internal parameters to an experimental system. Moreover, it
is possible to choose parameters for which the LB is more stable. The LB fluid must
already be (partially) set up using the electrokinetics agrid ... command, before the
diffusive species can be initialized. The variables density, D, and valency must be set
to properly initialize the diffusive species; the ext force is optional.
13.2.3. Boundaries
Syntax
electrokinetics 1 or 2 or 3 boundary 2 [charge_density charge density]
[shape shape args]
Required features:
1 ELECTROKINETICS
2 EK_BOUNDARIES
3 EK_REACTIONS
Description
The boundary command allows one to set up (internal or external) boundaries for the
electrokinetics algorithm in much the same way as the lbboundary command is used
for the LB fluid. The major difference with the LB command is given by the option
charge density, with which a boundary can be endowed with a volume charge density.
To create a surface charge density, a combination of two oppositely charged boundaries,
one inside the other, can be used. However, care should be taken to maintain the surface
charge density when the value of agrid is changed. Currently, the following shapes are
available: wall, sphere, cylinder, rhomboid, pore, stomatocyte, hollow cone, and spherocylinder. We refer to the documentation of the lbboundary command (Chapter 12) for
information on the options shape args associated to these shapes. In order to properly
set up the boundaries, the charge density and relevant shape args must be specified.
177
13.3. Output
13.3.1. Fields
Syntax
electrokinetics 1 or 2 or 3 print property 1 or 2 [vtk] filename [filename]
Required features:
1 ELECTROKINETICS
2 EK_BOUNDARIES
3 EK_REACTIONS
Description
The print parameter of the electrokinetics command enables simple visualization of
simulation data. A property of the fluid field can be exported into a file with name
filename in one go. Currently, supported values of the parameter property are: density,
velocity, potential , and boundary, which give the LB fluid density, the LB fluid velocity,
the electrostatic potential, and the location and type of the boundaries, respectively. The
boundaries can only be printed when the EK_BOUNDARIES is compiled in. The additional
option vtk can be used to directly export in the vtk format. The vtk format is readable
by visualization software such as paraview1 and mayavi22 . If the [vtk] option is not
specified, a gnuplot readable data will be exported.
Syntax
electrokinetics 1 or 2 or 3 species number print property [vtk] filename
[filename]
Required features:
1 ELECTROKINETICS
2 EK_BOUNDARIES
3 EK_REACTIONS
Description
This print statement is similar to the above command. It enables the export of diffusive
species properties, namely: density and flux , which specify the number density and flux
of species species number , respectively.
13.3.2. Local Quantities
Syntax
electrokinetics 1 or 2 or 3 node x y z velocity
Required features:
1 ELECTROKINETICS
2 EK_BOUNDARIES
3 EK_REACTIONS
Description
The node option of the electrokinetics command allows one to output the value of a
quantity on a single LB node. The node is addressed using three integer values which run
from 0 to dim x /agrid , dim y/agrid , and dim z /agrid , respectively. Thus far, only the
velocity of the LB fluid can be printed in the standard electrokinetics implementation.
For other quantities the lbnode command may be used.
1
2
http://www.paraview.org/
http://code.enthought.com/projects/mayavi/
178
Syntax
electrokinetics 1 or 2 or 3 species number node x y z density
Required features:
1 ELECTROKINETICS
2 EK_BOUNDARIES
3 EK_REACTIONS
Description
This command can be used to output the number density of the species number -th
diffusive species on a single LB node.
13.4. Catalytic Reaction
13.4.1. Concept
The electrokinetics solver implemented in ESPResSo can be used to simulate a system,
for which in addition to the electrokinetic equations, there is a (local) catalytic reaction
which converts one species into another.
If you are interested in using this implementation in ESPResSo for scientific purposes,
please contact J. de Graaf before you start your project.
Currently, a linear reaction is implemented which converts one species into two others,
in order to model the catalytic decomposition of hydrogen peroxide in the presence of a
platinum catalyst: 2H2 O2 → 2H2 O + O2 . The decomposition of H2 O2 is in reality more
complicated than the linear reaction introduced here, since it is assumed to proceed via
several intermediate complexed-states, but our model can be thought of as modeling
the rate-limiting step. If we assume that there are three non-ionic species with number
densities nk , where n0 = [H2 O2 ], n1 = [H2 O], and n2 = [O2 ], then we can write the
(electro)kinetic equations for this system as
∂nk
∂t
= − ∇ · ~jk + fk cnk ;
(13.6)
~jk = −Dk ∇nk + nk~vfl ;
(13.7)
X
∂~vfl
~
+ ~vfl · ∇~vfl ρfl = −kB T
∇nk
∂t
k
~ vfl + (η/3 + η b )∇(∇ · ~vfl );
+ η ∆~
∂ρfl
∂t
= − ∇ · (ρfl~vfl ) ,
(13.8)
(13.9)
which define relations between the following observables
nk
~jk
the number density of the particles of species k,
ρfl
the mass density of the fluid,
the number density flux of the particles of species k,
179
~vfl
the advective velocity of the fluid,
and input parameters
Dk
the diffusion constant of species k,
kB T
the thermal energy given by the product of Boltzmann’s constant kB
and the temperature T ,
η
the dynamic viscosity of the fluid,
ηb
the bulk viscosity of the fluid,
fk
the reaction constant f0 ≡ −1, f1 = 1 and f2 = 0.5 for the above reaction,
c
the reaction rate.
In this set of equations we have fully decoupled the number densities and the fluid mass
density. N.B. We have set the initial fluid mass density is not necessarily equal to the
sum of the initial species number densities. This means that some care needs to be taken
in the interpretation of the results obtained using this feature. In particular, the solution
of the Navier-Stokes equation exclusively models the momentum transport through the
(multicomponent) fluid, while the diffusive properties of the individual chemical species
are handled by Eqs. (13.6) and (13.7).
It is important to note that to ensure mass conservation the reaction must satisfy:
X
fk mk = 0,
(13.10)
k
where mk is the molecular mass of a reactive species. Unfortunately, the current electrokinetic implementation does not conserve mass flux locally. That is to say, the LB
fluid is compressible and the sum of the fluxes of the three species is not equal to zero in
the frame co-moving with the advective fluid velocity. It is therefore debatable whether
it is necessary to impose Eq. (13.10), since the EK algorithm itself does not conserve
mass density. However, we strived to be as accurate as possible and in future versions
of the EK algorithm the lack of incompressiblity will be addressed.
The reaction is specified by the second term on the right-hand side of Eq. (13.6). It
is important to note that this term can be set locally, as opposed to the other terms in
the equation system Eqs. (13.6)-(13.9), in our implementation, as will become clear in
the following. This has the advantage that catalytic surfaces may be modeled.
180
13.4.2. Initialization and Geometry Definition
Syntax
electrokinetics 1 or 2 or 3 reaction 3 [reactant_index reactant index ]
[product0_index product0 index ] [product1_index product1 index ]
[reactant_resrv_density reactant resrv density]
[product0_resrv_density product0 resrv density]
[product1_resrv_density product1 resrv density]
[reaction_rate reaction rate] [mass_reactant mass reactant]
[mass_product0 mass product0 ] [mass_product1 mass product1 ]
[reaction_fraction_pr_0 reaction fraction pr 0 ]
[reaction_fraction_pr_1 reaction fraction pr 1 ]
Required features:
1 ELECTROKINETICS
2 EK_BOUNDARIES
3 EK_REACTIONS
Description
The electrokinetics reaction command is used to set up the catalytic reaction between three previously defined the diffusive species, of which the i identifiers are given by
reactant index , product0 index , and product1 index , respectively. In the 1:2 reaction,
these fulfill the role of the reactant and the two products, as indicated by the naming convention. For each species a reservoir (number) density must be set, given by the
variables reactant resrv density, product0 resrv density, and product1 resrv density, respectively. These reservoir densities correspond to the initial number densities associated
with the reactive species. The reservoir densities, in tandem with reservoir nodes, see
below, can be used to keep the reaction from depleting all the reactant in the simulation box. The reaction rate variable specifies the speed at which the reaction proceeds. The three masses (typically given in the atomic weight equivalent) are used to
determine the total mass flux provided by the reaction, as described above, and are
also used to check whether the reaction ratios that are given satisfy the chemical requirement of mass conservation. Finally, the parameters reaction fraction pr 0 and
reaction fraction pr 1 specify what fractions of the product are generated when a given
quantity of reactant is catalytically converted. To use a chemical reaction, all options
for the electrokinetics reaction command must be specified.
Syntax
electrokinetics 1 or 2 or 3 reaction 3 region 3 [reaction_type reaction type]
[shape shape args]
Required features:
1 ELECTROKINETICS
2 EK_BOUNDARIES
3 EK_REACTIONS
Description
The region option of the electrokinetics reaction command allows one to set up
regions in which the reaction takes place with the help of the constraints that are available to set up boundaries. The integer value reaction type can be used to select the
reaction: 0 no reaction takes place for this region, 1 the catalytic reaction takes place
in this region, and 2 the region functions as a reservoir, wherein the species densities
are reset to their initial (or reservoir) concentrations. The rest of the command fol-
181
lows the same format of the electrokinetics boundary command. Currently, the
following shapes are available: box, wall, sphere, cylinder, rhomboid, pore, stomatocyte,
hollow cone, and spherocylinder. The box shape is a region specific command, which
can be used to set the entire simulation box to a specific reaction value. To use the
electrokinetics reaction region command, one must first set up a reaction, as described above. To successfully specify a region all the relevant arguments that go with
the shape constraints must be provided.
Parsing PDB Files
Syntax
electrokinetics 1 or 2 or 3 pdb-parse 2 pdb filename itp filename
Required features:
At the moment
this fails badly, if
you try to parse
incorrectly formatted files. This
will be fixed in
the future.
1 ELECTROKINETICS
2 EK_BOUNDARIES
3 EK_REACTIONS
Description
The electrokinetics pdb-parse feature allows the user to parse simple PDB files,
a file format introduced by the protein database to encode molecular structures. Together with a topology file (here itp filename) the structure gets interpolated to the
electrokinetics grid. For the input you will need to prepare a PDB file with a gromacs
force field to generate the topology file. Normally the PDB file extension is .pdb, the
topology file extension is .itp. Obviously the PDB file is placed instead of pdb filename
and the topology file instead of itp filename.
13.4.3. Reaction-Specific Output
Syntax
electrokinetics 1 or 2 or 3 print property 3 [vtk] filename [filename]
Required features:
1 ELECTROKINETICS
2 EK_BOUNDARIES
3 EK_REACTIONS
Description
The print parameter of the electrokinetics command can be used in combination with
the EK_REACTION feature to give advanced output options. Currently, supported values
of the parameter property are: pressure and reaction tags, which give the location and
type of the reactive regions and the ideal-gas pressure coming from the diffusive species,
respectively. To use this command a reaction must be set up.
182
14. Object-in-fluid
Please cite [12] (BibTeX-key cimrak in file doc/ug/citations.bib) if you use the
object-in-fluid implementation described below. For more details also see the
documentation at http://cell-in-fluid.fri.uniza.sk/oif-documentation or contact the
ˇ
Cell-in-fluid Research Group at University of Zilina.
Simulations using ESPResSo work mostly with objects (molecules, atoms, polymers,
colloids, crystals, . . . ) that are physically composed of points linked together with bonds.
These objects are like skeletons, without inner or outer volume.
The idea behind this module, is to use ESPResSo for objects that do have inner volume,
for example blood cells, magnetic beads, capsules, . . . The boundary of an object is
covered with triangular mesh. The vertices of the mesh are declared in ESPResSo as
particles. The edges of the mesh define elastic forces keeping the shape of the object.
The movement of object is achieved by adding forces to the mesh points.
Modelled elastic or rigid objects are immersed in the LB fluid flow. The fluid interacts
with an elastic object resulting in its deformation; this immediately generates forces
acting back on the fluid. The aim is to describe the immersed object using the notion of
particles, and to create bonds between these particles representing elastic or rigid forces.
The objects are composed of a membrane encapsulating the fluid inside the object.
For now, the inside fluid must have the same density and viscosity as the outside fluid.
The object is represented by its membrane (boundary), that is discretized using a triangulation. Such triangulation defines interacting particles distributed on the surface of
the immersed object [21]:
• between two particles, corresponding to the edges in the triangulation (modelling
the stretching of the membrane),
• between three particles, corresponding to the triangles of the triangulation (local
area, or local surface preservation of the membrane),
• between four particles, corresponding to two triangles from the triangulation sharing a common edge (bending of the membrane).
The object immersed in the fluid moves under the influence of the deforming forces,
defined through the bonds, and under the influence of the fluid motion. This interaction
is based on the frictional force between the fluid and the surface particles. Therefore the
object moves in the flow only if there is a nonzero difference between the fluid velocity
and the particle velocity. In other words, there has to be at least small flow through the
membrane, which is in most cases unphysical. However, this unphysical flow through
the membrane is probably negligible in larger scales.
183
14.1. Membranes
With this approach, it is easy to model also elastic sheets, or free membranes that do
not necessarily enclose a 3D object. In this case, area force global and volume force
interactions are not needed, since these two interactions are meant for closed immersed
objects.
14.2. Parameters
There are several parameters involved in this model. All of them should be calibrated
according to the intended application.
• Mass of the particles. Every particle has its mass, which influences the dynamics.
• Friction coefficient. The main parameter describing the fluid-particle interaction
is the friction parameter from the ESPResSo command lbfluid.
• Parameters of elastic moduli. Elastic behaviour can be described by five different elastic moduli: hyperelastic stretching, bending, local and global area preservation and volume preservation. Each of them has its own scaling parameter:
ks, kb, kal, kag, kv. Their mathematical formulations have been taken from [21].
The mass of the particles and the friction coefficient can be calibrated using the drag
coefficients of the ellipsoidal objects. These drag coefficients have known analytical
values and the mass and friction can be calibrated to fit this values. More details about
the calibration can be found in [12].
The elastic parameters are specific to the immersed objects. They correspond to
their physical values. More details about their mechanical and biological meaning is
presented in [14] specifically for red blood cells. However, the proper calibration to fit
the experimental data has been performed in [12].
14.3. Geometry
The membrane of the immersed object is triangulated. In doc/tutorials/03-object in fluid you can find an example using deformable objects in the fluid.
Triangulation can be obtained using various software tools. Two files are needed for mesh
input: mesh-nodes.dat and mesh-triangles.dat. The parameters of the mesh are the
184
number of particles on the surface of the immersed object, denoted by mesh_nnode, and
the number of triangular faces in the triangulation, denoted by mesh_ntriangle. These
parameters are obtained automatically from mesh-nodes.dat and mesh-triangles.datby
counting the number of lines in respective files.
The mesh-nodes.dat thus contains mesh_nnode lines with three real numbers separated by blank space, representing three coordinates of the corresponding particle. The
membrane is thus discretized into mesh_nnode particles with IDs starting from 0 to
mesh_nnode-1. The IDs are assigned in the same order as in the mesh-nodes.dat file.
The mesh-triangles.dat contains mesh_ntriangle lines with three nonnegative integers separated by blank space. Each line represents one triangle in the triangulation.
For algorithmic purposes it is crucial to have defined a correct orientation of the triangle. The orientation is defined using the normal vector associated with the triangle. The
important rule is that the normal vector of the triangle must point inside the immersed
object.
As an example, let us have one line in the file mesh-triangles.dat with numbers 4,
0 and 7. This means that particles with IDs 4, 0 and 7 form one triangular face of the
triangulation. The orientation is defined as follows: create two vectors v1 and v2 , such
that v1 is pointing from particle 4 to particle 0, and v2 is pointing from particle 4 to
particle 7. Be careful, the order of vectors and particles matters!
The normal vector n is computed as a vector product v1 × v2 . The direction of n
can be determined by the rule of right hand: the thumb points in the v1 direction, the
index finger in the v2 direction and the middle finger in the n direction. Following this
principle, all the lines in the mesh-triangles.dat files must be such that the normal
vectors of the corresponding triangles points inside the immersed object.
These two files are sufficient to describe the geometry and topology of the triangulation. The following geometric entities are necessary for the definition of bonded
interactions: position of the particles, edges, lengths of the edges, triangles, areas of
triangles, angles between two triangles sharing a common edge, surface of the immersed
object, volume of the immersed object. All these geometrical entities can be computed
using the information from the files mesh-nodes.dat and mesh-triangles.dat and the
computation is done in the script scripts/object_in_fluid.tcl.
The script scripts/object_in_fluid.tcl reads both mesh files, generates list of
edges, and computes all geometrical entities needed for definition of bonded interactions.
It then executes commands creating the particles, interactions and bonds.An example
of part command is as follows:
part 0 pos 3.0 3.0 6.0 type 1 mol 1 mass 1
Note, the is feature mol that used for the particles. We use this feature we distinguish
between different objects. The upper limit for the number of objects is 10000. However
it can be increased by changing the MAX_OBJECTS_IN_FLUID constant.
The following example shows an interaction.
inter 106 stretching_force 4.6 5.0
185
This command (”invisible” for the user who executes the scripts/object_in_fluid.tcl
script) creates an interaction for stretching with ID 106. Detailed description of the
available types of interactions is presented in Section 5.4.
14.4. Available commands
In order to use the object-in-fluid (OIF) commands and work with immersed objects, the
following ESPResSo features need to be compiled in: MASS, EXTERNAL_FORCES. We do not
specifically require LB, LB_BOUNDARIES, CONSTRAINTS, SOFT_SPHERE, AREA_FORCE_GLOBAL,
VOLUME_FORCE. They are most likely to be used (for objects immersed in fluid and interacting with boundaries and each other), but they are not necessary for the following
commands. For up-to-date overview of available oif commands see the OIF user guide
at cell-in-fluid.fri.uniza.sk/oif-documentation.
14.4.1. Initialisation
Syntax
oif_init
Description
Must be used before any other OIF command, initializes all global variables and lists,
does not take any arguments.
14.4.2. Information about object-in-fluid structures
Syntax
oif_info
Description
Prints information about whole framework, e.g. all global variables, currently available
templates and objects, etc. Does not take any arguments.
14.4.3. Templates for objects
Syntax
oif_create_template 1,2 template-id tid nodes-file nodes.dat
triangles-file triangles.dat [stretch x y z ] [ks ks value]
[kb kb value] [kal kal value] [kag kag value] 3 [kv kv value] 4
Required features: 1 MASS
4 VOLUME_FORCE
2 EXTERNAL_FORCES
3 AREA_FORCE_GLOBAL
Description
This command creates a template that will be used for all objects that share the same
elastic properties and have the same triangulation.
186
Arguments
• tid specifies a unique ID for each template. The first template has the ID 0. The
following ones need to be be numbered consecutively.
• nodes.dat input file, each line contains three real numbers. These are the x, y, z
coordinates of individual surface mesh nodes of the objects.
• triangles.dat input file, each line contains three integers. These are the ID numbers of the mesh nodes as they appear in nodes.dat. Note, the first node has ID
0.
• [stretch x y z ] coefficients by which the coordinates stored in nodes.dat will be
stretched in the x, y, z direction. The default values are 1.0 1.0 1.0.
• [ks ks value] elastic modulus for stretching forces
• [kb kb value] elastic modulus for bending forces
• [kal kal value] elastic modulus for local area forces
• [kag kag value] elastic modulus for global area forces
• [kv kv value] elastic modulus for volume forces
The three switches ks, kb and kal set elastic parameters for local interactions - ks for
edge stiffness, kb for angle preservation stiffness and kal for triangle surface preservation
stiffness. This stiffness can be either uniform over the whole object, or non-uniform. In
case of stretching modulus, we can have spring stiffness the same for all edges in the whole
object, or we can choose the value for every edge of the object separately. Analogically,
for kal and kb. Therefore, there are two options for setting ks, kal and kb stiffness.
Here is the explanation for ks:
• Uniform stiffness: To set uniform stiffness for all edges in the object, use ks
ks value
• Non-uniform stiffness: To set non-uniform stiffness, prepare a file filename with
number of lines equal to the number of edges of the triangulation. Each line should
contain a real number between 0 and 1, so called ”weight”. Then call ks filename
ksMin ksMax This command reads the weights weighti for each edge and the
stiffness for that edge is set to
ksi = ksM in ∗ (1 − weighti ) + ksM ax ∗ (weighti )
For bending stiffness, filename must contain the same number of lines as there are
edges in the object. However, for local area preservation, the stiffness constant is
linked to triangles. Therefore, filename must contain the same number of lines as
there are triangles in the object.
Warning: At least one elastic modulus needs to be set for the object.
187
14.4.4. Elastic objects
Syntax
oif_add_object 1,2, possibly 3 or 4 object-id oid template-id tid origin x y
z part-type type [rotate x y z ] [mass m]
Required features: 1 MASS
4 VOLUME_FORCE
2 EXTERNAL_FORCES
3 AREA_FORCE_GLOBAL
Description
Using a previously defined template tid , this command creates a new object. Features
AREA_FORCE_GLOBAL and VOLUME_FORCE are needed, if the template used the corresponding elastic moduli.
Arguments
• oid unique ID for each object, the first object has the ID 0. The following ones
should be numbered consecutively.
• tid object will be created using nodes, triangle incidences, elasticity parameters
and initial stretching saved in this template.
• origin x y z center of the object will be at this point.
• part-type type can be any integer starting at 0. All particles of one object have
the same part-type. One can have more objects with the same type of particles,
but this is not recommended, because the interactions between objects are set up
using these types.
• [rotate x y z ] angles in radians, by which the object is initially rotated around
the x, y, z axis. Default values are 0.0 0.0 0.0.
• [mass m] mass of one particle. Default value is 1.0
14.4.5. Mesh analysis
Syntax
oif_mesh_analyze nodes-file nodes.dat triangles-file
triangles.dat [orientation] [repair output file.dat method ]
[shift-node-ids output file.dat]
Description
This command is useful for some preparatory work with mesh before it is used for
creating elastic objects.
Arguments
• nodes-file nodes.dat - file with coordinates of the mesh nodes. The center of
the object should be as close to (0,0,0) as possible.
188
• triangles-file triangles.dat - file with incidences for all triangles. Each line of
this file contains three integer IDs (starting from 0) with indices of three vertices
forming one triangle.
• [orientation] checks whether all triangles of the surface mesh are properly oriented. For now, only works for convex (or almost convex) objects.
• [repair output file.dat method ] outputs the corrected triangles.dat file into output file.dat.
For now, only works for convex (or almost convex) objects. method needs to be
set to 1.
• [shift-node-ids output file.dat] subtracts 1 from all numbers in triangles.dat
and saves a new file output file.dat. This is useful, if the mesh generating software
starts numbering the particles from 1 instead of 0.
14.4.6. Output information about specific object
Syntax
oif_object_output 1,2, possibly 3 or 4 object-id oid [vtk-pos output file1 .dat]
[vtk-pos-folded output file2 .dat] [vtk-aff output file3 .dat]
[mesh-nodes output file4 .dat]
Required features: 1 MASS
4 VOLUME_FORCE
2 EXTERNAL_FORCES
3 AREA_FORCE_GLOBAL
Description
This command is used to output information about the object that can be used for
visualisation or as input for other simulations.
Arguments
• oid - the id of the object
• [vtk-pos output file1 .dat] outputs the mesh of the object to the desired output file1 .dat.
Paraview can directly visualize this file.
• [vtk-pos-folded output file2 .dat] the same as the previous option, however the
whole object is shift such that it is visualized within the simulation box. This
option is useful for simulating periodical processes when objects flowing out on
one side of simulation box are transferred to the opoosite side.
• [vtk-aff output file3 .dat] outputs affinity bonds that are currently activated. If
no bonds are present, the file will be generated anyway with no bonds to visualize.
Paraview can directly visualize this file.
• [mesh-nodes output file4 .dat] outputs the positions of the mesh nodes to output file4 .dat.
In fact, this command creates a new nodes.dat file that can be used by oif_object_set.
The center of the object is located at point (0,0,0). This command is aimed to
store the deformed shape in order to be loaded later.
189
14.4.7. Descriptive information about specific object
Syntax
oif_object_analyze 1,2, possibly 3 or 4 object-id oid [origin]
[pos-bounds bname] [approx-pos] [volume] [surface-area] [velocity]
[elastic-forces name(s) output file.dat] [f-metric name]
Required features: 1 MASS
4 VOLUME_FORCE
2 EXTERNAL_FORCES
3 AREA_FORCE_GLOBAL
Description
This command is used to output information about the properties of the object. Some
of these properties can also be visualized.
Arguments
• oid - the id of the object
• [origin] - outputs the location of the center of the object
• [pos-bounds bname] computes six extremal coordinates of the object. More precisely, runs through the all mesh points and remembers the minimal and maximal
x-coordinate, y-coordinate and z-coordinate. If bname is one of these: z-min,
z-max, x-min, x-max, y-min, y-max then the procedure returns one number according to the value of bname. If bname is all , then the procedure returns a list
of six numbers, namely x-min, x-max, y-min, y-max, z-min, z-max.
• [approx-pos] - outputs the approximate location of the center of the object. It is
computed as average of 6 mesh points that have extremal x, y and z coordinates
at the time of object loading.
• [volume] - outputs the current volume of the object
• [surface-area] - outputs the current surface of the object
• [velocity] - outputs the current average velocity of the object. Runs over all
mesh points and calculates their average velocity.
• [elastic-forces name(s) output file.dat] (example elastic-forces kv kb ks
out.vtk ) - name(s) can be one to five of the following: ks, kb, kal , kag, kv , in any
order. Arguments cannot repeat. Corresponding forces are computed for each
node by summing up all contributions. This computation can be considered as
a simple approach to elastic energy evaluation. If the output file has an extension .vtk, a vtk file is written with a scalar data fields that can be visualized in
ParaView to show the color coded force magnitudes on the surface of the object.
If more than one options were given, one can switch between the visualizations
in ParaView. If the file has any other extension (.txt, .dat, . . . ) a data file is
written with magnitudes of resulting forces in each node (five columns - some of
them will be zero, if not all arguments were given).
190
• [f-metric name] name can be one of the following: ks, kb, kal , kag, kv . There
is no output file as a second argument for these, because the output is just one
number - fmetric/”naive energy” computed as a sum of magnitudes of respective
elastic forces over all nodes of the object.
14.4.8. Setting properties for specific object
Syntax
oif_object_set 1,2, possibly 3 or 4 object-id oid [force x y z ] [origin x y z ]
[mesh-nodes mesh nodes.dat] [kill-motion] [un-kill-motion]
Required features: 1 MASS
4 VOLUME_FORCE
2 EXTERNAL_FORCES
3 AREA_FORCE_GLOBAL
Description
This command sets some properties of the object.
Arguments
• oid - the id of the object
• [force x y z ] - sets the force vector (x , y, z ) to all mesh nodes of the object.
Setting is done using ESPResSo command part $i set ext_force $x $y $z .
Note, that this command sets the external force in each integrate step. So if you
want to use the external force only in one iteration, you need to set zero external
force in the following integrate step
• [origin x y z ] - moves the object so that the origin has coordinates (x , y, z )
• [mesh-nodes mesh nodes.dat] - deforms the object such that its origin stays unchanged, however the relative positions of the mesh points are taken from file
mesh nodes.dat. The file mesh nodes.dat should contain the coordinates of the
mesh points with the origin’s location at (0,0,0). The procedure also checks
whether number of lines in the mesh nodes.dat file is the same as the number of
triangulation nodes of the object.
• kill-motion - stops all the particles in the object (analogue to the part pid
fix 1 1 1 command for single particles).
• un-kill-motion - releases the particles in the object (analogue to the part pid
unfix command for single particles).
191
15. Immersed Boundary Method for soft
elastic objects
Please contact the Biofluid Simulation and Modeling Group at the University of
Bayreuth if you plan to use this feature.
This section describes an alternative way to include soft elastic objects somewhat
different from the previous chapter. In the Immersed Boundary Method (IBM), soft
particles are considered as an infinitely thin shell filled with liquid (see e.g. [44, 13, 35]).
When the shell is deformed by an external flow it responds by elastic restoring forces
which are transmitted into the fluid. In the present case, the inner and outer liquid are
of the same type and are simulated using Lattice-Boltzmann.
Numerically, the shell is discretized by a set of marker points connected by triangles. The marker points are advected with exactly the local fluid velocity, i.e., they do
not possess a mass nor a friction coefficient (this is different from the Object-in-Fluid
method of the previous chapter). We implement these marker points as virtual particles in ESPResSo which are not integrated using the usual velocity-verlet scheme, but
instead are propagated using a simple Euler algorithm with the local fluid velocity (if
the IMMERSED_BOUNDARY feature is turned on).
To compute the elastic forces, three new bonded interactions are defined ibm triel,
ibm tribend and ibm volCons:
• ibm triel is a discretized elastic force with the following syntax
Syntax
inter ID ibm_triel ind1 ind2 ind3 max law
Required features:
IMMERSED_BOUNDARY
Description
where ind1 , ind2 and ind3 represent the indices of the three marker points making
up the triangle. The parameter max specifies the maximum stretch above which
the bond is considered broken. The final parameter law can be either
NeoHookean <k>
or
Skalak <k1> <k2>
which specifies the elastic law and its corresponding parameters (see e.g. [35]).
192
• ibm tribend is a discretized bending potential with the following syntax
Syntax
inter ID ibm_tribend ind1 ind2 ind3 ind4 method kb [flat|initial]
Required features:
IMMERSED_BOUNDARY
Description
where ind1 , ind2 , ind3 and ind4 are four marker points corresponding to two
neighboring triangles. The indices ind1 and ind3 contain the shared edge. Note
that the marker points within a triangle must be labelled such that the normal
vector ~n = (~rind2 − ~rind1 ) × (~rind3 − ~rind1 ) points outward of the elastic object.
The parameter method allows to specify different numerical ways of computing
the bending interaction. Currently, two methods are implemented, where the first
one ([TriangleNormals]) follows [35] and the second one ([NodeNeighbors]) follows
[24]. In both cases, kb is the bending modulus. The options [flat] or [initial]
specify whether the reference shape is a flat configuration or whether the initial configuration is taken as reference shape, this option is only available for the
[TriangleNormals] method.
• ibm volCons is a volume-conservation force. Without this correction, the volume of
the soft object tends to shrink over time due to numerical inaccuracies. Therefore,
this implements an artificial force intended to keep the volume constant. If volume
conservation is to be used for a given soft particle, the interaction must be added
to every marker point belonging to that object. The syntax is
Syntax
inter ID ibm_volCons softID kv
Required features:
IMMERSED_BOUNDARY
Description
where softID identifies the soft particle and kv is a volumetric spring constant [35].
193
16. External package: mbtools
mbtools1 is a set of tcl packages for setting up, analyzing and running simulations of
lipid membrane systems.
mbtools comes with a basic set of tabulated forces and potentials for lipid interactions
and some example scripts to help explain the syntax of the commands. If you make use
of mbtools or of these potentials please acknowledge us with a citation to:
* Cooke, I. R., Kremer, K. and Deserno, M. (2005): Tunable, generic model for fluid
bilayer membranes, Phys. Rev. E. 72 - 011506
16.1. Introduction
mbtools is located in the folder Espresso/packages/mbtools.
One of the main features of mbtools is the ability to easily create initial lipid configurations with interesting geometries. These include flat membranes, cylinders, spheres,
toroids, and randomly distributed gases. Each of these shapes is referred to as a geometry and any number of geometries can be combined in a single simulation. Once the
geometry has been chosen the user specifies the molecules which should be placed in
this geometry. For example one could choose sphere as a geometry and then define two
different lipids and/or a protein to be placed on the sphere. Within reason (e.g. size
restrictions) it should be possible to use any mixture of known molecule types on any
geometry. The molecule types available at present include proteins, lipids of any length,
and spherical colloids.
mbtools includes several miscellaneous utility procedures for performing tasks such
as warmup, setting tabulated interactions, designating molecules to be trapped and a
variety of topology related sorting or data analysis functions.
The analysis part of the mbtools package is designed to wrap together all the analysis
for a simulation into a single simple interface. At the beginning of the simulation the
user specifies which analyses should be performed by appending its name and arguments
to a variable, analysis_flags. After the analysis is setup one can then simply call do_analysis to perform all the specified proceedures. Analysis will store a data value each
time do_analysis is called. Then when a call to print_averages is made the average
of all stored values is printed to a file and the store of values is reset to nil.
1
This documentation was written by Ira R. Cooke and published on his website. It has been transcripted
by Tristan Bereau.
194
16.2. Installing and getting started
Since mbtools is provided as part of the espresso molecular dynamics simulation package
you will need to download and install Espresso before you can use it. Espresso can be
downloaded free from http://espressomd.org.
Once you have installed espresso you can find mbtools by looking inside the packages
subdirectory. Inside the packages/mbtools directory you will see a directory for each
of the mbtools subpackages as well as an examples directory. All of the examples scripts
should work out of the box except those involving colloids which require you to install
icover.sh (see documentation for hollowsphere molecule type). To run the simplebilayer
example cd to the examples directory and then type:
$ESPRESSO_SOURCE/$PLATFORM/Espresso scripts/main.tcl simplebilayer.tcl
The first part of this command is simply the full path to the appropriate espresso
executable on your machine when running on multiple processors). Obviously you will
need to have the $ESPRESSO_SOURCE and $PLATFORM environment variables set for it to
work. After this executable the relative paths to two tcl scripts are given. The first of
these main.tcl is given as an argument to espresso and is therefore interpreted first by
the espresso tcl interpreter. The second tcl script simplebilayer.tcl is in turn passed
as an argument to main.tcl.
Why separate the tcl commands into two files ?
This is really a matter of preference but if we keep all of the key commands and
complex coding in a single file main.tcl and delegate simple parameter setting to a
separate file it tends to be much easier to manage large numbers of jobs with regularly
changing requirements. Regardless of your personal preferences, the important point to
note is that all of the important commands are contained in main.tcl and you should
probably start there to get an understanding for how mbtools works.
Running the simplebilayer example should produce a directory called simplebilayer
which contains the output from your simulation. To view the results cd to the simplebilayer directory and look at the contents. The directory contains many files including:
• The configurations generated during warmup : warm.∗ .gz
• pdb files corresponding to warmup configurations : warm.vmd∗ .gz
• The configurations generated during the main run : simplebilayer.∗ .gz
• pdb files corresponding to main run configs : simplebilayer.vmd∗ .gz
• The most recently generated checkpoint file : checkpoint.latest.gz
• A directory containing the second most recent checkpoint file: checkpoint_bak
• A file containing the topology of the system : simplebilayer.top
• The original parameter file with which you ran the simulation : simplebilayer.tcl
195
• A original parameter file with which you ran the simulation : simplebilayer.tcl
• Files containing analysis output for example : time_vs_boxl_tmp
• Force and energy tables : ∗ .tab
• VMD script for visualising the warmup : warm_animation.script
• VMD script for visualising the main trajectory : vmd_animation.script
To visualise your results using the vmd scripts you need to make sure that you have
vmd installed properly and that you have the special vmd procedures used by the espresso
team (i.e. support for the loadseries command). Then you can visualise by typing:
vmd -e vmd_animation.script
16.3. The main.tcl script
The main.tcl file provided in the examples/scripts directory is a relatively complete
script written using mbtools. It is designed to run all of the examples provided but no
more. No doubt you will need to extend it for your own purposes.
16.3.1. Variables used by main.tcl
main.tcl expects the user to set various parameters in a parameters.tcl file (e.g.
simplebilayer.tcl). Some of these parameters have defaults and generally don’t need
to be worried about except for specific cases. In the following list variables that have no
default and therefore must be set in the parameter file are noted with an asterisk.
• thermo [Langevin] The type of thermostat to be used. Set to DPD for a dpd
thermostat. Any other value gives a langevin
• dpd gamma Required if you set the thermo to DPD
• dpd r cut Required if you set the thermo to DPD
• warmup temp [$systemtemp] The temperature at which the warmup is performed.
The default behaviour is to use the system temperature
• warmsteps [100] Number of integrate steps per warmup cycle
• warmtimes [20] Number of calls to integrate over which the warmup occurs
• free warmsteps [0] Warmup steps to be used for the warmup that occurs after
particles are freed of any temporary constraints.
• free warmtimes [0] Warmup times to be used for the warmup that occurs after
particles are freed of any temporary constraints.
196
• npt [off ] Whether to use the constant pressure barostat
• p ext The pressure you want to simulate at. Required if npt is set to on
• piston mass box mass. Required if npt is set to ”on”
• gamma 0 Required if npt is on. Usually set to 1 as for langevin gamma
• gamma v Required if npt is on. Box friction
• use vmd [offline] vmd mode
• mgrid [8] The number of meshpoints per side for dividing the bilayer plane into a
grid
• stray cut off [1000.0] Distance of the end tail bead from the bilayer midplane
beyond which a lipid is deemed to have strayed from the membrane bulk.
• ∗ systemtemp The temperature of the simulation during the main run
• ∗ outputdir Directory for output
• ∗ tabledir Directory where forcetables are kept
• ∗ ident a name for the simulation
• ∗ topofile the name of the file where the topology is written. Usually $ident.top
• ∗ tablenames A list of forcetable names to be used
• ∗ setbox l Box dimensions
• ∗ bonded parms A complete list of the bonded interactions required
• ∗ nb interactions A complete list of the non-bonded interactions required
• ∗ system specs A list of system specifications (see documentation for the setup_system command in 16.5)
• ∗ moltypes A list of molecule types (see documentation in 16.5)
• ∗ warm time step timestep to be used during warmup integration
• ∗ main time step timestep for the main integration run
• ∗ verlet skin skin used for verlet nesting list criterion
• ∗ langevin gamma langevin friction term
• ∗ int n times number of times to do main integration
• ∗ int steps number of steps in each main integration
197
• ∗ analysis write frequency How often to calculate the analysis
• ∗ write frequency How often to print out configurations
• vmdcommands a list of additional lines of commands to be written to the vmd_animation.script file
16.4. Analysis
The analysis package is designed to help organise the many possible analysis routines
that can be performed during a simulation. This documentation describes the basic user
interface commands and then all of the possible analysis routines. Instructions on how
to add a new analysis routine are given at the end of this section.
16.4.1. Basic commands
The following commands comprise the user interface to the analysis package.
At the start of a simulation all of the analysis that is to be performed is specified
using the setup_analysis command. Each time you want the analysis performed a call
to do_analysis should be made. One can then call print_averages to write results to
file.
::mbtools::analysis::setup_analysis : -outputdir.arg -suffix.arg
-iotype.arg -g.arg -str.arg
• commands [./] A tcl list where each element of the list is a string specifying the
name and complete argument list for a particular analysis to be carried out.
• outputdir [./] The directory where analysis output files will be created
• suffix [tmp] Suffix that will be appended to standard file names for analysis output
• iotype [a] The iotype that will be used when opening files for analysis. For an
explanation of the different iotypes see the documentation for the standard tcl
command open
• g [8] Number of grid points per side with which to divide the bilayer for height
profile analyses
• str [4.0] Distance of a tail bead from bilayer midplane beyond which a lipid is
deemed to be a stray lipid.
Sets up the analysis package for a simulation run or analysis run that is about to be
performed. This routine needs to be called before any analysis can be performed.
::mbtools::analysis::do_analysis :
198
Calls all of the analyze routines corresponding to commands setup in setup_analysis.
do_analysis should be called only after setup_analysis has already been called.
::mbtools::analysis::reset_averages
:
Calls all of the resetav routines corresponding to commands setup in setup_analysis.
These routines vary from command to command but they typically reset the storage
and counter variables used for analysis results. reset_averages is typically only called
internally by print_averages
::mbtools::analysis::print_averages :
Calls all of the printav routines corresponding to commands setup in setup_analysis.
These routines typically print results to a file buffer. After printing the reset_averages
routine is called internally. print_averages should be called only after setup_analysis
has already been called.
16.4.2. Available analysis routines
boxl
:
-verbose
: output ||
time_vs_boxl
Simply obtains the box dimensions from ESPResSo.
clusters : -alipid.arg -verbose : output || time_vs_clust,
sizehisto.[format %05d $time]
• alipid [1.29] Value for the area per lipid to be used in a making a rough calculation
of the area of clusters
Calls the espresso command analyze aggregation which groups molecules in the system into aggregates. Output to time_vs_clust is: maximum cluster size, minimum
cluster size, average size of clusters including those of size 2 or greater, standard deviation of clusters including those of size 2 or greater, number of clusters of size 2 or
greater, total average cluster size, total cluster size standard deviation, total number
of clusters, length of the interface between clusters, standard deviation of the interface
length, number of clusters for which length was calculate.
Additionally, at each call of print_averages the complete size histogram is printed
to a file with the formatted name sizehisto.[format %05d $time].
density_profile
: -nbins.arg -hrange.arg -beadtypes.arg
-colloidmoltypes.arg -r.arg -nogrid
-verbose : output || av_zprof
• nbins [100] Number of slices into which the height range is divided for the purpose
of calculating densities
199
• hrange [6] The maximum vertical distance from the bilayer midplane for which to
calculate densities. Note that the complete vertical range is therefore 2*varhrange
• beadtypes [0] A tcl list of the bead types for which to calculate a density profile
• colloidmoltypes [] A tcl list of molecule types identifying the molecules which are
colloids in the system. The default value is a null list
• r [0] A tcl list of sphere radii corresponding to the radii for each colloid type in the
system. If this is non-zero the density profile will be calculated in spherical shells
about the colloids in the system identified via colloidmoltypes or if colloidmoltypes
is not set then the system center of mass is assumed for the colloid/vesicle center
• nogrid If this is set a grid mesh will not be used to refine the density profile
calculation by taking into account vertical differences between mesh points
Calculates the number density of each of the beadtypes given in beadtypes as a function
of the vertical distance from the bilayer midplane. Lipids are also sorted according to
their orientation and assigned to upper or lower leaflets accordingly. Thus for a system
with 3 beadtypes we would obtain 6 columns of output corresponding to 0 (lower) 1
(lower) 2 (lower) 2 (upper) 1 (upper) 0 (upper) where the number refers to the bead
type and upper or lower refers to the bilayer leaflet.
energy : -verbose : output || time_vs_energy
Obtains the internal energies of the system from the analyze energy command of
ESPResSo.
flipflop : -verbose : output || time_vs_flip
Makes a call to the analyze get_lipid_orients command of ESPResSo and compares
this with a reference set of lipid orients obtained at the start of the simulation with
setup_analysis. Based on this comparison the number of lipids which have flipped
from their original positions is calculated
fluctuations : -verbose : output || powav.dat
Routine for calculating the power spectrum of height and thickness fluctuations for a flat
bilayer sheet. Uses the modes_2d routine in ESPResSo to calculate the height and thickness functions and perform the fft. See the documentation in the file fluctuations.tcl
for detail on what is calculated and how to obtain a stiffness value from the resulting
output. Note that this routine causes a crash if it detects a large hole in the bilayer.
localheights : -range.arg -nbins.arg -rcatch.arg -verbose :
output || av_localh
• range [1.0] Range of local height deviations over which to bin
200
• nbins [100] Number of slices to divide up the height range into for the purposes of
creating a profile
• rcatch [1.9] The distance about a single lipid to use a starting value for finding the
6 closest neighbours
For each lipid we calculate its 6 nearest neighbours and then calculate the height difference between the central lipid and these neighbours. Taking these 6 values for each lipid
we then create a histogram of number densities as a function of the height difference.
localorients : -range.arg -nbins.arg -verbose : output || av_localo
• range [1.0] Range of orientation deviations to consider
• nbins [100] Number of bins to use for histogram
Calculates the projection of the lipid orientation vector onto the xy plane for each lipid
and then bins the absolute values of these vectors.
orient_order : -verbose : output || time_vs_oop
Calculates the orientational order parameter S for each lipid through a call to the
espresso command analyze lipid_orient_order.
stress_tensor : -verbose : output || time_vs_stress_tensor
Calculates all 9 elements of the pressure tensor for the system through a call to the
espresso command analyze stress_tensor
pressure : -verbose : output || time_vs_pressure
Calculates the isotropic pressure through a call to analyze pressure. Results are
printed as a list of the various contributions in the following order: p inst, total , ideal ,
FENE , harmonic, nonbonded . Where p inst is the instantaneous pressure obtained
directly from the barostat.
stray : -verbose : output || time_vs_stray
Calculates the number of stray lipids based on a call to analyze get_lipid_orients.
16.4.3. Adding a new routine
To add a new analysis routine you should create a new file called myanalysis.tcl which
will contain all of your code. At the top of this file you should declare a namespace for
your analysis code and include all of the internal variables inside that namespace as
follows;
201
namespace eval ::mbtools::analysis::myanalysis {
variable av_myresult
variable av_myresult_i
variable f_tvsresult
variable verbose
namespace
namespace
namespace
namespace
export
export
export
export
setup_myanalysis
analyze_myanalysis
printav_myanalysis
resetav_myanalisis
}
Import your new file into the analysis package by adding a line like the following to
the analysis.tcl file.
source [file join [file dirname [info script]] myanalysis.tcl]
You then need to implement the following essential functions within your new namespace.
• ::mbtools::analysis::myanalysis::setup_myanalysis { args }
Typically you would use this function to initialise variables and open files.
Called by ::mbtools::analysis::setup_analysis. Arguments are allowed.
• ::mbtools::analysis::myanalysis::printav_myanalysis { void }
This function should print results to a file.
Called by ::mbtools::analysis::print_averages. Arguments are not allowed.
• ::mbtools::analysis::myanalysis::analyze_myanalysis { void }
This function performs the actual analysis and should update the storage and
averaging variables. Called by ::mbtools::analysis::do_analysis. Arguments
are not allowed.
• ::mbtools::analysis::myanalysis::resetav_myanalysis { void }
This function should update averages and reset variables accordingly depending
on your requirements.
Called by ::mbtools::analysis::reset_averages. Arguments are not allowed.
If any of these functions is not implemented the program will probably crash.
16.5. System generation
Package for setting up lipid membrane systems in a variety of geometrical shapes.
202
16.5.1. Basic commands
::mbtools::system_generation::setup_system
[iboxl] [moltypes]
:
[system_specs]
• system_specs This is a list of structures called system specifications. Each such
system specification in turn should be a list consisting of a geometry and a list
detailing the number of each molecule type i.e.
set system_spec { geometry n_molslist }
The geometry should be specified as a list with two elements. The first element
should be a string “geometry” identifying this list as a geometry. The second
element is a string containing the name of a geometry type mygeometry followed
by arguments to be passed to the routine create_mygeometry.
The n molslist should be specified as a list with two elements. The first element
should be a string “n molslist” identifying this list as an n molslist. The second
element is a list each element of which specifies a molecule type and the number
of such molecules.
• boxl A list containing the lengths of each of the box side lengths.
• moltypes A list, each element of which specifies a molecule type and type information. The exact format and requirements of this list are detailed for each molecule
separately (see below for a list of molecule types and their requirements) however
regardless of mol type the first two elements of the list must be a moltypeid and a
string specifying the moltype respectively.
Sets up the system including generating topologies and placing molecules into specified
geometries. Each geometry and list of molecules to be placed into that geometry are
grouped into a system spec.
Example:
The following code sets out the molecule types to be used in the simulation by setting
a list called moltypes. In this case two different lipid types are setup and assigned to
moltypeids 0 and 1 respectively. Moltype 0 will consist of three beads per lipid, the
first of which is of atomtype 0 and the second and third of which are of atomtype 1.
Bonds in the lipid will be of type 0 and 1 respectively.(see the ::mbtools::system_generation::place_lipid_linear function for further details).
set moltypes [list { 0 lipid { 0 1 1 } { 0 1 } }
{ 1 lipid { 0 2 2 2 } { 0 2 } } ]
We then construct system specs for a flat bilayer and a spherical bilayer and group
these into a system specs list.
First the spherical system specs
203
set geometry { geometry "sphere -shuffle -c { 0.0 0.0 15.0 } " }
set n_molslist { n_molslist { { 0 1000 } } }
lappend spherespec $geometry
lappend spherespec $n_molslist
The flat system spec
set geometry { geometry "flat -fixz" }
set n_molslist { n_molslist { { 1 3000 } } }
lappend bilayerspec $geometry
lappend bilayerspec $n_molslist
Now group together the systems pecs into a master list
lappend system_specs $spherespec
lappend system_specs $bilayerspec
Make the call to setup_system
::mbtools::system_generation::setup_system $system_specs
[setmd box_l] $moltypes
::mbtools::system_generation::get_trappedmols :
returns the internal list variable trappedmols which keeps track of all molecules that have
been trapped by their center of mass. This function should be called after setup and
would then typically be passed to the function ::mbtools::utils:trap_mols.
::mbtools::system_generation::get_userfixedparts :
returns the internal list variable userfixedparts which keeps track of all particles that
have been fixed in position during the setup. This is useful for later releasing particles
after warmup routines have been completed.
::mbtools::system_generation::get_middlebead :
returns the internal variable middlebead .
16.5.2. Available geometries
flat
:
-fixz -bondl.arg -crystal -half -pancake -shuffle
• fixz Fix the vertical positions of all particles. The ids of these particles are
added to the list of userfixedparts which can later be obtained through a call
to ::mbtools::system_generation::get_userfixedparts.
• crystal Sets lipids on a grid, instead of randomly.
204
• half Creates a halfbilayer (i.e. periodic only along one direction). Useful to measure a line tension.
• pancake Creates a spherical and flat bilayer. The diameter of the pancake cannot
exceed the box l.
• shuffle shuffle the topology prior to placing the lipids. This is required for a random
lipid distribution because otherwise the lipids will be placed on the sphere in the
order they appear in the topology
Creates a flat bilayer in the XY plane by random placement of lipids.
sphere : -c.arg -initarea.arg -bondl.arg -shuffle
• c [{0.0 0.0 0.0}] The location of the center of the sphere relative to the center
of the box
• initarea [1.29] An initial guess for the area per lipid. This guess is used to compute
initial sphere dimensions based on the number of lipids. This initial guess is then
iteratively refined until all lipids can be fit uniformly on the sphere.
• shuffle shuffle the topology prior to placing the lipids. This is required for a random
lipid distribution because otherwise the lipids will be placed on the sphere in the
order they appear in the topology
Creates a spherical vesicle by placing molecules in an ordered manner at uniform density
on the surface of the sphere. Molecules are assumed to have a uniform cross sectional
area and closely matched (though not identical) lengths. The radius of the vesicle will
depend on the number of lipids and the area per lipid.
sphere_cap : -r.arg -half -c.arg -initarea.arg -bondl.arg -shuffle
• r [10.0] The radius of the whole sphere where the cap is shaped
• half Create a half of sphere with the amount of molecules available
• c [{0.0 0.0 0.0}] The location of the center of the sphere relative to the center
of the box
• initarea [1.29] An initial guess for the area per lipid. This guess is used to compute
initial sphere dimensions based on the number of lipids. This initial guess is then
iteratively refined until all lipids can be fit uniformly on the sphere.
• shuffle shuffle the topology prior to placing the lipids. This is required for a random
lipid distribution because otherwise the lipids will be placed on the sphere in the
order they appear in the topology
205
Creates a spherical cap which is part of a vesicle of a radius r , by placing molecules in an
ordered manner at uniform density on the surface of the sphere. Molecules are assumed
to have a uniform cross sectional area and closely matched (though not identical) lengths.
If the option half is defined, the radius of the vesicle will depend on the number of lipids
and the area per lipid.
torus : -c.arg -initarea.arg -ratio.arg -bondl.arg -shuffle
• c [{0.0 0.0 0.0}] The location of the center of the torus relative to the center of
the box.
• initarea [1.29] An initial guess for the area per lipid. This guess is used to compute
initial radii based on the number of lipids. This initial guess is then iteratively
refined until all lipids can be fit uniformly on the torus.
• ratio [1.4142] Ratio of major toroidal radius to minor toroidal radius. Default
value is for the Clifford torus.
• shuffle shuffle the topology prior to placing the lipids. This is required for a random
lipid distribution because otherwise the lipids will be placed on the torus in the
order they appear in the topology.
Creates a toroidal vesicle by placing molecules in an ordered manner at uniform density
on the surface of the torus. Molecules are assumed to have a uniform cross sectional
area and closely matched (though not identical) lengths. The two radii of the torus will
depend on the number of lipids, the area per lipid and the ratio between radii.
cylinder : -c.arg -initarea.arg -bondl.arg -shuffle
• c [0.0 0.0 0.0]
• initarea [1.29]
• shuffle shuffle the topology prior to placing the lipids.
Creates a cylinder which spans the box along one dimension by placing molecules uniformly on its surface. Works in a similar way to the sphere routine.
random : -exclude.arg -inside.arg -shuffle -bondl.arg
• exclude.arg [] an exclusion zone definition suitable for passing to
::mbtools::utils::isoutside.
• inside.arg [] an inclusion zone definition suitable for passing to
::mbtools::utils::isoutside.
• shuffle shuffle the topology prior to placing the lipids.
206
Places molecules randomly in space with a (sortof) random orientation vector. If an
exclusion zone is defined, then no molecules will be placed such that their positions are
within the zone. If an inclusion zone if defined, then no molecules will be place outside
this zone. For instance,
set geometry { geometry "random -exclude { sphere { 0.0 0.0 0.0 } 4.0 }
-inside { cuboid { 0.0 0.0 0.0 } { 15.0 15.0 15.0 } }" }
will randomly place molecules within the volume between a sphere with a radius of 4.0
and a cuboid with dimension 15.0 × 15.0 × 15.0 at the origin.
readfile : -ignore.arg -f.arg -t.arg
• ignore.arg [] particle properties to be ignored during the file read.
• f .arg [] The file containing the configuration to be used for setup. Must be an
espresso blockfile with box length, particle and bonding information.
• t.arg [] The topology file corresponding to the file to be read.
• tol .arg [0.000001] Tolerance for comparison of box dimensions.
Use particle positions contained in a file to initialise the locations of particles for a
particular geometry. The box dimensions in the file and those set by the user are
compared and an error is returned if they are not the same to within a tolerance value
of tol . Even though we read from a file we also generate a topology from the nm olslist
and this topology is compared with the topology that is read in to check if the number
of particles are the same.
singlemol : -c.arg -o.arg -trapflag.arg -ctrap.arg
-trapspring.arg -bondl.arg
• c.arg [ 0.0 0.0 0.0 ] The molecule center. Exactly what this means depends on
the molecule type.
• o.arg [ 0.0 0.0 1.0 ] The orientation vector for the molecule. This is also molecule
type dependent
• trapflag.arg [ 0 0 0 ] Set this optional argument to cause a molecule to be trapped
by its center of mass. You should give three integers corresponding to each of the
three coordinate axes. If a value of 1 is given then motion in that axis is trapped.
• ctrap.arg [ ”” ] Set this optional argument to the central point of the trap. This
works much like an optical trap in that molecules will be attracted to this point
via a simple harmonic spring force
• trapspring.arg [ 20 ] The spring constant for the trap potential (harmonic spring).
Simply place a single molecule at the desired position with the desired orientation.
207
16.5.3. Adding a new geometry
To create a routine for setting up a system with a new type of geometry mygeom. Start
by creating a new file mygeom.tcl inside the system_generation directory. The new file
should declare a new namespace mygeom as a sub namespace of ::mbtools::system_generation and export the proceedure create_mygeom. Thus your mygeom.tcl file
should begin with the lines
namespace eval ::mbtools::system_generation::mygeom {
namespace export create_mygeom
}
Import your new file into the system_generation package by adding a line like the
following to the system_generation.tcl file
source [file join [file dirname [info script]] mygeom.tcl]
You then need to implement the create_mygeom proceedure within your new namespace as follows
::mbtools::system_generation::mygeom::create_mygeom
args
16.5.4. Available molecule types
lipid
: typeinfo : { moltypeid "lipid" particletypelist
bondtypelist }
• particletypelist A list of the particle types for each atom in the lipid. The particles
are placed in the order in which they appear in this list.
• bondtypelist A list of two bondtypeid s. The first id is used for bonds between
consecutive beads in the lipid. The second bondtypeid defines the pseudo bending
potential which is a two body bond acting across beads separated by exactly one
bead.
Places atoms in a line to create a lipid molecule.
hollowsphere : typeinfo : { moltypeid "hollowsphere"
sphereparticlelist bondtype natomsfill }
• sphereparticlelist A list of the particle types for each atom in the hollowsphere.
The atoms that make up the outer shell must be listed first followed by the atoms
that make up the inner filling.
• bondtype The typeid for bonds linking atoms in the outer shell.
• natomsfill Number of filler atoms. The atom types for these will be obtained from
the last natomsfill in the sphereparticlelist.
208
Creates a sphere of beads arranged such that they have an approximate spacing of bondl
and such that they optimally cover the sphere. The optimal covering is obtained using
the icover routines which are copyright R. H. Hardin, N. J. A. Sloane and W. D. Smith,
1994, 2000. Thus the routine will only work if you have installed icover and if you can
successfully run it from the command line in the directory that you started your espresso
job. These routines are serious overkill so if anybody can think of a nice simple algorithm
for generating a covering of the sphere let us know.
protein : typeinfo : { moltypeid "protein" particletypelist
bondtypelist }
• particletypelist A list of the particle types for each atom in the protein.
• bondtypelist A list of bondtypeids.
Create a protein molecule.
spanlipid : typeinfo : { moltypeid "protein" particletypelist
bondtypelist }
• particletypelist A list of the particle types for each atom in the lipid. Since this
is a spanning lipid the first and last elements of this list would typically be head
beads.
• bondtypelist A list of two bondtypeid s with the same meaning as explained above
for standard lipids.
Create a lipid which spans across the bilayer.
16.5.5. Adding a new molecule type
To add a new molecule type you need to define a proceedure which determines how the
atoms that make up the molecule should be placed. This proc will live directly in the
::mbtools::system_generation namespace. Examples can be found in place.tcl.
In order to register your new molecule type to allow placement in any geometry you
need to add a call to it in the function ::mbtools::system_generation::placemol.
Make sure that all arguments to your place_mymolecule routine are included in this
function call.
16.6. Utils
Useful utilities routines for various types. Includes file management, basic geometry and
math procedures.
209
16.6.1. Setup commands
::mbtools::utils::setup_outputdir : [outputdir] -paramsfile.arg
-tabdir.arg -tabnames.arg -startf.arg -ntabs.arg
• outputdir Complete path of the directory to be setup. At least the parent of the
directory must exist
• paramfile [] Name of a file to be copied to the output directory
• tabdir [] Full path name of the directory where forcetables are kept
• tabnames [] Complete list of forcetables to be used in the simulation. These will
be copied to the output directory
This routine is designed to setup a directory for simulation output. It copies forcetables
and the parameter file to the directory after creating it if necessary.
::mbtools::utils::read_startfile : [file]
• file Complete path of the file to be read. Should be an espresso blockfile.
Read in particle configuration from an existing file or simulation snapshot
::mbtools::utils::read_checkpoint : [dir]
• dir Directory containing the checkpoint file which must be called checkpoint.latest.gz.
Read in a checkpoint and check for success. Warn if the checkpoint does not exist.
::mbtools::utils::read_topology : [file]
• file Complete path of the file that contains the topology information.
Read in the topology from a file and then execute the analyze set "topo_part_sync"
command of ESPResSo.
::mbtools::utils::set_topology : [topo]
• topo A valid topology.
Set the given topology and then execute the analyze set "topo_part_sync" command
of ESPResSo.
::mbtools::utils::set_bonded_interactions : [bonded_parms]
• bondedp arms A list of bonded interactions. Each element of this list should contain
all the appropriate arguments in their correct order for a particular call to the
espresso inter command. See the espresso inter command for a list of possible
bonded interactions and correct syntax.
210
Set all the bonded interactions.
::mbtools::utils::set_nb_interactions : [nb_parms]
• nb parms A list of interactions. Each element of this list should contain all the
appropriate arguments in their correct order for a particular call to the espresso
inter command. See the espresso inter command for a list of possible non-bonded
interactions and correct syntax.
Set all the bonded interactions.
::mbtools::utils::init_random : [n_procs]
• n procs The number of processors used in this job.
Initialize the random number generators on each processor based on the current time
with a fixed increment to the time seed used for each proc.
::mbtools::utils::initialize_vmd : [flag] [outputdir]
[ident] -extracommands.arg
• flag Depending on the value of this parameter initialize vmd to one of its possible
states:
– interactive : VMD is started and a connection to espresso established for
immediate viewing of the current espresso process. With some luck this might
even work sometimes! If VMD doesn’t get a proper connection to espresso
then it will crash.
– offline : Just constructs the appropriate psf and vmd_animation.script
files and writes them to the output directory so that pdb files generated with
writepdb can be viewed with vmd -e vmd_animation.script.
– default : Any value other than those above for flag will just result in vmd not
being initialized.
• outputdir The directory where vmd output will be written.
• ident A basename to be be given to vmd files.
• extracommands [] A list of strings each of which will be written to the end of the
vmd_animationscript. Use this to give additional commands to vmd.
Prepare for vmd output.
211
16.6.2. Warmup commands
::mbtools::utils::warmup : [steps] [times] -mindist.arg
-cfgs.arg -outputdir.arg -vmdflag.arg -startcap.arg
-capgoal.arg
• steps number of integration steps used in each call to integrate.
• times number of times to call the integrate function during warmup.
• mindist [0] Terminate the warmup when the minimum particle distance is greater
than this criterion. A value of 0 (default) results in this condition being ignored.
If a condition is imposed this routine can become very very slow for large systems.
• cfgs [-1] Write out a configuration file every cfgs calls to integrate.
• outputdir [./] The directory for writing output.
• vmdflag [offline] If this flag is set to ”offline” (default) pdb files will be generated
for each configuration file generated.
• startcap [5] Starting value for the forcecap.
• capgoal [1000] For the purposes of calculating a cap increment this value is used
as a goal. The final forcecap will have this value.
Perform a series of integration steps while increasing forcecaps from an initially small
value.
16.6.3. Topology procs
::mbtools::utils::maxpartid
:
[topo]
• topo A valid topology.
Find the maximum particle id in a given topology.
::mbtools::utils::maxmoltypeid : [topo]
• topo A valid topology.
Find the maximum molecule type id.
::mbtools::utils::listnmols : [topo]
• topo A valid topology.
Construct a list with the number of molecules of each molecule type.
::mbtools::utils::minpartid : [topo]
212
• topo A valid topology.
Minimum particle id for the given topology.
::mbtools::utils::minmoltype : [topo]
• topo A valid topology/
Minimum molecule type id for this topology.
::mbtools::utils::listmoltypes : [topo]
• topo A valid topology.
Make a list of all the molecule types in a topology. Makes a check for duplication which
would occur for an unsorted topology.
::mbtools::utils::listmollengths : [topo]
• topo A valid topology.
Works out the length (number of atoms) of each molecule type and returns a list of these
lengths.
16.6.4. Math procs
::mbtools::utils::dot_product
:
A B
Returns A dot B
::mbtools::utils::matrix_vec_multiply : A B
Return the product of a matrix A with a vector B
::mbtools::utils::calc_proportions : ilist
Calculate the number of times each integer occurs in the list ilist.
::mbtools::utils::average : data from to
• data A list of numbers to be averaged
• from Optional starting index in data
• to Optional ending index in data
Calculate the mean of a list of numbers starting from from going up to to.
::mbtools::utils::stdev : data from to
213
• data A list of numbers to find the std deviation of
• from Optional starting index in data
• to Optional ending index in data
Calculate the standard deviation of a list of numbers starting from from going up to to.
::mbtools::utils::acorr : data
• data Data for which an autocorrelation is to be calculated
Calculate an autocorrelation function on a set of data.
::mbtools::utils::distance : pos1 pos2
• pos1 A position vector
• pos2 A position vector
Calculate the distance between two points whose position vectors are given.
::mbtools::utils::distance_min : pos1 pos2
• pos1 A position vector
• pos2 A position vector
Calculate the minimum image distance between two position vectors.
::mbtools::utils::min_vec : pos1 pos2
• pos1 A position vector
• pos2 A position vector
Calculate the minimum image vector from position vector2 to postition 1, i.e. pos1 pos2.
::mbtools::utils::normalize : vec
• vec The vector to be normalised
Normalize a vector
::mbtools::utils::scalevec : vec scale
• vec The vector to be scaled
• scale Scaling factor
Multiply all elements of a vector by a scaling factor
::mbtools::utils::uniquelist : original
• original A list possibly containing duplicate elements
Construct a list of all the unique elements in the original list removing all duplication.
214
16.6.5. Miscellaneous procs
::mbtools::utils::trap_mols
:
molstotrap
• molstotrap A list of trap values for molecules. This list would typically be obtained
by calling ::mbtools::get_trappedmols immediately after the system has been
setup.
Set the trap value for a list of molecules.
::mbtools::utils::isoutside : [pos] [zone]
• pos The point whose status is to be determined
• zone This will be a tcl list. The first element of the list must be a string with the
name of the zone type and subsequent elements will be further information about
the zone. Available zones are:
– sphere : center radius
– cuboid : center {L W H}
Determines whether the point at pos is outside the zone. Parameter center should be a
tcl list. Returns 1 if it is and 0 if it is not.
::mbtools::utils::calc_com : mol
• mol The molecule
Calculate the center of mass of a molecule.
::mbtools::utils::centersofmass_bymoltype : [moltypes]
• moltypes A list of molecule type ids
Determine the center of mass of every molecule whose type matches an item in the list
moltypes. Returns a nested list where each element in the list is itself a list of centers
of mass for a given moltype.
16.7. mmsg
mmsg is designed to provide a more controlled way of printing messages than the simple
puts commands of Tcl. It has an ability to turn on or off messages from particular
namespaces.
215
16.7.1. Basic commands
The following commands represent the standard interface for the mmsg package. For
consistency one should use these instead of a bare puts to standard out. mbtools makes
extensive use of these commands.
::mmsg::send : [namespace] [string] { [newline] }
• namespace A namespace. Typically this should be the current namespace which
one can get via namespace current
• string The message you want printed
• newline [yes] Set this to anything other than ”yes” and no carriage return will be
used after the message
The mmsg equivalent of puts. Designed for printing of simple status or progress messages.
::mmsg::err : [namespace] [string] { [newline] }
• namespace A namespace. Typically this should be the current namespace which
one can get via namespace current
• string The message you want printed
• newline [yes] Set this to anything other than ”yes” and no carriage return will be
used after the message
Prints error messages and causes program to exit.
::mmsg::warn : [namespace] [string] { [newline] }
• namespace A namespace. Typically this should be the current namespace which
one can get via namespace current
• string The message you want printed
• newline [yes] Set this to anything other than ”yes” and no carriage return will be
used after the message
Prints warning messages.
::mmsg::debug : [namespace] [string] { [newline] }
• namespace A namespace. Typically this should be the current namespace which
one can get via namespace current
• string The message you want printed
• newline [yes] Set this to anything other than ”yes” and no carriage return will be
used after the message
Prints debug messages.
216
16.7.2. Control commands
mmsg does several checks before it decides to print a message. For any given message
type it checks if that message type is allowed. It also checks to see if the namespace
given as an argument is in the allowable namespaces list. The default behaviour is to
print from the main mbtools namespaces and the global namespace
{ :: ::mbtools::system_generation ::mbtools::utils ::mbtools::analysis }
Note that children of these namespaces must be explicitly enabled. All message types
except debug are also enabled by default. The following commands allow this default
behaviour to be changed.
::mmsg::setnamespaces : namespacelist
• namespacelist A list of all namespaces from which messages are to be printed
Allows control over which namespaces messages can be printed from.
::mmsg::enable : type
• type A string indicating a single message type to enable. Allowable values are
”err”, ”debug”, ”send” and ”warn”
Allows particular message types to be enabled: For example one could enable debug
output with
mmsg::enable "debug"
::mmsg::disable : type
• type A string indicating a single message type to disable. Allowable values are
”err”, ”debug”, ”send” and ”warn”
Allows particular message types to be disabled: For example one could disable warning
output with
mmsg::enable "warn"
217
17. Under the hood
• Implementation issues that are interesting for the user
• Main loop in pseudo code (for comparison)
17.1. Internal particle organization
Since basically all major parts of the main MD integration have to access the particle
data, efficient access to the particle data is crucial for a fast MD code. Therefore the
particle data needs some more elaborate organisation, which will be presented here. A
particle itself is represented by a structure (Particle) consisting of several substructures
(e. g. ParticlePosition, ParticleForce or ParticleProperties), which in turn represent
basic physical properties such as position, force or charge. The particles are organised
in one or more particle lists on each node, called Cell cells. The cells are arranged by
several possible systems, the cellsystems as described above. A cell system defines a way
the particles are stored in ESPResSo, i. e. how they are distributed onto the processor
nodes and how they are organised on each of them. Moreover a cell system also defines
procedures to efficiently calculate the force, energy and pressure for the short ranged
interactions, since these can be heavily optimised depending on the cell system. For
example, the domain decomposition cellsystem allows an order N interactions evaluation.
Technically, a cell is organised as a dynamically growing array, not as a list. This
ensures that the data of all particles in a cell is stored contiguously in the memory.
The particle data is accessed transparently through a set of methods common to all cell
systems, which allocate the cells, add new particles, retrieve particle information and
are responsible for communicating the particle data between the nodes. Therefore most
portions of the code can access the particle data safely without direct knowledge of the
currently used cell system. Only the force, energy and pressure loops are implemented
separately for each cell model as explained above.
The domain decomposition or link cell algorithm is implemented in ESPResSo such
that the cells equal the ESPResSo cells, i. e. each cell is a separate particle list. For an
example let us assume that the simulation box has size 20 × 20 × 20 and that we assign 2
processors to the simulation. Then each processor is responsible for the particles inside
a 10 × 20 × 20 box. If the maximal interaction range is 1.2, the minimal possible cell
size is 1.25 for 8 cells along the first coordinate, allowing for a small skin of 0.05. If one
chooses only 6 boxes in the first coordinate, the skin depth increases to 0.467. In this
example we assume that the number of cells in the first coordinate was chosen to be 6
and that the cells are cubic. ESPResSo would then organise the cells on each node in
a 6 × 12 × 12 cell grid embedded at the centre of a 8 × 14 × 14 grid. The additional
218
cells around the cells containing the particles represent the ghost shell in which the
information of the ghost particles from the neighbouring nodes is stored. Therefore the
particle information stored on each node resides in 1568 particle lists of which 864 cells
contain particles assigned to the node, the rest contain information of particles from
other nodes.a
Classically, the link cell algorithm is implemented differently. Instead of having separate particle lists for each cell, there is only one particle list per node, and a the cells
actually only contain pointers into this particle list. This has the advantage that when
particles are moved from one cell to another on the same processor, only the pointers have
to be updated, which is much less data (4 rsp. 8 bytes) than the full particle structure
(around 192 bytes, depending on the features compiled in). The data storage scheme
of ESPResSo however requires to always move the full particle data. Nevertheless, from
our experience, the second approach is 2-3 times faster than the classical one.
To understand this, one has to know a little bit about the architecture of modern
computers. Most modern processors have a clock frequency above 1GHz and are able
to execute nearly one instruction per clock tick. In contrast to this, the memory runs
at a clock speed around 200MHz. Modern double data rate (DDR) RAM transfers up
to 3.2GB/s at this clock speed (at each edge of the clock signal 8 bytes are transferred).
But in addition to the data transfer speed, DDR RAM has some latency for fetching the
data, which can be up to 50ns in the worst case. Memory is organised internally in pages
or rows of typically 8KB size. The full 2 × 200 MHz data rate can only be achieved if
the access is within the same memory page (page hit), otherwise some latency has to be
added (page miss). The actual latency depends on some other aspects of the memory
organisation which will not be discussed here, but the penalty is at least 10ns, resulting in
an effective memory transfer rate of only 800MB/s. To remedy this, modern processors
have a small amount of low latency memory directly attached to the processor, the cache.
The processor cache is organised in different levels. The level 1 (L1) cache is built
directly into the processor core, has no latency and delivers the data immediately on
demand, but has only a small size of around 128KB. This is important since modern
processors can issue several simple operations such as additions simultaneously. The
L2 cache is larger, typically around 1MB, but is located outside the processor core and
delivers data at the processor clock rate or some fraction of it.
In a typical implementation of the link cell scheme the order of the particles is fairly
random, determined e. g. by the order in which the particles are set up or have been
communicated across the processor boundaries. The force loop therefore accesses the
particle array in arbitrary order, resulting in a lot of unfavourable page misses. In the
memory organisation of ESPResSo, the particles are accessed in a virtually linear order.
Because the force calculation goes through the cells in a linear fashion, all accesses to a
single cell occur close in time, for the force calculation of the cell itself as well as for its
neighbours. Using the domain decomposition cell scheme, two cell layers have to be kept
in the processor cache. For 10000 particles and a typical cell grid size of 20, these two
cell layers consume roughly 200 KBytes, which nearly fits into the L2 cache. Therefore
every cell has to be read from the main memory only once per force calculation.
219
18. Getting involved
Up to date information about the development of ESPResSo can be found at the web
page http://espressomd.org As the important information can change in time, we
will not describe its contents in detail but rather request the reader to go directly to the
URL. Among other things, one can find information about the following topics there:
• FAQ
• Latest stable release of ESPResSo and older releases
• Obtaining development version of ESPResSo
• Archives of both developers’ and users’ mailing lists
• Registering to ESPResSo mailing lists
• Submitting a bug report
18.1. Community support and mailing lists
If you have any questions concerning ESPResSo which you cannot resolve by yourself,
you may post a message to the mailing list. Instructions on how to register to the mailing
lists and post messages can be found on the homepage http://espressomd.org. Before
posting a question and waiting for someone to answer, it may be useful to search the
mailing list archives or FAQ and see if you can get the answer immediately. For several
reasons it is recommended to send all questions to the mailing lists rather than to contact
individual developers:
• All registered users get your message and you have a higher probability that it is
answered soon.
• Your question and the answers are archived and the archives can be searched by
others.
• The answer may be useful also to other registered users.
• There may not be a unique answer to your problem and it may be useful to get
suggestions from different people.
Please remember that this is a community mailing list. It is other users and developers
who are answering your questions. They do it in their free time and are not paid for
doing it.
220
18.2. Contributing your own code
If you are planning to make an extension to ESPResSo or already have a piece of your
own code which could be useful to others, you are very welcome to contribute it to
the community. Before you start making any changes to the code, you should obtain
the current development version of it. For more information about how to obtain the
development version, refer to the homepage http://espressomd.org.
It is also generally a good idea to contact the mailing lists before you start major
coding projects. It might be that someone else is already working on the problem or has
a solution at hand.
18.3. Developers’ guide
Besides the User guide, ESPResSo also contains a Developers’ guide which is a programmer documentation automatically built from comments in the source code and using
Doxygen. It provides a cross-referenced documentation of all functions and data structures available in ESPResSo source code. It can be built by typing
make dg
in the build directory. Afterwards it can be found in the subdirectory of the build
directory: doc/dg/html/index.html.
A recent version of this guide can also be found on the ESPResSo homepage http:
//espressomd.org.
18.4. User’s guide
If, while reading this user guide, you notice any mistakes or badly (if at all) described
features or commands, you are very welcome to contribute to the guide and have others
benefit from your knowledge.
For this, you should also checkout the development version as described on the homepage. As the user guide, like all ESPResSo code, is always in flow and changes are made
regularly, there are already many paragraphs marked with a “todo” box. To turn on
these boxes, edit the main file doc/ug/ug.tex and adapt the inclusion of the LATEX
package todonotes.
You can then build the user guide by typing
make ug
221
A. ESPResSo quick reference
part pid [pos x y z ] [type typeid ] [v vx vy vz ] [f fx fy fz ]
29
[bond bondid pid2 ...] [q charge] 1 [quat q1 q2 q3 q4 ] 2
[omega_body/lab x y z ] 2 [torque_body/lab x y z ] 2
[rinertia x y z ] 2 [[un]fix x y z ] 3 [ext_force x y z ] 3
[ext_torque x y z ] 2,3 [exclude pid2 ...] 4 [exclude delete pid2 ...] 4
[mass mass] 5 [dipm moment] 6 [dip dx dy dz ] 6 [virtual v ] 7,8
[vs_relative pid distance] 8 [vs_auto_relate_to pid ] 8 [temp T ] 9
[gamma g] 9 [rotation rot] 10 [solvation lA kA lB kB ] 11
[swimming ( ( v_swim v swim | f_swim f swim ) | off )] 12 [swimming
( ( v_swim v swim | f_swim f swim ) ( pusher | puller ) dipole_length dipole length rotational_friction rotational friction |
off )] 12,13
Required features: 1 ELECTROSTATICS 2 ROTATION 3 EXTERNAL_FORCES 4 EXCLUSION
5 MASS 6 DIPOLES 7 VIRTUAL_SITES_COM 8 VIRTUAL_SITES_RELATIVE
9 LANGEVIN_PER_PARTICLE 10 ROTATION_PER_PARTICLE 11 SHANCHEN
12 ENGINE 13 LB or LB_GPU
part pid print [( id | pos | type | folded_position | type | q | v | f
| torque_body | torque_lab | body_frame_velocity | fix | ext_force | ext_torque | bond | exclusions connections [range] |
swimming )]...
part
33
part pid delete
part deleteall
34
part auto_exclusions [range]
part delete_exclusions
34
Required features:
EXCLUSIONS
polymer num polymers monomers per chain bond length
[start pid ] [pos x y z ] [mode ( RW | SAW | PSAW ) [shield [trymax ]]]
[charge q] 1 [distance dcharged ] 1 [types typeidneutral [typeidcharged ]]
[bond bondid ] [angle φ [θ [x y z ]]] [constraints] 2
Required features:
1 ELECTROSTATICS
2 CONSTRAINTS
counterions N [start pid ] [mode ( SAW | RW ) [shield [trymax ]]]
[charge val ] 1 [type typeid ]
Required features:
222
1 ELECTROSTATICS
35
36
salt N+ N− [start pid ] [mode ( SAW | RW ) [shield [trymax ]]]
[charges val+ [val− ]] 1 [types typeid+ [typeid− ]] [rad r ]
Required features:
1 ELECTROSTATICS
diamond a bond length monomers per chain [counterions NCI ]
[charges valnode valmonomer valCI ] 1 [distance dcharged ] 1 [nonet]
Required features:
37
1 ELECTROSTATICS
icosaeder a monomers per chain [counterions NCI ]
[charges valmonomers valCI ] 1 [distance dcharged ] 1
Required features:
37
39
1 ELECTROSTATICS
crosslink num polymer monomers per chain [start pid ] [catch rcatch ]
[distLink link dist] [distChain chain dist] [FENE bondid ]
[trials trymax ]
40
copy_particles [set id1 id2 ...| range from to ...] [shift s x s y s z ]
40
223
42
constraint wall normal nx ny nz dist d type id [penetrable flag]
[reflecting flag] [only_positive flag] [tunable_slip flag]
constraint sphere center cx cy cz radius rad direction direction type
id [penetrable flag] [reflecting flag]
constraint cylinder center cx cy cz axis nx ny nz radius rad
length length direction direction type id [penetrable flag]
[reflecting flag]
constraint rhomboid corner px py pz a ax ay az b bx by bz
c cx cy cz direction direction type id [penetrable flag]
[reflecting flag]
constraint maze nsphere n dim d sphrad rs cylrad rc type id
[penetrable flag]
constraint pore center cx cy cz axis nx ny nz radius rad length length
type id
constraint stomatocyte center x y z orientation ox oy oz outer_radius Ro inner_radius Ri layer_width w direction direction
type id [penetrable flag] [reflecting flag]
constraint slitpore pore_mouth z channel_width c pore_width w
pore_length l upper_smoothing_radius us lower_smoothing_radius ls
constraint rod center cx cy lambda lambda 1
constraint plate height h sigma sigma 1
constraint ext_magn_field fx fy fz 2,3
constraint plane cell x y z type id
constraint mindist_position x y z
constraint hollow_cone center x y z orientation ox oy oz outer_radius Ro inner_radius Ri width w opening_angle alpha
direction direction type id [penetrable flag] [reflecting flag]
constraint spherocylinder center cx cy cz axis nx ny nz radius
rad length length direction direction type id [penetrable flag]
[reflecting flag]
constraint mindist_position x y z
Required features:
CONSTRAINTS
1 ELECTROSTATICS
2 ROTATION
3 DIPOLES
constraint delete [num]
45
constraint force n
45
constraint [num]
46
harmonic_force { x y z } k
46
Required features:
CUDA
part gc ( type | ( ( find | delete | status | number ) type ) )
Required features:
inter
224
49
GRANDCANONICAL
50
inter type1 type2 tabulated filename
Required features:
50
TABULATED
inter type1 type2 lennard-jones σ rcut [( cshift |auto ) [roff [rcap [ rmin ]]]] 51
Required features:
LENNARD_JONES
inter type1 type2 lj-gen σ rcut cshift roff e1 e2 b1 b2 [( rcap |auto ) λ δ] 51
Required features:
LENNARD_JONES_GENERIC
inter type1 type2 lj-cos σ rcut roff
inter type1 type2 lj-cos2 σ roff ω
Required features:
LJCOS
LJCOS2
inter type1 type2 smooth-step σ1 n k0 σ2 rcut
Required features:
ROTATION
inter
inter
inter
inter
inter
inter
bondid
bondid
bondid
bondid
bondid
bondid
velocity
57
GAY_BERNE
inter type1 type2 affinity α1 α2
Required features:
56
LJ_ANGLE
inter type1 type2 gay-berne 0 σ0 rcutoff k1 k2 µ ν
Required features:
56
GAUSSIAN
inter type1 type2 lj-angle σ rcut b1a b1b b2a b2b [rcap z0 δz κ 0 ]
Required features:
56
HERTZIAN
inter type1 type2 gaussian σ rcut
Required features:
55
HAT
inter type1 type2 hertzian σ Required features:
55
SOFT_SPHERE
inter type1 type2 hat Fmax rc
Required features:
55
BUCKINGHAM
inter type1 type2 soft-sphere a n rcut roffset
Required features:
54
MORSE
inter type1 type2 buckingham A B C D rcut rdiscont shift
Required features:
54
BMHTF_NACL
inter type1 type2 morse α rmin rcut
Required features:
53
SMOOTH_STEP
inter type1 type2 bmhtf-nacl A B C D σ rcut
Required features:
52
58
SHANCHEN
fene K ∆rmax [r0 ]
harmonic K R [rcut ]
quartic K0 K1 R [rcut ]
bonded_coulomb α
subt_lj reserved R
rigid_bond constrained bond distance positional tolerance
tolerance
59
60
60
60
61
61
225
inter
inter
inter
inter
inter
inter
inter
inter
inter
inter
inter
inter
bondid
bondid
bondid
bondid
bondid
bondid
bondid
bondid
bondid
bondid
bondid
bondid
tabulated bond filename
tabulated angle filename
tabulated dihedral filename
virtual_bond
stretching_force L0AB ks
bending_force θ0 kb
0
area_force_local SABC
kal
0
area_force_global S kag
volume_force V 0 kv
angle_harmonic K [φ0 ]
angle_cosine K [φ0 ]
angle_cossquare K [φ0 ]
Required features:
inter
inter
inter
inter
inter
226
68
69
70
71
71
72
72
73
ELECTROSTATICS
efield_caps ( total | induced | applied )
Required features:
67
ELECTROSTATICS
inter coulomb lB mmm2d maximal pairwise error [fixed far cutoff ]
[dielectric t m b ] [dielectric-contrasts ∆t ∆b ] [capacitor U ]
Required features:
66
ELECTROSTATICS
inter coulomb lB dh κ rcut
Required features:
65
66
ELECTROSTATICS
inter coulomb lB ewaldgpu tunealpha rcut ( Kcut | {Kcut,x Kcut,y Kcut,x } )
precision
Required features:
64
ELECTROSTATICS
inter coulomb lB ewaldgpu tune accuracy accuracy precision precision
K_max Kmax
Required features:
63
ELECTROSTATICS
inter coulomb [epsilon ( metallic | epsilon )] [n_interpol points]
[mesh_off xoff yoff zoff ]
inter coulomb lB ewaldgpu rcut ( Kcut | {Kcut,x Kcut,y Kcut,x } ) alpha
Required features:
63
ELECTROSTATICS
inter coulomb lB p3m ( tune | tunev2 ) [gpu] accuracy accuracy
[r_cut rcut ] [mesh mesh] [cao cao] [alpha α]
Required features:
62
62
BOND_ANGLE
bondid dihedral n K p
coulomb 0.0
coulomb
coulomb parameters
coulomb lB p3m [gpu] rcut ( mesh | {meshx meshy meshz } ) cao
alpha
Required features:
61
ELECTROSTATICS
73
inter coulomb lB mmm1d switch radius maximal pairwise error
inter coulomb lB mmm1d tune maximal pairwise error
Required features:
ELECTROSTATICS
inter coulomb lB mmm1dgpu switch radius [bessel cutoff ]
maximal pairwise error
inter coulomb lB mmm1dgpu tune maximal pairwise error
Required features:
CUDA
ELECTROSTATICS
75
ELECTROSTATICS
inter coulomb elc maximal pairwise error gap size
[far cutoff ] [noneutralization] [dielectric t m b ]
[dielectric-contrasts ∆t ∆b ] [capacitor U ]
Required features:
74
ELECTROSTATICS
inter coulomb lB memd localeps node node x node y node z dir X /Y /Z
eps ε
Required features:
74
MMM1D_GPU
inter coulomb lB memd f mass mesh [epsilon ε∞ ]
Required features:
73
75
ELECTROSTATICS
iccp3m n induced charges convergence convergence criterion areas areas
76
normals normals sigmas sigmas epsilons epsilons [eps_out eps out ]
[relax relaxation parameter ] [max_iterations max iterations ]
[ext_field ext field ]
Required features:
ELECTROSTATICS
77
dielectric sphere center cx cy cz radius r res res
dielectric wall normal nx ny nz dist d res res
dielectric cylinder center cx cy cz axis ax ay az radius r direction
d
dielectric pore center cx cy cz axis ax ay az radius r length l
smoothing_radius rs res res
dielectric slitpore pore_mouth z channel_width c pore_width w
pore_length l upper_smoothing_radius us lower_smoothing_radius ls
78
inter magnetic 0.0
inter magnetic
inter magnetic parameters
inter magnetic lB p3m rcut mesh cao alpha
Required features:
DIPOLES
inter magnetic lB p3m ( tune | tunev2 ) accuracy accuracy
[r_cut rcut ] [mesh mesh] [cao cao] [alpha α]
Required features:
79
DIPOLES
inter magnetic mdlc accuracy gap size [far cutoff ]
Required features:
79
80
DIPOLES
227
inter magnetic lB dawaanr
Required features:
DIPOLES
inter magnetic lB mdds n_cut value n cut
Required features:
DIPOLES
DPD
nemd
nemd
nemd
nemd
nemd
nemd
86
88
89
89
90
91
NPT
exchange n slabs n exchange
shearrate n slabs shearrate
off
92
profile
viscosity
Required features:
NEMD
cellsystem domain_decomposition [-no_verlet_list]
cellsystem nsquare
cellsystem layered n layers
228
84
INTER_DPD
thermostat npt_isotropic temperature gamma0 gammaV
Required features:
84
INTER_DPD
thermostat inter_dpd ignore_fixed_particles 0
Required features:
83
or TRANS_DPD
thermostat inter_dpd temperature
Required features:
83
COMFORCE
inter forcecap ( Fmax | individual )
setmd variable
setmd variable [value]+
thermostat
thermostat off
thermostat parameters
thermostat langevin temperature gamma
thermostat ghmc temperature n md phi [-no_flip | -flip | -random_flip]
[-no_scale | -scale]
thermostat dpd temperature gamma r cut [ WF wf tgamma tr cut TWF twf ]
Required features:
82
COMFIXED
inter typeid1 typeid2 comforce flag dir force fratio
Required features:
82
INTER_DPD
inter typeid1 typeid1 comfixed flag
Required features:
81
TUNABLE_SLIP
inter type1 type2 inter_dpd gamma r cut wf tgamma tr cut twf
Required features:
81
MAGNETIC_DIPOLAR_DIRECT_SUM
inter type1 type2 tunable_slip T γL rcut δt vx vy vz
Required features:
81
92
93
94
94
cuda list
cuda setdevice id
cuda getdevice
95
on_collision
on_collision off
on_collision [exception] bind_centers d bond1
on_collision [exception] bind_at_point_of_collision d bond1 bond2 type
reaction reactant_type rt catalyzer_type ct product_type pt range r
ct_rate k ct [eq_rate k eq] [react_once on/off ]
reaction off
reaction print
Required features:
a
95
CATALYTIC_REACTIONSa
The current implementation also requires the use of verlet lists and domain decomposition.
kill_particle_motion [rotation] 1
Required features:
1 ROTATION
kill_particle_forces [torques] 1
Required features:
96
98
1 ROTATION
99
system_CMS
system_CMS_velocity
100
galilei_transform
100
integrate steps [recalc_forces] [reuse_forces]
integrate set [nvt]
integrate set npt_isotropic pext piston [x y z ] [-cubic_box]
100
time_integration
time_integration steps
100
minimize_energy fmax steps gamma maxdisplacement
101
change_volume Vnew
change_volume Lnew ( x | y | z | xyz )
102
lees_edwards_offset offsetnew
102
Required features:
LEES_EDWARDS
velocities vmax [start pid ] [count N ]
102
sort_particles
103
parallel_tempering::main -rounds N -swap swap -perform perform
[-init init] [-values {Ti }] [-connect master ] [-port port]
[-load jnode ] [-resrate Nreset ] [-info info]
104
parallel_tempering::set_shareddata data
104
229
105
metadynamics
metadynamics set off
metadynamics set distance pid1 pid2 dmin dmax bheight bwidth fbound dbins
metadynamics set relative_z pid1 pid2 zmin zmax bheight bwidth fbound zbins
metadynamics print_stat current_coord
metadynamics print_stat coord_values
metadynamics print_stat profile
metadynamics print_stat force
metadynamics load_stat profile list force list
Required features: METADYNAMICS
analyze mindist [type list a type list b]
108
analyze distto pid
analyze distto x y z
analyze nbhood pid r catch
109
analyze nbhood x y z rc atch
analyze distribution part type list a part type list b
111
[rmin [rmax [rbins [log flag [int flag]]]]]
analyze radial_density_map xbins ybins xrange yrange
[axisofrotation centerofrotation beadtypelist [thetabins]]
112
analyze modes2d
analyze get_lipid_orients
analyze lipid_orient_order
analyze bilayer_set
analyze bilayer_density_profile
analyze cell_gpb Manningparameter outercellradius innercellradius
[accuracy [numberofinteractions]]
112
analyze get_folded_positions [-molecule] [shift x y z ]
113
analyze Vkappa [(
reset | read | set Vκ,1 Vκ,2 avk
) ]
112
113
113
114
analyze ( rdf | <rdf> ) part type list a part type list b [rmin rmax rbins]
114
analyze structurefactor type order
analyze vanhove type rmin rmax rbins [tmax ]
114
analyze
analyze
analyze
analyze
centermass part type
momentofinertiamatrix typeid
find_principal_axis typeid
gyration_tensor [typeid ]
114
115
115
116
analyze aggregation dist criteria s mol id f mol id
[min contact [charge criteria]]
116
analyze necklace pearl threshold back dist space dist first length
analyze holes typeidprobe mesh size
116
Required features:
230
LENNARD_JONES
116
analyze fluid temp 1 or 2 or 3
Required features:
1 LB
2 LB_GPU
117
3 ELECTROKINETICS
energy
energy ( total | kinetic | coulomb | magnetic )
energy bonded bondid
energy nonbonded typeid1 typeid2
117
analyze pressure
analyze pressure total
analyze pressure ( totals | ideal | coulomb |
tot_nonbonded_inter | tot_nonbonded_intra | vs_relative )
analyze pressure bonded bondid
analyze pressure nonbonded typeid1 typeid2
analyze pressure nonbonded_intra [typeid ]
analyze pressure nonbonded_inter [typeid ]
118
analyze stress_tensor
analyze stress_tensor total
analyze stress_tensor ( totals | ideal | coulomb |
tot_nonbonded_inter | tot_nonbonded_intra )
analyze stress_tensor bonded bondt ype
analyze stress_tensor nonbonded typeid1 typeid2
analyze stress_tensor nonbonded_intra [typeid ]
analyze stress_tensor nonbonded_inter [typeid ]
118
analyze local_stress_tensor periodic x periodic y periodic z range start x
range start y range start z range x range y range z bins x bins y
bins z
analyze set chains [chain start n chains chain length]
analyze set topo_part_sync
analyze set
119
analyze ( re | <re> ) [chain start n chains chain length]
121
analyze ( rg | <rg> ) [chain start n chains chain length]
122
analyze ( rh | <rh> ) [chain start n chains chain length]
122
analyze
analyze
analyze
analyze
120
analyze ( internal_dist | <internal_dist> ) [chain start n chains chain length]
123
analyze ( bond_dist | <bond_dist> ) [index index ]
[chain start n chains chain length]
123
analyze ( bond_l | <bond_l> ) [chain start n chains chain length]
124
analyze ( formfactor | <formfactor> ) qmin qmax qbins
[chain start n chains chain length]
124
analyze rdfchain rmin rmax rbins [chain start n chains chain length]
124
analyze ( <g1>| <g2>| <g3> ) [chain start n chains chain length]
analyze g123 [-init] [chain start n chains chain length]
125
231
analyze append
analyze remove [index ]
analyze replace index
analyze push [size]
analyze configs config
analyze configs
analyze stored
uwerr data nrep col [s tau] [plot]
uwerr data nrep f [s tau [f args]] [plot]
observable new name [parameters+]
observable id print [formatted]
observable id delete
observable new needs_profile_specs [other parameters] [ minx minx ]
[ maxx maxx ] [ miny miny ] [ maxy maxy ] [ minz minz ]
[ maxz maxz ] [ xbins xbins ] [ ybins ybins ] [ zbins zbins ]
observable new needs_radial_profile_specs [other parameters]
[ center <cx> <cy> <cx> ] [ maxr maxr ] [ minz minz ]
[ maxz maxz ] [ rbins rbins ] [ phibins phibins ] [ zbins zbins ]
correlation new obs1 id1 [obs2 id2 ] corr_operation
operation dt dt tau_max tau max [tau_lin tau lin]
[compress1 name [compress2 name] ]
correlation
correlation n_corr
correlation id autoupdate { start | stop}
correlation id update
correlation id finalize
correlation id write_to_file filename
correlation id print
correlation id print [ average1 | variance1 | correlation_time ]
correlation id print [ average_errorbars ]
correlation id write_checkpoint_binary filename
correlation id write_checkpoint_ascii filename
correlation id read_checkpoint_binary filename
correlation id read_checkpoint_ascii filename
blockfile channel write variable {varname1 varname2 ...}
blockfile channel write variable all
blockfile channel write tclvariable { varname1 varname2 ...}
blockfile channel write tclvariable all
blockfile channel write tclvariable reallyall
blockfile channel write particles what ( range | all )
blockfile channel write bonds range
blockfile channel write interactions
232
125
126
127
127
127
130
133
134
134
135
136
138
138
139
141
141
144
blockfile
blockfile
blockfile
blockfile
channel
channel
channel
channel
write
write
write
write
random
bit_random
seed
bitseed
145
blockfile channel write configs
145
blockfile channel write start tag
blockfile channel write end
blockfile channel write tag [arg]...
145
146
blockfile channel read start
blockfile channel read toend
blockfile channel read auto
blockfile channel read ( particles | interactions | bonds | variable |
seed | random | bitrandom | configs )
writemd channel [posx|posy|posz|vx|vy|vz|fx|fy|fz]...
146
readmd channel
147
writevsf channelId [( short | verbose )] [radius ( radii | auto )]
[typedesc typedesc]
148
writevcf channelId [( short | verbose )] [( folded | absolute )]
[pids ( pids | all )] [userdata userdata]
148
vtfpid pid
149
writevtk filename [( all | type )]
150
writepsf file [-molecule] NP MPC NC I Np S Nn S
151
writepdb file
writepdbfoldchains file chain start n chains chain length box l
writepdbfoldtopo file shift
151
readpdb pdb_file pdbfile type type first_id firstid
[ itp_file itpfile first_type fisttype]
[lj_with othertype epsilon sigma 1] [lj_rel_cutoff cutoff 1 ]
[fit_to_box]
152
Required features:
imd
imd
imd
imd
1 LENNARD_JONES
connect [port]
positions [( -unfolded |-fold_chains )]
listen seconds
disconnect
prepare_vmd_connection filename [start] [wait wait] [localhost]
[constraints] ...
prepare_vmd_connection [filename [wait [start [constraints]]]]
152
153
233
lbfluid [gpu] 2 [agrid agrid ] 1 or 2 [dens density ] 1 or 2 or 3
[visc viscosity] 1 or 2 or 3 [tau lb timestep] 1 or 2
[bulk_visc bulk viscosity] 1 or 2 or 3 [ext_force fx fy fz ] 1 or 2 or 3
[friction gamma ] 1 or 2 or 3 [couple 2pt/3pt ] 2
[gamma_odd gamma odd ] 1 or 2 or 3 [gamma_even gamma even] 1 or 2 or 3
[mobility] mobilities 3 [sc_coupling] coupling constants 3
Required features:
1 LB
2 LB_GPU
154
3 SHANCHEN
lbfluid print_interpolated_velocity x y z
155
lbfluid save_ascii_checkpoint filename lbfluid save_binary_checkpoint 157
filename lbfluid load_ascii_checkpoint filename lbfluid load_binary_checkpoint filename
thermostat lb 1 or 2 or 3 T
157
Required features:
1 LB
2 LB_GPU
3 SHANCHEN
lbnode x y z ( print | set ) args
Required features:
1 LB
2 LB_GPU
158
lbfluid remove_momentum
Required features:
1 LB
2 LB_GPU
3 SHANCHEN
lbfluid print [vtk] property filename [filename]
lbboundary shape shape args [velocity vx vy vz ]
lbboundary force [nboundary ]
Required features:
1 LB
2 LB_GPU
LB
167
LB_ELECTROHYDRODYNAMICS
countBonds particlel ist
findPropPos particlep ropertyl ist property
findBondPos particlep ropertyl ist
timeStamp path prefix postfix suffix
234
166
166
setmd mu_E µEx µEy µEz
Required features:
165
LB_BOUNDARIES
lbfluid cpu
lbfluid gpu
Required features:
158
3 SHANCHEN
170
171
171
171
B. Features
This chapter describes the features that can be activated in ESPResSo. Even if possible, it
is not recommended to activate all features, because this will negatively effect ESPResSo’s
performance.
Features can be activated in the configuration header myconfig.h (see section 3.4 on
page 27). Too activate FEATURE, add the following line to the header file:
#define FEATURE
B.1. General features
• PARTIAL_PERIODIC By default, all coordinates in ESPResSo are periodic. With
PARTIAL_PERIODIC turned on, the ESPResSo global variable periodic (see section 6.1 on page 86) controls the periodicity of the individual coordinates. Note
that this slows the integrator down by around 10 − 30%.
• ELECTROSTATICS This switches on the various electrostatics algorithms, such as
P3M. See section 5.7 on page 69 for details on these algorithms.
• DIPOLES This activates the dipole-moment property of particles; In addition, the
various magnetostatics algorithms, such as P3M are switched on. See section 5.7
on page 69 for details on these algorithms.
• ROTATION Switch on rotational degrees of freedom for the particles, as well as the
corresponding quaternion integrator. See section 4.1.1 on page 29 for details. Note,
that when the feature is activated, every particle has three additional degrees of
freedom, which for example means that the kinetic energy changes at constant
temperature is twice as large.
• ROTATION_PER_PARTICLE Allows to set whether a particle has rotational degrees
of freedom.
• LANGEVIN_PER_PARTICLE Allows to choose the Langevin temperature and friction
coefficient per particle.
• ROTATIONAL_INERTIA
• EXTERNAL_FORCES Allows to define an arbitrary constant force for each particle
individually. Also allows to fix individual coordinates of particles, e.g. keep them
at a fixed position or within a plane.
235
The list contains
all features, but
there are tons of
docs missing!
• CONSTRAINTS Turns on various spatial constraints such as spherical compartments
or walls. This constraints interact with the particles through regular short ranged
potentials such as the Lennard–Jones potential. See section 4.3 on page 42 for
possible constraint forms.
• TUNABLE_SLIP Switch on tunable slip conditions for planar wall boundary conditions. See section 5.9.1 on page 82 for details.
• MASS Allows particles to have individual masses. Note that some analysis procedures have not yet been adapted to take the masses into account correctly.
• EXCLUSIONS Allows to exclude specific short ranged interactions within molecules.
• COMFORCE Allows to pull apart groups of particles
• COMFIXED Allows to fix the center of mass of all particles of a certain type.
Docs missing
• MOLFORCES
How to use it?
• BOND_CONSTRAINT Turns on the RATTLE integrator which allows for fixed lengths
bonds between particles.
• VIRTUAL_SITES_COM Virtual sites are particles, the position and velocity of which
is not obtained by integrating equations of motion. Rather, they are placed using
the position (and orientation) of other particles. The feature VIRTUAL_SITES_COM
allows to place a virtual particle into the center of mass of a set of other particles.
See section 4.4 for details.
• VIRTUAL_SITES_RELATIVE Virtual sites are particles, the position and velocity of
which is not obtained by integrating equations of motion. Rather, they are placed
using the position (and orientation) of other particles. The feature VIRTUAL_SITES_RELATIVE
allows for rigid arrangements of particles. See section 4.4 for details.
• VIRTUAL_SITES_NO_VELOCITY
• VIRTUAL_SITES_THERMOSTAT
• THERMOSTAT_IGNORE_NON_VIRTUAL
• BOND_VIRTUAL
• MODES
• ADRESS
• METADYNAMICS
• LANGEVIN_PER_PARTICLE Allows to define the temperature and friction coefficient
for individual particles. See 4.1.1 for details.
236
• CATALYTIC_REACTIONS Allows the user to define three particle types to be reactant,
catalyzer, and product. Reactants get converted into products in the vicinity of a
catalyst according to a used-defined reaction rate constant. It is also possible to
set up a chemical equilibrium reaction between the reactants and products, with
another rate constant. See section 6.7 for details.
• OVERLAPPED
• COLLISION_DETECTION Allows particles to be bound oo collision. See section ??
• OLD_RW_VERSION This switches back to the old, wrong random walk code of the
polymer. Only use this if you rely on the old behaviour and know what you are
doing.
In addition, there are switches that enable additional features in the integrator or
thermostat:
• NEMD Enables the non-equilbrium (shear) MD support (see section 6.3 on page 93).
• NPT Enables an on–the–fly NPT integration scheme (see section 6.2.4 on page 92).
• DPD Enables the dissipative particle dynamics thermostat (see section 6.2.3 on
page 90).
• TRANS_DPD Enables the transversal dissipative particle dynamics thermostat (see
section 6.2.3 on page 91).
• INTER_DPD Enables the dissipative particle dynamics thermostat implemented as
an interaction, allowing to choose different parameters between different particle
types (see section 6.2.3 on page 91).
• INTER_RF
Documentation!
• DPD_MASS_RED Enables masses in DPD using reduced, dimensionless mass units.
• DPD_MASS_LIN Enables masses in DPD using absolute mass units.
• LB Enables the lattice-Boltzmann fluid code (see section 12 on page 165).
• LB_GPU Enables the lattice-Boltzmann fluid code support for GPU (see section 12
on page 165).
• SHANCHEN Enables the Shan Chen bicomponent fluid code on the GPU (see section 12 on page 165).
• LB_ELECTROHYDRODYNAMICS Enables the implicit calculation of electro-hydrodynamics
for charged particles and salt ions in an electric field.
237
B.2. Interactions
The following switches turn on various short ranged interactions (see section 5.1 on
page 50):
• TABULATED Enable support for user–defined interactions.
• LENNARD_JONES Enable the Lennard–Jones potential.
• LENNARD_JONES_GENERIC Enable the generic Lennard–Jones potential with configurable exponents and individual prefactors for the two terms.
• LJCOS Enable the Lennard–Jones potential with a cosine–tail.
• LJCOS2 Same as LJCOS, but using a slightly different way of smoothing the connection to 0.
• LJ_ANGLE Enable the directional Lennard–Jones potential.
• GAY_BERNE
• HERTZIAN
• MOL_CUT
• NO_INTRA_NB
• MORSE Enable the Morse potential.
• BUCKINGHAM Enable the Buckingham potential.
• SOFT_SPHERE Enable the soft sphere potential.
• SMOOTH_STEP Enable the smooth step potential, a step potential with two length
scales.
• BMHTF_NACL Enable the Born-Meyer-Huggins-Tosi-Fumi potential, which can be
used to model salt melts.
Some of the short range interactions have additional features:
• LJ_WARN_WHEN_CLOSE This adds an additional check to the Lennard–Jones potentials that prints a warning if particles come too close so that the simulation
becomes unphysical.
• OLD_DIHEDRAL Switch the interface of the dihedral potential to its old, less flexible
form. Use this for older scripts that are not yet adapted to the new interface of
the dihedral potential.
If you want to use bond-angle potentials (see section 5.5 on page 67), you need the
followig features.
238
• BOND_ANGLE
• BOND_ANGLEDIST
• BOND_ENDANGLEDIST
BOND ANGLEDIST and
BOND ENDANGLEDIST are
completely undocumented.
B.3. Debug messages
Finally, there are a number of flags for debugging. The most important one are
• ADDITIONAL_CHECKS Enables numerous additional checks which can detect inconsistencies especially in the cell systems. This checks are however too slow to be
enabled in production runs.
• MEM_DEBUG Enables an internal memory allocation checking system. This produces
output for each allocation and freeing of a memory chunk, and therefore allows to
track down memory leaks. This works by internally replacing malloc, realloc
and free.
The following flags control the debug output of various sections of Espresso. You will
however understand the output very often only by looking directly at the code.
• COMM_DEBUG Output from the asynchronous communication code.
• EVENT_DEBUG Notifications for event calls, i. e. the on_? functions in initialize.c.
Useful if some module does not correctly respond to changes of e. g. global variables.
• INTEG_DEBUG Integrator output.
• CELL_DEBUG Cellsystem output.
• GHOST_DEBUG Cellsystem output specific to the handling of ghost cells and the
ghost cell communication.
• GHOST_FORCE_DEBUG
• VERLET_DEBUG Debugging of the Verlet list code of the domain decomposition cell
system.
• LATTICE_DEBUG Universal lattice structure debugging.
• HALO_DEBUG
• GRID_DEBUG
• PARTICLE_DEBUG Output from the particle handling code.
239
• P3M_DEBUG
• ESR_DEBUG debugging of P3 Ms real space part.
• ESK_DEBUG debugging of P3 Ms k–space part.
• EWALD_DEBUG
• FFT_DEBUG Output from the unified FFT code.
• MAGGS_DEBUG
• RANDOM_DEBUG
• FORCE_DEBUG Output from the force calculation loops.
• PTENSOR_DEBUG Output from the pressure tensor calculation loops.
• THERMO_DEBUG Output from the thermostats.
• LJ_DEBUG Output from the Lennard–Jones code.
• MORSE_DEBUG Output from the Morse code.
• FENE_DEBUG
• ONEPART_DEBUG Define to a number of a particle to obtain output on the forces
calculated for this particle.
• STAT_DEBUG
• POLY_DEBUG
• MOLFORCES_DEBUG
• LB_DEBUG Output from the lattice–Boltzmann code.
• VIRTUAL_SITES_DEBUG
• ASYNC_BARRIER Introduce a barrier after each asynchronous command completion.
Helps in detection of mismatching communication.
• FORCE_CORE Causes ESPResSo to try to provoke a core dump when exiting unexpectedly.
• MPI_CORE Causes ESPResSo to try this even with MPI errors.
240
C. Sample scripts
In the directory ESPResSo/samples you find several scripts that can serve as samples
how to use ESPResSo.
lj liquid.tcl Simple Lennard-Jones particle liquid. Shows the basic features of ESPResSo:
How to set up system parameters, particles and interactions. How to warm up and
integrate. How to write parameters, configurations and observables to files. How
to handle the connection to VMD.
pe solution.tcl Polyelectrolyte solution under poor solvent condition. Test case for comparison with data produced by polysim9 from M.Deserno. Note that the equilibration of this system takes roughly 15000τ .
pe analyze.tcl Example for doing the analysis after the actual simulation run (offline
analysis). Calculates the integrated ion distribution P (r) for several different time
slaps, compares them and presents the final result using gnuplot to generate some
ps-files.
harmonic oscillator.tcl A chain of harmonic oscillators. This is a T = 0 simulation to
test the energy conservation.
espresso logo.tcl The ESPResSo-logo, the exploding espresso cup, has been created with
this script. It is a regular simulation of a polyelectrolyte solution. It makes use of
some nice features of the part command (see section 4.1 on page 29, namely the
capability to fix a particle in space and to apply an external force.
241
D. Maxwell Equations Molecular Dynamics
(MEMD)
In this chapter, we want to give a more thorough introduction to the MEMD (or
“Maggs”) algorithm for the calculation of Coulomb interactions that is implemented in
ESPResSo. For an even more detailed description, we refer to the publications [38, 43].
The method is intimately related to the Car–Parrinello approach, while being equivalent
to solving Maxwell’s equations with freely adjustable speed of light.
D.1. Equations of motion
Denoting the particle masses with mi , their charges with qi , their coordinates and momentum with ~ri and p~i respectively, the interparticle potential (of non-electromagnetic
type) with U , for the coupled system of charges and fields we write the following equations of motion
1
p~i
mi
∂U
~ ri ) − ζ p~i + f~i
p~˙i = −
+ qi E(~
∂~ri
mi
˙~
~
A = −E
~˙ = c2 ∇
~ × ∇
~ ×A
~ − 1 ~j,
E
0
~r˙i =
(D.1)
(D.2)
(D.3)
(D.4)
~ the vector-potential,
where 0 is the vacuum dielectric constant, c the speed of light, A
~
E the electric field, ~j the current density; ζ is the particle friction constant, and f~i is a
random force satisfying the standard fluctuation-dissipation theorem:
D
E
fiα (t)fjβ (t0 ) = 2ζkB T δij δαβ δ(t − t0 ),
(D.5)
where α and β denote Cartesian indices.
~ = ∇ × A the system of equations can be rewritten in a
If we introduce the vector B
~
form similar to the usual Maxwell equations. Currently in ESPResSo the version with B
~
and E is implemented.
242
D.2. Discretization
For implementation on the computer, the equations need to be discretized with respect
to both space and time.We consider a domain of physical space as being an affine space
and divide it into subdomains of contiguous cells of cubic shape. The charges live on
the vertices of our lattice which has the spacing a. The electric fields E(l) and vector
potentials A(l) live on the edges or links and are aligned with them. We need also the
~ which lives on the faces of the cube or on the
operator ∇ × . It gives the vector B,
plaquettes, Fig. D.1.
Figure D.1.: Spatial elements of a cell complex
In the implementation of the algorithm we assume that particles with masses mi and
charges qi live in the continuum (off–lattice approach). The charges are interpolated on
the lattice with grid spacing a using a linear interpolation scheme.
D.3. Initialization of the algorithm
The algorithm as it is implemented only calculates stepwise time updates of the exact
field solution. Therefore in order to start the simulation for the given random distribution
of charges we have to calculate the initial electrostatic field, i. e. the exact solution of
the electrostatic problem. We find a particular solution of Gauss’ law as the result of
the following recursive procedure (see Fig. D.2):
1. The charge in the plane z = zplane is
1 X
qplane =
q(~ri )δ(zi − zplane ),
Nz
(D.6)
i
Nz is the number of charges in plane z = zplane . Update the z-field according to
the formula
qplane
Ez2 = Ez1 +
;
(D.7)
0 a2
243
2. Subtract the charge qplane from the each charge on sites of zplane . The charge of
the wire y = ywire , z = zplane is
qwire =
1 X
q(~ri )δ(zi − zplane )δ(yi − ywire ),
Ny
(D.8)
i
Ny now meaning the number of charges in the wire. Update y-field
Ey2 = Ey1 +
qwire
;
0 a2
(D.9)
3. Subtract the charge qwire from the each charge on the sites of (ywire , zplane ). Update
x field
qvertex
Ex2 = Ex1 +
(D.10)
0 a2
This scheme is repeated until the fields are completely relaxed (i. e. the energy is
minimized). During repetition, the spatial dimensions are permutated to avoid a drift
in one direction.
Figure D.2.: Recursive solution of Gauss’ law
D.4. Time integrator
For the time discretization we have adopted the elegant solution which was found by
Rottler and Maggs [38] and allows to conserve both time–reversibility and phase–space
volume conservation:
1. Update the particle momenta by half a time step.
244
~ field by half a time step.
2. Update the B
3. Update the particle positions in x direction by half a time step.
4. Update the electric field in x direction by half a time step.
5. Update the particle positions in y direction by half a time step.
6. Update the electric field in y direction by half a time step.
7. Update the particle positions in z direction by half a time step.
8. Update the electric field in z direction by a full time step.
9. Update the particle positions in z direction by half a time step.
10. Update the electric field in y direction by half a time step.
11. Update the particle positions in y direction by half a time step.
12. Update the electric field in x direction by half a time step.
13. Update the particle positions in x direction by half a time step.
~ field by half a time step.
14. Update the B
15. Update the particle momenta by half a time step.
D.5. Self–energy
The interpolation of the charges onto the lattice gives rise to the artificial force exerted
on the particle by its own field. In order to cure this remedy, the direct subtraction of
the self–energy is introduced.
For the interpolated charge cloud the self–energy can be directly calculated. For the
simple cubic lattice in three dimensions the linear interpolation will give 8 charges which
are placed at the corners of the cube with edge length a (see Fig. D.3).
q7
q5
q8
q6
q3
q1
a
q4
q2
Figure D.3.: Linear interpolation scheme
245
Therefore in our case the self-energy is a symmetric bilinear form defined by the
matrix {αij }, the elements of which do not depend on the position of the charge. In our
algorithm the values of the coefficients are
αij =
X cos ~k(R
~ı − R
~ )
1
P3
3
~ aı )
4a0 L
ı=1 (1 − cos k~
(D.11)
~k
~ i coordinates of the interpowhere L is the number of lattice points per dimension, R
~
lated charges and k the wave vector. Those values are calculated during the initialization
step and are used in the calculation of the self-force. The value of the self-force which
has to be subtracted from the overall forces is given by the following ansatz
XX
∂Uself
∂qj
∂qi
~
Fself = −
.
(D.12)
=−
αij qi
+ qj
∂~r
∂~r
∂~r
i
j
D.6. For which systems to use the algorithm
Although it is not very well known by now, this algorithm is a promising alternative to
the often used Ewald-based methods. The main advantages and disadvantages shall be
named here. However, it is still best to understand the concept of the algorithm and
figure out for yourself, if it may be an option.
- The fields are not calculated for an arbitrary charge distribution, but updated
from the last solution. Therefore, particles should not move too much between
timesteps (less than a lattice cube).
- No procedure for error tuning yet. You have to adjust the parameters and determine the error yourself.
- Only 3D periodic systems are possible for now.
- With the given interpolation scheme, the short-range part of the potential is highy
underestimated when two particles are in the same lattice cube!
- The initialization routine scales with O(N 3 ) and takes a long time for larger (and
also inhomogenous) systems.
+ The algorithm is a local update scheme and spatially varying properties can be
applied (in the future).
+ Because of the locality, the algorithm itself scales O(N ) and has a big advantage
in speed for larger systems.
+ Because of the locality, it is highly parallelized.
+ It is fast.
246
The last item is of course dependent on the system properties. But if the charges are
evenly distributed and the system is not too sparse, this algorithm outperforms P3M
easily. Especially for systems with more than 1000 charges.
Of course, if the system is not dense enough, one will have to set the lattice spacing
in a way to avoid several particles in one cell and the mesh will be very fine for not so
many charges. Also, if you have lots of charges but your simulation should only run for
a short time, the initialization scheme takes too long in comparison.
But, if you have dense systems with more than 1000 charges or simulations that run
for many timesteps, this method is definitely an option.
247
E. The MMM family of algorithms
E.1. Introduction
Cleanup: References, mathematics
In the MMM family of algorithms for the electrostatic interaction, a convergence factor
approach to tackle the conditionally convergent Coulomb sum is used (even the authors
of the original MMM method have no idea what this acronym stands for). Instead
of defining the summation order, one multiplies each summand by a continuous factor
c(β, rij , nklm ) such that the sum is absolutely convergent for β > 0, but c(0, ., .) = 1. The
energy is then defined as the limit β → 0 of the sum, i. e. β is an artificial convergence
2
parameter. For a convergence factor of e−βnklm the limit is the same as the spherical
limit, and one can derive the classical Ewald method quite conveniently through this
approach [53]. To derive the formulas for MMM, one has to use a different convergence
factor, namely e−β|rij +nklm | , which defines the alternative energy
0
N
N
X X
X
qi qj e−β|pij +nklm |
1
˜ = 1 lim
E
=: lim
qi qj φβ (xij , yij , zij ).
2 β→0
|pij + nklm |
2 β→0
k,l,m i,j=1
i,j=1
φβ is given by φβ (x, y, z) = φ˜β (x, y, z) +
φ˜β (0, 0, 0), where
φ˜β (x, y, z) =
e−βr
r
X
(k,l,m)6=0
for (x, y, z) 6= 0 and φβ (0, 0, 0) =
e−βrklm
.
rklm
˜ exists, but differs for three dimensionally periodic systems by some multiThe limit E
ple of the square of the dipole moment from the spherical limit as obtained by the Ewald
summation[53]. From the physical point of view the Coulomb interaction is replaced by
˜ is then the energy in the
a screened Coulomb interaction with screening length 1/β. E
limit of infinite screening length. But because of the conditional convergence of the electrostatic sum, this is not necessarily the same as the energy of an unscreened system.
Since the difference to the Ewald methods only depends on the dipole moment of the
system, the correction can be calculated easily in linear time and can be ignored with
respect to accuracy as well as to computation time.
˜ = E, i.e. the convergence factor
For one or two dimensionally systems, however, E
approach equals the spherical summation limit of the Ewald sum, and MMM1D and
MMM2D do not require a dipole correction.
Starting from this convergence factor approach, Strebel constructed a method of computational order O(N log N ), which is called MMM [56]. The favourable scaling is obtained, very much like in the Ewald case, by technical tricks in the calculation of the far
248
formula. The far formula has a product decomposition and can be evaluated hierarchically similarly to the fast multipole methods.
For particles sufficiently separated in the z-axis one can Fourier transform the potential
along both x and y. We obtain the far formula as
X e2πfpq z + e2πfpq (λz −z)
λz
2
2πiu
qy
2πiu
px
y
x
e
φ(x, y, z) = ux uy
e
+ 2πux uy uz z − z +
.
6
fpq e2πfpq λz − 1
p,q6=0
p
where λx,y,z are the box dimensions, fpq =
(ux p)2 + (uy q)2 , fp = ux p, fq =
ux q, ωp = 2πux p and ωq = 2πuy q. The advantage of this formula is that it allows for a
product decomposition into components of the particles. For example
e2πfpq z = e2πfpq (zi −zj ) = e2πfpq zi e−2πfpq zj
etc. Therefore one just has to calculate the sum over all these exponentials on the
left side and on the right side and multiply them together, which can be done in O(N )
computation time. As can be seen easily, the convergence of the series is excellent as
long as z is sufficiently large. By symmetry one can choose the coordinate with the
largest distance as z to optimise the convergence. Similar to the Lekner sum, we need a
different formula if all coordinates are small, i. e. for particles close to each other. For
sufficiently small uy ρ and ux x we obtain the near formula as
pq z)
˜ y, z) = 2ux uy P cosh(2πf
φ(x,
e2πiuy qy e2πiux px +
fpq (e2πfpq λz −1)
Pp,q>0
4ux
(K0 (2πux pρl ) + KN (2πux pρ−l )) cos(2πux px)−
l,p>0
P b2n
2n +
2ux
<
(2πu
(z
+
iy))
y
2n(2n)!
n≥1
P
− 12
(ψ(2n) (1+ux x)+ψ(2n) (1−ux x)) 2n
ux
ρ −
(2n)!
n
n≥0
2 log(4π).
Note that this time we calculate φ˜ instead of φ, i. e. we omit the contribution of the
primary simulation box. This is very convenient as it includes the case of self energy
and makes φ˜ a smooth function. To obtain φ one has to add the 1/r contribution of the
primary box. The self energy is given by
˜ 0, 0) = 2ux uy
φ(0,
1
X
p,q>0
fpq
e2πfpq λz
X
+8ux
KN (2πux λy pl)+2ux ψ (0) (1)−2 log(4π).
−1
l,p>0
Both the near and far formula are derived using the same convergence factor approach,
and consequently the same singularity in β is obtained. This is important since otherwise
the charge neutrality argument does not hold.
To obtain the O(N log N ) scaling, some algorithm tricks are needed, which are not
used in MMM1D, MMM2D or ELC and are therefore not discussed here. For details,
see Strebel [56]. MMM is not implemented in ESPResSo.
249
E.2. MMM2D
In the case of periodicity only in the x and y directions, the far formula looks like
φ(x, y, z) = 4ux uy
2ux uy
e−2πfpq |z|
cos(ωp x) cos(ωq y)+
fpq
P
−2πfp |z|
e−2πfq |z|
cos(ωq y) + p>0 e fp
q>0
fq
P
p,q>0
P
cos(ωp x) −
2πux uy |z|
,
and the near formula is
˜ y, z) = 4ux P
φ(x,
l,p>0 (K0 (ωp ρl ) + K0 (ωp ρ−l )) cos(ωp x)− P
PNψ −1 1
b2n
1
+
2ux n≥1 2n(2n)!
< (2πuy (z + iy))2n + k=1
rk
r−k −
1 (2n)
(2n)
P
−2
(ψ (Nψ +ux x)+ψ (Nψ −ux x))
ux n≥0
(ux ρ)2n −
(2n)!
n
u
2ux log 4π uxy .
As said before, the energy obtained from these potentials is equal to the electrostatic
energy obtained by the spherical summation limit. The deeper reason for this is that in
some sense the electrostatic sum is absolutely convergent [5].
The near formula is used for particles with a small distance along the z axis, for
all other particles the far formula is used. Below is shown, that the far formula can
be evaluated much more efficiently, however, its convergence breaks down for small z
distance. To efficiently implement MMM2D, the layered cell system is required, which
splits up the system in equally sized gaps along the z axis. The interaction of all particles
in a layer S with all particles in the layers S-1,S,S+1 is calculated using the near formula,
for the particles in layers 1, . . . , S − 2, and in layers S + 2, . . . , N , the far formula is used.
The implementation of the near formula is relatively straight forward and can be
treated as any short ranged force is treated using the link cell algorithm, here in the
layered variant. The special functions in the formula are somewhat demanding, but
for the polygamma functions Taylor series can be achieved, which are implemented in
mmm-common.h. The Bessel functions are calculated using a Chebychev series.
The treatment of the far formula is algorithmically more complicated. For a particle
i in layer Si , the formula can product decomposed, as in
P
qi
e−2πfpq zi
qi
fpq
e−2πfpq zi
j∈IS ,S<Si −1 qi qj
e−2πfpq |zi −zj |
fpq
cos(ωp xi ) cos(ωq yi )
cos(ωp (xi − xj )) cos(ωq (yi − yj )) =
P
2πfpq zj
P
e2πfpq zj
j∈IS ,S<Si −1 qj e
cos(ωp xj ) cos(ωq yj )+
cos(ωp xi ) sin(ωq yi ) j∈IS ,S<Si −1 qj
cos(ωp xj ) sin(ωq yj )+
fpq
P
e−2πfpq zi
2πf
z
qi fpq sin(ωp xi ) cos(ωq yi ) j∈IS ,S<Si −1 qj e pq j sin(ωp xj ) cos(ωq yj )+
P
−2πfpq z
qi e fpq i sin(ωp xi ) sin(ωq yi ) j∈IS ,S<Si −1 qj e2πfpq zj sin(ωp xj ) sin(ωq yj ).
250
This representation has the advantage, that the contributions of the two particles are
decoupled. For all particles j only the eight terms
(±,s/c,s/c)
ξj
= qj e±2πfpq zj sin / cos(ωp xj ) sin / cos(ωq yj )
are needed. The upper index describes the sign of the exponential term and whether
sine or cosine is used for xj and yj in the obvious way. These terms can be used for all
expressions on the right hand side of the product decomposition. Moreover it is easy to
see from the addition theorem for the sine function that these terms also can be used to
calculate the force information up to simple prefactors that depend only on p and q.
(±,s/c,s/c)
Every processor starts with the calculation of the terms ξj
and adds them up
in each layer, so that one obtains
Ξ(±,s/c,s/c)
=
s
X
(±,s/c,s/c)
ξj
.
j∈Ss
Now we calculate
Ξ(l,s/c,s/c)
=
s
X
(+,s/c,s/c)
Ξt
t<s−1
and
Ξ(h,s/c,s/c)
=
s
X
(−,s/c,s/c)
Ξt
,
t>s+1
which are needed for the evaluation of the product decomposition. While the bottom
(l,s/c,s/c)
processor can calculate Ξs
directly, the other processors are dependent on its
(l,s/c,s/c)
results. Therefore the bottom processor starts with the calculation of its Ξs
and
(l,s/c,s/c)
(+,s/c,s/c)
sends up Ξs
and Ξs
of its top layer s to the next processor dealing with
the layers above. Simultaneously the top processor starts with the calculation of the
(h,s/c,s/c)
Ξs
and sends them down. After the communicated has been completed, every
(l/h,s/c,s/c)
(±,s/c,s/c)
processor can use the Ξj
and the ξj
to calculate the force rsp. energy
contributions for its particles.
In pseudo code, the far formula algorithm looks like:
1. for each layer s = 1, . . . , S
(±,s/c,s/c)
a) Ξs
=0
b) for each particle j in layer s
(±,s/c,s/c)
i. calculate ξj
(±,s/c,s/c)
ii. Ξs
(l,s/c,s/c)
2. Ξ3
(±,s/c,s/c)
+ = ξj
(+,s/c,s/c)
= Ξ1
3. for each layer s = 4, . . . , S
251
(l,s/c,s/c)
a) Ξs
(l,s/c,s/c)
4. ΞS−2
(l,s/c,s/c)
= Ξs−1
(+,s/c,s/c)
+ Ξs−2
(−,s/c,s/c)
= ΞS
5. for each layer s = (S − 3), ..., 1
(l,s/c,s/c)
a) Ξs
(l,s/c,s/c)
= Ξs+1
(−,s/c,s/c)
+ Ξs+2
6. for each layer s = 1, ..., S
a) for each particle j in layer s
(+,s/c,s/c) (l,s/c,s/c)
Ξs
i. calculate particle interaction from ξj
(−,s/c,s/c) (h,s/c,s/c)
Ξs
and ξj
For further details, see Arnold and Holm [5, 4], Arnold et al. [6].
E.2.1. Dielectric contrast
A dielectric contrast at the lower and/or upper simulation box boundary can be included
comparatively easy by using image charges. Apart from the images of the lowest and
topmost layer, the image charges are far enough to be treated by the far formula, and
can be included as starting points in the calculation of the Ξ terms. The remaining
particles from the lowest and topmost layer are treated by direct summation of the near
formula.
This means, that in addition to the algorithm above, one has to only a few things:
during the calculation of the particle and cell blocks ξ and Ξ, one additionally calculates
the contributions of the image charges and puts them either in a separate array or, for
the boundary layers, into two extra ξ cell blocks outside the simulation box. The entries
in the separate array are then added up over all processors and stored in the Ξ-terms of
the lowest/topmost layer. This are all modifications necessary for the far formula part.
In addition to the far formula part, there is an additional loop over the particles at the
boundary to directly calculate their interactions with their images. For details, refer to
Tyagi et al. [60].
E.3. MMM1D
In one dimensionally periodic systems with z being the periodic coordinate, the far
formula looks like
P
φ(ρ, z) = 4uz p6=0 K0 (ωρ) cos(ωz) − 2uz log( 2λρz ) − 2uz γ
P
Fρ (ρ, z) = 8πu2z p6=0 pK1 (ωρ) cos(ωz) + 2uρz
P
Fz (ρ, z) = 8πu2z p6=0 pK0 (ωρ) sin(ωz),
the near formula is
252
1 P
−2
(ψ(2n) (Nψ +uz z)+ψ(2n) (Nψ −uz z))
˜
φ(ρ, z) = −uz n≥0
(uz ρ)2n − 2uz γ+
(2n)!
n
PNψ −1 1
1
k=1
rk + r−k
1 P
−2
(ψ(2n) (Nψ +uz z)+ψ(2n) (Nψ −uz z))
F˜ρ (ρ, z) = −u3z n≥0
(uz ρ)2n−1 +
(2n)!
n
PNψ −1 ρ
ρ
+ r3
k=1
r3
k 1 −k
P
−2
(ψ(2n+1) (Nψ +uz z)+ψ(2n+1) (Nψ −uz z))
F˜z (ρ, z) = −u2z n≥0
(uz ρ)2n +
(2n)!
n
PNψ −1 z+kλz
z−kλz
+
,
k=1
r3
r3
k
−k
where ρ denotes the xy-distance of the particles. As for the two dimensional periodic
case, the obtained energy is equal to the one dimensional Ewald sum. Algorithmically,
MMM1D is uninteresting, since neither the near nor far formula allow a product decomposition or similar tricks. MMM1D has to be implemented as a simple NxN loop.
However, the formulas can be evaluated efficiently, so that MMM1D can still be used
reasonably for up to 400 particles on a single processor [3].
E.4. ELC
The ELC method differs from the other MMM algorithms in that it is not an algorithm
for the calculation of the electrostatic interaction, but rather represents a correction term
which allows to use any method for threedimensionally periodic systems with spherical
summation order for twodimensional periodicity. The basic idea is to expand the two
dimensional slab system of height h in the non-periodic z-coordinate to a system with
periodicity in all three dimensions, with a period of λz > h, which leaves an empty gap
of height δ = λz − h above the particles in the simulation box.
Since the electrostatic potential is only finite if the total system is charge neutral, the
additional image layers (those layers above or below the original slab system) are charge
neutral, too. Now let us consider the n-th image layer which has an offset of nλz to the
original layer. If nλz is large enough, each particle of charge q j at position (xj , yj , zj +
nλz ) and its replicas in the xy-plane can be viewed as constituting a homogeneous
q
charged sheet of charge density σj = λx jλy . The potential of such a charged sheet at
distance z is 2πσj |z|. Now we consider the contribution from a pair of image layers
located at ±nλz , n¿0 to the energy of a charge q i at position (xi , yi , zi ) in the central
layer. Since |zj − zi | < nλz , we have |zj − zi + nλz | = nλz + zj − zi and |zj − zi − nλz | =
nλz − zj + zi , and hence the interaction energy from those two image layers with the
charge qi vanishes by charge neutrality:
2πqi
N
X
j=1
σj (|zj − zi + nλz | + |zj − zi − nλz |) = 4πqi nλz
N
X
σj = 0.
j=1
253
The only errors occurring are those coming from the approximation of assuming homogeneously charged, infinite sheets instead of discrete charges. This assumption should
become better when increasing the distance nλz from the central layer.
However, in a naive implementation, even large gap sizes will result in large errors.
This is due to the order of summation for the standard Ewald sum, which is spherical,
while the above approach orders the cells in layers, called slab–wise summation. Smith
has shown that by adding to the Ewald energy the term
2πM 2
,
3
where M is the total dipole moment, one obtains the result of a slab–wise summation
instead of the spherical limit [53]. Although this is a major change in the summation
order, the difference is a very simple term. In fact, Smith shows that changes of the
summation order always result in a difference that depends only on the total dipole
moment.
Using the far formula of MMM2D, one can calculate the contributions of the additional layers up to arbitrarily precision, even for small gap sizes. This method is called
electrostatic layer correction, ELC. The advantage of this approach is that for the image
layers, z is necessarily large enough, so that all interactions can be represented using the
product decomposition. This allows for an order N evaluation of the ELC term.
The electrostatic layer correction term is given by
Ec = 2πMz2 −
Elc =
N
X
qi qj ψ(pi − pj ),
i,j=1
where
cosh(2πfpq z)
p,q>0 fpq (e2πfpq λz −1) cos(ωp x) cos(ωq y)+
P
cosh(2πf z)
2ux uy p>0 f (e2πfp λzp−1) cos(ωp x)+
p
P
cosh(2πf z)
2ux uy q>0 f (e2πfq λzq−1) cos(ωq y).
q
ψ(x, y, z) = 4ux uy
P
The implementation is very similar to MMM2d, except that the separation between
slices closeby, and above and below is not necessary.
E.5. Errors
Common to all algorithms of the MMM family is that accuracy is cheap with respect to
computation time. More precisely, the maximal pairwise error, i.e. the maximal error
of the ψ expression, decreases exponentially with the cutoffs. In turn, the computation
time grows logarithmically with the accuracy. This is quite in contrast to the Ewald
methods, for which decreasing the error bound can lead to excessive computation time.
For example, P3M cannot reach precisions above 10−5 in general. The precise form of
the error estimates is of little importance here, for details see Arnold et al. [6].
254
One important aspect is that the error estimates are also exponential in the nonperiodic coordinate. Since the number of closeby and far away particles is different for
particles near the border and in the center of the system, the error distribution is highly
non–homogenous. This is unproblematic as long as the maximal error is really much
smaller than the thermal energy. However, one cannot interpret the error simply as an
additional error source.
10
1
∆F∞
0.1
0.01
0.001
0.0001
1e-05
0
1
2
3
4
5
6
7
8
9
z
Figure E.1.: Error distribution of the ELC method.
Figure E.1 shows the error distribution of the ELC method for a gap size of 10% of
the total system height. For MMM2D and MMM1D the error distribution is less homogenous, however, also here it is always better to have some extra precision, especially
since it is computationally cheap.
255
F. Bibliography
[1] Andersen. Molecular-Dynamics Simulations At Constant Pressure And/Or Temperature. J. Chem. Phys, 72(4):2384–2393, 1980. ISSN 0021-9606.
[2] Hans C. Andersen. Rattle: A ”velocity” version of the shake algorithm for molecular
dynamics calculations. J. Comp. Phys., 51:24–34, 1983.
[3] A. Arnold and C. Holm. MMM1D: A method for calculating electrostatic interactions in 1D periodic geometries. J. Chem. Phys., 123(12):144103, 2005.
[4] Axel Arnold and Christian Holm. A novel method for calculating electrostatic
interactions in 2D periodic slab geometries. Chem. Phys. Lett., 354:324–330, 2002.
[5] Axel Arnold and Christian Holm. MMM2D: A fast and accurate summation
methodlimnb for electrostatic interactions in 2d slab geometries. Comput. Phys.
Commun., 148(3):327–348, 2002.
[6] Axel Arnold, Jason de Joannis, and Christian Holm. Electrostatics in Periodic Slab
Geometries I+II. J. Chem. Phys., 117:2496–2512, 2002.
[7] Axel Arnold, Olaf Lenz, Stefan Kesselheim, Rudolf Weeber, Florian Fahrenberger,
Dominic Roehm, Peter Kosovan, and Christian Holm. ESPResSo 3.1 – molecular dynamics software for coarse-grained models. In Michael Griebel, Christian
Rieger, and Marc Alexander Schweitzer, editors, Proceedings of the Sixth International Workshop on Meshfree Methods for Partial Differential Equations, Lecture
Notes in Computational Science and Engineering. Springer, Berlin, Germany, submitted.
[8] V. Ballenegger, A. Arnold, and J. J. Cerda. Simulations of non-neutral slab systems with long-range electrostatic interactions in two-dimensional periodic boundary conditions. J. Chem. Phys., 131(9):094107, 2009. doi: 10.1063/1.3216473. URL
http://link.aip.org/link/?JCP/131/094107/1.
[9] A. Brodka. Ewald summation method with electrostatic layer correction for interactions of point dipoles in slab geometry. Chem. Phys. Lett., 400:62–67, 2004.
[10] Juan J. Cerda, V. Ballenegger, O. Lenz, and C.Holm. P3M algorithm for dipolar
interactions. J. Chem. Phys., 129:234104, 2008.
[11] A. Chatterjee. Modification to Lees–Edwards periodic boundary condition for dissipative particle dynamics simulation with high dissipation rates. Molecular Simulation, 33(15):1233–1236, 2007.
256
[12] I. Cimr´ak, M. Gusenbauer, and T. Schrefl. Modelling and simulation of processes
in microfluidic devices for biomedical applications. Computers & Mathematics with
Applications, 64(3):278–288, 2012.
[13] Lindsay M Crowl and Aaron L Fogelson. Computational model of whole blood
exhibiting lateral platelet motion induced by red blood cells. Int. J. Numer. Meth.
Biomed. Engng., 26(3-4):471–487, 2010.
[14] M. Dao, C.T. Lim, and S. Suresh. Mechanics of the human red blood cell deformed
by optical tweezers. J. Mech. Phys. Solids, 51:2259–2280, 2003.
[15] M. Deserno. Counterion condensation for rigid linear polyelectrolytes. PhD thesis,
Universit¨at Mainz, 2000.
[16] M. Deserno and C. Holm. How to mesh up Ewald sums. i. J. Chem. Phys., 109:
7678, 1998.
[17] M. Deserno and C. Holm. How to mesh up Ewald sums. ii. J. Chem. Phys., 109:
7694, 1998.
[18] M. Deserno, C. Holm, and H. J. Limbach. Molecular Dynamics on Parallel Computers, chapter How to mesh up Ewald sums. World Scientific, Singapore, 2000.
[19] M Doi and S F Edwards. The theory of polymer dynamics. Oxford Science Publications, 1986.
[20] Burkhard D¨
unweg and Anthony J.C. Ladd. Lattice Boltzmann Simulations of
Soft Matter Systems. In Lattice Boltzmann Simulations of Soft Matter Systems,
Advances in Polymer Science, pages 1–78. Springer Berlin Heidelberg, 2008. doi:
10.1007/12 2008 4. URL http://dx.doi.org/10.1007/12_2008_4.
[21] M.M. Dupin, I. Halliday, C.M. Care, and L. Alboul. Modeling the flow of dense
suspensions of deformable particles in three dimensions. Phys Rev E Stat Nonlin
Soft Matter Phys., 75:066707, 2007.
[22] P.P. Ewald. Die berechnung optischer und elektrostatischer gitterpotentiale. Ann.
Phys., 64:253–287, 1921.
[23] Daan Frenkel and Berend Smit. Understanding Molecular Simulation. Academic
Press, San Diego, second edition, 2002.
[24] G Gompper and D M Kroll. Random surface discretizations and the renormalization
of the bending rigidity. Journal de Physique I, 6:1305–1320, 1996.
[25] Gary S. Grest and Kurt Kremer. Molecular dynamics simulation for polymers in
the presence of a heat bath. Phys. Rev. A, 33(5):3628–31, 1986.
[26] Owen A. Hickey, Christian Holm, James L. Harden, and Gary W. Slater. Implicit
Method for Simulating Electrohydrodynamics of Polyelectrolytes. Phys. Rev. Lett.,
105(14), SEP 29 2010. doi: {10.1103/PhysRevLett.105.148301}.
257
[27] R. W. Hockney and J. W. Eastwood. Computer Simulation Using Particles. IOP,
1988.
[28] F. H¨
ofling, Karl-Ulrich Bamberg, and Thomas Franosch. Anomalous transport
resolved in space and time by fluorescence correlation spectroscopy. Soft Matter, 7:
1358, 2011.
[29] Alan M. Horowitz. A generalized guided monte carlo algorithm. Phys. Lett. B, 268:
247 – 252, 1991.
[30] W. Humphrey, A. Dalke, and K. Schulten. VMD: Visual molecular dynamics. J.
Mol. Graphics, 14:33–38, 1996.
[31] C. Junghans, M. Praprotnik, and K. Kremer. Transport properties controlled by a
thermostat: An extended dissipative particle dynamics thermostat. Soft Matter, 4:
156, 2008.
[32] Stefan Kesselheim, Marcello Sega, and Christian Holm.
Applying to dna
translocation:
Effect of dielectric boundaries.
Computer Physics Communications, 182(1):33 – 35, 2011.
ISSN 0010-4655.
doi: 10.1016/j.
cpc.2010.08.014. URL http://www.sciencedirect.com/science/article/pii/
S001046551000305X. ¡ce:title¿Computer Physics Communications Special Edition for Conference on Computational Physics Kaohsiung, Taiwan, Dec 15-19,
2009¡/ce:title¿.
[33] Jiri Kolafa and John W. Perram. Cutoff errors in the ewald summation formulae
for point charge systems. Molecular Simulation, 9(5):351–368, 1992.
[34] A. Kolb and B. D¨
unweg. Optimized constant pressure stochastic dynamics. J.
Chem. Phys., 111(10):4453–59, 1999.
[35] Timm Kr¨
uger. Computer simulation study of collective phenomena in dense suspensions of red blood cells under shear. PhD thesis, Universit¨at Bochum, 2011.
[36] H. J. Limbach and C. Holm. Single-chain properties of polyelectrolytes in poor
solvent. J. Phys. Chem. B, 107(32):8041–8055, 2003.
[37] D Magatti and F Ferri. Fast multi-tau real-time software correlator for dynamic
light scattering. Applied Optics, 40(24):4011–4021, AUG 20 2001. ISSN 0003-6935.
doi: {10.1364/AO.40.004011}.
[38] A. C. Maggs and V. Rosseto. Local simulation algorithms for coulombic interactions.
Phys. Rev. Lett., 88:196402, 2002.
[39] Bernward A. Mann. The Swelling Behaviour of Polyelectrolyte Networks. PhD
thesis, Johannes Gutenberg-University, Mainz, Germany, December 2005.
258
[40] S. Marsili, G.F. Signorini, R. Chelli, M. Marchi, and P. Procacci. Orac: A molecular dynamics simulation program to explore free energy surfaces in biomolecular
systems at the atomistic level. J. Comp. Chem., 31:1106, 2009.
[41] B. Mehlig, D. W. Heermann, and B. M. Forrest. Hybrid monte carlo method for
condensed-matter systems. Phys. Rev. B, 45:679–685, 1992.
[42] P. Nikunen, M. Karttunen, and I. Vattulainen. How would you integrate the equations of motion in dissipative particle dynamics simulations. Com. Phys. Comm.,
153:407, 2003.
[43] Igor Pasichnyk and Burkhard D¨
unweg. Coulomb interactions via local dynamics:
A molecular-dynamics algorithm. J. Phys.: Condens. Matter, 16(38):3999–4020,
September 2004.
[44] Charles S Peskin. The immersed boundary method. Acta Numerica, 11:479–517,
2003.
[45] Jorge Ramirez, Sathish K. Sukumaran, Bart Vorselaars, and Alexei E. Likhtman.
Efficient on the fly calculation of time correlation functions in computer simulations.
J. Chem. Phys., 133(15):154103, OCT 21 2010. ISSN 0021-9606. doi: {10.1063/1.
3491098}.
[46] D. Roehm and A. Arnold. Lattice boltzmann simulations on GPUs with ESPResSo.
Eur. Phys. J. ST, 210:73–88, 2012.
[47] Michael Rubinstein and Ralph H. Colby. Polymer Physics. Oxford University Press,
Oxford, UK, 2003.
[48] K. Sch¨atzel, M. Drewel, and S Stimac. Photon-correlation measurements at large
lag times - improving statistical accuracy. Journal of Modern Optics, 35(4):711–718,
APR 1988. ISSN 0950-0340. doi: {10.1080/09500348814550731}.
[49] Heiko Schmitz and Florian Muller-Plathe. Calculation of the lifetime of positronium
in polymers via molecular dynamics simulations. J. Chem. Phys., 112(2):1040–1045,
2000. doi: 10.1063/1.480627. URL http://link.aip.org/link/?JCP/112/1040/
1.
[50] M. Sega, M. Sbragaglia, S. S. Kantorovich, and A. O. Ivanov. Mesoscale structures
at complex fluidfluid interfaces: a novel lattice boltzmann/molecular dynamics coupling. Soft Matter, 2013, in press. doi: 10.1039/C3SM51556G.
[51] X. Shan and H. Chen. Lattice boltzmann model for simulating flows with multiple
phases and components. Phys. Rev. E, 47:1815, 1993.
[52] J. Smiatek, M. P. Allen, and F. Schmidt. Tunable slip boundaries for coarse-grained
simulations of fluid flow. Eur. Phys. J. E, 26:115, 2008.
259
[53] E. R. Smith. Electrostatic energy in ionic crystals. Proc. R. Soc. Lond. A, 375:
475–505, 1981.
[54] T. Soddemann, B. D¨
unweg, and K. Kremer. A generic computer model for amphiphilic systems. Eur. Phys. J. E, 6:409, 2001.
[55] T. Soddemann, B. D¨
unweg, and K. Kremer. Dissipative particle dynamics: A useful
thermostat for equilibrium and nonequilibrium molecular dynamics simulations.
Phys. Rev. E, 68:046702, 2003.
[56] R. Strebel. Pieces of software for the Coulombic m body problem. Dissertation, ETH
Z¨
urich, 1999. URL http://e-collection.ethbib.ethz.ch/show?type=diss&nr=
13504.
[57] S. Succi. The lattice Boltzmann equation for fluid dynamics and beyond. Oxford
University Press, USA, 2001.
[58] A. P. Thompson, S. J. Plimpton, and W. Mattson. General formulation of pressure
and stress tensor for arbitrary many-body interaction potentials under periodic
boundary conditions. Journal of Chemical Physics, 131:154107, 2009.
[59] C. Tyagi, M. S¨
uzen, M. Sega, M. Barbosa, S. Kantorovich, and C. Holm. An
iterative, fast, linear-scaling method for computing induced charges on arbitrary
dielectric boundaries. J. Chem. Phys., 132:154112, 2010. doi: 10.1063/1.3376011.
[60] S. Tyagi, A. Arnold, and C. Holm. ICMMM2D: An accurate method to include
planar dielectric interfaces via image charge summation. J. Chem. Phys., 127:
154723, 2007.
[61] Sandeep Tyagi, Axel Arnold, and Christian Holm. Electrostatic layer correction
with image charges: A linear scaling method to treat slab 2d + h systems with
dielectric interfaces. J. Chem. Phys., 129(20):204102, 2008.
[62] Ulli Wolff. Monte carlo errors with less errors. Comput. Phys. Commun., 156:
143–153, 2004.
260
Index
Affinity interaction, 59
aggregation, 116
analysis, 111
aggregation, 116
bond distances internal first monomer,
124
bond lengths, 124
center of mass, 116
chains, 122
end-to-end distance of a chain, 122
energies, 118
finding holes, 117
fluid temperature, 118
form factor of a chain, 125
gyration tensor, 116
hydrodynamic radius of a chain, 123
internal distances within a chain, 124
local stress tensor, 121
minimal particle distance, 111
moment of inertia matrix, 116
particle distance, 111
particle distribution, 112
particles in the neighbourhood, 112
pearl-necklace structures, 117
pressure, 119
principal axis of the moment of inertia, 116
radial distribution function, 125
radial distribution function g(r), 114
radius of gyration of a chain, 123
stress tensor, 120
structure factor S(q), 115
topologies, 122
van Hove autocorrelation function G(r, t),
115
Analysis in the Core, 129
analyze (Tcl-command), 111
Anisotropic interactions, 57
bending force, 64
binary I/O, 148
blockfile (Tcl-command), 144
blocks, 146
BMHTF interaction, 54
bond distances internal first monomer,
124
bond lengths, 124
bond-angle interactions, 67
bonded coulomb bond, 61
bonded interaction type id, 60
bonded interactions, 60
bonded interactions oif, 63
box_l (global variable), 86
Buckingham interaction, 55
build directory, 24
cell_grid (global variable), 86
cell_size (global variable), 86
cellsystem (Tcl-command), 94
center of mass, 116
chains, 122
change_volume (Tcl-command), 102
configuration header, 27
configure, 15, 23
configure options, 24
constraint (Tcl-command), 42
copy_particles (Tcl-command), 40
correlation (Tcl-command), 136
Correlations, 135
Coulomb interactions, 69
counterions (Tcl-command), 36
crosslink (Tcl-command), 40
261
DAWAANR method, 81
Debye-H¨
uckel potential, 73
diamond (Tcl-command), 37
dielectric (Tcl-command), 79
Dielectric interfaces, 76, 78, 79
dihedral interactions, 68
Dipolar interactions, 79
Directional Lennard-Jones interaction, 57
DLC method, 81
domain decomposition, 94
DPD, 83, 90
DPD interaction, 83
dpd_gamma (global variable), 86
dpd_r_cut (global variable), 86
ELC method, 77
electrokinetics (Tcl-command), 174
Electrostatic interactions, 69
end-to-end distance of a chain, 122
energies, 118
energy unit, 12
EwaldGPU method, 72
features, 23, 27, 235
ADDITIONAL CHECKS, 239
ADRESS, 236
ASYNC BARRIER, 240
BMHTF NACL, 238
BOND ANGLE, 239
BOND ANGLEDIST, 239
BOND CONSTRAINT, 236
BOND ENDANGLEDIST, 239
BOND VIRTUAL, 236
BUCKINGHAM, 238
CATALYTIC REACTIONS, 237
CELL DEBUG, 239
COLLISION DETECTION, 237
COMFIXED, 236
COMFORCE, 236
COMM DEBUG, 239
CONSTRAINTS, 236
DIPOLES, 235
DPD, 237
DPD MASS LIN, 237
262
DPD MASS RED, 237
ELECTROSTATICS, 235
ESK DEBUG, 240
ESR DEBUG, 240
EVENT DEBUG, 239
EWALD DEBUG, 240
EXCLUSIONS, 236
EXTERNAL FORCES, 235
FENE DEBUG, 240
FFT DEBUG, 240
FORCE CORE, 240
FORCE DEBUG, 240
GAY BERNE, 238
GHOST DEBUG, 239
GHOST FORCE DEBUG, 239
GRID DEBUG, 239
HALO DEBUG, 239
HERTZIAN, 238
INTEG DEBUG, 239
INTER DPD, 237
INTER RF, 237
LANGEVIN PER PARTICLE, 235
LANGEVIN PER PARTICLE, 236
LATTICE DEBUG, 239
LB, 237
LB DEBUG, 240
LB ELECTROHYDRODYNAMICS,
237
LB GPU, 237
LENNARD JONES, 238
LENNARD JONES GENERIC, 238
LJ ANGLE, 238
LJ DEBUG, 240
LJ WARN WHEN CLOSE, 238
LJCOS, 238
LJCOS2, 238
MAGGS DEBUG, 240
MASS, 236
MEM DEBUG, 239
METADYNAMICS, 236
MODES, 236
MOL CUT, 238
MOLFORCES, 236
MOLFORCES DEBUG, 240
MORSE, 238
MORSE DEBUG, 240
MPI CORE, 240
NEMD, 237
NO INTRA NB, 238
NPT, 237
OLD RW VERSION, 237
OLD DIHEDRAL, 238
ONEPART DEBUG, 240
OVERLAPPED, 237
P3M DEBUG, 240
PARTIAL PERIODIC, 235
PARTICLE DEBUG, 239
POLY DEBUG, 240
PTENSOR DEBUG, 240
RANDOM DEBUG, 240
ROTATION, 235
ROTATION PER PARTICLE, 235
ROTATIONAL INERTIA, 235
SHANCHEN, 237
SMOOTH STEP, 238
SOFT SPHERE, 238
STAT DEBUG, 240
TABULATED, 238
THERMO DEBUG, 240
THERMOSTAT IGNORE NON VIRTUAL, 236
TRANS DPD, 237
TUNABLE SLIP, 236
VERLET DEBUG, 239
VIRTUAL SITES COM, 236
VIRTUAL SITES DEBUG, 240
VIRTUAL SITES NO VELOCITY,
236
VIRTUAL SITES RELATIVE, 236
VIRTUAL SITES THERMOSTAT, 236
FENE bond, 60
FFTW, 14
finding holes, 117
fluid temperature, 118
form factor of a chain, 125
gamma (global variable), 86
Gaussian interaction, 56
Gay-Berne interaction, 58
Generic Lennard-Jones interaction, 52
global area conservation, 66
global variables, 144
box_l, 86
cell_grid, 86
cell_size, 86
dpd_gamma, 86
dpd_r_cut, 86
gamma, 86
integ_switch, 86
lb_components, 86
local_box_l, 86
max_cut_bonded, 86
max_cut_nonbonded, 86
max_cut, 86
max_num_cells, 87
max_part, 87
max_range, 87
max_skin, 87
min_global_cut, 87
min_num_cells, 87
n_layers, 87
n_nodes, 87
n_part_types, 87
n_part, 87
node_grid, 87
npt_p_ext, 87
npt_p_inst, 87
nptiso_gamma0, 87
nptiso_gammav, 87
periodicity, 87
piston, 87
skin, 87
temperature, 87
thermo_switch, 88
time_step, 88
time, 88
timings, 88
transfer_rate, 88
verlet_flag, 88
verlet_reuse, 88
warnings, 88
gyration tensor, 116
263
harmonic bond, 60
harmonic-force (Tcl-command), 46
hat interaction, 56
Hertzian interaction, 56
hydrodynamic radius of a chain, 123
ICC?, 78
iccp3m (Tcl-command), 78
icosaeder (Tcl-command), 39
IMD, 153
imd (Tcl-command), 154
Installation, 23
integ_switch (global variable), 86
integrate (Tcl-command), 101
inter (Tcl-command), 50
Interaction DPD, 91
interactions, 50
Affinity, 59
bending force, 64
BMHTF, 54
bond-angle, 67
bonded, 60
bonded oif, 63
bonded coulomb, 61
Buckingham, 55
Coulomb, 69
DAWAANR method, 81
Debye-H¨
uckel, 73
dihedral, 68
Dipolar, 79
Directional Lennard-Jones, 57
DLC method, 81
DPD, 83
ELC method, 77
Electrostatic, 69
EwaldGPU, 72
FENE, 60
gaussian, 56
Gay-Berne, 58
Generic Lennard-Jones, 52
global area conservation, 66
harmonic, 60
hat, 56
hertzian, 56
264
Lennard-Jones, 51
Lennard-Jones cosine, 53
local area conservation, 65
Maggs method, 75
Magnetostatic, 79
MDDS method, 82
MEMD, 75
MMM1D, 74
MMM2D, 73
Morse, 55
non-bonded, 50
P3M, 70
quartic, 61
rigid bond, 62
smooth-step, 54
soft-sphere, 55
stretching force, 63
subtracted Lennard-Jones, 61
tabulated, 51
tabulated bond, 62
Tunable-slip boundary interactions,
82
volume conservation, 66
interactive mode, 27
internal distances within a chain, 124
label:DPDthermostat, 90
lb (Tcl-command), 165
lb_components (global variable), 86
Lees-Edwards Boundaries, 33, 103
lees_edwards_offset (Tcl-command),
103
length unit, 12
Lennard-Jones cosine interaction, 53
Lennard-Jones interaction, 51
local area conservation, 65
local stress tensor, 121
local_box_l (global variable), 86
Maggs method, 75
Magnetostatic interactions, 79
make, 15
max_cut (global variable), 86
max_cut_bonded (global variable), 86
max_cut_nonbonded (global variable), 86
max_num_cells (global variable), 87
max_part (global variable), 87
max_range (global variable), 87
max_skin (global variable), 87
Maxwell Equation Molecular Dynamics,
75
MDDS method, 82
MEMD, 75
metadynamics (Tcl-command), 109
min_global_cut (global variable), 87
min_num_cells (global variable), 87
minimal particle distance, 111
minimize_energy (Tcl-command), 102
MMM1D method, 74
MMM2D method, 73
moment of inertia matrix, 116
momentum exchange method, 93
Morse interaction, 55
MPI, 14
Multiple tau correlator, 139
myconfig.hpp, 27
n_layers (global variable), 87
n_nodes (global variable), 87
n_part (global variable), 87
n_part_types (global variable), 87
NEMD, 93
nemd (Tcl-command), 93
node_grid (global variable), 87
Non-bonded interactions, 50
npt_p_ext (global variable), 87
npt_p_inst (global variable), 87
nptiso_gamma0 (global variable), 87
nptiso_gammav (global variable), 87
observable (Tcl-command), 129
Observables, 129
oif (Tcl-command), 183
P3M method, 70
parallel_tempering (Tcl-command), 105
part (Tcl-command), 29
particle distance, 111
particle distribution, 112
particles in the neighbourhood, 112
pearl-necklace structures, 117
periodicity (global variable), 87
physical units, 12
piston (global variable), 87
polymer (Tcl-command), 35
prepare_vmd_connection (Tcl-command),
155
pressure, 119
principal axis of the moment of inertia,
116
quartic bond, 61
quick reference of Tcl-commands, 222
radial distribution function, 125
radial distribution function g(r), 114
radius of gyration of a chain, 123
random number generators, 145
random seed, 145
Rattle Shake algorithm, 62
readpdb (Tcl-command), 153
requirements, 14
rigid bond, 62
salt (Tcl-command), 37
setmd (Tcl-command), 86
shear boundaries, 103
shear viscosity, 103
shear-rate method, 93
skin (global variable), 87
smooth-step interaction, 54
soft-sphere interaction, 55
sort_particles (Tcl-command), 104
source directory, 24
stored configurations, 126, 146
stress tensor, 120
stretching force, 63
structure factor S(q), 115
subtracted Lennard-Jones bond, 61
tabulated bond interactions, 62
tabulated interaction, 51
Tcl global variables, 145
Tcl-commands
265
analyze, 111
blockfile, 144
cellsystem, 94
change_volume, 102
constraint, 42
copy_particles, 40
correlation, 136
counterions, 36
crosslink, 40
diamond, 37
dielectric, 79
electrokinetics, 174
harmonic-force, 46
iccp3m, 78
icosaeder, 39
imd, 154
integrate, 101
inter, 50
lb, 165
lees_edwards_offset, 103
metadynamics, 109
minimize_energy, 102
nemd, 93
observable, 129
oif, 183
parallel_tempering, 105
part, 29
polymer, 35
prepare_vmd_connection, 155
readpdb, 153
salt, 37
setmd, 86
sort_particles, 104
thermostat, 88
time_integration, 102
uwerr, 127
velocities, 104
writepdb, 152
writepdbfoldchains, 152
writepdbfoldtopo, 152
writepsf, 152
writevcf, 150
writevsf, 149
writevtk, 151
266
Tcl/Tk, 14
temperature (global variable), 87
thermo_switch (global variable), 88
thermostat (Tcl-command), 88
time (global variable), 88
time unit, 12
time_integration (Tcl-command), 102
time_step (global variable), 88
timings (global variable), 88
topologies, 122
transfer_rate (global variable), 88
Tunable-slip boundary interaction, 82
units, 12
uwerr (Tcl-command), 127
van Hove autocorrelation function G(r, t),
115
vcf, 149
velocities (Tcl-command), 104
verlet_flag (global variable), 88
verlet_reuse (global variable), 88
virtual sites, 46
volume conservation, 66
vsf, 149
vtf, 149
warnings (global variable), 88
whitespace, 144
writepdb (Tcl-command), 152
writepdbfoldchains (Tcl-command), 152
writepdbfoldtopo (Tcl-command), 152
writepsf (Tcl-command), 152
writevcf (Tcl-command), 150
writevsf (Tcl-command), 149
writevtk (Tcl-command), 151
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement