CSG Tutorial¶

(Based on the 2011-2017 CECAM meetings)

Please pick topics you are most interested in, since finishing the tutorial might take longer than one afternoon.

Installation¶

Building VOTCA¶

The simplest way is to use CMake:

prefix=~/votca
version=master # or 'stable' or 'v1.4.1'
git clone -b ${version} --recursive https://github.com/votca/votca.git
cd votca
mkdir build
cd build
cmake -DCMAKE_INSTALL_PREFIX=${prefix} ..
make -j5
make install

To build a gromacs version for VOTCA use

cmake -DBUILD_OWN_GROMACS=ON ..
make -j5
make install

Using the tutorials¶

All the tutorials are in the installation folder, i.e. ${prefix}/share/votca/csg-tutorials.

Downloading the manual¶

The corresponding version of the manual can be found on github (development version here).

Running GROMACS¶

The tutorial uses GROMACS, a molecular dynamics (MD) package, for generating the reference data. If you are not familiar with GROMACS, please read the following section for a brief overview on how to set up and run MD simulations using GROMACS. You can find all the GROMACS input files in the atomistic folder of each tutorial, e.g., for a box of atomistic water see csg-tutorials/spce/atomistic/.

Input files¶

You will need four files to run MD simulations * conf.gro - stores the coordinates of the molecule(s). It can be viewed with vmd. The default file name is conf.gro. * grompp.mdp - stores all simulation options, such as time step, number of simulation steps, etc. * topol.top - topology of the molecule * forcefield.itp - description of the atomistic force-field (not always needed)

During this tutorial, you will need to modify * nsteps - number of MD steps * nstxout - output frequency of coordinates to the trajectory file traj.trr * nstfout - output frequency of forces to traj.trr * nstxout-compressed - output frequency to the traj.xtc file, often used by the iterative Boltzmann inversion method * nstlog - output frequency of the md.log file * nstenergy - output frequency to the ener.edr file containing all thermodynamic information

MD simulations¶

To run MD simulations using GROMACS, first one must create a binary topology file topol.tpr using the gmx grompp program. Then run the MD integrator using the gmx mdrun program. topol.tpr contains conf.gro, grompp.mdp, topol.top, and forcefield.itp.

gmx grompp -c conf.gro # -c is needed if other than conf.gro file is used
gmx mdrun -v # -v (verbose) gives the estimate of the run time

Running other MD programs¶

In addition to GROMACS, VOTCA supports ESPResSo, Lammps, dl_poly, HOOMD-blue, and ESPResSo++. The interface to these is a bit more advanced, meaning VOTCA will allow you to do more crazy things and warn you less about settings, which might not make sense at all. Let’s have a look at csg-tutorials/spce/espressopp/. (ibi_lammps, ibi_espresso, ibi_dlpoly, and ibi_hoomd-blue are pretty similar)

Input files

Here is an example of the ESPResSo++ input files to run the CG MD simulations of the Iterative Boltzmann inversion procedure.

conf.gro - stores the coordinates of the molecule(s). It can be viewed with vmd. A pdb or xyz file would be okay, too.
spce.py is the simulation script which will be called by csg_inverse. It stores the whole simulation procedure.
topol.xml -topology of the molecule, defined in the initial condition. This is needed as most gro/pdb/xyz files have no molecule definition in them.

Mapping an atomistic trajectory onto a coarse-grained trajectory¶

SPC/E Water¶

Folder: csg-tutorials/spce/atomistic/

Have a look at the center of mass mapping file water.xml in this folder. Use csg_map to create a coarse-grained configuration (adjust the corresponding command in the run.sh script to map conf.gro instead of confout.gro). Visualize both configurations with vmd. (execute vmd conf.gro &)

Create a mapping file where the center of a coarse-grained water molecule is not the center of mass, but a different one, e.g., the charge center of the atomistic one.

Run a short MD simulation of a box of SPC/E water using GROMACS. (This is needed later for the different coarse-graining (CG) methods.) The input files are in the same folder (csg-tutorials/spce/atomistic/). Due to limited time, decrease the number of steps (nsteps) in grompp.mdp to a reasonable value (5000-10000) and adjust the output frequency of the trajectory, .log, and .edr files to, e.g., 50-100 (nstxout, nstfout, nstlog, nstenergy):

gmx grompp # combines conf.gro, topol.top, and grompp.mdp, and forcefield.itp (which is in the gromos43a1.ff folder of GROMACS) into topol.tpr
gmx mdrun -v # runs MD integrator. The trajectory is saved to traj.trr

Map this short atomistic trajectory to a CG trajectory using csg_map (adjust the corresponding line in the run.sh script to map traj.trr onto traj_cg.trr instead of conf(out).gro to conf_cg.gro.)

Iterative Boltzmann inversion (IBI) for SPC/E water¶

Here, a one-site coarse-grained (CG) model of a rigid 3-site water molecule (SPC/E model) is constructed (see the previous section of the tutorial) using the iterative Boltzmann inversion (IBI) method. The center of the CG bead is chosen to be the center of mass (COM) of a molecule. The target radial distribution function (RDF) is calculated from the CG bead coordinates obtained by mapping the reference atomistic trajectory. In the last step, a coarse-grained potential is obtained by matching the RDFs of the atomisitc and CG systems using the IBI method. For a more detailed description, look at the following publication.

Atomistic simulation¶

A short atomistic MD simulation has been already run in the last part of the previous section (see folder: csg-tutorials/spce/atomistic/). You can extract all thermodynamic information (total energy, kinetic energy, pressure, etc.) from the binary ener.edr file using gmx energy. Running the Extract_Energies.sh script does this for you (see the corresponding command line in the run.sh script). It creates an additional subfolder energies with all thermodynamic information:

./Extract_Energies.sh $equi #  The argument $equi is optional. If provided, analysis will start at the corresponding time frame (in GROMACS units, ps) (e.g. 1 to 5)

Calculation of RDF¶

Once again, check the mapping file water.xml. Atom names listed in the definition of the COM bead should correspond to those used in the conf.gro file. You can use csg_dump to check this:

csg_dump --top topol.tpr

Check the options file settings.xml. It contains the section of the corresponding non-bonded interaction (CG-CG). After this, calculate the center of mass RDF:

csg_stat --top topol.tpr --trj traj.trr --cg water.xml --options settings.xml (--nt 3 --begin $equi) # ( ) denotes additional options: --nt # number of threads to run calculation in parallel with more then one threads (e.g. 3), --begin # time frame in GROMACS units (ps) to start analysis (e.g. 1 to 5)

Compare your RDF with CG-CG.dist.tgt in csg-tutorials/spce/ibi/ which has been calculated with a much longer atomistic simulation run.

Running IBI¶

Now switch to the folder: csg-tutorials/spce/ibi/. Reduce the number of MD steps in grompp.mdp and adjust the equilibration time in the settings.xml file (cg.gromacs.equi_time) to a lower value (time frame in GROMACS units (ps) at which the analysis of the CG trajectory is started in each IBI step). Finally, start the IBI iterations:

csg_inverse --options settings.xml

At each iteration step, the current CG potential is CG-CG.pot.cur. Then, the CG-MD simulation is performed and the CG-CG RDF (CG-CG.dist.new) is determined. Finally, the CG potential is updated: CG-CG.pot.new.

Calculate the pressure after several iterations using gmx energy. You can do this by copying the Extract_Energies.sh script from the main folder (csg-tutorials/spce/ibi/) to the appropriate step folder (step_xxx) and executing it (type ./Extract_Energies.sh). Again, it will create a subfolder energies. You may notice that the pressure of the CG simulation is significantly too high. This can be adjusted by applying a pressure correction. To do so, add an appropriate post update option to the settings file, so that a (simple) pressure correction is applied. You can check the corresponding section of the manual.

Inverse Monte Carlo (IMC) for SPC/E water¶

Developing a CG potential with the inverse Monte Carlo (IMC) method works in a similar way as in the IBI example. The IMC procedure, again, requires the coarse-grained RDF of the single bead mapping based on the atomistic simulation as input. Therefore, in csg-tutorials/spce/imc, again, the RDF of a long atomistic simulation run is proveded: CG-CG.dist.tgt.

It is required to reduce the number of MD steps in grompp.mdp and to adjust the equilibration time in the settings.xml file (cg.gromacs.equi_time) to a lower value. IMC converges faster than IBI, but needs a better statistical sampling in each CG step. Therefore, the number of MD steps should be larger than in the IBI tutorial. Start the IMC iterations:

csg_inverse --options settings.xml

Again, the current CG potential is CG-CG.pot.cur, the CG-CG RDF is CG-CG.dist.new and the updated CG potential is: CG-CG.pot.new.

Calculate the pressure after several iterations using gmx energy and compare it with the IBI tutorial. You can do this by copying the Extract_Energies.sh script from the main folder (csg-tutorials/spce/ibi/) to the appropriate step folder (step_xxx) and executing it.

Relative entropy (RE) minimization for SPC/E water¶

Relative entropy (RE) minimization based coarse-graining of SPC/E water works similar to the IBI and the IMC example above. The reference atomistic simulation and the CG mapping are the same as in the IBI example. Again, in csg-tutorials/spce/re/, the RDF of a long atomistic simulation run is provided: CG-CG.dist.tgt. In this tutorial, the water-water CG potential is modeled using a cubic B-spline functional form. An initial guess for the cubic B-spline knot values is provided in CG-CG.param.init. At each iteration step, the CG potential table is computed from the current CG parameters (CG-CG.param.cur), the CG-MD simulation is performed, and the CG-CG RDF (CG-CG.dist.new) is determined. Finally, the new CG potential parameters (CG-CG.param.new) are computed using the relative entropy minimization algorithm. Reduce the number of MD steps in grompp.mdp, adjust the equilibration time in the settings.xml file (cg.gromacs.equi_time) and start the RE iterations:

csg_inverse --options settings.xml

Again, the current CG potential is CG-CG.pot.cur, the CG-CG RDF is CG-CG.dist.new and the updated CG potential is: CG-CG.pot.new.

Calculate the pressure after several iterations using gmx energy and compare it with the IBI tutorial. Again, you can do this by copying the Extract_Energies.sh script from the main folder (csg-tutorials/spce/ibi/) to the appropriate step folder (step_xxx) and executing it. For a more detailed description of the RE method, look at the following publication.

Force matching for SPC/E water¶

We will now derive a non-bonded CG potential for SPC/E using the force matching method.

Atomistic simulation¶

Basis for the force matching procedure is an atomistic MD simulation. All files are found in the atomistic folder (csg-tutorials/spce/atomistic/). If you have done the above tutorials, you have already generated the files of the atomistic md run. If not, then do so. Adjust the number of time steps to a reasonable value (5000-10000) and also choose an appropriate output frequency of the trajectory. Make sure, both, coordinates and forces are written to the trajectory file (nstxout and nstfout should have the same value in grompp.mdp).

Force matching (FM)¶

All files for running the actual force matching calculation can be found in csg-tutorials/spce/force_matching/. Have a look at the settings file (fmatch.xml). In the general force matching section (cg.fmatch), the number of frames to read in simultaneously (frames_per_block) and the type of LS solver (constrainedLS) are fixed (it is preferred to use constrained LS). The fmatch block of the interaction (cg.non-bonded.fmatch) contains the interaction range (min and max), the step size for the internal spline representation (step) and the output step (out_step). min and max have to be adjusted to be within the range of the RDF (see the calculation of RDF section of the IBI tutorial). Run the FM calculation (see also the corresponding line of the run.sh script):

csg_fmatch --top ../atomistic/topol.tpr --trj ../atomistic/traj.trr --options fmatch.xml --cg water.xml (--begin $equi ) # ( ) denotes additional option: , --begin # time frame in GROMACS units (ps) to start analysis (e.g. 1 to 5)

To obtain the CG potential, the CG force has to be integrated. (see the appropriate lines in the run.sh script):

csg_call table integrate CG-CG.force CG-CG.pot # integrates the table
csg_call table linearop CG-CG.pot CG-CG.pot -1 0 # multiplication of all table values by -1 (potential)

Change the spline grid (step), blocksize, and parameter constrainedLS. his should provide an overview of the whole procedure.

Running of CG simulation¶

To run a CG simulation with GROMACS, the potential has to be converted to a potential table, GROMACS can read (table_CG_CG.xvg). (Check the inverse section in the fmatch.xml for the corresponding options):

csg_call --options fmatch.xml --ia-name CG-CG --ia-type non-bonded convert_potential gromacs --clean input.pot table_CG_CG.xvg # calls convert_potential gromacs. Unsampled regions for distances smaller than the min value are extrapolated.

To run a CG simulation, you will need the conf.gro, topol.top, index.ndx and grompp.mdp files. You can use the ones of the ibi tutorial, and adjust the number of timesteps and output settings. Then run the simulation. Afterwards, you can calculate the RDF and thermodynamic data as explained in the IBI tutorial. You can also use the Extract_Energies.sh script of the IBI tutorial. When calculating the RDF from the CG simulation, you don’t need a mapping file and the –cg option can be omitted.

Compare the CG potential, the RDF and thermodynamics with the ones of the IBI, IMC or RE method (or any other method) and with the atomistic simulation. You will see that different methods lead to significantly different interaction potentials and a single site water model with a pair interaction potential is not capable of reproducing the RDF and thermodynamics at the same time. (Reason: three-body contributions are important but cannot be projected on a 2-body coarse-grained force-field. The incorporation of non-bonded 3-body interactions is work in progress. An extension to analytic non-bonded 3-body interactions will be released soon).

Visualization of IBI updates¶

Go to the folder csg-tutorials/spce/realtime. Execute the run.sh script.

Coarse-graining of liquid methanol¶

In the folder csg-tutorials/methanol/, you will find all relevant files to run an atomistic simulation of liquid methanol and obtain CG potentials with the IBI, IMC and FM method. Look at the SPC/E water tutorial to learn how to do this. You can compare the differences of the CG potentials, RDFs and thermodynamics between the different CG schemes and the atomistic simulation to the differences of the SPC/E water simulations. You will see that in the case of methanol, a pair potential is a better approximation to an ideal CG potential as in the SPC/E water case. The reason is that non-bonded 3-body effects are less important.

Coarse-graining of liquid hexane¶

Go to the folder csg-tutorials/hexane/. So far, we only considered single bead mappings. Hexane is a small alkane molecule. In this tutorial, a 3 bead CG mapping with one bond type and one angle type is chosen

Atomistic simulation¶

Go into the csg-tutorials/hexane/atomistic/ folder. Have a look at the mapping file hexane.xml. The hexane molecule is mapped to 3 beads with two different bead types with two bonds (of the same type) and one angle. You will find all relevant GROMACS input files in the folder. Have a look at the run.sh script. Again, adjust the number of time steps and the output frequencies in grompp.mdp and run an atomistic simulation. Extract the thermodynamic information (./Extract_Energies.sh) and calculate the 3 different RDFs (A-A, B-B, A-B) and the bond and angle distributions with csg_stat. In addition, you can compute the bond and angle distributions with csg_boltzmann (see run_boltzmann.sh) Compare the distributions to those in csg-tutorials/hexane/ibi_all obtained by a significantly longer atomistic MD run. You can map the (final) .gro file of the atomistic simuation to the CG one to get all necessary information for running the IBI procedure.

IBI for all interactions¶

Go to the folder csg-tutorials/hexane/ibi_all. The folder contains target RDFs and bond and angle distributions from a longer atomistic MD run. Have a look at the settings.xml file. It contains the sections for the non-bonded and bonded interactions. The three non-bonded interactions are updated every 3rd iteration step (first A-A, then B-B and then A-B, etc., see the cg.non-bonded.inverse.do_potential section). The bonded interactions are updated every iteration step. Adjust the number of time steps and output frequency in the grompp.mdp file and the equilibration time in the settings.xml file (cg.inverse.gromacs.equi_time) and start the IBI iterations. Calculate the pressure after several iterations using gmx energy. You can do this by copying the Extract_Energies.sh script from the main folder (csg-tutorials/spce/ibi/) to the appropriate step folder (step_xxx) and executing it.

IBI for non-bonded interactions only¶

Go to the folder csg-tutorials/hexane/ibi_nonbonded. The folder contains the same target RDFs as the csg-tutorials/hexane/ibi_all folder. In addition, it contains the tabulated bond and angle potentials (table_b1.xvg and table_a1.xvg). They are obtained by (non-iterative) Boltzmann inversion of the bond and angle target distribution functions of a longer MD simulation run. You can compute them with csg_boltzmann (see the run_boltzmann.sh script). A lack of statistics will become most apparent at lower values of the angle. In this case you might want to adjust the min value in the boltzmann_cmds file. Compare your results with the pre-computed ones (table_b1.xvg and table_a1.xvg) and with the ones in the subfolder step_001 in csg-tutorials/hexane/ibi_all Again, adjust the number of time steps and equilibration time and start the iteration process. Calculate the pressure and compare the thermodynamic properties as well as the obtained non-bonded potentials with the ones of the csg-tutorials/hexane/ibi_all tutorial. Calculate the bond and angle distributions (This can be done with csg_stat, using the settings.xml file and the hexane_cg.xml mapping file of the csg-tutorials/hexane/ibi_all folder.)

FM for all interactions together¶

Go to the folder csg-tutorials/hexane/force_matching. The folder contains the hexane mapping file with bond and angle interactions (hexane.xml) and the force matching options file (fmatch.xml). Have a look at both files and the run.sh script and start the force matching procedure. Basis is the atomistic trajectory with force output in csg-tutorials/hexane/atomistic. Integrate the force output to obtain the potentials and convert them to GROMACS tables. (see the run.sh script). Compare the obtained potentials to the IBI potentials. You can run CG simulations with the CG potentials, again, using the conf.gro, topol.top, index.ndx and grompp.mdp files from the csg-tutorials/hexane/ibi_all folder. Calculate the RDFs, bond and angle distributions. When comparing the results to the IBI potentials, you will see that in some cases, force matching can have problems with bonded interactions, especially if the functional form of the coarse-grained force field lacks essential interactions such as bond-angle or 3-body correlations. In such cases it can help to perform force matching only on the non-bonded contributions as was shown here.

FM for non-bonded interactions only¶

The files for the tutorial can be found in csg-tutorials/hexane/hybrid_force_matching/. The folder should contain all necessary files to reproduce the plots from the publication. To be able to parametrize only the non-bonded interactions via force matching, an atomistic trajectory has to be generated containing only forces contributing to the non-bonded interactions, meaning all other contributions need to be excluded. This is achieved by generating a second atomistic topoly file. Have a look at the topol.top file and compare it to the one in csg-tutorials/hexane/atomistic/. All bonded interactions have been deleted. Furthermore, all intramolecular interactions have been explicitly excluded. Generate the binary GROMACS topology file using this topol.top file and the conf.gro and grommp.mdp file of the reference atomistic trajectory (csg-tutorials/hexane/atomistic/). Then, generate the trajectory file with excluded bonded interactions using gmx mdrun with the -rerun option. (Have a look at the corresponding lines of the run.sh script). Have a look at the fmatch.xml file. It now only contains the non-bonded interactions. Start the FM calculation. Afterwards, intergrate the force output and convert the potentials to GROMACS tables (see the run.sh script). You can run the CG simulation, using the conf.gro, topol.top, index.ndx and grompp.mdp files and the bond and angle potential (table_b1.xvg and table_a1.xvg) from the csg-tutorials/hexane/ibi_nonbonded/ folder. Calculate the RDFs, bond and angle distributions and compare the results to the IBI results and FM of all interactions together.

Regularization of the inverse Monte Carlo method¶

For this tutorial go to the folder csg-tutorials/LJ1-LJ2/imc. Inverse Monte Carlo (IMC) needs a well defined cross-correlation matrix for which enough sampling is needed. If there is not enough sampling the algorithm might not converge to a stable solution. This might also happen if the initial potential guess for the iterative scheme is too far away from the real solution of the inverse problem. To overcome this deficiency and to stabilize the algorithm one could apply the so called Tikhonov regularization, which is a common technique to regularize ill-posed inverse problems. For further information on the Tikhonov regularization and/or ill-posed inverse problems in general don’t hesitate to have a look at the manual of VOTCA-1.4 to get a short overview or for a more detailed description at this publication or consult any book of choice on regularization of inverse problems.

This tutorial can be considered to be a proof of concept. It is based on the above mentioned publication. Here the user should get familiar with the application of the Tikhonov regularization and should see its benefit. The file run.sh will execute a preliminary run of 10 steps of iterative Boltzmann inverson (IBI) before the IMC method is applied. The users should figure out what happens if the preliminary IBI steps are skipped and should test different regularization parameters (e.g. 10,100.300,1000). The folder also contains a short python script which performs a singular value decomposition of the cross-correlation matrix (svd.py). Based on this decomposition one could get an educated guess on the order of the magnitude of the regularization parameter. It should be larger than the smallest singular values squared and smaller compared to the larger ones.

Additional tutorials¶

Have a look in the folder csg-tutorials. It contains additional tutorials on propane, methanol-water and urea-water mixtures. To do the tutorials, have a look at the corresponding run.sh scripts.

Advanced topics¶

Extending the scripting framework¶

Write a post update script, which smooths the tail of a potential by transforming dU(r) to s(r)dU(r) with

s(r) = 1 for r < rt
s(r) = 1-(rc-rt)-3(r - rt)2(3rc-rt-2r) for rt < r < rc
s(r) = 0 for r > rt

Hints: Start from skeleton.pl and use pressure_cor_simple.pl as a template.

Writing an analysis tool¶

VOTCA allows to write your own analysis code. There are many examples and two templates for serial and threaded analysis. If you are willing to learn how to write your own analysis in C++, ask for assistance.