openmm (c39b2)
OpenMM GPU acceleration interface to CHARMM
This module describes the interface of CHARMM with the OpenMM
development platform for GPU accelerated simulations. The current
interface supports molecular dynamics on CUDA or OPENCL supported
graphical processing units (GPUs). For a full list of hardware on
which the OpenMM libraries should run, see the OpenMM website
(https://simtk.org/home/openmm). The OpenMM methods are free and
available as pre-compiled libraries or source form. In addition, one
needs the NVIDIA drivers and CUDA toolkit installed on the
machine - please see OpenMM documentation for the basic procedures to
set-up and install these components, as well as which versions are required.
The CHARMM/OpenMM interface is under continuing development with
new CHARMM features being added all of the time. The current
implementation supports dynamics and energy calculations for periodic
and non-periodic systems using cutoffs, nocutoffs (for finite
systems), and PME/Ewald and cutoffs for periodic systems. Periodic
systems supported are only orthorhombic (a,b,c, alpha=beta=gamma=90.
Only Leapfrog Verlet integration and Langevin dynamics are
supported. Constant temperature molecular dynamics is also supported
through the Andersen heatbath method in the OpenMM module. Additionally,
constant pressure, constant temperature dynamics are available using
MC sampled barostat. Finally, we have provided access to the variable
timestep Verlet (Leapfrog) and Langevin integrators implemented in the
OpenMM module. SHAKE is supported as are all of the CHARMM
forcefields, e.g., CMAP.
Special Notice: The CHARMM/OpenMM interface is an evolving interface
with the OpenMM accelerated dynamics engine for GPU accelerated
molecular dynamics (See the News at www.charmm.org for a discussion of
the benchmarks and their performance). The functionality present
through the current CHARMM interface has been released prior to
"aging" in the CHARMM developmental version for a year because of the
important performance enhancements in provides through GPU
acceleration. The interface and associated modules have been well
tested, but are likely to contain yet undiscovered limitations,
compared with the full functionality in CHARMM. Additionally, we note
that this code operates in single precision as default, which may not be
acceptable for all applications. However, with the release of OpenMM 5.0
double and mixed precision models are available. At present, the CHARMM/OpenMM
interface accommodates molecular dynamics with and without periodic
boundary conditions, using all of the current CHARMM force fields, and
in NVE, NVT and NPT ensembles - although not using the same methods of
achieving these as in the rest of CHARMM. Users are forewarned to
carry out some pre-testing on their system prior to initiating long
runs on GPUs. As new features and methods are added to the
documentation.
* Setup | Setting up to use/compile CHARMM/OpenMM
* Usage | Usage and functional support of CHARMM/OpenMM
* Multi-GPU | Using multiple GPUs and platforms
* Block-OpenMM Running Block calculations on the GPU
* GBSAOBC2-OpenMM Running GBSA OBC2 Generalized Born Model OpenMM
* Examples | Examples of CHARMM/OpenMM usage
This module describes the interface of CHARMM with the OpenMM
development platform for GPU accelerated simulations. The current
interface supports molecular dynamics on CUDA or OPENCL supported
graphical processing units (GPUs). For a full list of hardware on
which the OpenMM libraries should run, see the OpenMM website
(https://simtk.org/home/openmm). The OpenMM methods are free and
available as pre-compiled libraries or source form. In addition, one
needs the NVIDIA drivers and CUDA toolkit installed on the
machine - please see OpenMM documentation for the basic procedures to
set-up and install these components, as well as which versions are required.
The CHARMM/OpenMM interface is under continuing development with
new CHARMM features being added all of the time. The current
implementation supports dynamics and energy calculations for periodic
and non-periodic systems using cutoffs, nocutoffs (for finite
systems), and PME/Ewald and cutoffs for periodic systems. Periodic
systems supported are only orthorhombic (a,b,c, alpha=beta=gamma=90.
Only Leapfrog Verlet integration and Langevin dynamics are
supported. Constant temperature molecular dynamics is also supported
through the Andersen heatbath method in the OpenMM module. Additionally,
constant pressure, constant temperature dynamics are available using
MC sampled barostat. Finally, we have provided access to the variable
timestep Verlet (Leapfrog) and Langevin integrators implemented in the
OpenMM module. SHAKE is supported as are all of the CHARMM
forcefields, e.g., CMAP.
Special Notice: The CHARMM/OpenMM interface is an evolving interface
with the OpenMM accelerated dynamics engine for GPU accelerated
molecular dynamics (See the News at www.charmm.org for a discussion of
the benchmarks and their performance). The functionality present
through the current CHARMM interface has been released prior to
"aging" in the CHARMM developmental version for a year because of the
important performance enhancements in provides through GPU
acceleration. The interface and associated modules have been well
tested, but are likely to contain yet undiscovered limitations,
compared with the full functionality in CHARMM. Additionally, we note
that this code operates in single precision as default, which may not be
acceptable for all applications. However, with the release of OpenMM 5.0
double and mixed precision models are available. At present, the CHARMM/OpenMM
interface accommodates molecular dynamics with and without periodic
boundary conditions, using all of the current CHARMM force fields, and
in NVE, NVT and NPT ensembles - although not using the same methods of
achieving these as in the rest of CHARMM. Users are forewarned to
carry out some pre-testing on their system prior to initiating long
runs on GPUs. As new features and methods are added to the
documentation.
* Setup | Setting up to use/compile CHARMM/OpenMM
* Usage | Usage and functional support of CHARMM/OpenMM
* Multi-GPU | Using multiple GPUs and platforms
* Block-OpenMM Running Block calculations on the GPU
* GBSAOBC2-OpenMM Running GBSA OBC2 Generalized Born Model OpenMM
* Examples | Examples of CHARMM/OpenMM usage
Top
SETTING UP AND BUILDING CHARMM/OPENMM
=====================================
To build CHARMM with the OpenMM interface to enable GPU accelerated molecular
dynamics one needs to first install the appropriate GPU drivers and software
support, e.g., NVIDIA drivers and CUDA toolkit. Additionally, the OpenMM
libraries need to be installed. Please see the OpenMM web pages and
documentation to accomplish this procedure (https://simtk.org/home/openmm).
OpenGL header files may need to be installed. For example, for Debian
and related Linux distributions. One may need to install the package
mesa-common-dev if it is not already installed.
Setup necesssary environment variables for load library and OpenMM path:
The following environment variables need to be setup: OPENMM_PLUGIN_DIR
and the library path. Assuming OpenMM has been installed in it's default
directory (/usr/local/openmm), then set the following environment variable:
Mac OSX or Linux
bash shell: export OPENMM_PLUGIN_DIR=/usr/local/openmm/lib/plugins
csh shell: setenv OPENMM_PLUGIN_DIR /usr/local/openmm/lib/plugins
These should be added to your .bashrc or .cshrc to ensure they are always
setup.
Additionally, one needs to tell the loader where the OpenMM libraries are
installed. This differs on Linux versus Mac OSX systems because of static
versus dynamic load library uses on these two OSs.
Linux (bash): export LD_LIBRARY_PATH=/usr/local/openmm/lib:$OPENMM_PLUGIN_DIR:$LD_LIBRARY_PATH
Linux (csh): setenv LD_LIBRARY_PATH /usr/local/openmm/lib:$OPENMM_PLUGIN_DIR:$LD_LIBRARY_PATH
Mac OSX (bash):
export DYLD_LIBRARY_PATH=/usr/local/openmm/lib:$OPENMM_PLUGIN_DIR:$DYLD_LIBRARY_PATH
Max OSX (csh):
setenv DYLD_LIBRARY_PATH /usr/local/openmm/lib:$OPENMM_PLUGIN_DIR:$DYLD_LIBRARY_PATH
OpenMM anisotropic barostat functionality is also available as an OpenMM plugin.
To enable the use of the plugin, one extra environment variable must be set at
runtime as follows.
Linux / Intel compilers
bash:
export CHARMM_PLUGIN_DIR=<CHARMM root directory>/lib/em64t/openmm_plugins
csh:
export CHARMM_PLUGIN_DIR=<CHARMM root directory>/lib/em64t/openmm_plugins
Linux / GCC compilers
bash:
export CHARMM_PLUGIN_DIR=<CHARMM root directory>/lib/gnu/openmm_plugins
csh:
export CHARMM_PLUGIN_DIR=<CHARMM root directory>/lib/gnu/openmm_plugins
Mac OSX
bash:
export CHARMM_PLUGIN_DIR=<CHARMM root directory>/lib/osx/openmm_plugins
csh:
export CHARMM_PLUGIN_DIR=<CHARMM root directory>/lib/osx/openmm_plugins
INSTALLING CHARMM/OpenMM
========================
Installing CHARMM with the OpenMM interface is straightforward on Linux and
Mac OSX:
Setup necesssary environment variables: Set OPENMM_PLUGIN_DIR as described
above, and set CUDATK to the location of the CUDA toolkit. Assuming OpenMM
and CUDA have been installed in their default directories
(/usr/local/openmm and /usr/local/cuda respectively), then set the following
environment variables:
Mac OSX or Linux
bash shell:
export OPENMM_PLUGIN_DIR=/usr/local/openmm/lib/plugins
export CUDATK=/usr/local/cuda
csh shell:
setenv OPENMM_PLUGIN_DIR /usr/local/openmm/lib/plugins
export CUDATK=/usr/local/cuda
These should be added to your .bashrc or .cshrc to ensure they are always
setup.
Linux / Intel compilers
install.com em64t openmm
Linux / GCC compilers
install.com gnu openmm
Mac OSX / Intel compilers
install.com osx ifort openmm
Mac OSX / GCC compilers
install.com osx gfortran openmm
The OpenMM anisotropic barostat plugin is available for use by default. The CharmmNB plugin allows CHARMM to set certain Ewald summation parameters directly. Without this plugin, by default, OpenMM calculates the parameters using an error tolerance. If OpenMM's default method is sufficient, one can add the keyword nonbplugin to the above install.com command lines.
Note: To enable compatibility with OpenMM version 5.1, please replace the openmm key work on the install.com command line above with openmm51. At present we are supporting features for OpenMM 5.2 and OpenMM 5.1
releases. These releases provide slightly different interfaces to Cuda-based
computations and precision. See comments below for more details.
SETTING UP AND BUILDING CHARMM/OPENMM
=====================================
To build CHARMM with the OpenMM interface to enable GPU accelerated molecular
dynamics one needs to first install the appropriate GPU drivers and software
support, e.g., NVIDIA drivers and CUDA toolkit. Additionally, the OpenMM
libraries need to be installed. Please see the OpenMM web pages and
documentation to accomplish this procedure (https://simtk.org/home/openmm).
OpenGL header files may need to be installed. For example, for Debian
and related Linux distributions. One may need to install the package
mesa-common-dev if it is not already installed.
Setup necesssary environment variables for load library and OpenMM path:
The following environment variables need to be setup: OPENMM_PLUGIN_DIR
and the library path. Assuming OpenMM has been installed in it's default
directory (/usr/local/openmm), then set the following environment variable:
Mac OSX or Linux
bash shell: export OPENMM_PLUGIN_DIR=/usr/local/openmm/lib/plugins
csh shell: setenv OPENMM_PLUGIN_DIR /usr/local/openmm/lib/plugins
These should be added to your .bashrc or .cshrc to ensure they are always
setup.
Additionally, one needs to tell the loader where the OpenMM libraries are
installed. This differs on Linux versus Mac OSX systems because of static
versus dynamic load library uses on these two OSs.
Linux (bash): export LD_LIBRARY_PATH=/usr/local/openmm/lib:$OPENMM_PLUGIN_DIR:$LD_LIBRARY_PATH
Linux (csh): setenv LD_LIBRARY_PATH /usr/local/openmm/lib:$OPENMM_PLUGIN_DIR:$LD_LIBRARY_PATH
Mac OSX (bash):
export DYLD_LIBRARY_PATH=/usr/local/openmm/lib:$OPENMM_PLUGIN_DIR:$DYLD_LIBRARY_PATH
Max OSX (csh):
setenv DYLD_LIBRARY_PATH /usr/local/openmm/lib:$OPENMM_PLUGIN_DIR:$DYLD_LIBRARY_PATH
OpenMM anisotropic barostat functionality is also available as an OpenMM plugin.
To enable the use of the plugin, one extra environment variable must be set at
runtime as follows.
Linux / Intel compilers
bash:
export CHARMM_PLUGIN_DIR=<CHARMM root directory>/lib/em64t/openmm_plugins
csh:
export CHARMM_PLUGIN_DIR=<CHARMM root directory>/lib/em64t/openmm_plugins
Linux / GCC compilers
bash:
export CHARMM_PLUGIN_DIR=<CHARMM root directory>/lib/gnu/openmm_plugins
csh:
export CHARMM_PLUGIN_DIR=<CHARMM root directory>/lib/gnu/openmm_plugins
Mac OSX
bash:
export CHARMM_PLUGIN_DIR=<CHARMM root directory>/lib/osx/openmm_plugins
csh:
export CHARMM_PLUGIN_DIR=<CHARMM root directory>/lib/osx/openmm_plugins
INSTALLING CHARMM/OpenMM
========================
Installing CHARMM with the OpenMM interface is straightforward on Linux and
Mac OSX:
Setup necesssary environment variables: Set OPENMM_PLUGIN_DIR as described
above, and set CUDATK to the location of the CUDA toolkit. Assuming OpenMM
and CUDA have been installed in their default directories
(/usr/local/openmm and /usr/local/cuda respectively), then set the following
environment variables:
Mac OSX or Linux
bash shell:
export OPENMM_PLUGIN_DIR=/usr/local/openmm/lib/plugins
export CUDATK=/usr/local/cuda
csh shell:
setenv OPENMM_PLUGIN_DIR /usr/local/openmm/lib/plugins
export CUDATK=/usr/local/cuda
These should be added to your .bashrc or .cshrc to ensure they are always
setup.
Linux / Intel compilers
install.com em64t openmm
Linux / GCC compilers
install.com gnu openmm
Mac OSX / Intel compilers
install.com osx ifort openmm
Mac OSX / GCC compilers
install.com osx gfortran openmm
The OpenMM anisotropic barostat plugin is available for use by default. The CharmmNB plugin allows CHARMM to set certain Ewald summation parameters directly. Without this plugin, by default, OpenMM calculates the parameters using an error tolerance. If OpenMM's default method is sufficient, one can add the keyword nonbplugin to the above install.com command lines.
Note: To enable compatibility with OpenMM version 5.1, please replace the openmm key work on the install.com command line above with openmm51. At present we are supporting features for OpenMM 5.2 and OpenMM 5.1
releases. These releases provide slightly different interfaces to Cuda-based
computations and precision. See comments below for more details.
Top
USAGE and IMPLEMENTATION
========================
USAGE: add keyword omm to dynamics command» dynamc or to energy call
(via the energy or gete commands» energy . For dynamics, this
gives you the default Verlet Leapfrog integrator with timestep specified in the dynamics
command. For energy/gete calls, you get all energy terms that are active with the non-bonded and
bonded terms being computed on the GPU. One can also include the various options noted below:
SUMMARY OF OPENMM COMMANDS
==========================
OMM [ openmm-control-spec ]
openmm-control-spec
on Sets omm_active to true and tells CHARMM all subsequent calls to energy,
dynamics or minimization will use OpenMM interface for calculation of
supported energies and forces. OpenMM context will be created later as needed
off Sets omm_active to false but retains any OpenMM context already created
clear Sets omm_ative to false and destroys the OpenMM Context
serialize [unit <unit #>] Serializes the current OpenMM system as an XML string. The string
is output to OUTU in the case of missing [unit <unit #>] and
is output to unit <unit #> otherwise
nb <on|off> Turns on/off use of the CharmmNBForce plugin which allows the PME/Ewald
parameters to be input directly into the OpenMM system instead of being
approximated from the OpenMM error tolerance
setting up GBSA OBC2
gbsa [<gboff/gbon> uueps <real> vveps <real>]
Enable gbsa module (subsequently turn it off w/ gboff or
on w/ gbon) set dielectric for solute (uueps, default 1) or
solvent (vveps, default 78.5).
Dynamics keyword options in CHARMM/OpenMM interface
keyword default action
=========================================================================================
omm false - dynamics keyword to access openMM interface
langevin false - dynamics keyword to turn on Langevin integration
andersen false - dynamics keyword to turn on Andersen heatbath
prmc false - dynamics keyword to turn on MC barostat
variable false - dynamics keyword to use variable timestep md
gamma 5.0 - Langevin friction coefficient in ps^-1
colfrq 1000 - Andersen heatbath coupling constant ps^-1
pref 1.0 - MC barostat reference pressure in atmospheres
prxx, pryy, przz 1.0 - MC barostat reference pressure in atmospheres
tens 0.0 - MC barostat reference surface tension (in dynes/cm^2)
iprsfrq 25 - MC barostat sampling frequency
vtol 1e-3 - Variable timestep error tolerence
NOTE: Coordinates, velocities and restart files can be written every NSAVC, NSAVV, ISVFRQ
timesteps to files specified by IUNCRD, IUNVEL and IUNWRI. Restarts can be used specifying
RESTSRT in the dynamics command with IUNREA also specified, like normal CHARMM runs.
WARNING: At present the energy file is not written, since OpenMM only returns the total energies
(TOTE, TOTKE, EPOT and TEMP) and VOLUME.
Constant Temperature Dynamics
=============================
omm langevin gamma <real> - runs Langevin dynamics with friction coefficient gamma (ps^1)
<5.0> at a temperature given by finalt in dynamics command.
omm andersen colfreq <integer> - runs constant T with Andersen collision frequency colfrq
<1000> at temperature given by finalt in dynamics command
Constant Pressure/Constant Temperature Dynamics
===============================================
Using either of the integrators noted above, one can run MC barostat-ed molecular dynamics by
adding:
omm langevin gamma <real> prmc pref <real> iprsfrq <integer> - runs Langevin dynamics with
<5.0> <1.0> <25> barostat with a reference
pressure of pref atmospheres
and MC volume move attempted
every iprsfq steps.
In addition, we have implemented a varient of the anisortopic barostat that enables constant surface
tension, constant surface area and related ensembles to be simulated. The freedom in changing the
size/shape of the box is related to the crystal space group chosen as well as the type of barostat.
The relevant commands are:
omm prmc przz <real> iprsfrq <integer> - runs constant normal pressure and
constant surface area
omm prmc prxx <real> pryy <real> iprsfrq <integer> - constant z dimension, constant pressure
in independent tangential x, y dimensions
omm prmc tens <real> iprsfrq <integer> - constant surface tension and constant volume
x,y degrees of freedom coupled to surface
tension, z changes to maintain constant volume
omm prmc tens <real> przz <real> iprsfrq <integer> - constant surface tension and normal pressure
Variable Timestep Molecular Dynamics
====================================
OpenMM has implemented a bounded error estimate driven variable timestep integration scheme
in which the size of the timestep is bounded by a specified error that would be associated with
the explicit Euler integrator. The timestep is chosen to satisfy the following relationship
error = dt^2 Sum_i ( |f_i|/m_i ),
where error is the desired maximum error in the step, given the current forces. From the
user-supplied error, the timestep follows from
dt = sqrt( error / Sum_i ( |f_i|/m_i ) )
Adding variable_timestep vtol <real> (default 1.0e-3) uses a variable timestep version of the
above integrators (Langevin or Leapfrog). One can run NVE dynamics with Leapfrog as well, but
this may be not useful.
One can also use the variable timestep algorithms with the barostat.
Energy Computations
===================
Energy terms supported through the CHARMM/OpenMM interface for computation
on the GPU include: BOND ANGL UREY DIHE IMPR VDW ELEC IMNB IMEL EWKS EWSE EWEX,
HARM and ETEN. However, these are returned from the CHARMM/OpenMM interface as just ENER,
i.e., the sum of the components. One can evaluate the individual components through
use of the SKIPE commands.
=================
The CHARMM/OpenMM interface supports a subset of the CONS HARM harmonic restraints.
Speciically, the default ABSOLUTE restraints with XSCALE=YSCALE=ZSCALE=1 are
supported. The COMP, WEIGHT and MASS keywords associated with this restraint are
also supported (» cons and testcase
c37test/3ala_openmm_restraints.inp)
The CHARMM/OpenMM interface supports the CONS RESDistance restraints.
(» cons and testcase c39test/omm_resdtest.inp)
The CHARMM/OpenMM interface supports the CONS DIHEdral restraints.
(» cons and testcase c39test/omm_consdihe.inp)
The CHARMM/OpenMM interface can carry-out energy calculations that combine
the forces for energy terms computed on the GPU (non-bonded (VDW/ELEC) and
bonded (BOND, ANGL, DIHE, IMPHI) with those from other CHARMM functionality.
At present, aside from doing a static energy/force evaluation, one cannot use
it is planned that we will support this functionality in the future).
NOTE: When using single precision arithmetic on the GPU, long NVE simulations may
have an energy drift on the order of 10^-2 * KBOLTZ * T / NDEGF per nanosecond.
However, with the release of OpenMM 5.0, both mixed and double precision models
are available in both CUDA and OpenCL.
As noted in the overview above, the CHARMM/OpenMM supports "no frills"
molecular dynamics for periodic and non-periodic systems. For non-periodic
systems cutoffs and no-cutoffs are supported. For cutoff based methods
a reaction field can be utilized. This is also true for periodic systems that
don't employ PME/Ewald methods. The cutoff method is keyed to the value of
the energy-related cutoff CTOFNB. If CTOFNB > 990 it's assumed that no
cutoffs are to be used and OpenMM computes all interactions for non-periodic
systems. If CTOFNB < 990, and other truncation methods (set note below) are not
specified then the solvent reaction field is used with a
cutoff switch such that the electrostatic energy for atom pair ij, u_ij, is
given by:
q_i*q_j / 1 \
u_ij = ---------- .| ---- + k_rf*r^2 -c_rf |
4*pi*eps_0 \ r_ij /
k_rf = (eps_solvwnt - 1)/(2*eps_solvent+1)/(r_cutoff)^3
c_rf = (3*eps_solvent)/(2*eps_solvent+1)/(r_cutoff)
where r_cutoff is the cutoff distance (CTOFNB) and eps_solvent is the dielectric
constant of the solvent. If eps_solvent >> 1, this causes the forces to go to
zero at the cutoff.
The CHARMM/OpenMM Generalized solvent reaction field can also be specified on any
nonbond/energy/dynamic command as:
energy omrf omrx <value> - default 1
Energy based cutoff methods
============================
Other energy based methods have recently been implemented. The CHARMM/OpenMM interface
now supports the folowing combinations of van der Waals and electrostatic methods:
Supported combinations of energy methods
----------------------------------------
**pme/ewald <specification> ctonnb <value> ctofnb <value> vatom vswitch/vfswitch
*noewald atom switch vatom vswitch ctonnb <value> ctofnb <value>
*noewald atom switch vatom vfswitch ctonnb <value> ctofnb <value>
*noewald atom fswitch vatom vfswitch ctonnb <value> ctofnb <value>
*noewald atom fshift vatom vswitch ctonnb <value> ctofnb <value>
*noewald atom fshift vatom vfswitch ctonnb <value> ctofnb <value>
*noewald atom omrxfld vatom vswitch/vfswitch ctonnb <value> ctofnb <value>
OpenMM now supports a vdW switching function pf the form S = 1 - 6x^5 + 15x^4 - 10x^3
where x = (r-r_switch)/(r_cutoff - r_switch) and r_cutoff = ctofnb and r_switch = ctonnb.
Also supported is the long range vdW correction that is keyed from the LRC command
Summary of supported non-bond truncation methods:
For van der Waals: VSWITCH, VFSWITCH, OMSW LRC
For electrostatics: SWITCH, FSWITCH, FSHIFT, OMRF [OMRX <RxnFld_dielectric (default 1) >]
*Note, non-Ewald/PME calculations are supported for periodic and non-periodic
systems.
**Note, see Ewald/PME discussion below.
Ewald and PME support
---------------------
This is deprecated, the CHARMM/OpenMM Interface now supports direct input
of Kappa, fftx, ffty, fftz from CHARMM energy/dynamic/nonbond command. If this is not required, one can install using the nonbplugin keyword described above.
Ewald and PME-based Ewald are both implemented. With PME-based Ewald the
OpenMM interface employs the cutoff (CTOFNB), the box length, and an
estimated error desired for the long-range electrostatic forces to
determine the number of number of grid points for the PME calculations,
FFTX, FFTY and FFTZ. However, to maintain consistency with CHARMM, the
the error estimate and KAPPA. Thus, KAPPA as set in the CHARMM energy/nonbond
or dynamics command may be over-ridden to ensure that FFTX(Y,Z) is maintained
as requested. The error estimate is defined as delta and is related to KAPPA
via the relaitonship
delta = exp[-(KAPPA*CUTNB)^2) (1)
In the current implementation, this relationship in the form
KAPPA = Sqrt(-ln(2*delta))/CUTNB (2)
is combined with
FFTX(Y,Z) = 2*KAPPA*box_x(y,z)/(3*delta^(1/5)) (3)
to eliminate KAPPA and is solved (via bisection) for a delta value that will
yield the user provided values of FFTX(Y,Z) and KAPPA is determined from
the relationship (2) above.
USAGE and IMPLEMENTATION
========================
USAGE: add keyword omm to dynamics command» dynamc or to energy call
(via the energy or gete commands» energy . For dynamics, this
gives you the default Verlet Leapfrog integrator with timestep specified in the dynamics
command. For energy/gete calls, you get all energy terms that are active with the non-bonded and
bonded terms being computed on the GPU. One can also include the various options noted below:
SUMMARY OF OPENMM COMMANDS
==========================
OMM [ openmm-control-spec ]
openmm-control-spec
on Sets omm_active to true and tells CHARMM all subsequent calls to energy,
dynamics or minimization will use OpenMM interface for calculation of
supported energies and forces. OpenMM context will be created later as needed
off Sets omm_active to false but retains any OpenMM context already created
clear Sets omm_ative to false and destroys the OpenMM Context
serialize [unit <unit #>] Serializes the current OpenMM system as an XML string. The string
is output to OUTU in the case of missing [unit <unit #>] and
is output to unit <unit #> otherwise
nb <on|off> Turns on/off use of the CharmmNBForce plugin which allows the PME/Ewald
parameters to be input directly into the OpenMM system instead of being
approximated from the OpenMM error tolerance
setting up GBSA OBC2
gbsa [<gboff/gbon> uueps <real> vveps <real>]
Enable gbsa module (subsequently turn it off w/ gboff or
on w/ gbon) set dielectric for solute (uueps, default 1) or
solvent (vveps, default 78.5).
Dynamics keyword options in CHARMM/OpenMM interface
keyword default action
=========================================================================================
omm false - dynamics keyword to access openMM interface
langevin false - dynamics keyword to turn on Langevin integration
andersen false - dynamics keyword to turn on Andersen heatbath
prmc false - dynamics keyword to turn on MC barostat
variable false - dynamics keyword to use variable timestep md
gamma 5.0 - Langevin friction coefficient in ps^-1
colfrq 1000 - Andersen heatbath coupling constant ps^-1
pref 1.0 - MC barostat reference pressure in atmospheres
prxx, pryy, przz 1.0 - MC barostat reference pressure in atmospheres
tens 0.0 - MC barostat reference surface tension (in dynes/cm^2)
iprsfrq 25 - MC barostat sampling frequency
vtol 1e-3 - Variable timestep error tolerence
NOTE: Coordinates, velocities and restart files can be written every NSAVC, NSAVV, ISVFRQ
timesteps to files specified by IUNCRD, IUNVEL and IUNWRI. Restarts can be used specifying
RESTSRT in the dynamics command with IUNREA also specified, like normal CHARMM runs.
WARNING: At present the energy file is not written, since OpenMM only returns the total energies
(TOTE, TOTKE, EPOT and TEMP) and VOLUME.
Constant Temperature Dynamics
=============================
omm langevin gamma <real> - runs Langevin dynamics with friction coefficient gamma (ps^1)
<5.0> at a temperature given by finalt in dynamics command.
omm andersen colfreq <integer> - runs constant T with Andersen collision frequency colfrq
<1000> at temperature given by finalt in dynamics command
Constant Pressure/Constant Temperature Dynamics
===============================================
Using either of the integrators noted above, one can run MC barostat-ed molecular dynamics by
adding:
omm langevin gamma <real> prmc pref <real> iprsfrq <integer> - runs Langevin dynamics with
<5.0> <1.0> <25> barostat with a reference
pressure of pref atmospheres
and MC volume move attempted
every iprsfq steps.
In addition, we have implemented a varient of the anisortopic barostat that enables constant surface
tension, constant surface area and related ensembles to be simulated. The freedom in changing the
size/shape of the box is related to the crystal space group chosen as well as the type of barostat.
The relevant commands are:
omm prmc przz <real> iprsfrq <integer> - runs constant normal pressure and
constant surface area
omm prmc prxx <real> pryy <real> iprsfrq <integer> - constant z dimension, constant pressure
in independent tangential x, y dimensions
omm prmc tens <real> iprsfrq <integer> - constant surface tension and constant volume
x,y degrees of freedom coupled to surface
tension, z changes to maintain constant volume
omm prmc tens <real> przz <real> iprsfrq <integer> - constant surface tension and normal pressure
Variable Timestep Molecular Dynamics
====================================
OpenMM has implemented a bounded error estimate driven variable timestep integration scheme
in which the size of the timestep is bounded by a specified error that would be associated with
the explicit Euler integrator. The timestep is chosen to satisfy the following relationship
error = dt^2 Sum_i ( |f_i|/m_i ),
where error is the desired maximum error in the step, given the current forces. From the
user-supplied error, the timestep follows from
dt = sqrt( error / Sum_i ( |f_i|/m_i ) )
Adding variable_timestep vtol <real> (default 1.0e-3) uses a variable timestep version of the
above integrators (Langevin or Leapfrog). One can run NVE dynamics with Leapfrog as well, but
this may be not useful.
One can also use the variable timestep algorithms with the barostat.
Energy Computations
===================
Energy terms supported through the CHARMM/OpenMM interface for computation
on the GPU include: BOND ANGL UREY DIHE IMPR VDW ELEC IMNB IMEL EWKS EWSE EWEX,
HARM and ETEN. However, these are returned from the CHARMM/OpenMM interface as just ENER,
i.e., the sum of the components. One can evaluate the individual components through
use of the SKIPE commands.
=================
The CHARMM/OpenMM interface supports a subset of the CONS HARM harmonic restraints.
Speciically, the default ABSOLUTE restraints with XSCALE=YSCALE=ZSCALE=1 are
supported. The COMP, WEIGHT and MASS keywords associated with this restraint are
also supported (» cons and testcase
c37test/3ala_openmm_restraints.inp)
The CHARMM/OpenMM interface supports the CONS RESDistance restraints.
(» cons and testcase c39test/omm_resdtest.inp)
The CHARMM/OpenMM interface supports the CONS DIHEdral restraints.
(» cons and testcase c39test/omm_consdihe.inp)
The CHARMM/OpenMM interface can carry-out energy calculations that combine
the forces for energy terms computed on the GPU (non-bonded (VDW/ELEC) and
bonded (BOND, ANGL, DIHE, IMPHI) with those from other CHARMM functionality.
At present, aside from doing a static energy/force evaluation, one cannot use
it is planned that we will support this functionality in the future).
NOTE: When using single precision arithmetic on the GPU, long NVE simulations may
have an energy drift on the order of 10^-2 * KBOLTZ * T / NDEGF per nanosecond.
However, with the release of OpenMM 5.0, both mixed and double precision models
are available in both CUDA and OpenCL.
As noted in the overview above, the CHARMM/OpenMM supports "no frills"
molecular dynamics for periodic and non-periodic systems. For non-periodic
systems cutoffs and no-cutoffs are supported. For cutoff based methods
a reaction field can be utilized. This is also true for periodic systems that
don't employ PME/Ewald methods. The cutoff method is keyed to the value of
the energy-related cutoff CTOFNB. If CTOFNB > 990 it's assumed that no
cutoffs are to be used and OpenMM computes all interactions for non-periodic
systems. If CTOFNB < 990, and other truncation methods (set note below) are not
specified then the solvent reaction field is used with a
cutoff switch such that the electrostatic energy for atom pair ij, u_ij, is
given by:
q_i*q_j / 1 \
u_ij = ---------- .| ---- + k_rf*r^2 -c_rf |
4*pi*eps_0 \ r_ij /
k_rf = (eps_solvwnt - 1)/(2*eps_solvent+1)/(r_cutoff)^3
c_rf = (3*eps_solvent)/(2*eps_solvent+1)/(r_cutoff)
where r_cutoff is the cutoff distance (CTOFNB) and eps_solvent is the dielectric
constant of the solvent. If eps_solvent >> 1, this causes the forces to go to
zero at the cutoff.
The CHARMM/OpenMM Generalized solvent reaction field can also be specified on any
nonbond/energy/dynamic command as:
energy omrf omrx <value> - default 1
Energy based cutoff methods
============================
Other energy based methods have recently been implemented. The CHARMM/OpenMM interface
now supports the folowing combinations of van der Waals and electrostatic methods:
Supported combinations of energy methods
----------------------------------------
**pme/ewald <specification> ctonnb <value> ctofnb <value> vatom vswitch/vfswitch
*noewald atom switch vatom vswitch ctonnb <value> ctofnb <value>
*noewald atom switch vatom vfswitch ctonnb <value> ctofnb <value>
*noewald atom fswitch vatom vfswitch ctonnb <value> ctofnb <value>
*noewald atom fshift vatom vswitch ctonnb <value> ctofnb <value>
*noewald atom fshift vatom vfswitch ctonnb <value> ctofnb <value>
*noewald atom omrxfld vatom vswitch/vfswitch ctonnb <value> ctofnb <value>
OpenMM now supports a vdW switching function pf the form S = 1 - 6x^5 + 15x^4 - 10x^3
where x = (r-r_switch)/(r_cutoff - r_switch) and r_cutoff = ctofnb and r_switch = ctonnb.
Also supported is the long range vdW correction that is keyed from the LRC command
Summary of supported non-bond truncation methods:
For van der Waals: VSWITCH, VFSWITCH, OMSW LRC
For electrostatics: SWITCH, FSWITCH, FSHIFT, OMRF [OMRX <RxnFld_dielectric (default 1) >]
*Note, non-Ewald/PME calculations are supported for periodic and non-periodic
systems.
**Note, see Ewald/PME discussion below.
Ewald and PME support
---------------------
This is deprecated, the CHARMM/OpenMM Interface now supports direct input
of Kappa, fftx, ffty, fftz from CHARMM energy/dynamic/nonbond command. If this is not required, one can install using the nonbplugin keyword described above.
Ewald and PME-based Ewald are both implemented. With PME-based Ewald the
OpenMM interface employs the cutoff (CTOFNB), the box length, and an
estimated error desired for the long-range electrostatic forces to
determine the number of number of grid points for the PME calculations,
FFTX, FFTY and FFTZ. However, to maintain consistency with CHARMM, the
the error estimate and KAPPA. Thus, KAPPA as set in the CHARMM energy/nonbond
or dynamics command may be over-ridden to ensure that FFTX(Y,Z) is maintained
as requested. The error estimate is defined as delta and is related to KAPPA
via the relaitonship
delta = exp[-(KAPPA*CUTNB)^2) (1)
In the current implementation, this relationship in the form
KAPPA = Sqrt(-ln(2*delta))/CUTNB (2)
is combined with
FFTX(Y,Z) = 2*KAPPA*box_x(y,z)/(3*delta^(1/5)) (3)
to eliminate KAPPA and is solved (via bisection) for a delta value that will
yield the user provided values of FFTX(Y,Z) and KAPPA is determined from
the relationship (2) above.
Top
Control over the number of GPU devices one uses, the platform for GPU-based
computations and the precision model is available through environment variables
Environment variable Setting Effect
OPENMM_DEVICE 0/1/0,1 Use device 0/1/0 and 1 (parallel)
OPENMM_PLATFORM OpenCL/CUDA/Reference Use OpenMM platform based on
OpenCL/Cuda/Reference(CPU)
OPENMM_PRECISION single/mixed/double Do calculations in single (fastest), mixed
or double precision (slowest) on platform
CUDA_COMPILER path to cuda compiler nvcc (usually /usr/local/cuda/bin)
Note: OpenMM chooses a default platform based on a guess for best performance if not is
specified with the environment variable. Also, the platform Reference is a cpu-based
platform for testing/validation purposes. Finally, the different precision models are not
supported in OpenMM 4.1.1 (only single precision).
Note: As of this release (02/15/2013) there is an issue with OpenCL on Mac OSX 10.7.8 and beyond,
the OpenMM team and Apple are discussing these problems, and there will hopefully be resolution soon.
Example (C-shell) setenv OPENMM_DEVICE 0,1 # Use both GPU devices
Note: OpenMM 4.1.1 supports parallel calculations (0,1) only for platform OpenCL.
OpenMM 5.0 adds parallel support for platform CUDA.
Control over the number of GPU devices one uses, the platform for GPU-based
computations and the precision model is available through environment variables
Environment variable Setting Effect
OPENMM_DEVICE 0/1/0,1 Use device 0/1/0 and 1 (parallel)
OPENMM_PLATFORM OpenCL/CUDA/Reference Use OpenMM platform based on
OpenCL/Cuda/Reference(CPU)
OPENMM_PRECISION single/mixed/double Do calculations in single (fastest), mixed
or double precision (slowest) on platform
CUDA_COMPILER path to cuda compiler nvcc (usually /usr/local/cuda/bin)
Note: OpenMM chooses a default platform based on a guess for best performance if not is
specified with the environment variable. Also, the platform Reference is a cpu-based
platform for testing/validation purposes. Finally, the different precision models are not
supported in OpenMM 4.1.1 (only single precision).
Note: As of this release (02/15/2013) there is an issue with OpenCL on Mac OSX 10.7.8 and beyond,
the OpenMM team and Apple are discussing these problems, and there will hopefully be resolution soon.
Example (C-shell) setenv OPENMM_DEVICE 0,1 # Use both GPU devices
Note: OpenMM 4.1.1 supports parallel calculations (0,1) only for platform OpenCL.
OpenMM 5.0 adds parallel support for platform CUDA.
Top
USING BLOCK IN CHARMM/OPENMM
============================
Many features of the CHARMM Block facility have been implemented on the GPU using the CHARMM/OpenMM
interface. However, this has been done with some restrictions on the manner in which the interaction
scaling is done.
Block is used as detailed in the Block documentaiton for setting up the scaling of different terms.
As with its implementation in CHARMM, one can scale the vdW, elec, bond, angle, dihed, impr terms
independent of one another, although negative values of the block coefficients are not allowed due
to a limitation in the manner in which the scaling is implemented. (Scaling is implemented by scaling
corresponding force constants as the system is being set-up, this the affected bond, angle, dihedral
and improper force constants are scaled by the block coefficient. The charges of atoms in a given block
are also scaled by their block coefficient and the L-J emin value is scaled by the block coefficient
squared, such that when the standard combination rules are applied the scaling of the interaction
is by the block coefficient to the first power. The intra-block terms are treated differently for
vdW and elec to ensure they match the CHARMM energies.)
The rmla command to remove specific terms from the block scaling also works as it does in CHARMM.
This implementaiton is fully suitable for running TI calculations or TP calculations. The trajectories
generated at a particular lambda value can be post-processed either using the GPU-based CHARMM/OpenMM
machinery, in which case the restrictions noted above apply and two separate passes must be carried out
to get the "reactant" and "product" energies needed to construct the TI integratnd or the TP free energy
increment. Alternatively, except if the OpenMM reaction field was employed, the trajectories can be
post-processed using the the CHARMM Block facility as detailed in block.doc
(» block ).
USING BLOCK IN CHARMM/OPENMM
============================
Many features of the CHARMM Block facility have been implemented on the GPU using the CHARMM/OpenMM
interface. However, this has been done with some restrictions on the manner in which the interaction
scaling is done.
Block is used as detailed in the Block documentaiton for setting up the scaling of different terms.
As with its implementation in CHARMM, one can scale the vdW, elec, bond, angle, dihed, impr terms
independent of one another, although negative values of the block coefficients are not allowed due
to a limitation in the manner in which the scaling is implemented. (Scaling is implemented by scaling
corresponding force constants as the system is being set-up, this the affected bond, angle, dihedral
and improper force constants are scaled by the block coefficient. The charges of atoms in a given block
are also scaled by their block coefficient and the L-J emin value is scaled by the block coefficient
squared, such that when the standard combination rules are applied the scaling of the interaction
is by the block coefficient to the first power. The intra-block terms are treated differently for
vdW and elec to ensure they match the CHARMM energies.)
The rmla command to remove specific terms from the block scaling also works as it does in CHARMM.
This implementaiton is fully suitable for running TI calculations or TP calculations. The trajectories
generated at a particular lambda value can be post-processed either using the GPU-based CHARMM/OpenMM
machinery, in which case the restrictions noted above apply and two separate passes must be carried out
to get the "reactant" and "product" energies needed to construct the TI integratnd or the TP free energy
increment. Alternatively, except if the OpenMM reaction field was employed, the trajectories can be
post-processed using the the CHARMM Block facility as detailed in block.doc
(» block ).
Top
USING GBSA OBC2 IN CHARMM/OPENMM
================================
OpenMM provides the facility to implement the GBSA OBC2 model from Onufriev, Bashford and Case. This interface
has been "opened" for use through the CHARMM/OpenMM interface. The relevant parameters to run these calculations
are the inherent atomic radii and scaling constants. These parameters for protein atoms contained in the
par/top_all36_prot models have been incorporated into the file toppar/openmm_gbsaobc2/charmm_all36_prot_gbsaobc.str.
Streaming this file just before calling gbsa in the CHARMM/OpenMM interface puts the radii and scale factors
into the wmain and wcomp arrays, respectively, making the subsequent call now prepared to upload the data
to the CHARMM/OpenMM interace and thus run Generalzied Born calculations through the interface.
A few support files are also included to 1) enable one to run the serialized CHARMM/OpenMM setup through the
python API in OpenMM (see file tool/OpenMMFiles/omm_gbsaobc-test.py) and 2) enable one to extract radii and
scaling factors for other Amber force fields (using the OpenMM supplied xml files for those force fields that
reside in OpenMMn.nn-Mac/python/simtk/openmm/app/data/) using the awk script getff.awk (see file
tool/OpenMMFiles/getff.awk).
» openmm
Test case: test/c39test/omm_gbsaobc_streamn-test.inp
USING GBSA OBC2 IN CHARMM/OPENMM
================================
OpenMM provides the facility to implement the GBSA OBC2 model from Onufriev, Bashford and Case. This interface
has been "opened" for use through the CHARMM/OpenMM interface. The relevant parameters to run these calculations
are the inherent atomic radii and scaling constants. These parameters for protein atoms contained in the
par/top_all36_prot models have been incorporated into the file toppar/openmm_gbsaobc2/charmm_all36_prot_gbsaobc.str.
Streaming this file just before calling gbsa in the CHARMM/OpenMM interface puts the radii and scale factors
into the wmain and wcomp arrays, respectively, making the subsequent call now prepared to upload the data
to the CHARMM/OpenMM interace and thus run Generalzied Born calculations through the interface.
A few support files are also included to 1) enable one to run the serialized CHARMM/OpenMM setup through the
python API in OpenMM (see file tool/OpenMMFiles/omm_gbsaobc-test.py) and 2) enable one to extract radii and
scaling factors for other Amber force fields (using the OpenMM supplied xml files for those force fields that
reside in OpenMMn.nn-Mac/python/simtk/openmm/app/data/) using the awk script getff.awk (see file
tool/OpenMMFiles/getff.awk).
» openmm
Test case: test/c39test/omm_gbsaobc_streamn-test.inp
Top
EXAMPLES
========
Molecular dynamics using NVE with PME in a cubic system (from JACS Benchmark):
set nsteps = 1000
set cutoff = 11
set ctofnb = 8
set ctonnb = 7.5
set kappa = 0.3308 ! Consistent with cutofnb and fftx,y,z
calc cutim = @cutoff
! Dimension of a box
set size 62.23
set theta = 90.0
! Dimension of a box
Crystal define cubic @size @size @size @theta @theta @theta
crystal build cutoff @cutim noper 0
image byseg xcen 0.0 ycen 0.0 zcen 0.0 select segid 5dfr end
image byres xcen 0.0 ycen 0.0 zcen 0.0 select segid wat end
! turn on faster options and set-up SHAKE
faster on
energy eps 1.0 cutnb @cutoff cutim @cutim -
ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64
shake fast bonh tol 1.0e-8 para
set echeck = echeck -1
open unit 20 write form restart.res
! Run NVE dynamics, write restart file
calc nwrite = int ( @nsteps / 10 )
! Run dynamics in periodic box
dynamics leap start timestep 0.002 -
nstep @nsteps nprint @nwrite iprfrq @nwrite isvfrq @nsteps iunwri 20 -
firstt 298 finalt 298 -
ichecw 0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck bycb -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 ntrfq @nsteps - !PME
omm ! Just turn on openMM, get Leapfrog Verlet, NVE
! Restart dynamics from current file
! Run dynamics in periodic box
dynamics leap restart timestep 0.002 -
nstep @nsteps nprint @nwrite iprfrq @nwrite isvfrq @nsteps iunwri 20 iunrea 20 -
firstt 298 finalt 298 -
ichecw 0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck bycb -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 ntrfq @nsteps - !PME
omm ! Just turn on openMM, get Leapfrog Verlet, NVE
!!!!!!!!!!!!!!!!!!!LANGEVIN HEATBATH NVT!!!!!!!!!!!!!!!!!!!!!
! Run NVT dynamics with Langevin heatbath, gamma = 10 ps^-1
! Run dynamics in periodic box
dynamics leap start timestep 0.002 -
nstep @nsteps nprint @nwrite iprfrq @nwrite isvfrq @nsteps iunwri 20 -
firstt 298 finalt 298 -
ichecw 0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck bycb -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 ntrfq @nsteps - !PME
omm langevin gamma 10 ! turn on openmm, set-up Langevin
! Run variable timestep Langevin dynamics with error tolerance of 3e-3
! Run dynamics in periodic box
dynamics leap start timestep 0.002 -
nstep @nsteps nprint @nwrite iprfrq @nwrite isvfrq @nsteps iunwri 20 -
firstt 298 finalt 298 -
ichecw 0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck bycb -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 ntrfq @nsteps - !PME
omm langevin gamma 10 variable vtol 3e-3 ! turn on openmm, set-up variable
! timestep Langevin dynamics
!!!!!!!!!!!!!!!!!!!LANGEVIN HEATBATH/MC BAROSTAT NPT!!!!!!!!!!!!!!!!!!!!!
! Run NPT dynamics with Langevin heatbath, gamma = 10 ps^-1
! Run dynamics in periodic box
dynamics leap start timestep 0.002 -
nstep @nsteps nprint @nwrite iprfrq @nwrite isvfrq @nsteps iunwri 20 -
firstt 298 finalt 298 -
ichecw 0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck bycb -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 ntrfq @nsteps - !PME
omm langevin gamma 10 - ! turn on openmm, set-up Langevin
mcpr pref 1 iprsfrq 25 ! set-up MC barostat at 1 atm, move attempt / 25 steps
! Run variable timestep Langevin dynamics with error tolerance of 3e-3
! Run dynamics in periodic box
dynamics leap start timestep 0.002 -
nstep @nsteps nprint @nwrite iprfrq @nwrite isvfrq @nsteps iunwri 20 -
firstt 298 finalt 298 -
ichecw 0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck bycb -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 ntrfq @nsteps - !PME
omm langevin gamma 10 variable vtol 3e-3 - ! turn on openmm, set-up variable
- ! timestep Langevin dynamics
mcpr pref 1 iprsfrq 25 ! set-up MC barostat at 1 atm, move attempt / 25 steps
!!!!!!!!!!!!!!!!!!!ANDERSEN HEATBATH NVT!!!!!!!!!!!!!!!!!!!!!
! Run NVT dynamics with Andersen heatbath, collision frequency = 250
! Run dynamics in periodic box
dynamics leap start timestep 0.002 -
nstep @nsteps nprint @nwrite iprfrq @nwrite isvfrq @nsteps iunwri 20 -
firstt 298 finalt 298 -
ichecw 0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck bycb -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 ntrfq @nsteps - !PME
omm andersen colfrq 250 - ! turn on openmm, set-up Andersen
mcpr pref 1 iprsfrq 25 ! set-up MC barostat at 1 atm, move attempt / 25 steps
! Run variable timestep Leapfrog w/ Andersen heatbath and error tolerance of 2e-3
! Run dynamics in periodic box
dynamics leap start timestep 0.002 -
nstep @nsteps nprint @nwrite iprfrq @nwrite isvfrq @nsteps iunwri 20 -
firstt 298 finalt 298 -
ichecw 0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck bycb -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 ntrfq @nsteps - !PME
omm andersen colfrq 250 variable vtol 3e-3 ! turn on openmm, set-up variable
! timestep Langevin dynamics
!!!!!!!!!!!!!!!!!!!ANDERSEN HEATBATH/MC BAROSTAT NPT!!!!!!!!!!!!!!!!!!!!!
! Run NPT dynamics with Andersen heatbath, collision frequency = 250
! Run dynamics in periodic box
dynamics leap start timestep 0.002 -
nstep @nsteps nprint @nwrite iprfrq @nwrite isvfrq @nsteps iunwri 20 -
firstt 298 finalt 298 -
ichecw 0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck bycb -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 ntrfq @nsteps - !PME
omm andersen colfrq 250 - ! turn on openmm, set-up Andersen
mcpr pref 1 iprsfrq 25 ! set-up MC barostat at 1 atm, move attempt / 25 steps
! Run variable timestep Leapfrog w/ Andersen heatbath and error tolerance of 2e-3
! Run dynamics in periodic box
dynamics leap start timestep 0.002 -
nstep @nsteps nprint @nwrite iprfrq @nwrite isvfrq @nsteps iunwri 20 -
firstt 298 finalt 298 -
ichecw 0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck bycb -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 ntrfq @nsteps - !PME
omm andersen colfrq 250 variable vtol 3e-3 - ! turn on openmm, set-up variable
- ! timestep Andersen dynamics
mcpr pref 1 iprsfrq 25 ! set-up MC barostat at 1 atm, move attempt / 25 steps
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!EXAMPLE ENERGY CALCULATIONS!!!!!!!!!!!!!!!!!!!!!!!!
! Use omm on/off/clear to set-up and carry-out energy calculations using CPU and/or GPU
! Energy calculation for periodic system use PME on CPU
energy eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64
! Same calculation using GPU throuhg CHARMM/OpenMM interface
energy eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 -
omm
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!EXAMPLE II!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! Energy calculation for periodic system use PME on CPU
energy eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64
omm on ! subsequent invocations of energy will use CHARMM/OpenMM interface
! Same calculation using GPU through CHARMM/OpenMM interface
energy eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64
omm off ! turn off use of GPU calculation but leave OpenMM "Context" intact
! Energy calculation for periodic system use PME on CPU
energy eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64
omm clear ! Deactivate (until next omm on) calculations using GPU
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!TEST CASES!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
The relevant test cases for the CHARMM/OpenMM functionality are:
Test case Purpose
c37test/
omm_acetate.inp Test whether CHARMM/OpenMM handles nbfixes correctly w/ periodic bcs.
omm_dynam-vts.inp Test CHARMM/OpenMM dynamics with various integrators and variable timestep integration
omm_dynamics.inp Test CHARMM/OpenMM dynamics with various integrators
omm_exception.inp Test whether CHARMM/OpenMM handles nbfixes correctly w/ periodic bcs
omm_modpsf.inp Test whether CHARMM/OpenMM senses psf changes and rebuilds OpenMM context
omm_nbexcl.inp Test whether CHARMM/OpenMM handles nb exclusions correctly
omm_nbfix.inp Test whether CHARMM/OpenMM handles nbfixes correctly
omm_nonperiodic.inp Test and compare CHARMM and CHARMM/OpenMM energy and forces for vacuum system
omm_periodic.inp Test and compare CHARMM and CHARMM/OpenMM energy and forces for solvated system
omm_restraints.inp Test restraint methods between CHARMM and CHARMM/OpenMM for vacuum system
c38test/
omm_block-periodic.inp Test BLOCK commands as implemented in CHARMM/OpenMM interface
omm_block-periodic2.inp Second test of BLOCK commands as implemented in CHARMM/OpenMM interface
omm_block1.inp Test of basic BLOCK commands as implemented in CHARMM/OpenMM interface
omm_fixed.inp Test implementation of fixed atoms in CHARMM/OpenMM interface
omm_go-model.inp Test Karanicolas/Brooks ETEN Go model w/o & w/ periodicity
omm_switch-nbfix.inp Test switch/shift w/ NBFixes functioanlity of CHARMM/OpenMM interface
omm_switch.inp Test switch/shift functioanlity of CHARMM/OpenMM interface
omm_switch14.inp Test switch/shift functioanlity of CHARMM/OpenMM interface
omm_switchpair.inp Test switch/shift functioanlity of CHARMM/OpenMM interface
c39test/
omm_block_ti.inp Test BLOCK-based TI calculations through CHARMM/OpenMM interface
omm_dynamics_baro2.inp Test anisotropic Monte Carlo based barostat plugin
omm_resdtest.inp Test CONS RESDistance restraint implmentation through CHARMM/OpenMM
omm_consdihetest.inp Test CONS DIHEdral restraint implementation through CHARMM/OpenMM
EXAMPLES
========
Molecular dynamics using NVE with PME in a cubic system (from JACS Benchmark):
set nsteps = 1000
set cutoff = 11
set ctofnb = 8
set ctonnb = 7.5
set kappa = 0.3308 ! Consistent with cutofnb and fftx,y,z
calc cutim = @cutoff
! Dimension of a box
set size 62.23
set theta = 90.0
! Dimension of a box
Crystal define cubic @size @size @size @theta @theta @theta
crystal build cutoff @cutim noper 0
image byseg xcen 0.0 ycen 0.0 zcen 0.0 select segid 5dfr end
image byres xcen 0.0 ycen 0.0 zcen 0.0 select segid wat end
! turn on faster options and set-up SHAKE
faster on
energy eps 1.0 cutnb @cutoff cutim @cutim -
ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64
shake fast bonh tol 1.0e-8 para
set echeck = echeck -1
open unit 20 write form restart.res
! Run NVE dynamics, write restart file
calc nwrite = int ( @nsteps / 10 )
! Run dynamics in periodic box
dynamics leap start timestep 0.002 -
nstep @nsteps nprint @nwrite iprfrq @nwrite isvfrq @nsteps iunwri 20 -
firstt 298 finalt 298 -
ichecw 0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck bycb -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 ntrfq @nsteps - !PME
omm ! Just turn on openMM, get Leapfrog Verlet, NVE
! Restart dynamics from current file
! Run dynamics in periodic box
dynamics leap restart timestep 0.002 -
nstep @nsteps nprint @nwrite iprfrq @nwrite isvfrq @nsteps iunwri 20 iunrea 20 -
firstt 298 finalt 298 -
ichecw 0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck bycb -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 ntrfq @nsteps - !PME
omm ! Just turn on openMM, get Leapfrog Verlet, NVE
!!!!!!!!!!!!!!!!!!!LANGEVIN HEATBATH NVT!!!!!!!!!!!!!!!!!!!!!
! Run NVT dynamics with Langevin heatbath, gamma = 10 ps^-1
! Run dynamics in periodic box
dynamics leap start timestep 0.002 -
nstep @nsteps nprint @nwrite iprfrq @nwrite isvfrq @nsteps iunwri 20 -
firstt 298 finalt 298 -
ichecw 0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck bycb -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 ntrfq @nsteps - !PME
omm langevin gamma 10 ! turn on openmm, set-up Langevin
! Run variable timestep Langevin dynamics with error tolerance of 3e-3
! Run dynamics in periodic box
dynamics leap start timestep 0.002 -
nstep @nsteps nprint @nwrite iprfrq @nwrite isvfrq @nsteps iunwri 20 -
firstt 298 finalt 298 -
ichecw 0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck bycb -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 ntrfq @nsteps - !PME
omm langevin gamma 10 variable vtol 3e-3 ! turn on openmm, set-up variable
! timestep Langevin dynamics
!!!!!!!!!!!!!!!!!!!LANGEVIN HEATBATH/MC BAROSTAT NPT!!!!!!!!!!!!!!!!!!!!!
! Run NPT dynamics with Langevin heatbath, gamma = 10 ps^-1
! Run dynamics in periodic box
dynamics leap start timestep 0.002 -
nstep @nsteps nprint @nwrite iprfrq @nwrite isvfrq @nsteps iunwri 20 -
firstt 298 finalt 298 -
ichecw 0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck bycb -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 ntrfq @nsteps - !PME
omm langevin gamma 10 - ! turn on openmm, set-up Langevin
mcpr pref 1 iprsfrq 25 ! set-up MC barostat at 1 atm, move attempt / 25 steps
! Run variable timestep Langevin dynamics with error tolerance of 3e-3
! Run dynamics in periodic box
dynamics leap start timestep 0.002 -
nstep @nsteps nprint @nwrite iprfrq @nwrite isvfrq @nsteps iunwri 20 -
firstt 298 finalt 298 -
ichecw 0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck bycb -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 ntrfq @nsteps - !PME
omm langevin gamma 10 variable vtol 3e-3 - ! turn on openmm, set-up variable
- ! timestep Langevin dynamics
mcpr pref 1 iprsfrq 25 ! set-up MC barostat at 1 atm, move attempt / 25 steps
!!!!!!!!!!!!!!!!!!!ANDERSEN HEATBATH NVT!!!!!!!!!!!!!!!!!!!!!
! Run NVT dynamics with Andersen heatbath, collision frequency = 250
! Run dynamics in periodic box
dynamics leap start timestep 0.002 -
nstep @nsteps nprint @nwrite iprfrq @nwrite isvfrq @nsteps iunwri 20 -
firstt 298 finalt 298 -
ichecw 0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck bycb -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 ntrfq @nsteps - !PME
omm andersen colfrq 250 - ! turn on openmm, set-up Andersen
mcpr pref 1 iprsfrq 25 ! set-up MC barostat at 1 atm, move attempt / 25 steps
! Run variable timestep Leapfrog w/ Andersen heatbath and error tolerance of 2e-3
! Run dynamics in periodic box
dynamics leap start timestep 0.002 -
nstep @nsteps nprint @nwrite iprfrq @nwrite isvfrq @nsteps iunwri 20 -
firstt 298 finalt 298 -
ichecw 0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck bycb -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 ntrfq @nsteps - !PME
omm andersen colfrq 250 variable vtol 3e-3 ! turn on openmm, set-up variable
! timestep Langevin dynamics
!!!!!!!!!!!!!!!!!!!ANDERSEN HEATBATH/MC BAROSTAT NPT!!!!!!!!!!!!!!!!!!!!!
! Run NPT dynamics with Andersen heatbath, collision frequency = 250
! Run dynamics in periodic box
dynamics leap start timestep 0.002 -
nstep @nsteps nprint @nwrite iprfrq @nwrite isvfrq @nsteps iunwri 20 -
firstt 298 finalt 298 -
ichecw 0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck bycb -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 ntrfq @nsteps - !PME
omm andersen colfrq 250 - ! turn on openmm, set-up Andersen
mcpr pref 1 iprsfrq 25 ! set-up MC barostat at 1 atm, move attempt / 25 steps
! Run variable timestep Leapfrog w/ Andersen heatbath and error tolerance of 2e-3
! Run dynamics in periodic box
dynamics leap start timestep 0.002 -
nstep @nsteps nprint @nwrite iprfrq @nwrite isvfrq @nsteps iunwri 20 -
firstt 298 finalt 298 -
ichecw 0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck bycb -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 ntrfq @nsteps - !PME
omm andersen colfrq 250 variable vtol 3e-3 - ! turn on openmm, set-up variable
- ! timestep Andersen dynamics
mcpr pref 1 iprsfrq 25 ! set-up MC barostat at 1 atm, move attempt / 25 steps
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!EXAMPLE ENERGY CALCULATIONS!!!!!!!!!!!!!!!!!!!!!!!!
! Use omm on/off/clear to set-up and carry-out energy calculations using CPU and/or GPU
! Energy calculation for periodic system use PME on CPU
energy eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64
! Same calculation using GPU throuhg CHARMM/OpenMM interface
energy eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64 -
omm
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!EXAMPLE II!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! Energy calculation for periodic system use PME on CPU
energy eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64
omm on ! subsequent invocations of energy will use CHARMM/OpenMM interface
! Same calculation using GPU through CHARMM/OpenMM interface
energy eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64
omm off ! turn off use of GPU calculation but leave OpenMM "Context" intact
! Energy calculation for periodic system use PME on CPU
energy eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64
omm clear ! Deactivate (until next omm on) calculations using GPU
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!TEST CASES!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
The relevant test cases for the CHARMM/OpenMM functionality are:
Test case Purpose
c37test/
omm_acetate.inp Test whether CHARMM/OpenMM handles nbfixes correctly w/ periodic bcs.
omm_dynam-vts.inp Test CHARMM/OpenMM dynamics with various integrators and variable timestep integration
omm_dynamics.inp Test CHARMM/OpenMM dynamics with various integrators
omm_exception.inp Test whether CHARMM/OpenMM handles nbfixes correctly w/ periodic bcs
omm_modpsf.inp Test whether CHARMM/OpenMM senses psf changes and rebuilds OpenMM context
omm_nbexcl.inp Test whether CHARMM/OpenMM handles nb exclusions correctly
omm_nbfix.inp Test whether CHARMM/OpenMM handles nbfixes correctly
omm_nonperiodic.inp Test and compare CHARMM and CHARMM/OpenMM energy and forces for vacuum system
omm_periodic.inp Test and compare CHARMM and CHARMM/OpenMM energy and forces for solvated system
omm_restraints.inp Test restraint methods between CHARMM and CHARMM/OpenMM for vacuum system
c38test/
omm_block-periodic.inp Test BLOCK commands as implemented in CHARMM/OpenMM interface
omm_block-periodic2.inp Second test of BLOCK commands as implemented in CHARMM/OpenMM interface
omm_block1.inp Test of basic BLOCK commands as implemented in CHARMM/OpenMM interface
omm_fixed.inp Test implementation of fixed atoms in CHARMM/OpenMM interface
omm_go-model.inp Test Karanicolas/Brooks ETEN Go model w/o & w/ periodicity
omm_switch-nbfix.inp Test switch/shift w/ NBFixes functioanlity of CHARMM/OpenMM interface
omm_switch.inp Test switch/shift functioanlity of CHARMM/OpenMM interface
omm_switch14.inp Test switch/shift functioanlity of CHARMM/OpenMM interface
omm_switchpair.inp Test switch/shift functioanlity of CHARMM/OpenMM interface
c39test/
omm_block_ti.inp Test BLOCK-based TI calculations through CHARMM/OpenMM interface
omm_dynamics_baro2.inp Test anisotropic Monte Carlo based barostat plugin
omm_resdtest.inp Test CONS RESDistance restraint implmentation through CHARMM/OpenMM
omm_consdihetest.inp Test CONS DIHEdral restraint implementation through CHARMM/OpenMM