ForceField Engine Options¶
Details of the ForceField engine can be set via its input block. Some option are specific to UFF and others to other force fields.
Common options¶
These options apply to any force field.
Type¶
There are a few predefined force field types, that, if used, require no other input.
Type [UFF | Amber95 | GAFF | Tripos5.2 | APPLE&P | UserDefined]
Type
- Type:
Multiple Choice
- Default value:
UFF
- Options:
[UFF, Amber95, GAFF, Tripos5.2, APPLE&P, UserDefined]
- Description:
Type of force field to be used
Non-bonded screening¶
The long range interaction (dispersion and Coulomb) are the most expensive to evaluate. This gives you the option to screen the interaction more aggressively.
NonBondedCutoff float
NonBondedCutoff
- Type:
Float
- Default value:
15.0
- Unit:
Angstrom
- Description:
Distance beyond which the non-bonded pair interactions (Coulomb and Van der Waals) will be ignored. The interactions are smoothly damped starting from 0.9*NonBondedCutoff. Has no effect on the Coulomb term for 3D-periodic systems, as Ewald summation is used.
It is usually a good idea to add some “skin” to the cutoff above when it’s used for computing a neighbor list for changing geometries (e.g. during molecular dynamics or geometry optimization). This way, the neighbor list will not need to be re-computed when atoms move a little. This may save some time because generating a neighbor list can be quite costly. The following option sets the thickness of the “skin”:
NeighborListSkin float
NeighborListSkin
- Type:
Float
- Default value:
2.5
- Unit:
Angstrom
- Description:
Thickness of the buffer region added to the NonBondedCutoff when building a neighbor list.
Note
This option also affects the cutoff used when generating a neighbor list in the real-space part of the Ewald summation but then it is added to the cutoff radius is used there.
Feedback¶
If you want to know more about the details of the force field you should crank up the verbosity.
Verbosity [Silent | Normal | Verbose | VeryVerbose]
Verbosity
- Type:
Multiple Choice
- Default value:
Silent
- Options:
[Silent, Normal, Verbose, VeryVerbose]
- Description:
Controls the verbosity of the engine.
Bonds usage¶
Bonds can be specified in the input, still you may not want to use those. Here are some options to control this.
BondsUsage [Input | None | Guess | Auto]
BondsUsage
- Type:
Multiple Choice
- Default value:
Auto
- Options:
[Input, None, Guess, Auto]
- Description:
Controls what bonds are used by the engine. The choice auto means: guess in case there are no bonds. Guessing only happens at the first MD step, or first geometry optimization step.
Ewald summation¶
For periodic systems the Ewald summation is performed for the Coulomb interaction. It has a couple of options:
EwaldSummation
Alpha float
Enabled Yes/No
GridSpacing float
RealSpaceCutoff float
Tolerance float
End
EwaldSummation
- Type:
Block
- Description:
Configures the details of the particle mesh Ewald (PME) summation of the Coulomb interaction.
Alpha
- Type:
Float
- Default value:
-1.0
- Unit:
1/Angstrom
- Description:
This parameter shifts the workload from real space (smaller alpha) to reciprocal space (larger alpha). Using a larger [Alpha] without decreasing [GridSpacing] may increase the error in the reciprocal-space contribution. Set to zero to disable the reciprocal-space Ewald part. Negative value means the [Alpha] will be determined automatically from the [Tolerance] and [RealSpaceCutoff] values.
Enabled
- Type:
Bool
- Default value:
Yes
- Description:
Set to false to use real-space pair summation instead of the Ewald, which is the default and the only option for molecules, 1D and 2D periodic systems.
GridSpacing
- Type:
Float
- Default value:
0.5
- Unit:
Angstrom
- Description:
Grid spacing in the particle mesh Ewald method. Smaller grid spacing will make the reciprocal energy calculation more accurate but slower. Using a larger [Alpha] value may require a smaller GridSpacing to be accurate.
RealSpaceCutoff
- Type:
Float
- Default value:
0.0
- Unit:
Angstrom
- Description:
Set the cutoff value for the real-space summation. Zero means the internal defaults will be used depending on the [Alpha] (if Alpha=0 then the cutoff will be set to 50 Bohr, otherwise to 20 Bohr).
Tolerance
- Type:
Float
- Default value:
1e-10
- Description:
Value of the error function that should be used to determine the cutoff radius for real-space Ewald summation if [Alpha] is set on input. Alternatively, if the [RealSpaceCutoff] is set but [Alpha] is not then the [Tolerance] value affects the [Alpha]. Larger values will make the real-space summation faster but less accurate.
Disabling energy terms¶
By default all force field energy terms are calculated, however, you can disable each one of them individually.
EnergyTerms
Angle Yes/No
Coulomb Yes/No
Dispersion Yes/No
Inversion Yes/No
Stretch Yes/No
Torsion Yes/No
End
EnergyTerms
- Type:
Block
- Description:
expert key, that allows you to disable specific energy terms.
Angle
- Type:
Bool
- Default value:
Yes
- Description:
Whether to use angle (bend) energy.
Coulomb
- Type:
Bool
- Default value:
Yes
- Description:
Whether to use coulomb energy.
Dispersion
- Type:
Bool
- Default value:
Yes
- Description:
Whether to use dispersion energy.
Inversion
- Type:
Bool
- Default value:
Yes
- Description:
Whether to use inversion energy.
Stretch
- Type:
Bool
- Default value:
Yes
- Description:
Whether to use stretch energy.
Torsion
- Type:
Bool
- Default value:
Yes
- Description:
Whether to use torsion energy.
Guessing or loading partial charges¶
The UFF forcefield has some very rudimentary partial charges guessing, only setting charges for atoms in water molecules. By default the partial charges in a force field calculation are zero. Essentially you will always need to specify atomic charges to make the results more realistic, either via the input or using one or the following options.
See also example LoadCharges, and ChargedMolecules.
GuessCharges¶
The simplest way is the use the GuessCharges key, that uses an engine that can calculate atomic charges. By default DFTB is used. DFTB is of course much more expensive than a forcefield, but if you run a MD calculation you can maybe afford a single DFTB calculation on the system.
GuessCharges Yes/No
GuessCharges
- Type:
Bool
- Default value:
No
- Description:
Use another engine to calculate/guess the charges to be used by the force field.
If you want to control the engine use the GuessChargesConfig key.
GuessChargesConfig
EngineType string
End
GuessChargesConfig
- Type:
Block
- Description:
Guess charges to be used by the forcefield
EngineType
- Type:
String
- Default value:
dftb
- Description:
Engine that can calculate or guess charges
LoadCharges¶
You have more control over the charge guessing, by loading the charges of another calculation. This way you can set any engine specific detail, such as the basis set, or functional.
You can load charges form a previous calculation to be used as force field charges.
LoadCharges
File string
Section string
Variable string
End
LoadCharges
- Type:
Block
- Description:
Load charges from a file to be used as forcefield charges
File
- Type:
String
- Description:
Name of the (kf) file
Section
- Type:
String
- Default value:
AMSResults
- Description:
Section name of the kf file
Variable
- Type:
String
- Default value:
Charges
- Description:
variable name of the kf file
Amber force field options¶
These options are relevant for the Amber and GAFF force fields:
AllowMissingParameters Yes/No
AllowMissingParameters
- Type:
Bool
- Default value:
No
- Description:
When parameters are not found for bonds, angles, dihedrals, or inversions, the first entry in the database will be used.
CheckDuplicateRules Yes/No
CheckDuplicateRules
- Type:
Bool
- Default value:
Yes
- Description:
The database could contain duplicate entries. For torsions this is a feature, and the potentials will be added. For all other terms this is no allowed, and if detected the program stops. One should fix the database or set the checking to false. As always the last entry will be used.
ForceFieldFile string
ForceFieldFile
- Type:
String
- Default value:
- GUI name:
Force field library
- Description:
Path to the force field parameter file
UFF options¶
The following options are only relevant for the UFF force field:
UFF
AtomTypesFile string
Database string
ElementsFile string
Library [UFF | UFF4MOF | UFF4MOF-II]
End
UFF
- Type:
Block
- Description:
Option for the UFF force filed.
AtomTypesFile
- Type:
String
- Default value:
mmatomtypes_db
- Description:
Expert option: Select the file that defines how UFF determines the atom types
Database
- Type:
String
- Default value:
general_db
- Description:
Expert option: Select the file that defines the UFF parameters per atom type
ElementsFile
- Type:
String
- Default value:
elements_db
- Description:
Expert option: Select the file that defines the elements known to UFF
Library
- Type:
Multiple Choice
- Default value:
UFF
- Options:
[UFF, UFF4MOF, UFF4MOF-II]
- GUI name:
Force field library
- Description:
Selects the used parameter library.
APPLE&P force field options¶
The ForceFieldFile key is mandatory and it should contain path to the APPLE&P forcefield file. This file is usually tailored for each system specifically.
Additionally, the following options are relevant for the APPLE&P force field.
DipoleConvergenceThreshold float
DipoleConvergenceThreshold
- Type:
Float
- Default value:
1e-06
- Unit:
eBohr
- Description:
Convergence criterion for induced point dipoles, in atomic units. When the length of every atomic delta_mu vector between two iterations becomes below the tolerance, the procedure is considered converged.
The repulsion/dispersion and Coulomb interaction between atoms connected by a bond or by a valence angle are excluded in APPLE&P. Those between atoms connected by a dihedral (the so called 1-4 neighbors) may be scaled down and the scaling factors can be changed using the following options:
APPLE&P
LongRangeCorrection Yes/No
MuMu14Scaling float
QMu14Scaling float
QQ14Scaling float
RD14Scaling float
End
APPLE&P
- Type:
Block
- Description:
Options for the APPLE&P force field.
LongRangeCorrection
- Type:
Bool
- Default value:
Yes
- GUI name:
Add long-range correction
- Description:
Add a long-range dispersion correction to the energy and pressure for 3D-periodic systems. This correction should be enabled only for a homogeneous liquid.
MuMu14Scaling
- Type:
Float
- Default value:
1.0
- GUI name:
Mu-Mu 3rd-neighbor scaling
- Description:
Scaling factor for dipole-dipole interactions between atoms connected to 3rd order (via a dihedral).
QMu14Scaling
- Type:
Float
- Default value:
0.2
- GUI name:
Q-Mu 3rd-neighbor scaling
- Description:
Scaling factor for charge-dipole interactions between atoms connected to 3rd order (via a dihedral).
QQ14Scaling
- Type:
Float
- Default value:
1.0
- GUI name:
Q-Q 3rd-neighbor scaling
- Description:
Scaling factor for charge-charge interactions between atoms connected to 3rd order (via a dihedral).
RD14Scaling
- Type:
Float
- Default value:
1.0
- GUI name:
RD 3rd-neighbor scaling
- Description:
Scaling factor for repulsion/dispersion interactions between atoms connected to 3rd order (via a dihedral).
Offloading calculations to LAMMPS¶
The ForceField engine can optionally offload the evaluation of energies and forces to LAMMPS to accelerate the calculation, possibly leveraging a GPU.
In this mode, the engine will still set up all force field parameters as usual, but instead of evaluating the potential directly in AMS, the engine converts the parameters into a a LAMMPS data file and then invokes LAMMPS as an external pipe worker.
As of AMS2025, this option is fully supported only for Type UFF
.
Setting up LAMMPS¶
The interface between AMS and LAMMPS relies on a modified LAMMPS version that integrates the amspipe library for communication via the AMSPipe protocol. We intend to make our modification available in upstream LAMMPS, but for now the modified version can be obtained from SCM’s GitHub page:
https://github.com/SCM-NV/lammps/tree/amspipe
We currently do not distribute binaries for the modified LAMMPS version, but it should be quite easy to build from source. Consult the LAMMPS documentation for a detailed description on building LAMMPS. For example, the following commands can be used to build LAMMPS including the necessary packages:
#!/bin/sh
# 1. build the amspipe library
git clone https://github.com/SCM-NV/amspipe.git
mkdir amspipe/build
cd amspipe/build
cmake -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON -DAMSPIPE_BUILD_DEMOS=OFF ..
cmake --build .
cmake --install . --prefix=../install
cd ../..
# 2. build modified LAMMPS
git clone https://github.com/SCM-NV/lammps.git -b amspipe
mkdir lammps/build
cd lammps/build
cmake ../cmake -DPKG_AMSPIPE=yes -Damspipe_ROOT=../../amspipe/install -DBUILD_MPI=no -DBUILD_OMP=yes -DPKG_OPENMP=yes -DPKG_GPU=yes -DPKG_MOLECULE=yes -DPKG_EXTRA-MOLECULE=yes -DPKG_KSPACE=yes
cmake --build .
-DBUILD_MPI=no -DBUILD_OMP=yes -DPKG_OPENMP=yes
are strongly recommended to enable OpenMP parallelization of the CPU code of LAMMPS. AMS currently cannot use the MPI framework to parallelize LAMMPS and linking to a MPI-enabled library can cause issues.-DPKG_GPU=yes
builds LAMMPS with GPU support. By default this builds the OpenCL version of the GPU package, which runs on both AMD and Nvidia cards. If you are targeting Nvidia cards only, we may consider switching to the Cuda version by adding-DGPU_API=cuda
(and optionally specifying also-DGPU_ARCH=sm_XX
). See the instructions for building the LAMMPS GPU package.-DPKG_MOLECULE=yes -DPKG_EXTRA-MOLECULE=yes -DPKG_KSPACE=yes
enable the potential terms required by UFF.
By default the ForceField engine will look for the modified LAMMPS binary simply as lmp
on $PATH
. It is therefore easiest just to extend $PATH
with the lammps/build
directory we made above:
export PATH="/path/to/modified/lammps/build:$PATH"
You may want to put this line into your $HOME/.profile
file to make this permanent. You can use the which
command to check that the correct lmp
binary is picked up:
$ which lmp
/path/to/modified/lammps/build/lmp
An alternative is to specify the path to the modified lmp
binary in the input of your job.
Engine ForceField
LAMMPSOffload
WorkerCommand exec /path/to/modified/lammps/build/lmp
End
End
Note
Prior to AMS2025, the LAMMPS interface in AMS was based on the lammps Python module. This proved to be a performance bottleneck and is no longer supported in AMS2025 in favor of the direct communication with the modified lmp
binary.
Input options¶
LAMMPSOffload
Enabled Yes/No
Input
UseGPU Yes/No
UseGPUForKSpace Yes/No
UseGPUForNeighbor Yes/No
UseOpenMP Yes/No
WorkerCommand string
End
LAMMPSOffload
- Type:
Block
- Description:
Offload the calculation to LAMMPS via AMSPipe.
Enabled
- Type:
Bool
- Default value:
No
- Description:
Enable offloading the force field evaluation to LAMMPS instead of handling it internally in AMS. This is currently only supported for Type=UFF.
Input
- Type:
Block
- Description:
Commands to be passed to LAMMPS to set up the calculation. If this is left empty, AMS will generate a set of commands to set LAMMPS up according to the settings of the ForceField engine. Any LAMMPS commands entered in this input block will be used to set LAMMPS up instead of those generated by AMS. To merge the AMS-generated lines with your customizations, include lines like ‘AMS somelammpskeyword’ anywhere in this block. Any such line will be replaced by the AMS-generated line for ‘somelammpskeyword’. Any text after ‘somelammpskeyword’ will be appended to the generated line verbatim, which can be used to modify the generated command by additional options. A special line ‘AMS everything’ will be replaced by the entire block of AMS-generated commands, except those overridden anywhere in this input block (defined manually or inserted using ‘AMS somelammpskeyword’. Any customized Input block should probably include ‘AMS read_data’ near or at the end to load the AMS-generated data file defining the system.
UseGPU
- Type:
Bool
- Default value:
No
- Description:
Accelerate LAMMPS calculations using a GPU. Requires a LAMMPS library built with the GPU package.
UseGPUForKSpace
- Type:
Bool
- Default value:
Yes
- Description:
When UseGPU is enabled, also use the GPU to accelerate reciprocal space electrostatic interactions. Disabling this can improve performance on less powerful GPUs.
UseGPUForNeighbor
- Type:
Bool
- Default value:
Yes
- Description:
When UseGPU is enabled, also use the GPU to accelerate neighbor searches. Disabling this can improve performance on less powerful GPUs.
UseOpenMP
- Type:
Bool
- Default value:
No
- Description:
Parallelize LAMMPS calculations using OpenMP threading. Requires a LAMMPS library built with the OMP package.
WorkerCommand
- Type:
String
- Default value:
exec lmp
- Description:
The command to execute to run the external worker. The command is executed in a subdirectory of the results directory. The LAMMPS input commands will be passed to the worker on standard input.