kspace_modify command

Syntax

kspace_modify keyword value ...
  • one or more keyword/value pairs may be listed

  • keyword = collective or compute or cutoff/adjust or diff or disp/auto or fftbench or force/disp/kspace or force/disp/real or force or gewald/disp or gewald or kmax/ewald or mesh or minorder or mix/disp or order/disp or order or overlap or scafacos or slab or splittol

    collective value = yes or no
    compute value = yes or no
    cutoff/adjust value = yes or no
    diff value = ad or ik = 2 or 4 FFTs for PPPM in smoothed or non-smoothed mode
    disp/auto value = yes or no
    fftbench value = yes or no
    force/disp/real value = accuracy (force units)
    force/disp/kspace value = accuracy (force units)
    force value = accuracy (force units)
    gewald value = rinv (1/distance units)
      rinv = G-ewald parameter for Coulombics
    gewald/disp value = rinv (1/distance units)
      rinv = G-ewald parameter for dispersion
    kmax/ewald value = kx ky kz
      kx,ky,kz = number of Ewald sum kspace vectors in each dimension
    mesh value = x y z
      x,y,z = grid size in each dimension for long-range Coulombics
    mesh/disp value = x y z
      x,y,z = grid size in each dimension for 1/r^6 dispersion
    minorder value = M
      M = min allowed extent of Gaussian when auto-adjusting to minimize grid communication
    mix/disp value = pair or geom or none
    order value = N
      N = extent of Gaussian for PPPM or MSM mapping of charge to grid
    order/disp value = N
      N = extent of Gaussian for PPPM mapping of dispersion term to grid
    overlap = yes or no = whether the grid stencil for PPPM is allowed to overlap into more than the nearest-neighbor processor
    pressure/scalar value = yes or no
    scafacos values = option value1 value2 ...
      option = tolerance
        value = energy or energy_rel or field or field_rel or potential or potential_rel
      option = fmm_tuning
        value = 0 or 1
    slab value = volfactor or nozforce
      volfactor = ratio of the total extended volume used in the
        2d approximation compared with the volume of the simulation domain
      nozforce turns off kspace forces in the z direction
    splittol value = tol
      tol = relative size of two eigenvalues (see discussion below)
    

Examples

kspace_modify mesh 24 24 30 order 6
kspace_modify slab 3.0
kspace_modify scafacos tolerance energy

Description

Set parameters used by the kspace solvers defined by the kspace_style command. Not all parameters are relevant to all kspace styles.


The collective keyword applies only to PPPM. It is set to no by default, except on IBM BlueGene machines. If this option is set to yes, LAMMPS will use MPI collective operations to remap data for 3d-FFT operations instead of the default point-to-point communication. This is faster on IBM BlueGene machines, and may also be faster on other machines if they have an efficient implementation of MPI collective operations and adequate hardware.


The compute keyword allows Kspace computations to be turned off, even though a kspace_style is defined. This is not useful for running a real simulation, but can be useful for debugging purposes or for computing only partial forces that do not include the Kspace contribution. You can also do this by simply not defining a kspace_style, but a Kspace-compatible pair_style requires a kspace style to be defined. This keyword gives you that option.


The cutoff/adjust keyword applies only to MSM. If this option is turned on, the Coulombic cutoff will be automatically adjusted at the beginning of the run to give the desired estimated error. Other cutoffs such as LJ will not be affected. If the grid is not set using the mesh command, this command will also attempt to use the optimal grid that minimizes cost using an estimate given by (Hardy). Note that this cost estimate is not exact, somewhat experimental, and still may not yield the optimal parameters.


The diff keyword specifies the differentiation scheme used by the PPPM method to compute forces on particles given electrostatic potentials on the PPPM mesh. The ik approach is the default for PPPM and is the original formulation used in (Hockney). It performs differentiation in Kspace, and uses 3 FFTs to transfer each component of the computed fields back to real space for total of 4 FFTs per timestep.

The analytic differentiation ad approach uses only 1 FFT to transfer information back to real space for a total of 2 FFTs per timestep. It then performs analytic differentiation on the single quantity to generate the 3 components of the electric field at each grid point. This is sometimes referred to as “smoothed” PPPM. This approach requires a somewhat larger PPPM mesh to achieve the same accuracy as the ik method. Currently, only the ik method (default) can be used for a triclinic simulation cell with PPPM. The ad method is always used for MSM.

Note

Currently, not all PPPM styles support the ad option. Support for those PPPM variants will be added later.


The disp/auto option controls whether the pppm/disp is allowed to generate PPPM parameters automatically. If set to no, parameters have to be specified using the gewald/disp, mesh/disp, force/disp/real or force/disp/kspace keywords, or the code will stop with an error message. When this option is set to yes, the error message will not appear and the simulation will start. For a typical application, using the automatic parameter generation will provide simulations that are either inaccurate or slow. Using this option is thus not recommended. For guidelines on how to obtain good parameters, see the How-To discussion.


The fftbench keyword applies only to PPPM. It is off by default. If this option is turned on, LAMMPS will perform a short FFT benchmark computation and report its timings, and will thus finish a some seconds later than it would if this option were off.


The force/disp/real and force/disp/kspace keywords set the force accuracy for the real and space computations for the dispersion part of pppm/disp. As shown in (Isele-Holder), optimal performance and accuracy in the results is obtained when these values are different.


The force keyword overrides the relative accuracy parameter set by the kspace_style command with an absolute accuracy. The accuracy determines the RMS error in per-atom forces calculated by the long-range solver and is thus specified in force units. A negative value for the accuracy setting means to use the relative accuracy parameter. The accuracy setting is used in conjunction with the pairwise cutoff to determine the number of K-space vectors for style ewald, the FFT grid size for style pppm, or the real space grid size for style msm.


The gewald keyword sets the value of the Ewald or PPPM G-ewald parameter for charge as rinv in reciprocal distance units. Without this setting, LAMMPS chooses the parameter automatically as a function of cutoff, precision, grid spacing, etc. This means it can vary from one simulation to the next which may not be desirable for matching a KSpace solver to a pre-tabulated pairwise potential. This setting can also be useful if Ewald or PPPM fails to choose a good grid spacing and G-ewald parameter automatically. If the value is set to 0.0, LAMMPS will choose the G-ewald parameter automatically. MSM does not use the gewald parameter.


The gewald/disp keyword sets the value of the Ewald or PPPM G-ewald parameter for dispersion as rinv in reciprocal distance units. It has the same meaning as the gewald setting for Coulombics.


The kmax/ewald keyword sets the number of kspace vectors in each dimension for kspace style ewald. The three values must be positive integers, or else (0,0,0), which unsets the option. When this option is not set, the Ewald sum scheme chooses its own kspace vectors, consistent with the user-specified accuracy and pairwise cutoff. In any case, if kspace style ewald is invoked, the values used are printed to the screen and the log file at the start of the run.


The mesh keyword sets the grid size for kspace style pppm or msm. In the case of PPPM, this is the FFT mesh, and each dimension must be factorizable into powers of 2, 3, and 5. In the case of MSM, this is the finest scale real-space mesh, and each dimension must be factorizable into powers of 2. When this option is not set, the PPPM or MSM solver chooses its own grid size, consistent with the user-specified accuracy and pairwise cutoff. Values for x,y,z of 0,0,0 unset the option.


The mesh/disp keyword sets the grid size for kspace style pppm/disp. This is the FFT mesh for long-range dispersion and ach dimension must be factorizable into powers of 2, 3, and 5. When this option is not set, the PPPM solver chooses its own grid size, consistent with the user-specified accuracy and pairwise cutoff. Values for x,y,z of 0,0,0 unset the option.


The minorder keyword allows LAMMPS to reduce the order setting if necessary to keep the communication of ghost grid point limited to exchanges between nearest-neighbor processors. See the discussion of the overlap keyword for details. If the overlap keyword is set to yes, which is the default, this is never needed. If it set to no and overlap occurs, then LAMMPS will reduce the order setting, one step at a time, until the ghost grid overlap only extends to nearest neighbor processors. The minorder keyword limits how small the order setting can become. The minimum allowed value for PPPM is 2, which is the default. If minorder is set to the same value as order then no reduction is allowed, and LAMMPS will generate an error if the grid communication is non-nearest-neighbor and overlap is set to no. The minorder keyword is not currently supported in MSM.


The mix/disp keyword selects the mixing rule for the dispersion coefficients. With pair, the dispersion coefficients of unlike types are computed as indicated with pair_modify. With geom, geometric mixing is enforced on the dispersion coefficients in the kspace coefficients. When using the arithmetic mixing rule, this will speed-up the simulations but introduces some error in the force computations, as shown in (Wennberg). With none, it is assumed that no mixing rule is applicable. Splitting of the dispersion coefficients will be performed as described in (Isele-Holder).

This splitting can be influenced with the splittol keywords. Only the eigenvalues that are larger than tol compared to the largest eigenvalues are included. Using this keywords the original matrix of dispersion coefficients is approximated. This leads to faster computations, but the accuracy in the reciprocal space computations of the dispersion part is decreased.


The order keyword determines how many grid spacings an atom’s charge extends when it is mapped to the grid in kspace style pppm or msm. The default for this parameter is 5 for PPPM and 8 for MSM, which means each charge spans 5 or 8 grid cells in each dimension, respectively. For the LAMMPS implementation of MSM, the order can range from 4 to 10 and must be even. For PPPM, the minimum allowed setting is 2 and the maximum allowed setting is 7. The larger the value of this parameter, the smaller that LAMMPS will set the grid size, to achieve the requested accuracy. Conversely, the smaller the order value, the larger the grid size will be. Note that there is an inherent trade-off involved: a small grid will lower the cost of FFTs or MSM direct sum, but a larger order parameter will increase the cost of interpolating charge/fields to/from the grid.

The PPPM order parameter may be reset by LAMMPS when it sets up the FFT grid if the implied grid stencil extends beyond the grid cells owned by neighboring processors. Typically this will only occur when small problems are run on large numbers of processors. A warning will be generated indicating the order parameter is being reduced to allow LAMMPS to run the problem. Automatic adjustment of the order parameter is not supported in MSM.


The order/disp keyword determines how many grid spacings an atom’s dispersion term extends when it is mapped to the grid in kspace style pppm/disp. It has the same meaning as the order setting for Coulombics.


The overlap keyword can be used in conjunction with the minorder keyword with the PPPM styles to adjust the amount of communication that occurs when values on the FFT grid are exchanged between processors. This communication is distinct from the communication inherent in the parallel FFTs themselves, and is required because processors interpolate charge and field values using grid point values owned by neighboring processors (i.e. ghost point communication). If the overlap keyword is set to yes then this communication is allowed to extend beyond nearest-neighbor processors, e.g. when using lots of processors on a small problem. If it is set to no then the communication will be limited to nearest-neighbor processors and the order setting will be reduced if necessary, as explained by the minorder keyword discussion. The overlap keyword is always set to yes in MSM.


The pressure/scalar keyword applies only to MSM. If this option is turned on, only the scalar pressure (i.e. (Pxx + Pyy + Pzz)/3.0) will be computed, which can be used, for example, to run an isotropic barostat. Computing the full pressure tensor with MSM is expensive, and this option provides a faster alternative. The scalar pressure is computed using a relationship between the Coulombic energy and pressure (Hummer) instead of using the virial equation. This option cannot be used to access individual components of the pressure tensor, to compute per-atom virial, or with suffix kspace/pair styles of MSM, like OMP or GPU.


The scafacos keyword is used for settings that are passed to the ScaFaCoS library when using kspace_style scafacos.

The tolerance option affects how the accuracy specified with the kspace_style command is interpreted by ScaFaCoS. The following values may be used:

  • energy = absolute accuracy in total Coulombic energy
  • energy_rel = relative accuracy in total Coulombic energy
  • potential = absolute accuracy in total Coulombic potential
  • potential_rel = relative accuracy in total Coulombic potential
  • field = absolute accuracy in electric field
  • field_rel = relative accuracy in electric field

The values with suffix _rel indicate the tolerance is a relative tolerance; the other values impose an absolute tolerance on the given quantity. Absolute tolerance in this case means, that for a given quantity q and a given absolute tolerance of t_a the result should be between q-t_a and q+t_a. For a relative tolerance t_r the relative error should not be greater than t_r, i.e. abs(1 - (result/q)) < t_r. As a consequence of this, the tolerance type should be checked, when performing computations with a high absolute field / energy. E.g. if the total energy in the system is 1000000.0 an absolute tolerance of 1e-3 would mean that the result has to be between 999999.999 and 1000000.001, which would be equivalent to a relative tolerance of 1e-9.

The energy and energy_rel values, set a tolerance based on the total Coulombic energy of the system. The potential and potential_rel set a tolerance based on the per-atom Coulombic energy. The field and field_rel tolerance types set a tolerance based on the electric field values computed by ScaFaCoS. Since per-atom forces are derived from the per-atom electric field, this effectively sets a tolerance on the forces, similar to other LAMMPS KSpace styles, as explained on the kspace_style doc page.

Note that not all ScaFaCoS solvers support all tolerance types. These are the allowed values for each method:

  • fmm = energy and energy_rel
  • p2nfft = field (1d-,2d-,3d-periodic systems) or potential (0d-periodic)
  • p3m = field
  • ewald = field
  • direct = has no tolerance tuning

If the tolerance type is not changed, the default values for the tolerance type are the first values in the above list, e.g. energy is the default tolerance type for the fmm solver.

The fmm_tuning option is only relevant when using the FMM method. It activates (value=1) or deactivates (value=0) an internal tuning mechanism for the FMM solver. The tuning operation runs sequentially and can be very time-consuming. Usually it is not needed for systems with a homogeneous charge distribution. The default for this option is therefore 0. The FMM internal tuning is performed once, when the solver is set up.


The slab keyword allows an Ewald or PPPM solver to be used for a systems that are periodic in x,y but non-periodic in z - a boundary setting of “boundary p p f”. This is done by treating the system as if it were periodic in z, but inserting empty volume between atom slabs and removing dipole inter-slab interactions so that slab-slab interactions are effectively turned off. The volfactor value sets the ratio of the extended dimension in z divided by the actual dimension in z. The recommended value is 3.0. A larger value is inefficient; a smaller value introduces unwanted slab-slab interactions. The use of fixed boundaries in z means that the user must prevent particle migration beyond the initial z-bounds, typically by providing a wall-style fix. The methodology behind the slab option is explained in the paper by (Yeh). The slab option is also extended to non-neutral systems (Ballenegger).

An alternative slab option can be invoked with the nozforce keyword in lieu of the volfactor. This turns off all kspace forces in the z direction. The nozforce option is not supported by MSM. For MSM, any combination of periodic, non-periodic, or shrink-wrapped boundaries can be set using boundary (the slab approximation in not needed). The slab keyword is not currently supported by Ewald or PPPM when using a triclinic simulation cell. The slab correction has also been extended to point dipole interactions (Klapp) in kspace_style ewald/disp.

Note

If you wish to apply an electric field in the Z-direction, in conjunction with the slab keyword, you should do it by adding explicit charged particles to the +/- Z surfaces. If you do it via the fix efield command, it will not give the correct dielectric constant due to the Yeh/Berkowitz (Yeh) correction not being compatible with how fix efield works.


The force/disp/real and force/disp/kspace keywords set the force accuracy for the real and space computations for the dispersion part of pppm/disp. As shown in (Isele-Holder), optimal performance and accuracy in the results is obtained when these values are different.

The disp/auto option controls whether the pppm/disp is allowed to generate PPPM parameters automatically. If set to no, parameters have to be specified using the gewald/disp, mesh/disp, force/disp/real or force/disp/kspace keywords, or the code will stop with an error message. When this option is set to yes, the error message will not appear and the simulation will start. For a typical application, using the automatic parameter generation will provide simulations that are either inaccurate or slow. Using this option is thus not recommended. For guidelines on how to obtain good parameters, see the Howto dispersion doc page.


Restrictions

none

Default

The option defaults are mesh = mesh/disp = 0 0 0, order = order/disp = 5 (PPPM), order = 10 (MSM), minorder = 2, overlap = yes, force = -1.0, gewald = gewald/disp = 0.0, slab = 1.0, compute = yes, cutoff/adjust = yes (MSM), pressure/scalar = yes (MSM), fftbench = no (PPPM), diff = ik (PPPM), mix/disp = pair, force/disp/real = -1.0, force/disp/kspace = -1.0, split = 0, tol = 1.0e-6, and disp/auto = no. For pppm/intel, order = order/disp = 7. For scafacos settings, the scafacos tolerance option depends on the method chosen, as documented above. The scafacos fmm_tuning default = 0.


(Hockney) Hockney and Eastwood, Computer Simulation Using Particles, Adam Hilger, NY (1989).

(Yeh) Yeh and Berkowitz, J Chem Phys, 111, 3155 (1999).

(Ballenegger) Ballenegger, Arnold, Cerda, J Chem Phys, 131, 094107 (2009).

(Klapp) Klapp, Schoen, J Chem Phys, 117, 8050 (2002).

(Hardy) David Hardy thesis: Multilevel Summation for the Fast Evaluation of Forces for the Simulation of Biomolecules, University of Illinois at Urbana-Champaign, (2006).

(Hummer) Hummer, Gronbech-Jensen, Neumann, J Chem Phys, 109, 2791 (1998)

(Isele-Holder) Isele-Holder, Mitchell, Hammond, Kohlmeyer, Ismail, J Chem Theory Comput, 9, 5412 (2013).

(Wennberg) Wennberg, Murtola, Hess, Lindahl, J Chem Theory Comput, 9, 3527 (2013).