forked from lijiext/lammps
241 lines
10 KiB
HTML
241 lines
10 KiB
HTML
<HTML>
|
|
<CENTER><A HREF = "http://lammps.sandia.gov">LAMMPS WWW Site</A> - <A HREF = "Manual.html">LAMMPS Documentation</A> - <A HREF = "Section_commands.html#comm">LAMMPS Commands</A>
|
|
</CENTER>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<HR>
|
|
|
|
<H3>kspace_style command
|
|
</H3>
|
|
<P><B>Syntax:</B>
|
|
</P>
|
|
<PRE>kspace_style style value
|
|
</PRE>
|
|
<UL><LI>style = <I>none</I> or <I>ewald</I> or <I>pppm</I> or <I>pppm/cg</I> or <I>pppm/tip4p</I> or <I>ewald/n</I> or <I>pppm/gpu</I> or <I>ewald/omp</I> or <I>pppm/omp</I> or <I>pppm/cg/omp</I> or <I>pppm/tip4p/omp</I> or <I>pppm/proxy</I> or <I>pppm/tip4p/proxy</I>
|
|
|
|
<PRE> <I>none</I> value = none
|
|
<I>ewald</I> value = accuracy
|
|
accuracy = desired relative error in forces
|
|
<I>pppm</I> value = accuracy
|
|
accuracy = desired relative error in forces
|
|
<I>pppm/cg</I> value = accuracy (smallq)
|
|
accuracy = desired relative error in forces
|
|
smallq = cutoff for charges to be considered (optional) (charge units)
|
|
<I>pppm/tip4p</I> value = accuracy
|
|
accuracy = desired relative error in forces
|
|
<I>ewald/n</I> value = accuracy
|
|
accuracy = desired relative error in forces
|
|
<I>pppm/gpu</I> value = accuracy
|
|
accuracy = desired relative error in forces
|
|
<I>ewald/omp</I> value = accuracy
|
|
accuracy = desired relative error in forces
|
|
<I>pppm/omp</I> value = accuracy
|
|
accuracy = desired relative error in forces
|
|
<I>pppm/tip4p/omp</I> value = accuracy
|
|
accuracy = desired relative error in forces
|
|
<I>pppm/proxy</I> value = accuracy
|
|
accuracy = desired relative error in forces
|
|
</PRE>
|
|
<PRE> <I>pppm/tip4p/proxy</I> value = accuracy
|
|
accuracy = desired relative error in forces
|
|
</PRE>
|
|
|
|
</UL>
|
|
<P><B>Examples:</B>
|
|
</P>
|
|
<PRE>kspace_style pppm 1.0e-4
|
|
kspace_style pppm/cg 1.0e-5 1.0e-6
|
|
kspace_style none
|
|
</PRE>
|
|
<P><B>Description:</B>
|
|
</P>
|
|
<P>Define a K-space solver for LAMMPS to use each timestep to compute
|
|
long-range Coulombic interactions or long-range 1/r^N interactions.
|
|
When such a solver is used in conjunction with an appropriate pair
|
|
style, the cutoff for Coulombic or other 1/r^N interactions is
|
|
effectively infinite; each charge in the system interacts with charges
|
|
in an infinite array of periodic images of the simulation domain.
|
|
</P>
|
|
<P>The <I>ewald</I> style performs a standard Ewald summation as described in
|
|
any solid-state physics text.
|
|
</P>
|
|
<P>The <I>pppm</I> style invokes a particle-particle particle-mesh solver
|
|
<A HREF = "#Hockney">(Hockney)</A> which maps atom charge to a 3d mesh, uses 3d FFTs
|
|
to solve Poisson's equation on the mesh, then interpolates electric
|
|
fields on the mesh points back to the atoms. It is closely related to
|
|
the particle-mesh Ewald technique (PME) <A HREF = "#Darden">(Darden)</A> used in
|
|
AMBER and CHARMM. The cost of traditional Ewald summation scales as
|
|
N^(3/2) where N is the number of atoms in the system. The PPPM solver
|
|
scales as Nlog(N) due to the FFTs, so it is almost always a faster
|
|
choice <A HREF = "#Pollock">(Pollock)</A>.
|
|
</P>
|
|
<P>The <I>pppm/cg</I> style is identical to the <I>pppm</I> style except that it
|
|
has an optimization for systems where most particles are uncharged.
|
|
The optional <I>smallq</I> argument defines the cutoff for the absolute
|
|
charge value which determines whether a particle is considered charged
|
|
or not. Its default value is 1.0e-5.
|
|
</P>
|
|
<P>The <I>pppm/tip4p</I> style is identical to the <I>pppm</I> style except that it
|
|
adds a charge at the massless 4th site in each TIP4P water molecule.
|
|
It should be used with <A HREF = "pair_style.html">pair styles</A> with a
|
|
<I>long/tip4p</I> in their style name.
|
|
</P>
|
|
<P>The <I>ewald/n</I> style augments <I>ewald</I> by adding long-range dispersion
|
|
sum capabilities for 1/r^N potentials and is useful for simulation of
|
|
interfaces <A HREF = "#Veld">(Veld)</A>. It also performs standard coulombic Ewald
|
|
summations, but in a more efficient manner than the <I>ewald</I> style.
|
|
The 1/r^N capability means that Lennard-Jones or Buckingham potentials
|
|
can be used with <I>ewald/n</I> without a cutoff, i.e. they become full
|
|
long-range potentials.
|
|
</P>
|
|
<P>Currently, only the <I>ewald/n</I> style can be used with non-orthogonal
|
|
(triclinic symmetry) simulation boxes.
|
|
</P>
|
|
<P>The <I>pppm/proxy</I> style is a special variant for calculations
|
|
in hybrid OpenMP/MPI parallel mode. It is functionally equivalent
|
|
with <I>pppm</I>, but it its force computation is being executed
|
|
as a single thread concurrently with a multi-threaded non-bonded
|
|
calculation for a pair style with <I>pppm/omp</I> suffix. For calcuations
|
|
across many multi-core nodes, this can have a performance benefit
|
|
over performing the real and reciprocal space part separately,
|
|
specifically when otherwise the time spent on the pair style
|
|
would slightly less than in <I>pppm</I> without threading.
|
|
</P>
|
|
<P>Note that the PPPM styles can be used with single-precision FFTs by
|
|
using the compiler switch -DFFT_SINGLE for the FFT_INC setting in your
|
|
lo-level Makefile. This setting also changes some of the PPPM
|
|
operations (e.g. mapping charge to mesh and interpolating electric
|
|
fields to particles) to be performed in single precision. This option
|
|
can speed-up long-range calulations, particularly in parallel or on
|
|
GPUs. The use of the -DFFT_SINGLE flag is discussed in <A HREF = "Section_start.html#start_2_4">this
|
|
section</A> of the manual.
|
|
</P>
|
|
<HR>
|
|
|
|
<P>When a kspace style is used, a pair style that includes the
|
|
short-range correction to the pairwise Coulombic or other 1/r^N forces
|
|
must also be selected. For Coulombic interactions, these styles are
|
|
ones that have a <I>coul/long</I> in their style name. For 1/r^6
|
|
dispersion forces in a Lennard-Jones or Buckingham potential, see the
|
|
<A HREF = "pair_lj_coul.html">pair_style lj/coul</A> or <A HREF = "pair_buck_coul.html">pair_style
|
|
buck/coul</A> commands.
|
|
</P>
|
|
<P>The specified <I>accuracy</I> determines the relative RMS error in per-atom
|
|
forces calculated by the long-range solver. It is set as a
|
|
dimensionless number, relative to the force that two unit point
|
|
charges (e.g. 2 monovalent ions) exert on each other at a distance of
|
|
1 Angstrom. This reference value was chosen as representative of the
|
|
magnitude of electrostatic forces in atomic systems. Thus an accuracy
|
|
value of 1.0e-4 means that the RMS error will be a factor of 10000
|
|
smaller than the reference force.
|
|
</P>
|
|
<P>The accuracy setting is used in conjunction with the pairwise cutoff
|
|
to determine the number of K-space vectors for style <I>ewald</I> or the
|
|
FFT grid size for style <I>pppm</I>.
|
|
</P>
|
|
<P>RMS force errors in real space for <I>ewald</I> and <I>pppm</I> are estimated
|
|
using equation 18 of <A HREF = "#Kolafa">(Kolafa)</A>, which is also referenced as
|
|
equation 9 of <A HREF = "#Petersen">(Petersen)</A>. RMS force errors in K-space for
|
|
<I>ewald</I> are estimated using equation 11 of <A HREF = "#Petersen">(Petersen)</A>,
|
|
which is similar to equation 32 of <A HREF = "#Kolafa">(Kolafa)</A>. RMS force
|
|
errors in K-space for <I>pppm</I> are estimated using equation 38 of
|
|
<A HREF = "#Deserno">(Deserno)</A>.
|
|
</P>
|
|
<P>See the <A HREF = "kspace_modify.html">kspace_modify</A> command for additional
|
|
options of the K-space solvers that can be set, including a <I>force</I>
|
|
option for setting an absoulte RMS error in forces, as opposed to a
|
|
relative RMS error.
|
|
</P>
|
|
<HR>
|
|
|
|
<P>Styles with a <I>cuda</I>, <I>gpu</I>, <I>omp</I>, or <I>opt</I> suffix are functionally
|
|
the same as the corresponding style without the suffix. They have
|
|
been optimized to run faster, depending on your available hardware, as
|
|
discussed in <A HREF = "Section_accelerate.html">Section_accelerate</A> of the
|
|
manual. The accelerated styles take the same arguments and should
|
|
produce the same results, except for round-off and precision issues.
|
|
</P>
|
|
<P>More specifically, the <I>pppm/gpu</I> style performs charge assignment and
|
|
force interpolation calculations on the GPU. These processes are
|
|
performed either in single or double precision, depending on whether
|
|
the -DFFT_SINGLE setting was specified in your lo-level Makefile, as
|
|
discussed above. The FFTs themselves are still calculated on the CPU.
|
|
If <I>pppm/gpu</I> is used with a GPU-enabled pair style, part of the PPPM
|
|
calculation can be performed concurrently on the GPU while other
|
|
calculations for non-bonded and bonded force calculation are performed
|
|
on the CPU.
|
|
</P>
|
|
<P>These accelerated styles are part of the USER-CUDA, GPU, USER-OMP, and
|
|
OPT packages respectively. They are only enabled if LAMMPS was built
|
|
with those packages. See the <A HREF = "Section_start.html#start_3">Making LAMMPS</A>
|
|
section for more info.
|
|
</P>
|
|
<P>See <A HREF = "Section_accelerate.html">Section_accelerate</A> of the manual for
|
|
more instructions on how to use the accelerated styles effectively.
|
|
</P>
|
|
<P><B>Restrictions:</B>
|
|
</P>
|
|
<P>A simulation must be 3d and periodic in all dimensions to use an Ewald
|
|
or PPPM solver. The only exception is if the slab option is set with
|
|
<A HREF = "kspace_modify.html">kspace_modify</A>, in which case the xy dimensions
|
|
must be periodic and the z dimension must be non-periodic.
|
|
</P>
|
|
<P>Kspace styles are part of the KSPACE package. They are only enabled
|
|
if LAMMPS was built with that package. See the <A HREF = "Section_start.html#start_3">Making
|
|
LAMMPS</A> section for more info.
|
|
</P>
|
|
<P>The <I>ewald/n</I> style is part of the USER-EWALDN package. It is only
|
|
enabled if LAMMPS was built with that package. See the <A HREF = "Section_start.html#start_3">Making
|
|
LAMMPS</A> section for more info.
|
|
</P>
|
|
<P>When using a long-range pairwise TIP4P potential, you must use kspace
|
|
style <I>pppm/tip4p</I> and vice versa.
|
|
</P>
|
|
<P><B>Related commands:</B>
|
|
</P>
|
|
<P><A HREF = "kspace_modify.html">kspace_modify</A>, <A HREF = "pair_lj.html">pair_style
|
|
lj/cut/coul/long</A>, <A HREF = "pair_charmm.html">pair_style
|
|
lj/charmm/coul/long</A>, <A HREF = "pair_lj_coul.html">pair_style
|
|
lj/coul</A>, <A HREF = "pair_buck.html">pair_style buck/coul/long</A>
|
|
</P>
|
|
<P><B>Default:</B>
|
|
</P>
|
|
<PRE>kspace_style none
|
|
</PRE>
|
|
<HR>
|
|
|
|
<A NAME = "Darden"></A>
|
|
|
|
<P><B>(Darden)</B> Darden, York, Pedersen, J Chem Phys, 98, 10089 (1993).
|
|
</P>
|
|
<A NAME = "Deserno"></A>
|
|
|
|
<P><B>(Deserno)</B> Deserno and Holm, J Chem Phys, 109, 7694 (1998).
|
|
</P>
|
|
<A NAME = "Hockney"></A>
|
|
|
|
<P><B>(Hockney)</B> Hockney and Eastwood, Computer Simulation Using Particles,
|
|
Adam Hilger, NY (1989).
|
|
</P>
|
|
<A NAME = "Kolafa"></A>
|
|
|
|
<P><B>(Kolafa)</B> Kolafa and Perram, Molecular Simualtion, 9, 351 (1992).
|
|
</P>
|
|
<A NAME = "Petersen"></A>
|
|
|
|
<P><B>(Petersen)</B> Petersen, J Chem Phys, 103, 3668 (1995).
|
|
</P>
|
|
<A NAME = "Pollock"></A>
|
|
|
|
<P><B>(Pollock)</B> Pollock and Glosli, Comp Phys Comm, 95, 93 (1996).
|
|
</P>
|
|
<A NAME = "Veld"></A>
|
|
|
|
<P><B>(Veld)</B> In 't Veld, Ismail, Grest, J Chem Phys, in press (2007).
|
|
</P>
|
|
</HTML>
|