Homec4science
Diffusion lammps (efficient_neuronet) Browse

lammps/doc/run_style.html
7e6755866f77efficient_neuronet

Last Change
Raw File

run_style.html
View Options

<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>run_style command
</H3>
<P><B>Syntax:</B>
</P>
<PRE>run_style style args
</PRE>
<UL><LI>style = <I>verlet</I> or <I>verlet/split</I> or <I>respa</I>
<PRE> <I>verlet</I> args = none
<I>verlet/split</I> args = none
<I>respa</I> args = N n1 n2 ... keyword values ...
N = # of levels of rRESPA
n1, n2, ... = loop factor between rRESPA levels (N-1 values)
zero or more keyword/value pairings may be appended to the loop factors
keyword = <I>bond</I> or <I>angle</I> or <I>dihedral</I> or <I>improper</I> or
<I>pair</I> or <I>inner</I> or <I>middle</I> or <I>outer</I> or <I>kspace</I>
<I>bond</I> value = M
M = which level (1-N) to compute bond forces in
<I>angle</I> value = M
M = which level (1-N) to compute angle forces in
<I>dihedral</I> value = M
M = which level (1-N) to compute dihedral forces in
<I>improper</I> value = M
M = which level (1-N) to compute improper forces in
<I>pair</I> value = M
M = which level (1-N) to compute pair forces in
<I>inner</I> values = M cut1 cut2
M = which level (1-N) to compute pair inner forces in
cut1 = inner cutoff between pair inner and
pair middle or outer (distance units)
cut2 = outer cutoff between pair inner and
pair middle or outer (distance units)
<I>middle</I> values = M cut1 cut2
M = which level (1-N) to compute pair middle forces in
cut1 = inner cutoff between pair middle and pair outer (distance units)
cut2 = outer cutoff between pair middle and pair outer (distance units)
<I>outer</I> value = M
M = which level (1-N) to compute pair outer forces in
<I>kspace</I> value = M
M = which level (1-N) to compute kspace forces in
</PRE>
</UL>
<P><B>Examples:</B>
</P>
<PRE>run_style verlet
run_style respa 4 2 2 2 bond 1 dihedral 2 pair 3 kspace 4
run_style respa 4 2 2 2 bond 1 dihedral 2 inner 3 5.0 6.0 outer 4 kspace 4
</PRE>
<P><B>Description:</B>
</P>
<P>Choose the style of time integrator used for molecular dynamics
simulations performed by LAMMPS.
</P>
<P>The <I>verlet</I> style is a standard velocity-Verlet integrator.
</P>
<HR>
<P>The <I>verlet/split</I> style is also a velocity-Verlet integrator, but it
splits the force calculation within each timestep over 2 partitions of
processors. See <A HREF = "Section_start.html#start_6">Section_start 6</A> for an
explanation of the -partition command-line switch.
</P>
<P>Specifically, this style performs all computation except the
<A HREF = "kspace_style.html">kspace_style</A> portion of the force field on the 1st
partition. This include the <A HREF = "pair_style.html">pair style</A>, <A HREF = "bond_style.html">bond
style</A>, <A HREF = "neighbor.html">neighbor list building</A>,
<A HREF = "fix.html">fixes</A> including time intergration, and output. The
<A HREF = "kspace_style.html">kspace_style</A> portion of the calculation is
performed on the 2nd partition.
</P>
<P>This is most useful for the PPPM kspace_style when its performance on
a large number of processors degrades due to the cost of communication
in its 3d FFTs. In this scenario, splitting your P total processors
into 2 subsets of processors, P1 in the 1st partition and P2 in the
2nd partition, can enable your simulation to run faster. This is
because the long-range forces in PPPM can be calculated at the same
time as pair-wise and bonded forces are being calculated, and the FFTs
can actually speed up when running on fewer processors.
</P>
<P>To use this style, you must define 2 partitions where P1 is a multiple
of P2. Typically having P1 be 3x larger than P2 is a good choice.
The 3d processor layouts in each partition must overlay in the
following sense. If P1 is a Px1 by Py1 by Pz1 grid, and P2 = Px2 by
Py2 by Pz2, then Px1 must be an integer multiple of Px2, and similarly
for Py1 a multiple of Py2, and Pz1 a multiple of Pz2.
</P>
<P>Typically the best way to do this is to let the 1st partition choose
its onn optimal layout, then require the 2nd partition's layout to
match the integer multiple constraint. See the
<A HREF = "processors.html">processors</A> command with its <I>part</I> keyword for a way
to control this, e.g.
</P>
<PRE>procssors * * * part 1 2 multiple
</PRE>
<P>You can also use the <A HREF = "partition.html">partition</A> command to explicitly
specity the processor layout on each partition. E.g. for 2 partitions
of 60 and 15 processors each:
</P>
<PRE>partition yes 1 processors 3 4 5
partition yes 2 processors 3 1 5
</PRE>
<P>When you run in 2-partition mode with the <I>verlet/split</I> style, the
thermodyanmic data for the entire simulation will be output to the log
and screen file of the 1st partition, which are log.lammps.0 and
screen.0 by default; see the "-plog and -pscreen command-line
switches"Section_start.html#start_6 to change this. The log and
screen file for the 2nd partition will not contain thermodynamic
output beyone the 1st timestep of the run.
</P>
<P>See <A HREF = "Section_accelerate.html">Section_accelerate</A> of the manual for
performance details of the speed-up offered by the <I>verlet/split</I>
style. One important performance consideration is the assignemnt of
logical processors in the 2 partitions to the physical cores of a
parallel machine. The <A HREF = "processors.html">processors</A> command has
options to support this, and strategies are discussed in
<A HREF = "Section_accelerate.html">Section_accelerate</A> of the manual.
</P>
<HR>
<P>The <I>respa</I> style implements the rRESPA multi-timescale integrator
<A HREF = "#Tuckerman">(Tuckerman)</A> with N hierarchical levels, where level 1 is
the innermost loop (shortest timestep) and level N is the outermost
loop (largest timestep). The loop factor arguments specify what the
looping factor is between levels. N1 specifies the number of
iterations of level 1 for a single iteration of level 2, N2 is the
iterations of level 2 per iteration of level 3, etc. N-1 looping
parameters must be specified.
</P>
<P>The <A HREF = "timestep.html">timestep</A> command sets the timestep for the
outermost rRESPA level. Thus if the example command above for a
4-level rRESPA had an outer timestep of 4.0 fmsec, the inner timestep
would be 8x smaller or 0.5 fmsec. All other LAMMPS commands that
specify number of timesteps (e.g. <A HREF = "neigh_modify.html">neigh_modify</A>
parameters, <A HREF = "dump.html">dump</A> every N timesteps, etc) refer to the
outermost timesteps.
</P>
<P>The rRESPA keywords enable you to specify at what level of the
hierarchy various forces will be computed. If not specified, the
defaults are that bond forces are computed at level 1 (innermost
loop), angle forces are computed where bond forces are, dihedral
forces are computed where angle forces are, improper forces are
computed where dihedral forces are, pair forces are computed at the
outermost level, and kspace forces are computed where pair forces are.
The inner, middle, outer forces have no defaults.
</P>
<P>The <I>inner</I> and <I>middle</I> keywords take additional arguments for
cutoffs that are used by the pairwise force computations. If the 2
cutoffs for <I>inner</I> are 5.0 and 6.0, this means that all pairs up to
6.0 apart are computed by the inner force. Those between 5.0 and 6.0
have their force go ramped to 0.0 so the overlap with the next regime
(middle or outer) is smooth. The next regime (middle or outer) will
compute forces for all pairs from 5.0 outward, with those from 5.0 to
6.0 having their value ramped in an inverse manner.
</P>
<P>Only some pair potentials support the use of the <I>inner</I> and <I>middle</I>
and <I>outer</I> keywords. If not, only the <I>pair</I> keyword can be used
with that pair style, meaning all pairwise forces are computed at the
same rRESPA level. See the doc pages for individual pair styles for
details.
</P>
<P>When using rRESPA (or for any MD simulation) care must be taken to
choose a timestep size(s) that insures the Hamiltonian for the chosen
ensemble is conserved. For the constant NVE ensemble, total energy
must be conserved. Unfortunately, it is difficult to know <I>a priori</I>
how well energy will be conserved, and a fairly long test simulation
(~10 ps) is usually necessary in order to verify that no long-term
drift in energy occurs with the trial set of parameters.
</P>
<P>With that caveat, a few rules-of-thumb may be useful in selecting
<I>respa</I> settings. The following applies mostly to biomolecular
simulations using the CHARMM or a similar all-atom force field, but
the concepts are adaptable to other problems. Without SHAKE, bonds
involving hydrogen atoms exhibit high-frequency vibrations and require
a timestep on the order of 0.5 fmsec in order to conserve energy. The
relatively inexpensive force computations for the bonds, angles,
impropers, and dihedrals can be computed on this innermost 0.5 fmsec
step. The outermost timestep cannot be greater than 4.0 fmsec without
risking energy drift. Smooth switching of forces between the levels
of the rRESPA hierarchy is also necessary to avoid drift, and a 1-2
angstrom "healing distance" (the distance between the outer and inner
cutoffs) works reasonably well. We thus recommend the following
settings for use of the <I>respa</I> style without SHAKE in biomolecular
simulations:
</P>
<PRE>timestep 4.0
run_style respa 4 2 2 2 inner 2 4.5 6.0 middle 3 8.0 10.0 outer 4
</PRE>
<P>With these settings, users can expect good energy conservation and
roughly a 2.5 fold speedup over the <I>verlet</I> style with a 0.5 fmsec
timestep.
</P>
<P>If SHAKE is used with the <I>respa</I> style, time reversibility is lost,
but substantially longer time steps can be achieved. For biomolecular
simulations using the CHARMM or similar all-atom force field, bonds
involving hydrogen atoms exhibit high frequency vibrations and require
a time step on the order of 0.5 fmsec in order to conserve energy.
These high frequency modes also limit the outer time step sizes since
the modes are coupled. It is therefore desirable to use SHAKE with
respa in order to freeze out these high frequency motions and increase
the size of the time steps in the respa hierarchy. The following
settings can be used for biomolecular simulations with SHAKE and
rRESPA:
</P>
<PRE>fix 2 all shake 0.000001 500 0 m 1.0 a 1
timestep 4.0
run_style respa 2 2 inner 1 4.0 5.0 outer 2
</PRE>
<P>With these settings, users can expect good energy conservation and
roughly a 1.5 fold speedup over the <I>verlet</I> style with SHAKE and a
2.0 fmsec timestep.
</P>
<P>For non-biomolecular simulations, the <I>respa</I> style can be
advantageous if there is a clear separation of time scales - fast and
slow modes in the simulation. Even a LJ system can benefit from
rRESPA if the interactions are divided by the inner, middle and outer
keywords. A 2-fold or more speedup can be obtained while maintaining
good energy conservation. In real units, for a pure LJ fluid at
liquid density, with a sigma of 3.0 angstroms, and epsilon of 0.1
Kcal/mol, the following settings seem to work well:
</P>
<PRE>timestep 36.0
run_style respa 3 3 4 inner 1 3.0 4.0 middle 2 6.0 7.0 outer 3
</PRE>
<HR>
<P><B>Restrictions:</B>
</P>
<P>The <I>verlet/split</I> style can only be used if LAMMPS was built with the
REPLICA package. See the <A HREF = "Section_start.html#start_3">Making LAMMPS</A>
section for more info on packages.
</P>
<P>Whenever using rRESPA, the user should experiment with trade-offs in
speed and accuracy for their system, and verify that they are
conserving energy to adequate precision.
</P>
<P><B>Related commands:</B>
</P>
<P><A HREF = "timestep.html">timestep</A>, <A HREF = "run.html">run</A>
</P>
<P><B>Default:</B>
</P>
<PRE>run_style verlet
</PRE>
<HR>
<A NAME = "Tuckerman"></A>
<P><B>(Tuckerman)</B> Tuckerman, Berne and Martyna, J Chem Phys, 97, p 1990
(1992).
</P>
</HTML>
c4science · Help