tx258/lammps-gran-kokkos
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

update docs, examples, and add performance numbers and version tags

Browse Source
This commit is contained in:
Axel Kohlmeyer
2023-04-07 23:22:01 -04:00
parent 70f1d17495
commit 37eb81799e
2 changed files with 56 additions and 45 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
doc/src/pair_lj_cut_sphere.rst
View File

@ -33,6 +33,8 @@ Examples
Description
"""""""""""
.. versionadded:: TBD
The *lj/cut/sphere* style compute the standard 12/6 Lennard-Jones potential,
given by
@ -70,17 +72,21 @@ is at :math:`2^{\frac{1}{6}} \sigma_{ij}`.
pairs that do not interact. However, if your system is highly
polydisperse (ratio > 2.0), the neighbor list build and force
computations may be inefficient. There are two ways to try and
speed up the computations.
speed up the simulations.
The first is to assign multiple atom types so that atoms of each
type are similar in size. E.g. if particle diameters range from 1
to 5 use 4 atom types, ensuring atoms of type 1 have diameters from
1.0-2.0, type 2 from 2.0-3.0, etc.
The first is to assign atoms to different atom types so that atoms of
each type are similar in size. E.g. if particle diameters range from
1 to 5 use 4 atom types, ensuring atoms of type 1 have diameters from
1.0-2.0, type 2 from 2.0-3.0, etc. This will reduce the number of
non-interacting pairs in the neighbor lists and thus reduce the time
spent on computing pairwise interactions.
The second is to try the :doc:`neighbor multi <neighbor>` command
which uses a different algorithm for buliding neighbor lists. This
will also require that you assign multiple atom types as in the
preceeding paragraph.
The second is to use the :doc:`neighbor multi <neighbor>` command
which enabled a different algorithm for building neighbor lists. This
will also require that you assign multiple atom types according to
diameters, but will in addition use a more efficient size-dependent
strategy to construct the neighbor lists and thus reduce the time
spent on building neighbor lists.
Here are example input script commands using both ideas for a
highly polydisperse system:
@ -95,22 +101,24 @@ is at :math:`2^{\frac{1}{6}} \sigma_{ij}`.
create_atoms 1 box
# create atoms with random diameters from bimodal distribution
variable switch atom random(0.0,1.0,345634)
variable diam atom (v_switch<0.75)*normal(0.4,0.075,325)+(v_switch>=0.7)*normal(1.2,0.2,453)
set group all diameter v_diam
# assign type 2 to atoms with diameter > 0.5
variable large atom 2.0*radius>0.5
# assign type 2 to atoms with diameter > 0.6
variable large atom (2.0*radius)>0.6
group large variable large
set group largea type 2
set group large type 2
pair_style lj/cut/sphere 2.5
pair_coeff * * 1.0
neighbor 0.3 multi
Using multiple atom types speeds up the calculation for this example
by more than a factor of 2, and using the multi-style neighbor list
build causes an additional speedup of about 20 percent.
Coefficients
""""""""""""

65
doc/src/pair_lj_expand_sphere.rst
View File

@ -33,6 +33,8 @@ Examples
Description
"""""""""""
.. versionadded:: TBD
The *lj/expand/sphere* style compute a 12/6 Lennard-Jones potential with
a distance shifted by :math:`\Delta = \frac{1}{2} (d_i + d_j)`, the
average diameter of both atoms. This can be used to model particles of
@ -72,19 +74,23 @@ is at :math:`2^{\frac{1}{6}} \sigma`.
pairs that do not interact. However, if your system is highly
polydisperse (ratio > 2.0), the neighbor list build and force
computations may be inefficient. There are two ways to try and
speed up the computations.
speed up the simulations.
The first is to assign multiple atom types so that atoms of each
type are similar in size. E.g. if particle diameters range from 1
to 5 use 4 atom types, ensuring atoms of type 1 have diameters from
1.0-2.0, type 2 from 2.0-3.0, etc.
The first is to assign atoms to different atom types so that atoms of
each type are similar in size. E.g. if particle diameters range from
1 to 5 use 4 atom types, ensuring atoms of type 1 have diameters from
1.0-2.0, type 2 from 2.0-3.0, etc. This will reduce the number of
non-interacting pairs in the neighbor lists and thus reduce the time
spent on computing pairwise interactions.
The second is to try the :doc:`neighbor multi <neighbor>` command
which uses a different algorithm for buliding neighbor lists. This
will also require that you assign multiple atom types as in the
preceeding paragraph.
The second is to use the :doc:`neighbor multi <neighbor>` command
which enabled a different algorithm for building neighbor lists. This
will also require that you assign multiple atom types according to
diameters, but will in addition use a more efficient size-dependent
strategy to construct the neighbor lists and thus reduce the time
spent on building neighbor lists.
Here are example input script commands using both ideas for a
Here are example input script commands using the first option for a
highly polydisperse system:
.. code-block:: c++
@ -97,21 +103,23 @@ is at :math:`2^{\frac{1}{6}} \sigma`.
create_atoms 1 box
# create atoms with random diameters from bimodal distribution
variable switch atom random(0.0,1.0,345634)
variable diam atom (v_switch<0.75)*normal(0.4,0.075,325)+(v_switch>=0.7)*normal(1.2,0.2,453)
variable diam atom (v_switch<0.75)*normal(0.2,0.04,325)+(v_switch>=0.7)*normal(0.6,0.2,453)
set group all diameter v_diam
# assign type 2 to atoms with diameter > 0.5
variable large atom 2.0*radius>0.5
# assign type 2 to atoms with diameter > 0.35
variable large atom (2.0*radius)>0.35
group large variable large
set group largea type 2
set group large type 2
pair_style lj/expand/sphere 2.5
pair_coeff * * 1.0 1.0
pair_style lj/expand/sphere 2.0
pair_coeff * * 1.0 0.5
neighbor 0.3 multi
neighbor 0.3 bin
Using multiple atom types speeds up the calculation for this example
by more than 30 percent, but using the multi-style neighbor list does
not provide a speedup.
Coefficients
""""""""""""
@ -122,14 +130,11 @@ file or restart files read by the :doc:`read_data <read_data>` or
:doc:`read_restart <read_restart>` commands, or by mixing as described below:
* :math:`\epsilon` (energy units)
* LJ cutoff ratio (unitless) (optional)
* :math:`\sigma` (distance units)
* LJ cutoff (distance units) (optional)
The last coefficient is optional. If not specified, the global LJ
cutoff ratio specified in the :doc:`pair_style command <pair_style>` is
used.
If a repulsive only LJ interaction is desired, the coefficient for the cutoff
ratio should be set to the minimum of the LJ potential using ``$(2.0^(1.0/6.0))``
cutoff specified in the :doc:`pair_style command <pair_style>` is used.
----------
@ -140,10 +145,10 @@ ratio should be set to the minimum of the LJ potential using ``$(2.0^(1.0/6.0))`
Mixing, shift, table, tail correction, restart, rRESPA info
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
For atom type pairs I,J and I != J, the epsilon coefficients and cutoff
ratio for the *lj/expand/sphere* pair style can be mixed. The default mixing
style is *geometric*. See the :doc:`pair_modify command <pair_modify>`
for details.
For atom type pairs I,J and I != J, the epsilon, sigma, and cutoff
coefficients for the *lj/expand/sphere* pair style can be mixed. The
default mixing style is *geometric*. See the :doc:`pair_modify command
<pair_modify>` for details.
The *lj/expand/sphere* pair style supports the :doc:`pair_modify shift <pair_modify>`
option for the energy of the Lennard-Jones portion of the pair interaction.
@ -169,8 +174,6 @@ The *lj/expand/sphere* pair style is only enabled if LAMMPS was built with the
EXTRA-PAIR package. See the :doc:`Build package <Build_package>` page
for more info.
The *lj/expand/sphere* pair style does not support the *sixthpower* mixing rule.
----------
Related commands