应用软件 | LAMMPS_19Nov2024

xwhuan

应用介绍

LAMMPS（Large-scale Atomic/Molecular Massively Parallel Simulator）是一个经典的分子动力学（Molecular Dynamics，MD）模拟软件，广泛应用于材料科学、化学、物理、生物学等领域，用于模拟原子、分子、颗粒、或连续介质的运动。

加载环境

平台安装多个版本的LAMMPS，新版的安装路径为/opt/app/lammps/版本号cpu 和/opt/app/lammps/版本号gpu

配置了环境加载脚本，执行即可加载所有软件环境。

## cpu版
source /opt/app/lammps/19Nov2024_cpu/lammps.sh

## gpu版
source /opt/app/lammps/19Nov2024_gpu/lammps.sh

命令行提交

先创建一个目录lammpstest并进入该目录
```
salloc 
ssh XXX
mkdir lammpstest
cd lammpstest
```

在该目录下创建如下测试文件lammps_demo.lj

# Simple Lennard-Jones simulation

# 1. Initialization
units           lj                    # 使用 Lennard-Jones 单位制
atom_style      atomic                # 每个原子仅有质量和坐标

# 2. Create simulation box and atoms
lattice         fcc 0.8442            # 创建一个面心立方晶格，密度约为 0.8442
region          box block 0 10 0 10 0 10  # 定义模拟区域大小
create_box      1 box                 # 为 1 种原子类型创建模拟箱
create_atoms    1 box                 # 在整个模拟箱中生成原子

# 3. Define force field
mass            1 1.0                 # 定义原子质量为 1.0
pair_style      lj/cut 2.5            # Lennard-Jones 力场，截断距离为 2.5
pair_coeff      1 1 1.0 1.0 2.5       # 力场参数：epsilon = 1.0, sigma = 1.0

# 4. Set simulation settings
velocity        all create 1.44 87287 loop geom  # 初始化速度，温度 1.44（LJ 单位）
neighbor        0.3 bin             # 邻居列表设置
neigh_modify    every 20 delay 0 check no # 每 20 步更新一次邻居列表

# 5. Define integration
fix             1 all nve             # 使用 NVE 积分器（能量守恒）
thermo          100                   # 每 100 步输出一次能量信息
thermo_style    custom step temp pe ke etotal # 自定义输出变量

# 6. Run simulation
run             500                   # 运行 500 步

执行命令

source /opt/app/lammps/19Nov2024_cpu/lammps.sh
lmp -in lammps_demo.lj

屏幕上查看结果

LAMMPS (2 Aug 2023 - Update 4)
OMP_NUM_THREADS environment is not set. Defaulting to 1 thread. (../comm.cpp:98)
  using 1 OpenMP thread(s) per MPI task
Lattice spacing in x,y,z = 1.6795962 1.6795962 1.6795962
Created orthogonal box = (0 0 0) to (16.795962 16.795962 16.795962)
  1 by 1 by 1 MPI processor grid
Created 4000 atoms
  using lattice units in orthogonal box = (0 0 0) to (16.795962 16.795962 16.795962)
  create_atoms CPU = 0.001 seconds
Generated 0 of 0 mixed pair_coeff terms from geometric mixing rule
Neighbor list info ...
  update: every = 20 steps, delay = 0 steps, check = no
  max neighbors/atom: 2000, page size: 100000
  master list distance cutoff = 2.8
  ghost atom cutoff = 2.8
  binsize = 1.4, bins = 12 12 12
  1 neighbor lists, perpetual/occasional/extra = 1 0 0
  (1) pair lj/cut, perpetual
      attributes: half, newton on
      pair build: half/bin/atomonly/newton
      stencil: half/bin/3d
      bin: standard
Setting up Verlet run ...
  Unit style    : lj
  Current step  : 0
  Time step     : 0.005
Per MPI rank memory allocation (min/avg/max) = 3.222 | 3.222 | 3.222 Mbytes
   Step          Temp          PotEng         KinEng         TotEng    
         0   1.44          -6.7733681      2.15946       -4.6139081    
       100   0.75715334    -5.7581426      1.1354461     -4.6226965    
       200   0.75139921    -5.7500924      1.126817      -4.6232753    
       300   0.73979146    -5.7326964      1.1094098     -4.6232866    
       400   0.73834687    -5.7300319      1.1072434     -4.6227884    
       500   0.72941121    -5.7153612      1.0938433     -4.6215179    
Loop time of 1.28074 on 1 procs for 500 steps with 4000 atoms

Performance: 168652.897 tau/day, 390.400 timesteps/s, 1.562 Matom-step/s
99.7% CPU use with 1 MPI tasks x 1 OpenMP threads

MPI task timing breakdown:
Section |  min time  |  avg time  |  max time  |%varavg| %total
---------------------------------------------------------------
Pair    | 1.0807     | 1.0807     | 1.0807     |   0.0 | 84.38
Neigh   | 0.15915    | 0.15915    | 0.15915    |   0.0 | 12.43
Comm    | 0.017768   | 0.017768   | 0.017768   |   0.0 |  1.39
Output  | 0.00015991 | 0.00015991 | 0.00015991 |   0.0 |  0.01
Modify  | 0.017599   | 0.017599   | 0.017599   |   0.0 |  1.37
Other   |            | 0.005328   |            |       |  0.42

Nlocal:           4000 ave        4000 max        4000 min
Histogram: 1 0 0 0 0 0 0 0 0 0
Nghost:           5657 ave        5657 max        5657 min
Histogram: 1 0 0 0 0 0 0 0 0 0
Neighs:         149883 ave      149883 max      149883 min
Histogram: 1 0 0 0 0 0 0 0 0 0

Total # of neighbors = 149883
Ave neighs/atom = 37.47075
Neighbor list builds = 25
Dangerous builds not checked
Total wall time: 0:00:01

作业提交模板

gpu版

在个人目录下创建作业脚本run.sh，内容如下：

申请合适的分区和显卡，gpu版需要加载nvidia和cuda的环境，
程序运行的代码还需要修改"- in"后的输入文件参数

#!/bin/bash
#SBATCH --job-name=lammps_gpu
## 选择适合的分区
#SBATCH --partition=gpu-a800
#SBATCH --gres=gpu:1
#SBATCH --ntasks-per-node=8

## 加载nvidia驱动环境
export PATH=/opt/app/nvidia/535.154.05/bin:$PATH
export LD_LIBRARY_PATH=/opt/app/nvidia/535.154.05/lib:/opt/app/nvidia/535.154.05/lib32:$LD_LIBRARY_PATH
## 加载CUDA环境
export PATH=/opt/app/cuda/12.2/bin:$PATH
export CUDA_PATH=/opt/app/cuda/12.2
export LD_LIBRARY_PATH=/opt/app/cuda/12.2/lib64:$LD_LIBRARY_PATH
## 加载环境
source /opt/app/lammps/19Nov2024_gpu/lammps.sh

## 运行程序， -in参数后为输入文件
mpirun -np 1 lmp -sf gpu -pk gpu 1 -in XXX

cpu版

在个人目录下创建作业脚本run.sh，内容如下：

申请合适的分区和节点后，修改"- in"输入文件即可提交作业

#!/bin/bash
#SBATCH --job-name=lammps_cpu
## 选择适合的分区
#SBATCH --partition=cpu
#SBATCH --nodes=1           
#SBATCH --ntasks-per-node=8 


## 加载环境
source /opt/app/lammps/19Nov2024_cpu/lammps.sh
## 运行程序， -in参数后为输入文件
mpirun -np 8 lmp -in XXX