GROMACS温控算法与耦合时间实操

下载16组完整输入、原始日志、能量数据与绘图脚本。所有例子使用同一盒216个SPC/E水分子（648原子），在鸢算科技集群P100上实际运行20 ps。GROMACS 2026.4使用CUDA 12.8、thread-MPI构建；每个任务使用1张GPU、2个CPU线程。建模与最小化见Packmol接GROMACS水盒工作流。

运行步骤#

先解压下载包，进入gromacs-water子目录。下面各个算例互相独立，每次从下载包根目录进入对应目录；为每个需要运行的目录保存一份下方的run.sh。start.gro包含坐标和速度，topol.top引用安装包自带SPC/E参数。下面的脚本在所选算例目录保存为run.sh：

#!/bin/bash
#SBATCH --partition=gpu-8p100-hs
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=2
#SBATCH --gres=gpu:1
#SBATCH --time=00:20:00
set -eo pipefail
source /etc/profile
module purge
module load gromacs/2026.4-p100-threadmpi
export OMP_NUM_THREADS=2
gmx grompp -f md.mdp -c start.gro -p topol.top -o md.tpr
gmx mdrun -deffnm md -ntmpi 1 -ntomp 2 -nb gpu -pme gpu -bonded cpu -update cpu
if grep -Eq '^pcoupl[[:space:]]*=[[:space:]]*no([[:space:]]|$)' md.mdp; then
  printf 'Potential\nTotal-Energy\nTemperature\nPressure\n0\n' | gmx energy -f md.edr -o thermo.xvg
else
  printf 'Potential\nTotal-Energy\nTemperature\nPressure\nVolume\n0\n' | gmx energy -f md.edr -o thermo.xvg
fi

sbatch run.sh
# 任务结束后检查最后几行日志
 tail -n 20 md.log

constraints = none不代表这里的水是柔性水：SPC/E拓扑含SETTLE，水分子内部几何仍保持刚性。continuation = yes与gen-vel = no保留输入速度。本组从同一GRO起点分叉，GRO速度有输出精度限制，并不是读取同一完整检查点的位级重启。

本例比较什么#

比较v-rescale、Nose–Hoover、Berendsen和随机动力学SD，再改变v-rescale的tau-t。SD使用integrator = sd，不再叠加另一套tcoupl温控。

nvt-v-rescale#

cd nvt-v-rescale
sbatch run.sh

完整 md.mdp：

integrator = md
dt = 0.001
nsteps = 20000
continuation = yes
cutoff-scheme = Verlet
coulombtype = PME
rcoulomb = 0.8
rvdw = 0.8
pbc = xyz
tcoupl = v-rescale
tc-grps = System
tau-t = 0.1
ref-t = 300
pcoupl = no
gen-vel = no
constraints = none
nstenergy = 10
nstxout-compressed = 100
fourierspacing = 0.12

nvt-nose-hoover#

cd nvt-nose-hoover
sbatch run.sh

完整 md.mdp：

integrator = md
dt = 0.001
nsteps = 20000
continuation = yes
cutoff-scheme = Verlet
coulombtype = PME
rcoulomb = 0.8
rvdw = 0.8
pbc = xyz
tcoupl = nose-hoover
tc-grps = System
tau-t = 0.1
ref-t = 300
pcoupl = no
gen-vel = no
constraints = none
nstenergy = 10
nstxout-compressed = 100
fourierspacing = 0.12

nvt-berendsen#

cd nvt-berendsen-demo
sbatch run.sh

完整 md.mdp：

integrator = md
dt = 0.001
nsteps = 20000
continuation = yes
cutoff-scheme = Verlet
coulombtype = PME
rcoulomb = 0.8
rvdw = 0.8
pbc = xyz
tcoupl = berendsen
tc-grps = System
tau-t = 0.1
ref-t = 300
pcoupl = no
gen-vel = no
constraints = none
nstenergy = 10
nstxout-compressed = 100
fourierspacing = 0.12

nvt-sd#

cd nvt-sd
sbatch run.sh

完整 md.mdp：

integrator = sd
dt = 0.001
nsteps = 20000
continuation = yes
cutoff-scheme = Verlet
coulombtype = PME
rcoulomb = 0.8
rvdw = 0.8
pbc = xyz
tcoupl = no
tc-grps = System
tau-t = 0.1
ref-t = 300
pcoupl = no
gen-vel = no
constraints = none
nstenergy = 10
nstxout-compressed = 100
fourierspacing = 0.12
ld-seed = 27129

thermostat-tau-0.05#

cd thermostat-tau-0.05
sbatch run.sh

完整 md.mdp：

integrator = md
dt = 0.001
nsteps = 20000
continuation = yes
cutoff-scheme = Verlet
coulombtype = PME
rcoulomb = 0.8
rvdw = 0.8
pbc = xyz
tcoupl = v-rescale
tc-grps = System
tau-t = 0.05
ref-t = 300
pcoupl = no
gen-vel = no
constraints = none
nstenergy = 10
nstxout-compressed = 100
fourierspacing = 0.12

thermostat-tau-0.5#

cd thermostat-tau-0.5
sbatch run.sh

完整 md.mdp：

integrator = md
dt = 0.001
nsteps = 20000
continuation = yes
cutoff-scheme = Verlet
coulombtype = PME
rcoulomb = 0.8
rvdw = 0.8
pbc = xyz
tcoupl = v-rescale
tc-grps = System
tau-t = 0.5
ref-t = 300
pcoupl = no
gen-vel = no
constraints = none
nstenergy = 10
nstxout-compressed = 100
fourierspacing = 0.12

实际结果#

Berendsen仅用于认识旧算法，不能产生正确的动能分布，不推荐用于需要严格系综的生产统计。它的预处理会明确警告；本包-demo目录为有意演示，在确认只有这一项警告后使用grompp -maxwarn 1。不要把-maxwarn 1作为通用解决错误的方法。温度平均值接近目标并不足以证明采样正确。

表中的标准差是轨迹自身的时间波动，不是平均值误差条。

提取与画图#

返回下载包根目录：

module purge
module load gpumdkit/1.5.8
python analyze.py

得到analysis/RESULTS.json和六幅PNG曲线。脚本按XVG列名读取数据；定容任务的体积从GRO盒矢量取得。NPT若重新提取体积，使用：

printf 'Volume\n0\n' | gmx energy -f md.edr -o volume.xvg

这些例子用于学会修改输入、提交任务、检查输出和解读曲线。换成蛋白、界面或其他力场时，需要重新选择合理的截断、耦合时间与模拟长度。