Energy drift in NVE (multiple time step algorithm)

User discussions
1

When I use the double precision GROMACS (2024.4) to run NVE simulations with multiple time step algorithm, I find that the kinetic energy drift speed was very fast, with a drift speed greater than 1000 J/mol · ns, while the kinetic energy drift speed of the mixed precision GROMACS simulation under the same parameter settings was less than one percent of the double precision. Why did this phenomenon occur? I would be extremely grateful for any of your opinions or suggestions.

Here are the MDP parameter settings I used when running NVE simulation:

; Run parameters
integrator              = md
nsteps                  = 10000000
dt                      = 0.001
mts                     = yes
mts-levels              = 2
mts-level2-forces       = longrange-nonbonded
mts-level2-factor       = 2
; Output control
nstxout                 = 10000
nstvout                 = 10000
nstenergy               = 10000
nstlog                  = 10000
; Bond parameters
continuation            = yes
constraint_algorithm    = lincs
constraints             = none
lincs_iter              = 1
lincs_order             = 4
; Nonbonded settings
cutoff-scheme           = Verlet
ns_type                 = grid
nstlist                 = 1
rcoulomb                = 1.2
rvdw                    = 1.2
DispCorr                = EnerPres
; Electrostatics
coulombtype             = PME
pme_order               = 4
fourierspacing          = 0.16
; Temperature coupling is off
tcoupl                  = no
; Pressure coupling is off
pcoupl                  = no
; Periodic boundary conditions
pbc                     = xyz
; Velocity generation
gen_vel                 = no
2

There can be a large cancellation of errors, in particular in mixed precision. You should lower the verlet-buffer-tolerance and also increase LINCS parameters, if LINCS is ued.

3

Thank you very much for your suggestion. I will give it a try.