Colvars atomsOfGroup and targetEquilSteps

LAMMPS LAMMPS Beginners
1

in my real application, im trying to calculate the PMF using umbrella sampling along rmsd of the alpha-carbons between two conformations of an enzyme solvated in various calcium concentrations.

since the systems generated by charmmgui are too large to attach, i made a minimal (non-working) example using similar rhodo benchmark (protein, charmm, pppm, npt).

# colvars-rhodo.in
shell cd ../lammps-alphataubio/bench
include in.rhodo
shell cd ../../colvars-rhodo
fix 3 all colvars colvars-rhodo.colvars output colvars-rhodo seed 12345 tstat 2
run 100

# colvars-rhodo.colvars

atoms {
  name foo
  atomNumbersRange 1-3
}

colvar {
  name quux
  rmsd {
#    atoms { atomsOfGroup foo }
    atoms { atomNumbersRange 1-3 }
    refPositions { (-6.80572,-9.10264,-32.2296) (-6.81746,-9.4978,-33.1915) (-5.86092,-9.04028,-31.7994) }
  }
}

harmonic {
  colvars quux
  centers 0
  targetCenters 10
  targetNumStages 9
  targetNumSteps 10
  targetEquilSteps 9
  outputFreq 1
  outputCenters on
  outputEnergy on
  writeTIPMF on
}

“Description: This parameter defines a unique name for this atom group, which can be referred to in the definition of other atom groups (including in other colvars) by invoking atomsOfGroup as a selection keyword.”

First, i cant figure out how to use atomsOfGroup to define a group of atoms and refer to it later. there’s no example in colvars manual, or lammps examples directory. Nothing in previous matsci posts either.

colvars: # atomsOfGroup = “foo”
colvars: Error: cannot find atom group with name foo.
ERROR on proc 0: Fatal error in the collective variables module.
(src/COLVARS/colvarproxy_lammps.cpp:294)
Last command: run 100

"Keyword targetEquilSteps⟨⟨Number of steps discarded from TI estimate⟩⟩
Context: harmonic
Acceptable values: positive integer
Description: Defines the number of steps within each stage that are considered equilibration and discarded from the restraint free energy derivative estimate reported reported in the output."

Second by “tugging” on a protein in an unphysical manner at each stage of steered MD, i want to do ~90% of equilibration steps per window to let the protein settle into realistic intermediate structures along the way before collecting PMF measurements. However i get “Error: keyword “targetequilsteps” is not supported, or not recognized in this context.” but im using targetequilsteps in the harmonic context as described in colvars manual.

colvars: Initializing a new “harmonic” instance.
colvars: # name = “harmonic1” [default]
colvars: # colvars = { quux }
colvars: # stepZeroData = off [default]
colvars: # outputEnergy = on
colvars: # outputFreq = 1
colvars: # timeStepFactor = 1 [default]
colvars: # writeTISamples = off [default]
colvars: # writeTIPMF = on
colvars: # centers = { 0 }
colvars: # targetCenters = { 10 }
colvars: # targetNumSteps = 10
colvars: # targetNumStages = 9
colvars: # outputAccumulatedWork = off [default]
colvars: # outputCenters = on
colvars: # forceConstant = 1 [default]
colvars: # decoupling = off [default]
colvars: # targetForceConstant = -1 [default]
colvars: The force constant for colvar “quux” will be rescaled to 1 according to the specified width (1).
colvars: Error: keyword “targetequilsteps” is not supported, or not recognized in this context.
ERROR on proc 0: Fatal error in the collective variables module.
(src/COLVARS/colvarproxy_lammps.cpp:294)
Last command: run 100

$ lmp_beluga -h

Large-scale Atomic/Molecular Massively Parallel Simulator - 7 Feb 2024 - Development
Git info (develop / patch_7Feb2024_update1-366-g069919ddcd)

OS: Linux "Rocky Linux 8.9 (Green Obsidian)" 4.18.0-513.24.1.el8_9.x86_64 x86_64

Compiler: GNU C++ 11.3.0 with OpenMP 4.5
C++ standard: C++17
MPI v3.1: Open MPI v4.1.4, package: Open MPI [email protected] Distribution, ident: 4.1.4, repo rev: v4.1.4, May 26, 2022

Accelerator configuration:

KOKKOS package API: CUDA OpenMP Serial
KOKKOS package precision: double
Kokkos library version: 4.3.0

FFT information:

FFT precision  = double
FFT engine  = mpiFFT
FFT library = KISS
KOKKOS FFT engine  = mpiFFT
KOKKOS FFT library = cuFFT

Installed packages:

ASPHERE CG-DNA CG-SPICA COLVARS EXTRA-DUMP KOKKOS KSPACE MC MOLECULE RIGID
2

Hi @alphataubio!

  1. Groups must currently be defined in the context of a colvar, so unless you have more than one colvar using the same group there is no point in using atomsOfGroup.
  2. targetEquilSteps is a keyword that only works when you change the force constant, but currently not when you change the restraint centers. To get a PMF from a steered-MD simulation, you either use the accumulated work (typically very crude on MD timescales), or extract snapshots from the SMD trajectory to initialize umbrella sampling windows.

Giacomo

3

for my applications in medicinal chemistry, i tried TI, ABF, steered-MD, umbrella sampling, and metadynamics.

the only one that works is metadynamics, all the other ones are a blooper reel of unphysical behaviors with the protein ripping itself apart when looking at the lammps trajectories.

due to very small and infrequent communication overhead between walkers, multiple-walkers metadynamics makes a lot of sense when scaling because each walkers can have it’s own node with lammps domain decomposition of ~80 cores, instead of decomposing a single metadynamics walker into a domain of 1600 cores with a lot of communication overhead for atoms and forces.