Dynamics in Event-driven simulations of the WCA potential

tags: Nucleation

Owners (the only one with the permission to edit the main text)

Frank, Antoine

---

Background

We are interested in testing the Event-driven molecular dynamics (EDMD) simulation method introduced in:
E. A. Peters, Rejection-free Monte Carlo sampling for general potentials., Physical Review E 85, 026703 (2012).

As a testing ground, we use the WCA (Weeks-Chandler-Andersen) model at low temperature.

Questions we have:

• How do dynamics compare between conventional molecular dynamics simulations and EDMD simulations. In particular, it would be good to compare diffusion and structural relaxation. This could be done in monodisperse systems or binary systems, or both.

• How does softness affect nucleation in the low-temperature limit. This is inspired by: [Dasgupta, Coli, and Dijkstra, Tuning the glass transition: enhanced crystallization of the laves phases in nearly hard spheres. ACS nano, 14, 3957-3968 (2020).] There, conventional MD simulations were used, but these cannot reach temperatures as low as can be reached with EDMD

• How does the simulation speed compare? (Monodisperse)

Preliminary work/data

I have added to the NAS all of my data in the /home/share/WCA_Data folder. There is a folder both for the data I kept locally ("Local") for most of my analyis, and a folder with the data from the cluster ("Ceres"). Some data will be duplicated between them.

Local data

Root folder:

The root folder contains the EDMD simulation code (mdbinary.c and associated code files). By default, the code reads an initial configuration file ("init.sph") and runs it at a temperature specified in "temperature.dat". Simulation length, snapshot frequency and other parameters can be adjusted inside the code.

There is also a Mathematica notebook with some analysis ("WCA.nb"), and its contents printed to a PDF file ("WCA.pdf"). This is not particularly well organized, but may be helpful for some reference plots.

Data:

Contains data copied from Ceres (this will be partially duplicated from the data in the "Ceres" folder). There are three subfolders containing different sets of simulations. In each, there is a file called "cpscript" which I used to synchronize data from the cluster. Additionally, there are some scripts to do analysis.

FQT and FQTscale:

Contains code for calculating the intermediate scattering function

$F(q,t)$ from snapshots generated on a logarithmic time scale. These are currently generated for the simulations in the "Data/FQT" folder. The "FQTscale" folder is the one I actually used, and has the additional feature of reading the intented value of

$|q|$ from the command line as well. See the script "fqtscript" in "Data/FQT/" to see how it is used (I think that one uses a local copy of the FQTscale code).

Grow

Contains code for creating configurations at a desired density / composition / number of particles.

HS

Contains code and data (copied from Ceres) for true hard spheres.

Lammps

Contains LAMMPS scripts and data from Ceres for simulating the WCA model using conventional MD. Also included is a variation on the FQT code that works on the LAMMPS file format.

MC

Seems to contain a Monte Carlo code for Lennard-Jones… Not sure I used it for anything, maybe just as a reference datapoint when setting up the codes.

MLOP

Contains code for the Neural-Network based order parameter developed in [Boattini, Ram, Smallenburg, and Filion,** Molecular Physics 116, 3066-3075 (2018).] The analysis is done in a python script, and there are some ugly bash scripts for translating between my file format (.sph) and the file format expected in the script (.xyz). It would probably be better to adapt the python code. (It might also be good to include a cell list in the neighbor detection routine if the analysis is slow…)

The code takes in a snapshot and produces a new one where particle are assigned types based on their local environment. It might be good to output the number of non-fluid particles as well.

Probably useful to consult the paper to see which particle labels correspond to which environments.

Mono

Is probably code for a monodisperse system. Not really used at this point

Ceres data

This contains various sets of simulations linked to the data already mentioned above.

Equil

Equilibration of different systems, so starting from a freshly grown configuration.

FQT

For measuring the structural relaxation

HS

For pure hard-sphere simulations

Lammps

For runs using the simulation package Lammps

Nucleate

Long runs where the goal is to see whether the system nucleates spontaneously for different densities and temperatures.

---

To do:

• Run some monodisperse simulations (fluids) in both EDMD and Lammps to get familiar

• Do some benchmarking: compare the time it takes to simulate

  ${\tau }_{\alpha }$ in Lammps vs. EDMD. (Dependence on Lammps time step, system size, etc.)

• Compare F(q,t) between the two methods for different densities, temperatures, maybe some different q's.

• Examine nucleation in the "Nucleate" folder using the MLOP python code.

• Try the FQT code on both

Points to be careful about

• WCA particles have a diameter which is larger than

  $\sigma$ (where
  $\sigma$ is the one used in the interaction potential.)

• The paper by Dasgupta et al. has the phase diagram, and they compare things in terms of an effective packing fraction.

Benchmarking

• As a function of system size N

Squares: soft-EDMD
Circles: LAMMPS

Relative :

• EDMD shell thickness dependency:

~ 2-fold increase in sim. rate for st going from 0.1 to 0.5 @ T = 1.0