- For our PtRu manuscript, we desired to include a degree-of-rate-control (DRC) plot in order to bolster our argument for why Pt75Ru25/C is experimentally observed to yield the highest NO3RR activity.
- Our previous work ran these microkinetics simulations at 0.1 V vs. RHE but did not include DRC results. We reran these calculations at 0.1 V vs. RHE with similar inputs and with DRC analysis turned on.
For reproducibility, here is a description of the tools and environment used to run the simulations.
- Calculations were performed on KNL nodes of the the Cori supercomputer at the National Energy Renewable Science Computing Center (NERSC).
- To run Docker images on NERSC, the Shifter tool was used. Shifter allows one to run programs inside Docker images without having to specify volume mapping options to access the local file system.
- Version 2.7.0 of the MKMCXX microkinetic modeling software package was used for all calculations. This version is no longer available at the author's website, but is available as a Docker image through Docker Hub.
- GNU Parallel 20190222 was used to parallelize individual microkinetic simulations. Its usage is shown in the code below.
- The Conda environment used in the Jupyter notebook that reads simulation results and produces plots is given in
conda-package-lock.yml.
Do the following while in the 0.1V-vs-RHE directory and with Python 2.7 accessible from the path as /usr/bin/python:
# Generate base simulation files
bash ./generate.sh
# Create run commands for just those simulations where
# all surface activation barriers are positive
/path/to/python2.7 ./get1_rate.py
# Execute the simulations
bash ./runTask-cori-knl.slurm
# Or, if on a Slurm cluster:
sbatch runTask-cori-knl.slurm- The file
input.mkmis a template for the input file for a single MKMCXX microkinetics simulation. Its syntax is demonstrated through an introductory tutorial here, as well as a series of example files included in the program download. Various options for the simulation are documented here.- In our code,
input.mkmcontains all of the parameters necessary to predict the NO3RR turnover frequency and product selectivities for a single set of N and O binding energies, at a given potential. - These parameters include the Hertz-Knudsen adsorption energies for adsorption/desorption of species to/from the surface (given in the lines labeled
HK;) as well as forward and backward Arrhenius barriers for surface reactions and some adsorption/desorption reactions (given in the lines labeledAR;).
- In our code,
- The file
Geninput.pyimplements the linear adsorbate and BEP scaling relationships. It takes in N and O binding energies (read through standard input) as well as an applied potential (hard-coded in thePotential = 0.1 * 96.485line). The convention for this simulation is that binding energies are in kJ/mol and that more positive binding energies indicate stronger adsorption. To change the applied potential to something other than 0.1 V vs. RHE, you should do a global search-and-replace operation on all files in this folder to replacePotential = 0.1 * ...withPotential = <new potential> * .... Through arithmetic and string substitution, this Python program takes a single set of N and O binding energies (and the applied potential) and produces a single MKMCXX input file with the correct Hertz-Knudsen and Arrhenius parameters for those binding energies and potential. - The file
generate.shestablishes a grid of N and O binding energies ranging from 200 kJ/mol to 700 kJ/mol in both variables. It then callsGeninput.pyfor each pair of N and O binding energies and places the generated MKMCXX input file into its own directory. GNU Parallel is employed here to accelerate the generation of these ~2k folders. - The file
get1_rate.py(which also contains the hard-coded applied potential) writes out commands to execute each one of the simulations. (This list of commands makes it easier to execute simulation jobs in parallel later on.) However, it writes out commands only for those simulations for which the specified N and O binding energy cause the Arrhenius barriers of all the surface reactions to be positive. It does this by hard-coding the BEP relationships for these reactions. With adsorbate scaling and BEP relations, it is possible to specify pairs of N and O binding energies that lead to unphysical, negative reaction barriers. This file ensures that those points are not considered for simulation. - The file
worker.shis the program that is ultimately executed once per simulation. It reads from standard input the name of the folder whose simulation is to be run, checks to ensure that MKMCXX has not already completed work in this folder, and then executes MKMCXX via the Docker image. - The file
runTask-cori-knl.slurmis a Slurm submission script specific to the Cori supercomputer at NERSC, but could be adapted for a generic computing cluster environment. This script loads GNU parallel and executes it on the command listcommands.txtwhich was generated byget1_rate.py. All simulations used KNL nodes on Cori, which have 68 physical cores and 4 hardware hyperthreads per core. Theinput.mkmtemplate uses theNPAR=4setting to instruct each MKMCXX instance to use 4 hyperthreads, and we run up to 66 such processes at any time. The outcomes of these jobs and their completion status is recorded to ajoblogfile in the submission directory, so that GNU Parallel can be quit and restarted at any time without repeating completed jobs. This enables us to set a low minimum wall time on Cori, enabling faster job turnaround and cheaper job cost through use of the flex discount.
- A Jupyter notebook (
01-jul-2020_ptru-kinetics-plots.ipynb) is used to parse and read results and produce DRC and TOF plots based on microkinetics data. This notebook makes heavy use of themkmcxxPython module, which is included. - The usage of this module to process results and produce plots appears below.
- Here is the generated degree of rate control plot:
- Here is the generated turnover frequency plot:
