[Feature] Partial deintercallation in VaspCatMatWorkChain

[ezpzbz]

[Feature] Partial deintercallation in VaspCatMatWorkChain

[ezpzbz]

After all, it is not a feature which want to have directly in the workchain. Still workchain can be improved to address some issues listed below but delegating the job of generating deintercalation composition within the workchain can cause unwanted issues.

potential issues:

• Number of configurations can be relatively high (hundreds of structures). If we design the workchain to submit them in sequentioal order, it would take forever to finish a set of calculation on one compound. If we go with parallel submission of jobs which was the initial idea, then we can easily the user job limit in many HPCs which will cause extra human intervention for re-playing paused processes.
• Generation of configurations via enumeration can lock the GIL. Consequently, it can cause whole workchain fail as I was facing the issue in heavy parsing of OUTCAR file.

possible solutions:

• I am working on some solution to select a subset of deintercalated compounds which is a work in progress based on featurizing the structures and select a limited number of diverse set among them. This way we can limit number of calculations via limiting number of structures.
• The other option in case of being forced to run calculation on all structures, would be having a control parameter to not submit jobs in parallel more than allowed number by HPC.
• Or I can test the aiida-fireworks plugin which is new and intended for buffering jobs on HPC. Briefly, we will submit all jobs at once, then this plugin will get them and will take care of sending them to nodes.

[ezpzbz]

Having this feature does not make much sense after all. So, I'm closing it.