Title:	Database for developing, testing, understanding, and launching vdW-DF-method progress
DNr:	NAISS 2023/6-306
Project Type:	NAISS Medium Storage
Principal Investigator:	Per Hyldgaard <hyldgaar@chalmers.se>
Affiliation:	Chalmers tekniska högskola
Duration:	2024-01-01 – 2025-01-01
Classification:	10304 10302 10407
Homepage:	http://fy.chalmers.se/~hyldgaar/SNIC
Keywords:

Abstract

This application concerns storage that we need to succeed at our overall goals: Developing the van der Waals (vdW) density functional (vdW-DF) method and using it for green technology solutions. Some background is provided in the report for SNIC2022/6-286 (which this proposal extends) as well as in the application for a NAISS-Large-Compute allocation (related to 2024). We seek medium-size storage for two purposes. First is to secure sufficient disk space (TBs and number of files) to actually utilize what will hopefully be a NAISS large-compute allocation relevant for our 2024 research -- as well as for use to use a smaller allocation of C3SE computing resources (expected to be granted to our Chalmers departments). Second is to allow us to continue to grow our vdW-DF database (primarily located at C3SE) in what is by now two types of ways. We need such collected storage to facilitate an easy overview of what functional-development idea works on what by which means -- and we therefore always keep what is considered core DFT results (structure, energies, forces and stresses, as well as their variation under for example strain) -- while constantly producing more by large-scale computing. Beyond that, and excitingly, we are also increasingly bringing a whole new dimension in using DFT for a in depth (if TB-costly) analysis that uses the full range of DFT-run outputs. In short, we now see clear benefits from now dramatically expanding the information-content of what we keep for some classes of individual database entries. The key observation is here that we have made formal progress (for example, a new formally exact DFT of electronegativity that we extract simply from postprocessing DFT output) and have emerging practice in converting such full-DFT-output into analysis of the nature of binding contributions and charge transfer. It costs TBs of disk space but for key classes of problems there are clear advantages to completing such analysis. This is, for example, true for ongoing work to improve DNA markers, something that we shall in 2024 intensify within the new Chalmers Spectral Design Initiative that we are starting. In 2024-2025 Chalmers NanoEI backs the planned January arrival of two postdocs that will start using our increasing collection of data, and who will also intensify the computation of additional DNA-related data. In addition we will in 2024 be working towards a full physics tuning of our brand new range-separated hybrid (RSH) vdWDFs, a tool that can control DFT errors because we now have a rigorous theory of electron transfer. The challenge is only computational and storage cost of completing the physics tuning, which means running jobs at different assumptions. We can then compare our new type of ionization predictions with traditional DFT comptations, to thus establish the value of what is otherwise one free parameter. We suggest a work flow (under Resource Usage), but it means that will need more storage this year (so we ask for more at PDC).