Skip to content

Latest commit

 

History

History
146 lines (126 loc) · 5.84 KB

meet-004_2023-11-17.md

File metadata and controls

146 lines (126 loc) · 5.84 KB

Phase Field Schema WG: Meet 3

hackmd-github-sync-badge

Meeting agenda, notes and actions for 2023-11-17.

Organizer : Daniel Wheeler

Attendees : - Daniel Wheeler (he/him)

  • Kasra Momeni (he/him)
  • Olga Wodo (she/her)
  • Trevor Keller (they/them)
  • Marvin Tegeler (he/him)
  • Steve DeWitt (he/him)

Highlevel Categories for Use Cases

There are a number of categories to think about when considering use cases.

  • Comparing the results of phase field simulations (e.g. order of accuracy comparisons)
  • Comparing the performance of numerical simulations (e.g. run time, memory usage)
  • Reusing a phase field simulation as part of larger workflow (e.g. coupled with CALPHAD or part of iterative loop in materials design)
  • Reusing a phase field simulations independently (structure-property mapping using generated phase field micorstrucutres, subsequent AI calibration on multiple different phase field simulations)
  • Reusing phase field simulations for subsequent related simulations, but extended for new phase field simualations (problem specific models with unique free energy or additional terms in free energy or governing equations, new application with similar equations)

Links

Agenda

  1. Any questions or items to raise for discussion
  2. Discuss AI use case (the pull-request)
  3. Kasra use case
    • In a hackpad and we can flesh this out together
  4. Flesh out schema ideas from use cases
    • Below is an extensive list of possible areas to address
    • Let's add more ideas and then whittle down to something manageable

Schema Ideas

  • Problem specification:
    • DOI (or not) link to a problem specification
    • Keywords related to the phase field application and typically used for materials ontologies (e.g. "additive-manufacturing" or "cobalt-alloy")
    • A phase field equation category (e.g. "cahn-hilliard" or "allen-cahn")
    • Number of phase fields
    • Number of components
    • Number of equations
    • Type of free energy equation?
  • Retrospective provenance metadata:
    • Memory usage, run time
    • mesh actually used, degrees of freedom
  • Prospective provenance metadata:
    • Computer architectures, nodes, threads etc
    • Control parameter settings, command line options, environment variables
    • parallel approach (shared versus distributed)
    • Numerical approach
      • meshing strategy, nominal degrees of freedom
      • non-linear solver strategy
      • linear solver strategy
        • preconditioner
        • iterative versus direct
      • time-stepping strategy
        • implicit, explicit, semi-implicit
      • discretization strategy
        • expected order of accuracy
        • FD, FV, FV, pseudospectral
      • Was the numerical scheme stable, or did it go to NaN?
  • Data Sharing:
    • Suggested naming conventions, formats for
      • time
      • phase field global quantities
      • field data
    • Suggested storage formats
    • URL or local paths to data files
    • DOI of data files
    • Domain / mesh file details if relevant for reading data
    • Formatting details
    • Address possible AI requirements
      • How to deal with 1000s of data files?
      • Some categorization of microstructure or link to a more details microstructure descriptor?
        • Number of phase or components
        • keywords: spinodal, laminar, dendritic
      • is the data intended for AI calibration?
      • Succinct stats about data sets
        • Number of features
        • Number of samples
        • Size
      • Post-processing techniques
      • What is the intention of the data set?
        • Is it balanced
        • Are there various categories?
      • Link to labels if relevant (global derived quantities, properties etc)
    • Does the dataset represent a time-evolution or a stationary solution?
  • Environment Execution:
    • Link to software framework (Codemeta) + repository + version
    • Facilitate execution
      • Link to container image
      • Link to environment generation files (e.g. environment.yml, shell.nix or Dockerfile)
    • Link to relevant input files + commit ID
    • implementation repository link + commit ID
  • Findability:
    • Implementation repository link + commit ID
    • DOI of implementation repository
    • DOI of output files

Action Items