Nek5000 / NekROM

Model-Order Reduction Framework for Nek5000

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

NekROM - Model Order Reduction Framework for Nek5000

This package include tools to help apply model-order reduction (MOR) to data produced by Nek5000 to generate reduced-order models (ROM). The generated ROMs can be run either in the Fortran driver embedded inside the Nek5000 userchk subroutine or can be run separately by provided driver scripts in Matlab, Python, and Julia. Users can also provide their own drivers which read the ROM operators and QOI factors from the ops/ and qoi/ directories.

Setup & Procedure

Set shell variables (in .bashrc for BASH users):

export MOR_DIR='/path/to/NekROM'
export PATH="$MOR_DIR/bin:$PATH"

Required files in NekROM case directory:

  • Nek5000 case files e.g., .rea, .map, SIZE
  • $caserom.usr, .usr file specific for NekROM cases (see $MOR_DIR/examples)
  • LMOR, specifies compile-time parameters
  • $case.mor, specifies run-time parameters
  • file.list, contains list of paths to the snapshots (relative path)

Optional file:

  • avg.list, contains list of paths to the average files

After ensuring the required files are in the case directory, run makerom $caserom to make a Nek5000 executable for ROM.

Parameters

Compile-time parameters (for setting memory allocation size) can be found in LMOR.

  • ls, maximum number of snapshots
  • lb, maximum number of total modes

run-time parameters can be found in $case.mor.

  • [general], header for general parameters
    • mode, off = offline, on = online, all = offline + online
    • field, v = velocity, t = temperature, vt = velocity + temperature
    • nb, number of POD modes (must be less than lb, default == lb)
  • [pod], header for pod parameters
    • type, l2 = L^2 POD modes, h10, H^1_0 POD modes
    • mode0, avg = average 0th mode, state = user-defined in ub,vb,wb,tb
    • augment, 0 = no ABM, 1 = 0th interactions, 2 = diagonals, 3 = 1 + 2
  • [qoi], header for qoi parameters
    • freq, frequency of QOI dump, if <1 freq=iostep
    • drag, drag based on OBJ data

Contribution

Our procedure for updating the code is exclusively through pull requests (no pushing). Please submit issues and PR to github.com/SEAL-UIUC/NekMOR. PRs should be the smallest coherent change to the code-base. Issue titles should describe the issue e.g., 'Error in x', 'Missing x', etc. PR titles should describe the modification made e.g., 'Fixed x', 'Improved x, etc. See conventions.txt for the coding style of this project when contributing.

Parameter File Support

In addition to the .rea support for setting internal parameters, .mor files are supported for par-like dictionary. The possible key/values are described in templates/mpar.template.

About

Model-Order Reduction Framework for Nek5000


Languages

Language:Fortran 92.2%Language:Julia 3.8%Language:Shell 1.9%Language:MATLAB 1.1%Language:Python 0.8%Language:NASL 0.3%