Run convergence studies for 3d flow simulations

[haakon-e]

Using the results of #19, run convergence studies and summarise results

[haakon-e]

How does number of cells grow in a grid?

• One possible sequence is: 100, 800, 6400, 51200.
• It is unclear at the moment what the limits of the computer is, and how this limit changes as we introduce 1) more grids (fractures), and 2) more equations (Biot, contact mechanics). What are common parameters?
• Currently, we're simulating with dt=1hour, source at center of domain with injection rate, 1/6 l/s What parameters are you changing?
• Currently, we're only varying the permeability

[haakon-e]

We are only able to run convergence studies for unstructured fixed-dimensional (3d) grids with incompressible flow.

This is because time-stepping is not stable for MPFA on unstructured grids (with relatively small time-steps).

Test case 1

Incompressible flow on unstructured grid with (mostly) unit parameters

• Grid sizes: 24, 192, 1536, 12288 (ref)
• No negative cells in any solutions
• See class: refinement.tests.test_run_convergence_study.TestRunModelForConvergenceStudy
  • And test method: test_0_frac_unit_domain.
• Parameters are equal to those in commit: 2922127

Results:

• Relative errors, increasing order of grid cells
• Test run time: 30m
  1. 1.33 (24 cells)
  2. 0.97 (192 cells)
  3. 0.60 (1536 cells)

[haakon-e]

See also #24

[haakon-e]

Note: The aforementioned simulations (with restrictions) have been run on non-optimized grids. These are known to probably be quite bad. As such, we might revisit this again in the future on optimized grids.