search

lanl / walkabout

Walkabout simulates random walk particle tracking.
http://walkabout.lanl.gov/
Other
2 stars 5 forks source link

Amanzi-Walkabout workflow results in divots along non-convex boundary #11

Open millerta opened 6 years ago

millerta commented 6 years ago

The current particleDriver workflow produces mesh points located at Amanzi cell centers and cell vertices at the boundaries. The result are missing elements along the non-convex top surface. This works ok for a box shaped (convex) mesh, but does not work for non-convex, non-Delaunay mesh. Any work around will result in missing tet elements, or inverted tet elements along the boundary. Walkabout Particles stop or have undefined behavior at these locations.

millerta commented 6 years ago

Consider splitting each hex (or prism) into tets for Walkabout (instead of the connect-delaunay scheme).

From Konstantin: (1) a splitting pattern with translational properties in x-, y- and z- direction may not be optimal from viewpoint of shape-quality of resulting tets. (2) quality of tets may degrade for topological hexes. (3) general (optimized) splitting pattern will require more code development inside or outside Amanzi since Amanzi does not have all tools typical for mesh generators.

Possible solutions: (a) Use external tools for mesh partitioning, e.g. MSTK, LaGriT, etc. (b) Use finite elements for hexes and prism to represent flow in Walkabout. (b) Impose rigid conditions on the input mesh that Amanzi will assume in underlying algorithms. (d) Use symmetric splitting: 24 for hexes and 12 for prisms

millerta commented 6 years ago

Preparing test directories for a new workflow. Sent Konstantin hex and tet 5, 6, 24 meshes. Full test directory with Amanzi input exodus files, Amanzi-Walkabout, Walkabout particle tracks: /scratch/nts/tam/ascem/walkabout/test_amanzi

lipnikov commented 6 years ago

Amanzi's pull request implementing new interface to Walkabout has been created.