October 1-5. 2018 ... Portage and Tangram can be made very efficient in a client application since: 1. They can utilize custom components from clients. 2. Interact with client mesh and data through ... Carlson, N. Truchas Reference Manual.
Portage + Tangram Parallel, Multi-material, Conservative Remapping on Unstructured Meshes A. Herring, G. Dilts, R. Ertl, C. Ferenbaugh, R. Garimella, E. Kikinzon, C. Malone, N. Ray, D. Shevitz, J. Velechovsky 27th International Meshing Roundtable October 1-5. 2018, Albuquerque, NM, USA
Targets a variety of applications – Conservative interpolation (remap) inside an Arbitrary Lagrangian-Eulerian (ALE) code – Remap between physics modules within a code – Link between two different codes
Specifically written for advanced architectures – On-node (shared-memory) parallelism – Distributed-memory parallelism
Templated on mesh and state classes for easy use in most codes
Reconstructs material interface inside each multi-material cell from cell-based volume fractions (and optionally centroids) Employs two methods of Piecewise Linear Interface Calculation (PLIC) – Volume of fluid (VOF) – Nested dissections for 3 or more materials – Moment of fluid (MOF) with Automatic Material Ordering
https://github.com/laristra/portage
https://github.com/laristra/tangram
Mesh-Free Remap
Toolbox Design
Remap from a 3D mesh to a 2D cutting plane Implemented using 3D particle-particle remap
Create your own from available or custom components
Portage and Tangram are: Open-source BSD licensed libraries Hybrid parallel (MPI + OpenMP) Templated on a mesh type
1. Convert 3D cell centers to a point set (blue) 2. Make 2D mesh, convert to another point set in 3D 3. Perform the mesh-free remap in 3D from blue to red 4. Copy data from red point set back to 2D mesh
Ideal to deal with non-conforming meshes Overall range of data matches reasonably well
Single-Material Remap Linear field ρ(x, y , z) = x + y − 3z + 10 Cubic to unstructured mesh [2] Conservative & exact
Multi-Material Remap in 2D & 3D Preservation of linear interfaces with MOF Preservation of linear material fields
Portage and Tangram are NOT: Customized for a specific application A canned solution for multi-material remap Portage and Tangram can be made very efficient in a client application since: 1. They can utilize custom components from clients 2. Interact with client mesh and data through lightweight wrappers
Applications of Portage Linked into LANL code xRage [3]
Mesh-Mesh Remap Intersection based (using R2D & R3D [1]) Multi-material nd 2 order accurate Conservative Bound preserving General polygonal/polyhedra meshes Automatic source mesh redistribution if it is not aligned with target mesh
Linking between Abaqus [4] and other LANL codes Used in Truchas [5] with a custom driver
Mesh-Mesh Remap Performance We observe good strong and weak scaling with MPI+OpenMP Intel Broadwell E5 2.1GHz 2*18 cores on node
MOF
References 1. Powel D. at. al. An exact general remeshing scheme applied to physically conservative voxelization. (JCP 2015) 2. Pouderoux J. et. al. 2D & 3D Voronoi Meshes Generation with ShaPo. (MultiMat 2017) 3. Gittings M. et. al. The RAGE radiation-hydrodynamic code. LA-UR-06-0027 (2008) 4. Steedman, D. Status of the Abaqus computational tool. LA-UR-06-5530 (2006) 5. Carlson, N. Truchas Reference Manual. LA-UR-13-28723 (2013)
Reconstruction Options in Tangram
VOF
– Multipart remapping by finding matching parts in meshes – Efficient remaps by caching intersections – Remaps in both directions by storing transposed mapping
Other types of remap and Tangram have also been shown to scale well