PyLith Development Plans, Feb 2014

Priorities for PyLith software development, such as new features and enhancements. This a draft for community comment (Feb 20, 2014).

This plan attempts to balance meeting short-term objectives of delivering high priority, new features and meeting long-term objectives of extending the code to solve a broader range of scientific problems.

Status: We have almost everything working. We are in the process of fixing a few bugs related to creating cohesive cells and running in parallel.

1. Replace C++ Sieve implementation of finite-element data structures with C DMPlex implementation. expert [99%]
   • DMPlex provides a simpler, more efficient implementation of the finite-element data structures that conforms to the PETSc data management (DM) interface. This provides tighter integration with the rest of PETSc. Additionally, this rewrite of the data structures results in a more efficient memory layout, resulting in better performance.
2. Switch from using Subversion to Git for version control.
3. Add ability to recursively refine a mesh.

This is the version that will be available for use at the June 2014 workshop. We are behind schedule for getting multiphysics done by then, and because this is a fixed deadline, we will probably aim to get some additional less ambitious features completed. Allowing different startup cases could slip to version 2.2.

1. Improve fault formulation for spontaneous rupture [10%]
   • Removes inner solve associated with updating Lagrange multipliers. This will significantly accelerate the nonlinear solve.
2. Higher order basis functions [0%]
   • Allow user to select order of basis functions independent of the mesh (which defines the geometry). This permits higher resolution for a given mesh.
3. Reorganize top-level code to conform to layout needed for multiphysics [0%]
   • Setup modular approach for specifying governing equations and computing residuals and Jacobians.
4. Reorganize top-level code to allow different startup cases [0%]
   • Elastic prestep
   • User-specified initial solution
   • Checkpoint via special spatial database?
5. Radial basis functions for spatial databases [0%]
6. Improved handling of buried fault edges [25%]

1. Multiphysics
   1. Incompressible elasticity via a pressure field
   2. Elasticity + heat flow
   3. Elasticity + fluid flow
2. GUI interface for specifying parameters
3. Switch to using PETSc time-stepping (TS) algorithms. [0%]
   • Replace simple Python-based time-stepping implementations with PETSc time-stepping algorithms that provide support for higher order discretization in time and real adaptive time stepping.
4. Multilevel nonlinear solve

1. Earthquake cycle modeling
   • Same mesh for dynamic and quasi-static parts (dynamic → quasi-static, quasi-static → dynamic, complete cycle)
2. Create strain hardening/softening 2-D and 3-D Drucker-Prager elastoplastic models.
3. Moment tensor point sources via equivalent body forces [5%]
   • Moment tensor point sources provide a mesh independent deformation source that is better suited for Green's function calculations than slip on a fault surface via cohesive cells.

• Major features
  1. Earthquake Cycle Modeling
     • Different meshes for dynamic and quasi-static parts
       • Requires interpolation of fields between different meshes/discretizations and may require extrapolation of solutions when quasi-static problems span a larger domain than the dynamic problems.
  2. Data assimilation
     • Use flexibility of multiphysics organization to support inclusion of data assimilation

• Minor features
  1. Use KD tree search algorithm to allow output of time histories at an arbitrary location
  2. Combined prescribed slip / spontaneous rupture fault condition
     • Use fault constitutive model to control slip on fault except during episodes of prescribed slip. Need some way to describe when to turn on/off prescribed slip.
  3. Use threading to accelerate integrations on multi-core machines.