Instrumenting Your Simulation with Catalyst for In-Situ Analysis

As simulation runs produce data faster than the filesystem can absorb, analyzing it while it is still in memory becomes increasingly important. On GPU and leadership-class systems, writing full state to disk at a useful cadence may no longer be viable. Instrumenting an existing simulation for in-situ analysis requires connecting Catalyst to the code, correctly describing the data, and configuring the build.

Catalyst supports this workflow through the Catalyst API and Conduit Mesh Blueprint. The webinar will focus on how these pieces come together in an AMReX code and how to troubleshoot common integration problems.

During this webinar, Kitware’s experts will:

• Walk through instrumenting an AMReX code from start to finish, from an existing time loop to live in-situ visualization. The demonstration will cover the Catalyst API and required CMake changes.
• Explain how to describe mesh and field data with the Conduit Mesh Blueprint. This includes the amrmesh channel for AMReX data and when arrays can be passed by pointer or deep-copied.
• Show how to keep Conduit consistent across the simulation, Catalyst, and ParaView. Learn how to recognize common failure signatures and troubleshoot problems using Blueprint verification, ghost-cell checks, and symbol checks.
• Generate a Catalyst script from a ParaView session and switch between extract-only and in-situ rendering pipelines at runtime.

The webinar will also address ghost-cell handling in in-situ AMR and current work at the VTK and AMReX level. It will close with a preview of instrumenting MFEM with Fides and cell-grid support for high-order elements.

Key Takeaways

• Instrument an existing simulation code by placing catalyst_initialize, catalyst_execute, and catalyst_finalize in the time loop.
• Describe mesh and field data with the Conduit Mesh Blueprint, including the amrmesh channel for AMReX data.
• Keep Conduit consistent across the simulation, Catalyst, and ParaView and recognize common failure signatures.
• Generate Catalyst scripts from ParaView, switch pipelines at runtime, and troubleshoot integration problems.

Who Should Attend

This intermediate-to-advanced webinar is intended for:

• Computational scientists and research software engineers maintaining production simulation codes.
• Simulation code architects, technical leads, and project PIs evaluating in-situ strategies.
• HPC application developers and performance engineers facing I/O bottlenecks at scale.
• Research scientists working with ensemble, UQ, optimization, or ML surrogate applications.
• Developers who previously evaluated Catalyst 1.
• Ascent and SENSEI users interested in Catalyst capabilities.

Attendees should be comfortable with C++ or Fortran, MPI, and CMake.