Space Charge (FFT): Invalid Particle (SP?)

[ax3l]

Might be related to single precision support in ImpactX #899, especially the s-to-t / t-to-s transforms.

In some of the space charge benchmarks that I perform right now (ImpactX development close to release 25.08), I see the following error:

RuntimeError: config 'impactx-6cpu-autovec' w/ 1000000 particles failed with: amrex::Abort::0::ParticleContainer::locateParticle(): invalid particle. !!!
SIGABRT
See Backtrace.0.2 file for details

File that causes this:
spacecharge_cheetah_impactx.py
Note: the issue does not appear in every run, I saw this in one of five in my laptop.

ImpactX build:

ImpactX build configuration:
  Version: 25.07 (25.07-15-g6326301b6459)
  C++ Compiler: GNU 14.3.0 
    /home/axel/micromamba/envs/benchmark-cpu/bin/x86_64-conda-linux-gnu-c++

  Installation prefix: /home/axel/micromamba/envs/benchmark-cpu
        bin: bin
        lib: lib
    include: include
      cmake: lib/cmake/ImpactX
     python: lib/python3.13/site-packages

  Build type: Release
  Build options:
    APP: ON
    COMPUTE: OMP
    SIMD: OFF
    IPO/LTO: OFF
    FFT: ON
    LIB: static
    MPI: OFF
    PRECISION: SINGLE
    PYTHON: ON
    OPENPMD: OFF

[ax3l]

While the above is a sporadically failing test with a relativistic (100 MeV) electron beam, it is easier to debug the algorithm for sub-relativistic cases and then come back to it, e.g., few-100 MeV proton beams.

I posted here a very simple test for development: #899 (comment)

and our CI also has aforementioned sub-relativistic examples we can study first.

We might be able to re-formulate/taylor the expressions in the transformations where we have terms like -1.0 + 1.000001 first:

• https://github.com/BLAST-ImpactX/impactx/blob/development/src/particles/transformation/ToFixedT.H
• https://github.com/BLAST-ImpactX/impactx/blob/development/src/particles/transformation/ToFixedS.H

[ax3l]

It looks like now in #899 we "only" have MLMG convergence issues left (and analysis tolerances).