search

Libbum / Wafer

Parallelized 3D FDTD Schrödinger Equation Solver
MIT License
21 stars 2 forks source link

Arbitrary NaNs #31

Open Libbum opened 7 years ago

Libbum commented 7 years ago

I think there may be some overflow somewhere with the new performance changes (from 2b6b3803e to 7455780), as at least twice I've hit random nan issues when starting to generate the 1st excited state.

For now just keep this in mind and keep an eye on it.

Libbum commented 7 years ago

This hasn't come up recently, so I'm closing for now. Will re-open if it occurs again.

Libbum commented 6 years ago

I think I've found a case where this can be reproduced. Boolean test grid at 30x30x30. Will investigate.

Libbum commented 6 years ago

Problem is in evolve. There are nans in phi after the fact. Digging deeper.

Libbum commented 6 years ago

noisy float

Libbum commented 6 years ago

On a 10x10x10 BTG there's a hit in normalise_wavefunction from #52. Seems the /= norm fails, meaning the .sqrt() is probably at fault.

Libbum commented 6 years ago

FYI, debug with RUST_BACKTRACE=1 cargo run > >(tee -a stdout.log) 2> >(tee -a stderr.log >&2)

Libbum commented 6 years ago

Yeah. It's evolve. It's the normalisation call from evolve. We need to do #48.