07-12, 16:24–16:46 (Europe/Amsterdam), Function (4.1)
We present SeismicWaves.jl, a package for 1D, 2D, and 3D wave simulations on multi-xPUs. SeismicWaves.jl supports forward and adjoint simulations to compute gradients with respect to model parameters for Full-Waveform Inversion (FWI) methods used in seismic tomography.
The package is based on finite-difference methods for solving the wave equation in parallel and distributed systems using ParallelStencil.jl and ImplicitGlobalGrid.jl, enabling seamless multi-xPU computing.
Rigorous benchmarks and synthetic inversions validate the solvers' correctness while achieving high memory throughput (up to 90% of peak) on modern GPUs and exhibiting good weak scaling on distributed systems.
In conclusion, SeismicWaves.jl provides easy-to-use and efficient solvers to bridge the gap between theory and practice for a variety of FWI applications and use cases, from fast prototyping on a laptop to high-resolution, large-scale simulations on supercomputing clusters.
I am a PhD student in Seismology at ETH Zurich, my main interests are in numerical methods for solving PDEs (especially for wave simulations) and Full-Waveform Inversion.