History of Wavesim

The Modified Born Series was developed at the University of Twente by Prof. Ivo Vellekoop, Gerwin Osnabrugge and Swapnil Mache, with help from Saroch Leedumrongwatthanakun and Maaike Benedictus. It was initially developed to simulate the anisotropic memory effect occurring in complex scattering structures such as biological tissue.

​

In benchmark experiments, the method was found to outperform pseudospectral time-domain (PSTD) by multiple orders of magnitude in both speed and accuracy. The method was initially developed for solving the Helmholtz equation and was later extended to Maxwell’s equations and Maxwell’s equations in birefringent materials and anisotropic metamaterials. 

For a long time, the bottleneck remained in the presence of wrapping artefacts. We solved this problem, however, by developing an alternating fast convolution strategy, where the artefacts are cancelled without significant penalty to the computation time or memory use.  

​

In 2023, a collaboration between the University of Twente, Rayfos Ltd and the Forth research institute was started to further develop Wavesim and help users switch to this new tool. As part of this effort, Rayfos implemented a CUDA-optimized version that outperforms the ‘reference’ implementation using the MATLAB parallel processing toolkit by a factor of 2. This accelerator is included for free with the version of Wavesim available on GitHub. 

​

​A Python version of Wavesim v0.1.0 - alpha was released in October 2024 for solving the Helmholtz equation in arbitrarily large media through domain decomposition. With this new framework, a complex 3D structure of 315 × 315 × 315 wavelengths (3.1⋅10^7) in size was simulated in just 379 seconds by solving over two GPUs. This was a factor of 1.93 increase over the largest possible simulation on a single GPU without domain decomposition. The open-source implementation is available on GitHub.

​​

This Forum is a place where researchers, engineers and scientists who are interested in the Wavesim simulation algorithm can discuss their needs and get community support for their applications.