preconDDA

4 views

Skip to first unread message

Maxim Yurkin

unread,

Dec 11, 2025, 6:16:41 AM (4 days ago) Dec 11

to ADDA questions and answers

Dear colleagues,

Since this discussion group is not only about the ADDA code, let me bring your attention to the recently published paper:
Lanier S. Learning to precondition: Reinforcement learning enhanced three-level circulant preconditioning for the Discrete Dipole Approximation, J. Quant. Spectrosc. Radiat. Transfer 350, 109741 (2026)
and the corresponding code preconDDA - https://github.com/Toastlovesjam/preconDDA .

It implements circulant preconditioning in DDA, which largely alleviates convergence problems of the iterative solver for large highly-contrast dielectric particles. The code works on Nvidia GPU, so it can be very fast whenever the simulation fits inside the GPU memory.

I recommend to give it a try, especially, if you hit a limit with ADDA (e.g. trying to simulate particles, say, with size parameter 30 and refractive index 2).

The corresponding ideas can also be implemented in ADDA (see https://github.com/adda-team/adda/issues/244), but it is not trivial, so we do not yet have a specific plan for that.