[arbitrandomuser]

Don't the different phases from all the random path more or less cancel out , and significant additions of phases only come from paths near the "classical" path? I wonder if this reduction would still be tractable on gpus to simulate diffraction

[dahart]

That’s what I remember from QED, the integrals all collapse to something that looks like a small finite-width Dirac impulse around the mirror direction. So the derivation is interesting and would be hard to simulate, but we can approximate the outcome computationally with extremely simple shortcuts. (Of course, with a long list of simplifying assumptions… some materials, some of the known effects, visible light, reasonable image resolutions, usually 3-channel colors, etc. etc.)

I just googled and there’s a ton of different papers on doing diffraction in the last 20 years, more that I expected. I watched a talk on this particular one last year: https://ssteinberg.xyz/2023/03/27/rtplt/