I’ve been lurking for a decade, but I have something worth sharing.I’m a physicist (Rice Space Physics) working on my own framework called Axiomatic Physical Homeostasis (APH), which models how Earth’s magnetosphere "relaxes" into stable shapes after solar storms.It occurred to me that Monte Carlo Path Tracing is doing it the hard way treating light like individual particles bouncing around. In plasma physics, we treat the field as a Stressed Fluid. I define a Geometric Stiffness (beta) for the vacuum and let it relax.The Experiment:- Seed: Shoot <1 ray per pixel (extremely noisy/stressed state).- Relax: Run a matrix multiplication on Tensor Cores minimizing the Geometric Stress of the light field.The Result:The light field snaps into the correct global illumination almost instantly. It preserves temporal inertia (no noise when moving the camera) because the field has mass.I’ve encrypted the whitepaper and put the Abstract/Proof on GitHub:Standard real-time ray tracing relies on Monte Carlo integration, where convergence requires millions of samples per pixel. WTS-RT (Wolf-Toffoletto-Schutza Ray Tracing) treats the radiance field as a continuous fluid governed by symplectic geometry. Using the WTS Relaxation Method originally for magnetospheric plasma modeling—the global illumination problem becomes an energy minimization problem. This achieves stable global illumination with sparse inputs (<1 ray per pixel), offloading healing of the light field to Tensor Core operations and decoupling rendering time from scene complexity.GitHub (Abstract + Encrypted Proof):https://github.com/aaronschutza/Researchhttps://github.com/aaronschutza/Research/blob/main/WTS_RTX.7z https://zenodo.org/communities/aph_physics
>>107809667fake and gay
>>107809667sir this is /v/ where we discuss apple and india>>>/sci/
>>107809667The screenshot doesn't look great.Any performance demos?Raytracing sucks (even for just one ray per pixel), because each ray needs the knowledge of the whole scene.This means a lot of random memory access with loads of cache misses.I don't see, how your approach solves this problem.
>>107809667>LLM-generatedGTFO
>>107809667schizo blabbering
>>107809667This is similar to using DLSS for raytracing. Only DLSS likely reconstructs the gaps better, because it's not schizophrenic.
>>107809667btfo schizo, nobody here cares for your bullshit
>>107809667I do care. Cool project, OP.
Tooker, is that you? I thought you were dead!
>>107809667I care OP, this is awesome
>>107809667I already achieved O(1) GI years ago when I made every frame take 1000ms to finish independently of what's being rendered
>>107809667That image looks way too noisy to use for anything. Got any working demos of this?
>>107809667Does it only work with raymarching or can you do a test like thishttps://www.youtube.com/watch?v=rG2aok2SdbU
>>107809667Why do some of these PDFs just randomly change language between English and Russian? Sometimes mid sentence?
>>107816488Yeah noticed that too. I was originally going to call it a Bogdanoff tier troll because it seemed high effort at first but really it's just an LLM having a stroke.
>>107809667>Penrose Tiles>room temperature superconductors>branes mentioned as well
>>107809667Huh, I'm not deep enough into light equation autism to be able to tell if this is schizo nonsense or legit, but seeing as you appear to be a very excitable young researcher, I'll give the perennial advice: do your homework before publishing outside your field. If you want to be taken seriously here, you should have a render of AT LEAST the Cornell light box. You'd probably want to also demonstrate nasty edge cases like caustics if you want to really want to show off. The sort of standard examples of standard problems that you see in papers on rendering techniques. That abstract art you have posted is doing you no favours. Assuming I'm not being trolled.
>>107818381Look at the repo he linked and these posts >>107816488>>107816551>>107816719it's either genuine mental illness induced manic ramblings or a really shitty troll
>>107818399Was too lazy, just checked out the """paper""" on zenodo. A shame. As it stands, we're going to see live AI rendering before we get a silver bullet to the light equation.
>>107809667was this the same guy a few weeks ago that got ai to invent the universe and gravity or something like that
>>107809667>I’ve encrypted the whitepaperFor what purpose
>>107819009So he can say "you heard it here first on Le 4Chan" for internet points if it gets picked up by a reputable journal. Or at least, that'd be the idea if OP were legit.
>>107809667Your paper doesn't explain how global illumination can result from a single ray (not path) sample per pixel (or 4x4 pixel block). The initial sampling is an information bottleneck that no postprocessing can overcome even in theory, very significant even for direct lighting from emissive surfaces. You can't get something out of nothing.
