D. Kirk and Prof. Slusallek discuss real-time raytracing

While on the subject of "realistic" rendering, a few years back there was a paper published on "wave tracing" where light wavefronts including phase etc were propogated through the scene. This would model reflections/shadows/translucency as well as diffraction etc.

The only niggling, teensy little problem was that run time was proportional to (the cube of?) the relative size of the scene and the wavelength of the light. In order to get a result in a reasonable amount of time, the wavelength was set to be, say, about 1/10th the dimension of the objects and so the image was dominated by diffraction patterns

 

Umm thats nice but what about the classicly light problems? Also how are you going to do composite materials?

 

Wow. I'm sure that would look pretty interesting! And heck, we'd get a good idea of what the world would look if we could see in the microwave spectrum

 

If you're interested, I think it was called "3D Graphics & the Wave Theory" and was published at SIGGRAPH 1981

 

what do you mean with classicly light problems? qed covers all light phenomena existing, and for composite materials, well you just have to calculate your impulse response or apply some different weighting inside the digital system representing your 'point' or whatever you call it.

if you have some constants for every material that is somehow interacted with, you shouldn´t have any problems.

 

Well, I don't know if you caught my joke before, diehaerte, but I have to say, the application of Quantum Electro-Dynamics to lighting is pretty ridiculous. "Classical" quantum mechanics is typically enough to describe the behavior of most solids (I've never heard of solid state physicists using even the Dirac equation to attempt to solve for properties of materials, let alone QED).

Suffice it to say, if you were going to really attempt to use QED to describe the behavior of light, you'd have to model things on vastly too small a scale for the computation to be even close to real-time. We're talking millions of years of computation per frame at current speeds, at least.

 

Depends on how you do it. You could use real atoms to model the atoms, arranged correctly (so say if you were modelling a chair, you'd arrange the atoms into a chair shape). You then might try using photons to calculate how the light would interact with those atoms. So for example you'd inject photons into the simulation environment at the location of the light-sources, then allow the photons to interact with the aforementioned atoms.

The beauty of this approach it this was is that it can be performed in a massively parallel manner, ie. you can use more than one photon at once. It's also quite easy for the artist to arrange the scene (eg. by simply moving the simulated "chair").

The challenge is how to record the outcome of the simulation. One idea would be some sort of sensor, maybe made from silicon or similar, divided into small squares to give you spatial resolution. You might put this sensor behind a small lenticular glass element in order to achieve good focus. Then simply recording the electric charge within each small square of silicon should allow you to form an image. Bingo!

Hmmm. Off to the Patent Office I go...

 

You'd never go so far as to use Quantum Mechanics for a realtime simulation. The calculations are just too intensive. Quantum Electro-Dynamics is a step further than using basic Quantum Mechanics. Each single photon interaction, when using QED, would consist of calculating a series of integrals, and the outcome would be some probability density.

Let me put it this way.

You're not going to calculate the properties of materials on the fly in a game. You're going to have those baked into the shaders, which determine how light interacts with surfaces. You're not really interested in why light interacts the way it does, just how it interacts. It wouldn't even be useful for game developers to attempt the calculations on how light would interact with real materials offline, as the calculations are far more intensive than game developers should bother to get into.

But the above is just Classical Quantum Mechanics, which has not been applied to materials of more atoms than on the order of a few hundred in any robust way. The calculations are just that complex. Imagine, for a moment, attempting to apply those simple calculations to a game world, and you may understand that it's just utterly unfeasible to use quantum-mechanical calculations to describe macroscopic phenomena. We just don't need to be that exact.

That said, Quantum Electrodynamics is a full step lower in level than the simple classical Quantum Mechanics calculations. We're talking about using as many calculations that Quantum Mechanics requires for describing an entire solid to describe just a single atom. It ain't going to happen. Anybody who says otherwise is not coming close to using real Quantum Mechanics in a realtime situation.

One may purport to use mathematics and techniques learned in Quantum Mechanics and Quantum Electrodynamics, but that's where it ends. No modelling of complex, real-world scenarios using real Quantum descriptions is going to happen.

 

Well until graphics chips come with a quantum computing unit and a improbability drive

 

/me wonders if I should have been a little bit more bloomin' obvious :?

 

I've got a really hot cup of tea ready.

 

that´s exactly what i´m saying. you calculate the impulse response for every material you are using well and then you basically have a table where you look them up. well all that talking is of no use, i hope me and my team will finish our implementation in some reasonable time.

and i guess you wouldn´t use quantum mechanics for solid state physics, spring-mass systems would be sufficient.

 

If your lookup table is a function defined as some generic f(x1,...,xn), how big is n?

LOL

 

Umm okay over how many wavelengths of light over how many angles over what temperatures what geometery of each molecule as even for a single compound you can have several stable/semi-stable confirmations. Okay next most interesting surfaces don't contain just one compound so how are you going to do something like wood?

How about a cloud of microscope particle in brownian motion how are you going to do that?

What about electronic exciting and emission for things such as fluorescence and pospherence?

Are you going to do the calculations of a single point or are you going to do molecule dynamics and then look at the statically distrubtion or what?

Do you even know what I'm talking about?

Just look at Folding at home the task is so big to try to work out just how the proteins roll up on themself is let they have to get millions of computers around the world to help. You think any developer is going to have the computer resources to do ab initio calculations ( fodling at home would be using force fields or semi-empircal cause it would just be to hard otherwise )? It takes me over 30 minutes to do the vibrational calculations for a 12 atom ( most of them were hydrogen too ) molecule and the complexity is around order N^4 so every time you double the number of atoms thats 16 times harder.

 

Spring-mass systems give heating properties and sound properties (to a rough approximation: works for some materials, not others), not visual properties. Visual properties, if you're going to calculate them, need quantum-mechanical calculations.

It's much easier to just measure the visual properties, considering most substances we use in everday life are very complex, and it would be nigh impossible to compute their properties.

 

Just as a side comment, most of these types of algorithms can be simplified to diagonalizing a matrix, which goes as N^3. I don't think you should have to deal with an N^4 algorithm if you're clever.

 

Yeah okay I'm wrong its order N^3 so thats every time you double the number of electrons ( not atoms that was my bad before ) you make it 8 times bigger.

 

Yep, which still makes it undoable for reasonable-sized systems. For example, some serious research is going on in DNA that is limited to no more than two base pairs in the calculation. That's just two steps in a ladder that's billions long in any real organism. While you may be able to do something with such a simulation, you're definitely not going to be able to predict all behaviors.

This is why many mathematicians and physicists are working diligently on searching for better approximations that reduce these calculations to order N^2 or, better yet, order N.

 

If you go from straight HF to MP2 then its a nN^4 where N=Number of basis functions, n=number of occupied orbitals if you go to MP5( which most software packages don't do ) then you get n^3xN^5 .

I think Chalnoth has pretty much said its useless after going past a few simple dna base pairs. Yet diehaerte is expecting to use it on all sorts of surfaces.