It's not - given an accurate geometry model and reflective window material, it'd be really easy to use a raytracer to test this:
* Renderers like Maxwell handle caustics nicely, and you could get away with rough photon-mapping (instead of time-consuming but very accurate MLT)
* Most renderers have very accurate physical sky light models which accurately model the intensity and position of the sun in the sky based on time of day and long/lat coords.
It'd be easy to python script a run of renders for each minute in the day (10 seconds for a render) over the next 5 years and then run an analysis of the rendered image for intensity on the ground. Just set the ground material to not be lit by any diffuse GI rays or direct lighting, and it would literally just render out the reflected sun on the ground (or off other buildings if you wanted to do it REALLY accurately).
It'd be time consuming and CPU intensive to do the rendering for all that: 2,630,880 renders for each minutes for 5 years, but you could probably work out an envelope you could skip (definitely the night time) after doing the first year.
7308 hours render time for the full 24-hour period for 5 years. But do-able, and very parallel over multiple machines.
You can do it on the back of an envelope. Radius of curvature sets the focal length and the min and max elevations of the sun are set by your latitude. If there's any elevation between the min and max where the reflection will hit the ground even close to the focal length of the building, you will have a problem.
It's the same optics calculations anyone designing a concentrating solar power system would do, and while you need to be careful if you want accurate answers, it's not complicated to see if it has the slightest chance of working.
You know, indirect lighting surely doesn't melt plastic, so you absolutely don't need Metropolis Light Transport here. And also you don't need to use photon maps. The computations to map the intensity of specular reflections from the building's windows would the equivalent to creating a photon map on a scene, not using the photon map to render a complete scene.
> You know, indirect lighting surely doesn't melt plastic ...
That depends on the light's intensity. If with the term "indirect" you include specular and diffuse reflection, then yes, in the right circumstances, the energy level can be spectacularly high.
But computer simulations need to be done very carefully, because the issue isn't visible light reflection, it's infrared reflection. For the building in question, the windows are glazed with a coating meant specifically to pass visible light to the interior but reflect infrared radiation (both to preserve building heat in the winter and reject heat in the summer).
What that means is that, as bad as the reflected visible glare seems to those passing by, the infrared energy levels are much higher, and easily explain the damage.
What? It's the indirect reflection of the sun, specularly reflected (and focused) off the building.
You need the light source (sun) reflected onto the ground (or whereever it's reflected) from off the building.
You can't use standard path-tracing / ray tracing for that, as that traces rays from the camera outwards, and wouldn't show the caustics (well, you might be lucky and get one or two fireflies, but that wouldn't be the result needed for analysis).
At the very least, you'd need BDPT. Photon mapping would be a much faster (but cruder) solution.
Basically what you're saying is you need classic ray-tracing from light sources, such as one can get with Pov-Ray and similar old-school ray-tracing programs.
Well, I'd define classic as standard "from camera" outwards.
POV Ray does classic (which won't give you what you need as it doesn't handle caustics) or photon mapping which will.
Bi-directional path tracing (tracing rays from both the camera and lights in the scene and joining the respective light paths in the middle), and Metropolis Light Transport (a progressive variation of BDPT) are the most accurate, but are often slow to converge to a noise-free result, mainly due to the fact they do handle caustics (reflections of bright light focal patterns) accurately, which introduces noise (over the standard GI noise).
But photon mapping would give reasonable results in a few seconds.
> Well, I'd define classic as standard "from camera" outwards.
Fair enough. I was thinking of optical ray tracing, not graphic ray tracing. In optical ray tracing the point is to analyze the paths of light beams through lenses and from mirrors, not the POV of a camera necessarily.
I've been able to get pretty useful optical ray tracing out of Pov-Ray:
Yeah, I work in the VFX industry, so I'm a bit biased :)
Cool - actually, that would be a lot less computationally expensive (assuming you just modelled the sun as a disc) tracing it backwards without any GI at all. Photon mapping's pretty similar though, you've just got a final gather process at the end of sending out the light rays. The difference is photon mapping records all the hits to all surfaces. If you limited it to specular ones (at least for the first ray bounce) you'd probably have something very similar.
> What? It's the indirect reflection of the sun, specularly reflected
I am pretty sure the standard use of "indirect lighting" in normal computer graphics jargon means light that has had at least one diffuse reflection on its path. Even Merriam-Webster lists this meaning [1].
You'd just need to run the ray-tracing from light sources algorithm, to map the intensity of the specular reflection of the building. This would be equivalent to just creating the photon map. Then you could just view the photon map directly, to see the intensity of the death ray. There would be no need to render the scene (although you could, if you wanted a photorealistic view).
* Renderers like Maxwell handle caustics nicely, and you could get away with rough photon-mapping (instead of time-consuming but very accurate MLT)
* Most renderers have very accurate physical sky light models which accurately model the intensity and position of the sun in the sky based on time of day and long/lat coords.
It'd be easy to python script a run of renders for each minute in the day (10 seconds for a render) over the next 5 years and then run an analysis of the rendered image for intensity on the ground. Just set the ground material to not be lit by any diffuse GI rays or direct lighting, and it would literally just render out the reflected sun on the ground (or off other buildings if you wanted to do it REALLY accurately).
It'd be time consuming and CPU intensive to do the rendering for all that: 2,630,880 renders for each minutes for 5 years, but you could probably work out an envelope you could skip (definitely the night time) after doing the first year.
7308 hours render time for the full 24-hour period for 5 years. But do-able, and very parallel over multiple machines.