Question-and-Answer Resource for the Building Energy Modeling Community
 | 1 | initial version |
Bear in mind that if your two planes are very large with respect to the distance between them, then there should be no change in the incident flux on the second surface with respect to distance. Your surfaces need to be small enough that the open ends start to dominate the cosine-weighted integral for illuminance. This implies that the distance between surfaces is on the order of or greater than the size of the ground glass surface.
That said, there may also be inaccuracies due to irradiance caching in this case, since -aa 0.15 will tend to use nearby values when it doesn't see much of a gradient. You can decrease to -aa 0.02 for a more reliable result. You should also make sure that you are sending enough samples to get a good integral. I would recommend -ad 700 -as 200 or so.
Finally, if you are not using the latest HEAD release, I suggest using -as 0 because I recently discovered and corrected a bias in the ambient super-sampling method that was introduced in version 5.1.
| | 2 | No.2 Revision |
Bear in mind that if your two planes are very large with respect to the distance between them, then there should be no change in the incident flux on the second surface with respect to distance. Your surfaces need to be small enough that the open ends start to dominate the cosine-weighted integral for illuminance. This implies that the distance between surfaces is on the order of or greater than the size of the ground glass surface.
That said, there may also be inaccuracies due to irradiance caching in this case, since -aa 0.15 will tend to use nearby values when it doesn't see much of a gradient. You can decrease to -aa 0.02 for a more reliable result. You should also make sure that you are sending enough samples to get a good integral. I would recommend -ad 700 -as 200 or so.
Finally, if you are not using the latest HEAD release, I suggest using -as 0 because I recently discovered and corrected a bias in the ambient super-sampling method that was introduced in version 5.1.
++++ New info: Thanks for posting your files. I used "rtrace -I+ -ab 1 -ad 4096 -aa 0" in my tests. There seem to be two issues with your experiment. The first is that there is nothing keeping your test points from seeing the sun directly, unless it is blocked by the diffuser. Further back points see the sun, so they have a larger value than points nearer to (and thus occluded by) the diffuser. You need a box around your space. All you have is a black wall that sits entirely behind your test points, and is therefore irrelevant.
I strongly recommend visualizing your scene in rvu as a sanity check before doing numerical analyses. This avoids most such issues. It may also help to visualize where your points are in space by replacing them with tiny spheres, as a further check.
| | 3 | No.3 Revision |
Bear in mind that if your two planes are very large with respect to the distance between them, then there should be no change in the incident flux on the second surface with respect to distance. Your surfaces need to be small enough that the open ends start to dominate the cosine-weighted integral for illuminance. This implies that the distance between surfaces is on the order of or greater than the size of the ground glass surface.
That said, there may also be inaccuracies due to irradiance caching in this case, since -aa 0.15 will tend to use nearby values when it doesn't see much of a gradient. You can decrease to -aa 0.02 for a more reliable result. You should also make sure that you are sending enough samples to get a good integral. I would recommend -ad 700 -as 200 or so.
Finally, if you are not using the latest HEAD release, I suggest using -as 0 because I recently discovered and corrected a bias in the ambient super-sampling method that was introduced in version 5.1.
++++ New info: Thanks for posting your files. I used "rtrace -I+ -ab 1 -ad 4096 -aa 0" in my tests. There seem to be two issues with your experiment. The first is that there is nothing keeping your test points from seeing the sun directly, unless it is blocked by the diffuser. Further back points see the sun, so they have a larger value than points nearer to (and thus occluded by) the diffuser. You need a box around your space. All you have is a black wall that sits entirely behind your test points, and is therefore irrelevant.
I strongly recommend visualizing your scene in rvu as a sanity check before doing numerical analyses. This avoids most such issues. It may also help to visualize where your points are in space by replacing them with tiny spheres, as a further check.
To save you from reading through all the comments, the main error was leaving off the -I+ option, which is needed to compute irradiance rather than radiance values. The initial simulation was basically sending parallel rays straight at the window, which of course doesn't change with distance (besides getting completely the wrong answer).
