I use Radiance for calculating daylight factors and illuminance levels in buildings.
I created a square sized building with one room (zone) at the middle of each façade and run a daylight simulation. As expected I got the about the same illuminance level in each of the rooms, but when I changed the lw parameter from 0.011 to 0.010 (or more precise from 0.0105 to 0.0104) I got a more than 10% sharp decrease of the illuminance in one of the rooms. In the other three rooms I got no significant change (less than 1%), which is what I expected for all the rooms. The results are consistent and not due to random incidences.
Is this due to a bug in Radiance?
I am not authorized to upload the files to show this case. Is there any other way I can do this bug reporting?
You should provide the output of “rtrace -version”, assuming that’s the tool you’re using. Also, the full set of options you are using. I’m curious what happens if you reduce the -lw option to something more typical, like 0.001?
It seems as though the number of sensors and the location of these influences the devation of the results.
If I decrease lw to 0.005 I still get about 10% lower results, but when I set lw=0.004, the difference seems to be gone. For lw = 0.001 the deviating room now gives slightly higher illuminance than the other rooms, but this difference is less significant.
Thank you for the additional information. In general, you should not expect a uniform progression towards the correct answer as calculation parameters are adjusted. The errors may oscillate to one side and another depending on the scene and parameters chosen. In other words, I would not classify this as a bug.
How did you decide on this set of parameters? In particular, I would not use -dt 0.05, but would set -dt 0.01 or smaller, and stick with a negative number of -lr to enable Russian-roulette sampling (as is the default in rtrace). Furthermore, specifying -u- makes the results repeatable, but may introduce some bias to the calculation.
Thanks for your answer. I just thought it was a bit strange that these deviating results appear only in one of the zones, although the building and the scene are totally symmetrical.
When I took away the -u- specification the deviation disappeared, so it seems to be connected to this repeatability function.
When I set -lr to a negative number the deviation also disappeared. I guess this can be expected, since it also changes the function of -lw.
Does a negative -lr value in general give more accurate illuminance values than a positive?
When -lr is negative and -lw is low the simulation time is about equal to having a positive number, but it becomes increasingly larger with higher values of -lw (e.g. -lw>=0.005). Why is that?
Our choice of parameters is a mixture of what we have read in the literature, recommendations from other users and our own experience. Most information that we found about settings of the rtrace parameters concern rendering, but I could imagine that there could be different criteria when trying to get accurate illumination values. Do you have any tip where to find useful information about this?
Once again, thanks a lot for your answers and tips. They are very useful to us.
The -lr value negative gives more accurate results in general. I would not bother with higher -lw settings, especially if it takes longer, as the result will be less reliable. I don’t understand why it takes longer in your case, to be honest.
The only reliable way to arrive at good settings is by trial and error, or through experience and understanding how each parameter affects the calculation. For this, Rendering with Radiance is still the best resource, although I can also suggest going through the tutorials from the second Radiance workshop.
Thanks a lot for your answers. I have another related question. Recently, when a scene with 12 zones was simulated, the illuminance levels for 5 of the rooms became very low. When it was resimulated these 5 zones got about twice as high average illuminance level as the first time. The other 7 zones got about the same result as before. The parameter setting was similar to the one given above. Do you have any idea how this could happen? Is there any chance that a simulation can be “truncated” before it is finished, but still produces results?
I really don’t know in your case. There are rare conditions where ambient light can “leak” into adjacent spaces if the -ar parameter is set too low, but this is speculation. More likely, the initial -ad setting was too low, as this is the most common cause of errors, but I would need to see your scene.