# Some trans oddities

I am struggling to understand some seemingly inconsistent results I'm getting with the trans material. We are doing some fairly complex studies of the daylight penetration in a bunch of art galleries. These galleries receive daylight that is filtered through some diffuse glass, and further modulated with sun control shades. It is the material properties of these shades that is giving me fits at the moment. In brief, certain shade transmissions give wacky results, while most perform as expected. To reduce the possibility of error, I have constructed a simple box room (16'x16'x9'), with a single 3' square aperture in the ceiling. this aperture has two panes of diffuse glass (as the real building will), between which I have a polygon that I assign various shade materials (using trans) to. I am then computing the horizontal illuminance on the floor with rtrace for each shade.

Results:

Test1 (no shade): EhFloor - 222 Lux
Test2 (10% shade): EhFloor - 23 Lux (10.41% of Test1, 4% off)
Test4 (5% shade): EhFloor - 12 Lux (5.4% of Test1, 8% off)

So far, so good!

(in all cases, the exterior horizontal illuminance is being predicted to within 10 Lux, so it's not even worth weighting the interior predictions for each individual calculation, IMHO.)

Now, we have two shades in the real building, both of which can be deployed at a time. So next I tried simulating a 10% and a 1% shade down, but I "cheated", by assigning the net transmission of the two shades to the single polygon (IOW, .10 * .01 = .001).

Test3 (10% + 1% shade as a single polygon): EhFloor = 0.4434 Lux (0.20% of test1, or 100% off!)

Thinking that maybe the aforementioned cheat was ill advised, I added a second polygon to the model, so I had two physical objects to apply shade materials to:

Test8 (10% + 1% shades, as two separate polygons): EhFloor - 0.5509 Lux (0.25% of Test1, or 148% off)

So much for that theory. thinking maybe there's a limit to how much you can remove from the daylight before the calculation goes awry (hey, I'm grasping at straws at this point), I tried another scenario:

Test7 (10% + 5% shades, as two separate polygons): EhFloor - 0.9656 Lux (0.43% of Test1, 13% off)

This is certainly close enough for Government work. But then:

Test9 (single 01% shade): EhFloor - 1.189 Lux (0.54% of test1, 46% off)

So, the 10% + 5% was accurate, which is cutting out more light together than a single 1% shade, but the 1% shade calc was fairly inaccurate. I thought maybe it was the inherent variability in the indirect calculation, so I ran test9 again and got exactly the same numbers! (I'm using rad with Q=M D=M V=H and -ab 3 for these tests, and it's a clear sky at noon at a latitude that pretty much places the sun directly over this aperture.)

The reason this entire test came about is because I can't seem to affect any further light reduction in my model past a certain point. I even tried making the shades totally opaque, and when I do that my model goes completely dark, as I expected, so the model seems to be "tight", and have no light leaks. Everything points to my shade material definitions but I can't find anything wrong with them, except for the fact that certain ones seem far less acurate than others. I have been trying to use Radiance as a relative performance evaluation tool on this project, but for this last part of the project where we are trying to knock out a very high percentage of the available light, I seem to be getting counter-intuitive results. Does anyone have any insight as to why these results are occurring?

Here are the trans parameters for the various shade materials (all are assumed to have a 95% transmitted specularity, as they are view preserving shades, even tho it probably does not matter too much because they are between two pieces of diffuse glass.):

0
7 .4 .4 .4 0 0 .25 .95

0
7 .25 .25 .25 0 0 .2 .95

0
7 .21 .21 .21 0 0 .047619 .95

## 10%+01% shades in single polygon
0
7 .201 .201 .201 0 0 .004975 .95

## 10%+5% shades as single polygon
0
7 .205 .205 .205 0 0 .02439 .95

Any help appreciated.

···

----

Rob Guglielmetti

Hi Rob,

probably this might have an effect: rad uses some default value for the -av
parameter, which specifies the constant ambient background, accounting for
the fact that only a finite number of ambient bounces are followed
(theoretically, in an absolute sense, an infinite number would be necessary)

For very low values of illumination the error of the -av approximation may of
course become obvious, the default -av certainly is some value correct for an
average scene with average conditions, this all is, as you might imagine, a
bit heuristic anyway. (In the RWR book, there's a short section about how to
determin the correct -av for a specific scene)

But for your extraordinary conditions try again with setting -av 0 0 0.(BTW: I
I do that often, the pictures look better...) You'll introduce some error
then, too, of course, but in your example with only 1% transmittance and sun
direcly overhead most light will stem from direct contribution, so completely
neglecting ambient light from the third bounce level onwards should produce
no significant error anymore.

Maybe it helps. Maybe not...

-Carsten

···

On Friday 09 May 2003 16:28, Rob Guglielmetti wrote:

(I'm using rad with Q=M D=M V=H and -ab 3 for these tests, and it's a

(I'm using rad with Q=M D=M V=H and -ab 3 for these tests, and it's a

Hi Rob,

probably this might have an effect: rad uses some default value for the -av
parameter, which specifies the constant ambient background, accounting for
the fact that only a finite number of ambient bounces are followed
(theoretically, in an absolute sense, an infinite number would be necessary)

For very low values of illumination the error of the -av approximation may of
course become obvious, the default -av certainly is some value correct for an
average scene with average conditions, this all is, as you might imagine, a
bit heuristic anyway. (In the RWR book, there's a short section about how to
determin the correct -av for a specific scene)

Hmmm. I have not been specifying an -av. Before you all load your guns for the firing squad, her me out. What I am doing is a series of full-day sun study calculations, for several says of the year under various building configurations. It amounts to hundreds of individual rtrace calculations. Determining an -av for each one would be time prohibitive. Besides, since I'm using a lot of ambient bounces, I thought I could get away with this. Early tests bore this out; I compared some early tests where I had not set an -av, to the same model run with an -av calculated using compamb, and they were very close. I assumed that by specifying four ambient bounces and waiting, that I was still getting valid results. Sort of a brute-force method.

Perhaps that works -- to a point? Maybe I'm getting to such low light levels that four is not enough. Maybe it's time to finally implement the trick Greg thought of for me in this case, where you use compamb to find a valid -av for mid afternoon, and then use the ground ambient value from gensky to weight that -av for the other times of the day. That way I can still put the whole thing into a script.

Does anyone have any guidelines for parameterization as it relates to daylight filtration? IOW, if I'm trying to measure what will likely be .001% of the available exterior daylight, is there some minimum number of ambient bounces I need to use, or am I trying to find too small a number now? Like I said, everything has tracked pretty logically so far, it's just these certain galleries where we are trying to get down to very low light levels that I'm getting these strange results...

But for your extraordinary conditions try again with setting -av 0 0 0.(BTW: I
I do that often, the pictures look better...) You'll introduce some error
then, too, of course, but in your example with only 1% transmittance and sun
direcly overhead most light will stem from direct contribution, so completely
neglecting ambient light from the third bounce level onwards should produce
no significant error anymore.

Well, right now all I'm doing is rtraces, I care more about the numbers than the look of the pictures, but that's a useful tip for the future. True in your example the -av 0 0 0 will probably not make a difference because I'm calculating direct contribution primarily, but in the real model I am computing values that are all about the indirect.

Rob Guglielmetti
[email protected]
www.rumblestrip.org

···

On Friday, May 9, 2003, at 05:31 PM, Carsten Bauer wrote:

On Friday 09 May 2003 16:28, Rob Guglielmetti wrote:

Hi Rob,

just for the records, :-), I did not accuse you of severe -av abuse by intent,
I just wanted to hint at the fact that the rad utility sets -av
automatically, unless you deliberately interfere ...

-cb

Actually, I don't think the rad util sets it at all. If you don't set it, it doesn't get used. that's why my ambient areas are always splotchy in my renderings. Dunno. I may be wrong. All I know is I have never set the av explicitly yet. Soon, I guess.

Hey, Brenda & I are going to London & Bath next month! I can't believe it took me 32 years to get to Europe, and now (after next month) I'll have been there three times in three years! I guess we never know what our future schedule holds...

I hope all is well. Thanks for sharing your Radiance insights with me. Are you using it at all nowadays?

Rob Guglielmetti
[email protected]
www.rumblestrip.org

···

On Sunday, May 11, 2003, at 05:46 AM, Carsten Bauer wrote:

just for the records, :-), I did not accuse you of severe -av abuse by intent,
I just wanted to hint at the fact that the rad utility sets -av
automatically, unless you deliberately interfere ...

Hi,

rad doesn't perform any calc to get an adequate av, it just sets a default
value ( 0.01 for ZONE=INTERIOR and 10 for ZONE=EXTERIOR)

-cb

Hi Rob,

Carsten is right about the ambient (-av) settings, but I don't think that's your problem. My guess is that you're running up against the ray weight limit Radiance imploys to trim its calculation, which is controlled by the -lw parameter. You should reset this to -lw 0 in your tests, or something very small such as -lw 1e-8 if -lw 0 takes too long.

Try it.
-Greg

From: Rob Guglielmetti <[email protected]>
Date: Fri May 9, 2003 9:28:25 AM US/Pacific
To: [email protected]

I am struggling to understand some seemingly inconsistent results I'm getting with the trans material. We are doing some fairly complex studies of the daylight penetration in a bunch of art galleries. These galleries receive daylight that is filtered through some diffuse glass, and further modulated with sun control shades. It is the material properties of these shades that is giving me fits at the moment. In brief, certain shade transmissions give wacky results, while most perform as expected. To reduce the possibility of error, I have constructed a simple box room (16'x16'x9'), with a single 3' square aperture in the ceiling. this aperture has two panes of diffuse glass (as the real building will), between which I have a polygon that I assign various shade materials (using trans) to. I am then computing the horizontal illuminance on the floor with rtrace for each shade.

....

Greg Ward wrote:

Hi Rob,

Carsten is right about the ambient (-av) settings, but I don't think that's your problem. My guess is that you're running up against the ray weight limit Radiance imploys to trim its calculation, which is controlled by the -lw parameter. You should reset this to -lw 0 in your tests, or something very small such as -lw 1e-8 if -lw 0 takes too long.

Try it.

Will do, thanks. Welcome back!

···

----

Rob Guglielmetti