Behavior when changing the -aa parameter

Hello.

I am studying the difference in computational load when varying the -aa parameter. However, I have two questions in my study.
In my study, -aa is varied between 0 and 1, and the other parameters are fixed at

-dp 256 -ds 0.25 -dt 0.25 -dc 0.5 -dr 1 -st 0.5 -lr 6 -lw 0.01 -ab 3 -ad 1000 -as 2048 -ar 300

(1) When -aa=0
My understanding was that when -aa=0, no completion (irradiance caching) is performed, which would be expected to increase the computational load compared to when completion is performed.
However, in my examination, the computation is completed earlier than any of -aa=0.1~1. The illuminance at the monitor point is also calculated and does not seem to be ending in error. Why is this?

In the case of (2) - aa=10^-5
I understand that when aa ≤ 10^-6, it is considered aa≃0, and we have considered a value very close to that. However, it crashes with the error statement [ rtrace: consistency - address not found in avlmemi ]. What should I do about this?

These are my questions.
I would appreciate it if you could tell me if it is the same.

(1) Setting -aa 0 turns off the indirect cache as you say, which changes major parts of the algorithm. For one thing, it changes the process for ray termination such that a -lw setting of 0.01 means less than 100 rays will be sent, and unless Russian roulette sampling is specified (-lr <= 0), those rays will spawn no further bounces even if -ab > 1. So, that’s why that calculation finishes in a reasonable time.

(2) As with most parameters in Radiance, there are ranges where parameters don’t work very well, and assigning -aa to something non-zero less than about 0.01 leads to an insane density of calculations with no real benefit. However, you should not have gotten the error you did, so I will look into that.

The reason we don’t restrict parameter values to a nominal range is that there are always exceptions, and we don’t wish to hamper anyone who needs to explore a calculation.

Cheers,
-Greg