Default values of FACPARMAX and RELPARMAX ?

[BJEANNOT0]

Hello,

I write because the manual, in section 4.6.7, says that values of 10.0 are often suitable for both RELPARMAX and FACPARMAX. However, I just noticed that their default values are respectively 1 and 1.1 in pestpp-glm

I think this may hinder the inversion process.
I know it because I just tried pestpp-glm on an inversion problem that I usually tackle with pestpp-ies (with ies_enforce_chglim =false). I observed that by default in pestpp-glm, relparmax=1 and facparmax=1.1, and that the objective function decreases very slowly compared to pestpp-ies. Besides, I keep getting this kind of warning messages :

"WARNING : change limit/ bound enforcement for lambda 1000resulted in less than 1% original upgrade vector length
change limit/bound enforcement for lambda 1000: upper change limit, PER_F00226, scaling factor : 5.63863e-06"

(I emphasize that PER_F00226 is the name of one of my parameters. Please also notice that a space is missing between "1000" and "resulted").

I will now proceed to make this pestpp-glm run again, but with relparmax=facparmax=10. I will keep you updated.

Best regards

[sjepsen395]

Hi,

I realize I'm late in this conversation. Today I encountered the same message and had to scratch my head about what exactly the change limits are on a log-transformed parameter whose initial value = 1.0. Log(initial value) = zero and if my math is correct anything times zero is zero. So I think the FACPARMAX and RELPARMAX must apply to the untransformed initial value which is 1.0 in my case. Maybe the default value of "1" that you saw is referring to the log of FACPARMAX? I don't recall seeing a mention on transform status of parameters in the mathematical definitions of FACPARMAX and RELPARMAX in the PEST User Manual Part I. I probably need to do a simple run using something like FACPARMAX = RELPARMAX = 1.175 and then study the .rec file so to make sure those pertain to untransformed parameters.

Steve