Dear @kmk961120
I spend quite a lot of time with my colleagues to solve the problem. We had some progress but I am not still convinced that the model works properly. I will start working on this problem when I am back.
I was focused on 2D only, but its interesting that your results for different polarization match in 3D. Did you do any simulations in 3D with different mesh sizes?
Here is the summary that I did:
a) I realized that medium monitor that stores the population is located at the center of the gain material. However, since the pump is strong and material is largely absorbent (a big portion of incoming light intensity gets absorbed on the first cell), cells behavior along material will not be identical. As a result while you are seeing a normal behavior in the central cell, the cells that receive light earlier behave non-normal and population inversion plots don’t make sense. You can check this by adding monitors at different locations and plotting population (Please see my simulation file at the end which includes three different location in one cell). One plausible answer that this model fails might be because this model is good only for materials thin enough to be considered 2D (only one or two mesh cells). I still need to do more work on this to convince myself!
b) I reduced the pump intensity. In the original file it was so strong that inversion plot had a sharp rise and some of the pump light was red shifted in the probe frequency. This generated light was strong and overlapping with delayed probe pulse and transmission results were not very reliable (as it includes both pump and probe light).
pump_prob_KX.fsp (254.7 KB)[quote=“kmk961120, post:3, topic:3932”]
Also, I got different transmission spectrum depending on the mesh size. (The overall structure is same but scale is different)
[/quote]
I agree and I noticed this before. Again, it looks like this model is not capable of capturing bulk material properties (where we expect identical behavior along the material).
Even with the above mentioned modifications, my results still vary with mesh size 
Please keep me updated if you got any new results and I hope we can resolve the issue.
Bests