Hi Andrew, I have to admit that I have never seen such a curve before. I could only guess since I don’t have your file. I suspect that you used freeform or even asphere with large zonal variation. The incident angle on image plane might be smaller in the higher RI region/the effective working number is smaller to collect more light. Another possibility is the distortion. If the pincushion distortion is high, it helps concentrate the energy and therefore improve the RI. Though, it’s rare to have such a huge variation over the fields.

I can have a better investigation if you can share me your file. If it’s confidential, you can send me the file by emailing support@zemax.com.

I would like to recommend you to read this material written by our excellent engineer Michael Cheng. He explained how RI is calculated in details. 

FAQ about Relative Illumination | Zemax Community