The molecule has ~30 atoms, so I probably can use QZ. The Multiwfn manuals says "It is worth to note that Mulliken analysis is incompatible with diffuse functions, if diffuse functions are employed, the analysis result will be meaningless." but I don't know any particular reference from journal. — roma ichenko 1 hour ago

For what reason are you using def2 rather than cc-pVXZ? — Nike Dattani 1 hour ago

Probably for no reason, I have a weird confidence that def2 is more appropriate for DFT. Most likely there is no big difference. Anyway the question is about diffuse functions. — roma ichenko 33 mins ago

In the work I do, I always use diffuse functions (the more functions, and types of functions, the more propensity to get an accurate description of the wavefunction). I don't deal with Hirschfield or Mulliken charges in my work, so I won't comment on how they're affected by diffuse functions, but I do remember we had this question where the OP said the diffuse functions posed a problem for something. I think your "strange results" might be less strange if you used the cc-pVDZ, TZ, QZ series and then did an extrapolation to the CBS limit. — Nike Dattani 18 mins ago

I'll probably try to look on basis set dependence and the influence of diffuse functions, but our HPC is down and I'm looking for some theoretical insights. — roma ichenko 3 mins ago