On the transfer of Shalika germs for G_2, revisited

In the summer of 1994 I wrote an unpublished manuscript, “On the transfer of Shalika germs for G\sb{2},” which I hoped to polish into a publication. It used Langlands’ geometric method, based on Igusa’s asymptotic expansions, to compare the \kappa-Shalika germs of the split exceptional group G\sb{2} over a p-adic field with germs on its two cuspidal endoscopic groups, PGL(3) and (SL(2)\times SL(2))/{\pm 1}. I think I showed it to Tom Hales, Magdy Assem, maybe Diana Shelstad, but that’s about it. Unfortunately, I was unable to finish the technical details (which I didn’t understand nearly as well as Tom or Madgy or Diana), dropped it, and moved onto an easier project! Another paper to add to my mathematical orphanage.

This month (Nov 2026) I revised it with two AI systems, Claude (Anthropic) and Gemini (Google), which are listed as co-authors. Their roles are described in the paper. The revision found a real gap in the 1994 argument. Its final step used a reduction lemma of Hales. That lemma needs an endoscopic group with a nontrivial connected center, and both cuspidal endoscopic groups of G\sb{2} are semisimple of full rank. So the lemma doesn’t apply.

Here is what the corrected and revised paper does:

  • It proves that the variety underlying Langlands’ partial resolution is a normal, Cohen-Macaulay toric variety.
  • It computes the Igusa constants of every branch of the special fiber.
  • For the 2-regular class and the biquadratic tori, it replaces Hales’ lemma with a direct proof: a birational map from the fiber to (\mathbb{P}^1)^4 shows that the germ doesn’t depend on the tangent direction, with an explicit matching constant.
  • For the 3-regular class and the cyclic cubic tori, the corresponding integral is reduced to an integral over a split quadric cone. Its expected value is stated as a conjecture.

Two statements about the Igusa expansion remain unproved, so these results are conditional. However, the matching of germs itself (claimed in the original paper) is now a theorem by other means, from Waldspurger, Ngô, and Gordon-Hales. What the Igusa approach in this corrected and revised version adds is explicit geometry. Finishing up the two Igusa expansion statements is left open. Probably more AI-aided exploration will yield them, but I’m moving on. I’m not sure adding a second proof of a result already known to be true is worth what I guessed was a significant amount of work.

The 1994 sources, the revision, referee reports, and verification scripts are on github (pdf and latex).

Leave a Reply