Zulip Chat Archive

Stream: Is there code for X?

Topic: Gelfand-Tornheim theorem


Jz Pan (Feb 14 2026 at 17:02):

How far could we add Gelfand-Tornheim theorem to mathlib? That is, if a field K has an infinite place (i.e. an AbsoluteValue of it to Reals, which is not non-archimedean),
then there exists an embedding K →+* ℂ such that the absolute value is equivalent to
the usual absolute value | | on .

Here is the idea of proof (which I gave to Aristotle AI):

  1. There exists z such that v z ≤ 1 and v (z + 1) > 1. In fact, since v is archimedean,
    there exists x and y such that v (x + y) > max (v(x), v(y)) so we may assume 0 < v(x) ≤ v(y),
    take z = x / y and we are done.

  2. The v(N) is unbounded, where N is the type of natural numbers. Suppose otherwise, i.e.
    there exists C such that v(n) < C for all natural numbers n, then consider
    v(z + 1) ^ n = v((z+1)^n) for sufficiently large n where z is in the step 1.
    Consider binomial expansion of (z+1)^n.
    Then we obtain that v(z + 1) ^ n = v((z+1)^n) < (n + 1) * C which is impossible when n goes infinity.

  3. In particular K is of characteristic zero, hence it is a Q-algebra, the restriction of v on
    Q gives an absolute value on Q. By Ostrowski's theorem (docs#Rat.AbsoluteValue.equiv_real_or_padic in mathlib),
    it can be deduced that the restriction of v is equivalent to the real absolute value.

  4. Taking completion with respect to the absolute value on the ring homomorphism Q -> K
    we obtain R -> \widehat{K}, hence \widehat{K} is a normed field and also a normed R-algebra.
    By docs#NormedAlgebra.Real.nonempty_algEquiv_or in mathlib, we know that \widehat{K} is isomorphic to R or C,
    in particular, K can be embedded into C preserving the absolute value (up to equivalence).
    This completes the proof.

The steps 1 - 3 was already formalized in https://github.com/acmepjz/lean-iwasawa/blob/master/Iwasawalib/NumberTheory/RamificationInertia/AbsoluteValue.lean (but not provided to AI). The last step is still missing, since it's not known by mathlib the obtained map K →+* ℂ preserving absolute value.

FYI This is the formal statement of the theorem I gave to Aristotle AI:

formal statement

After 5 days and countless time of retries, Aristotle AI returns a full formalized proof with 3300 lines (a quick scan reveals that the proof length may be reduced by 90%):

Aristotle AI output

Michael Stoll (Feb 14 2026 at 17:25):

That should not be hard to get from results in my Heights repo.
See in particular the end of Heights/Extension.lean there. You take the completion, show that it is an algebra over the reals (with compatible abs. value) and then apply Ostrowski's (other) Theorem.


Last updated: Feb 28 2026 at 14:05 UTC