What is known, and what is published, on the reverse mathematics of the nest of results called Hilbert's Theorem 90?

  • 2
    $\begingroup$ What is "reverse mathematics"? $\endgroup$ – Igor Rivin Jul 17 '13 at 23:48
  • 3
    $\begingroup$ @IgorRivin, reverse mathematics (en.wikipedia.org/wiki/Reverse_mathematics) is an area of logic which measures the strength of theorems by comparing their logical strength over a weak base theory. $\endgroup$ – Jason Rute Jul 18 '13 at 0:11
  • 2
    $\begingroup$ From Cris Conidis' research statement: "In the future I will examine and classify [in the setting of reverse mathematics] the number theoretic consequences of theorems from the (largely unexamined) number theoretic literature, such as Galois’ Theorem and Hilbert’s Theorem 90." His website: math.vanderbilt.edu/~conidis/Chris_J._Conidis/Home.html $\endgroup$ – Jason Rute Jul 18 '13 at 0:18
  • 1
    $\begingroup$ Theorem 90 is about finite cyclic extensions, so it's provable in $RCA_0$ (modulo some induction, which I don't expect). Kummer theorey might be much more interesting. Why are you interested in Theorem 90? $\endgroup$ – François G. Dorais Jul 18 '13 at 3:29
  • 1
    $\begingroup$ I am interested in it as a n example of cohomology pretty directly interpretable in PA. $\endgroup$ – Colin McLarty Jul 18 '13 at 9:01

It is provable in RCA0 that if $K$ is a finite Galois extension of $F$ with Galois group $G$ then $H^1(G,L^\times)$ is trivial (every cocycle $a:G \to L^\times$ is a coboundary).

The finite Galois theory in RCA0 was discussed by Friedman, Simpson and Smith in Countable algebra and set existence axioms [APAL 25 (1983), 141–181; MR0725732]. The executive summary is that finite Galois theory works fine in RCA0.

The usual proof of Theorem 90 works in RCA0. The linear independence of characters is provable using only $\Sigma^0_1$-induction since one can check linear dependence by testing at a fixed finite basis of $L$ over $K$. It follows that if $a:G \to L^\times$ is a cocycle then there is a $x \in L^\times$ such that $y = \sum_{\sigma \in G} a(\sigma)\sigma(x)$ is nonzero. Standard algebraic manipulations then show that $a(\sigma) = y/\sigma(y)$, i.e. $a$ is a coboundary.

Infinite Galois extensions are more subtle as discussed in my paper with Jeff Hirst and Paul Shafer. For Theorem 90, it seems that the issue is a definitional one more than anything else.

If $K$ is an infinite Galois extension of $F$ such that $F$ is a subset of $K$ (as opposed to being merely embeddable in $K$) then the Galois group $G$ of $K$ over $F$ is a separably closed subgroup of permutations of $K$. Then $G$ can be represented as a complete separable metric space in the usual manner and one can make sense of the idea of a continuous cocycle $a:G \to L^\times$. In that scenario, RCA0 does prove that $H^1(G,L^\times)$ is trivial since every continuous cocycle is a lifting of a cocycle from a finite Galois subextension.

In the general case where $F$ is merely embeddable in $K$, several issues arise. Galois theory still makes sense over WKL0 but the Galois group $G$ is not necessarily separably closed. It is unclear what a continuous cocycle would be in that case. So the main issue is to define $H^1(G,L^\times)$ before proving anything about it. Note that this case is impossible in ACA0 since we can always comprehend the range of an embedding. So, assuming ACA0, $H^1(G,L^\times)$ is always well defined and always trivial.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.