$\DeclareMathOperator\Gal{Gal}\DeclareMathOperator\SL{SL}\DeclareMathOperator\GL{GL}$Let $K/k$ be a Galois extension. Then one generalisation of Hilbert's Satz 90 states that $H^1(\Gal(K/k),\GL_n(K))=0$. From this it is not difficult to deduce that $H^1(\Gal(K/k),\SL_n(K))=0$.

In Serre's Galois cohomology book (III.3.1), there is a conjecture asserting that $H^1(\Gal(k^{\mathrm{sep}}/k),G(k^{\mathrm{sep}}))=0$ for any simply-connected semisimple $G$ over $k$ providing $k$ fulfills a cohomological dimension condition.

In any case, my question concerns with a particular choice of $K/k$:

Question: Letting $K/k=\mathbb{C}/\mathbb{R}$, do we have $H^1(\Gal(\mathbb{C}/\mathbb{R}),G(\mathbb{C}))=0$ for any simply-connected semisimple algebraic group $G$ defined over $\mathbb{R}$?

  • 8
    $\begingroup$ I'm not sure whether I misunderstood your question, but ⓐ to be clear, $\mathbb{R}$ does not satisfy the condition of having cohomological dimension $≤2$ (since its Galois group is finite cyclic nontrivial, and such have periodic cohomology), and ⓑ if $G$ is, for example, the simply-connected semisimple algebraic group $G_2$, then $H^1(k,G_2)$ classifies octonion algebras over $k$ (Serre, Galois Cohomology, III, appendix 2, 3.3), and for $k=\mathbb{R}$ there are two of them (the split octonions and usual octonions), so $H^1(\mathbb{R},G_2)$ is nontrivial. $\endgroup$
    – Gro-Tsen
    Commented May 28 at 8:32
  • $\begingroup$ PS: If you want a good introduction to the Galois cohomology of semisimple groups over the reals, Mikhail Borovoi has some nice papers (like this one) and talks about this topic. I'm a bit confused between the similarly-named ones, but if you write to him, I'm sure he can recommend where to start. $\endgroup$
    – Gro-Tsen
    Commented May 28 at 8:47
  • $\begingroup$ @Gro-Tsen Thank you! This answers my question. $\endgroup$
    – user148212
    Commented May 28 at 8:58
  • $\begingroup$ OK, then I'll copy it as an actual answer, then. $\endgroup$
    – Gro-Tsen
    Commented May 28 at 9:19

2 Answers 2


[Copied and extended from comments.]

First, it needs to be clarified in relation to the question that $\mathbb{R}$ does not satisfy the cohomological condition (viꝫ., having cohomological dimension $\leq 2$) in the hypothesis of the cited conjecture by Serre. Indeed, $\mathbb{R}$ has absolute Galois group cyclic of order $2$, and finite cyclic groups have periodic cohomology (so not at all finite cohomological dimension). So the following negative answer is not a counterexample to Serre's conjecture (😅).

Now a negative answer to the question posted is provided, for example, by the simply-connected semisimple algebraic group $G_2$ (the exceptional group of dimension $14$). In this case, the nonabelian Galois cohomology set $H^1(k, G_2)$ classifies octonion algebras over $k$ (Serre, Galois Cohomology, III, appendix 2, 3.3), as is to be expected since $G_2$ is the automorphism group of the octonions; but there are (exactly) two octonion algebras over the real numbers: the split octonions (which are the distinguished element of $H^1(\mathbb{R}, G_2)$) and the usual octonions. (Equivalently, since $G_2$ is also adjoint, $H^1(\mathbb{R}, G_2)$ classifies real forms of $G_2$ itself, of which there are exactly two: the split form, which is the automorphisms of the split octonions, and the compact form, which is the automorphisms of the usual octonions.) So $H^1(\mathbb{R}, G_2)$ has two elements: it is not trivial.

For more about the description of the Galois cohomology of semisimple groups over the reals, I recommend papers and talks by Mikhail Borovoi such as Borovoi & Timashev, Galois cohomology of real semisimple groups. For more about the Galois cohomology of $G_2$ and its relation to octonion algebras, see Springer & Veldkamp, Octonions, Jordan Algebras and Exceptional Groups (2000).


In addition to the answer of Gro-Tsen: Let $G=\operatorname{SU}(3)$. Then $H^1({\mathbb R},G)$ classifies matrices of Hermitian forms in 3 variables with determinant 1. There are two equivalence classes of such matrices: with representatives $\operatorname{diag}(1,1,1)$ and $\operatorname{diag}(-1,-1,1)$. Thus $\#H^1({\mathbb R},G)=2$. Similarly, $\#H^1({\mathbb R},\operatorname{SU}(2n+1))=n+1$.

In general, if $G$ is a compact connected $\mathbb R$-group with maximal torus $T$ and Weyl group $W={\mathcal N}_G(T)/T$, then there is a canonical isomorphism $$T({\mathbb R})_2/W\cong H^1({\mathbb R},G).$$ This is a result of Borel and Serre of 1964, repeated in Section III.4.5 of Serre's book "Galois cohomology". For a generalization to connected reductive $\mathbb R$-groups that are not necessarily compact, see my paper Galois cohomology of reductive algebraic groups over the field of real numbers, Commun. Math. 30 (2022), Issue 3, 191–201.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.