I am trying to prove the following statement of real Bott periodicity, on the level of actual spaces of Clifford module extensions (i.e., not equivalence classes of modules).

Let $W = \mathbb{R}^{\infty}$ (in the direct sum sense), equipped with the standard inner product. Fix in advance a set of skew-symmetric operators $e_1, e_2, \ldots $ on $W$ for which $e_i^2 = -I$ and $e_ie_j = -e_je_i$ for $i \neq j$. Let $X(n,W)$ be the space of operators $f_n$ on $W$ which anticommute with $e_1, \ldots, e_{n-1}$ have $f_n^2=-I$, and for which $\ker(f_n-e_n)^{\perp}$ is finite dimensional, and equip this space with the topology induced by the operator norm. I wish to show that $$ X(n,W) \cong X(n+8,W) $$ where $\cong$ denotes homeomorphism, or barring that, homotopy equivalence. Note that $X(n,W)$ is the space of orthogonal $Cl_n$-module structures on $W$ which restrict to the "standard" $Cl_{n-1}$-module determined by $e_1, \ldots, e_{n-1}$. Here $Cl_n$ denotes the real Clifford algebra on $n$ generators with negative definite quadratic form.

I hope to make use only of the well-known isomorphism of real Clifford algebras $Cl_{n+8} \cong Cl_{n}\otimes_{\mathbb{R}} Cl_{8} \cong Cl_{n}\otimes_{\mathbb{R}}\mathbb{R}(16)$. I'm aware that additional technical assumptions may be necessary, for instance regarding a "complete universe" of representations, but my hope is just to get the basic idea.

I would be content to show that the space of irreducible $Cl_n$-module extensions on $V$ of appropriate dimension is homeomorphic to the space of $Cl_{n+8}$-module extensions on $V \otimes \mathbb{R}^{16}$, with the map induced by tensoring with the canonical representation of the $16 \times 16$ real matrix algebra $\mathbb{R}(16)$.

The main paper I have been following is Behrens - Addendum to "A New Proof of Bott Periodicity".

  • 1
    $\begingroup$ A non-mathematical remark: an easy, and I think reasonably standard, way to write "$\mathbb R^\infty$ (in the direct sum sense)" is "$\mathbb R^{\oplus\infty}$". $\endgroup$
    – LSpice
    Dec 2, 2019 at 19:14
  • 2
    $\begingroup$ @LSpice And yet $\mathbb{R}^\infty$ is the standard notation in algebraic topology and related fields :). $\endgroup$ Dec 2, 2019 at 19:52

1 Answer 1


I believe I have the idea of a proof. Apologies for the mess - hopefully it is somewhat intelligible (and somewhat correct).

We follow the form of this isomorphism: $Cl_{n+8} \cong Cl_n\otimes_\mathbb{R}Cl_8$: \begin{cases} e_i \mapsto 1\otimes \eta_i, &\text{for } i=1,\ldots,8\\ e_j \mapsto e_{j-8}\otimes \eta_1\eta_2\cdots \eta_8, &\text{for } j=9,\cdots,n+8, \end{cases} where $\eta_1,\ldots,\eta_8$ are orthonormal generators of $Cl_8$. Taking $e_i$ now to denote the operators corresponding to these generators under the "standard" representation of $Cl_n$ on $W$ and $\eta_i$ the operators corresponding to the "standard" representation of $Cl_8$ on $\mathbb{R}^{16}$, we get a map $X(n,W) \to X(n+8,W\otimes_{\mathbb{R}}\mathbb{R}^{16})$, where $f_n$ maps to $f_n\otimes\eta_1\cdots\eta_8$. This is apparently continuous and injective, but not clearly surjective. So we take a different tack:

Let $O_{Cl_n}(W)$ denote the subgroup of $O(W)$ which fixes each of $e_1,\ldots,e_n$ under conjugation. Then we may identify $X(n,W)$, as the homogeneous space $O_{Cl_{n-1}}(W)/O_{Cl_n}(W)$ by considering the orbit of $e_n$ (actually for $n\equiv 3(\text{mod }4)$ there will be multiple homeomorphic orbits corresponding to inequivalent representations but let's ignore this for now).

Similarly $X(n+8,W\otimes\mathbb{R}^{16}) \cong O_{Cl_{n+7}}(W\otimes\mathbb{R}^{16})/O_{Cl_{n+8}}(W\otimes\mathbb{R}^{16})$. However any subgroup of $O(W\otimes\mathbb{R}^{16})$ which fixes $1\otimes\eta_1\ldots,1\otimes\eta_8$ must be of the form $O(W) \otimes I$ by the appropriate form of Schur's lemma: $\eta_1,\ldots,\eta_8$ give the only irreducible representation of the simple algebra $Cl_8$ on $\mathbb{R}^{16}$ up to equivalence, so the action of this subgroup must just "permute" these irreps, while acting on them only by $\pm 1$. So $O_{Cl_{n+7}}(W\otimes\mathbb{R}^{16}) \cong O_{Cl_{n-1}}(W)$ and $O_{Cl_{n+8}}(W\otimes\mathbb{R}^{16}) \cong O_{Cl_{n}}(W)$, thus $X(n+8,W\otimes\mathbb{R}^{16}) \cong X(n,W)$.


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.