If $X$ is a compact Hausdorff space, we can consider the Grothendieck ring of real vector bundles on $X$, $\mathit{KO}^0(X)$, and this extends to a generalized cohomology theory represented by a ring spectrum $\mathit{KO}$. Using complex vector bundles, we get another generalized cohomology theory, represented by $\mathit{KU}$. Using quaternionic vector bundles, we get a third one, represented by $\mathit{KSp} \simeq \Sigma^4\mathit{KO}$.

In the same way, one can define the Grothendieck ring of oriented vector bundles $\mathit{KSO}(X)$, spin vector bundles $\mathit{KSpin}(X)$, and so on for any $G$-structure. Do these constructions extend to spectra $\mathit{KSO}$, $\mathit{KSpin}$, and so on? If not, what issues arise?

The first concern I would imagine is a failure of Bott periodicity, but $\pi_k(\mathit{SO})$ and $\pi_k(\mathit{Spin})$ agree with $\pi_k(O)$ for $k\ge 2$, so in at least these cases some sort of construction might be possible.

An alternative approach sidestepping Bott periodicity would be to use algebraic $K$-theory: if $\mathsf{Vect}_{k}$ denotes the topological symmetric monoidal category of $k$-vector spaces, then $K(\mathsf{Vect}_{\mathbb R}) = \mathit{ko}$ and $K(\mathsf{Vect}_{\mathbb C}) = \mathit{ku}$. So if $\mathsf{Vect}^{\mathrm{or}}$ denotes the topological symmetric monoidal category of oriented real vector spaces, would it be reasonable to define $\mathit{kso}:= K(\mathsf{Vect}^{\mathrm{or}})$? Is this an interesting object? This could also generalize to $G$-structures.

Someone must have thought about this, but I can't find it written down anywhere.

Edit: Though I've accepted Denis' excellent answer, I would still be interested to learn of existing references which consider these kinds of cohomology theories/spectra.

  • 3
    $\begingroup$ $KSO(X)$ is just the kernel of the natural map from $KO(X)$ to $H^1(X,\mathbb Z/2)$. So probably if it can be defined, the spectrum is not far away from $KO$. $\endgroup$ – Will Sawin Apr 18 '17 at 2:04
  • 4
    $\begingroup$ Likewise, $KSpin$ is essentially only the $4$-connective cover of $KO$, only that we decided that we leave the $\pi_0$-unchanged. This happens in general if you take an $X$-structure classified by $BO\langle n\rangle BO$, just in case you want to consider $KString$ or so. What other interesting examples are there? $\endgroup$ – Lennart Meier Apr 18 '17 at 15:44

I am not sure if this is going to be a real answer to the question. However I believe these observations might be interesting.

Let me briefly sketch a way to describe a $G$-structure in (excessively) wide generality. Consider a fibration of spaces $\theta:X\to \coprod_n BO_n$. Then a $\theta$-structure on a vector bundle $V\to B$ is just a lift along $\theta$ of the map $X\to \coprod_n BO_n$ classifying $V$. Examples are orientations (with $X=\coprod_n BSO_n$), spin structures ($X=\coprod_n BSpin_n$), complex structures ($X=\coprod_n BU_n$) etc.

Now you want to define the sum of vector bundles with $\theta$-structure. In order to do so we need to require a little bit more of the map $\theta$. I'm going to assume that $X$ is an $E_\infty$-space and that $\theta$ is equipped with the structure of map of $E_\infty$-spaces. Then the abelian group of vector bundles with $\theta$-structure on some base $B$ is easily seen to be $$ \pi_0\mathrm{Map}(B,X)$$ Note that $\mathrm{Map}(B,X)$ has a canonical $E_\infty$-structure inherited from $X$. So it is natural to define the connective K-theory of vector bundles with $\theta$-structure as $$ K\theta^i(B) = \pi_{-i}\mathrm{Map}(B,X)^+$$ where $(-)^+$ is the group completion of $E_\infty$-spaces.

When the group completion of $X$ is given by a filtered colimit over the divisibility poset of $\pi_0X$ (for example $X=\coprod_n BO_n$, $X=\coprod_n BU_n$ or $X=\coprod_n BSp_n$) and $B$ is a finite space (that is the homotopy type of a CW complex with finitely many cells) you can rewrite the right hand side as $$ K\theta^i(B) = \pi_{-i}\mathrm{Map}(B,X^+)$$ since filtered colimits in spaces commute with finite limits. As Marc Hoyois notes in the comments, this is always true if every connected component of $X$ is simply connected. So $K\theta^i(-)$ restricted to finite spaces is a cohomology theory represented by the connective spectrum $X^+$.

Note that you might want a little more of this theory: you might want your $X^+$ to be an $E_\infty$-ring and not just a connective ring spectrum. This corresponds to the assumption that $X$ is an $E_\infty$-ring space and that $\theta$ respects this structure.

I do not know of a way to generalize nonconnective K-theory in this situation (of course one could just $K(1)$-localize, but that is a rather ad hoc procedure).

Also, the map $\theta$ is obviously just along for the ride, but since it allows us to think of $K\theta$ as "K-theory of vector bundles with structure" I think it helps the intuition.

| cite | improve this answer | |
  • $\begingroup$ This is nice! How do you see that you can pull the group completion inside when X is BO, BU and BSp? $\endgroup$ – James Cameron Apr 18 '17 at 5:16
  • 1
    $\begingroup$ @JamesCameron It follows from the fact that in the ∞-category of spaces finite limits commute with filtered colimits, and that in those cases the group completion can be given by a filtered colimit (since $\mathrm{Map}(B,-)$ is a finite limit). Now that I write it, I have the feeling that it should work for pretty much any $E_∞$-space $X$, let me see if I can straighten out the details.... $\endgroup$ – Denis Nardin Apr 18 '17 at 11:35
  • 4
    $\begingroup$ @DenisNardin I think that to express the group completion $X^+$ using filtered colimits you need to know that for every $x\in X$, there exists $m\geq 2$ such that the cyclic permutation of $x^m$ is homotopic to the identity (at least, I only know how to prove it in this case). These permutations live in $\pi_1$'s of $X$. In the $U_n$ and $Sp_n$ cases these are trivial, in the $O_n$ case we have $Z/2$ so any odd $m$ works. But this fails completely in other cases, eg when $X=\coprod_n BGL_n(R)$ for some nonzero commutative ring $R$. $\endgroup$ – Marc Hoyois Apr 18 '17 at 14:17
  • $\begingroup$ @MarcHoyois Hmm.. That is a good point. We can't expect to get algebraic K-theory spectra that easily. That's probably the best we can get in general. $\endgroup$ – Denis Nardin Apr 18 '17 at 14:35
  • $\begingroup$ This is really nifty. Thanks for writing it down! Time to play with some examples. $\endgroup$ – Arun Debray Apr 23 '17 at 4:49

Regarding preexisting references, I found a few examples of people using Grothendieck groups of oriented or spin vector bundles, but no cohomology theories or spectra. I'll leave them here in case anyone else is interested.

For $\mathit{KSpin}$:

For $\mathit{KSO}$:

  • Pejsachowicz, “Bifurcation of Fredholm maps II: the dimension of the set of bifurcation points.”
| 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.