For a finite CW complex $X$, the Chern character gives an isomorphism of finite-dimensional vector spaces: $$ ch : K^*(X)\otimes \mathbb{Q} \to H^*(X, \mathbb{Q}). $$ The vector space $V = H^*(X, \mathbb{Q})$ thus comes equipped with two natural maximal-rank lattices:

  • $L_H = H^*(X, \mathbb{Z}) / T$ (where $T$ is the torsion),
  • $L_K = ch(K^*(X) / T')$ (where $T'$ is the torsion).
What we can say about the relationship between $L_H$ and $L_K$?

Here is simple necessary relationship. Let $n=[\dim X/2]$; using denominators of size at most $n!$, the Chern character can be expressed as a linear combination of Chern classes. Chern classes are integral, so we must have: $$ n! \cdot L_K \subseteq L_H. $$

What other necessary conditions are known? Is there a known set of necessary and sufficient conditions? I.e., any $(V, L_H, L_K)$ satisfying them can be realised by some $X$.

For what it's worth, my motivation comes from spheres. Suppose we define a topological invariant $k$ to be the smallest natural number such that $k\cdot L_K \subseteq L_H$ (which presumably captures only a tiny bit of the relationship). Then the fact that $k=1$ for spheres allows a slick proof that the spheres admit no almost complex structure above dimension six. (In fact $L_K = L_H$ for spheres.)

Incidentally, the above ignores the $\mathbb{Z}/(2)$-grading. I haven't thought about this but I'd even be interested in the case when $X$ is even-dimensional with no odd-dimensional cohomology (e.g., a non-singular complex quadric).


I remember, dimly, work of J.Frank Adams,published in "Topology" (journal), in the early 1960's, on this kind of problem, Bruno Harris

  • $\begingroup$ Thank you for the lead! I will chase it down at the weekend. $\endgroup$ – Oliver Nash Oct 24 '18 at 20:30

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.