Intuitively, cohomology operations are ways to locally compute a cocycle $\alpha\in H^i(X, G)$ from any cocycle $\beta\in H^j(X, H)$. Formally, they are in one-to-one correspondence with homotopy classes of maps $K(G, i)\rightarrow K(H, j)$, which in turn correspond to elements $\theta \in H^j(K(G,i), H)$. Roughly speaking, $\alpha$ can be computed locally from $\beta$ by pulling back a fixed representative of $\theta$ along the map $X\rightarrow K(G, i)$ representing $\alpha$.
The cup product is a way to locally compute a cycle $\alpha\in H^{i+j}(X, G)$ from any two cycles $\beta\in H^i(X, G)$ and $\gamma\in H^j(X, G)$, and could therefore be interpreted as a "2-variable cohomology operation", which leads me to the following questions:
- Is there a general theory of multi-variable cohomology operations? I've come across "secondary cohomology operations", but that seems to be different from what I'm looking for.
- Are there $n$-variable cohomology operations besides combinations of the cup product and 1-valued (ordinary) cohomology operations?
- Elements of $H^i(X,G)\times H^j(X,H)$ are represented by maps $X\rightarrow K(G,i)\times K(H,j)$. So 2-variable cohomology operations $H^i(X,G)\times H^j(X,H)\rightarrow H^k(X, J)$ should be in one-to-one correspondence with elements in $H^k(K(G,i)\times K(H,j), J)$. Applying Kuenneth's formula to the latter seems to be a way towards arguing that the answer to the previous point is "no", but how does this work exactly?
- Characteristic classes (of the tangent bundle) are ways to compute a cycle just from the local presentation of the manifold. Does it make sense to interpret characteristic classes as 0-variable cohomology operations?
- Is the cup product with a characteristic class a cohomology operation? It certainly is a way to locally compute a cocycle from another cocycle. It is not a homomorphism of cohomology groups, but that's not a must for a cohomology operation according to some sources. Even though, I'm a bit confused about the last point: As a natural transformation between cohomology functors, it would have to be a group homomorphism, right? Or can one simply consider the cohomology functors to Set instead of AbelianGroup?