Let $M,N$ be topological manifolds such that $M$ does not admit a $PL$ structure and $N$ does. Is $M\#N$ still a triangulable manifold?
1 Answer
In high dimensions Galewski and Stern (and independently Matsumoto) proved that a manifold $M$ is triangulable iff $\beta \Delta(M)=0$. Here $\Delta \in H^4(M;\Bbb Z/2)$, and $\beta$ is the Bockstein corresponding to the short exact sequence $\ker \mu \to \Theta_{\Bbb Z} \to \Bbb Z/2$. The middle term is the homology cobordism group of 3-manifolds and the last map is the Rokhlin invariant of 3-manifolds.
These invariants are additive under connected sum. Therefore if $\beta \Delta(N)$ is nonzero, so is $\beta \Delta(M \# N)$ as long as the dimension is at least 6, and so $M \# N$ remains non-triangulable.
The only leftover mystery cases are dimensions 4 and 5. In dimension 4 triangulable manifolds are automatically PL and therefore smooth. So we see that $M \# N$ is smoothable only if $M$ and $N$ are not both definite manifolds of the same signature, but otherwise we are out of luck. In dimension 5, $M \# N$ is not triangulable if $M$ is but $N$ is not: $$\beta \Delta(M \#N) = \beta \Delta(M) + \beta \Delta(N) = \beta \Delta(N) \neq 0.$$ But eg $M \# M$ is always triangulable.