Skip to main content
13 events
when toggle format what by license comment
Jan 28, 2021 at 18:06 comment added Ian Agol For finitely generated subgroups of $GL_N(\mathbb{Z})$, I think one can improve the bound from $O(n^2)$ to $O(n\log^2n)$. The point is that for a word of length $n$, you can divide it into two subwords of length $\leq \lceil n/2\rceil$, which have length (in bits) bounded by $C n/2$. Multiplying these takes time $O(n/2 \log(n/2))$. en.wikipedia.org/wiki/… By induction, this gives $O(n \log^2n)$ I think.
Dec 20, 2020 at 23:43 comment added Moishe Kohan @dodd: Afaik, linearity is an open problem (for graph-manifolds without metrics of nonpositive curvature). As for Dehn function, it provides only an upper bound on complexity of the WP, as I explain, one can do better.
Dec 20, 2020 at 23:36 comment added markvs All fundamental groups of closed 3-manifolds are linear, so they have word problem in logspace. Sol has exponential Dehn function, Nil has Dehn function $\sim n^3$.
Dec 20, 2020 at 19:41 history edited Moishe Kohan CC BY-SA 4.0
added 188 characters in body
Dec 20, 2020 at 19:37 comment added Ian Agol Good point (and clearly stated in your answer).
Dec 20, 2020 at 19:32 comment added Moishe Kohan @IanAgol: Graph-manifold groups are automatic (apart from Nil and Sol), hence, have quadratic Dehn function for this reason.
Dec 20, 2020 at 19:15 comment added Ian Agol Some graph manifolds virtually fiber, and hence their fundamental groups are (virtually) the subgroup of a mapping class group, which is automatic. So they also have a quadratic time solution to the word problem. doi.org/10.2307/2118637 But I think that there are graph manifolds that are neither cubulated nor virtually fibered.
Dec 20, 2020 at 16:23 history edited Moishe Kohan CC BY-SA 4.0
added 118 characters in body
Dec 19, 2020 at 22:40 comment added Moishe Kohan @YCor: Right, I was careless. I will correct the answer once I am back to my computer.
Dec 19, 2020 at 7:07 comment added YCor If you multiply $n$ bounded matrices in $SL(3,Z)$ you get matrices of exponential size, i.e., exponents have $\sim n$ digits. The multiplication of a single such matrix with a single bounded matrix takes time $\sim n$. So it looks like the total time is likely to be $O(n^2)$ rather than $O(n)$.
Dec 19, 2020 at 3:29 vote accept Ben Cooper
Dec 19, 2020 at 3:18 comment added Ben Cooper Thank you, this is helpful
Dec 19, 2020 at 0:49 history answered Moishe Kohan CC BY-SA 4.0