For some $g \geq 2$, let $\Gamma_g$ be the fundamental group of a closed genus $g$ surface and let $S_g=\{a_1,b_1,\ldots,a_g,b_g\}$ be the usual generating set for $\Gamma_g$ satisfying the surface relation $[a_1,b_1]\cdots[a_g,b_g]=1$. Dehn proved the following famous theorem.

Theorem : Let $w$ be a reduced word in the generators $S_g$ such that $w$ represents the identity in $\Gamma_g$. Then $w$ contains a subword $r_1$ of length $k \geq 2g+1$ such that there exists a word $r_2$ of length $4g-k$ with $r_1 r_2$ a cyclic permutation of $[a_1,b_1]\cdots[a_g,b_g]$.

This leads to Dehn's algorithm for solving the word problem in a surface group. Namely, start with a word $w$ which is reduced. If $w$ does not contain a subword $r_1$ as in the Theorem, then $w$ does not represent the identity in $\Gamma_g$. Otherwise, we can replace the subword $r_1$ of $w$ with $r_2$, shortening $w$. After possibly reducing $w$, we do the above again. We stop if we have reduced $w$ to the empty word.

There are many sources for Dehn's algorithm; for instance, Stillwell's book on geometric topology and combinatorial group theory gives a proof using elementary topology, and Lyndon-Schupp give a proof based on small cancellation theory. However, many sources say that Dehn's original proof used hyperbolic geometry. I've heard people tell me that he studied the usual tessalation of the hyperbolic plane by regular $4g$-gons and applied something like Gauss-Bonnet; however, I've never managed to reconstruct this kind of proof. Can someone either spell it out or give a (modern) reference?

  • $\begingroup$ Its been a long time but I recall Dehn's proof being essentially identical to the small cancellation theory proof. Dehn is using the Cayley graph in the hyperbolic plane as he had no need for a general setup. $\endgroup$ Commented Jul 17, 2013 at 5:28
  • $\begingroup$ One little remark: you remplace $r_1$ by $r_2^{-1}$ to keep the stuff correct ... Of course you understand it but to keep things clean :) $\endgroup$
    – Olga
    Commented Jan 19, 2017 at 19:56
  • $\begingroup$ Another remark: shouldn't $r_j$ be the words in the alphabet of $a_j, b_j$ but also their inverses ?... $\endgroup$
    – Olga
    Commented Jan 19, 2017 at 20:03

2 Answers 2


Dehn gave at least three solutions of the conjugacy problem for surface groups, which can be found in my translation in the book Papers on Group Theory and Topology (Springer 1986), Papers 2, 4, and 5.

The first is based on an idea of Poincaré: lifting a curve to the universal cover, which is the disk model of the hyperbolic plane, replacing it by the hyperbolic straight line with the same endpoints, then projecting this line back to the surface as a "geodesic representative" of the original curve. Two curves are free homotopic (and the corresponding group elements are conjugate) if and only if they have the same geodesic representative.

This unpublished proof is conceptually simple, but it is not clear how to determine whether two curves have the same geodesic representative. Dehn's first published proof (Paper 4, 1912) worked out a way to do this, arriving in a roundabout way at Dehn's algorithm.

Shortly afterward (Paper 5, still in 1912) Dehn noticed that the algorithm follows easily from combinatorial properties of the tessellation of the universal cover, and the hyperbolic metric is irrelevant. This is essentially the proof we use today.

  • $\begingroup$ Is it possible you meant to write "Paper 5" in your last paragraph? Both 4 and 5 are from 1912, but I suspect you didn't intend to refer to the same paper twice. $\endgroup$
    – Michael
    Commented Apr 23, 2017 at 4:52
  • $\begingroup$ @Michael, thanks, I've made the correction. $\endgroup$ Commented Apr 23, 2017 at 22:57

In the paper "The combinatorial structure of cocompact discrete hyperbolic groups(here)", Cannon gave a proof of conjugacy problem for surface groups using hyperbolic geometry (Theorem 6). It's the the same proof mentioned by Stillwell in the second paragraph of his answer.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.