The positive braid monoid on $n$ strands is the monoid with generators $s_1$, $s_2$, ..., $s_{n-1}$ and relations $$s_i s_{i+1} s_i = s_{i+1} s_i s_{i+1} \qquad s_i s_j = s_j s_i \text{for}\ |i-j| \geq 2.\qquad (\ast)$$ For any element $\beta$ in the positive braid monoid, let $\Gamma(\beta)$ be the graph whose vertices are words for $\beta$, and whose edges correspond to using a single braid relation from the list $(\ast)$. For example, $\Gamma(s_1 s_1 s_2 s_1)$ looks like $$s_1 s_1 s_2 s_1 \longleftrightarrow s_1 s_2 s_1 s_2 \longleftrightarrow s_2 s_1 s_2 s_2.$$

This graph is a tree, but it is easy to find $\beta$ where this graph has cycles.

Main Question: What is a list of cycles which generates $\pi_1(\Gamma(\beta))$?

Here is the list of cycles which I conjecture works.

(1) If a word $w$ for $\beta$ can have two braid relations applied to it in non-overlapping positions, than we can apply the relations in either order, making a $4$-cycle in $\Gamma(\beta)$.

(2) If a word for $\beta$ contains a substring $s_p$, $s_q$, $s_r$, where $s_p$, $s_q$ and $s_r$ all commute, then we can glue in a hexagon, as in the figure below:

enter image description here

(3) If a word for $\beta$ contains a substring $s_i s_j s_{j+1} s_j$ where $s_i$ commutes with $s_i$, $s_j$, then we can glue in an octagon, as in the figure below:

enter image description here

(4) If a word for $\beta$ contains a reduced word for the longest element in $S_4$, then we can glue in a $14$-gon, as shown in the figure below.

enter image description here

Question: Does this list of cycles generate? If not, can we always generate $\pi_1(\Gamma(\beta))$ with a list of cycles like this, coming from subwords of some bounded length?

  • 4
    $\begingroup$ Is this a duplicate of mathoverflow.net/questions/432075 ? $\endgroup$ Feb 17 at 20:12
  • 1
    $\begingroup$ Yup, I do. I'm trying to understand these "weaves" that Cassals, Gorsky and company work with, which involve very nonreduced words. $\endgroup$ Feb 17 at 22:55
  • 1
    $\begingroup$ If it turned out that cycles supported on reduced subwords (like all the examples I have listed) generate $\pi_1$, that would be very good news for me. $\endgroup$ Feb 17 at 22:56
  • 1
    $\begingroup$ I skimmed the Tits paper doi.org/10.1007/978-1-4612-5648-9_35 . The result for reduced words is Proposition 4, and I believe that it is the same list of generators that I have conjectured: I believe that Tits' (a) is non-overlapping braid relation and (b) is moves supported on a rank $3$ subgroup which, in type $A$, means $A_1^3$, $A_2 \times A_1$ and $A_3$, which are the types I listed. $\endgroup$ Feb 18 at 13:28
  • 4
    $\begingroup$ Hi David, Ben and I also stumbled across this question and answered it (in some form) here: arxiv.org/abs/1405.4928. There is also a discussion in Ronan's book on buildings if I recall correctly. There is also a higher-categorical question (relations between relations between relations...) which should have a nice answer in terms of finite parabolic subgroups of certain ranks. (E.g. the finite parabolics of rank 3 give rise to the "relations between relations" that you see.) $\endgroup$ Feb 18 at 23:33

1 Answer 1


I found a published reference! This is the main result of:

Fukushi, Takeo, On a braid monoid analogue of a theorem of Tits., SUT J. Math. 47, No. 1, 45-53 (2011). ZBL1235.20036.

I'll keep my write up below.

My conjectured list generates $\pi_1$. This proof is basically the one in Lemma 3 and Proposition 4 of Tits' paper that Sam Hopkins referenced in the comments, plus some facts from Garside theory.

The Garside theory part:

I'll write $B_n^+$ for the positive braid monoid, $S_n$ for the symmetric group. For any subset $J$ of $[n-1]$, I write $(w_0)_J$ for the longest element in the corresponding parabolic subgroup of $S_n$ and $\Delta_J$ for the element of $B_n^+$ given by any reduced word for $(w_0)_J$.

We equip $B_n^+$ with right weak order, meaning that $u \preceq w$ if we can write $w = uv$ for $u$ and $v$ in $B_n^+$.

Theorem: $B_n^+$ is a meet semilattice with respect to $\preceq$, i.e., for any $x$ and $y$ in $B_n^+$, there is an element $x \wedge y$ of $B_n^+$ such that $z \preceq x$, $y$ if and only if $z \preceq x \wedge y$.

I found several sources attributing this to [Garside], but I couldn't find it there on a quick read. A proof can be found in [Epstein et al, Corollary 9.3.7].

We call $s_i$ a left descent of $w \in B_n^+$ if $s_i \preceq w$, in other words, if there is a word for $w$ starting with $s_i$.

Corollary: Let $J \subset [n-1]$ and let $w \in B_n^+$. Suppose that, for all $i \in J$, the braid generator $s_i$ is a left descent of $w$. Then $\Delta_J \preceq w$.

For $|J|=2$, this is Theorem H in [Garside]; I will need it for $|J|=2$ and $|J|=3$.

Proof Let $z = w \wedge \Delta_J$. Then we have $s_i \preceq z$ for all $i \in J$, and $z \preceq \Delta_J$. But the interval below $\Delta_J$ is isomorphic to the weak order on the Coxeter group $(S_n)_J$, and the only element in this weak order which is above $s_i$ for all $i \in J$ is $(w_0)_J$. So $z = \Delta_J$ and $w \succeq \Delta_J$. $\square$.

The part due to Tits:

I will need Lemma 3 from [Tits]. I think that Lemma 3 is missing a key word, which I have inserted in bold.

Lemma: Let $X$ and $X'$ be connected graphs and let $\phi : X \to X'$ be a surjective morphism of graphs, meaning that, for each edge $(a,b)$ of $X$, either $\phi(a) = \phi(b)$ or else $(\phi(a), \phi(b))$ is an edge of $X'$. Let $C$ be a collection of cycles in $X$. Suppose the following:

(1) For each vertex $a$ of $X$, the graph $\phi^{-1}(x)$ is connected, and those cycles of $C$ lying in $\phi^{-1}(a)$ generate $\pi_1(\phi^{-1}(a))$.

(2) For any edge $(a', b')$ in $X'$ and any two edges $e_1= (a_1, b_1)$, $e_2 =(a_2, b_2)$ in $X$ with $\phi(a_1) = \phi(a_2) = a'$ and $\phi(b_1) = \phi(b_2) = b'$, there is a cycle in the subgroup generated by $C$ of the form $e_1 f e_2^{-1} g$ where $f$ is a path $b_1 \leadsto b_2$ in $\phi^{-1}(b')$ and $g$ is a path $a_2 \leadsto a_1$ in $\phi^{-1}(a')$.

Then $C$ generates $\pi_1(X)$ if and only if $\phi(C)$ generates $\pi_1(X')$.

Using this lemma of Tits, I prove the result in almost exactly the same way that Tits proves his Proposition 4:

Proof of main result We induct on the length of $\beta$; the base case $\ell(\beta) = 0$ is clear. Let $J$ be the set of left descents of $\beta$; let $K$ be the complete graph on vertex set $J$, and define a map $\phi : \Gamma(\beta) \longrightarrow K$ sending a word $w$ for $\beta$ to its first letter. We now need to verify several things, in order to conclude by the Lemma.

The map $\Gamma(\beta) \to K$ is surjective. Surjectivity on vertives is obvious. For each edge $(i,j)$ of $K$, since $i$ and $j$ are both in $J$, we have $s_i \preceq \beta$ and $s_j \preceq \beta$. By the Corollary, this implies that $\Delta_{\{ i,j \}} \preceq \beta$, so there are words for $\beta$ of the form $(s_i s_j \cdots) u$ and $(s_j s_i \cdots) u$. Applying a braid operation to relate these two words gives us an edge of $\Gamma(\beta)$ lying above $(i,j)$. $\square$

The cycles $\phi(C)$ generate $\pi_1(K)$. Clearly, $\pi_1(K)$ is generated by its cycles of length $3$. For any $i$, $j$, $k \in J$, if $s_i$, $s_j$ and $s_k$ are all left descents of $\beta$, there there is a word for $\beta$ of the form $\Delta_{\{ i,j,k \}} u$. Using different reduced words for $\Delta$, we get one of the three pictures I drew above, so in each case the $3$-cycle $(i \leadsto j \leadsto k \leadsto i)$ is in the image of a cycle from $\Gamma(\beta)$. $\square$

For each vertex $i$ of $K$, the fundamental group of $\phi^{-1}(i)$ is generated by the cycles from $C$ lying in $\phi^{-1}(i)$.

Write $\beta = s_i \beta'$. The preimage of $i$ is precisely the graph $\Gamma(\beta')$, and the claim follows by induction.

Given an edge $(i, j)$ in $K$, and two edges $e_1=(u_1, v_1)$ and $e_2=(u_2, v_2)$ lying above $(i,j)$ there are paths as required.

Abbreviate $\Delta_{\{i,j\}}$ to $\delta$ and put $\beta = \delta \beta'$.

For $r \in \{ 1,2 \}$, $u_r$ must be joined to $v_r$ by a single braid relation which switches the first letter from $i$ to $j$. That means that $u_r$ and $v_r$ must be of the forms $(s_i s_j \cdots) x_r$ and $(s_j s_i \cdots ) x_r$, with $x_r$ a word for $\beta'$. Now, $\Gamma(\beta')$ is connected, so we can find a path $x_1 \leadsto x_2$ in $\Gamma(\beta')$; let $f$ be the induced path $(s_j s_i \cdots ) x_1 \leadsto (s_j s_i \cdots) x_2$ and let $g$ be the induced path $(s_i s_j \cdots) x_2 \leadsto (s_i s_j \cdots) x_1$. Then we can fill in $e_1 f e_2 g^{-1}$ using a collection of $4$ cycles involving non-overlapping braid moves. (For each $4$-cycle, one braid move is in the prefix $\delta$ and the other is in the suffix $\beta'$.) $\square$.

As should be evident, I have written this in a way which will generalize to other Artin groups, but I am confused about the state of Garside theory for other Artin groups, and I only care about braids right now, so I'll stop here.

[Epstein et al] Epstein, David B. A.; Cannon, James W.; Holt, Derek F.; Levy, Silvio V. F.; Paterson, Michael S.; Thurston, William P., Word processing in groups, Boston, MA etc.: Jones and Bartlett Publishers. xi, 330 p. (1992). ZBL0764.20017.

[Garside] Garside, F. A., The braid group and other groups, Q. J. Math., Oxf. II. Ser. 20, 235-254 (1969). ZBL0194.03303.

[Tits] Tits, Jacques, A local approach to buildings, The geometric vein, The Coxeter Festschr., 519-547 (1982). ZBL0496.51001.


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.