on Burau faithfulness for four strands

Overview

The Burau representation is a classical linear representation of the braid group over Laurent polynomials. Bharathram, Birman, and Brendle prove that the four-strand Burau representation is faithful (Bharathram, Birman & Brendle, 2026). Together with the known faithful cases $n\le 3$ and the known nonfaithful cases $n\ge 5$, this gives the complete classification:

$${\rho }_n\text{ is faithful exactly for }n\le 4.$$

The proof converts the kernel question into a controlled non-cancellation problem. A braid in the Burau kernel would force certain signed Laurent polynomials, computed from intersections of arcs in a punctured disk, to vanish. The paper develops enough disk-sequence topology to show that the required cancellation cannot occur in the four-strand case.

Faithfulness and the Burau Kernel

A representation of a group is faithful if it is injective. For the Burau representation this means

$${\rho }_n(\Phi )=I⟹\Phi =1$$

for every braid $\Phi \in B_n$. Equivalently,

$$\ker {\rho }_n=\{1\}.$$

In concrete terms, the Burau representation assigns to each braid a matrix with entries in $\mathbb{Z}[t,t^{-1}]$. Faithfulness asks whether this matrix-valued invariant distinguishes braids without ambiguity. If two braids have the same Burau matrix, then

$${\rho }_n(\Phi )={\rho }_n(\Psi )⟺{\rho }_n(\Phi {\Psi }^{-1})=I.$$

Thus faithfulness is precisely the assertion that no nontrivial braid is invisible to the representation.

This post follows the convention used by Bharathram, Birman, and Brendle: ${\rho }_n$ denotes the unreduced Burau representation

$${\rho }_n:B_n\to {\mathrm{GL}}_n(\mathbb{Z}[t,t^{-1}]).$$

Some classical references phrase the same problem for the reduced Burau representation. The two formulations are equivalent for faithfulness, but the unreduced convention is the natural one for the topological model used below.

Strand-Number Classification

Let $B_n$ be the braid group on $n$ strands. Burau introduced the representation in 1935 (Burau, 1935). Before the new theorem, every strand number except $n=4$ had already been settled.

Strands	Status	Reason
$n=1$	faithful	$B_1$ is trivial.
$n=2$	faithful	$B_2\cong \mathbb{Z}$, and the generator has nontrivial Burau image.
$n=3$	faithful	Magnus-Peluso proved this algebraically; the new paper also gives a topological proof (Magnus & Peluso, 1969; Bharathram, Birman & Brendle, 2026).
$n=4$	faithful	This is the theorem of Bharathram-Birman-Brendle.
$n\ge 5$	not faithful	Moody proved nonfaithfulness for $n\ge 9$, Long-Paton improved this to $n\ge 6$, and Bigelow proved $n=5$ (Moody, 1991; Long & Paton, 1993; Bigelow, 1999).

The stabilization property explains why the faithful range cannot reappear at larger $n$: once a nontrivial kernel element exists, adding an extra strand preserves kernel membership.

Topological Construction

Let $D_n$ be a disk with marked points

$$p_1,\dots ,p_n$$

in its interior. The braid group is identified with the mapping class group

$$B_n=\mathrm{Mod}(D_n),$$

where homeomorphisms preserve the boundary and may permute the marked points. This viewpoint lets a braid act directly on arcs in a punctured disk.

Fix a basepoint $p_{\ast }$ on $\partial D_n$. Let

$$x_1,\dots ,x_n$$

be the standard free generators of ${\pi }_1(D_n,p_{\ast })$, with $x_i$ looping once around $p_i$. The homomorphism

$${\pi }_1(D_n,p_{\ast })\to \mathbb{Z}$$

sends a word in the $x_i$ to its total exponent sum. Let ${\tilde{D}}_n$ be the corresponding cyclic cover, whose deck group is generated by $t$. The relative homology module

$$H_1({\tilde{D}}_n,\{{\tilde{p}}_{\ast }\})$$

is free of rank $n$ over $\mathbb{Z}[t,t^{-1}]$. The induced braid action on this module is the unreduced Burau representation. This is the standard topological construction of Burau (Birman & Brendle, 2005).

Stabilization of kernel elements

If $f:B_n\hookrightarrow B_{n+1}$ is induced by adding one puncture, then

$$\Phi \in \ker {\rho }_n⟹f(\Phi )\in \ker {\rho }_{n+1}.$$

The four-strand proof uses this contrapositive style of reasoning: a carefully chosen five-strand test can rule out a hypothetical four-strand kernel element.

Moody Polynomial

Figure 1: The fixed test arcs. The blue arc is the measuring arc from p₁ to p₂; the red arc is the base test arc from the boundary basepoint to p₃.

Figure notation

In the figures, blue denotes the fixed measuring arc $\alpha$, red denotes the test arc $\beta$ obtained from ${\beta }_{\ast }^3$ by a braid action, and gold denotes a point-pushing loop. The diagrams record red-blue intersections and the signed Laurent monomials associated with those intersections.

The first reduction replaces a matrix calculation by an intersection calculation. Fix the arcs shown in Figure 1. The blue arc $\alpha$ joins $p_1$ to $p_2$, and the red arc ${\beta }_{\ast }^3$ joins the boundary basepoint $p_{\ast }$ to $p_3$. For a braid $\Phi$, write

$$\beta =({\beta }_{\ast }^3)\Phi .$$

The question is whether the moved arc $\beta$ can intersect $\alpha$ in a way that is invisible to Burau.

Choose lifts $\tilde{\alpha },\tilde{\beta }\subset {\tilde{D}}_n$. The Moody polynomial is

$$\mathbb{M}(\alpha ,\beta )=\sum _{\ell \in \mathbb{Z}}(t^{\ell }\tilde{\alpha },\tilde{\beta })t^{\ell }\in \mathbb{Z}[t,t^{-1}],$$

where the coefficient is the algebraic intersection number in the cyclic cover.

There is also a planar description. Put

$$\alpha \cap \beta =\{q_1,\dots ,q_m\},$$

ordered along $\beta$ from $p_{\ast }$ to $p_3$. Each crossing contributes a signed monomial

$${ϵ}_i{t}^{k_i},{ϵ}_i\in \{-1,1\},$$

so that

$$\mathbb{M}(\alpha ,\beta )=\sum _{i=1}^m{ϵ}_i{t}^{k_i}.$$

The unsimplified polynomial records every crossing. The simplified polynomial combines equal powers of $t$. Therefore cancellation can occur only when two distinct crossings have the same exponent and opposite signs.

For a braid $\Phi$, define

$${\mathbb{M}}_{\Phi }=\mathbb{M}(\alpha ,({\beta }_{\ast }^3)\Phi ).$$

Moody obstruction

For $n\ge 3$, if ${\mathbb{M}}_{\Phi }\ne 0$, then $\Phi \notin \ker {\rho }_n$.

More generally, if for some $\Gamma \in B_n$,

$${\mathbb{M}}_{\Phi \Gamma }\ne {\mathbb{M}}_{\Gamma },$$

then $\Phi \notin \ker {\rho }_n$.

The second form is essential in the four-strand argument. The proof does not always detect $\Phi$ directly; instead, it appends a test braid $\Gamma$ and shows that $\Phi$ changes the Moody polynomial of that test.

Mechanism of the Moody Test

The Moody polynomial is a compressed homological pairing. The Burau representation is the action on the relative homology of the cyclic cover, and the lifted intersection pairing records how a moved test arc pairs with the fixed measuring arc. If a braid $\Phi$ lies in the Burau kernel, then it acts trivially on this homology module. Consequently, every homological intersection invariant computed from a test arc must agree before and after applying $\Phi$.

This is why the strengthened obstruction is useful. If

$${\mathbb{M}}_{\Phi \Gamma }\ne {\mathbb{M}}_{\Gamma },$$

then $\Phi$ cannot act trivially on Burau homology: the auxiliary braid $\Gamma$ has made the invisible action visible. The proof is allowed to choose such a $\Gamma$ because kernel membership would force equality for every test. In this sense, the argument is a separating-family argument: the Moody polynomials produced by suitable arcs and point-pushes separate a nontrivial Brunnian braid from the identity.

Disk Sequences and Cancellation

The exponents $k_i$ are controlled by the disks between consecutive crossings. Between $q_i$ and $q_{i+1}$ along $\beta$, let ${\beta }_i$ be the red subarc and let ${\alpha }_i$ be the blue subarc with the same endpoints. Together they bound a disk

$${\Delta }_i.$$

If ${\Delta }_i$ contains $k$ punctures, define

$$W_i=\{\begin{array}{ll}k,& {\beta }_i\text{ is clockwise around }{\Delta }_i,\\ -k,& {\beta }_i\text{ is counterclockwise around }{\Delta }_i.\end{array}$$

The list

$$W_1,\dots ,W_{m-1}$$

is the winding number sequence, and

$${\Delta }_1,\dots ,{\Delta }_{m-1}$$

is the disk sequence.

Bigelow exponent rule

If $k_i$ is the exponent contributed by $q_i$, then

$$W_i=k_{i+1}-k_i.$$

Consequently,

$$k_i=k_j⟺\sum _{\ell =i}^{j-1}{W}_{\ell }=0.$$

Repeated powers of $t$ can occur only across a consecutive block of disks with total winding zero. The proof therefore becomes a question about whether such a block can contain an odd number of sign reversals.

A disk ${\Delta }_i$ is sign-changing if adjacent crossings have opposite signs,

$${ϵ}_i=-{ϵ}_{i+1},$$

and sign-preserving otherwise. The parity condition is

$${\Delta }_i\text{ is sign-changing}⟺|\mathcal{P}({\Delta }_i)|\text{ is odd},$$

where $\mathcal{P}({\Delta }_i)$ is the set of punctures inside ${\Delta }_i$.

Parity criterion

If a braid satisfies the parity condition, then its Moody polynomial has no cancellations. Hence any such braid with $\iota (\alpha ,({\beta }_{\ast }^3)\Phi )>0$ is not in the Burau kernel.

Indeed, if $k_i=k_j$, then

$$\sum _{\ell =i}^{j-1}{W}_{\ell }=0.$$

Since $W_{\ell }=\pm |\mathcal{P}({\Delta }_{\ell })|$, this zero sum has even parity. Under the parity condition, the odd-puncture disks are exactly the sign-changing disks. Thus there are an even number of sign changes between $q_i$ and $q_j$, so

$${ϵ}_i={ϵ}_j.$$

Equal powers of $t$ have equal signs and cannot cancel.

The mechanism can be summarized as follows. Equal exponents in the Laurent polynomial are detected by zero total winding. Opposite coefficients are detected by sign changes. The parity condition synchronizes these two pieces of information: any zero-winding block contains an even number of sign changes, so equal exponents have equal signs. The proof therefore converts cancellation in $\mathbb{Z}[t,t^{-1}]$ into a parity statement about punctures in disks.

The Three-Strand Case

For $B_3$, every disk in a disk sequence contains $1$, $2$, or $3$ punctures. Minimal position forces the following behavior.

Punctures inside ${\Delta }_i$	Effect on signs
$1$ or $3$	sign-changing
$2$	sign-preserving

This is exactly the parity condition. Therefore every $3$-strand braid satisfies the parity criterion, and Moody polynomial cancellation cannot hide a nontrivial kernel element. This recovers the Magnus-Peluso faithfulness theorem for ${\rho }_3$ by a topological argument.

The obstruction in $B_4$ is now precise. Four-puncture disks introduce sign behavior that is not determined solely by parity, so the three-strand argument has to be refined rather than copied.

Reduction to Brunnian Braids

For a marked point $p_i$, the Birman exact sequence identifies the point-pushing subgroup $K_i$ with a punctured-disk fundamental group:

$$1\to {\pi }_1(D_{n-1},p_i)\stackrel{\mathrm{Push}}{\to }\mathrm{Mod}(D_n,p_i)\stackrel{\mathrm{Forget}}{\to }\mathrm{Mod}(D_{n-1})\to 1.$$

A loop based at $p_i$ gives a braid by pushing the marked point around the loop (Farb & Margalit, 2012).

The Brunnian subgroup is

$${\mathrm{Brun}}_n=\bigcap _{i=1}^n{K}_i.$$

Equivalently, a Brunnian braid becomes trivial after forgetting any one strand.

Figure 2: A nontrivial Brunnian four-braid. Long's normal-subgroup reduction allows the proof to concentrate on this subgroup.

Long’s theorem says that ${\rho }_n$ is faithful on $B_n$ if it is faithful on any nontrivial noncentral normal subgroup (Long, 1986). Since ${\mathrm{Brun}}_4$ is normal and noncentral, the proof reduces to the following target.

Brunnian target

If ${\rho }_4$ is faithful on ${\mathrm{Brun}}_4$, then ${\rho }_4$ is faithful on $B_4$.

This reduction also supplies intersections. Nontrivial Brunnian four-braids are pseudo-Anosov, so their images of ${\beta }_{\ast }^3$ must intersect $\alpha$ nontrivially (Whittlesey, 2000). Thus the Moody polynomial has crossings to measure.

Proper Point-Pushing Products

Point-pushing creates the braids used in the proof, but a push can also create artificial bigons between $\alpha$ and the new red arc. Bigons are a problem because disk sequences should be computed only after $\alpha$ and $\beta$ are in minimal position.

The paper isolates the local configurations that create such bigons:

• generalized bigons between the push loop $\Gamma$ and $\alpha$,
• generalized $\beta$-to-$\alpha$ trigons,
• generalized rectangles.

These configurations are called bigon-forming polygons. A product $\Phi \Gamma$ of point-pushing braids is proper if $\Gamma$ can be represented in minimal position with $\alpha$ and $\beta =({\beta }_{\ast }^3)\Phi$, no bigon-forming polygon occurs, and the initial and final pieces of $\Gamma$ do not undo the innermost disk around the pushed point.

Proper products preserve minimal position

If $\Phi \Gamma$ is a proper product, then $(\beta )\Gamma$ and $\alpha$ are in minimal position.

The definition is technical, but its purpose is simple: it ensures that a disk sequence after a push is a genuine minimal-position disk sequence rather than an artifact of unreduced bigons.

Four-Strand Disk Analysis

Figure 3: The special final push-map Γ₁. Ending a proper product with this loop sharply limits the disk types that can appear.

In $D_4$, sign behavior is no longer determined only by the number of punctures in a disk. In full generality, a disk in a four-strand disk sequence is sign-preserving exactly for

$$\{p_1,p_3\},\{p_1,p_4\},\{p_2,p_3\},\{p_2,p_4\},\{p_1,p_3,p_4\},\{p_2,p_3,p_4\},$$

and sign-changing exactly for

$$\{p_3\},\{p_4\},\{p_3,p_4\},\{p_1,p_2,p_3\},\{p_1,p_2,p_4\},\{p_1,p_2,p_3,p_4\}.$$

The parity condition is therefore not automatic.

The authors choose the specific push-map ${\Gamma }_1\in K_4$ shown in Figure 3. If a point-pushing braid has a proper product factorization

$$\Phi =\Psi {\Gamma }_1,$$

then the disk sequence becomes much more constrained.

Behavior	Possible puncture sets after the ${\Gamma }_1$ restriction
sign-preserving	$\{p_2,p_3\}$
sign-changing	$\{p_3\}$, $\{p_4\}$, $\{p_1,p_2,p_3\}$, $\{p_1,p_2,p_3,p_4\}$

Figure 4 gives the finite geometric list behind this table.

Figure 4: The disk types allowed after the beta component is isotoped off Γ₁. All listed types satisfy the parity condition except the disk containing all four punctures.

The four-strand problem is thus compressed into one exceptional disk type:

$$\{p_1,p_2,p_3,p_4\}.$$

This disk is sign-changing but has even cardinality, so it violates the parity condition.

Five-Strand Stabilization Test

To repair the exceptional disk type, embed

$$f:B_4\hookrightarrow B_5$$

by adding a fifth puncture $p_5$. If

$$\Phi ={\Phi }^{\prime }{\Gamma }_1$$

is a proper product in $B_4$, the authors construct a push-map

$$\Gamma \in K_5$$

such that both $f(\Phi )\Gamma$ and $\Gamma$ satisfy the parity condition.

Figure 5 shows the construction. The point $q$ is chosen inside the innermost bad four-puncture disk. A path ${\gamma }_1$ goes from $p_5$ to $q$ without crossing the red arc. A return path ${\gamma }_2$ exits through the red side of each bad four-puncture disk and then returns to $p_5$. The loop

$$\Gamma ={\gamma }_2\circ {\gamma }_1$$

pushes $p_5$ so that every bad four-puncture disk becomes a five-puncture disk.

Figure 5: The five-strand correction. The gold path enters the innermost bad four-puncture disk, and the magenta path exits through the red side of each bad disk, forcing those disks to include the new puncture p₅.

The exceptional type changes as follows:

$$\{p_1,p_2,p_3,p_4\}⇝\{p_1,p_2,p_3,p_4,p_5\}.$$

A five-puncture disk is sign-changing and has odd cardinality, so it satisfies the parity condition.

The loop $\Gamma$ is also chosen so that

$$\iota (\alpha ,({\beta }_{\ast }^3)(f(\Phi )\Gamma ))\ne \iota (\alpha ,({\beta }_{\ast }^3)\Gamma ).$$

Since both resulting Moody polynomials have no cancellations, this inequality of intersection counts forces

$${\mathbb{M}}_{f(\Phi )\Gamma }\ne {\mathbb{M}}_{\Gamma }.$$

The Moody obstruction gives

$$f(\Phi )\notin \ker {\rho }_5.$$

By stabilization of kernel elements, this implies

$$\Phi \notin \ker {\rho }_4.$$

Stabilization as a Testing Device

The five-strand step should not be read as an appeal to faithfulness in $B_5$, which is false. It is a testing device. If a four-strand braid $\Phi$ were in $\ker {\rho }_4$, then its stabilization $f(\Phi )$ would lie in $\ker {\rho }_5$. In that case the strengthened Moody obstruction would force

$${\mathbb{M}}_{f(\Phi )\Gamma }={\mathbb{M}}_{\Gamma }$$

for every five-strand test braid $\Gamma$.

The constructed $\Gamma$ violates exactly this equality. The role of the fifth puncture is to repair the one disk type where the four-strand parity argument fails. After that repair, both Moody polynomials have no cancellations, and the inequality of geometric intersection numbers becomes an inequality of Laurent polynomials. Thus the contradiction occurs inside $B_5$, but the contradicted hypothesis is the original assumption $\Phi \in \ker {\rho }_4$.

Completion of the Proof

Let $\Phi$ be a nontrivial element of ${\mathrm{Brun}}_4$. Since kernel membership is invariant under conjugation, it is enough to rule out the Burau kernel for a convenient conjugate of $\Phi$. The point-pushing group $K_4$ is free, and the proof chooses a conjugate whose freely reduced word does not cancel against the fixed final factor ${\Gamma }_1$. After this conjugation, one obtains a proper product factorization

$$\Phi ={\Phi }^{\prime }{\Gamma }_1.$$

The five-strand stabilization test applies, so $\Phi \notin \ker {\rho }_4$. Hence ${\rho }_4$ is faithful on ${\mathrm{Brun}}_4$, and Long’s normal-subgroup reduction upgrades this to all of $B_4$.

Main theorem

The Burau representation

$${\rho }_4:B_4\to {\mathrm{GL}}_4(\mathbb{Z}[t,t^{-1}])$$

is faithful.

The Jones braid group representation contains the reduced Burau representation as a summand, so the theorem also implies faithfulness of the Jones representation for $B_4$ (Jones, 1987; Bharathram, Birman & Brendle, 2026).

Remaining Problems

The faithfulness problem is now settled, but two structural questions remain open:

• Describe the kernel of ${\rho }_n$ explicitly for $n\ge 5$.
• Characterize the image of the Burau representation inside ${\mathrm{GL}}_n(\mathbb{Z}[t,t^{-1}])$.

The new paper is therefore not only a four-strand faithfulness proof. It also gives a disk-sequence method for detecting when apparent cancellation in a braid representation is geometrically impossible.

See Also

• on nonadditivity of unknotting number: Another low-dimensional topology result proved through an explicit geometric certificate.
• on self-avoiding walks and the honeycomb connective constant: Another example where local planar cancellations are converted into a global algebraic obstruction.
• on the lower bound of Ramsey number: Shares the theme of reducing a global existence question to a carefully engineered combinatorial certificate.

References

🐻  Bharathram, V., Birman, J.S. & Brendle, T.E. 2026. The Burau representation of the braid group is faithful for n=4.

🐻  Bigelow, S. 1999. The Burau representation is not faithful for n=5. Geometry and Topology 3, 397–404.

🐻  Birman, J.S. & Brendle, T.E. 2005. Braids: A Survey. In Handbook of Knot Theory, pp. 19–103. , Elsevier.

🐻  Burau, W. 1935. Über Zopfgruppen und gleichsinnig verdrillte Verkettungen. Abhandlungen aus dem Mathematischen Seminar der Universität Hamburg 11(1), 179–186.

🐻  Farb, B. & Margalit, D. 2012. A Primer on Mapping Class Groups, Princeton University Press,p.

🐻  Jones, V.F.R. 1987. Hecke algebra representations of braid groups and link polynomials. Annals of Mathematics 126(2), 335–388.

🐻  Long, D.D. 1986. A note on the normal subgroups of mapping class groups. Mathematical Proceedings of the Cambridge Philosophical Society 99(1), 79–87.

🐻  Long, D.D. & Paton, M. 1993. The Burau representation is not faithful for n \geq 6. Topology 32(2), 439–447.

🐻  Magnus, W. & Peluso, A. 1969. On a theorem of V. I. Arnold. Communications on Pure and Applied Mathematics 22, 683–692.

🐻  Moody, J.A. 1991. The Burau representation of the braid group Bₙ is unfaithful for large n. Bulletin of the American Mathematical Society 25(2), 379–384.

🐻  Whittlesey, K. 2000. Normal all pseudo-Anosov subgroups of mapping class groups. Geometry and Topology 4, 293–307.