Generalized Pentagon Equations

Abstract

Drinfeld defined the Knizhinik–Zamolodchikov (KZ) associator $\Phi_{\rm KZ}$ by considering the regularized holonomy of the KZ connection along the droit chemin $[0,1]$. The KZ associator is a group-like element of the free associative algebra with two generators, and it satisfies the pentagon equation.

In this paper, we consider paths on $\mathbb{C}\backslash\{z_{1},\dots,z_{n}\}$ which start and end at tangential base points. These paths are not necessarily straight, and they may have a finite number of transversal self-intersections. We show that the regularized holonomy $H$ of the KZ connection associated to such a path satisfies a generalization of Drinfeld’s pentagon equation. In this equation, we encounter $H$, $\Phi_{\rm KZ}$, and new factors associated to self-intersections, to tangential base points, and to the rotation number of the path.

1 Introduction

The pentagon equation was first introduced in the theory of braided monoidal categories by Mac Lane [8]. In a braided monoidal category $\mathcal{C}$ with tensor product $\otimes$, braiding $b$ and associativity constraint $a$ one has the following commuting pentagon diagram:

$$\begin{tikzcd}$$ (1)

and two hexagon diagrams.

The Mac Lane Coherence Theorem 1.1 (originally in [8], we cite the form in [6]) explains relation between morphisms in braided monoidal categories and elements of the braid group.

Theorem 1.1 ([6, Theorem 2.9.2]).

Let $X_{1},\dots,X_{n}\in\mathcal{C}$, $P_{1},P_{2}$ be two parenthesized products of $X_{1},\dots,X_{n}$ (in any order order) possibly with insertions of the unit object $I$, and $f,g:P_{1}\to P_{2}$ be two isomorphisms obtained by composing braiding, associativity and unit isomorphism and their inverses possibly tensored with identity morphisms. If the underlying braids are isotopic, then $f=g$.

Let $\mathfrak{g}$ be a complex quadratic Lie algebra (that is, a finite dimensional Lie algebra which possesses an invariant scalar product). Drinfeld [2, 3] turned the category $U\mathfrak{g}-{\rm mod}_{\hbar}$ of its finite dimensional representations into a braided monoidal category using solutions of the Knizhnik-Zamolodchikov (KZ) equations.

In more detail, the KZ connection for $n$ points on the complex plane

$$A_{\rm KZ}=\frac{1}{2\pi i}\sum_{i<j}t_{i,j}\,d\log(z_{i}-z_{j}).$$ (2)

takes values in the Drinfeld-Kohno Lie algebra of infinitesimal braids $\mathfrak{t}_{n}$ with generators $t_{ij}=t_{ji}$ with $i,j\in\{1,\dots,n\}$ and the relations

$$[t_{i,j},t_{k,l}]=0,\hskip 8.5359pt[t_{i,j}+t_{i,k},t_{j,k}]=0$$ (3)

for distinct indices $i,j,k,l$.

The braiding constraints are given by $b_{i,j}=\exp(t_{i,j}/2)$, and the associativity constraint is defined by the regularized holonomy of the KZ equation for $n=3,z_{1}=0,z_{3}=1$ and $z=z_{2}$ moving on the straight segment (droit chemin) $[0,1]$. It carries the name of the KZ associator:

$$\Phi_{\rm KZ}(t_{12},t_{23})={\rm Hol}^{\rm reg}\left(\frac{t_{12}}{2\pi i}\,d\log(z)+\frac{t_{23}}{2\pi i}\,d\log(z-1),[0,1]\right).$$

The KZ associator satisfies the pentagon equation in $U(\mathfrak{t}_{4})$:

$$\Phi_{\rm{KZ}}(t_{23},t_{34})\Phi_{\rm{KZ}}(t_{1,23},t_{23,4})\Phi_{\rm{KZ}}(t_{12},t_{23})=\Phi_{\rm{KZ}}(t_{12},t_{2,34})\Phi_{\rm{KZ}}(t_{12,3},t_{3,4}),$$ (4)

where $t_{i,jk}=t_{ij}+t_{ik}$.

In this paper, we generalize equation (4) to the case of an arbitrary number of marked points on the complex plane, and to paths which are not necessarily straight and which may have self-intersections. More precisely, we will consider the KZ equation on $\Sigma=\mathbb{C}\backslash\{z_{1},\dots,z_{n}\}$. For a point $z$ moving on $\Sigma$ we associate the KZ connection

$$A_{z}=\frac{1}{2\pi i}\sum_{i=1}^{n}t_{i,z}\,d\log(z-z_{i})$$ (5)

taking values in $\mathfrak{t}_{n+1}$. Consider a path $\gamma$ which starts at a tangential base point $(z_{i},v_{i})$ and ends at another tangential base point $(z_{j},v_{j})$, where $v_{i},v_{j}$ are nonvanishing vectors at $z_{i}$ and $z_{j}$, respectively. Denote by $\rm{rot}(\gamma)$ the rotation number of $\gamma$ with respect to the blackboard framing. We assume that $\gamma$ has a finite number of transverse self-intersection points. To each self-intersection point, one can associate the intersection number $\varepsilon_{l}\in\{1,-1\},l=1,\dots,m$ indicating the orientation of the frame formed by two tangent vectors to the path at that point.

The main object of our study is the regularized holonomy

$$H={\rm Hol}^{\rm reg}(A_{z},\gamma)$$

associated to the path $\gamma$. By adding one more point $w$ to the configuration, we can introduce several versions of the holonomy $H$ needed in the statement of the generalized pentagon equation:

$$H_{z}:=\mathrm{Hol}^{\mathrm{reg}}\left(\frac{t_{z,wj}}{2\pi i}\,d\log(z-z_{j})+\sum_{k\neq j}\frac{t_{k,z}}{2\pi i}\,d\log(z-z_{k}),\gamma\right),$$ (6a)

$$H_{w}:=\mathrm{Hol}^{\mathrm{reg}}\left(\frac{t_{iz,w}}{2\pi i}\,d\log(w-z_{i})+\sum_{k\neq i}\frac{t_{k,w}}{2\pi i}\,d\log(w-z_{k}),\gamma\right),$$ (6b)

$$H_{zw}:=\mathrm{Hol}^{\mathrm{reg}}\left(\sum_{k=1}^{n}\frac{t_{k,zw}}{2\pi i}\,d\log(z-z_{l}),\gamma\right).$$ (6c)

Note that all these versions are obtained from $H$ by various substitutions, namely if we write

$$H=H(t_{1,z},\dots,t_{n,z})$$

we have

$$H_{z}=H(t_{1,z},\dots,t_{(j-1),z},t_{jw,z},t_{(j+1),z},\dots,t_{n,z}),$$ (7a)

$$H_{w}=H(t_{1,w},\dots,t_{(i-1),w},t_{iz,w},t_{(i+1),w},\dots,t_{n,w}),$$ (7b)

$$H_{zw}=H(t_{1,zw},\dots,t_{n,zw}).$$ (7c)

Equipped with this notation, we can now state (a somewhat simplified version of) the generalized pentagon equation:

Theorem 1.2.

There exist elements $C_{l}\in\exp(\mathfrak{t}_{n+2})$ for $l=1,\dots,m$ such that the following identity holds in $U(\mathfrak{t}_{n+2})$,

$$\begin{array}[]{ll}\Phi_{\rm{KZ}}(t_{zw},t_{wj})\left(H_{zw}|v_{j}/v_{i}|^{t_{zw}/2\pi i}e^{\text{rot}(\gamma)t_{zw}}\right)\Phi_{\rm{KZ}}(t_{iz},t_{zw})&=\\ H_{z}\left(\prod_{l=1}^{m}C_{l}^{-1}e^{-\varepsilon_{l}t_{zw}}C_{l}\right)H_{w}.\end{array}$$ (8)

For the case of $n=2,z_{1}=0,z_{2}=1$ and $\gamma=[0,1]$, we recover Drinfeld’s pentagon equation (4). In the general case, equation (8) contains the holonomy $H$, the KZ associator $\Phi_{\rm KZ}$, as well as new terms $|v_{j}/v_{i}|^{t_{zw}/2\pi i}$ and $e^{\text{rot}(\gamma)t_{z,w}}$ depending on the tangential base points and on the rotation number of the path $\gamma$, as well as $(C_{l}^{-1}e^{-\varepsilon_{l}t_{z,w}}C_{l})$ which correspond to self-intersections

$$\gamma(t_{l})=\gamma(s_{l}),\quad 0\leq t_{l}<s_{l}\leq 1$$

of $\gamma$. We obtain the following partial information on $C_{l}$.

Theorem 1.3 (see Theorem 3.3).

We have,

$$C_{l}={\rm Hol}^{\rm reg}(A_{z},\gamma_{[0,t_{l}]}){\rm Hol}^{\rm reg}(A_{w},\gamma_{[1,s_{l}]})+O(t_{zw})$$ (9)

Note that Theorem 1.2 opens the following new perspective on the topic: one can use the full KZ connection for $n+2$ points to bring the configuration $\{z_{1},\dots,z_{n}\}$ to a tangential base point. Then, by Mac Lane Coherence Theorem equation (8) should reduce to some equality of braids (and eventually to a number of pentagon and hexagon constraints). This new interpretation should include combinatorial expressions for the rotation number, indices of intersection points and explicit presentations of the elements $C_{l}$. It may possibly be related to the work of G. Massuyeau [9] in which a similar geometric setup was considered for applications to the Goldman-Turaev Lie bialgebra. We hope to return to this interesting point of view in future work.

In a forthcoming work, we plan to apply Theorem 1.2 for computing van den Bergh double brackets of universal regularized holonomies ${\rm Hol}^{\rm reg}(A_{z},\gamma)$ (with values in a free associative algebra), and of Poisson brackets of the corresponding regularized holonomies for the Lie algebra $\mathfrak{g}={\rm gl}(N,\mathbb{C})$.

The structure of the paper is as follows. In Section 2.1, we discuss local solutions of the KZ equation near regular and tangential base points. In Section 2.2, we define regularized holonomies, and in Section 2.3 we recall their basic property under composition of paths. In Section 3.1, we follow Drinfeld and construct local solutions of the KZ equation near tangential base points. In Section 3.2, we analyse solutions of KZ equations self-intersection points of the path and prove the generalized Pentagon equation. Finally in subsection 3.3, we discuss the property of the terms $C_{l}$ in the equation.

Acknowledgements. We are grateful to D. Bar-Natan, F. Brown and B. Enriquez for interesting discussions. Reserach of A.A. and M.R. was supported in part by the grants number 208235 and 200400 and by the National Center for Competence in Research (NCCR) SwissMAP of the Swiss National Science Foundation. Research of A.A. was supported in part by the award of the Simons Foundation to the Hamilton Mathematics Institute of the Trinity College Dublin under the program “Targeted Grants to Institutes”.

2 Regularized holonomy

In this Section, we define regularized holonomies of the KZ equation and recall their basic properties.

2.1 Local solutions of the KZ equation

Let $\mathfrak{t}_{n+1}$ be the Drinfeld-Kohno Lie algebra in $n+1$ strands. Its generators are $t_{i,j}$ for $i,j\in\{1,\dots,n,z\}$ and $z$ stands for $n+1$, and the relations are given by (3). Recall that the generators $t_{i,z}$ for $i=1,\dots,n$ span a free Lie subalgebra $\mathfrak{f}_{n}$ of $\mathfrak{t}_{n+1}$.

Remark 2.1.

Notice that the symbol $z$ denotes both the last strand and the (coordinate of the) point moving on the surface.

For the surface $\Sigma=\mathbb{C}\backslash\{z_{1},\dots,z_{n}\}$, we consider the differential equation

$$d\Psi=A_{z}\Psi,$$ (10)

where $A_{z}$ is the flat connection given by (5) and solutions $\Psi(z)\in U\mathfrak{f}_{n}\cong\mathbb{C}\langle\langle t_{1,z},\dots,t_{n,z}\rangle\rangle$ take values in the (completion of) the universal envelo** algebra of $\mathfrak{f}_{n}$.

We will associate local solutions of the equation (10) to regular points $p\in\mathbb{C}\backslash\{z_{1},\dots,z_{n}\}$ and to tangential base points $(z_{i},v_{i})$, where $0\neq v_{i}\in\mathbb{C}$ is a tangent vector at $z_{i}$. To a regular point $p$, we associate the unique local solution $\Psi_{p}(z)$ which satisfies the initial condition $\Psi_{p}(p)=1$.

For a tangential base point $(z_{i},v_{i})$, let $B_{\varepsilon}(z_{i})$ be a small open disk of radius $\varepsilon$ around $z_{i}$, and let $l_{i}=\{z_{i}-sv_{i};s\in\mathbb{R}_{\geq 0}\}$ be a ray emanating from $z_{i}$ in the direction opposite to $v_{i}$. We denote by $D_{i}=B_{\varepsilon}(z_{i})\backslash l_{i}$ the simply connected domain obtained by deleting the ray $l_{i}$ from the disk $B_{\epsilon}(z_{i})$.

Lemma 2.2.

For $\varepsilon$ sufficiently small, there is a unique solution $\Psi_{z_{i},v_{i}}(z)$ of the differential equation (10) on the domain $D_{i}$ such that

$$\Psi_{z_{i},v_{i}}(z)=f\left(\frac{z-z_{i}}{v_{i}}\right)\exp\left(\frac{t_{i,z}}{2\pi i}\log\left(\frac{z-z_{i}}{v_{i}}\right)\right),$$ (11)

where $f$ is an analytic function on $B_{\varepsilon}(0)$ with $f(0)=1$.

Remark 2.3.

Solutions $\Psi_{z_{i},v_{i}}(z)$ have the following asymptotic behaviour for $z\to z_{i}$ in the domain $D_{i}$:

$$\Psi_{z_{i},v_{i}}(z)\sim_{z\to z_{i}}\left(1+O\left(\frac{z-z_{i}}{v_{i}}\right)\right)\exp\left(\frac{t_{i,z}}{2\pi i}\log\left(\frac{z-z_{i}}{v_{i}}\right)\right).$$

This can also be used as a definition of these solutions.

Proof.

For convenience of the reader, we recall the proof. Let $w=\frac{z-z_{i}}{v_{i}}$. By plugging the expression $f(w)w^{t_{iz}/2\pi i}$ into equation (10), we get

$$v_{i}^{-1}\left(f^{\prime}(w)+f(w)\frac{t_{i,z}}{2\pi iw}\right)=v_{i}^{-1}\frac{t_{i,z}}{2\pi iw}f(w)+\sum_{j\neq i}\frac{t_{j,z}}{2\pi i(z-z_{j})}f(w).$$ (12)

This yields

$$f^{\prime}(w)-\frac{1}{w}\left[\frac{t_{i,z}}{2\pi i},f(w)\right]=g(w),$$

where $g(w)=\sum_{k}g_{k}w^{k}$ is an analytic function near zero.

We are looking for a solution in the form $f(w)=\sum_{k}f_{k}w^{k}$ with $f_{0}=1$. Since $[t_{i,z},f_{0}]=[t_{i,z},1]=0$, the expression $\frac{1}{w}\left[\frac{t_{i,z}}{2\pi i},f(w)\right]$ is actually regular at $w=0$. Then, we find the coefficients $f_{k}$ for $k\geq 1$:

$$f_{k}=(k-{\rm ad}_{t_{i,z}/2\pi i})^{-1}g_{k-1}.$$

This implies that the convergence radius of the power series $f$ is greater or equal to the one of the power series $g$, and hence $f$ defines an analytic function near the origin, as required. ∎

Lemma 2.4.

Local solutions $\Psi_{p}(z)$ and $\Psi_{z_{i},v_{i}}(z)$ are group like elements of the (completed) universal envelo** algebra $U\mathfrak{f}_{n}$.

Proof.

For solutions at regular points, observe that the KZ connection is linear in generators of the Lie algebra $\mathfrak{t}_{n+1}$. Hence, $\Delta(A_{z})=A_{z}\otimes 1+1\otimes A_{z}$, where $\Delta$ is the standard coproduct. This implies that $\Delta(\Psi_{p})$ and $\Psi_{p}\otimes\Psi_{p}$ satisfy the same differential equation

$$d\hat{\Psi}(z)=(A_{z}\otimes 1+1\otimes A_{z})\hat{\Psi}(z)$$

with initial condition $\hat{\Psi}(p)=1\otimes 1$. Then, by the uniqueness property of solutions of first order ODEs we obtain $\Delta(\Psi_{p})=\Psi_{p}\otimes\Psi_{p}$, as required.

Similarly, for a local solutions $\Psi_{z_{i},v_{i}}$ we have

$$\Psi_{z_{i},v_{i}}(z)=f\left(\frac{z-z_{i}}{v_{i}}\right)\exp\left(\frac{t_{i,z}}{2\pi i}\log\left(\frac{z-z_{i}}{v_{i}}\right)\right),$$ (13)

and we observe that $\Delta(f(w))$ and $f(w)\otimes f(w)$ satisfy the same differential equation

$$\hat{f}^{\prime}(w)-\frac{1}{w}\left[\frac{t_{i,z}\otimes 1+1\otimes t_{i,z}}{2\pi i},\hat{f}(w)\right]=\sum_{j\neq i}\frac{v_{i}(t_{j,z}\otimes 1+1\otimes t_{j,z})}{2\pi i(z-z_{j})}\hat{f}(w).$$ (14)

An argument similar to the proof of existence of solutions implies uniqueness of an analytic solution of this equation with the initial value $\hat{f}(0)=1$. Hence, we can conclude as above.

∎

2.2 Holonomy maps

Let $\gamma:[0,1]\to\mathbb{C}$ be a smooth path with $\dot{\gamma}=\tfrac{d\gamma}{dt}\neq 0$ which may start and end at regular or marked points but with $\gamma(t)\in\Sigma=\mathbb{C}\backslash\{z_{1},\dots,z_{n}\}$ for all $0<t<1$. In case when $\gamma$ starts or ends at marked points (or both), we require that near that point $\gamma(t)$ be linear in $t$:

$$\begin{array}[]{ll}\gamma(t)=z_{i}+v_{i}t&{\rm for}\,\,0<t<\varepsilon;\\ \gamma(t)=z_{j}+v_{j}(1-t)&{\rm for}\,\,0<1-t<\varepsilon.\end{array}$$

Here $(z_{i},v_{i})$ is the tangential starting point of $\gamma$, $(z_{j},v_{j})$ is the tangential end point of $\gamma$, and $\varepsilon$ is small enough. The path $\gamma(t)$ may have a finite number of transverse self-intersection points, $\gamma(s_{l})=\gamma(t_{l})=a_{l}$ for $l=1,\dots,m$. Then, we also require that $\gamma(t)$ be a linear function near the points $s_{l}$ and $t_{l}$.

To each path $\gamma(t)$, we associate two local solutions $\Psi_{p}(z)$ and $\Psi_{q}(z)$ corresponding to its end points (regular or tangential). For regular end points, we have $p=\gamma(0),q=\gamma(1)$, and for tangential end points we have $(p=(\gamma(0),\dot{\gamma}(0)),q=(\gamma(1),-\dot{\gamma}(1))$. The solutions $\Psi_{p}(z)$ and $\Psi_{q}(z)$ can be analytically continued to a strip around the path $\gamma$. More precisely, we extend $\gamma\colon(0,1)\to\Sigma$ to a local diffeomorphism $\tilde{\gamma}\colon(0,1)\times(-\varepsilon,\varepsilon)\to\Sigma$. Note that the local solution $\Psi_{p}(z)$ defines a flat section of the pulled back connection $\tilde{\gamma}^{*}A_{z}$ on a neighborhood of $(0,0)\in[0,1)\times(-\varepsilon,\varepsilon)$ which can then be analytically continued to a solution on the entire strip. The same applies to the local solution $\Psi_{q}(z)$. By abuse of notation, we denote analytic continuations of local solutions by the same symbols. Note that if the path $\gamma$ has self-intersections, the self-intersection point has two pre-images in the strip $(0,1)\times(-\varepsilon,\varepsilon)$, and the values of analytically continued solutions at these points are different, in general.

Definition 2.5.

For a path $\gamma$, the (regularized) holonomy of $A_{z}$ along $\gamma$ is given by

$$\mathrm{Hol}^{\mathrm{reg}}(A_{z},\gamma):=\Psi_{q}^{-1}(z)\Psi_{p}(z)\in U\mathfrak{f}_{n},$$ (15)

where $z$ is any point in the strip around $\gamma$.

When both end points of $\gamma$ are regular, this definition coincides with the standard definition of a parallel transport of a connection (on a trivial bundle). In particular, in this case the holonomy is invariant under homotopies which preserve end points of the path. Similarly, for tangential end points the regularized holonomy is invariant under homotopies which preserve $(\gamma(0),\dot{\gamma}(0))$ for the starting point and $(\gamma(1),-\dot{\gamma}(1))$ for the end point.

Example 2.6.

Consider $\Sigma=\mathbb{C}\setminus\{0,1\}$, tangential base points $(0,1)$ and $(1,-1)$, and $\gamma(t)=t$ the straight segment (le droit chemin) connecting $0$ to $1$. We have two local solutions with asymptotic behavior

$$\Psi_{0,1}(z)\sim z^{\frac{t_{1,z}}{2\pi i}},\hskip 8.5359pt\Psi_{1,-1}(z)\sim(1-z)^{\frac{t_{2,z}}{2\pi i}}.$$

In this case, $\mathrm{Hol}^{\mathrm{reg}}(A_{z},[0,1])$ agrees with Drinfeld’s definition of the KZ associator

$$\Phi_{\rm KZ}(t_{1,z},t_{2,z})=\Psi_{1,-1}(z)^{-1}\Psi_{0,1}(z).$$

Lemma 2.7.

(Regularized) holonomies are group-like elements of the (completed) universal envelo** algebra $U\mathfrak{f}_{n}$.

Proof.

By Lemma 2.4, local solutions are group like. This property is preserved by analytic continuations. Since holonomies are given by ratios of two analytically continued local solutions, they are group like as well. ∎

2.3 Composition of paths and rotation number

Two paths $\gamma_{1}$ and $\gamma_{2}$ with end points (regular or tangential) $(p_{1},q_{1})$ and $(p_{2},q_{2})$ are called composable if $p_{2}=q_{1}$. In that case, we can define a composition of paths $\gamma_{2}\gamma_{1}$.

If the common end point $p_{2}=q_{1}$ is a regular point, the composition is defined up to homotopy by concatenation of paths (this allows to make the resulting path smooth). If the common end point is a marked point, the composition is defined modulo regular homotopy (that is, the first Reidemeister move is not allowed). In that case, there are two natural ways to define the composition. One considers a neighborhood of the marked point (see Fig. 1), and adds a clockwise or a counter-clockwise half-turn with respect to the blackboard orientation of $\mathbb{C}$ (see Fig. 2). In this paper, we chose to use the clockwise convention.

Figure 1: small region near the marked point

Figure 2: two ways of composing curves

(Regularized) holonomies are multiplicative under composition of paths:

Theorem 2.8.

For composable paths $\gamma_{1}$ and $\gamma_{2}$, we have a product formula for the corresponding (regularized) holonomies:

$${\rm Hol}^{\rm reg}(A,\gamma_{2}\gamma_{1})=\mathrm{Hol}^{\mathrm{reg}}(A,\gamma_{2})\mathrm{Hol}^{\mathrm{reg}}(A,\gamma_{1}).$$ (16)

Proof.

We have,

$$\mathrm{Hol}^{\mathrm{reg}}(A_{z},\gamma_{1}):=\Psi_{q_{1}}^{-1}(z)\Psi_{p_{1}}(z),\hskip 8.5359pt\mathrm{Hol}^{\mathrm{reg}}(A_{z},\gamma_{2}):=\Psi_{q_{2}}^{-1}(z)\Psi_{p_{2}}(z).$$

By choosing the same point $z$ in both equations (e.g. in the small neighborhood of $\gamma_{1}(1)=\gamma_{2}(0)$), we obtain

$$\mathrm{Hol}^{\mathrm{reg}}(A_{z},\gamma_{2})\mathrm{Hol}^{\mathrm{reg}}(A_{z},\gamma_{1})=\Psi_{q_{2}}^{-1}(z)\Psi_{p_{1}}(z)={\rm Hol}^{\rm reg}(A,\gamma_{2}\gamma_{1}),$$

as required. ∎

For future use, we will need a notion of rotation number of a path. It is only defined for paths up to regular homotopy. That’s why we will be using it for paths $\gamma$ with both end points being tangential end points. Hence, we have

$$(\gamma(0),\dot{\gamma}(0))=(z_{i},v_{i}),\hskip 8.5359pt(\gamma(1),\dot{\gamma}(1))=(z_{j},-v_{j})$$

for some marked points $z_{i},z_{j}$ and $v_{i}=\rho_{i}e^{i\varphi_{i}},v_{j}=\rho_{j}e^{i\varphi_{j}}$. We compute,

$$\int_{0}^{1}d\log(\dot{\gamma}(t))=\log\ \left|\frac{v_{j}}{v_{i}}\right|+2\pi i\ \rm{rot}_{\gamma},$$ (17)

where the imaginary part of this expression is the rotation number of $\gamma$. In particular, we have

$$\rm{rot}_{\gamma}\in\mathbb{Z}+\frac{1}{2}+\frac{1}{2\pi}(\varphi_{j}-\varphi_{i}).$$

Note that ${\rm rot}_{\gamma}\in\mathbb{R}$, and that one full turn in the anticlockwise direction corresponding to ${\rm rot}_{\gamma}=1$. For the composition of paths, we obtain

$${\rm rot}_{\gamma_{2}\gamma_{1}}={\rm rot}_{\gamma_{1}}+{\rm rot}_{\gamma_{2}}-\frac{1}{2},$$

where $-1/2$ on the right hand side comes from the extra clockwise half-turn in the definition of the path composition $\gamma_{2}\gamma_{1}$.

Note that one can consistently choose $v=1$ in all tangential base points. In that case, all rotation numbers of paths are half-integers: ${\rm rot}_{\gamma}\in\mathbb{Z}+1/2$. Furthermore, the regular fundamental group $\pi^{\rm reg}_{1}(\Sigma,(z_{i},v_{i}))$ surjects onto the ordinary fundamental group $\pi_{1}(\Sigma,z_{i})$.

3 Proof of the generalized pentagon equation

In this Section, we prove Theorem 1.2. We adapt Drinfeld’s original proof to the case of many marked points and to paths which may have self-intersections.

3.1 Asymptotic regions and local solutions

We consider the KZ connection for two moving points $z$ and $w$ in $\Sigma=\mathbb{C}\setminus\{z_{1},\dots,z_{n}\}$ with values in $\mathfrak{t}_{n+2}$:

$$A_{z,w}=\frac{1}{2\pi i}\left(\sum_{k=1}^{n}(t_{k,z}d\log(z-z_{k})+t_{k,w}d\log(w-z_{k}))+t_{z,w}d\log(z-w)\right),$$ (18)

and the corresponding KZ equation

$$d\Psi=A_{z,w}\Psi.$$ (19)

Let $\gamma:(0,1)\to\Sigma=\mathbb{C}\backslash\{z_{1},\dots,z_{n}\}$ be a path with a tangential starting point $(z_{i},v_{i})$, a tangential end point $(z_{j},v_{j})$, and a finite number of transverse self-intersections $a_{l}=\gamma(s_{l})=\gamma(t_{l})$ for $l=1,\dots,m$. As before, we assume that the path $\gamma$ is linear near end points and near self-intersection points.

The triangle $\{(t,s)|0\leq t\leq s\leq 1\}\subset\mathbb{R}^{2}$ in the $t-s$ plane is mapped to the set of ordered pairs of points on $\gamma$:

$$\gamma^{(2)}:(t,s)\mapsto(z=\gamma(t),w=\gamma(s)).$$

Following Drinfeld’s idea (see [3], and also [7]), we evaluate regularized holonomies on different parts of the closed path shown on Fig. 3. Since this path is contractible, the product of regularized holonomies is equal to 1, and this identity will give rise to the generalized pentagon equation.

Figure 3: $\Psi_{1},\dots,\Psi_{5}$ are 5 solutions of the KZ equation and $\bullet_{1},\bullet_{2}$ are two self intersection points of the curve in the (t,s) plane

In order to define regularized holonomies, we consider five asymptotic regions of the KZ equation associated to tangential base points:

$$1)\hskip 8.5359pt((z_{i}z)w)z_{j}\hskip 14.22636pt0<t\ll s\ll 1;$$

$$2)\hskip 8.5359pt(z_{i}(zw))z_{j}\hskip 14.22636pt0<s-t\ll s\ll 1;$$

$$3)\hskip 8.5359ptz_{i}((zw)z_{j})\hskip 14.22636pt0<s-t\ll 1-t\ll 1;$$ (20)

$$4)\hskip 8.5359ptz_{i}(z(wz_{j}))\hskip 14.22636pt0<1-s\ll 1-t\ll 1;$$

$$5)\hskip 8.5359pt(z_{i}z)(wz_{j})\hskip 14.22636pt0<t\ll 1,0<1-s\ll 1,$$

Here paranthesized expressions of the type $((z_{i}z)w)z_{j}$ label tangential base points and the corresponding asymptotic regions.

Consider the first asymptotic region. It is convenient to introduce the new coordinates $(u,v)$ by

$$u:=\frac{z-z_{i}}{w-z_{i}},\quad v:=\frac{w-z_{i}}{v_{i}}.$$

Let

$$D_{\varepsilon}:=\{(u,v)\in B_{\varepsilon}(0,0)\mid u\notin\mathbb{R}_{<0},v\notin v_{i}\mathbb{R}_{<0}\}$$

be a small polydisc around $(0,0)$ in $(u,v)$ coordinates with two branch cuts removed, so that $\log(u)$ and $\log(v)$ (taking real values on the rays opposite to branch cuts) are analytic functions on $D_{\varepsilon}$. Note that the first asymptotic region $0<t\ll s\ll 1$ maps to $D_{\varepsilon}$ by the map $\gamma^{(2)}$. Recall the following fact:

Lemma 3.1.

For $\varepsilon$ sufficiently small, there is a unique solution $\Psi_{1}$ of the KZ equation on $D_{\varepsilon}$ such that

$$\Psi_{1}(z,w)=f(u,v)\left(\frac{z-z_{i}}{v_{i}}\right)^{\frac{t_{iz}}{2\pi i}}\left(\frac{w-z_{i}}{v_{i}}\right)^{\frac{t_{iw}+t_{zw}}{2\pi i}},$$

where $f(u,v)$ is an analytic function with $f(0,0)=1$.

Proof.

See Section 2 in [3]. ∎

Remark 3.2.

We use the following shorthand notation for asymptotic behavior of the function $\Psi_{1}(z,w)$:

$$\Psi_{1}(z,w)\sim_{((z_{i}z)w)z_{j}}\left(\frac{z-z_{i}}{v_{i}}\right)^{\frac{t_{iz}}{2\pi i}}\left(\frac{w-z_{i}}{v_{i}}\right)^{\frac{t_{iw}+t_{zw}}{2\pi i}}.$$ (21)

Similarly, in the other four asymptotic regions we get the following solutions:

$$\Psi_{2}(z,w)\sim_{\left(z_{i}\left(zw\right)\right)z_{j}}\left(\frac{w-z}{v_{i}}\right)^{\frac{t_{zw}}{2\pi i}}\,\left(\frac{w-z_{i}}{v_{i}}\right)^{\frac{t_{iz}+t_{iw}}{2\pi i}},$$ (22)

$$\Psi_{3}(z,w)\sim_{z_{i}\left(\left(zw\right)z_{j}\right)}\left(\frac{w-z}{-v_{j}}\right)^{\frac{t_{zw}}{2\pi i}}\,\left(\frac{z-z_{j}}{v_{j}}\right)^{\frac{t_{zj}+t_{wj}}{2\pi i}},$$ (23)

$$\Psi_{4}(z,w)\sim_{z_{i}\left(z\left(wz_{j}\right)\right)}\left(\frac{w-z_{j}}{v_{j}}\right)^{\frac{t_{wj}}{2\pi i}}\left(\frac{z-z_{j}}{v_{j}}\right)^{\frac{t_{zw}+t_{zj}}{2\pi i}},$$ (24)

$$\Psi_{5}(z,w)\sim_{(z_{i}z)(wz_{j})}\left(\frac{z-z_{i}}{v_{i}}\right)^{\frac{t_{iz}}{2\pi i}}\left(\frac{w-z_{j}}{v_{j}}\right)^{\frac{t_{wj}}{2\pi i}}.$$ (25)

Note that in the expressions for $\Psi_{2}$ and $\Psi_{3}$ we are using tangential base points $(0,v_{i})$ and $(0,v_{j})$ for the $(zw)$ configuration.

3.2 Regularized holonomies

In this Section, we compute regularized holonomies for different parts of the closed path of Fig. 3. More precisely, we extend the asymptotic solutions $\Psi_{1},\dots,\Psi_{4}$ from above to the simply connected domain shown in Fig 3. For $\Psi_{5}$ we obtain solutions

$$\Psi_{5,l},\quad l=1,\dots,m+1,$$

where $\Psi_{5,l}$ coincides with the solution $\Psi_{5}$ in the region $\mathrm{V}_{l}$. Moreover, for each intersection point $l=1,\dots,m$ we will define two solutions $\Psi_{a_{l},u},$ and $\Psi_{a_{l},d}$ related by

$$\Psi_{a_{l},d}=\Psi_{a_{l},u}\exp(-\epsilon_{l}t_{z,w}).$$

With that the generalized pentagon equation will follow by identifying the terms in

$$\underbrace{\Psi_{4}^{-1}\Psi_{3}}_{\Phi_{\rm{KZ}}\atop\text{see \eqref{eq:Phi2}}}\underbrace{\Psi_{3}^{-1}\Psi_{2}}_{H_{zw}\left|\frac{v_{j}}{v_{i}}\right|^{\frac{t_{zw}}{2\pi}}e^{t_{zw}{\rm rot}_{\gamma}}\atop\text{see \eqref{eq:H_zw}}}\underbrace{\Psi_{2}^{-1}\Psi_{1}}_{\Phi_{\rm{KZ}}\atop\text{see \eqref{eq:Phi1}}}=\\ =\underbrace{\Psi_{4}^{-1}\Psi_{5,m+1}}_{H_{z}\atop\text{see \eqref{eq:H_z}}}\prod_{l=1}^{m}\left(\underbrace{\Psi_{5,l+1}^{-1}\Psi_{a_{l},u}}_{C_{l}^{-1}}\underbrace{\Psi_{a_{l},u}^{-1}\Psi_{a_{l},d}}_{e^{-\epsilon_{i}t_{zw}}\atop\text{see \eqref{eq:intterms}}}\underbrace{\Psi_{a_{l},d}^{-1}\Psi_{5,l}}_{C_{l}}\right)\underbrace{\Psi_{5,1}^{-1}\Psi_{1}}_{H_{w}\atop\text{see \eqref{eq:H_w}}}.$$

We are carrying out these identificatons in the following sections. The proof of $H_{z},H_{w},H_{zw}$ and $\Phi_{\rm{KZ}}$ contributions are similar to Section 2 of Drinfeld [3] and Lemma XIX.8.2 in [7]

3.2.1 Self-intersection points

In this Section, we consider solutions of (19) near self-intersection points. Suppose that the path $\gamma$ has $m$ transverse self-intersection points. We order these points in the counter-clockwise direction by their positions on the $s-t$ plane, see Fig. 3. If two self-intersection points are located on the same ray, we perturb the path by a small regular homotopy. For each self-intersection point, we denote by $\varepsilon_{l}$ the local intersection number and by $a_{l}=\gamma(s_{l})=\gamma(t_{l})$ the position of the intersection point, where $s_{l}>t_{l}$, $l=1,\dots,m$. By perturbing the curve $\gamma$ by a regular homotopy if needed, we can assume that $\gamma$ is a linear function for $t$ near $t_{l}$ and $s_{l}$ for $l=1,\dots,m$.

We denote by $L_{l}$ the line in the $z-w$ plane corresponding to the straight line

$$s=\frac{s_{l}-1}{t_{l}}t+1$$ (26)

in the $t-s$ plane which connects the points $(0,1)$ and $(t_{l},s_{l})$. Since $w-z=\gamma(s)-\gamma(t)$, it is convenient to introduce notation

$$u_{l}=\frac{d}{dt}\ \left(\gamma\left(\frac{s_{l}-1}{t_{l}}t+1\right)-\gamma(t)\right)|_{t=t_{l}}.$$ (27)

By assumptions, $\gamma(t)$ is linear in $t$ near $t_{l}$ and $s_{l}$ which implies

$$((z-w)\circ L_{l})(t)=a_{l}+u_{l}(t-t_{l}),\quad\text{for}\quad|t-t_{l}|\quad\text{sufficiently small}.$$ (28)

We denote by $B_{\varepsilon,a_{l}}(z-a_{l},w-z)$ the polydisk centered at $(0,0)$ with radius $\varepsilon$ and let $l_{a_{l}}=\{(z-a_{l},w-z);w-z=\gamma(s)-\gamma(t),s=\frac{s_{l}-1}{t_{l}}t+1,0\leq t\leq t_{l}\}$ denote a branch cut of the function $\log(w-z)$, and we denote

$$D_{\varepsilon,a_{l}}:=B_{\varepsilon,a_{l}}(z-a_{l},w-a_{l})\setminus l_{a_{l}}\cap B_{\varepsilon,a_{l}}(z-a_{l},w-a_{l}).$$

Similar to Theorem 4.6 in [1] (see also [4]), there is a unique solution $\Psi_{a_{l},u}(z,w)$ of the KZ equation on $D_{\varepsilon,a_{l}}$ such that

$$\Psi_{a_{l},u}(z,w)=f(z-a_{l},w-z)\exp\left(\log\left(\frac{w-z}{-u_{l}}\right)\frac{t_{zw}}{2\pi i}+\epsilon_{l}\frac{t_{zw}}{2}\right)$$

where $f$ is analytic on $B_{\varepsilon,a_{l}}$ and $f(0,0)=1$, where

$$\epsilon_{l}=\mathrm{sign}\left(\mathrm{Im}\left(\frac{\dot{\gamma}(s_{l})}{\dot{\gamma}(t_{l})}\right)\right)$$ (29)

is the sign of the intersection.

The normalization is chosen such that it satisfies the following property. If we extend $\Psi_{a_{l},u}$ continuously over the line $L_{l}$ from above, then on the line we have

$$\Psi_{a_{l},u}\sim(t-t_{l})^{t_{zw}},$$

Similarly, we can define a solution $\Psi_{a_{l},d}$ that has the desired asymptotics when we extend over $L_{l}$ from below, namely as analytic extension of

$$\Psi_{a_{l},d}(z,w)=f(z-a_{l},w-z)\exp\left(\log\left(\frac{w-z}{-u_{l}}\right)\frac{t_{zw}}{2\pi i}-\epsilon_{l}\frac{t_{zw}}{2}\right),$$

so that we have

$$\Psi_{a_{l},d}(z,w)=\Psi_{a_{l},u}(z,w)\,e^{-\varepsilon_{l}t_{zw}},$$

as well as

$$\lim_{(z,w)\mathrel{\downarrow}{(z_{0},w_{0})}}\Psi_{a_{l},u}(z,w)=\lim_{(z,w)\mathrel{\uparrow}{(z_{0},w_{0})}}\Psi_{a_{l},d}(z,w).$$

where $(z_{0},w_{0})$ is any point on the line $L_{l}$ and the limits are approaching the line from above and below, respectively. Recall that $\Psi_{5,l}$ and $\Psi_{5,l+1}$ are defined by the condition that they coincide with the asymptotic solution at $(t,s)=(0,1)$ in the regions $\mathrm{V}_{l}$ and $\mathrm{V}_{l+1}$, respectively. Thus we have

$$\lim_{(z,w)\mathrel{\downarrow}{(z_{0},w_{0})}}\Psi_{5,l+1}(z,w)=\lim_{(z,w)\mathrel{\uparrow}{(z_{0},w_{0})}}\Psi_{5,l}(z,w).$$

We now define the regularized holonomy along $L_{l}$ to be

$$C_{l}=\mathrm{Hol}^{\mathrm{reg}}(A_{z,w},L_{l}):=\Psi_{a_{l},u}^{-1}\Psi_{5,l+1}.$$ (30)

From the above limit considerations we obtain

$$C_{l} =\lim_{(z,w)\mathrel{\downarrow}{(z_{0},w_{0})}}\Psi_{a_{l},u}^{-1}(z,w)\Psi_{5,l+1}(z,w)$$

$$=\lim_{(z,w)\mathrel{\uparrow}{(z_{0},w_{0})}}\Psi_{a_{l},d}^{-1}(z,w)\Psi_{5,l}(z,w)$$

$$=\Psi_{a_{l},d}^{-1}\Psi_{5,l+1}.$$

We conclude that

$$\Psi_{5,l+1}^{-1}\Psi_{5,l} =\Psi_{5,l+1}^{-1}\Psi_{a_{l},u}\Psi_{a_{l},u}^{-1}\Psi_{a_{l},d}\Psi_{a_{l},d}^{-1}\Psi_{5,l}$$

$$=C_{l}^{-1}e^{-\epsilon_{l}t_{zw}}C_{l}.$$ (31)

3.2.2 $H_{z}$ and $H_{w}$

First, we consider the expression $\Psi_{5,1}(z,w)^{-1}\Psi_{1}(z,w)$, where the two solutions are analytically continued in the strip of the variable $w$ along the path $\gamma$.

At this point, it is convenient to introduce new functions

$$\tilde{\Psi}_{5,1}(z,w):=\left(\frac{z-z_{i}}{v_{i}}\right)^{\frac{-t_{iz}}{2\pi i}}\Psi_{5,1}(z,w),\hskip 8.5359pt\tilde{\Psi}_{1}(z,w):=\left(\frac{z-z_{i}}{v_{i}}\right)^{\frac{-t_{iz}}{2\pi i}}\Psi_{1}(z,w).$$

Observe that

$$\Psi_{5,1}(z,w)^{-1}\Psi_{1}(z,w)=\tilde{\Psi}_{5,1}(z,w)^{-1}\tilde{\Psi}_{1}(z,w),$$

and that functions $\tilde{\Psi}_{5,1}(z,w)$ and $\tilde{\Psi}_{1}(z,w)$ satisfy the same differential equation (the conjugate of the KZ equation by the factor $((z-z_{i})/v_{i})^{t_{iz}/2\pi i}$. Furthermore, both $\tilde{\Psi}_{5,1}(z,w)$ and $\tilde{\Psi}_{1}(z,w)$ are regular functions at $z=z_{i}$, and $\tilde{\Psi}_{5,1}(z_{i},w)$ and $\tilde{\Psi}_{1}(z_{i},w)$ are solutions of the KZ equation

$$d\tilde{\Psi}=\left(\frac{t_{iz,w}}{2\pi i}\,d\log(w-z_{i})+\sum_{k\neq i}\frac{t_{k,w}}{2\pi i}\,d\log(w-z_{k})\right)\tilde{\Psi}.$$ (32)

In fact,

$$\tilde{\Psi}_{5,1}(z_{i},w)=\Psi_{(z_{j},v_{j})}(w),\hskip 8.5359pt\tilde{\Psi}_{1}(z_{i},w)=\Psi_{(z_{i},v_{i})}(w).$$

We can now conclude

$$\begin{array}[]{lll}\Psi_{5,1}(z,w)^{-1}\Psi_{1}(z,w)&=&\tilde{\Psi}_{5,1}(z,w)^{-1}\tilde{\Psi}_{1}(z,w)\\ &=&\Psi_{(z_{j},v_{j})}(w)^{-1}\Psi_{(z_{i},v_{i})}(w)\\ &=&\mathrm{Hol}^{\mathrm{reg}}\left(\frac{t_{iz,w}}{2\pi i}\,d\log(w-z_{i})+\sum_{k\neq i}\frac{t_{k,w}}{2\pi i}\,d\log(w-z_{k}),\gamma\right)\\ &=&H_{w}.\end{array}$$ (33)

In a similar way, we have

$$\begin{array}[]{lll}\Psi_{4}(z,w)^{-1}\Psi_{5,m+1}(z,w)&=&\mathrm{Hol}^{\mathrm{reg}}\left(\frac{t_{z,wj}}{2\pi i}\,d\log(z-z_{j})+\sum_{k\neq j}\frac{t_{k,z}}{2\pi i}\,d\log(z-z_{k}),\gamma\right)\\ &=&H_{z}.\end{array}$$ (34)

3.2.3 $H_{zw}$

In this Section, we consider the expression $\Psi_{3}(z,w)^{-1}\Psi_{2}(z,w)$.

To start with, it is useful to consider the following simple differential equation

$$d\phi=t_{z,w}d\log(z-w)\phi.$$

We are interested in solutions of this equation for $z=\gamma(t),w=\gamma(s)$ with $t<s$. It convenient to introduce two normalized solutions $\phi_{i}(z,w)$ and $\phi_{j}(z,w)$,

$$\lim_{(s,t)\to(0,0)}\phi_{i}(z,w)\left(s-t\right)^{-\frac{t_{zw}}{2\pi i}}=1,\hskip 8.5359pt\lim_{(s,t)\to(1,1)}\phi_{j}(z,w)\left(s-t\right)^{-\frac{t_{zw}}{2\pi i}}=1.$$ (35)

The normalizations are chosen such that

$$\phi_{i}(z,w) =\left(\frac{w-z}{v_{i}}\right)^{\frac{t_{zw}}{2\pi i}}, \phi_{j}(z,w) =\left(\frac{w-z}{-v_{j}}\right)^{\frac{t_{zw}}{2\pi i}}$$

in the second and third asymptotic region, respectively.

Note that expressions $\phi_{i}(z,w)\left(s-t\right)^{-\frac{t_{zw}}{2\pi i}}$ and $\phi_{j}(z,w)\left(s-t\right)^{-\frac{t_{zw}}{2\pi i}}$ admit continuous extensions to the line $s=t$, we denote these extensions by $\varphi_{i}(t)$ and $\varphi_{j}(t)$, respectively. These functions are solutions of the following differential equation:

$$d\varphi=t_{zw}\,d\log\left(\lim_{s\to t}\frac{\gamma(s)-\gamma(t)}{s-t}\right)\,\varphi=t_{zw}\,d\log\dot{\gamma}(t)\,\varphi.$$

This implies

$$\phi_{j}(z,w)\phi_{i}^{-1}(z,w) =\phi_{j}(z,w)\left(s-t\right)^{-\frac{t_{zw}}{2\pi i}}\left(\phi_{i}(z,w)\left(s-t\right)^{-\frac{t_{zw}}{2\pi i}}\right)^{-1}$$

$$=\exp\left(t_{zw}\int_{0}^{1}d\log\dot{\gamma}(t)\right)$$

$$=\left|\tfrac{v_{j}}{v_{i}}\right|^{\frac{-t_{zw}}{2\pi}}e^{-t_{zw}{\rm rot}_{\gamma}},$$

where we used definition (17) in the last equation. Similarly to the previous Section, we define

$$\tilde{\Psi}_{2}(z,w)=\phi_{i}^{-1}(z,w)\Psi_{2}(z,w),\hskip 8.5359pt\tilde{\Psi}_{3}(z,w)=\phi_{j}^{-1}(z,w)\Psi_{3}(z,w).$$

Again, $\tilde{\Psi}_{2}(z,w)$ and $\tilde{\Psi}_{3}(z,w)$ satisfy the same differential equation (the conjugate of the KZ equation by the factor $(w-z)^{t_{zw}/2\pi i}$), they are regular at $w=z$, and their values $\tilde{\Psi}_{2}(z=w)$ and $\tilde{\Psi}_{3}(z=w)$ satisfy the equation

$$d\tilde{\Psi}=\left(\sum_{k=1}^{n}\frac{t_{k,zw}}{2\pi i}\,d\log(z-z_{l})\right)\tilde{\Psi}.$$

By considering the asymptotic behavior, we identify

$$\tilde{\Psi}_{2}(z=w)=\Psi_{(z_{i},v_{i})}^{1,\dots,n,zw}(z=w),\hskip 8.5359pt\tilde{\Psi}_{3}(z=w)=\Psi_{(z_{j},v_{j})}^{1,\dots,n,zw}(z=w)$$

. Using that $[t_{zw},t_{i,zw}]=0$ for all $i=1,\dots,n$, we conclude

$$\begin{array}[]{lll}\Psi_{3}(z,w)^{-1}\Psi_{2}(z,w)&=&\tilde{\Psi}_{3}(z=w)^{-1}\left|\frac{v_{j}}{v_{i}}\right|^{\frac{t_{zw}}{2\pi}}e^{t_{zw}{\rm rot}_{\gamma}}\,\tilde{\Psi}_{2}(z=w)\\ &=&\tilde{\Psi}_{3}(z=w)^{-1}\tilde{\Psi}_{2}(z=w)\,\left|\frac{v_{j}}{v_{i}}\right|^{\frac{t_{zw}}{2\pi}}e^{t_{zw}{\rm rot}_{\gamma}}\\ &=&{\rm Hol}^{\rm reg}\left(\sum_{k=1}^{n}\frac{t_{k,zw}}{2\pi i}\,d\log(z-z_{l}),\gamma\right)\left|\frac{v_{j}}{v_{i}}\right|^{\frac{t_{zw}}{2\pi}}e^{t_{zw}{\rm rot}_{\gamma}}\\ &=&H_{zw}\left|\frac{v_{j}}{v_{i}}\right|^{\frac{t_{zw}}{2\pi}}e^{t_{zw}{\rm rot}_{\gamma}}.\end{array}$$ (36)

3.2.4 $\Phi_{\rm KZ}$ contributions

Next, we consider the regularized holonomy $\Psi_{2}(z,w)^{-1}\Psi_{1}(z,w)$. In the asymptotic region $0<s,t\ll 1$, it is convenient to introduce

$$\tilde{\Psi}_{1,2}(z,w)=\left(\frac{w-z_{i}}{v_{i}}\right)^{-(t_{iz}+t_{iw}+t_{zw})/2\pi i}\Psi_{1,2}(z,w).$$

On the one hand, we have

$$\tilde{\Psi}_{2}(z,w)^{-1}\tilde{\Psi}_{1}(z,w)=\Psi_{2}(z,w)^{-1}\Psi_{1}(z,w).$$

And on the other hand, $w=z_{i}$ is not a singularity of the differential equation, $\tilde{\Psi}(w=z_{i})$ satisfies the following equation:

$$d\tilde{\Psi}=\left(t_{iz}d\log\left(\frac{z-z_{i}}{w-z_{i}}\right)+t_{zw}d\log\left(\frac{w-z}{w-z_{i}}\right)+\dots\right)\tilde{\Psi},$$

where $\dots$ stand for the terms regular in $(z-z_{i}),(w-z_{i})$. By making the change of variables

$$\zeta=\frac{z-z_{i}}{w-z_{i}},\hskip 8.5359pt1-\zeta=\frac{w-z}{w-z_{i}}$$

we identify the regularized holonomy with the Drinfeld associator

$$\Psi_{2}(z,w)^{-1}\Psi_{1}(z,w)=\Phi_{\rm KZ}(t_{iz},t_{zw}).$$ (37)

In a similar fashion, we obtain

$$\Psi_{4}(z,w)^{-1}\Psi_{3}(z,w)=\Phi_{\rm KZ}(t_{zw},t_{wj}).$$ (38)

3.3 Properties of $C_{l}$

We now have a rather explicit description of all the terms in the generalized pentagon equation with the exception of holonomies $C_{l}$. In this Section, we establish an important property of these regularized holonomies.

Consider the quotient of the Lie algebra $\mathfrak{t}_{n+2}$ by the Lie ideal generated by $t_{zw}$. Denote the quotient Lie algebra by $\tau=\mathfrak{t}_{n+2}/\langle t_{zw}\rangle$ and the canonical projection by $\pi:\mathfrak{t}_{n+2}\to\tau$. Observe that the image of generators $\pi(t_{i,z})$ and $\pi(t_{i,w})$ for $i=1,\dots,n$ span two commuting free Lie algebras with $n$ generators, that is $\tau\cong\mathfrak{f}_{z}\oplus\mathfrak{f}_{w}$. Then the image of the connection $A_{zw}$ splits into two parts:

$$\pi(A_{z,w})=\frac{1}{2\pi i}\left(\sum_{k=1}^{n}(\pi(t_{k,z})d\log(z-z_{k})+\pi(t_{k,w})d\log(w-z_{k}))\right)=\pi(A_{z})+\pi(A_{w}),$$

where $\pi(A_{z})$ is a connection on $\Sigma=\mathbb{C}\backslash\{z_{1},\dots,z_{n}\}$ corresponding to the point $z$, and similarly $\pi(A_{w})$ is a connection on $\Sigma$ corresponding to the point $w$.

Theorem 3.3.

We have,

$$\pi(C_{l})={\rm Hol}^{\rm reg}(\pi(A_{z}),\gamma_{[0,t_{l}]}){\rm Hol}^{\rm reg}(\pi(A_{w}),\gamma_{[1,s_{l}]}).$$ (39)

Proof.

Recall that

$$C_{l}=\Psi_{a_{l},d}^{-1}(z,w)\Psi_{5,l}(z,w),$$

where

$$\Psi_{a_{l},d}(z,w)\sim\left(\frac{w-z}{-u_{l}}\right)^{t_{zw}/2\pi i},\hskip 8.5359pt\Psi_{5,l}(z,w)\sim\left(\frac{z-z_{i}}{v_{i}}\right)^{t_{iz}/2\pi i}\left(\frac{w-z_{j}}{v_{j}}\right)^{t_{jz}/2\pi i}.$$

After taking a quotient, the self-intersection point $a_{l}$ becomes a regular (and not tangential) base point for $\pi(A_{zw})$, and we also have

$$\pi(\Psi_{a_{l},d})(a_{l},a_{l})=1.$$

Then,

$$\pi(\Psi_{a_{l},d})=\Psi_{a_{l}}(z)\Psi_{a_{l}}(w),$$

where $\Psi_{a_{l}}(z)$ and $\Psi_{a_{l}}(w)$ are local solutions corresponding to connections $\pi(A_{z})$ and $\pi(A_{w})$ with the standard normalization at the regular point $\Psi_{a_{l}}(a_{l})=1$.

Similarly, the asymptotic condition for $\Psi_{5,l}(z,w)$ splits into two independent factors:

$$\pi(\Psi_{5,l})(z,w)\sim\left(\frac{z-z_{i}}{v_{i}}\right)^{\pi(t_{iz})/2\pi i}\left(\frac{w-z_{j}}{v_{j}}\right)^{\pi(t_{jz})/2\pi i},$$

and hence

$$\pi(\Psi_{5,l})(z,w)=\Psi_{(z_{i},v_{i})}(z)\Psi_{(z_{j},v_{j})}(w).$$

We conclude,

$$\begin{array}[]{lll}\pi(\Psi_{a_{l},d}^{-1}(z,w)\Psi_{5,l}(z,w))&=&\pi(\Psi_{a_{l},d})^{-1}(z,w)\pi(\Psi_{5,l})(z,w))\\ &=&(\Psi_{a_{l}}(z)^{-1}\Psi_{a_{l}}(w)^{-1})(\Psi_{(z_{i},v_{i})}(z)\Psi_{(z_{j},v_{j})}(w))\\ &=&(\Psi_{a_{l}}(z)^{-1}\Psi_{(z_{i},v_{i})}(z))(\Psi_{a_{l}}(w)^{-1}\Psi_{(z_{j},v_{j})}(w))\\ &=&{\rm Hol}^{\rm reg}(\pi(A_{z}),\gamma_{[0,t_{l}]}){\rm Hol}^{\rm reg}(\pi(A_{w}),\gamma_{[1,s_{l}]}),\end{array}$$

as required. ∎