Riemann–Hilbert Problems and Painlevé VI

On this page:

what a Riemann–Hilbert problem (RHP) is (in the sense used in modern mathematical physics),
why it is a natural language for analytic continuation and monodromy,
how the Gauss hypergeometric equation can be encoded as an explicit matrix RHP with constant jumps,
how the story generalizes to Heun (four regular singular points), where an accessory parameter appears,
and why moving a singularity while keeping monodromy fixed produces isomonodromy and Painlevé VI.

1. What is a Riemann–Hilbert problem?

1.1 Boundary values and “jumps” across a contour

Let $\Sigma$ be an oriented contour in the complex plane (a union of smooth curves). An RHP asks for a function $\Phi(z)$ that is analytic on $\mathbb{C}\setminus \Sigma$ and whose limiting boundary values on $\Sigma$ satisfy a jump condition.

$\Phi_+(t)$ means the limit as $z\to t\in\Sigma$ from the left side of the oriented contour.
$\Phi_-(t)$ means the limit as $z\to t$ from the right side.

There are two common types:

Additive (scalar) jump:

\Phi_+(t)-\Phi_-(t)=f(t), \qquad t\in\Sigma.

Multiplicative (matrix) jump (the modern workhorse):

\Phi_+(t)=\Phi_-(t)\,J(t), \qquad t\in\Sigma,

where $J(t)$ is a given jump matrix.

In addition, one prescribes:

behavior at special points (endpoints of $\Sigma$ , singularities),
and typically a normalization (e.g. $\Phi(z)\to I$ as $z\to\infty$ ).

1.2 The simplest example: scalar RHP solved by a Cauchy integral

Consider $\Sigma=\mathbb{R}$ and the additive RHP

\Phi_+(x)-\Phi_-(x)=f(x), \qquad x\in\mathbb{R},

with a decay/normalization condition such as $\Phi(z)\to 0$ as $|z|\to\infty$ .

A canonical solution is the Cauchy transform

\Phi(z)=\frac{1}{2\pi i}\int_{-\infty}^{\infty}\frac{f(s)}{s-z}\,ds,\qquad \operatorname{Im} z\neq 0.

The Sokhotski–Plemelj formulas tell you how this integral behaves when you approach the real axis from above/below, producing exactly the prescribed jump. This is the prototype for how RHPs convert discontinuity data on a contour into analytic functions.

1.3 Why physicists care: Wiener–Hopf, inverse scattering, integrable systems

RHPs appear whenever:

you need to split something into parts analytic in complementary domains (Wiener–Hopf factorization),
you reconstruct a physical field from spectral data (inverse scattering),
you study long-time asymptotics by contour deformation (nonlinear steepest descent / Deift–Zhou method),
or you encode analytic continuation/monodromy constraints efficiently (isomonodromy and special functions).

A useful mental model is:

Analyticity + jump/monodromy data + normalization $\Rightarrow$ the solution.

This becomes especially powerful for matrix-valued problems, where jumps encode how a basis of solutions changes under analytic continuation.

2. Hypergeometric equation and monodromy

2.1 The Gauss hypergeometric equation

The Gauss hypergeometric equation is

z(1-z)\,y''+\bigl(c-(a+b+1)z\bigr)\,y'-ab\,y=0.

It has three regular singular points at $z=0,1,\infty$ . A basis of local solutions near a singular point can be analytically continued around loops encircling that singularity. After a loop, the basis comes back transformed by a constant $2\times 2$ matrix: this is monodromy.

RHPs are a systematic way of encoding that monodromy as jumps across cuts.

2.2 Rewrite as a $2\times 2$ Fuchsian system

Define a vector

u(z)=\begin{pmatrix}y(z)\\ z\,y'(z)\end{pmatrix}.

Then $u$ satisfies a first-order system

\frac{du}{dz}=A(z)\,u,\qquad A(z)=\frac{A_0}{z}+\frac{A_1}{z-1},

with constant residues

A_0=\begin{pmatrix}0&1\\ 0&1-c\end{pmatrix}, \qquad A_1=\begin{pmatrix}0&0\\ -ab&-(a+b-c+1)\end{pmatrix}.

Let $\Psi(z)\in GL(2,\mathbb{C})$ be a fundamental matrix solution of $\Psi' = A\Psi$ ; its columns are two independent solutions $u^{(1)},u^{(2)}$ .

The residue at infinity is

A_\infty:=-A_0-A_1=\begin{pmatrix}0&-1\\ ab&a+b\end{pmatrix},

whose eigenvalues are $a$ and $b$ (the exponents at $\infty$ ).

3. An explicit RHP whose solution gives a hypergeometric fundamental matrix

This section gives an RHP with:

an explicit contour (cuts),
constant jump matrices,
local exponent behavior at $0,1,\infty$ ,

such that solving the RHP produces a fundamental matrix for a hypergeometric system (up to a constant gauge).

3.1 Genericity assumptions (to avoid logarithms)

To keep the local forms diagonal (no log terms), assume a non-resonant regime such as

c\notin\mathbb{Z},\qquad c-a-b\notin\mathbb{Z},\qquad a-b\notin\mathbb{Z}.

If these conditions fail, similar formulas hold but local behavior may involve logarithms and the monodromy matrices may become non-diagonal (Jordan blocks).

3.2 Contour (branch cuts)

Choose the cut contour

\Sigma=\Sigma_0\cup\Sigma_1,\qquad \Sigma_0=(-\infty,0],\qquad \Sigma_1=[1,\infty),

both oriented left-to-right. Then $\mathbb{C}\setminus \Sigma$ is simply connected, so branches of $z^\alpha$ and $(1-z)^\beta$ are single-valued there.

3.3 Local exponent matrices

Encode the local exponents by diagonal matrices:

\Theta_0=\operatorname{diag}(0,\,1-c),\qquad \Theta_1=\operatorname{diag}(0,\,c-a-b),\qquad \Theta_\infty=\operatorname{diag}(a,\,b).

3.4 Connection matrix between bases at $0$ and $1$

A standard local basis near $z=0$ is

y_1(z)={}_2F_1(a,b;c;z),

y_2(z)=z^{1-c}{}_2F_1(a-c+1,\;b-c+1;\;2-c;\;z).

A standard local basis near $z=1$ is

\tilde y_1(z)={}_2F_1(a,b;\;a+b+1-c;\;1-z),

\tilde y_2(z)=(1-z)^{c-a-b}{}_2F_1(c-a,\;c-b;\;1+c-a-b;\;1-z).

These are related by a constant connection matrix $C_{01}$ :

\begin{pmatrix} y_1 & y_2 \end{pmatrix} = \begin{pmatrix} \tilde y_1 & \tilde y_2 \end{pmatrix}\,C_{01}, \qquad C_{01}=\begin{pmatrix} A & A_2\\ B & B_2\end{pmatrix},

where (in the generic non-resonant case)

A=\frac{\Gamma(c)\Gamma(c-a-b)}{\Gamma(c-a)\Gamma(c-b)}, \qquad B=\frac{\Gamma(c)\Gamma(a+b-c)}{\Gamma(a)\Gamma(b)},

A_2=\frac{\Gamma(2-c)\Gamma(c-a-b)}{\Gamma(1-a)\Gamma(1-b)}, \qquad B_2=\frac{\Gamma(2-c)\Gamma(a+b-c)}{\Gamma(a-c+1)\Gamma(b-c+1)}.

When you build fundamental matrices from $u=(y,zy')^T$ , the same constant coefficients appear (derivatives do not change the connection coefficients), so the same $C_{01}$ relates the corresponding fundamental matrices.

3.5 Constant jump matrices

Crossing a cut changes the argument of the fractional powers, producing constant monodromy factors.

Jump on $\Sigma_0=(-\infty,0)$ : $z^{1-c}$ picks up $e^{2\pi i(1-c)}$ , so

J_0=\begin{pmatrix}1&0\\ 0&e^{2\pi i(1-c)}\end{pmatrix}.

Jump on $\Sigma_1=(1,\infty)$ : in the local basis at $z=1$ , $(1-z)^{c-a-b}$ picks up $e^{2\pi i(c-a-b)}$ , so the diagonal jump there is

D_1=\begin{pmatrix}1&0\\ 0&e^{2\pi i(c-a-b)}\end{pmatrix}.

In the basis normalized at $z=0$ , this jump is conjugated by $C_{01}$ :

J_1=C_{01}^{-1}\,D_1\,C_{01}.

3.6 The RHP– ${}_2F_1$ (full statement)

Given $a,b,c$ (generic) and the contour $\Sigma=(-\infty,0]\cup[1,\infty)$ , find $\Psi:\mathbb{C}\setminus\Sigma\to GL(2,\mathbb{C})$ such that:

Analyticity: $\Psi$ is analytic on $\mathbb{C}\setminus\Sigma$ , with boundary values $\Psi_\pm$ on $\Sigma$ .
Jumps (piecewise constant):

\Psi_+(x)=\Psi_-(x)\,J(x),

where

J(x)=\begin{cases} J_0, & x\in(-\infty,0),\\ J_1, & x\in(1,\infty). \end{cases}

Local exponent behavior near $0$ :

\Psi(z)=G_0(z)\,z^{\Theta_0}, \qquad z\to 0,

with $G_0$ analytic and invertible near $0$ .

Local exponent behavior near $1$ :

\Psi(z)=G_1(z)\,(1-z)^{\Theta_1}\,C_{01}, \qquad z\to 1,

with $G_1$ analytic and invertible near $1$ .

Behavior at infinity:

\Psi(z)=\left(I+O\!\left(\frac{1}{z}\right)\right)\,z^{-\Theta_\infty}\,C_\infty, \qquad z\to\infty,

for some constant invertible $C_\infty$ (you can fix $C_\infty=I$ by choosing a canonical basis at $\infty$ ).

3.7 Why solving this RHP gives a hypergeometric system

Define

A(z)=\Psi'(z)\Psi(z)^{-1}.

Because the jumps are constant, differentiating $\Psi_+=\Psi_-J$ gives $\Psi'_+=\Psi'_-J$ , hence $A_+=A_-$ : $A(z)$ has no jump across $\Sigma$ . Therefore $A(z)$ extends meromorphically across the cuts.

The prescribed local behaviors imply that $A(z)$ has at most simple poles at $0$ and $1$ and decays like $O(1/z)$ at infinity, so

A(z)=\frac{\widehat A_0}{z}+\frac{\widehat A_1}{z-1}.

Moreover, the eigenvalues of $\widehat A_0,\widehat A_1,\widehat A_\infty$ match $(0,1-c)$ , $(0,c-a-b)$ , $(a,b)$ , so the resulting system is (up to a constant gauge) the hypergeometric Fuchsian system. Eliminating one component recovers the scalar hypergeometric equation.

This is the clean mechanism behind many “RHP $\Rightarrow$ differential equation” constructions: constant jumps force $\Psi'\Psi^{-1}$ to be rational.

4. From hypergeometric (3 points) to Heun (4 points)

Hypergeometric has 3 regular singular points ( $0,1,\infty$ ). Heun has 4 ( $0,1,t,\infty$ ). That single extra point is exactly where new complexity—and Painlevé VI—enters.

4.1 The Heun equation and the accessory parameter

A standard general Heun equation is

y''+\left(\frac{\gamma}{z}+\frac{\delta}{z-1}+\frac{\epsilon}{z-t}\right)y' +\frac{\alpha\beta\,z-q}{z(z-1)(z-t)}\,y=0, \qquad \epsilon=\alpha+\beta-\gamma-\delta+1.

It has regular singular points at $z=0,1,t,\infty$ .

Unlike the hypergeometric case, the local exponents do not uniquely determine the equation: there is an additional parameter $q$ , called an accessory parameter, which encodes global information (how local solutions are glued together / monodromy).

A useful rule of thumb for second-order Fuchsian equations on the Riemann sphere:

with $n$ regular singular points, there are $(n-3)$ accessory parameters;
so for $n=3$ (hypergeometric) there are none (rigidity),
and for $n=4$ (Heun) there is exactly one (non-rigidity).

4.2 The $2\times 2$ Fuchsian system with four poles

As before, you can rewrite the scalar equation (or work directly) in a $2\times 2$ Fuchsian form

\frac{\partial\Psi}{\partial z}= \left( \frac{A_0}{z}+\frac{A_1}{z-1}+\frac{A_t}{z-t} \right)\Psi, \qquad A_\infty:=-A_0-A_1-A_t.

Often one chooses a gauge so that $A_i\in\mathfrak{sl}_2$ and the eigenvalues of each residue are $\pm \theta_i/2$ . The parameters $\theta_0,\theta_1,\theta_t,\theta_\infty$ are the monodromy exponents (related to local Frobenius exponents).

5. The RHP for four regular singular points

Pick a cut system making the plane simply connected. For concreteness, if $t\in(0,1)$ , one convenient choice is

\Sigma(t)=(-\infty,0]\ \cup\ [t,1]\ \cup\ [1,\infty),

or any three non-intersecting arcs connecting the singular points to infinity.

An RHP encoding the monodromy is:

Unknown: $\Psi(z)$ analytic on $\mathbb{C}\setminus\Sigma(t)$ .
Jumps (constant on each component):

\Psi_+(z)=\Psi_-(z)\,M_0 \quad \text{on }(-\infty,0),

\Psi_+(z)=\Psi_-(z)\,M_t \quad \text{on }(t,1),

\Psi_+(z)=\Psi_-(z)\,M_1 \quad \text{on }(1,\infty).

Local exponent behavior:

\Psi(z)=G_i(z)\,(z-a_i)^{\Theta_i}\,C_i,\qquad z\to a_i\in\{0,1,t\},

with $G_i$ analytic invertible, $\Theta_i$ diagonal (encoding local exponents), and constant $C_i$ allowing for basis choices.

At infinity:

\Psi(z)=\left(I+O\!\left(\frac{1}{z}\right)\right)z^{-\Theta_\infty}C_\infty,\qquad z\to\infty.

The monodromy matrices satisfy a consistency condition (product around all singularities is identity):

M_\infty(M_1M_tM_0)=I \quad\Longleftrightarrow\quad M_\infty=(M_1M_tM_0)^{-1}.

Just as in the hypergeometric case, if the jumps are constant, then

A(z)=\Psi'(z)\Psi(z)^{-1}

has no jump across $\Sigma(t)$ and must be a rational function with simple poles at $0,1,t$ :

A(z)=\frac{\widehat A_0}{z}+\frac{\widehat A_1}{z-1}+\frac{\widehat A_t}{z-t}.

6. Why Painlevé VI appears: isomonodromy (moving $t$ with fixed monodromy)

Now comes the conceptual punchline.

6.1 Isomonodromic deformation

Let the position $t$ of one singularity vary, but demand that the monodromy data (equivalently, jump matrices)

M_0,\ M_1,\ M_t,\ M_\infty

remain constant.

In RHP language:

the contour endpoints move with $t$ ,
but the jump matrices do not.

In differential-equation language:

the residues $A_0(t),A_1(t),A_t(t)$ must adjust with $t$ so that analytic continuation data of solutions does not change.

This requirement yields the Schlesinger equations (the compatibility conditions of a Lax pair):

\frac{dA_0}{dt}=\frac{[A_t,A_0]}{t},\qquad \frac{dA_1}{dt}=\frac{[A_t,A_1]}{t-1},

\frac{dA_t}{dt}=-\frac{[A_t,A_0]}{t}-\frac{[A_t,A_1]}{t-1}.

These are nonlinear matrix ODEs, but they are integrable. For $2\times 2$ systems (rank 2), you can reduce them to a single scalar second-order equation: Painlevé VI.

6.2 Painlevé VI (PVI)

A common standard form of Painlevé VI is

\frac{d^2 y}{dt^2} = \frac{1}{2}\left(\frac{1}{y} + \frac{1}{y-1} + \frac{1}{y-t}\right)\left(\frac{dy}{dt}\right)^2 - \left(\frac{1}{t} + \frac{1}{t-1} + \frac{1}{y-t}\right)\frac{dy}{dt}

\quad + \frac{y(y-1)(y-t)}{t^2(t-1)^2} \left(\alpha + \beta\frac{t}{y^2} + \gamma\frac{t-1}{(y-1)^2} + \delta\frac{t(t-1)}{(y-t)^2}\right).

Here $(\alpha,\beta,\gamma,\delta)$ are determined by the monodromy exponents $\theta_0,\theta_1,\theta_t,\theta_\infty$ (i.e. by the conjugacy classes of the monodromy matrices). Roughly speaking, the function $y(t)$ can be realized as a geometrically natural coordinate on the moduli space of rank-2 Fuchsian connections with four poles (for example, it can be tied to an “apparent singularity” or a zero of a particular matrix element after a suitable gauge choice).

6.3 How this explains the Heun accessory parameter

For fixed $t$ , specifying only local exponents does not fix the Heun equation: you must also set the accessory parameter $q$ . From the monodromy viewpoint:

fixing monodromy data determines $q$ (implicitly and generally non-elementarily) as a function of $t$ and the monodromy.
under isomonodromy, monodromy is fixed while $t$ varies, so $q=q(t)$ becomes a dynamical object.

In fact, $q(t)$ is closely related to the isomonodromic Hamiltonian and to the logarithmic derivative of the Jimbo–Miwa–Ueno $\tau$ -function. At an intuitive level:

Hypergeometric (3 points): no accessory parameter; connection coefficients are explicit (Gamma functions).
Heun (4 points): one accessory parameter; matching a chosen monodromy typically forces nonlinear special functions (Painlevé VI) to appear.

This is why Heun functions are “one notch harder” than hypergeometric functions, and why isomonodromy becomes the natural organizing principle.

7. Summary and mental map

An RHP is an analytic function reconstruction problem from prescribed jumps across a contour.
For linear ODEs with regular singularities, “jump data” can be taken to encode monodromy of a fundamental matrix.
The Gauss hypergeometric equation (3 singularities) corresponds to a particularly rigid monodromy problem and admits an explicit RHP with two cuts and constant jumps; connection matrices are given by Gamma functions.
The Heun equation (4 singularities) introduces an accessory parameter, reflecting the fact that local exponents do not determine global monodromy.
Requiring monodromy to remain fixed while moving the fourth singularity $t$ yields the Schlesinger equations, which in rank 2 reduce to Painlevé VI.

8. What to read next (suggested paths)

Complex analysis → RHP: Cauchy integrals, Sokhotski–Plemelj formulas, singular integral equations, Wiener–Hopf factorization.
ODE → monodromy: Frobenius method, analytic continuation, monodromy/connection matrices, Fuchsian systems on the Riemann sphere.
Integrable systems: inverse scattering transform; the Deift–Zhou nonlinear steepest descent method for long-time asymptotics.
Isomonodromy: Schlesinger equations, Jimbo–Miwa–Ueno $\tau$ -function, Painlevé equations (especially PVI from 4-point Fuchsian systems).