---
title: 'Manhattan Curve: Thermodynamic & Geometric Analysis'
url: https://www.emergentmind.com/topics/manhattan-curve
type: topic
---

# Manhattan Curve: Thermodynamic & Geometric Analysis

Searching arXiv for recent and foundational papers on “Manhattan curve” to ground the article in the literature.
The Manhattan curve is a convex thermodynamic object attached to a pair of structures that carry comparable length data. In the classical surface setting, it is associated with two hyperbolic structures on the same surface and records the weighted exponential growth of closed geodesic lengths; in equivalent formulations it is the locus where a weighted Poincaré series has critical exponent \(1\), or where an associated pressure vanishes. In current arXiv literature, the same idea appears for pairs of left-invariant hyperbolic metrics on hyperbolic groups, as the \(n=1\) case of a higher-dimensional Manhattan manifold, and for pairs of hyperbolic holomorphic dynamical systems where geodesic lengths are replaced by logarithmic multiplier or Jacobian growth [1801.09826] [2104.13451] [2411.06375] [2508.12490].

## 1. Classical definition on hyperbolic surfaces

For a surface \(S\) with negative Euler characteristic and Fuchsian representations
\[
\rho_1,\rho_2:\pi_1(S)\to \mathrm{PSL}(2,\mathbb R),
\]
the cusped-surface formulation begins with the weighted Manhattan distance
\[
d_{\rho_1,\rho_2}^{a,b}(o,\gamma o)=a\,d(o_1,\rho_1(\gamma)o_1)+b\,d(o_2,\rho_2(\gamma)o_2),
\]
where \(o=(o_1,o_2)\in \mathbb H\times \mathbb H\). The associated Poincaré series is
\[
Q_{\rho_1,\rho_2}^{a,b}(s)=\sum_{\gamma\in \pi_1(S)}\exp\!\left(-s\,d_{\rho_1,\rho_2}^{a,b}(o,\gamma o)\right),
\]
with critical exponent \(\delta_{\rho_1,\rho_2}^{a,b}\). The Manhattan curve is then
\[
\mathcal C(\rho_1,\rho_2)=\{(a,b)\in D:\delta^{a,b}=1\},
\]
equivalently the boundary of the set where \(Q_{\rho_1,\rho_2}^{a,b}(1)<\infty\) [1801.09826].

This definition expresses the curve as a comparison object for two marked length spectra. At the level of closed geodesics, the relevant data are the lengths \(l_1[\gamma]\) and \(l_2[\gamma]\), and the weighted functional \(a\,l_1[\gamma]+b\,l_2[\gamma]\) determines the growth rate. The literature cited in the arXiv record places this construction in the line of work associated with Burger and Sharp, and the cusped case extends that theory from convex-cocompact settings to non-compact surfaces with cusps [1801.09826].

A parallel group-theoretic formulation replaces surface geodesic lengths by stable translation lengths or ordinary displacement growth. For a non-elementary hyperbolic group \(\Gamma\) and metrics \(d,d_\ast\in\mathcal D_\Gamma\), one defines
\[
\mathcal Q(a,b)=\sum_{[x]\in \mathrm{conj}(\Gamma)} \exp\!\bigl(-a\,\ell_\ast[x]-b\,\ell[x]\bigr)
\]
and
\[
\mathcal P(a,b)=\sum_{x\in\Gamma}\exp\!\bigl(-a\,d_\ast(o,x)-b\,d(o,x)\bigr).
\]
The Manhattan curve \(\mathcal C_M\) is the boundary of the convergence set of \(\mathcal Q\), and equivalently the graph \(b=\theta(a)\), where \(\theta(a)\) is the abscissa of convergence in \(b\) of \(\mathcal P(a,b)\) [2104.13451].

## 2. Pressure formalism and regularity theory

A central feature of the Manhattan curve is that it is not merely a convergence boundary. In the cusped-surface setting, under the hypotheses of boundary-preserving isomorphism and the extended Schottky condition, Dal’Bo–Peigné coding yields a topologically mixing countable Markov shift \((\Sigma^+,\sigma)\) with the BIP property and locally Hölder geometric potentials \(\tau,\kappa\) encoding the two length spectra. The Manhattan curve is exactly the pressure-zero locus
\[
\mathcal C(\rho_1,\rho_2)=\{(a,b)\in D:\ P_\sigma(-a\tau-b\kappa)=0\},
\]
and for each \((a,b)\in D\) there exists a unique \(t_{a,b}\) such that
\[
P_\sigma\big(-t_{a,b}(a\tau+b\kappa)\big)=0,
\qquad t_{a,b}=\delta^{a,b}.
\]
The same framework exhibits a genuinely cuspidal phase transition:
\[
P_\sigma\big(-t(a\tau+b\kappa)\big)=
\begin{cases}
\infty, & t<\dfrac{1}{2(a+b)},\\[1ex]
\text{real analytic}, & t>\dfrac{1}{2(a+b)}.
\end{cases}
\]
Within its domain of finiteness, the pressure is analytic, and the Manhattan curve is real analytic [1801.09826].

The pressure formulation persists in later generalizations. In complex dynamics, for \(f_1,f_2\) in the same hyperbolic component of \(\mathrm{End}(k,d)\), the weighted Poincaré series defines a critical exponent \(\delta_{f_1,f_2}^{a,b}\), and the Manhattan curve is
\[
\mathcal C(f_1,f_2)=\{(a,b)\in \mathbb R_{\ge 0}^2\setminus\{(0,0)\}:\ \delta_{f_1,f_2}^{a,b}=1\}.
\]
With \(\tau_i(x)=\log |\Jac f_i(h_i(x))|\), the curve is equivalently
\[
\mathcal C(f_1,f_2)=\{(a,b):\ \mathcal P(-a\tau_1-b\tau_2)=0\},
\]
which is the direct analogue of the surface-theoretic zero-pressure description [2508.12490].

These formulations explain why the Manhattan curve is routinely studied by thermodynamic formalism rather than by elementary orbit counting alone. The object is defined by growth, but its regularity and rigidity are extracted from pressure, equilibrium states, and symbolic or dynamical codings. This suggests that the term denotes not an arbitrary weighted-growth graph, but a specific critical-exponent/pressure locus.

## 3. Geometric meaning, slopes, and rigidity

The curve packages comparison data between two length structures, and its differential geometry carries explicit meaning. In the cusped-surface setting, if the curve is locally written as \(b=b(a)\), then
\[
b'(a)=-\frac{\int \tau\, d\mu_{-a\tau-b(a)\kappa}}{\int \kappa\, d\mu_{-a\tau-b(a)\kappa}},
\]
so the tangent slope is an average ratio of the two geometric potentials under the equilibrium state at that point. The same paper proves that the curve is convex and continuous in general, real analytic in the extended Schottky setting, strictly convex unless the two representations are conjugate in \(\mathrm{PSL}(2,\mathbb R)\), and a straight line if and only if they are conjugate [1801.09826].

For hyperbolic groups, the derivative takes a directly metric form. The function \(\theta(a)\) is continuously differentiable, and for every \((a,b)\in\mathcal C_M\),
\[
\theta'(a)=-\tau_{a,b},
\]
where \(\tau_{a,b}\) is the almost sure local intersection number determined by the corresponding boundary measure. The slope therefore records the typical asymptotic ratio between the two metrics along generic rays. The same work proves that \(\mathcal C_M\) is a straight line if and only if \(d\) and \(d_\ast\) are roughly similar, and if \(d\) and \(d_\ast\) are strongly hyperbolic then the Manhattan curve is twice continuously differentiable [2104.13451].

This rigidity has a quantitative form at the intercept \((0,v)\). The mean distortion
\[
(d_\ast/d)=\lim_{r\to\infty}\frac{1}{\#B(o,r)} \sum_{x\in B(o,r)}\frac{d_\ast(o,x)}{r}
\]
exists and satisfies
\[
\theta'(0)=-(d_\ast/d),
\qquad
(d_\ast/d)\ge \frac{v}{v_\ast}.
\]
Equality is equivalent to proportional stable translation lengths and to rough similarity of the two metrics [2104.13451].

In complex dynamics, the rigidity statement has the same formal shape but different content. The Manhattan curve \(\mathcal C(f_1,f_2)\) is a straight line if and only if the marked length spectra of \(f_1\) and \(f_2\) are proportional. Its special point with tangent slope \(-1\) controls the asymptotic correlation of multiplier spectra: if \((a_0,b_0)\) is the unique point on \(\mathcal C(f_1,f_2)\) where the tangent has slope \(-1\), then the correlation number is
\[
\alpha=a_0+b_0
\]
and appears in the asymptotic
\[
M_{T,\epsilon}(f_1,f_2)\sim C\,\frac{e^{\alpha T}}{T^{3/2}}.
\]
Thus the geometry of the curve is not decorative; it determines a concrete asymptotic counting exponent [2508.12490].

## 4. Hyperbolic groups, higher-dimensional analogues, and Manhattan manifolds

The hyperbolic-group version abstracts the surface picture to any non-elementary hyperbolic group equipped with left-invariant hyperbolic metrics quasi-isometric to a word metric. The intercepts
\[
(0,v),\ (v_\ast,0)\in \mathcal C_M
\]
are determined by the exponential volume growth rates of the two metrics, and the curve is the graph of a convex function \(\theta\). Beyond its \(C^1\) and, in stronger settings, \(C^2\) regularity, this framework connects the Manhattan curve to Patterson–Sullivan theory, topological flows, multifractal spectra, and large deviations. For example,
\[
\dim_H(E(\alpha),\rho)=\inf_{a\in\mathbb R}\{a\alpha+\theta(a)\},
\]
so the multifractal spectrum is the Legendre transform of the Manhattan curve [2104.13451].

A further generalization replaces a pair of metrics by an \(n\)-tuple of hyperbolic metric potentials \(D=(\psi_1,\dots,\psi_n)\) together with a reference \(\varphi\in H_\Gamma^{++}\). The resulting object is the Manhattan manifold
\[
\mathcal M_{D/\varphi}:=\{(v,y)\in\mathbb R^n\times\mathbb R:\ y=\mathcal O_{D/\varphi}(v)\},
\]
where \(\mathcal O_{D/\varphi}(v)\) is the critical exponent of
\[
\sum_{x\in\Gamma} e^{-\langle v,D(x)\rangle-s\varphi(o,x)}.
\]
When \(n=1\), this is exactly the usual Manhattan curve. The higher-dimensional theory proves that \(\mathcal O_{D/\varphi}\) is \(C^1\), strictly convex under independence, and that
\[
\mathrm{Int}(J_\varphi(D))=\{-\nabla\mathcal O_{D/\varphi}(v):v\in\mathbb R^n\},
\]
with \(-\nabla\mathcal O_{D/\varphi}\) a homeomorphism onto the interior of the joint translation spectrum [2411.06375].

This identifies the classical curve as the one-dimensional face of a broader convex-analytic duality. The Manhattan manifold plays for translation cones and joint translation spectra the role that the ordinary Manhattan curve plays for two-metric distortion.

## 5. Complex-dynamical formulation

In complex dynamics, the Manhattan curve is defined for a pair of hyperbolic rational maps or, more generally, hyperbolic holomorphic endomorphisms of \(\mathbb{CP}^k\). Fixing \(f_1,f_2\) in the same hyperbolic component and Hölder conjugacies \(h_i\), the weighted Poincaré series is
\[
P_{f_1,f_2}^{a,b}(s,x)
=
\sum_{n=1}^{\infty}\ \sum_{f_0^n(y)=x}
e^{-s\left(a\log|\Jac f_1^n(h_1(y))|+b\log|\Jac f_2^n(h_2(y))|\right)}.
\]
The corresponding critical exponent \(\delta_{f_1,f_2}^{a,b}\) gives the Manhattan curve
\[
\mathcal C(f_1,f_2)=\{(a,b):\delta_{f_1,f_2}^{a,b}=1\}.
\]
Here logarithmic Jacobian growth replaces geodesic length, in the spirit of Sullivan’s dictionary [2508.12490].

The paper proves the same structural properties familiar from surfaces. The curve has intercepts
\[
(2\mathrm{VD}_{f_1}(J(f_1)),0),
\qquad
(0,2\mathrm{VD}_{f_2}(J(f_2))),
\]
is real-analytic, is convex, and is a straight line if and only if the marked length spectra are proportional. In dimension \(1\), the intercepts become
\[
(\dim_H J(f_1),0),
\qquad
(0,\dim_H J(f_2)).
\]
The object therefore compares two dynamical systems by packaging their periodic-orbit Jacobian growth into the same convex thermodynamic curve [2508.12490].

Its importance in that setting is sharpened by counting results. For a hyperbolic endomorphism \(f\),
\[
N_T(f)\sim Li(T^{2\mathrm{VD}})
\]
when \(\log |\Jac f|\) is not cohomologous to a constant, and for two systems the simultaneous multiplier-counting function satisfies
\[
M_{T,\epsilon}(f_1,f_2)\sim C\,\frac{e^{\alpha T}}{T^{3/2}}.
\]
The correlation number \(\alpha\) is recovered from the curve by the tangent-slope-\(-1\) condition. This is the direct dynamical analogue of the classical correlation-number theorem for pairs of hyperbolic surfaces [2508.12490].

## 6. Terminological scope and common ambiguities

The named object “Manhattan curve” should be distinguished from several unrelated “Manhattan” constructions in the arXiv literature. One recurring source of confusion is the random walk on the Manhattan lattice. The paper “The mean square displacement of random walk on the Manhattan lattice” studies the exact function
\[
n\longmapsto \mathbb E_d[|X_n|^2]
\]
and shows that it is asymptotically linear in \(n\), but it explicitly states that it is not about a standard object called the Manhattan curve; at most, one may informally speak of a curve \(n\mapsto \mathbb E[|X_n|^2]\) [2206.12999].

A second source of ambiguity comes from computational geometry and rectilinear network design. Papers on generalized minimum Manhattan networks, minimum Manhattan networks, and bidirected minimum Manhattan networks study Manhattan paths, M-paths, and rectilinear networks under the \(L_1\)-metric. Their core objects are axis-parallel shortest paths and networks realizing such paths, not the thermodynamic or critical-exponent object called the Manhattan curve [1203.6481] [2403.11811] [1107.1359].

A third nearby usage occurs in multidimensional graph flows. “Manhattan and Chebyshev flows” defines the Manhattan flow number \(\Phi_d^1(G)\) via the annulus
\[
\{x\in\mathbb R^d:1\le \|x\|_1\le r-1\},
\]
and in dimension \(2\) the closest geometric boundary is the diamond-shaped \(\ell_1\)-sphere. The paper explicitly notes that the term “Manhattan curve” does not appear there [2510.22234].

A plausible implication is that the expression “Manhattan curve” is most precise when reserved for the convex critical-exponent or pressure-zero locus attached to a pair of hyperbolic structures, metrics, or dynamical systems, and that other Manhattan-metric objects are better named by the paper-specific terms already established in those literatures.

Source: https://www.emergentmind.com/topics/manhattan-curve