Bowen's Formula in Conformal Dynamics

• Bowen's Formula is a result that links the Hausdorff dimension of a limit set to the unique zero of the associated pressure function in conformal dynamical systems.
• The proof utilizes covering methods for upper bounds and mass distribution techniques for lower bounds, enhanced by transfer operator analysis.
• Extensions of the formula cover random, non-autonomous, and overlapping systems, as well as applications in complex dynamics using thermodynamic formalism.

Bowen’s formula is the statement that a geometrically defined dimension is recovered from the zero, or critical parameter, of a pressure function associated with a conformal dynamical system. In the deterministic conformal setting it identifies the Hausdorff dimension of a limit set with the unique real root of $P(s)=0$; in later formulations the same scheme appears for random conformal graph directed Markov systems, overlapping self-similar attractors via projection pressure, non-autonomous and tree iterated function systems, free semigroup actions, and several classes of rational or transcendental maps (Arima, 24 Jul 2025, Wang et al., 2010, Barański et al., 2010).

1. Classical statement and deterministic conformal settings

In the special case where $\Omega$ is a single point and $E$ is finite, a random conformal graph directed Markov system reduces to a conformal graph-directed iterated function system. In that case the classical Bowen formula states that the unique real root $s\ge 0$ of

$P(s)=0$

equals the Hausdorff dimension of the limit set $J$ (Arima, 24 Jul 2025).

An equivalent formulation appears for conformal iterated function systems on compact subsets of $\mathbb R^d$ satisfying the open-set and bounded-distortion conditions. If $\Lambda$ is the limit set, then

$\dim_H \Lambda=s^*,$

where $s^*$ is the unique real number for which the pressure vanishes. In the same setting this can also be written as the Moran equation

$\Omega$0

For autonomous systems this is the standard “dimension $\Omega$1 pressure zero” form of Bowen’s formula (Ono, 28 Feb 2026).

For non-autonomous conformal iterated function systems, the pressure is formed from the level sums

$\Omega$2

with lower and upper pressures $\Omega$3 and $\Omega$4. The Bowen dimension is

$\Omega$5

and Bowen’s formula asserts

$\Omega$6

when the root exists. In this framework $\Omega$7 is strictly decreasing and continuous on its finite domain, and under balancing and bounded-distortion hypotheses the zero is unique (Rempe-Gillen et al., 2012).

2. Pressure, geometric potentials, and thermodynamic formalism

The analytic content of Bowen’s formula is carried by a geometric potential and the associated pressure. For a random conformal graph directed Markov system, Roy and Urbański’s setting is described by a probability space $\Omega$8, an invertible ergodic $\Omega$9-preserving transformation $E$0, a finite vertex set $E$1, a countable edge set $E$2, and fiber maps

$E$3

that are $E$4 conformal contractions with fiberwise open-set condition, uniform cone, and bounded distortion properties. The geometric potential is

$E$5

and the random Ruelle operator is

$E$6

Under a summability condition this yields a relative topological pressure

$E$7

defined either from symbolic partition sums or from the asymptotics of $E$8 (Arima, 24 Jul 2025).

In rational graph-directed Markov systems the same thermodynamic pattern appears on the skew-product Julia set. With geometric potential

$E$9

the topological pressure $s\ge 0$0 is real-analytic, strictly decreasing, and satisfies

$s\ge 0$1

hence there is a unique $s\ge 0$2 such that $s\ge 0$3. The transfer operator $s\ge 0$4 produces a unique probability measure $s\ge 0$5, a positive density $s\ge 0$6, and the equilibrium state $s\ge 0$7 with

$s\ge 0$8

(Arimitsu et al., 2024).

This thermodynamic structure persists, with modifications, in several nonclassical settings. Countable alphabets require summability conditions such as

$s\ge 0$9

overlapping systems may replace classical pressure by projection pressure, and free semigroup actions use a Carathéodory–Pesin construction of pressure on arbitrary subsets (Arima, 24 Jul 2025, Wang et al., 2010, Xiao et al., 2020).

3. Standard proof architecture

Across the cited formulations, proofs of Bowen’s formula use the same core mechanisms (Arima, 24 Jul 2025, Rempe-Gillen et al., 2012, Ono, 28 Feb 2026, Arimitsu et al., 2024).

• Upper bounds by coverings: one covers the limit set by cylinders, inverse images of small balls, or maximal antichains. If the relevant partition sum or antichain sum tends to $P(s)=0$0 at exponent $P(s)=0$1, then $P(s)=0$2 vanishes, giving $P(s)=0$3.
• Lower bounds by mass distribution: one constructs measures $P(s)=0$4 or $P(s)=0$5 by assigning weight proportional to $P(s)=0$6 or $P(s)=0$7, passes to a weak-$P(s)=0$8 limit, and proves Frostman-type estimates such as $P(s)=0$9. This yields $J$0.
• Transfer operators and conformal measures: Perron–Frobenius or Ruelle operators provide eigenfunctions, eigenmeasures, random conformal measures, and invariant Gibbs or equilibrium states. These objects control both local scaling and distortion.
• Approximation by finite subsystems or repellers: infinite alphabets are approximated by finite subalphabets, and global pressures in transcendental dynamics are approximated by pressures on hyperbolic repellers. The finite models carry classical Bowen formulas, and the limiting argument transfers the result to the full system.

The same proof architecture also underlies multifractal refinements. In the random conformal graph directed Markov system setting, Gibbs measures for $J$1 and a Legendre transform argument produce Lyapunov spectra tied to $J$2 (Arima, 24 Jul 2025).

4. Randomness, countable alphabets, and variable branching

A major extension is the random conformal graph directed Markov system with countably many edges. For such a system the principal statement is that almost surely

$J$3

Equivalently,

$J$4

The proof relies on refined properties of random finitely primitive countable Markov shifts, including existence of fiberwise Perron–Frobenius eigenfunctions and measures, convexity and finiteness of $J$5 on $J$6, and the compact approximation property

$J$7

This allows the infinite system to be approximated by finite subsystems $J$8 (Arima, 24 Jul 2025).

Non-autonomous conformal iterated function systems require growth restrictions on the level alphabets and control of contraction variability. The cited hypotheses include subexponential growth,

$J$9

balanced systems,

$\mathbb R^d$0

and bounded distortion. Under such conditions Bowen’s formula holds, while counterexamples show the sharpness of the hypotheses: there are perfectly balanced systems with arbitrarily small $\mathbb R^d$1 for which $\mathbb R^d$2 (Rempe-Gillen et al., 2012).

Tree iterated function systems introduce a different combinatorics. Because branches need not line up level by level, the correct pressure substitute is not the ordinary level sum but the minimal sum over maximal antichains: $\mathbb R^d$3 For conformal TIFSs, if $\mathbb R^d$4 then $\mathbb R^d$5, and if additionally

$\mathbb R^d$6

and $\mathbb R^d$7, then $\mathbb R^d$8. Hence

$\mathbb R^d$9

The example in Section 5 of Ono’s paper shows that in the genuine tree setting the antichain optimization is essential (Ono, 28 Feb 2026).

5. Overlaps, projection pressure, and arbitrary subsets

For self-similar systems with overlaps, the classical pressure can fail to capture the geometry directly on the symbolic side. In the affine setting

$\Lambda$0

projection pressure replaces the classical pressure by summing over geometric partitions $\Lambda$1 and taking suprema over fibers of the canonical projection $\Lambda$2: $\Lambda$3 Its variational principle is

$\Lambda$4

where $\Lambda$5 is the projection entropy. If $\Lambda$6 is a compressed orthogonal matrix with $\Lambda$7, then the equation

$\Lambda$8

has a unique solution $\Lambda$9, and

$\dim_H \Lambda=s^*,$0

This formulation works even when the pieces $\dim_H \Lambda=s^*,$1 overlap and does not require open-set or finite-type separation assumptions (Wang et al., 2010).

A different extension treats arbitrary subsets for free semigroup actions by conformal maps. Xiao and Ma define topological pressure $\dim_H \Lambda=s^*,$2 through a Carathéodory–Pesin structure, using Bowen balls associated to semigroup words and the geometric potentials $\dim_H \Lambda=s^*,$3. If

$\dim_H \Lambda=s^*,$4

meaning that points of $\dim_H \Lambda=s^*,$5 have positive lower Lyapunov exponents and satisfy the tempered contraction condition, then there is a unique $\dim_H \Lambda=s^*,$6 such that

$\dim_H \Lambda=s^*,$7

A related formulation gives

$\dim_H \Lambda=s^*,$8

If the Lyapunov exponent is constant, then

$\dim_H \Lambda=s^*,$9

These results transplant Bowen’s formula from compact invariant sets to arbitrary subsets in a semigroup setting (Xiao et al., 2020, Xiao et al., 2020).

6. Complex-dynamical formulations and related usages

In transcendental meromorphic dynamics, Barański–Karpinska–Zdunik define, for $s^*$0 and $s^*$1,

$s^*$2

with lower and upper pressures obtained from $s^*$3. For $s^*$4, a spherical distortion theorem and the notion of GPS points imply that outside a Hausdorff-dimension-zero exceptional set the pressure $s^*$5 is well-defined and independent of $s^*$6. Writing

$s^*$7

Bowen’s formula becomes

$s^*$8

In this setting the target is the radial, or conical, Julia set rather than the full Julia set. Mayer–Urbański’s survey states the same conclusion in the form

$s^*$9

and also records that $\Omega$00 can vary real-analytically in analytic hyperbolic families under appropriate hypotheses (Barański et al., 2010, Mayer et al., 2020).

For rational graph-directed Markov systems, Arimitsu–Jaerisch–Sumi–Watanabe prove Bowen’s formula for the Julia set of a non-elementary, expanding, irreducible, and aperiodic system satisfying the backward separating condition. If $\Omega$01 is the unique zero of the geometric pressure, then

$\Omega$02

The proof uses topological exactness on the skew-product Julia set, density of repelling periodic points, and Ruelle–Walters theory for the geometric potential (Arimitsu et al., 2024).

For nicely expanding rational semigroups, the pressure $\Omega$03 is defined on the skew product using the geometric potential $\Omega$04. If the open-set condition holds, the unique root $\Omega$05 of

$\Omega$06

satisfies

$\Omega$07

Without the open-set condition one has only

$\Omega$08

The same paper also states that this recovers the classical dimension formula for contracting conformal iterated function systems without requiring a cone condition (Jaerisch et al., 2015).

A distinct usage of the name occurs in interval dynamics. For piecewise continuous self-maps of a compact interval, Calderón and Villar-Sepúlveda prove a Bowen orbit-separation formula for topological entropy: $\Omega$09 and this entropy does not depend on the metric generating the topology. This is not a pressure-dimension formula, but it is another standard formula associated with Bowen’s name (Calderón et al., 2024).

In all of these settings, the formula retains the same structural content: the relevant dimension or entropy is characterized by a critical value determined by pressure, separated-orbit growth, or an equivalent thermodynamic quantity. The differences lie in the choice of dynamical space, the form of the pressure, the admissible combinatorics, and the geometric object whose dimension is being measured.