Multiple Horizons and Nonequilibrium Temperatures

A static region bounded by several Killing horizons generally does not possess one equilibrium temperature. Each nonextremal horizon supplies a Euclidean regularity period $2\pi/\kappa_i$ relative to the same Killing clock. A single smooth thermal circle can remove every conical defect only when the relevant surface gravities agree, after a common normalization. Equal temperatures are necessary but may not be sufficient: state regularity, gauge potentials, boundary conditions, and backreaction remain separate tests.

Required background. Gravitational Actions and Boundary Terms supplies the boundary data and horizon terms needed to define each thermodynamic problem.

Helpful background. Quantum Fields in Curved Spacetime supplies regular stationary states; Einstein Equations supplies de Sitter and charged geometries; and Black-Hole Geometry and Mechanics supplies the distinct black-hole, cosmological, and Killing horizons.

The two-period obstruction

Consider a static patch

$$ds^2=-f(r)dt^2+\frac{dr^2}{f(r)}+r^2d\Omega_2^2$$

between simple roots $r_b\lt r_c$. The black-hole and cosmological horizons have positive temperature magnitudes

$$\kappa_b=\frac12\lvert f'(r_b)\rvert, \qquad \kappa_c=\frac12\lvert f'(r_c)\rvert, \qquad T_i=\frac{\kappa_i}{2\pi}.$$

After $t=-i\tau$, a neighborhood of horizon $i$ is a plane in polar coordinates only if

$$\tau\sim\tau+\beta_i, \qquad \beta_i=\frac{2\pi}{\kappa_i}.$$

There is only one Euclidean time coordinate. If $\kappa_b\ne\kappa_c$, choosing $\beta=\beta_b$ makes the black-hole tip smooth but leaves a conical defect at the cosmological tip, and conversely. Rescaling the common Killing field multiplies both surface gravities by the same constant, so their ratio and the obstruction are normalization independent.

In Schwarzschild–de Sitter, $f=1-2GM/r-r^2/\ell^2$, the two temperatures are unequal away from limiting cases. Gibbons and Hawking treat the horizon temperatures and the absence of ordinary mutual equilibrium directly Gibbons and Hawking 1977, §§II–IV. The statement is not that either horizon lacks a temperature; it is that one global KMS period cannot represent both.

One can see the inequality without solving a cubic. Eliminating $M$ from $f(r_b)=f(r_c)=0$ gives $L^2=r_b^2+r_b r_c+r_c^2$, where $L=\ell$. Consequently

$$\begin{aligned} \kappa_b&=\frac{(r_c-r_b)(2r_b+r_c)}{2L^2r_b},\\ \kappa_c&=\frac{(r_c-r_b)(r_b+2r_c)}{2L^2r_c},\\ \kappa_b-\kappa_c &=\frac{(r_c-r_b)^2(r_b+r_c)}{2L^2r_b r_c}\gt0 . \end{aligned}$$

For two distinct positive roots the black-hole horizon is therefore hotter in this common normalization. Equality arises only in the coalescing limit, where the static coordinates degenerate and a separate near-horizon scaling is needed. Charge can change this conclusion, as the lukewarm example below illustrates.

Results and open questions through 2026-08-19

The following matrix separates geometric regularity, state existence, and nonequilibrium interpretation at the evidence cutoff 2026-08-19.

Background or formulation	Evidence or result	Scope and limitation
Pure de Sitter static patch, Gibbons and Hawking 1977, §IV	One cosmological horizon fixes one period, and the regular Euclidean/Bunch–Davies Hadamard state is thermal for the central static observer	The static generator and observer normalization are specified; de Sitter-invariant non-Hadamard alpha vacua are not included, and this does not solve a two-temperature problem
Generic Schwarzschild–de Sitter, Gibbons and Hawking 1977, §§II and V and Teitelboim 2002, §1, PDF	The black-hole and cosmological surface gravities differ, so one Euclidean period cannot smooth both tips	Establishes an equilibrium obstruction for the common static clock, not the full time-dependent backreaction
Tuned charged “lukewarm” families, Romans 1992, §3.2 and §5, PDF	Parameters can satisfy $\kappa_b=\kappa_c$, removing the elementary two-period obstruction	Equality is a tuned condition and does not alone prove a unique globally regular interacting state or stability
Horizon-by-horizon constrained thermodynamics, Teitelboim 2002, §§2–4, PDF	One horizon is treated as the thermodynamic system while data at the other serve as boundary conditions	Produces consistent first laws for a selected region, but it is not one unconstrained global canonical ensemble
Degenerate or Nariai limit, Gibbons and Hawking 1977, §II and Romans 1992, §3.1, Eqs. (3.9)–(3.13), PDF	Coalescing roots require a rescaled near-horizon limit and can yield a finite limiting temperature	The limiting geometry differs from simply substituting a double root into the original static coordinates

Teitelboim formulates Schwarzschild–de Sitter thermodynamics by choosing which horizon is the system and which data are held fixed Teitelboim 2002, §§2–4, PDF. This supports a conservative position: well-defined horizon thermodynamics need not imply one global equilibrium ensemble. The more permissive position points to tuned equal-temperature solutions and special near-horizon limits as legitimate equilibrium saddles, but only within their declared parameter branch.

Equal temperature is not the whole state problem

A Hartle–Hawking-type state must be regular across the relevant future and past horizon branches and stationary under one automorphism group. The Kay–Wald analysis shows how strong uniqueness and thermal conclusions depend on a bifurcate Killing horizon, quasifree field, nonsingularity, and symmetry hypotheses Kay and Wald 1991, Parts I–II. It does not supply a universal state for arbitrary pairs of horizons.

For charged or rotating horizons, equilibrium also requires matching chemical potentials in the same gauge and frame. Equal $\kappa$ with unequal $\Phi_{\mathrm H}$ or $\Omega_{\mathrm H}$ still drives charge or angular momentum exchange for fields carrying the corresponding quantum numbers. Superradiant modes can obstruct a naive Gibbs construction even when a formal Euclidean metric is smooth.

State choice matters as well. A state regular at one future horizon may carry flux through the other. The renormalized stress tensor can then drive the parameters, so a stationary two-temperature background is at best an instantaneous approximation. “Both horizons radiate” is not by itself a closed backreaction solution: conservation, greybody transmission, quantum state, and motion of both horizons must be solved together.

Competing nonequilibrium interpretations

One interpretation treats the unequal temperatures as two reservoirs coupled through the static region. It expects a net flux, entropy production, and slow evolution toward another geometry. A second treats one horizon plus a constrained outer boundary as the system, declining to assign a global heat flow without a specified quantum state. A third uses conically singular or constrained Euclidean saddles as off-shell thermodynamic probes. These positions answer different questions and need not contradict one another.

The decisive axes are the background branch, matter model, quantum state, boundary condition, and backreaction scale. A free-field flux on a fixed Schwarzschild–de Sitter geometry does not establish the endpoint of the coupled semiclassical Einstein equation. Conversely, the absence of a globally smooth Euclidean section does not erase the local horizon temperatures measured by the corresponding stationary observers.

Limitations and open questionss

The sources assessed through the stated cutoff provide no universal theorem that equality of all horizon intensive potentials is sufficient for a globally regular, stable interacting state, and no controlled general evolution turns every unequal-temperature static patch into a unique endpoint. The sufficiency question and the generic endpoint therefore remain unresolved. The strongest current conclusion is conditional: a common smooth Euclidean equilibrium requires compatible periods and chemical potentials; otherwise the system must be constrained horizon by horizon or treated out of equilibrium.

The conclusions could change if a state-existence theorem covers interacting fields on multi-horizon geometries, if an independent semiclassical backreaction calculation determines the late-time branch for generic data, if a stability analysis invalidates a proposed lukewarm saddle, or if a revised contour prescription changes which conical or complex saddles contribute. Any of those events would change a different layer of the conclusion.

Exercise

Suppose two horizons generated by the same Killing field have $\kappa_1=0.10/L$ and $\kappa_2=0.08/L$. Can a constant rescaling of the generator produce one smooth Euclidean period for both? What additional check would still be needed if the surface gravities were equal?

Solution

No. A common rescaling multiplies both values by the same constant and leaves $\kappa_1/\kappa_2=1.25$, so the required periods remain unequal. If they were equal, one would still have to construct a quantum state regular on all relevant horizon branches and match any angular or electric chemical potentials, boundary conditions, and stability requirements. Equal temperatures remove only the elementary conical obstruction.

References

• Gibbons, G. W., and S. W. Hawking (1977), “Cosmological Event Horizons, Thermodynamics, and Particle Creation,” Physical Review D 15, 2738–2751. DOI.
• Kay, B. S., and R. M. Wald (1991), “Theorems on the Uniqueness and Thermal Properties of Stationary, Nonsingular, Quasifree States on Spacetimes with a Bifurcate Killing Horizon,” Physics Reports 207, 49–136. DOI.
• Romans, L. J. (1992), “Supersymmetric, Cold and Lukewarm Black Holes in Cosmological Einstein–Maxwell Theory,” Nuclear Physics B 383, 395–415. DOI. Open PDF.
• Teitelboim, C. (2002), “Gravitational Thermodynamics of Schwarzschild–de Sitter Space,” arXiv:hep-th/0203258. Open PDF.