Quantum Extremal Surfaces as Semiclassical Saddles

A quantum extremal surface is a codimension-two surface for which the first shape variation of the complete renormalized generalized entropy vanishes in every independent normal direction. The stationarity equation is semiclassical and geometric. It does not, by itself, choose the dominant saddle, assign an entropy to Hawking radiation, or prove information recovery.

Required background. Quantum Information supplies von Neumann entropy, subsystem or algebra choice, and the distinction between extremizing an entropy functional and selecting a fine-grained state entropy.

Helpful background. Quantum Fields in Curved Spacetime supplies state renormalization; Einstein Equations supplies the mean metric; Black-Hole Geometry and Mechanics supplies normal deformations and classical extremal surfaces; and Semiclassical Gravity fixes the perturbative regime of the saddle equation.

Units. This page restores $\hbar$ in generalized-entropy formulas while keeping $c=k_{\rm B}=1$; setting $\hbar=1$ recovers the site-wide convention.

The quantum extremality equation

Let $X$ be a smooth spacelike codimension-two surface with embedding $X^a(y)$, induced metric $\gamma_{ij}$, and two independent normal directions. Fix a global quantum state, a prescription assigning an outside algebra to each surface, anchoring and homology data when present, and a common regulator-renormalization prescription. The functional is

$$S_{\rm gen}[X] =S_{\rm grav}^{\rm ren}[X]+S_{\rm out}^{\rm ren}[X].$$

For an arbitrary compactly supported normal deformation $\delta X^a=\eta^a$, first variation gives

$$\delta_\eta S_{\rm gen} =\int_X d^{d-2}y\,\sqrt\gamma\, \eta^a(y)\, \mathcal E_a(y), \qquad \mathcal E_a \equiv \frac{1}{\sqrt\gamma}\frac{\delta S_{\rm gen}}{\delta X^a}.$$

The quantum extremality condition is therefore

$$\mathcal E_a(y)=0 \quad\text{for both normal components at every }y.$$

In a null basis $k^a,\ell^a$, this is equivalently $\Theta_k=\Theta_\ell=0$. Engelhardt and Wall introduced the prescription as the quantum deformation of classical area extremality and state its perturbative holographic setting in Engelhardt and Wall 2015, §§ 2.1–3.1, PDF.

Area and entropy variations

In Einstein gravity the gravitational variation is controlled by the area-gradient mean-curvature covector $K_a$. Fix its sign here by $\delta_\eta A=\int_X\sqrt\gamma\,K_a\eta^a$; authors who define the mean-curvature vector with the opposite sign must reverse that symbol throughout. Then

$$\mathcal E_a =\frac{K_a}{4G\hbar} +\frac{1}{\sqrt\gamma} \frac{\delta S_{\rm out}^{\rm ren}}{\delta X^a} +\mathcal E_a^{\rm higher}=0.$$

The higher-curvature contribution follows from the same renormalized action, including the nonstationary surface terms when needed, and cannot be omitted if it is of the retained order. The equation states a balance: classical mean curvature need not vanish because an entropy gradient can displace a nondegenerate classical extremal surface. The displacement is perturbative in $G\hbar$ only when the inverse geometric stability operator is controlled and species enhancement is absent. Conversely, taking $G\hbar\to0$ at fixed smooth entropy variation gives $K_a=0$, the classical extremal-surface equation.

The state and surface are coupled. Moving $X$ changes the outside algebra and its reduced state, so the entropy derivative is generally nonlocal. A finite counterterm can move a local term between the two displayed sectors, but the full $\mathcal E_a$ remains invariant when couplings and entropy subtraction are translated together.

Perturbation theory makes the displacement concrete. Write $X=X_0+G\hbar\,X_1+\cdots$, where $K_a[X_0]=0$. Linearizing the mean-curvature vector gives a Jacobi operator $\mathcal J_{ab}$, and the first correction obeys

$$\mathcal J_{ab}X_1^b =-\frac{4}{\sqrt\gamma} \left.\frac{\delta S_{\rm out}^{\rm ren}}{\delta X^a}\right|_{X_0} +\text{retained local terms}.$$

Boundary conditions specify the admissible homogeneous deformations; they do not automatically remove all zero modes. An invertible $\mathcal J$ on the gauge-fixed, anchored normal deformations gives a unique linear correction. A zero mode requires a solvability condition and can signal bifurcation or a modulus. This order-by-order construction is why a QES should be called a semiclassical saddle only inside the hierarchy that controls $X_1$.

The correction must also remain small compared with every geometric curvature radius.

For a one-direction illustration, let $A(x)=A_0+a_2x^2/2$ and $S_{\rm out}(x)=s_0+s_1x+O(x^2)$ near an area extremum, with $a_2\ne0$. The entropy balance gives $x_*=-4G\hbar s_1/a_2$ to leading order. Here $x$ has length units, $a_2$ is dimensionless in four dimensions, and $s_1$ has units of inverse length, so $x_*$ has the correct dimension. As $a_2\to0$ the estimate ceases to be uniformly small. This is a local Taylor model, not a complete QES: the second normal equation and anchoring data must still be satisfied.

Stationary point, saddle, and selected surface

Vanishing first variation establishes stationarity, not minimality. The Hessian

$$\delta^2 S_{\rm gen}[\eta,\eta] =\int_X\!\int_X \eta^a(y) \,\mathcal H_{ab}(y,y')\,\eta^b(y')$$

may have positive and negative directions. Boundary anchoring, homology, causal admissibility, contour choice, and comparison of competing stationary surfaces are separate parts of a physical prescription. The causal placement and barrier results in Engelhardt and Wall 2015, §§ 4–6, PDF assume perturbative semiclassical bulk dynamics and local signal propagation; they are not generic existence theorems for arbitrary spacetimes.

This distinction matters even in ordinary thermodynamics: an on-shell Euclidean geometry can be a saddle with a negative mode. Likewise, a QES equation can have several branches, no branch, or a branch outside the curvature and fluctuation regime. “Extremize” and “choose the globally relevant saddle” are different operations.

Checks and scientific boundary

Independent check: classical limit. Set the state entropy variation and retained higher-curvature terms to zero. The equation reduces to $K_a=0$, with correct dimension $[K_a]=L^{-1}$. This checks the normalization and the two-normal character of the equation.

Independent check: affine basis. Replacing $k^a\mapsto c k^a$ and $\ell^a\mapsto c^{-1}\ell^a$ rescales the two component equations but leaves their simultaneous zero set unchanged. A claimed surface that solves only one component is quantum marginal, not quantum extremal.

The semiclassical calculation establishes the renormalized variational equation, its state dependence, and its geometric consequences. Once one uses replica saddles to compute the fine-grained entropy of a specified radiation region, minimizes over island branches, or infers a Page curve and reconstruction map, the claim requires additional information-theoretic data. Those steps are treated in Black-Hole Information, and do not follow from the stationarity equation alone.

Exercise

In one normal direction let $S_{\rm gen}(x)=S_0+ax+bx^2/2+cx^3/6$ near $x=0$, with $b\lt 0$. Find the nearby stationary point to leading order when $a$ is small, and decide whether stationarity implies a minimum.

Solution

Solving $dS_{\rm gen}/dx=a+bx+O(x^2)=0$ gives $x_*=-a/b+O(a^2)$. At that point the second derivative is $b+O(a)$, which remains negative for sufficiently small $a$. The surface is stationary but locally maximal in this deformation direction, so quantum extremality alone does not select it as a minimum or dominant saddle.

References

• Engelhardt, N., and Wall, A. C. (2015). “Quantum Extremal Surfaces: Holographic Entanglement Entropy beyond the Classical Regime.” JHEP 01, 073. Open PDF.