Can a quantum spin chain encode black hole radial geometry?

One chain for every planar black brane

The object being encoded is not the full spacetime but a single radial differential equation. For a scalar field of mass m dual to a boundary operator of dimension Δ in an AdS d+1 black brane, the static homogeneous mode obeys an equation built from the operator L = −w² d/dw[(1−w) d/dw], where w runs from 0 at the boundary to 1 at the horizon. The key structural fact is that L carries no explicit d dependence; dimensionality enters only through how w and h = Δ/d relate to physical quantities.

The paper turns L into a lattice Hamiltonian by choosing an orthonormal basis ψn = √(2n+3) w² Pn(0,2)(2w−1), built from Jacobi polynomials. In that basis L is tridiagonal, so it becomes a chain with on-site energies εn = (n+1)(n+2)/2 and nearest-neighbor hoppings tn+1 = (n+1)(n+2)(n+3)/[2√((2n+3)(2n+5))]. Identifying site n with a single spin flip on a fully polarized vacuum gives a number-conserving spin chain whose one-magnon sector is exactly the radial problem. Because the same Hamiltonian serves every planar AdS d+1 Schwarzschild geometry, the construction is dimension-independent by design rather than by approximation.

The radial coordinate becomes a quasimomentum

The chain does not store the radial coordinate as a site label. Instead, a fixed radial position w appears as an oscillation across sites. Parameterizing w = cos²(q/2) with q in (0, π), the large-n asymptotics of the Jacobi polynomials give ψn(w) ≈ √(2/π) cos^{3/2}(q/2) sin^{−1/2}(q/2) cos[(n+3/2)q − π/4]. The horizon sits at q → 0 and the AdS boundary at q → π, so the exterior geometry is read off from a lattice quasimomentum.

This dictionary is what makes the encoding measurable rather than merely formal. The paper defines a coarse-grained, normalization-independent ratio Gr(w) of spin coherences Cij = ⟨σ+i σ−j⟩ over a block of sites, and shows Gr = cos(rq). The nearest-neighbor case r = 1 then fixes w = (1+G1)/2, and the exterior proper distance follows as ℓ(G1) = (2R/d) sinh⁻¹√[(1−G1)/(1+G1)]. In other words, a local spin-correlation measurement plus the external parameters d, R and horizon position z0 is enough to reconstruct the radial metric profile written in the paper. The endpoints G1 = 1 and G1 = −1 correspond to the horizon and the AdS boundary.

Analytic continuation reaches the singularity

Continuing the quasimomentum to q = iκ with κ ≥ 0 pushes the dictionary past the horizon, where w = cosh²(κ/2) and κ → ∞ approaches the singularity. Inside the horizon the radial direction is timelike, and the paper computes the proper time from horizon to singularity as τs = πR/d. This is a derived geometric result, not a hardware measurement, and it matches the value obtained previously from thermal one-point functions.

The same chain also reproduces a stability threshold. The spectrum of the lattice Hamiltonian is [1/4, ∞), and a homogeneous scalar solution corresponds to energy E = −h(h−1). Real h requires E ≤ 1/4, with the lower continuum edge reached at h = 1/2, that is Δ = d/2. Combined with m²R² = Δ(Δ−d), this gives m²R² = −d²/4, the Breitenlohner-Freedman bound in AdS d+1. The spectral edge of the spin chain and the gravitational stability bound are therefore the same object, which is a stronger statement than numerical agreement on a few eigenvalues.

Screening masses and exact inverse-moment sum rules

For a massless scalar with transverse momentum, the radial equation becomes a generalized eigenvalue problem, 4H|φj⟩ = Λj W^{2/d}|φj⟩ with Λj = (Mj/2πT)², where W is multiplication by w and Mj is a screening mass. Dimensionality now enters only through the power W^{2/d}, so one lattice operator pair covers all d. Rewriting this as an ordinary eigenvalue problem for Kd = (4H)^{−1/2} W^{2/d} (4H)^{−1/2} turns inverse spectral moments into traces, giving Σj (2πT/Mj)^{2p} = Tr Kd^p.

The first inverse moment evaluates exactly to [ψ(2/d) + γE]/[4(2/d − 1)] for d ≠ 2, which for d = 4 collapses to the simple relation Σj (2πT/Mj)² = log 2. The paper reports a second moment for d = 4, Σj (2πT/Mj)⁴ = 2 log²2 − π²/12. These are the screening-mass sum rules the abstract refers to, and the author states that they have not been established previously. Three checks anchor them: the d = 4 lowest eigenvalue converges to Λ0 ≈ 2.8969, giving M0/πT ≈ 3.4041, matching the known lowest massless-scalar screening mass in strongly coupled N = 4 supersymmetric Yang–Mills plasma; the d = 2 case reproduces the exact nonrotating BTZ spectrum Mj/πT = 4(j+1); and the large-j spacing matches the known WKB asymptotics of the continuum problem.

A 32-site run recovers proper time within 8%

The experiment truncates the chain to N = 32 sites and runs it on the 156-qubit IBM ibm_marrakesh superconducting processor. Preparing (|Ω⟩+|0⟩)/√2 uses the fully polarized vacuum as a phase reference, because number-conserving evolution leaves it invariant; the source propagator is then read from single-qubit observables as Un0(τ) = ⟨Xn(τ)⟩ + i⟨Yn(τ)⟩. The propagator is sampled at 81 times with Δτ = 0.05 up to τ = 4.0, with 4096 shots per circuit, across 162 response circuits submitted in nine jobs; the maximum transpiled depth was 344 and the maximum two-qubit gate count 992.

The measured real-time propagator is converted into a static radial response by a finite-time transform, and the thermal one-point coefficient c(h) is extracted by fitting over sites 8 ≤ n ≤ 22, a window chosen to sit in the asymptotic regime while avoiding the largest sites where finite-chain and hardware deviations are worst. The headline number comes from the unwrapped phase of R(h) along h = x − 0.2i, fitted linearly over 4 ≤ x ≤ 8: τs/R = 0.847 ± 0.017 against the exact π/4 ≈ 0.785, a relative deviation of about 7.8%.

That error bar covers finite-shot statistics only. The paper states explicitly that systematic hardware, finite-size, finite-time and reconstruction errors are not included, and that the deviation from the exact value exceeds the statistical uncertainty, so systematic contributions are present. The reconstructed c(h) agrees with the exact black brane result but deteriorates beyond the first pole at h = 2, where the reconstruction is most sensitive to those same effects. The author also notes that this implementation does not provide a quantum computational advantage; its role is to establish hardware realizability and to serve as a platform for interacting extensions.

At finite filling the low-energy theory is a c = 1 CFT

Second quantization changes what the spectral edge means. Under a Jordan–Wigner transformation the chain becomes an inhomogeneous free-fermion model, and adding a chemical potential shifts the one-particle spectrum to Eμ(λ) = 1/4 + λ² − μ. Occupation begins at μc = 1/4, where excitations disperse quadratically, so the same edge that encodes the BF bound in the one-particle problem becomes a z = 2 many-body quantum critical point.

At fixed filling ν = Nf/N, with u = n/N and a chemical potential scaling as μN = a²N², the semiclassical one-particle energy is E(u,q) = u² cos²(q/2). The local Fermi surface fixes a radial profile ρF(u) = 2 arcosh(u/a) for u ≥ a, with the region u < a completely filled. Linearizing about the Fermi surface away from the turning point u = a gives a c = 1 free-fermion CFT, and because the local Fermi velocity vF(u) = a√(u²−a²) varies along the chain, the natural coordinate is ξ(u) = (1/a) arcosh(u/a) = ρF(u)/(2a). Since the black brane proper radial distance is ℓ = Rρ/d, this gives ξ = dℓ/(2aR): the CFT's spatial coordinate is proportional to proper radial distance. The paper flags that the gravitational interpretation of this CFT remains to be understood, and that u = a is where the local linearization breaks down.

A fixed-geometry dictionary, not emergent gravity

The construction describes scalar fields in a fixed black brane geometry. It does not encode gravitational backreaction: making the encoded geometry respond dynamically to the matter sector would require an interacting or coupled quantum system beyond the fixed radial chain, and the paper calls whether even a restricted form of backreaction can be reproduced an open question. The work also focuses on static observables; finite-frequency bulk modes, transport coefficients and frequency-dependent correlators are named as a next step rather than delivered here.

The most decision-relevant comparison inside the paper is between what is exact and what is measured. The dimension-independent Hamiltonian, the BF bound at the spectral edge, the thermal one-point coefficient c(h), the screening sum rules including Σj (2πT/Mj)² = log 2 at d = 4, and the interior proper time τs = πR/d are analytic results checked against known continuum values. The hardware contribution is narrower: a 32-site truncation whose phase fit gives τs/R = 0.847 ± 0.017 versus π/4, with systematics excluded from the error bar. Readers weighing the claim should treat the exact lattice dictionary and the sum rules as the paper's durable contribution, and the processor run as evidence of realizability at a precision currently limited by systematic rather than statistical error.

Sources used in this answer