Universal scaling framework for parameterized quantum evolutions at criticality

AI summary

Overview

• Research area: Quantum many-body physics and variational quantum algorithms — specifically how well layered parameterized quantum evolutions can represent critical ground states, framed through renormalization-group (RG) and finite-resource scaling ideas borrowed from tensor-network theory.
• Technical level: Advanced. The paper assumes familiarity with RG, conformal field theory (CFT), finite-size scaling, tensor-network bond-dimension scaling, variational quantum eigensolvers (VQE), and Hamiltonian Variational Ansätze (HVA). The conceptual message is accessible, but the machinery is not.
• Scope (one sentence): The paper introduces a universal, ansatz-independent framework that assigns each finite-resource variational ansatz an emergent correlation length ξ_D growing as ξ_D ∝ D^κ, and uses the exponent κ to compare several parameterized quantum evolutions targeting the critical one-dimensional transverse-field Ising model (TFIM).

What This Paper Is About

Variational ansätze approximate complicated quantum ground states with limited resources, but at a critical point correlations extend across all length scales, so finite resources (such as a finite circuit depth D) inevitably cut off how far faithful correlations can reach. The authors ask a comparative question that existing studies — which focused on circuit-specific causal cones, finite-size performance, or particular hardware — could not answer: which variational architectures convert a given amount of resource into long-distance critical correlations most efficiently? Their answer is to import the finite-resource scaling perspective of tensor networks, define an emergent correlation length ξ_D for any ansatz, and measure its growth ξ_D ∝ D^κ as a single common figure of merit.

Key Contributions

1. A universal scaling framework for parameterized quantum evolutions. Building on finite-resource scaling from tensor-network theory, the authors assign every finite-resource ansatz an emergent correlation length ξ_D — the longest range over which it faithfully captures critical correlations — and define the efficiency exponent κ through ξ_D = A_ξ D^κ (with A_ξ a non-universal amplitude). Logarithmic growth ξ_D ∼ log D is flagged as inefficient; algebraic growth means polynomial resources suffice.
2. Two independent, cross-checking extraction protocols. κ is extracted (i) from the chord-corrected two-point spin–spin correlation function, whose envelope decays as exp(−r/ξ_D) beyond the critical power law, and (ii) from the variational energy error δE_abs(L,D) = ⟨ψ_var|H_T|ψ_var⟩ − E_0(L), using conformal perturbation theory with the appropriate logarithmic correction for Δ_Φ = 1. A bulk-energy-density criterion, L²δe_bulk compared against the universal CFT value πcv/24, is introduced to filter out finite-size-dominated data before the energy fit.
3. A systematic comparison across two design axes. The ansätze compared differ in interaction structure (nearest-neighbor versus long-range generators, with exponential "EXP" kernels f_λ(r) = λ^{r−1} and power-law "POW" kernels f_α(r) = r^{−α}) and in layer organization (separable HVA sub-layers versus combined single-exponential layers). Range parameters η_d are optimized independently at every layer, with gradients taken exactly via the Fréchet derivative.
4. A microscopic, quasiparticle interpretation of ξ_D. Analyzing optimized states in the fermionic quasiparticle basis of the target Hamiltonian, the authors show that finite depth leaves an unresolved window of width ξ_D^{−1} around the low-energy gap-closing modes responsible for long-distance correlations, recasting ξ_D as an infrared resolution scale and the variational problem as resolving the non-analytic occupation step of the critical ground state.

Main Findings

• All ansätze are compatible with algebraic growth, but exponents differ widely. Within the explored depth and system-size windows, every ansatz fits ξ_D ∝ D^κ, with fitted exponents ranging from κ ≃ 1 to κ ≃ 3 — a substantial spread in how efficiently architectures convert depth into long-range correlations.
• Exponential long-range interactions give the largest exponents. The EXP kernel family produces the largest fitted κ values, reaching the upper end (κ ≃ 3) of the observed range.
• Power-law long-range support alone confers no advantage. POW-kernel interactions remain much closer to nearest-neighbor behavior, showing that merely having long-range terms in the generator does not by itself improve critical-state scaling.
• Layer organization matters alongside interaction range. Combined-generator layers, which fold non-commuting generators into a single exponential, generally yield larger fitted exponents than their separable HVA counterparts.
• Benchmark case, TFIM HVA (separable, nearest-neighbor): from the correlation function at chain length L = 512, fitting over the bulk window r/L ∈ [5%, 40%] with binned maxima to isolate the smooth envelope, gives κ_C = 1.03 ± 0.03 (jackknife standard error over fitted depths); the deepest circuit is excluded from the fit as it falls outside the observed scaling trend.
• Independent energy-based estimate agrees. The energy-error analysis for the same ansatz yields κ_E = 1.06, consistent with κ_C, supporting the framework's central assumption that both observables are governed by the same emergent finite-depth correlation length.
• The two probes are mutually consistent and diagnostically useful. The correlator scaling function at fixed r/L = 0.18 collapses onto the same scaling variable extracted from the energy, and the bulk energy density shift collapses after rescaling in the finite-depth regime.
• Symmetry forbids the order parameter as a probe. Because both the generators and the initial state (the ℤ₂-even fully polarized state ⊗|↑⟩_n) respect the ℤ₂ spin-flip symmetry generated by P = ∏σ_n^z, ⟨σ_n^x⟩ = 0 identically; the leading allowed perturbation is therefore the ℤ₂-even energy operator ε with Δ_ε = 1, giving ν^{−1} = d − Δ_ε = 1 and placing the energy scaling in the logarithmically corrected case Δ_Φ = 1. The two-point correlator ⟨σ_m^x σ_n^x⟩ is used instead.
• Oscillations in the correlator are non-equilibrium artifacts. The chord-corrected correlation function shows oscillations not captured by asymptotic CFT scaling, attributed to the non-equilibrium structure of the optimized finite-depth state; binned maxima are used to extract the leading envelope.

Methodology in Plain English

The authors pick a target problem whose exact answer is known: the one-dimensional transverse-field Ising model at its critical point J/h = 1 with open boundary conditions, which is integrable (exact ground-state energy given analytically) and described at low energy by the Ising CFT with central charge c = 1/2. They build trial states by stacking D layers of unitaries, each layer being an exponential of a small set of Hermitian generators (transverse field, Ising coupling, or fermionic hopping/pairing terms), applied either as separate sub-layers (HVA style) or merged into one exponential. The total parameter count grows only as O(D). After a Jordan–Wigner transformation everything becomes quadratic, so a state on a chain of length L is fully specified by a 2L × 2L Dirac correlation matrix — meaning state evolution, observables, gradients, and the Fubini–Study metric can all be computed numerically exactly at large L, with no sampling noise. They then optimize the parameters, measure how correlations decay with distance, factor out the known critical power law using the chord distance d_chord(r,L) = (L/π)sin(πr/L), fit the leftover exponential envelope to get ξ_D at each depth, and finally fit ξ_D against D to read off κ. Because a finite ξ_D and a finite L compete as infrared cutoffs, they only trust the asymptotic scaling forms when L ≫ ξ_D, which is why they use the bulk energy density shift as a filter to discard data still dominated by finite-size effects.

Why This Matters

• Impact on research: The framework gives the variational-ansatz community a shared, architecture-agnostic language — borrowed from RG and tensor-network scaling — for asking "how efficient is this ansatz at criticality?" instead of comparing individual hardware demonstrations or causal-cone arguments. Because κ depends only on the emergent infrared scale, it can in principle be defined for other resources too, such as tensor-network bond dimension.
• Practical design guidance: The results argue against the intuition that long-range couplings automatically help; exponential kernels and combined-generator layers were the productive choices, while power-law kernels behaved close to nearest-neighbor circuits. That is directly actionable for ansatz design.

Real-world applications (as motivated by the paper's framing):

• Benchmarking and comparing candidate ansätze for critical-state preparation on near-term quantum simulators before committing to hardware.
• Guiding resource estimates for hybrid digital–analog quantum simulation platforms, where hardware-native many-body interactions are interleaved with on-site control.
• Informing ansatz choice on fermionic analog and hybrid platforms capable of long-range quadratic interactions, which the authors connect to the long-range fermionic generator family.
• Providing a diagnostic (the infrared resolution scale ξ_D^{−1}) for interpreting what a finite-depth experiment actually resolves near a critical point.

Industry relevance: Quantum-computing hardware and quantum-simulation efforts need principled answers to "how deep must the circuit be?" and "does adding long-range connectivity pay off?" This paper supplies a quantitative, cheap-to-evaluate figure of merit (κ) and an experimentally computable filtering criterion for when finite-size effects mask the answer.

Future Directions

• Extending the framework beyond the exactly solvable quadratic TFIM to interacting models and other universality classes, where the same ξ_D ∝ D^κ definition would need numerical state-vector or tensor-network evaluation rather than free-fermion correlation matrices.
• Establishing whether any architecture can beat algebraic growth (faster-than-algebraic scaling is stated not to be expected generically) and mapping which generator properties — as opposed to mere interaction range — drive the observed spread from κ ≃ 1 to κ ≃ 3.
• Understanding the failure of the deepest circuit in the TFIM HVA correlation fit, which the authors attribute to possible finite-size contamination, reduced envelope-extraction reliability as ξ_D approaches available separations, or convergence to a suboptimal local minimum.
• Turning the infrared-resolution picture into a design principle: using the unresolved window of width ξ_D^{−1} around gap-closing modes to target ansatz structure at the modes that actually limit critical-state preparation.
• Applying the bulk-energy-density filter and the two-probe (correlator and energy) consistency check as standard benchmarks for critical-state preparation on real devices.

Target Audience

Quantum many-body theorists, researchers in variational quantum algorithms and hybrid digital–analog quantum simulation, and tensor-network practitioners interested in transferring finite-entanglement scaling ideas to circuit-based ansätze. It is also relevant to quantum-computing experimentalists and hardware architects who need resource-efficiency arguments for circuit depth and connectivity, though the RG/CFT formalism makes it a demanding read for non-specialists.

Note on completeness: the supplied text is truncated during the discussion of Fig. 3 and does not report the fitted exponent values for the Kitaev, combined, EXP, and POW ansätze individually, the exact depth and system-size windows used in the fits, the optimized kernel parameters, or the full contents of Appendices B and C; those figures are therefore not stated here.

Authors’ abstract

Variational ansätze are a cornerstone of quantum many-body physics, providing compact approximations to complex ground states using finite resources. Recent quantum-technology advances have introduced a new class based on layered parameterized evolutions. Assessing whether they can represent critical ground states is challenging: correlations span all length scales, while finite circuit depth limits how far they extend. Building on finite-resource scaling from tensor networks, we assign each ansatz an emergent correlation length $ξ_D$, the longest range over which it faithfully captures critical correlations. Its growth with refinement parameter $D$, $ξ_D \propto D^κ$, defines an exponent $κ$ measuring how efficiently an architecture converts resources into long-distance correlations. Applying this framework to the critical transverse-field Ising model, with $D$ the circuit depth of parameterized evolutions, we compare ansätze with nearest-neighbor and long-range generators, and layers where generators act separately or combined. All ansätze are compatible with algebraic growth, but fitted exponents range from $κ\simeq1$ to $κ\simeq3$. Exponential interactions give the largest exponents, while power-law interactions stay close to nearest-neighbor behavior, showing long-range support alone gives no scaling advantage. Combined-generator layers generally outperform separable ones, so layer organization matters alongside interaction range. Finally, a quasiparticle analysis shows finite depth leaves an unresolved window of width $ξ_D^{-1}$ around the low-energy modes responsible for long-distance correlations. The emergent correlation length thus acts as an infrared resolution scale, providing a benchmark for critical-state preparation and a guide for designing resource-efficient variational architectures.