Quantum Computing Solution of the Bethe-Salpeter Equation for Relativistic Scalar Bound States via Tensor-Network VQE

AI summary

Overview

Research area: Quantum computing applied to quantum field theory — specifically gate-based variational quantum simulation of the relativistic two-body bound-state problem (Bethe–Salpeter equation).

Technical level: Advanced. The paper assumes familiarity with the Bethe–Salpeter equation, Wick rotation, O(4) partial-wave projection, Pauli-string decompositions, VQE, and matrix-product states.

Scope (one sentence): The paper reduces the homogeneous Bethe–Salpeter equation for two massive relativistic scalars in the ladder approximation to a symmetric matrix eigenvalue problem of dimension N = 2^n, solves it with an MPS-ansatz VQE at N = 16 (n = 4 qubits), and then argues — as an explicitly stated negative result — that the problem sits in a classically tractable regime over the tested sizes.

What This Paper Is About

The Bethe–Salpeter equation is the relativistic, field-theoretic analogue of the bound-state Schrödinger equation, and solving it numerically is expensive, especially beyond the ladder approximation or directly in Minkowski space. The author asks whether a gate-based quantum eigensolver can handle it, using the simplest non-trivial case: two identical massive scalars exchanging a massive scalar in the ladder approximation. The goal is to build the full pipeline — Euclidean reduction, discretization, Pauli decomposition, tensor-network VQE — benchmark it against classical diagonalization, and then honestly assess whether any quantum advantage is plausible for this problem.

Key Contributions

1. A complete gate-based pipeline for the covariant BSE. The homogeneous BSE is Wick-rotated to Euclidean space, projected onto the O(4) S-wave channel, discretized with Gauss–Legendre quadrature into a real symmetric matrix eigenvalue problem, and recast as an n-qubit Pauli Hamiltonian suitable for VQE.

2. A benchmarked VQE solution with an MPS tensor-network ansatz. At N = 16 (n = 4 qubits) the VQE reproduces the classically computed maximum eigenvalue — which encodes the minimum coupling for binding — to better than 1% mean relative error, with the best grid point (L, η) = (2, 0.01) reaching 4.8 × 10⁻³.

3. The first entanglement quantification of the BSE amplitude in qubit encoding (to the author's knowledge). The Schmidt spectrum is analyzed across bipartitions, showing low, area-law-like entanglement over the tested sizes; bond dimension χ = 2 already reproduces more than 99% of the ground-state fidelity.

4. A critical, explicitly negative assessment of quantum advantage. Three independent barriers are analyzed — exponential Pauli overhead, approximately size-independent low entanglement, and decreasing VQE gradient scales for the tested ansatz (consistent with generic barren-plateau concerns at larger n) — leading to the conclusion that the studied problem is classically tractable and that genuine advantage would require physically motivated extensions such as 2D Minkowski-space BSE, N-body bound states, or non-ladder kernels.

Main Findings

• VQE accuracy at N = 16: Mean relative error in the maximum eigenvalue is below 1% versus classical diagonalization, with the best grid point (L, η) = (2, 0.01) giving 4.8 × 10⁻³, after convergence in fewer than 300 ADAM steps.

• Low, area-law-like entanglement: The von Neumann entropy of the BSE amplitude stays far below its maximal value across the tested sizes, and an MPS bond dimension of χ = 2 already reproduces more than 99% of the ground-state fidelity — the empirical justification for the MPS ansatz.

• Exponential Pauli overhead: For the n = 4 example, exactly (4⁴ + 2⁴)/2 = 136 of the 4⁴ = 256 Pauli strings have nonzero coefficients, a fraction of 0.531. The general count is (4^n + 2^n)/2, because real-symmetric Hamiltonians eliminate only the odd-Y sector, leaving I/X/Z strings plus even-Y combinations.

• Shallow depth suffices: A depth scan over L = 1, …, 8 shows the dominant improvement occurs between L = 1 and L = 2 (Δλ_max ≈ 9.7 × 10⁻⁵ for N = 16), while further layers change λ_max only at the 10⁻⁷ level. The production setting is L = 2, giving 12 variational parameters at n = 4.

• Classical tractability (the central negative result): Over the tested range the problem is well described by low-bond-dimension tensor networks and efficiently handled by MPS/Lanczos methods, so the quantum workflow does not outperform classical diagonalization here.

• Wick–Cutkosky benchmark agreement: In the deep-bound limit (M/m = 0.001, μ → 0, N = 128), the discretized calculation gives α₁ = 1.9949 against the exact value α₁^WC = 2, a relative error of 0.26%. Eigenvalue ratios α_n/α₁ match the exact O(4) pattern n(n+1)/2 with deviations below 2% for n ≤ 4 at N = 128.

• Massive exchange weakens binding: At N = 64 and p_max = 25m, the ratio α_min(μ = 0.15)/α_min(μ → 0) ranges from 1.038 to 1.056 across the tabulated masses, i.e. a massive exchange boson requires 3.8–5.6% more coupling than the massless case.

• Single-channel saturation near threshold: The exact angle-resolved Wick–Cutkosky coupling α₁ = 2/(1 − η²) diverges as η = M/2m → 1, but the angular-averaged single-channel formulation instead saturates to a finite value, because it omits the ℓ = 2, 4, … partial waves that drive the divergence.

• Classical reference values at N = 16 (μ/m = 0.15): M/m = 1.2, 1.4, 1.6, 1.8 correspond to λ_max = 2.4325 × 10⁻³, 2.2563 × 10⁻³, 2.0835 × 10⁻³, 1.9178 × 10⁻³ and g²_min = 411.09, 443.21, 479.97, 521.44. The amplitude at M/m = 1.6 rises from near zero, peaks near p ≈ 1.3m, then decreases monotonically.

Methodology in Plain English

The author starts from the full four-dimensional relativistic bound-state equation and simplifies it in stages until it becomes something a quantum computer can address.

First, the equation is rotated from Minkowski to Euclidean space (a standard trick that turns oscillatory integrals into well-behaved ones) and averaged over the four-dimensional angular sphere S³, keeping only the lowest, spherically symmetric partial wave (ℓ = 0). This reduces a 4D integral equation to a one-dimensional radial equation depending only on the momentum magnitude p.

Second, that radial integral is replaced by a finite sum using Gauss–Legendre quadrature on the interval [0, p_max] (p_max = 20m for the main discretization, 25m for the Wick–Cutkosky comparison table). Absorbing the quadrature weights into the wavefunction and applying a further rescaling by the inverse propagator makes the problem a real, symmetric matrix eigenvalue problem of dimension N. Choosing N = 2^n lets the eigenvector be stored directly in the amplitudes of an n-qubit register.

Third, the matrix is expanded in the n-qubit Pauli basis, and its sign is flipped so that its largest eigenvalue becomes the lowest eigenvalue of the VQE Hamiltonian — minimizing the energy then yields −1/g²_min, the inverse of the minimum coupling needed for binding.

Fourth, the trial state is a deliberately shallow, real-amplitude circuit: Hadamard gates on every qubit to create a uniform superposition, followed by L sweeps of nearest-neighbor blocks consisting of a CNOT and two R_Y rotations. The gate set is restricted to R_Y and CNOT because the target eigenvector can be chosen real. The parameter count is 2(n−1)L. Optimization uses the parameter-shift rule for exact gradients, random initialization in [−π, π] with a fixed seed, and ADAM with step size 0.03, at most 1000 steps, and an early-stopping tolerance of 10⁻⁸ after 80 burn-in iterations. Production results use L-BFGS-B on the exactly evaluated Rayleigh quotient (a noiseless, classically optimized version of the same objective), with the PennyLane VQE implementation used as validation.

All simulations run on PennyLane's default.tensor backend in MPS mode with the quimb library, with a maximum internal bond dimension χ_max = 16 — far above what n = 4 requires, so the representation does not artificially truncate the state. No quantum hardware is used anywhere in the paper; the Pauli-measurement overhead discussed in the complexity analysis is a prospective cost for future sampling-based hardware, not a cost of these simulations.

Why This Matters

Impact on research. This is, to the author's knowledge, the first gate-based VQE treatment of the covariant Bethe–Salpeter equation, complementing earlier work that solved the BSE on a D-Wave annealer via a QUBO reformulation and light-front VQE studies of pion structure on IBM hardware. It establishes a reusable Pauli-Hamiltonian framework that future BSE variants can build on, and it supplies the first entanglement quantification of the BSE amplitude in qubit encoding. Its most distinctive contribution is arguably the negative result: rather than claiming advantage, it identifies precisely which structural features (exponential Pauli overhead, size-independent low entanglement, shrinking gradient scales) place this problem in the classically tractable regime — information that is directly useful for deciding where to invest quantum-algorithm effort in nuclear and particle physics.

Real-world applications (domains where the BSE is the working tool):

• Positronium and hydrogen spectroscopy in quantum electrodynamics.
• Meson and baryon structure in quantum chromodynamics, including covariant quark–diquark treatments of the nucleon in ladder approximation.
• Exciton physics in condensed matter, relevant to semiconductors and optoelectronic materials.
• General relativistic bound-state problems beyond the ladder approximation and directly in Minkowski space.

Industry relevance. For gate-based quantum-computing vendors and algorithm developers, the paper is a concrete case study in matching ansatz structure (MPS, area-law entanglement) to problem structure, and in the practical use of tensor-network simulators (PennyLane, quimb) as development backends. For the semiconductor and materials modeling sector, the exciton application of the BSE connects the method to problems of commercial interest. For the high-performance computing and scientific software sector, the result clarifies that classical MPS/Lanczos solvers remain the right tool at these problem sizes.

Future Directions

• Two-dimensional Minkowski-space BSE. Working directly in Minkowski space removes the Wick-rotation simplification and is identified as a physically motivated setting where genuine quantum advantage may become plausible.
• N-body bound states. Extending from two constituents to few-body and many-body bound states enlarges the Hilbert space in a way that may outrun classical methods.
• Non-ladder kernels. Going beyond the ladder approximation adds genuinely new physics and additional entanglement structure; whether the observed area law survives this is an open question.
• Entanglement and gradient scaling at larger n. The paper's entanglement analysis covers only the tested sizes, and the observed decreasing gradient scales are consistent with barren-plateau concerns; a systematic study of both, plus a rigorous layer-depth convergence study (the choice L = 2 is explicitly described as pragmatic rather than optimized), is needed before any scalability claim can be made.

Target Audience

Quantum-computing researchers working on variational algorithms and tensor-network methods; theoretical and computational physicists in particle, nuclear, and hadron physics who use the Bethe–Salpeter equation; condensed-matter theorists interested in exciton binding; and research groups evaluating where quantum simulation of quantum field theory might realistically deliver advantage. Readers who benefit most will have a working knowledge of both relativistic bound-state formalism and near-term quantum algorithms; the paper's candid negative framing also makes it valuable to funding agencies and research strategists assessing the maturity of the field.

Authors’ abstract

We present a gate-based quantum computing solution of the homogeneous Bethe-Salpeter equation (hBSE) for the bound state of two massive relativistic scalar particles interacting via ladder-approximation scalar exchange. After Wick rotation to Euclidean space and O(4) S-wave partial-wave projection, the hBSE is reduced to a symmetric matrix eigenvalue problem of dimension N = 2^n. We decompose the resulting Hamiltonian into a sum of n-qubit Pauli operators and solve it with the Variational Quantum Eigensolver (VQE) using a Matrix Product State (MPS) tensor-network ansatz. For N=16 (n=4 qubits), the VQE recovers the maximum eigenvalue - which encodes the minimum coupling constant for binding - to better than 1 % mean relative error compared to classical diagonalization. Entanglement analysis of the BSE amplitude shows low, area-law-like entanglement over the tested sizes, which motivates the MPS ansatz. A critical analysis of three independent barriers - exponential Pauli overhead, approximately size-independent low entanglement, and decreasing VQE gradient scales for the tested ansatz, consistent with generic barren-plateau concerns at larger n - reveals that the specific problem studied here lies in a classically tractable regime: over the tested range it is well described by low-bond-dimension tensor networks and efficiently handled by MPS/Lanczos methods. This negative result provides, to our knowledge, the first entanglement quantification of the BSE amplitude in qubit encoding, establishes the Pauli Hamiltonian framework for future BSE variants, and identifies 2D Minkowski-space BSE, N-body bound states, and non-ladder kernels as physically motivated extensions where genuine quantum advantage may become plausible.