A rigorous hybridization of variational quantum eigensolver and classical neural network

TL;DR

This work analyzes neural post-processing for variational quantum algorithms and proves that diagonal non-unitary post-processing (DNP) cannot satisfy self-contained training, polynomial resource scaling, and variational consistency simultaneously under finite sampling, due to normalization-induced instabilities and exponential sampling costs. To address this, the authors introduce a normalization-free, norm-preserving approach, the unitary variational quantum-neural hybrid eigensolver (U-VQNHE), which applies a diagonal unitary post-processing that preserves the Rayleigh–Ritz bound and improves robustness. Theoretical results show exponential resource requirements for DNP in both constant-depth and unitary-2-design circuit regimes, while numerical experiments on transverse-field Ising models demonstrate that U-VQNHE achieves higher accuracy and stability than VQE and DNP variants. The work provides a principled guideline for scalable quantum–neural hybrids by enforcing normalization-by-construction and opens avenues for further exploration of unitary post-processing and its combinations with adaptive quantum circuits.

Abstract

Neural post-processing has been proposed as a lightweight route to enhance variational quantum eigensolvers by learning how to reweight measurement outcomes. In this work, we identify three general desiderata for such data-driven neural post-processing -- (i) self-contained training without prior knowledge, (ii) polynomial resources, and (iii) variational consistency -- and show that current approaches, such as diagonal non-unitary post-processing (DNP), cannot satisfy these requirements simultaneously. The obstruction is intrinsic: with finite sampling, normalization becomes a statistical bottleneck, and support mismatch between numerator and denominator estimators can render the empirical objective ill-conditioned and even sub-variational. Moreover, to reproduce the ground state with constant-depth ansatzes or with linear-depth circuits forming unitary 2-designs, the required reweighting range (and hence the sampling cost) grows exponentially with the number of qubits. Motivated by this no-go result, we develop a normalization-free alternative, the unitary variational quantum-neural hybrid eigensolver (U-VQNHE). U-VQNHE retains the practical appeal of a learnable diagonal post-processing layer while guaranteeing variational safety, and numerical experiments on transverse-field Ising models demonstrate improved accuracy and robustness over both VQE and DNP-based variants.

Figures (4)

• Figure 1: VQNHE implementation for a 7-site TFIM using 7 qubits. (a) Training dynamics of the neural network within VQNHE. The vertical axis shows the loss function---the expectation value of the Hamiltonian---plotted on a logarithmic scale. Quantum circuit evaluations are performed using the Qiskit sampler with 500 shots per circuit. The inset highlights the region between the lowest vanilla VQE energy and the exact ground state energy, noting that values below this range are nonphysical. (b) Final neural-network outputs after 200 training epochs. The horizontal axis indexes bit strings (in decimal), while the vertical axis shows the corresponding network outputs. Blue lines denote measured bit strings, and red lines denote unmeasured ones. Most outputs cluster near $10^{-6}$, whereas a few extreme values (above $10^{-2}$; red) emerge. Since nearly all bit strings are sampled by the quantum circuits, these red points correspond to strings that contribute only to the numerator.
• Figure 2: Sketch of proof of exponential decay of Bhattacharyya coefficient in a finite-depth circuit. Each layer acts locally on neighboring qubits, forming a backward light-cone (red wedge) of finite width $\ell=O(d)$. (a) The circuit is divided into blocks $B_i$, with buffer regions in between each pair of blocks of length $g \gtrsim \ell$. Because this light-cone remains bounded as the total system size grows, the Bhattacharyya coefficient between two independently drawn circuit outputs approximately factorizes across independent blocks, leading to the exponential decay. (b) The buffer region has been discarded via classical stochastic channel. The remaining blocks have almost no overlap to the other blocks. The formal proof can be found in Appendix \ref{['app:exp_decay_proof']}.
• Figure 3: Training curves of the neural network in VQNHE with constraints applied, for transverse-field Ising models of various sizes and shot numbers. (a)--(c) show 10-, 12-, and 14-qubit systems, respectively, where the neural-network output is constrained to $f(s)\in[1/r, r]$ for $r$ values from $1.5$ to $5.5$. For each $r$, the number of circuit shots is set to $N = \frac{9r^4}{4\epsilon^2}$Zhang2022, ensuring deviations smaller than $\epsilon=0.05$. The shaded blue regions indicate the range between the VQE baseline and the exact ground-state energy. (d)--(e) correspond to 12-qubit systems trained under fixed-shot conditions: (d) $N=135{,}056$ shots per circuit, satisfying the accuracy bound at $r=3.5$, and (e) $N=10{,}000$ shots per circuit. In these plots, the shaded red or blue regions indicate the area between the VQE baseline and the exact energy.
• Figure 4: Comparison of the training performance of VQNHE and U-VQNHE. (a) Energy difference from the ground-state energy for TFIM instances with increasing numbers of qubits, using 10,000 measurements per circuit. The neural-network output is bounded within $[1/3,3]$, as before. The red dashed line shows the gap between VQE and the ground-state energy. As the number of qubits increases, the deviation between VQNHE and the exact ground-state energy generally worsens. (b) Results for a 12-site TFIM using a two-layer ansatz and 10,000 measurements per circuit. The region between the exact ground-state energy and the VQE result is shown in light blue. The solid blue line shows the result from U-VQNHE, while the dotted red line shows that of VQNHE with constrained DNP (with range parameter $r=3$). VQNHE violates the variational bound, leading to nonphysical solutions, whereas U-VQNHE remains within the physical region while improving upon VQE.

Theorems & Definitions (17)

• Definition 1: Desiderata for data-driven neural post-processing
• Remark 1: Why variational consistency becomes nontrivial for DNP
• Proposition 1: Support mismatch yields an unbounded empirical objective
• proof
• Proposition 2: Coupon-collector scaling for support inclusion
• proof
• Theorem 1: Exponential decay of the Bhattacharyya coefficient
• proof : Proof sketch
• Theorem 2: Bhattacharyya coefficient for unitary 2-design states
• proof : Proof sketch