A rigorous hybridization of variational quantum eigensolver and classical neural network
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866910026933731328 |
|---|---|
| author | Kim, Minwoo Park, Kyoung Keun Lee, Kyungmin Bang, Jeongho Kim, Taehyun |
| author_facet | Kim, Minwoo Park, Kyoung Keun Lee, Kyungmin Bang, Jeongho Kim, Taehyun |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_17295 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | A rigorous hybridization of variational quantum eigensolver and classical neural network Kim, Minwoo Park, Kyoung Keun Lee, Kyungmin Bang, Jeongho Kim, Taehyun Quantum Physics 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. |
| title | A rigorous hybridization of variational quantum eigensolver and classical neural network |
| topic | Quantum Physics |
| url | https://arxiv.org/abs/2602.17295 |