# Molecular energies from sparse quantum prolate diagonalization on a trapped-ion quantum processor

Quantum.world

> Scope (one sentence): The paper introduces "sparse QPD," a compressed-sensing variant of quantum prolate diagonalization that reduces the number of distinct quantum circuits needed for a chemically…

**arXiv**: 2611.22834

**Published**: 2026-11-17

**Authors**: Davide Castaldo, Marvin Kronenberger, Marek Miller, Sélène Forget, Gemma C. Solomon, Kresten Lindorff-Larsen, Matthias Christandl, Markus Reiher

## AI summary

## Overview

**Research area:** Quantum computing for quantum chemistry — specifically single-ancilla (hybrid quantum–classical) phase-estimation algorithms for computing molecular electronic energies, combined with compressed sensing and validated on trapped-ion hardware.

**Technical level:** Advanced. The paper assumes familiarity with second-quantized electronic Hamiltonians, filter diagonalization, prolate spheroidal wave functions, Trotterized time evolution, and compressed-sensing/matrix-completion theory.

**Scope (one sentence):** The paper introduces "sparse QPD," a compressed-sensing variant of quantum prolate diagonalization that reduces the number of distinct quantum circuits needed for a chemically accurate energy estimate from 𝒪(N_s) to 𝒪(log N_s) without increasing circuit depth, and demonstrates it on the Quantinuum Helios trapped-ion processor for a two-orbital active-space model of an arginine–tryptophan residue pair in water.

## What This Paper Is About

Computing molecular energies accurately on quantum hardware is limited by the cost of running many deep circuits, so the field has moved toward "single-ancilla" methods that sample a system's autocorrelation function C(t) with short circuits and let classical post-processing extract the eigenvalues. Quantum prolate diagonalization (QPD) is one such method; this paper asks whether the *number of samples* (i.e., distinct circuits) required by QPD can be reduced by exploiting the fact that molecular Hamiltonians have sparse spectra. The authors answer yes, by borrowing a compressed-sensing signal-recovery algorithm, and then test the resulting method under realistic shot noise, Trotter error, and hardware noise on a real trapped-ion device.

## Key Contributions

1. **A compressed-sensing extension of QPD ("sparse QPD").** The authors combine the original QPD formulation [Phys. Rev. Res. 8(3), 033177 (2026)] with matrix completion / compressed sensing, using the low-rank Hankel structure of the sampled autocorrelation signal to recover QPD matrix elements from far fewer samples.
2. **A sample-complexity reduction from 𝒪(N_s) to 𝒪(log N_s).** Here N_s is the number of sampled points of the autocorrelation function (equivalently, distinct circuits). The authors state this reduction comes without increasing circuit depth at chemical accuracy, and report up to an order of magnitude reduction in total algorithmic cost.
3. **A full accounting of error sources within the QPD accuracy bound.** The paper generalizes the error matrices δA and δB in the QPD accuracy bound (Theorem 2 of the earlier QPD work) so that they contain not only prolate-sampling truncation error and shot noise, but also quantum hardware noise, the compressed-sensing reconstruction error, and Trotter error. A Cramér–Rao lower bound supporting the numerical analysis is derived in the appendix, following earlier Fisher-information approaches.
4. **A hardware demonstration on Quantinuum Helios.** Using a two-orbital active-space-embedded Hamiltonian for an interacting arginine–tryptophan residue pair in water and first-order Trotterized Hamiltonian simulation, the authors resolve energies to chemical accuracy with circuits containing up to 448 two-qubit gates (1221 gates in total), reaching energy errors below 1 kcal/mol on hardware. They also characterize performance on the Helios-1E emulator.

## Main Findings

- **Sample complexity drops exponentially in the exponent:** sparse sampling reduces the number of autocorrelation samples from 𝒪(N_s) to 𝒪(log N_s), where N_s is the number of sampled points (distinct circuits), with no increase in circuit depth at chemical accuracy.
- **Chemical accuracy achieved on hardware:** on the Quantinuum Helios trapped-ion quantum computer, sparse QPD resolves the Hamiltonian energies to chemical accuracy (defined in the paper as an error within 1.6 mHa), with circuits of up to 448 two-qubit gates.
- **Sub-kcal/mol errors on device:** the authors report energy errors below 1 kcal/mol on hardware for circuits containing up to 448 two-qubit gates (1221 gates in total), using standard first-order Trotterization.
- **Total cost reduction of up to an order of magnitude:** combining sparse sampling with the existing QPD algorithm yields up to an order of magnitude reduction of the total algorithmic cost.
- **Favorable comparison to prior methods (as reported for original QPD):** the paper recalls that QPD achieves chemically accurate estimates for weakly and strongly correlated systems using evolution times more than one order of magnitude shorter than standard quantum phase estimation, with at least three orders of magnitude fewer shots per sample than other quantum subspace methods.
- **Robustness to realistic noise:** the analysis explicitly includes shot noise, Trotter error, and hardware noise in the accuracy bound, and the hardware experiment is performed under these realistic conditions rather than in an ideal simulator.
- **Scaling outlook from numerics:** for a suitably large quantum processor, the authors demonstrate numerically that the method extends to larger systems in the weakly correlated regime.
- **Not reported in the available text:** absolute energies in Hartree, qubit counts, total shot budgets, wall-clock runtimes, and the specific numerical error values in mHa for the hardware experiment are not given in the excerpt analyzed here (the text is truncated mid-discussion of the Hadamard test).

## Methodology in Plain English

The starting point is a simple identity: the autocorrelation function C(t) = ⟨Φ|e^(−iHt)|Φ⟩ is a sum of oscillating terms, one per eigenstate, whose frequencies are the energies you want. A quantum computer can measure samples of C(t) at chosen times using a Hadamard test (a single ancilla qubit, short Trotterized evolution circuits), and a classical computer then extracts the frequencies.

QPD turns this into a matrix eigenvalue problem. It builds a small subspace out of "filtered" versions of the initial state, using prolate spheroidal wave functions as filters because these leak very little signal from outside the energy window of interest. The Hamiltonian and overlap matrices in this subspace can be written entirely in terms of the measured C(t) samples — collected into a Hermitian Toeplitz matrix — plus purely classical, signal-independent quantities (filter values, their derivatives, and integration weights) computed once.

The new ingredient is that the Toeplitz matrix of samples has low-rank Hankel structure, because molecular Hamiltonians have sparse spectra: only a few eigenstates carry appreciable weight. Compressed sensing exploits this by measuring only a small, cleverly chosen subset of the samples and reconstructing the rest, in the same spirit as earlier compressed-sensing phase-estimation proposals (ground-state estimation in Ref. [yi2024quantum], multiple eigenvalues in Ref. [castaldo2025heisenberg]). Because only the number of circuits shrinks — not the evolution time or the depth of each circuit — the depth-related cost is untouched.

Finally, the authors quantify what can go wrong. The QPD error bound separates three effects: the intrinsic error of representing the Hamiltonian in a small trial subspace (controlled by the trial-space dimension M and the product c = W_f T, with M = c/π used here), the quality of the initial state (leakage from out-of-band eigenstates scaling with |a_k|²), and everything else lumped into the error matrices δA and δB — now including hardware noise, reconstruction error, and Trotter error. The demonstration uses an active-space-embedded two-orbital Hamiltonian for an arginine–tryptophan residue pair interacting in water, run on the Quantinuum Helios device and its emulator.

## Why This Matters

**Impact on research.** The paper attacks the part of the quantum-chemistry cost budget that is often ignored: not circuit depth or qubit count, but the sheer number of distinct circuit executions. Showing that sample count can be reduced from 𝒪(N_s) to 𝒪(log N_s) without deepening circuits reframes compressed sensing as a practical add-on to single-ancilla phase estimation rather than a theoretical curiosity. Delivering chemical accuracy on a trapped-ion processor with up to 448 two-qubit gates, while explicitly accounting for shot noise, Trotter error, and hardware noise, provides a realistic benchmark for near-term quantum chemistry rather than an idealized simulation.

**Real-world applications (implications of this line of work):**

- **Biomolecular modeling:** the demonstrated system is an interacting arginine–tryptophan residue pair in water, representative of the non-covalent interactions that govern protein structure and binding.
- **Drug discovery and lead optimization:** faster quantum estimates of interaction energies could feed into high-throughput screening workflows, the motivation the paper cites for ab initio quantum-chemistry on hardware.
- **Catalysis and reaction energetics:** accurate low-lying spectra of active-space models are the input to reaction-barrier and mechanism studies.
- **Materials and molecular design:** the weakly correlated regime in which the authors numerically demonstrate scaling covers many closed-shell organic and main-group systems.

**Industry relevance.** Because the method reduces the number of circuit executions rather than depth, it directly targets quantum-computer wall-clock time and queue consumption — the cost metric that matters for cloud-accessed trapped-ion hardware. The use of an active-space-embedded Hamiltonian also shows how quantum resources can be confined to a small correlated fragment while a classical embedding handles the rest, a pattern that maps onto industrial workflows.

## Future Directions

- **Scaling to larger molecular systems.** The authors numerically demonstrate extension to larger systems only in the weakly correlated regime and only for a "suitably large quantum processor"; actual hardware demonstrations beyond the two-orbital active space remain open.
- **Strongly correlated and transition-metal systems.** The paper notes that original QPD was validated for strongly correlated systems, but the hardware demonstration here is a small active-space model; whether sparse QPD retains its sample savings under hardware noise in strongly correlated cases is an open question.
- **Tighter integration with error mitigation and correction.** Since hardware noise enters the error matrices δA and δB alongside reconstruction and Trotter error, separating and suppressing these contributions — for example with better Trotterizations or error mitigation — is a natural next step.
- **Initial-state preparation and out-of-band leakage.** The accuracy bound shows the estimate degrades with the squared overlap of out-of-band eigenstates; improved state preparation, or filter designs that suppress this leakage further, would relax a key accuracy constraint.
- **Reconstruction under noise.** The compressed-sensing recovery was originally developed for noiseless or mildly noisy signals; characterizing how far the 𝒪(log N_s) guarantee survives realistic device noise, and how many samples are needed in practice versus asymptotically, is left for follow-up work.

## Target Audience

Quantum-chemistry algorithm developers and quantum-computing researchers working on near-term, hybrid phase-estimation and quantum-subspace methods; computational chemists interested in active-space embedding and in what current trapped-ion hardware can actually deliver for molecular energies; and applied mathematicians working on compressed sensing, prolate spheroidal functions, or filter-diagonalization signal processing. The paper is not introductory — readers should already be comfortable with second quantization, Trotterization, and generalized eigenvalue problems.

## Authors’ abstract

We introduce sparse quantum prolate diagonalization (sparse QPD), which combines the original formulation of QPD [Phys. Rev. Res., 8(3), 033177, (2026)] with compressed sensing to exploit Hamiltonian spectrum sparsity for molecular systems. We show that sparse sampling reduces the sample complexity of QPD from $\mathcal{O}(N_s)$ to $\mathcal{O}(\log N_s)$, where $N_s$ is the number of sampled points of the autocorrelation function (i.e.\ distinct circuits), without…

[The full abstract on arXiv](https://arxiv.org/abs/2611.22834)

[Read the original paper](https://arxiv.org/abs/2611.22834)

---

Canonical: https://quantum.world/paper/molecular-energies-from-sparse-quantum-prolate-diagonalization-on-a-2611.22834
