Nonorthogonal variational quantum simulation for quantum chemistry

AI summary

Overview

Research area: Quantum computing for quantum chemistry — variational quantum simulation (VQS) of real- and imaginary-time many-body dynamics.

Technical level: Advanced. The paper derives parameter equations of motion from McLachlan's variational principle, works in second quantization, and assumes familiarity with Pauli-string Hamiltonians, the Hadamard test, and quantum Fisher information.

Scope: The paper proposes and analyzes NOVQS, a framework that represents an evolving molecular wavefunction as a linear combination of several shallow, nonorthogonal parameterized quantum circuits, and benchmarks it against conventional single-circuit VQS on hydrogen chains and the nitrogen molecule.

What This Paper Is About

Simulating how electrons move in molecules over time is a central task in chemistry, but exact classical methods become intractable as the wavefunction grows, and entanglement growth during time evolution makes tensor-network and neural-network ansätze increasingly costly. Quantum computers could help, but Trotterization requires circuit depths that scale with evolution time and the number of Hamiltonian terms, putting it beyond near-term hardware. This paper asks whether the expressive power of a variational quantum simulation can be increased without increasing circuit depth — by combining several shallow circuits into one nonorthogonal superposition rather than making a single circuit deeper.

Key Contributions

1. A nonorthogonal variational quantum simulation (NOVQS) framework. The wavefunction is written as a sum over $N_c$ parameterized quantum circuits with complex coefficients $c_j(t) = a_j(t)e^{ib_j(t)}$, and equations of motion are derived from McLachlan's variational principle for both real-time evolution (Eq. 16) and imaginary-time evolution (Eq. 21), using two virtual parameters $\theta_0^R$ and $\theta_0^I$ to absorb normalization and global phase.

2. A compact, hardware-friendly ansatz for electronic structure. Inspired by fermionic simulation, the ansatz uses fSim gates in a dense brickwork pattern followed by $R_Z$ rotations. A single Trotter step has circuit depth $N_q + 1$ with $N_q(N_q-1)/2$ fSim gates, where $N_q$ is the number (even) of spin orbitals. Each fSim gate decomposes into at most 3 entangling gates, and the ansatz allows arbitrary spin-orbital pairs to interact with only linear depth and nearest-neighbor connectivity. Parameters per circuit are $N_{p,j} = N_q^2 N_l$ for $N_l$ layers.

3. Three protocols for measuring the Fisher matrix and force vector. Besides the standard Hadamard test, the authors give an ancilla-free Hadamard test that exploits particle-number conservation (constant electron number $N_e$) to remove the ancilla qubit, and a classical-shadow-based procedure, both aimed at reducing circuit depth and ancilla overhead. Total circuit count under the Hadamard-test route scales as $N_{\text{total}} = \mathcal{O}(N_p^2 N_d^2 + N_c N_h N_p N_d + N_c^2 N_h)$.

4. Error analysis and resource estimation, plus numerical demonstrations on hydrogen chains and the nitrogen molecule comparing NOVQS against conventional VQS.

Main Findings

• Shallow ensembles can beat deep single circuits: Numerical simulations of hydrogen chains and the nitrogen molecule show that a collection of shallow, or even single-layer, PQCs can match or outperform a much deeper single PQC in conventional VQS.

• Circuit number versus depth trade-off: The authors identify an explicit trade-off between the number of circuits $N_c$ and the depth per circuit, meaning expressivity can be purchased either with more circuits or with deeper ones at comparable aggregate cost.

• Systematic expressivity control: Increasing the number of PQCs systematically enhances the ansatz's expressive power, providing a dial that does not require deeper circuits.

• Hamiltonian ansatz is impractical for chemistry: The standard variational Hamiltonian ansatz is poorly suited to molecular electronic structure because the Hamiltonian contains $\mathcal{O}(N_q^4)$ terms, potentially producing prohibitive depth even for one layer — motivating the fSim-based dense local ansatz.

• Correspondence with classical variational Monte Carlo: The NOVQS equations coincide in form with stochastic reconfiguration (imaginary time) and time-dependent VMC (real time); the difference is that (t)VMC estimates the metric and force by Monte Carlo sampling, whereas NOVQS measures them directly on a quantum computer. NOVQS is thus framed as the quantum-computing counterpart of classical (t)VMC.

• Not reported in the available content: Specific numerical values for energies, fidelities, infidelities, Trotter errors, basis-set sizes, chain lengths, qubit counts, or shot counts from the hydrogen-chain and nitrogen-molecule simulations. The text is truncated before the numerical results section, so no quantitative benchmarks can be quoted.

Methodology in Plain English

The authors start from the standard second-quantized electronic Hamiltonian and the Schrödinger equation. Rather than approximating the time-evolution operator with Trotter products — whose depth grows with time and with the number of Hamiltonian terms — they keep the evolving state inside a small, fixed parameterized circuit and move the time dependence into the circuit parameters.

Their key twist is that the ansatz is not one circuit but a weighted sum of several circuits, all started from the same initial state, with complex weights. Because the circuits are different unitaries applied to the same state, the resulting components are generally not orthogonal — hence "nonorthogonal." To evolve, they apply McLachlan's principle: at each time step, choose the parameter velocities that make the variational state best satisfy the Schrödinger equation. This reduces to solving a linear system $M\dot{\theta} = V$, where $M$ is the quantum Fisher information matrix (a positive-definite covariance matrix) and $V$ is the force vector; for imaginary time, the sign of the force term flips. Two auxiliary parameters handle normalization and phase so the unnormalized combination can be used directly.

Because the gates are of the form $e^{-iH\theta}$, each derivative of the circuit is another circuit with an inserted Pauli operator, so every entry of $M$ and $V$ reduces to overlaps of the form $\langle \Psi_0 | U | \Psi_0 \rangle$, measurable by Hadamard-test-type circuits. The authors then replace the ancillary-qubit Hadamard test with ancilla-free variants (valid because particle number is conserved) and classical-shadow estimation, cutting the depth and qubit overhead that would otherwise block near-term execution.

Finally, they run classical numerical simulations of real- and imaginary-time evolution for hydrogen chains and $N_2$, comparing NOVQS with different numbers of circuits against VQS with a single deeper circuit at matched aggregate depth.

Why This Matters

Impact on research. The work connects two previously separate threads — nonorthogonal multistate ansätze developed for ground-state VQE (NOVQE, quantum subspace expansion) and variational quantum simulation of dynamics. It gives the first variational principle and measurement protocols for the dynamical case, and it establishes a concrete, tunable knob (number of circuits) for trading quantum resources against expressivity. It also sharpens the analogy between quantum-circuit variational simulation and classical time-dependent variational Monte Carlo, making the two directly comparable as representations of the same variational manifold.

Real-world applications (as motivated by the paper's framing of chemistry and materials simulation):

• Modeling photochemistry and excited-state dynamics, where real-time evolution of molecular electronic structure is the natural target.
• Ground-state and finite-temperature property estimation via imaginary-time evolution, which the paper treats with the same equation-of-motion machinery.
• Strongly correlated molecular systems such as hydrogen chains and the nitrogen molecule, used here as the demonstration cases.
• Materials and catalysis problems where electronic correlation and entanglement growth defeat tensor-network or neural-network ansätze in higher dimensions.

Industry relevance. The ansatz is explicitly designed for near-term processors with limited depth and nearest-neighbor connectivity on a two-dimensional lattice; fSim gates have already been demonstrated experimentally on superconducting and semiconductor spin qubits (as cited in the paper). This makes the proposal relevant to hardware groups optimizing native two-qubit gate sets for chemistry workloads, and to quantum-chemistry software efforts that need algorithms whose accuracy improves with circuit count rather than qubit coherence time.

Future Directions

• Hardware demonstration. The paper is entirely numerical; implementing NOVQS on superconducting or spin-qubit devices, where fSim is native, is the obvious next test, including the ancilla-free and classical-shadow measurement routes under real noise.
• Quantifying the trade-off. The manuscript identifies a circuit-number-versus-depth trade-off but the visible content does not specify the quantitative crossover points (how many shallow circuits equal one deeper circuit at fixed total depth) or how this scales with molecule size; a systematic scaling study with $N_q$ is implied but not reported here.
• Measurement cost reduction. The Hadamard-test route's $\mathcal{O}(N_p^2 N_d^2)$ circuit count for the metric-tensor elements grows steeply with parameter number; tighter shadow-based or grouped-measurement strategies, and the role of the classical-shadow protocol, remain to be characterized.
• Noise and error mitigation. Error analysis is described as provided, but the interplay between the truncated-circuit advantage and device error rates — and whether the nonorthogonal combination amplifies sampling noise — is an open question raised by removing the ancilla.

Target Audience

Researchers in quantum algorithms and quantum chemistry who already understand variational circuits and second quantization — particularly those working on near-term quantum simulation of molecular dynamics, on nonorthogonal/multistate ansätze, and on the boundary between quantum-circuit VQS and classical time-dependent variational Monte Carlo. Experimentalists designing native-gate chemistry benchmarks on superconducting or spin-qubit hardware will also find the ansatz specification and resource counts directly useful.

Authors’ abstract

Dynamical simulation of quantum many-body systems is a central task in quantum chemistry and requires efficient wavefunction representations. Although tensor-network and neural-network quantum states have achieved considerable success in ground-state calculations, entanglement growth hinders their application to quantum dynamics. Quantum computing may offer a route to quantum advantage, but algorithms such as Trotterization generally require fault-tolerant quantum computers, whereas near-term hardware supports only limited circuit sizes. Variational quantum simulation (VQS) instead represents the evolving state with a single shallow, fixed-size parameterized quantum circuit (PQC). Here, we introduce nonorthogonal variational quantum simulation (NOVQS), which applies linear combinations of parameterized quantum states to real- and imaginary-time evolutions. We design a shallow, hardware-friendly ansatz tailored to second-quantized electronic-structure Hamiltonians, together with resource-efficient protocols for measuring the matrices and vectors in the parameter equations of motion. Error analysis and resource estimation are also provided. Numerical simulations of hydrogen chains and the nitrogen molecule demonstrate that a collection of shallow, or even single-layer, PQCs can match or outperform a much deeper PQC in VQS. In particular, we identify a trade-off between circuit number and depth. NOVQS therefore provides a flexible route to enhancing wavefunction expressivity under circuit-depth constraints, making it a promising framework for quantum dynamics simulations on near-term quantum processors.