Optimizing Unitary Coupled Cluster Wave Functions on Quantum Hardware: Error Bound and Resource-Efficient Optimizer

Technical Summary: Optimizing Unitary Coupled Cluster Wave Functions on Quantum Hardware

Problem Statement
The paper addresses the challenges of optimizing Unitary Coupled Cluster (UCC) wave functions on near-term quantum hardware to compute the ground state of many-body systems. While the Variational Quantum Eigensolver (VQE) is a standard hybrid approach, it suffers from three primary limitations:

1. Decoherence: Limits circuit depth and expressivity.
2. Measurement Overhead: Estimating variational energy requires a number of shots scaling as $1/\epsilon^2$ for statistical error $\epsilon$.
3. Optimization Hardness: VQE landscapes often exhibit "barren plateaus" (exponentially flat cost functions) and local minima, making global convergence difficult. Furthermore, a small energy gradient in VQE does not guarantee proximity to the true ground state energy.

To address these issues, the authors investigate the Projective Quantum Eigensolver (PQE), a method introduced in a previous work (Ref. 1) that formulates ground state search as a root-finding problem rather than an energy minimization problem. PQE optimizes parameters to drive the "residues" (projections of the Schrödinger equation onto excited states) to zero, rather than minimizing the energy directly.

Methodology
The paper provides a rigorous mathematical analysis of the PQE algorithm and proposes a novel optimization strategy.

1. Mathematical Framework:

   • The algorithm seeks to solve $r_\nu(t) = \langle \Phi_\nu | \hat{U}^\dagger(t) \hat{H} \hat{U}(t) | \Phi_0 \rangle = 0$ for all excited determinants $\nu \neq 0$.
   • The authors derive a general error bound for the algorithm using Kato's inequality and Temple's inequality. They establish a relationship between the squared sum of residues ($\sum r_\mu^2$) and the energy error ($|E(t) - E_{gs}|$) and the overlap with the ground state.
   • Unlike VQE, where a small gradient norm does not imply small energy error, the authors prove that a small residual norm provides formal guarantees on the accuracy of the energy and the quality of the wavefunction overlap.

2. Optimization Strategy:

   • The standard PQE update (from Ref. 1) approximates a modified Newton-Raphson step using a fixed denominator based on Møller-Plesset perturbation theory ($\Delta_\mu$).
   • The authors analyze the convergence properties of quasi-Newton methods using the Newton-Kantorovich theorem. They identify that the standard update rule (based on the Jacobian at the initial Hartree-Fock point) may fail when the solution is far from the initial guess (e.g., in strongly correlated regimes).
   • They propose a hybrid optimizer that dynamically switches between update rules:
     • Far from the solution: Uses a gradient-descent-like update (or a simplified Jacobian approximation) to ensure stability and progress from the initial state.
     • Near the solution: Switches to an approximate Newton-Raphson update (using a refined Jacobian approximation) to achieve faster, quadratic-like convergence.

Key Contributions

1. Rigorous Error Bounds: The paper derives tight mathematical bounds relating the off-diagonal coefficients (residues) of the Hamiltonian to the energy error and wavefunction overlap. This provides formal convergence guarantees absent in VQE.
2. Convergence Criterion: Based on the derived bounds, the authors formulate a well-informed stopping criterion ($\epsilon_T$) that allows users to estimate the energy error without knowing the exact ground state.
3. Novel Optimizer: A new residue-based optimizer is introduced that adapts its update rule based on the distance to the solution, leveraging the Newton-Kantorovich theorem to ensure robustness.
4. Numerical Benchmarks: Extensive simulations on $H_4$, $H_6$, $BeH_2$, and $LiH$ dissociation curves.

Results

• Error Bound Tightness: Numerical tests on an $H_4$ molecule show that the derived error bound tracks the true energy error within less than one order of magnitude, significantly tighter than previous bounds based on the residual 1-norm.
• Optimizer Performance:
  • The new hybrid optimizer demonstrates superior robustness compared to the standard PQE update (Ref. 1), particularly for large bond distances where static correlation is high.
  • In terms of resource efficiency, the new method requires significantly fewer measurements (and thus less runtime) to reach a target accuracy compared to both the standard PQE and VQE optimized with the Broyden–Fletcher–Goldfarb–Shanno (BFGS) method.
• Convergence: The method successfully converges to the ground state energy with fewer optimization steps than VQE, avoiding the issues of barren plateaus associated with energy minimization.

Significance
The paper claims that by shifting from variational energy minimization to projective root-finding, PQE offers a more reliable path to ground state preparation on near-term hardware. The primary significance lies in:

• Providing formal guarantees that small residues imply small energy errors, addressing the uncertainty inherent in VQE optimization.
• Offering a resource-efficient alternative that reduces the measurement overhead and optimization steps required for chemical accuracy.
• Establishing a mathematical foundation (via Newton-Kantorovich analysis) for designing robust optimizers that can handle the non-linearities of strongly correlated systems where standard linear approximations fail.

The authors position this work as a step toward practical quantum advantage in the near term, providing a method that is less susceptible to the specific optimization pathologies of VQE while remaining feasible for current hardware constraints.