⚛️ quantum physics

Variationally Optimized Imaginary-time Polynomial Filters for Ground State Projection

This paper introduces a variational imaginary-time evolution framework using optimized polynomial filters derived from an operator-level action principle, which significantly enhances the accuracy, stability, and success probability of ground-state preparation on near-term quantum devices compared to standard Taylor-Trotter-Suzuki approaches.

Original authors: Bahman Seifi, Ibsal Assi, J. P. F. LeBlanc

Published 2026-08-31

📖 4 min read🧠 Deep dive

CC0 1.0

Original authors: Bahman Seifi, Ibsal Assi, J. P. F. LeBlanc

Original paper dedicated to the public domain under CC0 1.0 (http://creativecommons.org/publicdomain/zero/1.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Finding the lowest energy state of a quantum system is like finding the deepest valley in a vast, foggy mountain range. For physicists, this "ground state" holds the key to understanding how materials behave, how chemical reactions proceed, and what new phases of matter might exist. In the classical world of traditional computers, solving for this state becomes impossible as the system grows, because the number of possibilities explodes exponentially, quickly overwhelming even the most powerful machines. Quantum computers offer a different path, operating natively within these vast spaces of possibility. However, using them to find the ground state is not straightforward. One powerful method involves a process called imaginary-time evolution, which acts as a filter to wash away all the excited, higher-energy states, leaving only the calm, stable ground state behind. The challenge has been that this filter is mathematically difficult to build on a quantum computer, often requiring a delicate balance between accuracy and the likelihood of the process actually succeeding.

A team of researchers has now developed a new way to build this filter that significantly improves both its precision and its reliability. Their work focuses on a technique known as ancilla-based imaginary-time evolution. In this setup, the quantum computer uses a helper qubit, called an ancilla, to guide the main system toward its lowest energy state. The process works by repeatedly applying a short, controlled step that nudges the system closer to the ground state. If the helper qubit is measured in a specific way after each step, the system is updated; if not, the attempt is discarded and the process starts over. This "post-selection" is crucial but comes with a cost: as the number of steps increases, the chance of successfully completing the entire sequence drops dramatically, often making the method impractical for large systems.

The researchers started with a standard approach that uses fixed mathematical formulas to approximate these steps. They found that while these formulas work well for very small, cautious steps, they become unstable and inaccurate when the steps are made larger to speed things up. To solve this, they replaced the fixed formulas with a flexible, adjustable method. Instead of using a rigid, pre-calculated recipe, they treated the coefficients of the mathematical steps as variables that could be tuned. By applying a principle that seeks the most efficient path for the system to evolve, they derived new, optimized values for these steps. This new approach, which they call an action-based variational framework, keeps the exact same physical circuit structure as the old method but changes the internal parameters to be much smarter.

The results of their simulations show a marked improvement. When they tested this new method on a model of magnetic spins, known as the transverse-field Ising model, it converged to the correct ground state energy much faster and more smoothly than the standard method. In tests with larger systems, the traditional approach began to fail, overshooting the target energy or becoming unstable, while the new variational method remained robust. Perhaps most importantly, the new method increased the probability of success for each individual step. Because the overall success of the process depends on the product of all these individual steps, even a small increase in the per-step success rate leads to a massive improvement in the total chance of finishing the calculation. In their tests, this led to an order-of-magnitude enhancement in the final success probability, meaning the quantum computer could achieve the same result with far fewer attempts.

The researchers also provided a way to predict how large a step can be taken without breaking the simulation. By calculating simple properties of the system's energy landscape before running the quantum computer, they can determine a safe time step that balances speed and accuracy. This gives experimentalists a practical guide for setting up their machines. The work demonstrates that by optimizing the mathematical description of the evolution rather than just the hardware, it is possible to make quantum ground-state preparation more efficient and reliable. This approach does not require more qubits or deeper circuits, making it a practical route for near-term quantum devices to tackle complex problems that were previously out of reach. The findings suggest that with the right mathematical tuning, the probabilistic nature of quantum computing can be managed more effectively, bringing us closer to using these machines for real-world scientific discovery.

Technical Summary: Variational Optimization of Ancilla-Based Imaginary Time Evolution for Quantum Ground State Preparation

Problem Statement
Determining the ground state ( $|GS\rangle$ ) of a quantum many-body system is fundamental for predicting equilibrium properties. While classical methods like exact diagonalization (ED) are limited by exponential Hilbert-space scaling, and other classical approaches (e.g., Quantum Monte Carlo, tensor networks) face issues such as the sign problem or ansatz expressiveness, quantum computers offer a native route to circumvent these bottlenecks. A primary challenge in quantum ground-state preparation is implementing Imaginary-Time Evolution (ITE), defined by the non-unitary operator $e^{-\tau H}$ . Since this operator cannot be directly realized as a deterministic unitary gate, existing probabilistic implementations often rely on fixed-coefficient approximations, such as truncated Taylor expansions combined with Trotter-Suzuki (TS) decompositions. These standard methods face a critical trade-off: reducing approximation error requires smaller time steps or higher expansion orders, which increases circuit depth and, crucially, lowers the post-selection success probability. This trade-off limits the practical efficiency of ITE on near-term quantum devices.

Methodology
The authors propose an Action-Based Variational Approach (ABVA) to optimize ancilla-assisted probabilistic ITE. The framework retains the standard circuit architecture—using a single ancilla qubit to implement a Linear Combination of Unitaries (LCU) via post-selection—but replaces fixed analytical coefficients with variationally optimized parameters derived from an operator-level action principle.

Variational Framework: The method is grounded in an action principle $S = \int d\tau \text{Tr}[U^\dagger(\dot{U} + HU)]$ . By extremizing this action with respect to the parameters of a variational ansatz, the authors derive equations of motion for the coefficients.
Variational Polynomial Ansatz (ABVPA): For the imaginary-time update, the standard first-order Taylor expansion ( $1 - \tau H$ ) is replaced by a generalized linear polynomial $c_0(\tau) + c_1(\tau)H$ . The coefficients $c_0$ and $c_1$ are determined by the quantum geometric tensor and generalized forces, depending on static moments of the Hamiltonian ( $\mu_n = \text{Tr}[H^n]$ ).
Variational Trotter-Suzuki Ansatz (ABVTSA): Similarly, the real-time propagator block within the LCU circuit, typically implemented via a second-order TS decomposition ( $e^{-iA\delta t/2}e^{-iB\delta t}e^{-iA\delta t/2}$ ), is replaced by a variational form $e^{id_0(t)A}e^{id_1(t)B}e^{id_2(t)A}$ . The parameters $d_i(t)$ are optimized to minimize the error relative to the exact propagator.
Implementation: The optimized coefficients modify the ancilla preparation angles and the rotation parameters within the controlled-unitary blocks. Crucially, this optimization does not alter the circuit depth or the number of gates; it only changes the parameter values.
Error Bounds: The authors derive rigorous, state-independent error bounds for both the linear filtering (ABVPA) and the real-time decomposition (ABVTSA). These bounds depend solely on static Hamiltonian properties (trace moments), providing a classical pre-computation guideline for selecting an optimal time step $\delta\tau$ that balances accuracy and success probability.

Key Contributions

Action-Based Variational Optimization: The introduction of a variational framework that optimizes the coefficients of the ITE filter and the real-time propagator without increasing circuit complexity.
Enhanced Stability and Accuracy: Demonstration that the variational parameters allow for larger time steps while maintaining stability, a regime where standard Taylor expansions fail.
Improved Success Probability: By optimizing the filter, the method increases the per-step and cumulative post-selection success probability, reducing the number of circuit shots required to achieve a target statistical precision.
Resource Efficiency: The approach maintains a single-ancilla requirement per step, avoiding the overhead of multi-ancilla schemes, while offering superior performance to fixed-coefficient methods.

Results
The method was benchmarked on the one-dimensional transverse-field Ising model (QIM) with $N=5$ and $N=8$ qubits.

Convergence: In simulations, the ABVA implementation converged to the ground-state energy faster and more smoothly than the standard TS-Taylor ITE. While the standard method exhibited deviations and overshoots at larger time steps (e.g., $\tau=1.8$ ), the ABVA remained stable and valid.
Fidelity: The ABVA achieved higher fidelity to the exact ground state, particularly in the late-time regime. The distribution of fidelity values was narrower and closer to unity compared to the standard method, indicating reduced sensitivity to approximation errors.
Success Probability: The per-step success probability was consistently higher for ABVA. Consequently, the cumulative success probability over the full evolution was significantly improved (up to an order-of-magnitude enhancement in final success probability in some regimes), reducing the measurement overhead.
Error Analysis: Numerical evaluation confirmed that the chosen time steps ( $\delta\tau = 0.4$ ) lay within the validity regime of the variational approximations, with global errors significantly lower than those of the standard TS-Taylor approach.

Significance and Claims
The paper claims that the ABVA framework provides a practical route to higher-fidelity ground-state preparation on near-term quantum devices. By shifting the burden from static analytical approximations to parameter optimization, the method alleviates the inherent trade-off between approximation accuracy and post-selection success probability. The authors emphasize that this is achieved without increasing circuit depth or ancillary overhead, making it a resource-efficient improvement over existing gate-based probabilistic ITE schemes. The work suggests that variational polynomial filtering can bridge the gap between rigorous energy-filtering methods and the hardware constraints of current quantum processors. The authors remain modest regarding scalability, noting that while the method improves the balance of accuracy and probability, systems with small spectral gaps or complex Hamiltonians may still require careful tuning of time steps or higher-order approximations, which would introduce additional resource trade-offs.

More like this