QB Ground State Energy Estimation Benchmark

Nicole Bellonzi, Joshua T. Cantin, Mohammad Reza Jangrouei, Alexander Kunitsa, Jason Necaise, Nam Nguyen, John Penuel, Maxwell D. Radin, Jhonathan Romero Fontalvo, Rashmi Sundareswara, Linjun Wang, Thomas Watts, Yanbing Zhou, Michael C. Garrett, Adam Holmes, Artur F. Izmaylov, Matthew Otten · 2025

Read QB Ground State Energy Estimation Benchmark by Nicole Bellonzi, Joshua T. Cantin, Mohammad Reza Jangrouei, Alexander Kunitsa, Jason Necaise, Nam Nguyen, John Penuel, Maxwell D. Radin, Jhonathan Romero Fontalvo, Rashmi Sundareswara, Linjun Wang, Thomas Watts, Yanbing Zhou, Michael C. Garrett, Adam Holmes, Artur F. Izmaylov, Matthew Otten (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-10873.txt.

Introduction

Nicole Bellonzi, Joshua Cantin, Mohammad Reza Jangrouei and colleagues built an open benchmark for ground state energy estimation in 2025. Finding a molecule's lowest energy is central to quantum chemistry, and both classical and quantum solvers claim progress on it.

The benchmark compares three methods across a shared set of problems: a heat-bath configuration interaction method, the density matrix renormalisation group, and a quantum phase estimation approach. The classical interaction method solved almost everything, the renormalisation group shone on weakly entangled systems, and the quantum method was limited by hardware and algorithms. The authors note many test problems were designed around classical strengths, and they propose harder, strongly correlated cases for balance.

Catalog record

CREATORNicole Bellonzi, Joshua T. Cantin, Mohammad Reza Jangrouei, Alexander Kunitsa, Jason Necaise, Nam Nguyen, John Penuel, Maxwell D. Radin, Jhonathan Romero Fontalvo, Rashmi Sundareswara, Linjun Wang, Thomas Watts, Yanbing Zhou, Michael C. Garrett, Adam Holmes, Artur F. Izmaylov, Matthew Otten
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE70 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-10873

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

QB Ground State Energy Estimation Benchmark Nicole Bellonzi, Joshua T. Cantin, Mohammad Reza Jangrouei, Alexander Kunitsa, Jason Necaise, Nam Nguyen, John Penuel, Maxwell D. Radin, Jhonathan Romero Fontalvo, Rashmi Sundareswara, Linjun Wang, Thomas Watts, Yanbing Zhou, Michael C. Garrett, Adam Holmes, Artur F. Izmaylov, Matthew Otten Abstract Ground State Energy Estimation (GSEE) is a central problem in quantum chemistry and condensed matter physics, demanding efficient algorithms to solve complex electronic structure calculations. This work introduces a structured benchmarking framework for evaluating the performance of both classical and quantum solvers on diverse GSEE problem instances. We assess three prominent methods – Semistochastic Heat-Bath Configuration Interaction (SHCI), Density Matrix Renormalization Group (DMRG), and Double-Factorized Quantum Phase Estimation (DF QPE) –highlighting their respective strengths and limitations. Our results show that fully optimized SHCI achieves near-universal solvability on the benchmark set, DMRG excels for low-entanglement systems, and DF QPE is currently constrained by hardware and algorithmic limitations. However, we observe that many benchmark Hamiltonians are drawn from datasets tailored to SHCI and related approaches, introducing a bias that favors classical solvers. To mitigate this, we propose expanding the benchmark suite to include more challenging, strongly correlated systems to enable a more balanced and forward-looking evaluation of solver capabilities. As quantum hardware and algorithms improve, this benchmarking framework will serve as a vital tool for tracking progress and identifying domains where quantum methods may surpass classical techniques. The QB-GSEE benchmark repository is openly available at https://github.com/isi-usc-edu/qb-gsee-benchmark [ 1 ] . By maintaining a scalable and open resource, we aim to accelerate innovation in computational quantum chemistry and quantum computing. 1 Introduction Ground State Energy Estimation (GSEE) has long served as a cornerstone problem in chemistry and physics, enabling precise calculations of chemical reaction rates and material properties. However, despite its centrality, GSEE also highlights fundamental computational challenges and limitations, particularly in balancing accuracy, efficiency, and scalability. For decades, GSEE has driven advances in physics and computing, with algorithms evolving to address its complexities. Exact solutions remain exponentially hard to compute due to the rapid growth of the Hilbert space with system size, making them impractical for large systems. As a result, heuristic algorithms, ranging from approximations like Density Functional Theory (DFT) to methods like semistochastic heat-bath configuration interaction (SHCI), have become essential. These heuristics often provide useful and accurate results but operate with varying degrees of reliability, often lacking systematic guarantees of performance. The effectiveness of these approaches can sometimes be more art than science, relying on empirical tuning and problem-specific adaptations rather than universally applicable principles. Despite these challenges, emerging technologies offer promising alternatives to classical heuristics. Quantum computing, in particular, represents a paradigm shift in computational science, with its potential for exponential advancements over the coming decade [ 2 , 3 , 4 ] . Among its many applications, quantum computing offers a fundamentally new approach to GSEE by naturally representing and manipulating quantum states, directly encoding wavefunctions within their qubits. Algorithms such as Variational Quantum Eigensolver (VQE) and Quantum Phase Estimation (QPE) leverage quantum principles to estimate ground-state energies more efficiently, offering the potential for polynomial or exponential speedups over classical methods. Additionally, quantum computers provide a pathway to solve problems with high accuracy, capturing all system correlations and surpassing the approximations required by classical methods like density functional theory. However, the field remains in its early stages, with significant challenges in validating and benchmarking quantum algorithms for GSEE. Limited qubit coherence times, noise, and resource constraints hinder practical implementation, making it difficult to determine whether quantum algorithms truly outperform classical alternatives in meaningful scenarios. To effectively assess progress in GSEE and quantum computing, robust benchmarking frameworks are essential. Existing repositories such as HamLib [ 5 ] , VarBench [ 6 ] , and g2 [ 7 , 8 ] offer valuable datasets for evaluating both classical and quantum algorithms, yet they often lack comprehensive reference energies and detailed performance breakdowns, limiting their ability to explain when and why solvers succeed or fail. This makes it difficult to compare methods meaningfully across diverse Hamiltonian instances. A number of studies have also defined benchmarks that seek to evaluate the performance of a quantum processor on tasks related to GSEE [ 9 , 10 , 11 , 12 , 13 , 14 , 15 ] . While these benchmarks provide a way to compare present-day Noisy Intermediate Scale Quantum (NISQ) processors, it is challenging to relate the benchmark results to utility: in many cases the tasks being benchmarked represent only a subroutine within GSEE and, moreover, are derived from problem instances that are not representative of real-world use cases. In addition, metrics for evaluating quantum computers can be misleading when the total time to solution is not accounted for or when circuit compilation tricks allow a solver to achieve a performance on benchmark tasks that is not representative of the performance on high-utility problem instances [ 16 , 17 ] . To help guide future advancements in both classical and quantum GSEE approaches, this work introduces an open-source GSEE benchmark that incorporates hardness metrics, systematically measuring algorithmic performance and identifying cases where GSEE algorithms excel or struggle. By integrating diverse Hamiltonian problem sets, advanced feature extraction techniques, and machine learning-driven performance analyses, this benchmark can provide deeper insights into solver behavior, scalability, and generality. This paper describes the benchmark and demonstrates its application to the Semistochastic Heat-Bath Configuration Interaction (SHCI), Density Matrix Renormalization Group (DMRG), and resource estimates for Double-Factorized Quantum Phase Estimation (DF QPE). This modular framework, shown in Figure 1.1 and available on GitHub [ 1 ] , integrates Hamiltonian problem instances, evaluates their features, and applies machine learning techniques to analyze algorithm performance. Figure 1.1 : Overview of the main components of the GSEE benchmark repository [ 1 ] . In Section 2 , we will introduce the benchmarking framework and dive into the benchmark results, highlighting the performance of SHCI, DMRG, and DF QPE as key solvers. Section 3 explores key insights from these results and outlines future directions to enhance benchmarking methodologies and expand the dataset. In Section 4.2 , we break down the essential problem features that define computational challenges and solver capabilities. Section 4.3 details the diverse set of problem instances used for evaluation, spanning real-world chemistry and physics applications. Finally, Section 4.4 introduces our machine learning-driven approach to predicting solvability, leveraging latent space analysis to map out the boundaries of what can—and cannot—be solved efficiently. 2 Results The QB GSEE benchmark is designed to provide a rigorous and scalable evaluation of algorithms for ground-state energy estimation, integrating both classical and quantum approaches within a unified assessment framework. At its core, the benchmark consists of three interdependent components: a structured problem instance database, an automated feature computation module, and a comprehensive performance analysis pipeline. Each component is designed to facilitate a direct comparison of algorithmic performance while ensuring scalability, modularity, and extensibility. The problem instance database serves as the foundation of the benchmark, housing a diverse collection of Hamiltonians that span different domains of computational chemistry and condensed matter physics. These problem instances, detailed in Section 4.3 , are categorized into two primary types: benchmark instances, which correspond to well-characterized problems with reliable classical reference solutions 1

Read the complete text