Jielun Chen of the California Institute of Technology and Garnet Kin-Lic Chan have developed a framework for achieving quantum speedups in simulating the behavior of correlated electrons, a challenge at the heart of modern chemistry. Their work distinguishes itself by targeting calculations precisely where widely used classical methods already perform well, rather than seeking advantage where those methods fail.

The researchers describe how to obtain these speedups for correlated electronic structure and dynamics, focusing on problems where classical heuristics scale linearly with the number of atoms. They write, potentially enabling simulations of systems with thousands of atoms using moderate quantum resources.

Correlated Electronic Structure & Dynamics Target Regime

Researchers are achieving quantum simulation speedups not by tackling problems beyond the reach of conventional computers, but by improving upon calculations already handled efficiently by existing classical methods. This approach, detailed in their work, focuses on “correlated electronic structure and dynamics,” a specific area of quantum chemistry where classical heuristics often succeed. The team’s framework aims for substantial quantum speedup precisely where these classical methods are most effective, a departure from typical quantum advantage research.

This strategy addresses a fundamental challenge in demonstrating quantum advantage; proving the limits of classical algorithms is difficult and constantly evolving. The work bypasses the need to identify problems where classical methods fail outright, instead focusing on enhancing the performance of already successful techniques. By targeting calculations at the same level of approximation as these heuristics, the researchers sidestep the difficulty of demonstrating superiority in scenarios where classical algorithms are continually improving.

This allows for a more robust and reliable demonstration of quantum speedup. Within an excitation basis encoding that yields the correlated analog of exponential compression in simulating independent particles, this enables relevant quantum simulations of correlated electrons in systems of thousands of atoms with moderate fault-tolerant resources. The team’s calculations target electronic structure beyond mean-field theory, such as density functional theory or Hartree-Fock theory, in scenarios where polynomial-cost classical heuristics provide accurate results. The framework’s efficiency is tied to the correlation length, and the researchers focus on cases where m ≤ 3.

m-Exciton Bethe-Salpeter Equation & Coupled Cluster Methods

Calculations of excitation energies and dynamics using the m-exciton Bethe-Salpeter equation, and ground-state energies via m-fold coupled cluster excitations, represent a specific focus within this new framework, even as it potentially accelerates many methods. A key assumption underpinning this efficiency is the locality of correlation, meaning that distant particle-hole pairs contribute negligibly to the overall correlation energy. This allows for coupled cluster methods with computational cost scaling linearly with system size, a significant simplification. The resulting equations, as detailed in the work, provide a pathway toward more efficient simulations of complex systems.

Classical Heuristic Scaling with Correlation Length L_c

Quantum algorithms targeting electronic structure can achieve speedups of up to the 18th power of the correlation length, $L_{c}$ , according to new work focused on systems where classical methods already succeed. This scaling, or O(L_c^(18)), represents a relative improvement over classical sparse techniques and is also expressed as a function of the 19th power in polynomial degree d and the 7th power in interaction range $R_{c}$ .

This focus on low-excitation rank in fluctuations from mean-field retains standard classical assumptions, avoiding the need for fine-tuning or reliance on worst-case classical exponential complexity. The research emphasizes asymptotic scaling rather than system-specific benchmarks.

Quantum Speedup via Geometric Locality & Range R_c

The research acknowledges the inherent difficulty in definitively proving quantum advantage, as classical algorithms continually improve, but frames its approach as addressing this challenge directly. The framework focuses on simulating a limited number of effective excitations, keeping the number of excitations, m, at three or less, reflecting the practical constraints of most applications and simplifying the computational demands.

The efficiency stems from recognizing that while quasiparticles and quasiholes may interact over long distances due to Coulomb interactions, the fluctuations of these particles are correlated over a shorter range, a property exploited in existing linear-scaling classical algorithms. Linear algebra computations, including solving eigenvalue problems and evolving systems in time, are central to both the m-exciton Bethe-Salpeter equation and m-excitation coupled cluster equations, but the new approach imbues these with geometric locality, limiting the scope of calculations.

QSVT Implementation & Block Encoding Efficiency

Quantum simulations using the quantum singular value transform, or QSVT, can sidestep the need to store extensive “light cone” data, a significant reduction in computational demand for modeling correlated electrons. This approach operates on a block encoding of a Hamiltonian scaled by a factor α, ensuring the matrix norm divided by α remains less than or equal to one, and uses sparse block-encoding schemes to minimize computational cost.

The total cost of applying a polynomial through QSVT is directly linked to the block-encoding cost, the scaling factor α, and the system size denoted as ‘d’. The efficiency of this method hinges on a qubit encoding strategy where the number of qubits required scales with the number of excitations, specifically using O(2m log L) qubits for m-fold excitations, where L represents a system parameter.

While prior quantum simulation work in this excitation basis often focused on small physical systems to limit qubit requirements, this framework prioritizes keeping the number of excitations, ‘m’, small, mirroring the approach of successful classical heuristics, to enable simulations of larger systems.

This focus on small ‘m’ allows for efficient state encoding, even as system size increases, and contrasts with attempts to achieve exact solutions requiring m proportional to L. The quantum cost of obtaining a solution state using this method is approximately the sum of the initialization cost, the block-encoding cost multiplied by ‘sd’ (where ‘s’ is a scaling factor and ‘d’ is related to the condition number of the linear operator), all divided by the measurement error ε’. However, the long-range Coulomb interaction and projection into the excitation basis complicate the expectation of immediate geometric locality and sparsity in the linear algebra. This means the method’s advantage isn’t simply derived from a naturally sparse matrix structure, but from the specific way it manages the complexity of electron correlations.

Correlation Volume Dependence in Linear Algebra Costs

The researchers found the total cost to multiply a linear operator scales with the correlation volume and operator sparsity, but importantly, does not depend on overall system size at the initial state. The amount of input data required ranges from a constant value for simple, parametrically defined systems to a value proportional to the volume and correlation length for more complex, non-translationally invariant problems. This data requirement never scales exponentially with the number of excitations, forming the basis for high-order polynomial speedups when limiting the number of excitations to three or fewer.

Speedup Factor: O(L_c^(18)) & Polynomial Degree d

The quantum speedup is defined as the ratio between correlation volume and block-encoding cost, a metric where operator sparsity and polynomial degree largely cancel between quantum and classical calculations. For multiexciton eigenvalue determination, the quantum cost factor includes an inverse relationship with initial state overlap, γ. However, the researchers note that eliminating exponential advantages in system size, dependent on excitation number and dimensionality, would diminish the primary source of polynomial speedup in $L_{c}$ .

“Technically, the matrix elements are a summation of poly(m) terms with magnitude O(1), but throughout the paper we focus on m ≤ 3 so we drop this factor,” the paper states, clarifying a simplification within their calculations.

Applicability to Systems of Thousands of Atoms (m ≤ 3)

This compression stems from an encoding strategy that mirrors techniques used to efficiently simulate non-interacting particles, but applied specifically to scenarios with few quasiparticle excitations. The researchers acknowledge that proving the limits of classical heuristics is a constantly evolving challenge, and their framework addresses this difficulty by focusing on scenarios where polynomial-cost classical methods are accurate.

This focus on small excitation numbers, rather than attempting to solve for all possible excitations, is key to achieving this efficiency and scaling to larger systems. The quantum speedup is expressed in terms of the effective correlation length $L_{c}$ , related to the interaction range $R_{c}$ .

👉 More information
🗞 Framework for Robust Quantum Speedups in Practical Correlated Electronic Structure and Dynamics
✍️ Jielun Chen and Garnet Kin-Lic Chan
🧠 DOI: http://link.aps.org/doi/10.1103/v2ms-wmz1

More like this