Cost gaps of 10^(-4) achieved in quantum power demand simulations

Researchers at TISI Inc., Osaka University, and RIKEN have achieved cost gaps of 10^(-4) in simulations optimizing electric power demand using a new quantum framework. The work introduces Pauli Correlation Encoding (PCE), a qubit-efficient method representing complex problems with fewer qubits than conventional approaches. PCE encodes binary variables through multi-body correlations, enabling compact representations and a continuous relaxation that allows for scaling to larger systems. These results establish PCE as a physically motivated framework for large-scale combinatorial optimization.

Pauli Correlation Encoding for Power Demand Optimization

TISI Inc. The simulations, conducted in collaboration, tested problem sizes ranging from m = 18 to 10,296 variables, revealing consistent performance across a broad spectrum of complexity. The core of this optimization framework is Pauli Correlation Encoding (PCE), a method that embeds multiple binary variables into correlated expectation values of Pauli operators. This encoding allows for a combinatorial increase in effective problem size with a limited number of qubits, a critical advantage for tackling large-scale optimization challenges.

Unlike traditional one-qubit-per-variable encodings, PCE compresses information, enabling the treatment of dense, large-scale instances with fewer quantum resources. The researchers propose a two-stage hybrid formulation, initially averaging the problem over time to establish a starting point for a more detailed, time-resolved optimization process. The interplay between continuous relaxation and discretization governs the performance of PCE, according to the study.

The effective resolution of the correlator representation dictates how reliably improvements in the continuous loss function translate into better discrete solutions. Larger systems exhibited more consistent behavior, indicating the potential for scalability as quantum computing hardware advances. This suggests that while the method is effective across a range of problem sizes, maintaining sufficient resolution in the correlator representation is important for achieving optimal results. The team further validated the approach on a trapped-ion quantum processor, obtaining high-quality solutions despite the presence of noise and limitations in sampling.

This demonstration of robustness is significant, as it suggests that the PCE framework can function effectively even with the imperfections inherent in current quantum hardware. “PCE provides a qubit-efficient variational representation for fully connected quadratic optimization problems with real-valued and nonuniform coefficients,” the study states, highlighting the method’s adaptability to complex, real-world scenarios.

Two-Stage Hybrid Formulation for Time-Resolved Optimization

The formulation of a time-averaged problem is an important initialization step for a subsequent time-resolved optimization process, a technique designed to capture the statistical structure of uncertainty in available demand reduction. This initial averaging, performed over a time window of length n_T, establishes parameters that enhance the efficiency of the later, more detailed optimization stage.

The averaged cost function, denoted as C_T(x), is calculated as the average of the hourly costs over the specified time window, enabling a balanced benchmark setting while preserving the inherent correlations within the power procurement problem. This two-stage approach is underpinned by a specific mathematical definition of the target power, = 1/2∑_{i =, ensuring the procurement process aims to secure approximately half of the total available demand reduction.}

The researchers use normalized loss gap measures, Δ, to compare optimization performance across varying problem sizes and solution methods, with Δ L_T defined as L_T -, referencing an optimal cost baseline. This rigorous comparison allows for a precise evaluation of the method’s effectiveness as the scale of the optimization problem increases. The transition from continuous to discrete solutions is central to the observed optimization behavior, with binary variables relaxed to continuous variables y(θ) ∈^m through the use of Pauli correlators generated by a parameterized quantum circuit.

Qubit Efficiency with Correlation Order k = n/2

The choice of correlation order in Pauli Correlation Encoding (PCE) significantly impacts qubit efficiency, with the highest number of Pauli correlators achieved when k equals floor(n/2). This configuration maximizes the number of expectation values used to represent classical variables, offering a pathway to compact representations without a one-to-one qubit-variable mapping. Each Pauli correlation operator, constructed from subsets of qubits and assigned a Pauli type (X, Y, or Z), defines a specific interaction within the quantum system, enabling the encoding of multi-body correlations.

PCE’s ability to represent variables through expectation values of these k-body operators allows for scalability beyond what direct binary encodings provide. While maximizing representational capacity through k = n/2 offers advantages, the complexity of the correlator space that a finite-depth variational circuit must explore also increases. Determining whether this increased capacity necessitates systematically deeper or more expressive ansatz remains an open question.

Numerical simulations demonstrate near-optimal performance across problem sizes ranging from m = 18 to 10,296, with normalized cost gaps on the order of ${10}^{–4}$ relative to solutions with certified optimality. The hyperparameters ${\alpha }_{sc}$ and β, selected via grid search to minimize the total cost $C_T$, further refine the optimization process.

“Solving Model 1 provides variational circuit parameters that capture the dominant structure of the cost landscape and are used to initialize the subsequent time-dependent optimization,” the researchers report, highlighting the two-stage approach employed. The ability to effectively navigate this complex correlator space, balancing representational power with computational feasibility, will be important for realizing the full potential of PCE in larger systems.

Scaling of Problem Size with Qubit Number in PCE

The number of qubits required to represent a problem grows slowly with problem size when using Pauli Correlation Encoding (PCE), a key factor in achieving consistently low normalized cost gaps across simulations. This contrasts with direct binary encodings, where qubit demands increase linearly, and allows for tackling significantly larger optimization challenges. The work demonstrates that this encoding maintains effectiveness as problem complexity increases, a key attribute for real-world applications.

Larger systems exhibit more consistent behavior in this regard, suggesting a smoothing of the optimization landscape with increasing scale. This interplay between continuous relaxation and discretization is a key factor in the scalability observed in the simulations. It highlights the importance of balancing representational power with computational complexity.

Cost Gap of 10^(-4) Achieved in Numerical Simulations

Numerical simulations have now demonstrated optimization performance approaching certified optimality, achieving normalized cost gaps as low as ${10}^{–4}$ across problem instances. The simulations assessed performance relative to solutions generated by the Gurobi optimizer, establishing a benchmark with a relative optimality gap tolerance of ${10}^{–4}$ to define minimum and maximum cost values for normalization.

The researchers quantified the quality of their solutions using a normalized cost gap, calculated as (Δ C_T)/W_T, a metric ranging from 0 to 1 that reflects the relative optimality of a given solution. Analysis of larger instances revealed that the quantum processing unit (QPU) results converged toward the mean of a random distribution, indicating that the initialization process effectively became randomized at the level of post-processed cost.

This behavior, consistent with correlator distributions and match probability analyses, suggests that even with noise, the proposed method maintains viability on current quantum hardware. Examining the distributions of continuous variables and corresponding Pauli correlators at iterations achieving minimum decoded binary cost provided insight into the structure of optimal solutions prior to post-processing.

Variational Quantum Algorithms and Qubit Limitations

Numerical simulations demonstrate near-optimal performance across problem sizes ranging from m = 18 to 10,296, with normalized cost gaps on the order of ${10}^{–4}$ relative to solutions with certified optimality. This approach provides a continuous relaxation of the discrete optimization problem, enabling compact representations with a reduced qubit count. The authors further propose a two-stage hybrid formulation, in which a time-averaged problem provides initialization for a time-resolved optimization.

More like this

Quantum AlgorithmsResearchers Unlock Accurate Spectra Via Optimised Density Functionals

Quantum AlgorithmsResearchers Find Lower-Noise Circuits Beat Faster Fourier Transforms

Quantum SimulationResearchers Simulate Quantum Systems Using Periodic Drives

Quantum PhysicsImmanuel Bloch wins Wolf Foundation’s Wolf Prize for quantum simulation work