Researchers Detail Costs of Simulating Quantum Systems with New Framework

Aaron Sander and colleagues at Technical University of Munich, in collaboration with Weierstrass Institute for Applied Analysis and Stochastics, Munich Quantum Software Company GmbH, Software Competence Centre Hagenberg GmbH, and Freie University Berlin, present a cost-resolved approach for matrix product state-based quantum trajectory simulations. The analysis reveals that seemingly equivalent stochastic unravelings can sharply alter the overall computational expense. By quantifying this trade-off with dimensionless inflation factors, the researchers provide a practical protocol for selecting unravelings that optimise performance under specific hardware constraints, establishing unraveling choice as a key hardware-aware simulation design problem.

Cost-resolved analysis optimises quantum simulations via bond dimension and sampling control

The researchers of Munich and collaborating institutions achieved a 30% decrease in runtime for complex simulations of open quantum systems. This improvement addresses previous limitations stemming from the trade-off between trajectory complexity and statistical convergence, which previously hindered accurate modelling of larger systems. Open quantum systems are of paramount importance in modelling realistic quantum hardware, where interactions with the environment introduce noise and decoherence, and in understanding the many-body dynamics of complex materials. Simulating these systems accurately requires significant computational resources, and optimising these simulations is crucial for advancing both quantum computing and materials science. A novel framework decomposes total computational cost into memory, runtime, and sampling effort, revealing how physically equivalent simulation approaches redistribute these costs.

Bond dimension inflation (α) and sampling inflation (κ), two dimensionless factors, quantify this trade-off, demonstrating their relationship to memory usage, runtime, and the number of simulations needed for reliable results. The bond dimension represents the amount of entanglement that needs to be tracked during the simulation; higher bond dimensions capture more complex correlations but also increase memory requirements and computational cost. Statistical convergence refers to the need to run multiple trajectories (simulations) to obtain statistically meaningful results. The researchers found that different stochastic unravelings, methods for representing the quantum system’s evolution as a series of probabilistic trajectories, can lead to vastly different values of α and κ. Benchmarks across various noise channels, including amplitude damping and dephasing, showed that the most computationally efficient unraveling, the method used to break down the quantum simulation, is not static, but shifts depending on the strength of the noise, the resolution of the time-steps, the system size, and available parallel processing power. For instance, stronger noise levels may necessitate unravelings that prioritise sampling effort over minimising bond dimension, while larger systems may benefit from unravelings that reduce memory usage at the expense of increased runtime. Selecting an unraveling represents a hardware-dependent design choice, extending optimisation beyond minimising trajectory entanglement. Detailed analysis highlights that the optimal strategy isn’t universally about minimising entanglement, but about balancing computational resources to suit the specific simulation and available hardware. This is particularly important given the limited resources available on near-term quantum devices and classical simulators.

Entanglement reduction is not universally beneficial for quantum simulation efficiency

Simulating open quantum systems, those interacting with their environment, is vital for building realistic models of both fragile quantum computers and the complex behaviour of materials. These interactions lead to decoherence, the loss of quantum information, and noise, which can significantly impact the performance of quantum algorithms and the properties of materials. Accurately modelling these effects requires sophisticated simulation techniques. A subtle but significant tension exists within existing simulation techniques, however. While scientists have long focused on minimising entanglement, a measure of quantum correlation, reducing it doesn’t automatically equate to faster overall calculations. Traditional approaches often prioritise minimising the bond dimension, effectively limiting the amount of entanglement tracked during the simulation, with the assumption that this reduces computational cost. However, this can lead to increased sampling effort, requiring a larger number of trajectories to achieve the same level of statistical accuracy. This work demonstrates that reducing entanglement can sometimes increase the overall computational cost if it necessitates a significant increase in the number of trajectories required for convergence.

The cost-resolved approach decomposes the total simulation expense into memory usage, runtime, and statistical sampling, revealing that different, yet physically valid, simulation methods redistribute these costs rather than reduce them overall. This decomposition allows for a more nuanced understanding of the trade-offs involved in choosing a particular simulation method. Dimensionless inflation factors, α and κ, are introduced in this work, offering a means to compare simulation approaches under specific hardware limitations. A value of α greater than one indicates that the chosen unraveling leads to a larger bond dimension than necessary, increasing memory usage. Similarly, a value of κ greater than one indicates that more trajectories are needed to achieve the same level of accuracy compared to a more efficient unraveling. Establishing a framework to balance computational demands moves beyond minimising entanglement in quantum simulations. The approach reveals that the most efficient simulation isn’t necessarily the one with the least entanglement, but the one that best utilises available computational resources. This is particularly relevant as simulations become more complex and require greater computational power, demanding a more nuanced approach to optimisation. The findings have implications for the design of quantum algorithms and the development of new materials, as they provide a means to optimise simulations for specific hardware platforms and achieve more accurate and efficient results. The ability to tailor simulations to available resources is crucial for accelerating progress in both quantum computing and materials science, particularly as systems grow in size and complexity.

The research demonstrated that physically equivalent methods for simulating quantum systems do not always reduce computational cost, but instead redistribute it between memory usage, runtime, and the number of simulations needed. This means minimising entanglement alone does not guarantee the most efficient simulation. Researchers quantified this trade-off using inflation factors, allowing comparison of different simulation approaches based on available computational resources. The authors established a protocol for determining the optimal simulation method using pilot studies, which considers system size and noise strength.