Miaolan Xie and her collaborators, Matt Menickelly of Argonne National Laboratory and Stefan M. Wild of Lawrence Berkely National Laboratory, recently published a paper in the Journal of Optimization Theory and Applications titled “A Stochastic Quasi-Newton Method in the Absence of Common Random Numbers.” The work introduces Q-SASS, a new optimization method designed for problems where randomness cannot be controlled.

 

While the paper uses quantum chemistry, such as finding the ground-state energy of molecules, as a primary benchmark, Xie emphasized that the method’s applications extend far beyond chemistry. Q-SASS is designed to improve Variational Quantum Algorithms (VQAs), which are considered leading candidates for useful applications on near-term quantum computers.

According to Xie, these algorithms have potential real-world use cases in areas such as finance, logistics and supply chain management, machine learning, and physical experiments. More broadly, Q-SASS applies to any optimization problem involving black-box simulations where the user cannot control random noise, in other words, where "Common Random Numbers"(CRN) are not available.

 

“In many simulation-based optimization problems, you cannot control the source of randomness,” Xie explained. “Standard stochastic quasi-Newton methods often fail or become inefficient in these settings.” Q-SASS provides a way to navigate this noise efficiently, regardless of whether the underlying problem comes from chemistry, finance, or engineering.

 

A key distinction of the work, Xie noted, is its reliability, adaptivity, and efficiency. The authors establish high-probability complexity bounds, providing theoretical guarantees on how fast the algorithm converges even in noisy environments. This allows advanced quasi-Newton methods to be applied in settings that more closely reflect the stochastic nature of real-world problems.