LAKE WALES, Fla. ‐ Quantum annealing hardware like D-Wave's latest $15 million 2048-Qubit unit will not fit the real world problems at supply chain optimization software vendor ServicePower, according to Alex Syrichas, a research scientist there and a doctoral candidate under research professor Alan Crispin at Manchester Metropolitan University. Their problems routinely use from 10,000 to as many as 1 million variables -- each represented by a virtual Qubit -- and Syrichas has found a way to simulate their quantum annealing using parallelization on standard server farms to make up for the slower speed compared to D-Wave's specialized hardware.
Syrichas, whose doctoral thesis is on using software simulations of quantum annealing, found commercial success with ServicePower, where CEO Marne Martin recruited him to adapt it to field service and logistics applications even before finishing his thesis.
Syrichas's thesis, specifically on using quantum annealing simulations to solve the traveling salesman problem, is one of the simplest formulations of an non-deterministic polynomial-time (NP Hard) problem not solvable on convention computers, until now. The traditional approach would be to calculate all the possible routes among, say, a dozen cities, finding the optimal one by comparing the length of each. Unfortunately that would take an inordinate amount of time to calculate. Quantum annealing, on the other hand, compares all route lengths simultaneously (using a superposition of values for its variables) thus immediately coming up with a close-to-optimal solution. Repeated runs, each time resetting the starting variables to the last closest-to-optimal values from the previous run, eventually finds a route close enough to optimal to use.
"The routing problem for field service and logistics has lots of other variables beyond the traveling-salesman problem, such as time windows and from where the goods are being transferred," Martin told EE Times in an exclusive interview. "We are the first to use MMU's quantum annealing algorithm on these routining problems The first three patents filed on it by ServicePower are part of a knowledge transfer partnership [KTP] initiative with Syrichas and Crispin at MMU."
ServicePower has since filed a fourth patent on how to adapt the solution to other problems.
Just like hardware quantum annealing at D-Wave, ServicePower's software version iterates until an optimal solution is found ‐ but with as many variables as necessary on standard servers
As a part of the United Kingdom's Network Quantum Information Technology program, Crispin cites MMU's Path Integral Monte Carlo algorithm as the key. It uses random stochastic sampling to simulate the quantum tunneling hardware postulated by Richard Feynman and pioneered by D-Wave into superconducting hardware.
Besides field service and logistics, ServicePower is also adapting the algorithm to other NP-Hard problems in the financial sector, machine learning, simulations of molecular interactions, protein folding, aerospace simulations, software validation, home health care and even finding cures for cancer.
ServicePower's first non-field-service/logistics application is Zeus, which allows a wider class of vertical applications to make use of its quantum annealing algorithms.
— R. Colin Johnson, Advanced Technology Editor, EE Times