What happens when a quantum computer has 56 qubits, but the experiment you want to run needs 64? One option is to wait for a bigger machine. This paper explores another possibility. Sometimes the program takes up more space than the calculation itself needs. In a simulated experiment involving 16 interacting spins, which behave like tiny magnets, we rewrote the program to use just 32 qubits. That put it within the machine’s capacity and cut the number of operations involving pairs of qubits by 68%.
Think of a delivery driver visiting eight cities. To record the next stop, you could print all eight city names and tick one box. Or you could simply write the city’s number. Either way, the driver knows where to go. Our software can spot a similar opportunity inside certain quantum programs, even ones written by someone else. In an eight-city routing example, it reduced the quantum memory needed by 57% and could prove that every possible route still had the same chance of appearing.
The paper also asks what we can learn from the answers a quantum computer gives us. The same program can produce a different answer each time it runs. Imagine trying to identify the shortest round trip from a handful of suggested routes. Two suggestions might look different until you notice that one simply follows the other backwards. The paper shows how to count separately what a suggestion reveals about which roads to take and what it reveals about which way round they were listed.
That distinction has a practical consequence. In one set of tests, the task was to identify the best arrangement of facilities from just twelve suggestions. Changing how the quantum program was trained raised success from 64% to 74% when the software reading the answers knew the program’s usual patterns. For software that judged each suggestion by cost alone, the same change lowered success from 44% to 27%. For someone designing a quantum experiment, it pays to ask two questions at the outset: can we express the problem compactly enough to fit the machine, and what will we actually be able to learn from the answers it gives us?