Filtered-state preparation could cut quantum phase estimation runtime by more than two orders of magnitude

[Caitlin_75]

Quantum Zeitgeist covered a paper in Quantum Science and Technology from researchers at Sungkyunkwan University in South Korea and Xanadu in Canada, published on September 25. It tackles one of the less talked about bottlenecks in quantum algorithms, which is preparing a starting state that is close enough to the answer you are looking for. The headline claim is that their filtered approach reduced runtime by more than two orders of magnitude in simulations of certain problems, with overlap improving by more than a factor of one hundred

Some background helps here. Quantum phase estimation is a key algorithm for finding things like the ground-state energy of molecules and materials. Its chance of success depends on how much the initial state overlaps with the target state, and that success probability scales with the square of the overlap. For large or strongly correlated systems the overlap tends to shrink rapidly as the system grows, which the paper links to what is known as the orthogonality catastrophe

The team's framework applies a spectral filter to the starting state to boost its overlap with the target, then weighs that gain against the cost of implementing the filter and the chance of the filtering step failing. They studied Gaussian filters and a modified Krylov-subspace filter. A standard Krylov filter can give accurate energies but suppress unwanted components so hard that the success probability drops, so the modified version adds a tunable parameter to balance the two. In their comparisons the modified Krylov filter came out with the lowest overall cost

The numerical tests used Fermi-Hubbard models in a strongly correlated regime, with the ratio of on-site repulsion to hopping set at 10. The authors also point out a limitation. The advantage of the filtered approach can shrink if the gap between ground and excited states closes too quickly

This is squarely a chemistry and materials simulation paper, not a cryptography one. It still matters to the bigger picture, because the more efficient quantum algorithms get, the fewer resources a useful machine needs. Every improvement like this nudges forward the point where quantum computers deliver something valuable

The data behind the findings is openly available, which is a nice touch. I would be interested to hear whether people doing quantum chemistry see state preparation or circuit depth as the bigger bottleneck in practice

Quantum Algorithms Gain From Filtered-state Preparationquantumzeitgeist.com