comparing Trotter-Suzuki and qDRIFT on real IBM hardware during my APL quantum research
This started in the Quantum Education in Science and Technology program at JHU APL’s ASPIRE internship, where I was introduced to programming cloud-based quantum computers. With a small team I compared two algorithms for simulating quantum magnetism: First Order Trotter-Suzuki decomposition (FOTS), the deterministic workhorse, and qDRIFT, a newer randomized compilation routine that weights operations by their importance to the simulation.
We built both in Qiskit and simulated the dynamics of a one-dimensional Ising model, then ran the same experiments on real IBM Quantum devices. The interesting result was about noise: with fixed initial states and observables, qDRIFT had more algorithmic error than FOTS in noiseless simulation, but on actual hardware the gap mostly closed, meaning qDRIFT’s results were less affected by the noise of a real machine. And when the Heisenberg model had one dominant ferromagnetic coupling, qDRIFT outperformed FOTS outright. Since qDRIFT’s gate count scales with the sum of Hamiltonian coefficients rather than the number of terms, that tradeoff matters for chemical systems with millions of terms.
We presented this at the IEEE Integrated STEM Education Conference in March 2023. The poster and extended abstract are on my site.