Hubbard series 3 — IBM Nighthawk, 72 fermion modes and our local timing benchmark.
Can we, as a student/hobby project, use a quantum computer to beat our own classical computer on recorded computation time? This third Hubbard series follows that concrete question: from fermion theory and mapping to 72 modes on IBM Nighthawk, error mitigation and a local timing ratio of approximately 20x.
The claim is deliberately precise: approximately twenty times less registered QPU time than the kernel time of our current classical chi64 baseline. This is not yet a matched-accuracy or end-to-end advantage. Both limitations are explained alongside the results.
Reading order
- Part 1: A laptop as our benchmark: the Nighthawk project
- Part 2: What does U=8 dynamics mean? The full fermion model
- Part 3: From a 2D lattice to hardware: mapping and circuit depth
- Part 4: Small pilots: is the problem shots or the circuit?
- Part 5: Readout, ZNE, PEC and TFLO: what did we actually use?
- Part 6: Our 6×6 results: charge, spin and doublons
- Part 7: Approximately 20x: our local timing milestone explained
- Part 8: How close are we to Google and the Bonsai results?
- Part 9: From a fast run to a reliable answer
Earlier series and source material
The nine articles are available through the links above. The associated GitHub repository remains private; access to the research archive requires permission. The earlier Hubbard series remain available.