IBM, RIKEN and Nvidia improve iron-sulfur cluster simulation accuracy 79-fold

Started by Oliver98, Today at 03:34 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

Topic: IBM, RIKEN and Nvidia improve iron-sulfur cluster simulation accuracy 79-fold   Views(Read 43 times)
Active members in this topic:
Oliver98(1)

Oliver98

A hybrid quantum chemistry method has taken a big step in accuracy. Quantum Computing Report says IBM, RIKEN and Nvidia achieved a 79 times energy accuracy gain for a 72 qubit active space of an iron sulfur cluster

The key change was starting sample based quantum diagonalisation from broken symmetry states. The run used IBM's ibm_kingston with 88 active qubits and about 25 minutes of quantum processor time

The classical side used RIKEN's ROQUO system with 144 Nvidia GB200 GPUs, plus the Miyabi-G and Fugaku supercomputers. That is a large classical workload alongside the quantum run

The extrapolated energy came within about 0.33 milliHartree of the DMRG reference, whereas earlier runs stalled more than 500 milliHartree away. The headline figure of 4.3 milliHartree differs from the table, the work was funded by Japan's NEDO, and it is an arXiv preprint

Iron sulfur clusters are vital in biology and notoriously hard to simulate, so a 79 times improvement matters. Choosing a better starting point rather than more hardware is a clever lesson. The mismatched figures need clearing up before anyone celebrates too much

Are smarter algorithms more important than bigger quantum computers right now? And how close is quantum chemistry to real usefulness?