Randomized Krylov Quantum Diagonalization
randomized time-evolution circuits that make sample-based quantum diagonalization practical for large molecules
collaborators: IBM Research, TCS Research, Prof. Phani Motamarri
Ongoing work.
Sample-based quantum diagonalization is only as good as the configurations the quantum computer provides. SqDRIFT generates them from time-evolution circuits under the molecular Hamiltonian, built by randomly sampling Hamiltonian terms (the qDRIFT method). Because these circuits approximate a Krylov space of the Hamiltonian, the method inherits convergence guarantees that heuristic circuit ansätze lack. For large active spaces, however, faithful time evolution is far too deep for today’s processors.
This project aims to make SqDRIFT practical for large, strongly correlated molecules on current hardware: sampling the terms that matter most for the state at hand, keeping the circuits shallow, choosing the evolution times, and feeding the best configurations from the classical diagonalization back into later rounds of sampling. We test it on [4Fe-4S] on IBM Heron processors and, in simulation, on the iron-molybdenum cofactor (FeMoco) at 152 qubits.
