Quantum Blogs and Articles

Neural Networks Boost Accuracy of Quantum Simulations for Complex Materials

Researchers at ETH Zürich have introduced a neural‑based selected configuration (NQS‑SC) method that outperforms the traditional NQS‑VMC approach for electronic structure calculations. In benchmark tests on molecules with strong static correlation, NQS‑SC reduced ground‑state energies by an average of 0.027 Hartree and produced wave‑function coefficients with significantly lower variance. The technique leverages the neural network to deterministically select key configurations, offering systematic accuracy improvements absent in VMC. Nonetheless, both methods still struggle with dynamical correlation, highlighting the need for hybrid solutions.