Session 2-3

Abstract

To explore the nature of many-body electrons in quantum materials, numerical algorithms have been developed in parallel with advances in computational infrastructure. In recent years, machine learning has begun to engage with and augment these approaches. Notably, numerical simulations based on a generalized entangled wave function called resonating-valence-bond ansatz [1,2] incorporating neural networks have facilitated access to excitation spectra of quantum many-body systems [3]. These spectra have recently attracted renewed interest as observables relevant to the estimation of quantum entanglement among electrons in quantum materials.

By leveraging recent advances in numerical methods, we have extended the method of generator coordinates [4,5], based on the highly flexible variational-wave-function framework [6,7], by incorporating artificial neural networks. This approach enables us to simulate a wide range of spectroscopic responses, including electron energy loss spectroscopy, inelastic neutron scattering, photoemission and (inverse) photoemission spectroscopy, and resonant inelastic x-ray scattering spectra in quantum materials.

We applied the present method to typical copper oxide superconductors. By simulating neutron scattering spectra of the cuprates, we clarified the material dependence of the energy scale of spin excitations and found that it does not correlate with the optimal superconducting critical temperatures. By combining with electron energy loss and photoemission spectroscopy spectra, we can elucidate entangled nature of the many-body electrons in the cuprates.