Machine Learning Techniques for Solving Quantum Many-Body Problems (QuanticLearn)

Summary

The quantum many-body problems (QMBs) are of fundamental importance in various fields such as condensed matter physics, in particular the physics of nanoelectronic devices with applications to quantum information technology. The main goal of the project is to develop machine learning techniques, based on artificial neural networks (ANNs), which shall provide efficient solutions to QMB problems, with the main focus on quantum transport and entanglement. The training and validation sets are formulated as 2-particle scattering problems in an effective mass description and, complementary, as atomistic descriptions of nanoelectronic devices at DFT-NEGF level. One objective is to provide accurate predictions regarding the many-body scattering wavefunctions and transmission functions using the ANNs. Formulating the 2-particle scattering formalism based on the R-matrix method, entangled systems shall be investigated, in particular the implementation of quantum sorters. Furthermore, an efficient framework shall be constructed based on DFT-NEGF approach to quantum transport combined with ML techniques (DFT-NEGF-ML). As test-cases, the quantum transport in hybrid perovskite nanoFETs shall be investigated, which have been in the focus in the past few years. The methods developed during this project shall also enable the possibility to design quantum neuron devices and the investigation of learning schemes in quantum neural networks.

Rezumat

Problemele multi-particula in mecanica cuantica au o importanta fundamentala in multe domenii, intre care fizica dispozitivelor nanoelectronice cu aplicatii in tehnologia informatiei cuantice. Scopul principal al proiectului este acela de a dezvolta tehnici de invatare automata, utilizand retele neurale artificiale (ANNs), care pot pune la dispozitie solutii eficiente in rezolvarea problemelor multi-particula, in particular a problemelor de transport. Seturile de antrenare si validare sunt formulate sub forma unor probleme de imprastiere bi-particula si, complementar, sub forma unor sisteme descrise atomistic la nivel DFT-NEGF. Unul din obiectivele proiectului este predictia starilor de imprastiere si a functiilor de transmisie utilizand retele neurale. Prin extinderea formalismului de imprastiere la cazul bi-particula, utilizand metoda matricii R, vor fi investigate sisteme cu entanglement si, in particular, sisteme de tip quantum sorter. De asemenea, proiectul propune realizarea unui cadru pentru rezolvarea eficienta a problemelor de tip DFT-NEGF, prin tehnici de invatare automata (machine learning, ML) formuland o abordare de tip DFT-NEGF-ML. Cazurile de lucru vor fi selectate din cadrul nanotranzistorilor perovskitici, care prezinta un interes actual ridicat. Metodele dezvoltate pe parcursul proiectului vor crea posibilitatea dezvoltarii de nanodispozitive neurale cuantice, precum si investigarea schemelor de antrenare specifice retelelor neurale cuantice.