Probing quantum entanglement in many-body systems via inverse statistical methods of classical physics
le mercredi 7 octobre 2020 à 11h00
Séminaire LPMMC
Personne à contacter : Vincent Rossetto ()
Lieu : Salle de lecture 2, maison des Magistères
Résumé : Quantum simulators and computers of intermediate scale, composed of a few tens of individual degrees of freedom, are currently being developed on various experimental platforms (e.g. trapped atoms or ions, superconducting qubits). In these systems, many-body entanglement represents the fundamental resource, allowing them to outperform their classical counterparts for specific tasks, but for the same reason also making their exact simulation by classical machines intractable. Proving that quantum entanglement is indeed present among the individual qubits is therefore a central task for certifying these devices. Such a certification has to rely on statistical correlations measured among the qubits (forming a certain data set), and on controlled assumptions about their internal working. However, today, no scalable method exists for proving that a generic data set is incompatible with a non-entangled state. In this talk, I will present a new approach, inspired by inverse problems of statistical physics and data science, to solve this major problem of quantum information science. Specifically, I will show that this problem is equivalent to building a certain classical model (e.g. an Ising model), whose equilibrium Boltzmann distribution fits the data set of interest -- or showing that no such classical model exists. I will introduce the general concept of entanglement witnesses (Bell's inequalities, EPR-steering criteria...), and show how solving inverse statistical problems leads to optimal, data-tailored entanglement witnesses. Most importantly, this can be achieved in a scalable way under generic assumptions, leveraging on scalable techniques of statistical physics (e.g. Monte-Carlo methods).
Laboratoire de Physique et Modélisation des Milieux Condensés
