BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
X-WR-TIMEZONE;VALUE=TEXT:Europe/Zurich
BEGIN:VEVENT
DTSTAMP:20260924T092951Z
UID:a3f3ceb6-02d7-4be8-88bf-085390ef8b1c
DTSTART:20260806T110000
DTEND:20260806T130000
DESCRIPTION:Neural quantum states provide accurate variational representati
 ons of strongly correlated ground states\, but computing dynamical quantit
 ies is expensive\, usually requiring many retrainings of the full network.
  We study a hybrid architecture that combines the expressivity of NQS with
  the interpretability of tensor networks\, confining all the physics of ex
 citations to a small\, linearly solvable set of parameters: the network en
 codes the complex\, long range correlations of the ground state\, while a 
 small tensor network head attached to the encoder output represents locali
 zed excitations on top of it. Since the head enters linearly\, it is optim
 ized by solving an eigenvalue problem\, giving the excitations at all mome
 nta at the same time\, without retraining the rest of the network.\n\nWe a
 pply this construction to large two dimensional spin systems\, extracting 
 excitation spectra and dynamical structure factors at negligible cost comp
 ared to methods requiring explicit real time evolution. Finally\, we show 
 that the same framework can be extended to real time dynamics and fermioni
 c systems\, whenever the relevant physics is accurately captured by a line
 ar perturbation of the ground state.
LOCATION:ETH Zürich\, Hönggerberg HIT E 41.1
ORGANIZER:Niccolò Baldelli
SUMMARY:Efficient computation of dynamical structure factors using Hybrid T
 ensor Network-Neural Network architectures
END:VEVENT
END:VCALENDAR