BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260605T154541Z
LOCATION:Bldg. 6 - Room 102
DTSTART;TZID=Europe/Stockholm:20260701T150000
DTEND;TZID=Europe/Stockholm:20260701T153000
UID:submissions.pasc-conference.org_PASC26_sess161_msa218@linklings.com
SUMMARY:Random Sketches, Precise Electrons: A Randomized Approach to Densi
 ty Functional Theory Eigenproblems
DESCRIPTION:Moritz Gubler (Paul Scherrer Institute)\n\nElectronic-structur
 e methods form the computational foundation of modern materials science an
 d quantum chemistry, enabling predictions of molecular properties, reactio
 n mechanisms, and solid-state behavior from first principles. At the heart
  of these methods lies the iterative solution of large Hermitian eigenprob
 lems, arising from density functional theory, Hartree-Fock, and related ap
 proaches.\n\nRandomization provides a practical way to reduce the dimensio
 nality of large linear algebra objects while preserving the structure that
  matters for convergence. By forming a small random sketch\x9d of the sear
 ch space, one can approximate projections and orthogonalization through a 
 reduced representation, lowering memory movement and synchronization witho
 ut changing the mathematical goal of the iteration.\n\nThis structure make
 s mixed-precision a natural fit. Sketch space operations are carried out i
 n double precision to keep the reduced coordinates and projected operators
  stable, while vectors and updates in the full space are stored and applie
 d in single precision.\n\nIn this PASC-funded project, we apply these idea
 s to a randomized eigensolver for large Hermitian eigenproblems from elect
 ronic-structure calculations. The resulting algorithm is intended to run e
 fficiently on graphics processing units, leveraging mixed precision capabi
 lities of modern hardware to accelerate electronic-structure simulations.\
 n\nDomain: Chemistry and Materials, Computational Methods and Applied Math
 ematics\n\nSession Chair: Michael Herbst (EPFL)\n\n
END:VEVENT
END:VCALENDAR
