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:20190719T085743Z
LOCATION:HG E 1.1
DTSTART;TZID=Europe/Stockholm:20190612T111500
DTEND;TZID=Europe/Stockholm:20190612T114500
UID:submissions.pasc-conference.org_PASC19_sess109_pap_jan116@linklings.co
 m
SUMMARY:Analytical PAW Projector Functions for Reduced Bandwidth Requireme
 nts
DESCRIPTION:Paper\nComputer Science and Applied Mathematics, Chemistry and
  Materials, Climate and Weather\n\nAnalytical PAW Projector Functions for 
 Reduced Bandwidth Requirements\n\nBaumeister, Tsukamoto\n\nLarge scale ele
 ctronic structure calculations require modern high performance comput
 ing (HPC) resources and, as important, mature HPC applications that c
 an make efficient use of those. Real-space grid-based applications of
  Density Functional Theory (DFT) using the Projector Augmented Wave m
 ethod (PAW) can give the same accuracy as DFT codes relying on a
  plane wave basis set but exhibit an improved scalability on distribu
 ted memory machines. The projection operations of the PAW Hamiltonian
  are known to be the performance critical part due to their limi
 tation by the available memory bandwidth. We investigate on the utili
 ty of a 3D factorizable basis of Hermite functions for the localized 
 PAW projector functions which allows to reduce the bandwidth requirem
 ents for the grid representation of the projector functions in projec
 tion operations. Additional on-the-fly sampling of the 1D basis funct
 ions eliminates the memory transfer almost entirely. For an quan
 titative assessment of the expected memory bandwidth savings we show 
 performance results of a first implementation on GPUs. Finally, we su
 ggest a PAW generation scheme adjusted to the analytically given proj
 ector functions.<br /><br />Full paper: https://doi.org/10.1145/3324989.33
 25717
END:VEVENT
END:VCALENDAR

