BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Theory - Chemistry Research Interest Group
SUMMARY:Uncertainty-Driven Construction of Markov Models f
rom Accelerated Molecular Dynamics - Dr Thomas D S
winburne\, Aix-Marseille University
DTSTART;TZID=Europe/London:20190206T141500
DTEND;TZID=Europe/London:20190206T151500
UID:TALK113914AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/113914
DESCRIPTION:A common way of representing the long-time dynamic
s of materials is in terms of a Markov chain that
specifies the transition rates for transitions bet
ween metastable states. This chain can either be u
sed to generate trajectories using kinetic Monte C
arlo\, or analyzed directly\, e.g.\, in terms of f
irst passage times between distant states. While a
number of approaches have been proposed to infer
such a representation from direct molecular dynami
cs (MD) simulations\, challenges remain. For examp
le\, as chains inferred from a finite amount of MD
will in general be incomplete\, quantifying their
completeness is extremely desirable. Second\, mak
ing the construction of the chain as computational
ly affordable as possible is paramount. \n\nI will
talk about some recent work [1] to address these
two questions. We first quantify the local complet
eness of the chain in terms of Bayesian estimators
of the yet-unobserved rate\, and its global compl
eteness in terms of the residence time of trajecto
ries within the explored subspace. We then systema
tically reduce the cost of creating the chain by m
aximizing the increase in residence time against t
he distribution of states in which additional MD i
s carried out and the temperature at which these a
re respectively carried out. Using as example the
behavior of vacancy and interstitial clusters in m
aterials\, we demonstrate that this is an efficien
t\, fully automated\, and massively-parallel schem
e to efficiently explore the long-time behavior of
materials. We also show how accommodation of exch
ange\, rotation\, reflection and translation symme
tries can massively enhance sampling efficiency.\n
\n[1] TD Swinburne and D Perez\, Self-optimized co
nstruction of transition rate matrices from accele
rated atomistic simulations with Bayesian uncertai
nty quantification\, Physical Review Materials 201
8
LOCATION:Department of Chemistry\, Cambridge\, Unilever lec
ture theatre
CONTACT:Lisa Masters
END:VEVENT
END:VCALENDAR