BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:CD Technical Meeting (ML16): Surrogate models for materials in fus
 ion power plants
DTSTART:20260909T100000Z
DTEND:20260909T103000Z
DTSTAMP:20260909T105600Z
UID:indico-event-817@indico.ukaea.uk
DESCRIPTION:Speakers: Eszter Szekely (UKAEA)\n\nMachine Learning\, Uncerta
 inty Quantification and Data Science \n \n\nTo accelerate dynamical simu
 lations of irradiated metals\, we combine coarse-graining techniques with 
 data-driven surrogates\, specifically under conditions that will occur in 
 working fusion power plants which are expensive to test experimentally. Si
 mulating longer timescales will bring us closer to the scales of the lifec
 ycle of fusion power plants. We use autoregressive methods fitted to data 
 gained from coarse-graining molecular dynamics (MD) simulations.  \n \n
 \n\n1. Fusion-relevant conditions \n\n\nFuture fusion power plants could 
 be a safe and low-carbon energy source that would allow humanity to mitiga
 te climate change. Achieving fusion on Earth requires temperatures excee
 ding those of the Sun’s core\, creating extreme thermal conditions. Add
 itionally\, the wall materials of the power plant will need to withstand i
 rradiation by high-energy neutrons causing structural and compositional ch
 anges over time. Tungsten is a prime candidate wall material due to its hi
 gh melting point\, low erosion rate and low tritium fuel retention. Howe
 ver\, its behaviour under prolonged neutron irradiation and high heat flu
 x is not fully understood yet. As the fusion conditions are hard to reach 
 experimentally\, computational studies are valuable for the design choices
  of the power plant. \n\n\n \n\n\nState-of-the-art MD simulations provi
 de good descriptions of irradiation effects in metals [2]. In this project
 \, we create surrogates for these simulations to reduce computational cost
 s greatly and to make it feasible to simulate longer timescales. \n\n\n
  \n\n\n2. Forecasting coarse-grained atomistic descriptors \n\n\nAtomi
 stic descriptors (D) are geometric representations of atomic environments.
  They have been successfully applied as basis functions for developing che
 mical models [1\,3]. Here\, we reduce the dimensions of D by coarse-graini
 ng descriptors ( ˜D) in a space-averaged way following Ref. [4]\, but int
 roducing voxelisation. ˜D have been shown to be a good latent space to pr
 opagate MD using vector autoregression (VAR) models\, and also to fit mode
 ls predicting global physical properties of interest [ 4]. Thus\, the surr
 ogates forecast simulation states with negligible computational costs\, an
 d can predict various global observables of engineering interest [4]. How
 ever\, Ref. [4]’s studies were limited to forecasting MD under fixed ph
 ysical conditions.  \n\n\n \n\n\nHere\, we study the applicability of t
 he method for building models forecasting dynamical simulations of irradi
 ation\, transferable to different physical conditions. We build our data
 sets combining techniques suggested in Ref. [2] and [4]. As our target dy
 namical systems are more complex\, we improve the methodology for accur
 ate forecasting. First\, we apply voxelised descriptors which gives ac
 cess to semi-local observables along with the global ones\, and makes it
  possible to fine-tune the level of coarse-graining in the surrogates. F
 urther\, we study how applying more advanced ML models and using differen
 t timescales for the fits affect the accuracy of the predictions. We val
 idate our models comparing the predicted observables to physical proper
 ties computed from MD simulations. \n\n\n \n\n\nAcknowledgments \n\n\nT
 his work has been funded by the Fusion Futures Programme. As announced by 
 the UK Government in October 2023\, Fusion Futures aims to provide holis
 tic sup- \n\n\nport for the development of the fusion sector. This work 
 used the ARCHER2 UK National Supercomputing Service (https://www.archer2.
 ac.uk) and the Pitagora Supercomputer hosted by CINECA under the TADFRM
  project. \n\n\n \n\n\nReferences \n\n\n[1] Albert P. Bartók. Gaussian
  Approximation Potential: an interatomic potential derived from first pri
 nciples Quantum Mechanics. Ph.D. thesis\, University of Cambridge\, Engin
 eering Dept\, Trumpington St\, Cambridge CB2 1PZ\, 2010. \n\n\n[2] Max B
 oleininger\, Daniel R. Mason\, Andrea E. Sand\, and Sergei L. Dudarev. Mi
 crostructure of a heavily irradiated metal exposed to a spectrum of atomi
 c recoils. \n\n\nScientific Reports\, 13(1):1684\, Jan 2023. \n\n\n[3] T
 huong T. Nguyen\, Eszter Székely\, Giulio Imbalzano\, Jörg Behler\, Gá
 bor Csányi\, Michele Ceriotti\, Andreas W. Götz\, and Francesco Paesani
 . Comparison of permutationally invariant polynomials\, neural networks\,
  and Gaussian approximation potentials in representing water interaction
 s through many-body expansions. J. Chem. Phys.\, 148(24):241725\, 2018. \
 n\n\n[4] Thomas D Swinburne. Coarse-graining and forecasting atomic mater
 ial simulations with descriptors. Physical Review Letters\, 131(23):23610
 1\, 2023. \n\n\n \n\n\nhttps://indico.ukaea.uk/event/817/
URL:https://indico.ukaea.uk/event/817/
END:VEVENT
END:VCALENDAR