BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Asia/Hong_Kong
X-LIC-LOCATION:Asia/Hong_Kong
BEGIN:STANDARD
TZOFFSETFROM:+0800
TZOFFSETTO:+0800
TZNAME:HKT
DTSTART:19911015T033000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20251218T030657Z
LOCATION:Meeting Room S221\, Level 2
DTSTART;TZID=Asia/Hong_Kong:20251215T145000
DTEND;TZID=Asia/Hong_Kong:20251215T150000
UID:siggraphasia_SIGGRAPH Asia 2025_sess111_papers_1003@linklings.com
SUMMARY:Precise Gradient Discontinuities in Neural Fields for Subspace Phy
 sics
DESCRIPTION:Mengfei Liu, Yue Chang, and Zhecheng Wang (University of Toron
 to); Peter Yichen Chen (MIT CSAIL); and Eitan Grinspun (University of Toro
 nto)\n\nMany physical phenomena exhibit discontinuities in their spatial d
 erivatives—such as folds in creased materials or interfaces in heterogeneo
 us solids—making their accurate representation essential for high-fidelity
  simulation. Traditional approaches address such discontinuities by aligni
 ng mesh discretizations with the interface, but this tight coupling betwee
 n geometry and simulation limits generalization: changes in discontinuity 
 geometry require remeshing, which alters the system’s discrete operators a
 nd prevents consistent reuse of reduced-order basis. Since these basis are
  typically derived from mesh-dependent operators, applying reduced-order m
 odeling across varying geometries remains a fundamental challenge.Neural r
 epresentations offer an alternative by encoding basis functions as continu
 ous neural fields, enabling generalization across shape variations. Howeve
 r, their inherent continuity makes it difficult to represent functions wit
 h discontinuous gradients. While recent work has explored discontinuities 
 in function values, modeling continuous functions with discontinuous deriv
 atives has remained largely unexplored.We introduce a neural field constru
 ction capable of capturing gradient discontinuities while maintaining cont
 inuity in the function itself. Our approach augments input coordinates wit
 h a non-trainable, smoothly clamped distance function within a lifting fra
 mework, allowing the gradient discontinuity to be encoded explicitly. We s
 how that this construction yields higher-quality basis functions compared 
 to traditional neural fields and supports reduced-order simulation across 
 families of shapes with heterogeneous materials and creases—capabilities n
 ot demonstrated by prior work. Furthermore, our method can be combined wit
 h previous techniques that model function-value discontinuities via liftin
 g, enabling the simulation of examples with simultaneous cuts and creases.
 \n\nRegistration Category: Full Access, Full Access Supporter\n\nSession C
 hair: Ligang Liu (University of Science and Technology of China)\n\n
END:VEVENT
END:VCALENDAR