PDEformer-2 Extends Foundation Models to Two-Dimensional PDEs
A computational-graph representation and coordinate queries let one pretrained model address diverse 2D PDE forms after training on a 40 TB dataset.
Sep 15, 20264 min2507.15409
Numerical algorithms for problems in analysis and algebra.
A computational-graph representation and coordinate queries let one pretrained model address diverse 2D PDE forms after training on a 40 TB dataset.
Mapping equal-mass 1D states into CDT space makes translations affine, yielding zero 2-width for linear transport and O(n⁻¹) shock-tolerant bounds.
A modal DG solver stores the Vlasov distribution in tensor-train form, reaching 10⁴-scale compression on weakly nonlinear benchmarks.