We establish conditions for which graph Laplacians Δλ,ϵ on compact, boundaryless, smooth submanifolds M of Euclidean space are semiclassical pseudodifferential operators (ΨDOs): essentially, that the graph Laplacian's kernel bandwidth (bias term) ϵ decays faster than the semiclassical parameter h, i.e., h≫ϵ and we compute the symbol. Coupling this with Egorov's theorem and coherent states ψh localized at (x0,ξ0)∈T∗M, we show that with Uλ,ϵt:=e−itΔλ,ϵ spectrally defined, the (co-)geodesic flow Γt on T∗M is approximated by ⟨Uλ,ϵ−tOph(a)Uλ,ϵtψh,ψh⟩=a∘Γt(x0,ξ0)+O(h). Then, we turn to the discrete setting: for Δλ,ϵ,N a normalized graph Laplacian defined on a set of N points x1,…,xN sampled i.i.d. from a probability distribution with smooth density, we establish Bernstein-type lower bounds on the probability that ∣∣Uλ,ϵ,Nt[u]−Uλ,ϵt[u]∣∣L∞≤δ with Uλ,ϵ,Nt:=e−itΔλ,ϵ,N. We apply this to coherent states to show that the geodesic flow on M can be approximated by matrix dynamics on the discrete sample set, namely that with high probability, ct,N−1∑j=1N∣Uλ,ϵ,Nt[ψh](xj)∣2u(xj)=u(xt)+O(h) for ct,N:=∑j=1N∣Uλ,ϵ,Nt[ψh](xj)∣2 and xt the projection of Γt(x0,ξ0) onto M.
@article{arxiv.2112.10748,
title = {On a quantum-classical correspondence: from graphs to manifolds},
author = {Akshat Kumar},
journal= {arXiv preprint arXiv:2112.10748},
year = {2022}
}
Comments
This is a companion paper to "Manifold learning via quantum dynamics"