AI & ML
Projection geometry and relaxed quasi-orthogonality for inf-sup stable Galerkin methods
arXiv cs.NA0 views
arXiv:2609.16265v2 Announce Type: replace Abstract: We give an elementary, coordinate-free proof that uniformly inf-sup stable nested Petrov-Galerkin methods on Hilbert spaces satisfy relaxed general quasi-orthogonality. More precisely, for any window of $N$ consecutive levels, the sum of squared Galerkin increments is bounded by the squared error at the beginning of the window times $N^\sigma$, where $\sigma<1$. The exponent $\sigma$ and the prefactor can be chosen explicitly in terms of the uniform projection bound. The proof uses Hilbert-space projection estimates, dyadic iteration, and duality, without relying on matrix factorizations. As a corollary, we obtain explicit growth bounds for finite block-LU factors. In the symmetric indefinite case, the positive and negative spectral splittings of the Galerkin detail spaces yield a characterization of full quasi-orthogonality in terms of uniform boundedness of the associated finite-window sign operators. We also construct a self-adjoint involution and a nested, uniformly inf-sup stable Galerkin sequence for which full general quasi-orthogonality fails. For every $0<\alpha<1$, the construction yields a hierarchy whose finite-window constants have growth of exact order $N^\alpha$, along with an explicit target whose Galerkin increments are not square summable.
Source:https://arxiv.org/abs/2609.16265PreviousDynamical Low-Rank Filters for Data AssimilationNextNumerical Methods for Dynamical Low-Rank Approximations of Stochastic Differential Equations -- Part II: Stochastic discretization
Related posts

The Download: the agentic web arrives and the drive for safer roads
0MIT Technology Review

Stop counting tokens, start counting what ships
0MIT Technology Review
OpenAI is finally merging ChatGPT’s chat and Work tabs
0TechCrunch

10 Open-Source Robotics Repositories for 3D Perception, VLA Policies, and Sensor Calibration
0dev.to
My self-modifying system tried to wire a gate-passing module into its own protection enforcer. It was refused three times.
0dev.to
Last night an AI agent performed brain surgery on a mouse
0Hacker News