Code and Data for "Conditioning with Prescribed Subspaces under Inverse-Trace Normalization"
Year: 2026
Type: software
Abstract: This record contains the reproducibility materials for the manuscript "Conditioning with Prescribed Subspaces under Inverse-Trace Normalization" by Fangyue Liang. The package supports the study of condition-number minimization for positive definite additive loadings subject to a weighted inverse-trace budget, a prescribed leading subspace, and an output spectral gap. It includes Python implementations and frozen inputs for evaluating explicit scalar-block exactness thresholds, comparing elementary and refined bounds, checking the necessary inverse-gap family, and reconstructing non-scalar examples. The supplied computations also cover a fixed-readout low-rank approximation example and its factor implementation, with saved factors, data splits, preprocessing maps, fitted heads, and independent verification scripts. The archive provides high-precision numerical checks, symbolic diagnostics, parameter settings, hash manifests, and recorded results. A documented reproduction driver regenerates the current mathematical and factor-interface calculations from the included inputs and compares the outputs with the archived results. This workflow uses the supplied frozen readout heads; training source and separate dependencies are also retained. Manuscript and computational-supplement LaTeX sources and figure-generation code connect the computational evidence to the paper. See README_Zenodo.txt for an overview, README.md inside the archive for setup and execution instructions, and LICENSE_SCOPE.txt for the licensing of original and third-party materials.
Author: Liang, Fangyue
Institution: Tsinghua University
Language: English
Topic: Matrix Theory and Algorithms
Subfield: Computational Theory and Mathematics
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7235148560
