{"id":"arxiv/2607.12969","url":"https://arxiv.org/abs/2607.12969","tags":["math.NA"],"score":0,"title":"Low-Precision Rank Compensation for Matrices and Tensor Trains","domain":"math.NA","source":"arxiv","authors":["Ibrahim El Mountasser","Khalide Jbilou"],"reviews":0,"subject":"math.NA","summary":"Lower numerical precision reduces storage and memory traffic but raises the perturbation floor. We study rank compensation: reinvesting saved memory in a larger approximation rank. For matrices, the singular-value error identity yields a directly testable sufficient condition requiring the additional singular component to offset the perturbation from storing the rank-augmented approximation in lower precision. On ten SuiteSparse matrices, all 100 truncation-dominated configurations (50 FP32 and 50 FP16) are certified non-increases and strict accuracy wins, with mean error ratio $0.963$ and…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-07-14","reactions":{}}