{"id":"arxiv/2608.17307","url":"https://arxiv.org/abs/2608.17307","tags":["math.OC"],"score":0,"title":"A note on the complexity of random subspace model-based methods for derivative-free optimization","domain":"math.OC","source":"arxiv","authors":["Coralia Cartis","Lindon Roberts"],"reviews":0,"subject":"math.OC","summary":"We demonstrate that, with a suitable rescaling, using Johnson-Lindenstrauss transforms (JLTs) in the random subspace model-based derivative-free optimization (DFO) algorithm from [Cartis & Roberts, Math. Prog. 199 (2023)] achieves an improved first-order worst-case evaluation complexity bound compared to 'unscaled' JLTs. This improved bound does not require any modification to the original algorithm or complexity analysis, and matches the best-known bound in terms of dimension dependence for model-based DFO (e.g. achieved by [Scheinberg & Chaudhry, ICM 2026] using Haar matrices in a similar…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-08-18","reactions":{}}