{"id":"arxiv/2609.32729","url":"https://arxiv.org/abs/2609.32729","tags":["math.OC","cs.LG"],"score":0,"title":"Revisiting AdaGrad in Stochastic Convex Optimization: Last Iterates, High Probability, and Lower Bounds","domain":"math.OC","source":"arxiv","authors":["Weiming Ou","Xiao Wang"],"reviews":0,"subject":"math.OC","summary":"AdaGrad and AdaGrad-Norm are widely used adaptive methods, but their precise behavior in stochastic convex optimization remains less understood. We first show that AdaGrad-Norm and AdaGrad do not admit any universal \\textbf{last-iterate} rate, even under sub-Gaussian noise and bounded iterates. We then prove that bounded variance alone is too weak: even with bounded iterates, it cannot yield \\textbf{high-probability average-iterate} rates, and without bounded iterates it may not even guarantee convergence in expectation. Besides, we construct tight $Ω(\\log T/\\sqrt T)$ lower bounds for both…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-09-26","reactions":{}}