1: \begin{abstract}

2:   We consider stochastic optimization over $\ell_p$ spaces using

3:   access to a first-order oracle.  We ask: {What is the minimum

4:     precision required for oracle outputs to retain the unrestricted

5:     convergence rates?}  We characterize this precision for every

6:   $p\geq 1$ by deriving information theoretic lower bounds and by

7:   providing quantizers that (almost) achieve these lower bounds.  Our

8:   quantizers are new and easy to implement.  In particular, our

9:   results are exact for $p=2$ and $p=\infty$, showing the minimum

10:   precision needed in these settings are $\Theta(d)$ and $\Theta(\log

11:   d)$, respectively. The latter result is surprising since recovering

12:   the gradient vector will require $\Omega(d)$ bits.

13: \end{abstract}

14: