Title:Correction of Samplable Additive Errors

Abstract:We study the correctability of efficiently samplable errors. Specifically, we consider the setting in which errors are efficiently samplable without the knowledge of the code or the transmitted codeword, and the error rate is not bounded. We show that, assuming the existence of one-way functions, there is a samplable distribution $Z$ over $\{0,1\}^n$ with entropy $n^\epsilon$ for $0 < \epsilon < 1$ that is not correctable by efficient coding schemes. Then, we show that there is an oracle relative to which there is a samplable $Z$ with entropy $\omega( \log n)$ that is not correctable by efficient syndrome decoding for rate $R > \omega(\log n/n)$. Also, we show that there is an oracle relative to which there is a samplable $Z$ with entropy $\omega( \log n)$ that is not correctable by efficient coding schemes for rate $R < 1 - \omega(\log n /n)$. For flat distributions $Z$, we show that if $Z$ forms a linear subspace of dimension $m$, there is a linear code that corrects $Z$ with rate $R \leq 1 - m/n$. For general flat distributions $Z$, there is a (non-explicit) linear code that corrects $Z$ with error $\epsilon$ for rate $R \leq 1 - m/n - O(\log(1/\epsilon)/n)$, and no coding scheme can correct $Z$ with error $\epsilon$ for rate $R > 1 - m/n + \log(1/(1-\epsilon))/n$. Finally, we show that the existence of one-way functions is necessary to derive the impossibility results.