Long code (mathematics)

This is the approved revision of this page, as well as being the most recent.

In theoretical computer science and coding theory, the long code is an error-correcting code that is locally decodable. Long codes have an extremely poor rate, but play a fundamental role in the theory of hardness of approximation.

Definition

Let $f_1,\dots,f_{2^n} : \{0,1\}^k\to \{0,1\}$ for $k=\log n$ be the list of all functions from $\{0,1\}^k\to\{0,1\}$. Then the long code encoding of a message $x\in\{0,1\}^k$ is the string $f_1(x)\circ f_2(x)\circ\dots\circ f_{2^n}(x)$ where $\circ$ denotes concatenation of strings. This string has length $2^n=2^{2^k}$.

The Walsh-Hadamard code is a subcode of the long code, and can be obtained by only using functions $f_i$ that are linear functions when interpreted as functions $\mathbb F_2^k\to\mathbb F_2$ on the finite field with two elements. Since there are only $2^k$ such functions, the block length of the Walsh-Hadamard code is $2^k$.

An equivalent definition of the long code is as follows: The Long code encoding of $j\in[n]$ is defined to be the truth table of the Boolean dictatorship function on the $j$th coordinate, i.e., the truth table of $f:\{0,1\}^n\to\{0,1\}$ with $f(x_1,\dots,x_n)=x_j$. Thus, the Long code encodes a $(\log n)$-bit string as a $2^n$-bit string.

Properties

The long code does not contain repetitions, in the sense that the function $f_i$ computing the $i$th bit of the output is different from any function $f_j$ computing the $j$th bit of the output for $j\neq i$. Among all codes that do not contain repetitions, the long code has the longest possible output. Moreover, it contains all non-repeating codes as a subcode.