Abstract:We show that one-way functions exist if and only there exists an efficient distribution relative to which almost-optimal compression is hard on average. The result is obtained by combining a theorem of Ilango, Ren, and Santhanam and one by Bauwens and Zimand.
What problem does this paper attempt to address?
The core problem that this paper attempts to solve is the relationship between the existence of one - way functions (OWF) and the average complexity of almost - optimal compression. Specifically, the author attempts to prove that the following two propositions are equivalent:
1. **Existence of one - way functions**: That is, there exists a polynomial - time computable function \( f: \{0,1\}^* \to \{0,1\}^* \) such that for any probabilistic polynomial - time algorithm Inverter, with a sufficiently large input length \( n \), the success rate of inverting \( f \) is very low.
2. **Almost - optimal compression is difficult under certain distributions**: That is, there exists a polynomial - time samplable distribution \( D \) such that for any probabilistic polynomial - time compression algorithm Compress, with a sufficiently large input length \( n \), it is difficult to compress the string \( x \) to a length close to its Kolmogorov complexity \( K(x) \).
### Specific problem description
#### Existence of one - way functions
A one - way function refers to a function that is easy to compute but difficult to invert. Formally, a one - way function \( f \) satisfies the following conditions:
- There exists a polynomial - time algorithm that can compute \( f(x) \) given an input \( x \).
- For any probabilistic polynomial - time algorithm Inverter, for any length \( n \) and any constant \( q\in\mathbb{N} \), we have:
\[
\Pr_{x \leftarrow U_n, \text{Inverter}}[\text{Inverter}(1^n, f(x)) \in f^{-1}(f(x))] \leq \frac{1}{n^q}
\]
where \( U_n \) represents the uniform distribution of \( n \)-bit length.
#### Difficulty of almost - optimal compression
Almost - optimal compression refers to compressing a string \( x \) to a length close to its Kolmogorov complexity \( K(x) \). Formally, if there exists a polynomial - time samplable distribution \( D \), such that for any probabilistic polynomial - time compression algorithm Compress, for any length \( n \) and constant \( c \), we have:
\[
\Pr_{x \leftarrow D_n, \text{Compress}}[\text{Compress}(x) \text{ outputs a program of } x \text{ of length } \leq K(x)+c\log^2 n] \leq \frac{1}{n}
\]
### Main contributions of the paper
By combining the results of Ilango, Ren and Santhanam [IRS21, IRS22] and Bauwens and Zimand [BZ23], the paper proves that the above two propositions are equivalent. Specifically, the paper shows the proofs in the following two directions:
1. **Deriving the difficulty of almost - optimal compression from the existence of one - way functions**: Assuming the existence of one - way functions, it is possible to prove that almost - optimal compression is difficult under certain distributions by constructing a pseudorandom generator (p.r.g.).
2. **Deriving the existence of one - way functions from the difficulty of almost - optimal compression**: Assuming that almost - optimal compression is difficult under certain distributions, it can be proved that there is no effective inversion algorithm, thus deriving the existence of one - way functions.
### Summary
The main goal of this paper is to establish the equivalence relationship between the existence of one - way functions and the average complexity of almost - optimal compression, thus providing new perspectives and tools for cryptography and computational complexity theory.