We know that A function $f:\Bbb Z_2^n \longrightarrow\Bbb Z_2^m$ is a (strong) one-way function (OWF), if:
- $f$ can be computed by a PT algorithm. Equivalently, there exists a PPT algorithm that on input $x$ outputs $f(x)$, for all $x\in \Bbb Z_2^n$.
- For all PPT adversaries $A$, there exists a negligible function $\operatorname{negl}_A(n)$ such that, for all large enough $n$, we have $$ \operatorname{Pr}\left[f(x)=f(z): x \stackrel{\\\$}{\leftarrow}\Bbb Z_2^n,\ z \leftarrow A(f(x))\right] \leq \operatorname{negl}_A(n) .$$ In other words, in terms of Games, for all PPT adversaries $A$ who play the Game $\operatorname{OWF}$ (getting $f(x)\in \Bbb Z_2^m$ and returning some pre-image $z\in \Bbb Z_2^n$), where
Game $\operatorname{OWF}_f$
Procedure Initialize
$x \stackrel{\\\$}{\leftarrow}\Bbb Z_2^n$
Return $f(x)$
Procedure Finilize($z$)
Return ($f(x)=f(z)$)
there exists a negligible function $\operatorname{negl}_A(n)$ such that, for all large enough $n$, we have $$\Pr[\operatorname{OWF}_f^A\implies \text{True}]\leq \operatorname{negl}_A(n). $$
My Question here is how can we interpret formally the negation of this definition? So, what is a non-OWF $f$?
My answer is that there exist an adversary $A$ such that the probability that $A$ wins the above Game is non-negligible (some people use the term noticeable).
Does this mean that the probability that, for any $x\in \Bbb Z_2^n$ (even if this $x$ is predefined by us), $A$ returns a correct pre-image $z$ of $f(x)$ is non-negligible?
In addition, does it mean that in this case we can feed $A$ with any $y\in \Bbb Z_2^m$, it can return us some $z\in \Bbb Z_2^n$ such that $y=f(z)$ with some noticeable probability? I think that this happens only if $f$ is surjective.
Thank you.
Update. So, to be even more precise, assume that we need to implement the following reduction; We want to show that if a function $f$ is not an OWF, then we can factor any $N\in \Bbb Z^+$. Check Prop 3 of 2.4.2 of these notes. Then, we write a pseudocode to define an adversary $B$ against $N$, which invokes $A$. And it says that $A$ takes $N$ and returns some value.
I can understand that $A$ is an algorithm, so we can put in it any value. But, in $A(N)$, how do we precisely employ the fact that $f$ is not an OWF?
I would say that if $f$ is surjective, then $N=f(x)$ for some specific $x$. So, if $A$ succeeds means that it can find a valid pre-image $z$ of $f(x)$.
Is the event of picking a specific $x_0$ from $\Bbb Z_2^n$ and feed $A$ with $f(x_0)$ a subset of the event that we pick uniformly at random an $x$ from $\Bbb Z_2^n$ and feed $A$ with $f(x)$?
