Pseudorandom Function (PRF): Difference between revisions
No edit summary |
No edit summary |
||
Line 10: | Line 10: | ||
=== Syntax === | === Syntax === | ||
A Pseudorandom Function (PRF) is a tuple of functions <math>(\mathsf{Gen}, | A Pseudorandom Function (PRF) is a tuple of functions <math>(\mathsf{Gen}, F)</math>, with respect to a keyspace <math>\mathcal{K}</math>, domain <math>\mathcal{D}</math>, and range <math>\mathcal{R}</math>, such that: | ||
* <math>\mathsf{Gen}(1^{\lambda}) \to k</math>, takes a security parameter, and outputs a key <math>k\in \mathcal{K}</math>, | * <math>\mathsf{Gen}(1^{\lambda}) \to k</math>, takes a security parameter, and outputs a key <math>k\in \mathcal{K}</math>, | ||
* <math> | * <math>F_k(x) \to y</math>, takes a key <math>k\in \mathcal{K}</math> and input <math>x\in \mathcal{D}</math>, and outputs an element <math>y\in \mathcal{R}</math>. | ||
Generally, we assume that \$|\mathcal{D}|\$ is "large," in the sense that it grows exponentially with the security parameter. If instead <math>|\mathcal{D}|</math> bounded by some polynomial in the security parameter, then the primitive is a "small-domain" PRF. | |||
=== Security === | === Security === | ||
A PRF is ''secure'' if for all efficient <math>D</math>, there exists a negligible function <math>\nu</math>, such that | |||
<math> | |||
\big| \Pr [D^{F_k}(1^{\lambda}) = 1: k\gets \mathsf{Gen}(1^{\lambda})] - | |||
\Pr [D^{R}(1^{\lambda}) = 1 : R \gets \mathcal{F}[\mathcal{D},\mathcal{R}]] \big| \le \nu(\lambda). | |||
</math> | |||
=== Weak Security === | === Weak Security === | ||
A PRF is ''weakly secure'' if for all efficient <math>D</math>, and all polynomials <math>s(\lambda)</math>, there exists a negligible function <math>\nu</math>, such that | |||
<math> | |||
\big| \Pr_{x_i \gets \mathcal{D}} [D((x_i,F_k(x_i))_{i\in [s(\lambda)]}) = 1] - | |||
\Pr_{x_i \gets \mathcal{D}} [D((x_i,R(x_i))_{i\in [s(\lambda)]}) = 1] \big| \le \nu(\lambda). | |||
</math> | |||
The only different in this definition is that we sample $s(\lambda)$ inputs ''at random'' instead of allowing a distinguisher to choose them. This is strictly weaker as a distinguisher could always query their oracle at random locations. | |||
== Relationship to other primitives == | |||
* [[PRG]]s imply the existence of [[PRF]]s, due to [[ccref#ggm84|[GGM84]]] (this means that [[OWF]]s imply the existence of [[PRF]]s, in combination with [[ccref#hill99|[HILL99]]]) | |||
* [[PRF]]s imply the existence of [[PRP]]s, due to [[ccref#lr88|[LR88]]] | |||
There are also many folklore results surrounding pseudorandom functions: | |||
* [[PRP]]s over are [[PRF]]s, due to the switching lemma [[folklore]] | |||
== Sufficient assumptions == | |||
* [[DDH]] implies [[PRF]]s | |||
* [[LWE]] implies [[PRF]]s | |||
== | == Variations == |
Revision as of 22:48, 27 June 2024
A Pseudorandom Function (PRF) is a primitive originally defined in [cite].
TOOD: information definition
Formal Definition
Syntax
A Pseudorandom Function (PRF) is a tuple of functions , with respect to a keyspace , domain , and range , such that:
- , takes a security parameter, and outputs a key ,
- , takes a key and input , and outputs an element .
Generally, we assume that \$|\mathcal{D}|\$ is "large," in the sense that it grows exponentially with the security parameter. If instead bounded by some polynomial in the security parameter, then the primitive is a "small-domain" PRF.
Security
A PRF is secure if for all efficient , there exists a negligible function , such that
Weak Security
A PRF is weakly secure if for all efficient , and all polynomials , there exists a negligible function , such that
The only different in this definition is that we sample $s(\lambda)$ inputs at random instead of allowing a distinguisher to choose them. This is strictly weaker as a distinguisher could always query their oracle at random locations.
Relationship to other primitives
- PRGs imply the existence of PRFs, due to [GGM84] (this means that OWFs imply the existence of PRFs, in combination with [HILL99])
- PRFs imply the existence of PRPs, due to [LR88]
There are also many folklore results surrounding pseudorandom functions:
Sufficient assumptions