Equivalence of NBU PRG and PRG
From CRYPTUTOR
We show that two definitions of pseudorandom generators are equivalent:
- The standard definition, based on computational indistinguishability, which we call the IND-PRG definition.
- An alternative definition based on next-bit unpredictability, which we call the NBU-PRG definition.
IND-PRG implies NBU-PRG
The IND-PRG definition is a general definition that includes all ways of distinguishing the pseudorandom distribution from random, including the method in the NBU-PRG experiment.
Suppose G is an IND-PRG. For any NBU-PRG adversary A, we can give to A the first i − 1 bits of a string and check if its guess for the ith bit is correct. This process distinguishes G's output distribution from the uniform distribution with the same advantage as A in the NBU-PRG experiment. Since we assumed G was an IND-PRG, this advantage must be negligible.
NBU-PRG implies IND-PRG
Let
be an NBU-PRG. We use a hybrid argument to establish that it also must satisfy the IND-PRG definition. We define the following hybrid distributions over
-bit strings:
Observe that
is the uniform distribution, while
is G's output distribution
. Since
is a polynomial in the security parameter k, all that remains is to show that adjacent hybrids are computationally indistinguishable.
Suppose for contradiction that
and
are not computationally indistinguishable. Then some adversary A can distinguish between the distributions with non-negligible advantage
. Let us construct an NBU-PRG adversary
to predict the ith bit of G:
:
- On input
, generate truly random bits
.
- Give
to A.
- If A outputs 0, output
, otherwise output its complement (as our guess for the ith bit of G).
We now analyze the advantage of
in guessing the ith bit of G. Let
be the event that it guesses correctly.
So
, and
has non-negligible advantage in the NBU-PRG experiment -- a contradiction. Thus the adjacent hybrid distributions
and
are computationally indistinguishable, as desired.
, generate truly random bits
.
to
, otherwise output its complement (as our guess for the 
