In my last post I promised a simple proof that there are no deterministic extractors for SV sources. The proof is due to Salil Vadhan, Avi Wigderson and myself and its technique has been used to obtain impossibility results on doing cryptography with SV sources. We show that even more restricted classes of sources are not amenable to deterministic extraction:
Definition For , a source
of length
is
-imbalanced if for every
,
.
A source which is
-imbalanced is an SV source with bias
in a very strong sense: every bit
has bias at most
not only conditioned on the first
bits
but also conditioned on all
other bits
. The following lemma shows that there are no deterministic extractors for imbalanced sources and therefore also for SV-sources.
Lemma For every function and every
, there is a
-imbalanced source
such that
has bias at least
.
Proof There exists a set of size exactly
such that
is constant on
, say taking value
. Consider the source
that with probability
outputs a uniformly selected element of
and with probability
outputs a uniformly selected element of
. By construction,
with probability at least
, so it has bias at least
. Moreover,
assigns every string in
probability mass either
or
. Thus
is
-imbalanced.
Nice! Despite having taught this proof in a class (!), I was not aware of its origin.