A distribution
is flat
if
.
Flat distributions are commonly used in practice. For example,
the most frequently used model for undirected random graphs
without self-loops is
with
.
It consists of all graphs with vertex set
in which vertices
are labeled and the edges are
chosen independently with probability
, which is a
function of
. So a graph
with vertex set
and
edges in this model occurs with probability
.
If we assume also that
satisfies
for some fixed
,
then
is flat [Gur91a], since
.
Similar distributions for directed graphs are also flat for the
same reason [WB95]. Most of the NP-complete
graph problems have other
parameters as inputs-for example, rather than
taking just a graph as input, the input may be
, where
is often
a bound on the
number of vertices
or the number of edges of some structure
the graph may have that the
problem is speaking of (clique, etc.).
So their distributions are flat.
Gurevich [Gur87] showed that
dealing with flat distributions to get
average-case completeness results
using p-time or ap-time reductions is impossible unless
.
Proof.a problem
with a flat distribution
is complete
under ap-time reductions.
Then
. Let
be an arbitrary decision problem in NTIME
for some polynomial
.
For any
, let
.
Then
is in NP. Define
by
for the appropriate constant
and
if
for any
.
Then
, which implies that
there is an ap-time reduction
from
to
. Hence, deciding whether
can be done by deciding whether
.
Since
is computable in time polynomial on
-average,
computing
can be carried out in time
, due
to the particular selection of
.
Also,
for some
and
for all
, where
is polynomial on
-average and
so
again due to the definition of
.
Thus,
and so
for some polynomial
. Since
is flat,
there is a
such that
.
This implies that
.
For reductions
that are restricted to be one-one and length-preserving
within a polynomial (i.e., there is a polynomial
such that,
for all
,
),
Wang and Belanger [WB95] showed that the
above incompleteness theorem is true
without any assumption.
We note that all the natural NP-complete problems are indeed
complete under such reductions.
Proof.to the contrary that
there were an ap-time reduction
reducing DH1
to a problem
with flat distribution
,
where
is one-one and length-preserving within a polynomial.
Then
is weakly dominated by
with respect to
.
Since
is one-one,
is weakly dominated by
.
Since
is flat, there is an
such that
for all but finitely many
.
Since
is length-preserving within a polynomial,
there is a constant
such that, for all
,
.
When
is sufficiently large and greater than
,
. So
.
But
, which cannot
be weakly dominated by
, a contradiction.