Information Technology Reference
In-Depth Information
j
1
The quotient of these two binomial products is
so the number of subsets
that become lethal upon redistribution again exceeds the number of those that be-
come non-lethal. Now we claim that M k <
<
1
,
j
N 1 , so it is better to put a single unit
dose. Indeed M k = 2 k k while N 1
= 2 k 2
k
1 . So putting a dose 1
/
j for j
>
1 is only
1
j .
optimal once h
2
We have an amount of poison h that we distribute over the biscuits, putting a
dose h i in the i -th biscuit. A k -subset V is lethal if and only if h
1.
We number the biscuits in decreasing order of their doses, putting the most poi-
sonous biscuit first, i.e., h 1 ≥ ··· ≥
(
V
)= i V h i
be the family of poisonous k -
subsets. We want to distribute the poison in such a way that
h 2 k 1 .Let
P
P
has maximum car-
dinality. We adopt hypergraph terminology. We say that V
∈ P
is an edge, and
deg
P (
i
)
is equal to the number of edges that contains i .
2 2 k 2
1 .
1
1
Lemma 2. If h
<
2
then deg
P (
2 j
+
1
)
j
+
1
k
Proof. By the decreasing dosage of poison
2 j
+
1
(
2 j
+
1
)
h 2 j + 1
h 1 + ··· +
h 2 j + 1
h
<
1 ,
+
j
1thenlet V
V c
+
<
+
=
∪{
+
and so h
h 2 j + 1
2. If V is any k -subset that contains 2 j
2 j
.Inotherwords, V is the neighbor of V in the odd graph
}
O k that is connected by
1
V
the edge that has 2 j
+
1 as the odd one in. Then h
(
V
)+
h
(
)=
h
+
h 2 j + 1 <
2. So if
V is poisonous then V is not, and we conclude that deg
P (
2 j
+
1
)
is at most half of
thedegreeof2 j
+
1 in the complete hypergraph on all k subsets. The degree of the
complete hypergraph is 2 k 2
k
1 .
1
Lemma 3. If h
<
2
then the number of lethal edges is at most
j
+
1
2 j
j
2 k
2 j
i
2 k
k
1
2
2 j
1
2 j
1
+
.
k
j
k
i
i
=
j
+
1
Proof. We maximize the number of edges V under the constraint that the hypergraph
has maximal
i 2 j + 1 deg
(
i
)
, which by the previous lemma is bounded by
2 k
n
2 j
2
.
2
k
1
The greedy solution is to first take all k -subsets that have no elements in
{
2 j
+
1
,
etc, until the sum of the degrees exceed the given bound. We need to show that this
happens exactly when we have taken all k -subsets that contain
,...,
2 k
1
}
,thentotakeall k -subsets that have one element in
{
2 j
+
1
,...,
2 k
1
}
>
j elements from
{
and half of the k -subsets that contain exactly j elements from this set.
In other words, we need to show that
1
,...,
2 j
}
 
Search WWH ::




Custom Search