Next: Bounds for
Up: Time-Stamped Graphs and Their
Previous: Graphs
The Restricted Question
In this section we consider the most restrictive parameter, giving us the
following problem.
Problem 4.1
Let
be a digraph with
vertices.
Is there a graph
in the class
with
vertices
and a(n injective) time-stamp function
such that
?
We say that
is realizable if the answer to Problem 4.1
is yes. For the remainder of this paper, we assume that the class
is the class of all graphs. (Hence in light of the observation in
Section 1, we can allow
to be non-injective in our proofs without
any loss of strength.)
Unlike the more relaxed Problem 1.1,
the answer to Problem 4.1 is often no.
For example, suppose
is an undirected connected simple graph
(i.e., viewed as a digraph, all of its arcs occur in antiparallel pairs).
If there exists a vertex
such
that the subgraph of
induced by the neighbours of
is not connected
or
deg
, then
is not realizable as an associated influence
digraph. This will be proved below.
So it is natural to consider the following definition and problem.
Definition 4.2
Given a digraph
, let
where the minimum is taken over all time-stamped graphs
.
Problem 4.3
Find
for a given digraph
.
We note that
is well-defined, as we will give an upper bound
for
in Proposition 5.1.
As an application, let
be a digraph on the set of mathematicians. Then if
is not the collaboration
influence digraph for mathematicians, is it perhaps an
induced subdigraph of the influence digraph for
mathematicians and physicists?
Clearly a digraph
is realizable if and only if
. We now give
examples in which
. Although we are dealing with directed graphs, we
wonder whether the special case when
is generated from an undirected graph
would be easier. These examples consider the general case
when
directed and the special case when
is an undirected graph
(each undirected edge representing two arcs).
Proposition 4.4
Suppose
is a directed graph with at least three vertices
whose underlying undirected graph2
is connected3. If there exists a vertex
such that
, then
.
Proof.
Suppose

. Let

be a time-stamped graph giving rise to

.
Then

and

(and hence

and

)
have the same vertex set. Clearly
deg

for every vertex

. Suppose
deg

.
Note that

is connected since

is connected.
Clearly

must have degree 1 in

as
deg

.
Now,

has at least three vertices,

is connected, and

has degree 1. Therefore
there must be a path of length 2 of the form

. Note that

and

influence each other. Now, either

influences

or vice versa,
depending on the time-stamps on the edges

and

.
In either case, this implies
deg

, a contradiction.
Corollary 4.5
Suppose
is an undirected connected graph. If there exists a vertex
such that
, then
.
Proposition 4.6
Suppose
is a directed graph
whose underlying undirected graph
is connected.
If there exists a vertex
such
that the subgraph of
induced by the neighbours of
has
components, then
.
Proof.
Let the vertex set of

be

.
Let

with vertex set

be a time-stamped graph such that

.
Let the components of the subgraph of

induced by the neighbours of

be

.
For

, there is a non-decreasing path

in

between

and a vertex

in

, and we may assume that every internal
vertex (if any) belongs to

.
If

is of length at least 2, let

be the first internal vertex on

from

to

in

. (We note that although

may be a
non-decreasing path from

to

,

will still be selected according
to this rule.) Call

the
slave of

and

the
master of

.
We claim that at most one of

is of length 1. Suppose not, say

and

are of length 1. Then
we have a path of length 2 from

to

in

. Hence
either

influences

or vice versa, a contradiction to the
definition of

and

.
Although

may be an internal vertex of

, we claim that

,
that is, no slave can serve two masters. If this is not the case,
we would have a path of length 2 from

to

in

, a contradiction.
Hence we have constructed

distinct elements of

, as desired.
Corollary 4.7
Suppose
is an undirected connected graph.
If there exists a vertex
such
that the subgraph of
induced by the neighbours of
has
components,
then
.
Given these stringent necessary conditions, one may wonder whether there are
large classes of graphs that are realizable. The next proposition gives such
a class: a complete graph with the edges of a clique removed. (The graph
is
often described as the join of
and an independent set of
vertices, or as a complete multipartite graph with one part of size
and
parts of size 1.)
Proof.
Let

,

,
and

. Note that

or

may be empty.
Construct the graph

with vertex set

as follows: For all

, put two edges between

and

,
where one has time-stamp 1 and the other has time-stamp 3;
and put one edge between

and

with
time-stamp 2 for all

. Then it is easy to check that

.
Remark 4.9
We note that the graph constructed in Proposition 4.8 will not
be more complication even if we want
to be injective. Instead of
the general technique of computing transitive closure given in
Section 1, the construction in
the proof of Proposition 4.8 will work if
the time-stamps are all within
of their stated values.
Proposition 4.8 gives us a nice class of realizable graphs,
but these graphs are dense. So one might
think that if
, then
must be ``close'' to complete. This
is not the case, as we already have a class of digraphs
with
vertices and
arcs having
, namely, the
associated
influence digraphs of
that achieve
; see Theorem 2.1.
For an undirected example, let
be the graph whose vertices are
where
with undirected edges from
to
and
, as well as from
to
for all
(everything modulo
). Then
where
and
has
an edge with time-stamp approximately 2 between
and
,
an edge with time-stamp approximately 1 between
and
, and
an edge with time-stamp approximately 3 between
and
for all
(everything modulo
).
Next: Bounds for
Up: Time-Stamped Graphs and Their
Previous: Graphs
Eddie Cheng
2003-05-01