NORTH- HOLLAND

Changing and Unchanging of the Radius of a Graph

Ronald D. Dutton and Sirisha R. Medidi

Department of Computer Science
University of Central Florida
Orlando, Florida 32816

and

Robert C. Brigham
Departments of Mathematics and Computer Science
University of Central Florida
Orlando, Florida 32816

Dedicated to Professor John Maybee on the occasion of his 65th birthday.

Submitted by J. Richard Lundgren

ABSTRACT

Harary introduced the concept of changing and unchanging of a graphical

invariant i, asking for characterizations of graphs G = (V, E) for which i(G - v),
i(G - e) or i(G + e) either differ from i(G) or are equal to i(G) for all ‘u E V,
e E E, or e E ??, respectively, where z is the set of edges in the complement of
G. This paper examines changing and unchanging properties of the radius of a
graph, an important parameter in network design.

1. Introduction

Harary [lo] introduced the concept of changing and unchanging of a

graphical invariant i, with interest in determining those graphs for which
i(G - v) = i(G) and i(G - v) # i(G) for all vertices v of G, i(G - e) =
i(G) and i(G - e) # i(G) for all edges e of G, and i(G + e) = i(G) and
i(G+e) # i(G) for all edges e of c, the complement of G. This concept has

LINEAR ALGEBRA AND ITS APPLICATIONS 217:67-82 (1995)

@ Elsevier Science Inc., 1995 0024-3795/95/%9.50

655 Avenue of the Americas, New York, NY 10010 SSDI 0024-3795(94)00153-5
 R. D. DUTTON, S. R. MEDIDI AND R. C. BRIGHAM

been investigated for several invariants [l-4, 7, 12-191. Here we consider

the invariant radius, indicated by r(G). Often, when the context is clear,
the argument G associated with an invariant will not be used.
Let dG(x, y) be the distance between vertices x and y in a graph G.
Again, when there is no danger of confusion, the subscript G will be
dropped. The eccentricity of vertex x is defined by e(x) = maxvEv d(x, v),
the radius by T = minzEv e(x), and the diameter by d = maxzEv e(x). The
center of G, denoted C, is the set of vertices having eccentricity r. We will
use n to represent IVI, the number of vertices of G.
In networks, the radius of the graph model represents a measure of the
shortest possible time required to broadcast a message from a single vertex,
namely one in C, to all other vertices. Changing and unchanging studies
then reflect how this parameter can vary as a result of equipment failure
(vertex or edge removal) or network enhancement (edge addition).
We will consider the effects which can occur by removing a single ver-
tex or edge and by adding a single edge, as well as finding those graphs
responding in a fixed way for all vertices or edges. For the latter, it is
helpful to define the following sets:

V+ E {G : r(G - v) > r(G) Vu E G},

V= E {G : r(G - v) = r(G) ‘dv E G},

V- z {G : r(G -v) < r(G) Vu E G},

R+ E {G : r(G - e) > r(G) t/e E G},

R= E {G : r(G - e) = r(G) Ve E G},

A= E {G:r(G+e)=r(G)VeEG},

A- c {G : r(G + e) < r(G) Ve E c},

In naming these sets, R is intended to be a mnemonic aid for removal of an

edge and A for addition of an edge. Note that one could define R- and A+
in analogous ways, but such sets are empty, since the radius cannot decrease
when an edge is removed or increase when an edge is added. Throughout,
we assume the original graph G is connected and thus has finite radius.
However, removal of a vertex or edge may disconnect the graph, which we
interpret as an increase in the radius to infinity.
Within a single graph G the radius can increase, remain the same,
or decrease when a vertex is removed; increase or remain the same when
an edge is removed; and remain the same or decrease when an edge is
added. For example, in Figure 1, r(G) = 2, r(G - x) = 00, r(G - y) = 2,
r(G - z) = 1, r(G - vx) = 00, r(G - wz) = 2, r(G + wy) = 2, and
RADIUS OF A GRAPH 69

w
W

x
1.

Y
V

FIG. 1.

T(G + cm) = 1.
The following useful fact is well known and easy to establish.

PROPOSITION 1. For any graph G, T = d if and only if C = V.

Graphs satisfying the conditions of Proposition 1 are called self-centered,

and Buckley [5] provides a survey of associated results.
Sections 2, 3, and 4 deal with vertex removal, edge removal, and edge
addition, respectively. In Section 5 we give a Venn diagram showing how
the sets defined above interrelate.
2. Vertex Removal

2.1. T(G - v) > r(G)

It is shown in Buckley and Harary [6, p. 981 that V+ = 0. We draw
that conclusion by approaching the problem in a manner which sets the
stage for much of the work in this paper. A k-depth spanning tree (k-DST)
of a graph G is a spanning tree of height k. It must be true that k > r,
and if k = r, such trees must be rooted at a center vertex. A breadth first
search algorithm beginning with any vertex in C will always produce an
T-DST. Such breadth first search spanning trees usually are not unique,
even when rooted at the same vertex, nor does every T-DST of a graph
occur as a breadth first spanning tree. Define L to be the set of vertices of
G which appear as leaves in at least one T-DST of G. Then I = V - L is
the set of vertices which are interior vertices of every T-DST.

THEOREM 2. LetG=(V,E) and v E V. Then r(G - w) > r(G) if and

only if v E I and there is no (r + l)-DST which has only v at level r + 1.

Proof Assume r(G - TJ) > r(G). If w E L or such a (r + l)-DST

exists, then there is a k-DST of G - 21with k I r(G). Next assume that
2, E I and there is no (r + l)-DST having only w at level T + 1. Suppose
T(G - w) 5 r(G). Then there is a k-DST, k 5 T, of G - v. Either 2, can
be added as a leaf at a level at most T, violating 21 E 1, or adding 21 as
70 R. D. DUTTON, S. R. MEDIDI AND R. C. BRIGHAM

a leaf extends the tree to level T + 1 with only v on level T + 1, again a

contradiction. W

Note that the above proof implies that v E L whenever r(G---‘u) < r(G),
since v E I means r (G - v) 2 r(G). The fact that V+ = 0 follows
immediately from Theorem 2, since every tree has at least two leaves. It
further follows that 1115 n - 2 with equality if and only if G is the path
P, on n 2 2 vertices.
Cycles have the property that ICI = n and 111= 0, while trees have
ICI 5 2 and 1112 n - 2. These and other observations lead to the following:

CONJECTURE. Let G be any graph. Then ICI + 1115 n.

The remainder of this section is devoted to results related to the con-

jecture.

OBSERVATION 3. The following are equivalent:

(i) ICI + 14 5 n,
(ii) ICI 5 ILI, and
(iii) IC r711 5 IL - Cl.

Proof The equivalence of (i) and (ii) follows immediately from the
definition of L and I. The equivalence of (ii) and (iii) is a consequence of
C=(CnL)u(CnI) and L= (CnL)u(L-C). ??

Let S be any subset of V, and let As be the largest degree of any vertex
in S.

THEOREM 4. For any graph G, (I( 5 n - A,.

Proof. An r-DST rooted at a center vertex of degree AC has AC sub-

trees rooted at its children, so IL/ > A C, and since I = V - L, we obtain
1115 n-A,. ??

Theorem 4 trivially implies )I) < n - 6, where 6 is the minimum degree

of the graph. Let A = A” be the maximum degree.

THEOREM 5. For any graph, A 5 max(lL(,n - JCJ- 1).

Proof. Let v be a vertex with degree A > n- ICI. Then either v E C or

v has a neighbor z E C. If v E C, then, by Theorem 4, IL1 > A. Otherwise,
RADIUS OF A GRAPH 71

consider a breadth first search spanning tree rooted at 2. Observe that (1)
every child of ZI is a root of a subtree containing at least one leaf, (2) every
horizontal edge between v and a neighbor of x corresponds to another
subtree rooted at the common neighbor, which contains yet another leaf,
and (3) since v $! C, there is a neighbor of x at distance two from v which
is the root of a subtree with at least one more leaf. Thus, (LI 2 A. ??

Let x(G) be the chromatic number of the graph G.

THEOREM 6. For any connected graph, at least two of the following

hold:

(1) x I m=(lLI,n - (LI - l),

(2) X I max(lCj,n - ICI - l), and
(3) ICI I I-4.

ProoJ All three properties hold for odd cycles and complete graphs,
so by Brook’s theorem [9, p. 1281 we may assume x 5 A. Observe that we
need only show that the falseness of (1) implies (2) and (3) and then the
falseness of (2) implies (3). Suppose first that x > max(lLI,n - IL\ - 1).
Then A > IL1 and, by Theorem 5, A < n - ICI - 1 so x < n - ICI - 1.
Furthermore, since n - IL\ - 1 < x 5 n - ICI - 1, it follows that ICI < ILI.
Thus, (2) and (3) must be true. Next assume x > max(lCI,n - ICI - 1).
Then A > n - ICI - 1 and, with Theorem 5, that implies x 5 A 5 ILI.
Since JC( < x, we conclude that ICI < ILI. W

THEOREM 7. For any connected graph, at least two of the following

hold:

(1) ICI I ma4lGn - IL1- I),

(2) x I max(lCI,n - ICI - l), and
(3) x 5 I-4.

Proof As in the proof to Theorem 6, we may assume x <_ A. Suppose

ICI > max((LI, n - JLI - 1). Then, by Theorem 6, x 5 max(lLI, n - ILJ - 1)
and x I max(lCI, n - ICI - 1). Suppose x > ILI. Then A > ILJ and, by
Theorem5, x 2 A 5 n-(Cl-1 < n-(n-ILI--1)--l = ILI, acontradiction.
Next, assume x > max(lCI, n - ICI - 1). Then A > n - ICI - 1 implies, by
Theorem 5, that x 5 A 5 ILI. W
72 R. D. DUTTON, S. R. MEDIDI AND R. C. BRIGHAM

2.2. r(G - v) < r(G)

F’rom the comment following Theorem 2 we know, for r(G - v) to be
less than r(G), that w must be in L. In fact, for some r-DST, v must be the
only leaf node on level T. This and the related results listed below follow
directly from Gliviak [8] as reported in Buckley and Harary [6, pp. 98-991.
Define, for 0 I i 5 d, Ni(z) = {z : d~(z,x) = i}.

THEOREM 8. Let v be an arbitrary vertex of a graph G. Then r(G -

w) < r(G) if and only ifN,.(z) = {v} for some z E C(G).

COROLLARY 9. Ifr(G - w) < r(G),then r(G - w) = r(G) - 1.

COROLLARY 10. Ifr(G - v) < r(G), then C(G - v) 5 C(G).

COROLLARY 11. IfG=(V,E)cV-, thenC(G)=V.

The antipodal graph of G = (V, E) has vertex set V with two vertices
adjacent if and only if their distance is equal to the diameter d.

THEOREM 12. A graph is in V- if and only if its antipodal graph is

(n/2)Kz.

2.3. r(G - v) = r(G)

From Theorems 2 and 8, it is clear that r(G - w) = r(G) if and only
if v E L and any r-DST with TJon level r has at least one other vertex on
level r, or v E I and G has a (r + l)-DST with w being the only vertex on
level r + 1. When this holds for every vertex, we obtain a characterization
of graphs in V=.

THEOREM 13. A graph G = (V, E) is in V= if and only if N,.(z) 2 2

for each z E C(G) and, for each v E I, there is an (r + l)-DST having only
v onlevelr+l.

3. Edge Removal

3.1. r(G - e) > r(G)

Analogous to the situation of Section 2.1, defining IE to be the set of
edges appearing in every r-DST leads immediately to the following theorem.

THEOREM 14. Let G be a graph with edge e. Then r(G - e) > r(G) if
and only if e E IE.
RADIUS OF A GRAPH 73

It follows that any graph G has at most n - 1 edges whose removal in-
creases the radius. This leads immediately to the following characterization
of R+ and is attributed to Gliviak [8] in Buckley and Harary [6, p. 971.

THEOREM 15. A graph G is in R+ if and only if G is a nontrivial tree.

The following conjecture is similar to the one given in Section 2.1.

CONJECTURE. For any graph G, ]C] + ]IE] I n + 1 with equality if

and only if G is a tree with two central vertices, i.e., ICI = 2.

3.2. r(G - e) = r(G)

It follows from Theorem 14 that r(G - e) = r(G) if and only if e $ 1~.
Thus we have a characterization of R=.

THEOREM 16. A graph G is in R= if and only if IE = 0.

An important special case arises when r = d. It will be convenient

to consider a breadth first search spanning tree “rooted at an edge e.”
Such trees are formed similarly to a normal breadth first search spanning
tree rooted at a vertex, except that two adjacent vertices a and b are
initially selected and placed on level 0 of the “tree,” along with their joining
edge. Thereafter, the expansion places on level i those vertices which are
at distance i from one of a and b and at distance i or i + 1 from the other.
Any such search partitions the vertices of G into sets A, S, and B according
to whether they are closer to a than to b, the same distance from a and b,
or closer to b than to a, respectively. We consider a E A and b E B. Note
that S # 0 if and only if e is in an induced odd cycle. The following facts
are easily verified.

OBSERVATION17. For a graph G, let T be any breadth first search tree

rooted at edge e = ab . Then

(i) if t E A (respectively B) an d as on level i of T, then there is no edge

xt with x on level i - 1 and x E B U S (respectively A U S), and
(ii) if t E C(G) and r(G - e) > r(G), then dG(t,a) # dG(t,b).

THEOREM 18. Every graph G with at least three vertices and having
r =d is in R=.

Proof. Since r = d, every vertex is in C(G). Harary and Norman [l l]

74 R. D. DUTTON, S. R. MEDIDI AND R. C. BRIGHAM

showed that all center vertices lie in a single block of G, so we may assume,
since n 2 3, that G has no bridges. Suppose for some edge e = ab that
r(G - e) > r(G), and let T be a breadth first search tree rooted at e.
Employing the notation described above and Observation 17 (ii), we have
5’ = 0. Let TA be the subtree rooted at a, and TB the one rooted at
b. Observation 17(i) shows that all graph edges between TA and TB are
horizontal. Suppose x is a vertex in TA with a horizontal edge to TB. Then
vertex x can reach b without using e in the same number of steps as it would
take using e. Thus, the removal of e does not increase the distance between
x and any vertex, a contradiction, since x is a center vertex. It follows that
no such edges exist and that e is a bridge, again a contradiction, and the
theorem is proven. ??

Theorem 18 is useful in showing that, with the exception of Kz, V- is

a subset of R=. This will be used in Section 5.

THEOREM 19. V- - {Kz} 2 R=.

Proof Let G E V- . If G # K2 and has a bridge, it has a cut vertex,

implying G 6 V- . Thus, G E V- - {Kz} implies G is bridgeless. Since
G’s antipodal graph is (n/2)Kz, we have r = d.
The result now follows from Theorem 18. ??

4. EDGE ADDITION

We have been unable to formulate satisfactory characterizations of ei-

ther A= or A-. Medidi [17] includes characterizations based on distance
requirements forced by the definition of radius, but structural results ap-
pear difficult. This section discusses what is known.
It is possible to give a characterization of graphs in A- which have
radius 2. We consider a star Kl,, to be nontrivial when m > 0.

THEOREM 20. A graph G of radius 2 is in A- if and only if G is the

complement of the disjoint union of at least two nontrivial stars.

Proof Let G be a graph whose complement ?? is a disjoint union of

at least two nontrivial stars. Any leaf vertex of 5 has degree JV1- 2 in G
and is at distance two from the single vertex to which it is nonadjacent.
It follows that adding any edge to G reduces the radius from 2 to 1, and
thus G E A-. Now assume r(G) = 2 and G E A-. Then the set of vertices
with degree at most IV1 - 3 must form a complete subgraph. Otherwise,
it would be possible to place an edge between two such vertices and not
RADIUS OF A GRAPH 75

0
t
I

t 2
.
.
.
r-3
+
r-2

r- 1

r,

r+l
E3 .
.
.
(c- 1)’

f,
I

(r+ 1)

E‘IG.
2.

reduce the radius. Thus this set is independent in ??. All other vertices
must have degree IV1 - 2 in G, since r > 1, and thus are of degree 1 in ??.
It follows that ?? is a disjoint union of nontrivial stars, and there must be
at least two of them, since r(G) = 2. ??

The situation becomes more difficult when r > 3. We first show there
are graphs in A- with r < d, contrary to our early suspicions.

THEOREM 21. IfG E A-, then d 5 2r - 2, and this bound is sharp.

Proof Let z and y be vertices such that d(z, y) = d. Recall that

N,(z) = {t: dG(z,t) = i} for 0 5 i 5 d. Let z E C(G + zy). There is
a k such that z E Ark(x). Since G E A-, r(G + sy) I r - 1. We may
assume the sets J/i(x) are cyclically arranged with the new edge sy joining
No(x) to N&). Th en z can reach, in at most r - 1 steps, only vertices
in Nk-+I), Nk+-i)+i,. . . , %++I) where subscript arithmetic is taken
modulo d + 1. Thus the number of N,(z) sets d + 1 is at most 2r - 1, that
76 R. D. DUTTON, S. R. MEDIDI AND R. C. BRIGHAM

is, d 5 2r - 2.
z

m x

FIG. 3.
Y

The graph of Figure 2 shows sharpness. It is easily checked that it has

radius r, has diameter 2r - 2, and is in A-. The vertices are labeled in a
way which emphasizes the diameter. ??

We suspect, but have been unable to show, there are graphs in A- that
have diameter i, for every i satisfying r < i 5 2r - 2 and for any r 2 2.
The statement “G E A= if and only if every r-DST of G is in A=”
is false. Although necessity clearly holds, sufficiency is disproven by the
graph G, one of an infinite family, in Figure 3. For G,r = 3 but adding
edge xy reduces the radius to 2, implying G $ A=. Yet the only r-DST is
G - XZ, which clearly is in A=.

5. INTERSECTION RELATIONSHIPS BETWEEN V=, V-, R+, R=,

A=,A-

This section considers when two or three of the sets V=, V-, R+, R=,
A=, and A- have nonempty intersections. Certainly any intersection of
four of the sets is empty. The set V+ is not discussed, since it also is
empty. There would seem to be desirable network modeling features for
graphs in V=, V- , R=, and A-, so graphs in two or more of these sets
should be particularly interesting. We have already seen, in Theorem 19,
that V- - {Kz} c R=. It is straightforward to show that R+ fl V= =
R+ n A- = 0 and R+ f~ V- = {Kz}. Observe that empty graphs satisfy
the definition for membership in both R+ and R=, as do complete graphs
for A= and A-. This ambiguity is resolved by defining empty graphs to be
in R= and not R+, and complete graphs to be in A= and not A-. With
this understanding, it is trivial that V= n V- = R+ n R= = A= n A- = 0.
Figure 4 is a Venn diagram showing the possible nonempty intersections,
and reflecting the comments of the previous paragraph. The alphabetic
labels represent the nonempty families of graphs described below. One
section, marked by a “?,” remains unsettled and provides an open question:
RADIUS OF A GRAPH 77

D///I// ////,//,,,,,, n

? A?=andA-
? V?=

ES-
? R?=
?? R+

FIG.4.

Is V=nA- G R=?
a: Define A,, p 2 2, to be a cycle C’s, with a pendant vertex z adja-
cent to vertex x on the cycle. Then A, $ V= U V- , since x is a cut
vertex, and A, $ R=, since e = x.z is a bridge, nor is it in R+, since it is
not a tree (hereafter, we will not repeat this obvious argument concerning
membership in R+). Finally, AP $ A=, since adding xy, for any y $ N(x),
reduces T, and A, $ A-, since adding ZV, w E N(x) - {z}, does not change
the radius.
b: Define BP, p 2 4, to be a cycle Cz, with a chord between two
vertices which are distance two apart. Label the vertices sequentially by
~0,~1,...,~2p-l, and let the chord be e = ‘~1212~_i. Then r(Bp) = p - 1,
and it is easy to see that BP E V=. Removing e increases the radius, so
BP $ R=. Adding up-lv,+i leaves the radius unchanged, so B, 4 A-, and
adding 2rswpreduces the radius from p - 1 to [p/2], showing BP +! A=.
c: A preliminary lemma will be useful. Let H be a graph having vertex
x, and G be obtained from H by replacing x with adjacent vertices xi
and x2, each joined to all original neighbors of x. We say G is obtained
from H by splitting x. Note that if G is obtained from H by successively
splitting every vertex of H, then G is simply the composition H[K2] (see
Harary [9, p. 221).

LEMMA 22. Let G be obtained from H by splitting x. Then r(H) =

r(G - x1) = r(G - x2) = r(G).

Proof Clearly H, G - x1, and G - 22 are isomorphic. If r(G - xi) <

78 R. D. DUTTON, S. R. MEDIDI AND R. C. BRIGHAM

@yp K4 K4 K4 K4
\ /

FIG.5.

r(G) and t E C(G-xl), then in G the only vertex which can be at distance
T from t is zi, a contradiction, since x2 also must be at distance r from
t. On the other hand, if r(G - xi) > r(G), then zr is an internal vertex
of every T-DST of G. But then so is x2. This is a contradiction, since an
T-DST can be formed in which 21 (or ~2) is a leaf. W

COROLLARY 23. Let G = H[K2]. Then G E V=.

Define a graph CL = (V, E),p > 2, with V = {ai, bi, Q, di : 1 5 i 5 p}

and E = {aiaj,bibj,cicj,didj : i # _i} U {aibj,cidj : i # j} U {biq,diai :
1 5 i 5 p}. Then let C, = CL[Kz]. In particular, C2 is the graph of
Figure 5. The graph C, is in V= by Corollary 23. Since the radius of C,
is 3, remains 3 when any edge is removed, and drops to 2 when any edge
is added, it follows that C, E V= n R= n A-.
d: Define the graph DP, p 2 2, by first creating a cycle C2, with vertices
labeled cyclically by ~0, wi, . . . , w~~_I and then adding a vertex x adjacent
to vzP- 1, VO,and wi. The radius of D, is p, every vertex is in the center, and
adding any edge reduces the radius. Thus D, E A-. Removing z leaves
the radius unchanged, so D, 6 V- and removing up reduces the radius, so
D, $ V=.
e: Even cycles and hypercubes are in V- n A- n R=.
f: Let FP, p > 3, be the graph discussed in Theorem 21 to illustrate
membership in A-. Then FP has diameter d = 2p - 2 with r = p. Since
removing vertex 1 or edge 01 increases the radius, FP 6 V- U V= U R=.
g: Odd cycles on at least seven vertices are in V= n R= but not
A= u A-.
h: Consider K2,*, p 2 3, and let the smaller bipartite set be {WI, ~2).
RADIUS OF A GRAPH 79

FIG. 6.

Removing any edge leaves the radius 2, so Kz,+, E R=. Removing vi or v2

or adding edge 211212 reduces the radius, while removing any other vertex
or adding any other edge does not change the radius. Hence KQ, $ V= u
V-uA=uA-.
i: Define the graph IP, p 2 3, by IP = CL - albz, where CL is described
under family bc, above. The graph 13 is shown in Figure 6.
The radius of IP is 3 and remains 3 when any edge is removed, so
IP E R=. Its antipodal graph is 2pKz, in which the edges join ai to ci and
bi to di, so IP E V-. Adding edge a albz leaves the radius unchanged,
while adding albl reduces it to 2. Hence IP $ A= u A-.
j: DefinethegraphJ,=(V,E),p>2,byV={z,y}U{zi,yi:11i<
p} and E = {my} U {ziyi, ZQ, yiy : 1 5 i 5 p}. It is easily checked that the
radius of JP is 2 and remains 2 if any vertex is removed or edge added, so
JP E V= n A=. Removing edge xy increases the radius to 3, so JP $ R=.
k: Km,+,,3 5 m 5 p, has radius 2 and is in A= n V= n R=.
1: Let L,,p > 3, be a wheel with 2p vertices on its cycle and two
opposing spokes removed. The radius of L, is 2 and remains 2 when any
edge is removed or added, implying L, E A= U R=. It also remains 2 when
a degree 2 vertex is removed, but increases to p when the center vertex is
removed, showing L, 4 V= U V- .
m: K2 is clearly in A= n R+, and we have seen that it is the only
member of V- not in R=.
o: Let the graph O,,p 2 3, be formed from 2K, by adding an edge
e = xy joining the complete graphs. Clearly 0, has radius 2, and adding
any edge leaves the radius at 2, so 0, E A=. Since x is a cut vertex and e
is a bridge, 0, $ V= U V- U R=.
p and q: We have seen in Theorem 15 that R+ is the family of nontrivial
trees. We now characterize the trees in R+ - A=. Those remaining, the set
R+ n A=, include the nontrivial stars. We call a tree central if its center
has one element. Otherwise it is bicentml. Let us define T to be the set
of trees T satisfying either (i) T is central with center {z} and contains a
path 2120~~that is a subpath of every path having length at least 2r - 1,
80 R. D. DUTTON, S. R. MEDIDI AND R. C. BRIGHAM

FIG. 7

FIG.8.

or (ii) T is bicentral with center {z, y} and contains a path zuzy that is a
subpath of every path having length 2r - 1. We now show that the defined
set T is actually R+ - A=.

THEOREM 24. R+ -A== T.

Sketch ofproof. We will outline the proof for central trees only, leaving
certain details to the reader. A proof for bicentral trees is similar and
will be omitted entirely. Any central tree T @ T must have the form
shown in Figure 7. A triangular figure represents a subtree rooted at the
indicated vertex, e.g., T,, and the enclosed number represents the depth
of the subtree, as measured from its root. Since T(T + uy) < T, we have
T q,!A=. Therefore, T E R+ - A=.
Now suppose T is a central tree in R+ - A=. If T E T, then either T,
must have a leaf at level r - 1 or greater or T, must have a leaf at level
r - 2 or greater, or both. Therefore T has one of the forms shown in Figure
8. Let e = ab be an edge such that r(T + e) < r(T), and let z E C(T + e).
Such an edge e exists because T $ A=. It is straightforward to show that
this implies at least one of a and b, say b, is in TV. Now consider form (1)
in Figure 8, and assume a E T,. Then by examining the possible locations
for z, that is, in T,, T,, T,, or T,, we find in every case that z is distance
at least r from some vertex, a contradiction. Repeating the argument for
a E T,, Tz, and Tg (with b’s level assumed to be at least as large as a’s)
implies form (1) cannot occur. Applying the same technique to form (2)
gives rise to similar contradictions unless z E {u, w}, in which case b # y,
since otherwise there is a leaf of TY not reachable in r - 1 steps from either
u or w. We claim y has a child s which is an ancestor of b and all leaves
of TY on levels r - 2 and T - 1. If not, the form is as shown in Figure 9.
RADIUS OF A GRAPH 81

FIG. 9.

FIG. 10.

No matter where a is located, neither u nor w can reach all leaves of T,

in fewer than r steps, a contradiction which shows the correct form is that
given by Figure 10. In this form, all paths of length at least 2r - 1 include
the subpath syxw, so T E T.
A similar argument for bicentral trees completes the proof.

The authors wish to express their appreciation to the referees for many
help&l suggestions.

REFERENCES

1 D. Bauer, F. Harary, J. Nieminen, and C. L. Suffel, Domination alternation

sets in graphs, Discrete Math. 47:153-161 (1983).
2 R. C. Brigham, P. Z. Chinn, and R. D. Dutton, Vertex domination critical
graphs, Networks 18:173-179 (1988).
3 R. C. Brigham and R. D. Dutton, An extremal problem for edge domination
insensitive graphs, Discrete Appl. Math. 20:113-125 (1988).
4 R. C. Brigham and R. D. Dutton, Changing and unchanging invariants: The
edge clique cover number, Congr. Namer. 70:145-152 (1990).
5 F. Buckley, Self-centered graphs, in Graph Theory and Its Applications: East
and West, Proceedings of the First China-USA International Graph Theory
Conference, Ann. New York Acad. Sci. 576, 1989, pp. 71-78.
6 F. Buckley and F. Harary, Distance in Graphs, Addison-Wesley, Redwood
City, Calif., 1990.
 R. D. DUTTON, S. R. MEDIDI AND R. C. BRIGHAM

7 J. Carrington, F. Harary, and T. W. Haynes, Changing and unchanging of

the domination number of a graph, J. Combin. Math. Combin. Comput.
9:57-63 (1991).
8 F. Gliviak, On radially critical graphs, in Recent Advances in Graph Theory,
Proc. Sympos. Prague, Academia Praha, Prague, 1995, pp. 207-221.
9 F. Harary, Graph Theory, Addison-Wesley, Reading, Mass. 1969.
10 F. Harary, Changing and unchanging invariants for graphs, Bull. Malaysian
Math. Sot. 5:73-78 (1982).
11 F. Harary and R. Z. Norman, The dissimilarity characteristic of Husimi trees,
Ann. Math. 58: 134-141 (1953).
12 T. W. Haynes, On /c-y-Insensitive domination, Ph.D. Dissertation, Univ. of
Central Florida, Orlando, 1988.
13 T. W. Haynes, R. C. Brigham, and R. D. Dutton, Extremal 2-a-insensitive
graphs, Congr. Numer. 67:158-166 (1988).
14 T. W. Haynes, L. M. Lawson, R. C. Brigham, and R. D. Dutton, Changing
and unchanging of the graphical invariants: Minimum and maximum degree,
maximum clique size, node independence number and edge independence
number, Congr. Numer. 72~239-252 (1990).
15 L. M. Lawson and T. W. Haynes, Changing and unchanging of the node
covering number of a graph, Congr. Numer. 77:157-162 (1991).
16 L. M. Lawson and T. W. Haynes, The effects of graph modifications on
edge independence and edge covering numbers, J. Combin. Math. Combin.
Comput. to appear.
17 S. Medidi, The Effect of Node Deletions, Edge Additions, and Edge Deletions
on Graphs: Node Connectivity, Edge Connectivity, Diameter, and Radius,
Master’s Project, Dept. of Computer Science, Univ. of Central Florida, 1992.
18 D. P. Summer and P. Blitz, Domination critical graphs, J. Combin. Theory
Ser. B 34:65-76 (1983).
19 H. B. Walikar and B. D. Acharya, Domination critical graphs, Nat. Acad.
Sci. Lett. 1:7&72 (1979).

Received 20 July 1993; final manuscript accepted 12 June 1994