c 19XX Cambridge University Press

Combinatorics, Probability and Computing (19XX) 00, 111.

DOI: 10.1017/S0000000000000000 Printed in the United Kingdom

On the Independent Domination Number

of Random Regular Graphs

W. Duckworth and N.C. Wormald

Department of Computing
Macquarie University
Sydney, NSW 2109, Australia
and
Department of Combinatorics & Optimization
University of Waterloo
Waterloo ON, Canada N2L 3G1

Received 1 July 2003

A dominating set D of a graph G is a subset of V (G) such that for every vertex
v V (G), either in v D or there exists a vertex u D that is adjacent to v.
We are interested in finding dominating sets of small cardinality. A dominating set
I of a graph G is said to be independent if no two vertices of I are connected by
an edge of G. The size of a smallest independent dominating set of a graph G is
the independent domination number of G. In this paper we present upper bounds
on the independent domination number of random regular graphs. This is achieved
by analysing the performance of a randomised greedy algorithm on random regular
graphs using differential equations.

1. Introduction
Throughout this paper we consider simple graphs that are undirected, unweighted and
contain no loops or multiple edges. A graph G is said to be d-regular if every vertex
in V (G) has degree d (i.e. each vertex is adjacent to precisely d other vertices in G).
When discussing any graph G, we let n denote the cardinality of V (G) and for d-regular
graphs on n vertices, we assume dn to be even to avoid parity problems. For other basic
graph-theoretical definitions we refer the reader to, for example, Diestel [3].
A dominating set D of a graph G is a subset of the vertices of G such that for every
vertex v V (G), either in v D or there exists an edge uv E(G) with u D. We are
interested in finding dominating sets of small cardinality. The domination number of a
graph G, which we denote by (G), is the size of a smallest dominating set of G.

Research supported by the Australian Research Council while the author was affiliated with the
Department of Mathematics and Statistics, University of Melbourne. Currently supported by the
Canada Research Chairs program.
2 W. Duckworth and N.C. Wormald

For an arbitrary graph G the problem of determining (G) is one of the core NP-
hard optimisation problems in graph theory and this problem remains NP-hard even
for planar graphs of maximum degree 3 [5]. Johnson [7] showed that for general graphs
on n vertices, (G) is approximable within 1 + log n. Raz and Safra [11] showed that
(G) is not approximable within c log n for some c > 0. When restricted to graphs of
bounded degree d 3, Papadimitriou and Yannakakis [10] showed that the problem of
Pd+1
determining (G) is APX-complete and is approximable within 12 + i=1 i1 .
A dominating set I of a graph G is said to be independent if no two vertices of I are
connected by an edge of G. The independent domination number of a graph G, denoted
by i (G), is the size of a smallest independent dominating set of G. Halldorsson [6]
showed that for an arbitrary graph G, i (G) is not approximable within n1 for any
> 0. Note that for a d-regular graph G it is simple to verify that determining i (G)
is approximable within (d + 1)/2. For graphs with bounded degree d, Alimonti and
Calamoneri [1] improved upon this trivial bound when d is small and Kann [8] showed
that this problem is APX-complete for bounded degree graphs.
We consider random d-regular graphs that are generated uniformly at random (u.a.r.),
and need some associated notation. We say that a property B = Bn of a random graph
holds asymptotically almost surely (a.a.s.) if the probability that B holds tends to 1 as n
tends to infinity. When d-regular graphs are the objects of consideration, this is modified
so that n is restricted to even numbers if d is odd. For other basic random graph theory
definitions we refer the reader to Bollobas [2].
We [4] improved upon an earlier result of Molloy and Reed [9] by showing that for a
random cubic (i.e. 3-regular) graph G, i (G) a.a.s. satisfies 0.2641n i (G) 0.27942n.
The upper bound was achieved by using differential equations to analyse the performance
of a randomised greedy algorithm that is based on repeatedly choosing vertices of current
minimum degree and deleting edges. The lower bound was calculated by means of a direct
expectation argument.
Zito [16] presented upper and lower bounds on i (G) when G is a random d-regular
graph and gave explicit values for these bounds when 3 d 7. The lower bounds
were, again, calculated by means of a direct expectation argument whilst the upper
bounds were calculated by using differential equations to analyse the performance of a
randomised algorithm that is based on repeatedly choosing vertices of a particular degree
and deleting edges. Note that for d = 3, the upper bound in [16] is larger than the upper
bound result presented in [4].
In this paper we analyse the average-case performance of a simple heuristic, which
is a random greedy algorithm, that gives upper bounds on i (G) when G is a random
d-regular graph. This algorithm is an extension of that for d = 3 presented in [4] and
improves all upper bounds presented in [16]. Using the direct expectation argument
presented in [16], we also evaluate corresponding lower bounds on i (G) for several small
values of d.
In the following section we give a description of our algorithm and in Section 3 we
outline the method used for its analysis. Our analysis uses a theorem of the second
author [15] which we restate in Section 3. The results of this paper are encompassed by
the following theorem, the proof of which is given in Section 4.
 On the Independent Domination Number of Random Regular Graphs 3

Theorem 1.1. Let d 3 be fixed. Then for a random d-regular graph G on n vertices,
the size of a minimum independent dominating set is asymptotically almost surely less
than Cu (d)n, where the constant Cu (d) is given in Table 1.

The constant Cu (d) referred to in Theorem 1.1 arises from the solution of particular
sets of differential equations. In Table 1, corresponding lower bounds are also given by
evaluating constants C` (d) from the argument in [16] (the details of which are restated
in the final section).

Table 1 Bounds on i (G) when G is a random d-regular graph on n vertices.

d C` (d)n Cu (d)n

03 0.2641n 0.27942n
04 0.2236n 0.24399n
05 0.1959n 0.21852n
06 0.1755n 0.19895n
07 0.1596n 0.18329n
08 0.1468n 0.17037n
09 0.1362n 0.15948n
10 0.1273n 0.15015n
15 0.0976n 0.11783n
20 0.0803n 0.09830n
30 0.0606n 0.07526n
40 0.0494n 0.06181n
50 0.0420n 0.05285n

2. Prioritising choices
Consider the following algorithm that greedily finds a dominating set of a graph G.
Repeatedly choose a vertex u randomly to add to a set I. After each vertex is chosen,
remove u and its neighbours from G along with all their incident edges. Once no vertices
remain, the set I is a dominating set in G. It is not difficult to see that the set I is also
a maximal independent set and therefore an independent dominating set.
We modify this algorithm slightly by giving priority, in the choice of u, to the vertices
of current minimum degree in G, and add to I, not always u, but sometimes one of its
neighbours. To be precise, the new algorithm is the following. Repeatedly choose a vertex
u from those vertices of current minimum degree in G. If u has a neighbour of degree
strictly larger than that of u, select v from those vertices of current maximum degree
amongst the neighbours of u. Otherwise, select v to be u. Add v to I and remove v from
G along with its neighbours and all their incident edges.
The deletion of the edges incident with the neighbours of v may cause the current
degree of a vertex in G (that is not a vertex in I or adjacent to a vertex of I in G) to
be decreased to zero. We refer to such vertices as accidental isolates. Due to the priority
assigned to vertices of current minimum degree, accidental isolates that are created are
immediately selected in the subsequent steps to be part of I.
4 W. Duckworth and N.C. Wormald

Note that no two accidental isolates may be connected by an edge of G. This, and the
fact that all edges incident with neighbours of vertices of I in G are deleted, ensures that
I is an independent dominating set in G.
Define an operation to be the process of selecting a vertex, u, of current minimum
(non-zero) degree, the selection of a vertex, v, to add to I, the deletion of all edges
incident with neighbours of v and adding all accidental isolates that are created to I. In
particular, an operation of Type r is an operation in which u has degree r.
The algorithm above for finding a small independent dominating set of d-regular graphs
is a direct extension of the algorithm in [4] that finds a small independent dominating
set of cubic graphs. The algorithm in [4] is analysed as follows. Letting variables Yi
(i = 0, . . . , 3) denote the number of vertices of current degree i, the expected values
of Yi are estimated throughout the algorithm for each i using differential equations. It
is shown that with high probability the variables are concentrated near their expected
values. The analysis in [4] has major complications arising from the fact that priority is
given to vertices currently of minimum degree. We call such an algorithm a prioritised
algorithm.
We do not analyse exactly the algorithm presented above, but a similar one, using a
technique introduced by the second author [15]. This approach approximates the perfor-
mance of a prioritised algorithm by analysing associated deprioritised algorithms. These
algorithms entirely avoid prioritising by using a randomised mixture of operations. The
particular mixture used for any step is prescribed in advance but changes over the course
of the algorithm in order to approximate the prioritised algorithm.
One of the main objectives of using this new technique is to reduce the number of
conditions that are required to be checked. Arguments in [4] require steps involving
branching processes and large deviation inequalities. They also require checking complex
conditions regarding derivatives, especially at the transition between phases, which are
those points at which the smallest commonly occurring degree undergoes a sudden
shift.

3. Use of deprioritised algorithms

The operations and priorities described in the prioritised algorithm given in Section 2 can
be analysed using [15, Theorem 1]. This provides us with a set of differential equations
whose solution describes the state of a deprioritised version of the algorithm during its
execution. From this, we deduce asymptotically almost sure bounds on the size of the
independent dominating set at the end of the algorithm. (We claim, but do not prove,
that the deprioritised algorithm produces the same asymptotic result as the original
prioritised algorithm.)
The standard model for random d-regular graphs is as follows. Take a set of dn points
in n buckets labelled 1, 2, . . . , n, with d points in each bucket, and choose u.a.r. a pairing
P = p1 , . . . , pdn/2 of the points such that each pi is an unordered pair of points and each
point is in precisely one pair pi . The resulting probability space of pairings is denoted by
Pn,d . Form a d-regular pseudograph on n vertices by placing an edge between vertices i
and j for each pair in P having one point in bucket i and one in bucket j.
 On the Independent Domination Number of Random Regular Graphs 5

In order to prove that a property is a.a.s. true of a uniformly distributed random d-

regular (simple) graph, it is enough to prove that it is a.a.s. true of the pseudograph
corresponding to a random pairing (see Bollobas [2] and Wormald [13]).
As in [14], we redefine this model slightly by specifying that the pairs are chosen
sequentially. The first point in a random pair may be selected using any rule whatsoever,
as long as the second point in that random pair is chosen u.a.r. from all the remaining
free (unpaired) points. This preserves the uniform distribution of the final pairing.
When a pair has been determined in the sequential process, we say that it has been
exposed. By exposing pairs in the order which an algorithm requests their existence, the
generation of the random pairing may be combined with the algorithm (as in [4, 12, 14]).
In this way, the algorithm such as the one in the previous section, which deletes edges,
may be described in terms of operations incorporated into the pairing generation. The
definition of the operations may be extended to do whatever other tasks the algorithm
needs to carry out.
The algorithm proper acts upon the final (pseudo)graph of the generation process, but
the set of exposed pairs builds up this final graph during the course of the generation
process which incorporates the algorithm. The order in which the edges are deleted
corresponds to the order in which the pairs were exposed.
The setting of [15, Theorem 1] requires a number of definitions, and may be described
as follows. It concerns a class of processes applied to the random pairing. As described
above, this may be defined in terms of the generation algorithm which exposes pairs.
The beginning of the generation algorithm is the empty pairing G0 . The pairing Gt+1
is obtained from Gt by applying an operation which may expose some of the pairs; the
degree of a bucket is the number of points it contains in exposed pairs. The operation,
opt , which is applied to Gt must be one of some prespecified set of operations, Opi ,
i = 1, . . . , d, where Opi consists of selecting a bucket u of degree d i (vertex of degree
i) in Gt u.a.r., and then applying some specified set of tasks, resulting in Gt+1 . A subset
I of V (G) E(G) is selected during the operations, with I0 = initially, and I = It for
the pairing Gt .
For 1 i d, let Yi = Yi (t) denote the number of buckets of degree d i in Gt , and
let Yd+1 = Yd+1 (t) denote cardinality of the set It . Put Y(t) = Y1 (t), . . . , Yd+1 (t)).
We refer the reader to [15, Theorem 1] for motivation for the following definitions, and
provide a little explanation below. Let y denote (y1 (x), . . . , yd+1 (x)). Given functions
fi,r (x, y), define

k (x, y) = fdk1,dk (x, y) ,

(1)
k (x, y) = fdk1,dk1 (x, y) ,

where

t Y(t)
x= , y(x) = . (2)
n n
6 W. Duckworth and N.C. Wormald

We will consider the equations

dyi
= F (x, y, i, k) (3)
dx
where
k k
k +k fi,dk (x, y) + k +k fi,dk1 (x, y) k d2
F (x, y, i, k) = (4)

fi,1 (x, y) k =d1
and work with the parameters of fi,` in the domain

D = {(x, y) : 0 x d, 0 yi d for 1 i d + 1, yd } (5)

for some pre-chosen value of > 0. The behaviour of the process will be described in
terms of the function y = y(x) = (y1 (x), . . . , yd+1 (x)) defined as follows, with reference
to an initial value x = x0 = t0 /n of interest:

yi (x0 ) = Yi (t0 )/n, i = 1, . . . , d + 1, and inductively for k 1, y

is the solution of (3) with initial conditions y(xk1 ) = y(xk1 ),
extending to all x [xk1 , xk ], where xk is defined as the infimum
of those x > xk1 for which at least one of the following holds: (6)
k 0 and k < d 1; k + k and k < d 1; ydk 0; or the
solution is outside D or ceases to exist.

The interval [xk1 , xk ] is called phase k. This inductive definition of y continues for
phases k = 1, 2, . . . , m, where
m denotes the smallest k for which either k = d 1, or any of the
termination conditions for phase k in (6) hold at xk apart from
(7)
xk = inf{x xk1 : k 0}.

It turns out that the intervals called phases have nonempty interior provided
k > 0 and k + k > at (xk1 , y(xk1 )) (1 k min{d 2, m}), (8)

fd1,d1 > 0 at (x0 , y(x0 )),

0 0
fdk,dk k + fdk,dk1 fdk1,dk > 0 at (xk1 , y(xk1 ))+ 1 < k
k min{d 2, m}),

0
fdk,dk > 0 at (xk1 , y(xk1 )) 1 < k m,

0
f1,1 > 0 at (xd2 , y(xd2 ))+ if m = d 1,
(9)
with f 0 denoting df (x,y(x))
dx and (x, y(x)) +
and (x, y(x))
referring to the right-hand and
left-hand limits as functions of x.
 On the Independent Domination Number of Random Regular Graphs 7

We may now restate [15, Theorem 1] which we will use in the following section in
connection with the independent dominating set algorithm.

Theorem 3.1 ([15]). Let d 3. Assume that for some fixed > 0 the operations Opr
satisfy

E Yi (t + 1) Yi (t) | Gt {opt = Opr } = fi,r (t/n, Y1 /n, . . . , Yd+1 /n) + o(1) (10)

for some fixed functions fi,r (x, y1 (x), . . . , yd+1 (x)) and for i = 1, . . . , d + 1, r = 1, . . . , d,
with the convergence in o(1) uniform over all t and Gt for which Yr (t) > 0 and Yd (t) > n.
Assume furthermore that

(i) there is an upper bound, depending only upon d, on the number of pairs exposed, and
on the number of elements added to I (i.e. |It+1 | |It |), during any one operation;
(ii) the functions fi,r are rational functions of x, y1 , . . . , yd+1 with no pole in D defined
in (5);
(iii) there exist positive constants C1 , C2 and C3 such that for 1 i < d, everywhere on
D , fi,r C1 yi+1 C2 yi when r 6= i, and fi,r C3 yi+1 for all r.

Define y as in (6), set x0 = 0, define m as in (7), and assume that (8) and (9) both
hold. Then there is a randomised algorithm on Pn,d for which a.a.s. there exists t such
that |It | = nyd+1 (xm ) + o(n) and Yi (t) = nyi (xm ) + o(n) for 1 i d. Also yi (x) 0
for xk1 x xk , 1 i d k 1 (1 k m).

Some of these definitions can be explained easily. The algorithm in Section 2 works
by deleting edges; the edges deleted correspond to pairs exposed in the corresponding
pairing generation algorithm as described above. In particular, a vertex of degree i in
the original algorithm corresponds to a bucket of degree d i in the pairing version; we
use vertex degree and bucket degree to distinguish these complementary measures. The
algorithm gives higher priority to the buckets of highest degree (vertices of lowest degree).
The phase is determined by the set of bucket degrees which are reasonably common
(meaning, roughly, more than cn buckets have that degree for some c > 0). Phase k
corresponds to a period in which the smallest such common vertex degree is d k (i.e.
largest common bucket degree is k). At such a time, vertices of degree d k 1, when
created, will immediately be used up, by being chosen for u in the subsequent steps,
until the minimum positive vertex degree returns to d k. So phase k basically consists
of a mixture of two operations: Opdk and Opdk1 . The functions and represent
respectively the expected net increase in Yk+1 in an Opdk , and the expected net decrease
in Yk+1 in an Opdk1 . From these quantities, one may estimate the proportions of these
operations being performed at any stage. The randomised algorithm referred to in the
theorem uses roughly the same mixture of operations. This in turn lets us calculate the
expected changes in the variables, and the result is (4), which leads to the differential
equation (3) analogous to the equations derived in [4].
8 W. Duckworth and N.C. Wormald

4. Algorithm analysis
Motivated by the algorithm for finding a small independent dominating set as described
in Section 2, we specify the tasks in Opr , 1 r d. Here Opr must first select a random
bucket, u, of degree d r. The set of randomised tasks consists of choosing a bucket
v, exposing pairs in v and all neighbouring buckets and adding v and any accidental
isolates to I (which denotes the independent dominating set). An accidental isolate,
when dealing with the pairing generation, refers to any bucket (other than v and its
neighbours) that attains degree d. According to the rule given in the description of the
algorithm, v is chosen randomly from the buckets neighbouring u of strictly smaller
degree (recalling the relationship between bucket degree and vertex degree), if there are
any, and otherwise v = u.
We may verify the hypotheses of Theorem 3.1. First we will show that (10) holds when
Yd (t) > n (for any > 0). From here onwards in this description, v-degree refers to
vertex degree, so v-degree i means bucket degree d i. Let yi denote Yi /n and
q,r = (Sq1 )r (Sq1
1 )r
where
b
X d
X
jyj
Sba = Pj , Pj = and s= iyi .
j=a
s i=1
Then, when performing an instance of Opr , the probability that the neighbours of u have
maximum v-degree exactly q is q,r + o(1) (see [4] for similar arguments). Similarly, for
Opr , the probability that u has b neighbours of v-degree q, given that the maximum
v-degree amongst the neighbours of u is q, is b,q,r /q,r + o(1) where

b r
b,q,r = (Pq ) (Sq1
1 )rb .
b
Also, the expected number of neighbours of u that have v-degree j, 1 j q 1, given
that u has b neighbours of v-degree q and q is the maximum v-degree of all neighbours
of u, is b,j,q,r /b,q,r + o(1) where

b r
b,j,q,r = (Pq ) (Sq1
1 )rb1 (r b)Pj .
b
Let q denote the maximum v-degree of the neighbours of u. If q > r, Opr selects a
bucket, v, u.a.r. from the neighbours of u of v-degree q to add to the dominating set.
This is Case A. Otherwise (all neighbours of u have v-degree at most r), Opr selects u
to add to the dominating set. This is Case B. For both cases, once the dominating set
bucket has been chosen, all pairs incident with the chosen bucket and its neighbours are
exposed.
The pairs incident with u are always exposed and the effect, on the expected change
in Yi , of changing the v-degree of u to 0 is just
i=r
where, here and in the following, R = 1 if the statement R is true, 0 otherwise.
 On the Independent Domination Number of Random Regular Graphs 9

For Case A, the effect, on the expected change in Yi , of changing the v-degrees of v
and its neighbours (other than u) by exposing all pairs incident with the neighbours of
v, is just
i=q + (q 1)i + o(1)
where
Pd
i = Pi + i x=2 (x 1)Px and

i = Pi + Pi+1 i+1d .
For each of the other neighbours of u, the effect, on the expected change in Yi , of changing
its v-degree from j to j 1 is just
i=j1 i=j .
For Case B, the expected number of neighbours of u of degree j r is
Pj
r(Sr1 )r + o(1)
Sr1
and the effect, on the expected change in Yi , of the changing v-degree of one such neigh-
bour w, and all its neighbours (other than u), by exposing all pairs incident with w, is
just
i=j + (j 1)i + o(1).
So we have that (10) holds with
d
X
fi,r = i=r + q,r (i=q + (q 1)i )
q=r+1

d
X r
X q1
X
+ b,q,r (b 1)(i=q1 i=q ) + b,j,q,r (i=j1 i=j ) (11)
q=r+1 b=1 j=1

r
X
+ r(Sr1 )r1 Pj ((j 1)i i=j ).
j=1

It also follows that (10) also holds for i = d + 1 with fd+1,r defined as 1 + f0,r , since in
each Opr , an extra vertex is added to the independent dominating set I and the expected
number of accidental isolates is f0,r as defined in (11).
Hypothesis (i) of Theorem 3.1 is immediate since in any operation only the pairs
involving points in one bucket and its neighbours are exposed, and a bounded number
of vertices are added to I (as there are certainly less than d2 accidental isolates). The
functions fi,r satisfy (ii) because from (11) their (possible) singularities satisfy s = 0,
which lies outside D since in D , s yd . Hypothesis (iii) follows from (11) again
using s yd and the boundedness of the functions yi (which follows from the
boundedness of D ). Thus, defining y as in (6) with t0 = 0, Yd (0) = n and Yi (0) = 0 for
i 6= d, we may solve (3) numerically to find m, verifying (8) and (9) at the appropriate
points of the computation.
10 W. Duckworth and N.C. Wormald

It turns out that these hold for each d in Table 1, and that in each case m = d 1, for
sufficiently small > 0. For such , the value of yd+1 (xm ) may be computed numerically
(the result is shown as the constant Cu (d) in Table 1), and then by Theorem 3.1, this is
the asymptotic value of the size of the independent dominating set I at the end of some
randomised algorithm. So the conclusion is that a random d-regular graph a.a.s. has an
independent dominating set of size at most nyd+1 (xm ) + o(n). Note also that (by the
theorem) yi (x) 0 in phase k for 1 i d k 1, and by the nature of the differential
equation, yi (x) will be strictly positive for i > d k. So by (6) and (7), the end of the
process (for arbitrarily small) occurs in phase d 1 when y1 becomes 0.

5. Lower bounds
Zito [16] considered the expected number of independent dominating sets of size n for a
random d-regular graph on n vertices, d 3. He showed that this expectation is at most
( d(12)
)n
O(1) (1 2) 2 (d)d ((1 + x)d 1)1
n
(1 )1 dd xd

where x takes the value that minimises

((1 + x)d 1)1
.
xd
Finding the value of below which this expression tends to zero gives a lower bound on
the expected size of an independent dominating set of a random d-regular graph on n
vertices. Thus giving the constants C` (d) reported in table 1.

References
[1] P. Alimonti and T. Calamoneri. Improved approximations of independent dominating set
in bounded degree graphs. Lecture Notes in Computer Science, 1197:216, 1997. Springer-
Verlag.
[2] B. Bollobas. Random graphs. Academic Press Inc., Harcourt Brace Jovanovich Publishers,
London, 1985.
[3] R. Diestel. Graph theory. Springer-Verlag, New York, 2000.
[4] W. Duckworth and N.C. Wormald. Minimum independent dominating sets of random cubic
graphs. Random Structures and Algorithms, 21(2):147161, 2002.
[5] M.R. Garey and D.S. Johnson. Computers and intractability. W.H. Freeman and Co., San
Francisco, California, 1979.
[6] M.M. Halldorsson. Approximating the minimum maximal independence number. Informa-
tion Processing Letters, 46(4):169172, 1993.
[7] D.S. Johnson. Approximation algorithms for combinatorial problems. Journal of Computer
and System Sciences, 9:256278, 1994.
[8] V. Kann. On the approximability of NP-complete optimisation problems. Doctoral thesis,
Department of Numerical Analysis, Royal Institute of Technology, Stockholm, 1992.
[9] M. Molloy and B. Reed. The dominating number of a random cubic graph. Random Struc-
tures & Algorithms, 7(3):209221, 1995.
[10] C.H. Papadimitriou and M. Yannakakis. Optimization, approximation and complexity
classes. Journal of Computer and System Sciences, 43(3):425440, 1991.
 On the Independent Domination Number of Random Regular Graphs 11

[11] R. Raz and S. Safra. A sub-constant error-probability low-degree test and a sub-constant
error-probability PCP characterization of NP. Proceedings of the 29th Annual ACM Sym-
posium on the Theory of Computing, pages 475484, 1997.
[12] N.C. Wormald. Differential equations for random processes and random graphs. The Annals
of Applied Probability, 5(4):12171235, 1995.
[13] N.C. Wormald. Models of random regular graphs. In Surveys in combinatorics, 1999 (Can-
terbury), pages 239298. Cambridge University Press, Cambridge, 1999.
[14] N.C. Wormald. The differential equation method for random graph processes and greedy
algorithms. In M. Karonski and H. Promel, editors, Lectures on Approximation and Ran-
domized Algorithms, pages 73155. PWN, Warsaw.
[15] N.C. Wormald. Analysing greedy algorithms on regular graphs. In M.N.J. Nesetril and
O. Serra, editors, Electronic Notes in Discrete Mathematics, volume 10. Elsevier Science
Publishers, 2001.
[16] M. Zito. Greedy algorithms for minimisation problems in random regular graphs. Lecture
Notes in Computer Science, 2161:524536, Springer-Verlag, 2001.