Journal of Algebra 296 (2006) 462–479

www.elsevier.com/locate/jalgebra

Zero-divisor graphs of non-commutative rings

S. Akbari ∗ , A. Mohammadian
Institute for Studies in Theoretical Physics and Mathematics, PO Box 19395-5746, Tehran, Iran
Department of Mathematical Sciences, Sharif University of Technology, PO Box 11365-9415, Tehran, Iran
Received 9 October 2004
Available online 9 January 2006
Communicated by Paul Roberts

Abstract
In a manner analogous to the commutative case, the zero-divisor graph of a non-commutative ring
R can be defined as the directed graph Γ (R) that its vertices are all non-zero zero-divisors of R in
which for any two distinct vertices x and y, x → y is an edge if and only if xy = 0. We investigate
the interplay between the ring-theoretic properties of R and the graph-theoretic properties of Γ (R).
In this paper it is shown that, with finitely many exceptions, if R is a ring and S is a finite semisimple
ring which is not a field and Γ (R)  Γ (S), then R  S. For any finite field F and each integer n
2,
we prove that if R is a ring and Γ (R)  Γ (Mn (F )), then R  Mn (F ). Redmond defined the simple
undirected graph Γ (R) obtaining by deleting all directions on the edges in Γ (R). We classify all
ring R whose Γ (R) is a complete graph, a bipartite graph or a tree.
 2005 Published by Elsevier Inc.

Keywords: Zero-divisor; Non-commutative ring; Directed graph; Matrix ring

1. Introduction

The concept of zero-divisor graph of a commutative ring with identity was introduced
by Beck in [6] and has been studied in [1–5]. Recently Redmond in [9] has extended this
concept to any arbitrary ring. Throughout the paper, R denotes a ring (not necessarily with

* Corresponding author.
E-mail addresses: [email protected] (S. Akbari), [email protected] (A. Mohammadian).

0021-8693/$ – see front matter  2005 Published by Elsevier Inc.

doi:10.1016/j.jalgebra.2005.07.007
 S. Akbari, A. Mohammadian / Journal of Algebra 296 (2006) 462–479 463

identity) and D(R) denotes the set of all zero-divisors. If X is either an element or a subset
of R, then the left annihilator of X is the left ideal Ann (X) = {a ∈ R | aX = 0} and the
right annihilator of X, denoted by Annr (X), is similarly defined. For any subset Y of R,
we define Y ∗ = Y \{0}. The zero-divisor graph of R, denoted by Γ (R), is a directed graph
with the vertex set D(R)∗ in which for any two vertices x and y, x → y is an edge if and
only if x = y and xy = 0. Note that if x and y are two distinct vertices and xy = yx = 0,
then there are two directed edges between x and y and we say there is a multiple edge
between x and y. We show that if Γ (R) has more than one vertex and it has no multiple
edges, then R is isomorphic to one of two specified non-commutative rings of order four.
Also for a ring R, we define a simple undirected graph Γ (R) with the vertex set D(R)∗ in
which two vertices x and y are adjacent if and only if x = y and either xy = 0 or yx = 0.
Note that for a commutative ring R, the definition of the zero-divisor graph of R in [4]
coincides with the definition of Γ (R).
In [9] it has been shown that for any ring R, every two vertices in Γ (R) are connected
by a path of length at most 3. Note that using the proof of this result in commutative case,
one can establish that for any arbitrary ring R, if there exists a path between two vertices x
and y in the directed graph Γ (R), then the length of the shortest path between x and y is
at most 3. Moreover, in [9] it is shown that for any ring R, if Γ (R) contains a cycle, then
the length of the shortest cycle in Γ (R), is at most 4.
For every directed graph Γ and any vertex a of Γ , the number of the edges of Γ of
the form x → a is called the in-degree of a and the number of the edges of Γ of the form
a → y is called the out-degree of a. A subset Ω of the vertices of a directed graph Γ is
called a clique if x → y and y → x are two edges of Γ , for each pair of distinct vertices x
and y in Ω. Two directed graphs Γ1 and Γ2 are said to be isomorphic if there is a bijective
map ϕ between the vertex set of Γ1 and the vertex set of Γ2 such that for any two vertices
x and y of Γ1 , x → y is an edge in Γ1 if and only if ϕ(x) → ϕ(y) is an edge in Γ2 .
For an undirected graph G, the degree of a vertex v in G is the number of edges incident
to v. A graph G is complete if there is an edge between every pair of the vertices. A subset
X of the vertices of a graph G is called independent if there is no edge with two endpoints
in X. A graph G is called bipartite if its vertex set can be partitioned into X and Y such that
every edge of G has one endpoint in X and other endpoint in Y . A graph G is said to be
star if G contains one vertex in which all other vertices are joined to this vertex and G has
no other edges. Two graphs G1 and G2 are said to be isomorphic if there is a bijective map
ψ between the vertex set of G1 and the vertex set of G2 such that the adjacency relation is
preserved.
For any field F , we denote Mn (F ) the ring of all n × n matrices over F . Also for any
i and j , 1  i, j  n, we denote by Eij the element of Mn (F ) whose (i, j )-entry is 1
and other entries are 0. Moreover, a ring R is called semisimple if its Jacobson radical is
zero. By the Wedderburn–Artin Theorem, we know that every semisimple Artinian ring
is isomorphic to the direct product of finitely many full matrix rings over division rings.
If e is an idempotent element in R (e2 = e), by Peirce decomposition [8, p. 308], we can
write R = eRe ⊕ eR(1 − e) ⊕ (1 − e)Re ⊕ (1 − e)R(1 − e) (if R has no identity, then
eR(1 − e) denotes the subring {ex − exe | x ∈ R}, and similar definitions are used for the
sets (1 − e)Re and (1 − e)R(1 − e)). An idempotent element e is said to be non-trivial if
464 S. Akbari, A. Mohammadian / Journal of Algebra 296 (2006) 462–479

e = 0, 1. A ring R is called reduced if it has no non-zero nilpotent elements. Also a ring R

is called a null ring if R 2 = {0}. We denote the finite field of order q by Fq .
In this article we prove that for any finite field F and each integer n
2, if R is a
ring and Γ (R)  Γ (Mn (F )), then R  Mn (F ). This shows that the zero-divisor graph of
Mn (F ) determines the ring up to isomorphism. Using later results, we obtain that if S is
a finite semisimple ring which is not a field and Γ (R)  Γ (S), then apart from finitely
many exceptions, we have R  S. In relation to the graph Γ (R), we prove that this graph
is complete if and only if either R  Z2 × Z2 or D(R)2 = {0}. Also we establish that there
are just two finite non-commutative rings R for which Γ (R) is bipartite. We show that if
Γ (R) has no cycle, then either Γ (R) is a star graph or R is isomorphic to one of the rings
Z2 × Z4 or Z2 × Z2 [x]/(x 2 ). Furthermore we prove that for any left Artinian ring R, if
Γ (R) has at least four vertices, then Γ (R) is a star graph if and only if R is isomorphic to
the direct product of a division ring and a ring of order 2.

2. Some properties of the zero-divisor graphs of non-commutative rings

In this section we characterize rings with respect to their zero-divisor graphs. We start
with some properties of the zero-divisors of a ring. The following remark is useful in our
proofs.

Remark 1. For a ring R, we note that if D(R) is finite, then R is either a finite ring or
a domain. Moreover, if R is not a domain, then |R|  |D(R)|2 . To see this, suppose that
D(R) is finite and non-zero. Let x ∈ D(R)∗ . Since Ann (x) and Rx are contained in D(R),
these sets are finite. Since [R : Ann (x)] = |Rx|, we conclude that |R| = |Ann (x)||Rx| 
|D(R)|2 .

Lemma 2. Let R be a left Artinian ring and R = D(R). Then R has identity and any
element of R\D(R) is unit.

Proof. Let a ∈ R\D(R). So the map φ : R → R, defined by φ(x) = xa, is a left R-module
monomorphism. Since R is a left Artinian ring and φ is injective, it is well known that φ
is surjective. Thus there is an element e ∈ R such that ea = a. Now since a ∈ / D(R) and
(xe − x)a = 0, we conclude that xe = x, for each x ∈ R. Also we have a(ex − x) = 0.
This yields that ex = x, for any x ∈ R. Hence 1 = e is the identity of R. Moreover, since
φ is surjective, there is an element a  ∈ R such that a  a = 1. Again we have a ∈
/ D(R) and
(aa  − 1)a = 0, so aa  = 1. Thus a is a unit, as desired. 2

Theorem 3. Let R be a ring such that |D(R)| = p is a prime number. If R is not a reduced
ring, then either R is isomorphic to Zp2 , Zp [x]/(x 2 ) or the null ring of order p.

Proof. Since D(R) is finite, by Remark 1, R is also finite. If R has no identity, then
by Lemma 2 we have R = D(R). Thus the additive group of R is cyclic. Clearly, this
implies that R is the null ring of order p. So assume that R has identity. Let J (R) be the
Jacobson radical of R. If J (R) = {0}, then by the Wedderburn–Artin Theorem, we may
 S. Akbari, A. Mohammadian / Journal of Algebra 296 (2006) 462–479 465

write R = Mn1 (Fq1 ) × · · · × Mnr (Fqr ), where qi ’s are prime powers and n1 , . . . , nr , r are
natural numbers. Since R is not reduced, at least one of ni ’s, say n1 , is greater than 1. By
Lemma 2, all non-invertible elements of R are zero-divisors, so we find that

 r
n2i 
r
 ni  n   n n −1 

D(R)
= qi − qi − 1 qi i − qi · · · qi i − qi i .
i=1 i=1

This implies that

 
n (n −1)/2 n (n −1)/2

r
n (n +1)/2

r
 ni  ni −1 
q1 1 1 · · · qr r r qi i i − q i − 1 qi − 1 · · · (qi − 1) = p.
i=1 i=1

Now since p is a prime and n1 > 1, we conclude that q1 = p. But the number given in the
above parenthesis is greater than 1, a contradiction. Hence J (R) = {0}. It is well known
R
that any unit of the ring J (R) is of the form u + J (R), where u is a unit of R. Since R is
R
finite, this implies that any zero-divisor of J (R) is of the form z + J (R), where z is a zero-
divisor of R. On the other hand, for any z ∈ D(R) and j ∈ J (R), we have (z + j ) + J (R) =
z + J (R) and hence z + j ∈ D(R). So z + J (R) ⊆ D(R). Thus we obtain that |D(R)| =
|D( J (R)
R
)||J (R)|. It follows from |D(R)| = p and J (R) = {0} that |D( J (R) R
)| = 1 and
|J (R)| = p. So J (R) is a domain and hence its characteristic is a prime number. Since
R

J (R) is a nilpotent ideal of order p, we conclude that J (R)2 = {0}. Therefore |R| is a
R
power of char J (R) . Since |R| is divisible by p, we find that char J (R)
R
= p. By Remark 1,
we have |R|  p 2 . Furthermore, | J (R)
R
|
p and so |R| = p 2 . If char R = p 2 , then R  Zp2 .
Suppose that char R = p and a is an element of J (R)∗ . Then R = {λ + µa | λ, µ ∈ Zp }.
Now since a 2 = 0, it is easily checked that R  Zp [x]/(x 2 ) and the proof is complete. 2

Corollary 4. Let R be a ring. If Γ (R) has one vertex, then either R is isomorphic to one
of the rings Z4 , Z2 [x]/(x 2 ) or the null ring of order 2.

Proof. By Remark 1, R is finite. Since every finite reduced ring is isomorphic to the direct
product of finitely many fields, any finite reduced ring is either a field or has at least three
zero-divisors. Hence R is not reduced and by Theorem 3 the assertion is proved. 2

Theorem 5. Let R be a ring. Then Γ (R) is a complete graph if and only if either R 
Z2 × Z2 or D(R)2 = {0}. Moreover in the latter case D(R) is an ideal of R.

Proof. For one direction, the proof is straightforward. For the other direction assume that
Γ (R) is a complete graph. First suppose that there exists a zero-divisor e with e2 = 0. We
show that e2 = e. If not, then since the vertices e and e2 are adjacent, we have e3 = ee2 = 0.
Hence e2 (e − e2 ) = 0 and so e − e2 is a non-zero zero-divisor different from e. Now since
Γ (R) is a complete graph, we conclude that e2 = e(e − e2 ) = 0, a contradiction. So e
is an idempotent element and hence we can write R = eRe ⊕ eR(1 − e) ⊕ (1 − e)Re ⊕
(1 − e)R(1 − e). We show that eR(1 − e) = {0}. Let x be an element of eR(1 − e)∗ .
Since x(e + x) = 0, e + x is a vertex of Γ (R). But e and e + x are not adjacent, which is
466 S. Akbari, A. Mohammadian / Journal of Algebra 296 (2006) 462–479

a contradiction. Similarly (1 − e)Re = {0}. Since none of the vertices of eRe∗ is adjacent
to e, we have eRe = {0, e}. Now we show that (1 − e)R(1 − e) is a domain. Suppose
yy  = 0, for some y, y  ∈ (1 − e)R(1 − e)∗ . Since (e + y)y  = 0, e + y is a vertex of Γ (R).
But e and e + y are not adjacent, a contradiction. So (1 − e)R(1 − e)∗ is an independent
set. This yields that (1 − e)R(1 − e)  Z2 . Hence in this case R  Z2 × Z2 .
Next suppose that z2 = 0, for any z ∈ D(R). Clearly this implies that D(R) is an ideal
of R. Assume that there exist two zero-divisors a and b such that ab = 0. Since Γ (R)
is a complete graph, ba = 0. Hence b ∈ Ann (a + b)\Annr (a + b). Now since D(R) =
Ann (a + b) ∪ Annr (a + b) is an additive group, D(R) = Ann (a + b). Thus a(a + b) = 0
and so ab = 0, a contradiction. Therefore D(R)2 = {0} and the proof is complete. 2

The following theorem has been established in [5] for every commutative ring with
identity. Now using Theorem 5 we generalize it for any arbitrary ring.

Theorem 6. Let R be a ring which is not isomorphic to Z2 × Z2 . If one of the graphs Γ (R),
Γ (R[x]), and Γ (R[[x]]) is a complete graph, then two other graphs are also complete
graphs.

Proof. Since Γ (R) is an induced subgraph of Γ (R[x]) and Γ (R[x]) is an induced sub-
graph of Γ (R[[x]]), it is enough to prove that if Γ (R) is a complete graph, then Γ (R[[x]])
is also a complete graph. Suppose that Γ (R) is complete. Since R  Z2 × Z2 , Theo-
rem 5 implies that D(R) is an ideal of R with D(R)2 = {0}. To complete  the proof, one
needs only verify that D(R[[x]]) ⊆ D(R)[[x]]. Assume that f (x) = ∞ n=0 fn x∞
n is an el-

ement of D(R[[x]])\D(R)[[x]]. Then there exists a non-zero element g(x) = n=0 gn x n

R[[x]]
such that f (x)g(x) = 0. Since D(R)R
is a domain, then D(R)[[x]]  D(R)
R
[[x]] is also a do-
main. Since f (x)g(x) = 0 and f (x) ∈ / D(R)[[x]], we conclude that g(x) ∈ D(R)[[x]].
Assume that r and s are the smallest indices such that fr ∈ / D(R) and gs = 0. Since
f (x)g(x) = 0 and
 g(x) ∈ D(R)[[x]] and D(R) 2 = {0}, we have f g(x) = 0, for each i < r.
i
Thus (f (x) − r−1 n=0 fn x )g(x) = 0 and therefore fr gs = 0. It follows from gs = 0 that
n

fr ∈ D(R), a contradiction. 2

Note that in [5] it has been shown that if R = Z2 × Z2 , then Γ (R[x]) is not complete,
while Γ (R) is a complete graph with two vertices. It has been also proved that between
each pair of the vertices of Γ (R[[x]]) there is a path of the length at most 2.

Example 7. Let R = a0 b0 | a, b ∈ Z2 . Then Γ (R) is a directed graph with three vertices
and the edges E12 → E11 and E12 → E11 + E12 . Note that Γ (R) is the smallest zero-
divisor graph associated to a non-commutative ring.

Theorem 8. Let R be a ring such that Γ (R) has no multiple edges. If Γ (R) has more than
one vertex, then R  R or R  Rop .

Proof. If R is a reduced ring, then for any x, y ∈ R, xy = 0 if and only if yx = 0. Thus

in this case if there is an edge between two vertices x and y, then there is a multiple
 S. Akbari, A. Mohammadian / Journal of Algebra 296 (2006) 462–479 467

edge between x and y. So R is not reduced. Suppose that a is a non-zero nilpotent el-
ement of R and n is the smallest natural number such that a n = 0. If n
3, then there
is a multiple edge between a and a n−1 , a contradiction. Hence a 2 = 0. Without loss of
generality, suppose that a → e is an edge in Γ (R) (if there is no such edge, then we
consider R op instead of R). Since ea = 0 and a(ea) = (ea)a = 0, we have ea = a. Thus
(e − e2 )a = a(e − e2 ) = 0. Now since ea = ae, e2 = e. So e is a non-trivial idempo-
tent and hence we can write R = eRe ⊕ eR(1 − e) ⊕ (1 − e)Re ⊕ (1 − e)R(1 − e). Since
(1−e)R(1−e) ⊆ Ann (e)∩Annr (e), we conclude that (1−e)R(1−e) = {0}. Assume that
x ∈ eR(1 − e) and y ∈ (1 − e)Re are non-zero elements. We have yx ∈ (1 − e)R(1 − e), so
yx = 0 and therefore xy = 0. Since x(xy) = (xy)x = 0 and y(xy) = (xy)y = 0 and x = y,
we get a contradiction. Now since e = 1, exactly one of the sets eR(1 − e) and (1 − e)Re
is non-zero. Suppose eR(1 − e) = {0} and (1 − e)Re = {0}. Since eR(1 − e) is a null ring,
we obtain that |eR(1 − e)| = 2. Let eR(1 − e) = {0, b}. Since b(eRe) = {0} and Γ (R)
has no multiple edges, eRe ∩ Ann (b) = {0}. Using R = eRe ⊕ eR(1 − e), we conclude
that Ann (b) = eR(1 − e). Also since eb = b and Rb ⊆ Ann (b), we obtain Rb = {0, b}.
Now the equation [R : Ann (b)] = |Rb| implies that |R| = 4. Thus R = {0, e, b, e + b}.
We have e2 = e, b2 = 0, eb = b, and be = 0. Now it is easy to see that R  R. Simi-
larly, if eR(1 − e) = {0} and (1 − e)Re = {0}, then we find that R  Rop and the proof is
complete. 2

In the next theorem, we determine all left Artinian rings R for which Γ (R) is a star
graph with at least four vertices.

Theorem 9. Let R be a left Artinian ring and Γ (R) has at least four vertices. Then Γ (R)
is a star graph if and only if R is isomorphic to the direct product of a division ring and a
ring of order 2.

Proof. For one direction, the proof is clear. For the other direction let a be that vertex
of Γ (R) which is adjacent to all other vertices. First suppose that a 2 = 0. Since a 2 is
also adjacent to all other vertices of Γ (R), a 2 = a and therefore we can write R = aRa ⊕
aR(1−a)⊕(1−a)Ra ⊕(1−a)R(1−a). If (1−a)R(1−a) = {0}, then for any elements x
and y of the set aR(1−a)∗ ∪(1−a)Ra ∗ we have either xy = 0 or yx = 0. Since a = 1, this
set cannot be empty and hence it has just one element. Without loss of generality, assume
that aR(1 − a) = {0, b} and (1 − a)Ra = {0}. We have b(a − b) = 0. So a − b is a vertex
of Γ (R) which is not adjacent to a, a contradiction. Thus (1 − a)R(1 − a) = {0}. Since all
vertices of the set (1 − a)R(1 − a)∗ are adjacent to all vertices of aRa ∗ ∪ aR(1 − a)∗ ∪
(1 − a)Ra ∗ and noting that Γ (R) is a star graph, we conclude that aRa = {0, a} and
aR(1 − a) = (1 − a)Ra = {0}. Therefore aRa and (1 − a)R(1 − a) are two ideals of R.
Furthermore, because of Γ (R) is a star graph, (1 − a)R(1 − a)∗ is an independent set.
Suppose that (1 − a)R(1 − a) is not a domain. Since Γ ((1 − a)R(1 − a)) is a connected
graph [9, Theorem 3.2] with no edge, the ring (1 − a)R(1 − a) has exactly one non-zero
zero-divisor, say c. Clearly c2 = 0 and so (a + c) − c is an edge of Γ (R), a contradiction.
Hence (1 − a)R(1 − a) is a domain. Now since R = {0, a} ⊕ (1 − a)R(1 − a), the ring
(1 − a)R(1 − a) is a left Artinian domain and so it is a division ring. Moreover, a 2 = a and
therefore in this case R is isomorphic to the direct product of a division ring and Z2 .
468 S. Akbari, A. Mohammadian / Journal of Algebra 296 (2006) 462–479

Next suppose that a 2 = 0. Assume that R has a non-trivial idempotent, say e. We

have R = eRe ⊕ eR(1 − e) ⊕ (1 − e)Re ⊕ (1 − e)R(1 − e). Since all vertices of the
set eR(1 − e)∗ ∪ (1 − e)Re∗ ∪ (1 − e)R(1 − e)∗ are adjacent to e, this set is a subset of {a}.
Moreover e = 1, so eR(1 − e) ⊕ (1 − e)Re ⊕ (1 − e)R(1 − e) = {0, a} and thus we have
R = eRe ⊕ {0, a}. (∗)
Also, since Γ (R) is a star graph, eRe∗ is an independent set. If we repeat the above proof as
we did for (1 − a)R(1 − a) before, we find that eRe must be a domain. Now if eR(1 − e) =
{0, a}, then Ra ⊆ Ann ({e, a}). This implies that Ra = {0, a}. Moreover in this case R =
eRe ⊕ eR(1 − e). Thus aR = {0}. This shows that {0, a} is an ideal of R. So by (∗),
R
we have {0,a}  eRe and thus eRe is a left Artinian domain and so it is a division ring.
Assume that za = 0, for some z ∈ eRe. If z = z−1 in the division ring eRe, then a =
ea = (z z)a = 0, a contradiction. By (∗), we conclude that Ann (a) = {0, a}. Since [R :
Ann (a)] = |Ra| = 2, we have |R| = 4. This is a contradiction, since Γ (R) has at least
four vertices. By a similar argument, (1 − e)Re = {0} and therefore (1 − e)R(1 − e) =
{0, a}. Since {0, a} is an ideal of R, by (∗), eRe is a division ring and so in this case R is
isomorphic to the direct product of a division ring and the null ring of order 2.
Now suppose that R has no non-trivial idempotent. Let x ∈ D(R) \ {0, a}. Since R is
a left Artinian ring with no non-trivial idempotent, by [8, Corollary 19.19], R is a left
Artinian local ring. So there exists the smallest natural number n such that x n = 0. If
n
4, then since x—x n−1 and x 2 —(x n−2 − a) are two edges in Γ (R), we easily get a
contradiction. If x 2 = 0, then x and a − x are adjacent in Γ (R), so we again obtain a
contradiction. Thus n = 3 and x 3 = 0. Hence x 2 = a, for any x ∈ D(R)\{0, a}. Therefore
x(xy) = ay = y 3 = 0, for all zero-divisors x and y. So we conclude that xy = a, for any
x, y ∈ D(R)\{0, a}. Now let b, c, d be distinct vertices of degree 1 in Γ (R). We have
bc = bd = cb = cd = a and so b(c − d) = c(b − d) = 0. Since b2 = 0 and c2 = 0, we
conclude that c − d = b − d = a, a contradiction. This completes the proof. 2

Let R be a ring of order 2, p be a prime, and n be a natural number. Then Γ (R × Fpn ) is

a star graph with p n vertices, if R has identity; and it is a star graph with 2p n − 1 vertices,
if R has no identity. So by Corollary 4 and Theorem 9 we get the following.

Corollary 10. A finite star graph can be realized as Γ (R) if and only if it has p n or 2p n − 1
vertices, for some prime p and integer n
0.

Example 11. Theorem 9 is not valid for an arbitrary ring. To see this let Z x, y be the
polynomial ring with non-commuting variables and I = (x 2 , 2x, xy, yx − x). Assume that
R = Z x, y /I. Then it is easy to see that Γ (R) is a star graph and x + I is that vertex
which is adjacent to all other vertices of the graph. We note that R has no non-trivial
idempotent and so R is not isomorphic to the ring given in Theorem 9.

Theorem 12. Let R be a left Artinian ring which is not a division ring and Γ (R) has no
cycle of length 3. If Γ (R) is not a star graph, then R is isomorphic to either the direct
product of two division rings or the direct product of a division ring and one of the rings
Z4 or Z2 [x]/(x 2 ).
 S. Akbari, A. Mohammadian / Journal of Algebra 296 (2006) 462–479 469

Proof. If R is a local ring, then D(R) is equal to the Jacobson radical of R, and hence it is
a nilpotent ideal. This implies that there is a vertex adjacent to all other vertices of Γ (R).
Hence either Γ (R) has a cycle of length 3 or Γ (R) is a star graph. Thus assume that R is
not a local ring. By [8, Corollary 19.19], R has a non-trivial idempotent, say e. We have

R = eRe ⊕ eR(1 − e) ⊕ (1 − e)Re ⊕ (1 − e)R(1 − e). (∗)

If (1 − e)R(1 − e) = {0}, then ((1 − e)Re)(eR(1 − e)) = {0} and so every three ver-
tices of {e} ∪ eR(1 − e)∗ ∪ (1 − e)Re∗ are mutually adjacent. This implies that the set
eR(1 − e)∗ ∪ (1 − e)Re∗ has at most one vertex. Since e = 1, by (∗) we conclude that this
set is not empty. So without loss of generality, let eR(1 − e) = {0, x} and (1 − e)Re = {0}.
Since R = eRe ⊕ eR(1 − e), we have xR = {0}. This yields that x is adjacent to all other
vertices of Γ (R). Hence again either Γ (R) has a cycle of length 3 or Γ (R) is a star graph.
Thus we have (1 − e)R(1 − e) = {0}. Since the vertices of {e}, (1 − e)R(1 − e)∗ and
eR(1 − e)∗ ∪ (1 − e)Re∗ are mutually adjacent, we have eR(1 − e) = (1 − e)Re = {0}.
Therefore by (∗), we can write R  R1 × R2 , where R1 = eRe and R2 = (1 − e)R(1 − e).
If |D(R1 )|
3, then by [9, Theorem 3.2], Γ (R1 ) is a connected graph with at least two
vertices. So there are two distinct elements u, v ∈ D(R1 )∗ such that uv = 0. Now if w is
an arbitrary element of R2 , then the vertices (u, 0), (v, 0) and (0, w) of Γ (R) are mutu-
ally adjacent, a contradiction. This implies that |D(R1 )|  2 and |D(R2 )|  2. If there are
two elements a ∈ D(R1 )∗ and b ∈ D(R2 )∗ , then a 2 = b2 = 0. So the vertices (a, 0), (0, b)
and (a, b) are mutually adjacent, a contradiction. Thus at least one of the rings R1 and R2
is domain. If both R1 and R2 are domains, then R is the direct product of two division
rings. Otherwise we may assume that |D(R1 )| = 1 and |D(R2 )| = 2. Now R1 is a division
ring and by Corollary 4 and Theorem 9, R2 is isomorphic to either Z4 or Z2 [x]/(x 2 ), as
desired. 2

Theorem 13. Let R be a ring which is not a domain and Γ (R) has no cycle. If Γ (R) is
not a star graph, then either R  Z2 × Z4 or R  Z2 × Z2 [x]/(x 2 ).

Proof. Suppose that Γ (R) is not a star graph. So there exists a path a—x—y—b in Γ (R)
of length 3. Since two vertices a and y are not adjacent, we have x 2 ∈/ {a, y}. If x 2 ∈
/ {0, x},
then a—x—y—x —a is a cycle in Γ (R), a contradiction. Hence x ∈ {0, x} and similarly
2 2

y 2 ∈ {0, y}. If x 2 = y 2 = 0, then x—y—(x + y)—x is a cycle in Γ (R), a contradiction.

Thus either x 2 = x or y 2 = y. Hence R has a non-trivial idempotent. Now using the second
paragraph of the proof of Theorem 12, R is isomorphic to either the direct product of two
domains or the direct product of a domain and one of the rings Z4 or Z2 [x]/(x 2 ). Since
Γ (R) has no cycle and is not a star graph, the assertion clearly holds. 2

3. The zero-divisor graphs of matrix rings

We begin this section with the following lemma.

Lemma 14. Let F be a finite field and n
2. Suppose a ∈ Mn (F ) is a non-zero matrix and
rank a = k < n. Then the in-degree and the out-degree of a in Γ (Mn (F )) are |F |n(n−k) − ε,
470 S. Akbari, A. Mohammadian / Journal of Algebra 296 (2006) 462–479

2
and the degree of a in Γ (Mn (F )) is equal to 2|F |n(n−k) − |F |(n−k) − ε, where ε = 1,
unless a 2 = 0 and in this case ε = 2.

Proof. It is well known that there exist two invertible matrices u and v in Mn (F ) such that
uav = ki=1 Eii . We have, (uav)x = 0 (respectively, x(uav) = 0), for some x ∈ Mn (F ), if
and only if the first k rows (respectively, columns) of x are zero. Hence dimF Ann (a) =
dimF Ann (uav) = n(n − k) and dimF Annr (a) = dimF Annr (uav) = n(n − k). Thus
the in-degree and the out-degree of a in Γ (Mn (F )) are |F |n(n−k) − ε, where ε = 1, unless
a 2 = 0 and in this case ε = 2. Moreover, clearly

dimF Ann (a) ∩ Annr (a) = dimF Ann (uav) ∩ Annr (uav) = (n − k)2 .
2
So |Ann (a) ∪ Annr (a)| = |F |n(n−k) + |F |n(n−k) − |F |(n−k) . Therefore the degree of a in
2
Γ (Mn (F )) equals 2|F |n(n−k) − |F |(n−k) − ε, where ε = 1, unless a 2 = 0 and in this case
ε = 2. 2

Next we show that for a finite semisimple ring which is not a field, the ring R is deter-
mined by the graph Γ (R).

Remark 15. Now we want to calculate the maximum degree and the minimum degree of
Γ (Mn (F )), for any finite field F and any n
2. By Lemma 14, the rank of any vertex with
the maximum degree in Γ (Mn (F )) is 1. Since clearly there is a matrix with rank 1 whose
2
square is not 0, the maximum degree of Γ (Mn (F )) is 2|F |n(n−1) − |F |(n−1) − 1. Now let a
be a vertex with the minimum degree in Γ (Mn (F )). By Lemma 14, we have rank a = n−1.
On the other hand we know that the rank of any matrix x ∈ Mn (F ) with x 2 = 0, is not more
than n/2. Hence if n
3, then we conclude that a 2 = 0 and by Lemma 14 the degree of a is
2|F |n − |F | − 1. Suppose that n = 2. Obviously by Lemma 14 the degree of any non-zero
nilpotent matrix x ∈ M2 (F ) as a vertex of Γ (M2 (F )) is 2|F |2 − |F | − 2. So in this case
we have a 2 = 0 and the degree of a is 2|F |n − |F | − 2.

Theorem 16. Let R and S be two finite semisimple rings which are not fields. If Γ (R) 
Γ (S), then R  S.

Proof. By the Wedderburn–Artin Theorem, we can write R = Mn1 (F1 ) × · · · × Mnr (Fr )
and S = Mm1 (K1 ) × · · · × Mms (Ks ), where F1 , . . . , Fr and K1 , . . . , Ks are finite fields
and n1 , . . . , nr and m1 , . . . , ms are natural numbers. Clearly we may assume that r
s.
Indeed each vertex in Γ (R) with minimum degree has exactly one non-unit component.
Without loss of generality, suppose that a = (a1 , u2 , . . . , ur ) is a vertex with minimum
degree in Γ (R), where a1 is a zero-divisor and ui ’s are units. Also we may assume that
a  = (a1 , u2 , . . . , ur ) is the vertex of Γ (S) corresponding to a, where a1 is a zero-divisor
and ui ’s are units. We claim that n1 = m1 and F1  K1 .
If n1 = 1, then a1 = 0 and the set of all neighbors of a, that is {(u1 , 0, . . . , 0) |
u1 ∈ F1∗ }, is independent. Suppose that m1
2. Clearly the set of all neighbors of a 
is {(z1 , 0, . . . , 0) | z1 ∈ (Ann (a1 ) ∪ Annr (a1 ))∗ }. We know that there are two invert-
ible matrices p and q such that the m1 th row of pa1 and the m1 th column of a1 q
 S. Akbari, A. Mohammadian / Journal of Algebra 296 (2006) 462–479 471

are 0. Assume that the ith column of a1 is non-zero. Now for an index j = i, we have
Eim1 (pa1 ) = 0 and (a1 q)Em1 j = 0. Since a1 (Eim1 p) = 0, we conclude that Eim1 p =
qEm1 j . But (qEm1 j )(Eim1 p) = 0, so obviously we can find two neighbors of a  which are
adjacent. This contradiction implies that m1 = 1. Since the degree of a is |F1 | − 1 and the
degree of a  is |K1 | − 1, we conclude that |F1 | = |K1 | and F1  K1 .
So assume that n1
2 and m1
2. First suppose that r = s = 1. If n1 = 2, then by
Remark 15, the difference between the maximum degree and the minimum degree of Γ (R)
is 1. Since Γ (R)  Γ (S), by Remark 15 we conclude that m1 = 2. Now by considering
the maximum degree of two graphs Γ (R) and Γ (S), we find that 2|F1 |2 − |F1 | − 1 =
2|K1 |2 − |K1 | − 1. This implies that |F1 | = |K1 | and so F1  K1 . Thus we can assume that
n1
3 and m1
3. By Remark 15 we have a12 = 0 and so in this case the degree of a is
2|F1 |n1 − |F1 | − 1. Similarly we find that the degree of a  is 2|K1 |m1 − |K1 | − 1. Thus

2|F1 |n1 − |F1 | = 2|K1 |m1 − |K1 |. (∗)

Moreover, by Remark 15 the maximum degree of the vertices of Γ (R) is 2|F1 |n1 (n1 −1) −
|F1 |(n1 −1) − 1 and the maximum degree of the vertices of Γ (S) is 2|K1 |m1 (m1 −1) −
2

|K1 |(m1 −1) −1. This yields that 2|F1 |n1 (n1 −1) −|F1 |(n1 −1) = 2|K1 |m1 (m1 −1) −|K1 |(m1 −1) .
2 2 2

Now since |F1 | and |K1 | are prime powers, using Eq. (∗), it is not hard to see that n1 = m1
and |F1 | = |K1 | and so F1  K1 .
Next suppose that r
2. Since the set of all neighbors of a is {(z1 , 0, . . . , 0) | z1 ∈
(Ann (a1 ) ∪ Annr (a1 ))∗ }, the degree of a is 2|F1 |n1 − |F1 | − 1. Clearly in this case (∗)
holds. Assume that A is the set of all vertices with the minimum degree of Γ (R) that
have at least one common neighbor with a. Since each pair of the vertices in Γ (Mn1 (F1 ))
has at least one common neighbor [12, Example 2.8], A is the set of all vertices whose
first components are vertices with the minimum degree of Γ (Mn1 (F1 )) and their other
components are units. Let B be the set of all vertices of Γ (R) that are adjacent to at least
one vertex of A. Since clearly every vertex of Γ (Mn1 (F1 )) is adjacent to a vertex with
the minimum degree in Γ (Mn1 (F1 )), we obtain that B is the set of all vertices with the
property that each vertex has a non-zero zero-divisor as its first component and all its other
components are 0. Moreover, the vertices with the minimum degree among all vertices
adjacent to all vertices of B have the form (0, v2 , . . . , vr ), where vi ’s are units. Let C
denote the set of these vertices. Since C is not empty and Γ (R)  Γ (S), we have s
2.
2
Moreover, the degree of each vertex of C is |F1 |n1 − 1 and by the fact Γ (R)  Γ (S), we
2 2
conclude that |F1 |n1 − 1 = |K1 |m1 − 1. Again since |F1 | and |K1 | are prime powers, by (∗)
we have n1 = m1 and |F1 | = |K1 |. So F1  K1 and the claim is proved.
Now let D be the set of all vertices not in C and adjacent to all vertices of B. Indeed, D
is the set of all vertices with the property that each vertex has zero as its first component
and at least one of its other components is zero-divisor. Assume that R1 = Mn2 (F2 ) ×
· · · × Mnr (Fr ) and S1 = Mm2 (K2 ) × · · · × Mms (Ks ). Now it is easily checked that the
induced subgraph on D is isomorphic to Γ (R1 ). Again since Γ (R)  Γ (S), we conclude
that Γ (R1 )  Γ (S1 ). If R1 is a field, then S1 is also a field. In this case, the set C has
|F1 | − 1 vertices and by Γ (R)  Γ (S), we obtain R1  S1 . Otherwise, by induction on r,
we have R1  S1 , and this completes the proof. 2
472 S. Akbari, A. Mohammadian / Journal of Algebra 296 (2006) 462–479

In [2] it has been proved that for any finite commutative ring R with identity, with
finitely many exceptions, if Γ (R) is isomorphic to the zero-divisor graph of a reduced
ring S, then R  S. Now we generalize it to any arbitrary ring.

Theorem 17. Let R and S be two finite rings which are not fields. If S is reduced and
Γ (R)  Γ (S), then R  S, except in the following cases.

(i) S  Z2 × Z2 and either R is isomorphic to Z9 , Z3 [x]/(x 2 ) or the null ring of order 3;

(ii) S  Z6 and R is isomorphic to one of the rings R, Rop , Z8 , Z2 [x]/(x 3 ), Z4 [x]/
(2x, x 2 − 2) or 2Z8 = {0, 2, 4, 6};
(iii) S  Z2 × Fpn and R is isomorphic to the direct product of the null ring of order 2
and Fq m , where p and q are primes and p n = 2q m − 1.

Proof. Since S is reduced, by the Wedderburn–Artin Theorem, there are finite fields
F1 , . . . , Fs such that S  F1 × · · · × Fs . Clearly for any x ∈ D(S)∗ , there exists x  ∈ D(S)∗
such that xx  = 0 and x + x  is a unit. This yields that the vertices x and x  in Γ (S)
have no common neighbor. If R is a reduced ring, then R is a semisimple ring and Theo-
rem 16 implies that R  S. So assume that there exists a ∈ D(R)∗ such that a 2 = 0. Since
Γ (R)  Γ (S), there exists a vertex b adjacent to a such that a and b have no common
neighbor.
First suppose that ab = 0 and ba = 0. Assume that bj = 0 and bj −1 = 0 for some
integer j . Since ba = 0, then a + bj −1 ∈ / {0, b}. So the vertex a + bj −1 is adjacent to both
a and b, a contradiction. Hence b is not nilpotent. Since R is finite, there exist two integers
m > n such that bm = bn . Assume that n is the smallest possible positive integer with this
property. If n
2, then bm−1 − bn−1 is a common neighbor of a and b, a contradiction.
So n = 1 and hence e = bm−1 is a non-trivial idempotent. Since b and e have the same
neighbors, a and e have no common neighbor. Furthermore since ea ∈ Ann ({a, b}), we
have ea = a. Now noting that ae = 0, ea = a and R = eRe ⊕ eR(1 − e) ⊕ (1 − e)Re
⊕ (1 − e)R(1 − e), we find that a ∈ eR(1 − e). On the other hand (1 − e)R(1 − e) ⊆
Ann ({a, e}) and thus (1 − e)R(1 − e) = {0}. This implies that (1 − e)Re ⊆ Ann (a) ∩
Annr (e). Since a and e have no common neighbor, (1 − e)Re = {0}. This yields that

R = eRe ⊕ eR(1 − e) (∗)

and so a is adjacent to all vertices of Γ (R). Hence by [4, Theorem 2.5], Γ (S) and therefore
Γ (R) is a star graph. We have ab = ba, so R is a non-commutative ring. Now Theorem 9
and the Wedderburn’s Little Theorem [8, p. 203] yield that Γ (R) has at most three vertices.
By (∗), e is a left identity which is a zero-divisor and hence R has no identity. Also by
Lemma 2, we conclude that |R|  4. Thus we have R = {0, e, a, e + a} and it is not hard
to verify that R  R. So Γ (R) is a star graph with three vertices. By [3, Example 2.1(a)],
since S is reduced, we have S  Z6 . Similarly, if ba = 0 and ab = 0, then R  Rop and
again S  Z6 .
Next suppose that ab = ba = 0. Assume that bk = 0 and bk−1 = 0, for some integer k. If
k = 2, then {a − b, b − a} ⊆ Ann ({a, b}) and so we conclude that a − b = b and b − a = a.
Since Γ (R) is a connected graph [9, Theorem 3.2] and a and b have no common neighbor,
 S. Akbari, A. Mohammadian / Journal of Algebra 296 (2006) 462–479 473

if Γ (R) has more than two vertices, then there is a vertex x, adjacent to exactly one of
the vertices a and b, and x = a, b. But a = 2b and b = 2a, a contradiction. Thus Γ (R)
has exactly two vertices and by Remark 1, we have |R|  9. If |R| = 3, then R is the null
ring of order 3. If |R|
4, then noting that |D(R)| = 3, by Lemma 2, R has identity. Since
3a = 0 and D(R)2 = {0}, we obtain that |R| = 9. Clearly, the subring generated by a and
1 is equal to R and so R is commutative. Hence by [3, Example 2.1(a)], R is isomorphic
to Z9 or Z3 [x]/(x 2 ) and S  Z2 × Z2 .
So assume that k
3. Since bk−1 a = bk−1 b = 0, we have a = bk−1 . This implies that a
is the only vertex adjacent to b. Now we claim that for any two vertices u, v in Γ (S), if v
is adjacent to u and all other neighbors of u, then u is a vertex of degree 1. To prove this,
suppose that there exists a vertex w = v adjacent to u. The vertex v − w is adjacent to u,
so by assumption, v is adjacent to both w and v − w and this implies that v 2 = 0. But since
S is a reduced ring, we reach a contradiction. This proves the claim. Since Γ (R)  Γ (S),
the above argument shows that for any two vertices x, y in Γ (R), if y is adjacent to x and
all other neighbors of x, then x is a vertex of degree 1. Note that a = bk−1 is adjacent to
bk−1 + bk−2 and its all other neighbors. Thus by the above argument a = bk−1 is the only
vertex adjacent to bk−1 + bk−2 . We have b2 (bk−1 + bk−2 ) = 0. Now if k
4, then since
ab = 0 and k is the smallest integer which bk = 0, a = b2 and so we conclude that bk−1 +
bk−2 = b2 . Multiplying this equation by b2 yields that k = 4. But we have b3 + b2 = b2 ,
a contradiction. Therefore k = 3. Suppose az = 0, for some z ∈ R \ {0, a, b}. Since a and b
have no common neighbors, bz = 0. On the other hand a(bz) = b(bz) = 0. Therefore, since
b2 = 0, we have bz = a = b2 . Hence b(z − b) = 0 and since a(z − b) = 0, we conclude
that z − b = a. Thus Annr (a) = {0, a, b, a + b}. By a similar argument we obtain that
Ann (a) = {0, a, b, a + b}. Since bR is an additive subgroup of Annr (a), |bR| = 2 or 4.
Furthermore since a = b2 , we have Annr (b) = {0, a}. If |bR| = 2, then |R| = 4 and so
R = {0, a, b, a + b}. Clearly in this case we have R  2Z8 . So assume that |bR| = 4. This
implies that |R| = 8. Now since a = b2 , b3 = 0 and a and b have no common neighbor,
it is easy to see that a is the only vertex adjacent to a + b. So a is a vertex with degree
2 which is adjacent to two vertices of degree 1. Now since Γ (R) is a connected graph [9,
Theorem 3.2], we conclude that Γ (R) is a star graph with three vertices. Since |D(R)| = 4,
by Lemma 2, R has identity. If R is non-commutative, then by [7] R is isomorphic to an
upper triangular matrix ring over Z2 and therefore R has five zero-divisors, a contradiction.
Thus R is a commutative ring with identity. Hence by [3, Example 2.1(a)], in this case R
is isomorphic to one of the rings Z8 , Z2 [x]/(x 3 ), or Z4 [x]/(2x, x 2 − 2) and S  Z6 .
Now suppose that b is not nilpotent. Since R is finite, there exist integers m > n which
bm = bn . Assume that n is the smallest possible positive integer with this property. If n
2,
then since bm−1 − bn−1 ∈ Annr ({a, b}) and b2 = 0, we conclude that bm−1 − bn−1 = a.
Thus a is adjacent to bn−1 and all neighbors of bn−1 . This implies that a is the only
vertex adjacent to bn−1 . But bn−1 is also adjacent to bm−n+1 − b. Thus bm−n+1 − b = a =
bm−1 − bn−1 . So bm−n+2 = b2 and by the minimality of n, we obtain that n = 2. Now
we claim that a and b2 have no common neighbor. Assume that there is a vertex y = a
in Γ (R) such that b2 y = 0. Since a is the only vertex adjacent to b, by = 0 and yb = 0.
Also since a(by) = 0 and a and b have no common neighbor and b2 = 0, we conclude that
by = a and so ay = (bm−1 − b)y = −a = 0. On the other hand b(yb) = ab = 0. Again,
since (yb)a = 0 and a and b have no common neighbor and b2 = 0, we have yb = a. This
474 S. Akbari, A. Mohammadian / Journal of Algebra 296 (2006) 462–479

implies that yb2 = ab = 0 and hence ya = y(bm−1 − b) = −a = 0. Also if there is a vertex

y  = a such that y  b2 = 0, then by a similar argument we find that the both ay  and y  a
are non-zero. This proves the claim. Now let g = bm−1 , if n = 1; and g = b2m−4 , if n = 2.
Clearly g is a non-trivial idempotent. Also since bn g = gbn = bn , it is easily checked that
Ann (g) = Ann (bn ) and Annr (g) = Annr (bn ). So we conclude that a is the only vertex
adjacent to g. Therefore, noting that ag = ga = 0, we obtain gR(1 − g) = (1 − g)Rg = {0}
and (1 − g)R(1 − g) = {0, a}. Hence R = gRg ⊕ {0, a} and so a is adjacent to all vertices
of Γ (R). Thus by [4, Theorem 2.5], Γ (S) and therefore Γ (R) is a star graph. So by
Theorem 9 and noting that S is reduced, we find S  Z2 × Fpn and R is isomorphic to the
direct product of the null ring of order 2 and Fq m , where n, m are natural numbers and p, q
are prime numbers. Now by considering the number of the vertices of Γ (R) and Γ (S), we
find that p n = 2q m − 1 and so case (iii) occurs and the proof is complete. 2

Now by combining the previous theorem, Theorems 9 and 12, we are able to determine
all finite rings R for which the graph Γ (R) is bipartite.

Corollary 18. Two rings R and Rop are the only finite non-commutative rings whose Γ (R)
and Γ (Rop ) are bipartite.

Remark 19. We would like to determine all rings whose zero-divisor graphs have at most
four vertices. First assume that Γ (R) has at most three vertices. Corollary 4 characterizes
all rings whose zero-divisor graphs have exactly one vertex. Since Γ (Z2 × Z2 ) and Γ (Z6 )
are star graphs with two and three vertices, respectively, so Theorem 17 determines all
rings whose zero-divisor graphs are a star with at most three vertices. Thus it is enough
to assume that Γ (R) is the complete graph with three vertices. By Theorem 5, we have
D(R)2 = {0}. If R = D(R), then R is a null ring of order 4; otherwise using Lemma 2,
R is a local ring. By [3, Example 2.1(a)], if R is a commutative ring with identity such
that Γ (R) is a complete graph with three vertices, then R is isomorphic to one of the
rings F4 [x]/(x 2 ), Z4 [x]/(x 2 +x +1), Z4 [x]/(2, x)2 , or Z4 [x, y]/(x, y)2 . Hence we should
determine all non-commutative local rings R with |D(R)| = 4. Using Remark 1, we have
|R|  16. Since R has identity and D(R) is a nilpotent ideal of order 4, |R| is a power
of 2. So |R| = 8 or 16. If |R| = 8, then since D(R)2 = {0}, the subring generated by D(R)
and 1 is equal to R which contradicts the non-commutativity of R. Therefore |R| = 16.
Since R is a local ring, D(R) is the Jacobson radical of R and D(R) R
= F4 . Let F be a
subset with four elements of R such that 0 ∈ F and D(R) = {u + D(R) | u ∈ F}. If z is
R

an element of D(R)∗ , then it is easy to see that R = {u + vz | u, v ∈ F}. This implies that
for each t ∈ F, there exists an element σ (t) ∈ F such that zt = σ (t)z. We show that σ
induces a non-trivial automorphism of D(R) R
= F4 . For two elements t1 , t2 ∈ F, we have
σ (t1 + t2 )z = z(t1 + t2 ) = (σ (t1 ) + σ (t2 ))z. Thus σ (t1 + t2 ) − (σ (t1 ) + σ (t2 )) ∈ D(R) and
so σ ((t1 +D(R))+(t2 +D(R))) = σ (t1 +t2 )+D(R) = (σ (t1 )+D(R))+(σ (t2 )+D(R)).
Similarly one can see that σ is a multiplicative homomorphism. Also if σ (t) ∈ D(R) for
some t ∈ F, then zt = 0 and thus t = 0. This implies that σ is injective and therefore it
is an automorphism of D(R) R
, as desired. Since R is a non-commutative ring and F4 has
 S. Akbari, A. Mohammadian / Journal of Algebra 296 (2006) 462–479 475

exactly two automorphisms, we conclude that σ (t) = t 2 for any t ∈ F. Now consider the
map φ : R → D(R)
R
defined by

u + D(R) v + D(R)
φ(u + vz) = .
0 + D(R) u2 + D(R)

It is not hard to see that R is isomorphic to

 
a b


R= a, b ∈ F4 .
0 a2

Now suppose that Γ (R) has four vertices. By Theorem 3, if R is not a reduced ring, then
either R is isomorphic to one of the rings Z25 , Z5 [x]/(x 2 ) or the null ring of order 5. So
assume that R is reduced. If R is a direct product of three or more fields, then |D(R)|
7.
Thus R is a direct product of two finite fields. Now since Γ (R) has four vertices, it is easily
checked that R is isomorphic to one of the rings Z2 × F4 or Z3 × Z3 . Therefore we have
determined all rings whose zero-divisor graphs have at most four vertices.

Now we want to prove that if R is a full matrix ring over a finite field, then R is deter-
mined by its zero-divisor graph.

Theorem 20. Let F be a finite field and n
2. If R is a ring such that Γ (R)  Γ (Mn (F )),
then R  Mn (F ).

Proof. Suppose that Ω is an arbitrary clique which is maximal among all cliques contain-
ing the vertices of the maximum in-degree in Γ (Mn (F )). By Lemma 14, any element of Ω
has rank 1 and its square is non-zero, so ω2 = trace(ω)ω = 0, for any ω ∈ Ω. Hence, not-
ing that the elements of Ω are pairwise commuting, Ω is simultaneously diagonalizable.
So there is an invertible matrix h such that for each ω ∈ Ω, hωh−1 = λEii , for some i,
1  i  n, and λ ∈ F ∗ , where i and λ depend on ω. Since the product of any two matri-
ces of Ω is zero, by the maximality of Ω, there are elements λ1 , . . . , λn in F ∗ such that
hΩh−1 = {λ1 E11 , . . . , λn Enn }. Moreover since Γ (R)  Γ (Mn (F )), every clique X in
Γ (R) which is maximal among all cliques containing the vertices of maximum in-degree
has the following properties:

(i) For any vertex a ∈ X, there is a vertex u ∈/ X such that u → x is an edge of Γ (R), for
each x ∈ X \ {a}, and the directed edge u → a is not in Γ (R);
(ii) For any vertices x ∈ X and x  ∈
/ X, if (X \ {x}) ∪ {x  } is a clique of Γ (R), then u → x
is an edge if and only if u → x  is an edge, for any vertex u = x, x  ;
(iii) There is no vertex u such that x → u is an edge of Γ (R), for all x ∈ X, and there is
no vertex v such that v → x is an edge of Γ (R), for all x ∈ X.

Now we claim that there is a clique which is maximal among all cliques containing
the vertices of maximum in-degree of Γ (R) whose elements are idempotent. Let A be an
arbitrary clique which is maximal among all cliques containing the vertices of maximum
476 S. Akbari, A. Mohammadian / Journal of Algebra 296 (2006) 462–479

in-degree in Γ (R). We show that A has no nilpotent element. Assume that there is a nilpo-
tent element a ∈ A and suppose that m is the smallest positive integer such that a m = 0.
By the maximality of A, a m−1 ∈ A. Suppose that a m−1 = a. For any vertex u = a m−1
of Γ (R), if u → a is an edge, then u → a m−1 is also an edge, which contradicts (i). Thus
a m−1 = a and so a 2 = 0. Assume that b is an element of A \ {a}. By (i), we have a + b = 0.
Now for any vertex u = a + b, if u → x is an edge of Γ (R), for each x ∈ A \ {a + b},
then u → a + b is also an edge. Hence (i) implies that a + b ∈ / A. On the other hand
(A \ {b}) ∪ {a + b} is a clique of Γ (R). Therefore by (ii), u → a + b is an edge if and
only if u → b is an edge, for any vertex u ∈ / {a + b, b}. This implies that if u → b is an
edge, then u → a is an edge, for each vertex u = a, which contradicts (i). So none of the
elements of A is nilpotent. Now since Γ (R) is a finite graph, R is finite and hence for any
t ∈ A there is a positive integer m(t) such that t m(t) is a non-trivial idempotent, see [11,
p. 55]. By the maximality of in-degree of t, Ann (t) = Ann (t m(t) ) and thus t m(t) is a ver-
tex with the maximum in-degree of Γ (R). Thus {t m(t) | t ∈ A} = {e11 , . . . , enn } is a clique
which is maximal among all cliques containing the vertices of the maximum in-degree in
Γ (R) whose elements are idempotents, so the claim is proved.
Suppose that there is z ∈ D(R)∗ such that (e11 + · · · + enn )z = 0. Multiplying this
equation by eii , we have eii z = 0, for any i, 1  i  n. Thus eii → z is an edge in Γ (R), for
any i, which contradicts (iii). Similarly, we observe that Ann (e11 + · · · + enn ) = {0}. Since
e11 + · · · + enn is an idempotent element which is not a zero-divisor, 1 = e11 + · · · + enn is
the identity of R.
To complete the proof, we need other properties of the graph Γ (Mn (F )). Let Ω be an ar-
bitrary clique which is maximal among all cliques containing the vertices of maximum in-
degree of Γ (Mn (F )) and without loss of generality suppose that Ω = {λ1 E11 , . . . , λn Enn },
where λl ’s are contained in F ∗ . For any i = j , 1  i, j  n, assume that Wij is the set of
all vertices w such that w → λr Err and λs Ess → w are edges in Γ (Mn (F )), for any
1  r, s  n, r = j , and s = i. It is not hard to see that Wij = {λEij | λ ∈ F ∗ }. Now, for
any i = j , 1  i, j  n, suppose that Vij ’s are the sets with the above properties associ-
ated to the clique {e11 , . . . , enn } in Γ (R). Since Γ (R)  Γ (Mn (F )), Vij ’s have all of the
properties of Wij ’s. Note that there is no a directed edge from eii to a vertex of Vij , and
also there is no a directed edge from a vertex of Vij to ejj , for any i, j . Furthermore for
any vij ∈ Vij and vkl ∈ Vkl , vij → vkl is an edge of Γ (R) if and only if j = k. Moreover,
we show that for any distinct indices i, j, k we have Vij Vj k ⊆ Vik . To see this, let vij ∈ Vij
and vj k ∈ Vj k be two arbitrary elements. We know that vij vj k = 0. For any 1  r, s  n,
r = k, and s = i, by the definitions of Vij and Vj k , we have vj k err = 0 and ess vij = 0. This
implies that vij vj k → err and ess → vij vj k are edges in Γ (R). Thus by the definition,
vij vj k ∈ Vik . Note that with a similar proof we find that for any i = j , the difference of
two distinct elements of Vij is contained in Vij .
We claim that for any j , 2  j  n, if vij is a vertex in Vij , then e11 v1j = v1j . By the
definition, we know that for any v1j ∈ V1j , e11 v1j ∈ V1j . Also, if e11 v1j = e11 v1j , for some
 
v1j = v1j in V1j , then e11 (v1j − v1j ) = 0, which is a contradiction, since v1j − v1j  ∈V .
1j
Therefore by the finiteness of V1j , for any vertex v1j ∈ V1j , there is a vertex ṽ1j such that
e11 ṽ1j = v1j . Thus e11 v1j = e11 (e11 ṽ1j ) = e11 ṽ1j = v1j and so the claim is established.
For any j , 2  j  n, fix a vertex e1j ∈ V1j . Now with a similar proof to the above, we
show that for any i = j , 2  i  n and 1  j  n, there is a unique vertex eij ∈ Vij such
 S. Akbari, A. Mohammadian / Journal of Algebra 296 (2006) 462–479 477

that e1i eij = e1j . We know that for any vij ∈ Vij , e1i vij ∈ V1j . If e1i vij = e1i vij , for some
vij = vij in Vij , then e1i (vij − vij ) = 0, which is a contradiction, since vij − vij ∈ Vij . But
Vij is a finite set, so there is a unique vertex eij ∈ Vij such that e1i eij = e1j .
Finally, we prove that {eij | 1  i, j  n} is a set of elements of R with the properties
that e11 + · · · + enn = 1 and eij ekl = δj k eil , for all indices i, j, k, l. For this, it is enough to
show that eij ej k = eik , for any indices i, j, k. As in the previous paragraph, there exists a
vertex ej k ∈ Vj k such that eij ej k = eik . Thus e1k = e1i eik = e1i (eij ej k ) = e1j ej k . Hence,
by the uniqueness of ej k , we conclude that ej k = ej k and so eij ej k = eik .
Now by [10, Proposition 1.1.3], we have R  Mn (S), for some finite ring S. To com-
plete the proof, we prove that S is a field. To get a contradiction, let yy  = 0, for some
y, y  ∈ D(S)∗ . Clearly, the in-degree of the vertex y  E11 is more than |S|n(n−1) − 1
and by Lemma 14, the maximum in-degree in Γ (Mn (F )) is |F |n(n−1) − 1, so we have
|F |n(n−1) − 1 > |S|n(n−1) − 1. This implies that |F | > |S|. Furthermore we know that
the in-degree of the vertex I − E11 is equal to |S|n − 1 and by Lemma 14, the mini-
mum in-degree in Γ (Mn (F )) is |F |n − 2. This follows that |S|n − 1
|F |n − 2 and so
0 < |F |n − |S|n  1, which is a contradiction. Hence by the Wedderburn’s Little Theorem
[8, p. 203], S is a field. Now by Lemma 14, the maximum in-degree of all vertices of the
graphs Γ (Mn (S)) and Γ (Mn (F )) are |F |n(n−1) − 1 and |S|n(n−1) − 1, respectively. This
yields that |S| = |F | and so S  F . Therefore R  Mn (F ) and the proof is complete. 2

Now we want to generalize the previous theorem.

Theorem 21. Let R be a ring and S be a finite semisimple ring which is not a field. If
Γ (R)  Γ (S), then R  S, unless R and S are isomorphic to one of the commutative
rings given in Theorem 17.

Proof. By the Wedderburn–Artin Theorem, we may write S = Mn1 (F1 ) × · · · × Mns (Fs ),
where Fi ’s are finite fields, n1 , . . . , ns , s are natural numbers. Note that if either all ni ’s
are equal 1 or s = 1, then by Theorems 17 and 20, the assertion is proved. So suppose that
at least one of ni ’s and s is more than 1. Without loss of generality, assume that n1
2.
Let x be the vertex of Γ (R) corresponding to the vertex y = (I, 0, . . . , 0) in Γ (S). Since
all edges incident to y in Γ (R) are multiple, x has this property too. We show that x is
not nilpotent. To get a contradiction, assume that x k = 0, where k is the smallest positive
integer with this property. We know that there is no vertex in Γ (S) adjacent to y and all
other neighbors of y. If x k−1 = x, then x k−1 is adjacent to x and all other neighbors of
x in Γ (R), a contradiction. Thus x k−1 = x and so x 2 = 0. Moreover, since every vertex
in Rx\{0, x} is adjacent to x and all other neighbors of x, we have Rx ⊆ {0, x}. Since
there is no a vertex adjacent to all other vertices of Γ (S), this property holds for Γ (R).
Hence Rx = {0, x} and so [R : Ann (x)] = 2. Since x 2 = 0, we conclude that the maxi-
mum in-degree of the vertices of Γ (R) and the in-degree of x differ at most 1. This is a
contradiction, since y clearly does not have this property as a vertex of Γ (S). This shows
that x is not nilpotent.
Now since R is finite, there is a natural number m such that e = x m is a non-trivial
idempotent, see [11, p. 55]. For some z ∈ R, if ez = 0, then x(x m−1 z) = 0 and since
all edges incident to x in Γ (R) are multiple, we have (x m−1 z)x = 0. By repeating this
478 S. Akbari, A. Mohammadian / Journal of Algebra 296 (2006) 462–479

procedure, we obtain that ze = zx m = 0. Thus we have Annr (e) ⊆ Ann (e). The converse
is similarly proved and we find that Ann (e) = Annr (e). This shows that all edges incident
to e in Γ (R) are multiple. Hence eR(1 − e) = (1 − e)Re = {0} and so

R = eRe ⊕ (1 − e)R(1 − e). (∗)

Assume that g is the vertex in Γ (S) corresponding to e. Since e is adjacent to all neighbors
of x in Γ (R), g is adjacent to all neighbors of y in Γ (S). This yields that the j th component
of g is zero, for each j
2. Also it is straightforward to see that for any matrix a ∈
Mn1 (F1 ), the sets Ann (a) \ Annr (a) and Annr (a) \ Ann (a) are non-empty. Now since
all edges incident to g in Γ (S) are multiple, we conclude that g = (g1 , 0, . . . , 0), for some
unit g1 ∈ Mn1 (F1 ). Let Ag be the set of all vertices in Γ (S), adjacent to all neighbors
of g such that their degrees are more than the degree of g. Clearly the first component
of every vertex of Ag is a non-zero zero-divisor and its other components are 0. Also
let Bg be the set of all vertices in Γ (S) adjacent to g and adjacent to at least one of the
neighbors of g. Hence the first component of any vertex of Bg is 0 and at least one of
its other components is a zero-divisor. Now if the induced subgraphs of Γ (S) on the Ag
and Bg are denoted by ΓAg and ΓBg , respectively, then we have ΓAg  Γ (Mn1 (F1 )) and
ΓBg  Γ (Mn2 (F2 ) × · · · × Mns (Fs )).
Assume that Ae and Be denote the sets of the vertices in Γ (R) corresponding to Ag
and Bg , respectively. By (∗), clearly (1 − e)R(1 − e)∗ is the set of all neighbors of e
in Γ (R). Suppose that a = a1 + a2 is an arbitrary element of Ae , where a1 ∈ eRe and
a2 ∈ (1 − e)R(1 − e). So, by the definition of Ae , a1 + a2 is adjacent to all vertices of
(1 − e)R(1 − e)∗ in Γ (R) and therefore for any x ∈ R we have either a2 x = 0 or xa2 = 0.
Since Γ (R) has no vertex adjacent to all other vertices of the graph, we conclude that
a2 = 0. Moreover since the degree of a is more than the degree of e in Γ (R), we find
that a = a1 is a non-zero zero-divisor in the ring eRe. So clearly we have ΓAe  Γ (eRe),
where ΓAe is the induced subgraph of Γ (R) on Ae . Suppose that b = b1 + b2 is an arbi-
trary element of Be , where b1 ∈ eRe and b2 ∈ (1 − e)R(1 − e). Since b is adjacent to e
in Γ (R), we have b1 = 0. Also since at least one of the neighbors of b in Γ (R) is con-
tained in (1 − e)R(1 − e)∗ , we obtain that b = b2 is a non-zero zero-divisor in the ring
(1 − e)R(1 − e). Thus clearly we have ΓBe  Γ ((1 − e)R(1 − e)), where ΓBe is the in-
duced subgraph of Γ (R) on Be .
Now since Γ (R)  Γ (S), we conclude that Γ (eRe)  Γ (Mn1 (F1 )) and so by The-
orem 20, we obtain eRe  Mn1 (F1 ). Furthermore, we have Γ ((1 − e)R(1 − e)) 
Γ (Mn2 (F2 ) × · · · × Mns (Fs )). By repeating this procedure l times, where l = |{ni |
ni
2}|, we find a semisimple ring T such that R  T × R1 and S  T × S1 , where
S1 is a reduced ring and Γ (R1 )  Γ (S1 ). Since Γ (R)  Γ (S), we have |D(R)| = |D(S)|.
Hence we find

|T |
D(R1 )
+
D(T )
|R1 | −
D(T )

D(R1 )










= |T |
D(S1 )
+
D(T )
|S1 | −
D(T )

D(S1 )
.
 S. Akbari, A. Mohammadian / Journal of Algebra 296 (2006) 462–479 479

Therefore since |D(R1 )| = |D(S1 )|, we conclude that |R1 | = |S1 |. Now it is easy to see
that for all exceptions given in Theorem 17, |R1 | = |S1 |. Hence by Theorem 17, R1  S1
and the proof is complete. 2

Example 22. The assumption that S is semisimple is necessary in the previous theorem.
To see this consider the following example.
    
a b c
2a b




a, b, c, d ∈ Z2 .
R= a, b ∈ Z4 and S = 0 a d
0 0
0 0 a

It is easy to verify that Γ (R)  Γ (S). However, R and S have different characteristics
and so they are not isomorphic.

Acknowledgments

The authors are indebted to the Institute for Studies in Theoretical Physics and Mathe-
matics (IPM) for support. The research of the first author was in part supported by a grant
from IPM (No. 83050211). The authors thank a referee for numerous helpful comments.

References

[1] S. Akbari, H.R. Maimani, S. Yassemi, When a zero-divisor graph is planar or a complete r-partite graph,
J. Algebra 270 (2003) 169–180.
[2] S. Akbari, A. Mohammadian, On the zero-divisor graph of a commutative ring, J. Algebra 274 (2004) 847–
855.
[3] D.F. Anderson, A. Frazier, A. Lauve, P.S. Livingston, The zero-divisor graph of a commutative ring, II, in:
Lecture Notes in Pure and Appl. Math., vol. 220, Dekker, New York, 2001, pp. 61–72.
[4] D.F. Anderson, P.S. Livingston, The zero-divisor graph of a commutative ring, J. Algebra 217 (1999) 434–
447.
[5] M. Axtell, J. Coykendall, J. Stickles, Zero-divisor graphs of polynomials and power series over commutative
rings, Comm. Algebra, in press.
[6] I. Beck, Coloring of commutative rings, J. Algebra 116 (1988) 208–226.
[7] K.E. Eldridge, Orders for finite noncommutative rings with unity, Amer. Math. Monthly 75 (5) (1968) 512–
514.
[8] T.Y. Lam, A First Course in Noncommutative Rings, Springer, New York, 2001.
[9] S.P. Redmond, The zero-divisor graph of a non-commutative ring, Internat. J. Commutative Rings 1 (4)
(2002) 203–211.
[10] L.H. Rowen, Ring Theory, vol. I, Academic Press, Boston, MA, 1988.
[11] S. Singh, Q. Zameeruddin, Modern Algebra, third reprint, Vikas Publishing House Pvt. Ltd., Dehli, 1995.
[12] T. Wu, On directed zero-divisor graphs of finite rings, Discrete Math. 296 (2005) 73–86.