Subrings of matrix rings

Grigore Călugăreanu

Abstract

When comparing several classes of rings, matrix examples are frequently

used. In the sequel, we intend to put some order in this matter.
For a positive integer n ≥ 2 and a nonzero commutative ring with identity
R, denote by S = Mn (R) the (full) matrix ring (or unital R-algebra). Our goal
is to classify some of the subrings of S, and among these, subrings with identity
and commutative subrings, respectively. Actually these will be subalgebras
and will be described by determining subrings among finitely generated free
R-submodules.
What follows is a natural way of presenting such subrings.
We denote by Eij and call matric unit, the n× n matrix which has all entries
zero, excepting the entry on the i-th row and j-th column, which is 1 (the name
is not standard; we just want somehow to emphasize that these matrices are
not ring units, i.e., invertible matrices). Notice that Eij Ekl = δjk Eil . Then
obviously any matrix decomposes as a linear combination of all the n2 matric
n
P
units as follows A = [aij ] = aij Eij . Actually, it is easy to see that S =
i,j=1
Mn (R) is a free left (or right) R-module on the n2 element basis {Eij } (that
is, these form a linearly independent generating set). As finitely generated free
R-modules, Mn (R) ∼
2
= Rn are isomorphic R-algebras.
Since nonzero commutative rings have IBN (Invariant Basis Number, see
[1]), any two bases on a finitely generated free module have the same (finite)
number of elements.

1 Subalgebras generated by matric units

It is easy to determine the subrings among the free submodules generated by
some matric units. Consider N = {1, 2, ..., n} and for a binary relation ρ ⊆
N × N , we consider the R-submodule Mρ generated by {Eij : (i, j) ∈ ρ}. Since
there is no problem with the additive subgroup, in order to have a subring, only
closure under multiplication is needed. Owing to the above multiplication rules
for matric units, we obtain

Lemma 1 Mρ is a subring if and only if ρ is transitive (i.e., ρ ◦ ρ ⊆ ρ).

1
Proof. The condition is clearly sufficient. Conversely, let (i, j) and (j, k) be
two pairs in ρ. Since Mρ consists in linear combinations of Eij ’s with (i, j) ∈ ρ,
both Eij and Ejk belong to Mρ . Hence so is their product Eik = Eij Ejk and
(i, k) ∈ ρ, as desired.
Similarly
Lemma 2 Mρ is a subring with identity if and only if ρ is a preorder (i.e.,
reflexive and transitive).
Proof. Mρ is a subring with identity iff In = E11 + E22 + ... + Enn ∈ Mρ , and
this happens iff all Eii ∈ Mρ .
A well-known example is the natural (total) order on N given by (i, j) ∈ ρ
iff i ≤ j: this way we obtain the subring with identity of all the upper triangular
matrices.   
 a b 0 
For a commutative ring R with identity,  c 0 0  : a, b, c ∈ R is, in
0 0 0
 
M3 (R), the R-submodule generated by {E11 , E12 , E21 }. It is not subring and
does not contain the identity (matrix I3 ).
It is easy to give examples of idempotents and zero square elements generated
by matric units: Eii + Eik , k 6= i are idempotents and Eii + Eik + Eki + Ekk ,
k 6= i are idempotents. Moreover Eii + σ are idempotents, where σ means any
sum of matric units on the i-th row (or column), outside the diagonal. Actually
such σ’s are zero square and Eii σ = σ, σEii = 0.

2 Subalgebras generated by linear combinations

of matric units
When it comes to free submodules generated by some given linearly independent
linear combinations of matric units, a description is somehow similar. These are
also (finitely generated) free subalgebras of Mn (R) ∼
2
= Rn .
Linear combinations of matric units are also determined by binary relations
in N × N , just listing the pairs from the lower script. For instance, E11 + E12 +
E21 + E22 is determined by {(1, 1), (1, 2), (2, 1), (2, 2)}.
If B = {ρk }sk=1 ⊆ P(N × N ) determine some generators, then the resulting
R-submodule is a subring iff B is closed under composition: ρk ◦ ρl ∈ B ∪ ∅ for
every 1 ≤ k, l ≤ s.
Recall that (P(N ×N ), ◦, ∅) is a semigroup with zero (and a complete lattice
with respect to ⊆). Thus
Proposition 3 For a subset B ⊆ P(N × N ), the R-submodule generated as
above is a subring if and only if B is a subsemigroup (with zero) of (P(N ×
N ), ◦, ∅).
Proof. Indeed, all we need is ρ, τ ∈ B implies ρ ◦ τ ∈ B or = ∅.
As for subrings with identity, if we denote ∆M the equality relation on a set
M , we have

2
Proposition 4 For a subsemigroup (with zero) B of (P(N × N ), ◦, ∅), the R-
submodule generated as above is a subring with identity if and only if B contains
a partition of ∆N ×N . [This amounts to: there is a partition π = B1 ∪ ... ∪ Bm
of N = {1, 2, ..., n} such that all ∆Bj ∈ B].
P P
Proof. Indeed, only for such linear combinations a1 Eii + ... + am Eii ,
i∈B1 i∈Bm
we recapture the identity In = E11 + E22 + ... + Enn , by taking a1 = a2 = ... =
am = 1.
As an example,  for n = 3,  consider the set of matrices
 a b c 
 0 a d  : a, b, c, d ∈ R . Such matrices can be presented as linear com-
0 0 a
 
binations a(E11 + E22 + E33 ) + bE12 + cE13 + dE23 (here I3 = E11 + E22 + E33 ).
Therefore, this is a subalgebra generated by E11 + E22 + E33 , E12 , E13 , E23 , in
the subalgebra of all the upper triangular matrices (which is generated by all
E11 , E22 , E33 , E12 , E13 , E23 ) (in order to check it is also a subring, just use the
previous Propositions).

3 Special remarks
In the sequel, to simplify the wording, matric units Eij with i 6= j will be
called outside the diagonal. Sometimes it will be useful to decompose a matrix
n
P P
A= aii Eii + aij Eij , that is the diagonal and outside the diagonal (and
i=1 i6=j
i < j for upper triangular matrices).
Further, in presenting this way a subring (with identity), two or more matric
units are said to be connected if they have the same coefficient (otherwise, a
matric unit will be called isolated). In the example above, the matric units on
the diagonal are connected and the matric units outside the diagonal are not
connected. That is, connected matric units yield the linear combinations which
generate the subalgebra we consider (E11 + E22 + E33 in the example above).
When dealing with subrings S with identity of full matrix rings Mn (R) (i.e.,
In ∈ S), notice the following:
(i) if a matric unit on the diagonal is not connected with other matric units
on the diagonal, it must be isolated (i.e., it cannot be connected with any matric
unit outside the diagonal);
(ii) if some matric units on the diagonal are connected, matric units outside
the diagonal cannot belong
 to this connection.
 
 a a 0 
The set of matrices  0 a 0  : a ∈ R = {a(I3 + E12 ) : a ∈ R} is a
0 0 a
 
subring, but has no identity (as subalgebra, it has only one generator E11 +
E22 + E33 + E12 ).
(iii) a matric unit (or more) on the diagonal, which is connected to some
other matric units on the diagonal, may be connected to some matric units

3
outside the diagonal, butunder
 a different connection.
 

 a + b b 0 0 

0 a 0 0
  
For example, the set    : a, b ∈ R i.e., linear combina-

 0 0 a 0  

0 0 0 a
 
tions aI4 + b(E11 + E12 ) with idempotent E11 + E12 .
As observed above, if not isolated,
P all matric units Pon the diagonal must be
connected with each other, like, a1 Eii +...+am Eii , and these m linear
i∈B1 i∈Bm
combinations act like δ, i.e., identity or zero, when multiplied  with matric units
P Ejk if j ∈ B1
outside the diagonal. That is, for instance ( Eii )Ejk = ,
i∈B1 0 if j ∈ / B1
and similarly on the left. This means that when finding conditions which assure
an R-module to be a subring with identity, the matric units connected on the
diagonal are no concern with respect to closure under multiplication (when mul-
tiplied by each other, the linear combinations on the diagonal act as idempotent
or zero).   
 a1 a1 ... a1
 

 a2 a2 ... a2 
 

Other examples. (a) The set  . . . : ai ∈ R, 1 ≤ i ≤ n
 
 .. .. ... ..  

 

 
an an ... an
 
given in [3], as (general) Armendariz but not (general) reduced (sub)ring, has
no identity (it is generated by independent linear combinations: {E11 + E12 +
... +E1n , E21 + E22 + ... + E  2n , ..., En1 +  En2 + ... + Enn }.

 a a 12 a 13 a 14 

0 a a a
 
23 24 
 P
(b)   |a, a ij ∈ R , aI4 + i<j aij Eij , or

 0 0 a a 34  

 0 0 0 a
 
 
 a c 0 0
 

c a 0 0 
 
(c)    : a, b, c ∈ R , a(E11 +E22 )+b(E33 +E44 )+c(E12 +E21 ),
 0 0 b 0 
 

0 0 0 b
 
or   

 a 0 c 0 

0 a d 0
  
(d)   : a, b, c, d ∈ R , a(E11 +E22 +E33 )+bE44 +cE13 +dE23 .


 0 0 a 0  

0 0 0 b
 

An upper triangular Toeplitz matrix over  R is given as 


 a 1 a 2 a 3 ... a n−2 a n−1 a n 



  0 a1 a2 ... an−3 an−2 an−1  


   
0 0 a ... a a a
 

  1 n−4 n−3 n−2  

 .. .. .. . . .. .. ..  : a ∈ R, 1 ≤ i ≤ n , that is, gen-
  

 .
 . . . . . .  i
 
 0 0 0 ...

 a1 a2 a3 



   
 0 0 0 ...

 0 a 1 a 2
 


 
0 0 0 ... 0 0 a1
 

4
erated by the sums E11 + E22 + ... + Enn , E12 + E23 + ... + En−1,n , ..., E1,n−1 +
E2n , E1n . Here again, this is a subring with identity.

4 Commutative subrings
Further, a binary relation ρ on N will be called zero square if ρ2 = ∅ (in the
semigroup with zero (P(N ×N ), ◦, ∅)). Notice that in this case ρ∩∆N = ∅ (i.e.,
ρ does not contain equal pairs). We call a set of matric units {Eij , (i, j) ∈ ρ}
independent, if ρ is zero square (since i 6= j, these are matric units outside
the diagonal), and dependent otherwise. In this case for any (i, j), (k, l) ∈ ρ,
Eij Ekl = 0n . Notice that any linear combination of independent matric units
is zero square (more, any two such linear combinations have zero product).
As examples, every matric unit outside the diagonal is zero square. So is
E12 + E13 . More general, any outside sum of matric units on the same row (or
same column) is zero square.
What follows refers to commutative subrings.

Proposition 5 If a subring S consists only in symmetric matrices, it is com-

mutative.

Proof. Since S is closed under multiplication, for any two (symmetric) matrices
A, B ∈ S, the product is also symmetric. But this happens if and only if
AB = BA.
Further

Proposition 6 If a subring S consists only in matrices with scalar diagonal,

and outside the diagonal the matric units are independent (connected or not),
then S is commutative.

Proof. Indeed, such matrices are sums X + Y , X ′ + Y ′ with scalar X, X ′ , so

central, and Y.Y ′ = Y ′ .Y = 0. This way (X + Y )(X ′ + Y ′ ) = (X ′ + Y ′ )(X + Y ).
  

 a 0 b c 

0 a 0 d
  
Examples. (1) F =   : a, b, c, d ∈ R , i.e., linear combi-


 0 0 a 0  

0 0 0 a
 
nations aI4 + bE13 + cE14 + dE24 . Notice that here the zero square elements are
only the combinations bE13 + cE14 + dE24 . The relation {(1, 3), (1, 4), (2, 4)} is
zero square.
 So F is commutative.
 

 a b x y 

0 a b z 
 
(2) T4 =    : a, b, x, y, z ∈ R . Here the diagonal is scalar, but

 0 0 a b  

0 0 0 a
 
outside the diagonal we have matric units that are dependent:
{(1, 2), (2, 3), (3, 4)} is not zero square.

5
  
0 1 0 0
 0 0 1 0  2
Denoting N = E12 + E23 + E34 =   0 0 0 1 , we obtain N = E13 +


0 0 0 0
E24 , N 3 = E14 and N 4 = 04 . This way, an arbitrary matrix in T4 can be
written as A = aI4 + bN + xE13 + yN 3 + zE24 . For instance, E13 is zero square
but not central: N E13 = 04 6= E14 = E13 N . Here E12 , E23 , E34 ∈ / T4 (only
the sum E12 + E23 + E34 ∈ T4 ). So T4 is not commutative. However, it is
semicommutative (see [2]).

The following easy properties of matric units are useful when searching for
not central nilpotent elements.

Proposition 7 Let S be a subring of Mn (R) for any ring with identity R.

(i) If i, j, k are distinct among {1, 2, ..., n} and Eij , Ejk ∈ S then Eij is a
zero square matrix which is not central.
(ii) If a matric unit Eij ∈ S (i 6= j) is central then for all k, l ∈ {1, 2, ..., n},
Eki , Ejl ∈/ S, i.e., if there is a central matric unit on the i-th row and j-th
column in S, there cannot be other matric units on the j-th row nor on the i-th
column in S.
2
Proof. (i) For i 6= j, Eij = 0n and Eij Ejk = Eik 6= 0n = Ejk Eij .
(ii) By contradiction, suppose there exists Eki ∈ S. If k = j, then Eji ∈ S
(the symmetric) and Eij is not central: Eij Eji = Eii 6= Ejj = Eji Eij . If k 6= j,
then Eki Eij = Ekj 6= 0 = Eij Eki and again Eij is not central.
Remark. Any given matric unit Ers (r 6= s) does not commute with Esr :
indeed Ers Esr = Err 6= Ess = Esr Ers .

The question, ”how can subrings of Rn , which are not subalgebras, be

described ?”, is not addressed here.

References
[1] Lam T. Y. Lectures on rings and modules. Graduate Texts in Math. Springer,
1999.
[2] Liu Z. K. Semicommutative Subrings of Matrix Rings. Journal of Mathe-
matical Research & Exposition 26, 2 (2006), 264–268.
[3] Wang W. K. A Class of Maximal General Armendariz Subrings of Matrix
Rings. Journal of Mathematical Research & Exposition 29, 1, (2009) 185–
190.