Determinants of Blo ck Matrices
John R. Silvester
1 Intro duction
! !
e f a b
. Their sum and N = Let us rst consider the 2 2 matrices M =
g h c d
and pro duct are given by
! !
a + e b + f ae + bg af + bh
M + N = and MN = :
c + g d + h ce + dg cf + dh
Here the entries a; b; c; d; e; f ; g ; h can come from a eld, such as the real numb ers, or more
n n
generally from a ring, commutative or not. Indeed, if F is a eld, then the set R = F
of all n n matrices over F forms a ring non-commutative if n 2, b ecause its
elements can b e added, subtracted and multiplied, and all the ring axioms asso ciativity,
distributivity, etc. hold. If a;b;:::;h are taken from this ring R , then M; N can be
2 2 2n 2n
thought of either as memb ers of R 2 2 matrices over R or as memb ers of F . It
is well-known fact, which we leave the reader to investigate, that whether we compute
with these matrices as 2n 2n matrices, or as 2 2 \blo ck" matrices where the blo cks
a;b;::: are n n matrices, i.e., elements of R makes no di erence as far as addition,
subtraction and multiplication of matrices is concerned. See for example [2], p. 4, or
2 2 2n 2n
[6], pp. 100{106. In symb ols, the rings R and F can be treated as b eing identical:
2 2 2n 2n 2 n n 2 2n 2n
R = F , or F = F . More generally, we can partition any mn mn matrix
m n n m mn mn
as an m m matrix of n n blo cks: F = F .
The main p oint of this article is to lo ok at determinants of partitioned or blo ck
matrices. If a; b; c; d lie in a ring R , then provided that R is commutative there is a
determinant for M, which we shall write as det , thus: det M = ad bc, which of
R R
course lies in R . If R is not commutative, then the elements ad bc, ad cb, da bc,
da cb may not b e the same, and we do not then know which of them if any mightbe
n n
a suitable candidate for det M. This is exactly the situation if R = F , where F is a
R
eld or a commutative ring and n 2; so to avoid the diculty we take R to be, not
n n n n
the whole of the matrix ring F , but some commutative subring R F . The usual
2 2 m m
theory of determinants then works quite happily in R , or more generally in R , and
m m n n
for M 2 R we can work out det M, which will b e an elementofR . But R F , so
R
det M is actually a matrix over F , and we can work out det det M, which will b e an
R F R
n n m m m n n m mn mn
elementofF . On the other hand, since R F ,wehave M 2 R F = F ,
so we can work out det M, which will also b e an element of F . Our main conclusion is
F
that these two calculations give the same result:
n n
Theorem 1. Let R be a commutative subring of F , where F is a eld or a commu-
m m
tative ring, and let M 2 R . Then
det M = det det M: 1
F F R 1
!
A B
For example, if M = where A, B, C, D are n n matrices over F which all
C D
commute with each other, then Theorem 1says
det M = det AD BC: 2
F F
Theorem 1 will be proved later. First, in section 2 we shall restrict attention to the
case m = 2 and give some preliminary and familiar results ab out determinants of blo ck
diagonal and blo ck triangular matrices which, as a by-pro duct, yield a pro of by blo ck
matrix techniques of the multiplicative prop erty of determinants. In section 3 we shall
prove something a little more general than Theorem 1 in the case m =2; and Theorem
1 itself, for general m, will b e proved in section 4.
2 The multiplicative prop erty
!
A B
n n 2n 2n
Let M = , where A, B, C, D 2 F , so that M 2 F . As a rst case,
C D
!
A O
, a blo ck-diagonal supp ose B = C = O, the n n zero matrix, so that M =
O D
matrix. It isawell-known fact that
!
A O
det = det A det D: 3
F F F
O D
The keen-eyed reader will notice immediately that, since
det A det D = det AD; 4
F F F
equation 3 is just 2 in the sp ecial case where B = C = O. However, we p ostp one
this step, b ecause with care we can obtain a proof of the multiplicative prop erty 4 as
aby-pro duct of the main argument.
One way of proving 3 is to use the Laplace expansion of det M by the rst n rows,
F
which gives the result immediately. A more elementary pro of runs thus: generalize to
the case where A is r r but D is still n n. The result is now obvious if r = 1, by
expanding by the rst row. So use induction on r , expanding by the rst row to p erform
the inductive step. Details are left to the reader. This result still holds if we know only
that B = O, and the pro of is exactly the same: we obtain
!
A O
det = det A det D: 5
F F F
C D
By taking transp oses, or by rep eating the pro of using columns instead of rows, we also
obtain the result when C = O, namely,
!
A B
= det A det D: 6 det
F F F
O D 2
In order to prove 4 we need to assume something ab out determinants, and we shall
assume that adding a multiple of one row resp ectively, column to another row resp ec-
tively, column of a matrix do es not alter its determinant. Since multiplying a matrix
on the left resp ectively, right by a unitriangular matrix corresp onds to p erforming a
number of such op erations on the rows resp ectively, columns, it do es not alter the de-
terminant. A uni triangular matrix is a triangular matrix with all diagonal entries equal
to 1. We shall also assume that det I =1, where I is the n n identity matrix. So
F n n
now observe that
! ! ! ! !
I I A B C D I I I O
n n n n n
= ; 7
C D A B I I O I O I
n n n n
! !