Axiomatic approach to the determinant function

Recall the formula for determinants of k k matrices for k = 1, 2, 3: det[a] = a, det a1 b1 a2 b2 = a1 b2 a2 b1 and

a1 b1 c1 det a2 b2 c2 = a1 b2 c3 a1 b3 c2 a2 b1 c3 +a3 b1 c2 +a2 b3 c1 a3 b2 c1 . a3 b3 c3 Our approach to determinants of n n matrices is via their properties (rather than via an explicit formula as above). Let d be a function that associates a scalar d(A) R with every n n matrix A with entries in R. We use the following notation. If the columns of A are A1 , A2 , . . . , An , we write d(A) = d([A1 , A2 , . . . , An ]) simply as d(A1 , A2 , . . . , An ). Thus d can be regarded as a function on Rn1 Rn1 , the n-fold product of n1 column vectors.
1 / 17

Axioms for the determinant function

(i) d is called multilinear if for each j = 1, 2, . . . , n, scalars , and n 1 column vectors A1 , . . . , Aj 1 , Aj +1 , . . . , An , B, C , we have
d(A1 , . . . , Aj 1 , B + C, Aj +1 , . . . , An ) = d(A1 , . . . , Aj 1 , B, Aj +1 , . . . , An ) + d(A1 , . . . , Aj 1 , C, Aj +1 , . . . , An ).

(ii) d is called alternating if d(A1 , A2 , . . . , An ) = 0 whenever Ai = Aj for some i = j, i, j = 1, . . . , n. (iii) d is called normalized if d(I ) = d(e1 , e2 , . . . , en ) = 1, where ej is the j th basic column vector, j = 1, . . . , n. (iv) A normalized, alternating and multillinear function d on the set of all n n matrices is called a determinant function of order n.

2 / 17

Formula for the determinant of a 2 2 matrix

Suppose d is an alternating multilinear normalized function on 2 2 matrices A = [A1 , A2 ]. We show that a1 b1 d(A1 , A2 ) = d , = a1 b2 a2 b1 . a2 b2 Write the rst column as A1 = a1 e1 + a2 e2 and the second column as A2 = b1 e1 + b2 e2 . Using the axioms for a determinant function, we obtain d(e1 , e2 ) + d(e2 , e1 ) = d(e1 + e2 , e1 + e2 ) = 0, and d(A1 , A2 ) = d(a1 e1 + a2 e2 , b1 e1 + b2 e2 ) = d(a1 e1 + a2 e2 , b1 e1 ) + d(a1 e1 + a2 e2 , b2 e2 ) = d(a1 e1 , b1 e1 ) + d(a2 e2 , b1 e1 ) +d(a1 e1 , b2 e2 ) + d(a2 e2 , b2 e2 ) = a2 b1 d(e2 , e1 ) + a1 b2 d(e1 , e2 ) = (a1 b2 a2 b1 )d(e1 , e2 ) = a1 b2 b1 a2 . Exercise: Can you do this for a 3 3 matrix A = [A1 , A2 , A3 ]?
3 / 17

Properties of a determinant function

We show that there can be only one determinant function of order n. Lemma: Suppose that d(A1 , A2 , . . . , An ) is a multilinear alternating function on the set of all n n matrices. Then (a) If some Aj = 0, then d(A1 , A2 , . . . , An ) = 0.
(b) d(A1 , A2 , . . . , Aj , Aj +1 , . . . An ) = d(A1 , A2 , . . . , Aj +1 , Aj , . . . , An ). (c) d(A1 , A2 , . . . , Ai , . . . , Aj , . . . , An ) = d(A1 , A2 , . . . , Aj , . . . , Ai , . . . , An ).

Proof: (a) Let Aj = 0. For any column C , by the multilinearity, d(A1 , A2 , . . . , Aj , . . . , An ) = d(A1 , A2 , . . . , C C, . . . , An ) = d(A1 , A2 , . . . , C, . . . , An ) d(A1 , A2 , . . . , C, . . . , An ) = 0. (b) Put Aj = B, Aj +1 = C. By the alternating property,
0 = d(A1 , A2 , . . . , B + C, B + C, . . . , An ) = d(A1 , A2 , . . . , B, B + C, . . . , An ) + d(A1 , A2 , . . . , C, B + C, . . . , An ) = d(A1 , A2 , . . . , B, C, . . . , An ) + d(A1 , A2 , . . . , C, B, . . . , An )

Hence d(A1 , A2 , . . . , B, C, . . . , An ) = d(A1 , A2 , . . . , C, B, . . . , An ). (c) follows from (b) by 2(j i) 1 transpositions.

4 / 17

Uniqueness of the determinant function

Lemma: Suppose f is a multilinear alternating function on n n matrices and f (e1 , e2 , . . . , en ) = 0. Then f is identically zero. Proof: Let A = [aij ] be an n n matrix with columns A1 , . . . , An . For j = 1, . . . , n, write Aj as Aj = a1j e1 + a2j e2 + + anj en . Since f is multilinear, we obtain f (A1 , . . . , An ) =
h

ah(1)1 ah(2)2 ah(n)n f (eh(1) , eh(2) , . . . , eh(n) ),

where the sum is over all functions h : {1, 2, . . . , n} {1, 2, . . . , n}. Since f is alternating, we obtain f (A1 , . . . , An ) =
h

ah(1)1 ah(2)2 ah(n)n f (eh(1) , eh(2) , . . . , eh(n) ),

where the sum is now over all bijections h : {1, 2, . . . , n} {1, 2, . . . , n}.
5 / 17

Uniqueness of the determinant function

By using part (c) of the lemma above, we see that f (A1 , . . . , An ) =
h

ah(1)1 ah(2)2 ah(n)n f (e1 , e2 , . . . , en ),

where the sum is over all bijections h : {1, 2, . . . , n} {1, 2, . . . , n}. Since f (e1 , . . . , en ) = 0, we obtain f (A) = 0. Theorem: Let f be an alternating multilinear function of order n, and let d be a determinant function of order n. Then for every n n matrix A = [A1 , A2 , . . . , An ], we have f (A1 , A2 , . . . , An ) = d(A1 , A2 , . . . , An )f (e1 , e2 , . . . , en ). In particular, if f is also a determinant function, then f (A1 , A2 , . . . , An ) = d(A1 , A2 , . . . , An ).
6 / 17

Proof of the uniqueness of a determinant function

Proof: Consider the function g given by
g (A1 , A2 , . . . , An ) = f (A1 , A2 , . . . , An ) d(A1 , A2 , . . . , An )f (e1 , e2 , . . . , en ).

Since f, d are alternating and multilinear, so is g. Since g (e1 , e2 , . . . , en ) = 0, the result follows from the previous lemma. Notation: We shall denote the determinant of A by det A. Setting det[a] = a gives the existence for n = 1. Assume that we have proved the existence of the determinant function det of order n 1. The determinant function of order n can be dened in terms of the determinant function of order n 1 as follows. Let Aij denote the (n 1) (n 1) matrix obtained from an nn matrix A by deleting the ith row and j th column of A for i, j = 1, . . . , n.
7 / 17

Existence of the determinant function

Theorem The function f on nn matrices A = [aij ] given by f (A) := a11 det A11 a12 det A12 + + (1)n+1 a1n det A1n is multilinear, alternating, and normalized, and hence it is the determinant function of order n. (Expansion by the rst row) Proof: Fix j {1, . . . , n 1}. Suppose the columns Aj and Aj +1 of A are equal. Then A1i has two equal columns except when i = j, j + 1. By induction det(A1i ) = 0 if i = j and i = j + 1. Thus f (A) = (1)j +1 a1j det(A1j ) + (1)j +2 a1(j +1) det(A1(j +1) ). Since Aj = Aj +1 , we have a1j = a1j +1 and A1j = A1j +1 . Thus f (A) = 0. Therefore the function f is alternating. Let A = I = [e1 , e2 , . . . , en ]. Then a12 = = a1n = 0, so that f (A) = 1 det(A11 ) = det(e1 , e2 , . . . , en1 ) = 1. The multilinear property of the function f can be similarly proved.
8 / 17

Determinants of triangular and elementary matrices

In a similar manner, we have the expansion by row k = 2, . . . n: k+j a det A det A = n kj kj for A = [aij ]. j =1 (1) Theorem: (i) Let U be an upper triangular or a lower triangular matrix. Then det U is equal to the product of the diagonal entries of U . (ii) Let E := I + Eij be an elementary matrix of the type I, where R and i = j . Then det E = 1. (iii) Let E := I + Eij + Eji Eii Ejj be an elementary matrix of type II, where i = j . Then det E = 1. (iv) Let E := I + ( 1)Eii be an elementary matrix of type III, where 0 = R. Then det E = . Proof: (i) Let U = [uij ] be upper triangular. Arguing as in the lemma used for proving the uniqueness of the determinant function, we obtain det U = h uh(1)1 uh(2)2 uh(n)n , where the sum is over all bijections h : {1, 2, . . . , n} {1, 2, . . . , n}.
9 / 17

Multiplicity of det
Since U is upper triangular the only choice of the bijection h yeilding a nonzero term is the identity bijection (and this gives a plus sign). (ii) follows from part (i). (iii) E is obtained from the identity matrix by exchanging columns i and j . The result follows since the determinant function is alternating. (iv) follows form part (i). Theorem: (Multiplicativity of det) For any nn matrices A and B , det(AB ) = det A det B. Proof: Since AB = A[B1 , . . . , Bn ] = [AB1 , . . . , ABn ], we need prove det(AB1 , AB2 . . . , ABn ) = det A det B. Keep A xed, and for an arbitrary nn matrix B , dene f (B1 , B2 , . . . , Bn ) = det(AB1 , AB2 , . . . , ABn ).
10 / 17

det(AB ) = det A det B

The function f is alternating, since f (B1 , . . . , Bi , . . . , Bi , . . . , Bn ) = det(AB1 , . . . , ABi , . . . , ABi , . . . , ABn ) = 0. To show that the function f is also multilinear, let C and D be n1 column vectors, and let , R. Then
f (B1 , . . . , C + D, . . . , Bn ) = = det(AB1 , . . . , A(C + D), . . . , ABn ) det(AB1 , . . . , AC + AD, . . . , ABn )

= det(AB1 , . . . , AC, . . . , ABn ) + det(AB1 , . . . , AD, . . . , ABn )

= f (B1 , . . . , C, . . . Bn ) + f (B1 , . . . , D, . . . , Bn )

Since f is alternating and multilinear, an earlier theorem gives f (B1 , B2 , . . . , Bn ) = det(B1 , . . . , Bn )f (e1 , e2 , . . . , en ). But f (e1 , e2 , . . . , en ) = det(Ae1 , . . . , Aen ) = det(A1 , . . . , An ) = det A. Thus det(AB1 , AB2 , . . . , ABn ) = det(B1 , . . . , Bn ) det A = det A det B , as desired.
11 / 17

Determinant and invertibility

Lemma: (i) Let A be a square matrix. Then A is invertible if and only if det A = 0, and in that case, 1 det A1 = . det A (ii) Suppose A and B are square matrices such that AB = I . Then A and B are invertible and B = A1 . Proof: (i) Let A be invertible. Then (det A)(det A1 ) = det AA1 = det I = 1, so that det A = 0. Suppose A is not invertible. Then there is a nonzero column vector x such that Ax = 0. Hence a linear combination of the column vectors of A having at least one nonzero coecient is equal to the zero column vector. Thus some column of A is a linear combination of other columns of A. It now follows from the multilinearity and alternating properties of the detetminant function that det A = 0. (ii) det A det B = det(AB ) = det(I ) = 1. So det A = 0 and A is invertible. Now B = (A1 A)B = A1 (AB ) = A1 is also invertible.
12 / 17

Determinant of the transpose of a matrix

Theorem: For any n n matrix A, det At = det A. Proof: If A is not invertible, then At is also not invertible, and det A = 0 = det At . Let now A be invertible. Then there are elementary matrices E1 , . . . , Ek such that A = E1 Ek , and so det A = det E1 det Ek . Now the transpose of an elementary matrix is also an elementary matrix of the same type, and has the same determinant. Hence t E t ) = det E t det E t = det E det E = det A. det At = det(Ek 1 k 1 1 k Corollary: The determinant is a multilinear, alternating and normalized function of the rows of a square matrix, and so for A = [aij ], we have the following expansion by column k , where k = 1, . . . , n:
n

det A =
i=1

(1)k+i aik det Aik .

13 / 17

Computation of determinant by Gauss-Jordan Method

Let A be an n n matrix. Reduce A to its row canonical form (rcf) U by elementary row operations (E.R.O.s) of type I (denoted by Ri + Rj with i = j ), of type II (denoted by Ri Rj ), and of type III (denoted by i Ri with 0 = i ), that is, by premultiplying A by elementary matrices of types I, II and III. To obtain U from A, if r row interchanges Ri Rj have been used and if 1 , . . . , p are the nonzero scalars used for the row operations i Ri , then det U = (1)r 1 p det A. This follows from the result about the determinants of elementary matrices and from the multiplicativity of the determinant function proved earlier. If u11 , . . . , unn are the diagonal entries of U , then det U = u11 unn . Hence det A = (1)r (1 p )1 u11 unn .
14 / 17

The cofactor matrix

Denition: Let A = [aij ] be an n n matrix. For i, j = 1, . . . , n, the cofactor of the entry aij , denoted by cof aij , is dened by cof aij = (1)i+j det Aij . If n = 1, we dene cof a11 = 1. The cofactor matrix of A, denoted by cof A, is the matrix cof A = [cof aij ]. Expansion of det A by row k and by column k can be written as
n n

aik cof aik = det A =

i=1 j =1

akj cof akj

for each k = 1, . . . , n.

Theorem: For any square matrix A, A(cof A)t = (det A)I = (cof A)t A. 1 In particular, if det A is nonzero, then A1 = (cof A)t . det A
15 / 17

Matrix inverse using its cofactor matrix

Proof: The (i, j )th entry of C := (cof A)t A is cij =
n k=1 akj

cof aki .

If i = j, then cij is the expansion of det A by the j th column of A. Let i = j . Consider the matrix B = [bij ] obtained by replacing the ith column of A by the j th column of A. Then B has a repeated column. Expanding det B by its ith column, cij = n k=1 bki cof bki = det B = 0. The other equation A(cof A)t = (det A)I is proved similarly. Cramers Rule: Consider the system a11 a12 a1n a21 a22 a2n . . . . . . . . . . . . an1 an2 ann

x1 x2 . . . xn

b1 b2 . . . bn

of n linear equations in the n unknowns x1 , x2 , . . . , xn .

16 / 17

Cramers Rule
Suppose the coecient matrix A = [aij ] is invertible. Let Cj be the matrix obtained from A by replacing its j th column by the right side b = [b1 , b2 , . . . , bn ]t . Then the unique solution of the system Ax = b is given by det Cj for j = 1, . . . n. xj = det A Proof: Let A1 , . . . , An be the columns of A. If x = [x1 , . . . , xn ]t satises Ax = b, then b = x1 A1 + x2 A2 + + xn An . Since the determinant function is multilinear and alternating, we obtain det Cj = det[A1 , . . . , Aj 1 , b, Aj +1 , . . . , An ] = xj det A. Hence xj = det Cj for j = 1, 2, . . . , n. det A
17 / 17