Definition. Given n vectors v1 , · · · , vn ∈ F n , the determinant det(v1 , . . . , vn ) is a scalar in F such that (1) Multilinearity: det is linear in each argument det(v1 , . . . , vi−1 , u + w, vi , . . . , vn ) = det(v1 , . . . , vi−1 , u, vi , . . . , vn ) +det(v1 , . . . , vi−1 , w, vi , . . . , vn ) det(v1 , . . . , vi−1 , cu, vi , . . . , vn ) = c det(v1 , . . . , vi−1 , u, vi , . . . , vn ) (2) Alternating: det changes sign when you switch neighboring arguments det(v1 , . . . , vi , vi+1 , . . . , vn ) = −det(v1 , . . . , vi+1 , vi , . . . , vn ) (3) Normalization: if {e1 , . . . , en } is the standard basis then det(e1 , . . . , en ) = 1 If A is an n × n-matrix with columns v1 , . . . , vn , we define det(A) = det(v1 , . . . , vn ). Proposition. The determinant is uniquely defined. a b Example. det = ad − bc. We need to check that the axioms for the c d determinant hold: (1) Multilinearity det a1 + a2 c1 + c2 b d = (a1 + a2 )d − b(c1 + c2 ) a1 c1 b d a2 c2 b d = (a1 d − bc1 ) + (a2 d − bc2 ) = det + det ; a b 1 + b2 det = a(d1 + d2 ) − (b1 + b2 )c c d1 + d2 a b1 a b2 = (ad1 − b1 c) + (ad2 − b2 c) = det + det ; c d1 c d2 sa b a b det = (sa)d − b(sc) = s(ad − bc) = s det ; sc d c d a sb a b det = a(sd) − (sb)c = s(ad − bc) = s det . c sd c d (2) Alternating c d a b det = cb − ad = −(ad − bc) = −det a b c d (3) Normalization det 1 0 0 1 =1·1−0·0=1 Proposition. Given n × n-matrices A, B, det(AB) = det(A)det(B) Proof. Suppose B has columns v1 , . . . , vn . Then AB has columns Av1 , . . . , Avn . Thus, det(AB)/det(A) = det(Av1 , . . . , Avn )/det(A) Since the properties of the determinant uniquely define it, we need to check that they hold for det(Av1 , . . . , Avn )/det(A). 1 2 (1) Multilinearity: This follows from linearity of matrix multiplication. det(Av1 , . . . , A(u + w), . . . , Avn )/det(A) = det(Av1 , . . . , Au + Aw, . . . , Avn )/det(A) = (det(Av1 , . . . , Au, . . . , Avn ) + det(Av1 , . . . , Aw, . . . , Avn ))/det(A) = det(Av1 , . . . , Au, . . . , Avn )/det(A) + det(Av1 , . . . , Aw, . . . , Avn )/det(A); det(Av1 , . . . , A(cv), . . . , Avn )/det(A) = det(Av1 , . . . , c(Av), . . . , Avn )/det(A) = c det(Av1 , . . . , Av, . . . , Avn )/det(A). (2) Alternating: det(Av1 , . . . , Avi , Avi+1 , . . . , Avn )/det(A) = −det(Av1 , . . . , Avi+1 , Avi , . . . , Avn )/det(A) (3) Normalization: The columns of A are Ae1 , . . . , Aen , so by definition det(A) = det(Ae1 , . . . , Aen ) and thus det(Ae1 , . . . , Aen )/det(A) = 1. Thus, det(AB)/det(A) = det(B) or det(AB) = det(A)det(B) Corollary. If A is invertible then det(A−1 ) = 1 det(A) . Proof. By normalization the determinant of the identity matrix is det(I) = 1, so 1 = det(I) = det(AA−1 ) = det(A)det(A−1 ) ⇒ det(A−1 ) = 1/det(A) Proposition. An n × n-matrix A is invertible iff det(A) 6= 0. Proof. If the matrix is invertible then det(A) = 1/det(A−1 ) 6= 0 On the other hand, recall that a matrix is invertible iff its columns {v1 , . . . , vn } are linearly independent. So if A is not invertible so that {v1 , . . . , vn } is linearlly dependent, then by a homework problem, there is an i such that vi ∈ span{v1 , . . . , vi−1 , vi+1 , . . . , vn }. Write vi = a1 v1 + · · · + ai−1 vi−1 + ai+1 vi+1 + · · · + an vn By multilinearity we know that determinant is linear in the ith entry so a1 det(v1 , . . . , vi−1 , v1 , vi+1 , . . . , vn ) .. . det(v1 , . . . , vn ) = +ai−1 det(v1 , . . . , vi−1 , vi−1 , vi+1 , . . . , vn ) +ai+1 det(v1 , . . . , vi−1 , vi+1 , vi+1 , . . . , vn ) .. . +an det(v1 , . . . , vi−1 , vn , vi+1 , . . . , vn ) And since each term is a determinant with repeated entries, each term equals 0 and thus det(A) = det(v1 , . . . , vn ) = 0. Definition. The characteristic polynomial of a square matrix is the polynomial pA (λ) = det(A − λI). 3 Example. A = 1 1 1 0 . pA (λ) = det 1 1 1 0 −λ 1 0 0 1 1 1 λ 0 1−λ 1 = det − = det = (1−λ)(−λ)−1·1 = λ2 −λ−1 1 0 0 λ 1 −λ We know that the λ is an eigenvalue for A iff A − λI is not invertible. Thus eigenvalues are roots of the characteristic polynomial. For example A has eigenvalues which we can find by the quadratic formula. √ 1± 5 λ= 2