# Professor John Nachbar Econ 4111 February 19, 2008

Deﬁnite Matrices
1 Basic Deﬁnition

An N × N symmetric matrix A is positive deﬁnite iﬀ for any v = 0, v Av > 0. For example, if a b A= b c then the statement is that for any v = (v1 , v2 ) = 0, v Av = [ v1 v2 ] = [ v1 v2 ] a b b c av1 + bv2 bv1 + cv2 v1 v2

2 2 = av1 + 2bv1 v2 + cv2

>0
2 + 2bv v + cv 2 is called a quadratic form.) An N × N symmetric (The expression av1 1 2 2 matrix A is negative deﬁnite iﬀ −A is positive deﬁnite. The deﬁnition of a positive semideﬁnite matrix relaxes > to ≥, and similarly for negative semideﬁniteness. If N = 1 then A is just a number and a number is positive deﬁnite iﬀ it is positive. For N > 1 the condition of being positive deﬁnite is somewhat subtle. For example, the matrix 1 3 A= 3 1

is not positive deﬁnite. If v = (1, −1) then v Av = −4. Loosely, a matrix is positive deﬁnite iﬀ (a) it has a diagonal that is positive and (b) oﬀ diagonal terms are not too large in absolute value relative to the terms on the diagonal. I won’t formalize this assertion but it should be plausible given the example. The canonical positive deﬁnite matrix is the identity matrix, where all the oﬀ diagonal terms are zero.

2

Inverses of deﬁnite matrices.

Theorem 1. If A is positive deﬁnite then A is invertible and A−1 is positive deﬁnite.

1

3 Checking deﬁniteness. If A is positive deﬁnite then v Av > 0 for all v = 0. It remains to show that A−1 is positive deﬁnite. if A is negative deﬁnite then A−1 exists and is negative deﬁnite. If A is symmetric and invertible then (A−1 ) = (A )−1 = A−1 . where the second equality comes from the fact that A−1 is symmetric and the last inequality follows from the fact that A is positive deﬁnite. hence Av = 0 for all v = 0. Thus. with the sign negative iﬀ the number of columns in the submatrix of A is odd. 2 . hence A is invertible.Proof. To see this. Likewise. Similarly. (A−1 ) = (A )−1 . a31 a32 a33 then the leading principal submatrices are a11 . Consider v A−1 v for any v = 0. a leading principal submatrix of A is a submatrix formed by deleting all but the ﬁrst n rows and columns. Since A is negative deﬁnite iﬀ −A is positive deﬁnite. Theorem 2. It is messy but easy to implement on a computer. hence A−1 is symmetric. . if a11 a12 a13   A =  a21 a22 a23  . Theorem 2 implies that a matrix is negative deﬁnite iﬀ the principal minors alternate in sign. (A−1 ) A = I . hence A has full rank. Omitted. There is a mechanical check for deﬁniteness of a matrix. A matrix is positive deﬁnite iﬀ all of the leading principal minors are positive.   a11 a12 a21 a22 and A itself. note that [A (A−1 ) ] = A−1 A = I . For any invertible matrix A. Proof. Given an N × N matrix A. Hence A (A−1 ) = I = I . A leading principal minor is the determinant of a leading principal submatrix. and since multiplying an n × n matrix by -1 multiplies the determinant by (−1)n . Then v A−1 v = v A−1 AA−1 v = (A−1 v ) A(A−1 v ) > 0.

4 Deﬁniteness on subspaces. if K = 1. which will be the typical case in our applications. The border preserving leading principal minors are the determinants of these submatrices. while retaining the associated components of the border. Omitted. If the columns of B are linearly independent then T has dimension N − K . Theorem 3. with the sign positive iﬀ K is even. Let B by an N × K matrix.        a11 a12 a13 b11 b12 a21 a22 a23 b21 b22 a31 a32 a33 b31 b32 b11 b21 b31 0 0 b12 b22 b32 0 0     . plane. 3 . A border preserving leading principal submatrix of this bordered matrix is the matrix formed by deleting all but the ﬁrst n ≥ K rows and columns of A. For example. T is a vector subspace in RN : T is a line. or hyperplane though the origin. There is a mechanical check for this that parallels that for (full) deﬁniteness. Assume that B has full rank (K ). all the border preserving principal minors must be negative. T = {v ∈ RN : B v = 0}. Proof. then A is positive deﬁnite on T . If v Av > 0 for all v ∈ T . if the bordered matrix is. A symmetric matrix A is positive deﬁnite on T iﬀ the border preserving leading principal minors of the associated bordered matrix all have the same sign.   then the border preserving leading principal submatrices are      a11 a12 b11 b12 a21 a22 b21 b22 b11 b21 0 0 b12 b22 0 0      and the original bordered matrix. with K ≤ N . Let T = {v ∈ RN : B v = 0}. Thus. Consider the (N + K ) × (N + K ) bordered matrix. Deﬁniteness on a subspace shows up in the second order conditions for constrained optimization. v = 0. Consider the set That is. Let B be N × K where K ≤ N . T consists of all vectors that are at right angles to the columns of B . A B B 0 .

The sign of the minor for the next smallest matrix. is positive. if K = 1. In particular.For negative deﬁnite matrices. 4 . except that the minors alternate in sign. and so on. with the sign negative iﬀ the number of columns in the submatrix of A is odd. then the smallest submatrix has one column and the sign of the corresponding minor is negative. with two columns. one gets the same result.