6 negative semi definite matrices
play

6. Negative (Semi-)Definite Matrices Daisuke Oyama Mathematics II - PowerPoint PPT Presentation

6. Negative (Semi-)Definite Matrices Daisuke Oyama Mathematics II May 1, 2020 Some Facts from Linear Algebra Let M R N N . M is said to be nonsingular if there exists A R N N such that MA = AM = I . In this case, A is called


  1. 6. Negative (Semi-)Definite Matrices Daisuke Oyama Mathematics II May 1, 2020

  2. Some Facts from Linear Algebra Let M ∈ R N × N . ▶ M is said to be nonsingular if there exists A ∈ R N × N such that MA = AM = I . In this case, A is called the inverse matrix of M and denoted by M − 1 . ▶ The following are equivalent: ▶ M is nonsingular. ▶ rank M = N . ▶ | M | ̸ = 0 . ▶ { z ∈ R N | Mz = 0 } = { 0 } . ▶ 0 is not a characteristic root of M . 1 / 34

  3. Some Facts from Linear Algebra Let M ∈ R N × N . ▶ The equation in λ , | M − λI | = 0 , is called the characteristic equation of M . ▶ The characteristic equation of M has N solutions in C (counted with multiplicity). ▶ The solutions to the characteristic equation of M are called the characteristic roots of M . ▶ If λ 1 , . . . , λ N are the characteristic roots of M , then | M | = ∏ N n =1 λ n . ▶ If M is nonsingular and λ 1 , . . . , λ N are its characteristic roots, then λ − 1 1 , . . . , λ − 1 N are the characteristic roots of M − 1 . 2 / 34

  4. Some Facts from Linear Algebra Let M ∈ R N × N . ▶ λ ∈ C is an eigenvalue of M if there exists z ∈ C N with z ̸ = 0 such that Mz = λz. In this case, z is called an eigenvector of M that corresponds (or belongs) to λ . ▶ λ is an eigenvalue of M if and only if it is a characteristic root of M . 3 / 34

  5. Some Facts from Linear Algebra Let M ∈ R N × N be a symmetric matrix. ▶ All the eigenvalues (hence characteristic roots) of M are real. ▶ Each eigenvalue of M has real eigenvectors. ▶ ∃ U ∈ R N × N orthogonal (i.e., U T U = UU T = I ) such that   λ 1 O ... U T MU = (= diag( λ 1 , . . . , λ N )) ,     O λ N where λ 1 , . . . , λ N ∈ R are the eigenvalues of M . ▶ If M is nonsingular, then M − 1 is symmetric. 4 / 34

  6. Negative (Semi-)Definite Matrices Definition 6.1 ▶ M ∈ R N × N is negative semi-definite if z · Mz ≤ 0 for all z ∈ R N . ▶ M ∈ R N × N is negative definite if z · Mz < 0 for all z ∈ R N with z ̸ = 0 . ▶ M ∈ R N × N is positive definite ( positive semi-definite , resp.) if − M is negative definite (negative semi-definite, resp.). 5 / 34

  7. Remark ▶ In many math books, negative definiteness is defined only for symmetric matrices, or for quadratic forms ∑ N i,j =1 a ij z i z j . (Any quadratic form is written as z · Mz for some symmetric M .) ▶ Sometimes, matrices (not necessarily symmetric) that are negative definite in our sense are called negative quasi-definite. 6 / 34

  8. Example: Negative (Semi-)Definiteness of Jacobi Matrices Let X ⊂ R N be a non-empty open convex set. Suppose that f : X → R N is differentiable. 1. ( y − x ) · ( f ( y ) − f ( x )) ≤ 0 for all x, y ∈ X if and only if Df ( x ) is negative semi-definite for all x ∈ X . 2. If Df ( x ) is negative definite for all x ∈ X , then ( y − x ) · ( f ( y ) − f ( x )) < 0 for all x, y ∈ X , x ̸ = y . ▶ For N = 1 , “ ( y − x ) · ( f ( y ) − f ( x )) ≤ 0 ( < 0 ) for all x, y ∈ X ” implies that f is nonincreasing (strictly decreasing). ▶ Cf. Proposition 5.20. 7 / 34

  9. Example: Negative (Semi-)Definiteness of Hesse Matrices Let X ⊂ R N be a non-empty open convex set. Suppose that f : X → R N is differentiable and ∇ f is differentiable. 1. f is concave if and only if D 2 f ( x ) is negative semi-definite for all x ∈ X . 2. If D 2 f ( x ) is negative definite for all x ∈ X , then f is strictly concave. ▶ Proposition 5.21. 8 / 34

  10. Characterizations of Negative (Semi-)Definiteness Proposition 6.1 Let M ∈ R N × N . 1. M is negative definite ⇒ M + M T is negative definite. ⇐ 2. Suppose that M is symmetric. M is negative definite ⇐ ⇒ all the characteristic roots of M are negative. 3. M is negative definite ⇒ M is nonsingular and M − 1 is negative definite. = 9 / 34

  11. Proof 1. For any z ∈ R N , z T ( M + M T ) z = 2 z T Mz . 2. Since M = U T diag( λ 1 , . . . , λ N ) U for some U orthogonal (hence nonsingular), z T Mz < 0 for all z ∈ R N \ { 0 } ⇒ ( Uz ) T diag( λ 1 , . . . , λ N )( Uz ) < 0 for all z ∈ R N \ { 0 } ⇐ n =1 λ n ( y n ) 2 = y T diag( λ 1 , . . . , λ N ) y < 0 ⇒ ∑ N ⇐ for all y ∈ { Uz | z ∈ R N \ { 0 }} = R N \ { 0 } ⇐ ⇒ λ 1 , . . . , λ N < 0 . 10 / 34

  12. 3. Suppose Mz = 0 . Then z T ( M + M T ) z = 0 . Thus, if M is negative definite (and so is M + M T ), we must have z = 0 . Take any z ∈ R N , z ̸ = 0 . Let x = M − 1 z ( ̸ = 0 ). Then z = Mx . Then we have z T M − 1 z = ( Mx ) T M − 1 ( Mx ) = x T M T x = x T Mx < 0 . 11 / 34

  13. Characterizations of Negative (Semi-)Definiteness Proposition 6.2 Let M ∈ R N × N be symmetric. 1. M is negative semi-definite ⇒ ∃ B ∈ R N × N such that M = − B T B . ⇐ 2. M is negative definite ⇒ ∃ B ∈ R N × N nonsingular such that M = − B T B . ⇐ 12 / 34

  14. Proof ▶ The “if” part: Suppose that M = − B T B . Then for any z ∈ R N , z T Mz = − z T B T Bz = −∥ Bz ∥ 2 ≤ 0 . ▶ If B is nonsingular and z ̸ = 0 , then ∥ Bz ∥ ̸ = 0 . 13 / 34

  15. Proof ▶ The “only if” part:   λ 1 O ... Since M is symmetric, we have U T MU =     O λ N for some U orthogonal (hence nonsingular). If M is negative semi-definite, then λ 1 , . . . , λ N ≤ 0 . √− λ 1   O ...  U T . ▶ Let B =   √− λ N  O   λ 1 O ...  U T = M . Then − B T B = U    O λ N ▶ If M is negative definite, then λ 1 , . . . , λ N < 0 , so that B is nonsingular. 14 / 34

  16. Characterizations of Negative (Semi-)Definiteness Proposition 6.3 Let M ∈ R N × N be symmetric. M is negative definite ⇒ ( − 1) r | r M r | > 0 for all r = 1 , . . . , N . ⇐ ▶ r M r ∈ R r × r is the r × r submatrix of M obtained by deleting the last N − r columns and rows of M , which is called the leading principal submatrix of order r of M . ▶ | r M r | is called the leading principal minor of order r of M . ▶ r M ∈ R r × N will denote the r × N submatrix of M obtained by deleting the last N − r rows of M . 15 / 34

  17. Proof ▶ The “only if” part: If M is negative definite, then r M r is negative definite and its characteristic roots λ 1 , . . . , λ r are all negative, and thus, ( − 1) r | r M r | = ( − λ 1 ) × · · · × ( − λ r ) > 0 . ▶ The “if” part: by induction: Trivial for N = 1 . ▶ Assume that the assertion holds for N − 1 . Suppose that ( − 1) r | r M r | > 0 for all r = 1 , . . . , N . Then L = N − 1 M N − 1 is negative definite by the induction hypothesis. Hence, ▶ L is nonsingular, and B T ˜ ▶ L = − ˜ B for some nonsingular ˜ B . 16 / 34

  18. Proof ( L ) b ▶ Write M = , where b ∈ R ( N − 1) × 1 . b T a NN L − 1 b ( I N − 1 ) ▶ Let U = . 0 T 1 ( L ) 0 Then one can verify that M = U T U , 0 T c where c = a NN − b T L − 1 b . ▶ Thus, | M | = c | L | . But by assumption, ( − 1) N | M | > 0 and ( − 1) N − 1 | L | > 0 , so that c < 0 . ( ˜ ) B 0 ▶ Let B = √− c U , which is nonsingular, 0 T B T ˜ where L = − ˜ B . Then M = − B T B . Hence, M is negative definite. 17 / 34

  19. Note ▶ “ ( − 1) r | r M r | ≥ 0 for all r = 1 , . . . , N ” does not imply that M is negative semi-definite. ▶ For example, ( 0 ) 0 M = 0 1 satisfies this condition ( ( − 1) | 1 M 1 | = ( − 1) 2 | M | = 0 ), but is not negative semi-definite. 18 / 34

  20. Characterizations of Negative (Semi-)Definiteness Proposition 6.4 Let M ∈ R N × N . ▶ Suppose that M is symmetric. M is negative semi-definite ⇒ ( − 1) r | r M π ⇐ r | ≥ 0 for all r = 1 , . . . , N and for all permutations π of { 1 , . . . , N } . ▶ If (not necessarily symmetric) M is negative semi-definite, then ( − 1) r | r M π r | ≥ 0 for all r = 1 , . . . , N and for all permutations π of { 1 , . . . , N } . 19 / 34

  21. Application to Concave Functions ∂ 2 f Denote f ij ( x ) = ∂x j ∂x i ( x ) . ▶ f ( x 1 , x 2 ) is strictly concave = D 2 f ( x 1 , x 2 ) is negative definite ⇐ ∀ ( x 1 , x 2 ) � � f 11 f 12 ⇒ ( − 1) f 11 > 0 and ( − 1) 2 � � ⇐ � > 0 ∀ ( x 1 , x 2 ) � � f 21 f 22 � ⇒ f 11 < 0 and f 11 f 22 − ( f 12 ) 2 > 0 ⇐ ∀ ( x 1 , x 2 ) ▶ f ( x 1 , x 2 ) is concave ⇒ D 2 f ( x 1 , x 2 ) is negative semi-definite ⇐ ∀ ( x 1 , x 2 ) � � f 11 f 12 ⇒ ( − 1) f 11 ≥ 0 , ( − 1) 2 � � ⇐ � ≥ 0 , � � f 21 f 22 � � � f 22 f 21 ( − 1) f 22 ≥ 0 , and ( − 1) 2 � � � ≥ 0 ∀ ( x 1 , x 2 ) � � f 12 f 11 � ⇒ f 11 ≤ 0 , f 22 ≤ 0 , and f 11 f 22 − ( f 12 ) 2 ≥ 0 ⇐ ∀ ( x 1 , x 2 ) 20 / 34

  22. Characterizations of Negative (Semi-)Definiteness Proposition 6.5 Let M ∈ R N × N be symmetric, and B ∈ R N × S with S ≤ N be such that rank B = S . Let W = { z ∈ R N | B T z = 0 } . ▶ M is negative definite on W if and only if � � r M r r B ( − 1) r � � � > 0 � ( r B ) T � 0 � for all r = S + 1 , . . . , N . ▶ M is negative semi-definite on W if and only if � r M π r B π � ( − 1) r � r � � ≥ 0 � ( r B π ) T � 0 � for all r = S + 1 , . . . , N and for all permutations π of { 1 , . . . , N } . 21 / 34

Recommend


More recommend