fisica

© All Rights Reserved

7 views

fisica

© All Rights Reserved

- Mat Lab Cheat Sheet
- Course Information
- PCA.xls
- Shear Building - seismic
- Paper - On the Dynamic Analysis of Rotor-bearing Systems
- Matrix Eigenvalues and Eigenvectors
- Mazri - Vol 5 No 1 (2)
- Akira Takahashi and Shaul Mukamel- Anharmonic oscillator modeling of nonlinear susceptibilities and its application to conjugated polymers
- Eigen-04
- MAT423_2013_01
- Digital Resonators
- T2D CN M1 V2
- Matrix Method for Diff Equ
- Paper 29-Recognition of Facial Expression Using Eigenvector Based Distributed Features and Euclidean Distance Based Decision Making Technique
- Matrix Properties
- statika-noteeng (1)
- s1-7
- Modal Analysis - Solved example
- ReconfigurableDeltaOperatorEigenstructureAssignmentForATaillessAircraft_Sobel2009.pdf
- Two Examples on Eigenvector and Eigenvalue

You are on page 1of 4

www.elsevier.com/locate/laa

Wilfred Kaplan

Department of Mathematics, University of Michigan, 3868 East Hall, Ann Arbor, MI 48109-1109, USA

Received 23 October 1999; accepted 4 April 2000

Submitted by R.A. Brualdi

Abstract

The paper presents necessary and sufficient conditions that a symmetric matrix be coposi-

tive or strictly copositive. The conditions are given in terms of the eigenvalues and eigenvec-

tors of the principal submatrices of the given matrix. 2000 Elsevier Science Inc. All rights

reserved.

AMS classification: 15A18; 15A48; 15A63; 49N10; 65K99

x > 0. It is termed strictly copositive if it is copositive and equality holds only for

x = 0. There is an extensive literature on such matrices; see for example [15,710]

and references cited therein.

In this paper, we give necessary and sufficient conditions for these two classes of

matrices. The conditions can be derived fairly easily from Theorems 3.1 and 3.2 of

[2]. However, the statement of the conditions and the very short direct proofs should

be of interest.

In the proofs we denote by Q(x) the quadratic form x T Ax, by kxk the Euclidean

norm of x, and by Rnpos the set of all x in Rn such that x > 0, x =/ 0.

tive if and only if every principal submatrix B of A has no eigenvector v > 0 with

associated eigenvalue 6 0.

E-mail address: wilkap@math.1sa.umich.edu (W. Kaplan).

0024-3795/00/$ - see front matter 2000 Elsevier Science Inc. All rights reserved.

PII: S 0 0 2 4 - 3 7 9 5 ( 0 0 ) 0 0 1 3 8 - 5

204 W. Kaplan / Linear Algebra and its Applications 313 (2000) 203206

Proof. Let the principal submatrices of A have the property stated in the theorem

and let Q(x0 ) 6 0 for some x0 in Rnpos . Some, but not all, components of x0 may be

0. For proper numbering, we can assume that x0 = (a1 , . . . , am , 0, . . . , 0)T , where

1 6 m 6 n, and ai > 0 for i = 1, . . . , m. The vector x0 cannot be the unique one

at which Q has a nonpositive value; among all such vectors, we can choose one at

which m has its least value, and we assume that our x0 has this property.

We consider the case m > 1. Let y be an arbitrary vector of Rm and let

Q0 (y) = Q(y1 , . . . , ym , 0, . . . , 0).

Let y0 = (a1 , . . . , am )T . We consider the function Q0 (y) restricted to the set E:

kyk = 1, y > 0. By multiplying y0 by a positive scalar, we can assure that y0 is

in E. The vector y0 must then be in the relative interior of E, since a vector in

the relative boundary of E would have less than m nonzero components. Thus, the

function Q0 (y) on E has positive values on the relative boundary of E and has a

nonpositive value at one point in the relative interior. Accordingly, the function must

have a negative or zero absolute minimum at some vector v in the relative interior, so

that v > 0. But the vector v would be an eigenvector of a principal submatrix of A

with a negative or a zero associated eigenvalue. By hypothesis, there can be no such

vector. Accordingly, Q(x) > 0 in Rnpos .

For the case m = 1, the principal submatrix [a11 ] would have the eigenvector

v = (1) with a nonpositive eigenvalue and the same conclusion follows.

Conversely, let Q(x) > 0 in Rnpos and let some principal submatrix B of A have

an eigenvector v > 0 with a nonpositive associated eigenvalue. We can assume that

B is obtained from A by deleting the rows and columns following the mth, and write

v = (a1 , . . . , am )T , x0 = (a1 , . . . , am , 0, . . . ., 0)T . Then Q(x0 ) = v T Bv 6 0, con-

trary to hypothesis. Therefore, no principal submatrix of A can have an eigenvector

v > 0 with a negative or a zero associated eigenvalue.

Thus, Theorem 1 is proved.

only if every principal submatrix B of A has no eigenvector v > 0 with associated

eigenvalue < 0.

inequalities.

Q(x) = 2x12 + 2x22 + 2x32 2x1x2 2x1 x3 + 6x2 x3 .

The principal matrices of order 1 are the diagonal elements, all equal to 2, and these

matrices have only the positive eigenvalue 2. Of those of order 2, two have the ei-

genvalues 1 and 3, and the remaining one has the eigenvalues 5 and 1, with eigen-

vectors c(1, 1)T associated to the eigenvalue 1. The principal submatrix of order

3 is A itself, which has the eigenvalues 5.5616, 1.4384 and 1. The eigenvectors

W. Kaplan / Linear Algebra and its Applications 313 (2000) 203206 205

associated with the eigenvalue 1 are multiples of (0, 1, 1)T. In no case is there a

nonpositive eigenvalue with associated eigenvector v > 0. Accordingly, A is strictly

copositive.

For this example, it is not difficult to verify directly that Q(x) > 0 for x > 0,

x=/ 0. The matrix A is not positive definite and some entries are negative; if it were

positive definite or all entries were positive, then of course it would have to be strictly

copositive.

Geometric interpretation: The idea behind the proofs arose from simple geometric

reasoning. We illustrate this by considering the case n = 3 and a matrix A, which has

one negative eigenvalue and two positive eigenvalues. After a rotation of axes, the

set on which Q(x) 6 0 can be represented by an inequality

ax12 + bx22 cx32 6 0,

where a, b, and c are positive. The set is bounded by the two nappes of an elliptical

cone; we denote by E1 , and E2 the solids bounded by the two nappes. If we reverse

the rotation, one of the solids, say E1 , may intersect the first octant at an interior point

P of E1 ; this occurs precisely when A is not copositive. If this occurs, but no such

point P exists on a coordinate plane, then Q(x) is nonnegative on the boundary of

the first octant but takes on negative values at some interior points of the first octant;

this leads to a negative eigenvalue of A with positive eigenvector; if such a point

P exists on a coordinate plane but not on a coordinate axis, then similar reasoning

leads to a negative eigenvalue with a positive eigenvector of a principal submatrix

of A having order 2; if such a point P exists on a coordinate axis, then we have a

similar statement for order 1.

Remark. Copositive matrices occur in optimization theory. See for example [3,

Chapter 3], [4,5], [6, p. 133] and [9, Chapter 2]. The tests of Theorems 1 and 2 are

of little practical value for such applications when the matrix has a large order, since

a matrix of order n has 2n 1 principal submatrices.

References

[1] L.-E. Anderson, G. Chang, T. Elfving, Criteria for copositive matrices using simplices and barycen-

tric coordinates, Linear Algebra Appl. 220 (1995) 930.

[2] R.W. Cottle, G.J. Habetler, C.E. Lemke, On classes of copositive matrices, Linear Algebra Appl.

3 (1970) 295310.

[3] R.W. Cottle, J.-S. Pang, R.E. Stone, The Linear Complementarity Problem, Academic Press, New

York, 1992.

[4] G. Danninger, Role of copositivity in optimality criteria for nonconvex optimization problems,

J. Optim. Theory Appl. 75 (1998) 535558.

[5] G. Danninger, I.M. Bomze, Using copositivity for global optimality criteria in concave quadratic

programming problems, Math. Programming 62 (1993) 575580.

206 W. Kaplan / Linear Algebra and its Applications 313 (2000) 203206

[6] W. Kaplan, Maxima and Minima with Applications, Wiley, New York, 1999.

[7] T.S. Motzkin, Copositive quadratic forms, National Bureau of Standards Report. 1818, 1952,

pp. 1122.

[8] T.S. Motzkin, Quadratic forms positive for nonnegative variables not all zero, Notices Amer. Math.

Soc. 12 (1965) 224.

[9] K. Murty, Linear Complementarity, Linear and Nonlinear Programming, Heldermann, Berlin, 1988.

[10] H. Vliaho, Criteria for copositive matrices, Linear Algebra Appl. 81 (1986) 1934.

- Mat Lab Cheat SheetUploaded byRod Owens
- Course InformationUploaded bysachinp2006
- PCA.xlsUploaded byViacheslav Baykov
- Shear Building - seismicUploaded byArut MV
- Paper - On the Dynamic Analysis of Rotor-bearing SystemsUploaded byVILLANUEVA_DANIEL2064
- Matrix Eigenvalues and EigenvectorsUploaded byrodwellhead
- Mazri - Vol 5 No 1 (2)Uploaded byMazri Yaakob
- Akira Takahashi and Shaul Mukamel- Anharmonic oscillator modeling of nonlinear susceptibilities and its application to conjugated polymersUploaded byGlade680
- Eigen-04Uploaded byMangisi Haryanto Parapat
- MAT423_2013_01Uploaded byHaire Kahfi Maa Takaful
- Digital ResonatorsUploaded byKashif Mahmood
- T2D CN M1 V2Uploaded bySeba Pñaloza Dz
- Matrix Method for Diff EquUploaded bySarvesh Deepak Koyande
- Paper 29-Recognition of Facial Expression Using Eigenvector Based Distributed Features and Euclidean Distance Based Decision Making TechniqueUploaded byEditor IJACSA
- Matrix PropertiesUploaded bymgpadalkar
- statika-noteeng (1)Uploaded byRavi Khandelwal
- s1-7Uploaded byrichk98
- Modal Analysis - Solved exampleUploaded byMansoor
- ReconfigurableDeltaOperatorEigenstructureAssignmentForATaillessAircraft_Sobel2009.pdfUploaded bycmpmarinho
- Two Examples on Eigenvector and EigenvalueUploaded byNo Al Prian
- 10.1007_s10470-008-9201-xUploaded byPramod Srinivasan
- mm-34Uploaded bySomasekhar Chowdary Kakarala
- optimazation Appendix AUploaded bybakhtyar21
- Euler WaveEQUploaded byThale1905
- 76057.pdfUploaded bysandeepuae
- SHREYAS CommunicationUploaded byPedro Sabino
- VI Sem. B.sc. Mathematics - Core Course - Number Theory and Linear AlgebraUploaded byMohan Rao
- L6Uploaded byrajpd28
- GPTutorial.pdfUploaded byVictor Rodriguez
- Lecture 32 Systems of Differential EquationsUploaded byJohnny Pinny

- The Firebombing of Tokyo in March 1945Uploaded byRoberto Carreira
- 7 Occena vs MarquezUploaded byToni Calsado
- Dmitri Warren de Klerk - 2653786 - ThesisUploaded byrgv2417arora
- Investor Presentation July2008Uploaded bymixedbag
- Baldwin Brothers, Inc., in No. 15691, J. Robert Baldwin and Ruth D. Baldwin, in No. 15692, Tenth Street Building Corporation of Erie, in No. 15693, Baldwin Builders, Inc., in No. 15694, Arthur W. Baldwin and Nancy F. Baldwin, in No. 15695 v. Commissioner of Internal Revenue, 361 F.2d 668, 3rd Cir. (1966)Uploaded byScribd Government Docs
- 8W Ch8 PS Stock Valuation StudentsUploaded byMuhammet Dalgıç
- Voice Input _ Glass _ Google DevelopersUploaded bySurcahyono
- Catalina Press ReleaseUploaded bysleeplessnight
- cfdUploaded byVivek Harris
- TD-W8151N User GuideUploaded bySteveWillman
- 2013 47 Summer Wiring Matters Medical LocationsUploaded bypavlone
- Improving Compressed Air-SourcebookUploaded byJonathan FuckYouScribd
- Promotions _ Kenny Rogers ROASTERS.pdfUploaded byBernard Tay
- Industrial Automation Cable.pdfUploaded bymelgarcia829
- tlcUploaded byAftab Hussain Siddiqui
- The Bull's Eye - April 2010Uploaded bydbhsbullseye
- Review of CRDA Draft Development PlanUploaded byYogesh J. Parekh
- Wincor Nixdorf ConfigProgManual_englishUploaded byalfi56k
- Road Project 2011-12Uploaded byanutam.cool
- Contacts LogUploaded byValéria Rosa Brascher
- ICSA Firewall TestingUploaded byGeorge Zecheru
- 0452_s11_ms_11Uploaded byAthul Tomy
- fm 200 tank size.pdfUploaded byporkovan
- Pavarotti El Grande (1)Uploaded bygeorgesculigia
- Comparison of Headspace and Direct Injection Techniques mUploaded byDr. John Jack Driscoll
- syllabus winter 17Uploaded byapi-350693115
- Iti openingUploaded byKumudRanjan
- WPS specUploaded byshifa
- Recruitment & SelectionUploaded bylakshmanlakhs
- Service Diagnosis Split Pocket BookUploaded byHugo Machado