A = - 2>,<, ü>,. " i

Similar documents
Banach Journal of Mathematical Analysis ISSN: (electronic)

Elementary linear algebra

MATH 423 Linear Algebra II Lecture 33: Diagonalization of normal operators.

arxiv: v1 [math.fa] 17 Dec 2016

Throughout these notes we assume V, W are finite dimensional inner product spaces over C.

Mathematik-Bericht 2010/5. An eigenvalue estimate and its application to non-selfadjoint Jacobi and Schrödinger operators

Math 108b: Notes on the Spectral Theorem

Estimates for norms of resolvents and an application to the perturbation

Salah Mecheri. Communicated by S. L. Troyanski

THE ALTERNATIVE DUNFORD-PETTIS PROPERTY FOR SUBSPACES OF THE COMPACT OPERATORS

EXPLICIT UPPER BOUNDS FOR THE SPECTRAL DISTANCE OF TWO TRACE CLASS OPERATORS

Quantum Computing Lecture 2. Review of Linear Algebra

AN EXTREMUM PROPERTY OF SUMS OF EIGENVALUES' HELMUT WIELANDT

BOUNDED MATRICES AND SOME ALGEBRAS OF OPERATORS DONAL P. O'DONOVAN

REPRESENTATION THEORY WEEK 7

ON THE SIMILARITY OF CENTERED OPERATORS TO CONTRACTIONS. Srdjan Petrović

Math 408 Advanced Linear Algebra

Chapter 2. Linear Algebra. rather simple and learning them will eventually allow us to explain the strange results of

Upper triangular forms for some classes of infinite dimensional operators

Singular Value Inequalities for Real and Imaginary Parts of Matrices

Perturbations of functions of diagonalizable matrices

e jk :=< e k, e j >, e[t ] e = E 1 ( e [T ] e ) h E.

Remarks on Fuglede-Putnam Theorem for Normal Operators Modulo the Hilbert-Schmidt Class

AN IMPROVED ESTIMATE IN THE METHOD OF FREEZING ROBERT E. VINOGRAD

THE HYPEROCTANT PROPERTY IN ORTHOMODULAR AC-LATTICES'

A NOTE ON A BASIS PROBLEM

AN ELEMENTARY PROOF OF THE SPECTRAL RADIUS FORMULA FOR MATRICES

SOME INEQUALITIES FOR ENTIRE FUNCTIONS SABUROU SAITOH. For any entire functions <p(z) and \p(z) on C with finite norm. 1 // /(z) 2exp(- z 2) *<*j

THE SPECTRAL DIAMETER IN BANACH ALGEBRAS

Free products of topological groups

ADJOINT FOR OPERATORS IN BANACH SPACES

QUASI-UNIFORMLY POSITIVE OPERATORS IN KREIN SPACE. Denitizable operators in Krein spaces have spectral properties similar to those

TRIANGULAR TRUNCATION AND NORMAL LIMITS OF NILPOTENT OPERATORS

j=1 x j p, if 1 p <, x i ξ : x i < ξ} 0 as p.

Quantum Physics II (8.05) Fall 2002 Assignment 3

Derivations on Commutative Normed Algebras.

ANOTHER CLASS OF INVERTIBLE OPERATORS WITHOUT SQUARE ROOTS

Lecture notes: Applied linear algebra Part 1. Version 2

arxiv: v2 [math.oa] 19 Sep 2010

SOME HILBERT SPACE EXTREMAL PROBLEMS

Convexity of the Joint Numerical Range

THE NORM OF A HERMITIAN ELEMENT IN A BANACH ALGEBRA

Chapter 6 Inner product spaces

2 8 P. ERDŐS, A. MEIR, V. T. SÓS AND P. TURÁN [March It follows thus from (2.4) that (2.5) Ak(F, G) =def 11m 1,,,,(F, G) exists for every fixed k. By

FORMAL TAYLOR SERIES AND COMPLEMENTARY INVARIANT SUBSPACES

Review of Linear Algebra

Operators with numerical range in a closed halfplane

COMPLETENESS THEOREM FOR THE DISSIPATIVE STURM-LIOUVILLE OPERATOR ON BOUNDED TIME SCALES. Hüseyin Tuna

Linear algebra 2. Yoav Zemel. March 1, 2012

Section 17: The Eigenvalue Problem

Diagonalization by a unitary similarity transformation

Kernel Method: Data Analysis with Positive Definite Kernels

CS 246 Review of Linear Algebra 01/17/19

1 Matrices and vector spaces

2 K cos nkx + K si" nkx) (1.1)

Maximizing the numerical radii of matrices by permuting their entries

Finite and infinite dimensional generalizations of Klyachko theorem. Shmuel Friedland. August 15, 1999

THE SOLUTION OF 3y2 ± 2" = x3

On John type ellipsoids

x 3y 2z = 6 1.2) 2x 4y 3z = 8 3x + 6y + 8z = 5 x + 3y 2z + 5t = 4 1.5) 2x + 8y z + 9t = 9 3x + 5y 12z + 17t = 7

Singular value inequality and graph energy change

SINGULAR VALUE INEQUALITIES FOR COMPACT OPERATORS

Sufficient conditions for functions to form Riesz bases in L 2 and applications to nonlinear boundary-value problems

SINGULAR INTEGRAL OPERATORS ON THE UNIT CIRCLE 1

Symmetric and self-adjoint matrices

Fundamentals of Engineering Analysis (650163)

The lattice of closed ideals in the Banach algebra of operators on a certain dual Banach space

A NOTE ON QUASI ISOMETRIES II. S.M. Patel Sardar Patel University, India

Linear maps preserving AN -operators

^-,({«}))=ic^if<iic/n^ir=iic/, where e(n) is the sequence defined by e(n\m) = 8^, (the Kronecker delta).

1 Smooth manifolds and Lie groups

Compression, Matrix Range and Completely Positive Map

AFFINE MAPPINGS OF INVERTIBLE OPERATORS. Lawrence A. Harris and Richard V. Kadison

j=1 [We will show that the triangle inequality holds for each p-norm in Chapter 3 Section 6.] The 1-norm is A F = tr(a H A).

We continue our study of the Pythagorean Theorem begun in ref. 1. The numbering of results and remarks in ref. 1 will

arxiv: v1 [math.fa] 1 Oct 2015

dominant positive-normal. In view of (1), it is very natural to study the properties of positivenormal

2. reverse inequalities via specht ratio To study the Golden-Thompson inequality, Ando-Hiai in [1] developed the following log-majorizationes:

ON JAMES' QUASI-REFLEXIVE BANACH SPACE

Operator-valued extensions of matrix-norm inequalities

ON A CLASS OF COVARIANCE OPERATORS U =

ON THE NUMERICAL RANGE OF AN OPERATOR

LINEAR ALGEBRA BOOT CAMP WEEK 4: THE SPECTRAL THEOREM

TWO CHARACTERIZATIONS OF POWER COMPACT OPERATORS

LOCALLY ^-CLOSED SPACES AND RIM /»-CLOSED SPACES

NORMS ON SPACE OF MATRICES

THE ANGLE OF AN OPERATOR AND POSITIVE OPERATOR PRODUCTS 1

ON THE OPTIMAL ASYMPTOTIC EIGENVALUE BEHAVIOR OF WEAKLY SINGULAR INTEGRAL OPERATORS FERNANDO COBOS, SVANTE JANSON, AND THOMAS KÜHN

Clarkson Inequalities With Several Operators

NON POSITIVELY CURVED METRIC IN THE SPACE OF POSITIVE DEFINITE INFINITE MATRICES

Assignment 10. Arfken Show that Stirling s formula is an asymptotic expansion. The remainder term is. B 2n 2n(2n 1) x1 2n.

Part IA. Vectors and Matrices. Year

A CLASS OF SCHUR MULTIPLIERS ON SOME QUASI-BANACH SPACES OF INFINITE MATRICES

COMPOSITION OF LINEAR FRACTIONAL TRANSFORMATIONS IN TERMS OF TAIL SEQUENCES

CONVEXITY OF INTEGRAL MEANS OF SUBHARMONIC FUNCTIONS

ON MATRIX VALUED SQUARE INTEGRABLE POSITIVE DEFINITE FUNCTIONS

Kotoro Tanahashi and Atsushi Uchiyama

ON FREE ABELIAN /-GROUPS1

A PRIMER ON SESQUILINEAR FORMS

A DECOMPOSITION THEOREM FOR FRAMES AND THE FEICHTINGER CONJECTURE

Transcription:

proceedings of the american mathematical society Volume 81, Number 3, March 1981 SPECTRA OF OPERATORS WITH FIXED IMAGINARY PARTS ANDRZEJ POKRZYWA Abstract. The aim of this paper is to obtain the best bound for the distance between the eigenvalues of a Hermitian matrix B and the real parts of eigenvalues of a matrix B + A, where A is also Hermitian, in the terms of eigenvalues of A. A similar problem in infinite-dimensional Hubert space is also considered. This paper was inspired by the papers of Kahan [4], [5] and Gohberg [1]. The obtained results may be regarded as the generalizations of some results of these authors. A solution of the problem of Kahan, which consists of computing the best constant in the inequality \\Z Z*\\ < Kn\\Z + Z*\\ for all n X n matrices with real spectrum, is obtained (Corollary 2). Notations. Let H denote a complex Hubert space with the norm and the scalar product <, >. L(H) denotes the algebra of all bounded linear operators acting in H. For an A G L(H), o(a) denotes the spectrum of A. For a compact A G L(H), sx, s2,... denote the eigenvalues of VAA*, repeated according to multiplicity and arranged in decreasing order. Finite-dimensional case. In this section we assume that H is «-dimensional and that A = A* is an operator in H with eigenvalues X, > A2 > > A,. Lemma 1. There exists B = B* G L(H) such that o(a + ib) c R and Proof. Let {t}}" be an orthonormal basis of H. Define operator S by the formula n S - -<, ex)e + 2 <, <?,>, -! 2 The vectors vk = 2"_,exp(((2A: l)/n)jm)ej are mutually orthogonal eigenvectors of S of norm n. When the basis {e,} is suitably chosen then l " A = - 2>,<, ü>,. " i Received by the editors May 25, 1979. AMS (MOS) subject classifications (1970). Primary 15A42, 15A60, 47A55; Secondary 15A18, 15A57. 359 1981 American Mathematical Society 0002-9939/81/0000-0103/$02.SO

360 andrzej pokrzywa We shall show that the operator n B = i 2 sign( -jxaej, ek)(, e }ek kj=l satisfies the thesis of our lemma. Since A = 2J««i<j4&, ek} <, ej~)ek we see that the matrix of the operator n A + ib = 2 <Aek, ek}(, ek}ek + 2 {Aep ek}(, e )ek k~\ k<j is triangular in the basis {e,}. Therefore the eigenvalues of A + ib are Xk(A + ib) = (Aek, ek} = (1/«)27A,.. Hence a(a + ib) c R. Note that (Aej, ek} - / - 2 \<*y>»,>. ** ) = - S A, exp( (* ""/>"*) Now it is easy to verify that if we define the numbers >y and the operator T by the relations " I 2s 1 \ "~1 bj = -i 2 A, exp- Jm\, T - <, e,>e + 2 <, et+1>«t, i=i \ «/ _i then 5 = (1/h)2"~11,T-/. Since the numbers A^T) = exp(2 tori/«) are the eigenvalues of T, the eigenvalues of B are Hence M*) - -^ 2,(A,(r)y-1 2 \ ctg iii ^i!.. «7=1 " *=1 2w PH = max A,(5) - i 2 K ctg ^-U k n,_, in Lemma 2. If B = B* C L(H) and a(a + ib) c R f/ien ll^ll < 7 2 \ ctg 2* ~! g. Proof. By the theorem on triangular matrix form there exists an orthonormal basis {e,}" of H such that ((A + ib)ek, e } = 0 = ((A - ib)ejt ek) for k <j. This implies that /'( te,, ek} = <yle,, e > and i'<äe\, e,> = -(Aek, e ) for A: </. Since {A + ibek, ek) c o(a + ib) c R and A, B are self adjoint, (Bek, ek} =0. Hence B = 2,,*<5fc> eft><, «,>«* = -2,-* sign(./ - k)(aej, ek><, ey>ek. Setting <, e,)e,- =, we may write fi = /2sign(A:-y) ^^, (1) j, k Since B = B* there is a unit eigenvector / of B such that \\B\\ =!< /,/>. If we set <,/>/= F then tr BF = <5/,/>. Using properties of trace and (1) we see

SPECTRA OF OPERATORS 361 that (± \\B\\ =) tr BF = tr(/2 sign(a: - j)ekaejf\ V j,k I = tri A 2 i sign(â; - j)e}fek\ =txag, (2) where we have set i 2y^ sign(a: j)ejfek = G. G is a selfadjoint operator. Let to, > «2 > > w be its eigenvalues. It is shown in [1] that w,-=-«+,_,, j = 1, 2,...,«, (3) and that if / *= 0 for all j then for j < [/i/2] 2*_, arg(wy. + i\\ej\\2) = (2y - l)w/2. This means that ^^^^ arctgillvllv«,) < arc tgj(^ WHV^-arctg^, since the function arc tg is concave in the interval [0, oo]. Since tangent is an increasing function in [0, tt/2) we obtain the inequality tg((2/ l)w/(2n)) < \/{na,), equivalent to uj < - ctg V- i it, j = 1, 2,..., n/2. (4) n ~ 2n By continuity of eigenvalues (4) holds also when E}f = 0 for some j. It follows from (3) that tr G = 0. Let fi > -\; then Sj{A +,i) - A,. + p. (5) Let x be the normalized eigenvector of G, Gxj = UjX,. Since G is selfadjoint {xj}" is an orthonormal basis for H. Thus, using Abel transformation, we may write tr AG = tr(^ + /i)g - 2 <(-* + m)gx,, *,-> = 2 «/(^ + /*) *;> */> n -1 j n = 2 («, - w,+ i) 2 <(^ + io*ar> **>+«2 «z4 + /*)**> **> (6) y=l Ä-1 *-l The Ky-Fan lemma [2, Lemma II.4.1] and (5) imply that 2 <(A + mk> **> < 2 %( +»O- 2 (\ + m)- *=i *:=i *-i Note also that if j = n then in the above inequality we have in fact the equality without the modulus. Hence n-1 7 n trag< 2 («>- S+i) 2 (A* + ft) + a 2 (A* + Ñ y-l *=1 *-l = 2 "A + ft) = 2 tyo,-. 7=1 7-1

362 ANDRZEJ POKRZYWA Writing the just obtained inequality with -A instead of A we obtain by (3) the inequality -tr AG < 2,<o,.(-An+1_,) = 2yAyw,. This with (2), (3) and (4) shows that I"/2] * = \trag\ < 2 V; = 2 (Ay - K+i-jh The lemma is proved. 7 7 = 1 1 ["4?] y v 2/-1 1, 2/-1 < - 2 (\ - W>* ^-* = - 2\ ctg J w. Theorem 1. Suppose that B is a selfadjoint operator in H. Let {&}*, {/%}" 6e /«e eigenvalues of B, B + ia, respectively, arranged in such a way that ßj > ßj+x, Re fij > Re ij+, forj = 1,2,...,«1. Then /Î,.- Re M> < - 2 K ctg -^^r-». " j=i z/1 Proof. Following Kahan [5] and identifying the operators with matrices we may assume that B + ia is an upper triangular matrix and that B + ia = D + iz, where D is a real diagonal matrix, Z is an upper triangular matrix with real spectrum. Hence the numbers Re a are eigenvalues of D. Since B D = i{z - Z*)/2 it follows from Weyl's inequality that \ß} - Re júj < \\B - D\\ = Im Z. Since Re Z = (Z + Z*)/2 = A the thesis follows from Lemma 2. For a subset F of the complex plane let Re F = (Re À; X G F). The following corollaries follow easily from the obtained results. Corollary 1. max 1 " 2s 1 dist(a(5), Re a(b + ia)) = - 2 À, ctg -^-it. b = b'sl(h) " nx s 2n (dist(, ) denotes the Hausdorff distance of sets.) Corollary 2. Kn = max{ Z - Z* / Z + Z* ; Z G L(#), ct(z) c R, Z # 0} 2 ["/2] 2,-1 "»? Ctg-Jn-^ K = max{dist o(fi), Re o(b + ic); B = B*,C = C*, C < 1}. Using the inequalities one may see that r(2s+l)*/(2n) IT ( 2s + I 2s - 1 \ i-)v(2n) ctgx <^(ctg~^r*+ ctg-5 } w 2s 1 / (s + i)»/". ~Ctg - < I' CtgX (1<Í<[«/21), «2«^-i)w/ L J/ 1 W 2,. 7T ^ ^ 2 7T 2..7T 7 Ctg ^ ~ ^ ln Sm T < *" < «Ctg ^ ~ ~Z m Sm 7 ft Zft 77 Zft ft ft TT ft and that A^/ln n» 2/tt.

SPECTRA OF OPERATORS 363 Infinite-dimensional case. Let H be a separable infinite-dimensional Hubert space, and let A be a selfadjoint compact operator in H. Xf, XT, j = 1, 2,..., denote the positive eigenvalues of A and -A, respectively, repeated according to multiplicity and arranged in decreasing order. If there are only «positive (negative) eigenvalues of A we set A,+ = 0 (A/ = 0) for j > n. Theorem 2. 2 sup dist(a( ), Re o(b + ia)) = - 2 (V + K)/ (2s ~ 0- í = i'e (H) ff i-1 Proof. It follows from the Macaev theorem [3, Theorem III.4.2], or from Corollary 1 that "sup" is not less than the right-hand side. Thus it suffices to prove that if B = B* G L(H) then dist(a(ä), Re o(b + ia)) < - 2 (V + K)/ (2s ~ 0- (7) * *-i It follows from the Weyl-von Neumann theorem [6, Theorem X.2.1] that there exists a compact selfadjoint operator K such that the operator B + K has a pure point spectrum. Then there exists a sequence {Pn)\x' of orthogonal projections in H converging strongly to the identity operator such that Pn is n dimensional and commutes with B + K. Define the operators Bn, Cn by the formulas B - (1 - Pn)B{\ - Pn) + PnBPn, Cn = Bn + ipnapn. Since Bn - B = P (K - KPJ + (K - P K)P it follows from the compactness of K that \\Bn - B\\ -h» 0. Since A is compact too we see that \\C - (B + ia)\\ -+ 0. The operators B, Cn are compact perturbations of B; therefore their essential spectra are identical. These facts and the perturbation theorems [6, Chapter IV, 3] imply that dist(o(b), o(bn)) -» 0, dist(a(5 + ia), o(c )) -> 0. (8) Note further that a(bn) = a((l - Pn)B\0_P)H) u a(pnb\ph), and o(c ) = a((l - P )B\0_Pm)H) U a(pn(b + ia)\pnh). Consequentíy, dist(o(b ), Re o(cn)) < áist(a(p B\PmH), Re a(p (B + ia)\p >H)) 1 [n/2] 2v - 1 2 < 7 2 (a/ + x;)ctg ^ ^ < 2 (V + a;)/ (2, - l), (9) since they'th positive (negative) eigenvalue of P A\PH, if it exists, is not greater (less) than X + (-X/), and ctg((2s - l)/2ft)w < 2n/(2s l)7r. The desired inequality (7) now follows from (8) and (9). References 1. I. C. Gohberg, On connections between Hermitian components of nilpotent matrices and on an integral of triangular truncation, Bui. Akad. Stiince RSS Moldoven. 1 (1963), 27-37. MR 35 #2168. (Russian) 2. I. C. Gohberg and M. G. Krein, Introduction to the theory of linear nonselfadjoint operators in Hilbert space, "Nauka", Moscow, 1965; English transi., Transi. Math. Monos., vol. 18, Amer. Math. Soc, Providence, R. I, 1969. MR36 #3137; MR39 #7447.

364 ANDRZEJ POKRZYWA 3. _, Theory of Volterra operators and its applications, "Nauka", Moscow, 1967; English transi., Transi. Math. Monos., vol. 24, Amer. Math. Soc, Providence, R. L, 1970. MR 36 #2007; MR 41 #9041. 4. W. Kahan, Every n X n matrix Z with real spectrum satisfies \\Z Z*\\ < \\Z + Z*\\ (log2 n + 0.038), Proc Amer. Math. Soc 39 (1973), 235-241. MR 47 # 1833. 5. _, Spectra of nearly Hermitian matrices, Proc. Amer. Math. Soc. 48 (1975), 11-17. MR 51 #5627. 6. T. Kato, Perturbation theory for linear operators, Die Grundlehren der math. Wissenschaften, Bd. 132, Springer-Verlag, New York, 1966. MR 34 #3324. Institute of Mathematics, Poush Academy of Sciences, 00-950, Warsaw, P. O. B. 137, Poland