ebook img

Lyapunov functions for periodic matrix-valued Jacobi operators PDF

0.3 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Lyapunov functions for periodic matrix-valued Jacobi operators

Lyapunov fun tions for periodi matrix-valued Ja obi operators 7 0 0 ∗ † 2 Evgeny Korotyaev Anton Kutsenko n a February 2, 2008 J 6 1 ] Abstra t P We onsider periodi matrix-valued Ja obi operators. The spe trum of this operator S is absolutely ontinuous and onsists of intervals separated by gaps. We de(cid:28)ne the . h Lyapunov fun tion, whi h is analyti on an asso iated Riemann surfa e. On ea h sheet t a the Lyapunov fun tion has the standard properties of the Lyapunov fun tion for the m s alar ase. We show that this fun tion has (real or omplex) bran h points, whi h we [ all resonan es. We prove that there exist two types of gaps: i) stable gaps, i.e., the 1 endpoints are periodi and anti-periodi eigenvalues, ii) unstable (resonan e) gaps, i.e., v the endpoints are resonan es (real bran h points). We show that some spe tral data 8 4 determine the spe trum ( ounting multipli ity) of the Ja obi operator. 4 1 0 7 1 Introdu tion and main results 0 ℓ2(Z)m / h Consider self-adjoint matrix-valued Ja obi operators J a ting on and are given by mat (Jy)n = anyn+1+bnyn+a⊤n−1yn−1, n ∈ Z, yn ∈ Cm,y = (yn)n∈Z ∈ ℓ2(Z)m,m > 1, (1.1) a ,b = b⊤ m m deta = 0 v: wnherZe n n n are p-periodi sequen es oσf(th)e × real matri es and n 6 for all i ∈ . It is well known that the spe trum J of J is absolutely ontinuous and onsists X [λ+ ,λ−],λ+ < λ− 6 λ+,n = 1,...,N < of non-degenerated intervals n−1 n n−1 n n ∞. These intervals r γ = (λ−,λ+) > 0 a n n n are separated by the gaps with the length . Introdu e the fundamental m m ϕ = (ϕn(z))n∈Z,ϑ = (ϑn(z))n∈Z × matrix-valued solutions of the equation a y +b y +a⊤ y = zy , ϕ ϑ 0, ϕ ϑ I , (z,n) C Z, n n+1 n n n−1 n−1 n 0 ≡ 1 ≡ 1 ≡ 0 ≡ m ∈ × (1.2) I m m 2m 2m M m p where is the identity × matrix. We de(cid:28)ne the monodromy × matrix and T ,n > 1 n the tra e by ϑ (z) ϕ (z) M (z) = p p , T (z) = TrMn(z). p (cid:18) ϑ (z) ϕ (z) (cid:19) n p (1.3) p+1 p+1 ∗ Institut fu(cid:4)r Mathematik, Humboldt Universit(cid:4)at zu Berlin, Rudower Chaussee 25, 12489, Berlin, Ger- many, e-mail: evgenymath.hu-berlin.de † Department of Mathemati s of Sankt-Petersburg State University, Russia e-mail: ku enkoarambler.u 1 M D Introdu e the modi(cid:28)ed monodromy matrix and the determinant by a 0 M = P M P−1, P = 0 , D(z,τ) = det(M(z) τI ), τ,z C. 0 p 0 0 (cid:18) 0 I (cid:19) − 2m ∈ (1.4) m τ = τ (z),j N M(z) N = 1,,..,m j j 2m m Let ∈ be the eigenvalues of , where { }. An eigenvalue M(z) of is alled a multiplier. Theorem (Lyapunov-Poin ar(cid:1)e) i) The following identities hold 0 I M⊤JM = J, J = m , (cid:18) I 0 (cid:19) (1.5) m − D(z,τ) = τ2mD(z,τ−1), all z,τ C,τ = 0. ∈ 6 (1.6) τ(z) d > 1 z C z R τ−1(z) ii) Iτf(z) is a multiplier of multipli ityd for some ∈M(z)(,ozr C∈ ), then 2m (or ) isτ±a1(mz)u,ljtipliNer of muσlt(ipl)i =ity m. MzoreCov:erτ,(eza) h= 1 ∈ , has exa tly multipτli(ezr)s j ∈ m and J ∪j=z1{ C∈ | τj(z)| = 1}. τ′(z) = 0 iii) If is a simple multiplier for some ∈ and | | , then 6 . m = 1 It is well knoσw(n t)ha=t thpe[λs+pe ,tλru−m] of the sλ +ala<r (λ− 6 λ)+Ja< o.b..i <opeλr−at6orλJ+is a<bsλo−lutely λ o±ntinuous and J ∪1 n−1 n , where [λ0+ ,λ−1],n 1N p p−1 pγan=d n are 2-periodi eigenvalues. The intervals n−1 n ∈ p are separated by gaps n (λ−,λ−) γ > 0 γ γ = 0 n n of lengths | n| . If a gap n is degenerate, i.e. | n| , then the orresponding σ σ ∆(z) = ϕp+1(z)+ϑp(z) segments σn(, )n=+1 mλergeC. :Re∆ a(zll)tha[t t1h,e1]Lyapunov fun( tio1n)p−n∆(λ±) = 1 2 nand Nthe spe trum J { ∈ ∈ − }. Note that − n for all ∈ p. We re all well-known fa ts (see e.g. [P℄ or [KKu℄). F p Theorem 1.1. A real polynoamia>l 0,bis thRe,LnyapZunovFfu(nz) t=ionczfpo+r sOo(mzpe−s1) alarz -periodi n n Ja obico>pe0rator Fwi′t(hz n)u=m0b,ers( 1)p−jF(z∈) > 1 ∈ i(cid:27)j N zas< ..→. <∞z for j j p−1 1 p−1 some and − for all ∈ and for some . ∆ t t [ 1,1] c > 0 ∆ Remark. 1) Zeros of − for any (cid:28)xed ∈ − and a onstant determine . σ( ) = σ( ) , ∆ = ∆ ∆,∆ 2) J J forsomeJa obioperatorsJ J i(cid:27) , where arethe orresponding Lyapuano=v 1fu,nb et=ion0s. n Z ∆ = Te(z) e T e 3) If n n for all ∈ , then p 2 , where p is the Chebyshev polynomial. Before we des ribe the ontent of our paper, we brie(cid:29)y omment on ba kground litera- ture for matrix-valued Ja obi operators. Inverse spe tral theory for s alar periodi Ja obi operators is well understood, see a book [T℄ and papers [BGGK℄, [K℄, [KKu℄, [P℄, [vM℄ and referen es therein. The orrespondingtheoryforperiodi Ja obimatri eswithmatrix-valued oe(cid:30) ients is still largely a wide open (cid:28)eld. Some new parti ular results were re ently ob- tained by Gesztesy and oauthors [CG℄,[CGR℄,[GKM℄ and see referen es therein. Note that for (cid:28)nite Ja obi matri es with matrix-valued oe(cid:30) ients the omplete solution of the inverse problem was given re ently by Chelkak and Korotyaev [CK℄. M(z) D(τ,z) = 0 The eigenvalues of are the zeros of the equation . This is an algebrai τ 2m z equation in of degree , where thτe( zo)e,(cid:30)j ieNnts are polynomials of . It is well known (see j 2m e.g. Chapter 8, [Fo℄) that the roots ∈ onstitute one or several bran hes of one 2 C or several analyti fun tions that have only algebrai singularities in . Thus the number M(z) N z e of eigenvalues of is a onstant with the ex eption of some spe ial values of (see below the de(cid:28)niτtio(nz),ojf a rNesonan e). There is a (cid:28)niNte n=um2mber of su h points on the plane. j 2m e If the fun tions ∈ are all distin t, then . If some of them are identi al, N < 2m M(z) e then and is permanently degenerate. m = 1 If , then the Riemann surfa e for the multipliers has 2 sheets, but the Lyapunov m > 2 fun tion is entire. Similarly, in the ase it is mor∆e (ozn)v=eni1e(nτt(fzo)r+usτ−to1( zo)n),stjru tNthe Riemann surfa e for the Lyapunov fun tions given by j 2 j j ∈ m, m whi hσh(as) = shmeetsz(:se∆e t(hz)e eq[ua1t,io1n] (1.7))#. ANote that the Lyapunov-Poin ar(cid:1)e TAheorem gives J j=1{ j ∈ − }. Let be a number of elements of a set . S ∆ ,s = 1,..,p 6 m p s 0 s TheoreRm,1p.2.>T1hereexistanalyti fun tions onthe -sheetedRiemann surfa e s s having the folωlowingNpro,pseretieNs: , ω = N i) There exist disjoint subsets s ⊂ m ∈ p0 s m su h that all bran hes of ∆ ,s N ∆ (z) = 1(τ (z)+τ−1(z))S, j ω s ∈ p0 have the form j 2 j j ∈ s and satisfy e D(z,τ) p0 τ +τ−1 = Φ (z,ν), Φ (z,ν) = (ν ∆ (z)), ν = , τ = 0, s s j (2τ)m − 2 6 (1.7) Y1 jY∈ωs z,τ C Φ (z,ν) ν,z C s for any ∈ , where the fun tions are some polynomials with respe t to ∈ . ∆ = ∆ i ω ,j ω Φ = Φ ∆ = ∆ i j k s k s k s If for some ∆∈,j ∈N , then and . Y = (α,β) R ii) Let some bran h j ∈ m be real analyti on seome ienterval ⊂ and 1 < ∆ (z) < 1 z Y ∆′(z) = 0 z Y − j foρr,ρan,sy ∈N ,.#Tωhe>n 2 j 6 for ea h ∈ . iii) All fun tions s ∈ p0 s given by (1.8) are polynomials and N0 ρ = ρ , ρ (z) = (∆ (z) ∆ (z))2, z C, where ρ = 1, if #ω = 1. s s i j s s − ∈ (1.8) Y1 i<jY,i,j∈ωs (λ−,λ+) n n iv) Ea h endpo∆int,ojf aNgap is a periodρi (or anti-periodi )eigenvalue or a real bran h j m point oτf0s,ojmeN ∈ , whi h is a zero ofA(b=elo(awasu. .ah )p−o1ints are alled resonan es). v) Let j ∈ m be eigenvalues of a matrix p 1 2 p . Then following asymptoti s hold zp ∆j(z) = 2 (τj0 +O(z−m1 )), τj(z) = zpτj0 +O(z−m1 )), j ∈ Nm, (1.9) ρs(z) = zp|ωs|(|ωs|−1)(cs +O(z−m1 )), cs = 2|ωs|(1−|ωs|) (τk0 −τj0)2, s ∈ Np0 (1.10) Y k<j,k,j∈ωs z τ0 = τ0 j,k ω ,j = k c = 0 ρ (z) = as | | → ∞ Moreover, if j 6 k for all ∈ s 6 , then s 6 and s z|ωs|(|ωs|−1)(c +O(z−1)) z s as | | → ∞. z z ρ 0 0 De(cid:28)nition. The number is a resonan e of J, if is a zero of given by (1.8). 3 a = diag a ,a ,...,a , b = diag b ,b ,...,b , n Z n 1,n 2,n m,n n 1,n 2,n m,n Example. Leat > 0,b { R,(j,n) N }Zand { = m } for all ∈ and for some j,n j,n ∈ ℓ2(∈Z) m × . Then the operator J ⊕1 Jj, where Jj is a s alar Ja obi operator, a ting in and is given by ( jy)n = aj,nyn+1 +aj,n−1yn−1 +bj,nyn, n Z, y = (yn)n∈Z ℓ2(Z). J ∈ ∈ ∆ ∆ j j j In this ase ea h is the standard Lyapunov fun tion for J , and the properties of is ρ = 1 ω = j j well known, see Theorem 1.1 and [T℄. Thus , sin e { }. Below we show the following identity (see (3.2)) mp D(z,τ) = cτm (z λ (τ)), z,τ C, τ = 0, c = ( 1)mdetA , j p − ∈ 6 − (1.11) Yj=1 λ (τ) D(z,τ) = 0 τ Λ(τ) = j wλhe(rτe),n Nare ,zeτrosCof the polynomialσ( ) = mp σfor (cid:28)xed σ. =Deλ(cid:28)n(Se1)t,hSe1s=et τ C : { n ∈ mp} ∈ . Then we obtain J j=1 j, where j j { ∈ τ = 1 N0 = 0,.S..,m m > 0 | | } is the bounded lose set. Ler m { ∆} f,ojr N . We des ribe minimum j m spe tral data whi h determine all Lyapunov fun tions ∈ for some Ja obi operator. κ R,j N0 cosκ = cosκ j = n TheorΛem=1Λ.3(e.iκL0e)t k ∈ Λ ∈Λ(meiκsaj)t,is#fyΛ = (mn 6 j)p+1j,fjor aNll 6 . 0 j j m i) Let andlet Λ⊂,j N0 − ∈ forDso(m,e)Ja obi operator J. Then∆th,ej speN tral data j ∈ σm(, d)etermine the polynomial · · , all Lyapunov j m fun tions ∈ and the spe trum J ( ounted a ording to its multipli ity). Λ Λ #Λ = #Λ 1 1 1 1 1 ii) Let ⊂ and let − . Then there exist in(cid:28)nitely many Ja obi operators, Λ Λ Λ j = 2,...m having tehe same spe tral deata 0, 1, j, , but di(cid:27)erent determinants. eikκj =e1 k N λ (eiκj),n N n mp Example. Note that if m = 2for somκe =∈0,κ, t=heπn Λ ∈ are k-periodi 0 1 0 eigenvalues. Consider the ase . Let . Then is a set of all periodi Λ #Λ = 2p,#Λ = p+1 1 0 1 eigenvaluesaΛnd= λisasetofsλomeσa(nti)-periodi eigenvaluessu h that κ = π . Let the set 2 { }, where ∈ ΛJ ,Λis ,aΛsome four-periodi eigenvalued,eit.e(.M, (2z) 2.τIDu)e to Theorem 1.3, thee spe traledata 0 1 2 determine the polynomial p − m , σ( ) ∆ ,∆ 1 2 the spe trum J and the Lyapunov fun tions . p > 3 mp λs(τ),s N Tτ heCorem 1.4. i) Let . Then ea h sum Pj=1 j ∈ p−1 does not depend on ∈ . In parti ular, the following identities hold mp p mp p λ (τ) = Trb , λ2(τ) = Tr(b2 +2a a⊤), all τ C. n n n n n n ∈ (1.12) Xn=1 Xn=1 Xn=1 Xn=1 ii) The following estimate is ful(cid:28)lled: pm λ2(τ) > 2pm(detA )p2m, n p (1.13) Xn=1 where the identityκholdRs,tjrueNi(cid:27) bn ≡ 0, aconas⊤nκ==(dceotsAκp)p2mIm fojr=alnl n ∈ Np. j m n j iλii) Le(teniκujm) =ber2scos 1∈(κ +∈2πn) satisfy(j,n,k)6 N0 Nfo0r all N6 .bTh=en0,thaeae⊤ig=enIvalues nm+k p j for all ∈ m × p−1 × p i(cid:27) n n n m for n N p a = I all ∈ pm and n=1 n m. Q 4 We des ribe a priori estimates = max a (j,k) b (j,k) a = a (j,k) , ∞ n n n n Proposition1.5. LetkJk {| |, | |}, wherethematri es { } b = b (j,k) n n { }. Then the following estimates hold 6 max λ+ , λ− = 6 (4m 1) , kJk∞ {| 0| | N|} kJk − kJk∞ (1.14) λ− +λ+ λ− λ+ p + | N 0| 6 N − 0 6 (4m 1) , if Trb = 0. ∞ ∞ j kJk 2 2 − kJk (1.15) Xj=1 Remark. A priori estimates for s alar Ja obi operators were obtained in [BGGK℄, [K℄, [KKr℄. The plan of our paper is as follows. In Se t. 2 we prove Theorem 1.2. In Se t. 3 we prove Theorem 1.3, 1.4 and Proposition 1.5. In Se tion 4 we onsider examples for the ase m = p = 2 , where we onstru t the omplex and real resonan es. In Se t. 5 we shortly re all the well known results about the properties of point spe trum and the absen e of singular ontinuous spe trum. In the proof we use [CK℄, [BBK℄. 2 The Lyapunov fun tions y = (yn)n∈Z anyn+1 +bnyn + We re all well-known results. For any solution of the equation a⊤ y = zy n−1 n−1 n we de(cid:28)ne y 0 I f = n , and = m , n Z. n (cid:18) y (cid:19) Tn (cid:18) a−1a⊤ a−1(z b ) (cid:19) ∈ (2.1) n+1 − n n−1 n − n ϑ ϕ f f = f M = n n Then n satis(cid:28)es n Tn n−1. Thus the matrix-valued fun tion n (cid:18) ϑ ϕ (cid:19) n+1 n+1 M = M ,M = I M = n sna>tis1(cid:28),eswherne njT=n1Xjn−=1Xn.0..X1 f2omr.maTthrii sesgiXvejs. tWhee rmewonriotderTonmyinmthaetrfioxrm n Qj=1Tj, Q I 0 0 I a⊤ 0 = m m n−1 = P−1R 0P , Tn (cid:18) 0 a−n1 (cid:19)(cid:18) −Im z −bn (cid:19)(cid:18) 0 Im (cid:19) n nTn n−1 (2.2) where I 0 0 I P = a⊤ I , R = P m = a⊤ a−1, 0 = m . n n ⊕ m n n(cid:18) 0 a−n1 (cid:19) n ⊕ n Tn (cid:18) −Im z −bn (cid:19) (2.3) Then using (2.2), we get n n n M = P−1R 0P = P−1 R 0 P = P−1M P , M = R 0. n j jTj j−1 n jTj 0 n n 0 n jTj (2.4) Yj=1 (cid:16)Yj=1 (cid:17) Yj=1 R , j > 1 j j Matri es T , satisfy 0 I R⊤JR = J, ( 0)⊤J 0 = J, J = m , j j Tj Tj (cid:18) I 0 (cid:19) (2.5) m − 5 whi h yields M⊤JM = J, M−1 = JM⊤J, n Z. n n n − n ∈ (2.6) Using (2.4) we obtain a⊤ϑ (a⊤)−1 a⊤ϕ a−1ϑ⊤a a−1ϑ⊤ M = P M P−1 = n n 0 n n , M⊤ = 0 n n 0 n+1 n n n 0 (cid:18) ϑ (a⊤)−1 ϕ (cid:19) n (cid:18) ϕ⊤a ϕ⊤ (cid:19) (2.7) n+1 0 n+1 n n n+1 and ϕ⊤ ϕ⊤a M−1 = JM⊤J = n+1 − n n . n − n (cid:18) −a−01ϑ⊤n+1 a−01ϑ⊤nan (cid:19) (2.8) M M−1 = I M−1M = I n n 2m n n 2m Due to and we dedu e that a⊤ϑ (a⊤)−1ϕ⊤ a⊤ϕ (a⊤)−1ϑ⊤ = I , ϑ (a⊤)−1ϕ⊤ = ϕ (a⊤)−1ϑ⊤, n n 0 n+1 − n n 0 n+1 m n 0 n n 0 n (2.9) ϑ⊤a ϕ ϑ⊤ a ϕ = a , ϕ⊤ a⊤ϕ = ϕ⊤a⊤ϕ , ϑ⊤ a⊤ϑ = ϑ⊤a⊤ϑ . n n n+1 − n+1 n n 0 n+1 n n n n n+1 n+1 n n n n n+1 (2.10) Re all the asymptoti s of fundamental solutions ϕ = znA +O(λn−1), ϑ = O(zn−1), n > 1, A = (a a ...a )−1 n+1 n n+1 n 1 2 n (2.11) z M p as → ∞. Substituting last asymptoti s into we obtain 0 0 M (z) = zp +O(zp−1). p (cid:18) 0 A (cid:19) (2.12) p D(z,τ) It is well known that the determinant satis(cid:28)es 2m 2m D(z,τ) = det(M(z) τI) = τ2m + τ2m−jξ (z) = (τ τ (z)), z,τ C, j j − − ∈ (2.13) Xj=1 Yj=1 s−1 1 1 ξ = 1, ξ = T , ξ = (T +T ξ ), ...,ξ = T ξ ,.., 0 1 1 2 2 1 1 s s−j j − −2 −s (2.14) X0 ξ = ξ j N j 2m−j m see p.331-333 [RS℄. TheTid,ennt>ity1(1.6) gives for all ∈ . Re all that the n Chebyshev polynomials satisfy: [n] τn +τ−n 2 (n j 1)! T (ν) = = 2m−1 c νn−2j, c = ( 1)jn − − 2n−2j−m, n n,j n,j 2 − (n 2j)!j! (2.15) X0 − ν = τ+τ−1 τn+τ−n = T (ν) 2 , see [AS℄. Then (2.13) and the identity 2 n yield D(τ,z) (τm +τ−m) (τm−1 +τ1−m) m ξ = +ξ +...+2−mξ = j T (ν), (2τ)m 2m 1 2m m 2m−1 j (2.16) Xj=0 6 1 m τn +τ−n m TrMn(z) = j j = T (∆ (z)), z C. n j 2 2 ∈ (2.17) X1 Xj=1 The substitution of (2.15) into the equality (2.16) gives m D(τ,z) Φ(ν,z) = = φ (z)νm−j, (2τ)m j (2.18) X0 ξ 1 φ = 1, φ = c ξ = , φ = c +c ξ , φ = c ξ +c ξ ,..., 0 1 m−1,0 1 2 m,1 m−2,0 2 3 m−1,1 1 m−3,0 3 2 (2.19) φ = c +c ξ +c ξ +...+c ξ , 2n m,n m−2,n−1 2 m−4,n−2 4 m−2n,0 2n (2.20) φ = c ξ +c ξ +c ξ +...+c ξ . 2n+1 m−1,n 1 m−3,n−1 3 m−5,n−2 5 m−2n−1,0 2n+1 (2.21) a = I ,b = 0 n Z 0 n m n LDe0t for all ∈ and denote the orreDsp0(oτn,dzi)ng=J(aτ 2o+bi1oper2aτtTor(bzy))Jm . If is the orresponding determinant, then we obtain − p 2 and Φ0 = D0(τ,z) = (ν T (z))m = mCm( T (z))jνm−j, CN = N! (2τ)m − p 2 0 j − p 2 where m (N−m)!m!. P Φ(ν,z) = D(τ,z) = mφ (z)νm−j Proof of Theorem 1.2. i) We have the identity (2τ)m 0 j , see φ ∆ (z),..,∆ P(z) j 1 m (Φ2(.z1,8ν),)w=he0re the polyznomCials are given by (2.19)-(2.21). are the roots of for (cid:28)xed ∈mν.m−Tjhwen(zt)he statemwent i) follows from thze well-known about the zeros of a polynomial 0 j , where j is a polynomials in , see [Fo℄. ∆′(z)P= 1(1 τ−2(z))τ′(z) = 0,z Y ii) We have j 2 − 6 ∈ , sin e by the Lyapunov-Poin ar(cid:1)e τ′(z) = 0 z Y Theorem, 6 for all ∈ . f = τn + α τn−1 + .. + α , g = b τs + 1 n 0 iii) Re all that the resultant for polynomials β τs−1 +..+β 1 s is given by 1 α .. α 0 0 .. 0 1 n  0 1 α .. α 0 .. 0   1 n n . . . . . . . .  lines    0 ... 0 1 α .. α  R(f,g) = det 1 n  .   β0 β1 .. βs 0 .. 0   (2.22)    0 β β .. β 0 .. 0    0 1 s  s  . . . . . . . .   lines    0 ... 0 β β .. β  0 1 s      f τ ,..,τ 1 n The dis riminant of the polynomial with zeros is given by Disf = (τi τj)2 = ( 1)n(n2−1)R(f,f′). − − Yi<j DisΦ (λ,τ) = (∆ (λ) ∆ (λ))2 = ( 1)Nj(Nj−1)R(Φ ,Φ′) Thus we have j i<s,i,s∈ωj i − s − 2 j j is polyno- Φ (λ,τ) Q ρ = N0Dis Φ miailv,)siEn ae h jgap hasistthheefpoormlynγonm=ial.(λT−nh,eλn+nt)h=e fu∩nmj= t1iγonn,j, whQere1 γn,j =j(iλs−np,jo,λly+nn,jo)m⊂ialR. is ∆ (z) / [ 1,1] z γ ∆ (λ− ) = some (cid:28)nite interval su h that j ∈ − for all ∈ n,j. Note that j n,j ± or ∆ (λ+ ) = 1 λ± ∆ (z) j n,j ± or n,j is the bran h point of j , otherwise we have a ontradi tion. 7 F v) Using (2.12) we de(cid:28)ne by F(t) = tpM = F +tF (t), F = 0 A , t = z−1, p 0 1 0 m p ⊕ F F (t) = const+O(t) t 0 τ (t),j N 1 1 j 2m where is a some maFtr(itx)polynomial anτd0,j N as → A. Let τ (∈t) be the eigenvalues of . Re all that j ∈ m are eigenvalues of p. Theen, j are τ1/mj given by onvergent Puiseux series (the Taylor series in ) e ∞ k k τ (t) = τ0 + α tmj, t < r, m = m, m N, j w ,s N j j j,k | | j j ∈ ∈ s ∈ k Xk=1 X1 e r > 0 w = N ,w = N w ,...,w = N N for some , see p=4, [RS℄, where 1 m1 2 m1+m2 \ 1 k m \ m−mk τ0 = τ0 j,j′ w τ (t),j w and j j′ for all ∈ s. Here j ∈ s are the bran hes of one or more multivalued τ = 0 analyti fun tions with at worst algeebrai bran h points at and satisfy τj(z) = zpτj(t) = zp(τj0 +O(z−m1s)) = τj0zp +O(zp−m1s), j ∈ ws. τ0 e τ (t),j N t m = 1 If ea h eigenvaluτe(zj)i=s sτim0zpple+, Oth(eznp−e1a) ,hj j N ∈ m is anal2y∆ti f=orτs(mza)l+l , i1.e.,= τs0zp +. Thus we obtain j j e∈ m, whi h yields j j τj(z) j O(zp−1) ρ s and using (1.8), we get asymptoti s for . 3 Inverse Problems In order to prove Theorem 1.3 we need the Lemma 3.1-3.3. W = W s s s W = cosjκ Lemmκar 3.C1,.sL>et1 { cor,sj}κj,rr==0 cboesaκr′ × martr=ixrw′ ith omdpeotnWent=s 0 r,j r for some ∈ , where 6 for all 6 . Then 6 . cosjz = T (cosz) T j j Proof. UsWing = cosjκ = T (co,swκhe)re theWCyhe=by0shev polynomialsy = (ayre)gsiveCnsb+y1,(y2.=150), we obtain r,j r j Pr(x.)L=et s y Tfo(rx)s,oPme=v0e tor n 0 ∈ cosκ ,6 j . Then we dedu e that a polynomial j=0 j j 6 has distin tzeros j ∈ N0 degPP6 degT = s W s s , whi h gives a ontradi tion, sin e . Thus is invertible. c = ( 1)mdetA ξ (z),j N p j m Lemma 3.2. Let − . The polynomials ∈ , given by (2.13), satisfy ξ (z) = ξ (z) = O(zpj), ξ (z) = czpm +O(zpm−1) as z , , j 2m−j m → ∞ (3.1) mp m−1 D(z,τ) c (z λ (τ)) = = (τm +τ−m)+ξ (z)+ (τm−j +τj−m)ξ (z). − n τm m j (3.2) nY=1 Xj=1 Proof. Using (1.6), (2.13) and Viete's formulas, we get j ξ (z) = ξ (z) = ( 1)j τ (z), 1 6 j 6 m. j 2m−j − is (3.3) 16i1<X..<ij62mYs=1 8 τ (z) = τ0zp + o(zp) j N Asymptoti s (1.9) give j j , ∈ m and the Lyapunov-Poin ar(cid:1)e Theorem τ (z) = 1 = O(z−p) z τ0 = 0 yields m+j τj(z) as → ∞, sin e ea h j 6 . Then, using (3.3), we obtain m ξ (z) = O(zpj), ξ (z) = ( 1)m τ0zpm +o(zpm) = ( 1)mdetA zpm +O(zpm−1). j m − j − p Yj=1 Identity (3.2) follows from last identity and (1.6), (2.13). We need the following Lemma about polynomials. (z )k Ck (c )s Cs+1 s,k > 1 Lemma 3.3. Let n 1 ∈ and j 0 ∈ for some integers . Then there exists r(z) g a unique polynomial satisfying for some polynomial the following relations s k r(z) = zk c zj+O(zk−1) as z and r(z) = h(z)g(z), h(z) = (z z ), j n → ∞ − (3.4) Xj=0 nY=1 = g : degg 6 s , dim = s +1 s s Proof. Intgrodu e the linear spa e of polynomia:ls P C{s+1 } P . s s Note that ∈ P . De(cid:28)ne the linear operator A P → by 1 dn ( g) = h(z)g(z) , n = k +j 1, j N , g . A j n!dzn |z=0 − ∈ s+1 ∈ Ps (cid:16) (cid:17) g ( g) = c , j N h(z)g(z) = zk s c zj+O(zk−1) We rewriteA inthe form A j j−1 ∈ s+1, where j=0 j z : Cs+1 diPm = dimCs+1 = s s as → ∞. We dedu e that A P → is an isomorphism, sin e P s + 1 g = 0 g = 0 g = −1(c )s r(z) = h(z)g(z) r(z) = zk s ancdzijf+AO(zk−1,)thenz . If A n 0 and r , then j=0 j as → ∞. Moreover, the polynomial is unique, sin e A is an P isomorphism. c = ( 1)mdetA ζ (z) c−1 p 0 Pζ r(zo)of ocf−T1ξhe(zo)rejm 1N.3. Re all that − degζ 6.pjDe(cid:28)ne the pζol(yzn)o=mialspj ζ ≡zn , sojmeζ≡j,n ∈ Rj. In,tro∈du emt.heLseemtsmKas3=.2{ypi(emlds−s−1)j+1,...,pa(nmd−thse)n}, js ∈ N0mP−1,n=K0 mj,n= {0fo}r, whi h satisfy m K = N0 , K K = ,for all r = s. ∪s=j s p(m−j) r ∩ s ∅ 6 (3.5) h = τ−j +τj j Using (3.2) and (3.5) and setting , we obtain pm m m D(z,τ) q(z,τ) = = (z λ (τ)) = h ζ (z) = h ζ (z) n m−j j j m−j cτm − (3.6) nY=1 Xj=0 Xj=0 m p(m−j) m m m m = h ζ zn = h ζ zn = h ζ zn j m−j,n j m−j,n j m−j,n Xj=0 Xn=0 Xj=0 Xs=j nX∈Ks Xj=0 Xs=j nX∈Ks m s m s m pm = h ζ zn = h ζ zn = η (τ)zn = η (τ)zn, j m−j,n j m−j,n n n Xs=0 Xj=0 nX∈Ks Xs=0 nX∈KsXj=0 Xs=0 nX∈Ks Xn=0 9 where s η (τ) = h ζ , n K , s N0 . n j m−j,n ∈ s ∈ m (3.7) Xj=0 τ = eiκr,r N0 Substituting ∈ s into the identity (3.7) we get η = 2W y , η = (η (eiκr))s , y = (ζ )s , W = (cosjκ )s , n s n n n r=0 n m−j,n j=0 s r r,j=0 (3.8) for all e(n,s) ∈ Ks × Ne0m. Due to Lemma 3.1, ea h matrix Ws,s ∈ N0m is invertible. Thus (3.8), (3.7) give η (τ) =< (τj +τ−j)s ,(2W )−1η >, (n,s) K N0 n j=0 s n ∈ s × m (3.9) < , > Rs e ηn( ),n Ns ηwhe=re(η (·ei·κr))iss a s alar produ t in . Then the fun tions · ∈η ( ),anre dNetermined by n n r=0. We will use this fa t to determine all fun tions n · ∈ m. In order teo des ribe re overing we need the following simple fa t. 0 6 k 6 m 1 η (eiκj),(n,j) N0 N0 Lemmηa 3n.4. Lekt K Λ− . Then numbers nκ j N0 ∈ pm × k, the fun - tions n, ∈ ∪j=0 j, the set k+1 and the numbers j, ∈ m determine the numbers η (eiκj),(n,j) N0 N0 η n k+1K n ∈ pm × k+1, and the fun tions n, ∈ ∪j=0 j. Proof. Let Sk+1 = (ηn(eiκk+1))n∈∪kj=0Kj = (ηn(eiκk+1))pnm=p(m−k−1)+1. Using (3.6), we see Λ q(z,eiκk+1) k+1 that the elements of are some zeros of the polynomial and re all that #Λ = p(m k 1) + 1 S k+1 k+1 − − . Also, usiqn(gz,(e3iκ.6k)+,1)we see that the omponentsΛof ve tor S k+1 k+1 are the oe(cid:30) ients of the polynomiqa(lz,eiκk+1) . Then, due to Lemma 3.3,(η (eiκakn+d1))pm n n=0 uniquelydeterminethepolyno(mηia(eliκj)) ajnd,Nus0ing(3.6), wedetermine η (τ). Substituting obtained values n n∈Kk+1, ∈ k+1 into the (3.9), we determine n , n K k+1 ∈ . q(z,τ) = D(z,τ) Now we des ribe theΛre overing pro edure. Re all thqa(zt,eiκ0) (ηcτm(ei.κ0S))upbmstituting the elements of the set 0 intηo t(hτe),(n3.6)K, we determine and n nk=0. NT0hen (3.9) determineηs(aτll)funn tionms Kn = N∈0 0. Step by step, usiqn(gz,Lτe)mma 3.4 for ∈ m−1, we determine n , ∈ ∪j=0 j pm. Then (3.6) gives . Using (2.13) we get cq(z,τ) = D(z,τ) = τ2m + O(τ2m−1) τ q(z,τ) as → ∞, then the polynomial uniquely c D(z,τ) determine onstant aa,nbd,n Z. a > 0,n Z = m ii) Assume that n n ∈ are diagonal, and ea h n ∈ . Then J ⊕1 Jj, ∆ D j j j where J are s alar Ja obi operators. Let be,jtheNLyapunov fun tions and let be the orresponding determinant for the operators Jj ∈ m. Then we dedu e that e m D(z,τ) = D , D (z,τ) = τ(τ +τ−1 2∆ (z)). j j j − (3.10) Yj=1 e e ∆ (z) ∆′ (z0) = 0 ( 1)p−j∆ (z0) = B∆y T(zh0e)or>em1,1.1j, theNLyapunov funz t0io<n...m< z0satis(cid:28)es m j andq (−z) = s(∆m (zj) |cosmκ )j +| cosκ ,s∈> 1p−1 for some 1 p. Then the polynomial s m − 0 0 satis(cid:28)es 1 ( 1)p−jq (z0) = s ∆ (z0) +( 1)p−j 1+ cosκ > 1, (q )′(z0) = 0, j N . − s j | m j | − − s 0 s j ∈ p−1 (cid:16) (cid:16) (cid:17) (cid:17) 10

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.