Upload
m-i
View
212
Download
0
Embed Size (px)
Citation preview
January 23, 2003 11:33 WSPC/152-CCM 00090
Communications in Contemporary MathematicsVol. 5, No. 1 (2003) 101–118c© World Scientific Publishing Company
BOUNDS FOR THE SPECTRUM OF ANALYTIC
QUASINORMAL OPERATOR PENCILS
M. I. GIL’
Department of Mathematics, Ben Gurion University of the Negev,P.0. Box 653, Beer-Sheva 84105, Israel
Received 23 April 2001Revised 1 May 2002
We consider a class of pencils (operator valued functions of a complex argument) in aseparable Hilbert space. Bounds for the λ-nonlinear spectrum are suggested. Applica-tions to differential operators, integral operators with delay and infinite matrix pencilsare also discussed.
Keywords: Linear operators; pencils; spectrum; integral and differential operators;infinite matrices.
Mathematics Subjects Classification 2000: 47A55, 47A75, 47G10, 47G20
1. Introduction and Statement of the Main Result
Numerous mathematical and physical problems lead to analytic pencils (ana-
lytic operator-valued functions of a complex argument), cf. [1, 6, 14, 16, 26], and
references therein. A lot of papers and books are devoted to the spectrum of
analytic operator pencils functions, in particular, polynomial operator pencils, (see
for instance, [13, 18, 24, 25, 27]). Mainly, the completeness of the root vectors
and asymptotic distributions of the eigenvalues are considered. However, in many
applications, for example, in numerical mathematics and stability analysis, bounds
for the spectrum are very important. But the bounds are investigated considerably
less than the asymptotic distributions and the completeness. Below we establish
bounds for the spectrum of a class of operator pencils. We also discuss applications
of these bounds to differential operators, integral operators with delay and matrix
pencils. Our results are new even in the case of the usual (linear) spectral problems,
cf. [4, 5, 15, 21, 22].
Note also that our results below are intimately connected with the spectral
problems, considered in the books [11, Chap. X] and [3, Chap. V].
Let H be a separable Hilbert space with a scalar product (·, ·), the norm ‖ · ‖and the unit operator I. For a linear operator A, σ(A) is the spectrum. Recall that
101
Com
mun
. Con
tem
p. M
ath.
200
3.05
:101
-118
. Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
MO
NA
SH U
NIV
ER
SIT
Y o
n 05
/20/
13. F
or p
erso
nal u
se o
nly.
January 23, 2003 11:33 WSPC/152-CCM 00090
102 M. I. Gil’
a linear operator V is a Volterra one if it is quasinilpotent (that is, σ(V ) = {0})and compact, cf. Gohberg and Krein [12].
In the present paper we consider an operator-valued function A(λ) of the type
A(λ) = D(λ) + V+ + V−(λ ∈ C) (1.1)
where V± are Volterra operators in H with the properties pointed below and D(λ)
is a normal operator in H of the form
D(λ) =
∫ ∞−∞
e(λ, t)dP (t) (λ ∈ C)
where e(λ, t) for any real finite t is an entire function of λ, and for any λ ∈ C is
a continuous function of t ∈ R. Moreover, for each finite λ, D(λ) is bounded. In
addition, P (t)(−∞ ≤ t ≤ ∞) is a maximal resolution of the identity. That is, P (·)is a left-continuous orthogonal resolution of the identity defined on (−∞,∞) with
the property: any gap P (t0 + 0) − P (t0) of P (·) (if it exists) is one-dimensional,
cf. Brodskii [2], Gohberg and Krein [12], and Gil’ [7, p. 69]. In addition, we assume
that
P (t)V+P (t) = V+P (t) and P (t)V−P (t) = P (t)V− (t ∈ R) . (1.2)
Furthermore, λ is a regular point of A(·) ifA(λ) is boundedly invertible. The comple-
ment of the set of all regular points to the closed complex plane is the (λ-nonlinear)
spectrum of A(·) and is denoted by Σ(A(·)).So for a linear operator A0, σ(A0) =
∑(A(·)) with A(λ) = A0 − λI.
Clearly, D(λ) is invertible if and only if
ρ(D(λ)) ≡ inft∈γ(P )
|e(λ, t)| > 0 (1.3)
where γ(P ) denotes the set of all points of the growth of P (·). That is,
P (t2)− P (t1) = 0 (t1 < t2)
if and only if the segment (t1, t2] does not belong to γ(P ).
Let Y be an ideal of linear compact operators in H with a norm | · |Y , such that
|CB|Y ≤ ‖C‖|B|Y and |BC|Y ≤ ‖C‖|B|Y for an arbitrary bounded linear operator
C in H and a B ∈ Y . Assume that Y has the following property: any Volterra
operator V ∈ Y satisfies the inequalities
‖V k‖ ≤ θk|V |kY (k = 1, 2, . . .) (1.4)
where constants θk are independent of V and
k√θk → 0 (k →∞) .
For instance, if Y is the Hilbert–Schmidt ideal, then (1.4) holds with θk = 1/√k!
(see [7, Sec. 2.3]). We also will check that the von Neumann–Schatten ideal has
property (1.4). It is assumed that
V+, V− ∈ Y (1.5)
Com
mun
. Con
tem
p. M
ath.
200
3.05
:101
-118
. Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
MO
NA
SH U
NIV
ER
SIT
Y o
n 05
/20/
13. F
or p
erso
nal u
se o
nly.
January 23, 2003 11:33 WSPC/152-CCM 00090
Bounds for the Spectrum of Analytic Quasinormal Operator Pencils 103
and ideal Y has property (1.4). Furthermore, for a Volterra operator V ∈ Y , denote
JY (V,m, z) ≡m−1∑k=0
z−1−kθk|V |kY (z > 0) .
Set
W±(λ) ≡ D−1(λ)V± (λ /∈ Σ(D(·))) .
Due to [7, Lemma 3.2.4], and according to (1.2), the operators W±(λ) are
quasinilpotent for any λ /∈ Σ(D(·)).A natural number ni(V ) will be called the indicator of the nilpotentness of a
nilpotent operator V , if V ni(V ) = 0 but V ni(V )−1 6= 0. If V is quasinilpotent but
not nilpotent, then we put ni(V ) =∞. Set
ν±(λ) ≡ ni(W±(λ)) .
Everywhere below one can replace ν±(λ) by ∞. Now we are in a position to
formulate the main result of the paper
Theorem 1.1. Let A(λ) be given by (1.1). In addition, under conditions (1.2) and
(1.5), for a λ /∈ Σ(D(·)), let
ζ(A(λ)) ≡ max
{1
JY (V−, ν−(λ), ρ(D(λ)))− ‖V+‖,
1
JY (V+, ν+(λ), ρ(D(λ)))− ‖V−‖
}> 0 . (1.6)
Then λ is a regular point of operator-valued function A(·) represented by (1.1).
Moreover,
‖A−1(λ)‖ ≤ 1
ζ(A(λ))ρ(D(λ)). (1.7)
The proof of this theorem is presented in the next section.
2. Proof of Theorem 1.1
We need the following simple
Lemma 2.1. Let
A0 = I +W1 +W2 , (2.1)
where W1,W2 are bounded linear operators in H. In addition, W1 is a quasinilpotent
operator. Then under the condition
ψ0 ≡ ‖(I +W1)−1W2‖ < 1 , (2.2)
the operator A0 is boundedly invertible, and
‖A−10 ‖ ≤
‖(I +W1)−1‖1− ψ0
.
Com
mun
. Con
tem
p. M
ath.
200
3.05
:101
-118
. Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
MO
NA
SH U
NIV
ER
SIT
Y o
n 05
/20/
13. F
or p
erso
nal u
se o
nly.
January 23, 2003 11:33 WSPC/152-CCM 00090
104 M. I. Gil’
Proof. Since W1 is a quasinilpotent operator, I + W1 is invertible. According to
(2.1) we have
A0 = (I +W1)(I + (I +W1)−1W2) . (2.3)
Moreover, under condition (2.2), the operator I + (I +W1)−1W2 is also invertible
and
‖(I + (I +W1)−1W2)−1‖ ≤∞∑k=0
‖((I +W1)−1W2)k‖ ≤∞∑k=0
ψk0 = (1− ψ0)−1 .
Due to (2.3)
‖A−10 ‖ ≤ ‖(I + (I +W1)−1W2)−1‖‖(I +W1)−1‖ .
This proves the required result.
For the sake of brevity, put
jY (V ) = JY (V, ni(V ), 1) =
ni(V )−1∑k=0
θk|V |kY (2.4)
for a Volterra operator V from ideal Y with property (1.3). Everywhere below j−1Y (·)
means 1jY (·) .
Lemma 2.2. Let W1,W2 be Volterra operators from ideal Y with property (1.4).
Then under the condition
ζ(A0) ≡ max{j−1Y (W1)− ‖W2‖, j−1
Y (W2)− ‖W1‖} > 0 , (2.5)
the operator A0 represented by (2.1) is boundedly invertible. Moreover, the inverse
operator satisfies the inequality ‖A−10 ‖ ≤ ζ−1(A0).
Proof. Due to (1.4), we have
‖(I +W1)−1‖ ≤∞∑k=0
‖W k1 ‖ =
ni(W1)−1∑k=0
‖W k1 ‖
≤ni(W1)−1∑k=0
θk|W1|kY = jY (W1) .
Hence, ψ0 ≤ ‖W2‖jY (W1). So the condition
‖W2‖jY (W1) < 1 (2.6)
implies (2.2). Now Lemma 2.1 yields the invertibility of A0 with the estimate
‖A−10 ‖ ≤
jY (W1)
1− ‖W2‖jY (W1)=
1
j−1Y (W1)− ‖W2‖
. (2.7)
Com
mun
. Con
tem
p. M
ath.
200
3.05
:101
-118
. Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
MO
NA
SH U
NIV
ER
SIT
Y o
n 05
/20/
13. F
or p
erso
nal u
se o
nly.
January 23, 2003 11:33 WSPC/152-CCM 00090
Bounds for the Spectrum of Analytic Quasinormal Operator Pencils 105
But condition (2.6) is equivalent to the following one: j−1Y (W1) − ‖W2‖ > 0.
Exchanging W1 and W2, we can assert that the condition j−1Y (W2) − ‖W1‖ > 0
yields the invertibility and estimate
‖A−10 ‖ ≤
1
j−1Y (W2)− ‖W1‖
.
This, (2.5) and (2.7) imply the required result.
Proof of Theorem 1.1. Due to (1.1),
A(λ) = D(λ)(I +W+(λ) +W−(λ)) . (2.8)
Let us apply Lemma 2.2 to the operator-valued function I + W+(λ) + W−(λ).
According to (2.4), condition (2.5) means that at least one of the following
inequalities
j−1Y (W−(λ)) > ‖W+(λ)‖, or j−1
Y (W+(λ)) > ‖W−(λ)‖
holds. Equivalently,
jY (W−(λ))‖W+(λ)‖ < 1 or (and) jY (W+(λ))‖W−(λ)‖ < 1 . (2.9)
Since D(λ) is normal,
‖D−1(λ)‖ ≤ 1
ρ(D(λ)). (2.10)
Thus, ‖W+(λ)‖ ≤ ‖V+‖ρ−1(D(λ)) and
jY (W±(λ)) ≤ν(λ)−1∑k=0
θkρ−k(D(λ))|V±|kY .
Hence,
‖W+(λ)‖jY (W−(λ)) ≤ ‖V+‖ρ−1(D(λ))jY (W−(λ))
= ‖V+‖ν(λ)−1∑k=0
θk|V−|kYρk+1(D(λ))
= ‖V+‖JY (V−, ν−(λ), ρ(D(λ))) .
Similarly
‖W−(λ)‖jY (W+(λ)) ≤ ‖V−‖JY (V+, ν+(λ), ρ(D(λ))) .
Thus, condition (1.6) implies at least one of the inequalities (2.9). Now Lemma 2.2
yields the inequality
‖(I +W+(λ) +W−(λ))−1‖ ≤ ζ−1(A(λ)) .
Taking into account relations (2.8) and (2.10) we arrive at the required result.
Com
mun
. Con
tem
p. M
ath.
200
3.05
:101
-118
. Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
MO
NA
SH U
NIV
ER
SIT
Y o
n 05
/20/
13. F
or p
erso
nal u
se o
nly.
January 23, 2003 11:33 WSPC/152-CCM 00090
106 M. I. Gil’
3. Spectrum Localization
Theorem 1.1 implies
Theorem 3.1. Let A(λ) be given by (1.1). Then, under conditions (1.2) and (1.5),
for any µ ∈ Σ(A(·)), either µ ∈ Σ(D(·)), or the both inequalities
‖V+‖JY (V−, ν−(µ), ρ(D(µ)) ≥ 1 and ‖V−‖JY (V+, ν+(µ), ρ(D(µ)) ≥ 1 (3.1)
are true.
It is easy to show that the latter result is exact in the following sense: if either
V− = 0, or (and) V+ = 0, then due to the Corollary 3.1,
Σ(A(·)) = Σ(D(·)) . (3.2)
Now put
ν± ≡ supλ/∈Σ(D(·))
ν±(λ) = supλ/∈Σ(D(·))
ni(W±(λ)) .
In the sequel one can replace ν± by ∞.
Corollary 3.2. Let A(λ) be given by (1.1). In addition, let (1.2) and (1.5) hold
with V+ 6= 0 and V− 6= 0. Then each of the following equations:
‖V+‖JY (V−, ν−, z) = 1 and ‖V−‖JY (V+, ν+, z) = 1 (3.3)
has a unique positive root zup(Y ) and zlow(Y ), respectively. Moreover, any µ ∈Σ(A(·)) satisfies the inequality
ρ(D(µ)) ≤ min{zup(Y ), zlow(Y )} . (3.4)
Indeed, inequalities (3.1) imply
‖V+‖JY (V−, ν−, ρ(D(µ)) ≥ 1 and ‖V−‖JY (V+, ν+, ρ(D(µ)) ≥ 1 .
Comparing these inequalities with Eq. (3.3), we arrive at the required result.
To estimate the quantities zup(Y ), zlow(Y ), let us consider the equation
∞∑k=1
akzk = 1 (3.5)
where the coefficients ak are nonnegative and have the property
γ0 ≡ 2 maxk
k√ak <∞ .
Lemma 3.3. The unique nonnegative root z0 of Eq. (3.5) satisfies the estimate
z0 ≥ 1/γ0.
Proof. Set in (3.5) z = xγ−10 . We have
1 =∞∑k=1
akγ−k0 xk . (3.6)
Com
mun
. Con
tem
p. M
ath.
200
3.05
:101
-118
. Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
MO
NA
SH U
NIV
ER
SIT
Y o
n 05
/20/
13. F
or p
erso
nal u
se o
nly.
January 23, 2003 11:33 WSPC/152-CCM 00090
Bounds for the Spectrum of Analytic Quasinormal Operator Pencils 107
But∞∑k=1
akγ−k0 ≤
∞∑k=1
2−k = 1
and therefore, the unique positive root x0 of (3.6) satisfies the inequality x0 ≥ 1.
Hence, z0 = γ−10 x0 ≥ γ−1
0 .
Note that the latter lemma generalizes the well-known result for algebraic
equations, cf. Ostrowski [19, p. 277].
Lemma 3.3 gives us the inequalities
zup(Y ) ≤ 2 maxj=1,2,...
j
√θj−1|V−|j−1
Y ‖V+‖
and
zlow(Y ) ≤ 2 maxj=1,2,...
j
√θj−1|V+|j−1
Y ‖V−‖, with θ0 = 1 .
Now Theorem 3.2 implies
Corollary 3.4. Under conditions (1.1), (1.2) and (1.5) for any µ ∈ Σ(A(·)) the
estimate ρ(D(µ)) ≤ ψY (A) is valid, where
ψY (A) ≡ 2 min
{max
j=1,2,...
j
√θj−1|V−|j−1
Y ‖V+‖, maxj=1,2,...
j
√θj−1|V+|j−1
Y ‖V−‖}.
4. Pencils with Hilbert Schmidt Nilpotent Parts
Let N1(K) be the Hilbert–Schmidt norm of a Hilbert–Schmidt operator (HSO) K:
N1(K) ≡ [Trace K∗K]1/2 .
Throughout this section it is assumed that
W±(λ) = D−1(λ)V±(λ /∈ Σ(D(·))) are Volterra Hilbert–Schmidt operators .(4.1)
For a Volterra Hilbert–Schmidt operator V , put
J2(V,m, z) =m−1∑k=0
Nk1 (V )
zk+1√k!
(z > 0) .
Recall that ν±(λ) ≡ ni(W±(λ)) ≤ ∞.
Theorem 4.1. Under conditions (1.2) and (4.1), for a λ /∈ Σ(D(·)), let
ζH(A(λ)) ≡ max
{1
J2(V−, ν−(λ), ρ(D(λ)))− ‖V+‖ ,
1
J2(V+, ν+(λ), ρ(D(λ)))− ‖V−‖
}> 0 . (4.2)
Com
mun
. Con
tem
p. M
ath.
200
3.05
:101
-118
. Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
MO
NA
SH U
NIV
ER
SIT
Y o
n 05
/20/
13. F
or p
erso
nal u
se o
nly.
January 23, 2003 11:33 WSPC/152-CCM 00090
108 M. I. Gil’
Then λ is a regular point of the operator-valued function A(·), represented by (1.1).
Moreover,
‖A−1(λ)‖ ≤ 1
ρ(D(λ))ζH (A(λ)).
Proof. Due to [7, Lemma 2.3.1], we have
‖V j‖ ≤ N j1 (V )√j!
(j = 1, 2, . . .) (4.3)
for any Volterra HSO V . Now the required result follows from Theorem 1.1.
Theorem 4.1 yields
Corollary 4.2. Let A(λ) be given by (1.1). Then, under conditions (1.2) and (4.1),
for any µ ∈ Σ(A(·)) we have either µ ∈ Σ(D(·)), or the both inequalities
‖V+‖J2(V−, ν−(µ), ρ(D(µ))) ≥ 1 and ‖V−‖J2(V+, ν+(µ), ρ(D(µ))) ≥ 1
are true.
Furthermore, for a constant a > 0, the Schwarz inequality implies,( ∞∑k=0
ak√k!
)2
=
( ∞∑k=0
2k/2ak
2k/2√k!
)2
≤∞∑j=0
2−j∞∑k=0
2ka2k
k!= 2 exp[2a2] .
Hence
J2(V±,m, z) ≤√
2z−1ez−2N2
1 (V±) . (4.4)
Now Theorem 4.1 implies
Corollary 4.3. Let relations (1.2) and (4.1) hold. In addition, for a λ /∈ Σ(D(·)),let
ζH1(A(λ)) ≡ max
{√1/2ρ(D(λ)) exp
[− N2
1 (V+)
ρ2(D(λ))
]− ‖V−‖ ,
√1/2ρ(D(λ)) exp
[− N2
1 (V−)
ρ2(D(λ))
]− ‖V+‖
}> 0 .
Then λ is a regular point of operator-valued function A(·), represented by (1.1).
Moreover,
‖A−1(λ)‖ ≤ 1
ρ(D(λ))ζH1(A(λ)).
Theorem 4.4. Under conditions (1.1), (1.2) and (4.1), let V+ 6= 0 and V− 6= 0.
Then each of the following equations:
‖V+‖J2(V−, ν−, z) = 1 and ‖V−‖J2(V+, ν+, z) = 1 (4.5)
Com
mun
. Con
tem
p. M
ath.
200
3.05
:101
-118
. Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
MO
NA
SH U
NIV
ER
SIT
Y o
n 05
/20/
13. F
or p
erso
nal u
se o
nly.
January 23, 2003 11:33 WSPC/152-CCM 00090
Bounds for the Spectrum of Analytic Quasinormal Operator Pencils 109
has a unique positive root zup(HS) and zlow(HS), respectively. Moreover, for any
µ ∈ Σ(A(·)), the inequality
ρ(D(µ)) ≤ min{zup(HS), zlow(HS)} ≤ ψ2(A)
is true, where
ψ2(A) = 2 min
{max
j=1,2,...
j
√‖V+‖N j−1
1 (V−)√(j − 1)!
, maxj=1,2,...
j
√‖V−‖N j−1
1 (V+)√(j − 1)!
}. (4.6)
Proof. Theorem 3.2 yields
ρ(D(µ)) ≤ min{zup(HS), zlow(HS)} .
But due to Lemma 3.3, min{zup(HS), zlow(HS)} ≤ ψ2(A).
We need the following simple lemma.
Lemma 4.5. The unique positive root z0 of the equation
zez = a (a = const > 0) (4.7)
satisfies the estimate
z0 ≥ ln[1/2 +√
1/4 + a] . (4.8)
If, in addition, the condition a ≥ e holds, then
z0 ≥ ln a− ln ln a . (4.9)
Proof. Since z ≤ ez − 1(z ≥ 0), we arrive at the relation a ≤ e2z0 − ez0 . Hence,
ez0 ≥ r1,2, where r1,2 are the roots of the polynomial y2 − y − a. This proves
inequality (4.8).
Furthermore, if the condition a ≥ e holds, then z0ez0 ≥ e and z0 ≥ 1. Now (4.7)
yields ez0 ≤ a and z0 ≤ ln a. So
a = z0ez0 ≤ ez0 ln a .
Hence, inequality (4.9) follows.
According to (4.4) and (4.5), we have zup(HS) ≤ zup and zlow(HS) ≤ zlow
where zup and zlow are unique positive roots of equations√
2‖V+‖z−1 exp [z−2N21 (V−)] = 1 (4.10)
and√
2‖V−‖z−1 exp [z−2N21 (V+)] = 1 ,
respectively. Clearly, (4.10) is equivalent to the equation
2‖V+‖2z−2 exp [2z−2N21 (V−)] = 1 .
Com
mun
. Con
tem
p. M
ath.
200
3.05
:101
-118
. Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
MO
NA
SH U
NIV
ER
SIT
Y o
n 05
/20/
13. F
or p
erso
nal u
se o
nly.
January 23, 2003 11:33 WSPC/152-CCM 00090
110 M. I. Gil’
Substitute in the latter equation z2 = 2N21 (V−)x−1. Then with the notation
aup ≡N2
1 (V−)
‖V+‖2and alow ≡
N21 (V+)
‖V−‖2
we have xex = aup. Now Lemma 4.5 implies
z2up ≤
2N21 (V−)
ln [1/2 +√
1/4 + aup].
Similarly,
z2low ≤
2N21 (V+)
ln[1/2 +√
1/4 + alow].
Denote
δ(A) ≡√
2 min
{N1(V−)
ln1/2[1/2 +√
1/4 + ‖V+‖−2N21 (V−)]
,
N1(V+)
ln1/2[1/2 +√
1/4 + ‖V−‖−2N21 (V+)]
}. (4.11)
Then
min{zup(HS), zlow(HS)} ≤ min{zup, zlow} ≤ δ(A) .
Clearly, δ(A)→ 0, if either N1(V−)→ 0, or N1(V+)→ 0.
Thus, Theorem 4.4 implies
Corollary 4.6. Under conditions (1.1), (1.2) and (4.1), for any µ ∈ Σ(A(·)), the
inequality ρ(D(µ)) ≤ δ(A) is true, where δ(A) is given by (4.11).
5. Operators with von Neumann Schatten Nilpotent Parts
In this section it is assumed that V± are Volterra operators belonging to the
Neumann–Schatten ideal C2p with some integer p > 1. That is,
Np(V±) ≡ [Trace (V ∗±V±)p]1/2p <∞ . (5.1)
For a natural k, denote
θ(p)k =
1√[k/p]!
where [x] means the integer part of x > 0. For a Volterra operator V ∈ C2p(p > 1),
put
J2p(V,m, z) =m−1∑k=0
θ(p)k Nk
p (V )
zk+1(p > 1, z > 0) .
Recall that ν±(λ) ≡ ni(V±(λ)) ≤ ∞.
Com
mun
. Con
tem
p. M
ath.
200
3.05
:101
-118
. Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
MO
NA
SH U
NIV
ER
SIT
Y o
n 05
/20/
13. F
or p
erso
nal u
se o
nly.
January 23, 2003 11:33 WSPC/152-CCM 00090
Bounds for the Spectrum of Analytic Quasinormal Operator Pencils 111
Theorem 5.1. Under conditions (1.2) and (5.1), for a λ /∈ Σ(D(·)), let
ζ2p(A(λ)) ≡ max
{1
J2p(V−, ν−(λ), ρ(D(λ)))− ‖V+‖ ,
1
J2p(V+, ν+(λ), ρ(D(λ)))− ‖V−‖
}> 0 . (5.2)
Then λ is a regular point of operator-valued function A(·), represented by (1.1).
Moreover,
‖A−1(λ)‖ ≤ 1
ρ(D(λ))ζ2p(A(λ)).
Proof. For any Volterra operator V ∈ C2p we have V p ∈ C2. According to (4.3)
‖V jp‖ ≤ N j1 (V p)√j!
≤Npjp (V )√j!
(V ∈ C2p; j = 1, 2, . . .) .
Hence, for any k = m+ jp(m = 0, . . . , p− 1; j = 0, 1, 2, . . .),
‖V k‖ = ‖Vm+pj‖ ≤ ‖Vm‖N j
1 (V p)√j!
≤Nm+pjp (V )√j!
. (5.3)
Consequently,
‖V k‖ ≤ θ(p)k Nk
p (V ) (k = 1, 2, . . .) . (5.4)
Now the required result is due to Theorem 1.1.
Theorem 5.1 implies
Corollary 5.2. Under conditions (1.1), (1.2) and (5.1), for any µ ∈ Σ(A(·)), either
µ ∈ Σ(D(·)), or the both inequalities
‖V+‖J2p(V−, ν−(µ), ρ(D(µ))) ≥ 1 and ‖V−‖J2p(V+, ν+(µ), ρ(D(µ))) ≥ 1
are true.
Theorem 3.2 and relation (5.4) yield
Theorem 5.3. Let A(λ) be given by (1.1). In addition, under conditions (1.2) and
(5.1), let V+ 6= 0 and V− 6= 0. Then for any µ ∈ Σ(A(·)), the inequality
ρ(D(µ)) ≤ min{zup(C2p), zlow(C2p)}
is true, where zup(C2p) and zlow(C2p) are the unique positive roots of the equations
‖V+‖J2p(V−, ν−, z) = 1 and ‖V−‖J2p(V+, ν+, z) = 1 , (5.5)
respectively.
Com
mun
. Con
tem
p. M
ath.
200
3.05
:101
-118
. Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
MO
NA
SH U
NIV
ER
SIT
Y o
n 05
/20/
13. F
or p
erso
nal u
se o
nly.
January 23, 2003 11:33 WSPC/152-CCM 00090
112 M. I. Gil’
To estimate zup(C2p), zlow(C2p) we can apply Lemma 3.3. Note that according
to (5.3) one can replace Eq. (5.5) by the following ones:
‖V+‖p−1∑j=0
∞∑k=1
N j+pkp (V−)
zj+pk+1√k!
= 1 and ‖V−‖p−1∑j=0
∞∑k=1
N j+pkp (V+)
zj+pk+1√k!
= 1 . (5.6)
Denote
ψ2p(A) = 2 min
maxj=1,2,...,p−1;k=0,1,...
j+kp+1
√‖V+‖N j+kp
p (V−)√k!
,
maxj=1,2,...,p−1;k=0,1,...
j+kp+1
√‖V−‖N j+kp
p (V+)√k!
.
According to (5.6), Theorem 3.2 and Lemma 3.3 yield the inequality
ρ(D(µ)) ≤ ψ2p(A) for any µ ∈ σ(A(·))
provided relations (1.1), (1.2) and (5.1) hold.
6. A Matrix Differential Operator on R1
Let us consider in space L2(R1,Rn) the matrix differential operator T defined by
(Tu)(x) =d2u(x)
dx2+ θ
du(x)
dx+ ψu(x) (x ∈ R) (6.1)
with the domain {v ∈ L2(R1,Rn) :
d2v
dx2∈ L2(R1,Rn)
}where
θ = diag [θj ]nj=1, ψ = (ψjk)nj,k=1
are constant complex n × n-matrices. Applying the Fourier transformation to
operator (6.1), we arrive at the following operator pencil in Cn:
A(λ) = λ2I + λθ + ψ(λ ∈ C) (6.2)
where I is the unit matrix. In the considered case one can take
D(λ) = λ2I + λθ + diag[ψjj ] = diag[λ2 + λθj + ψjj ] .
In addition,
V+ =
0 ψ12 · · · ψ1n
0 0 · · · ψ2n
· 0 · · · ·0 0 · · · 0
and V− =
0 · · · 0 0
ψ21 · · · 0 0
· · · · · ·ψn1 · · · ψn,n−1 0
Com
mun
. Con
tem
p. M
ath.
200
3.05
:101
-118
. Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
MO
NA
SH U
NIV
ER
SIT
Y o
n 05
/20/
13. F
or p
erso
nal u
se o
nly.
January 23, 2003 11:33 WSPC/152-CCM 00090
Bounds for the Spectrum of Analytic Quasinormal Operator Pencils 113
are nilpotent n× n-matrices. So
N21 (V−) =
n∑j=1
j−1∑k=1
|ψjk|2; N21 (V+) =
n∑j=1
n∑k=j+1
|ψjk|2 (6.3)
and ν±(λ) = ni(D−1(λ)V±) ≤ n. Take the maximal spectral function P (t) =
{Qk}nk=1, where Qk are the projectors on the k first coordinates:
Qkh = (h1, . . . , hk, 0, . . . , 0)
for any vector h = (h1, . . . , hn) ∈ Cn. Clearly, conditions (1.2) hold. Obviously,
J2(V±, ν±(λ), z) ≡ν±(λ)−1∑k=0
Nk1 (V±)
zk+1√k!≤ J2(V±, n, z) ≡
n−1∑k=0
Nk1 (V±)
zk+1√k!
(z > 0) .
So instead of (4.5), one can consider the equations
‖V+‖CnJ2(V−, n, z) = 1, ‖V−‖CnJ2(V+, n, z) = 1 (6.4)
where ‖ · ‖Cn is the Euclidean norm. Denote by ylow the unique positive root of the
first equation in (6.4),and by yup the unique positive root of the second equation.
By virtue of Lemma 3.3, we can assert that
min{yup, ylow} ≤ min
{max
k=0,...n−1
k+1
√‖V+‖Nk
1 (V−), k+1
√max
k=0,...n−1‖V−‖Nk
1 (V+)
}.
Clearly, for any k < n,
k+1
√‖V+‖Nk
1 (V−) = Nk1 (V−)
k+1
√‖V+‖N1(V−)
≤ Nk1 (V−)
k+1
√N1(V+)
N1(V−).
Similarly,
k+1
√‖V−‖Nk
1 (V+) ≤ N1(V+)k+1
√N1(V−)
N1(V+).
If N1(V+) ≥ N1(V−), then
k+1
√‖V+‖Nk
1 (V−) ≤ N1(V+)n
√N1(V−)
N1(V+).
If N1(V+) ≤ N1(V−), then
k+1
√‖V−‖Nk
1 (V+) ≤ N1(V+)n
√N1(V+)
N1(V−).
Thus min{yup, ylow} ≤ qn(A), where
qn(A) =n
√min{N1(V+)Nn−1
1 (V−), N1(V−)Nn−11 (V+)} . (6.5)
Due to Corollary 3.4, for any µ ∈ Σ(A(·)), we have
ρ(D(µ)) = minj|µ2 + µθj + ψjj | ≤ qn(A) .
Com
mun
. Con
tem
p. M
ath.
200
3.05
:101
-118
. Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
MO
NA
SH U
NIV
ER
SIT
Y o
n 05
/20/
13. F
or p
erso
nal u
se o
nly.
January 23, 2003 11:33 WSPC/152-CCM 00090
114 M. I. Gil’
So the spectrum of the matrix pencil A(λ) = λ2 I + λθ + ψ lies in the set
{µ ∈ C : |µ2 + µθj + ψjj | ≤ qn(A), j = 1, 2, . . . , n} .Hence, it follows that if
| − t2 + itθj + ψjj | > qn(A) . (6.6)
for all t ∈ R and j = 1, 2, . . . , n, then A(it) is invertible, and thus operator T
defined by (6.1) is invertible. Assume that θj , ψjj are real and
ψjj < −qn(A) (j = 1, . . . , n) .
Then due to (6.6) operator T defined by (6.1) is invertible. Recall that qn(A) is
defined by (6.5) and (6.3).
7. Infinite Matrix Pencils
7.1. The general case
Let {ek}∞k=1 be an orthogonal normed basis in H, and
D(λ) = diag [ak(λ)]∞k=1 (7.1)
be the diagonal matrix in that basis, where ak(λ)(k = 1, 2, . . .) are entire functions.
In addition, V+ and V− are an upper triangular compact matrix and a lower
triangular compact one, respectively:
(V+ek, ej) = ajk for j < k, (V+ek, ej) = 0 for all j ≥ k ;
(V−ek, ej) = ajk for j > k, (V−ek, ej) = 0 for all j ≤ k . (7.2)
Put as above A(λ) = D(λ) + V+ + V−. So
(A(λ)ej , ej) = aj(λ), (A(λ)ej , ek) = ajk = const (j 6= k) .
It is simple to see that the spectrum Σ(D(·)) of the diagonal matrix D(·) is the
closure of the set of all the roots of aj(λ), j = 1, 2, . . .. Take the maximal spectral
function P (t) = {Pk}∞k=1, where Pk are defined by
Pk =k∑j=1
(., ej)ej .
Clearly, conditions (1.2) hold. Moreover,
ρ(D(λ)) ≡ infk=1,2,...
|ak(λ)| .
In particular, let V± be Hilbert–Schmidt matrices. That is,
N21 (V−) =
∞∑j=1
j−1∑k=1
|ajk|2 <∞; N21 (V+) =
∞∑j=1
∞∑k=j+1
|ajk|2 <∞ . (7.3)
Then due to Corollary 4.6, the spectrum of A(λ) is included in the set
{z ∈ C : |aj(z)| ≤ δ(A); j = 1, 2, . . .} ,where δ(A) is defined by (4.11).
Com
mun
. Con
tem
p. M
ath.
200
3.05
:101
-118
. Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
MO
NA
SH U
NIV
ER
SIT
Y o
n 05
/20/
13. F
or p
erso
nal u
se o
nly.
January 23, 2003 11:33 WSPC/152-CCM 00090
Bounds for the Spectrum of Analytic Quasinormal Operator Pencils 115
7.2. A quadratic pencil
In many cases, the Cauchy problem for the second order differential equations leads
to the pencil
A(λ) = λ2I +Bλ+ C (7.4)
where B is a normal compact operator, C is an arbitrary compact one, cf. [27].
Let {ek} be the set of normed eigenvectors of B and λk(B) be the corresponding
eigenvalues. Let
cjk = (Cek, ej) (j, k = 1, 2, . . .)
be the matrix entries of C in the basis {ek}. Then A(λ) can be written in the form
(1.1) with the notation (7.1), (7.2), where
aj(λ) = λ2 + λj(B)λ + cjj
and ajk = cjk (j 6= k). Assume that V+, V− are Hilbert–Schmidt operators. So
condition (7.3) holds. Let s1j , s2j be the roots of the equations aj(λ) = λ2+λj(B)λ+
cjj = 0. Then
D(λ) = diag [(λ− s1j)(λ− s2j)] .
Hence, Σ(D(·)) is the closure of the sequences {s1j , s2j j = 1, 2, . . .}. Moreover,
ρ(D(λ)) = infj=1,2,...
|(λ− s1j)(λ− s2j)| .
For any µ ∈ Σ(A(·)) due to Corollary 4.6 we have
infj=1,2,...
|(µ− s1j)(µ− s2j)| ≤ δ(A)
where δ(A) is defined by (4.11). So Σ(A(·)) lies in the set
{z ∈ C : |(z − s1j)(z − s2j)| ≤ δ(A); j = 1, 2, . . .} .
8. Differential Delay Equations with Fredholm Operators
Let us consider in L2[0, 1] an integro-differential equation with delay
ut(x, t) + a(x) + b(x)u(x, t− 1) + (Fu)(x, t) = 0(t ≥ 0 x ∈ [0, 1]) (8.1)
where a(x), b(x) are real bounded measurable functions, defined on [0, 1], and
(Fu)(x, t) =
∫ 1
0
K(x, s)u(s, t)ds .
Here K(·, ·) is a real Hilbert–Schmidt kernel. Equations of the type (8.1) arise in
various applications, cf. Pao [20].
By virtue of application of the Laplace transformation to Eq. (8.1) we get the
operator pencil in L2[0, 1] defined by
[A(λ)h](x) = (λI + a(x) + b(x)e−λ)h(x)
+ (Fh)(x) (x ∈ [0, 1]; h ∈ L2[0, 1]; λ ∈ C) . (8.2)
Com
mun
. Con
tem
p. M
ath.
200
3.05
:101
-118
. Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
MO
NA
SH U
NIV
ER
SIT
Y o
n 05
/20/
13. F
or p
erso
nal u
se o
nly.
January 23, 2003 11:33 WSPC/152-CCM 00090
116 M. I. Gil’
Here I denotes the unit operator in L2[0, 1]. Put
D(λ) = Iλ+ a(x) + b(x)e−λ; (V+h)(x) =
∫ 1
x
K(x, s)h(s)ds
and
(V−h)(x) =
∫ x
0
K(x, s)h(s)ds (h ∈ L2[0, 1]) .
For 0 ≤ t ≤ 1, define the maximal spectral function P (t) by
(P (t)h)(x) = 0 for t < x ≤ 1 and (P (t)h)(x) = h(x) for 0 ≤ x ≤ t .
In addition, put
P (t) = I for t > 1 and P (t) = 0 for t < 0 .
Clearly, conditions (1.2) are fulfilled. Moreover,
N21 (V+) ≡
∫ 1
0
∫ 1
x
K2(x, s) ds dx <∞ ,
N21 (V−) ≡
∫ 1
0
∫ x
0
K2(x, s) ds dx <∞ .
For any µ ∈ Σ(A(·)) due to Corollary 4.6 we have
ρ(D(µ)) ≡ infx|µ+ a(x) + b(x)e−µ| ≤ δ(A) (8.3)
where δ(A) is defined by (4.11). So in the considered case Σ(A(·)) lies in the set
{z ∈ C : |z + a(x) + b(x)e−z | ≤ δ(A); 0 ≤ x ≤ 1} .
We will say that A(λ) is a stable pencil, if its spectrum lies in open left half-plane.
Let us derive conditions, that provide the stablility of the pencil defined by
(8.2). Assume that
a(x) > |b(x)|+ δ(A) (x ∈ [0, 1]) . (8.4)
Let µ = α+ iβ ∈ Σ(A(·)) with real α, β. Then due to (8.3),
infx|α+ a(x) + b(x)e−α cosβ| ≤ δ(A) .
If α ≥ 0, then
infxα+ a(x)− |b(x)| ≤ δ(A) .
This contradicts (8.4). So the inequality α ≥ 0 is impossible. Thus, the pencil,
defined by (8.2) is stable. This means that, under condition (8.4), all the solutions
of the Eq. (8.1) tend to zero in the norm L2[0, 1].
About other bounds for the spectrum of integral and integro-differential opera-
tors, and their applications to the stability of parabolic integrodifferential equations
see also [8–10].
Com
mun
. Con
tem
p. M
ath.
200
3.05
:101
-118
. Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
MO
NA
SH U
NIV
ER
SIT
Y o
n 05
/20/
13. F
or p
erso
nal u
se o
nly.
January 23, 2003 11:33 WSPC/152-CCM 00090
Bounds for the Spectrum of Analytic Quasinormal Operator Pencils 117
Acknowledgment
I am very grateful to the referee for his very helpful remarks. The research of the
author was supported by the Israel Ministry of Secience and Technology.
References
[1] F. V. Atkinson, H. Langer and R. Mennichen, Sturm–Liouville problems withcoefficients which depend analytically on the eigenvalue parameter, Acta Sci. Math.(Szeged) 57 (1993) 25–44.
[2] M. S. Brodskii, Triangular and Jordan Representations of Linear Operators, Transl.Math. Monogr., v. 32, Amer. Math. Soc. Providence, R.I., 1971.
[3] O. Diekmann, S. A. van Gils and S. M. Verduyn Lunel, Delay Equations: Functional-,Complex- and Nonlinear Analysis, Applied Mathematical Sciences, v. 110, Springer-Verlag, Berlin, 1995.
[4] D. E. Edmunds and W. D. Evans, Spectral Theory and Differential Operators.Clarendon Press, Oxford, 1990.
[5] Y. Egorov and V. Kondratiev, Spectral Theory of Elliptic Operators. BirkhauserVerlag, Basel, 1996.
[6] H. O. Fattorini, Second Order Linear Differential Equations in Banach Spaces, North-Holland, Amsterdam, 1983.
[7] M. I. Gil’, Norm Estimations for Operator-valued Functions and Applications. MarcelDekker, Inc. New York, 1995.
[8] M. I. Gil’, Invertibility conditions and bounds for spectra of matrix integral operators,Monatsh. Math. 129 (2000) 15–24.
[9] M. I. Gil’, Stability of Fredholm type Integro-Parabolic Equations, J. Math. Anal.Appl. 244 (2000) 318–332.
[10] M. I. Gil’, Invertibility and positive invertibility conditions of integral operators inL∞, J. Integral Equations Appl. 13 (2001), 1–14.
[11] I. Gohberg, S. Goldberg and M. A. Kaashoek, Classes of Linear Operators, OperatorTheory. Advances and Applications, v. 49, Birkhauser, Basel, 1990.
[12] I. Gohberg and M. G. Krein, Theory and Applications of Volterra Operators in HilbertSpace, Trans. Math. Monographs, v. 24, Amer. Math. Soc., R.I., 1970.
[13] M. A. Kaashoek and S. M. Verduyn Lunel, Characteristic matrices and spec-tral properties of evolutionary systems, Trans. Amer. Math. Soc. 334(2) (1992)479–517.
[14] V. Kolmanovskii and A. D. Myshkis, Introduction to the Theory and Applications ofFunctional Differential Equations, Kluwer Academic Publishers, Dordrecht, 1998.
[15] H. Konig, Eigenvalue Distribution of Compact Operators, Birkhauser Verlag, Basel-Boston-Stuttgart, 1986.
[16] A. E. Lifschitz, Magnetohydrodynamics and Spectral Theory, Kluwer AcademicPublishers, Dordrecht, 1989.
[17] J. Locker, Spectral Theory of Nonselfadjoint Two Point Differential Operators. Amer.Math. Soc., Mathematical Surveys and Monographs, v. 73, R.I., 1999.
[18] A. S. Markus, Introduction to the Spectral Theory of Polynomial Operator Pencils,Amer. Math. Soc., Providence, 1988.
[19] A. M. Ostrowski, Solution of Equations in Euclidean and Banach Spaces, AcademicPress, New York, London, 1973.
[20] C. V. Pao, Nonlinear Parabolic and Elliptic Equations, Plenum Press, New York,1992.
Com
mun
. Con
tem
p. M
ath.
200
3.05
:101
-118
. Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
MO
NA
SH U
NIV
ER
SIT
Y o
n 05
/20/
13. F
or p
erso
nal u
se o
nly.
January 23, 2003 11:33 WSPC/152-CCM 00090
118 M. I. Gil’
[21] A. Pietsch, Eigenvalues and s-Numbers, Cambridge University Press, Cambridge,1987.
[22] S. Prossdorf, Linear Integral Equations, Itogi Nauki i Tekhniki, Fundamental’nyenapravleniya 27, VINITI, Moscow, 1988, in Russian.
[23] Yu. Safarov and D. Vassiliev, The Asymptotic Distribution of Eigenvalues of PartialDifferential Operators, Amer. Math. Soc., R.I., 1997.
[24] L. Rodman, An Introduction to Operator Polynomials, Birkhauser Verlag, Basel-Boston-Stuttgart, 1989.
[25] Chr. Tretter, On λ-Nonlinear Boundary Eigenvalue Problems, Akademie Verlag,Berlin, 1993.
[26] J. Wu, Theory and Applications of Partial Functional Differential Equations,Springer-Verlag, New York, 1996.
[27] S. Yakubov and Y. Yakubov, Differential Operator Equations, Chapman &Hall/CRC, Boca Raton–New York–Washington, 2000.
Com
mun
. Con
tem
p. M
ath.
200
3.05
:101
-118
. Dow
nloa
ded
from
ww
w.w
orld
scie
ntif
ic.c
omby
MO
NA
SH U
NIV
ER
SIT
Y o
n 05
/20/
13. F
or p
erso
nal u
se o
nly.