Upload
sharynsir
View
237
Download
0
Embed Size (px)
Citation preview
8/6/2019 T2 N1
1/132
i
i
ii
.2, 1
2010
i
.., .. i i,
2-i i i . . . . . . . . . . . . . . . . . . . . . . . 4
.. i i
i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
i .. i ii . . . . . . 24
i .. ii i ii ii i i i-
i i . . . . . . . . . . . . . . . . . . . . . . 35
.., i .. i i
i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
.., .. ii i 72
.I., .., .. ii i
i i .. i i ii
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
.., i .. i i
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
.., .., i ., .. i
i i- i i-
i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
i .., .. i ii i i
i i . . . . . . . . . . . . . . . . . . . . . . . . . . 109
I. ., i . . ii i
i . . . . . . 119
8/6/2019 T2 N1
2/132
Carpathian
mathematical
publications
Scientific journal
V.2, 1
2010
Contents
Voloshyn H.A., Maslyuchenko V.K. On approximation of the separately continuous
functions 2-periodical in relation to the second variable . . . . . . . . . . . . . . 4
Voloshyna T.V. An analogue of Bernsides lemma for finite inverse symmetric semi-
group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Gavrylkiv V.M. On representation of semigroups of inclusion hyperspaces . . . . . . . 24
Hryniv R.O. Analyticity and uniform stability in the inverse spectral problem for
impedance SturmLiouville operators . . . . . . . . . . . . . . . . . . . . . . . . . 35
Zagorodnyuk A.V., Kravtsiv V.V. Symmetric polynomials on the product of Banach
spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Zatorsky R.A., Semenchuk A.V. Periodic recurrent fractions of third degree . . . . . . 72
Kopach M.I., Obshta A.F., Shuvar B.A. On applications of iteration algorithms and
Skorobagatkos branching fractions to approximation of roots of polynomials in
Banach algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
Lopushansky O.V., Oleksienko M.V. A Poisson type formula for Hardy classes on
Heisenbergs group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Prykarpatsky A.K., Artemovich O.D., Popowicz Z., Pavlov M.V. Riemann type
algebraic structures and their differential-algebraic integrability analysis . . . . . . 96
Skaskiv O.B., Kuryliak A.O. Direct analogues of Wimans inequality for analytic
functions in the unit disc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Chuchman I. Ya., Gutik O. V. Topological monoids of almost monotone injective
co-finite partial selfmaps of the set of positive integers . . . . . . . . . . . . . . . . 119
8/6/2019 T2 N1
3/132
.2, 1
2010
.., .. ,
2- . . . . . . . . . . . . . . 4
..
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
.. . . . 24
..
. . . . . . . . . . . . . . . . 35
.., ..
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
.., .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
.., .., ..
..
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
.., ..
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
.., .., ., ..
- -
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
.., ..
. . . . . . . . . . . . . . . . . . . . . . . . . 109
. ., . . -
. 119
8/6/2019 T2 N1
4/132
i i Carpathian Mathematical
ii. .2, 1 Publications. V.2, .1
517.51
.., ..
I I,
2-I I I
.., .. i i, 2-
i i i // i i ii. 2010. .2,
1. C. 414.
i i , -
i X i i i i f : X R R,
2-i i i, i ii
i fn : X R R, i fx
n= fn(x, ) : R R i
i i fxn fx R x X.
i ii
ii i i [8, c. 98], : i i f : [0, 1]2 R i i-
i i fn : [0, 1]2 R, i i i ii
i , ii ii fxn = fn(x, ) x [0, 1] i-
i i fx = f(x, ) ii [0, 1]. i [1]
i Bn : C[0, 1] C[0, 1], i i g C[0, 1]
i
(Bng)(y) =n
k=0
Ckngk
n
yk(1 y)nk,
0 y 1, , i i X i i f : X [0, 1] R i
fn(x, y) = (Bnfx)(y),
x X, y [0, 1] i ii i i,
fxn fx [0, 1] x X. hn h Y ,
, ii ii hn h, hn h = supxY
|hn(y) h(y)| 0 n .
2000 Mathematics Subject Classification: 54C30, 65D15.i i : i i i, , .
c .., .., 2010
8/6/2019 T2 N1
5/132
i i 5
ii (. [2]) i i i -
i i. i , -
ii
i [8, c. 398]. i -
, i X i i i
i f : XR R, 2-i i i, i i-i i fn : XR R, if
xn : R R
i i, i fxn fx R x X. ,
i
f : X R R, i 2-i i i, , -
i i ii ii, i i
ii.
i [3].
1 i ii
i C(Y) i i g : Y R, i-
i Y, i Tp ii,
i
qy(g) = |g(y)|, y Y.
i (C(Y), Tp) Cp(Y).
Y Cu(Y) i i
(C(Y), ) ii
g = maxyY
|g(y)|,
Tu i Cu(Y), i ii ii
C(Y).
A : C(Y) C(Y) pu-, i -
i A : Cp(Y) Cu(Y). pp-, uu- i up-
i.
C2 i i 2-i i
g : R R. i i
g = maxyR
|g(y)| = max0y2
|g(y)|,
C2 i ii ii Tu. i C2 -
i i Tp ii. iTp i Tu C2 iii
i i A : C2 C2. , A : C2 C2
pu-, i i A : (C2, Tp) (C2, Tu).
S = {z C : |z| = 1} i, i -
C. i, S i, i
8/6/2019 T2 N1
6/132
6 .., ..
Cu(S) i Cp(S). i i i h C(S) i g = U h C2,
g(y) = h(eiy)
y R. i, i U : C(S) C2 -
ii, ii i Cu(S) i (C2, ) i-
ii i Cp(S) i (C2, Tp).
ii C2 i C(S).
ii C0[0, 2] C[0, 2], -
i g C[0, 2], g(0) = g(2). i R : C2 C0[0, 2],
Rg = g|[0,2] ii, ii,
C2 i C0[0, 2] i ii . , i C2 -
i i ii C0[0, 2] C[0, 2].
i f : X Y Z i (x, y) X Y fx(y) = fy(x) =
f(x, y) i (x) = fx. i : X ZY -
f i f i. i i X Y C(X Y)
CC(X Y)
i (i)
i f : X Y R.
i (., ,
[7, 2.1.2]).
1. X iY ii , f : XY R i i : X RY
i f i. i:
) f CC(X Y) (X) C(Y) i i : X Cp(X) ;
) Y i, f C(X Y) (X) C(Y) i i : X Cu(X) .
i iii f i , 1
CC(X Y) C(X, Cp(Y)) i i-
: X Cp(Y), i p-, C(X Y)
C(X, Cu(Y)) i i : X Cu(Y), i -
u-.
i X CC2(X R) i i
i i f : X R R, i 2-i i
i. ii U : C(S) C2 i CC(X S), i CC2(XR) C(XR) C(X S), i
i i 1.
2. X i i, f : XR R i i : X RR i f i. i:
) f CC2(X R) (X) C2 i : X C2 p-;
) f CC2(X R) C(X R) (X) C2 i : X C2 u-.
i T i i ii g : R R,
8/6/2019 T2 N1
7/132
i i 7
i
g(y) = a0 +n
k=1
(ak cos ky + bk sin ky).
, T ii C2.
CT(X R) i f : X R R, i
fy : X R i y R, ifx : R R ii x X. i, CT(X R) ii
CC2(X R).
2 i ii i
ii i An : C2 T T,
Ang g R i g C2.
= p, u i = p, u. ii i An : C2 T
-, i An -. A : C2 C2
A : CC2(X R) CC2(X R), i i f CC2(X R) iii i f = Af, X R
f(x, y) = (Afx)(y).
(
x) =
fx
i (x) = fx x X, f CC2(X R) i f = Af.i, i = A . i i , i u-, p- i A pu- u- i A
uu-. i, 2, .
3. ) A : C2 C2 pu- , f CC2(X R) i
f =
Af. i
f CC2(X R) C(X R).
) A : C2 C2 uu- , f CC2(X R) C(X R) if = Af. i f CC2(X R) C(X R). i .
4. (An)n=1 ii i An : C2 C2, f CC2(X R)i fn = Anf. i:
) (An)n=1 pu- ii i T, fn
CT(X R) C(X R) n i fxn fx R x X;
) f C(XR) i(An)n=1 uu- ii i T,
fn CT(X R) C(X R) n i fxn f
x R x X.
8/6/2019 T2 N1
8/132
8 .., ..
. i An(C2) T, fn CT(XR). i i
fn 3. i,
fxn = An((x)) (x) = fx R
x X, ii i An .
3
i i ii ii:
Dn(t) =sin(2n + 1) t
2
2sin t2
=1
2+
nk=1
cos kt,
i :
Kn(t) = 2n + 1
sin(n + 1) t22sin t
2
2 = 1n + 1
nk=0
Dk(t),
i . ii Dn n- -
(Sng)(x) =a0
2+
nk=1
(ak cos kx + bk sin kx)
i g C2 [8, c. 381]:
(Sng)(x) = 1
0
(g(x + t) + g(x t))Dn(t)dt,
i i [7, c.751]:
(Fng)(x) =1
n + 1
nk=0
(Skg)(x) =1
0
(g(x + t) + g(x t))Kn(t)dt.
i [7, c. 526; c. 752], Sng g i-
i g C2, i Fng
g R
i g C2. Fn . i Fng T n i
Fng g R i g C2, (Fn)n=1 ii
i i T i ii.
i i g C2:
(Fng)(x) =1
g(x + t)Kn(t)dt =1
g(y)Kn(y x)dy.
, ii i 2-i.
8/6/2019 T2 N1
9/132
i i 9
i , , Kn i i:
1) Kn(0) =n+12
;
2) Kn(t) 0 i Kn(t) = Kn(t) i t R;
3)1
K
n(t)dt = 1.
i, Fn : C2 C2 uu-.
i, Fn ii i i g C2 i i x R
|Fng(x)| 1
|g(x + t)|Kn(t)dt g
Kn(t)dt = g,
i ii Fng g, i Fn i (C2, ). i Tp Tu, Fn i up-
. , .
5. ) Fn : C2 T uu- i-i i T i ii.
) i i X i i 2-
i i i i f : X R R i
fn(x, y) = (Fnfx)(y)
, fxn T - n N i x X fxn f
x R
i x X.
i, Fn pu-.
. K C2, K(t) 0 R, K(0) > 0 i t0 R. i i
(g) =
g(t)K(t t0)dt
i (C2, Tp).
. a [, ), t0 = a + 2m i m. = K(0)
2. , 0 < < K(0). i i K i,
i > 0, a + < i K(t) , i a t a + . n
bn = a+n
i i gn C2, i ii [, ]
(, 0), (a, 0), (a + bna2
, n), (bn, 0) (, 0). ,
gn(t) 0 t R, bn a n . , i K(t t0) = K(t a)
t R,
(gn) =
gn(t)K(t t0)dt =
gn(t)K(t a)dt
8/6/2019 T2 N1
10/132
10 .., ..
=
bna
gn(t)K(t a)dt
bna
gn(t)dt = n(bn a)
2=
2> 0
n. , (gn) 0. i , i
i Tp ii.
i .
6. Fn pp-, , i pu-.
. i
(Fng)(x) =1
g(y)Kn(y x)dy,
i K = 1
Kn C2 i K(0) =n+12
> 0, i ,
x R i
x(g) = (Fng)(x) =
g(y)Kn(y x)dy
p- C2. i Fn pp-, pp-
i i p-i i ii x,
.
4
i ii [0, 2] n + 1 i xk =2kn+1
. 0 a0 x1i ak = a0 + kd k = 1, . . . , n, d =
2n+1
. a0, a1, . . . , an ii ii
ii [0, 2], ak+1 ak = d i xk ak xk+1 k = 0, 1, . . . , n.
i g C2, n i x R
(Jng)(x) =2
n + 1
nk=0
g(ak)Kn(x ak),
Kn . i Kn i, i Jng
i. , i Jn : C2 T,i, , ii. i Jn , i
Jng i ig. i, Jng g R i
g C2 [4, c.36]. i i
i i
(Jng)(x) =1
g(y)Kn(y x)dn+1(y),
n+1(y) = kd, ak < y ak+1, i d =2
n+1
.
8/6/2019 T2 N1
11/132
i i 11
7. ) Jn : C2 T pu- i-i T.
) i i X i f CC2(X R)
i
fn(x, y) = (Jnfx)(y)
i CT(X R), fxn fx R x X.
. ) i y1, . . . , yn R i i 1, . . . , n C2
(Ag)(y) =n
k=1
g(yk)k(y)
i Ag C2, i A : C2 C2, , -
, ii. g C2
Ag maxk=1,...,n
|g(yk)| = maxk=1,...,n
qyk(g),
=n
k=1
k. i , A pu- [5, .12].
i, Jn . ,
pu-i. i Jng g R g C2 i imJn T,
) .
) ) i 4.
5 i
ii pu- ii i Gn : C2
T T, i Bn : C[0, 1] C[0, 1] (. .1).
[8, c. 398].
P[a, b] i i ii ii [a, b]. Bni C[0, 1] P[0, 1] i pu-. Bnf f [0, 1]
f C[0, 1].
ii i : [0, 1] [1, 1], (x) = 2x 1, -
i U : C[1, 1] C[0, 1], U f = f . i y = (x) -
i x = 1(y) = 12
(y + 1), ii, U
U1 : C[1, 1] C[0, 1], i U1g = g 1.
Cn = U1BnU.
i i i ii, U1(P[0, 1]) P[1, 1], , im Cn P[1, 1].
U i U1 pp- i uu-, , Cn pu-
i , i i C[1, 1] P[1, 1]. i , Cnf U1U f = f
[1, 1] f C[1, 1], U1 uu-.
8/6/2019 T2 N1
12/132
12 .., ..
C+2
C2
ii i () i C2. -
(t) = arccos t. i : [1, 1] [0, ] i
i 1(x) = cos x. V f = f . V : C[0, ] C[1, 1]
. V1g = g 1. V i V1 pp-
i uu-i. i , g P[1, 1] i f = V1g, f(x) = g(cos x)
0 x . i , f - ii [0, ] , T+[0, ],
i i T C+2.
R : C2 C[0, ], Rf = f|[0,] i P : C[0, ] C2,
iii i i g C[0, ] i f C+2,
f|[0,] = g. R i P pp- i uu-, P Rf = f,
f C+2.
Dn = P V1CnV R.
Dn : C2 C2 pu-, V R : C2 C[1, 1] pp-
, Cn : C[1, 1] C[1, 1] pu- i P V1
:C[1, 1] C2 uu-.
im Dn P V1(im Cn) P V
1(P[1, 1]) P(T+[0, ]) T C+2
i f C+2
Dnf P V1V Rf = P Rf = f R.
K : C2 C+2 i L : C2 C
2,
(Kf)(x) =f(x) + f(x)
2i (Lf)(x) =
f(x) f(x)
2 i f C2 i x R. i, K + L = I, I .
K i L pp- i uu-. M : C2 C2
, N : C2 C2 ,
(M f)(x) = f(x)sin x i (N f)(x) = f(x)cos x
f C2 i x R. M N pp- i uu-i. ,
M(T) T i N(T) T, M(C2) C+2.
En = M2DnK + M DnM L.
i, En : C2 C2 pu- , im En T.
f C2 , Kf C+2 i M Lf C
+2.
Enf = M2DnKf + M DnM LfM
2Kf + MMLf = M2(Kf + Lf) = M2f R.
W : C2 C2, W f = f , (x) = x +2
, i
W1, W1g = g 1, 1(x) = x 2
. -
i, W i W1 pp- i uu-, W(T) T i W1(T)
T.
Fn = W1EnW.
8/6/2019 T2 N1
13/132
i i 13
, im Fn W1(imEn) W
1(T) T. Fn pu- i
f C2
Fnf = W1EnW fW
1M2W f R,
W1M2W f(x) = W1M2f(x + 2
) = W1[f(x + 2
)sin2 x]
= f(x
2+
2)sin2(x
2) = f(x)cos2 x = N2f(x).
,
Fnf N2f R.
i
Gn = En + Fn.
Gn pu-, En i Fn . i, im Gn T,
im En T im Fn T.i, f C2
Gnf = Enf + FnfM2f + N2f = (M2 + N2)f = f R,
M2 + N2 = I, sin2 x + cos2 x = 1 i x R.
, .
8. Gn : C2 T pu- ii T.
i , Gn - Bn. :
Gn = En + Fn
= M2DnK + M DnM L + W1M2DnKW + W
1M DnM LW
= M2P V1CnV RK + M P V1CnV R M L + W
1M2P V1CnV RK W
+W1M P V1CnV RMLW = M2P V1U1BnU V R K
+M P V1U1BnU V R M L + W1M2P V1U1BnU V R K W
+W1
M P V
1
U
1
BnUV RMLW.
En = M2P V1U1BnU V R K + M P V
1U1BnU V R M L
i
Gn = En + W1EnW.
, [7, c.698] i
i C2 i,
ii- ii ii i ,
i pu- ii i An : C2 T.
8/6/2019 T2 N1
14/132
14 .., ..
i
1. .., .. i i i i // . i.
i. -. . 2007. . 336-337. . 52-59.
2. .., .., .. i i
i // . . i i i i i.
i. I-i. 2010. . 2-3.
3. .., .. i i, 2-i i-
i // . . i i i i
i. i. I-i. 2010. . 27-28.
4. . . .2. M.: , 1965. 538 .
5. .. iii i . ii: , 2002. 72 .
6. .. i i i i // . . i.-. .
ii. 1999. 345 .
7. .. . .III. .-.: . -
.-. -, 1949. 783 .
8. .. . T 2. C.---: ,
2005. 464 .
i i i ii i ,
ii,
i 7.04.2010
Voloshyn H.A., Maslyuchenko V.K. On approximation of the separately continuous functions
2-periodical in relation to the second variable, Carpathian Mathematical Publications, 2, 1
(2010), 414.
Using Jacksons and Bernsteins operators we prove that for every topological space X and
an arbitrary separately continuous function f : X R R, 2-periodical in relation to the
second variable, there exists such sequence of jointly continuous functions fn : X R R
such that functions fxn
= fn(x, ) : R R are trigonometric polynomials and fx
n fx on R
for every x X.
.., .. , 2-
// -
. 2010. .2, 1. C. 414.
,
X f :
X R R, 2- ,
fn : X R R,
fxn
= fn(x, ) : R R fxn fx R
x X.
8/6/2019 T2 N1
15/132
i i Carpathian Mathematical
ii. .2, 1 Publications. V.2, .1
512.53
..
I I
I
.. i i i-
// i i ii. 2010. .2, 1. C. 1523.
i
i i i.
S i i. S: ab aa1 = ab1 a1a = a1b. i ii
. H H S H := {h S : H h}. H = H, H .
i i i X i i i X i - IS(X) ISn, |X| = n. i IS(X) - i X. i i iS i i i i IS(X) i i X. i i i i- i i , iii .
i ii H i i S - i H :=
(s, t) S S
st1 H. ii i (Hs), ss1 H, i i ii H. i X i H i(S) : x X i s S xH(s) = (xs). : SIS(X) i S
2000 Mathematics Subject Classification: 20M18, 20M20, 20M30.i i : , i i.
c .., 2010
8/6/2019 T2 N1
16/132
16 ..
i iiH. i [1] i, - i i S i i i i i ii.
N = {1, 2, . . . n}, ISn i i i N,S(M) i M. ISn dom i ran -
ii i . i i i , - ii . i i rank i def ii.
i i ii i i .
1 ([1]). i M N i i G S(M) iiH = G ISM i ii ISn. i ,
i ii ISn .
i ii H = G ISM |M| = k,|G| = m, [S(M) : G] = k!
m= r. ii i i ISn -
i ii H [ISn : H].
2 ([1]). i iISn i- ii H = G ISM (Hg), g S(N), i i
ii Ag S(N) i A = GS(M), iii
i i ii i A .
[ISn : H] = S(N) : G S(M) =n!
m (n k)!= Ck
n
[S(M) : G] = Ckn
r.
i i ii iii H.
, i , [3]. : ISn IS(X) i i ISn i X
i i ii H. -i 1 = {(x, y) ISn ISn|(x) = (y)} i i iISn. ii i ii i i i iii H [2].
3 ([1]). i iISn i i i ii H = G ISM i i
i i, |M| = 1. i
ISn i ISn i i
{1, 2, . . . , n}.
|(ISn| = |ISn| =ni=0
(Cin)2 i!.
1 ([2]). i ii H = G ISM, |M| = k, Ik = { ISn|rank < k} ii i
8/6/2019 T2 N1
17/132
i i i 17
1. t Ik (t) i -
i i i i ii
H = G ISM .
. k = n, M = N i H = G S(N). 1, i i ISn i i. (ISn) i .
1
() ii i .
2. i i ISn ii
ISn
() +1
n
ISn
def = |ISn| . (1)
. , |ISn| =n
k=0
Ckn2k!. 1n
ISn
def .
i i. k in k. ii i k ii ISn i
Ckn2 k!.
1
n
ISn
def =1
n
nk=0
ISn
rank=k
def =1
n
nk=0
(n k)
Ckn2
k!.
ISn
() i i. i
k Ckn . i ISn i k i i i ii. i i i L N i k i
ISn
dom=L
().
St(i) = { ISn| dom = L, (i) = i}. i
ISn
dom=L
() =iL
|St(i)| = |L| |St(i)| = k Ck1n1 (k 1)! =k
n Ckn k!.
i i L N Ckn ,
ISn
() =n
k=0
kn
Ckn2 k!.
iISn
() +1
n
ISn
def =n
k=0
k
n
Ckn2 k! +
nk=0
n k
n
Ckn2 k!
=n
k=0
k + n k
n
Ckn2 k! =
nk=0
Ckn2 k! = |ISn|.
8/6/2019 T2 N1
18/132
8/6/2019 T2 N1
19/132
i i i 19
1) , k = n i H = G S(N). i, 5 i [2], |(ISn)| = 1 + n!m = 1 + r, m = |G|, r = [S(N) : G]. 2 ii i i r. 4 i [2] i ISn i , - i i i 1.
(ISn)
() +1
[ISn : H]
(ISn)
def =
(Sn)
() +1
r r = 1 +
(Sn)
().
4 i [2], i i i ISn i i ii H i i i i i S(N) i G.
(Sn)
()
. i S(N) i i ii S(N) i G .
(Sn)
() = 1 |(Sn)| = |Sn/ G| = r.
, ii (2) .2) 1 < k < n i G = S(M). i, 5 i [2]
|(ISn)| = 1 + (Ckn)
2 +n
i=k+1
(Cin)2 i!.
2 ii i i Ckn.
i i , M Cknk i N. 1 i ISn k ii i 1. i, i Ckni i.
, i i i k. X1 , i k M M. 3 i 4 t ISn k i (t) X1 i i i i, domt = M, i i (t) i X1 ii i i, domt = rant = M. t ISn k
i (t) i i Ckn i i def (t) =
Ckn 1. i ii i 1 i i i
k, i i i i i , X1 i i (t) i (ISn), domt = rant = M.
t ISn k + 1. ISn i Ckk+1 = k + 1 k, t i . (t) Ckk+1 = k + 1 i. ii i , t ISn i, k + 1 i n, (t) Cki i Ckn C
ki .
8/6/2019 T2 N1
20/132
20 ..
ii (2) :
1
Ckn
(ISn)
def =1
Ckn
Ckn +
Ckn2
Ckn 1
+
Ck+1n2
(k + 1)!
Ckn Ckk+1
+ . . . +
C
n1
n2
(n 1)!
C
k
n C
k
n1
+ (C
n
n)
2
(n)!
C
k
n C
k
n
= 1 + (Ckn)2 +
ni=k+1
(Cin)2i! Ckn
1
Ckn
ni=k+1
(Cin)2 i! Cki .
ii
(ISn)
(). i Ckn -
i . X1. St(X1) ={ (ISn)|X
1 = X1}. X1 i (t) i (ISn),
rank t = k. 3 i [2] i ISn i k -i ii i1. t i > k i X(t)1 = X1.
i 4 (M)t = M. I k! i i t ki -i M. i i k , i Ciknk , i t , Ciknk , i i-ii i (i k)! .
|St(X1)| = 1 +n
i=k+1
(Ciknk)2 (i k)! k!.
,
(ISn)
() = Ck
n |St(X1)| = Ck
n
1 +
n
i=k+1
(Cik
nk)2
(i k)! k!
.
i ,
(ISn)
() = Ckn +1
Ckn
ni=k+1
(Cin)2 i! Cki .
, ii (2) .3) 1 < k < n i G i S(M) i r.
5 i [2]
|(ISn)| = 1 + r (Ckn)
2 +n
i=k+1
(Cin)2 i!.
ii i i r Ckn. i , M Ckn k i N, i r . i Ckn i L1, . . . , LCkn r i i.
1 i ISn k iii 1. i i r Ckni i.
8/6/2019 T2 N1
21/132
i i i 21
, i i i k. X1 , i M M. 3 i 4 t ISn k i (t) X1 i i ii, domt = M, i i (t) i X1 i, domt = rant = M. t ISn k i-
(t) i r , iLi. i k, i i i i -i , r i i ii i 1, i ii Li i i i r i (ISn). G = A(M), r = 2 i i i i i, i . , k = 4 i G = K4, ii i r = 6 i , 6 i i (ISn) i 6 i. i i (ISn), domt = rant = M.
i , t ISn i, k + 1 i n (t) r Cki i r (Ckn C
ki ).
i i ii (2) :
1
rCkn
(ISn)
def =1
rCkn
rCkn + r2(Ckn)
2(Ckn) +n
i=k+1
(Cin)2i!r(Ckn C
ki )
= 1 + r(Ckn)2 +
ni=k+1
(Cin)2i! rCkn
1
Ckn
ni=k+1
(Cin)2 i! Cki .
ii
(ISn)(). i rCkn
i . X1. St(X1) = { (ISn)|X
1 = X1}. X1 i (t) i (ISn),
rank t = k. 3 i [2] i ISn i k - i ii i 1. t i > k iX
(t)1 = X1. i 4 (M
)t = M i tMx
Gx. i t iM |Gx| = |G| = m = k!
r. i i k , i -
Ciknk , i t , Ciknk , i
iii i (i k)! .
|St(X1)| = 1 +n
i=k+1
(Ciknk)2 (i k)!
k!
r.
,
(ISn)
() = rCkn |St(X1)| = rCkn
1 +n
i=k+1
(Ciknk)2 (i k)!
k!
r
= rCkn +1
Ckn
n
i=k+1
(Cin)2 i! Cki .
8/6/2019 T2 N1
22/132
22 ..
, ii (2) .4) 1 < k < n i G i S(M)
i r. 4 i [2] |(ISn)| = 1 +ni=k
(Cin)2 i!. 2, ii
i i r Ckn. i , M Ck
nk i N, i r
. i Ckn i L1, . . . , LCkn r i i.
i, 1 , i ISn k ii i 1. i, i rCkni i.
, i i i k. X1 , i M M. 3 i 4 t ISn k i (t) X1 i i ii, domt = M, i i (t) i X1 i,
domt = rant = M. t ISn k i(t) i r , i Li.
i i i , t ISn i, k +1 i n (t) r Cki i r (C
knC
ki ).
1
rCkn
(ISn)
def =1
rCkn
rCkn + (Ckn)
2k!r(Ckn 1) +n
i=k+1
(Cin)2i!r(Ckn C
ki )
= 1 + k!(Ckn)2 +
ni=k+1
(Cin)2i! k!Ckn 1Ckn
ni=k+1
(Cin)2 i! Cki .
i k, i i i i i , k! i i i ii i1. i ii Li i r i k! i i (ISn), i k M i . i ii , S(M) i i r i. i X1 iii t ISn k, X
(t)1 = X1, i
k!r
= |G| = m. ii
i t ISn i k, i , X(t)1 = X1, i . , ii i
(ISn)
() = rCkn
m +n
i=k+1
(Ciknk)2 (i k)! m
= k!Ckn +1
Ckn
ni=k+1
(Cin)2 i! Cki .
ii (2) i .
8/6/2019 T2 N1
23/132
i i i 23
i
1. .. i i i i i-
// i i. i i. 1998. . 2. . 16-21.
2. .. i i i i i-
ISn // i i. i: i.-. . 2006. . 1. . 9-16.
3. ., . . .: , 1972. . 1. 283 . . 2.
422 .
4. .. // . . . 1962. . 28, 3.
. 164-176.
i i ii i ,
, .
i 20.05.2010
Voloshyna T.V. An analogue of Bernsides lemma for finite inverse symmetric semigroup , Car-
pathian Mathematical Publications, 2, 1 (2010), 1523.
An analogue of Bernsides lemma for transitive permutation representations of finite inverse
symmetric semigroup is obtained.
.. -
// . 2010. .2, 1. C. 1523.
.
8/6/2019 T2 N1
24/132
i i Carpathian Mathematical
ii. .2, 1 Publications. V.2, .1
512+515.12
Gavrylkiv V.M.
ON REPRESENTATION OF SEMIGROUPS OF INCLUSION
HYPERSPACES
Gavrylkiv V.M. On representation of semigroups of inclusion hyperspaces, Carpathian Mathe-
matical Publications, 2, 1 (2010), 2434.
Given a group X we study the algebraic structure of the compact right-topological semi-
group G(X) consisting of inclusion hyperspaces on X. This semigroup contains the semigroup
(X) of maximal linked systems as a closed subsemigroup. We construct a faithful represen-tation of the semigroups G(X) and (X) in the semigroup P(X)P(X) of all self-maps of the
power-set P(X). Using this representation we prove that each minimal left ideal of (X) is
topologically isomorphic to a minimal left ideal of the semigroup pTpT, where by pT we denote
the family of pretwin subsets ofX.
Introduction
After discovering a topological proof of Hindman theorem [8] (see [10, p.102], [9]), topo-logical methods become a standard tool in the modern combinatorics of numbers, see [10],
[11]. The crucial point is that any semigroup operation defined on a discrete space X can
be extended to a right-topological semigroup operation on (X), the Stone-Cech compacti-
fication of X. The extension of the operation from X to (X) can be defined by the simple
formula
A B =
A X : {x X : x1A B} A
, (1)
where A, B are ultrafilters on X. Endowed with the so-extended operation, the Stone-
Cech compactification (X) becomes a compact right-topological semigroup. The algebraic
properties of this semigroup (for example, the existence of idempotents or minimal left ideals)have important consequences in combinatorics of numbers, see [10], [11].
The Stone-Cech compactification (X) of X is the subspace of the double power-set
P(P(X)), which is a complete lattice with respect to the operations of union and intersection.
In [7] it was observed that the semigroup operation extends not only to (X) but also to the
complete sublattice G(X) ofP(P(X)) generated by (X). This complete sublattice consists
of all inclusion hyperspaces over X.
2000 Mathematics Subject Classification: 20M30; 20M12; 22A15; 22A25; 54D35.Key words and phrases: binary operation, semigroup, right-topological semigroup, representation, self-linked
set, twin set, pretwin set, minimal left ideal.
c Gavrylkiv V.M., 2010
8/6/2019 T2 N1
25/132
On representation of semigroups of inclusion hyperspaces 25
By definition, a family F of non-empty subsets of a discrete space X is called an inclusion
hyperspace if F is monotone in the sense that a subset A X belongs to F provided
A contains some set B F. Besides the operations of union and intersection, the set
G(X) possesses an important transversality operation assigning to each inclusion hyperspace
F G(X) the inclusion hyperspace
F = {A X : F F (A F = )}.
This operation is involutive in the sense that (F) = F.
It is known that the family G(X) of inclusion hyperspaces on X is closed in the double
power-set P(P(X)) = {0, 1}P(X) endowed with the natural product topology. The induced
topology on G(X) can be described directly: it is generated by the sub-base consisting of
the sets
U+ = {F G(X) : U F } and U = {F G(X) : U F}
where U runs over subsets of X. Endowed with this topology, G(X) becomes a Hausdorff
supercompact space. The latter means that each cover of G(X) by the sub-basic sets hasa 2-element subcover. Let also N2(X) = {A G(X) : A A
} denote the family of all
linked inclusion hyperspaces on X and (X) = {F G(X) : F = F} the family of all
maximal linked systems on X.
By [6], both the subspaces (X) and N2(X) are closed in the space G(X). Observe that
U+ (X) = U (X) and hence the topology on (X) is generated by the sub-basis
consisting of the sets
U = {A (X) : U A}, U X.
The extension of a binary operation from X to G(X) can be defined in the same
manner as for ultrafilters, i.e., by the formula (1) applied to any two inclusion hyperspacesA, B G(X). In [7] it was shown that for an associative binary operation on X the space
G(X) endowed with the extended operation becomes a compact right-topological semigroup.
The structure of this semigroup was studied in details in [7]. In particular, it was shown
that for each group X the minimal left ideals of G(X) are singletons containing invariant
inclusion hyperspaces. Besides the Stone-Cech extension, the semigroup G(X) contains
many important spaces as closed subsemigroups. In particular, the space (X) of maximal
linked systems on X is a closed subsemigroup of G(X). The space (X) is well-known in
General and Categorial Topology as the superextension of X, see [12].
We call an inclusion hyperspace A G(X) invariant if xA = A for all x X. It follows
from the definition of the topology on G(X) that the set
G(X) of all invariant inclusion
hyperspaces is closed and non-empty in G(X). Moreover, the set
G(X) coincides with the
minimal ideal of G(X), which is a closed semigroup of right zeros. The latter means that
A B = B for all A, B
G(X).
The minimal ideal
G(X) contains the closed subset
N2(X) = N2(X)
G(X) of invariant
linked systems on X. The subset max
N2(X) of maximal invariant linked systems on X is
denoted by
(X). It can be shown that
(X) is a closed subsemigroup of
N2(X). By [2,
2.2], this semigroup has cardinality |
(X)| = 22|X|
for every infinite group X.
8/6/2019 T2 N1
26/132
26 Gavrylkiv V.M.
The thorough study of algebraic properties of semigroups of inclusion hyperspaces and
the superextensions of groups was started in [7] and continued in [1], [2] and [3]. In this
paper we construct a faithful representation of the semigroups G(X) and (X) in the semi-
group P(X)P(X) of all self-maps of the power-set P(X) and show that the image of (X) in
P(X)P(X) coincides with the semigroup (X,P(X)) of all functions f : P(X) P(X) that
are equivariant, monotone and symmetric in the sense that f(X \ A) = X \ f(A) for allA X. Using this representation we prove that each minimal left ideal of (X) is topo-
logically isomorphic to a minimal left ideal of the semigroup pTpT, where by pT we denote
the family of pretwin subsets of X. A subset A of a group X is called a pretwin subset if
xA X\ A yA for some x, y X.
1 Right-topological semigroups
In this section we recall some information from [10] related to right-topological semi-
groups. By definition, a right-topological semigroup is a topological space S endowed with
a semigroup operation : S S S such that for every a S the right shift ra : S S,
ra : x x a, is continuous. If the semigroup operation : S S S is (separately)
continuous, then (S, ) is a (semi-)topological semigroup.
From now on, S is a compact Hausdorff right-topological semigroup. We shall recall some
known information concerning ideals in S, see [10].
A non-empty subset I of S is called a left (resp. right) ideal if SI I (resp. IS I). If
I is both a left and right ideal in S, then I is called an ideal in S. Observe that for every
x S the set SxS = {sxt : s, t S} (resp. Sx = {sx : s S}, xS = {xs : s S}) is
an ideal (resp. left ideal, right ideal) ideal in S. Such an ideal is called principal. An ideal
I S is called minimalif any ideal ofS that lies in I coincides with I. By analogy we defineminimal left and right ideals ofS. It is easy to see that each minimal left (resp. right) ideal
I is principal. Moreover, I = Sx (resp. I = xS) for each x I. This simple observation
implies that each minimal left ideal in S, being principal, is closed in S. By [10, 2.6], each
left ideal in S contains a minimal left ideal.
We shall use the following known fact, see [3, Lemma 1.1].
Proposition 1.1. If a homomorphism h : S S between two semigroups is injective on
some minimal left ideal of S, then h is injective on each minimal left ideal of S.
2 The function representation of the semigroup G(X)
In this section given a group X we introduce the function representation : G(X)
P(X)P(X) of the semigroup G(X) in the semigroup P(X)P(X) of all self-maps of the power-
set P(X) of X. The semigroup P(X)P(X) endowed with the Tychonov product topol-
ogy is a compact right-topological semigroup naturally homeomorphic to the Cantor cube
({0, 1}X)P(X) = {0, 1}XP(X). The sub-base of the topology ofP(X)P(X) consists of the sets
x, A+ = {f P(X)P(X) : x f(A)},
x, A = {f P(X)P(X) : x / f(A)}.
8/6/2019 T2 N1
27/132
On representation of semigroups of inclusion hyperspaces 27
Given an inclusion hyperspace A G(X) consider the function
A : P(X) P(X), A(A) = {x G : x1A A}
called the function representation of A.
Proposition 2.1. A function : P(X) P(X) coincides with the function representationA of some (invariant) inclusion hyperspace A G(X) if and only if is
1) equivariant in the sense that (xA) = x(A) for any A X and x X;
2) monotone in the sense that (A) (B) for any subsets A B of X;
3) () = , (X) = X (and (P(X)) {, X}).
Proof. To prove the only if part, take any inclusion hyperspace A G(X) and consider
its function representation A.
It is equivariant because
A(xA) = {y X : y1xA A} = {xy : y1A A} = x A(A)
for any x X and A X.
Also it is monotone because
A(A) = {x G : x1A A} {x G : x1B A} = A(B)
for any subsets A B of X.
It is clear that A() = and A(X) = X.
If A is invariant, then for every A A we get A(A) = X and for each A P(X) \ A
we get A(A) = .
To prove the if part, fix any equivariant monotone map : P(X) P(X) with () =
and (X) = X and observe that the family
A = {x1A : A X, x (A)}
is an inclusion hyperspace with A = . If (P(X)) {, X}, then the inclusion hyper-
space A is invariant.
Remark 2.1. If X is a left-topological group and A is the filter of neighborhoods of the
identity element e of X, then the functional representations A and A have transparent
topological interpretations: for any subset A X the set A(A) coincides with the interior
of a set A X while A(A) with the closure of A in X!
The correspondence : A A determines a map : G(X) P(X)P(X) called the
function representation of the semigroup G(X).
Theorem 1. The function representation : G(X) P(X)P(X) is a continuous injective
semigroup homomorphism.
8/6/2019 T2 N1
28/132
28 Gavrylkiv V.M.
Proof. To check that is a semigroup homomorphism, take any two inclusion hyperspaces
X, Y G(X) and let Z = X Y. We need to check that Z(A) = X Y(A) for every
A X. Observe that
Z(A) = {z G : z1A Z} = {z G : {x G : x1z1A Y} X } =
= {z G : Y(z1A) X } = {z G : z1Y(A) X } = X(Y(A)).
To see that is injective, take any two distinct inclusion hyperspaces X, Y G(X).
Without loss of generality, X \ Y contains some set A X. It follows that e X(A) but
e / Y(A) and hence X = Y.
To prove that : G(X) P(X)P(X) is continuous we first define a convenient sub-base of
the topology on the spaces P(X) and P(X)P(X). The product topology ofP(X) is generated
by the sub-base consisting of the sets
x+ = {A X : x A} and x = {A X : x / A}
where x X. On the other hand, the product topology on P(X)P(X) is generated by thesub-base consisting of the sets
x, A+ = {f P(X)P(X) : x f(A)} and x, A = {f P(X)P(X) : x / f(A)}
where A P(X) and x X.
Now observe that the preimage
1(x, A+) = {A G(X) : x A(A)} = {A G(X) : x1A A} = (x1A)+
is open in G(X). The same is true for the preimage
1
(x, A
) = {A G(X) : x / A(A)} = {A G(X) : x1
A / A} = (X\ x1
A)
which also is open in G(X).
3 The semigroup (X, P(X)) and its projections (X, F)
Since for a group X the function representation : G(X) P(X)P(X) is an isomorphic
embedding, instead of the semigroup (X) we can study its isomorphic copy (X,P(X)) =
((X)) P(X)P(X). Our strategy is to study (X, P(X)) via its projections (X, F) onto
the faces P(X)F of the cube P(X)P(X), where F is a suitable subfamily ofP(X).
Given a subfamily F P(X) byprF : P(X)
P(X) P(X)F, prF : f f|F,
we denote the projection ofP(X)P(X) onto its F-face P(X)F. Let
F = prF : (X) P(X)F
and
(X, F) = F((X)) = prF((X,P(X)) = (prF )((X)).
Now we detect functions f : F P(X) belonging to the image (X, F). Let us call a
family F P(X)
8/6/2019 T2 N1
29/132
On representation of semigroups of inclusion hyperspaces 29
X-invariant if xF F for every F F and every x X;
symmetric if for each A F we get X\ A F.
Theorem 2. A function f : F P(X) defined on a symmetric X-invariant subfamily
F P(X) belongs to the image (X, F) = F((X)) if and only if
1) f is equivariant;
2) f is monotone;
3) f is symmetric in the sense that f(X\ A) = X\ f(A) for each A F.
Proof. To prove the only if part, take any maximal linked system L (X) and consider
its function representation f = L : P(X) P(X).
By Proposition 2.1, the function f is equivariant and monotone. Consequently, the
restriction f|F satisfies the items (1), (2). To prove the third item, take any set A F andobserve that
f(X\ A) = {x X : x1(X\ A) L} = {x X : X\ x1A L} =
= {x X : x1A / L} = X\ {x X : x1A L} = X\ f(A).
This completes the proof of the only if part.
To prove the if part, take any function f : F P(X) satisfying the conditions 1)3)
and consider the family
Lf = {x1A : A F, x f(A)}.
We claim that this family is linked. Assuming the converse, find two sets A, B F andtwo points x f(A) and y f(B) with x1A y1B = . Then yx1A X \ B and
hence yx1f(A) f(X \ B) = X \ f(B) by the properties 1)3) of the map f. Then
x1f(A) X \ y1f(B), which is not possible because the neutral element e of the group
X belongs to x1f(A) y1f(B).
Enlarge the linked family Lf to a maximal linked family L (X). We claim that
L|F = f. Indeed, take any set A F and observe that
f(A) {x X : x1A Lf} {x X : x1A L} = L(A).
To prove the reverse inclusion, observe that for any x X \ f(A) = f(X \ A) we get
x1(X\ A) = X\ x1A Lf L. Since L is linked, x1A / L and hence x / L(A).
A subfamily F P(X) is called -incomparable if for any subset A, B F the inclusion
A B implies the equality A = B. In this case each function f : F P(X) is monotone,
so the characterization Theorem 2 simplifies as follows.
Corollary 3.1. A function f : F P(X) defined on a -incomparable symmetric X-
invariant subfamilyF P(X) belongs to the image (X, F) = F((X)) if and only if f is
equivariant and symmetric.
8/6/2019 T2 N1
30/132
30 Gavrylkiv V.M.
A subfamily F P(X) is called -invariantifL(F) F for every maximal linked system
L (X). In this case (X, F) FF is a subsemigroup of the right-topological group FF of
all self-maps ofF.
Now we see that Theorem 1 implies
Proposition 3.1. For any -invariant subfamilyF P(X) the map
F = prF : (X) (X, F) FF
is a continuous semigroup homomorphism and (X, F) is a compact right-topological semi-
group.
4 Self-linked sets in groups
Our strategy in studying minimal left ideals of the semigroup (X) consists in findinga relatively small -invariant subfamily F P(X) such that the function representation
F : (X) (X, F) is injective on some (equivalently all) minimal left ideals of (X).
The first step in finding such a family F is to consider the family of self-linked sets in X.
Definition 4.1. A subset A of a group X is self-linked if xA yA = for all x, y X.
Self-linked sets in (finite) groups were studied in details in [1]. The following simple
characterization can be easily derived from the definitions.
Proposition 4.1. For a subset A X the following conditions are equivalent:
1) A is self-linked;
2) the family of shifts {xA : x X} is linked;
3) AA1 = X;
4) A belongs to an invariant linked system A
N2(X);
5) A belongs to a maximal invariant linked system A
(X) = max
N2(X).
The following proposition was first proved in [3, 4.1]. Here we present a short proof for
completeness.
Proposition 4.2. For any invariant linked system L0
N2(X) the upper set
L0 = {L (X) : L L0}
is a closed left ideal in (X).
8/6/2019 T2 N1
31/132
On representation of semigroups of inclusion hyperspaces 31
Proof. Let A, B (X) be maximal linked systems with L0 B. Then for every subset
L L0 we get
L =
xX
x(x1L) A L0 A B
which means that L0 A B.
To show that L0 is closed in (X), take any maximal linked system L (X) \ L0and find a set A L0 with A / L. Since L is maximal linked, X \ A L. Consequently,
(X\ A) is an open neighborhood of L that does not intersect L0.
Observe that any linked system L N2(X) extending an invariant linked system L0
N2(X) lies in the inclusion hyperspace L0 . It turns out that sets from L
0 \ L0 have a specific
structure described in the following theorem.
Theorem 3. For any maximal invariant linked system L0
(X) and any A L0 \ L0there are points a, b X such that aA X\ A bA.
Proof. Fix a subset A L0 \ L0. We claim that
aA A = (2)
for some a X. Assuming the converse, we would conclude that the family {xA : x X}
is linked and then the invariant linked system L0 {xA : x X} is strictly larger than L0,
which impossible because of the maximality of L0.
Next, we find b X with
A bA = X. (3)
Assuming that no such a point b exist, we conclude that for any x, y X the union xAyA =X. Then (X \ xA) (X \ yA) = X \ (xA yA) = , which means that the family
{X \ xA : x X} is linked and invariant. We claim that X \ A L0 . Assuming the
converse, we would conclude that X \ A misses some set L L0. Then L A and hence
A L0 which is not the case. Thus X\ A L0 and hence {X\ xA : x X} L
0 because
L0 is invariant. Since L0 {X \ xA : x X} is an invariant linked system containing L0,
the maximality of L0 guarantees that G \ A L0 which contradicts A L0 .
Unifying the equalities (2) and (3) we get the required inclusions
aA X\ A bA.
5 Twin and pretwin sets in groups
Having in mind the sets appearing in Theorem 3 we introduce the following two notions.
Definition 5.1. A subset A of a group X is called
a twin subset if X\ A = xA for some x X;
a pretwin subset if xA X\ A yA for some x, y X.
8/6/2019 T2 N1
32/132
32 Gavrylkiv V.M.
By T and pT we denote the families of twin and pretwin subsets of X, respectively.
Proposition 5.1. The familiespT andT are -invariant.
Proof. Take any maximal linked system L (X) and consider its function representation
f = L : P(X) P(X), which is equivariant, monotone, and symmetric according to
Theorem 2.To show that the family pT is -invariant, take any pretwin set A pT and find two
points x, y X with xA X \ A yA. Applying to those inequalities the monotone
equivariant symmetric function f we get
xf(A) = f(xA) f(X\ A) = X\ f(A) f(yA) = yf(A),
which means that f(A) is pretwin.
If a set A is twin, then X \ A = xA for some x X and then X \ f(A) = f(X \ A) =
f(xA) = xf(A), which means that f(A) is a twin set.
Propositions 5.1 and 3.1 imply that (X,T) and (X, pT) both are compact right-
topological semigroups. The importance of the family pT is explained by the following
Theorem 4. For every maximal invariant linked system L0
(X) the restriction pT|L0 :
L0 (X, pT) is a topological isomorphism of the compact right-topological semigroups.
Proof. Since pT is continuous and the semigroups (X) and (X, pT) are compact. It
suffices to check that the restriction pT|L0 is bijective.
To show that it is surjective, take any function f (X, pT), which is equivariant,
monotone, and symmetric according to Theorem 2.By the proof of Theorem 2, the family
Lf = {x1A : A pT, x f(A)}
is linked. We claim that so is the family L0 Lf. Assuming the opposite we could find
disjoint sets A Lf and B L0. Since A is pretwin, xA X \ A yA for some x, y X.
Now we see that
B X\ A yA X\ yB,
which is not possible as B is self-linked and hence meets its shift yB.
Now extend the linked family L0 Lf to a maximal linked family L (X) and showthat L|pT = f (repeating the argument of the proof of Theorem 2).
Next, we show that the restriction pT|L0 is injective. Take any two distinct maximal
linked systems X, Y L0. It follows that there is a set A X \ Y . This set belongs to
L0 \L0 and hence is pretwin by Theorem 3. Now the definition of the function representation
yields that e X(A) \ Y(A), witnessing that pT(X) = pT(Y).
Since the function representation pT is injective on the left ideal L0 of (X), it is
injective on some minimal left ideal of (X) and hence is injective on each minimal left ideal
of (X), see Proposition 1.1. In such a way we prove
8/6/2019 T2 N1
33/132
On representation of semigroups of inclusion hyperspaces 33
Corollary 5.1. The function representation pT : (X) (X, pT) is injective on each
minimal left ideal of (X). Consequently, each minimal left ideal of (X) is topologically
isomorphic to a minimal left ideal of the semigroup (X, pT).
6 Acknowledgments
The author express his sincere thanks to Taras Banakh for help during preparation of
the paper.
References
1. Banakh T., Gavrylkiv V., Nykyforchyn O. Algebra in superextensions of groups, I: zeros and commuta-
tivity, Algebra Discrete Math, 3 (2008), 1-29.
2. Banakh T., Gavrylkiv V. Algebra in superextension of groups, II: cancelativity and centers, Algebra
Discrete Math, 4 (2008), 1-14.
3. Banakh T., Gavrylkiv V. Algebra in the superextensions of groups, III: minimal left ideals, Mat. Stud.,
31, 2 (2009), 142-148.
4. Bilyeu R.G., Lau A. Representations into the hyperspace of a compact group, Semigroup Forum 13 (1977),
267-270.
5. Engelking R. General Topology, PWN, Warsaw, 1977.
6. Gavrylkiv V. The spaces of inclusion hyperspaces over noncompact spaces, Mat. Stud., 28, 1 (2007),
92-110.
7. Gavrylkiv V. Right-topological semigroup operations on inclusion hyperspaces, Mat. Stud., 29, 1 (2008),
18-34.
8. Hindman N., Finite sums from sequences within cells of partition of N, J. Combin. Theory Ser. A 17
(1974), 1-11.
9. Hindman N., Ultrafilters and combinatorial number theory, Lecture Notes in Math. 751 (1979), 49-184.
10. Hindman N., Strauss D. Algebra in the Stone-Cech compactification, de Gruyter, Berlin, New York,
1998.
11. Protasov I. Combinatorics of Numbers, VNTL, Lviv, 1997.
12. Teleiko A., Zarichnyi M. Categorical Topology of Compact Hausdorff Spaces, VNTL, Lviv, 1999.
13. Trnkova V. On a representation of commutative semigroups, Semigroup Forum, 10, 3 (1975), 203-214.
Vasyl Stefanyk Precarpathian National University,
Ivano-Frankivsk, Ukraine.
Received 25.05.2010
8/6/2019 T2 N1
34/132
34 Gavrylkiv V.M.
i .. i ii // i -
i ii. 2010. .2, 1. C. 2434.
i i i
G(X), i i ii i X. i
i i (X) i ii.
i G(X) (X) ii P(X)P(X) i i-
i- P(X) . , ii i i i (X) i i ii i
i i pTpT.
.. // -
. 2010. .2, 1. C. 2434.
-
G(X), X.
(X)
. G(X) (X)
P(X)P(X) - P(X) .
, (X) pTpT.
8/6/2019 T2 N1
35/132
i i Carpathian Mathematical
ii. .2, 1 Publications. V.2, .1
517.98
Hryniv R.O.
ANALYTICITY AND UNIFORM STABILITY IN THE INVERSE
SPECTRAL PROBLEM FOR IMPEDANCE STURMLIOUVILLE
OPERATORS
Hryniv R.O. Analyticity and uniform stability in the inverse spectral problem for impedance
SturmLiouville operators, Carpathian Mathematical Publications, 2, 1 (2010), 3558.
We prove that the inverse spectral mapping reconstructing the impedance function of theSturmLiouville operators on [0, 1] in impedance form from their spectral data (two spectra or
one spectrum and the corresponding norming constants) is analytic and uniformly stable in a
certain sense.
1 Introduction
The main goal of this paper is to establish analyticity and uniform continuity of solutions
to the inverse spectral problems for a certain class of SturmLiouville operators on [0, 1] inthe so-called impedance form. Namely, the spectral problems of interest are
(a2(x)y(x)) = a2(x)y(x), x [0, 1], (1)
subject to suitable boundary conditions, e.g., the Neumann ones
y(0) = y(1) = 0 (2)
or NeumannDirichlet ones
y(0) = y(1) = 0. (3)
Here a > 0 is an impedance function, which will be supposed to belong to the Sobolev space
W12 (0, 1), so that the logarithmic derivative := (log a) (called the logarithmic impedance
below) is in L2(0, 1). Without loss of generality we may assume that a(0) = 1, so that
a(x) = expx
0(s) ds
. Such spectral problems arise in many applications, e.g., in modelling
propagation of sound waves in a duct [44], torsional vibrations of the earth [17] or longitudinal
vibrations in a thin straight rod [13].
2000 Mathematics Subject Classification: Primary 34A55, Secondary 34L40, 47E05.Key words and phrases: inverse spectral problems, SturmLiouville operators, impedance, analyticity, uni-
form stability.
c Hryniv R.O., 2010
8/6/2019 T2 N1
36/132
36 Hryniv R.O.
The corresponding differential operators SN and SD given by the differential expression
(y) := a2(a2y) and boundary conditions (2) and (3) respectively are self-adjoint in the
weighted Hilbert space L2
(0, 1); a2 dx
and have simple discrete spectra accumulating at
+. We denote by 0 = 0 < 1 < the eigenvalues of SN and by 0 < 0 < 1 < those of SD. The inverse spectral problem is to reconstruct the impedance function a or its
logarithm from the spectra of SN and/or SD.For the standard SturmLiouville operators, i.e., those generated by the differential ex-
pression
d2
dx2+ q,
with q a real-valued locally integrable potential, it was proved by Borg [7] in 1946 that,
generically, knowledge of the spectrum corresponding to one set of boundary conditions (e.g.
Neumann ones or NeumannDirichlet ones) does not allow to unambiguously determine q.
(An exceptional situation where this is possible was pointed out by Ambartzumyan [5] in
1929.) However, two such spectra do uniquely determine q.
The same holds true for the inverse spectral problem of reconstructing the impedance
function a of the operators SN or SD. In fact, these operators are unitarily equivalent to self-
adjoint operators TN and TD acting in L2(0, 1) and generated by the differential expression
() := 1a
d
dxa2
d
dx
1
a=
ddx
+ d
dx
(4)
and the boundary conditions
y[1](0) = y[1](1) = 0 (5)
and y[1](0) = y(1) = 0 (6)
respectively. Here and hereafter f[1](x) := f(x) (x)f(x) shall denote the quasi-derivativeof a function f. Moreover, for a W22 (0, 1) the differential expression () can be recast inthe potential form
() = d2
dx2+ + 2
with potential q = + 2. For a W12 (0, 1) the reduction to the potential form is stillpossible, but the potential q becomes a distribution from W12 (0, 1) [39]. SturmLiouville and
Schrodinger operators with singular potentials (that are, e.g., point interactions, measures,or distributions) have been widely studied; we refer the reader, e.g., to the books [1, 3]
and to review paper [40] where additional references can be found. Inverse problems for
distributional potentials in the space W12 (0, 1) have also been successfully treated; see,
e.g., [24,41].
This suggests the following method of solving the inverse spectral problem for impe-
dance SturmLiouville operators under consideration: first, one recasts the problem (1) in
the potential form, then uses one of the algorithms reconstructing the potential q from
the spectral data
(n), (n)
of TN and TD, and, finally, finds by solving the Riccati
differential equation + 2 = q. However, this equation may not possess global solutions on
8/6/2019 T2 N1
37/132
Uniform stability in the inverse problem 37
[0, 1], whence it is desirable to find a way to reconstruct the impedance a or its logarithmic
derivative directly from the spectral data for the operators TN and TD.
In the papers [2, 6, 8, 32, 35] several approaches to reconstruction of the impedance a W12 (0, 1) were suggested and the corresponding spectral data were completely described.
These necessary and sufficient conditions require that the spectra (n) and (n) must
(i) interlace, i.e., that n < n < n+1 for all n Z+, and
(ii) satisfy the asymptotic relationsn = n + 2n,
n = (n +
12
) + 2n+1,
where the sequence (n) belongs to 2.
Moreover, the induced mapping from the spectral data
(n), (n)
into the impedance
function a providing a solution to the inverse spectral problem was shown in [6] and [32] to
be locally continuous in a certain sense. In particular, this yields local stability of the inversespectral problem; see also similar stability results for the related problem of reconstructing
the potential q in [4, 7, 16, 1921, 31, 33, 34, 3638, 46]. Here we introduce a metric on the
set of the spectral data
(n), (n)
by e.g. identifying such data with the sequence (n) in
the representation of item (ii) above. Typically, this local stability states that, for a fixed
M > 0, there are positive and L with the following property: if potentials q1 and q2(resp., logarithmic impedances 1 and 2) are such that q1 M and q2 M (resp.,1 M and 2 M) and the corresponding spectral data 1 :=
(1,n), (1,n)
and
2 :=
(2,n), (2,n)
satisfy 1 2 , then
q1 q2 L1 2 (7)(resp., then
1 2 L1 2) (8)for a suitable norm . For instance, local stability results with respect to the L2(0, 1)-normwere established in [32, 38] in the regular case q L2(0, 1), and in [6,8,32] for impedanceSturmLiouville operators. In [16,33] the case L(0, 1) was treated; earlier Hochstadt in [20,
21] proved stability if only finitely many eigenvalues in one spectrum are changed. The
papers [19,36] studied to what extent only finitely many eigenvalues in one or both spectra
determine the potential, and the latter problem in the non-self-adjoint setting was recentlydiscussed in [31]. Also, stability of the inverse spectral problems on semi-axis was proved
in [30,37], and the inverse scattering problem on the line was studied in [10,18].
However, the above results cannot be considered satisfactory, as they refer to the norm
of the potential q (resp. of the logarithmic impedance ) to be recovered and thus specify
neither the allowed noise level nor the Lipschitz constant L. Therefore we need a global
stability result that asserts (7) whenever the spectral data 1 and 2 run through bounded
sets N and with L only depending on N.
Recently, such a uniform stability in the inverse spectral problem for SturmLiouville
operators on [0, 1] was established by Shkalikov and Savchuk [43]. They considered operators
8/6/2019 T2 N1
38/132
38 Hryniv R.O.
with real-valued potentials from the Sobolev spaces Ws2 (0, 1) with s > 1. (For negatives, such potentials are distributions; see [40] for the review on SturmLiouville operators
with distributional potentials.) Their approach for solving the inverse spectral problem was
based on the so called Prufer angle and used extensively the implicit function theorem. In our
work [22] analyticity and global stability of the inverse spectral mapping for s [1, 0] wasestablished using a different approach that generalizes the classical method due to Gelfandand Levitan [12] and Marchenko [29] and has been successfully applied to reconstruction of
SturmLiouville operators with singular potentials in [24,25].
The main aim of this paper is to prove analyticity and Lipschitz continuity on bounded
subsets of the inverse spectral mapping
(n), (n) for the class of the SturmLiouville
operators in impedance form with logarithmic impedance L2(0, 1). To this end we usethe approach of [2] to the inverse spectral problem for impedance SturmLiouville operators
based on the Krein equation [27] and further develop the methods of [22]. Also, we discuss the
analogous properties in the inverse spectral problem of reconstruction of from the Neumann
spectrum (n) and the corresponding norming constants n defined in Subsection 2.1.
We mention that the methods of [2] could be used to treat logarithmic impedances
belonging to Lp(0, 1) with p [1, ). However, apart from some technicalities caused bymore complicated properties of the Fourier transform in Lp(0, 1) for p = 2, the approachwould remain the same and we decided to sacrifice the generality to simplicity of presentation.
See Section 5 for discussion of possible generalizations.
The paper is organised as follows. In the next section, we state the main results of the
paper and recall the method of reconstructing the impedance SturmLiouville operators from
their spectral data using the GelfandLevitanMarchenko and Krein equations. In Section 3,
we show analyticity and uniform continuity in the inverse problem of reconstructing the
logarithmic impedance from the spectrum of the operator TN() and the sequence of thecorresponding norming constants. Reconstruction from two spectra (those of TN() and
TD()) is discussed in Section 4; there the problem is reduced to the one studied in Section 3
by showing that the norming constants depend analytically and Lipschitz continuously on
these spectra. The last Section 5 discusses some ways of extending the results to a wider class
of operators. Finally, three appendices contain auxiliary results on some related nonlinear
mappings in L2(0, 1), on relation between some analytic functions of sine type and their
zeros, and on the special Banach algebra that were used in the proofs.
2 Preliminaries and main results
In this section we state the main results of the paper and recall the method of solution of
the inverse spectral problem based on the GelfandLevitanMarchenko [28] and Krein [27]
equations. All the missing details can be found in [2].
2.1 Spectral data
Throughout this subsection, designates a fixed real-valued function in L2(0, 1). We denote
by n and n, nZ+, the eigenvalues of the operators TN() and TD() respectively defined
8/6/2019 T2 N1
39/132
Uniform stability in the inverse problem 39
via (4)(6) and recall that these eigenvalues interlace, i.e., n < n < n+1 for all n Z+,and satisfy the relations
n = n + 2n,
n = (n +
12
) + 2n+1 (9)
with some 2(Z+)-sequence = (n).
For = 2 C, the equation ()u = 2u subject to the initial conditions u(0) = 1 andu[1](0) = 0 has the solution
c(x, ) = cos x +
x0
k(x, t)cos tdt, (10)
where k is the kernel of the so called transformation operator. Clearly, cos x is a solution
of the unperturbed equation (0)u = 2u with = 0; it is mapped into the solution c(, )for a generic by means of the transformation operator via (10). The function k vanishes
for a.e. (x, t) [0, 1]2
with x < t and, for every x [0, 1], k(x, ) belongs to L2(0, 1) andthe mapping x k(x, ) is continuous from [0, 1] into L2(0, 1). Also, there exists a kernel k1with similar properties such that
c[1](x, ) = sin x x0
k1(x, t)sin tdt; (11)
we recall that f[1] := f f is the quasi-derivative of a function f.Set 2n :=
n and 2n+1 :=
n, n Z+. Then c(, 2n) is an eigenfunction of
the operator TN() corresponding to the eigenvalue n = 22n, and we call the number
1
n := 1/(2c(, 2n)2
) the norming constant for this eigenvalue. It is known [2] that
n = 1 + n, (12)
where the sequence := (n)nZ+ belongs to 2. Moreover, the norming constants ncan be determined from the spectra of the operators TN() and TD() as follows. We set
C() := c(1, ) and S() := c[1](1, ); due to (10) and (11) these are entire functions of
exponential type 1 with zeros n and
n respectively. The Hadamard canonical
products for S and C are
S() = 2
n=1
22n 22n2
, C() =
n=0
2n+1 22(n + 1
2)2
, (13)
so that S and C are uniquely determined by their zeros. Then we have (cf. [2])
n = 2nS(2n)C(2n)
, (14)
where the dot denotes the derivative in .
Here and hereafter,f
shall stand for the L2(0, 1)-norm of a function f.
8/6/2019 T2 N1
40/132
40 Hryniv R.O.
2.2 The main results
We introduce the set N of pairs
(n)nZ+, (n)nZ+
with the following properties:
the sequences (n) and (n) strictly interlace, i.e., n < n < n+1 for all n Z+;
the sequence := (k)kZ+, with 2n :=
n n and 2n+1 := n (n +1
2),belongs to 2.
In this way every element :=
(n), (n)
ofN is identified with a sequence (n) in 2thus inducing a metric on N. Namely, if1 and 2 are elements ofN and 1 := (1,n) and
2 := (2,n) are the corresponding 2-sequences of remainders, then
distN(1,2) := 1 22.
In what follows, 0 shall stand for the element ofN corresponding to = 0; then we get
distN(,0) =
(n)
2 .
According to [2], every element ofN gives the eigenvalue sequences of the operators
TN() and TD() for a unique real-valued function L2(0, 1) and, conversely, for everyreal-valued L2(0, 1) the spectra of the corresponding SturmLiouville operators TN()and TD() form an element ofN. When the logarithmic impedance varies over a bounded
subset of L2(0, 1), then the corresponding spectral data
(n), (n)
remain in a bounded
subset ofN. Moreover, the Prufer angle technique (cf. [41,42]) yields then a positive d such
that all the corresponding spectral data
(n), (n)
are d-separated, i.e., that n n dand n+1 n d for every n Z+. Summarizing, we conclude that the uniform stabilityof the inverse spectral problem we would like to establish is only possible on bounded sets
of spectral data in N that are d-separated for some d > 0.This motivates the following definition.
Definition 2.1. For d (0, /2) and r > 0, we denote byN(d, r) the set of all Nthat are d-separated and satisfy distN(,0) r.
In these notations, the first main result of the paper reads as follows.
Theorem 1. For every d (0, /2) and r > 0, the inverse spectral mapping
N(d, r)
L2(0, 1) (15)
is analytic and Lipschitz continuous.
See [9] for analyticity of mapping between Banach spaces. In fact, as in [22], we prove
first the analyticity and Lipschitz continuity of the inverse spectral problem of reconstructing
from the Neumann spectrum (n) and the norming constants (n) (see Theorem 2 below),
and then derive Theorem 1 by showing that the norming constants depend analytically and
Lipschitz continuously on the two spectra.
More exactly, we denote by L the family of strictly increasing sequences := (n)nZ+such that 2n :=
n
n form an element of2 and pull back the topology on L from that
8/6/2019 T2 N1
41/132
Uniform stability in the inverse problem 41
of 2 by identifying such with (2n) 2. For d (0, ) and r > 0, we denote by L(d, r)the closed convex subset ofL consisting of sequences (n)nZ+ such that (2n)2 r andn+1 n d for all n Z+. Next, we write A for the set of sequences := (n)nZ ofpositive numbers such that the sequence (n) with n := n 1 belongs to 2. This inducesthe topology of 2 on A; we further consider closed subsets A(d, r) ofA consisting of all
(n) satisfying the inequalities n d for all n Z and the relation (n)2 r.It is known [2] that, given an element (,) L A, there is a unique real-valued
L2(0, 1) such that is the sequence of eigenvalues and the sequence of normingconstants for the SturmLiouville operator TN(). Some further properties of the induced
mapping are described in the following theorem.
Theorem 2. For every d (0, ) and d (0, 1) and every positive r and r, the inversespectral mapping
L(d, r) A(d, r) (,) L2(0, 1)
is analytic and Lipschitz continuous.
2.3 Solution of the inverse spectral problem using the Krein equa-
tion
The classical algorithm of reconstructing the potential q = + 2 of a SturmLiouville op-
erator uses the so called GelfandLevitanMarchenko (GLM) equation relating the spectral
data (,) and the transformation operator K, see e.g. the monographs [28,29] for details.
The derivation of the GLM equation sketched below follows the reasoning of [24], to which
we refer the reader for further details.First we notice that due to the asymptotics of n and n the series in
h(s) := 1 + 2
n=0
n cos(22ns) cos(2ns)
(16)
converges in L2(0, 1) (in fact, h is an even function on (1, 1)). Next, denote by F an integraloperator in L2(0, 1) with kernel
f(x, t) := 12hx+t2 + hxt2 . (17)
Starting with the resolution of identity for the operator TN(),
I = 2
n=0
n( , cn)cn,
with cn = c( , 2n) being the eigenfunction corresponding to the eigenvalue n = 22n, andusing the relations (10) and the definition of F, after straightforward transformations one
arrives at the equality
I = (I + K)(I+ F)(I+ K).
8/6/2019 T2 N1
42/132
42 Hryniv R.O.
Actually, the above equality rewritten in terms of the kernels k and f of the operators K
and F produces the GLM equation,
k(x, t) + f(x, t) +
x0
k(x, s)f(s, t) ds = 0, x > t. (18)
Given the spectral data and thus the kernel f, one solves the GLM equation for the kernelk and then determines the potential q from the relation
q(x) = 2d
dxk(x, x). (19)
However, this approach does not work for impedance SturmLiouville operators under
consideration since formula (19) is then meaningless: indeed, the kernel k is not regular
enough to have a well-defined restriction k(x, x) to the diagonal and the potential q = + 2
is a distribution rather then a regular function. Instead, one can use the method of Krein
that reconstructs the function
L2(0, 1) directly. The original method was suggested
by Krein [27] for smooth functions and was further developed for the class of impedanceSturmLiouville operators with Lp(0, 1), p [1, ) in [2].
Namely, with the function h of (16), one considers a different GLM-type integral equation
(called the Krein equation)
r(x, t) + h(x t) +x0
r(x, s)h(s t) ds = 0, 0 < t < x < 1, (20)
of which the GLM equation (18) is the even part (in the sense that if r is a solution to (20),
then the function
k(x, t) := 12
r(x, xt2 ) + r(x, x+t2 )
solves (18)). It can be proved (see the next section) that equation (20) possesses a unique
solution r and, moreover, the function satisfies the equality
= r(, 0). (21)
This formula will be the basis of the reconstruction algorithm and stability analysis.
3 Stability of the inverse spectral problem: norming constants
In this section, we prove Theorem 2 on analytic and Lipschitz continuous dependence of
the logarithmic potential determining the impedance SturmLiouville operator TN() on
its eigenvalues n and norming constants n.
We shall study the correspondence between the data (,) L(d, r) A(d, r) andthe functions through the chain of mappings
(,) h r ,
in which h is the function of (16), r is the kernel solving the Krein equation (20), and, finally,
is given by (21).
8/6/2019 T2 N1
43/132
Uniform stability in the inverse problem 43
Lemma 3.1. The mapping
L(d, r) A(d, r) (,) h L2(0, 1)
is analytic and Lipschitz continuous.
Proof. We have h = 1 + h + h,, where
h(s) := 2
n=0
[cos(22ns) cos(2ns)], h,(s) := 2
n=0
n cos(22ns);
recall that the numbers 2n = 2n n and n := n 1 form sequences in 2 that inducethe topology ofL and A.
Introduce the function f L2(0, 1) whose Fourier coefficients are f(0) = 0 and
f(n) =
f(n) := 2n
for n N; then we have h = 1(f) with the mapping 1 of Lemma A.1. Therefore thefunction h depends analytically and Lipschitz continuously on f in bounded sets. Since
the mapping sending (2n) 2 into f L2(0, 1) is linear and quasi-isometric in the sensethat f =
2(2n), we conclude that the mapping h is analytic and Lipschitz
continuous on bounded sets.
Next, let g be the function in L2(0, 1) whose Fourier coefficients are
g(n) = g(n) := n
for n Z+. Then h, = 2(f, g) with 2 being the mapping of Lemma A.2. Theproperties of 2 and of the mapping (n) g then establish the required dependenceof h, on (,). The lemma is proved.
Solubility of the Krein equation crucially relies on the following property of the convolu-
tion operator H = H(,) defined via
(Hf)(x) :=
1
0
h(x t)f(t) dt,
with the function h of (16).
Lemma 3.2. For every d (0, ), d (0, 1), and positive r and r, there exists > 0 withthe following property: if (,) is an arbitrary element ofL(d, r) A(d, r) and h is thefunction of (16), then for the corresponding convolution operator H we have I+ H I.
Proof. Observing that
1
0
cos2n(x
t)f(t) dt = cos 2nx
1
0
cos2ntf(t) dt + sin 2nx 1
0
sin2ntf(t) dt
8/6/2019 T2 N1
44/132
44 Hryniv R.O.
and that the functions 1,
2sin2nx,
2cos2nx, n N, form an orthonormal basis ofL2(0, 1), we find that
((I + H)f, f) = (f, f) + 2 limk
k
n=0n
|(f, cos22ns)|2 + |(f, sin22ns)|2]
|(f, 1)|2 2 limk
kn=1
|(f, cos2ns)|2 + |(f, sin2ns)|2]= 2
n=0
n|(f, cos22ns)|2 + |(f, sin22ns)|2]
= 20|(f, 1)|2 +
n=1
n|(f, e22nis)|2 + |(f, e22nis)|2].
It follows from the results of [14, Ch. VI], [45, Ch. 4] that the system
E :=
e22nis}nN {1}
e22nis}nN
is a Riesz basis of L2(0, 1). Moreover, it was shown in [23] that there exists m = m(d, r) > 0
that gives a lower bound of E for every L(d, r). Since the inclusion A(d, r)implies that n d for all n Z, we get
((I + H)f, f) = 20|(f, 1)|2 +
n=1
n|(f, e22nis)|2 + |(f, e22nis)|2] dmf2,
and the proof is complete.
To study solubility of the Krein equation (20), we shall regard it as a relation between
the corresponding integral operators. To this end we recall several notions that will be used.
The ideal S2 of HilbertSchmidt operators in L2(0, 1) consists of integral operators whose
kernels are square integrable on := [0, 1] [0, 1]. The linear set S2 becomes a Hilbertspace under the scalar product
A, B2 := tr(AB) :=10
10
a(x, y)b(x, y) dxdy,
where a and b are the kernels of A and B respectively; in particular, AS2 := A, A1/22 isthe corresponding norm.
As an example, the inequality
10
10
|h(x y)|2 dxdy 2h2
implies that the convolution operator H belongs to S2 and, moreover, H2S2 2h2.Denote by S+2 the subspace ofS2 consisting of all HilbertSchmidt operators with lower-
triangular kernels. In other words, AS2 belongs to S
+
2
if the kernel a of A satisfies
8/6/2019 T2 N1
45/132
Uniform stability in the inverse problem 45
a(x, y) = 0 for a.e. 0 x < y 1. For an arbitrary A S2 with kernel a the cut-off a+ ofa given by
a+(x, y) =
a(x, y) for x y,0 for x < y
generates an operator A+
S+2 , and the corresponding mapping
P+ : A
A+ turns out to
be an orthoprojector in S2 onto S+2 , i.e. (P+)2 = P+ and P+A, B2 = A, P+B2 for allA, B S2; see details in [15, Ch. I.10].
With these notations, the Krein equation (20) can be recast as
R + P+H+ P+(RH) = 0 (22)
or
(I+ P+H)R = P+H,where P+X is the linear operator in S2 defined by P+X Y = P+(Y X) and I is the identityoperator in S2. Therefore solubility of the Krein equation and continuity of its solutionson H is strongly connected with the properties of the operator P+H.
Lemma 3.3. For every X B, the operator P+X is bounded in S2. Moreover, for everyconvolution operator H from the set
H := {H = H(,) | (,) L(d, r) A(d, r)} S2
the operatorI+ P+H is invertible in B(S+2 ) and the inverse (I+ P+H)1 depends analyticallyand Lipschitz continuously on H H in the topology ofS2.
Proof. Boundedness ofP+X is a straightforward consequence of the inequality
P+X YS2 Y XS2 XBYS2 ,
cf. [14, Ch. 3]. Assume next that I+ X I in L2(0, 1); then for Y S+2 we find that
(I+ P+X )Y, Y2 = Y, Y2 + Y X , Y 2 = tr
Y(I + X)Y
.
Since Y(I+ X)Y Y Y and the trace is a monotone functional, we get
(I+ P+X )Y, Y2 Y, Y2,
i.e., I+ P+X Iin S+2 .Applying now Lemma 3.2, we conclude that for every H H it holds I+ P+H Iwith
of that lemma depending only on d, d, r, and r; therefore, I+ P+H is boundedly invertiblein B(S+2 ) and (I+ P+H)1B(S+
2) 1.
Since P+H depends linearly on H, it follows that the mapping H (I+ P+H)1 from S2 intoB(S+2 ) is analytic and Lipschitz continuous on the set H. The proof is complete.
8/6/2019 T2 N1
46/132
46 Hryniv R.O.
Corollary 3.1. For every H H, the Krein equation (22) has a unique solutionR := (I+ P+H)1P+H S+2 ;
moreover, R depends analytically and Lipschitz continuously in S+2 on H H S2.It follows that the kernel r(x, t) of R is square integrable in the domain and depends
analytically and Lipschitz continuously in L2() on H. However, we need to know that
r( , 0) is well defined and belongs to L2(0, 1).To this end we use the Krein equation to find that
r(x, t) = h(x t) 10
r(x, s)h(s t) ds
as a function of x depends continuously in L2(0, 1) on t [0, 1]. Indeed, since the shiftf() f( t) is a continuous operation in L2(R), h( t) enjoys the required property.Next, since the kernels r and h belong to L2(), we find that1
0
10
r(x, s)h(s t) ds2 dx
10
dx
10
|r(x, s)|2 ds10
|h(s t)|2 ds
210
|h(s)|2 ds10
10
|r(x, s)|2 dsdx < .
(23)
Thus the function
1
0
r(x, s)h(s t) ds (24)
of the variable x [0, 1] belongs to L2(0, 1); moreover, continuity of the shifts h( t) andestimate (23) show that function (24) depends continuously in L2(0, 1) on t [0, 1]. Wethus conclude that indeed r(, t) depends continuously in L2(0, 1) on t [0, 1]. In particular,r(x, 0) is a well-defined function in L2(0, 1).
Finally, we again use the Krein equation and (21) to get the relation
(x) = r(x, 0) = h(x) 10
r(x, s)h(s) ds.
The integral on the right-hand side is a bilinear expression in h and r. In view of the analytic
dependence of r on h stated in Corollary 3.1 and estimates (23), this yields analyticity andLipschitz continuity of r(x, 0) on h L2(0, 1). On account of Lemma 3.2, the proof ofTheorem 2 is complete.
4 Reconstruction from two spectra
We recall that the norming constants n for the SturmLiouville operator TN() can be
determined from the spectra (n) and (n) of TN() and TD() by the formula (14),
n =
2n
S(2n)C(2n)
,
8/6/2019 T2 N1
47/132
Uniform stability in the inverse problem 47
where the entire functions S and C are given by the canonical products (13) over n = 22n
and n = 22n+1 respectively. This induces a mapping from the spectral data :=
(n), (n) N into the norming constants := (n) A. In this section, we shall
establish Theorem 1 by proving the following result.
Theorem 3. For every d
(0, /2) and r > 0, the mapping
N(d, r) A (25)
is analytic and Lipschitz continuous; moreover, there exist positive constants d and r such
that the range of this mapping belongs toA(d, r).
By definition, A consists of elements of the commutative unital Banach algebra A intro-
duced in Appendix C. We observe that the metrics on A agrees with the norm of A, and
thus the results of Appendix C yield the following statement.
Proposition 4.1. For every positive d and r, the set A(d, r) consists of invertible elementsofA. Moreover, the mapping 1 is analytic and Lipschitz continuous in A on A(d, r),and its range lies in A
(1 + r)1, rd1
.
In view of Proposition 4.1, it suffices to prove Theorem 3 with replaced by 1.
The elements of the sequence 1 are 1n = S(2n)C(2n)/2n. We shall show that thesequences
:=
(1)n+1S(2n)/2n
nZ+, :=
(1)nC(2n)
nZ+
form elements ofA. Thus Theorem 3 will be proved if we show that the mappings
N(d, r) A, N(d, r) A (26)
enjoy the properties required therein for the mapping (25).
To begin with, integral representations (10) and (11) of the solution c( , ) and its quasi-derivative c[1]( , ) yield the formulae
S() = sin 10
k1(1, t)sin tdt, (27)
C() = cos + 1
0
k(1, t)cos tdt (28)
for the functions S and C. Therefore both expressions S(2n)/2n and C(2n) can berecast in the form
cos 2n +
10
g(t)cos 2nt dt
with g(t) = tk1(1, t) for the former expression and g(t) = k(1, t) for the latter. The sequences
and have therefore similar structures; namely, their n-th element equals
cos 2n + (
1)n
1
0
g(t)cos 2nt dt (29)
8/6/2019 T2 N1
48/132
48 Hryniv R.O.
for respective g; here, as usual, 2n := 2n n.Clearly, the mapping (2n) (cos 2n 1) is analytic in 2. Its Lipschitz continuity
follows from the inequality | cos x cos y| |x y|; also, the inequality 1 cos x x2/2yields the estimate
(cos 2n 1)2 12(2n)22. (30)Set
g(s) :=
g(1 2s), s [0, 1
2),
g(2s 1), s [12
, 1];
then straightforward transformations give
vn := (1)n10
g(t)cos 2nt dt = (1)n10
g(s)ei2n(12s) ds
=
1
0
g(s)ei2n(12s)e2ins ds.
(31)
Therefore the above number vn gives the n-th Fourier coefficient of the function u := (f, g),
where is the mapping of Lemma A.3 and f is the function introduced in the proof of
Lemma 3.1. It follows from Lemma A.3 that the sequence (u(n))nZ of Fourier coefficients
ofu depends analytically and boundedly Lipschitz continuously in 2 on f and g. We prove
in the lemma below that the functions k(1, ) and k1(1, ) (and thus the correspondingtransformates g) depend in the same manner on = (,) N(d, r).
Lemma 4.1. The mappings
N(d, r) (,) k(1, ) L2(0, 1),
N(d, r) (,) k1(1, ) L2(0, 1)
are analytic and Lipschitz continuous.
Proof. Since both mappings can be treated similarly, we only consider the second one. By
definition, we have S(2n)/2n = 0, and thus the numbers 2n = n + 2n, n Z, arezeros of the odd entire function S()/ of (27). The required properties of the mapping
k1(1, ) follow now from the results of [26]; see Appendix B.
The above reasoning justifies the inclusion1
A
as well as analyticity and Lipschitzcontinuity of the mappings of (26). It remains to prove that there exist positive d and r
such that, for every N(h, r), the corresponding elements and belong to A(d, r).Existence of such an r follows from the uniform estimates of the 2-norms of the sequences
(cos 2n 1) of (30) and the fact thatnZ+
|vn|2 (f, g)2,
see (31) and the discussion following it. Indeed, in view of Lemma A.3 the function u =
(f, g) remains in the bounded subset ofL2(0, 1) when f and g vary over bounded subsets
8/6/2019 T2 N1
49/132
Uniform stability in the inverse problem 49
of L2(0, 1), and the latter is the case when runs over N(d, r) by the definition of the
functions f and g and Lemma 4.1.
Next, in view of formula (13) and the interlacing property of n and n, the numbers
n = (1)nS(2n)/2n and n = (1)nC(2n) are all of the same sign and thus are allpositive in view of the asymptotic relation (29). The uniform positivity of n and n (and
thus existence of a positive d
such that 1/n = nn d
) follows immediately from thelemma below.
Lemma 4.2. For every d (0, /2) and r > 0 we have
sup(,)
supnZ+
log |S(2n)/2n| < , sup(,)
supnZ+
log |C(2n)| < ,
where S and C are constructed via (13) from the sequences and, and the suprema are
taken over (,) N(d, r).Proof. We assume first that n
= 0. By (13), we have
S(2n)/2n = 222n
2n2
kN, k=n
22k 22n2k2
.
Dividing both sides by
cos n =d sin z
dz
z=n
= 2
kN, k=n
k2 n2k2
,
we conclude that
|S(2n)/2n| =2
2n2n2
kN, k=n
2
2k 2
2n2(k2 n2);
for n = 0 the direct calculations give
lim0
|S()/| = 2kN
22k2k2
.
Recall that 2k := 2k k and set
an,k :=2n 2k(n
k)
,
with a0,0 = 1 and an,n = 0 if n N; then22k 22n
2(k2 n2) =
1 + an,k
1 + an,k
and2
|S(2n)/2n| =kZ
(1 + an,k).
In what follows, all summations and multiplications over the index set Z will be taken in the principal value
sense and the symbol V.p. will be omitted.
8/6/2019 T2 N1
50/132
50 Hryniv R.O.
Since the sequence (n) is 2d-separated for every (,) N(d, r), we have 1 + ak,n 2d/for all n Z+ and all k Z. Therefore, with
K := maxx1+2d/
log(1 + x) xx2
< ,
we get the estimate logkZ
(1 + an,k)
kZ
an,k + K
kZ
a2n,k, (32)
provided the two series converge.
Clearly, k=n
1
n k = 0,
and thus kZ
an,k
=
1
k=n
2kk n
r
3
by the CauchyBunyakovskiSchwarz inequality (recall that
kZ 22k r2 by the definitionof the set N(d, r) and
k=n(k n)2 = 2/3). Next, the inequality
a2n,k 222k
2(k n)2 +222n
2(k n)2for k = n yields
kZ
a2n,k 4r2k=n
1
2(k n)2 =4r2
3.
It follows from (32) that logkZ
(1 + an,k) (3r + 4Kr2)/3,
where the constant K only depends on d.
Similarly, we find that
|C(2n)| =
kZ+
22k+1 22n2(k + 1
2)2
= kZ+
22k+1 22n2(k + 1
2) 2n2
and then mimic the above reasoning to establish the other uniform bound. The lemma is
proved.
Proof of Theorem 3. Combining the results of Lemmata 4.1 and 4.2, we conclude that the
mappings (26) enjoy all the properties stated in Theorem 3, and thus so does the mapping
(,) 1. In virtue of Proposition 4.1 this completes the proof of the theorem.Proof of Theorem 1. Analyticity and Lipschitz continuity on bounded sets of the inverse
spectral mapping
N L2(0, 1)is the direct consequence of those for the mappings (25) and (15) established in Theorems 3
and 2 respectively.
8/6/2019 T2 N1
51/132
Uniform stability in the inverse problem 51
5 Some extensions
The results proved above for the class of impedance SturmLiouville operators with real-