Eigenfunctions of the Cosine and Sine Transforms

A description of the eigensubspaces of the cosine and sine operators is given. The spectrum of each of these two operators consists of two eigen- values 1, -1 and their eigensubspaces are infinite{dimensional. There are many possible bases for these subspaces, but most popular are the ones construct...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Журнал математической физики, анализа, геометрии
Datum:2013
1. Verfasser: Katsnelson, V.
Format: Artikel
Sprache:English
Veröffentlicht: Фізико-технічний інститут низьких температур ім. Б.І. Вєркіна НАН України 2013
Online Zugang:https://nasplib.isofts.kiev.ua/handle/123456789/106768
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Назва журналу:Digital Library of Periodicals of National Academy of Sciences of Ukraine
Zitieren:Eigenfunctions of the Cosine and Sine Transforms / V. Katsnelson // Журнал математической физики, анализа, геометрии. — 2013. — Т. 9, № 4. — С. 476-495. — Бібліогр.: 7 назв. — англ.

Institution

Digital Library of Periodicals of National Academy of Sciences of Ukraine
id nasplib_isofts_kiev_ua-123456789-106768
record_format dspace
spelling Katsnelson, V.
2016-10-04T20:01:55Z
2016-10-04T20:01:55Z
2013
Eigenfunctions of the Cosine and Sine Transforms / V. Katsnelson // Журнал математической физики, анализа, геометрии. — 2013. — Т. 9, № 4. — С. 476-495. — Бібліогр.: 7 назв. — англ.
1812-9471
https://nasplib.isofts.kiev.ua/handle/123456789/106768
A description of the eigensubspaces of the cosine and sine operators is given. The spectrum of each of these two operators consists of two eigen- values 1, -1 and their eigensubspaces are infinite{dimensional. There are many possible bases for these subspaces, but most popular are the ones constructed from the Hermite functions. We present other "bases" which are not discrete orthogonal sequences of vectors, but continuous orthogo- nal chains of vectors. Our work can be considered to be a continuation and further development of the results obtained by Hardy and Titchmarsh: "Self-reciprocal functions"(Quart. J. Math., Oxford, Ser. 1 (1930)).
Представлено описание собственных подпространств косинус- и синус-опе- раторов. Спектр каждого из этих двух операторов состоит из двух собственных значений 1, -1 , а их собственные подпространства бесконечномерные. Есть много возможных базисов этих подпространств, но наиболее популярными являются те, которые построены из функций Эрмита. Мы представляем другие "базисы", которые являются не дискретными ортогональными последовательностями векторов, а непрерывными ортогональными цепочками векторов. Работу можно считать продолжением и дальнейшим развитием результатов, приведенных Харди и Титчмаршем в статье "Self-reciprocal functions" (Quart. J. Math., Oxford Ser. 1 (1930)).
I thank Armin Rahn for his careful reading of the manuscript and his help in improving English in this paper.
en
Фізико-технічний інститут низьких температур ім. Б.І. Вєркіна НАН України
Журнал математической физики, анализа, геометрии
Eigenfunctions of the Cosine and Sine Transforms
Article
published earlier
institution Digital Library of Periodicals of National Academy of Sciences of Ukraine
collection DSpace DC
title Eigenfunctions of the Cosine and Sine Transforms
spellingShingle Eigenfunctions of the Cosine and Sine Transforms
Katsnelson, V.
title_short Eigenfunctions of the Cosine and Sine Transforms
title_full Eigenfunctions of the Cosine and Sine Transforms
title_fullStr Eigenfunctions of the Cosine and Sine Transforms
title_full_unstemmed Eigenfunctions of the Cosine and Sine Transforms
title_sort eigenfunctions of the cosine and sine transforms
author Katsnelson, V.
author_facet Katsnelson, V.
publishDate 2013
language English
container_title Журнал математической физики, анализа, геометрии
publisher Фізико-технічний інститут низьких температур ім. Б.І. Вєркіна НАН України
format Article
description A description of the eigensubspaces of the cosine and sine operators is given. The spectrum of each of these two operators consists of two eigen- values 1, -1 and their eigensubspaces are infinite{dimensional. There are many possible bases for these subspaces, but most popular are the ones constructed from the Hermite functions. We present other "bases" which are not discrete orthogonal sequences of vectors, but continuous orthogo- nal chains of vectors. Our work can be considered to be a continuation and further development of the results obtained by Hardy and Titchmarsh: "Self-reciprocal functions"(Quart. J. Math., Oxford, Ser. 1 (1930)). Представлено описание собственных подпространств косинус- и синус-опе- раторов. Спектр каждого из этих двух операторов состоит из двух собственных значений 1, -1 , а их собственные подпространства бесконечномерные. Есть много возможных базисов этих подпространств, но наиболее популярными являются те, которые построены из функций Эрмита. Мы представляем другие "базисы", которые являются не дискретными ортогональными последовательностями векторов, а непрерывными ортогональными цепочками векторов. Работу можно считать продолжением и дальнейшим развитием результатов, приведенных Харди и Титчмаршем в статье "Self-reciprocal functions" (Quart. J. Math., Oxford Ser. 1 (1930)).
issn 1812-9471
url https://nasplib.isofts.kiev.ua/handle/123456789/106768
citation_txt Eigenfunctions of the Cosine and Sine Transforms / V. Katsnelson // Журнал математической физики, анализа, геометрии. — 2013. — Т. 9, № 4. — С. 476-495. — Бібліогр.: 7 назв. — англ.
work_keys_str_mv AT katsnelsonv eigenfunctionsofthecosineandsinetransforms
first_indexed 2025-11-27T01:26:23Z
last_indexed 2025-11-27T01:26:23Z
_version_ 1850790849315602432
fulltext Journal of Mathematical Physics, Analysis, Geometry 2013, vol. 9, No. 4, pp. 476–495 Eigenfunctions of the Cosine and Sine Transforms V. Katsnelson The Weizmann Institute of Science Rehovot 76100, Israel E-mail: victor.katsnelson@weizmann.ac.il; victorkatsnelson@gmail.com Received October 29, 2012, revised December 17, 2012 A description of the eigensubspaces of the cosine and sine operators is given. The spectrum of each of these two operators consists of two eigen- values 1, −1 and their eigensubspaces are infinite–dimensional. There are many possible bases for these subspaces, but most popular are the ones constructed from the Hermite functions. We present other ”bases” which are not discrete orthogonal sequences of vectors, but continuous orthogo- nal chains of vectors. Our work can be considered to be a continuation and further development of the results obtained by Hardy and Titchmarsh: ”Self-reciprocal functions”(Quart. J. Math., Oxford, Ser. 1 (1930)). Key words: Fourier transform, cosine-sine transforms, eigenfunctions, Melline transform. Mathematics Subject Classification 2010: 47A38 (primery); 47B35, 47B06, 47A10 (secondary). 1. The cosine transform C and the sine transform S are defined by the formulas (Cx)(t) = √ 2 π ∫ R+ cos(tξ)x(ξ) dξ, t ∈ R+ , (1a) (Sx)(t) = √ 2 π ∫ R+ sin(tξ) x(ξ) dξ, t ∈ R+ , (1b) where R+ is the positive half-axis, R+ = {t ∈ R : t > 0}. For x ∈ L1(R+), the integrals in (1) are well defined as Lebesgue integrals. If x(t) ∈ L2(R+) ∩ L1(R+), then the Parseval equalities hold: ∫ R+ |(Cx)(t)|2dt = ∫ R+ |x(t)|2dt, (2a) ∫ R+ |(Sx)(t)|2dt = ∫ R+ |x(t)|2dt. (2b) c© V. Katsnelson, 2013 Eigenfunctions of the Cosine and Sine Transforms Thus, the transforms C and S can both be considered as the linear operators de- fined on the linear manifold L1(R+)∩L2(R+) of the Hilbert space L2(R+), map- ping this linear manifold into L2(R+) isometrically. Since the set L1(R+)∩L2(R+) is dense in L2(R+), each of these operators can be extended to an operator defined on the whole space L2(R+), which maps L2(R+) into L2(R+) isometrically. We retain the notation C and S for the extended operators. In an even broader con- text, the transformation (1) can be considered for those x, for which the integrals on the right–hand sides are meaningful. Considered as operators in the Hilbert space L2(R+), the operators C and S are self-adjoint operators which satisfy the equalities C2 = I, S2 = I, (3) where I is the identity operator in L2(R+). Each of the spectra σ(C) and σ(S) of these operators consists of two points: +1 and −1. By Cλ and Sλ we denote the spectral subspaces of the operators C and S, respectively, corresponding to the points λ = 1 and λ = −1 of their spectra. These spectral subspaces are eigensubspaces: C1 = {x ∈ L2(R+) : Cx = x}, C−1 = {x ∈ L2(R+) : Cx = −x}; (4a) S1 = {x ∈ L2(R+) : Sx = x}, S−1 = {x ∈ L2(R+) : Sx = −x}. (4b) Moreover, two orthogonal decompositions hold L2(R+) = C1 ⊕ C−1, L2(R+) = S1 ⊕ S−1 . (5) The spectra of the operators C and S are highly degenerated: the eigensubspaces Cλ and Sλ are infinite-dimensional. Many bases are possible in these subspaces. The best known are the bases formed by the Hermite functions hk(t) restricted onto R+. The Hermite functions hk(t) are defined as hk(t) = e t2 2 dk (e−t2) dtk , t ∈ R, k = 0, 1, 2, . . . . (6) It is known that the system {hk}k=0,1,2, ... forms an orthogonal basis in the Hilbert space L2(R). The properties of the Hermite functions hk as eigenfunctions of the Fourier transform were established by N. Wiener, [1, Ch. 1]. In [1], N. Wiener developed the L2-theory of the Fourier transform which was based on these prop- erties of the Hermite functions. The Hermite functions hk are originally defined on the whole real axis R. The restrictions hk |R+ of the Hermite functions hk onto R+ are considered as the vectors of the Hilbert space L2(R+). Each of two systems {h2k |R+ }k=0,1,2, ... and Journal of Mathematical Physics, Analysis, Geometry, 2013, vol. 9, No. 4 477 V. Katsnelson {h2k+1|R+ }k=0,1,2, ... is an orthogonal basis in L2(R+). The systems { h4l|R+ } l=0,1,2, ... ,{ h4l+2|R+ } l=0,1,2, ... , { h4l+1|R+ } l=0,1,2, ... , and { h4l+3|R+ } l=0,1,2, ... are orthogonal bases of the eigensubspaces C1, C−1, S1 and S−1, respectively. We present other ”bases” which are not discrete orthogonal sequences of vectors, but continuous orthogonal chains of (generalized) vectors. This is the main goal of this paper. Our work may be considered as a further development of the results given in [2] by Hardy and Titchmarsh. (The contents of [2] and [3] were reproduced in the book [4].) 2. First we discuss the eigenfunctions of the transforms C and S in the broad sense. These transforms are of the form x → Kx, where (Kx)(t) = ∫ R+ k(tξ)x(ξ)dξ, (7) and k is a function of one variable defined on R+. (It should be mentioned that some operational calculus related to the operators of the form (7) was developed in [5].) R e m a r k 1. If the integral (7) does not exist as a Lebesgue integral, i.e., the function k(tξ)x(ξ) of the variable ξ is not summable, then a meaning can be attached to the integral (7) by some regularization procedure. We use the regularization procedure ∫ R+ k(tξ)x(ξ)dξ = lim ε→+0 ∫ R+ e−εξk(tξ)x(ξ)dξ, (8) and the regularization procedure ∫ R+ k(tξ)x(ξ)dξ = lim R→+∞ R∫ 0 k(tξ)x(ξ)dξ . (9) If for some a ∈ C both integrals ∫ R+ k(tξ)ξ−adξ and ∫ R+ k(tξ)ξa−1dξ (10) have a meaning for every positive t, then, changing the variable tξ → ξ, we obtain K t−a = κ(a) ta−1, K ta−1 = κ(1− a) t−a, (11) 478 Journal of Mathematical Physics, Analysis, Geometry, 2013, vol. 9, No. 4 Eigenfunctions of the Cosine and Sine Transforms where κ(a) = ∫ R+ k(ξ)ξ−adξ, κ(1− a) = ∫ R+ k(ξ)ξa−1dξ. (12) Equalities (11) mean that the subspace (two-dimensional if a 6= 1/2) generated by the functions t−a and ta−1 is invariant with respect to the transformation K and that the matrix of this operator in the basis t−a, ta−1 is: ∥∥∥ 0 κ(1−a) κ(a) 0 ∥∥∥ . Thus, assuming that κ(a) 6= 0, κ(1− a) 6= 0, we obtain that the functions √ κ(1− a)t−a + √ κ(a)ta−1 and √ κ(1− a)t−a − √ κ(a)ta−1 (13) are the eigenfunctions of the transform K corresponding to the eigenvalues λ+ = √ κ(a)κ(1− a) and λ− = − √ κ(a)κ(1− a), (14) respectively. To find the eigenfunctions of the form (13) for cosine and sine transforms C and S, we have to calculate the constants (12) corresponding to the functions kc(τ) = √ 2 π cos τ and ks(τ) = √ 2 π sin τ, (15) which generate the kernels of these integral transforms. This is accomplished in the following Lemma 1. Let ζ belong to the strip 0 < Re ζ < 1. Then 1. ∞∫ 0 (cos s) sζ−1 ds = ( cos π 2 ζ ) Γ(ζ) , (16a) ∞∫ 0 (sin s) sζ−1 ds = ( sin π 2 ζ ) Γ(ζ) , (16b) where Γ is the Euler Gamma-function and the integrals in (16) are under- stood in the sense ∞∫ 0 { cos s sin s } s ζ−1 ds = lim R→+∞ R∫ 0 { cos s sin s } s ζ−1 ds = lim ε→+0 +∞∫ 0 e−εs { cos s sin s } s ζ−1 ds . Journal of Mathematical Physics, Analysis, Geometry, 2013, vol. 9, No. 4 479 V. Katsnelson 2. The above limits exist uniformly with respect to ζ from any fixed compact subset of the strip 0 < Re ζ < 1. 3. Given δ > 0, then for any ζ from the strip δ < Re ζ < 1 − δ and for any R ∈ (0, +∞), ε ∈ (0,+∞), the estimates ∣∣∣∣ R∫ 0 { cos s sin s } sζ−1ds ∣∣∣∣ ≤ C(δ)e π 2 |Im ζ|, (17a) ∣∣∣∣ +∞∫ 0 e−εs { cos s sin s } sζ−1ds ∣∣∣∣ ≤ C(δ)e π 2 |Im ζ| (17b) hold, where C(δ) < ∞ does not depend on ζ, R and ε. We omit the proof of Lemma 1. This lemma can be proved by a standard method using integration in the complex plane. According to Lemma 1, the integrals ∞∫ 0 { kc(s) ks(s) } s−a ds and ∞∫ 0 { kc(s) ks(s) } sa−1 ds, where kc and ks, (15), are the functions generating the kernels of the integral transformations C and S, exist for every a such that 0 < Re a < 1, or, amounting to the same, 0 < Re (1− a) < 1. The constants κc(a) and κc(1 − a), corresponding to the function kc(τ) =√ 2 π cos τ , are: κc(a) = √ 2 π sin πa 2 Γ(1− a), κc(1− a) = √ 2 π cos πa 2 Γ(a). (18a) The constants κs(a) and κs(1−a), corresponding to the function ks(τ) = √ 2 π sin τ , are: κs(a) = √ 2 π cos πa 2 Γ(1− a), κs(1− a) = √ 2 π sin πa 2 Γ(a). (18b) 3. Later, we will have to transform the expression (18) for the constants κc 480 Journal of Mathematical Physics, Analysis, Geometry, 2013, vol. 9, No. 4 Eigenfunctions of the Cosine and Sine Transforms and κs using the following identities for the Euler Gamma-function Γ(ζ): Γ(ζ + 1) = ζΓ(ζ) , see [6] , 12.12, (19a) Γ(ζ)Γ(1− ζ) = π sinπζ , see [6] , 12.14, (19b) Γ(ζ)Γ ( ζ + 1 2 ) = 2 √ π 2−2ζΓ(2ζ), see [6] , 12.15. (19c) Lemma 2. The following identities hold: √ 2 π ( cos π 2 ζ ) Γ(ζ) = 2 ζ− 1 2 Γ ( ζ 2 ) Γ ( 1 2 − ζ 2 ) , (20a) √ 2 π ( sin π 2 ζ ) Γ(ζ) = 2 ζ− 1 2 Γ ( 1 2 + ζ 2 ) Γ ( 1− ζ 2 ) . (20b) P r o o f. From (19b) it follows that cos π 2 ζ = π Γ ( 1 2 − ζ 2 ) Γ ( 1 2 + ζ 2 ) . From (19c) it follows that Γ(ζ) = π− 1 2 Γ ( ζ 2 ) Γ ( 1 2 + ζ 2 ) 2ζ−1 . Combining the last two formulas, we obtain (20a). Combining the last formula with the formula sin π 2 ζ = π Γ ( ζ 2 ) Γ ( 1− ζ 2 ) , we obtain (20b). Lemma 3. The values κc(a), κc(1−a), κs(a), κs(1−a), which appear as the coefficients of the linear combinations (13), are: κc(a) = 2 1 2 −a Γ(1 2 − a 2 ) Γ(a 2 ) , κc(1− a) = 2a− 1 2 Γ(a 2 ) Γ(1 2 − a 2 ) , (21a) κs(a) = 2 1 2 −a Γ(1− a 2 ) Γ(1 2 + a 2 ) , κs(1− a) = 2a− 1 2 Γ(1 2 + a 2 ) Γ(1− a 2 ) . (21b) 4. From the expressions (21) we can see that the products κc(a)κc(1 − a) and κs(a)κs(1− a) do not depend on a, κc(a)κc(1− a) = 1, κs(a)κs(1− a) = 1 0 < Re a < 1. Journal of Mathematical Physics, Analysis, Geometry, 2013, vol. 9, No. 4 481 V. Katsnelson Theorem 1. Let a ∈ C, 0 < Re a < 1, a 6= 1 2 , and κc(a), κc(1 − a), κs(a), κs(1− a) be the values appeared in (21). Then: 1. The functions E+ c (t, a) = √ κc(1− a)t−a + √ κc(a)ta−1, (22a) E− c (t, a) = √ κc(1− a)t−a − √ κc(a)ta−1 (22b) of the variable t ∈ R+ are the eigenfunctions (in the broad sense) of the co- sine transform C corresponding to the eigenvalues +1 and −1, respectively, E+ c (t, a) = lim R→∞ √ 2 π R∫ 0 cos(tξ) E+ c (ξ, a) dξ, E− c (t, a) = − lim R→∞ √ 2 π R∫ 0 cos(tξ) E− c (ξ, a) dξ. 2. The functions E+ s (t, a) = √ κs(1− a)t−a + √ κs(a)ta−1, (23a) E− s (t, a) = √ κs(1− a)t−a − √ κs(a)ta−1 (23b) of the variable t ∈ R+ are the eigenfunctions (in the broad sense) of the sine transform S corresponding to the eigenvalues +1 and −1, respec- tively, E+ s (t, a) = lim R→∞ √ 2 π R∫ 0 sin(tξ) E+ s (ξ, a) dξ, (24) E− s (t, a) = − lim R→∞ √ 2 π R∫ 0 sin(tξ) E− s (ξ, a) dξ. (25) For the fixed t ∈ (0,∞), the limits exist uniformly with respect to a, from any compact subset of the strip 0 < Re a < 1. R e m a r k 2. In (22) and (23), the values of the square roots √ κ(a) and√ κ(1− a) should be chosen such that their products equal 1. 482 Journal of Mathematical Physics, Analysis, Geometry, 2013, vol. 9, No. 4 Eigenfunctions of the Cosine and Sine Transforms R e m a r k 3. For a = 1 2 , there is only one eigenfunction E ( t, 1 2 ) = 2t− 1 2 . R e m a r k 4. Since E+ c (t, a) = E+ c (t, 1− a), E− c (t, a) = −E− c (t, 1− a), (26a) E+ s (t, a) = E+ s (t, 1− a), E− s (t, a) = −E− s (t, 1− a), (26b) each eigenfunction appears in the family {E± c,s(t, a)}0<Re a<1 twice. To avoid this redundancy, we should consider the family where only one of the points a or 1−a appears. 5. If 0 < Re a < 1 and x(t) is any of the eigenfunctions of the form either (22), or (23), then the integral ∫ R+ |x(t)|2 diverges. Thus, none of these eigenfunctions belong to L2(R+). This integral diverges both at the points t = +0 and t = +∞. However, this integral diverges variously for a with Re a = 1 2 and for a with Re a 6= 1 2 . If Re a = 1 2 , then the integrals diverge logarithmically both at t = +0 and t = +∞. If Re a 6= 1 2 , then the integrals diverge more strongly: powerwisely. We try to construct the eigenfunctions of the operator C (of the operator S) from L2 as the continuous combinations of the eigenfunctions of the form (22) (of the form (23)). Our hope is that the singularities of the ”continuous linear combinations” of eigenfunctions, which are in some sense an averaging of the eigenfunctions of the family, are weaker than the singularities of individual eigenfunctions. These continuous linear combinations should not include the eigenfunctions of the form (22) and (23) with a : Re a 6= 1 2 . The singularities of eigenfunctions with a : Re a 6= 1 2 are too strong and can not disappear by averaging. Thus, we have to restrict ourselves to a’s of the form a = 1 2 +iτ, τ ∈ R. Considering the case Re a = 1 2 in more detail, we introduce special notation for the eigenfunctions E± c,s(t, 1 2 + iτ)}: e+ c (t, τ) = 1 2 √ π E+ c (t, 1 2 + iτ), e−c (t, τ) = 1 2i √ π E− c (t, 1 2 + iτ), (27a) e+ s (t, τ) = 1 2 √ π E+ s (t, 1 2 + iτ), e−s (t, τ) = 1 2i √ π E− s (t, 1 2 + iτ). (27b) (We include the normalizing factor 1 2 √ π in the definition of the functions e±c,s.) According to (21), (22), the functions e±c,s(t, τ) can be expressed as e+ c (t, τ) = 1 2 √ π ( t− 1 2 −iτ c(τ) + t− 1 2 +iτ c(−τ) ) , (28a) e−c (t, τ) = 1 2i √ π ( t− 1 2 −iτ c(τ)− t− 1 2 +iτ c(−τ) ) , (28b) Journal of Mathematical Physics, Analysis, Geometry, 2013, vol. 9, No. 4 483 V. Katsnelson e+ s (t, τ) = 1 2 √ π ( t− 1 2 −iτ s(τ) + t− 1 2 +iτ s(−τ) ) , (29a) e−s (t, τ) = 1 2i √ π ( t− 1 2 −iτ s(τ)− t− 1 2 +iτ s(−τ) ) , (29b) where c(τ), s(τ) are the ”phase factors” c(τ) = 2i τ 2 exp { i arg Γ(1 4 + i τ 2 ) } , −∞ < τ < ∞, (30a) s(τ) = 2i τ 2 exp { i arg Γ(3 4 + i τ 2 ) } , −∞ < τ < ∞. (30b) In (30), exp{i arg Γ(ζ)} = Γ(ζ) |Γ(ζ)| . Since c(τ) = c(−τ), s(τ) = s(−τ) for real τ , the values of the functions e+ c (t, τ), e−c (t, τ), e+ s (t, τ), e−s (t, τ) are real for t ∈ (0,∞), τ ∈ (0,∞). R e m a r k 5. The parameter τ , which enumerates the families {e±c (t, τ}, {e±s (t, τ}, runs over the interval (0,∞). There is no need to consider negative τ . (See Remark 4). 6. Let us introduce four integral transforms T+ c , T−c , T+ s , T−s . For φ(t) ∈ L1(R+) and t > 0, let us define (T+ c φ)(t) = ∫ R+ e+ c (t, τ)φ(τ) dτ, (T−c φ)(t) = ∫ R+ e−c (t, τ)φ(τ) dτ, (31a) (T+ s φ)(t) = ∫ R+ e+ s (t, τ)φ(τ) dτ, (T−s φ)(t) = ∫ R+ e−s (t, τ)φ(τ) dτ. (31b) Lemma 4. If φ(τ) ∈ L1(R+), and x(t) = (Tφ)(t), where T is any of the above–introduced four transformations T±c,s, then the function x(t) is continuous on the interval (0,∞) and the estimate |x(t)| ≤ 1√ π ‖φ‖ L1(R+) · t− 1 2 , 0 < t < ∞, (32) holds. P r o o f. Let e(t, τ) be any of the four above–introduced functions e+ c (t, τ), e−c (t, τ), e+ s (t, τ), e−s (t, τ). The function e(t, τ) is continuous with respect to t at each t > 0, τ > 0 and satisfies the estimate |e(t, τ | ≤ 1√ π t− 1 2 , 0 < t < ∞, 0 < τ < ∞. (33) Now Lemma 4 is a consequence of the standard results of the Lebesgue integration theory. 484 Journal of Mathematical Physics, Analysis, Geometry, 2013, vol. 9, No. 4 Eigenfunctions of the Cosine and Sine Transforms Theorem 2. Let φ(τ) be a function satisfying the condition ∞∫ 0 |φ(τ)|eπ 2 τ dτ < ∞, (34) and x+ c (t) = (T+ c φ)(t), x−c (t) = (T−c φ)(t), (35) x+ s (t) = (T+ s φ)(t), x−s (t) = (T−c φ)(t). (36) Then the functions x+ c (t), x−c (t) are the eigenfunctions (in the broad sense) of the cosine transform C, and the functions x+ s (t), x−s (t) are the eigenfunctions (in the broad sense) of the sine transform S, i.e., x+ c (t) = lim R→∞ √ 2 π R∫ 0 cos(tξ) x+ c (ξ) dξ, (37a) x−c (t) = − lim R→∞ √ 2 π R∫ 0 cos(tξ) x−c (ξ) dξ (37b) and x+ s (t) = lim R→∞ √ 2 π R∫ 0 sin(tξ) x+ s (ξ) dξ, (38a) x−s (t) = − lim R→∞ √ 2 π R∫ 0 sin(tξ) x−s (ξ) dξ (38b) for every t ∈ (0,∞). In particular, in (37), (38) the limits exist. P r o o f. According to Theorem 1 and (27), e+ c (t, τ) = lim R→∞ √ 2 π R∫ 0 cos(tξ)e+ c (ξ, τ) dξ for every t, τ. Multiplying by φ(τ) and integrating with respect to τ , we obtain x+ c (t) = √ 2 π ∞∫ 0 ( lim R→∞ R∫ 0 cos(tξ)e+ c (ξ, τ) dξ ) φ(τ) dτ. Journal of Mathematical Physics, Analysis, Geometry, 2013, vol. 9, No. 4 485 V. Katsnelson From (17) we obtain the estimate ∣∣∣∣ R∫ 0 cos(tξ)e+ c (ξ, τ) dξ ∣∣∣∣ ≤ Ct− 1 2 e π 2 τ , ∀ R < ∞, τ ∈ R+, t ∈ R+, where the value C < ∞ does not depend on R, τ, t. This estimate and condition (34) for the function φ(t) allow us to apply the Lebesgue dominated convergence theorem ∞∫ 0 ( lim R→∞ R∫ 0 cos(tξ)e+ c (ξ, τ) dξ ) φ(τ) dτ = lim R→∞ ∞∫ 0 ( R∫ 0 cos(tξ)e+ c (ξ, τ) dξ ) φ(τ) dτ. (39) Thus, x+ c (t) = lim R→∞ √ 2 π ∞∫ 0 ( R∫ 0 cos(tξ)e+ c (ξ, τ) dξ ) φ(τ) dτ. On the other hand, using estimate (33) for e+ c (ξ, τ), we can justify the change of order of integration in the series integral which appears on the right–hand side of the above equality. For any finite R, ∞∫ 0 ( R∫ 0 cos(tξ)e+ c (ξ, τ) dξ ) φ(τ) dτ = R∫ 0 cos(tξ) ( ∞∫ 0 e+ c (ξ, τ)φ(τ) dτ ) dξ = R∫ 0 cos(tξ)x+ c (ξ) dξ. Finally, we obtain the equality x+ c (t) = limR→∞ R∫ 0 cos(tξ)x+ c (ξ) dξ, i.e., equality (37a) for the function x+ c . Equality (37b) for the function x−c and equalities (38) for the functions x+ s , x+ s can be obtained analogously. R e m a r k 6. In Theorem 2 we assume that the function φ satisfies condition (34). Assuming only that ∞∫ 0 |φ(τ)| dτ < ∞, we can not justify equality (39). To 486 Journal of Mathematical Physics, Analysis, Geometry, 2013, vol. 9, No. 4 Eigenfunctions of the Cosine and Sine Transforms apply the Lebesgue dominated convergence theorem, we need the estimate sup R∈(0,∞) τ∈(−∞,∞) ∣∣∣∣ R∫ 0 (cos ξ) · ξ−1 2+iτ dξ ∣∣∣∣ < ∞. We are, however, able to establish (17), but this estimate is not strong enough. The question of whether equalities (37), (38) hold under the assumption ∞∫ 0 |φ(τ)| dτ < ∞ remains open. 7. Our considerations in the context of the L2-theory on the operators C and S are based on the L2-theory for the Melline transform. (See the article ”Melline Transform” on page 192 of [7, Vol. 6] and references there.) The Melline transform M is defined by (Mf)(ζ) = ∞∫ 0 f(t)tζ−1 dt. If the function f(t) ∈ L2(R+) is compactly supported in the open interval (0,∞), then the function Φ(ζ) = (Mf)(ζ) of the variable ζ is defined in the whole complex ζ-plane and is holomorphic. The function f(t) can be recovered from the function Φ = Mf by the formula f(t) = 1 2πi ∫ Re ζ=c Φ(ζ) t−ζ dζ, where c is an arbitrary real number. Moreover, the Parseval equality ∞∫ 0 |f(t)|2dt = 1 2π ∫ Re ζ= 1 2 |Φ(ζ)|2 |dζ| holds (from which we recognize the significance of the vertical line Re ζ = 1 2). Thus the Melline transform M generates the linear operator defined on the set of all compactly supported functions f from L2(R+) which maps this set iso- metrically into the space L2 ( Re ζ = 1 2 ) of the functions defined on the vertical line Re ζ = 1 2 and which are square-integrable. Since the set of all compactly supported functions f is dense in L2(R+), this operator can be extended to an isometrical operator defined on the whole L2(R+) which maps L2(R+) isomet- rically into L2 ( Re ζ = 1 2 ) . We will continue to denote this extended operator by M. Journal of Mathematical Physics, Analysis, Geometry, 2013, vol. 9, No. 4 487 V. Katsnelson It turns out that the operator M maps the space L2(R+) onto the whole space L2 ( Re ζ = 1 2 ) . The inverse operator M−1 is defined everywhere on L2 ( Re ζ = 1 2 ) . If Φ ∈ L2 ( Re ζ = 1 2 ) , then the function f(t) = (M−1Φ)(t) is defined as an L2(R+)-function and can be expressed as f(t) = 1 2π ∫ ∞ −∞ Φ ( 1 2 + iτ ) t− 1 2 −iτdτ, 0 < t < ∞. (40a) Furthermore, the function Φ ( 1 2 + iτ ) = (Mf)(1 2 + iτ ) can be expressed as Φ ( 1 2 + iτ ) = ∞∫ 0 f(t) t− 1 2 +iτdt, −∞ < τ < ∞. (40b) The pair of formulas (40a) and (40b) together with the Parseval equality ∞∫ 0 |f(t)|2dt = 1 2π ∞∫ −∞ ∣∣Φ( 1 2 + iτ )∣∣2 dτ (40c) make up the most import part of the L2-theory of the Melline transform. 8. Developing the L2-theory of the cosine and sine transforms, we first of all prove Lemma 5. Let φ(t) ∈ L1(R+) ∩ L2(R+). Then ∫ R+ |(Tφ)(t)|2dt = ∫ R+ |φ(τ)|2dτ , (41) where T is any of the above–introduced (see (31)) four transformations T±c,s. P r o o f. The proof is based on the Parseval equality for the Melline transform. We present the transformations T±c,s as the inverse Melline transforms. Given a function φ(τ) defined for τ ∈ (0,∞), we introduce the functions Φ+ c ( 1 2 + iτ ) = √ π c(τ) φ(|τ |), (42a) Φ−c ( 1 2 + iτ ) =1 i sign (τ) √ π c(τ) φ(|τ |), (42b) 488 Journal of Mathematical Physics, Analysis, Geometry, 2013, vol. 9, No. 4 Eigenfunctions of the Cosine and Sine Transforms and Φ+ s ( 1 2 + iτ ) = √ π s(τ) φ(|τ |), (43a) Φ−s ( 1 2 + iτ ) =1 i sign (τ) √ π s(τ) φ(|τ |), (43b) which are defined for τ ∈ (−∞,∞). Here c(τ), s(τ) are the ”phase factors” introduced in (30). It is clear that ∣∣Φ( 1 2 + iτ )∣∣ = √ πφ(|τ)|, thus ∞∫ −∞ ∣∣Φ( 1 2 + iτ )∣∣2 dτ = 2π ∞∫ 0 |φ(τ)|2dτ, where Φ is any of the four functions Φ+ c , Φ−c , Φ+ s , Φ−s . Comparing (31a), (28a) and (42a), we can see that the function (T+ c φ)(t) can be interpreted as the inverse Melline transform of the function Φ+ c , (T+ c φ)(t) = 1 2π ∞∫ −∞ t− 1 2 −iτΦ+ c ( 1 2 + iτ ) dτ. (44a) The Parseval equality transform, as applied to the inverse Melline transform of the function ϕ+ c (τ), yields ∞∫ 0 |(T+ c φ)(t)|2dt = 1 2π ∞∫ −∞ ∣∣Φ+ c ( 1 2 + iτ )∣∣2dτ = ∞∫ 0 |φ(τ)|2dτ. This is equality (41) for the transform T+ c . The functions T−c φ, T+ s φ, T−s φ can also be interpreted as the inverse Melline transforms: (T−c φ)(t) = 1 2π ∞∫ −∞ t− 1 2 −iτΦ−c ( 1 2 + iτ ) dτ (44b) and (T+ s φ)(t) = 1 2π ∞∫ −∞ t− 1 2 −iτΦ+ s ( 1 2 + iτ ) dτ, (45a) (T−s φ)(t) = 1 2π ∞∫ −∞ t− 1 2 −iτΦ−s ( 1 2 + iτ ) dτ. (45b) The Parseval equalities, as applied to the inverse Melline transform of the func- tions Φ−c , Φ+ s and Φ−s , yield equalities (41) for the transforms T−c , T+ s and T−s , respectively. Journal of Mathematical Physics, Analysis, Geometry, 2013, vol. 9, No. 4 489 V. Katsnelson 9. According to Lemma 5, the operators T+ c , T−c , T+ s , T−s are linear operators each of which is defined on the linear manifold L1(R+) ∩ L2(R+) of the Hilbert space L2(R+) and which maps this linear manifold into L2(R+) isometrically. Since the set L1(R+)∩L2(R+) is dense in L2(R+), each of these operators can be extended to an operator defined on the whole space L2(R+), which maps L2(R+) into L2(R+) isometrically. We will continue to write T+ c , T−c , T+ s and T−s for the extended operators. We now consider the operators T+ c , T−c , T+ s , T−s as the operators defined on all of L2(R+), mapping L2(R+) into L2(R+) isometrically and acting on the functions φ(t) ∈ L1(R+) ∩ L2(R+) according to (31). Theorem 3. 1. The range of values of the operator T+ c is the eigensubspace C+1 of the operator C; 2. The range of values of the operator T−c is the eigensubspace C−1 of the operator C; 3. The range of values of the operator T+ s is the eigensubspace S+1 of the operator S; 4. The range of values of the operator T−s is the eigensubspace S−1 of the operator S. R e m a r k 7. Since the operators T+ c , T−c , T+ s , T−s act isometrically from L2(R+) into L2(R+), the equalities (T+ c )∗T+ c = I, T+ c (T+ c )∗ = P+ c , CT+ c = T+ c ; (46a) (T−c )∗T−c = I, T−c (T−c )∗ = P−c , CT−c = −T−c (46b) and (T+ s )∗T+ s = I, T+ s (T+ s )∗ = P+ s , ST+ s = T+ s ; (47a) (T−s )∗T−s = I, T−s (T−s )∗ = P−s , ST−s = −T−s (47b) hold, where P+ c , P−c , P+ s and P−s are orthogonal projectors from L2(R)+ onto the eigensubspaces C+1, C−1,S+1 and S−1, respectively, and (T+ c )∗, (T−c )∗, (T+ s )∗, (T−s )∗ are the operators Hermitian-conjugated to the operators (T+ c ), (T−c ), T+ s ), (T−s ) with respect to the standard scalar product in the Hilbert space L2(R+). In particular, the operators (T+ c )∗, (T−c )∗, (T+ s )∗ and (T−s )∗ are generalized inverses ? of the operators T+ c , T−c , T+ s and T−s , respectively. ? In the sense of Moore–Penrose, for example. 490 Journal of Mathematical Physics, Analysis, Geometry, 2013, vol. 9, No. 4 Eigenfunctions of the Cosine and Sine Transforms It is worth mentioning that ( (T+ c )∗ x ) (τ) = ∫ R+ e+ c (t, τ)x(t) dt, ( (T−c )∗x ) (τ) = ∫ R+ e−c (t, τ)x(t) dt, (48a) ( (T+ s )∗x ) (τ) = ∫ R+ e+ s (t, τ) x(t) dt, ( (T−s )∗x ) (τ) = ∫ R+ e−s (t, τ) x(t) dt. (48b) Theorem 3 is a consequence of the following Lemma 6. Let a function x(t) belong to L2(R+), and x̂c(t) and x̂s(t) be the cosine and sine Fourier transforms of the function x: x̂c(t) = √ 2 π ∞∫ 0 x(s) cos(ts) ds, (49a) x̂s(t) = √ 2 π ∞∫ 0 x(s) sin(ts) ds. (49b) Let Φx(ζ), Φx̂c(ζ) and Φx̂s(ζ) be the Melline transforms of the functions x, x̂c and x̂s, respectively. (All three functions x, x̂c, x̂s belong to L2(0,∞), so their Melline transforms exist and they are the L2 functions on the vertical line Re ζ = 1 2 .) Then for ζ : Re ζ = 1 2 , the equalities Φx̂c(ζ) = Φx(1− ζ) · 2 ζ− 1 2 Γ ( ζ 2 ) Γ ( 1 2 − ζ 2 ) , (50a) Φx̂s(ζ) = Φx(1− ζ) · 2 ζ− 1 2 Γ ( 1 2 + ζ 2 ) Γ ( 1− ζ 2 ) (50b) hold. P r o o f. It is enough to prove Eqs. (50) assuming that the functions x(t), x̂c(t), x̂s(t) are continuous and belong to L2(R+) ∩ L1(R+): the set of these functions x is dense in L2(R) and all three transforms, cosine, sine and Melline transforms, act continuously from L2 to L2. Under these extra assumptions on the functions x(t), x̂c(t), x̂s(t), the Melline transforms Φx(ζ), Φx̂c(ζ), Φx̂c(ζ) are defined everywhere on the vertical line Re ζ = 1 2 and are continuous functions. For such x, Eqs. (50) will be established for every ζ : Re ζ = 1 2 . Journal of Mathematical Physics, Analysis, Geometry, 2013, vol. 9, No. 4 491 V. Katsnelson We fix ζ : Re ζ = 1 2 . The Melline transform Φx̂c(ζ) is Φx̂c(ζ) = lim R→∞ R∫ 0 x̂c(t)tζ−1dt. Substituting (49a) for x̂c(t) into the last formula, we obtain Φx̂c(ζ) = lim R→∞ R∫ 0 (√ 2 π ∞∫ 0 x(s) cos(ts) ds ) tζ−1 dt. (51) For fixed finite R, we change the order of integration R∫ 0 ( ∞∫ 0 x(s) cos(ts) ds ) tζ−1 dt = ∞∫ 0 x(s) ( R∫ 0 cos(ts) tζ−1 dt ) ds . The change of order of integration is justified by Fubini’s theorem. Changing the variable ts = τ , we get R∫ 0 cos(ts) tζ−1 dt = s−ζ Rs∫ 0 cos(τ) τ ζ−1 dτ . Thus R∫ 0 (√ 2 π ∞∫ 0 x(s) cos(ts) ds ) tζ−1 dt = ∞∫ 0 x(s)s−ζ (√ 2 π Rs∫ 0 cos(τ) τ ζ−1 dτ ) ds . (52) According to Lemma 1, for every s > 0, lim R→∞ Rs∫ 0 cos(τ) τ ζ−1 dτ = ( cos π 2 ζ ) Γ(ζ) . The value ρ∫ 0 (cos τ)τ ζ−1 dτ , considered as a function of ρ, vanishes at ρ = 0, is a continuous function of ρ, and has a finite limit as ρ →∞. Therefore there exists 492 Journal of Mathematical Physics, Analysis, Geometry, 2013, vol. 9, No. 4 Eigenfunctions of the Cosine and Sine Transforms a finite M(ζ) < ∞ such that the estimate holds ∣∣∣ ρ∫ 0 (cos τ)τ ζ−1 dτ ∣∣∣ ≤ M(ζ), where the value M(ζ) does not depend on ρ. In other words, ∣∣∣∣∣ Rs∫ 0 cos(τ) τ ζ−1 dτ ∣∣∣∣∣ ≤ M(ζ) < ∞ ∀ s,R : 0 ≤ s < ∞, 0 ≤ R < ∞ . By the Lebesgue dominated convergence theorem, lim R→∞ ∞∫ 0 x(s)s−ζ (√ 2 π Rs∫ 0 cos(τ) τ ζ−1 dτ ) ds = ∞∫ 0 x(s)s−ζ (√ 2 π ∞∫ 0 cos(τ) τ ζ−1 dτ ) ds . (53) Taking into account equalities (51), (53) and using (16a) and (20a), we reduce the last equality to the form Φx̂c(ζ) = ∞∫ 0 x(s) s−ζ ds · 2 ζ− 1 2 Γ ( ζ 2 ) Γ ( 1 2 − ζ 2 ) . To obtain (50a) from the previous equality, we need only to consider that ∞∫ 0 x(s) s−ζ ds = Φx(1− ζ) . Equality (50b) can be proved in a similar way. R e m a r k 8. Equalities (50) can be presented in the form Φx̂c ( 1 2 + iτ ) = Φx ( 1 2 − iτ ) · c 2(τ), (54a) Φx̂s ( 1 2 + iτ ) = Φx ( 1 2 − iτ ) · s 2(τ), (54b) where c(τ) and s(τ) were introduced in (30). P r o o f. [Proof of Theorem 3] Let xc(t) be defined by (49a). The equality Cx = x, i.e., the equality xc(t) = x(t) for the functions xc(t), x(t), is equivalent to the equality Φx̂c ( 1 2 + iτ ) = Φx ( 1 2 + iτ ) Journal of Mathematical Physics, Analysis, Geometry, 2013, vol. 9, No. 4 493 V. Katsnelson for their Melline transforms. According to Lemma 6, (54a), the last equality can be reduced ? to the form Φx ( 1 2 − iτ ) · c(τ) = Φx ( 1 2 + iτ ) · c(−τ), −∞ < τ < ∞. (55a) Analogously, the equalities Cx = −x, Sx = x and Sx = −x for the functions x(t) are equivalent to the equalities Φx ( 1 2 − iτ ) · c(τ) = −Φx ( 1 2 + iτ ) · c(−τ), −∞ < τ < ∞, (55b) and Φx ( 1 2 − iτ ) · s(τ) = Φx ( 1 2 + iτ ) · s(−τ), −∞ < τ < ∞, (56a) Φx ( 1 2 − iτ ) · s(τ) = −Φx ( 1 2 + iτ ) · s(−τ), −∞ < τ < ∞. (56b) Thus each of the equalities Cx = x, Cx = −x, Sx = x, Sx = −x for the function x(t), 0 < t < ∞, is equivalent to the symmetry condition for its Melline transform Φx ( 1 2 + iτ ) , −∞ < τ < ∞. These symmetry conditions, which appear as conditions (55), (56), can be presented in the form Φx ( 1 2 + iτ ) = √ π c(τ) φ(|τ |), −∞ < τ < ∞, Φx ( 1 2 + iτ ) = 1 i sign (τ) √ π c(τ) φ(|τ |), −∞ < τ < ∞, and Φx ( 1 2 + iτ ) = √ π s(τ) φ(|τ |), −∞ < τ < ∞, Φx ( 1 2 + iτ ) = 1 i sign (τ) √ π s(τ) φ(|τ |), −∞ < τ < ∞, where φ(τ) is the function defined for 0 < τ < ∞. Comparing these expressions for the function Φx ( 1 2 + iτ ) with the expressions (28), (29) for the eigenfunctions e+ c (t, τ), e−c (t, τ), e+ s (t, τ), e−s (t, τ), we can see that in each of the four cases the inversion formula x(t) = 1 2π ∞∫ −∞ t− 1 2+iτΦx ( 1 2 + iτ ) dτ for the Melline transform can be presented in terms of the function φ(τ) as x(t) = ∞∫ 0 e+ c (t, τ) φ(τ) dτ, x(t) = ∞∫ 0 e−c (t, τ) φ(τ) dτ, (57a) x(t) = ∞∫ 0 e+ s (t, τ) φ(τ) dτ, x(t) = ∞∫ 0 e−s (t, τ) φ(τ) dτ, (57b) ? Remember that c−1(τ) = c(−τ). 494 Journal of Mathematical Physics, Analysis, Geometry, 2013, vol. 9, No. 4 Eigenfunctions of the Cosine and Sine Transforms respectively. Now the symmetries (55), (56) of the function Φx ( 1 2 +iτ ) are hidden in the structure of the functions e+ c , e−c , e+ s , e−−. Thus, the equalities Cx = x, Cx = −x and Sx = x, Sx = −x for the functions x are equivalent to the representability of x in one of the four forms (57), i.e., in the form x = T+ c φ,x = T−c φ, x = T+ s φ and x = T−s φ, respectively (with φ ∈ L2(R+)). Acknowledgements. I thank Armin Rahn for his careful reading of the manuscript and his help in improving English in this paper. References [1] N. Wiener, The Fourier Integral and Certain of its Applications. Cambridge Univ. Press, Cambridge, 1933. [2] G.H. Hardy and E.C. Titchmarsh, Self-Reciprocal Functions. — Quart. J. Math., Oxford, Ser. 1 (1930), 196–231. [3] G.H. Hardy and E.C. Titchmarsh, Formulae Connecting Different Classes of Self- Reciprocal Functions. — Proc. London Math. Soc., Ser. 2 33 (1931), 225–232. [4] E.C. Titchmarsh, Introduction to the Theory of Fourier Integrals. Clarendon Press, Oxford, 1937. Third Edition. Chelsea Publishing Co., New York, 1986. [5] J.L. Burchnall, Symbolic Relations Assotiated with Fourier Transforms. — Quart. J. Math. 3 (1932), 213–223. [6] E.T. Whittaker and G.N. Watson, A Course of Modern Analysis. Fourth Edition. Cambridge Univ. Press, Cambridge, 1927. [7] Encyclopaedia of Mathematics. Kluver Acad. Publ., Dordracht–Boston–London, 1995. Journal of Mathematical Physics, Analysis, Geometry, 2013, vol. 9, No. 4 495