Publication Cover
Applicable Analysis
An International Journal
Volume 99, 2020 - Issue 16
1,672
Views
10
CrossRef citations to date
0
Altmetric
Articles

Trigonometric and cylindrical polynomials and their applications in electromagnetics

Pages 2807-2822 | Received 04 Jan 2019, Accepted 14 Feb 2019, Published online: 27 Feb 2019

ABSTRACT

A development of the theory of trigonometric polynomials (TPs) is considered that involves generalization of the notion of TP and extension of the methods and results to cylindrical polynomials (CPs). On the basis of the proposed augmentation of TPs and CPs, a general approach is presented to the analysis of guided waves and resonances in electromagnetics and beyond. The method employs the known explicit forms of dispersion equations (DEs) describing eigenoscillations and normal waves in layered structures and is based on the development of the theory of generalized TPs and CPs performed in the study. The approach enables one to complete rigorous proofs of existence and determine domains of localization of the TP and CP zeros and the DE roots and validate iterative numerical solution techniques.

COMMUNICATED BY:

2010 MATHEMATICS SUBJECT CLASSIFICATIONS:

1. Introduction

Trigonometric polynomials (TPs) arise in many areas of pure and applied mathematics: algebra of polynomials and complex analysis [Citation1,Citation2], Fourier analysis, approximation theory, numerical methods and interpolation [Citation3] to name a few, as well as in various applications including signal processing [Citation4]. TPs occur (see e.g. [Citation5] and references therein) in the search for eigenvalues of non-selfadjoint Sturm–Liouville problems on the semi-axis, whole axis, or an interval with piecewise constant coefficients in the equations and local or non-local boundary or transmission conditions or conditions at infinity containing spectral parameter.

Analysis of TPs including the existence and distribution of their (real or complex) zeros constitutes a specific direction of real and complex analysis developed in many classical works. Information about the existence and location of zeros is essential property of TPs (as it is the case of any polynomial). The proofs of the existence and investigations of distribution of the TP zeros constitute complicated mathematical problems addressed during the last two centuries by many famous mathematicians: Littlewood [Citation2,Citation6,Citation7], Erdelyi [Citation8,Citation9], Lanczos [Citation10], and others [Citation11].

Let us give an example (taken from an extended collection of problems [Citation12]) and a typical sophisticated task of the TP theory. A 'regular' TP (1) LN(θ)=j=0NCjcosjθ+Sjsinjθ(1) where coefficients Cj and Sj do not vanish simultabeously has exactly 2N zeros. However, generally, the occurrence of zeros of a TP may constitute a severe problem. Indeed, a cosine TP (2) LM(θ)=j=1Mcosnjθ(2) was considered by Littlewood in [Citation2] who posed the following problem: if the nj are integral and all different, what is the lower bound on the number of real zeros of TP (Equation2)? Possibly M−1, or not much less. No progress has been made on this problem in the last half century until 2008 when it has been shown [Citation13] that there exists a cosine polynomial (Equation2) with the nj integral and all different so that the number of its real zeros in the period [π,π) is O(M5/6logM). Below we make another contribution to the analysis of the Littlewood TPs and prove a statement (Corollary 5.3) estimating the number of zeros for a certain subfamily of TPs (Equation2). Figure  gives an example of a TP (Equation2) of order 100 with the estimated number 21005/6 of zeros located in [π,π) to be around 93.

Figure 1. TP L100(x)=m=1100cosnmx, x(0,π), with randomly distributed digits 1nm106.

Figure 1. TP L100(x)=∑m=1100cos⁡nmx, x∈(0,π), with randomly distributed digits 1≤nm≤106.

Many approaches in the TP theory deal with introduction of various TP families and obtaining different sufficient conditions providing the existence of the TP zeros valid for each particular family or class of TPs. In many cases, the developed approaches enable one to describe distribution of zeros and determine their number (over a period). Let us give two examples illustrating such methods. In [Citation9] it is examined the size of a real TP of degree at most n having at least k zeros in K:=R (mod 2π) (counting multiplicities). In [Citation14] the following sufficient conditions are obtained for the existence of zeros of a family of cosine TPs (3) CM(x)=12+m=1Mamcosmx,aM=0,(3) where {am}, a0=1, is a monotonically decreasing (am+1am<0, m=0,,M1) and convex (am+22am+1+am<0, m=0,,M2) number sequence: TP (Equation3) has one simple zero on every interval (mπ/M,(m+1)π/M), m=1,,M1.

An example of the TP generalization according to [Citation11] has the form (4) TM(x)=12a0+m=1Mamcosmθ,aM0,(4) where θ=Ψ(x) is a continuous function with positive (or negative, but not both) first derivative for x real. In particular, Tk(x)=cos(karccosx), x[1,1], are the Chebyshev polynomial of degree k.

Cylindrical polynomials (CPs) have much in common with TPs; however, their theory is much less developed. CPs occur when particular solutions to Maxwell equations are sought in polar or cylindrical coordinates in domains where boundary and interface contours or surfaces possess circular symmetry. Standard reference books on cylindrical functions (see e.g. [Citation15–17]) consider two-term sums of products of two cylindrical functions occurring, e.g. in the form [Citation18,Citation19] Jn(κ1x)Yn(κ2x)Jn(κ2x)Yn(κ1x) (n=0,1,2,) of cross-products of Bessel functions. Extension to the greater number of factors in the products or greater number of product terms has not been performed, to the best of our knowledge.

This study follows and contribute to the three main directions of the TP and CP theories: creating a universal form of generalized TPs; extension of the methods of the TP analysis to the class of CPs involving cylindrical functions of different kinds and order; and obtaining new sufficient conditions providing the existence and description of distribution of zeros of certain families of TPs and CPs. The latter employs non-periodic TPs and analysis of their parameter dependence.

Another objective is application of the obtained results in the electromagnetic field theory, namely, to the determination of real and complex oscillations and waves. The latter are reduced to multi-parameter eigenvalue problems [Citation20,Citation21] and then in many cases (when, e.g. the Helmholtz equation admits closed-form solutions using separation of variables) to dispersion equations (DEs). Their complete study is a sophisticated task which requires deep analytical–numerical investigations. In this work, a major attention is paid to creating an introduction to a general method that enables obtaining (sufficient) conditions of the existence and description of localization of the DE roots providing thus justification for the methods of determination of oscillations and waves in multi-layered structures which can be easily implemented in calculations. The method is based on the studies of the so-called generalized TPs (GTPs) and generalized CPs (GCPs) aimed in particular to finding zeros of GTPs and GCPs using different approaches.

The approach proposed in this study makes use of the method [Citation22,Citation23] employing GCPs applied in [Citation23] to the rigorous analysis and determination of real waves in a dielectric waveguide (DW) and a Goubau line (GL) of circular cross section. The technique involves recursive procedures of the determination of zeros of TPs, GTPs, and GCPs.

2. Trigonometric polynomials

TPs are weighted linear combinations (5) TM(x)=q=12m=0MqPq(m)(x)Uq(nq,mx)orEM(t)=m=0MP(m)(t)einmt(5) of complex exponents or trigonometric functions U1(y)=siny (q=1) or U2(y)=cosy (q=2), where M=M1+M2 (for TM), nq,m and nm are integers forming integer 'frequency' parameter vectors κ¯q,M=(nq,1,,nq,M), and Pq(m)(x) and P(m)(x) (m=1,2,,Mq, q=1,2) are constants or bounded continuous functions considered on the whole line, half-line x>0, or an interval. A cosine TP (Equation2) considered by Littlewood has all P(m)(x)=1 and is therefore 2π-periodic. One also separates the cases M1=M2=M0 and n1,m=n2,m=m or nm=m and all Pq(m)(x)=const (m=1,2,,M0) giving classical (2π-periodic) TPs, even (a cosine TP with all P1(m)(x)=0) or odd (a sine TP with all P2(m)(x)=0). A 'regular' TP (Equation1) constitutes, as well as TPs (Equation2)–(Equation4), examples of such polynomials; several more are presented in the appendix.

Together with TPs (Equation5) we will study more general cases of non-periodic TPs considering wieghted trigonometric sums (TSs) (6) GM(x)=q=12m=0MqPq(m)(x)Uq(κq,mx)(6) with generally non-integer ‘frequency’ parameter vectors κ¯q,M=(κq,1,,κq,M), q=1,2. TS (Equation6) is a TP having the form (Equation5) of the order M=M1+M2.

Many standard examples of TPs and TSs calculated in the closed form for which it is possible to determine all zeros explicitly are considered in the appendix.

3. Cylindrical polynomials

In this study, we call CPs the weighted linear combinations (7) CM(x)=m=0MQ(m)(x)Im(κmx)(7) of cylindrical functions Im, where each Im is either Bessel, Jp, or Neumann, Yp, function (p=0,1,2,), κm are real numbers forming parameter vectors κ¯M=(κ1,,κM), and Q(m)(x) (m=0,1,,M) are constants or bounded continuous functions considered on the whole line, half-line x>0, or an interval. Unlike classical 2π-periodic TPs and TSs (EquationA1)–(EquationA6) with integer parameter vectors, CPs (Equation7) are non-periodic (and in this respect they are similar to TPs (Equation6)) although every term in (Equation7) is an almost-periodic function. Namely, the following statements hold: (i) every function Im has a countable sequence of positive simple zeros {zk(m)} increasing w.r.t. index k; (ii) the distance dk(m)=zk+1(m)zk(m) (k=0,1,2,) between two neighboring zeros of Im tends to a constant (in particular, to π for the Bessel functions) as x; (iii) there is a number r>0 such that r=mink=0,1,dk(m) (m=1,2,); and (iv) zeros of every two different cylindrical functions Jm(x), Jn(x), Ym(x), Yn(x) (mn), or Jm(x), Yn(x) (m,n=0,1,) alternate.

Analysis of CPs including the existence and distribution of their (real or complex) zeros constitutes a specific set of complicated mathematical problems that has been rarely addressed in the literature as compared with TPs (where one may refer, e.g. to [Citation1,Citation2,Citation10,Citation12]).

4. Generalized TPs and CPs

Consider weighted linear combinations (8) TN,M(x)=m=1MQ(m)(x)U(m,Nm)(x)(8) or (9) GN,M(x)=m=1MP(m)(x)W(m,Nm)(x)(9) of the products U(m,Nm)=Πn=1NmU1,2(κn,T(m)x) or W(m,Nm)(x)=Πn=0NmIn(m)(κn,G(m)x) of trigonometric, U1,2, or cylindrical, In(m), functions, where N=m=1MNm, P(m)(x) and Q(m)(x) are constants or bounded continuous functions for x>0 and κn,T(m) and κn,G(m) are real parameters forming the ‘frequency’ parameter vectors κ¯T(m)=(κ1,T(m),,κNm,T(m)) and κ¯G(m)=(κn,G(1),,κn,G(Nm)), m=1,2,,M.

In this study, functions (Equation8) or (Equation9) are referred to, respectively, as generalized TPs (GTPs) or GCPs of order (N,M). In particular, for TPs (Equation5) and (Equation6) (10) GM(x)=m=1M1P1(m)(x)sin(κ1,mx)+m=0M2P2(m)(x)cos(κ2,mx)=TM,M(x)(10) all Nm=1 so that they are GTPs of order (M,M) with M=M1+M2.

In the studies of cylindrical functions (e.g. in [Citation15–17]) zeros are considered of the two-term sums of products of two cylindrical functions; that is, of G4,2(x)=m=12W(m,2)(x), where W(m,2)(x)=In1,m(m)(κn1,m(m)x)In2,m(m)(κn2,m(m)x) with n1,2,m denoting the order of two cylindrical functions In1,m(m) and In2,m(m) that have either different kind, or the same kind and different order or different parameters κn1,m(m) and κn2,m(m) (m=1,2). Such two-term sums of products, e.g. Jn(κ1x)Yn(κ2x)Jn(κ2x)Yn(κ1x) (n=0,1,2,) as in [Citation19] are called cross-product Bessel functions.

Lemma 4.1

A product U(m,N)(x)=Πn=1NU1,2(κn,T(m)x) of trigonometric functions U1,2 can be represented as a GTP of order (M,M) with a certain M=M(N), that is, as a TP (Equation6) of order M.

Proof.

The first step of the induction proof is to check basic trigonometric identities sinxsiny=0.5(cos(xy)cos(x+y)), sinxcosy=0.5(sin(xy)+sin(x+y)), and cosxcosy=0.5(cos(xy)+cos(x+y)) and to apply them to the triple products, e.g. sinxsinysinz=0.5(sinzcos(xy)sinzcos(x+y))=0.25(sin(zx+y)+sin(x+yz)+sin(z+xy)+sin(xyz)) so that Πn=13U1(κnt)=0.25m=14U1(κnt)=T4,4(x)=G4(x) with x=κ1t, y=κ2t, z=κ3t, κ1=κ1+κ2+κ3, κ2=κ1+κ2κ3, κ3=κ1κ2+κ3, and κ4=κ1κ2κ3.

Assuming now that (11) Πn=1N1U1,2(κnx)=m=0M(N1)PqU1,2(κnx),(11) one can easily verify, after converting each product of two trigonometric functions to a sum, that Πn=1NU1,2(κnx)=U1,2(κNx)Πn=1N1U1,2(κnx)=m=0M(N)PmU1,2(κnx).

Corollary 4.2

A GTP (Equation8) of order (N,M) can be represented as a TP (Equation6) of a certain order M1=M1(M,N).

Proof.

Every product U(m,Nm) in (Equation8) can be written according to Lemma 4.1 as a GTP of order (Mm,Mm) with a certain Mm=Mm(Nm). Consequently, GTP (Equation8) of order (M,N) can be represented as a sum of TPs (Equation6) and finally as a TP (Equation6) of the order M=m=1MMm.

Lemma 4.3

Denote by Δm(1)(α)=(xs(1,m)(α),xs(1,m+1)(α)) and Δl(2)(β)=(xc(1,l)(β),xc(1,l+1)(β)), m,l=1,2,, intervals formed by two pairs of neighboring positive zeros of sinαx and cosβx given by (EquationA9) and (EquationA10) with α,β0. Then for any three distinct positive numbers x1>0, x, and x2, x1<x<x2, there are α>0, β>0, and integers m and l such that xΔm(1)(α) and Δm(1)(α)(x1,x2) or xΔl(2)(β) and Δl(2)(β)(x1,x2).

Proof.

Endpoints of Δm(1)(α) and Δl(2)(β) form equidistant (uniform) grids of points on the half-line x>0 separated by d1=πα1 and d2=πβ1. Therefore, there is an interval between certain neighboring grid points of each grid that contains one, two, or three of the numbers x1, x, and x2 (we assume that x does not coincide with any of the grid points). Choosing α or β such that d1,2<min{xx1,x2x}<x2x1 we see that there are integers m and l such that xΔm(1)(α) or xΔl(2)(β) and Δ,m,l(1,2)(x1,x2).

5. Zeros of GTPs and GCPs

In order to formulate sufficient conditions that guarantee the existence of zeros of GTPs and GCPs and describe their localization, we will use the following

Lemma 5.1

Let fj(x)C[a,b], j=1,2,3 (continuous in a closed interval [a,b]), f2(a)=f2(b)=0, and f1(a)f1(b)<0 or, equivalently, there is an s(a,b) such that f1(s)=0; then the equation f(x)f2(x)f3(x)+f1(x)=0 has a root x(a,b).

Proof.

Indeed, f(a)=f1(a) and f(b)=f1(b) so that f(a)f(b)<0 and therefore f(x)=0 has a root x=x(a,b).

5.1. GTPs

We have already noted that TPs (EquationA1)–(EquationA6) have each infinitely many zeros for any number of their terms N=2,3, and that infinitely many zeros of 'neighboring' TPs SN, SN1, CN, CN1, and TN, TN1 (N=2,3,) alternate. This statement can be in a certain sense generalized for arbitrary GTP using Lemma 5.1, validating simultaneously a recursive procedure of proving the existence and determining the location of zeros of TPs, GTPs, and GCPs.

Theorem 5.2

There are (in general, non-integer) frequency vectors κ¯q,M=(κq,1,,κq,M), q=1,2, such that a TP (Equation6) of order 2M where P(m)(x) are arbitrary constants or bounded continuous functions that do not vanish on the half-line x>0 has a positive zero.

Proof.

The first step of the induction proof is to call that TPs of orders 1 and 2, sin(κ1,1x), cos(κ2,1x), and P1sinκ1,1x+P2cosκ2,1x, have each infinitely many positive zeros for any P1,2=const, P12+P220, and κq,10, q=1,2. The latter statement is easily extended to the case when P1,2=P1,2(x) are any two functions satisfying the condition of the theorem. Next, assume that TP G2M2(x) in the form (Equation6) of order 2(M1) has a positive zero xM1 and consider TPs of order 2M−1 (12) G1,2M1(x)=P1(M)(x)sin(κ1,Mx)+G2M2(x),(12) (13) G2,2M1(x)=P2(M)(x)cos(κ2,Mx)+G2M2(x).(13) Applying Lemmas 4.3 and 5.1 with (a,b) being an interval formed by two neighboring zeros of sin(κ1,Mx) or cos(κ2,Mx) and choosing κ1,M or κ2,M in line with Lemma 4.3 with respect to xM1(a,b) we prove the statement of the theorem.

Statement of Theorem 5.2 is valid for cosine TP (Equation2) considered by Littlewood in the form stronger than that for non-periodic TPs and GTPs. Namely, one can estimate the number of zeros on half-period (0,π] and establish the existence of infinitely many positive zeros of TP (Equation2).

Corollary 5.3

TP (Equation2) of order M2 where n1 and n2, n2>n1, are two aribtrary (different) integers, has at least 2n2 zeros on half-period (0,π] and infinitely many positive zeros.

Proof.

The first step of the induction proof is to call that TPs (Equation2) of orders 1 and 2, L1(x)=cosn1x and L2(x)=cosn21x+cosn22x=2cos((n22+n21)/2)xcos((n22n21)/2)x, have each infinitely many positive zeros (14) xn1(1,k)=(2k1)π2n1,xn+(2,l)=(2l1)πn+,xn(3,m)=(2m1)πn,k,l,m=1,2, ,(14) for any integers n1, n21, and n22, n22>n21, with n±=n22±n21; in particular, L1(x) has n1 zeros and L2(x) has n22=n+/2+n/2 (when parity of n+ and n is the same) or n22+1=(n++1)/2+(n+1)/2 (parity of n+ and n is different) zeros on half-period (0,π]. Thus, zeros of L2(x) form a finite set of distinst points on the interval (0,π). To show details of the proof of the induction step, consider L3(x)=cosn3x+L2(x). According to Lemma 4.3, one can choose n3 such that intervals Δn3(s)=(xn3(1,s),xn3(1,s+1)), s=1,2,,n22eo (where n22eo=n22 or n22eo=n22+1) between two neighboring zeros of cosn3x contain each one zero of L2(x). This means, in line with Lemma 5.1, that L3(x) has a zero on each Δn3(s). Consider now LM(x)=cosnMx+LM1(x) and assume that {xM(p)}p=1p=LM1 are LM1 zeros of LM1(x) on (0,π). According to Lemma 4.3, one can choose nM such that intervals between two neighboring zeros of cosnMx contain each one zero of LM1(x). Therefore, Lemma 5.1 implies that the corollary is proved.

Figures  and  illustrate the proofs and display examples of TPs (Equation2) of different orders. Comparing Figures  and  one observes a clear similarity between CPs and TPs with respect to the statements of Theorems 5.2 and 5.4 and Lemma 4.3.

Figure 2. TPs L3(x)=cos3x+cos7x+cos20x (M=3, n22=7, n21=3, n3=20) with L2(x)=cos3x+cos7x=2cos2xcos5x, x(0,π); L2(x) (o) has 7 zeros located together with 7 zeros of L3(x) (*) between pairs of neighboring zeros of cos20x.

Figure 2. TPs L3(x)=cos⁡3x+cos⁡7x+cos⁡20x (M=3, n22=7, n21=3, n3=20) with L2(x)=cos⁡3x+cos⁡7x=2cos⁡2xcos⁡5x, x∈(0,π); L2(x) (o) has 7 zeros located together with 7 zeros of L3(x) (*) between pairs of neighboring zeros of cos⁡20x.

Figure 3. TPs L9(x)=m=19cosnmx (lower curve with greater number of oscillations) with κ¯M=(n1,,n9)=(7,13,2,15,8,4,11,17,6), L8(x)=L9(x)cos6x (upper curve with greater number of oscillations), and cos6x, x(0,π); zeros of TPs alternate and L9(x) and L8(x) have each 11 zeros located between pairs of neighboring zeros of cos6x.

Figure 3. TPs L9(x)=∑m=19cos⁡nmx (lower curve with greater number of oscillations) with κ¯M=(n1,…,n9)=(7,13,2,15,8,4,11,17,6), L8(x)=L9(x)−cos⁡6x (upper curve with greater number of oscillations), and cos⁡6x, x∈(0,π); zeros of TPs alternate and L9(x) and L8(x) have each 11 zeros located between pairs of neighboring zeros of cos⁡6x.

Figure 4. CPs C3(x)=J0(3x)+J0(7x)+J0(20x) with C2(x)=J0(3x)+J0(7x), x(0,π); C2(x) (o) has 4 zeros located together with 6 zeros of C3(x) (*) between pairs of neighboring zeros of J0(20x).

Figure 4. CPs C3(x)=J0(3x)+J0(7x)+J0(20x) with C2(x)=J0(3x)+J0(7x), x∈(0,π); C2(x) (o) has 4 zeros located together with 6 zeros of C3(x) (*) between pairs of neighboring zeros of J0(20x).

5.2. GCPs

According to Corollary 4.2, Theorem 5.2 can be applied to GTP (Equation8). However it is reasonable to give an independent proof which provides sufficient conditions of the existence of zeros of both GTPs (Equation8) and GCPs (Equation9).

Theorem 5.4

There are ‘frequency’ vector sets ΘT(M)=m=1Mκ¯T(m) and ΘG(M)=m=1Mκ¯G(m), where κ¯T(m)=(κ1,T(m),,κNm,T(m)) and κ¯G(m)=(κ1,G(1),,κNm,G(Nm)) (m=1,2,,M) have real components, such that GTP (Equation8) and GCP (Equation9) of order (N,M) (N=m=1MNm) where Q(m)(x) and P(m)(x) (m=1,2,M) are arbitrary constants or bounded continuous functions that do not vanish on the half-line x>0 have each a positive zero.

Proof.

First, note that at M=1, GTP (Equation8) and GCP (Equation9) of order (1,N1) are single products having each (as a function of x) infinitely many positive zeros forming countable sets TN1(1)=n=0N1tn(1) and ZN1(1)=n=0N1Zn(1) being unions of the infinite sets tn(1) and Zn(1) of the zeros τn,k(1) and zn,k(1) (k=1,2,) of all U1,2(κn,T(1)x) and In(1) that enter these products.

Second, for any M=2,3, and m=1,2,M, the products U(m,Nm)(x) and W(m,Nm)(x) comprising Nm factors have each infinitely many positive zeros forming countable sets TNm(m)=n=0Nmtn(m) and ZNm(m)=n=0NmZn(m) being unions of the infinite sets tn(m) and Zn(m) of the zeros τn,k(m) and zn,k(m) (k=1,2,) of all U1,2(κn,T(m)x) and In(m). Elements of tn(m) and Zn(m) (and of TNm(m) and ZNm(m)) depend on the parameter vectors κ¯T(m) and κ¯G(m) (m=1,2,M).

Perform the next step of induction and represent GTP (Equation8) and GCP (Equation9) as (15) TN,M(x)=Q(M)(k(M)x)U(M,NM)(x)+TN,M1(x),(15) (16) GN,M(x)=P(M)(k(M)x)W(M,NM)(x)+GN,M1(x),(16) where N=m=1M1Nm, U(M,NM)(x) and W(M,NM)(x) vanish at the endpoints of the intervals un,k(M)=(a,b)=(τn,k(M),τn,k+1(M)) and In,k(M)=(a,b)=(zn,k(M),zn,k+1(M)) (n=1,2,NM) between every two their neighboring zeros.

Assume that the parameter vector sets ΘT(M1) and ΘG(M1) in (Equation15) and (Equation16) are such that GTP TN,M1(x) and GCP GN,M1(x) of order (M1,N) have each a zero xT>0 or xG>0. Choose ΘT(M1) and ΘG(M1) according to Lemma 4.3 with (a,b) being intervals un,k(M1) or In,k(M1) formed by two neighboring zeros of U(M,NM)(x) or W(M,NM)(x) with index pairs (n,k) or (n,k) such that xT(a,b) or xG(a,b). Next, use Lemma 5.1 to conclude that GTP (Equation8) and GCP (Equation9) of order (M,N) have each a zero on these intervals proving thus the statement of the theorem.

6. Application to DEs in electromagnetic field theory

6.1. Waveguides with circularly symmetric layered media

When electromagnetic wave propagation is considered [Citation20,Citation24] in waveguides having circular symmetry, like DWs or GLs formed by several concentric layers of media, all the field components of symmetric and non-symmetric waves are expressed [Citation21,Citation24] via a potential function φ(r) which is sought generally as a linear combination of cylindrical functions of order m=0,1,2,. This fact enables one to reduce finally the determination of the wave propagation constants to the solution of DEs involving CPs and GCPs. Further stages of the mathematical model are described in the next section.

6.1.1. Sturm–Liouville problems

Analysis of running (normal) waves in waveguides filled with circularly symmetric layered media is reduced [Citation21,Citation22,Citation25] to the non-selfadjoint Sturm–Liouville problems for the Bessel equation with piecewise constant coefficients on the semi-axis (17) Lφ1rddrrdφdr+k~2φ=0,r>0,[φ]r=a=α1(γ)dφdrr=a=0,(17) (18) φC1[0,+)C2(0,a)C2(a,+),φ(r)0,r, {(DW)},Lφ=0,r>a,φ(a)=0,[φ]r=b=α2(γ)dφdrr=b=0,(18) (19) φC1[a,+)C2(a,b)C2(b,+),φ(r)0,r, {(GL)};(19) here k~2=k02β2,r>a,εk02β2,r<a {(DW)},k~2=k02β2,r>b,εk02β2,a<r<b {(GL),} β is the wave propagation constant (spectral parameter of the problem), ε and k0 are, respectively (real or complex) permittivity of the media and wavenumber of vacuum (a real parameter), and α1,2(γ) with γ=β/k0 are given expicitly for a particular structure.

Finding eiganvalues of (Equation17) and (Equation18) (propagation constants of running waves) is reduced [Citation20–26] to the solution of functional equations w.r.t. spectral parameter γ or λ=γ2 usually called DEs. When real spectrum of problems (Equation17) and (Equation18) is considered, the quantities are sought as real-valued functions of real spectral parameter γ (or λ) varying on a certain interval γ(γ1,γ2).

For multi-layered DWs or GLs when the number M>1 of dielectric layers (that is, discontinuities of the coefficient in the differential equations entering (Equation17) and (Equation18)) may be arbitrary, transmission conditions in (Equation17) and (Equation18) comprise 2M jump relations with different α1,2(j)(γ) at the layer boundaries r=aj, aj+1>aj>0, for DW or r=bj, bj+1>bj>a>0 (j=1,2,M1), for GL.

The general results of the classical Sturm–Liouville theory concerning the existence and distribution of (real or complex) spectrum for this type of problems are not applicable because the boundary (transmission) conditions depend on the spectral parameter; this dependence which is specified by concrete functions obtained explicitly governs the presence or absence of spectrum.

The explicit expressions for the DEs obtained in [Citation22–24] for multi-layered DWs and GLs (open or shielded) or structures formed by plane-parallel layers of media [Citation27,Citation28] show that these DEs involve GCPs where weight coefficients P(m)(x) and Q(m)(x) are constants or bounded continuous functions for x>0 determined explicitly and κn,T(m) and κn,G(m) forming the parameter vectors κ¯T(m)=(κ1,T(m),,κNm,T(m)) and κ¯G(m)=(κn,G(1),,κn,G(Nm)), m=1,2,,M are real quantities expressed in terms of parameters of the particular structure (Figure ).

Figure 5. Zeros of a GCP Fg (curve with highest oscillation) given by (Equation21) situated between neighboring zeros of Φ0 (curve with a negative starting value) and Φ1 (curve with a positive starting value).

Figure 5. Zeros of a GCP Fg (curve with highest oscillation) given by (Equation21(21) Fg(x)≡PD(qw)Φ1(x)−qxΦ0(x)=0(GL),(21) ) situated between neighboring zeros of Φ0 (curve with a negative starting value) and Φ1 (curve with a positive starting value).

6.1.2. Single-layer DW and GL

The DEs for single-layer DW and GL can be represented [Citation21–23,Citation25,Citation26] in the general form involving GCPs (Equation9) (20) Fd(x)PD(w)J1(x)+xJ0(x)=0(DW),(20) (21) Fg(x)PD(qw)Φ1(x)qxΦ0(x)=0(GL),(21) where Jm, Ym and Km (m=0,1) are the Bessel, Neumann, and Macdonald functions, PD(w)=ε(wK0(w)/K1(w)), Φ0=J0(qx)Y0(x)J0(x)Y0(qx) and Φ1=J0(x)Y1(qx)J1(qx)Y0(x) are Bessel cross-products [Citation19], and the waveguide geometric and material parameters x=κεγ2, γ=βk0, κ=k0a, u=κε1 (β, ε, and k0 are, respectively, the wave progation constant, permittivity, and free-space wavenumber), w=u2x2=κγ21, and q=(b/a)>1 (a and b are characteristic dimensions of DW and GL).

For a GCP Fg in (Equation21) we can apply this reasoning and Lemma 5.1 by setting f1(x)=qxΦ0(x) and f2(x)=Φ1(x). Then Fg(x) has a zero between every two neighboring zeros hk(1) and hk+1(1) of Φ1(x) as soon as q>1 is such that a zero of Φ0(x) belongs to the interval (hk(1)hk+1(1)). The latter condition can be satisfied because zeros hk(j)=hk(j)(q) of Φj(x) alternate for different j=1,2 and form sequences of points decreasing with respect to q [Citation21,Citation22,Citation26]. The conclusion concerning the existence and location of the zeros of Fg between neighboring alternating zeros of Φ0(x) and Φ1(x) is perfectly illustrated by Figure .

Straightforward analysis of (Equation20) and (Equation21) demonstrates that functions Fd and Fg entering DEs have distinct common features: they are sums of (products of) cylindrical functions Jm and Φm each having infinitely many alternating simple positive zeros. The latter yields an immediate proof (illustrated by Figure  and verified below) of the (sufficient conditions) providing the existence of real roots of the DEs located between zeros of Jm and Φm (m=0,1). The existence, localization, and number of the DE roots are governed actually by a number of zeros of Jm or Φm that are inside the domain x(0,u) of Fd and Fg; that is, by the value of parameter u.

Figure 6. Example for a GCP Fd(x) in DE (Equation20): plots of J0(x) (upper curve with no oscillations), J1(x) (lower curve with no oscillations), and Fd(x) (curve with one oscillations) at ε=5 and κ=2 (u=κε1=4) displaying a zero of Fd(x) between neighboring zeros of J1(x) and neighboring zeros of J0(x) and J1(x).

Figure 6. Example for a GCP Fd(x) in DE (Equation20(20) Fd(x)≡PD(w)J1(x)+xJ0(x)=0(DW),(20) ): plots of J0(x) (upper curve with no oscillations), J1(x) (lower curve with no oscillations), and Fd(x) (curve with one oscillations) at ε=5 and κ=2 (u=κε−1=4) displaying a zero of Fd(x) between neighboring zeros of J1(x) and neighboring zeros of J0(x) and J1(x).

Figure 7. An example showing alternating zeros of Φ0(x) (curve starts at −0.3), Φ1(x) (curve starts above 0.2), and G2,4(G)=PGΦ0Φ1 (curve starts above 0).

Figure 7. An example showing alternating zeros of Φ0(x) (curve starts at −0.3), Φ1(x) (curve starts above 0.2), and G2,4(G)=PGΦ0−Φ1 (curve starts above 0).

6.1.3. Multi-layered waveguides

For M-layer DWs or GLs with M>1 DE takes the form [Citation29] of the determinant equation (22) FM(γ)detA(γ)=0,A=apq(γ)p,q=12M,(22) where ζ(γ) is a one-to-one function of the spectral parameter considered on a certain interval γ(γ1,γ2) and apq(γ)=I(pq)(κG(p,q)ζ(γ)) are cylindrical functions, either of different order or type or with different parameter factors κG(p,q) (p,q=1,2,2M).

Thus, solution to Equation (Equation22) reduces to the determination of zeros of a GCP (Equation9) obtained as a result of calculation of the determinant. The GCP involves products Πn=0Np,qIn(p,q)(κn,G(p,q)ζ(γ)) of the cylindrical functions entering (Equation22).

A is a block-diagonal matrix obtained explicitly in [Citation29] which yields the possibility of recursive computation of its determinant and obtaining formulas similar to (Equation15) (23) FM=I(11)(κG(1,1)ζ)FM11I(12)(κG(1,2)ζ)FM12(23) using low-order minors FM11,2 and cofactors so that at the first (and each subsequent) recursive step the expression for FM(γ) contains two terms. This enables one to apply the procedure outlined in the proof of Theorem 5.4: Assume that parameter κG(1,1) in (Equation23) is such that an interval uk(11) formed by two neighboring zeros of I(11)(κG(1,1)ζ) with index k contains one particular zero of I(12)(κG(1,2)ζ). Such a choice of parameter is possible because each I(1p)(κG(1,p)ζ), p=1,2, has infinitely many zeros forming almost periodical sequences and each zero (as well as the distance between any two neighboring zeros) is a monotonically decreasing function of κG(1,p).

Next, we use Lemma 5.1 to conclude that FM in (Equation23) has a zero on the interval uk(11)

6.2. Plane-parallel layered guiding structures

Determining resonant states and eigenfrequencies of plane-parallel layered dielectrics in free space, between parallel perfectly conducting planes, or in a waveguide of rectangular cross section is considered in terms of non-selfadjoint Sturm–Liouville problems (Equation17) for the equation Lφφ+k~2φ=0 on the line with piecewise constant coefficients; eigenfunctions φ(r) are sought as a linear combination of trigonometric functions

and the problem in question is reduced [Citation27,Citation28] to DEs involving complex-valued TPs (24) G2(z)=G2(M)(z)=q=12Pq(z)Uq(κqz);(24) here z is a real or complex variable associated with one particular layer in an M-layer structure (e.g. z=κεsγ2 with s being the index of the layer with permittivity εs, 1sM) and P1,2(z) are complex-valued functions depending on all the problem parameters. In particular, for the single-layer structure (comprising one dielectric slab), we have [Citation27] (25) G2(z)=G2(1)(z)=P1(z)sin(tz)+cos(tz),P1(z)=iZ(C,z)=i2zC+Cz,(25) where t>0 and C>0 are (real) parameters. In [Citation27], it is proved that G2(1)(z) is an entire even function, the DE G2(1)(z)=0 has no real zeros and has infinitely many complex zeros located in pairs in the first and third quadrants in the complex z-plane. This result is extended to M-layer plane-parallel structures. Explicit (tedious) expressions for P1,2(z) in the cases of two and three layers may be found in [Citation27].

7. Conclusion

We have developed the theory of TPs and proposed a generalization of the notion of TP that can be applied to the analysis of CPs.

The proposed method is based on the introduction and analysis of GTPs and GCPs. The technique can be applied in electromagnetics using the explicit forms of DEs expressed as weighted sums of products of trigonometric and cylindrical functions that describe eigenoscillations and normal waves in layered structures. The approach enables one to complete rigorous proofs of existence and determine domains of localization of the DE roots and validate iterative numerical solution techniques.

The obtained results complete mathematical theory of DEs for multi-layered waveguides possessing circular or plane-parallel symmetry and can be extended to more general structures as well as to determination of complex waves.

Disclosure statement

No potential conflict of interest was reported by the author.

References

  • Milovanovic GV, Mitrinovic DS, Rassias ThM. Topics in polynomials: extremal problems, inequalities, zeros. Singapore: World Scientific; 1994.
  • Littlewood JE. Some problems in real and complex analysis, heath mathematical monographs. Lexington (MA); 1968.
  • Davis PJ. Interpolation and approximation. New York: Blaisdell; 1968.
  • Dumitrescu DA. Positive trigonometric polynomials and signal processing applications. Springer; 2007.
  • Chanane B. Computing the spectrum of non-selfadjoint Sturm Liouville problems with parameter-dependent boundary conditions. J Comp Appl Maths. 2007;206(1):229–237. doi: 10.1016/j.cam.2006.06.014
  • Littlewood JE. On the real roots of real trigonometrical polynomials (II). J Lond Math Soc. 1964;39:511–552. doi: 10.1112/jlms/s1-39.1.511
  • Littlewood JE. The real zeros and value distributions of real trigonometrical polynomials. J Lond Math Soc. 1966;41:336–342. doi: 10.1112/jlms/s1-41.1.336
  • Erdelyi T. Polynomials with littlewood-type coefficient constraints. In: Chui CK, Schumaker LL, Stoeckler J, editors. Approximation theory X. Nashville, TN: Vanderbilt University Press; 2001. p. 1–40.
  • Borwein P, Erdelyi T. Trigonometric polynomials with many real zeros and a littlewood-type problem. Proc Am Math Soc. 2000;129(3):725–730. doi: 10.1090/S0002-9939-00-06021-4
  • Lanczos C. Trigonometric interpolation of empirical and analytical functions. J Math Phys. 1938;17:123–199. doi: 10.1002/sapm1938171123
  • Oliveira-Pinto F. Experiments with a class of generalized trigonometric polynomials. Calcolo. 1984;21(3):253–268. doi: 10.1007/BF02576536
  • Polya G, Szegö G. Aufgaben und Lehrsätze aus der Analysis. 3rd ed. Berlin: Springer; 1964.
  • Borwein P, Erdelyi T, Ferguson R, et al. On the zeros of cosine polynomials: solution to a problem of littlewood. Ann Math. 2008;167:1109–1117. doi: 10.4007/annals.2008.167.1109
  • Runovsky VK. On real zeros of an even trigonometric polynomial. Mat Zametki. 1988;43(3):334–336.
  • Watson GN. Theory of bessel functions. 2nd ed. New York: Cambridge University Press and Macmillan Co.; 1948.
  • Jahnke E, Emde F. Tables of functions. 4th ed. New York: Dover Publications; 1945.
  • Abramowitz M, Stegun IA., editors. Handbook of mathematical functions with formulas, graphs, and mathematical tables. New York: Dover; 1972.
  • NIST Digital Library of Mathematical Functions. Release 1.0.20 of 2018-09-15. F. W. J. Olver, A. B. Olde Daalhuis, D. W. Lozier, B. I. Schneider, R. F. Boisvert, C. W. Clark, B. R. Miller, and B. V. Saunders, eds.
  • Ek JM, Thielman HP, Huebschman EC. On the zeros of cross-product bessel functions. J Math Mech. 1966;16(5):447–452.
  • Snyder AW, Love J. Optical waveguide theory. Springer; 1983.
  • Shestopalov Y, Kuzmina E, Samokhin A. On a mathematical theory of open metal-dielectric waveguides. FERMAT. 2014;5.
  • Shestopalov Y, Kuzmina E. Symmetric surface complex waves in Goubau line. Cogent Eng. 2018;5(1):1–16.
  • Shestopalov Y, Kuzmina E. Complex waves in multi-layered metal-dielectric waveguides. Proc. 2018 Int. Conf. Electromagnetics in Advanced Application (ICEAA), 20th Edition, Cartagena de Indias, Colombia: September 10–14, 2018. p. 122–125.
  • Weinstein LA. Open resonators and open waveguides. Golem Press; 1969.
  • Shestopalov Y. Complex waves in a dielectric waveguide. Wave Motion. 2018;57:16–19. doi: 10.1016/j.wavemoti.2018.07.005
  • Ismail MEH, Muldoon M. Monotonicity of the zeros of a cross-product of bessel functions. SIAM J Math Anal. 1978;9(4):759–767. doi: 10.1137/0509055
  • Shestopalov Y. Resonant states in waveguide transmission problems. PIER B. 2015;64(1):119–143. doi: 10.2528/PIERB15083001
  • Shestopalov Y. Singularities of the transmission coefficient and anomalous scattering by a dielectric slab. J Math Phys. 2018;59(3):033507. doi: 10.1063/1.5027195
  • Karchevskii EM, Beilina L, Spiridonov AO, et al. Reconstruction of dielectric constants of multi-layered optical fibers using propagation constants measurements, arXiv: 1512.06764v1 [math.NA], Dec. 2015.

Appendix

The Fejer kernel (A1) Fn(x)=1nk=0n1Dk(x)=1nsinnx2sinx22,(A1) the Dirichlet kernel (A2) Dn(x)=k=nneikx=1+2k=1ncos(kx)=sinxn+12sinx2(A2) and other trigonometric sums like (A3) EN(x)=k=0N1eikx=CN(x)+iSN(x)=ei(1/2)(N1)xsinNx2sinx2,(A3) a cosine TP with all P1(k)(x)=0, P2(k)(x)=1, and the parameter vector κ¯2,N1=(1,,N1) (A4) CN(x)=k=0N1cos(kx)=sinNx2sinx2cos(N1)x2,(A4) a sine TP with all P1(k)(x)=1, P2(k)(x)=0, and the parameter vector κ¯1,N1=(1,,N1) (A5) SN(x)=k=1N1sin(kx)=sinNx2sinx2sin(N1)x2,(A5) a TP with all P1(k)(x)=P2(k)(x)=1 (A6) TN(x)=k=0N1(sinkx+coskx)=q=12k=0N1Uq(kx)=SN(x)+CN(x)=2sin(N1)x2+π4sinNx2sinx2,(A6) ‘shifted’ TPs (EquationA4), (EquationA5) with all P1(k)(x)=sina and P2(k)(x)=cosa (A7) C~N(x;a)=k=0N1cos(a+kx)=q=12k=0N1(1)q1U3q(a)U3q(kx)=CN(x)cosaSN(x)sina=cosa+(N1)x2sinNx2sinx2,(A7) and all P1(k)(x)=cosa and P2(k)(x)=sina (A8) S~N(x;a)=k=0N1sin(a+kx)=q=12k=0N1Uq(a)U3q(kx)=CN(x)sina+SN(x)cosa=sina+(N1)x2sinNx2sinx2,(A8) are examples of TPs which can be calculated in the closed form involving only products of trigonometric functions. This enables one to obtain all zeros of TPs (EquationA1)–(EquationA6) xC,N(1,m)=(2m1)πN1,m=1,2,;xC,N(2,l)=2lπN,xS,N(2,l)=xC,N(2,l),l=1,2,;xS,N(1,n)=2nπN1,n=1,2,;xT,N(1,s)=2πN1s14,s=1,2,;xT,N(2,p)=xC,N(2,p),p=1,2, (limiting ourselves only to positive zeros, n=1,2,,).

Zeros (A9) xc(1,m)(k)=(2m1)π2k,k=1,N1,m=1,2,,(A9) (A10) xs(2,l)(k)=lπk,k=1,2,N1,l=1,2,,(A10) of all intermediate terms coskx and sinkx in TPs (EquationA1)–(EquationA6) alternate with xC,N(1,m), xC,N(2,l), xS,N(1,n), xT,N(1,s), and xT,N(2,p). In fact, excluding, e.g. merging zeros of (EquationA6) xT,N(1,s)=xT,N(2,p) when (A11) N(4s1)=l(4N1),N=4N,(A11) which is valid particularly for p=s=N, N=1,2,, we have that zeros of (EquationA6) are all different if condition (EquationA11) does not hold. Another valuable observation is that infinitely many zeros of SN, SN1, CN, CN1, and TN, TN1 (N=2,3,) alternate.

Also, the distances dT,N(1)=xT,N(1,s+1)xT,N(1,s)=(2π/(N1)) and dT,N(2)=xT,N(2,p+1)xT,N(1,p)=(2π/N) between neigboring zeros of CN(x), SN(x), and TN(x) are less than the distance dcs,k(12)=xs,k(2,m)xc,k(1,m)=(π/2k) between neigboring zeros of coskx and sinkx (k=1,N1) for N>4k+1.