1. Introduction
When light is used for steering micromachine components and generating multiple optical traps for microscopic particles, laser beam modes are a convenient tool, as they conserve their structure both upon free-space propagation and at the focus of a spherical lens. Another benefit is that mode symmetry can be adjusted by varying certain parameters and passing from symmetry in the Cartesian coordinates to circular symmetry via elliptical symmetry. Ince-Gaussian (IG) modes can be utilized for this purpose, as they are transformed into either Hermite-Gaussian (HG) beams or Laguerre-Gaussian (LG) beams when the ellipticity parameter is changed.
An initial solution to the Whittaker equation in the form of IG functions was obtained by F.M. Arscott in [
1,
2]. In W. Miller’s book [
3], this solution was obtained in separated variables in elliptical coordinates. The applications of IG beams in optics research were considered in [
4]. In [
5], expressions for the amplitudes of IG beams were obtained for
p = 0, 1, and 2, independent of the ellipticity parameter. In [
6], elegant IG beams were investigated. These works have created an impetus for the broad adoption of IG beams, along with other well-known laser modes (LG and HG), in optics. In [
7], IG beams were generated with a digital hologram. In [
8], vector (classically entangled) IG beams were generated as superpositions of a right-handed, circularly polarized, even IG beam and a left-handed, circularly polarized, odd IG beam. IG beams with quantum entanglement were generated in [
9]. Nonlinear transformation of IG beams by spontaneous parametric down-conversion was investigated in [
10,
11]. In [
12], IG beams were generated also parametrically, but without conversion. In [
13], elegant IG beams were studied in a parabolic medium. Propagation of IG beams in uniaxial crystals was studied in [
14]. IG beams are used for underwater data transmission [
15]. Propagation of IG beams in a turbulent atmosphere was considered in [
16].
As seen from the above brief review of works dealing with IG beams, they have been actively studied in optics. However, no analytical representation of these beams via LG or HG modes with an explicit dependence on the ellipticity parameter has yet been proposed. In this work, in an effort to fill in this gap, we derive a number of particular analytical formulae for even and odd IG modes with the indices p = 3, 4, 5, and 6, which are expressed via LG and HG modes and exhibit explicit dependence on the ellipticity parameter ε. This explicit dependence of the expansion coefficients on the ellipticity parameter makes it possible to control the intensity pattern of IG beams by continuously varying this parameter. In this work, for the first time in the context of optics, we consider a situation wherein the ellipticity parameter ε can be not only positive, but negative as well. We analyze how the IG beam changes with changes in the sign of this parameter. The simultaneous representation of IG modes via several LG modes and several HG modes, as proposed herein, suggests that when the ellipticity parameter ε approaches zero, any IG mode is converted into a specific LG mode (more exactly, to its real or imaginary part), and when the ellipticity parameter ε approaches infinity, the IG mode aligns with a specific HG mode.
2. Solution of the Paraxial Equation in Elliptic Coordinates
The paraxial Helmholtz equation is given by [
3]:
where
are the Cartesian coordinates;
=
is a 2D vector;
z is the coordinate along the optical axis;
k is the wavenumber of light; and
is the complex amplitude of a monochromatic light field. Below, we use the following dimensionless coordinates:
,
,
, with
being the waist radius of the Gaussian beam. Then, in dimensionless variables, Equation (1) reads as follows:
Its simplest solution with a finite energy is a conventional Gaussian beam:
It is known that separation of variables in Equation (2) into Cartesian and polar coordinates makes it possible to obtain two solution families—HG and LG beams, respectively [
17]:
with
.
Both families are examples of structurally stable light fields, i.e., at any propagation distance
z, the transverse intensity pattern of such fields aligns—up to scale—with that in the initial plane
:
Thus, it will suffice to study both families at , omitting the last argument for brevity: and .
We also introduce normalized variants of each solutions family:
When Equation (2) is solved, the variables are separated in the elliptic coordinates
[
3]; then, a family of IG beams, which are also structurally stable, is obtained:
Here,
and the parameter
is supposed to be arbitrary. Substitution of Equation (7) into Equation (2) yields equations for the functions
and
:
Both Equation (9) reduce to the Ince equation, with its canonical form being given by [
1,
2,
3]:
where
p is an integer nonnegative number;
is a positive ellipticity parameter, as mentioned in the Introduction; and
is some constant. In works [
1,
2,
3], it is supposed that
ε > 0. This assumption can also be seen from comparison of Equations (9) and (10), as
ε = 2
σ2 > 0. Below, in the current work, we show that
ε in the solution (10) can be of any sign.
The constant
λ in Equation (9) appears on separating the variables
and
in Equation (2). For Equation (10), the constant is a real number, introduced for the existence of solution
in the form of a trigonometric polynomial of degree
p (called the Ince polynomial). Once the form of the solution is restricted in this way, we find that
should be a root of a certain polynomial of degree (
p + 1), whose coefficients depend on
(i.e., root of a characteristic equation). Thus, in total, (
p + 1) roots appear as
, and they are all real-valued. In addition, if these roots are sorted in ascending order at
, their ascending order does not change with increasing values of
ε (for all positive
ε, the characteristic equation has no multiple roots). Thus, by choosing a root
and by constructing a trigonometric Ince polynomial
, solution (7) can be written via the solutions to Equation (10) as follows:
where
. It can be shown that the characteristic equation is factorized and splits into two equations. If
p is even, then there are
p/2 odd Ince polynomials (they are written via the sines of multifold angles and usually indexed as
) and
p/2 + 1 even Ince polynomials (they are written via the cosines of multifold angles and indexed as
). If
p is odd, there are equally (
p + 1)/2 even and odd Ince polynomials (indexed as
). Thus, even and odd IG modes (11) are usually written as
where
and
are even and odd Ince polynomials, respectively, and indices
p and
q are nonnegative integers (
).
Unfortunately, a tradition of writing the even and odd IG modes differently and studying them separately leads to unjustified complications in the designations. The index q is chosen to be of the same parity as p, i.e., . If p is odd, then this list stops at , but if p is even., it stops at for modes and at for modes . This separation is probably related to the initial form of the Ince polynomials , as expressed via or . If, instead of the sines and cosines, exponentials were used and if the even and odd IG modes were not separated, then, for any fixed p, the index q would range from 0 to p, and respective numbers would be sorted in ascending order. This change would make it possible to investigate the IG modes in a unified way, similarly to the Hermite-Gaussian modes; despite the Hermite polynomials being both even and odd, the family of the HG modes is not split into even and odd subfamilies. However, at this point, we will adhere to the commonly used designations.
As the IG modes are structurally stable, below, we write them at
, omitting the argument
z for brevity, but specifying explicitly the dependence on the parameter
:
. Such notation is convenient when studying limiting cases
and
, when the IG modes reduce to the LG and HG modes, respectively. In addition, similarly to Equation (6), we use the IG modes with and without normalization:
preferring the non-normalized variant,
, in cumbersome cases, as for the IG modes, it is generally much simpler than the normalized one.
3. Expansions of the IG Modes by the HG and LG Modes at Small Values of p
The simplest case is when
. Then,
and the even IG mode reduces to the conventional Gaussian beam:
The following three modes are also independent of the parameter
ε and were given in [
4,
5]:
3.1. Cases with the Characteristic Polynomial of Degree 2
Below, without derivation, we give explicit analytical expressions for those IG modes, for which the characteristic polynomial reduces to a quadratic polynomial and the respective roots can be expressed in square radicals.
It is worth noting that if the initial definition
is dropped, the parameter
(ellipticity) can be considered to be a negative number (as observed by Edward Ince [
18]). Therefore, in addition to the limiting cases of IG modes with
and
, we also consider a case wherein
.
In
Table 1, the following designations are used:
and
are the normalized Hermite-Gaussian and Laguerre-Gaussian modes (6), and
are the normalized Ince-Gaussian modes (13).
The coefficients
here are positive, monotonically increasing functions of
, as shown below:
We should note a similarity among the IG modes with equal indices p and parity flags . For instance, to obtain mode from mode , it is sufficient to swap the superposition coefficients and change the sign of one of them. It turns out that a similar property is valid in a more general case. Namely, if the whole set of IG modes except one is known for fixed values p and , this unknown mode can be obtained by manipulating the expansion coefficients of the modes from the set. When the characteristic polynomial is quadratic, there are only two modes and the “manipulation” is just a swapping of the coefficients with a sign change. For more complicated cases, manipulations are also more complicated. However, they are incomparably simpler than obtaining the dependence of the superposition coefficients on the ellipticity parameter.
3.2. Cases with the Characteristic Polynomial of Degree 3
For mode
with
, we obtain the following characteristic equation:
whose roots
can be found by using the Cardano formula. In ascending order, they read as follows:
with
Asymptotic expansions of all three roots at small and large values of
are given in
Table 2.
These formulae are very useful for investigating the limiting cases and .
For
, the even, non-normalized IG modes are given by
with
and
We do not present the normalized version in order to avoid cumbersome fractions, and we give the normalized IG modes only in the limiting cases (
Table 3).
For numerical computation of the IG modes, when the ellipticity parameter is neither very small nor very large, the Cardano formula with expansions (19) should be employed, with the normalizing multiplier added if necessary. If, however, the ellipticity parameter is close to the limiting values, then the asymptotic expressions of the characteristic roots are more convenient. For instance, if and , we obtain , , and, thus, .
For many values of , the roots of the Equation (17) can be easily found without using the Cardano formula; for example, this is the case for . It is enough to choose some integer as a root and find which it corresponds to (relative to , Equation (17) is quadratic). In particular, for , we obtain , , and .
In addition, it can be shown that a simple interrelation exists between the IG modes constructed for the parameters
ε and −
ε:
The first two formulae here are written for the even value of p, whereas the last is written for an odd value of p. This result demonstrates once again that the case of negative ε is no less important in investigating the IG modes than the case of positive ε. In particular, Equation (21) indicates that the IG modes with negative values of the parameter ε are also orthogonal to each other, similarly to the modes with positive values of ε.
The proof of the relationships (21) is based upon the following property of the characteristic polynomials:
Now, we consider the case wherein
. For the even IG modes, the characteristic equation is given by
We again apply the variant of the Cardano formula wherein the cubic equation has three real-valued roots, and these roots in the ascending order read as follows:
where
Asymptotic expansions of all three roots at small and large values of
are given in
Table 4.
For
, even IG modes (without normalization) are given by
with
and
For the limiting cases, the normalized IG modes are shown in
Table 5.
Equation (22) establishes relationships between the roots of the characteristic polynomials:
. Therefore, as per Equation (21), odd IG modes are obtained from the even IG modes by swapping the indices of the HG modes, replacing
with
, and adding a multiplier
:
with
and
Similarly to the above considered case for
, for many values of
ε, the roots of the characteristic polynomials can be found without using the Cardano formula. For instance, for the even IG modes, the following values can be chosen:
, whereas for the odd IG modes—the values
. In particular, for
, we obtain the roots
and
. Therefore, for example,
There is one more case with the cubic characteristic equation, the case
for the odd IG modes:
Correspondingly, the odd, non-normalized IG modes are given by
with
and
Sometimes cubic Equation (28) can be solved without using the Cardano formula; for instance, it can be solved for
. In particular, for
, we obtain the roots
. Therefore,
3.3. Even IG Modes at p = 6
Now we consider the case
p = 6 for the even IG modes. The characteristic polynomial is quartic:
Its roots
can be obtained by the Ferrari formula. We use the variant of this formula presented in [
19]. Let us introduce an auxiliary variable
with
Then, four roots of Equation (32) are expressed in ascending order via
, as follows:
Asymptotic expansions of all roots at small and large values of
are written in
Table 6.
Then, the non-normalized IG modes are expressed via the LG and HG modes in the following way:
with
and
For the limiting cases, the normalized IG modes are shown in
Table 7.
It is obvious that when index
p, which determines the degree of the characteristic polynomial, increases, finding all of its roots in a simple form and without numerical methods becomes increasingly impractical. Nevertheless, it is easy to observe that for
, one obtains the roots
, whereas the value
yields the roots
. Thus, for example,
3.4. Numerical Methods and IG Modes
It is known that for all characteristic polynomials , regardless of the index p and the parity flag , the coefficient for any term of the form is an integer. Therefore, the series expansions of the roots by powers of ε (both at and at ) have expansion coefficients that are rational numbers (see the above formulae as examples of such expansions). Consequently, asymptotic expansions of the IG modes by the LG and HG modes have coefficients that can be represented as series of the powers of ε (if , then by powers ε, ε2, …; if , then by the negative powers), and the coefficients in these series are also rational numbers.
For instance, let us consider mode
, writing it in the following form:
Here,
and
If in the expansion coefficients at and , we retain only terms up to ε3 and ε−4, respectively, then we obtain the following asymptotics.
In particular, as , then, as a limiting case, we obtain an identity that is already well known: .
Here, in the limiting case, we obtain a HG mode: .
4. Applying the Padé Approximants for Approximate Computation of IG Modes
In this section, we demonstrate using a concrete example how the Padé approximants can be employed to compute the IG modes. In the limiting case, IG mode reduces to a HG mode:
(third row in
Table 1). To obtain an approximate expansion of mode
from the HG modes that is suitable for the whole range
, we use the Padé approximants [
20]. The Padé approximant of some function
is given by the expression
where the coefficients
and
are chosen so that at small values
, it has several first terms of the Taylor expansion by the powers of
, exactly the same as those of the function
, whereas at large
, several first terms of the asymptotic expansion by the powers
are the same as those of the function
.
Collecting Equations (37) and (38) into the one, we get
where exact representations of the functions of
are given by Equation (36), whereas their approximations in the form of Equation (39) depend on the choice of the parameters
L and
M. Since
then the Padé approximants of the functions
and
should have the following form:
For each fixed value of the parameter
L, the coefficients of the rational functions in Equation (41) can be found by solving systems of equations. These equations are obtained if we expand functions (41) into series in powers of
at
and in powers of
at
, and then make equal these expansions to those already known from Equations (37) and (38). This approach is easy to implement in modern computer algebra systems. We here give only the simplest results when the obtained fractions are not very cumbersome yet:
Differences between the exact functions
and
and their Padé approximants are depicted in
Figure 1. In particular, the maximal difference
is 0.0187 for
and 0.0063 for
; while the maximal difference
is 0.0631 for
and 0.0224 for
. Here, the expansions
coincide with the expansions
up to the terms
and
, inclusively, whereas for
and for
the coincidence is up to the terms
and
. Increasing the parameter
L makes it possible to obtain increasingly accurate approximations of the IG mode, although this approach is accompanied by more cumbersome fractional rational expressions for the coefficients.
It is worth noting that using the Padé approximants does not require knowing the exact expressions for the functions and . Instead, it suffices to know the first few terms of the series expansions of these functions (in powers of at and in powers of at ). These expansions are obtained from the expansions of the roots of the characteristic equation; to expand the roots by the powers of or , well-known conventional methods are used.
In this section, the use of the Padé approximants is demonstrated for mode , which is expanded into the series of HG modes. It is easy to see that this approach is suitable for any other IG mode. Its value for numerical computations of modes becomes obvious for large values of the index p because, in this case, roots of the characteristic equation cannot be explicitly expressed in radicals.