OSA's Digital Library

Optics Express

Optics Express

  • Editor: Michael Duncan
  • Vol. 10, Iss. 16 — Aug. 12, 2002
  • pp: 728–739
« Show journal navigation

Stable aggregates of flexible elements give a stronger link between rays and waves

M.A. Alonso and G.W. Forbes  »View Author Affiliations


Optics Express, Vol. 10, Issue 16, pp. 728-739 (2002)
http://dx.doi.org/10.1364/OE.10.000728


View Full Text Article

Acrobat PDF (351 KB)





Browse Journals / Lookup Meetings

Browse by Journal and Year


   


Lookup Conference Papers

Close Browse Journals / Lookup Meetings

Article Tools

Share
Citations

Abstract

A recently proposed ray-based method for wave propagation is used to provide a meaningful criterion for the validity of rays in wave theory. This method assigns a Gaussian contribution to each ray in order to estimate the field. Such contributions are inherently flexible. By means of a simple example, it is shown that superior field estimates can result when the contributions are no longer forced to evolve like parabasal beamlets.

© 2002 Optical Society of America

1. Introduction

It is usually said that rays result from considering the limit of small wavelength [1

1. M. Born and E. Wolf, Principles of Optics. Electromagnetic Theory of Propagation, Interference and Diffraction of Light (Cambridge, Cambridge, 7th edition, 1999), pp. 116–129.

]. For nonzero wavelengths, however, the meaning of a ray is unclear, and the choice of a set of rays that is to be associated with a wave field can be ambiguous. Rays, nevertheless, are often used successfully to estimate the propagation of a variety of waves, and there are several non-equivalent procedures for doing so. For some of these methods, rays represent infinitely localized conduits of power, while for others they correspond to extended plane waves. Each of these approaches has its own domain of validity. Overviews of different ray-based wave propagation techniques are presented, for example, in two books by Kravtsov and Orlov [2

2. Yu. A. Kravtsov and Yu. I. Orlov, Geometrical Optics of Inhomogeneous Media (Springer, Berlin, 1990). [CrossRef]

,3

3. Yu. A. Kravtsov and Yu. I. Orlov, Caustics, Catastrophes and Wave Fields (Springer, Berlin, 2nd Edition, 1999). [CrossRef]

].

In a recent series of papers [4–7

4. G.W. Forbes and M.A. Alonso, “Using rays better. I. Theory for smoothly varying media,” J. Opt. Soc. Am. A 18, 1132–1145 (2001). [CrossRef]

], we introduced the core elements of a new ray-based method for the propagation of waves. The field estimate given by this method takes the form of a sum of Gaussian contributions where each member is associated with a ray but also holds information about the neighboring rays. It turns out that there is an inherent flexibility in the width of these contributions: the sum is structured so that the result is insensitive to this width. For this reason, we coined the term “stable aggregates of flexible elements” (SAFE) to describe the method.

2. Sums of Gaussian contributions

For simplicity, consider the case of propagation in two dimensions. The field U considered here satisfies the Helmholtz equation.

[2+k2n2(x,z)]U(x,z)=0,
(1)

where k = 2π/λ and n is the refractive index. We assume that this field propagates essentially (but not necessarily paraxially) towards larger z. (That is, it is assumed that rays do not turn around in z.) A one-parameter family of rays can be associated with this two-dimensional wave field. The rays can be written as x = X(ξ; z), where ξ smoothly identifies the rays. The initial conditions for these rays, namely X(ξ;0) and ∂X/∂z (ξ;0), must be determined from the specified initial field U(x,0). The rays are then traced through the medium by using standard ray equations:

Xz(ξ;z)=P(ξ;z)H(ξ;z),Pz(ξ;z)=12H(ξ;z)n2x[X(ξ;z),z],
(2)

where P and H correspond, respectively, to n(X, z) times the x and z components of a unit vector locally tangent to the ray. Hence

H(ξ;z)=n2[X(ξ;z),z]P2(ξ;z)>0.
(3)

Also of interest is the optical length of these rays, measured from a common phase front. This function, denoted by L, satisfies

Lz(ξ;z)=n2[X(ξ;z),z]H(ξ;z),Lξ(ξ;z)=P(ξ;z)Xξ(ξ;z).
(4)

Effectively, L is the eikonal evaluated at [X(ξ; z), z].

The field estimates considered here take the form of a sum over all rays:

Uγ(x,z)=w(ξ,γ;z)g(ξ,γ;x,z)dξ.
(5)

Here, w is a weight function and g is the Gaussian contribution given by

g(ξ,γ;x,z)=k2πexp(kγ2[xX(ξ;z)]2+ik{L(ξ;z)+[xX(ξ;z)]P(ξ;z)}).
(6)

Notice that, at any given z, the contribution in Eq. (6) is centered at x = X(ξ; z) and has a transverse phase that is linear and matches that of a plane wave propagating in the ray’s local direction. The width of the Gaussian is evidently determined by γ. We now summarize two ways in which the form in Eq. (5) can be forced to satisfy Eq. (1) asymptotically:

a) Gaussian beamlets

The conventional option exploits the linearity of the wave equation. That is, each independent contribution wg is forced to be an asymptotic solution of Eq. (1), i.e. to evolve like a parabasal Gaussian beamlet around the ray identified by ξ. As discussed in the Appendix of Ref. 4

4. G.W. Forbes and M.A. Alonso, “Using rays better. I. Theory for smoothly varying media,” J. Opt. Soc. Am. A 18, 1132–1145 (2001). [CrossRef]

, the weight and width of the contributions must then evolve according to

z[Hw(ξ;z)]=w2H52(P2n2xiγn2),
(7)
γz(ξ;z)=iH3[γ2n2+iγPn2x14(n2x)2+H222n2x2],
(8)

where w, γ, P, and H are functions of ξ and z, and n and its derivatives are evaluated at [X(ξ; z), z]. The beamlets that result from these expressions are asymptotic solutions of the wave equation, valid under what is known as the parabasal approximation, that is, the assumption that the beamlet remains well localized around its central ray. The field estimate for z > 0 results from the sum of the independent beamlets.

b) SAFE

The novel approach proposed in Ref. 4

4. G.W. Forbes and M.A. Alonso, “Using rays better. I. Theory for smoothly varying media,” J. Opt. Soc. Am. A 18, 1132–1145 (2001). [CrossRef]

is to force only the sum of all contributions in Eq. (5) to be an accurate solution Eq. (1). Integration by parts over ξ can then be used to mix different terms that could not be combined in the beamlet approach. The simple form of the weight function that results was shown in Ref. 4

4. G.W. Forbes and M.A. Alonso, “Using rays better. I. Theory for smoothly varying media,” J. Opt. Soc. Am. A 18, 1132–1145 (2001). [CrossRef]

to be

w(ξ,γ;z)=[γX(ξ;z)+iP(ξ;z)H(ξ;z)]12[a0(ξ)+O(k1)],
(9)

where the primes denote partial derivatives with respect to ξ. These derivatives mean that the weight of each Gaussian contribution depends not only on the corresponding ray but also on its closest neighbors. This is consistent with the intuitive idea that rays carry information in a shared fashion. It turns out that |a 0|2 (which is constant along a ray) is proportional to the power carried by the ray bundle.

Uγ(0)γ=O(k1)Uγ(0).
(10)

This insensitivity is a remarkable feature of the combination of Eqs. (5) and (9).

Notice that the calculation of SAFE’s basic field estimate requires no information other than that given by the ray family (including the conserved weight a 0). For the beamlet summation case, on the other hand, one must also trace the width and weight of each beam. The bounds for the variation of γ mentioned above [which follow from the explicit form of the right-hand side of Eq. (10) as presented in Ref. 4

4. G.W. Forbes and M.A. Alonso, “Using rays better. I. Theory for smoothly varying media,” J. Opt. Soc. Am. A 18, 1132–1145 (2001). [CrossRef]

] have a simple geometric interpretation in terms of the ray family alone. This interpretation and the meaning of insensitivity in general are illustrated with a simple example in the following section. We also show that the collective behavior of the contributions in SAFE can give more accurate wave models than a sum of Gaussian beamlets that each obeys Eqs. (7) and (8). To many, this is counterintuitive.

3. Example: Modes of a quadratic gradient-index waveguide

To illustrate the meaning of the insensitivity condition and to demonstrate that SAFE can outdo sums of Gaussian beamlets, we consider the gradient-index waveguide used in the example from Ref. 6

6. M.A. Alonso and G.W. Forbes, “Using rays better. III. Error estimates and illustrative applications in smooth media,” J. Opt. Soc. Am. A 18, 1357–1370 (2001). [CrossRef]

, where

n2(x)=n02(1ν2x2),
(11)

with n 0 and ν both positive and real. The modes of this waveguide are Hermite-Gaussians of the form

Um(x,z)=Hm(k¯vx)exp(k¯ν2x22)exp[ik¯vz(12m+1k¯)12],
(12)

where = kn 0/ν and Hm is the Hermite polynomial of order m. When m > M = (-1)/2, the argument of the square root within the second exponential becomes negative. In this case, the positive imaginary root must be chosen, so the corresponding modes decay exponentially in z, i.e. they are evanescent. For 0 ≤ mM, the modes in Eq. (12) are homogeneous, and they linearly accumulate a phase as they propagate in z.

3.1 Matching the initial condition and achieving insensitivity

In the example in Ref. 6

6. M.A. Alonso and G.W. Forbes, “Using rays better. III. Error estimates and illustrative applications in smooth media,” J. Opt. Soc. Am. A 18, 1357–1370 (2001). [CrossRef]

we considered a broad initial field composed of several modes. Here, however, our goal is to determine what type of evolution for the width of the Gaussian contributions leads to better results. It turns out that an individual homogeneous mode of this medium is ideal for this purpose. A set of initial conditions for a ray family that are associated with a Hermite-Gaussian initial field (without regard of the medium in which this field was to propagate) like U m(x,0) was proposed in Ref. 5

5. M.A. Alonso and G.W. Forbes, “Using rays better. II. Ray families to match prescribed wave fields,” J. Opt. Soc. Am. A 18, 1146–1159 (2001). [CrossRef]

. These conditions are given by

X(ξ;0)=X0cosξ,P(ξ;0)=P0sinξ,
(13)

where X0=ν1(2m+1)k¯ is the spatial half-width of U m(x,0), and P0=n0(2m+1)k¯ is the corresponding half-width of its Fourier transform. Here, ξ, varies over an interval of length 2π. It follows from Eqs. (2) and (11) that these rays propagate according to

X(ξ;z)=X0cos(ξn0H0νz),P(ξ;z)=P0sin(ξn0H0νz),
(14)

where H0=n01(2m+1)k¯ is the constant value that H(ξ; z) takes for this example. The corresponding ray family is overlaid onto the intensity profile of the mode for m = 8 in Fig. 1.

Fig. 1. Intensity profile (grayscale) and rays (red lines) for the waveguide mode in Eq. (12) with m = 8. This illustration is applicable even if its aspect ratio is changed to make the ray angles either more or less extreme.

Notice that the high intensity regions coincide roughly with the ray caustics, and that the central fringes can be thought of as interference patterns associated with the ray crossings. Further, to within a shift in z, all rays are identical and have a uniform weight, i.e. a 0(ξ) = c. The optical length of each ray follows from Eqs. (4):

L(ξ;z)=X0P02[sin(ξn0H0vz)cos(ξn0H0vz)(ξn0H0vz)]+H0z.
(15)

Unlike X and P, this function is not periodic in ξ. It can be seen from Eqs. (5), (6), and (9), however, that the values of X 0 and P 0 given after Eq. (13) guarantee that the field contribution wg is periodic when the square root in Eq. (9) is chosen to vary continuously in ξ. That is, by admitting only these discrete values for X 0 and P 0, this “quantization” condition ensures that the resulting estimate is independent of the chosen interval.

It was found in Ref. 5

5. M.A. Alonso and G.W. Forbes, “Using rays better. II. Ray families to match prescribed wave fields,” J. Opt. Soc. Am. A 18, 1146–1159 (2001). [CrossRef]

that, provided γ is chosen to be γ 0 = n 0 ν, the initial condition U m(x,0) is met exactly by the estimate that results upon substituting Eq. (6) and the leading order of Eq. (9) into Eq. (5) with z = 0. Here, the ray family is the one discussed above. Although the match is exact, we claim that this construction gives a meaningful connection between rays and waves only if the estimate is insensitive to changes in γ over a significant interval around γ 0. (We only consider here real values for γ, although this parameter can be complex in general as discussed in Refs. 4–6

4. G.W. Forbes and M.A. Alonso, “Using rays better. I. Theory for smoothly varying media,” J. Opt. Soc. Am. A 18, 1132–1145 (2001). [CrossRef]

, provided its real part is positive for the contributions to be localized.) On account of Eq. (10), insensitivity evidently holds for sufficiently large values of k, but such observations are not useful in practice. However, we have derived an explicit insensitivity condition. In the present case, insensitivity demands that kX0P0=2m+1 must be much larger than unity, as we now explain.

At a given z, the ray family can be represented by the parametric curve [X(ξ; z), P(ξ; z)], for varying ξ, in the (x,p) plane referred to as phase space. It is seen from Eqs. (14) that, for the present example, this curve is always an ellipse enclosing an area of πX 0 P 0 = π(2m + 1)/k. The geometric criterion for insensitivity found in Ref. 4

4. G.W. Forbes and M.A. Alonso, “Using rays better. I. Theory for smoothly varying media,” J. Opt. Soc. Am. A 18, 1132–1145 (2001). [CrossRef]

is best expressed in a (dimensionless) scaled space with coordinates (kγx,kγp). Specifically, γ must be chosen to maximize the tightest local radius of curvature of the curve (kγX,kγP). This radius of curvature must be much greater than unity for Uγ(0) to be insensitive to γ. The value γ 0 = n 0 ν turns the ellipse into a circle with a radius of 2m+1. Varying γ deforms the circle into an ellipse, although the enclosed area is fixed at π(2m + 1). Clearly, a larger circle can withstand stronger deformations before its peak curvature becomes problematic.

It is impressive to see that credible results emerge even when the peak curvature comes to well within an order of magnitude of unity. For example, Fig. 2 shows, for m=8, the deformation of the circle and the corresponding variation of the field estimates (and resulting errors) associated with variation of γ. Notice that the errors gather in x around the caustics (at |x| = X 0 in this case) for larger γ, and around the momentum caustics (at x = 0) for smaller γ. Although γ is varied by a factor of four for the animation — hence the ray localization varies by a factor of two —the heights of the intensity peaks change by no more than about five percent. Of course, this insensitivity is far more pronounced for higher order modes (where it is usual to expect ray methods to apply).

Fig. 2. (156 KB) Movie illustrating insensitivity in the construction of even a low-order waveguide mode, namely m = 8. (a) The deformation of the scaled phase space curve caused by changing γ by a factor of 4. (The value of γ is indicated by the blue dot.) (b) The corresponding SAFE estimate (black curve), as well as the associated error (red curve) magnified by a factor of 1000. These results are valid regardless of the number of propagating modes supported by the waveguide, i.e regardless of M.

3.2 Propagation

We now consider the estimation of the propagated modes by using the two approaches considered in Section 2.

a) Gaussian beamlets

Γτ(ξ;τ)=i{1Γ2(ξ;τ)+ρ2[cos(ξτ)+isin(ξτ)Γ(ξ;τ)]2},
(16)

where ρ = P 0/H 0 = [( m + 1/2)/(M - m)]1/2 is a measure of the non-paraxiality of the mode. Notice that ρ is small only when mM. That is, only the low-order modes in a waveguide that supports many propagating modes are in fact paraxial. The solution to this equation is given by

Γ(ξ;τ)=Ω(τ)cos(ξτ)isin(ξτ)cos(ξτ)iΩ(τ)sin(ξτ),Ω(τ)=1+iρ2τ.
(17)

The first of these expressions corresponds to a “breathing” (or phase space rotation) of the Gaussians, while the second one describes a linear stretching (or phase space shearing), as will be shown later. Similarly, the solution to Eq. (7) for this case is given by

w(ξ;H0τn0ν)=ciρ[cos(ξτ)+iΓ(ξ;τ)sin(ξτ)]
=ciρcos(ξτ)iΩ(τ)sin(ξτ).
(18)

By using Eqs. (17) and (18) in Eq. (5), we find the corresponding estimate for the modes at different z. In this simple example, the rays are known in closed form. In general, however, rays must be traced numerically, so we can only keep track of a finite, discrete set of them. To simulate this, we replace the integral in Eq. (5) by a sum at uniform intervals in ξ of 2π/N. Fig. 3a shows, for m = 8 and several values of N, the evolution in z of the sums of beamlets compared to the exact result. Notice that, with the chosen initial conditions, the initial field is well matched by the estimate even for a relatively small number of contributions. As z increases, however, the numerical errors caused by the discretization become apparent. Further, it turns out that these are not the only sources of error. Even for very large N, where the sum approaches the integral and the result retains a Hermite-Gaussian form (due to an intrinsic symmetry of this problem), the amplitude and phase go drastically astray. These variations are shown in Fig. 4. The amplitude (which should be constant) initially increases with z, and then decays roughly as z -0.53. The source of this fundamental error is explained by the diagram in Fig. 3b, which shows schematically the region of phase space occupied by each beamlet. As z increases, these Gaussians rotate and shear linearly, reaching spatial spreads well beyond the parabasal regime. The rate of deterioration of this estimate is then dictated by the rate of shearing of each beamlet, which as seen from the second expression in Eq. (17) is given by ρ 2 τ/z = γ 0 P02/H03. This means that it takes roughly (2πρ 2)-1 ray periods for the estimate to fail. That is, modes that are closer to the paraxial limit (ρ → 0) have longer-lived estimates. This makes sense because, in the paraxial limit, the medium in Eq. (11) becomes the optical analogue of the harmonic oscillator, the beamlets are analogous to coherent states, and the parabasal approximation made in the derivation of Eqs. (7) and (8) is exactly valid. Beyond the paraxial limit, the beamlets do not individually retain their Gaussian profile.

Fig. 3. (1.44 MB) Movie showing (a) the deterioration of the sums of Gaussian beamlets for the m = 8 waveguide mode, and (b) the evolution of the effective phase space areas occupied by a set of these beamlets. The animation covers propagation from z = 0 out to z = 7.5 (where the rays are periodic in z with period 2π.). The sums in (a) include 16 (red), 64 (green), and 256 (blue) beamlets, and are compared with the exact profile (black line). By contrast, SAFE propagates this mode exactly, and 16 rays then give sufficient accuracy for many purposes.
Fig. 4. Amplitude (green curve) and phase (red curve) of the ratio between an estimate made up of a continuous superposition of Gaussian beamlets and the corresponding exact result given in Eq. (12) with m = 8. Notice that the phase discrepancy goes to -π/4, and the slope of the amplitude curve for high z goes approximately to -0.53.

b) SAFE

It follows from Eqs. (14) that, for any z, the phase space curve is always the same ellipse, and rays just circulate around it. Therefore, the most convenient choice for γ is always γ 0 = n 0 ν. By changing the ray parameter to ξ′ = ξ - n 0 ν z/H 0, the estimate in Eq. (5) becomes independent of z, except for a phase factor due to the last term in Eq. (15). This phase matches exactly the linear phase in Eq. (12), and the estimate matches, at all z, the exact result. That is, for this example, SAFE can always match the exact result. Further, the result is equally insensitive at all z, so the grip between rays and waves holds indefinitely. Even when a discrete number of contributions is used, the estimate is quite accurate, as seen from the initial frame of Fig. 3 (where z = 0). This frame characterizes SAFE’s results at all z. Notice also that the non-paraxiality of the problem has no impact on the validity of SAFE.

3.3 Discussion

While this example might appear to be very special in that SAFE is able to match the exact result for all z, it is nevertheless representative of a more general behavior for (at least) smooth waveguides in which n is independent of z. Any guided ray in such medium has a periodic trajectory, which in phase space describes a closed loop (symmetric around the x axis) parametrized by z. Rays belonging to different loops have different periods, meaning that the waveguide is anharmonic. [The only harmonic optical waveguide is the one in Eq. (11) within the paraxial approximation.] The ray families associated to the modes of the waveguide correspond to all the rays that compose one of these phase space loops (i.e. identical rays shifted in z) with the “quantization” condition that the area enclosed by the loop is π(2m + 1)/k. A suitable value (or range of values) for γ can be chosen from the geometric criterion for insensitivity, and the desired mode can then be constructed approximately. Its profile is independent of z. If, on the other hand, we let each Gaussian propagate as a beam, the anharmonicity will cause it to stretch in phase space in a form similar to that in Fig. 3b, causing the estimate to deteriorate under propagation. Therefore, for smooth waveguides, it is always more effective to make γ independent of z instead of forcing it to evolve as the width of a Gaussian beamlet. This constancy of γ gives an analogue to Heller’s “frozen Gaussian approximation” used in semiclassical quantum mechanics [8]. There are other options, however, considered in Refs. 4–6

4. G.W. Forbes and M.A. Alonso, “Using rays better. I. Theory for smoothly varying media,” J. Opt. Soc. Am. A 18, 1132–1145 (2001). [CrossRef]

where γ is allowed to depend on ξ, and this can help to further enhance SAFE’s estimates for more complex waveguides.

Suppose now that the initial field is not a mode, and the corresponding phase space curve is not one of the unchanging phase space tracks mentioned above. As z increases, the anharmonicity causes the resulting curve to get wrapped up in a spiral-like form, where each turn becomes increasingly similar to one of the natural loops. Again, the corresponding propagated beamlets would be highly stretched in phase space, causing errors in the estimate. SAFE, on the other hand, can handle such cases well, as seen in the example in Ref. 6

6. M.A. Alonso and G.W. Forbes, “Using rays better. III. Error estimates and illustrative applications in smooth media,” J. Opt. Soc. Am. A 18, 1357–1370 (2001). [CrossRef]

. We can therefore expect SAFE to give superior results to those of beamlet summation for a wide variety of cases, where the parabasal approximation used in the derivation of Eqs. (7) and (8) becomes invalid, due to the beamlets’ expansion.

4. Propagation through an interface

Following the approach in Ref. 7

7. G.W. Forbes, “Using rays better. IV. Refraction and reflection,” J. Opt. Soc. Am. A 18, 2557–2564 (2001). [CrossRef]

, the transmitted and reflected fields U t and U r resulting from the incidence of U i = U m at the interface are estimated from the corresponding transmitted and reflected ray families by using the prescription in Eqs. (5), (6), and (9). By requiring that the boundary conditions, namely U i γ0 + U r γ0 = U t γ0 and z(Uγ0i+Uγ0r)=zUγ0t at z = z 1, are asymptotically satisfied, it follows that the weights used in these estimates are given by the following intuitive relations:

a0t(ξ)=t(ξ)a0i(ξ)=t(ξ)c,a0r(ξ)=r(ξ)a0i(ξ)=r(ξ)c.
(19)

Here, the transmission and reflection factors are similar to the TE Fresnel coefficients:

t(ξ)=2H0H1(ξ)H0+H1(ξ),r(ξ)=H0H1(ξ)H0+H1(ξ).
(20)

with

H1(ξ)=n12P2(z1;ξ)=H0q2(1+ρ2)ρ2sin2(ξn0H0νz1),
(21)

where q = n 1/n 0. Note that the first part of Eq. (21) follows from the fact that, due to Snell’s law, P is conserved across a flat interface perpendicular to z. In the last part of Eq. (21), we used Eq. (14) as well as n02 = P02 + H02. Notice also that, for q<ρ1+ρ2 (i.e. n 1 < P 0), the rays hitting the interface in the vicinity of the z axis are totally reflected. As a result, the corresponding transmitted rays have a (positive) imaginary H 1, i.e. the transmitted field has evanescent components. Estimates of the resulting wavefield when the incident mode has m = 8 and ρ = 1 are shown in Fig. 5 as an animation where q = n 1/n 0 is changed in time.

Fig. 5. (1.76 MB) Movie showing the intensity of the field excited when the mode in the waveguide considered in Sec. 3 is incident upon an interface with a homogeneous medium. The axes are scaled to units of X 0. The red rays are incident from the left, the blue rays are reflected by the interface, and the green rays are transmitted. Additional transmitted rays (in orange) have been added in the lower right corner to clarify the ray’s caustic structure. The initial frame has q = 0.71, and the final frame has q = 1.0, which means that the index is matched initially at the edge of the mode, but ends up matched at the center. The interference between the incident and reflected fields reveals interesting features of their relative phase and strength (in keeping with the properties of the familiar Fresnel reflection coefficient).

As mentioned above, one of the great advantages of SAFE is the possibility of estimating its own level of error. This follows from explicitly considering the top two orders in Eq. (9):

w(ξ,γ;z)=[γX(ξ;z)+iP(ξ;z)H(ξ;z)]12[a0(ξ)+a1(ξ,γ;z)ik+O(k2)].
(22)

Notice that a 1 can depend on z and γ as well as on ξ. For the incident field we let a 1(ξ, γ 0;z) = 0 since, with γ = γ 0 and a 0 = c, SAFE gives an exact reconstruction of U m (so no correction is needed). As with a 0 earlier, the values of a 1 at z 1 for the transmitted and reflected field estimates are found from the boundary conditions (as in Ref. 7

7. G.W. Forbes, “Using rays better. IV. Refraction and reflection,” J. Opt. Soc. Am. A 18, 2557–2564 (2001). [CrossRef]

) to be given by

a1t(ξ,γ0;z1)=ha1r(ξ,γ0;z1)=cn0ν4H02t2h52(ihexp[2i(ξn0H0νz1)]
+{th32+i(r2)sin(ξn0H0νz1)exp[i(ξn0H0νz1)]}h)
(23)

where h(ξ) = H 1(ξ)/H 0. [A simple prescription for finding the functions in Eq. (23) at other values of γ is given in Ref. 6

6. M.A. Alonso and G.W. Forbes, “Using rays better. III. Error estimates and illustrative applications in smooth media,” J. Opt. Soc. Am. A 18, 1357–1370 (2001). [CrossRef]

. With this, an optimal value of γ can be found by minimizing the error estimates discussed below. We do not bother with this here, however.]

The results in Eq. (23) can be used as corrections to a 0 in order to improve the accuracy of the estimates, roughly doubling the number of significant digits. Alternatively, they can provide a valuable measure of error of the estimates. It was shown in Ref. 6

6. M.A. Alonso and G.W. Forbes, “Using rays better. III. Error estimates and illustrative applications in smooth media,” J. Opt. Soc. Am. A 18, 1357–1370 (2001). [CrossRef]

that an rms relative error of SAFE’s basic estimate is asymptotically given by

εγ(z)=[U(x,z)Uγ(0)(x,z)2dxU(x,z)2dx]12=1k[H12(ξ)a1(ξ,γ;z)2dξH12(ξ)a0(ξ)2dξ]12+O(k2).
(24)

By using Eqs. (19), (20), and (23) in Eq. (24), the RMS errors εγt and εγ r for the transmitted and reflected field estimates are given by

Tεγt(z1)=Rεγr(z1)=ρ242π(2m+1)[t2h3(ihexp[i(ξn0H0νz1)]
+{th32exp[i(ξn0H0νz1)]+i(r2)sin(ξn0H0νz1)}h)|2dξ]12,
(25)

where T and R are given by

T=12πh1t2dξ,R=12πr2dξ.
(26)

For the estimates shown in Fig. 5, the expected errors at the interface are shown in Fig. 6. The rms error for the transmitted field is evidently only a few percent for the majority of cases shown in Fig. 5. Notice that that, because the field for z<0 is generally dominated by the incident field, the relative error in the total field is typically much less than the value indicated by the orange curve in Fig. 6. It is also interesting that the reflected field is no longer propagated without error by SAFE. This follows from the fact that a0r (ξ) is not constant.

Fig. 6. Plots of the rms relative errors at the interface itself for the transmitted (green) and reflected (orange) field estimates associated with Fig. 5. Notice that the error climbs rapidly as q approaches a value that represents the onset of internal reflection in the waveguide. This is due to the presence, in this case, of rays normal to the z axis, which represent a problem for SAFE. Since the problematic rays travel in the x direction, the error peaks disappear as we move away from the interface.

5. Concluding remarks

Acknowledgments

MAA acknowledges the funding by the Dirección General de Asuntos del Personal Académico IN112300 Optica Matemática project of the Universidad Nacional Autónoma de México. GWF thanks both the Australian Research Council and Macquarie University (Sydney, Australia).

References and links

1.

M. Born and E. Wolf, Principles of Optics. Electromagnetic Theory of Propagation, Interference and Diffraction of Light (Cambridge, Cambridge, 7th edition, 1999), pp. 116–129.

2.

Yu. A. Kravtsov and Yu. I. Orlov, Geometrical Optics of Inhomogeneous Media (Springer, Berlin, 1990). [CrossRef]

3.

Yu. A. Kravtsov and Yu. I. Orlov, Caustics, Catastrophes and Wave Fields (Springer, Berlin, 2nd Edition, 1999). [CrossRef]

4.

G.W. Forbes and M.A. Alonso, “Using rays better. I. Theory for smoothly varying media,” J. Opt. Soc. Am. A 18, 1132–1145 (2001). [CrossRef]

5.

M.A. Alonso and G.W. Forbes, “Using rays better. II. Ray families to match prescribed wave fields,” J. Opt. Soc. Am. A 18, 1146–1159 (2001). [CrossRef]

6.

M.A. Alonso and G.W. Forbes, “Using rays better. III. Error estimates and illustrative applications in smooth media,” J. Opt. Soc. Am. A 18, 1357–1370 (2001). [CrossRef]

7.

G.W. Forbes, “Using rays better. IV. Refraction and reflection,” J. Opt. Soc. Am. A 18, 2557–2564 (2001). [CrossRef]

8.

E.J. Heller, “Frozen Gaussians: A very simple semiclassical approximation,” J. Chem. Phys. 75, 2923–2931 (1981). [CrossRef]

9.

M.M. Popov, “A new method of computation of wave fields using Gaussian beams,” Wave Motion 4, 85–97 (1982). [CrossRef]

10.

V.M. Babich and M.M. Popov, “Gaussian summation method (review),” Izvestiya Vysshikh Zavedenii, Radiofizika 32, 1447–1466 (1989).

11.

M.J. Bastiaans, “The expansion of an optical signal into a discrete set of Gaussian beams,” Optik 57, 95–102 (1980).

12.

A.N. Norris, “Complex point-source representation of real point sources and the Gaussian beam summation method,” J. Opt. Soc. Am. A 12, 2005–2010 (1986). [CrossRef]

13.

P.D. Einziger, S. Raz, and M. Shapira, “Gabor representation and aperture theory,” J. Opt. Soc. Am. A 3, 508–522 (1986). [CrossRef]

14.

P.D. Einziger and S. Raz, “Beam-series representation and the parabolic approximation: the frequency domain,” J. Opt. Soc. Am. A 5, 1883–1892 (1988). [CrossRef]

15.

B.Z. Steinberg, E. Heyman, and L.B. Felsen, “Phase-space beam summation for time-harmonic radiation from large apertures,” J. Opt. Soc. Am. A 8, 41–59 (1991). [CrossRef]

16.

J.M. Arnold, “Phase-space localization and discrete representation of wave fields,” J. Opt. Soc. Am. A 12, 111–123 (1995). [CrossRef]

OCIS Codes
(080.2710) Geometric optics : Inhomogeneous optical media
(080.2720) Geometric optics : Mathematical methods (general)
(350.7420) Other areas of optics : Waves

ToC Category:
Focus Issue: Rays in wave theory

History
Original Manuscript: June 19, 2002
Revised Manuscript: July 30, 2002
Published: August 12, 2002

Citation
Miguel Alonso and G. Forbes, "Stable aggregates of flexible elements give a stronger link between rays and waves," Opt. Express 10, 728-739 (2002)
http://www.opticsinfobase.org/oe/abstract.cfm?URI=oe-10-16-728


Sort:  Journal  |  Reset  

References

  1. M. Born and E. Wolf, Principles of Optics. Electromagnetic Theory of Propagation, Interference and Diffraction of Light (Cambridge, Cambridge, 7th edition, 1999), pp. 116-129.
  2. Yu. A. Kravtsov andYu. I. Orlov, Geometrical Optics of Inhomogeneous Media (Springer, Berlin, 1990). [CrossRef]
  3. Yu. A. Kravtsov andYu. I. Orlov, Caustics, Catastrophes and Wave Fields (Springer, Berlin, 2nd Edition, 1999). [CrossRef]
  4. G.W. Forbes and M.A. Alonso, �??Using rays better. I. Theory for smoothly varying media,�?? J. Opt. Soc. Am. A 18, 1132-1145 (2001). [CrossRef]
  5. M.A. Alonso and G.W. Forbes, �??Using rays better. II. Ray families to match prescribed wave fields,�?? J. Opt. Soc. Am. A 18, 1146-1159 (2001). [CrossRef]
  6. M.A. Alonso and G.W. Forbes, �??Using rays better. III. Error estimates and illustrative applications in smooth media,�?? J. Opt. Soc. Am. A 18, 1357-1370 (2001). [CrossRef]
  7. G.W. Forbes, �??Using rays better. IV. Refraction and reflection,�?? J. Opt. Soc. Am. A 18, 2557-2564 (2001). [CrossRef]
  8. E.J. Heller, �??Frozen Gaussians: A very simple semiclassical approximation,�?? J. Chem. Phys. 75, 2923-2931 (1981). [CrossRef]
  9. M.M. Popov, �??A new method of computation of wave fields using Gaussian beams,�?? Wave Motion 4, 85-97 (1982). [CrossRef]
  10. V.M. Babich and M.M. Popov, �??Gaussian summation method (review),�?? Izvestiya Vysshikh Zavedenii, Radiofizika 32, 1447-1466 (1989).
  11. M.J. Bastiaans, �??The expansion of an optical signal into a discrete set of Gaussian beams,�?? Optik 57, 95-102 (1980).
  12. A.N. Norris, �??Complex point-source representation of real point sources and the Gaussian beam summation method,�?? J. Opt. Soc. Am. A 12, 2005-2010 (1986). [CrossRef]
  13. P.D. Einziger, S. Raz, and M. Shapira, �??Gabor representation and aperture theory,�?? J. Opt. Soc. Am. A 3, 508-522 (1986). [CrossRef]
  14. P.D. Einziger and S. Raz, �??Beam-series representation and the parabolic approximation: the frequency domain,�?? J. Opt. Soc. Am. A 5, 1883-1892 (1988). [CrossRef]
  15. B.Z. Steinberg, E. Heyman, and L.B. Felsen, �??Phase-space beam summation for time-harmonic radiation from large apertures,�?? J. Opt. Soc. Am. A 8, 41-59 (1991). [CrossRef]
  16. J.M. Arnold, �??Phase-space localization and discrete representation of wave fields,�?? J. Opt. Soc. Am. A 12, 111-123 (1995). [CrossRef]

Cited By

Alert me when this paper is cited

OSA is able to provide readers links to articles that cite this paper by participating in CrossRef's Cited-By Linking service. CrossRef includes content from more than 3000 publishers and societies. In addition to listing OSA journal articles that cite this paper, citing articles from other participating publishers will also be listed.

Supplementary Material


» Media 1: MOV (155 KB)     
» Media 2: MOV (1435 KB)     
» Media 3: MOV (1761 KB)     

« Previous Article  |  Next Article »

OSA is a member of CrossRef.

CrossCheck Deposited