Professional Documents
Culture Documents
Received April 28, 2006; accepted June 7, 2006; posted June 20, 2006 (Doc. ID 70311)
Two compact analytical descriptions of Fresnel diffraction patterns from polygonal apertures under uniform
illumination are detailed. In particular, a simple expression for the diffracted field from constituent edges is
derived. These results have fundamental importance as well as specific applications, and they promise new
physical insights into diffraction-related phenomena. The usefulness of the formulations is illuminated in the
context of a virtual source theory that accounts for two transverse dimensions. This application permits cal-
culation of fractal unstable-resonator modes of arbitrary order and unprecedented accuracy. © 2006 Optical
Society of America
OCIS codes: 050.1220, 050.1960, 140.3410, 260.1960, 140.0140.
再 冎
fundamental importance and have specific applications
冕冕
where standard methods, such as fast Fourier transform
kU0 k
(FFT) techniques, fail. For example, in unstable- U共x,y兲 = dd exp i 关共x − 兲2 + 共y − 兲2兴 ,
resonator-mode calculations, both (paraxial beam) ABCD 2iL 2L
⍀
matrix modeling and existing semi-analytical methods
can give accurate results only in limited parameter re- 共1兲
gimes in which the Fresnel number of the resonator is
low. Consequently, a complete and detailed study of the where k is the wavenumber of the incident plane wave
fractal laser modes arising from unstable cavities8–13 has and ⍀ denotes the aperture area (see Fig. 1). 共x , y兲 and
not previously been possible. A specific advantage of our 共 , 兲 are the observation- and aperture-plane coordinate
formalisms over, for example, FFT-based methods is their axes, respectively. The paraxial approximation, employed
ability to calculate the fine details of only a small portion in the derivation of the Fresnel diffraction integral (1),
of one, or many, complex diffraction patterns. It is this strictly requires the inequality L3 kb4 / 8 to be satisfied,
property that allows us to apply our results in the calcu- where b is the largest characteristic length associated
lation of fractal laser modes of unprecedented accuracy. with the aperture.
c共j兲 =
1
2
+ f共j兲sin 冉 冊
2
j2 − g共j兲cos 冉 冊
2
j2 , 共7a兲
s共j兲 =
1
2
− f共j兲cos 冉 冊
2
j2 − g共j兲sin 冉 冊
2
j2 , 共7b兲
where c共j兲 = 兰0j cos共2 / 2兲d and s共j兲 = 兰0j sin共2 / 2兲d
are the familiar Fresnel cosine and sine integrals, respec-
tively. The auxiliary Fresnel functions can be evaluated
Fig. 1. Schematic diagram illustrating the coordinate system efficiently using rational approximations of the required
used to describe diffraction patterns from an aperture ⍀ illumi-
accuracy.19 When j is nonnegative, one may choose
nated by a plane wave of amplitude U0.
1 + 0.926j
Within an edge-wave representation of diffraction pat- f共j兲 = , 共8a兲
terns, the field in the observation plane is regarded as the 2 + 1.792j + 3.104j2
superposition of the transmitted plane wave and an edge-
wave field E共x , y兲 arising from the boundary of the 1
aperture:4 g共j兲 = . 共8b兲
2 + 4.142j + 3.492j2 + 6.67j3
U共x,y兲 = U0 + E共x,y兲. 共2兲
When j ⬍ 0, the auxiliary functions are evaluated using
Here, is equal to unity if the point 共x , y兲, lies within the
bright geometrical shadow of the aperture, and is zero if it
is outside this region. We now express the edge-wave field
冋
f共− 兩j兩兲 = f共兩j兩兲 − 2 c共兩j兩兲sin 冉 冊
2
j2 − s共兩j兩兲cos 冉 冊册
2
j2 ,
冉 冊册
U共x,y兲 = dudv exp i 共u + v 兲 . 2 2
共3兲
冕
共1 + i兲2 2 1 u2
⍀
+ duS关v1共u兲,v2共u兲兴exp i u2 ,
To facilitate the analytical evaluation of the area integral 1+i u1
2
(3), we introduce the following S-function: 共10兲
S共1, 2兲 = − 1 +
1
共1 + i兲
冕
2
1
冉 冊
dv exp i v2 .
2
共4兲
where v1共u兲 and v2共u兲 are parametric expressions describ-
ing the shape of the aperture in the relative coordinate
system (see Fig. 2). The edge-wave field E共x , y兲 is then
The j共j = 1 , 2兲 limits denote the boundary of an infinite given by Eq. (2). In many cases it proves to be more effi-
slit aperture in one direction. Physically, S represents the cient to consider a specific aperture shape as broken down
edge-wave pattern from this slit when the slit is illumi- into an assembly of subapertures. Babinet’s principle can
nated by a unit plane wave, and this can be written as a then be employed to generate diffraction patterns. Here
superposition of two components (from the constituent we focus our attention on apertures that are regular poly-
edges): gons. For a polygon with N sides, the most obvious decom-
S共1, 2兲 = D共1兲 + D共− 2兲, 共5兲
where
D共j兲 = −
i
1+i
*共j兲exp i j2 ,
2
冉 冊 共6a兲
and
共j兲 ⬅ f共j兲 + ig共j兲. 共6b兲
The auxiliary Fresnel functions, f and g, and defined Fig. 2. Schematic diagram of an aperture of arbitrary shape in
through the relative coordinates 共u , v兲.
2770 J. Opt. Soc. Am. A / Vol. 23, No. 11 / November 2006 Huang et al.
B. Line-Integral Method
In this formulation, we define two vectors in complex-
number notation, p = x + iy and q = + i, and recast the
Fresnel diffraction integral (1) as
U共p兲 =
kU0
2iL
冕冕 ⍀
d2q exp i 冉 k
2L
兩p − q兩2 . 冊 共13兲
冋 册
can be calculated with Eq. (10). Babinet’s principle can
then be stated as
U共x,y兲 = U0 + E共x,y兲, 共11a兲
U共p兲 = U0 −
1
2
冖 冉
exp i
k
2L
兩q − p兩2 冊 q−p
兩q − p兩2
· ndl ,
where 共14兲
N
where n is the unit outward normal vector to the edge
E共x,y兲 ⬅ 兺 E 共x,y兲,
j=1
j 共11b兲 (within the aperture plane), dl is the line element along
the edge, and denotes the aperture boundary. The pa-
For an isosceles triangle [e.g., triangle 1 in Fig. 3(a)] of rameter is equal to unity if point p lies within the bright
half-angle = / N, whose vertex is at the origin and geometric projection of this aperture, and it is zero if p is
whose perpendicular length a is aligned along the -axis, outside this region.
the edge-wave contributions are given by For a polygonal aperture (see Fig. 4), the position of a
冋
vertex Aj is given by the position vector qj. It is also con-
E共x⬘,y⬘兲 = U0 关1 − + S共x⬘,x⬘ − A兲兴 venient to define vectors Lj = qj+1 − qj, tj = Lj / Lj, and nj (de-
noting the unit outward normal of side Lj). The position
冉 冊册
vector of a general point on the side of length Lj can then
+
1
冕 x⬘
1 + i x⬘−A
duS共w1,w2兲exp i u2
2
,
be represented as q = qj + tjl, where l is the distance from
the vertex Aj. Using these constructs, it is possible to re-
write Eq. (14) in a form more suitable for describing the
共12a兲 diffraction pattern from an N-sided regular polygon:
冋 册
where N
1
w1共u,x⬘,y⬘, 兲 = y⬘ − 共u − x⬘兲tan , 共12b兲 U共p兲 = U0 − 兺 I 共q ,t 兲
2 j=1
j j j , 共15兲
冕 冉 冊
Lj
wave pattern of each of the other component triangles k q j + t jl − p
Ij共qj,tj兲 = exp i 兩qj + tjl − p兩2 • njdl.
[e.g., triangles 2, 3 , . . ., and 8 in Fig. 3(a)] is through rota- 2L 兩qj + tjl − p兩2
0
tional transformation. A rotational transform of an angle
in the aperture plane is equivalent to considering the 共16a兲
system with new observation-plane coordinates x⬘
= x cos + y sin and y⬘ = y sin − x cos . After some manipulation, Eq. (16a) can be simplified to
However, we find that a more efficient approach to the
N-sided polygonal aperture problem is to replace triangle
ensembles with combinations of triangles and trapezoids
[see Fig. 3(b)]. In cases with an even number of sides, one
can also exploit the presence of a central rectangle. For
example, calculations for an octagonal aperture require
just two integrations—one for each trapezoid on either
side of this rectangle. In contrast, adding together eight
triangles, each with half-angle = / 8 and with a diffrac-
tive contribution resembling Eq. (12), could require eight
integrations (one per triangle). Triangle ensembles are
therefore not necessarily efficient computational con-
structs, though the physical symmetry of the problem is Fig. 4. Geometrical constructs used in the formulation of the
more explicitly preserved when they are used. line-integral method.
Huang et al. Vol. 23, No. 11 / November 2006 / J. Opt. Soc. Am. A 2771
共qj−p兲·tj
dl
l + 兩共qj − p兲 · nj兩
2 2 冉 冊
exp i
k
2L
l2 .
共16b兲
The edge-wave field, in the line-integral representation, is
then expressed as
N
U0
E共p兲 =
2 j=1 兺 I 共q ,t 兲.
j j j 共17兲
C. Fresnel Diffraction Patterns where, as earlier, a is the radius of the inscribed circle of
The S-function and line-integral approaches have been the polygon. Equation (21) contains two terms; the first is
used to calculate the diffraction patterns for N-sided the conventional Fresnel number for a circular aperture,
regular-polygonal apertures with N = 3 to N = 10. The nu- and the second has been proposed to account for addi-
merical integrations are calculated using an extended tional pattern detail arising from the geometrical nature
trapezoidal rule. When denoting x0 , x1 , . . . , xN as a se- of the aperture, where
2772 J. Opt. Soc. Am. A / Vol. 23, No. 11 / November 2006 Huang et al.
f共N兲 = 0.30618N2 − 0.10533N − 0.68095. 共22兲 accuracy of the line-integral approach, which can be the
most time-consuming, is essentially limited only by ma-
chine precision.
For example, Feff can vary with either a change in the Our original motivation for the reformulations of the
number of aperture sides (see Fig. 5) or a change in L for Fresnel diffraction problem was the complete failure of
a fixed aperture (see Fig. 6). FFT approaches in two-dimensional (2D) transverse
unstable-resonator mode calculations when moderate-to-
D. Comparison of Methods high cavity Fresnel numbers were considered.26 The reso-
The diffraction patterns calculated using the S-function lution of this problem, which involves the superposition of
and line-integral approaches have been checked indepen- sections from very many distinct diffraction patterns of
dently using the standard FFT method.23 The three meth- widely varying size, is detailed in the following section.
ods have their own distinct advantages and disadvan-
tages.
For a single, complete, and low Feff pattern that is 3. EXAMPLE OF APPLICATION
sampled with a uniform transverse grid, the FFT ap- The virtual source (VS) method26–30 is a semi-analytical
proach is most efficient. This is because the other meth- technique that unfolds an unstable cavity into a sequence
ods can require many numerical integrations. For ex- of equivalent virtual apertures. Accurate approximations
ample, a decahedron aperture many involve four of the cavity mode profiles can be obtained using a
integrations using the S-function method (assuming ap- weighted summation of edge waves diffracted from these
propriate aperture decomposition), while the line-integral apertures. In particular, computation of 2D unstable-
method could involve one integration per polygon side. cavity eigenmodes using this approach requires knowl-
However, FFT methods involve computation of complete edge of only sections of the patterns from these apertur-
diffraction patterns and further require the use of spatial ing elements. Moreover, the sampling density of the final
grids that sample a sufficient amount of the dark back- mode pattern dictates the resolution of each of these pat-
ground surrounding each pattern. When one requires ac- tern sections. FFT methods can be used for calculating
curate knowledge of the detailed structure of one or many edge-wave patterns (as an alternative to Fresnel integral
higher Feff patterns or sections of high resolution patterns approximations). This works well in one-dimensional VS
(e.g., dense and/or nonuniform sampling of a small area), codes. However, the FFT approach becomes impractical
or simply calculation of the optical field at a single trans- when fully 2D virtual apertures are involved (due to the
verse point, then the efficiency of the other two methods demands placed on computer memory and processing
tends to be far greater. time). A distinct advantage over existing (fully numerical)
Implementation of the FFT approach in such contexts Fox–Li techniques31 for calculating unstable-cavity laser
can also lead to reduced accuracy in the data acquired. modes is that virtual source approaches enable the simul-
Deployment of rational approximations in the S-function taneous calculation of a whole family of modes. In con-
method introduces some truncation errors, but high- trast, the Fox–Li algorithm needs to be manipulated non-
accuracy approximations24,25 can be used without entail- trivially when dealing with different higher-order modes,
ing great computational overhead. Finally, the numerical and in each application only a single pattern can be
obtained.30
We consider a geometrically unstable resonator with a
single polygonal aperturing element whose edge is repre-
sented by the equation h共 , 兲 = 0. After the cavity is un-
folded, the edge of the kth virtual aperture of the system
can be described by the equation
hk ⬅ h共M−k, M−k兲 = 0, 共23兲
␣Ns共␣ − 1兲
+
Ns
兺␣
k=1
−k
册
Ek共x,y兲 , 共25兲
integral method, and 共xc , yc兲 represents the coordinates of ertures is given in Ref. 30.
an arbitrary point on the boundary of the system aper- To demonstrate our 2DVS method, first we calculate
ture. For polygonal apertures, we choose this to be the the lowest order mode of an unstable resonator with the
point farthest from the center of the aperture. e0 is the same (low Fresnel number) parameters used by Berry
planar field amplitude at the point 共xc , yc兲 in the output [see Fig. 4(a) of Ref. 32]. Our result, shown in Fig. 7, is
plane. In Eq. (25), ␣ is the mode eigenvalue satisfying the completely consistent with Berry’s asymptotic prediction
polynomial equation27–29 that employs trigonometric approximations of the Fresnel
Ns integrals. Second, we apply our method to calculate the
␣Ns+1 + 兺 关E 共x ,y 兲 − E
k=0
k c c k+1共xc,yc兲兴␣
Ns−k
= 0, 共26兲 lowest-order modes of resonators with higher Fresnel
numbers and with triangular, pentagonal, hexagonal, and
decahedral apertures. These results are shown in the up-
where E0共xc , yc兲 = −1. The accuracy of the 2DVS calcula-
per row of Fig. 8. The lower row of panels in Fig. 8 shows
tion increases as Ns → ⬁. Southwell suggested that to en-
a magnification of the central region of each pattern. For
sure a reliable result, the number of virtual sources must
each configuration, the generic fractal signature of a high
be chosen so that Ns ⬎ log共250Neq兲 / log共M兲, where the
degree of fine structure within larger scale structure is
equivalent Fresnel number Neq of an unstable resonator
clearly evident. To the best of our knowledge, these fractal
is given by23
冉 冊冉 冊
eigenmodes are the most detailed and accurate patterns
a2 M2 − 1 that have so far been obtained from theoretical calcula-
Neq = 共27兲 tion. Opportunities are thus opened up for detailed com-
B 2M
parisons with more recent experiment results33 and for
and is the wavelength of the intracavity field. An alter- modeling novel applications of such fractal light.
native expression for the required number of virtual ap-
4. CONCLUSIONS
We have presented two complementary analytical de-
scriptions of Fresnel diffraction patterns from polygonal
apertures. Two different, but entirely equivalent, math-
ematical routes have been taken to formulate this prob-
lem, and supporting numerical work has verified that
they produce identical results.
The two analytical forms for the edge waves concerned
provide a potential physical framework for interpreting
diffraction-related phenomena in the Fresnel regime.
They also make possible an accurate generalization of
Southwell’s virtual source method to include two trans-
verse dimensions.26,32 This can allow calculation of 2D
eigenmode patterns with arbitrary order and to unprec-
edented accuracy. Further applications may, for example,
be possible in the area of understanding excess quantum
Fig. 7. (Color online) Lowest-order mode of an unstable resona- noise,34,35 where the transverse symmetry of a resonator
tor with triangular aperture (M = 4, Neq = 7.4604, R = 1 mm). is known to have a significant effect.14–16
Fig. 8. (Color online) Lowest-order modes of unstable resonators (M = 1.5, R = 1 mm) with (a) triangular 共Neq = 12.5兲, (b) pentagonal
共Neq = 32.725兲, (c) hexagonal 共Neq = 37.5兲, and (d) decahedral 共Neq = 45.225兲 apertures. Note that a redefinition of a to be the radius of the
circumcircle of each polygon yields a value of 50 for Neq in each of these four configurations.
2774 J. Opt. Soc. Am. A / Vol. 23, No. 11 / November 2006 Huang et al.