IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 59, NO.

10, OCTOBER 2011

3835

Huygens Principle for Complex Spheres

Thorkild B. Hansen, Member, IEEE, and Gerald Kaiser
AbstractHuygens relations that express wave elds of primary sources in terms of Huygens sources on a spherical surface remain valid when the sphere radius is a complex number in a certain bounded domain. Such Huygens relations are derived for frequency and time-domain acoustic and electromagnetic elds. Interior and exterior source congurations are combined to obtain dual-sphere Huygens relations, in which the primary sources are enclosed by one sphere and the observation points by another. The Huygens sources are complex point sources that exhibit directivity, which for certain parameter ranges makes it possible to achieve high accuracy with only a small fraction of the Huygens surface included in the integration. In general, for a dual-sphere conguration with xed physical dimensions, the fraction of the spheres required to achieve a given accuracy diminishes with frequency, up to a certain frequency limit. Beyond this upper frequency limit the size of the required regions on the two spheres remain roughly constant. The upper frequency limit is increased when the imaginary part of the complex sphere radius is increased. A similar result holds in the time domain with respect to diminishing pulse width. Index TermsElectromagnetic theory, electromagnetic propagation, electromagnetic transient analysis.

I. INTRODUCTION UYGENS principle states that the solution of the wave equation radiated by bounded primary sources can be represented outside the source region as a superposition of spherical Huygens wavelets radiated by secondary sources (the Huygens sources) on a surface (the Huygens surface) enclosing the primary sources [1], [2], ([3], p. 132). For acoustic elds the Helmholtz representation ([3], p. 377), ([4], Sections 2.2.7, 2.4.7) gives an exact formulation of Huygens principle in which the Huygens sources are determined from the acoustic pressure and its normal derivative on the Huygens surface. For electromagnethtic elds the Stratton-Chu formulas (sometime called the Larmor-Tedone formulas) ([5], Sec. 8.14), and the Kottler-Franz formulas [6], [7] give exact formulations of Huygens principle with the Huygens sources determined from the electromagnetic eld on the Huygens surface; see also ([4], Sections 2.1.9, 2.3.8). All these variations of Huygens principle employ non-directional Huygens sources that radiate isotropic spherical waves (acoustics) or dipole elds (electromagnetics). This lack of directivity makes it necessary to integrate over the entire Huygens
Manuscript received July 13, 2010; revised January 18, 2011; accepted March 18, 2011. Date of publication August 04, 2011; date of current version October 05, 2011. This work was supported by the Air Force Ofce of Scientic Research. T. B. Hansen is with Seknion Inc., Boston, MA 02115 USA (e-mail: thorkild. hansen@att.net). G. Gaiser is with the Center for Signals & Waves, Austin, TX 78712 USA (e-mail: kaiser@wavelets.com). Digital Object Identier 10.1109/TAP.2011.2163764

surface to get accurate eld values, even when all observation points are located in a small region to one side of the surface. In this paper we consider a spherical Huygens surface and prove that the Huygens representations for both acoustic and electromagnetic elds remain valid when the sphere radius is a complex number in a certain bounded domain of the complex plane. With a complex sphere radius the Huygens sources radiate complex source point beams (sometimes called Gaussian beams), which can have signicant directivity [8][10]. A major purpose of the paper is to investigate the properties of such complex-sphere Huygens relations in both frequency and time domains. In particular, we shall investigate how the directivity of the complex source point beams helps reduce the size of the integration areas on the sphere required to achieve a given accuracy of the computed elds. The rst exact spherical Huygens representation involving complex point source beams was obtained by Norris [11], who expressed the eld of a single real point source at the origin in terms of complex source point beams emanating from a sphere with radius , where and are real. Heyman [12] derived the analogous time-domain formula using the analyticsignal Fourier transform. Norris and Hansen subsequently generalized these formulas to arbitrary bounded acoustic sources both in the frequency domain [13] and time domain [14]. In this work, the weights of the complex source point beams are expressed in terms of the standard spherical expansion coefcients for the outgoing spherical multipoles. We also mention the work of Dezhong [15] who sketched out Huygens relations for a closed surface. The expansions of [13] have recently been augmented to include standing spherical multipoles and used to accelerate the computation of matrix-vector products in method of moments scattering calculations [16]. The eld expressions in [11][14], [16] can be written both as Huygens representations (the sphere contains all primary sources) and as more conventional multipole representations . Huygens relations for frequency-domain electromagnetic elds were obtained recently in the excellent work of Tap, Pathak, and Burkholder [17], [18]. These relations, which require the values of the elds on a surface in complex space, have been used to accelerate matrix-vector product computations in the method of moments [17]. In the present paper we derive explicit expressions that shed light on the high-frequency efciency of some of these Huygens relations. In [19] we obtained Huygens relations for frequency and time-domain acoustic elds in terms of the acoustic eld and its normal derivative on a sphere with complex radius. In the present paper we derive the analogous formulas for electromagnetic elds. In particular, we use the Gaussian regularization procedure ([19], Sec. 5) to get Huygens relations in the time domain. The paper [19] contains detailed physical interpreta-

0018-926X/$26.00 2011 IEEE

3836

IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 59, NO. 10, OCTOBER 2011

Fig. 1. The interior-source conguration. The sphere is centered at the origin. All sources are in the interior and all observation points in the exterior and .

Fig. 2. The exterior-source conguration. The sphere is centered at the origin. All sources are in the exterior and all observation points in the interior and .

tions, whereas the emphasis of the present paper is on numerical properties. The paper is organized as follows. Section II deals with the frequency domain and Section III deals with the time domain. The standard Huygens representations for a real spherical Huygens surface are presented in Section II.A. Both the interior and exterior source problems are considered with the Huygens sources expressed in terms of primary volume sources. In Section II.B we prove rigorously that these Huygens formulas remain valid when the sphere radius is a complex number in a certain bounded domain. Section II.C discusses properties of the complex point source beams. In Section II.D we combine the interior and exterior source Huygens formulas to obtain a dual-sphere Huygens relation, in which the primary sources are enclosed by one sphere and the observation points are enclosed by another sphere. Section II.E investigates the high-frequency efciency of the dual-sphere Huygens relation. In Section III.A we use the analytic-signal Fourier transform and a Gaussian time signal to derive the Huygens formulas for interior and exterior sources. The corresponding dual-sphere formulas are derived in Section III.B. In Section III.C we discuss the efciency of the dual-sphere time-domain relations. Section IV presents conclusions. II. HUYGENS REPRESENTATIONS IN THE FREQUENCY DOMAIN The standard rectangular coordinates are denoted by with unit vectors , and , so that a general point in space can be expressed as . The spherical coordinates are related to the rectangular coordinates through , and with the radial spherical unit vector given by . We will make extensive use of the unit vectors and , which are simply evaluated at and , respectively. In all frequency-domain analysis we assume time dependence with . For the acoustic eld the source-free regions of space is an unbounded linear lossless uid that can be fully described by the pressure . The acoustic sources are characterized by the source function , so that the pressure satises the inhomogeneous Helmholtz equation , where is the wave number, is the wavelength, and is the wave speed. The eld with source function is thus a generic scalar eld that satises the Helmholtz equation. The acoustic eld can therefore be expressed as (1)

where is the distance from the source point to the observation point , and is the the source region. As in [19], the subscript indicates that measures the distance from an emission point (source point). For the electric and magnetic elds and the sourcefree region is free space with permeability and permittivity , where the electric and magnetic elds can be expressed in terms of a volume current as in ([4], (2.335)(2.336))

(2)

(3)

where is the identity dyadic, and is the wave number. Again, is the region in which the sources reside. A. Huygens Representations for Sphere With Real Radius This section presents frequency-domain Huygens representations for a single spherical surface centered at the origin with real radius . We consider two congurations. The rst, called the interior-source conguration, has all sources in the interior of the sphere and all observation points in the exterior of the sphere, as shown in Fig. 1. We let denote the minimum radius of a sphere centered at the origin that contains all points of . The second, called the exterior-source conguration, has all sources in the exterior and all observation points in the interior, as shown in Fig. 2. We let denote maximum radius of a sphere centered at the origin that does not contain any points of . The acoustic eld at the observation point can be expressed through the Helmholtz representation in terms of its values and normal derivative on the sphere as

(4) where is the distance from a point on the sphere to the observation point. As in [19], the subscript indicates that is a reception point (observation point). To make

HANSEN AND KAISER: HUYGENS PRINCIPLE FOR COMPLEX SPHERES

3837

(4) valid for both interior and exterior sources, we have introduced the sign factor given by and for interior sources and exterior sources, respectively. The integration over the unit sphere can be expressed in spherical coordinates as . The acoustic pressure and its normal derivative can be determined from the volume source through (1). The electromagnetic analogs of (4) that express the electric and magnetic elds in terms of their tangential components on the sphere are the Kottler-Franz formulas [6], [7], ([4], (2.345)(2.346)) which state that1

Fig. 3. Region of analyticity in the complex -plane. For interior sources the when and . For exterior sources the gray gray region is region is when and .

(5) where is a unit vector pointing from a point on the sphere to the observation point. The formula for the magnetic eld is given by (5) with and interchanged, and replaced by . The tangential electric and magnetic elds on the sphere can be determined from the volume current through (2) and (3), respectively. For interior sources the Huygens representations hold when the observation point is outside the sphere and all source points are inside the sphere . This implies that must be in the range . Similarly, for exterior sources must be in the range . These conditions also imply that the right-hand sides of (4) and (5) are constant functions of on portions of the real axis, a property that will be essential for the generalization to complex in Section II.B. For interior sources the far eld can be computed by using to get the far-eld approximation for the propagator . Hence, except for the phase factor , the propagator is isotropic. The formulas of this section could be derived straightforwardly using Greens second identity and a free-space Greens function (for the electromagnetic eld one would use a dyadic Greens second identity and the dyadic free-space Greens function). This same approach could also be used for more general Huygens surfaces and more general Greens functions. B. Huygens Representations for Sphere With Complex Radius We now make the radius of the sphere complex in the Huygens representations (4) and (5). Specically, we let where is a positive constant. At this point we do not know if the formulas (4) and (5) remain valid under this substitution. With (1), (2), and (3) substituted into (4) and (5) the distances and are complex. Such distances have been investigated in detail in many papers [9], [10], [19], so we merely summarize their properties here. The square root is dened to have a non-negative real part and its
1We use the Kottler-Franz formulas because they only require tangential elds. The Stratton-Chu formulas also require normal components.

branch cut is placed along the negative real axis. In real 3D space the branch cut results in a disk of radius , centered at , and tangent to the real sphere ; see ([19], Fig. 3). The propagator satises the homogeneous Helmholtz equation everywhere in space except on this disk where its sources reside. These sources are highly singular along the rim of the disk [23][25]. Hence, when is complex the Huygens sources for the representations (4) and (5) occupy the spherical shell (spherical shells are shown in Figs. 7 and 8 below). The imaginary parts of the the complex distances satisfy (6) when the sources are in the interior, and (7) when the sources are in the exterior. The lower bound for in (6) is derived in the Appendix. The same derivation can be used to obtain the lower bound for in (7). In (6), only if and only if . Similarly in (7), only if and only if . The far-eld approximation for the propagator is , which shows that the Huygens sources are now directive with their maxima attained in the direction and minima attained in the direction . Below we shall compare the directivity of the Huygens sources to the directivity of a generic source. We next prove the validity of the Huygens expressions for complex using the approach of [19]. Since the unit sphere is nite, the right-hand sides of (4) and (5) are analytic functions of when the integrands are analytic for all ([4], p. 126). The integrands are analytic away from the branch cuts of the two square roots that dene and . For interior sources no branch cut is encountered if where (see Fig. 3) is the open region bounded by the line (ensuring that the branch cut of is avoided) and the circle (ensuring that the branch cut of is avoided). For exterior sources no branch cut is encountered if where is open the region bounded by the line and the circle . The regions and are shown in Fig. 3. The right-hand sides of (4) and (5) are thus analytic functions of in the regions and . Since these regions contain

3838

IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 59, NO. 10, OCTOBER 2011

the part of the real axis where the right-hand sides of (4) and (5) are constant (see the discussion near the end of Section II.A), the right-hand sides of (4) and (5) are constant throughout and , and the proof is complete. As expected, (4) and (5) are valid when the spherical shell given by contains neither source nor observation points. For more general Huygens surfaces and more general Greens functions, the proof must take into account other possible singularities, such as those that could be present in the Jacobian. The complex Huygens relations do not hold for all regular surfaces2 for which the real Huygens representation holds because the integral expression is not necessarily analytic in a sufciently large domain. The source disks of the present Huygens relations are tangent to the real sphere . One can also use the approach of the present paper to derive Huygens relations in which the source disks are tilted with respect to the sphere normal. Such formulas are identical to (4) and (5) except for the fact that is replaced by a more complicated expression. Tilted disks could potentially be useful for modeling higher-order probes in spherical scanning (see [21] and [22] for related work). C. Properties of the Single-Sphere Huygens Representations We begin by comparing the directivity of the the propagator , whose sources reside on a disk of radius , to the directivity of a generic source that also resides on that disk. Consider an -directed complex point dipole source located at , whose source disk has radius , is centered at the origin, and is perpendicular to . The normalized far eld is

Fig. 4. The normalized far elds of a complex point dipole source and a constant current source. The two sources reside on the same disk of radius .

(8)

where the last expression holds for small and is the identity dyadic. Next consider an -direct surface current that is constant on the disk and zero everywhere else (we refer to this as the constant current). The far eld of the constant current is given by

main beam of the constant current is narrower than the main beam of the complex point dipole source. When the opposite is true. This behavior is conrmed by Fig. 4 which shows the two far elds, normalized by the common factor , for , and . Note that normalized far eld of the complex point dipole attains its minimum at , whereas the normalized far eld of the constant current is symmetric with respect to . This example illustrates that when the primary source is large and roughly equal in size to the disk , the directivity of each complex point source may be much lower than the directivity of the primary source. In Section II.D we sharpen the representation by using two spheres; the rst sphere encloses the primary source and the second sphere encloses a collection of observation points. Hence, one set of complex point sources transmit the eld from the rst sphere and another set of complex point sources receive that eld at the second sphere. This transmission-reception process improves the directivity. We next explain a key property of the representations (4) and (5) that make them suitable as building blocks for fast computation schemes. Inserting (1), (2), and (3) into (4) and (5) reveals that the integrands of the Huygens representations contain the factor . The inequalities (6) and (7) give the following bounds (10)

(9) for interior sources, and where is the Bessel function of order 1 and the last expression holds for small .3 The factor in front of is for the complex point dipole and for the constant current. Hence, when the
2Kellogg

(11) for exterior sources. The lower bounds are negative in both (10) and (11), so the sphere can be divided into two regions: (i) a region where , and (ii) a region where . The two regions are shown in Fig. 5 for the case , and . Notice that the region with (the darker of the two) is in the direction to

[20] denes a regular surface; see also ([4], Ch. 2).

3Except for the factor , the electromagnetic far-eld patterns (8) and (9) are identical to the scalar far-elds patterns for a complex point source and a scalar constant source. This similarity between the scalar and electromagnetic cases carries over to all the formulas of this paper. As a consequence, the numerical properties of the scalar and electromagnetic formulas are identical.

HANSEN AND KAISER: HUYGENS PRINCIPLE FOR COMPLEX SPHERES

3839

Fig. 7. The dual-sphere conguration with the emission sphere (right) enand the reception sphere (left) enclosing the closing the source region . The radii of the inner and outer boundaries of the observation point region , respectively. spherical shells are and Fig. 5. on the light part of the sphere, and on the dark part of the sphere. Here , and .

From (1) and (4) it follows that the acoustic eld at the observation point in the region is given in terms of the eld on the reception sphere by

(12) where . The eld on the reception sphere can be expressed in terms of the eld on the emission sphere through

Fig. 6. Error of the computed eld as a function of cap angle and wavelength for the conguration in Fig. 5.

(13) where . Finally, the eld on the emission sphere can be expressed in terms of the volume source through (14) with . Similarly, the electric eld at the observation point can be expressed in terms of the tangential elds on the reception sphere as

the observation point. This can be proven always to be the case [19]. To gain further insight we compute the eld for the conguration shown in Fig. 5 for a range of frequencies. We include only those points of the sphere that lie inside a spherical cap with angle centered on the positive -axis ( corresponds to including the entire sphere, and corresponds to including only points on the sphere with ). Since , the integrand becomes more and more peaked near the point where attains its minimum, as increases. This minimum is near the positive -axis in the middle of the dark region in Fig. 5. Fig. 6 shows the error of the computed eld for different values of the wavelength and the cap angle . As the wave length decreases, a smaller and smaller cap angle is required to achieve a given accuracy. We shall use the exponential behavior of the integrand along with a second complex sphere to achieve a fast computation method in Section II.E. D. Dual-Sphere Huygens Representation We next consider the dual-sphere conguration in Fig. 7 where one sphere (the emission sphere) encloses the source region and another sphere (the reception sphere) encloses the region that contains the observation points. The emission sphere is centered at and the reception sphere is centered at . The minimum radius of a sphere centered at that encloses is denoted . For simplicity we let the two spheres have the same complex radius . Fig. 7 shows the spherical shells (bounded by spheres of radii and ) in which the Huygens sources reside.

(15) where . The electric eld on the reception sphere can be expressed in terms of the tangential elds on the emission sphere as

(16)

3840

IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 59, NO. 10, OCTOBER 2011

where . The magnetic elds are given by (15) and (16) with interchanged with , and replaced with . Finally, the electric and magnetic elds on the emission sphere can be expressed in terms of the volume sources as

Moreover, it can be shown that (21) where the lower and upper bounds are attained only when and , respectively. In other words, only when the two disk centers are closest to each other, and only when the two disk centers are farthest away from each other. The bounds (19), (20), and (21) show that and vary much less than . Moreover, for large and only attain values in short intervals near . Therefore, as a rst approximation, we neglect the variation of and and investigate which portions of the spheres can be omitted solely based on the behavior of . Later we discuss the effect of and . Dene the two angles and by

(17) and (18)

where

E. Efciency of the Dual-Sphere Huygens Representation We shall now investigate the possibility of using the dualsphere Huygens formula to obtain a method for matrix-vector multiplications in the method of moments. Assume that a surface scatterer (for example a hard acoustic body or a perfect electric conductor) with unknowns is divided into groups that each contain unknowns, as in a single-level fast multipole approach [26]. As usual, interactions between adjacent groups are computed in the standard way. Each group is surrounded by a complex sphere, so the dualsphere Huygens representation can be used to compute interacnontions between non-adjacent groups. Since there are adjacent group pairs, the matrix-vector multiplication cannot be performed in operations unless the computation of the Huygens sources on one sphere, due to the Huygens sources on operations. a non-adjacent sphere, can be performed in this is possible. We shall now examine under what conditions , so For a surface scatterer the group diameter is points to discretize a Huygens sphere. Hence, if it takes all points on the two spheres were included, it would take operations to compute the Huygens sources on one sphere due to the Huygens sources on a non-adjacent sphere. Therefore, to achieve the goal, we must be able to omit large parts of the two spheres without introducing appreciable errors. This may indeed be possible due to the exponential decay of the integrand, which is a direct consequence of the nonzero imaginary parts of the complex distances. From (6) and (7) it follows that in the dual-sphere conguand ration the imaginary parts of the complex distances satisfy

(22) when and so that when . To estimate the angle of the spherical caps required on each sphere to achieve a given accuracy, we expand for small and

(23) and are azimuthal angles with respect to the diwhere rection . Taking the imaginary part of (23) gives (24), shown at the bottom of the next page, which in turn proves

(25) Since we shall assume that (26) so that (25) becomes (27)

(19) and (20) is a positive constant. The condition (26) is met when and when . Let be a given positive threshold value for the integrand, so that only the points on the two spheres where where

HANSEN AND KAISER: HUYGENS PRINCIPLE FOR COMPLEX SPHERES

3841

are included. From (27) this condition can be expressed for small and as

(28) showing that all required points are included when the maximum values for and are set to4 (29) If is set to be independent of frequency, (29) implies that both and decay as as . The fraction of a unit sphere area that lies within a cap of angle is for small . Hence, the area to be included on each sphere to achieve a given accuracy diminishes as as . For the surface scatterer introduced above, the number of points required to discretize a Huygens sphere is and the diameter of a group is , so that . Therefore, the number of points required on each sphere is . With this result we can indeed compute the Huygens sources on one sphere, due to the Huygens sources on a non-adjacent sphere, in operations as desired. We numerically validate these properties of the dual-sphere Huygens relation by computing the eld of a constant source residing on a disk of radius at observation points on a disk of the same radius. The conguration is shown in Fig. 8. Both disks have and both spheres have and . The emission sphere and the constant source disk are both centered at , and the reception sphere and the observation disk are both centered at . The gure is exactly to scale and shows that both the disks and the spheres are close to each other. We include only those points on the two spheres that lie on the spherical caps with angle corresponding to a threshold value (see Fig. 8). The maximum eld error is dened to be the maximum relative error of the eld computed at all points on the observation disk. Numerical results are shown in Fig. 9 for , corresponding to the wavelength being in the range . The top plot conrms that the fraction of each sphere required to achieve a given accuracy (see bottom plot) diminishes with increasing frequency. The middle plot conrms that
4We could possibly obtain a more efcient representation by letting . a function of

Fig. 8. The eld due to the constant source disk enclosed by the emission sphere is computed at the observation disk enclosed by the reception sphere. and both spheres have and Both disks have radius . The emission sphere and the constant source disk are both centered , and the reception sphere and observation disk are both centered at . The threshold is and the gure is exactly to scale. at

Fig. 9. Numerical results for the conguration in Fig. 8. The top plot shows that the faction of the surface of each sphere included in the eld computation is roughly constant as decreases with . The middle plot shows that a function of . The bottom plot shows the maximum relative error.

be

approaches a constant, so that the cap area falls off as . Hence, for this source-receiver conguration operating in the stated frequency range, the dual-sphere Huygens relation does indeed perform as predicted; see also numerical results presented in [17], [18]. It appears that there are signicant differences between the dual-sphere Huygens relations of the present paper and the representations used in [17], [18]. For example, in [18] it is stated that in a method-of-moments application the beam coefcients

(24)

3842

IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 59, NO. 10, OCTOBER 2011

for each basis function are computed by matching elds over a test sphere enclosing the beams. With the current Huygens representation the beam coefcients would be computed directly from the electric and magnetic elds of the basis functions at discrete points on the complex Huygens surface. The efciency of the dual-sphere representation improves with increasing . However, as increases beyond a certain value, the two sets of disks touch and the representation breaks down. Therefore, we considered xed physical dimensions (including xed ) and increasing frequency. However, we neglected the fact that and vary on the spheres in the derivation of the estimate (29) for the cap angles. As we shall now see, the variations of these variables can be very important at high frequencies and in some cases result in an upper frequency limit, beyond which the cap angles no longer diminish. Fortunately, the upper frequency limit can always be increased by increasing (if permitted by geometrical constraints). It is easiest to illustrate this phenomenon by investigating the variation of the integrand on the emission sphere for the case where the eld is computed on the reception sphere at with only two point sources (located at and ) present in . The factor of importance is then (30) where and are the complex distances associated with and , respectively. The factor attains its maximum value at , where and can attain any value in accordance with (19). To illustrate, consider the emitting and reception spheres in Fig. 8 and let the two point sources be located at and . Here, corresponds to where . We nd that attains its maximum at and attains its maximum at . Consequently, the factor in (30) has two maxima spaced 3.4 apart. Numerical simulations reveal that the upper frequency limit for this conguration is (corresponding to ) and the minimum achievable cap angle on the emission sphere is about 5 . Similar considerations hold for the reception sphere. Despite the upper frequency limit, the dual-sphere Huygens relation can still be useful for method of moments computations provided it is used in a regime where the variations of and are unimportant, as demonstrated by Fig. 9. Also, if allowed by geometrical constraints, the upper frequency limit can always be increased by making larger. One signicant difference between the fast multipole method [26] and the dual-sphere Huygens approach is that the group interactions are computed one-by-one in the dual-sphere approach whereas they are compute all-at-once in the fast multipole method. Hence, the dual-sphere approach can be implemented as an on-the-y algorithm rather than by storing all the matrix entries.

III. HUYGENS REPRESENTATIONS IN THE TIME-DOMAIN In this section we apply the analytic-signal Fourier transform ([4], Section 5.3) to obtain time-domain analogs of the complex Huygens representations.5 The transformation to the time domain is not straightforward because the distances are complex. For example, grows exponentially as when and as when . As a consequency, the integrands of (4) and (5) grow exponentially with on part of the sphere even when the spectrum of the . Thus, one cannot immediately source is bounded as interchange the integration over frequency with the integration over the sphere because the integral over frequency at any given point on the sphere may diverge.6 To overcome this problem we regularize by assuming the source has Gaussian time dependence [19] (31) where the duration is related to the standard deviation by . Since the parameter can be chosen arbitrarily can be made arbitrarily close to a delta function, so small, no generality is lost by specializing to . In particular, every continuous signal can be expressed as the limit of a superposition of translated versions of (32) We can now perform the translation to the time domain by use of the analytic signal dened for a general time function as (33) where is a complex time and (34) is the standard Fourier transform of . The original time signal can be recovered from the analytic signal through . The domain of analyticity for depends on the properties of . For most functions (including the delta function), the do. For the main of analyticity includes the lower half plane Gaussian time function (31) the Fourier transform is , so that the domain of analyticity of plane. is the entire complex

5Real time-domain Huygens relations that involve only the acoustic eld (not its normal derivative) and only the tangential electric eld (not the magnetic eld) can be found in [27] and [28], respectively. 6The exponential growth of the integrand is canceled by the spatial integration over the sphere to ensure that the Huygens representations produce the correct . elds as

HANSEN AND KAISER: HUYGENS PRINCIPLE FOR COMPLEX SPHERES

3843

The explicit expression for

is

The time-domain analog of (4) can then be expressed as

(35) where erfc is the complementary error function and is the Faddeeva function [29]. Since the signal is Gaussian, we can introduce additional durations and and corresponding imaginary time shifts and to get the reproducing relation Similarly, the time-domain analog of (5) is

(41)

(36)

and . when To facilitate the factorization of the Huygens relations we introduce durations ( , and ) and imaginary time shifts ( , and ) associated with the three complex distances , and . The imaginary time shifts are introduced to improve the numerical properties of time-convolution integrals. We found in ([19], Section 8) that convolution integrals involving are easiest to compute if the complex time arguments of the two time functions are near the real axis. Accordingly, we assume that the frequency spectra of all source functions are proportional to at each point in space, so that (1), (2), and (3) give (42) (37) integrates with respect to the variable . The formula for the magnetic eld is given by interchanged with , and replaced with (42) with . The bounds (6) and (7) for the imaginary parts of the complex distances lead us to choose the imaginary time shifts as and . Notice that the formulas (41) and (42) are written in factor form: one factor is associated with the emission region, another with the reception region. The two factors are joined at the complex sphere. where

(38) and

(39) B. Dual-Sphere Huygens Representation A. Huygens Representations for Sphere With Complex Radius For the single-sphere formulation we have only two complex lengths and , and the associated durations and imaginary time shifts satisfy (40) For the dual-sphere formulation we have three complex lengths , and , and therefore three associated durations and complex time shifts, which satisfy (43) in the region can be obThe acoustic eld at a point tained in three steps as follows. First compute the eld on the

3844

IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 59, NO. 10, OCTOBER 2011

emission sphere from (37). Then compute a eld with duration and complex time on the reception sphere: shift

Finally compute the electric eld at the reception point from the eld on the reception sphere as

(47) The magnetic elds are given by (46) and (47) with interreplaced with . changed with , and In light of the bounds (19), (20), and (21) for the imaginary parts of the complex distances, it is near optimal to set and (see ([19], Section 8)). C. Efciency of the Dual-Sphere Huygens Representation We next investigate if the dual-sphere Huygens representation could be useful for fast computations in the time domain. Unlike the analogous frequency-domain representation studied in Section II.E, the time-domain representation cannot rely on all beams getting sharper with frequency. In fact, with Gaussian time dependence there is a large frequency component near zero frequency, at which the directivity of the beams vanish. However, as we shall see, the time-domain representation still exhibits compressive behavior as the duration diminishes. Assume that there is only one point source inside the emission sphere so that the exact eld at a point on the reception sphere is

(44)

Finally compute the desired eld with duration at the reception point

(45)

The analogous electromagnetics formulas are as follows. Begin by computing and (48) which can also be expressed in terms of the complex Huygens representation [19] from (38) and (39). Then compute the electric eld on the reception sphere

(49) where and . Let us rst study the amplitude of the integrand of (49) on in ([19], Fig.

(46)

the emission sphere. The plot of the function 8) shows that the maximum magnitude of

HANSEN AND KAISER: HUYGENS PRINCIPLE FOR COMPLEX SPHERES

3845

Fig. 10. The magnitude of } as a function of imaginary time for different values of the duration and real time.

Fig. 11. The cap angle on the emission sphere required to achieve a given accuracy of the eld on the reception sphere diminishes with .

occurs when and is at its maximum. From the bounds (19), (20), and (21) we have

(50) Since is independent of , we simply need to

determine how decays as goes from a positive value ) to a negative (corresponding to value (corresponding to ). Fig. 10 shows that falls off faster from its maximum value, as gets smaller. In other words, as gets smaller the integrand on the emission sphere gets more peaked, and we can reduce the included area of the sphere without increasing the error. We conrm this fact by computing the eld (49) using only points that lie on a spherical cap of angle on the emission , and . The sphere. Let singe emission point is given by , and the reception point is determined by the angle through . The error to be of a eld computed at is dened

cap angle required to achieve a given accuracy diminishes with . Hence, we achieve compression even though the signal has a signicant frequency component near zero, where the complex point sources lose their directivity. Compression is still achieved because the average positive frequency of the signal grows with diminishing , and thus the average directivity of the complex point sources used by the Huygens representation grows with diminishing . Similar to the upper frequency limit discussed in Section II.E, there is a lower duration limit for , beyond which the required cap angles do not diminish. Further studies are required before we can say whether the complex Huygens relation is useful for fast computations in the time domain. IV. CONCLUSION Through rigorous analytical continuation we derived frequency-domain Huygens relations for acoustic and electromagnetic elds for interior and exterior source congurations involving a complex sphere. These formulas were combined to obtain dual-sphere Huygens relations, in which the sources were enclosed by one sphere and the observation points by another. Time-domain analogs were obtained by applying the analytic-signal Fourier transform with a Gaussian time regularization procedure. Whenever the variations of the imaginary parts of the internal complex distances ( and in Fig. 7) of the dual-sphere Huygens relation are negligible, the area required on each sphere to achieve a given accuracy diminishes as as increases. When these conditions are satised, the dual-sphere Huygens relation leads to a technique for computing matrixvector products in the method of moments. However, for xed physical dimensions and varying frequency, there exists in general an upper frequency limit beyond which the variations of the imaginary parts of the internal complex distances are no longer negligible. In this regime the

(51)

where is the exact eld. With this error denition we suppress the large relative errors that often occur at early and late times when the eld values are very low. Fig. 11 shows as a function of cap angle for two observation points on the reception sphere and ve different durations. As expected from the analysis of the integrand, the

3846

IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 59, NO. 10, OCTOBER 2011

required areas on the two spheres do not diminish with , and the dual-sphere Huygens relation does not result in a technique. Fortunately, the upper frequency limit can be increased by increasing , provided this is allowed by geometrical constraints (i.e., the shells in Fig. 7 do not intersect). Similarly, for the dual-sphere Huygens relation in the time domain, there exists a parameter regime in which the areas required on the two spheres to achieve a given accuracy diminish with the pulse width. However, there is a lower limit for the pulse width, beyond which the areas no longer diminish. The lower limit for the pulse width can be decreased by increasing . It still remains to be seen if the dual-sphere Huygens relation works well enough in a sufciently large parameter regime to be a reliable alternative to other fast methods such as the fast multipole method [26].

Squaring (56) and rearranging terms gives

(57) which has the two solutions: and . The rst of these does not come into play since . Hence, the lower bound of is attained only at with (58) which can be inserted into (53) to get the desired expression (59) It may be of interest to know that when . REFERENCES
[1] C. Huygens, Trait de la Lumire. Leyden, 1690. [2] B. B. Baker and E. T. Copson, The Mathematical Theory of Huygens Principle, 3rd ed. New York: Chelsea, 1987. [3] M. Born and E. Wolf, Principles of Optics, 6th ed. Cambridge: Cambridge Univ. Press, 1993. [4] T. B. Hansen and A. D. Yaghjian, Plane-Wave Theory of Time-Domain Fields. New York: IEEE Press, 1999. [5] J. A. Stratton, Electromagnetic Theory. New York: McGraw-Hill, 1941. [6] F. Kottler, Elektromagnetische Theorie der Beugung an schwarzen Shirm, Ann. Phys., vol. 71, pp. 457508, 1923. [7] W. Franz, Zur Formulierung des Huygensschen Prinzips, Z. Naturforschg., vol. 3a, pp. 500506, 1948. [8] J. B. Keller and W. Streifer, Complex rays with an application to Gaussian beam, J. Opt. Soc. Amer., vol. 61, pp. 4043, 1971. [9] G. A. Deschamps, Gaussian beam as a bundle of complex rays, Electron. Lett., vol. 7, pp. 684685, 1971. [10] L. B. Felsen, Complex source point solution of the eld equations and their relation to the propagation and scattering of Gaussian beams, Symposia Mathematica, vol. 18, pp. 3956, 1976. [11] A. N. Norris, Complex point-source representation of real point sources and the Gaussian beam summation method, J. Opt. Soc. Am., vol. 3, no. 12, pp. 20052010, Dec. 1986. [12] E. Heyman, Complex source pulsed beam representation of transient radiation, Wave Motion, vol. 11, pp. 337349, 1989. [13] A. N. Norris and T. B. Hansen, Exact complex source representations of time-harmonic radiation, Wave Motion, vol. 25, pp. 127141, 1997. [14] T. B. Hansen and A. N. Norris, Exact complex source representations of transient radiation, Wave Motion, vol. 26, pp. 101115, 1997. [15] Y. Dezhong, Study of Huygens principle, Int. J. Infrared Milli. Waves, vol. 16, no. 4, pp. 831838, 1995. [16] T. B. Hansen, Efcient eld computation using Gaussian beams for both transmission and reception, Wave Motion, 2010, doi:10.1016/j. wavemoti.2010.10.009. [17] K. Tap, Complex source point beam expansions for some electromagnetic radiation and scattering problems, Ph.D. dissertation, The Ohio State Univ., Columbus, 2007. [18] K. Tap, P. H. Pathak, and R. J. Burkholder, Exact complex source point beam expansions for electromagnetic elds, IEEE Trans. on Antennas and Propagation vol. 59, no. 9, pp. 33793390, Sep. 2011. [19] T. B. Hansen and G. Kaiser, Generalized Huygens principle with pulsed-beam wavelets, J. Phys. A: Math. Theor., vol. 42, no. 47, Nov. 27, 2009, paper # 475403. [20] O. D. Kellogg, Foundations of Potential Theory. New York: Springer, 1967.

APPENDIX A LOWER BOUND FOR This Appendix derives the expression in (6) for the lower bound of in the case of interior sources. Without loss of generality we let and consider a single source point . Then,

(52) where since and . From ([30], p. 109) it follows that the solution , with non-negative real part, which satises the equation with real and is . Hence, (53) Since of for both and , the lower bound occurs at an interior point determined by (54) We have

(55) which can be used in (54) to get (56)

HANSEN AND KAISER: HUYGENS PRINCIPLE FOR COMPLEX SPHERES

3847

[21] T. B. Hansen, Complex point sources in probe-corrected cylindrical near-eld scanning, Wave Motion, vol. 43, pp. 700712, Jun. 2006. [22] T. B. Hansen, Complex-point dipole formulation of probe-corrected cylindrical and spherical near-eld scanning of electromagnetic elds, IEEE Trans. Antennas Propag., vol. 57, pp. 728741, Mar. 2009. [23] G. Kaiser, Physical wavelets and their sources: Real physics in complex spacetime, Top. Rev. J. Phys. A vol. 36, p. R29R338, 2003 [Online]. Available: http://arxiv.org/abs/math-ph/0303027 [24] G. Kaiser, Making electromagnetic wavelets, J. Phys. A: Math. Gen. vol. 37, p. 592947, 2004 [Online]. Available: http://arxiv.org/ abs/math-ph/0402006 [25] G. Kaiser, Making electromagnetic wavelets: II. Spheroidal shell antennas, J. Phys. A: Math. Gen. vol. 38, p. 495508, 2005 [Online]. Available: http://arxiv.org/abs/math-ph/0408055 [26] R. Coifman, V. Rokhlin, and S. Wandzura, The fast multipole method for the wave equation, IEEE Antennas Propag. Mag., vol. 35, pp. 712, Jun. 1993. [27] T. B. Hansen, Formulation of spherical near-eld scanning for timedomain acoustic elds, J. Acoust. Soc. Am., vol. 98, pp. 12041215, Aug. 1995. [28] T. B. Hansen, Formulation of spherical near-eld scanning for timedomain electromagnetic elds, IEEE Trans. Antennas Propag., vol. 45, pp. 620630, Apr. 1997. [29] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions, 9th Printing. New York: Dover, 1970. [30] H. E. Jensen, Matematisk Analyse. Lyngby: Technical Univ. Denmark Press, 1984, vol. 1.

Thorkild B. Hansen (M91) received the Ph.D. degree in electromagnetics from the Technical University of Denmark. From 1991 to 1997, he was with the Air Force Research Laboratory (formerly, Rome Laboratory) of Hanscom Air Force Base, MA, working on techniques for analyzing electromagnetic waves and antennas. He joined Schlumbergers underground radar project in 1997 and transferred with the project to Witten Technologies in 2000. He is currently developing techniques for RFID and near-eld communications at Seknion, Inc. He is coauthor of Plane-Wave Theory of Time-Domain Fields, a featured book of IEEE Press in 1999. Dr. Hansen received the R.W.P. King Prize in 1992 and the S.A. Schelkunoff Prize in 1995 for publications on electromagnetic wave propagation. The underground radar imaging technology he helped develop at Schlumberger and Witten Technologies won the 2002 NOVA Award for innovation in construction and Wall Street Journals 2004 Technology Innovation Award in Software.

Gerald Kaiser received Ph.D. degrees in physics and mathematics. He is Professor Emeritus at the University of Massachusetts, where he taught mathematical sciences, and heads the Center for Signals and Waves in Austin, TX. The ideas of acoustic and electromagnetic wavelets were introduced in his 1994 book A Friendly Guide to Wavelets. He is currently writing a book summarizing his work on the subject since 1994. Further information is available at www.wavelets.com.