This action might not be possible to undo. Are you sure you want to continue?
Wave propagation in media having negative permittivity and permeability
Richard W. Ziolkowski
Department of Electrical and Computer Engineering, University of Arizona, 1230 East Speedway Boulevard, Tucson, Arizona 85721-0104
Department of Electrical Engineering–Physical Electronics, Tel-Aviv University, Tel-Aviv 69978, Israel Received 3 May 2001; published 30 October 2001 Wave propagation in a double negative DNG medium, i.e., a medium having negative permittivity and negative permeability, is studied both analytically and numerically. The choices of the square root that leads to the index of refraction and the wave impedance in a DNG medium are determined by imposing analyticity in the complex frequency domain, and the corresponding wave properties associated with each choice are presented. These monochromatic concepts are then tested critically via a one-dimensional ﬁnite difference time domain FDTD simulation of the propagation of a causal, pulsed plane wave in a matched, lossy Drude model DNG medium. The causal responses of different spectral regimes of the medium with positive or negative refractive indices are studied by varying the carrier frequency of narrowband pulse excitations. The smooth transition of the phenomena associated with a DNG medium from its early-time nondispersive behavior to its late-time monochromatic response is explored with wideband pulse excitations. These FDTD results show conclusively that the square root choice leading to a negative index of refraction and positive wave impedance is the correct one, and that this choice is consistent with the overall causality of the response. An analytical, exact frequency domain solution to the scattering of a wave from a DNG slab is also given and is used to characterize several physical effects. This solution is independent of the choice of the square roots for the index of refraction and the wave impedance, and thus avoids any controversy that may arise in connection with the signs of these constituents. The DNG slab solution is used to critically examine the perfect lens concept suggested recently by Pendry. It is shown that the perfect lens effect exists only under the special case of a DNG medium with ( ) ( ) 1 that is both lossless and nondispersive. Otherwise, the closed form solutions for the ﬁeld structure reveal that the DNG slab converts an incident spherical wave into a localized beam ﬁeld whose parameters depend on the values of and . This beam ﬁeld is characterized with a paraxial approximation of the exact DNG slab solution. These monochromatic concepts are again explored numerically via a causal two-dimensional FDTD simulation of the scattering of a pulsed cylindrical wave by a matched, lossy Drude model DNG slab. These FDTD results demonstrate conclusively that the monochromatic electromagnetic power ﬂow through the DNG slab is channeled into beams rather then being focused and, hence, the Pendry perfect lens effect is not realizable with any realistic metamaterial. DOI: 10.1103/PhysRevE.64.056625 PACS number s : 41.20.Jb, 42.25.Bs, 42.79.Bh
Several recent papers have exposed the usefulness of metamaterials that produce negative indices of refraction 1–7 . Metamaterials are artiﬁcially constructed materials having electromagnetic properties not generally found in nature. Examples include photonic band gap structures 2,8 and double negative DNG media 1,3–7 , i.e., metamaterials having negative permittivity and negative permeability. Pendry 3 has proposed the intriguing possibility that a DNG medium could lead to a negative index of refraction and might overcome known problems with common lenses to achieve a ‘‘perfect’’ lens that would focus the entire spectrum, both the propagating as well as the evanescent spectra. Pendry’s analysis followed much of the original work of Veselago 1 . In this paper we use both analytical and numerical techniques to understand more completely the mathematics and wave physics associated with DNG medium propagation and scattering problems and apply the results to explore further this ‘‘perfect’’ lens concept. In particular, we elucidate the choice of the square root
1063-651X/2001/64 5 /056625 15 /$20.00
that leads to the negative angles associated with Snell’s law and the focusing effects Pendry associated with the DNG medium. Pendry’s analysis depends on a certain deﬁnition of the ‘‘forward’’ wave direction, as implied by a speciﬁc choice of the square root for the wave number and the wave impedance. Moreover, inconsistent choices for the propagating and evanescent spectra were made. We apply two wellknown analytic continuation arguments to the choice of the square root. The ramiﬁcations of these choices are clariﬁed in homogeneous DNG media and with the scattering of a plane wave from a DNG medium interface. Comparisons are achieved by deﬁning correctly the sense of polarization; the ﬁeld energies and satisfaction of Poynting’s theorem; and the reﬂection and transmission coefﬁcients for both square root choices. This analysis also allows us, unlike Pendry, to deﬁne the square root consistently for the propagating and the evanescent spectra. One-dimensional ﬁnite difference time domain FDTD simulations of the same DNG wave problems are given. The DNG medium is realized with lossy Drude models for both the electric permittivity and magnetic permeability. This
©2001 The American Physical Society
i. Consequently.0 e ikz i t . from the expressions for Z that the sign of the square root in Z is uniquely deﬁned by k. the exact solution also proves that this occurs for only one particular lossless. However. II B – II D . the wave speed v . H. particularly if the DNG medium has a large negative index or becomes lossy or dispersive. the waves in these more general DNG media are found to coalesce into. The negative index result agrees with the recent work by Smith et al.e. An analytical.RICHARD W. deˆ ﬁning the wave vector k kk by introducing the direction of ˆ ˆ ˆ propagation k where here k z). the free space wave impedance Z 0 number k 0 ¯ / 0 . respectively. 1 and the power ﬂow 056625-2 .or right-handed materials. the derived closed form expression predicts two perfect foci: one at a point within the slab and one at a point beyond the slab. though. a realistic DNG material must in fact be dispersive. PLANE WAVES IN A HOMOGENEOUS DOUBLE NEGATIVE MEDIUM A. 1 is E0 2 1 e Re 2 Z* S 1 2 Re E H* ˆ z 2(Im k)z . Moreover. The possibility of a negative index of refraction material that does not violate causality is also proved analytically and conﬁrmed with these FDTD numerical experiments. This agrees with the intuitive ray picture given in 3 . for example. exact frequency domain solution to the scattering of a wave from a DNG slab is given and is used to characterize several physical effects. the normalized permittivity / 0 . and the nor0 / 0 . in which the wave ˆ propagates in the k direction but has E. The LH plane wave terminology was introduced in 1 for the case. The DNG slab acts like a converter from a pulsed cylindrical wave to a pulsed beam. 4 where the speed of light c 1/ 0 0 . As pointed out in 4 . We ﬁnd analytically that the perfect lens effect does not exist in general. and the polarization of the wave. the analytical and FDTD results demonstrate not only that Pendry’s perfect lens effect is inaccessible with any metamaterial specially designed to realize the DNG medium. Time harmonic plane waves FDTD approach not only supports the steady state phenomena associated with the frequency domain analysis. This is demonstrated with simulations of the ﬁeld intensity and the Poynting ﬂux. we do ﬁnd that. but. ( )/ 0 1 so for a matched medium with ( )/ 0 that n( ) 1. we demonstrate conclusively that the solution is independent of the analytic continuation choices of the square root. In fact. In particular. These include the concepts of causality. even though Pendry’s analysis of the focusing of the evanescent spectrum Eqs. 2 k 3 Z E0 H0 k 1 v Z0 . and Re k forming a left-handed coordinate system so that S •Re k 0 as opposed to a right-handed system for which this projection would be positive. ¯ /¯ . Furthermore.H 0 . the free space wave /c. but also that such a metamaterial could be used to achieve the localized beam ﬁelds for a variety of applications. 5 The deﬁnitions of the wave speed and wave impedance consist of square roots whose proper signs need to be deﬁned. we will call a plane wave that has S •Re k 0 simply a LH or a RH plane wave. Consequently. . Consider an x-polarized plane wave propagating along the z axis: E H E 0 .0. It is shown conclusively that the square root choice leading to a negative index of refraction and positive wave impedance is the correct one. 1a 1b i t the wave number k. there are some differences in the interpretation of the physics that we will elucidate. the DNG medium strongly channels the ﬁelds propagating in it in a paraxial sense along the propagation axis. then regardless of the deﬁnitions of these square roots we have for the DNG medium E0 H0 Re k and S •Re k 0. Rather. 2 – 4 affects the ﬁeld structure described by Eq.0 e ikz 0. 6 To avoid confusion with the terminology associated with chiral left. This subject will be dealt with in Secs. The average Poynting’s malized permeability ¯ vector at the angular frequency corresponding to Eq. and the wave impedance Z being given by the expressions k k 0 n. However. 1 v n c ¯¯ ¯¯ . By studying in the frequency domain within a rigorous context a spectral analysis of the Green’s function for the scattering of a plane wave from a DNG medium slab. it is conﬁrmed that the Pendry perfect lens effect does not occur in general. ZIOLKOWSKI AND EHUD HEYMAN PHYSICAL REVIEW E 64 056625 II.. as we will demonstrate. Using two-dimensional FDTD simulations of the scattering of a line-source-generated cylindrical wave from a DNG slab realized with a double Drude medium. it is used to critically examine the perfect lens concept suggested by Pendry. care must be exercised in interpreting the meaning of the rays in the DNG slab. the ﬁnal result that both the propagating and evanescent spectra are focused is correct. the power ﬂow associated with the wave. localized beam ﬁelds that channel the power ﬂow. 4–6 . as predicted by the analytic solution. c n . 13 – 21 of 3 is wrong because of the inconsistent choice of the square root. We note. nondispersive case for which the index of refraction is 1. but also demonstrates the causal transient behaviors. The deﬁnition of the square roots in Eqs.
note that the sign of the square root for depends on the sign chosen for n. 1 – ˆ ˆ z while S of Eq. ¯ ei . and the electromagnetic power ﬂow associated with both concepts. note with Eq. we ﬁnd that. Re n 0. i. 10 Re that with Eq. PHYSICAL REVIEW E 64 056625 described by Eq. the wave directions associated with reﬂection and transmission from an interface. /4 . 3 /4. 8 ¯ ei . in order to deﬁne the reﬂection and transmission coefﬁcients. /2. 5 .H0 . Explicitly. 9 and 10 . Applying instead ˆ ˆ z while S of Eq. 13 The upper and lower signs in Eqs. 9 . and Im n 0 and note that with Eq.WAVE PROPAGATION IN MEDIA HAVING NEGATIVE .. It is found that physically based expressions for the electromagnetic waves can be obtained that are independent of either of the analytic continuation deﬁnitions. ¯ r /¯ r . The wave directions In a DNG medium.0 . v ) and hence S • v 0. we see that v ˆ points in the z direction. . one needs to deﬁne the ‘‘incoming’’ and the ‘‘outgoing’’ solutions with respect to the interface. and we use the time variable t explicitly whenever confusion may occur . Thus. 10 to the solution in Eqs. 9a 9b /4. 5 versus the ˆ direction v of Re k. respectively. 12a d f e 14 ¯ r /¯ r 1 i and in the lossless limit . 2 – 4 is n 1 2 1 2 /2. 9 . Causality in a linear dispersive medium implies that if ( ) and ( ) are transformed to the time domain via the inverse Fourier transformation 1 2 i t For small loss we may obtain explicit expressions for n and . Since in addition the time domain solutions are real.5 /4 . the wave is LH with respect to both the ˆ wave vector k and and the direction of propagation z. Again. the direction of increase of the real part of the phase. which gives a relation between the time variation of the ﬁeld energy in a region of space and the power ﬂux through the surface surrounding that region. it follows that the frequency domain constitutive parameters should satisfy 056625-3 . This question becomes critical in problems involving transitions through interfaces since. /2. Such a discussion introduces a general characterization of plane wave properties and relies on Poynting’s theorem. Deﬁnitions of the square roots n ¯ r¯ r . We summarize the plane wave properties by considering the direction of the Poynting vector S in Eq. The question of what is the proper deﬁnition of a square root is then replaced with the question of what is the ‘‘outgoing’’ wave direction. but with Eq. 11 F t n ¯ r¯ r 1 ¯i 1 i 2 ¯r 1 2 ¯i ¯r ¯i ¯r ¯i ¯r . 7a 7b one choice for the square roots in Eqs. one has to explore the analytic properties of the dispersion relation k( ) over the entire frequency domain as implied by the requirement that the medium will be causal. i. 5 the root in Eq. In both cases the solution is ˆ ˆ LH with respect to (E0 . the wave is LH with respect to the wave vector ˆ and RH with respect to z. if 0 ¯i then ¯r and 0 ¯i ¯r . 10a 10b Since the choice of the square root deﬁnes the causal properties of the wave solution.e. 0. . 12 and 13 correspond to the deﬁnitions in Eqs. This issue will arise when we discuss the power and energy results asssociated with Poynting’s theorem. . Re n 0. they are not independent. C. 10 . 12b then their time domain counterparts (t) and (t) are strictly causal in the sense that (t) 0 and (t) 0 for t 0 for simplicity we do not introduce special symbols for time domain constituents. Causal medium /2. Henceforth we shall consider two alternative deﬁnitions of the square roots that are based upon analytic continuation considerations. 5 points in the 4 we see that v ˆ direction. while the alternative choice is n 1 2 1 2 The effects of the deﬁnitions of the square roots are best characterized by their effects on the direction of causality. ei .e. 9 Re 0. the relative constitutive parameters satisfy ¯ ¯ Expressing n n e i n. and Im n 0. which is also the direction of exponential dez cay of the wave due to the loss term in k. B. .. which is also the direction of exponential decay of the wave. D. . Applying the root in Eq. Expressing ¯ ¯ r i¯ i and ¯ ¯ r i ¯ i .
the function n( ) and. the DNG medium has to be dispersive so one should use Eq.. 15c E• t P 0 H• t M Thus all the singularities poles and branch points and the associated branch cuts of ( ) and ( ) must lie in Im 0. 19 is expected to be positive if the polarization and magnetization ﬁelds are in the same direction as the electric and magnetic ﬁelds. Thus. choosing the positive root for Im pe .e. 2 pe 2 L 0 1 i e 2 0e .. Z( ) in 0 i. one ﬁnds there that Discrepancies between Eqs. Since ˆ ˆ z is the outward-pointing normal on the entrance face. the RHS of Eq. m2 17b where all the coefﬁcients are positive to ensure analyticity 0 and represent losses that may be taken to be in Im small. We therefore pay special attention to these examples.. 15a ( ) i ( ) and ( ) i ( ) . F. E. In contrast. * for real . m 16b We provide a slight variation of this for discussion purposes: 2 pe 0 1 i e1 i 2 pm . both analytically and numerically. 20 0 1 i . 20 . e2 17a 0 1 i m1 i . A similar relation applies for imaginary part. One readily veriﬁes 0e roughly smaller than pe . Furthermore. 15c and then continuing the square root analytically along the real 0) to the regions axis more generally. the right-hand side RHS of Eq. Im As an example. In addition. 18 2 where e e1 e2 and e1 e2 . Eq. These models are actually Lorentz medium models. 3 and 4 should be analytic in Im branch cut introduced by the square root there should be in 0). nondispersive medium is 1 2 1 2 0 1 i 2 pm .e. 19 in any considerations of the power ﬂow for that case. 0. if an electromagnetic ﬁeld enters a DNG region. e. 15b 0 0 As noted in 7 and by many others. Thus. and should be analytic in As discussed above. The form of has the unit outward-pointing normal n Poynting’s theorem in a homogenous. 9 gives a LH plane wave with respect to k that is ˆ ˆ propagating in the k z direction and gives a positive power ﬂow negative ﬂux into the volume V. pm . n Eq. the Eqs.RICHARD W. On the other hand. ZIOLKOWSKI AND EHUD HEYMAN PHYSICAL REVIEW E 64 056625 * . consider the lossy Drude medium. as they would be in the case of the Drude medium. thereby. 19 and 20 are immediately apparent. As pointed out in 4 . 10 gives a LH plane wave that will have negative power ﬂow into the volume V through the surface because its ﬂux is positive. 19 or Eq. the quantity U em in Eq. 20 is negative. Energy and power and are analytic functions for Im →0 and as → →0 0. Interface and slab problems From the discussions in previous sections it follows that the behavior of the wave scattered from a DNG interface or slab can serve as an another indicator of the correct choice of the square root. which ˆ . Let the entrance face to this region be the plane z 0 and let the exit face be z so that for a ﬁnite time the pulse has no interactions with the exit face. suggested in 3 in connection with the DNG slab 2 pe dV. the correct form of Poynting’s theorem in a dispersive medium is 9 ˆ S•n d V V 0 E• t E 0 H• t H for Im 0. → 0 and → 0 as implied by Eq. let the volume V be a large semi-inﬁnite space in the x-y directions. 20 is then negative. pm i. where and are approximately negative pe with small positive imaginary part . nondispersive DNG medium. In a homogeneous. 056625-4 . e 16a V ˆ S•n d t E 2 H 2 dV V t U em .g. In particular. 9 . in Im . because the two square root choices lead to different results for the left-hand side LHS of either Eq. This implies that square roots for v and Z in this range are described properly by Eq. the power ﬂow into the medium becomes another identiﬁer of the correct choice of the square root. 19 where the volume V is bounded by the surface V. with Im L that Re L 0 for 0 (Im L 0 is relatively small for real if the coefﬁand its cients are small .
0 10 rad/s. However.0 1011 rad/s does not produce a DNG medium at the target frequency. a Real part of the relative Drude .0 1011 rad/s. . one-dimensional 1D time domain simulations of both the one-dimensional interface and slab cases are considered in this regard next. Consequently. We also obtain for this case that S of the transmitting wave in the DNG medium propagates away from the interface.0 1011 rad/s. 9 or Eq. 1 b . 1 a . p 2. b Imaginary part of the relative Drude .0. on the other hand. there is no loss in generality in the results for the choice of the square root. Z Z0 r 01 t 01 1 r 01 21 Consider. p 5. 1.0 10 rad/s where 1. As noted previously. or optical frequencies with the appropriate frequency values in the Drude models. and p 1.0 1011 rad/s. 1 a and 1 b . in fact. This target frequency value f 0 was chosen arbitrarily. As shown in Fig.665 1011 rad/s. model for the angular plasma frequency p 1. 1. but the phase of this wave decreases away from the interface so that the wave on the DNG side seems to be propagating toward the interface.0 10 rad/s. p 056625-5 . hence.0 1011 rad/s. III. p 5. The fact that r 01 and t 01 blow up does not rule out the choice of the square root in Eq. for example. We considered a low loss. .0 108 rad/s when of Re(1 FIG. while the cases with p 2. PHYSICAL REVIEW E 64 056625 The reﬂection and transmission coefﬁcients for a wave impinging normally from the free space side on a DNG interface are given by Z Z0 .WAVE PROPAGATION IN MEDIA HAVING NEGATIVE . the case with p 1. 10 . For the choice in Eq.0 1012 rad/s where 1. These material models are incorporated into a standard FDTD simulator with the corresponding auxiliary differential equations for the polarization and magnetization currents. 1D FDTD SIMULATIONS OF THE DNG INTERFACE AND SLAB PROBLEMS With the square root choices and their consequences in hand. matched DNG medium with so that ¯ ¯ 1 . The angular frequency has been normalized to the target frequency 0 2 f 0 . the series of multiple internal interactions can be summed up in closed form even in this pathological limit and the ﬁnal result provides a regular expression for the ﬁeld which is given in Eqs. All of the results can be achieved in a similar fashion at any set of microwave. p 2.0 108 rad/s. where f 0 30 GHz. and p 1.0. Thus in order to address the issue of the ‘‘correct’’ or ‘‘physical’’ choice for the square root it is suggested to study ﬁrst the problem of an interface with a matched DNG under transient excitation. the phase of the transmitted wave increases propagates away from the interface but S propagates toward the interface. In particular.0.0 108 rad/s. we simulated the interaction of a plane wave with the DNG slab in 1D and the interaction of a line source ﬁeld with a DNG slab in 2D using a ﬁnite difference time domain simulator 10. we obtain Z Z 0 and r 01 0. millimeter. p 5.11 . the Drude DNG medium modeled with the FDTD simulator is temporally dispersive. 43 and 46 below.0 1012 rad/s are given. p 5. Plots pe pm p and e m ) and Im(1 ) for 1. For the choice of k in Eq. It should be pointed out that this ﬁnal result is. in this case we have Z Z 0 . p 1. r 01 and t 01 blow up. model for the 11 angular plasma frequency p 1.0 1011 rad/s. Moreover. respectively. Note that there is no possibility of inserting a constant negative permittivity or permeability into a leapfrog staggered grid FDTD simulator and having it run stably. the losses in all cases are small in comparison to the real parts.665 1011 rad/s. a ‘‘matched medium’’ where ¯ ¯ . For the DNG slab. The DNG slab was modeled by the lossy Drude medium 16 . Z Z0 2Z . in Figs. independent of the choice of the square root in either Eq. 10 . Furthermore. and 27. the values of Re(1 ) at the target frequency are approximately 1. 10 since this special case can be treated in a proper limit sense.665 1011 rad/s.665 1011 rad/s.0 1012 rad/s do. As shown in Fig. 6. 9 . p 1. this auxiliary material approach allows one to isolate the medium physics from the ﬁeld physics for postprocessing purposes. p 11 12 2.
26 10.5 En x 1 i 1/2 . 23 The initial ﬁeld is launched from a total-ﬁeld/scattered-ﬁeld TF-SF plane within the mesh 10 . x x n 1/2 1/2) t….n t). 2 pH y 1 0.n t…. This required averaging of the electric ﬁeld and electric current values in the discrete equations.0x on .5 1 0.0/6.0 (mT p t)/mT p and x o f f t (m n)T p /mT p . The Courant number must be less than or equal to 1 for the FDTD simulator to be stable for free space problems. 2 pE x . 1D FDTD simulations PHYSICAL REVIEW E 64 056625 The 1D time domain equations solved with the FDTD simulator for the matched DNG medium were tE x Hn y 1/2 i 1/2 Hn y 1/2 i 1/2 0 t z z . 22 i En i x 1 n J 2 x t 0 1/2 z Hn y 1/2 i 1/2 1/2 Hn y 1/2 i 1/2 where K y has been normalized by 0 to make the magnetic current equation dual to the electric current deﬁnition. this is an exact truncation. the magnetic ﬁeld is taken at the cell center for half-integer time steps. The electric and magnetic currents were located together at the cell centers in order to achieve the matched medium conditions numerically. 056625-6 .0x on 6.(n 1/2) t…. tK y Ky 0 . and the peak of its frequency spectrum occurs at f 0 1/T p . The 1D FDTD grid was terminated at both ends with a one-way wave operator absorbing boundary condition.0 3 4 5 10.0x 3 f f o 15. Two types of unit amplitude pulse were used.5 1 0.0x 4 f f o 6.0x 5 f f . ZIOLKOWSKI AND EHUD HEYMAN A. 5-10-5 pulse or a cw signal 5-1000-5 pulse was used to probe the DNG slab. i 1/2 t n K i 1/2 t y t t Hn y 1/2 tJ x Jx 1 0 0 tH y zE x Ky . When the Drude model is included. and then a smooth windowed turn-off for m cycles.0 3 t T p /2 T p /2 1 t T p /2 T p /2 2 3 for 0 t T p for t T p . o where x on 1.5 t x i En x 1 i 1 . the discretized equay tions in the FDTD simulator are i 1/2 t n J t x Jn x 1/2 i 1/2 z . En i 1 x En i x 1 0 Kn y Kn y 1 i 1/2 1 0. broad bandwidth pulse f t 0 7. either a 20-cycle. The other input signal is the multiple cycle m-n-m pulse: g on t sin sin f t 0 0t 0t 0t for 0 t mT p for mT p t for t m n Tp m n m Tp 25 for m n T p t g o f f t sin m n m Tp . If J n 1/2(i 1/2) J x „(i 1/2) z. This single cycle pulse has a broad bandwidth. One was a single cycle. The m-n-m pulse is a sinusoidal signal that has a smooth windowed turn-on for m cycles. The electric ﬁeld is taken at the cell edge for integer time steps. a constant amplitude for n cycles. it has an adjustable bandwidth through the total number of cycles m n m) centered at the frequency f 0 .5 i 1/2 1 2 1 0 p t En 2 1 0. where T p 2 / 0 1/f 0 is the period of one cycle and the three-derivative smooth window functions are given by the expressions g on t gof f t 1. For the cases considered below. leapfrog in time approach.(n E n (i) E x (i z. the Courant condition is more complicated but is well known. 24 where T p is the length of time the pulse has a nonzero value.0x on 15.0 7. H y (i 1/2) H y „(i 1/2) z. For a Courant number c t/ z 1.5 0 2 p zH y Jx . and K n (i 1/2) K y „(i 1/2) z. hence.RICHARD W. A recent review 12 discusses the various discretization schemes for a variety of dispersive media. These equations are discretized with the standard staggered grid.0. Jn x 3/2 i 1/2 1 0.
The peaks in Fig.95 z/c 9. DNG medium results FIG. PHYSICAL REVIEW E 64 056625 FIG. those for the energy quantity E x H y t are shown in Fig. B. From Fig. but RH with respect to ˆ the direction of propagation z. and both negative permittivity and permeability. Although there is some peaking in Fig. The energy received at a ﬁxed observation point behind the DNG slab is plotted as a function of the simulation time. 4.5 10 4 smaller than the incident ﬁeld. 4. Unless otherwise noted. The time history of the Poynting vector term E x H y was measured at three points in the DNG medium. the front face of the DNG slab at z i z where i 1200.WAVE PROPAGATION IN MEDIA HAVING NEGATIVE . The E x ﬁeld was measured at i 610. 3. opposite polarities for the single negative parameter cases. Thus. This is conﬁrmed in Fig. First. The steady state values for the term E x H y were successively less at each of the three observation points. 1. The DNG medium is lossy and does absorb some of the energy. The measured electric ﬁeld has the correct. 1500. The total ﬁeld/scattered ﬁeld plane was set at z i z where i 600. 4. and a DNG medium. Several quantities were measured in the 600-cell DNG slab case. This was tested further with the cw incident pulse. the 1D FDTD problem space for the simulation results discussed below was taken to be 5000 cells long where z 3.0 10 5 m 0 /300 where the target frequency f 0 c/ 0 30. The time histories of the electric ﬁeld E x measured in front of the total ﬁeld/scattered ﬁeld plane for a negative permittivity medium. 2. the DNG slab was 600 or 1200 cells thick. 2. 4 as is the free space result. the reﬂected pulses on the right. The corresponding time step was t 0. The results for E x H y for the 20-cycle incident pulse are shown in Fig. Values for thin and thick slabs are compared to reference values. Next. a negative permeability medium. The energy quantity E x H y t was measured just in front of the TF-SF plane at i 610 and just after the slab at i 1810. 3. negative permeability only. The results are shown in Fig. The slab was deﬁned by the DNG cases shown in Figs. The single cycle pulse was launched toward a 600-cell slab that had negative permittivity only.0 GHz. . essentially at the expected ‘‘matched’’ level for the discretization used. The DNG medium can dramatically slow down a wave propagating through it. 3. and the back face of the DNG slab at z i z where either i 1800 or i 2400. Figure 4 also demonstrates that the DNG medium slows down the propagation speed of the pulse considerably. 056625-7 . the term E x H y decreases in the DNG slab. The incident pulse appears on the left. The wave in the DNG slab was deﬁnitely LH with respect to k. The reﬂected pulse in the DNG case was 1. The energy calculation for a 1200-cell DNG slab is also provided in Fig.5 10 14 s 95 fs. . the propagation characteristics of the pulses in the DNG were investigated. These broad bandwidth components have different propagation characteristics from the main part of the pulse as conﬁrmed by the cw runs. 3 are due to the ‘‘turn-on’’ and ‘‘turn-off’’ portions of the incident pulse. The term E x H y was measured in time at three points i 1205. the matched nature of the FDTD equations 23 was tested with a simple interface problem. There is no gain in the signals. The FIG. 1795. 10 cells in front of the TF-SF plane. 3 we found that E x H y was positive.
we have vg As expected. . respectively.0 10 8 1. matched DNG medium with ¯ ¯ . 5 for the 20-cycle 11 rad/s. 4 at which 50% of the energy has been collected at the observation point. the agreement is very good. ZIOLKOWSKI AND EHUD HEYMAN PHYSICAL REVIEW E 64 056625 steady state wave front propagates at the group velocity of the target frequency f 0 . Here. only propagation in the forward direction was observed in the slabs for time prior to the waves interacting with the back side of the DNG slab. In fact. The negative index was also calcu- 056625-8 . 31 be labeled P RHS . since .0343c. we have 2 / 2 2. hence.26%. The difference between the analytical and the numerical times is 0. for the low loss. The expected values were (600/0. 5 b . 27b To further conﬁrm the signs of Poynting’s vector and the energy. by calculating P J t and 0 M K t. the energy U em was found to be negative over the extent of the pulse duration. 19 for the 600 z DNG slab under consideration becomes z 1800 z z 1200 z 600 z 0 0 E• t E 0 H• t M c n 1 2 pe i i e e /2 2¯ 2 pm i i m m /2 2¯ 1 E xH y E• t P 0 H• t H 28a c n 1 2 pe 2¯ 2 pm 2¯ 1 dz. if we let the DNG medium’s permittivity and permeability exhibit small losses.3334c. the relationship between the causal propagation and the negative index of refraction predictions for the DNG medium was considered.95 8924 t.124 600) t/0. the wave group speed 29 is.0 1012 rad/s DNG media. one has approximately that vg c 1 2 p 2 1 .e. P RHS P LHS LHS P max 2 For the 2. This has been previously conﬁrmed numerically for related media in 7 . 5. The power in the medium as calculated by the LHS of Eq. Selecting the point on the curves in Fig. hence. This can occur only if E x H y is a positive quantity. let the power in the medium as calculated by the RHS of Eq.m .e.. 0.0 1011 rad/s. 32 is shown in Fig. the general formula for the group velocity is vg k c 1 n 1 c „ n e … m 1 1 27a . Positive energy ﬂows into the DNG medium to cause the term t U em to increase. and for the long DNG slab is 8901 t. and 0. . The corresponding result for the RHS of Eq. 20 was calculated for the DNG slab as 600 z U em 1/2 0 0E P •E 0 H M •H dz. several more quantities were calculated. 5 a .e. Moreover.e. Let the LHS maximum instantaneous value of P LHS be P max . and 1. In particular. The ability to distinguish the powers across the two faces occurs because the wave speed in the DNG slab is so slow. For the Drude DNG 17 . it is in the z direction and that v g ( ) c/n i. In particular. it is the negative of the difference in the ﬂux through the DNG slab.665 1011 rad/s. i. one veriﬁes that for p pe v g ( ) 0 i. it is desired to avoid the earliest time points since the envelope of the excitation pulse contains a variety of high frequency components whereas late times require dealing with issues of the medium’s relaxation rate.RICHARD W. The wave simply slowed down in the DNG slab and ﬁnally made its way out of it. 31 . but negative across the exit face.0 10 rad/s slab. The results for this quantity are given in Fig.. This further conﬁrms the fact that the wave in the DNG medium is ˆ LH with respect to k but RH with respect to the direction of ˆ propagation z. Note also that the phase velocity v p c/n( ). A precise measure of the speed from the simulation results is quite difﬁcult. The normalized square difference between these two power calculations. pulse excitation of the p 5. 31 will be labeled P LHS .95 4462 t and (1200/0. Moreover. it is very slow . 0.124c. 29 The calculated results are shown in Fig. No change in the direction of causality was observed in the 1D FDTD simulation results. since the slab was matched to free space.. C. e. both expressions yield positive power values in the DNG slab. then we have the result that the power ﬂow is increasing as the wave enters the medium while the energy density becomes more negative. 28b where the last approximation applies if the loss term is small. 30 2 ¯ ¯ where in the last expression the susceptibility terms ¯ 1 ¯ 1 e and m have been used. one ﬁnds that the difference in the number of time steps between the reference and the short DNG slab is 4412 t.. Notice that P LHS is positive across the entrance face.124 1200) t/0. the RHS of Eq. For our target frequency f 0 . Index of refraction Finally. its predicted values agree quite precisely with the instantaneous power ﬂow difference calculations. Despite the spatial integration averaging and the time derivatives associated with P RHS . On the other hand. ¯ ¯ 1 i. no reﬂections from this back side of the DNG slab were observed.
. Note. 6 a . a The negative ﬂux of the instantaneous Poynting vec11 tor through the DNG slab with p 5. a Early-time results. The later-time results at the same points are shown in Fig. Causality. E n FDTD ln . . 33 The values Re n FDTD 1.0 108 rad/s to give n r ( f 0 ) 1. i.665 10 rad/s and 3 at the target frequency f 0 1. The negative index of refraction nature of the DNG medium is clearly seen. The agreement is very good and further conﬁrms the presence of the negative index of refraction for the DNG medium. The electric ﬁelds at the center of the slab and 10 cells beyond it. that in contrast to the description in 4 .00 and Im n FDTD 1. The negative index of refraction expresses itself after only a few cycles.e. We E then considered the quantity 1 ik 0 z 2 z 1 ˜ x z2 . The discrete Fourier transforms were calculated and are labeled as ˜ x (z.665 1011 rad/s and 1.t). To calculate the index of referaction. FDTD predicted electric ﬁeld time histories at two points: point 1 with z 130 z.06 10 30 Ghz.WAVE PROPAGATION IN MEDIA HAVING NEGATIVE .00 and n i 1. lated explicitly. were sampled. The time histories contained 20 000 FDTD time steps and were then zero-padded out to 32 768 time points.0 10 rad/s and 1. It is clear that the response of the ﬁeld at the ﬁrst point z 1 130 z occurs before the one at the second point z 2 140 z. The cw was driven into a DNG medium with 11 1. however. 6. The DNG medium is speciﬁed by p 2.0 108 rad/s so that n r p 2.07 10 3 were obtained for f f 0 . E ˜ x z1 . The very early-time responses of these measurements are shown in Fig.0 108 rad/s is plotted against the number of FDTD time steps for the 20-cycle pulse excitation. i. 50 cells into the DNG medium. b late-time results. PHYSICAL REVIEW E 64 056625 FIG.t) and E x (z 140 z. .00. This demonstrates that causality in the direction of wave propagation is indeed preserved in the DNG medium. the wave in the medium travels away from the source in a direction opposite to the direction determined by the negative index.e. ). The response at z 2 140 z leads the one at z 1 130 z. Fourier transforms of the two electric ﬁeld time histories were taken. 5. b Normalized square difference between the negative ﬂux of the instantaneous Poynting vector through the slab and the integrated local power density in the same slab. FIG. 6 b .. the wave propagating away from 056625-9 . and point 2 with z 140 z. E x (z 130 z.
The wave physics in a DNG medium is thus described by all of the expressions associated with that choice. 40a k 2 z G r. i (z d) 43a ei 0 z0 g 2i 1 2i T 0 Z0 i e tZ 0 (z re 2i T cos 0 z d i Z0 sin Z z d . 9 . 42b ˆ I 0 “ zG TM . and Z appear with opposite signs for either choice. z d. both R and T are independent of these deﬁnitions because they are even functions of . Because R and T are symmetrical in and in Z. 43b 43c ei 0 d z0 ) T. It is sufﬁcient to explore the focusing properties for the TM and TE Green’s functions. Z0 Z 41 are continuous at z i .2. This formulation is therefore a convenient starting point to explore the effects of the DNG slab.z .r r r 34 Z . 40b subject to the continuity conditions G TM and G TE and 1 1 TM are continuous at z i . 42a ˆ “ “ zG TM .r where k t 1 2 2 dk x dk y e i(k x x k y y) g z.7. The ABCD matrix analysis in 14 . It can be synthesized by plane wave superposition where d. The solution for g is now given by ei ei 0 z z0 2i 0 z0 R 0 ei 0( z0 z) 2i (z d) . does remain intact despite the presence of what appears to be a wave propagating in the DNG medium toward the source. For example. the solution 41 – 43 is independent of either analytic continuation choice of the sign in the square root for the DNG slab. i. because Z is. Henceforth we shall consider only G TM and remove the superscript TM . is given by Eqs. 3.3 or 5.2 2 subject to the same boundary conditions as in Eq. Sec. 1 and 4c E r H r I0 1 i z cos i sin 1 2 . 0 t k 2n 2 k 2. 0 t Z0 0 0 .56 there for R and T. Note that r and t are subject to the deﬁnition of the square roots. 5. Z0 Z t 1 r 2Z 0 . 0 z d.. zG zG TE 35a 35b The reﬂection and transmission coefﬁcients of the interface from the double negative medium into air are r Z0 Z .z . 3. Sec. FOCUSING PROPERTIES OF A DOUBLE NEGATIVE SLAB G r. The 1D FDTD numerical results essentially conﬁrm that the correct choice for the square root is indeed Eq. On the other hand. z 0 0. 38 k 2 k 2 and x y d2 dz 2 k 2 k 2 g z. These functions may be used to describe the ﬁeld of any other source conﬁguration. the ﬁeld due to a longitudinal dipole source J r ˆ zI 0 r r 36 13 .0.k t t z z 39 We consider the focusing of point sources through a DNG slab of width d located between z 0 and z d. 2.z 0 ). where z i are the points of any discontinuity in the medium. 35b . and in particular Eq. Sec. 37a 37b T kt cos i sin 1 2 Z/Z 0 Z 0 /Z .k t . gives the reﬂection and transmission coefﬁcients from the slab: R kt 1 i 2 sin Z/Z 0 Z 0 /Z Z/Z 0 Z 0 /Z 1 located at r (0. 0 z 0. ZIOLKOWSKI AND EHUD HEYMAN PHYSICAL REVIEW E 64 056625 the source.e.RICHARD W. The spectral wave numbers and wave impedances in free space and in the slab are 0 k 2 k 2. as is the ﬁeld term in the slab region. deﬁned by the equations 13 . IV. 056625-10 .
46 do not vanish identically at a given point and the focusing effect is lost. . z f 2 2d z0 . while for the choice 10 one has Z and t . for either choice Z 0 so that r of the square root. The ﬁnal ﬁeld description within this DNG slab consists of a single wave species whose phase propagation direction is backward with focus at point z z 0 . denoting the standard deﬁnition i. with positive real part of the square root in Eq. 40b as ˜ . . Our derivation is complete and is independent of the square root deﬁnitions. ˜ k 0˜ n k2 t 2k 0˜ n . A. regardless of the choice of the square root. z 0. 2d) 47a 47b e i[ 0 0 (z z0 . in this case sin 1 sin n inc trans inc . Thus. we shall explore the possible focusing of the paraxial spectrum components small k t /k 0 ) for which 0 k0 k2 t 2k 0 . 0 z d. 47a and 47b vanish at the points zf1 z0 . 10 .. Then one observes that the arguments of the exponentials in Eqs. Consequently. One may associate the arrows with the energy direction as in Pendry’s paper but one may alternatively choose to associate them with the phase progression. giving the opposite direction. T cos ˜ d i sin ˜ d 1 e i˜ d .’’ Note also that Pendry’s choices for the square roots in 3 for the propagating and evanescent spectra were not consistent. 0 46a 1 2i e 0 i( 0 z 0 ˜ z) g 2i 1 2i e 0 i˜ d e i ˜ (z d) . 44 where ¯ ¯ 1. Here the phases in Eq. if z 0 d. Thus. consider the more general case in Eq. Next. 1 so that the incident and transmitted rays are located on the same side of the normal to the interface rather than on opposite sides.e. Yet the arrow directions on these rays are open for interpretation. The ray picture is thus as seen in Fig. Inside the slab. as did Pendry in 3 . respond to the deﬁnition of the square root in Eq. However.e. i. 0 z d. Eq. As a result. 48 Thus. does not assume a speciﬁc choice of the square root and thus of the ‘‘ray directions. however. 9 and in Eq. 46b . z d. respectively. the phase progression is negative as implied by Eq. Eq. the lossless. we ﬁnd that → ˜ and Z→ Z 0 where the upper and lower signs cor- ei ei 0 z z0 2i 0 z0 . Different choices were made for these two spectra. disper1 so that ˜ sionless case for which ¯ ¯ 0 . The analysis above. 42 becomes R 0. 43 yields then. respectively.WAVE PROPAGATION IN MEDIA HAVING NEGATIVE . 46b 46c e i[ 0 0 (z z0 d) ˜ d] . Equations 46b and 46c become of 3 . one does ﬁnd that the slab solution is completely independent of the choice of the square root. Now consider. the ﬁeld for z d focuses again at z 2d z 0 . Solution in a matched DNG slab PHYSICAL REVIEW E 64 056625 Speciﬁcally. 50 056625-11 . 49 1 2i g 1 2i e 0 i 0( z0 z) . z d. if one takes a slab matched to free space so that ¯ ¯.. The presence of the foci results from the negative index of refraction which causes the transmitted angle to be negative. 45 44 Note explicitly that for the choice 9 one has Z Z 0 so that r 0 and t 1. This observation applies for both the propagating spectrum and the evanescent spectrum. then the ﬁeld in the slab and the transmitted ﬁeld beyond the slab have perfect foci at z f 1 and z f 2 .
51b g 1 e 2ik 0 ik 0 [z d(1 ˜ )] n e i(k t /2k 0 )[z 2 Thus. 57 i 2 . Solution in a lossy DNG slab Thus the beam parameters in the lossy DNG medium and. 58a g 1 e 2ik 0 exp ik 0 [z z0 d(1 ˜ )] n i k2 t 2k 0 z z0 d 1 ˜r n ˜ 2 n id ˜i n ˜ 2 n . as ˜ → . 1 e 2ik 0 g ik 0 ( z 0 ˜ z) n e z0 i(k t /2k 0 )( z 0 2 ˜ z/n ) .. Expressions 52 for the foci are still valid but they become complex and hence there are no real foci and the lens effect is completely lost. dispersion1 is not physically reless DNG medium such as n( ) alizable. with a waist Wa b/k at z a 56 and collimation length 2b. in particular. the waist width may readily be obtained by comparing Eq. 58a and 57 . Expresn a z 0 ˜ r and its collimation length b sion 58b represents a beam that propagates in free space The ﬁeld due to this source is a globally exact beam solution of the wave equation that is conﬁned essentially near the z axis. the paraxial expression for the spectrum of the ﬁelds in Eq. a paraxial focus at z f 1 d/2 can be achieved only if n ˜ z 0 d/(2n )→0 and hence z f 2 →d. Comparing n n deﬁnition of n the quadratic terms in Eqs. Here. However. 38 . respectively. Expressions 46b and n 46c become. one concludes that this expression represents a beam that propagates along the negative z axis in a medium with ˜ and has its waist at n n z 0 ˜ i 0. 51 is rewritten here as g 1 e 2ik 0 exp ik 0˜ ( z n ˜ z 0 /n ) i k2 t 2k 0˜ n z Recalling next that 4 shows that a lossless. 0 z d.e. Finally. however. ˜ denotes the standard positive real part square root of n n i. of course. In that case. let us return to the focusing issue from the perspective of the DNG Drude medium. we consider the lossy DNG Drude slab. and the focusing effect is indeed lost. Furthermore. Also note that for a DNG slab with a large negative n index of refraction. However.RICHARD W. 0 z d. 53 Again. 45 . i. these paraxial focus locations would be very difﬁcult to distinguish from any interface effects. referring to the spectral representation in Eq. one has 2 p This expression is readily recognized as a Gaussian beam ﬁeld that propagates along the z axis. one might expect that the resulting ﬁelds will take on beamlike forms. which is generated by a source at r 0.e... that z f 1 d and z f 2 d. in this case the dispersive nature of the medium contributes signiﬁcantly to the loss of any focusing. 55 provided. the phases in Eq. 55 is given by g 1 e 2ik 0 i[k 0 (k t /2k 0 )][ (z a) ib] 2 2 0 2 c2 i 1 2 2 p . since the foci are located at complex points. 50 with ˜ r 0 and ˜ i 0. Note that these ‘‘paraxial foci’’ become the ‘‘perfect foci’’ of Eq. 54 z0 ˜ r n i z0 ˜ i n . the paraxial components may focus at G n z f1 ˜ z0 . ˜ z f 2 d 1 1/n z0 . i. b 0. We therefore consider a so-called complex source beam. we consider only the paraxial regime near this axis where the solution due to the source 54 being located in uniform free space is given by 056625-12 . To this end Eq. ˜ is positive in the lossless DNG medium and has a negative imaginary part for the lossy case . following the deﬁnition of ˜ as explained before Eq. z d. 48 when ˜ →1. z0 ˜ d(1 1/n )] 51a . 52 e ik r r r 1 2 r z a x2 y2 z a ib ib 1 exp ik z a ib . ZIOLKOWSKI AND EHUD HEYMAN PHYSICAL REVIEW E 64 056625 where. a point source located at a complex position produces a Gaussian beam ﬁeld locally 15 . 51 with Eq.a ib .0. 58b ˜ where we have used explicitly ˜ ˜ r in i . recalling the n n ˜ in Eq. 57 . 46 do not vanish identically at a given point.e. B. z d.
The DNG slab is outlined and the source is located at the intersection of the horizontal and vertical lines.210 .WAVE PROPAGATION IN MEDIA HAVING NEGATIVE . FIG. The magnetic and electric currents were taken at the cell centers.0 10 rad/s DNG slab. The DNG slab is outlined and the source is located at the intersection of the vertical and horizontal lines.0 DNG slab.808 and J 110. The slab was thus d 100 z thick in the direction of propagation and the source-to-slab distance was z 0 50 z d.0 was Drude slab with p 2. The electric ﬁeld intensity over the FDTD simulation 11 space at t 3900 t for the rad/s and p 5. In both types of ﬁgure. a lossless DNG 0. the electric ﬁeld was taken at the cell centers.0 10 rad/s DNG slab.01 cm long. The magnetic ﬁelds were taken at the cell edges. The electric ﬁeld intensity over the FDTD simulation 11 rad/s and space at t 3700 t for the p 1. The electric ﬁeld intensity over the FDTD simulation space at t 3900 t for the lossless p 2. staggered grid approach. In the intensity ﬁgures below gray-scale contours represent different levels. and cw pulses were used to investigate the wave propagation effects in the DNG slab.0 108 rad/s DNG slab. These equations were again discretized with a standard leapfrog in time. some of the contours in regions of expected low values are unavoidably black in order to achieve the targeting lines and the slab outline. 056625-13 . corresponding to 0 /100. Thus. The electric ﬁeld intensity over the FDTD simulation 11 space at t 3700 t for the rad/s and p 5. The DNG slab was located in the cell region K 12. 9.95 z/( 2c) 223. 20cycle. The FDTD grid was terminated with a 10-celllayer. and H z . except for the lossless case in which the p 2. for the ˆ ˆ x and z directions. The darker lighter regions correspond to higher lower intensity levels.224ps. The time step was set to be t 0. . Thus. The index labels were K and J. Square FDTD cells were used. the darker lighter regions correspond to higher lower magnitude values. Note that the line source takes a time derivative of whatever excitation signal is speciﬁed. The cell sides were x z 0. the conditions speciﬁed by the analytical solutions for the foci 48 inside and outside the DNG slab geometry were met. C. along the positive z axis and has its waist at a d(1 ˜ r / ˜ 2 ) z 0 and its collimation length b ˜ n dn i / ˜ 2 n n 0. The line source was located at K 410. The DNG slab is outlined and the source and the paraxial foci locations are located at the intersection of the horizontal and vertical lines.0 10 1. the appropriate averaging occurred here in the magnetic ﬁeld and current equations. respectively. two-time-derivative Lorentz material 2TDLM ABC 16–18 .665 1011 rad/s and considered. PHYSICAL REVIEW E 64 056625 FIG. In an attempt to view the perfect lens foci. FIG. 10. in contrast to the 1D case.0. The amplitude of the excitation was normalized by 1/350 10 6 to compensate for the change in the incident pulse amplitude caused by the time derivative. The line source was driven with the cw pulse for all of the cases reported below.917 10 15 s 0. J 60 in all cases. H x .0 10 8 1.0 10 8 1. The single cycle.665 1011 rad/s case was used with 0.655 1011 rad/s and 0. 1. The DNG medium parameters were again those shown in Fig. The DNG slab is outlined and the source is located at the intersection of the vertical and horizontal lines. In the Poynting vector ﬁgures below. 8. 2D FDTD simulations The 2D FDTD simulator solved the equivalent TM set for E y . Note that the line source was placed at 50 cells from the front of FIG. 7. ˆ The x-z FDTD simulation space was 820 cells in the x ˆ direction for the lossless DNG Drude and 620 cells in the z case and 820 cells 320 cells for the lossy DNG Drude cases reported below. .
7. almost identical found for this ¯ r ¯ r results were obtained for the slightly lossy p 2. The slabs that were simulated were all located in the cell region K 12. No steady state foci were found. As predicted. IV A . 12.0 1011 rad/s. 11.0 10 8 10 rad/s was considered. 13. 056625-14 . FIG. This channeling of the wave energy within the DNG slab is further described by viewing the FIG. and they were not growing in the z direction as steady state conditions were obtained. despite numerous attempts with the lossless cases. they begin to diverge as though they orginate from an extended line source at the rear face of the slab. particularly any misplaced foci. The electric ﬁeld intensity over the FDTD simulation 12 rad/s and space at t 3950 t for the p 1. The ﬁrst paraxial focus should therefore appear near the center of the DNG slab. Figs. these locations vary dramatically over time.0 with p 1.180 . there is an enhanced ﬁeld level where it is expected just beyond the rear face.808 and J 80. the conditions speciﬁed by the analytical solutions for the paraxial ˜ n foci 52 are z f 1 ˜ z 0 60 z and z f 2 d(1 1/n ) z 0 130 z from the front of the DNG slab. The electric ﬁeld intensity E 2 in the FDTD simulation space at t y 3900 t is given in Fig.190 . This behavior is symptomatic of the highly dispersive nature of this DNG medium as discussed in Sec. In fact. The slabs were thus d 100 z thick in the direction of propagation and the source-to-slab distance was z 0 20 z d.0 10 1.0 10 rad/s. and 1. For all 10 rad/s and of these cases. Thus. 10 and 11 do show a distinctive beamlike channeling of the waves near the normal as the permittivity and permeability become more negative.0 108 rad/s DNG slab. 9 represents the case where the slab has positive permittivity and permeability at the target frequency f 0 . The DNG slab was located in the cell region K 12. While there are no focal points.0 10 rad/s were considered. The term E y H x is plotted over the FDTD simulation 11 rad/s and space at t 3750 t for the p 5. these observation times represent enough time for many cycles to propagate through the slab and enter back into free space beyond the slab. One does observe a paraxial focus near the center of the DNG slab. 50 cells beyond the front of the slab.0 108 rad/s DNG slab. 7. respectively. 12 8 1. For all of these cases. but were RH with respect to the propagaˆ ˆ tion axis z. Figs. The DNG slab is outlined and the source is located at the intersection of the vertical and horizontal lines. the conditions for foci or paraxial foci speciﬁed by the analytical solutions for the DNG slab geometry were not met. it was desired to determine the structure of the ﬁelds for cases when no paraxial foci should appear in the DNG slab. one ﬁnds that there are no focal points in any of these cases. IV A . t 3700 t. ZIOLKOWSKI AND EHUD HEYMAN PHYSICAL REVIEW E 64 056625 FIG.RICHARD W.0 10 8 1. It was found that the ﬁelds were LH with respect to k. While something like large amplitude ﬁelds occur in the neighborhood of the expected foci as shown in Fig. As noted in Sec. Pendry’s lens effect was not 1 case.0 p 5. In contrast. These results conﬁrm that a large negative index of refraction. The defo1011 rad/s and cusing is inherent in any realistic DNG medium. Finally. and the second should appear 10 cells beyond its back face. The locations of the source and the paraxial foci are noted. The numerical intensity of the electric ﬁeld E 2 in the y FDTD simulation space at t 3900 t is plotted in Fig.808 and J 70.665 1. this DNG slab with the hope that the focus in the slab would occur at the center of the slab.0 10 1. and t 3950 t. the lossy DNG 11 Drude slab with rad/s and 1. Slabs 11 5. the paraxial focus beyond the rear face is very difﬁcult to distinguish from the behavior of the ﬁeld as the wave exits the slab. Thus. dispersive DNG slab completely lacks the presence of any foci. As the waves reenter free space from the DNG slab. 10 and 11 represent the results for DNG slabs at the target frequency f 0 . Recall that Fig. The slab was thus d 120 z thick in the direction of propagation and the source-to-slab distance was z 0 10 z d. Nonetheless. The numerical intensity of the electric ﬁeld E 2 y in the FDTD simulation space is plotted in Figs. The FDTD problem space was increased in size in this case with the hope of seeing the wave dynamics better. The term E y H x is plotted over the FDTD simulation 12 space at t 3750 t for the rad/s and p 1.0 10 rad/s DNG slab. and that the focus outside the slab would occur at 50 cells beyond the back of the slab along the z axis. 8.0 108 rad/s DNG slab. In an attempt to view the paraxial foci. sometimes even completely disappearing. The DNG slab is outlined and the source is located at the intersection of the vertical and horizontal lines. 9–11 for these three DNG slabs at the times t 3700 t. The DNG slab is outlined and the source is located at the intersection of the vertical and horizontal lines. This is particularly obvious in both the regions 0 z d and z d.
The FDTD simulations further conﬁrmed this conclusion. particularly when the index of refraction had large negative values. T.W. 77 2001 . Radiation and Scattering Prentice Hall.O. Computational Electrodynamics. IEEE Trans. Phys. Inc. Kroll. lossless DNG medium having ¯ ¯ 1. It was demonstrated that the choice of the square root that produces a negative index of refraction and positive wave impedance is the correct choice. IEEE Trans. 12. Phys. Pendry. Ziolkowski.R. Wittwer and R. NJ. 200 2000 . Phys. T. This behavior is readily seen in the time slice shown in Fig. Mag. Analytical continuation based choices of the square roots associated with the index of refraction and the wave impedance were introduced. 3 J. 14 A. PHYSICAL REVIEW E 64 056625 Poynting vector component E y H x which is plotted throughout the simulation space at t 3750 t for the p 5.A. 12 and 13 the waves along the z axis are traveling at speeds v g c. The waves away from the axis travel faster than those on this axis. 10 A. however. Taﬂove. Rev. which further explains the channeling of the wave energy along the propagation axis. D.A. Taﬂove Artech House. Wittwer and R. 8 J.. 15 G. 046604 2001 . 78. 23 1971 . Englewood Cliffs. 6 R. Smith. 5 R. 250 2001 . Science 292. 85. 1998 . Radiation and Scattering of Waves IEEE Press. Joannopoulos. IEEE Trans. lossy DNG medium. Lett. 056625-15 . Felsen and N. The presence of the negative index of refraction was shown to be not in disagreement with causality. 49. and S. Shelby. The DNG slab solution was shown to be independent of the choice of the square roots. and J. V. 7 R. Numerical FDTD simulations of these 1D and 2D cases were provided. 17 D. 7. Note that in Figs. Rev. NJ. Electromagnetic Wave Propagation.L. R. The analytical solution for a matched DNG slab demonstrated that the Pendry ‘‘perfect lens’’ effect can be realized only in the presence of a nondispersive. Smith.G. n( ) 1.. 13. Nemat-Nasser.WAVE PROPAGATION IN MEDIA HAVING NEGATIVE . Lett. Phys. These simulations did. edited by A. This possibility was correctly predicted by the analytical solution. Notomi. 13 L. Young and R. Princeton. 489 2001 . and its presence was veriﬁed with the FDTD simulations.C.A. Piscataway. Veselago.W. Kosaka. 16 D. 2933 2000 . Tamamura. IEEE Antennas Propag. Ziolkowski. Norwood. 10. No focal points either within the slab or in its exterior were found in any of the FDTD simulations. 1994 . Kawakami. Deschamps. Ziolkowski. and S.0 1012 rad/s case in Fig. 2 H. MA. 9 J. 85. Photonic Crystals: Molding the Flow of Light Princeton University Press. 18 D.D.e. Ishimaru. and S. Shelby. Rev. The DNG slab converted the cylindrical wave into a beamlike ﬁeld as it propagated through the DNG slab.W. 12 and for the p 1. Winn. Schultz. . Marcuvitz. Wittwer and R. 1995 . 43. Meade. Lett.B. The FiniteDifference Time-Domain Method Artech House. Schultz. Sov.B. Electron. 1991 . D. NJ. Lett. CONCLUSIONS We have investigated the propagation of electromagnetic waves in DNG media from both analytical and numerical points of view. A.C. Nor- wood.W. This channeling effect may have many practical considerations and is being investigated further.D. 1986 . 48. 1995 . . 61 2001 .N. 4 D.C. Electromagnetic Wave Theory John Wiley & Sons. Rev. B 58. 3966 2000 . 48. New York. Tomita. One can see that the power ﬂow is positive in the DNG slab and is being concentrated ˆ along the z axis. 11 Advances in Computational Electrodynamics: The FiniteDifference Time-Domain Method. Phys. Antennas Propag. Results for the 1D plane wave scattering from a DNG interface and a DNG slab and the 2D line source cylindrical wave excitation of the DNG slab were presented. Nelson. T. 509 1966 .R. S. Usp.A. 12 J. Antennas Propag. i. The wave physics associated with each of these choices was clariﬁed. 1 V. R10 096 1998 . Kong. Appl. M.C. E 63.0 1011 rad/s case in Fig. Phys. Inc. Ziolkowski. 192 2000 . The lens effect was shown not to exist for any realistic dispersive.. Kawashima. MA. Microwave Theory Tech. show a channeling or paraxial focusing of the wave energy due to the presence of a DNG slab. Sato. Smith and N.R.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue listening from where you left off, or restart the preview.