Home Search Collections Journals About Contact us My IOPscience Dispersive pulse dynamics and associated pulse velocity measures This article has been downloaded from IOPscience. Please scroll down to see the full text article. 2002 J. Opt. A: Pure Appl. Opt. 4 S125 (http://iopscience.iop.org/1464-4258/4/5/359) View the table of contents for this issue, or go to the journal homepage for more Download details: IP Address: 132.198.11.131 The article was downloaded on 23/02/2012 at 18:54 Please note that terms and conditions apply. INSTITUTE OF PHYSICS PUBLISHING JOURNAL OF OPTICS A: PURE AND APPLIED OPTICS J. Opt. A: Pure Appl. Opt. 4 (2002) S125–S134 PII: S1464-4258(02)32839-3 Dispersive pulse dynamics and associated pulse velocity measures Kurt Edmund Oughstun and Natalie Anne Cartwright College of Engineering and Mathematics, University of Vermont, Burlington, VM 05405-0156, USA E-mail: oughstun@emba.uvm.edu and ncartwri@emba.uvm.edu Received 17 January 2002 Published 14 August 2002 Online at stacks.iop.org/JOptA/4/S125 Abstract The classical theory of dispersive signal propagation by Sommerfeld and Brillouin proposed in 1914 provided the first proof within the classical Maxwell–Lorentz theory that an electromagnetic signal could not propagate faster than the vacuum speed of light in a causal dielectric, also introducing the signal velocity and the precursor fields (or forerunners). Modern asymptotic theory has extended these results, providing a physically meaningful definition of the signal velocity while showing the critical role that the precursor fields play in the observed pulse dynamics. Their role in defining the observed pulse velocity is presented and compared with the classical group velocity as well as with other pulse velocity measures that yield superluminal results. Keywords: Electromagnetic pulse propagation, dispersion, ultrashort pulse dynamics, ultrashort pulse velocity measures 1. Introduction Ever since the advent of the special theory of relativity [1] in 1905, the velocity of light in a dispersive, absorptive medium has presented a troublesome dilemma to the electromagnetic and optical physics community. Although Lorentz’s classical model [2] of dielectric dispersion is causal, the group velocity was found to yield noncausal results in regions of anomalous dispersion described by that model. With Sommerfeld’s proof [3] within the classical Maxwell–Lorentz theory that an electromagnetic signal could not propagate faster than the vacuum speed of light in a Lorentz model dielectric, Brillouin [4, 5] then introduced the signal and energy velocities within that medium as a replacement for the group velocity. As was pointed out by Brillouin [5], a group of waves had been originally defined by Rayleigh [6, 7] ‘as moving beats . . . following each other in a regular pattern. A signal is a short isolated succession of wavelets, with the system at rest before the signal arrived and also after it has passed. . . . In general, the signal velocity will differ from the group velocity, especially if the phase velocity is strongly frequency dependent and if the absorption cannot be ignored.’ These early results, including the later work by Baerwald [8], have been borne out and improved upon by the modern asymptotic theory [9–11]. Together with Loudon’s energy velocity [12], a correct physical model of dispersive pulse propagation has been presented [13, 14]. In spite of these results, the group velocity has become one of the most misused concepts in modern physics, engineering and applied mathematics. Because of its inherent simplicity, the group velocity description of dispersive pulse propagation, which is based, in part, on the slowly-varying-envelope approximation, originally introduced by Born and Wolf [15] in the context of partial coherence theory, is widely accepted and employed throughout physics [16] and applied mathematics [17] with central importance in electromagnetics [18], acoustics [19] and optics [20, 21]. Unfortunately, the approximations that are introduced by this description have largely been ignored, resulting in the current controversy regarding superluminal velocities in dispersive materials [22]. The purpose of this paper is to briefly review past and current research as well as to provide a solid theoretical foundation on which to excuse experimental measurements claiming superluminal pulse velocities. The description is based upon the exact Fourier–Laplace integral representation [11] of a linearly polarized, plane wave electromagnetic pulse propagating in the positive z-direction 1464-4258/02/050125+10$30.00 © 2002 IOP Publishing Ltd Printed in the UK S125 with electric and magnetic field vectors 1 E y (z, t ) = f˜(ω) exp{i[k̃(ω)z − ωt]}dω E0 2π C c E 0 n(ω) f˜(ω) exp{i[k̃(ω)z − ωt]}dω Bx (z, t ) = − 2πc C (1) respectively, where E (z, t) = 1̂ y E y (z, t) and B (z, t) = 1̂x Bx (z, t). Here z = z −z 0 denotes the propagation distance into the positive half-space z z 0 from the input plane at z = z 0 where the initial field behaviour E y (z 0 , t ) = E 0 f (t) is specified with electric field strength E 0 . The nondimensional function f (t) describes the initial pulse structure with temporal frequency spectrum ∞ ˜f (ω) = f (t) exp(iωt) dt (2) Real & Imaginary Parts of the Complex Index of Refraction K E Oughstun and N A Cartwright 100 10–1 10–2 1015 1016 1017 1018 −∞ where ω is the angular frequency. The contour C appearing in equation (1) is the Bromwich contour ω = ω + ia, where ω varies from −∞ to +∞ and where the quantity a is greater or equal to the abscissa of absolute convergence for the particular pulse shape considered [11, 23]. The quantity k̃(ω) ≡ β(ω) + iα(ω) = (ω/c)n(ω) (3) denotes the complex wavenumber in the dispersive medium with complex index of refraction n(ω) = n r (ω) + in i (ω) = (µε(ω)/(µ0 ε0 ))1/2 (4) where n r (ω) = Re {n(ω)} is the real index of refraction, n i (ω) = Im {n(ω)} is the imaginary part of the complex index of refraction and β(ω) = Re {k̃(ω)} is the propagation factor and α(ω) = Im {k̃(ω)} the attenuation factor for a timeharmonic plane wave with fixed angular frequency ω in the dispersive material. Here Re {∗} denotes the real part, and Im {∗} denotes the imaginary part of the quantity inside the brackets. The exponential factor appearing in equation (1) may then be rewritten as exp{i[k̃(ω)z − ωt]}= exp[−α(ω)z] exp{i[β(ω)z − ωt]} (5) where the first exponential term on the right-hand side of this expression describes the attenuation and the second term describes the phase change upon propagation. Finally, notice that both cgs and MKS units are employed here through the use of a conversion factor that appears in double brackets ∗ in the equation affected. If this factor is included in the equation it is then in cgs units provided also that ε0 = µ0 = 1, while the equation is in MKS units if this factor is omitted (i.e. replaced by unity). If no such factor appears, then the equation is proper in both systems of units. The analysis is presented for the case of a single-resonance Lorentz model dielectric with (relative) dielectric permittivity ε(ω)/ε0 = 1 − b2 ω2 − ω02 + 2iδω charge and m e the mass of an electron, and where δ is the phenomenological damping constant. With relative magnetic permeability µ/µ0 = 1, the complex index of refraction is given by 1/2 b2 . (7) n(ω) = 1 − 2 ω − ω02 + 2iδω Brillouin’s choice √ of the material parameters (ω0 = 4 × 1016 r s−1 , b = 20 × 1016 r s−1 and δ = 0.28 × 1016 r s−1 ) which describe a highly absorptive material, are used in the numerical examples presented here; however, the results are not restricted to this highly absorptive case. The general results obtained here are also applicable to more general models of the dispersion relation than that given in equation (6), such as that given by a multiple-resonance Lorentz model dielectric. The frequency dispersion of the real and imaginary parts of the complex index of refraction described by equation (7) for Brillouin’s choice of the medium parameters are illustrated in figure 1. The material absorption band isdefined over domain the approximate angular frequency [ ω02 − δ 2 , ω12 − δ 2 ], where ω1 = ω02 + b2 . The material dispersion is normal (n r (ω) increases with increasing frequency) over the approximate angular frequency domains [0, ω02 − δ 2 ) and ( ω12 − δ 2 , ∞) above and below the absorption band, respectively, while it is anomalous (n r (ω) decreases with increasing frequency) over the approximate angular frequency domain [ ω02 − δ 2 , ω12 − δ 2 ] containing the absorption band. The corresponding frequency dispersion of the attenuation and propagation factors α(ω) and β(ω), respectively, are depicted by the solid curves in figure 2. (6) where ω0 is the undamped resonance frequency, b = [(4π/ε0 )N qe2 /m e ]1/2 the plasma frequency with number density N of Lorentz oscillators, where qe denotes the absolute S126 Figure 1. Angular frequency dependence of the real (solid curve) and imaginary (dashed curve) parts of the complex index of refraction for a single-resonance Lorentz model dielectric with undamped resonance frequency ω0 = 4 × 1016 r s−1 , plasma √ frequency b = 20 × 1016 r s−1 and phenomenological damping constant δ = 0.28 × 1016 r s−1 . 2. The phase velocity The phase velocity describes the rate at which the phase fronts appearing in the Fourier–Laplace integral representation of Real & Imaginary Parts of the Complex Wavenumber (r/m) Dispersive pulse dynamics and associated pulse velocity measures frequency of the initial pulse, and the propagation factor is given by β(ω) ≈ (ω/c)n r (ωc ). (9) 109 With the further usual approximation that the attenuation factor is nondispersive so that α(ω) ≈ α(ωc ), the propagated field vectors given in equation (1) may be directly evaluated as 108 E y (z, t) ≈ E 0 f [t − (β(ωc )/ωc )z] exp[−α(ωc )z] c Bx (z, t) ≈ − n r (ωc ) f [t − (β(ωc )/ωc )z] c × exp[−α(ωc )z]. 107 106 105 1015 1016 1017 1018 (10) The electromagnetic pulse then propagates undistorted in shape, but attenuated in amplitude, at the phase velocity v p (ωc ) = ωc /β(ωc ). 3. The group velocity Relative Phase, Group & Energy Velocity Figure 2. Angular frequency dependence of the real (solid curve) and imaginary (dashed curve) parts of the complex wavenumber k̃(ω) = β(ω) + iα(ω) for the single-resonance Lorentz model dielectric considered in figure 1. The dotted curves illustrate the frequency dependence of the linear β (1) (ω) and quadratic β (2) (ω) approximations to the real part of the complex wavenumber about the angular frequency value ωc = 0.75ω0 . 10 0 β (1)(ω) = β(ωc ) + β (ωc )(ω − ωc ) 1016 1017 1018 Figure 3. Angular frequency dependence of the relative phase velocity ν p (ω)/c (dashed curve), relative group velocity νg (ω)/c (alternating dashed-dotted curve) and relative energy velocity ν E (ω)/c (solid curve) for the single-resonance Lorentz model dielectric considered in figure 1. equation (1) propagate through the dispersive medium. The phase velocity for this electromagnetic wave field is obtained from the second term on the right-hand side of equation (5) as v p (ω) = ω . β(ω) (8) The angular frequency dispersion of the relative phase velocity v p /c is illustrated by the dashed curve in figure 3. The phase velocity describes the pulse velocity only in the simplest nonphysical situation when the index of refraction of the medium is nondispersive (excluding the trivial, physically realizable case when the medium is the ideal vacuum). In that case, n(ω) ≈ n(ωc ), where ωc is a characteristic oscillation (11) with α(ω) ≈ α(ωc ), where β (ω) ≡ ∂β(ω)/∂ω. Here ωc describes a characteristic oscillation frequency of the initial pulse; for example, for a pulse envelope modulated sine wave, f (t) = u(t) sin(ωc t + ψ) where u(t) is the envelope function and ψ is a phase constant, ωc = 2π fc is the fixed carrier frequency of the initial pulse. With this substitution, the propagated field vectors given in equation (1) may be directly evaluated as E y (z, t) ≈ E 0 f (t − β (ωc )z) exp[−α(ωc )z] × cos[β(ωc )z − ωc t] c Bx (z, t) ≈ − E 0 f (t − β (ωc )z) exp[−α(ωc )z] c × [n r (ωc ) cos(β(ωc )z − ωc t) − n i (ωc ) sin(β(ωc )z − ωc t)]. 10 –1 1015 A slightly more accurate description of the material dispersion is provided by the linear dispersion approximation (12) The pulse phase then travels with the phase velocity given in equation (8) while the pulse structure propagates undistorted in shape, attenuated with propagation distance, at the group velocity 1 vg (ω) = . (13) β (ω) The angular frequency dispersion of the group velocity is illustrated by the alternating dashed-dotted curve presented in figure 3. Note that the group velocity becomes negative in the region of anomalous dispersion; this negative branch of the group velocity curve is not depicted in this graph. The accuracy of the linear dispersion approximation β (1)(ω) given in equation (11) is illustrated in figure 2 when ωc = 0.75ω0 . As can be seen, this approximation is accurate only over a very small neighbourhood about ωc . The local approximation does improve as ωc is decreased towards zero frequency while it gets worse as ωc is increased into the region of anomalous dispersion. The local approximation also improves as ωc increases above the absorption band. A somewhat more accurate description of the material dispersion is provided by the widely used quadratic dispersion approximation β (2) (ω) = β(ωc )+β (ωc )(ω − ωc )+ 12 β (ωc )(ω − ωc )2 (14) S127 K E Oughstun and N A Cartwright with α(ω) ≈ α(ωc ), where β (ω) ≡ ∂ 2 β(ω)/∂ω2 describes the so-called group velocity dispersion (GVD). With this substitution, the propagated field vectors given in equation (1) become E0 exp[−α(ωc )z] E y (z, t) ≈ [2πβ (ωc )z]1/2 ∞ f (t ) × Re exp[i(β(ωc )z − ωc t + 3π/4)] −∞ (β (ωc )z + t − t) dt × exp −i (15) 2β (ωc )z cE 0 exp[−α(ωc )z] Bx (z, t) ≈ − c[2πβ (ωc )z]1/2 ∞ g(t ) × Re exp[i(β(ωc )z − ωc t + 3π/4)] −∞ (β (ωc )z + t − t) dt × exp −i 2β (ωc )z where ∞ 1 n(ω) f (ω) exp(−iωt) dω. (16) g(t) = 2π −∞ The pulse phase then propagates with the phase velocity v p (ωc ) while the pulse itself propagates with the group velocity vg (ωc ), where the pulse shape is proportional to the Fresnel transform of the initial pulse shape. The propagated pulse structure then depends upon the timescale parameter [24] TF ≡ |2πβ (ωc )z|1/2 (17) which depends upon the value of the GVD and corresponds to the principal Fresnel zone in the analogous slit diffraction problem. If T > 0 denotes the initial temporal pulse width, then for sufficiently small propagation distances z 0, the inequality T > TF will be satisfied and the pulse shape evolves in the same fashion as the diffracted field in the near field of the diffracting slit aperture, while for sufficiently large propagation distance, the inequality T < TF will be satisfied and the pulse shape approaches that given by the Fourier transform of the initial pulse shape in the same manner as the diffracted field in the Fraunhofer region. Similar results are obtained if, instead of just the propagation factor, the complex wavenumber k̃(ω) is approximated by a finite number of terms in its Taylor series expansion about ωc . The quadratic dispersion approximation given in equation (14) is then replaced by k̃ (2) (ω) = k̃(ωc ) + k̃ (ωc )(ω − ωc ) + 12 k̃ (ωc )(ω − ωc )2 . (18) This form of the approximation results in a frequencydependent behaviour in the absorption factor. However, it also results in a complex group velocity ṽg (ω) = 1 k̃ (ω) (19) that is complex valued almost everywhere along the entire real frequency axis, being real valued at ω = 0 and ω = ±∞. The physical meaning of a complex-valued velocity remains to be given. The accuracy of the quadratic dispersion approximation β (2) (ω) given in equation (14) is also illustrated in figure 2 S128 when ωc = 0.75ω0 . It is seen to be accurate only over a slightly larger neighbourhood about ωc than is the linear approximation given in equation (11). It has long been assumed that increased accuracy over a larger frequency domain about the characteristic pulse frequency ωc may be obtained by using higher-order approximations to describe the material dispersion. In 1975 Anderson et al stated in the abstract of their paper [25] that ‘the evolution of slowly varying wave pulses in strongly dispersive and absorptive media is studied by a recursive method. It is shown that the resulting envelope function may be obtained by including correction terms of arbitrary dispersive and absorptive orders’. This sentiment is extended in the 1990 text by Butcher and Cotter [26] which states that ‘to describe pulse propagation in dispersive media, in general, we must retain the secondorder dispersion, and for ultrashort pulses or those with a wide frequency spectrum it may sometimes be necessary to also include higher-order terms’. Continuing on in this tradition, Akhmanov et al state in their 1992 text [27] that ‘one can analyse how the dispersion of a medium affects a propagating pulse for any higher-order approximation of the dispersion theory. Naturally, the higher-order approximations make the quantitative picture of dispersive spreading more precise although its basic features obtained for the second- and thirdorder approximations remain unchanged.’ Unfortunately, these assertions, although physically appealing, are not valid in the ultrashort pulse, ultrawideband signal regime. As first proved by Oughstun and Xiao [28, 29], ‘with the exception of a small neighbourhood about some characteristic frequency of the initial pulse, the inclusion of higherorder terms in the Taylor series approximation of the complex wavenumber in a causally dispersive, attenuative medium beyond the quadratic approximation is practically meaningless from both the physical and mathematical points of view’. 4. The energy velocity A quantity of fundamental importance to both the analysis and interpretation of propagation phenomena in a causally dispersive medium is the velocity of energy transport (or the energy velocity) of a monochromatic field. This physical velocity is defined as the ratio of the time-average value of the Poynting vector to the total time-average electromagnetic energy density stored in both the field and the medium. The original derivation of this quantity for a single-resonance Lorentz model dielectric by Brillouin [4, 5] neglected to include that portion of the electromagnetic energy that is stored in the excited Lorentz oscillators of the medium and consequently was in error. Loudon [12] was the first to derive a correct expression for the energy velocity in a single-resonance Lorentz medium. This momentous research would later prove to be critical in providing a correct physical description of dispersive pulse dynamics. Consider a multiple-resonance Lorentz model dielectric with complex index of refraction n(ω) = 1 − j b2j ω2 − ω2j + 2iδ j ω 1/2 (20) Dispersive pulse dynamics and associated pulse velocity measures where each separate resonance line is described by the parameter triple (ω j , δ j , b j ) of undamped resonance frequency, phenomenological damping constant and plasma frequency, respectively, with summation index j extending over the set of resonance lines present in the dielectric material. The time-average value of the electromagnetic energy density stored in the Lorentz oscillators is then found to be given by [11, 30] 1 ε0 b2j (ω2 + ω2j ) 2 Uosc = 4π 4 |E| (ω2 − ω2 )2 + 4δ 2 ω2 j j (21) j where ω is the angular frequency of oscillation of the timeharmonic plane wave field with electric field strength E. The time-average value of the energy density stored in the timeharmonic plane wave electromagnetic field is given by 1 ε0 2 2 2 (22) U f = 4π 4 (n r (ω) + n i (ω) + 1)|E| . With use of the identity [11] n r2 (ω) − n 2i (ω) = 1 − b2j (ω2 − ω2j ) j (ω2 − ω2j )2 + 4δ 2j ω2 (23) the total time-average electromagnetic energy density stored in the coupled field–medium system is found to be U = U f + Uosc 1 ε0 b2j ω2 |E|2 n 2 + . = r 4π 2 2 2 2 2 2 j (ω − ω j ) + 4δ j ω (24) The time-average value of the magnitude of the Poynting vector for the monochromatic plane wave field is given by [11] 2 c 1 2 (25) |S| = 4π 2µ c n r (ω)|E| . 0 The time-average velocity of energy transport in a multiple resonance Lorentz model dielectric is then given by the ratio of these two quantities, so that [11, 30] v E (ω) ≡ |S| = U n r (ω) + 1 nr (ω) c b2j ω2 j (ω2 −ω2j )2 +4δ 2j ω2 . (26) c n r (ω) + ωn i (ω)/δ 5. The signal velocity The signal velocity in a Lorentz model dielectric was originally introduced by Brillouin [4, 5] and its description was later improved upon by Baerwald [8]. Its original physical interpretation has been criticized by Smith [33] as being impractical from an experimental point of view since ‘the difficulty of arriving at a workable definition for the signal velocity is that a pulse of radiation is not a point, i.e. the motion of the pulse cannot be equated to the motion of a point associated with the pulse’. This criticism is not surprising since the original definition of the signal velocity is based upon the intricacies of the asymptotic method of steepest descent, the signal arrival being defined by the instant the path of steepest descent crossed the simple pole singularity appearing in the spectrum of the initial signal. The modern asymptotic theory [9–11] has redefined the signal velocity in light of Olver’s saddle point method [34] which proved that the path of steepest descent was irrelevant in determining the signal velocity. The signal velocity is defined for the fundamental canonical problem of the step function modulated signal f (t) = u(t) sin(ωc t) (28) where ωc = 2π fc is a fixed carrier frequency and where the envelope function u(t) is the Heaviside unit step function (u(t) = 0 for t < 0 and u(t) = 1 for t > 0). With this substitution in equation (1), the integral representation for the propagated signal becomes E0 Re ũ(ω − ωc ) exp[(z/c)φ(ω, θ )] dω E y (z, t) = 2π C E0 n(ω − ωc )ũ(ω − ωc ) (29) Bx (z, t) = −c Re 2πc C × exp[(z/c)φ(ω, θ )] dω with complex phase function This result directly reduces to the expression v E (ω) = in v E (ω) corresponds to the angular frequency region where the time-average electromagnetic energy density Uosc stored in the Lorentz oscillators is near its local maximum value. This linear result then complements the nonlinear result of minimal propagation velocity observed in self-induced transparency [31, 32]. (27) given by Loudon [12] for the case of a single-resonance Lorentz model dielectric. The frequency dependence of the energy velocity in a single-resonance dielectric is illustrated by the solid curve presented in figure 3. Notice first that v E (ω) c for all real ω with v E → c as ω → ∞. The energy velocity is seen to attain a minimum value just above the resonance frequency ω0 near to the frequency value where n i (ω) attains its maximum value, and remains small through the region of anomalous dispersion. Naturally, this occurs over each absorption band in a dielectric material described by multiple resonance frequencies. Each minimum region φ(ω, θ ) ≡ i(c/z)(k̃(ω)z − ωt) = iω(n(ω) − θ) (30) where θ = ct/z is a nondimensional space-time value. Although this analysis of the signal velocity is presented in terms of the Heaviside unit step function signal, the expressions appearing in equation (29) remain valid for any pulse that can be expressed in the form given in equation (28) with fixed carrier frequency ωc . If the initial time behaviour E(z 0 , t ) = E 0 f (t) of the field at the plane z = z 0 is zero for all time t < 0 and if the model of the material dispersion is causal, then the propagated field (29) identically vanishes for all θ < 1 with z > 0, so that f (t) = 0 for t < 0 ⇒ E y (z, t ) = Bx (z, t ) = 0 for all θ < 1. (31) S129 K E Oughstun and N A Cartwright A proof of this important result was first given by Sommerfeld [3] for the Heaviside unit step function signal (28) in a single-resonance Lorentz model dielectric, and was later extended [9, 11] to an arbitrary plane wave pulse. The dynamical field evolution then occurs for θ 1 and asymptotic methods of analysis are typically required as z → ∞. The first step in the asymptotic analysis of the integral representation (29) of the propagated field for θ 1 is to determine the set of saddle points of the complex phase function where dφ/dω = 0, so that n(ω) + ωn (ω) − θ = 0. (32) The roots of this saddle point equation then give the saddle point locations in the complex ω-plane. Since the saddle point equation depends upon the space-time parameter θ = ct/z, the saddle points will then evolve with time at any fixed propagation distance z. Because of the symmetry relations [11] n(−ω) = n ∗ (ω∗ ) and φ(−ω, θ ) = φ ∗ (ω∗ , θ) satisfied by a causal medium, if ω j (θ ) is a saddle point, then so also is −ω∗j (θ ). The physical significance of the saddle points can be appreciated through consideration of the defining relation (z/c)φ(ω, θ ) = i(k̃(ω)z − ωt). Upon differentiating this expression with respect to ω one obtains (z/c)φ (ω, θ ) = i((∂ k̃(ω)/∂ω)z − t). Since φ (ω, θ ) = 0 at the saddle points of the complex phase function, then z/t = (∂ k̃(ω)/∂ω)−1 ω=ω j = ṽg (ω j ) and the complex group velocity is found to be real valued at the saddle points. With the saddle point locations known for θ 1, the next step in the analysis is to express the integral representation (29) in terms of an integral I (z, θ ) with the same integrand but with a new contour of integration P(θ ) to which the original contour C may be deformed [9–11]. By Cauchy’s residue theorem, the integral representation (29) of E y (z, t ) and the contour integral I (z, θ ) are related by E y (z, t ) = I (z, θ ) − Re {2πi(θ )} (33) where 1 ũ(ω − ωc ) exp[(z/c)φ(ω, θ )] ω=ω p 2π p (34) is the sum of the residues of the poles that were crossed in the deformation from C to P(θ ), and where 1 ũ(ω − ωc ) exp[(z/c)φ(ω, θ )] dω. I (z, θ ) = 2π P(θ ) (35) Similar expressions are obtained for the associated magnetic field component. For the asymptotic evaluation of the contour integral I (z, θ ) as z → ∞, the path P(θ ) is taken as a union of Olver-type paths [9–11] with respect to a subset of the set of saddle points of φ(ω, θ ) such that P(θ ) evolves continuously for all θ 1. Not all saddle points in this set may be appropriate in the asymptotic description because the Olver-type paths with respect to them may not be deformable to the original contour C owing, for example, to the presence of the branch cuts of φ(ω, θ); such saddle points are said to be inaccessible, otherwise they are said (θ ) = S130 Res to be accessible. The dominant accessible saddle point (or points) refers to the saddle point (or points) that has the largest value of Re {φ(ω, θ )} at it, and hence, has the least exponential attenuation associated with it. In comparison, Brillouin’s interpretation [4, 5] of this asymptotic method required that the contour of integration C be deformed so that it lay along the entire path of steepest descent through the accessible saddle points of the complex phase function. Olver’s theorem [34] proved this requirement unnecessary with important consequences regarding the physical significance of whether or not a particular pole singularity is crossed in deforming the contour C to P(θ ). If ω j (θ ) and −ω∗j (θ ) are the dominant accessible firstorder saddle points at a particular value of θ and if they are isolated from each other as well as from all other saddle points of the complex phase function φ(ω, θ ) at that value of θ , then the nonuniform asymptotic approximation of I (z, θ ) as z → ∞ is obtained from Olver’s theorem [34] as [9, 11] 1/2 c ũ(ω j − ωc ) I (z, θ ) ∼ Re − 2πzφ (2) (ω j , θ ) × exp[(z/c)φ(ω j , θ )] 1/2 c + − ũ(ω∗j − ωc ) 2πzφ (2) (−ω∗j , θ ) (36) × exp[(z/c)φ(ω∗j , θ )] . The dynamical evolution of the saddle points then provides a nearly complete description of the dynamical evolution of the transient field behaviour associated with dispersive pulse propagation. The residue contribution (34) is nonzero only if ũ (ω − ωc ) has poles. For the Heaviside unit step function signal (θ ) = 0 for θ < θ S (θ ) = exp(−α(ωc )z) sin(β(ωc )z − ωc t) for θ > θ S (37) where θ S denotes the space-time point at which the Olver-type path P(θ ) crosses the simple pole singularity at ω = ωc . This contribution to the asymptotic behaviour of the propagated field describes the steady-state behaviour of the signal. The arrival of this signal contribution is determined by the dynamics of that dominant saddle point that becomes exponentially negligible in comparison to this pole contribution. In a single-resonance Lorentz model dielectric the asymptotic theory [9–11] shows that the propagated field described in equation (29) may be expressed in the form E y (z, t ) = A S (z, t ) + A B (z, t ) + Ac (z, t) (38) as z → ∞, with a similar expression for Bx (z, t ). The asymptotic behaviour of the component field A S (z, t ) is due to the pair of distant saddle points [9, 11] ∼ ω± S PD (θ ) = ±ξ(θ ) − iδ(1 + η(θ )) (39) with ξ(θ ) = (ω02 − δ 2 + b2 θ 2 /(θ 2 − 1))1/2 and η(θ ) = (δ 2 /27 + b2 /(θ 2 − 1))/ξ 2 (θ ), and is referred to as the first or Sommerfeld precursor field. The front of the Sommerfeld precursor arrives at θ = 1 with zero amplitude and an infinite instantaneous angular frequency. As θ increases from Dispersive pulse dynamics and associated pulse velocity measures unity the amplitude rapidly builds to a maximum value and thereafter decays as the attenuation factor increases and the instantaneous oscillation frequency chirps downward towards the upper frequency edge of the material absorption band. The asymptotic behaviour of the component field A B (z, t ) is due to the near saddle points [9, 11] ω+S PN (θ ) ∼ = i(|ψ(θ )| − (2/3)δζ (θ )) (40a) 0.15 (a) 0.1 0.05 0 – 0.05 for 1 < θ θ1 , while for θ θ1 , ∼ ω± S PN (θ ) = ±ψ(θ ) − (2/3)iδζ (θ ) (40b) where ζ (θ ) = (3/2)ς(θ ) and ψ(θ ) = (ω02 (θ 2 − θ02 )/(θ 2 − θ02 + 3αb2 /ω02 )−δ 2 ς 2 (θ ))1/2 with ς(θ ) = (θ 2 −θ02 +2b2 /ω02 )/(θ 2 − θ02 + 3αb2 /ω02 ), and is referred to as the second or Brillouin precursor field. Here – 0.1 4 0.04 5 6 7 8 9 t (s) 10 11 12 13 14 x10–15 (b) 0.03 θ0 = n(0) = (1 + b2 /ω02 )1/2 (41) θ1 ∼ = θ0 + 2δ 2 b2 /(θ0 ω02 (3αω02 − 4δ 2 )) 0.01 with α = 1 − δ 2 (4ω12 + b2 )/(3ω02 ω12 ). As the near saddle point ω+S PN (θ ) moves down the imaginary axis for 1 < θ θ1 , as described by equation (40a), the Brillouin precursor is nonoscillatory over this space-time domain and reaches a peak amplitude near the space-time point θ = θ0 where there is no exponential attenuation. As θ increases above θ1 the Brillouin precursor becomes oscillatory with an instantaneous angular frequency that chirps upward towards the lower frequency edge of the absorption band with decreasing amplitude as the attenuation factor monotonically increases. The final contribution Ac (z, t ) appearing in equation (38) is due to the simple pole singularity given by equation (37). This contribution to the asymptotic behaviour of the propagated field describes the steady state behaviour of the propagated signal that oscillates at the input carrier frequency ω = ωc . The Sommerfeld precursor arrives at the vacuum speed of light c and dominates the initial evolution of the propagated field. The Brillouin precursor becomes exponentially dominant over the Sommerfeld precursor at the space-time point θ = θ S B where [9, 11] θ S B ≈ θ0 − 4δ 2 b2 3θ0 ω04 (42) at which point the oscillation frequency of the Sommerfeld precursor has decreased to the value [9, 11] ωS B ∼ = ξ(θ S B ) ≈ ω0 2 1/2 b2 5δ 2+ 2 + ω0 3ω02 0.02 . (43) The dynamical evolution of the propagated signal due to an input Heaviside unit step function signal then separates into two cases depending upon the value of the carrier frequency ωc in comparison with the value ω S B that is a characteristic of the dispersive material. For 0 < ωc < ω S B the dynamical field evolution is first dominated by the Sommerfeld precursor over the space-time domain 1 θ < θ S B , is next dominated by the Brillouin precursor over the space-time domain θ S B < θ < θc and is finally dominated by the pole contribution for all θ > θc , as 0 – 0.01 – 0.02 – 0.03 – 0.04 5 6 7 8 t (s) 9 10 11 x10–15 Figure 4. Dynamical field evolution due to an input unit step function modulated signal with (a) below the resonance carrier frequency (ωc = ω0 /2) at five absorption depths (z = 5z d ), and (b) above the resonance carrier frequency (ωc = 2.5ω0 > ω S B ) at five absorption depths (z = 5z d ) in the single-resonance Lorentz model dielectric considered in figure 1. illustrated in figure 4(a), where θc denotes the space-time value at which the exponential attenuation of the Brillouin precursor first equals and thereafter remains greater than that at the carrier frequency ωc . The propagated signal is then characterized by several distinct features: the arrival of the Sommerfeld precursor front at θ = 1 which travels at the velocity v S = c, the peak of the Brillouin precursor at θ = θ0 which travels at the velocity v B = c/θ0 = c/n(0), and the arrival of the main signal at θ = θc which travels at the main signal velocity vc (ωc ) = c θc (44) where 0 < vc v B . For ωc > ω S B the dynamical field evolution is first dominated by the Sommerfeld precursor over the space-time domain 1 θ < θc1 , by the pole contribution over the spacetime domain θc1 < θ < θc2 , by the Brillouin precursor over the space-time domain θc2 < θ < θc , and is finally dominated by the pole contribution for all θ > θc , as illustrated in figure 4(b). Here θc1 denotes the space-time value at which the exponential attenuation of the Sommerfeld precursor first equals and thereafter remains greater than that at the carrier frequency ωc , and θc2 denotes the space-time value at which the exponential attenuation of the pole contribution first equals and S131 K E Oughstun and N A Cartwright intensity centroid of the pulse. A variant of this velocity measure which is defined in terms of the Poynting vector of the pulse has recently been introduced by Peatross, Glasgow and Ware [35, 36]. The temporal centre of the Poynting vector S (z, t ) of a plane wave pulse propagating in the positive z direction is defined as ∞ ∞ t S (z, t) dt ẑ × S (z, t) dt (47) tz ≡ ẑ × Relative Signal & Energy Velocities 1 SB 0.9 0.8 0.7 0.6 0.5 0.4 −∞ 0.3 0.2 0.1 0 0 5 10 15 −∞ where ẑ denotes the unit vector along the propagation direction. If tz 0 denotes the temporal centroid of the Poynting vector of the initial pulse at the input plane at z = z 0 , then the pulse centro-velocity is given by [35] vcentro = Figure 5. Angular frequency dependence of the relative energy velocity ν E (ωc )/c (solid curve) and the relative anterior presignal velocity νc1 (ωc )/c, posterior presignal velocity νc2 (ωc )/c, and main signal velocity νc (ωc )/c branches (open circles) for the single-resonance Lorentz model dielectric considered in figure 1. then becomes greater than that for the Brillouin precursor for a finite θ interval. The propagated signal is now characterized by several distinct features: the arrival of the Sommerfeld precursor front at θ = 1 which travels at the velocity v S = c, the first arrival of the signal at θ = θc1 which travels at the anterior presignal velocity c vc1 (ωc ) = (45) θc1 where c > vc1 > c/θ S B , the arrival of the Brillouin precursor at θ = θc2 which travels at the posterior presignal velocity c vc2 (ωc ) = (46) θc2 where c/θ S B > vc2 > c/θ0 , and finally the arrival of the main signal which travels at the main signal velocity given by equation (44). The frequency dependence of the signal velocity in a single-resonance Lorentz model dielectric is illustrated in figure 5. Each sequence of open circles describes a separate branch of the signal velocity obtained from a series of numerical calculations given in table 9.1 of [11]: the anterior presignal velocity vc1 (ωc ), the posterior presignal velocity vc2 (ωc ) and the main signal velocity vc (ωc ). The solid curve in the figure describes Loudon’s energy velocity v E (ωc ), which is seen to form an upper envelope to the signal velocity values [9, 11]. Based upon this important result, a new physical description of dispersive pulse dynamics has been given [11, 13, 14] in terms of the energy velocity and attenuation of a time-harmonic plane wave in the causally dispersive dielectric. This new description accurately describes all of the features that occur in the evolution of an ultrawideband pulse as it propagates through a dispersive material, including the precursor fields, and reduces to the group velocity description in the limit of zero loss. 6. The pulse centroid velocity In 1970 Smith [33] proposed a new pulse velocity measure called the centro-velocity which described the motion of the S132 z − z0 . tz − tz 0 (48) The integrals appearing in equation (47) may be transformed into the angular frequency domain upon application of the Parseval–Plancherel theorem with the result that [35] tz = T {Ẽ (z, ω)} ∞ ≡ − i ẑ × ∂ Ẽ (z, ω)/∂ω × H̃ ∗ (z, ω) dω × ẑ × −∞ ∞ −∞ S̃ (z, ω) dω −1 . (49) The Poynting vector centroid delay may then be expressed in the form (50) tz − tz 0 = G z + Rz 0 where G z ≡ (∂β(ω)/∂ω)z ∞ S̃(z, ω)(∂β(ω)/∂ω) dω = z −∞ ∞ −∞ S̃(z, ω) dω (51) is the group delay energy centroid, and where Rz 0 ≡ T {Ẽ (z 0 , ω) exp(−α(ω)z)} − T {Ẽ (z 0 , ω)} (52) is called the pulse reshaping delay. As stated in the abstract of their paper [35], Peatross, Glasgow and Ware state that this ‘result provides a context wherein group velocity is always meaningful even for broad band pulses and when the group velocity is superluminal or negative. The result imposes superluminality on sharply defined pulses’. The asymptotic representation of the propagated field given in equation (38) shows that, as an ultrawideband pulse propagates into a Lorentz model dielectric, the field becomes dominated by the Brillouin precursor whose peak amplitude travels at the velocity v B = c/θ0 = c/n(0). (53) Hence, as the propagation distance increases, the pulse centrovelocity asymptotically approaches this characteristic velocity of the Brillouin precursor, so that lim vcentro = v B . z→∞ (54) The evolution of the centro-velocity for an input ten-cycle rectangular envelope pulse for a wide range of input pulse Dispersive pulse dynamics and associated pulse velocity measures 1 relative centrovelocity 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0 1 2 3 4 5 6 7 8 9 10 Figure 6. Behaviour of the relative centro-velocity νcentro (ωc )/c with relative propagation distance z/z d for an input ten-cycle rectangular envelope pulse for several values of the input pulse carrier frequency ωc in the single-resonance Lorentz model dielectric considered in figure 1. frequencies in a single-resonance Lorentz model dielectric is illustrated by the family of curves in figure 6. Each open circle in the figure for z/z d 1 describes a data point for the centro-velocity determined from a numerical experiment that was based upon two independent techniques (an asymptotic code and an FFT-based code). Each asterisk for z/z d = 0 describes the relative group velocity value vg (ωc )/c at that particular frequency. For each carrier frequency considered, the centro-velocity asymptotically approaches the subluminal limiting value v B /c = 1/θ0 = 0.667 set by the peak velocity of the Brillouin precursor as the propagation distance increases in the mature dispersion regime (typically z/z d 1 outside of the absorption band). Superluminal centroid velocity results are only observed for carrier frequency values within the absorption band for sufficiently small propagation distances. 7. Discussion The velocity of an electromagnetic pulse as it propagates through a causally dispersive material is a fundamental problem in physics that has remained unresolved for nearly a century. Nevertheless, a significant body of published research has now been established that shows that: (1) any study of the pulse velocity of an ultrawideband pulse in a dispersive material must give careful consideration to the material dispersion, including the attenuation, over the entire frequency domain, and (2) any conclusions regarding either ultrawideband or ultrashort pulse dynamics, including the pulse velocity, that are based upon the quadratic or any higher-order dispersion approximation should be viewed with extreme scepticism for propagation distances that exceed a single absorption depth (at some characteristic frequency of the initial pulse) in the material. The detailed results presented here show that the only physically meaningful velocity measures for dispersive pulse propagation are the energy velocity, signal velocity and the pulse centro-velocity. Although the energy velocity has only been derived for a monochromatic plane wave in a Lorentz model dielectric, it is an essential ingredient in the new physical description of dispersive pulse dynamics [11, 13, 14]. This description of the pulse evolution in terms of the attenuation and energy velocity of each spectral component present in the initial pulse spectrum provides a detailed representation of each feature present in the propagated pulse, including the precursor fields. The signal velocity describes the point at which the propagated pulse begins to predominantly oscillate at the input carrier frequency of the pulse [37]; consequently, it can be measured in a carefully defined laboratory experiment. This causal velocity measure is fundamental to the problem of signal (and hence information) propagation through a dispersive material or system. Claims of superluminal pulse propagation [38–41] are primarily based upon the group velocity approximation which has been proven [28, 29] to be invalid in the mature dispersion limit that is typically achieved when the propagation distance exceeds a single absorption depth in the dispersive material. Superluminal peak velocities can briefly occur for such temporally unbounded pulse envelope functions as the Gaussian envelope function [42–44]. For such an ultrashort Gaussian envelope pulse, the peak velocity is found [43, 44] to move along the group velocity curve in such a manner that as the propagation distance increases, the peak velocity approaches the energy velocity [43, 44]. Although the peak velocity may briefly exceed the vacuum speed of light (or may even be negative) during the initial evolution of the pulse, this is primarily due to pulse reshaping and does not entail the transmission of either energy or information superluminally [22, 45]. That any energy or information is not transmitted superluminally follows immediately through consideration of a unit step function modulated Gaussian pulse. In that case, the exact result expressed in equation (31) directly applies and no information can have been transmitted faster than the vacuum speed of light. Furthermore, as pointed out by Landauer [22], ‘an emerging peak is not necessarily related to the incident peak in a causative physical way’. Such is the case in ultrashort Gaussian pulse propagation [44] where the oscillation frequency at the peak amplitude point changes continuously as the pulse evolves. Finally, the pulse centro-velocity recently introduced by Peatross, Glasgow and Ware [35] provides a convenient measure of the pulse velocity for temporal pulses with a sharply defined envelope function. However, their assertion that superluminality is imposed on sharply defined pulses requires further careful study. The delta function pulse is certainly sharply defined, yet its propagated field identically vanishes over the entire space-time domain ct/z < 1. The same result holds for a rectangular envelope pulse of arbitrarily short initial temporal width [46]. These results then establish that: (1) superluminal peak velocities in causally dispersive materials can momentarily occur for sufficiently small propagation distances in the immature dispersion regime when ω0 ωc ω1 , (2) however, as the propagation distance increases into the mature dispersion regime, the pulse centro-velocity for an ultrashort pulse becomes subluminal and approaches the subluminal velocity of the dominant precursor field. As S133 K E Oughstun and N A Cartwright tempting as it may seem, pulse reshaping does not imply either superluminal energy or superluminal information transmission. In the light of these results, (3) superluminal energy and information transfer is not physically possible within the framework of the Maxwell– Lorentz theory in linear, causally dispersive systems. Acknowledgments The research presented in this paper has been supported, in part, by the United States Air Force Office of Scientific Research under grant no 49620-01-1-0306. The first author (KEO) would like to express his gratitude to Professor JeanJacques Greffet for his successful effort in organizing the highly productive meeting on Electromagnetic Optics in Paris this past summer that resulted in the preparation of this paper. It was my great pleasure to finally meet both him and Professor Rodney Loudon during this meeting. References [1] Einstein A 1905 Zur Elektrodynamik Bewegter Körper Ann. Phys., Lpz. 17 891–921 [2] Lorentz H A 1906 Versuch einer Theorie der Electrischen und Optischen Erscheinungen in Bewegten Körpern (Leipzig: Teubner) see also Lorentz H A 1909 The Theory of Electrons (Leipzig: Teubner) [3] Sommerfeld A 1914 Über die Fortpflanzung des Lichtes in disperdierenden Medien Ann. Phys., Lpz. 44 177–202 [4] Brillouin L 1914 Über die Fortpflanzung des Licht in disperdierenden Medien Ann. Phys., Lpz. 44 204–40 [5] Brillouin L 1960 Wave Propagation and Group Velocity (New York: Academic) [6] Lord Rayleigh 1877 On progressive waves Proc. London Math. Soc. IX 21–6 [7] Lord Rayleigh 1881 On the velocity of light Nature XXIV 52–5 [8] Baerwald H G 1930 Über die fortpflanzung von signalen in dispergierenden systemen Ann. Phys. 7 731–60 [9] Oughstun K E and Sherman G C 1988 Propagation of electromagnetic pulses in a linear dispersive medium with absorption (the Lorentz medium) J. Opt. Soc. Am. B 5 817–49 [10] Oughstun K E and Sherman G C 1989 Uniform asymptotic description of electromagnetic pulse propagation in a linear dispersive medium with absorption (the Lorentz medium) J. Opt. Soc. Am. A 6 1394–420 [11] Oughstun K E and Sherman G C 1994 Electromagnetic Pulse Propagation in Causal Dielectrics (Berlin: Springer-Verlag) [12] Loudon R 1970 The propagation of electromagnetic energy through an absorbing dielectric J. Phys. A: Math. Gen. 3 233–45 [13] Sherman G C and Oughstun K E 1981 Description of pulse dynamics in Lorentz media in terms of the energy velocity and attenuation of time-harmonic waves Phys. Rev. Lett. 47 1451–4 [14] Sherman G C and Oughstun K E 1995 Energy velocity description of pulse propagation in absorbing, dispersive dielectrics J. Opt. Soc. Am. B 12 229–47 [15] Born M and Wolf E 1959 Principles of Optics (Oxford: Pergamon) [16] Eckart C 1948 The approximate solution of one-dimensional wave equations Rev. Mod. Phys. 20 399–417 [17] Whitham G B 1974 Linear and Nonlinear Waves (New York: Wiley) [18] Jackson J D 1975 Classical Electrodynamics (New York: Wiley) S134 [19] Lindsay R B 1960 Mechanical Radiation (New York: McGraw-Hill) [20] Bloembergen N 1965 NonlinearOptics (New York: Benjamin) [21] Akhmanov S A, Vysloukh V A and Chirkin A S 1992 Optics of Femtosecond Laser Pulses (New York: AIP) [22] Landauer R 1993 Light faster than light? Nature 365 692–3 [23] Stratton J A 1941 Electromagnetic Theory (New York: McGraw-Hill) [24] Jones J 1974 On the propagation of a pulse through a dispersive medium Am. J. Phys. 42 43–6 [25] Anderson D, Askne J and Lisak M 1975 Wave packets in an absorptive and strongly dispersive medium Phys. Rev. A 12 1546–52 [26] Butcher P N and Cotter D 1990 The Elements of Nonlinear Optics (Cambridge: Cambridge University Press) ch 2 [27] Akhmanov S A, Vysloukh V A and Chirkin A S 1992 Optics of Femtosecond Laser Pulses (New York: AIP) ch 1 [28] Oughstun K E and Xiao H 1997 Failure of the quasimonochromatic approximation for ultrashort pulse propagation in a dispersive, attenuative medium Phys. Rev. Lett. 78 642–5 [29] Xiao H and Oughstun K E 1999 Failure of the group velocity description for ultrawideband pulse propagation in a double resonance Lorentz model dielectric J. Opt. Soc. Am. B 16 1773–85 [30] Oughstun K E and Shen S 1988 Velocity of energy transport for a time-harmonic field in a multiple-resonance Lorentz medium J. Opt. Soc. Am. B 5 2395–8 [31] McCall S L and Hahn E L 1969 Self-induced transparency Phys. Rev. 183 457–85 [32] Allen L and Eberly J H 1975 Optical Resonance and Two-Level Atoms (New York: Wiley) ch 4 [33] Smith R L 1970 The velocities of light Am. J. Phys. 38 978–84 [34] Olver F W J 1970 Why steepest descents? SIAM Rev. 12 228–47 [35] Peatross J, Glasgow S A and Ware M 2000 Average energy flow of optical pulses in dispersive media Phys. Rev. Lett. 84 2370–3 [36] Ware M, Glasgow S A and Peatross J 2001 Role of group velocity in tracking field energy in linear dielectrics Opt. Express 9 506–18 [37] Oughstun K E, Wyns P and Foty D P 1989 Numerical determination of the signal velocity in dispersive pulse propagation J. Opt. Soc. Am. A 6 1430–40 [38] Chu S and Wong S 1982 Linear pulse propagation in an absorbing medium Phys. Rev. Lett. 48 738–41 [39] Laude V and Tournois P 1999 Superluminal asymptotic tunneling times through one-dimensional photonic bandgaps in quarter-wave-stack dielectric mirrors J. Opt. Soc. Am. B 16 194–8 [40] Milonni P W, Furuya K and Chiao R Y 2000 Quantum theory of superluminal pulse propagation Opt. Express 8 59–65 [41] Dogariu A, Kuzmich A, Cao H and Wang L J 2001 Superluminal light pulse propagation via rephasing in a transparent anomalously dispersive medium Opt. Express 8 344–50 [42] Garrett C G B and McCumber D E 1970 Propagation of a gaussian light pulse through an anomalous dispersion medium Phys. Rev. A 1 305–13 [43] Oughstun K E and Balictsis C M 1996 Gaussian pulse propagation in a dispersive, absorbing dielectric Phys. Rev. Lett. 77 2210–13 [44] Balictsis C M and Oughstun K E 1997 Generalized asymptotic description of the propagated field dynamics in gaussian pulse propagation in a linear, causally dispersive medium Phys. Rev. E 55 1910–21 [45] Diener G 1996 Superluminal group velocities and information transfer Phys. Lett. A 223 327–31 [46] Oughstun K E and Sherman G C 1990 Uniform asymptotic description of ultrashort rectangular optical pulse propagation in a linear, causally dispersive medium Phys. Rev. A 41 6090–113