LIMITATIONS ON THE PRINCIPLE OF STATIONARY PHASE WHEN IT IS APPLIED TO TUNNELING ANALYSIS ∗ A. E. Bernardini Instituto de F´ısica Gleb Wataghin, UNICAMP, PO Box 6165, 13083-970, Campinas, SP, Brasil. (Dated: February 1, 2008) 7 0 Abstract 0 2 Using a recently developed procedure - multiple wave packet decomposition - here we study n a J the phase time formulation for tunneling/reflecting particles colliding with a potential barrier. 2 To partially overcome the analytical difficulties which frequently arise when the stationary phase 1 v method is employed for deriving phase (tunneling) time expressions, we present a theoretical exer- 0 1 cise involving a symmetrical collision between two identical wave packets and an one-dimensional 0 1 rectangular potential barrier. Summing the amplitudes of the reflected and transmitted waves 0 7 - using a method we call multiple peak decomposition - is shown to allow reconstruction of the 0 / h scattered wave packets in a way which allows the stationary phase principle to be recovered. p - t n PACS numbers: 03.65.Xp a u q : v i X r a ∗ Electronic address: alexeb@ifi.unicamp.br 1 Recently, a series of experimental results [1, 2, 3, 4], some of them confirming the pos- sibility of superluminal tunneling speeds for photons, have revived an interest in the tun- neling time analysis [5, 6, 7, 8, 9]. On the theoretical front, people have tried to introduce quantities that have the dimension of time and can somehow be associated with the pas- sage of the particle through the barrier or, strictly speaking, with the definition of the tunneling time. These proposals have led to the introduction of several time definitions [5, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], some of which can be organized into three groups. (1) The first group comprises a time-dependent description in terms of wave pack- ets where some features ofanincident packet andthe comparable featuresof thetransmitted packet are utilized to describe a quantifiable delay as a tunneling time [9]. (2) In the sec- ond group the tunneling times are computed based on averages over a set of kinematical paths, whose distribution is supposed to describe the particle motion inside a barrier. In this case, Feynman paths are used like real paths to calculate an average tunneling time with the weighting function exp[iSx(t)/~], where S is the action associated with the path x(t) (where x(t) represents the Feynman paths initiated from a point on the left of the barrier and ending at another point on the right of it [21]). The Wigner distribution paths [16], and the Bohm approach [22, 23] are included in this group. (3) In the third group we notice the introduction of a new degree of freedom, constituting a physical clock for the measurements of tunneling times. This group comprises the methods with a Larmor clock [11] or an oscillating barrier [24]. Separately, standing on itself is the dwell time defined by the interval during which the incident flux has to exist and act, to provide the expected accumulated particle storage, inside the barrier [7]. There is no general agreement [5, 8] among the above definitions about the meaning of tunneling times (some of the proposed tunneling times are actually traversal times, while others seem to represent in reality only the spread of their distributions) and about which, if any, of them is the proper tunneling time [5]. In the context where we intend to work on, the tunneling mechanism is embedded by theoretical constructions involving analytically- continuous gaussian, or infinite-bandwidth step pulses to examine the tunneling process. Nevertheless, such holomorphic functions do not have a well-defined front in a manner that the interpretation of the wave packet speed of propagation becomes ambiguous. Moreover, infinite-bandwidthsignalscannotpropagatethroughanyrealphysical medium(whosetrans- fer function is therefore finite) without pulse distortion, which also leads to ambiguities in 2 determining the propagation velocity during the tunneling process. For instance, some of the barrier traversal time definitions lead, under tunneling time conditions, to very short times, which can even become negative. It can precipitately induces an interpretation of violation of simple concepts of causality. Otherwise, negative speeds do not seem to create problems with causality, since they were predicted both within special relativity and within quantum mechanics [18]. A possible explanation of the time advancements related to the negative speeds can come, in any case, from consideration of the very rapid spreading of the initial and transmitted wave packets for large momentum distribution widths. Due to the similarities between tunneling (quantum) packets and evanescent (classical) waves, exactly the same phenomena are to be expected in the case of classical barriers[47]. The existence of such negative times is predicted by relativity itself based on its ordinary postulates [5], and they appear to have been experimentally detected in many works [26, 27]. In this extensively explored scenario, the first group quoted above contains the so-called phase times [28, 29, 30] which are obtained when the stationary phase method (SPM) [35] is employed for obtaining the times related to the motion of the wave packet spatial centroid. Generically speaking, the SPM essentially enables us to parameterize some subtleties of several quantum phenomena, such as tunneling [2, 9, 16], resonances [31, 32, 33], incidence- reflection and incidence-transmission interferences [34] as well as the Hartman effect [38] and its superluminal traversal time interpretation [5, 7, 19]. In fact, it is the simplest and most usual approximation method for describing the group velocity of a wave packet in a quantum scattering process represented by the collision of a particle with a potential barrier [5, 7, 12, 29, 38, 39]. Inthefollowing study wewill concentrate onsomeincompatibilities thatappear when the SPM is utilized for deriving tunneling times. After quantifying the restrictive conditions for the use of the method, at the end of our analysis, we discuss a theoretical exercise involving a symmetrical collision between two identical wave packets and an one-dimensional rectan- gular potential barrier. We demonstrate that by summing the amplitudes of the reflected and transmitted waves in the scope of what we denominate a multiple peak decomposition analysis [39], we can recompose the scattered wave packets in a way that the analytical conditions for the SPM applicability are totally recovered. The SPM can be successfully applied for describing the movement of the center of a wave packet constructed in terms of a momentum distribution g(k−k ) which has a pronounced 0 3 peak around k . By assuming that the phase that characterizes the propagation varies 0 smoothly around the maximum of g(k − k ), the stationary phase condition enables us to 0 calculate the position of the peak of the wave packet (highest probability region to find the propagating particle). With regard to the tunneling phenomenon, the method is usually applied to find the position of a wave packet that traverses a potential barrier. For the case in which we consider a rectangular potential barrier V(x), V(x) = V if x ∈ [−L/2, L/2] 0 and V(x) = 0 if x ∈/[−L/2, L/2], V x∈ [−L/2, L/2] o  V(x) = (1)      0 x∈/[−L/2, L/2]    it is well known that the transmittedwave packet solution (x ≥ L/2) calculated by means  of the Schroedinger formalism is given by [40] w dk k2 ψT(x,t) = g(k−k )|T(k,L)| exp ik(x−L/2)−i t+iΘ(k,L) . (2) 2π 0 2m Z0 (cid:20) (cid:21) In case of tunneling, the transmitted amplitude and the phase shift are respectively given by −1 w4 2 |T(k,L)| = 1+ sinh2[ρ(k)L] , (3) 4k2ρ2(k) (cid:26) (cid:27) and 2k2−w2 Θ(k,L) = arctan tanh[ρ(k)L] , (4) kρ(k) (cid:26) (cid:27) for which we have made explicit the dependence on the barrier length L, and we have adopted ρ(k) = (w2 −k2)21 with w = (2mV )12 and ~ = 1. By not considering any eventual 0 distortion that |T(k,L)| could cause to the supposedly symmetric function g(k − k ), the 0 stationary phase condition is indiscriminately applied to the phase (2) leading to d k2 k(x−L/2)− t+Θ(k,L) = 0 dk 2m (cid:26) (cid:27)(cid:12)k=kmax (cid:12) k dΘ(k,L) (cid:12) ⇒ x−L/2− max t+ (cid:12) = 0. (5) m dk (cid:12)k=kmax (cid:12) (cid:12) The above result is frequently adopted for calculating (cid:12)the transit time tT of a transmitted wave packet when its peak emerges at x = L/2, t = m dΘ(k,α(L )) = 2mL w4 sinh(α)cosh(α)− 2km2ax −w2 km2axα (6) T kmax dk (cid:12)(cid:12)k=kmax kmaxα ( 4km2ax (w2−km2ax)(cid:0)+w4 sinh2((cid:1)α) ) (cid:12) (cid:12) (cid:12) 4 1 where we have defined the parameter α = (w2 −k2 )2 L. The concept of opaque limit is max introduced when we assume that k is independent of L and then we make α tend to ∞ max [19]. In this case, the transit time can be rewritten as 2m OL t = . (7) T k ρ(k ) max max Intheliterature,thevalueofk isfrequentlyapproximatedbyk ,themaximumofg(k−k ), max 0 0 which, in fact, does not depend on L and could lead us to the superluminal transmission time interpretation [8, 19, 46]. To clear up this point, we notice that when we take the so called opaque limit in Eq. (7), with L going to ∞ and w fixed as well as with w going to ∞ and L fixed, with k < w in both cases, the expression (7) leads to times corresponding to 0 a transmission process performed with velocities larger than c [19]. Such a superluminal interpretation was extended to the study of quantum tunneling through two successive barriers separated by a free region [20]. In this approach, the total traversal time should be independent of the barrier widths and of the distance between the barriers. In a subsequent analysis, the same technique was applied to a problem with multiple successive barriers where the tunneling process was designated as a highly non-local phenomenon [46]. It would be perfectly acceptable to consider k = k for the application of the stationary max 0 phase condition if the momentum distribution g(k−k ) centered at k was not modified by 0 0 any boundary condition. That is the case of the incident wave packet before the collision with the potential barrier. In this sense, and in the context of the above quoted theoretical results, our criticism is concerned with the way of obtaining all the above results for the transmitted wave packet. It has not taken into account the bounds and enhancements imposed by the analytical form of the transmission coefficient. To perform the correct analysis, we should calculate the correct value of k to be max substituted in Eq. (6) before taking the opaque limit. We are thus obliged to consider the relevant amplitude for the transmitted wave as the product of a symmetric momentum distribution g(k − k ), which describes the incoming wave packet, by the modulus of the 0 transmission amplitude T(k,L), which is a crescent function of k. The maximum of this product representing the transmission modulating function would be given by the solution of the equation ′ ′ g (k −k ) |T(k,L)| g(k−k ) |T(k,L)| 0 + = 0. (8) 0 g(k−k ) |T(k,L)| (cid:20) 0 (cid:21) 5 Obviously, the peak of the modified momentum distribution is shifted to the right of k 0 so that k has to be found in the interval ]k ,w[. Moreover, we can demonstrate by the max 0 numerical results of Table 1 that k presents an implicit dependence on L. For obtaining max the Table 1 data we have found the maximum of g(k−k )|T(k,L)| by assuming a gaussian 0 1 distribution g(k−k ) = a2 4 exp −a2(k−k0)2 almost completely comprised in the interval 0 2π 4 [0,w]. (cid:16) (cid:17) h i By increasing the value of L with respect to the wave packet width a, the value of k max obtained from the numerical calculations to be substituted in Eq. (6) also increases up to L reaches certain values for which the modified momentum distribution becomes unavoidably distorted. In this case, the relevant values for k are concentrated in the neighborhood of the upper boundary value w. We shall show in the following that the value of L, which sets up the distortion the momentum distribution can be analytically obtained in terms of a. Now, if we take the opaque limit of α by fixing L and increasing w, the above results immediately ruin the superluminal interpretation upon the result of Eq. (6), since tOL tends T to ∞ when k is substituted by w. Otherwise, when w is fixed and L tends to ∞, the max parameter α calculated at k = w becomes indeterminate. The transit time t still tends to T ∞ but now it exhibits a peculiar dependence on L, which can be easily observed by defining the auxiliary function sinh(α)cosh(α)−α G(α) = . (9) sinh2(α) When α ≫ 1, the transmission time assumes infinite values 2mL 2m tα = G(α) ⇒ tα ≈ → ∞. (10) T wα T w (w2−k2)12 −1 with an asymptotic dependence on (w2 −k2) 2. Only when α tends to 0 we have an explicit linear dependence on L given by 2mL G(α) 4mL t0 = lim = (11) T w α→0 α 3w (cid:26) (cid:27) In addition to the above results, the transmitted wave must be carefully studied in terms of the ratio between the barrier extension L and the wave packet width a. For very thin barriers, i. e. when L is much smaller than a, the modified transmitted wave packet presents substantially the same form of the incident one. For thicker barriers, but yet with L < a, the peak of the gaussian wave packet modulated by the transmission coefficient is shifted to higher energy values, i. e. k > k increases with L. For very thick barriers, i. e. max 0 6 TABLE I: The values of k numerically obtained in correspondence with the increas- ing of the barrier extension L. Thevalues are calculated in terms of the wave packet width a for different values of the potential barrier height expressed in terms of wa. We have fixed the incoming momentum by setting k a= 1. 0 wa 1.5 2.0 4.0 6.0 8.0 10 20 L/a 0.00 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 0.10 1.0235 1.0648 1.3799 1.6769 1.8547 1.9397 2.0051 0.20 1.0794 1.1825 1.6571 1.9178 2.0000 2.0204 2.0203 0.30 1.1478 1.3001 1.8430 2.0289 2.0562 2.0551 2.0342 0.40 1.2196 1.4116 1.9874 2.1025 2.0986 2.0857 2.0484 0.50 1.2921 1.5194 2.1155 2.1668 2.1399 2.1170 2.0628 0.60 1.3649 1.6266 2.2429 2.2314 2.1828 2.1495 2.0775 0.70 1.4383 1.7360 2.3819 2.3002 2.2281 2.1834 2.0925 a 0.80 ∗ 1.8489 2.5466 2.3751 2.2761 2.2188 2.1078 0.90 ∗ 1.9646 2.7627 2.4578 2.3272 2.2558 2.1234 1.00 ∗ ∗ 3.1137 2.5504 2.3818 2.2947 2.1392 aForthe valuesofLmarkedwith∗,wecandemonstratebymeans ofEqs.(12-13)thatthe modulated momentum distribution has already been completely distorted. In this case, the maximum has no meaning in the context of the applicability of the method of stationary phase. when L > a, we are able to observe that the form of the transmitted wave packet is badly distorted with the greatest contribution coming from the Fourier components corresponding to the energy w just above the top of the barrier in a kind of filter effect. We observe that the quoted distortion starts to appear when the modulated momentum distribution presents a local maximal point at k = w which occurs when d [g(k−k ) |T(k,L)|] > 0. Since dk 0 k=w the derivative of the gaussian function g(k−k ) is negative at k = w, the p(cid:12)revious relation 0 (cid:12) 7 gives − g′(w−k0) < lim T′(k,L) = wL2 1+ w3L2 < wL2 (12) g(w−k0) k→w(cid:20)T(k,L)(cid:21) 4 (cid:16)1+ wL2(cid:17) 3 4 (cid:16) (cid:17) which effectively represents the inequality a2 wL2 3 k (w−k )< ⇒ L > a 1− 0 . (13) 2 0 3 2 w r (cid:18) (cid:19) Due to the filter effect, the amplitude of the transmitted wave is essentially composed by the plane wave components of the front tail of the incoming wave packet that reaches the first barrier interface before the peak arrival. Meanwhile, only whether we had cut the momentumdistributionoffatavalueofk smallerthanw,i. e. k ≈ (1−δ)w,thesuperluminal interpretation of the transition time (7) could be recovered. In this case, independently of the way that α tends to ∞, the value assumed by the transit time would be approximated by tα ≈ 2m/wδ, which is a finite quantity. Such a finite value would confirm the hypothesis T of superluminality. However, the cut off at k ≈ (1 − δ)w increases the amplitude of the tail of the incident wave as we can observe in Fig. 1. It means that the contribution of wave packet tail for the final composition of the transmitted wave is put on the same level with the contribution of the peak of the incident wave. Consequently, an ambiguity in the definition of the arrival time is created. To summarize, at this point we are particularly convinced that the use of a step- discontinuity to analyze signal transmissions in tunneling processes deserves a more careful analysis than the immediate application of the stationary phase method. The point is that we cannot find an analytic-continuation between the above-barrier case solutions and the below-barriercase solutions. By assuming the factualinfluence ofthe amplitude of the trans- mitted wave, we may introduce an alternative analysis where we consider the possibility of using the multiple peak decomposition technique developed for the above barrier diffusion problem [39]. By means of such an experimentally verifiable exercise, we shall be able to understand how the filter effect can analytically affect the calculations of transit times in the tunnel process. In the framework of the multiple peak decomposition [39], we suggest a suitable way for comprehending the conservation of probabilities for a very particular scattering configura- tion where the asymmetric aspects above discussed can be totally eliminated. In order to recover the scattered momentum distribution symmetry conditions for accurately applying 8 1.0 k =0.6w cutoff k =0.7w cutoff k =0.8w cutoff k =0.9w cutoff k =1.0w cutoff 2 )| 0 0, ( c n YI 0.5 | 0.0 -2 -1 0 1 2 x/a FIG. 1: Dependence of the wave packet shape on the cut off value of a momentum distribution centered around k =0.5w with the values of k comprised between 0 and k . o cutoff the SPM, we assume a symmetrical colliding configuration of two wave packets traveling in opposite directions. By considering the same rectangular barrier V(x), we solve the Schroedinger equation for a plane wave component of momentum k for two identical wave packets symmetrically separated from the origin x = 0. At time t = −(mL)/(2k ) chosen for 0 mathematical convenience, we assume that they perform a totally symmetric simultaneous collision with the potential barrier. The wave packet reaching the left(right) side of the barrier is represented by +∞ ψL (R)(x,t) = dkg(k−k )φL (R)(k,x) exp[−iEt]. (14) 0 Z0 Here we have assumed that the limits of the above integral can be naturally extended from the interval [0,w] to the interval [0,∞] as a first approximation. Its range of validity can be controlled by the choice of the width ∆k of the momentum distribution g(k −k ) (with 0 k > 0) with ∆k enhanced by the barrier’s height (V ). By assuming that φL (R)(k,x) are 0 0 9 Schroedinger equation solutions, at the time t = −(mL)/(2k ), i. e. when the wave packet 0 peaks simultaneously reach the barrier, we can write φL (R)(k,x) = exp[±ikx]+RL (R)(k,L)exp[∓ikx] x < −L/2(x > L/2), 1 B φL (R)(k,x) = φ2L (R)(k,x) = αBL (R)(k)exp[∓ρx]+βBL (R)(k)exp[±ρx] −L/2 < x < L/2,   φL (R)(k,x) = TL (R)(k,L)exp[±ikx] x > L/2(x < −L/2). 3 B    where the upper(lower) sign is related to the index L(R). By assuming the conditions for the continuity of φL ,R and their derivatives at x = −L/2 and x = L/2, after some mathematical manipulations, we can easily obtain RL ,R(k,L) = exp[−ikL] exp[iΘ(k,L)][1−exp[2ρ(k)L]] (15) B 1−exp[2ρ(k)L]exp[iΘ(k,L)] (cid:26) (cid:27) and TL ,R(k,L) = exp[−ikL] exp[ρ(k)L][1−exp[2iΘ(k,L)]] , (16) B 1−exp[2ρ(k)L]exp[iΘ(k,L)] (cid:26) (cid:27) L where Θ(k,L) is given by the Eq. (4) and R (k,L) and TR(k,L) as well as RR(k,L) and B B B L T (k,L) are intersecting each other. By analogy with the procedure of summing amplitudes B that we have adopted in the multiple peak decomposition scattering [39], such a pictorial configurationobliges usto sumthe intersecting amplitudeof probabilities beforetaking their squared modulus in order to obtain RL ,R(k,L)+TR,L (k,L) = exp[−ikL] exp[ρ(k)L]+exp[iΘ(k,L)] = exp{−i[kL+ϕ(k,L)]} B B 1+exp[ρ(k)L]exp[iΘ(k,L)] (cid:26) (cid:27) (17) with 2kρ(k) sinh[ρ(k)L] ϕ(k,L) = arctan . (18) w2+(k2−ρ2(k))cosh[ρ(k)L] (cid:26) (cid:27) From Eq. (17), it is important to observe that, differently from the previous standard tun- neling analysis, by adding the intersecting amplitudes at each side of the barrier, we keep the original momentum distribution undistorted since |RL ,R(k,L)+TR,L (k,L)| is equal to one. B B At this point we recover the most fundamental condition for the applicability of the SPM. It allows us to accurately find the position of the peak of the reconstructed wave packet composed by reflected and transmitted superposing components. The phase time interpre- tation can be, in this case, correctly quantified in terms of the analysis of the new phase ϕ(k,L). By applying the stationary phase condition to the recomposed wave packets, the maximal point of the scattered amplitudes g(k−k )|RL ,R(k,L)+TR,L (k,L)| are accurately 0 B B 10