Calculation of band structures by a discrete variable representation based on Bloch functions Hervé Le Rouzoa兲 Laboratoire de Photophysique Moléculaire du CNRS, Laboratoire associé à l’Université Paris-Sud, Université Paris-Sud, bât. 210, 91405 Orsay Cedex, France 共Received 7 July 2004; accepted 11 June 2005兲 The propagation of waves in periodic media is restricted to allowed bands of energy or frequencies separated by forbidden gaps. We propose a discrete variable representation, based on Bloch functions, that is suitable for the calculation of band structures in one-dimensional systems. The solutions are obtained by a single diagonalization without any integration. The method only needs the values of the potential at grid points. Applications to the standard Krönig-Penney and Mathieu potentials, and to the nonanalytical case of a soft Coulomb potential, show that very good accuracy is achieved with moderate grid sizes. © 2005 American Association of Physics Teachers. 关DOI: 10.1119/1.1994858兴 I. INTRODUCTION The propagation of various kinds of waves in periodic materials exhibits general features such as the existence of band gaps, that is, energy ranges in which waves cannot propagate. These forbidden band gaps are separated by allowed bands, which together constitute the band structure of the system. These band structures have first been recognized for electrons in solids.1 Recently, their study has received increasing interest for electromagnetic waves in periodic dielectric media, or photonic crystals,2 and for acoustic waves in phononic crystals.3 In this article, we compute band structures with the aid of a discrete variable representation 共DVR兲 based on Bloch functions. We consider the time independent Schrödinger equation for a one-dimensional particle of mass − ប 2 d 2 + V共x兲共x兲 = E 共x兲. 2 dx2 共1兲 The solution of eigenvalue problems of the form given in Eq. 共1兲 for an eigenstate 共x兲 with energy E is a major goal in quantum physics and chemistry. Usually, 共x兲 is expanded into a finite set of basis functions n共x兲: N 共x兲 = 兺 cnn共x兲. 共2兲 n=1 The unknown coefficients cn can be determined variationally, for example, by either the Rayleigh-Ritz method4 if is a bound state 共E is an unknown discrete value兲, or the variational R-matrix method5 if is a free state 共E is continuous兲. In both cases, we first compute the matrix representation of the Hamiltonian operator with matrix elements 冓冏 Hmn = m − 冏冔 ប2 d2 + V共x兲 n = Tmn + Vmn . 2 dx2 Am. J. Phys. 73 共10兲, October 2005 II. THE GENERIC DISCRETE VARIABLE REPRESENTATION The discrete variable representation is defined by a set of N basis functions n共x兲, n共x兲, 共n = 1,2,…,N兲 http://aapt.org/ajp 共4兲 together with a N-point quadrature, with real abscissae xk and positive real weights wk: 共xk,wk兲, 共k = 1,2,…,N兲. 共5兲 The basis functions and grid points have to describe the desired solution 共x兲 of Eq. 共1兲, on the interval 共a , b兲, which is possibly infinite. The basis functions are orthonormal on the interval 共a , b兲: 共3兲 To be able to keep only a few terms in Eq. 共2兲, the basis set must be sophisticated, which might lead to difficulties in evaluating the matrix elements in Eq. 共3兲. A discrete variable representation bypasses this difficulty because the Hamiltonian matrix evaluation does not require any integration: the kinetic energy matrix T is known analytically, and the potential is simply computed at the grid points to produce a diag962 onal matrix. The method was developed for molecular quantum mechanics by Light and co-workers in the 1980s.6 In this paper, we solve Eq. 共1兲 for an electron in a crystal lattice, that is, for a spatially periodic potential, by constructing a discrete variable representation based on Bloch basis functions. The Fourier basis discrete variable representation was introduced for periodic problems.7–10 Other trigonometric basis sets also are available.11,12 In Sec. II a brief derivation of the generic discrete variable representation is presented. In Sec. III we treat an arbitrary periodic potential and derive explicit formulas for Bloch basis functions. In Sec. IV we apply the Bloch-discrete variable representation to the Krönig-Penney and Mathieu potentials, for which analytical solutions are known, as well as to the more difficult soft Coulomb periodic potential. Our summary and conclusion are given in Sec. V. 具兩m兩n典 = 冕 b a N m* 共x兲n共x兲dx ⯝ 兺 wkm* 共xk兲n共xk兲 k=1 = ␦mn , 共6兲 where *n denotes the complex conjugate of n. Suppose that it is possible to construct from the set n共x兲 new basis functions um共x兲 that satisfy13 © 2005 American Association of Physics Teachers 962 um共xk兲 = ␦mk 冑w k 共7兲 . Each function um共x兲 is localized around the discrete grid point xm; it is a continuous analog of the Dirac delta function ␦共x − xm兲. The matrix representation of any multiplicative op- We can express the matrix representation of the kinetic energy operator Tmn in Eq. 共3兲 in the new basis where V̂ is diagonal. By definition 冓冏 冏冔 um erator, say the potential V̂, can be shown to be diagonal: Vmn = 具um兩V̂兩un典 共8a兲 N N ␦mk d2 * u n = 兺 w ku m 共xk兲u⬙n共xk兲 = 兺 wk u⬙n共xk兲 冑 dx2 wk k=1 k=1 = 冑wmu⬙n共xm兲. 共14兲 We take two derivatives of Eq. 共12兲 and find N N N * = 兺 w ku m 共xk兲V共xk兲un共xk兲 = 兺 wk k=1 k=1 ␦mk ␦nk 冑wk V共xk兲 冑wk 共8b兲 N = 兺 ␦mkV共xk兲␦nk = V共xm兲␦mn . 共8c兲 k=1 u⬙n共x兲 = 冑wn 兺 *k 共xn兲⬙k共x兲, so that the matrix representation of the kinetic energy operator in the basis um is N Tmn = − As a special case, the representation of the position operator x̂ is Xmn = 具um兩x̂兩un典 = xm␦mn . 共9兲 Equation 共9兲 often is used as the starting point for the derivation of popular discrete variable representations that rely on orthogonal polynomials and Gaussian quadratures.14 In addition to giving a diagonal representation of V̂, Eq. 共7兲 leads to the attractive property that the potential matrix V is given without any integration. Only a simple evaluation at the grid points is needed, in contrast to other methods where the calculation of potential matrix elements is a major difficulty, especially for the Coulomb potential. Now we show how the functions um can be expressed in terms of the original basis set n. Let c̃n be the unknown coefficients in the expansion ប2 冑wmwn 兺 *k 共xn兲⬙k共xm兲. 2 k=1 um共x兲 = 兺 c̃nn共x兲. 共10兲 n=1 Because the basis n is orthonormal, we have using Eq. 共7兲 共16兲 In most cases, the finite sum in Eq. 共16兲 can be evaluated once in closed analytical form for the basis functions n. Given the discrete variable representation of the operators V̂ and T̂, the energies and wave functions are obtained by diagonalization of the Hamiltonian matrix, Eq. 共3兲. Let Ei be the eigenvalue associated with the eigensolution i共x兲 N i共x兲 = 兺 cnun共x兲, 共17兲 n=1 where the cn form the ith eigenvector of the Hamiltonian matrix. With the aid of Eq. 共7兲, the numerical values of i共x兲 at the grid points xk are readily obtained as N N i共xk兲 = 兺 cnun共xk兲 = 兺 cn n=1 N 共15兲 k=1 n=1 ␦nk = ck 冑w k 冑w k . 共18兲 In other words, except for weight factors, the eigenvectors of the Hamiltonian matrix directly give the values of the wave function at the grid points. N c̃n = 具兩n兩um典 = 兺 wk*n共xk兲um共xk兲 III. BLOCH-DISCRETE VARIABLE REPRESENTATION FOR PERIODIC POTENTIALS k=1 N = 兺 wk*n共xk兲 k=1 ␦mk 冑w k = 冑wm*n共xm兲. 共11兲 We substitute Eq. 共11兲 into Eq. 共10兲 and obtain the desired basis functions N um共x兲 = 冑wm 兺 *n共xm兲n共x兲. 共12兲 n=1 Note that the initial basis set, Eq. 共4兲, and the quadrature rule, Eq. 共5兲, must be consistent.11 For instance, letting x = xk into Eq. 共12兲 and using Eq. 共7兲 provides a necessary relation between the weight wk and the values of all basis functions at the grid point xk 冉兺 N wk = n=1 963 兩n共xk兲兩 2 冊 −1 . Am. J. Phys., Vol. 73, No. 10, October 2005 共13兲 Let the potential be a P-periodic function: V共x + P兲 = V共x兲. Floquet’s theorem15 states that we can find a solution to Eq. 共1兲 of the form 共x兲 = eixF共x兲, 共19兲 where F共x兲 has the same periodicity as V共x兲 and is the propagation constant. Solutions of the form of Eq. 共19兲 are called Bloch wave functions.16 Bloch functions are similar to plane waves exp共ikx兲; the difference is the presence of the periodic modulation F共x兲. However, is not the same as k: Whereas k is a good quantum number, is not because p̂ = −iប / x does not commute with the Hamiltonian. For this reason, ប is known as a quasi-momentum.17 Because F共x兲 is periodic, it can be expanded in a Fourier series with the fundamental angular frequency = 2 / P. It follows that 共x兲 can be expanded as Hervé Le Rouzo 963 ⬁ 共x兲 = e ix 兺 ⬁ c ne inx n=−⬁ = 兺 cnei共+n兲x , 共20兲 n=−⬁ where the cn are coefficients to be determined. In view of the expansion of Eq. 共20兲, it is natural to build a discrete variable representation with Bloch basis functions of the form n共x兲 = ei共+n兲x 冑P , 共n = − M,…,− 1,0,1,…,M兲. 共21兲 In Eq. 共21兲, einx is the usual Fourier basis, and eix accounts for the Bloch factor. It is easy to verify that Eq. 共21兲 defines a set of N = 2M + 1 orthonormal basis functions over the range 共0 , P兲. The quadrature adopted for the Bloch discrete variable representation is the simple rectangular rule wk = ⌬, xk = k⌬, 共k = 1,2,…,2M + 1兲 共22兲 with the step size ⌬= P . 2M + 1 共23兲 Fig. 1. The Krönig-Penney 共full line兲 and Mathieu 共dashed line兲 potentials given in Eqs. 共29兲 and 共30兲. M 共24兲 The sum over k can be evaluated analytically.18 We obtain 冋 Tmm = M共M + 1兲 ប2 2 + 2 3 2 Tmn = ប2 共− 1兲m−nei共m−n兲⌬ 2 冋 ⫻ − 册 共25a兲 DVR Hmn 共兲 = ប2 共− 1兲m−nei共m−n兲⌬ 2 册 cos关共m − n兲⌬⬘兴 i + 2 , sin共共m − n兲⌬⬘兲 2 sin2关共m − n兲⌬⬘兴 共26兲 For the special case of = 0 and P = 2, Eq. 共25a兲 and 共25b兲 give the discrete variable representation obtained in Ref. 9 for the simple Fourier basis set einx / 冑2 over the range 共0 , 2兲, namely, ប2 M共M + 1兲 Tmm = 3 2 cos关共m − n兲/共2M + 1兲兴 ប2 . 共− 1兲m−n Tmn = 2 sin2关共m − n兲/共2M + 1兲兴 2 共27a兲 共27b兲 Finally, we see that for a periodic potential with angular frequency = 2 / P, sampled at N = 2M + 1 regularly spaced points over one period, the Bloch discrete variable representation of the Hamiltonian is 964 Am. J. Phys., Vol. 73, No. 10, October 2005 + 2 i sin关共m − n兲⌬⬘兴 册 cos关共m − n兲⌬⬘兴 , 2 sin2关共m − n兲⌬⬘兴 共28b兲 where m , n = 1 , 2 , … , N, and ⌬ = P / N and ⌬⬘ = / N. The N ⫻ N Hamiltonian matrix HDVR共兲 in Eq. 共28a兲 and 共28b兲 can be readily computed in terms of . We emphasize that the nature of the basis sets, the original Bloch functions and their localized analogs, is transparent in Eq. 共28a兲 and 共28b兲, and is no longer required because the eigenvectors of HDVR共兲 are the amplitudes of the wave functions at the grid points up to the factor 1 / 冑⌬ 关Eq. 共18兲兴. A. The Krönig-Penney and Mathieu potentials where . 2M + 1 冋 ⫻ − 共28a兲 IV. APPLICATION TO MODEL POTENTIALS 共25b兲 ⌬⬘ = 册 M共M + 1兲 ប2 2 + 2 + V共m⌬兲 3 2 Because the modulus of any Bloch basis function is equal to 1 / 冑P, Eq. 共13兲 gives the weights wk = ⌬ in accordance with Eq. 共22兲. We substitute the Bloch basis functions of Eq. 共21兲 and the quadrature parameters of Eq. 共22兲 into Eq. 共16兲 and obtain 1 ប2 Tmn = 兺 共 + k兲2ei共+k兲共m−n兲⌬ . 2 2M + 1 k=−M 冋 DVR Hmm 共兲 = There are very few periodic potentials for which analytical solutions are available. The simplest is an infinite array of rectangular wells, the Krönig-Penney potential.19 The sinusoidal Mathieu potential is more physical 共the potential and its derivatives are continuous兲, but the calculation is more involved.20 In this section, we treat both the Krönig-Penney and Mathieu potentials on the same footing, and show that they contain the same basic physics. Both potentials have period P and oscillate between 0 and 2Vm if we take VKP共x兲 = 再 2Vm − P/4 ⬍ x ⬍ P/4 0 P/4 ⬍ x ⬍ 3P/4 冎 共29兲 for the Krönig-Penney potential, and 冉 冊 V M 共x兲 = Vm + Vmcos 2x P 共30兲 for the Mathieu potential. These potentials are depicted in Fig. 1. The dispersion relation, E = E共兲, is one of the most important properties of a material; it is the key link between the crystal structure and its physical properties. Even in one diHervé Le Rouzo 964 Fig. 2. Exact dispersion curves E = Ei共兲 and discrete variable representation values 共circles兲 for five equally spaced values in the first Brillouin zone. The horizontal dashed lines are the upper bounds 共2Vm兲 of the potentials. The parameters are P = 5 bohr and Vm = 1.5 hartree. mension, E共兲 exhibits many features of energy bands in actual periodic lattices. In the reduced representation that is adopted here 共 苸 关0 , / P兴兲, E共兲 is a multi-valued function of . For the Krönig-Penney and Mathieu potentials, the relation between E and is a transcendental equation of the form cos P = f共E , P , Vm兲, where f is a complicated function 共an infinite determinant for the Mathieu equation兲. However, it is possible to solve these relations numerically to obtain the exact energies Ei, one in each band, for a given propagation constant and a band index i. The associated wave functions will be denoted as i共x兲. The same potential parameters will be used throughout, namely, P = 5 bohr and Vm = 1.5 hartree: All values are given in atomic units.21 Correspondingly, the reduced zone for is 关0 , / P兴 = 关0 , 0.628兴 a.u. The exact dispersion curves, E = Ei共兲, in the reduced zone scheme are given in Fig. 2 for the Krönig-Penney and Mathieu potentials, for band indices ranging from i = 1 to i = 6. These dispersion curves19,20 will be used to check the discrete variable representation results. The dispersion curves exhibit forbidden energy bands, separating regions of allowed energy where the solutions i共x兲 remain bound as x → ± ⬁. Both potentials support two narrow, low energy, bands. For further comparison with the discrete variable representation results, the exact energies for = 0.2 a.u. and the same potential parameters are given in Table I. Table I. Comparison of exact and discrete variable representation values of the energies 共hartree兲 of the Bloch states with = 0.2 a.u. in the four lowest bands for the Mathieu and Krönig-Penney potentials. The discrete variable representation calculations are done with N = 51 and N = 201 points, respectively. The potential parameters are P = 5 bohr and Vm = 1.5 hartree. Mathieu potential Krönig-Penney potential Energy Ei Exact DVR error Exact DVR error E1 E2 E3 E4 0.714 78 2.039 68 2.896 48 4.294 67 2.456⫻ 10−7 5.291⫻ 10−6 1.782⫻ 10−5 5.339⫻ 10−5 0.442 13 1.693 99 3.106 00 4.324 94 0.003 22 0.011 14 0.009 44 0.006 86 965 Am. J. Phys., Vol. 73, No. 10, October 2005 Fig. 3. Exact density probabilities 兩i共x兲兩2 over two periods of the Mathieu and Krönig-Penney potentials for = 0.2. The baselines are drawn at the exact energies Ei given in Table I. The potentials are plotted to help visualize the vertical energy scale. In Fig. 3 we present the exact density probabilities 兩i共x兲兩2 of the Mathieu and Krönig-Penney potentials for = 0.2 a.u. As expected, the states 1 and 2, belonging to the two lowest narrow bands, correspond to levels that are close to that of an isolated well. Higher states still feel the presence of the potential wells, but for increasing energy, they tend toward the traveling wave eix, for which the density probability is constant everywhere. For discrete variable representation calculations a real value of is typically chosen initially, the same for all basis orbitals. Then the set of associated energies Ei is obtained from a single diagonalization. More precisely, the steps are as follows: 1. Choose an odd number N = 2M + 1 of basis functions. This choice determines N grid points xk according to Eq. 共22兲. 2. For a given in the reduced zone, construct the N ⫻ N Hamiltonian matrix HDVR共兲 according to Eq. 共28a兲 and 共28b兲. To do so requires the computation of V共xk兲, where V is given by Eq. 共29兲 or Eq. 共30兲. 3. Diagonalize the discrete variable representation Hamiltonian matrix to obtain the energies Ei and the corresponding values of the wave functions i共xk兲 on the grid. We chose N = 51 grid points for the discrete variable representation calculations for the Mathieu potential, and N = 201 for the similar calculations for the Krönig-Penney potential. The errors in Ei are reported in Table I. We see that the discrete variable representation energies are excellent for the Mathieu potential, but less precise for the Krönig-Penney potential in spite of a larger grid. As is well known in the theory of Fourier series, the reason for this discrepancy is that discontinuous potentials require more harmonics, which means more discrete variable representation points are needed to represent the wave functions near the discontinuities. For the dispersion curves, the discrete variable representation calculations are done for five equidistant values spanning the reduced zone as shown in Fig. 2. We see that there is good agreement between the exact dispersion curves and the discrete variable representation energies. We now compare the discrete variable representation wave Hervé Le Rouzo 965 Fig. 5. The soft Coulomb potential in Eq. 共31兲. The parameters are P = 10 bohr and Q = a = 1 a.u. Fig. 4. Wave function moduli 兩i共x兲兩: Exact 共solid curves兲 and discrete variable representation 共circles兲 for = 0.2. The discrete variable representation calculations are done with N = 51 points for the Mathieu potential, and with N = 201 points for the Krönig-Penney potential 共in the latter case, only one third of the points are plotted兲. functions with the exact results.19,20 Except at the band edges, where the solutions are periodic standing waves 共the famous Mathieu’s functions when V = V M 兲, the wave functions of the Bloch form, Eq. 共19兲, oscillate aperiodically because the exp共ix兲 factor modulates the periodic function F共x兲 as we move from one cell 共or period兲 to the next. However, the modulus of the Bloch functions 兩共x兲兩 = 兩F共x兲兩 is the same in every cell: All that is modulated is the phase of the wave functions. In Fig. 4 the exact and discrete variable representation values of 兩i共x兲兩 for the three lowest states are plotted for = 0.2 a.u. The agreement between the exact 共curves兲 and discrete variable representation 共circles兲 wave functions is very good for the Mathieu potential. For the Krönig-Penney potential, a larger grid size is needed to obtain a similar agreement. B. The soft Coulomb periodic potential V共x兲 = −1 / 共1 + x2兲1/2 supports bound and free states similar to those of the true Coulomb potential.22 We shall take Q = a = 1 and P = 10 共in atomic units兲. The corresponding periodic potential is depicted in Fig. 5. The atomic energies of the bound states of the isolated atom22 are reported in Table II, along with the band edges computed by the Bloch discrete variable representation with N = 101 grid points. The discrete variable representation band structure is given in Fig. 6. In Fig. 7 the discrete variable representation density probabilities 兩i共x兲兩2 for = max / 2 ⯝ 0.157 a.u. are presented. The states 1 and 2, belonging to the two lowest narrow bands, correspond to levels that are close to those of an isolated well. The third state has a significant density probability outside the wells, that is, between atoms: This state belongs to the conduction band. V. CONCLUSION We have shown that the discrete variable representation with underlying Bloch basis functions is suitable for the calculation of band structures. The method can be applied to an arbitrary one-dimensional periodic potential. The parameters are the 共odd兲 number N of grid points and the propagation constant . The Hamiltonian matrix is computed using Eq. To illustrate the capabilities of the Bloch discrete variable representation, we consider a potential whose band structure cannot be solved analytically. We construct a P-periodic soft Coulomb potential with V共x兲 = − Q2 共31兲 冑a2 + 共x − P/2兲2 , for 0 艋 x 艋 P. Such quasi-Coulombic potentials have been used in multiphoton physics because the atomic potential Table II. Energies for the soft Coulomb potential in Eq. 共31兲 with parameters Q = a = 1 a.u. For the isolated atom, the exact values are from Ref. 22; for the crystal with period P = 10 bohr, the edges of the four lowest bands were computed with N = 101 points. Atomic energies Crystal band edges Energy Ei Exact Lower Upper E1 E2 E3 E4 −0.669 82 −0.274 94 −0.151 49 −0.092 70 −0.670 28 −0.296 20 −0.173 92 0.023 033 −0.669 29 −0.246 15 0.004 753 4 0.345 14 966 Am. J. Phys., Vol. 73, No. 10, October 2005 Fig. 6. The dispersion curves E = Ei共兲 for the soft Coulomb potential calculated by the Bloch discrete variable representation with N = 101 points. Hervé Le Rouzo 966 rials with general anisotropy,” Phys. Rev. B 69, 094301-1–10 共2004兲. See any elementary quantum physics text such as K. Gottfried, Quantum Mechanics 共W. A. Benjamin, New York, 1966兲. 5 H. Le Rouzo, “Variational R-matrix method for quantum tunneling problems,” Am. J. Phys. 71, 273–278 共2003兲, and references therein. 6 J. C. Light, I. P. Hamilton, and J. V. Lill, “Generalized discrete variable approximation in quantum mechanics,” J. Chem. Phys. 82, 1400–1409 共1985兲; J. V. Lill, G. A. Parker, and J. C. Light, “The discrete variablefinite basis approach to quantum scattering,” ibid. 85, 900–910 共1986兲; Z. Bacic and J. C. Light, “Highly excited vibrational levels of ‘floppy’ triatomic molecules: A discrete variable representation-distributed Gaussian basis approach,” ibid. 85, 4594–4604 共1986兲; J. C. Light and T. Carrington, Jr., “Discrete variable representations and their utilization,” Adv. Chem. Phys. 114, 263–310 共2000兲. 7 A. S. Dickinson and P. R. Certain, “Calculation of matrix elements for one-dimensional quantum-mechanical problems,” J. Chem. Phys. 49, 4209–4211 共1968兲. 8 R. Meyer, “Trigonometric interpolation method for one-dimensional quantum-mechanical problems,” J. Chem. Phys. 52, 2053–2059 共1970兲. 9 D. T. Colbert and W. H. Miller, “A novel discrete variable representation for quantum-mechanical reactive scattering via the S-matrix Kohn method,” J. Chem. Phys. 96, 1982–1991 共1992兲. 10 J. T. Muckerman, “Some useful discrete variable representations for problems in time-dependent and time-independent quantum mechanics,” Chem. Phys. Lett. 173, 200–205 共1990兲. 11 D. Baye, “Constant-step Lagrange meshes for central potentials,” J. Phys. B 28, 4399–4412 共1995兲. 12 H. Karabulut and E. L. Silbert III, “Trigonometric discrete variable representations,” J. Phys. B 30, L513–L516 共1997兲. 13 In the context of Lagrange interpolation, Eq. 共7兲 is sometimes called the “Lagrange condition.” See D. Baye and P.-H. Heenen, “Generalised meshes for quantum mechanical problems,” J. Phys. A 19, 2041–2059 共1986兲. 14 R. M. Whitnell and J. C. Light, “Symmetry-adapted discrete variable representations,” J. Chem. Phys. 89, 3674–3680 共1988兲. 15 G. Floquet, “Sur les équations différentielles linéaires à coefficients périodiques,” Ann. Sci. Ec. Normale Super. 12, 47–88 共1883兲. See ⬍http:// www.numdam.org⬎. 16 Note that because Eq. 共1兲 is linear and second order, two independent solutions can be found for a given energy E共兲 : 共x兲 and 共−x兲 are the independent solutions, unless P = N, where N is an integer. In the latter case, the two solutions become identical and represent standing waves 共the Mathieu functions for V = V M 兲 corresponding to the band edges. 17 Note that we can derive the exact relation for the electron velocity in Bloch states: v = 具p̂典 / = E共兲 / ប. This result should be compared with the velocity v = 具p̂典 / = បk / for free electrons. 18 The calculation is tedious and follows the derivation suggested in Ref. 9. 19 R. A. Smith, Wave Mechanics of Crystalline Solids 共Chapman and Hall, London, 1961兲, pp. 134–144. 20 N. W. McLachlan, Theory and Application of Mathieu Functions 共Clarendon Press, Oxford, 1951兲; M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions 共Dover, New York, 1965兲, pp. 721–749. 21 In atomic units: = ប = 兩e兩 = 1. Atomic units of length and energy are 1 bohr= 0.529 177 Å and 1 hartree = 27.211 65 eV, respectively. 22 J. H. Eberly, Q. Su, and J. Javanainen, “High-order harmonic production in multiphoton ionization,” J. Opt. Soc. Am. B 6, 1289–1298 共1989兲; Q. Su and J. H. Eberly, “Model atom for multiphoton physics,” Phys. Rev. A 44, 5997–6008 共1991兲; S. L. Haan, R. Grobe, and J. H. Eberly, “Numerical study of autoionizing states in completely correlated two-electron systems,” Phys. Rev. A 50, 378–391 共1994兲. 23 R. G. Littlejohn and M. Cargo, “Multidimensional discrete variable representation bases: Sinc functions and group theory,” J. Chem. Phys. 116, 7350–7361 共2002兲. 4 Fig. 7. Discrete variable representation density probabilities 兩i共x兲兩2 over two periods of the soft Coulomb potential for = max / 2 with N = 101 points. The energies of the states 共in hartree兲 are E1 = −0.669 79, E2 = −0.278 14, E3 = −0.107 73, and E4 = 0.166 27. 共28a兲 and 共28b兲 and then diagonalized. The eigenvalues En are the desired energies, one in each band n. Another attractive feature is that the components of the nth eigenvector directly give the values of the nth wave function at the grid points. The generalization of the Bloch discrete variable representation to two and three dimensions is simple if the Cartesian products of one-dimensional basis functions are taken to be the basis.9 The procedure for creating more general multidimensional discrete variable representations has been recently addressed.23 ACKNOWLEDGMENTS It is a pleasure to thank Dr. Eric Charron, Dr. Andrew Mayne, and Dr. Georges Raseev of the Laboratoire de Photophysique Moléculaire in Orsay, for stimulating discussions and helpful comments on the manuscript. a兲 Electronic address: herve.le-rouzo@ppm.u-psud.fr C. Kittel, Introduction to Solid State Physics 共Wiley, New York, 1996兲, 7th ed., Chap. 7. 2 K. M. Ho, C. T. Chan, and C. M. Soukoulis, “Existence of a photonic gap in periodic dielectric structures,” Phys. Rev. Lett. 65, 3152–3155 共1990兲; J. D. Joannopoulos, R. D. Meade, and J. N. Winn, Photonic Crystals: Molding the Flow of Light 共Princeton University Press, NJ, 1995兲; F. Szmulowicz, “Analytic, graphical, and geometric solutions for photonic band gaps,” Am. J. Phys. 72, 1392–1396 共2004兲. 3 S. Yang, J. H. Page, Z. Liu, M. L. Cowan, C. T. Chan, and P. Sheng, “Ultrasound tunneling through 3D phononic crystal,” Phys. Rev. Lett. 88, 104301-1–4 共2002兲; T. Wu, Z. Huang, and S. Lin, “Surface and bulk acoustic waves in two-dimensional phononic crystal consisting of mate1 967 Am. J. Phys., Vol. 73, No. 10, October 2005 Hervé Le Rouzo 967