Group velocity

Roger Grimshaw July 2, 2002

Waves often propagate as a packet, or group, within which there are several wave crests. A common illustration is given by the wave pattern formed when a stone is thrown into a pond. An axially symmetric ring of waves on the water surface propagates outwards as a wave group; within the group, however, one can see that the wave crests propagate through the group, apparently from the rear to the front. The speed of the waves crests, called the phase velocity, is different from that of the group as a whole, this speed being called the group velocity. For water waves, the phase velocity is greater than the group velocity, except for the very short waves dominated by surface tension. This important distinction between phase and group velocity arises in all physical systems which support waves. The concept of group velocity appears to have been rst formulated by Hamilton in 1839, quoted by Havelock (1914). The rst recorded observation of the group velocity of a (water) wave is due to Russell in 1844 (note also his remark that the sound of a cannon travels faster than the command to re it , Russell (1885)).However, our present understanding of group velocity is usually attributed to Stokes, who used it as the topic of a Smiths Prize examination paper in 1876. For simplicity, rst consider a linearised system, with just a single active spatial variable x, and with time represented by t. Then a sinusoidal wave has the representation, u(x, t) = a cos (kx t + ) . (1)

Here k is the wavenumber, is the wave frequency, while a is the constant amplitude and is a constant phase. Thus, this sinusoidal wave has a wavelength = 2/k and a wave period T = 2/. Equation (1) is a kinematic expression, valid for all physical systems which support waves. The dynamics of the system in question will result in the dispersion relation, = (k) , (2) dening the frequency as a function of wavenumber. The phase velocity c = /k is likewise a function of wavenumber. To obtain the group velocity, consider the argument rst advanced by Stokes (1876) and later in more general form by Rayleigh (1881). Consider the linear superposition of two sinusoidal waves, each of the form (1 with frequencies , wavenumbers k K, and with equal amplitudes a and phases . The result can be written in the form u(x, t) = 2a cos (kx t) cos (Kx t) . (3)

This expression can then be interpreted as a wave packet, or group, with a dominant wavenumber k and frequency , and with a group velocity of /K. Taking the limit K 0, one sees that the group velocity is given by d . (4) cg = dk It is obtained by dierentiation of the dispersion relation (2). A more general argument, which links the origin of a wave packet to initial conditions, uses Fourier superposition to represent the solution of an initial-value problem in the form,

u(x, t) =

F (k) exp (i(kx t)) dk + c.c ,

(5)

where F (k) is the Fourier transform of u(x, 0) and c.c stands for complex conjugate. Thus, the initial conditions determine the Fourier transform, each component of which evolves independently with frequency related to the wavenumber k through the dispersion relation (2). To obtain a wave packet, we suppose that the initial conditions are such that F (k) has a dominant component centred at k = k0 . The dispersion relation is then approximated by = 0 + b1 (k k0 ) + b2 (k k0 )2 , where b1 = d = cg , dk and b2 = 1 d2 . 2 dk 2 (6) (7)

Here, both b1 , b2 are evaluated at k = k0 . The expression (5) then becomes u(x, t) A(x, t) exp (i(k0 x 0 t)) + c.c. , (8)

where

A(x, t) =

F (k0 + ) exp (i((x cg t))) ib2 2 t) d ,

(9) where the variable of integration has been changed from k to = k k0 . Here the sinusoidal factor exp (i(k0 x 0 t)) is a carrier wave with a phase velocity 0 /k0 , while the (complex) amplitude A(x, t) describes the wave packet. Since the term proportional to 2 in the exponent in (9) is a small correction term, it can be seen that to leading order, the amplitude A propagates with the group velocity cg , while the afore-mentioned small correction term gives a dispersive correction term proportional to t1/2 . Indeed, it can be shown that A(x, t) F (k0 ) 2 |b|t
1 2

exp (

i signb) , 4

where

x = cg t .

(10)

This same result can be obtained directly from (5) by using the method of stationary phase, valid here in the limit when t (see, for instance, Lighthill (1978) or Whitham (1974)). In this case, it is not necessary to assume also that F (k) is centred at k0 , and so wave packets are the generic long-time outcome of initial-value problems. It is useful to note that the group velocity appears naturally in the kinematic theory of waves (Lighthill (1978) and Whitham (1974)). Thus, let the wave eld be dened asymptotically by u(x, t) A(x, t) exp (i(x, t)) + c.c., (11)

where A(x, t) is the (complex) wave amplitude, and (x, t) is the phase, which is assumed to be rapidly-varying compared to the amplitude. Then it is natural to dene the local wave frequency and wavenumber by = , k= . (12) t x Note that the expression (10) has the required form (11). Then crossdierentiation leads to the kinematic equation for the conservation of waves, k + = 0. (13) t x

But now if we suppose the dispersion relation (2) holds for the frequency and wavenumber dened by (12), then we readily obtain k k + cg = 0, t x (14)

with a similar equation for the frequency. Thus, both the wavenumber and frequency propagate with the group velocity, a fact which can also be seen in (10). Equation (14) is itself a simple wave equation, which can be readily integrated by the method of characteristics, or rays. It is important to note that the group velocity cg is a function of k, so that (14) is a nonlinear equation. Next, suppose that the physical system contains several spatial variables, represented by the vector x. Then the phase variable (kx t) in (1, 5) is replaced by (k.x t), where k is the vector wavenumber. The dispersion relation (2) is replaced by = (k) . (15)

The phase velocity is now c, which has a magnitude /|k| and is in the direction of k. The denition of group velocity, replacing (4) is cg =
k .

(16)

Thus the group velocity can dier from the phase velocity in both magnitude and direction. A striking example of the latter arises for internal waves, whose group velocity is perpendicular to the phase velocity (see Lighthill (1978)). Because the group velocity is the velocity of the wave packet as a whole, it is no surprise to nd that it can also be identied with energy propagation. Indeed, it can be shown that in most linearised physical systems, an equation of the following form can be derived, E + t .(cg E) = 0 , (17)

where E is the wave action density, and is proportional to the square of the wave amplitude |A|2 , with the factor being a function of the wavenumber k. Typically, the wave action is just the wave energy density divided by the frequency, at least in inertial frames of reference. So far the discussion has remained within the realm of linearised theory, and it remains to mention the consequences of nonlinearity.

Analogous denitions and concepts can be developed, and, at least for weakly nonlinear waves, the main outcome is that a dependence on wave amplitude (more strictly, on |A|2 ) appears in the dispersion relation. This has the consequence that the phase and group velocities both inherit a weak dependence on the wave amplitude. It can be shown that, at least within the connes of weakly nonlinear theory, in the case of just a single active spatial dimension, the complex wave amplitude A is governed by the nonlinear Schrdinger (NLS) equation, o i( A A 2A + cg )+b2 2 + |A|2 A = 0 . t x x (18)

Here recall that b2 is dened in (7) and is a nonlinear coecient which is system-dependent. If A is assumed to depend only on t then (18) has the plane wave solution A = A0 exp (i|A0 |2 t) , (19) which shows that the nonlinear coecient in (18) has the physical interpretation that |A0 |2 is the nonlinear correction to the wave frequency. It can be shown that this plane wave is stable (unstable) according as b2 < (>)0, see Whitham (1974). In the unstable case, b2 > 0, a perturbed plane wave will evolve into one or more solitons, where the soliton family is given by (see Zakharov and Shabat (1972)), . b2 (20) Here x0 is a phase constant determining the location of the soliton at t = 0, and the amplitude is the free parameter. Note that whereas in linear theory, the wave packet prole is determined by the initial conditions, the inuence of even weak nonlinearity, when as in (18) it is balanced by weak dispersion, results in the generic prole sech. A(x, t) = exp (i2 t)sech((x cg t x0 )) , where 2 = See also Modulated waves; Nonlinear Schrdinger equations; Wave o packets, linear and nonlinear

References
Hamilton, W.R. 1839. Researches respecting vibration, connected with the theory of light, Proc. Roy. Irish Acad. 1: 267, 341.

Havelock, T.H. 1914, The Propagation of Disturbances in Dispersive Media Cambridge, C.U.P. Lighthill, M.J. 1978. Waves in Fluids, Cambridge, C.U.P. Stokes, G.G. 1876. Problem 11 of the Smiths Prize examination papers (Feb. 2, 1876), in Mathematical and Physical Papers, volume 5, p.362: Johnson Reprint Co., New York, 1966. Russell, J.Scott. 1844. Report on Waves. Brit. Assoc. Reports, p. 369: 311-390. Russell, J.Scott. 1885 The Wave of Translation in the Oceans of Water Air and Ether. London. Whitham, G.B. 1974. Linear and Nonlinear Waves, New York: Wiley. Zakharov, V.E. & Shabat, A.B. 1972. Exact theory of two-dimensional self-focusing and one-dimensional self-modulation of waves in nonlinear media. Soviet Physics, JETP, 34: 62-69