Professional Documents
Culture Documents
Mark Andrews∗
Department of Physics, Australian National University, ACT 0200, Australia
(Dated: February 3, 2008)
We discuss four general features of force-free evolution: (1) The spatial spread of any packet
changes with time in a very simple way. (2) Over sufficiently short periods of time (whose duration
is related to the spread in momentum of the packet) the probability distribution moves but there
is little change in shape. (3) After a sufficiently long period (related to the initial spatial spread)
the packet settles into a simple form simply related to the momentum distribution in the packet.
In this asymptotic regime, the shape of the probability distribution no longer changes except for
its scale, which increases linearly with the time. (4) There is an infinite denumerable set of simple
wave packets (the Hermite-Gauss packets) that do not change shape as they evolve.
arXiv:0801.0188v1 [quant-ph] 31 Dec 2007
contains a factor exp[ı(p − p̄)2 t/2m~] that will be very time scale for changes in shape (including the changes
close to unity throughout the region where φ(p) has any in spatial scale), and tx gives the time scale to move to
significant magnitude. Hence the asymptotic period where there is no further change
in shape but |ψ| expands uniformly with time.
−ıp̄2 t ∞
Z
1 ı p̄t
ψ(x, t) ≈ √ exp( ) exp[ p(x − )]φ(p) dp,
2π~ 2m~ −∞ ~ m
−ıp̄2 t p̄t IV. THE EVOLUTION OF THE SPREAD ∆x
= exp( ) ψ(x − , 0). (5)
2m~ m
The fact that every free packet spreads exactly as in
The wave function translates, with speed p̄/m, and no Eq. (3) has appeared in at least one textbook[5], but
change in shape, just a change in phase. A more rigorous the derivation is simpler if the concept of the total time
approach to the error in this approximation is given in derivative of an operator is used[6]. The total time
Section VII.
derivative, dt  of any operator  is defined to be
Here we have taken the initial time to be t = 0, but
this analysis can be applied at any time. Since ∆p does dt  = ∂t  + (ı/~)[Ĥ, Â], (8)
not change, the time scale m~/2∆2p for changes in shape
applies throughout the evolution. where Ĥ is the Hamiltonian operator. For the free
If the wave function has any discontinuity then ∆p = particle this leads to the operator equations of motion
∞ and there will be very rapid changes. An example of mdt x̂ = p̂ and dt p̂ = 0. From the general property
this behavior will be considered in Section VI. that dt hÂi = hdt Âi, it follows that mdt X̂ = P̂ and
dt P̂ = 0, where X̂ = x̂ − hx̂i and P̂ = p̂ − hp̂i. Fur-
thermore, mdt X̂ 2 = 2R̂, where R̂ = (P̂ X̂ + X̂ P̂ )/2, and
III. THE EVOLUTION FOR LARGE TIMES
mdt R̂ = P̂ 2 . Since hP̂ 2 i is constant, it follows that
For the propagator in Eq. (1), write the term (x − mhR̂i = hP̂ 2 it + mhR̂it=0 = hP̂ 2 i(t − tmin ), (9)
x0 )2 in the exponent as x2 − x̄2 − 2(x − x̄)x0 + (x0 − x̄)2 ,
with x̄ = hx̂it=0 . We expect ψ(x0 , 0) to be small if (x0 − where tmin = −mhR̂it=0 /hP̂ 2 i. Now we can integrate
x̄)2 ∆2x and therefore, for all times (future or past) mdt hX̂ 2 i = 2hR̂i to give
with |t| m∆2x /~, the term in (x0 − x̄)2 will make a
negligible change to the phase of the exponential. Hence hX̂ 2 i = m−2 hP̂ 2 i(t − tmin )2 + hX̂ 2 imin , (10)
r
m ım 2 which is equivalent to Eq. (3). Thus, for any wave packet,
(x − x̄2 ) ×
ψ(x, t) ≈ exp
2πı~t 2~t one can calculate the time of minimum spread from hP̂ 2 i
Z ∞
ım and hR̂it=0 . A wave packet has its minimum spread at
(x − x̄)x0 ψ(x0 , 0) dx0
exp −
−∞ ~t the time when hR̂i = 0.
r
m ım 2 If the wave function ψ is real at some time, the ı in
(x − x̄2 ) φ m(x − x̄)/t ,
= exp (6) Schrõdinger’s equation ∂t ψ = −ı(~/2m)∂x2 ψ shows that
ıt 2~t
ψ will immediately become R complex. AlsoR hp̂i = 0 be-
cause hp̂i = ψ p̂ψ dx = R(p̂ψ)∗ ψ dx = R− (p̂ψ)ψ dx =
R
where
Z ∞ −hp̂i. Similarly, hp̂x̂i = ψ p̂x̂ψ dx = (p̂ψ)∗ x̂ψ dx =
1
φ(p) = √ exp(−ıpx0 /~)ψ(x0 , 0) dx0 ,
R
(7) − (p̂ψ) (x̂ψ) dx=−hx̂p̂i. That is, hR̂i = 0 and therefore
2π~ −∞
the wave packet has its minimum spread at that instant.
which is the initial momentum wave function of the This also shows that a free wave packet can be real only
packet. A more rigorous derivation of the error in for one instant. In fact, the probability is symmetric
this approximation is given in Section VII. Historically, in time about that instant, because if ψ(x, t) satisfies
this asymptotic form was important in the theory of Schrõdinger’s equation then so does ψ ∗ (x, 2t0 − t) and
scattering.[4] since these two functions are equal at t = t0 if ψ(x, t0 ) is
The asymptotic probability distribution is therefore real, they must be equal at all times.
Given a wave packet ξ that is at rest (in the sense
(m/t) |φ m(x− x̄)/t |2 . This has a simple interpretation:
m(x − x̄)/t is the momentum required for the particle to that hp̂iξ = 0 and therefore hx̂iξ is constant) we can
be at position x at time t, if it left x̄ at time t = 0. So the make a moving packet by multiplying the wave func-
probability of finding the particle at x is proportional to tion by exp(ıpx/~). With ψ(x, t0 ) = exp(ıpx/~)ξ(x, t0 ),
the probability that it had the right momentum to get it follows that p̂ ψ = p ψ + exp(ıpx/~)p̂ ξ and therefore
there (ignoring the fact that the initial distribution was hp̂iψ = p. Similarly, hp̂2 iψ = p2 + hp̂2 iξ and hence
spread over a distance of order ∆x about x̄). (∆p )ψ = (∆p )ξ . Also, p̂ (xψ) = p xψ + exp(ıpx/~)p̂ (xξ)
Introduce the times tp = m~/2∆2p and tx = 2m∆2x /~. and therefore hp̂x̂ + x̂p̂iψ = 2phx̂iξ + hp̂x̂ + x̂p̂iξ . Thus,
Then tp 6 tx (from the uncertainty relation) and equal- hR̂iψ = hR̂iξ and it follows that the evolution of the
ity implies a Gaussian packet. In general, tp gives the spatial spread is the same as for ξ. The evolution of ψ
3
can be obtained from that of ξ by applying a Galilean for any function f (x, t). Now we can apply this result
transformation[7]: repeatedly, starting with χ, to give
ı t − t0 t − t0 (t + ıτ )n 1
ımx2 x2
ψ(x, t) = exp (px − p2 ) ξ(x − p , t). (11) χn (x, t) ∝ p exp[ 2 ] ∂xn exp(− 2 ), (16)
~ 2m 2m ~(t + ıτ ) γ
(t − ıτ )
It is easy to check that this ψ satisfies Schrõdinger’s equa-
tion if ξ does, and that ψ(x, t0 ) = exp(ıpx/~) ξ(x, t0 ). where we have introduced the time-varying length scale
γ = [~(t2 + τ 2 )/mτ ]1/2 . The generating relation (Ro-
drigue’s formula) for the Hermite polynomials is
V. THE HERMITE-GAUSS WAVE PACKETS
(d/dξ)n exp(−ξ 2 ) = (−1)n Hn (ξ) exp(−ξ 2 ), (17)
A vast array of time-dependent solutions of
Schrõdinger’s equation (SE) can be found through and using this in Eq. (16) gives
the use of invariant operators[6]. For any operator Â, 1
(t+ıτ )n γ n ımx2 −x2/γ 2 x
χn (x, t) ∝ p exp[ 2 ]e Hn ( ). (18)
(Ĥ − ı~∂t )Âψ = ([Ĥ, Â] − ı~ ∂t Â)ψ = −ı~(dt Â)ψ. (12) (t − ıτ ) ~(t+ıτ ) γ
If dt  = 0 at all times, we say that  is invariant. If Separating the real and imaginary parts of the first ex-
an invariant operator is applied to any solution of SE, ponent, the normalized sequence of solutions to SE is
then from Eq. (12) the result will also satisfy SE. Af-
ter being multiplied by a suitable time-dependent factor π −1/4 2 2 x
(if necessary), eigenfunctions of invariant operators will χn (x, t) = √ n eı(θ+(n−1)β) e−x /2γ Hn ( ), (19)
γ 2 n! γ
satisfy SE and the eigenvalue will be constant. [If the
eigenfunctions are degenerate, appropriate linear com- where θ = tx2 /2τ γ 2 and eıβ = [(t − ıτ )/(t + ıτ )]1/2 . The
binations, with time-dependent coefficients, may be re- probability distributions of the first few are:
quired.]
In the present case of a free particle, a useful invariant 1
|χ(x, t)|2 = √ exp(−x2 /γ 2 ) (20)
operator is γ π
b̂ = mx̂ − p̂(t − ıτ ), τ real (13) 2 x2
|χ1 (x, t)|2 = √ exp(−x2 /γ 2 ) (21)
γ π γ2
The eigenfunction of b̂ with eigenvalue zero (so that hx̂i = 1
hp̂i = 0) is easily found using p̂ = −ı~∂x : |χ2 (x, t)|2 = √ (1 − 2x2 /γ 2 )2 exp(−x2 /γ 2 ).(22)
2γ π
mτ 1/4 1 ımx2 These do not change their shape as they evolve; only
χ(x, t) = ( ) p exp( ). (14)
π~ (t − ıτ ) 2~(t − ıτ ) the scale of x changes with time. It would serve little
purpose to show a graph of these probability distributions
This is Schrõdinger’s well-known free Gaussian wave
because they have the same spatial form as do the energy
packet. The factor (t − ıτ )−1/2 does not come from the
eigenfunctions of the harmonic oscillator. The relation to
eigenvalue equation b̂ χ = 0, but it is clear from the prop- the harmonic oscillator is made clearer in Section VII.
agator in Eq. (1) that this factor is required to satisfy SE. The fact that the Hermite-Gauss packets do not change
Propagators always satisfy SE and with x0 = 0 the prop- shape is consistent with the asymptotic results in Section
agator in Eq. (1) has the same form as χ with τ = 0. III because the Fourier transform of a Hermite-Gaussian
Furthermore, any solution of SE is still a solution if a is also a Hermite-Gaussian.
constant (such as −ıτ ) is added to t; thus χ satisfies SE. Z ∞
Applying b̂† = mx̂ − p̂(t + ıτ ) to the eigenfunction χ 1 2 2
√ e−z /2 Hn (z)e−ıpz dz = ı−n e−p /2 Hn (p). (23)
will give another solution to SE (because b̂† is also in- 2π −∞
variant), and clearly this must have the form of the same
Gaussian multiplied by x. Applying b̂† again gives a so-
lution that has the form of the same Gaussian multiplied VI. EXAMPLES
by a quadratic polynomial in x, and so on. We will now
show that the sequence of polynomials generated in this Our first example is a wave packet which is smooth
way are Hermite polynomials. (infinitely differentiable) everywhere and goes to zero ex-
First write exp[ımx2 /2~(t − ıτ )] in Eq. (14) as ponentially at large distances, but does change shape as
exp[ımx2 /2~(t + ıτ )] exp[−mτ x2 /~(t2 + τ 2 )] and then, it evolves. We use the second derivative of Schrõdinger’s
since b̂† = mx + ı~(t + ıτ )∂x , packet χ in Eq.(14). Thus χ̄2 (x, t) ∝ ∂x2 χ(x, t), which
also must satisfy SE. It has the form
1 2 1 2
2 ımx 2 ımx
b̂† exp[ ] f = ı~(t + ıτ ) exp[ ] ∂x f (15) 2
x2
~(t + ıτ ) ~(t + ıτ ) χ̄2 (x, t) = N κ3 (2κ2 x2 − 1)e−κ , (24)
4
0.0
x 0.0
x$t
0 1 2 3 4 0 1 2 3 4
FIG. 1: The probability distribution |ψ(x, t)|2 for the wave FIG. 2: The probability distribution for the same wave packet
packet in Eq.(24), whose initial probability distribution is as in Fig. 1 for some times (t = 3, 4, 6, 16) when the packet is
shown by the solid curve. Only half of the packet is shown, approaching its asymptotic form. In the asymptotic region,
because it remains symmetric. The probability is shown t|ψ(x, t)|2 is a function of x/t only, as in Eq.(25), so we use
in the period before the asymptotic regime for the times x/t as the horizontal variable in this figure, which hides the
t = 0, 0.1, 0.2, 0.4, 0.6, 0.8, 1.0 in units of τ . The distance x substantial spreading of these packets. For all t > 7tx ≈ 16τ
is in units of (~τ /m)1/2 . the exact distribution is indistinguishable from the asymp-
totic form; so the graph for t = 16 stands for all later times.
!Ψ"x, t#!2
(30)
This is a rigorous bound on the change in the wave func-
0.5 tion (apart from the shift p̄t/m). For |δψ| to be small
compared with ψ with |ψ|2 ∼ 1/2∆x (from the normal-
ization of ψ), we require t∆4p /πm~3 1/4∆2x and hence
t∆2p /πm~3 1/4∆2x ∆2p 6 1/~2 (from the uncertainty re-
x lation). That is, we require t πm~/∆2p , in agreement
0.0 0.2 0.4 0.6 0.8 1.0 with Section II.
(b) For a more rigorous approach to the error in
FIG. 3: The probability distribution |ψ(x, t)|2 for the initially the asymptotic approximation, write the exact form of
square wave packet in Eq.(26), for several values of the time Eq. (6) as
(t = 0, 0.001, 0.01, 0.1) in the early period when the packet r
shows complicated behavior, but does not spread appreciably. m ım 2 2 m
The distance x is in units of a, the width of the initial packet. ψ(x, t) = exp (x −x̄ ) φ (x−x̄) +δψ (31)
ıt 2~t t
The times are in units of ma2 /~ and are indicated by the
numbers against the graphs. where
r Z ∞
0.5 m ım 2
0.15 δψ = exp[ 2
(x − x̄ )] f (x0 )g(x0 )dx0 , (32)
0.2 2πı~t 2~t −∞
0.1
with
t !Ψ"x, t#!
0
0.10 ım exp[− ım 2
~t (x − x̄) ] − 1
2 f (x0 ) = exp[ (x̄ − x)x0 ] (33)
~t x0 − x̄
g(x0 ) = (x0 − x̄)ψ(x0 , 0). (34)
0.05 Now the Cauchy-Schwarz inequality gives
p
|δψ|2 6 m3 /π~3 t3 ∆2x . (35)
0.00
x$t This is a rigorous bound to the error in the asymp-
0 5 10 15 totic approximation. For |δψ| to be small compared
with ψ we require |δψ|2 1/2∆x , which leads to t
FIG. 4: The probability distribution for the initially square (4/π)1/3 m∆2x , in agreement with Section II.
packet for some times (t = 0.1, 0.2, 0.5) when the packet is If the wave packet has discontinuities, as in the square
approaching its asymptotic form. We show t|ψ(x, t)|2 as a wave in Eq. (26), this derivation of the error is applicable
function of x/t. For all t > 0.5 the exact distribution is in- only when the initial time is taken to be the instant when
distinguishable from the asymptotic form; so the graph for the discontinuities exist. At any other time, ∆x does not
t = 0.5 stands for all later times.
exist (because m2 d2t hX̂ 2 i = 2hP̂ 2 i, as in Section IV, and
hP̂ 2 i = ∞.) The divergence of hx̂2 i can easily be seen
VII. FURTHER CONSIDERATIONS for the asymptotic wave function of the initially square
packet from Eq. (28).
The material above is simple enough and sufficiently (c) The relation of the Hermite-Gauss solutions for free
complete to be presented in an elementary course in particle motion to the energy eigenfunctions of the har-
quantum mechanics, but some related topics may be monic oscillator can be further elucidated as follows. Sep-
helpful and may serve as extension material for students. arating the real and imaginary parts of the exponent in
(a) For a more rigorous approach to the short-time χ in Eq. (14) gives
behavior, write the exact form of Eq. (5) as ψ(x, t) = (mτ /π~)1/4 ıtx2 x2
exp(−ıp̄2 t/2m~) ψ(x − p̄t/m, 0) + δψ, where χ(x, t) = p exp( ) exp(− ), (36)
(t − ıτ ) 2τ γ 2 2γ 2
1 −ıp̄2 t
δψ(x, t) = √ exp( )× (29) with γ = [~(t2 + τ 2 )/mτ ]1/2 . Similarly to the derivation
2π~ 2m~
Z ∞ of Eq. (15), one can show that
ı p̄t −ı(p − p̄)2
√
exp[ p(x − )] exp[ ] − 1 φ(p) dp.
−∞ ~ m 2m~ e−ıθ b̂ eıθ = ı 2~mτ eıβ â, (37)
6
√
where â = (x̂/γ+ıγ p̂/~)/ 2 and, as before, θ = tx2 /2τ γ 2 VIII. CONCLUSION
and β = ((t − ıτ )/(t + ıτ ))1/2 . Then [â, ↠] = 1 and
â is the type of operator familiar from the theory of To see the big picture of the free evolution of a wave
the harmonic oscillator. Schrõdinger’s packet χ, with packet, it helps to extend the evolution back to the dis-
b̂ χ = 0, corresponds to the ground state of â, with tant past as well as forward. The centroid hx̂i of the
â exp(−x2 /2γ 2 ) = 0. Then the sequence of states ob- packet moves with constant speed hp̂i/m and we know
tained by successive application of ↠to exp(−x2 /2γ 2 ) is from Section IV that every packet will contract and
the sequence of oscillator states exp(−x2 /2γ 2 )Hn (x/γ) then expand, so that ∆x will take its minimum value
and corresponds to the sequence of solutions of the free ∆min at some time t0 . Far from the minimum, for
SE obtained by successively applying b̂† to χ. |t − t0 | th = m∆min /∆p , ∆x will vary linearly with
The Hermite-Gauss solutions can be readily general- time, from Eq. (3). We also know, from Section III, that
ized to have non-zero values of hx̂i and hp̂i. Instead of for |t − t0 | tx = 2m∆2min /~ the shape of the wave
b̂ χ = 0 leading to the solution for χ(x, t) in Eq. (14), packet will not change, but the scale will change linearly
start with b̂ χ = βχ, where β = mhx̂i − hp̂i(t − ıτ ), and with time (consistent with the linear change in ∆x ). Note
then repeatedly apply b̂† . that th = (tx tp )1/2 , where tp = m~/2∆2p . Thus th is the
It is not surprising that this Hermite-Gauss sequence harmonic mean of tx and tp and tp 6 th 6 tx .
of solutions exists for the free particle; a similar se- Therefore, changes in shape (other than in scale) can
quence can be found for any Hamiltonian that is at most occur only in the waist region around t0 and such changes
quadratic in p̂ and x̂, even when the coefficients of the cannot be rapid over times of the order of tp , if ∆p exists.
terms in the Hamiltonian vary with the time.[8] But even in this waist region, there need not be changes
(d) The spatial derivative of any solution of the free in shape, as exemplified by the Hermite-Gauss packets.
SE is also a solution. Taking successive derivatives of
Schrõdinger’s packet χ in Eq. (14) gives a sequence of
APPENDIX: COMMENT ON MITA’S PAPER
packets that also have the form of a Hermite polynomial
multiplied by the original Gaussian and by a phase factor,
but in this case the probability distributions do change The only example considered in detail by Mita[1] is
shape as they evolve. the square wave packet, as in Eq. (26). We have shown
From Rodrigue’s Eq. (17), and defining χ̄n = ∂xn χ, that in this case, |ψ(x, t)| evolves to approach the form
sin(βx/t)/(βx/t) where β = am/2~, and this packet is
2
x2 not Gaussian. All that has been achieved in Mita’s note is
χ̄n ∝ κn+1 e−κ Hn (κx), (38)
to impose a Gaussian fit to an arbitrary wave packet [by
where κ = [m/2ı~(t − ıτ )]1/2 . But the Fourier transform making a quadratic approximation to the logarithm in
of this combination of a Gaussian and an exponential is Mita’s Eq.(14)]; if the packet is nothing like a Gaussian,
different to that in Eq. (23): the fit will be poor.
Z ∞ For Mita’s parameters m = ~ = a = 1, the exact distri-
1 2 (−ı)n 2 bution differs from the approximate one [from Eq. (28)]
√ e−z Hn (z)e−ıpz dz = √ e−p /4 pn . (39) by a negligible amount for all t > 0.5. Mita’s Fig. 3(c)
2π −∞ 2
shows only that, if you don’t look too closely, (sin x/x)2
The momentum distribution of χ̄n is not a Hermite- is not very different from a Gaussian shape; if the x-axis
Gaussian and therefore these packets change shape as were extended further, the next maximum in the exact
they evolve. The first example in Section VII illustrates distribution would be revealed – it does not decrease in
the case n = 2. relative size for later times.