Where Does the Spread σ(t) of a Wave Packet Come From?

— Derived using only the same tools as Ehrenfest’s theorem (for second- and third-year undergraduates)

In the simulator of a particle thrown upward, the “blur” around the trajectory is drawn with the formula σ(t)=σ02+(ℏt2mσ0)2. This page derives, step by step, where this formula comes from. The tools we use are exactly the same as those used to derive Ehrenfest’s theorem, and no new mathematics is needed.

The conclusion first If the potential is at most linear in position (a uniform gravitational field is exactly this case), the momentum spread σp is constant in time, and only the position spread grows with time, by just the part due to “the velocity not being fixed” (this assumes no initial correlation between position and momentum; with a correlation, the packet may shrink for a while before it spreads). Moreover, the force term cancels out cleanly in the course of the derivation, so gravity has no influence at all on how the packet spreads.

0. Setup

A particle of mass m moves under a uniform force. The Hamiltonian is

H= p22m +V(x) , V(x) =Fx (F is a constant; it is the slope of V, and the force is −F)

For the thrown-up problem, read x as the vertical direction y, so that V=mgy, that is, F=mg (the force is −∂V∂y=−mg, pointing downward). The horizontal direction is the special case F = 0, so it is included in the discussion below as it stands.

What we want to measure as the “blur” is the standard deviation of the position.

σx(t)2 ≡ ⟨x2⟩ − ⟨x⟩2

1. Intuition First (the 30-Second Version)

A wave packet has neither its position nor its momentum fixed to a single value. The uncertainty principle places a lower bound on the momentum spread that a wave packet prepared with width σ0 must have. For the minimum-uncertainty Gaussian wave packet, in which equality holds, this gives

Δp= ℏ2σ0 ⇒ Δv= Δpm = ℏ2mσ0

That is, the packet has a velocity spread of just this amount. If the velocity is uncertain by Δv, then after t seconds the position is uncertain by an additional Δv⁢t. The original width σ0 and this additional part arise from independent causes, so the standard deviations add as the sum of their squares, giving

σ(t)2 = σ02 + (Δv⁢t)2 = σ02 + (ℏt2mσ0)2

That already gives the answer. However, the questions “is a sum of squares really correct?” and “does gravity really have no effect?” cannot be answered by this argument. Doing that properly is the job of the next section.

2. A Careful Derivation — Using Only Ehrenfest’s Tools

2.1 We use just this one equation

For an operator A^ with no explicit time dependence, the time evolution of the expectation value is

(1) ddt ⟨A^⟩ = iℏ ⟨ [H^, A^] ⟩

Applying this to A^ = x^ and p^ gave exactly Ehrenfest’s theorem. Applying the same equation to x^2, p^2, x^p^+p^x^ as well is what we do this time.

2.2 The commutators we need

[x^,p^]=iℏ alone gives the following (please work them out by hand to check).

A^ [H^,A^] Relation used
x^ −iℏ⁢p^m [p^2,x^]=−2iℏp^
p^ +iℏF [x^,p^]=iℏ
x^2 −iℏ⁢x^p^+p^x^m [p^,x^2]=−2iℏx^
p^2 +2iℏFp^ [x^,p^2]=2iℏp^
x^p^+p^x^ −2iℏ⁢p^2m+2iℏFx^ Combine the two rows above

2.3 The five equations

Substituting these into Eq. (1) gives the following.

(2) d⟨x⟩dt = ⟨p⟩m d⟨p⟩dt = −F  ← Ehrenfest’s theorem d⟨x2⟩dt = ⟨xp+px⟩m d⟨p2⟩dt = −2F⟨p⟩ d⟨xp+px⟩dt = 2⟨p2⟩m−2F⟨x⟩

This is the crux of the matter this time. No higher-order quantity such as ⟨x3⟩ appears on the right-hand side. Since these five equations close among themselves, they can be solved just as they stand (a stroke of luck that occurs only when V is at most quadratic; more on this later).

2.4 Rewriting in terms of “fluctuations”

What we want to know is the spread around the mean, so we define the following three quantities.

σx2 = ⟨x2⟩−⟨x⟩2  … spread in position σp2 = ⟨p2⟩−⟨p⟩2  … spread in momentum C = 12⟨xp+px⟩−⟨x⟩⟨p⟩  … correlation between position and momentum

Differentiating with Eq. (2) (the terms in ⟨x⟩ and ⟨p⟩ cancel exactly):

(3) dσp2dt = −2F⟨p⟩ − 2⟨p⟩⁢ (−F) = 0
(4) dCdt = ⟨p2⟩m −F⟨x⟩ − ⟨p⟩2m +F⟨x⟩ = σp2m
(5) dσx2dt = 2Cm
⚠ Please look at this In both Eqs. (3) and (4), the terms containing F cancel out exactly. That is, no force appears at all in the equations for the fluctuations. Whether or not there is gravity, the packet spreads in the same way. Here we see that the force acts only on ⟨x⟩ and ⟨p⟩ (that is, on the motion of the center).

2.5 Solving the equations

Just integrate in the order Eqs. (3) → (4) → (5).

(6) σp2(t) = σp02  (constant) C(t) = C0+σp02tm σx2(t) = σ02+ 2C0tm+ σp02t2m2

Finally, we fix the initial state. For the minimum-uncertainty Gaussian wave packet (an ordinary Gaussian with a real envelope), there is no initial correlation between position and momentum, that is, C0=0, and σp0=ℏ2σ0 holds as well. Substituting these into Eq. (6) gives

(7) σ(t)= σ02+ (ℏt2mσ0)2

The formula expected from the intuition of Section 1 has now been derived properly. The reason we may “add them as a sum of squares” is that C0=0 in Eq. (6). That is the answer. Conversely, for a wave packet that is already spreading or shrinking from the start (C0≠0), the cross term in the middle remains, and the packet may shrink for a while before spreading.

3. Reading the Formula

3.1 When does the spreading start?

While t is small, Eq. (7) stays almost at σ0, and then, beyond a certain time, grows linearly. That crossover time is

τ= 2mσ02ℏ (at this time σ=2⁢σ0 )

The order of magnitude of this τ decides everything.

Situationσ0 τ=2mσ02/ℏ
Electron1 nm1.7×10−14 s (17 fs)
Proton1 nm3.2×10−11 s
1 g object1 µm1.9×1019 s ≈ 40 times the age of the universe

If a 1 g object is positioned to within 1 µm, it takes 40 times the age of the universe before its wave packet visibly spreads. This is the real reason why “everyday objects appear to obey classical mechanics.” It is not that quantum mechanics is wrong; the spreading is simply far too slow.

3.2 Choosing σ0 involves a trade-off

The two terms of Eq. (7) work in opposite directions as functions of σ0. If the packet is prepared narrow, the first term is small but the second term is large; if it is prepared wide, the reverse holds. Therefore, there is a way of preparing the packet that minimizes the spread at a given time T. Differentiating with respect to σ02 and setting the result to 0 gives

(8) σ0best = ℏT2m ⇒ σ(T)min = ℏTm

In that case, τ=T exactly. In other words, the optimal way to prepare the packet is to set it up so that the “time when it starts to spread” coincides with the “time when it lands.” In the figure below, vary σ0 to check this for yourself.

Left: how σ(t) grows (the dotted lines are the two limits, σ0 and Δv·t). Right: how the spread at landing, σ(T), changes with σ0 (log–log). The bottom of the valley is given by Eq. (8).

3.3 What happens when the mass is changed?

From Eq. (8), the relation σ(T)∝1m follows. Even if the mass is made 106 times smaller, the blur grows only by a factor of 103. Because the dependence is this gentle, the enormous mass difference between everyday objects and electrons translates directly into a difference in whether something “looks classical.”

Particlem [kg] σ(T) σ(T)/R Appearance
Baseball0.1451.0×10−17 m 10−16Completely classical (smaller than an atomic nucleus)
Microparticle10−91.2×10−13 m 10−12Completely classical (smaller than a single atom)
C601.2×10−243.6 µm 3.5×10−5Classical (the blur is 10−5 of the flight distance)
Electron9.1×10−314.1 mm 0.040The trajectory is visible but blurred
10−6 × electron9.1×10−374.1 m 40No longer a trajectory

For v0 = 1 m/s and θ = 45°, T = 144 ms and the range is R = 10.2 cm. σ0 is the optimal value from Eq. (8).

4. Relation to Ehrenfest’s Theorem — The Most Interesting Part

The condition derived in lecture notes (11) was that, if V varies so slowly over the extent of the wave packet that its change can be neglected (exactly, if the force can be written as a polynomial of at most first degree), the motion of the wave packet follows Newton’s equation of motion. This means that the approximation

⟨ dVdx (x) ⟩ ≃ dVdx (⟨x⟩)

is required (the notes write the coordinate as x in one dimension, but here we read it as y, the direction of gravity). In a uniform gravitational field, however, dVdy=mg is constant everywhere, so this approximation is not even an approximation. The two sides are equal from the start. In other words,

In a uniform gravitational field, Ehrenfest’s theorem is not an approximation but exact d⟨p⟩dt=−mg holds exactly. Furthermore, using ⟨p⟩=md⟨y⟩dt we obtain d2⟨y⟩dt2=−g, so the acceleration of the mean position does not depend on the mass. Therefore no matter how light the particle is, its mean position lies exactly on the classical parabola. What changes when the mass is reduced is not the “trajectory” but only the “blur around the trajectory.”

It is not that the trajectory of a light particle “bends” or “deviates.” The center follows Newton’s law to the very end, and it is the cloud around that center that becomes thicker. When you lower the mass in the simulator, this is why the red parabola keeps staying in the middle of the cloud.

Why gravity does not affect the spread

There are two ways to say it, and both mean the same thing.

What happens for a general V? The five equations (2) closed because V was at most quadratic. For example, if V∝x3, then d⟨p2⟩dt contains a term ⟨x2p+⋯⟩, and the equation for that term contains still higher-order terms, and so on without end, so the set never closes. We therefore need an approximation: since the spread is small, we may assume ⟨V′(x)⟩≃V′(⟨x⟩) to hold. This is precisely the condition of lecture notes (11). Only when the potential can be written as a polynomial of at most second degree in position (a free particle, a uniform force, a harmonic oscillator, and sums of these) do we have a special case in which no approximation is needed.

5. Checking Numerically

The derivation above was done by hand, so we have checked it independently by numerical calculation. We solved the Schrödinger equation directly with the split-operator method (the split-step Fourier method), measured σx and σp from the wave function, and compared them with Eq. (6). We deliberately chose m≠1, g≠0, and an initial correlation C0≠0 (a chirped wave packet).

t σx (numerical) σx from Eq. (6) Difference σp (numerical)
0.50.847555170.84755517 2×10−110.96168833
1.01.162033631.16203363 7×10−110.96168833
2.01.853411661.85341166 2×10−100.96168833
3.02.571854952.57185495 3×10−100.96168833

The two agree to 10 digits. Also look at the rightmost column. σp has not changed through the 8th decimal place. Although a force is acting, the spread of the momentum does not change at all: Eq. (3) can be seen here exactly as it stands.

6. Summary

  1. Applying Eq. (1) for expectation values also to x2, p2, and xp+px gives the three equations, Eqs. (3), (4), (5), that close on the fluctuations alone.
  2. No force appears in those equations. That is why gravity does not affect the spread.
  3. Solving them gives σ(t)2=σ02+(ℏt2mσ0)2. Its content is the sum of the squares of “the initial width” and “the part due to the undetermined velocity × time.”
  4. The spreading starts at τ=2mσ02/ℏ — for a 1 g object, that is 40 times the age of the universe.
  5. In a uniform gravitational field, Ehrenfest’s theorem is exact, so the center follows the classical parabola regardless of the mass. The mass changes only the thickness of the blur around it.

Mochizuki Research Group, Tokyo University of Science / From lecture notes (11), “Ehrenfest’s Theorem” (in Japanese). The numerical verification used the split-step Fourier method (natural units, dt = 10−4, grid of 8192 points). The formulas are typeset in the browser’s native MathML (no external libraries and no network communication at all).