53
Part One · Quantum mechanics

Path integrals

A thrown ball follows one path. Quantum mechanics, as Richard Feynman reformulated it in 1948, says something stranger: to get from here to there, a particle takes every path at once, straight, curved, looping, absurd. Each path contributes a little arrow whose angle is its action divided by ℏ\hbar. Add all the arrows and you get the quantum amplitude. And when ℏ\hbar is small compared with the action, almost all the arrows cancel, except those near one special path: Newton’s.

On the plate: many paths between two points, each carrying a spinning arrow.

Slicing time

Start from the propagator K(xb,tb;xa,ta)=⟨xb∣e−iH(tb−ta)/ℏ∣xa⟩K(x_b, t_b; x_a, t_a) = \langle x_b|e^{-iH(t_b - t_a)/\hbar}|x_a\rangle, the amplitude to go from xax_a to xbx_b. Chop the time into NN short steps and insert a complete set of positions between each pair (Chapter 28). For a short step, the exponential splits into kinetic and potential parts, as in the solver of Chapter 31, and each piece contributes eiL Δt/ℏe^{iL\,\Delta t/\hbar}, with LL the Lagrangian of Chapter 12. Multiply them all:

K=∫D[x(t)]  eiS[x]/ℏ,S=∫L dtK = \int\mathcal{D}[x(t)]\;e^{iS[x]/\hbar}, \qquad S = \int L\,dt

The integral is over every function x(t)x(t) joining the endpoints. Paul Dirac noticed in 1933 that the short-time amplitude goes as eiS/ℏe^{iS/\hbar}; Feynman made it a complete formulation. Slice the time into 8 steps: each path becomes a zigzag, and the path integral becomes an ordinary integral over the intermediate positions. Established

Adding the arrows

Take a ball thrown upward and consider a family of paths: the classical arc plus a bulge of size aa. Each has action S(a)=Scl+c a2S(a) = S_{\text{cl}} + c\,a^2: the classical path is where the action is stationary. Add their arrows one by one, from large negative aa to large positive, head to tail. The plate draws the chain, which curls into a double spiral, the Cornu spiral of optics.

Set ℏ\hbar to 0.30 (in units of the action’s curvature). Far from a=0a = 0, neighbouring paths differ in phase by a lot and their arrows wind in tight circles that go nowhere. Near a=0a = 0 the phase hardly changes, the arrows line up, and they make the whole length of the chain. The paths that matter, highlighted on the left, are those within about ℏ/c\sqrt{\hbar/c} of the classical one. Shrink ℏ\hbar, and the band closes onto Newton’s path.

Newton, from stationary phase

This is the stationary-phase argument. As ℏ→0\hbar \to 0, the only paths that survive are those where small changes do not change SS to first order: δS=0\delta S = 0. That is exactly the principle of least action of Chapter 12, whose solutions obey the Euler–Lagrange equations, Newton’s laws. Classical mechanics is not replaced by quantum mechanics; it is the part of it where one path dominates the sum.

The next correction, the spread of nearby paths around the classical one, gives the WKB approximation of Chapter 47. Every semiclassical formula is a path integral evaluated near its stationary points.

The free particle, exactly

For a free particle every slice is a Gaussian integral, and the path integral can be done exactly:

K(x,T;0,0)=m2πiℏT  exp⁡ ⁣(imx22ℏT)K(x, T; 0, 0) = \sqrt{\frac{m}{2\pi i\hbar T}}\;\exp\!\left(\frac{imx^2}{2\hbar T}\right)

The exponent is iScl/ℏiS_{\text{cl}}/\hbar, the classical action of the straight path. The plate shows the real part of KK at time 1.0: fringes that get closer together farther out, because reaching a distant point in a fixed time needs high momentum and short wavelength. Integrating KK against a Gaussian start reproduces the spreading packet of Chapter 32 exactly. The harmonic oscillator can be done exactly too, and every packet in it returns after one period, as Chapter 31’s solver showed.

Imaginary time

Replace time tt by −iτ-i\tau. The oscillating eiS/ℏe^{iS/\hbar} becomes a real, decaying weight e−SE/ℏe^{-S_E/\hbar}, and the free path integral becomes the mathematics of diffusion (Chapter 5): random walks, as on the plate. Paths that return to their start after imaginary time βℏ\beta\hbar add up to

Z=Tr e−βH=∮D[x]  e−SE[x]/ℏZ = \mathrm{Tr}\,e^{-\beta H} = \oint\mathcal D[x]\;e^{-S_E[x]/\hbar}

the partition function of statistical mechanics (Chapter 19), with β=1/kBT\beta = 1/k_BT. Quantum mechanics in imaginary time is statistical mechanics in one more dimension. Because the weights are real and positive, computers can sample them at random, and this is how lattice QCD computes the proton’s mass from the theory of quarks and gluons, in Part Five. Established

What you now have

The path integral from time-slicing, the sum of phasors and the Cornu spiral, stationary phase and the principle of least action, the exact free propagator, and the link between imaginary time and statistical mechanics. In Part Five, fields replace paths and the same sum becomes the starting point for quantum field theory and Feynman diagrams.

· · ·

Three questions to carry with you.

Why do far-from-classical paths contribute little?tap
Their neighbours have very different phases, so their arrows cancel.
How does classical mechanics emerge?tap
As ħ → 0, only paths with δS = 0 survive the cancellation: the principle of least action.
What does imaginary time turn the path integral into?tap
A real, positive sum over random walks: the partition function of statistical mechanics.

And one to work out. An electron travels freely 0.1 mm in 1 µs, at 100 m/s. Its classical action divided by ℏ\hbar is m Δx2/(2ℏT)m\,\Delta x^2/(2\hbar T). What is it, in radians? (m=9.109×10−31m = 9.109\times10^{-31} kg, ℏ=1.0546×10−34\hbar = 1.0546\times10^{-34} J s.)

S/ħ =

Sources: P. A. M. Dirac, Phys. Z. Sowjetunion 3, 64 (1933). R. P. Feynman, Rev. Mod. Phys. 20, 367 (1948). R. P. Feynman and A. R. Hibbs, Quantum Mechanics and Path Integrals, emended edition by D. F. Styer (2010). L. S. Schulman, Techniques and Applications of Path Integration. Checked by code: Cornu sums for S = c a² converge to √(πħ/c)·e^(iπ/4); free-propagator convolution reproduces σ(t) of Chapter 32; S/ħ for the electron example (0.1 mm in 1 µs) = 43.2; a 1 g ball on the same trip, 4.7×10²⁸.

Next · Chapter 54
Interpretations: Copenhagen, many-worlds, Bohm, QBism, collapse models

Every interpretation predicts the same numbers, except one family. It is being tested deep underground in Italy.

Turn the page