69
Part Two · Quantum fields and particles

Loops and divergences

In the late 1930s, physicists computed the first correction to the electron’s properties from QED and got infinity. Not a large number: infinity. For a decade it looked as if quantum field theory was broken. It wasn’t. The infinities turned out to sit only in quantities that can never be measured on their own, and to cancel in every prediction that can. Learning to see that was the hardest step in twentieth-century physics, and it starts with one loop.

On the plate: two predictions that each depend on an arbitrary cutoff. Move the cutoff. Their difference does not move.

Where infinity comes from

A closed loop in a Feynman diagram carries a momentum kk that nothing fixes, so it is integrated over all values (Chapter 64). At large kk, propagators fall like 1/k21/k^2 or 1/k1/k, while the integral ∫d4k\int d^4k grows like k4k^4. If the powers don’t win, the integral diverges at high momentum: an ultraviolet divergence, from arbitrarily short distances.

Count powers and you get the superficial degree of divergence DD. In QED, D=4−Eγ−32EeD = 4 - E_\gamma - \tfrac32E_e, where EγE_\gamma and EeE_e count external photons and electrons. Only diagrams with few external lines can diverge: the photon’s self-energy (D=2D = 2), the electron’s (D=1D = 1), and the vertex (D=0D = 0). Three kinds of infinity, no more, however many loops you draw. Established

Taming the integrals

To handle a divergent integral, first make it finite, then see what survives. One way is a cutoff Λ\Lambda: stop at momenta above it. The modern way, introduced by Gerard ’t Hooft and Martinus Veltman in 1972, is dimensional regularization: compute in d=4−2ϵd = 4 - 2\epsilon dimensions, where the integral converges, and watch the answer as ϵ→0\epsilon \to 0. Divergences appear as poles, from the Gamma function’s Γ(ϵ)≈1/ϵ−γ\Gamma(\epsilon) \approx 1/\epsilon - \gamma. It respects gauge symmetry, which a crude cutoff does not.

Two tools do most of the work. Feynman parameters combine denominators, 1AB=∫01dx[xA+(1−x)B]2\frac{1}{AB} = \int_0^1\frac{dx}{[xA + (1-x)B]^2}, and a shift of kk then makes the integral spherically symmetric, and doable.

The vacuum polarization loop

A photon can briefly become an electron–positron pair and back. That loop, Π(q2)\Pi(q^2), diverges logarithmically. With a cutoff it reads, roughly, Π(q2)≈α3πln⁡Λ2m2\Pi(q^2) \approx \frac{\alpha}{3\pi}\ln\frac{\Lambda^2}{m^2} plus a finite function of q2q^2. Move the cutoff to 10 (log₁₀ of Λ/m\Lambda/m).

The plate shows Π\Pi at two momenta, Q=10mQ = 10m and Q=1000mQ = 1000m. Both slide with the cutoff. Their difference stays fixed, at α3πln⁡(104)=0.0071\frac{\alpha}{3\pi}\ln(10^4) = 0.0071 for large QQ. Measurable things, like how the strength of the force changes between two distances, are differences of this kind. The infinity is the same at every momentum, so it never enters a comparison.

The clue in the infinity

The divergent part of the photon loop has exactly the form of a term already in the Lagrangian, FμνFμνF_{\mu\nu}F^{\mu\nu}. So it can be absorbed into a redefinition of the electron’s charge. The divergent parts of the electron loops absorb into its mass and the normalization of its field. Nothing new is needed. The charge you write in the Lagrangian, the “bare” charge, is not the charge you measure, and only the measured one appears in predictions. That is renormalization, the next chapter. The infinity was telling us that the bare parameters were never observable.

What you now have

Ultraviolet divergences from loops, the superficial degree of divergence and QED’s three divergent functions, cutoffs and dimensional regularization, Feynman parameters, and the vacuum polarization loop with its cutoff-independent differences.

· · ·

Three questions to carry with you.

Why do loop integrals diverge?tap
The loop momentum is unconstrained, and at large k the integrand does not fall fast enough to beat d⁴k.
Why is dimensional regularization preferred?tap
It keeps gauge symmetry, which simple cutoffs break.
Why doesn’t the cutoff affect predictions?tap
It enters only as a constant that is the same at every momentum, absorbed into the measured charge, mass and field normalization.

And one to work out. Using D=4−Eγ−32EeD = 4 - E_\gamma - \tfrac32E_e, what is the superficial degree of divergence of the electron self-energy (two external electrons, no external photons)?

D =

Sources: G. ’t Hooft and M. Veltman, Nucl. Phys. B 44, 189 (1972). M. D. Schwartz, Quantum Field Theory and the Standard Model, ch. 16. M. E. Peskin and D. V. Schroeder, ch. 7. Checked by code: renormalized one-loop vacuum polarization at Q = 10, 100, 1000 m equals (α/3π)(ln Q²/m² − 5/3) to 10⁻⁶ at large Q; difference between Q = 10m and 1000m fixed.

Next · Chapter 70
Renormalization

The electron's charge depends on how closely you look: 1/137 at long range, about 1/128 at the Z mass.

Turn the page