Regularization (Dimensional, Cutoff)

Core Idea

Regularization is a mathematical procedure that makes divergent loop integrals finite by introducing a regulator parameter. Cutoff regularization imposes a maximum momentum; dimensional regularization continues spacetime to d = 4 - epsilon dimensions where integrals converge. The regulator is removed after renormalization absorbs the divergences into physical parameters.

Explainer

Divergent loop integrals are mathematically meaningless as written -- you cannot extract a finite number from an infinite integral without first making it finite. Regularization is the procedure that accomplishes this by introducing a parameter that controls the divergence. The two most common methods are cutoff regularization and dimensional regularization, each with its own advantages.

Cutoff regularization is the most intuitive: impose a maximum momentum |k| < Lambda on all loop integrals. Every integral becomes finite, and divergences appear as powers of Lambda (quadratic divergences as Lambda², logarithmic as ln Lambda). The physical interpretation is appealing: Lambda represents the energy scale above which the theory may need modification. The drawback is that a hard cutoff breaks Lorentz invariance and can violate gauge invariance, generating spurious terms (like a photon mass) that must be carefully subtracted. It is conceptually useful but technically cumbersome for gauge theories.

Dimensional regularization works by analytically continuing the number of spacetime dimensions from 4 to d = 4 - epsilon. In d dimensions, integrals that diverge at d = 4 become convergent for sufficiently small d, and the divergences reappear as poles in 1/epsilon as d -> 4. The method is purely algebraic: there is no need to interpret non-integer dimensions geometrically. Its great virtue is that it preserves both Lorentz invariance and gauge invariance automatically, making it the standard tool for gauge theory calculations. A notable feature is that power-law divergences vanish in dimensional regularization, leaving only logarithmic divergences as 1/epsilon poles.

After regularization, the divergences are explicit and parameterized. The next step -- renormalization -- absorbs these divergences into redefinitions of the bare parameters (mass, coupling constant, field normalization). The renormalized parameters are then fixed by experiment. The final physical predictions are independent of the regularization scheme: cutoff and dimensional regularization give the same answers for all observables once renormalization conditions are imposed. The regulator is a scaffolding that is removed after the building is complete.