A hard momentum cutoff Lambda makes all loop integrals finite by restricting |k| < Lambda. Why is this simple and intuitive approach problematic for gauge theories like QED?
ABecause a momentum cutoff is not Lorentz invariant and can break gauge invariance — the cutoff treats spatial and temporal momenta differently and can generate gauge-non-invariant terms that would give the photon an unphysical mass
BBecause a cutoff makes the integrals too convergent, losing physical information
DBecause the cutoff Lambda has units of energy, which is inconsistent with natural units
A sharp cutoff |k| < Lambda breaks Lorentz invariance (it is not invariant under boosts) and can violate gauge invariance (Ward identities may not be satisfied). In QED, a cutoff can generate a term proportional to Lambda^2 A_mu A^mu, which is a photon mass term forbidden by gauge invariance. While the cutoff gives correct physical results if you carefully subtract these artifacts, it is technically cumbersome. Dimensional regularization avoids these problems entirely because it preserves Lorentz invariance and gauge invariance by construction.
Question 2 Multiple Choice
In dimensional regularization, you compute loop integrals in d = 4 - epsilon dimensions. Divergences appear as poles in 1/epsilon. What does 'spacetime with a fractional number of dimensions' actually mean mathematically?
AYou literally work in a spacetime with 3.99 spatial dimensions
BIt is a formal analytic continuation: the rules of integration (Gaussian integrals, angular integrals) are defined as algebraic functions of d, and you evaluate them at d = 4 - epsilon without needing a geometric interpretation of non-integer dimensions
CYou compactify the extra dimensions on a very small manifold
DYou add a small imaginary part to the number of dimensions
Dimensional regularization is an algebraic technique, not a geometric one. Key integration formulas (like the d-dimensional Gaussian integral and the area of the d-dimensional unit sphere) are well-defined analytic functions of d for any complex value. You evaluate Feynman integrals using these formulas, treat d as a continuous parameter, and take d -> 4 at the end. The divergences manifest as poles at d = 4 (i.e., 1/epsilon poles when d = 4 - epsilon). No one needs to visualize 3.99-dimensional space — the method is purely algebraic.
Question 3 True / False
Dimensional regularization automatically sets power-law divergences (like quadratic divergences proportional to Lambda^2) to zero, unlike cutoff regularization.
TTrue
FFalse
Answer: True
In dimensional regularization, integrals like integral d^dk / (k^2 + m^2) that would give quadratic divergences with a cutoff evaluate to expressions involving only logarithmic poles (1/epsilon) and finite terms — there are no power-law terms. This is because scaleless integrals (those with no mass scale in the integrand) vanish in dimensional regularization by analytic continuation. This is both a virtue (it simplifies calculations and preserves gauge invariance) and a subtlety (you can miss physics associated with quadratic sensitivity to high scales, such as the hierarchy problem). Cutoff regularization makes quadratic divergences explicit.
Question 4 Short Answer
Explain why the choice of regularization scheme should not affect physical predictions, and what property of the theory guarantees this.
Think about your answer, then reveal below.
Model answer: Physical predictions (cross sections, decay rates, mass ratios) are independent of the regularization scheme because the regulator is an intermediate mathematical device that is removed after renormalization. The divergent parts are absorbed into the bare parameters (mass, charge, wave function normalization) regardless of how the divergence is parameterized (as Lambda^2 with a cutoff or as 1/epsilon with dimensional regularization). Renormalization conditions — measured values of physical quantities like the electron mass and charge at a specific scale — fix the finite parts of the renormalized parameters. Once these conditions are imposed, all other predictions are uniquely determined and regulator-independent. This is guaranteed by the renormalizability of the theory.
If two regularization schemes gave different physical predictions, at least one would disagree with experiment, and the theory would be ambiguous. Renormalizability ensures this does not happen: the divergent structure is universal (the same poles appear in any scheme), and the finite parts are fixed by experiment. Different schemes may differ in intermediate steps but must agree on all observables.