Regularization

The loop integrals of interacting QFT are divergent as written. Before they can be renormalized, the divergences must be made finite and parametrized by a regulator — a temporary modification of the theory that renders every integral finite while tracking the divergence through a controllable parameter. This page surveys the standard schemes and explains why physical results are independent of the choice.

Conventions: $\hslash =c=1$, ${\eta }_{\mu \nu }=\mathrm{diag}(+1,-1,-1,-1)$.

The role of a regulator

A regulator deforms the theory so that loop integrals converge, introducing a parameter (a cutoff $\Lambda$, a dimension $ϵ$, a regulator mass) that recovers the original divergent theory in a limit. The divergences reappear as poles or powers of that parameter. The essential principle:

Physical observables must be independent of the regulator once the theory is renormalized. The regulator is scaffolding — it makes intermediate steps finite, and must disappear from final answers.

A good regulator preserves as many symmetries of the theory as possible (Lorentz invariance, gauge invariance), because a regulator that breaks a symmetry forces extra work to restore it — or signals a genuine anomaly.

Momentum cutoff

The most physical regulator: simply forbid loop momenta above a scale $\Lambda$,

$$\int \frac{d^4{\ell }_E}{(2\pi {)}^4}⟶{\int }_{|{\ell }_E|<\Lambda }\frac{d^4{\ell }_E}{(2\pi {)}^4}.$$

Divergences appear as powers of $\Lambda$: the ${\varphi }^4$ self-energy becomes $\sim {\Lambda }^2$, the vertex $\sim \ln \Lambda$. The cutoff has a clear physical interpretation — $\Lambda$ is the scale beyond which the theory is replaced by new physics — which makes it the natural language of the Wilsonian RG and effective field theory. Its drawback: a hard cutoff breaks Lorentz and gauge invariance, so it is cumbersome for gauge-theory calculations.

Pauli–Villars regularization

Subtract a fictitious heavy field of mass $M$ from each propagator:

$$\frac{1}{{\ell }^2-m^2}⟶\frac{1}{{\ell }^2-m^2}-\frac{1}{{\ell }^2-M^2}=\frac{m^2-M^2}{({\ell }^2-m^2)({\ell }^2-M^2)}.$$

The extra $1/{\ell }^2$ falloff at large $\ell$ makes the integral converge; the divergence reappears as $\ln M$ (or powers) as $M\to \infty$. Pauli–Villars preserves Lorentz and (abelian) gauge invariance, making it cleaner than a cutoff for QED. It is awkward for non-abelian theories, where a mass term for the regulator field is not gauge invariant.

Dimensional regularization — the workhorse

The modern default. Continue the number of spacetime dimensions from $4$ to $d=4-ϵ$ (with $ϵ$ complex), where loop integrals converge for suitable $\mathrm{Re}ϵ$:

$$\int \frac{d^4\ell }{(2\pi {)}^4}⟶{\mu }^{ϵ}\int \frac{d^d\ell }{(2\pi {)}^d},$$

with $\mu$ an arbitrary renormalization scale inserted to keep couplings at their four-dimensional mass dimension. The master integral is exactly computable:

$$\int \frac{d^d{\ell }_E}{(2\pi {)}^d}\frac{1}{({\ell }_E^2+\Delta {)}^n}=\frac{1}{(4\pi {)}^{d/2}}\frac{\Gamma (n-d/2)}{\Gamma (n)}{\Delta }^{d/2-n}.$$

Ultraviolet divergences appear as poles $1/ϵ$ from the $\Gamma$ function at $d\to 4$; for the logarithmic vertex,

$$\ln \Lambda ⟷\frac{1}{ϵ}-\frac{1}{2}\ln \Delta +\text{const}.$$

Its decisive advantages:

• Preserves Lorentz and gauge invariance manifestly (the regulator is just the dimension), which is why it is essential for non-abelian gauge theories.
• No power divergences: pure power-law divergences (like ${\Lambda }^2$) are set to zero in dim reg, so only the physical logarithmic running survives explicitly — clean, but it obscures the hierarchy problem, which is about those quadratic terms.
• Introduces the scale $\mu$ that becomes the running scale of the renormalization group.

The one subtlety: ${\gamma }^5$ and the chiral anomaly are dimension-specific, so dim reg needs care for chiral theories ('t Hooft–Veltman scheme).

Minimal subtraction ($\overline{\text{MS}}$)

Dimensional regularization pairs with a subtraction scheme that defines which finite parts are absorbed along with the $1/ϵ$ poles:

• MS (minimal subtraction): subtract only the $1/ϵ$ pole.
• $\overline{\text{MS}}$ (modified MS): also subtract the universal $-{\gamma }_E+\ln 4\pi$ that always accompanies the pole. This is the near-universal choice in modern high-precision QCD and electroweak calculations.

The scheme is part of the definition of the renormalized couplings; different schemes give different intermediate numbers but identical physical predictions once related by finite renormalization.

Lattice regularization

Replace continuous spacetime by a discrete Euclidean lattice of spacing $a$; the shortest wavelength is $2a$, so momenta are cut off at $\sim \pi /a$. This is the only regulator that is fully nonperturbative — it defines the path integral without reference to Feynman diagrams — and it is the basis of lattice field theory. Its cost: it breaks continuous Lorentz/rotational symmetry (recovered as $a\to 0$) and complicates chiral fermions (the doubling problem).

Comparison

Summary

• A regulator makes divergent loops finite, at the cost of a parameter that carries the divergence.
• Physics must be regulator-independent after renormalization.
• Dimensional regularization ($d=4-ϵ$, poles $1/ϵ$) is the workhorse: it preserves gauge invariance and introduces the RG scale $\mu$.

Where this leads

• Absorbing the poles into counterterms: renormalization.
• The scale $\mu$ and running: the renormalization group.
• The cutoff viewpoint: Wilsonian RG and effective field theory.

References

• Peskin & Schroeder, An Introduction to Quantum Field Theory, Ch. 7.5, 10.
• 't Hooft & Veltman, Nucl. Phys. B 44, 189 (1972).
• Weinberg, The Quantum Theory of Fields, Vol. 1, Ch. 11–12.
• Srednicki, Quantum Field Theory, Ch. 14.