Dimensional Regularization and Minimal Subtraction

Dimensional regularization defines loop integrals by analytic continuation from a complex spacetime dimension, conventionally $d=4-2\epsilon$. Minimal subtraction then defines renormalized parameters by removing only the Laurent poles in $\epsilon$; modified minimal subtraction, $\overline{\rm MS}$, removes the recurring combination $1/\epsilon-\gamma_E+\ln4\pi$. The regulator and the subtraction scheme are distinct choices.

The method is powerful because it preserves translation and Lorentz covariance of loop integrals and turns logarithmic ultraviolet sensitivity into poles whose residues are local. It is also easy to misuse. The scale factor in the measure, the $4\pi$ and Euler–Mascheroni constants, the ultraviolet or infrared origin of every pole, and the continuation of chiral or intrinsically four-dimensional objects must all be stated.

Required background. QFT Regulator Families and Their Tradeoffs supplies the regulator decision and symmetry qualifications. Dimensional Regularization as an Amplitude Tool supplies Wick rotation, tensor reduction, and amplitude-level continuation.

Helpful background. One-Loop Integral Families and Analytic Functions is useful for branch choices and analytic continuation after the ultraviolet subtraction has been defined.

Continuation in dimension introduces a renormalization scale

The Euclidean integration measure has mass dimension $d$. To keep a coupling expressed in its four-dimensional units, introduce a scale $\mu$. For the logarithmic one-loop integral used below, write

$$\int_k \equiv \mu^{2\epsilon} \int\frac{d^d k_E}{(2\pi)^d}, \qquad d=4-2\epsilon.$$

The power of $\mu$ is fixed by dimensional analysis. In four-dimensional quartic scalar theory,

$$\lambda_0 = \mu^{2\epsilon}Z_\lambda\lambda,$$

whereas a gauge or Yukawa coupling is conventionally written with $\mu^\epsilon$. This scale is not a hard momentum boundary. It is the coordinate scale needed to compare renormalized parameters after analytic continuation.

The continuation is defined by analytic formulae for rotationally invariant integrals and tensor contractions. A central Euclidean identity is

$$\mu^{2\epsilon} \int\frac{d^d\ell}{(2\pi)^d} \frac{1}{(\ell^2+\Delta)^\alpha} = \frac{\mu^{2\epsilon}}{(4\pi)^{d/2}} \frac{\Gamma(\alpha-d/2)}{\Gamma(\alpha)} \Delta^{d/2-\alpha},$$

initially in a convergence domain and elsewhere by analytic continuation. Dimensional regularization is the collection of these continued rules, not the instruction to treat a literal noninteger-dimensional spacetime as an experimentally accessible geometry. Collins gives a systematic construction and states the properties required for momentum shifts, scaling, and tensor algebra Collins 1984/2023, ch. 4, pp. 62–87.

Laurent expansion of the logarithmic master integral

Set $\alpha=2$ and $d=4-2\epsilon$. Then

$$\begin{aligned} J(\Delta) &\equiv \mu^{2\epsilon} \int\frac{d^d\ell}{(2\pi)^d} \frac{1}{(\ell^2+\Delta)^2}\\ &= \frac{1}{(4\pi)^2} \Gamma(\epsilon) \left(\frac{4\pi\mu^2}{\Delta}\right)^\epsilon. \end{aligned}$$

Expand all three sources of finite terms:

$$\begin{aligned} \Gamma(\epsilon) &=\frac1\epsilon-\gamma_E+\mathcal O(\epsilon),\\ (4\pi)^\epsilon &=1+\epsilon\ln4\pi+\mathcal O(\epsilon^2),\\ \left(\frac{\mu^2}{\Delta}\right)^\epsilon &=1+\epsilon\ln\frac{\mu^2}{\Delta} +\mathcal O(\epsilon^2). \end{aligned}$$

Multiplication gives

$$J(\Delta) = \frac{1}{16\pi^2} \left[ \frac1\epsilon -\gamma_E +\ln4\pi +\ln\frac{\mu^2}{\Delta} \right] +\mathcal O(\epsilon).$$

It is convenient to define

$$\frac{1}{\bar\epsilon} \equiv \frac1\epsilon-\gamma_E+\ln4\pi.$$

The bar belongs to the whole pole combination; it is not a second regulator. Losing any one of these finite terms changes the scheme.

Massive nonexceptional scalar bubble

Consider

$$B(p^2;m^2) = \mu^{2\epsilon} \int\frac{d^d k}{(2\pi)^d} \frac{1}{(k^2+m^2)((k+p)^2+m^2)}$$

in Euclidean signature, with $m>0$. Feynman parametrization and the shift $\ell=k+(1-x)p$ give

$$\frac{1}{AB} = \int_0^1dx\, \frac{1}{[xA+(1-x)B]^2},$$

and therefore

$$B(p^2;m^2) = \int_0^1dx\,J(\Delta_x), \qquad \Delta_x=m^2+x(1-x)p^2.$$

Substitution yields

$$B(p^2;m^2) = \frac{1}{16\pi^2} \left[ \frac{1}{\bar\epsilon} -\int_0^1dx\, \ln\frac{\Delta_x}{\mu^2} \right] +\mathcal O(\epsilon).$$

Because $m>0$ and the Euclidean momentum is nonexceptional, the pole is ultraviolet. It is independent of $p^2$ and is therefore a local two-point ambiguity. The parameter integral contains the finite momentum dependence. Analytic continuation to Minkowski kinematics adds the appropriate $-i0$ and branch structure, but does not change the local ultraviolet pole.

MS and modified MS are finite-distance schemes

Let $K$ denote the operation that selects the counterterm part of the Laurent series. For the bubble,

$$K_{\rm MS}B = \frac{1}{16\pi^2}\frac1\epsilon,$$

whereas

$$K_{\overline{\rm MS}}B = \frac{1}{16\pi^2}\frac{1}{\bar\epsilon}.$$

The corresponding renormalized integrals are

$$\begin{aligned} B_{\rm MS}(p^2;\mu) &=\frac{1}{16\pi^2} \left[ -\gamma_E+\ln4\pi -\int_0^1dx\, \ln\frac{\Delta_x}{\mu^2} \right],\\ B_{\overline{\rm MS}}(p^2;\mu) &=-\frac{1}{16\pi^2} \int_0^1dx\, \ln\frac{\Delta_x}{\mu^2}. \end{aligned}$$

Thus MS and $\overline{\rm MS}$ differ by the finite local constant

$$\frac{-\gamma_E+\ln4\pi}{16\pi^2}.$$

Equivalently, define

$$\bar\mu^2 \equiv 4\pi e^{-\gamma_E}\mu^2.$$

An MS expression written with $\mu$ can then be rewritten in the same form as an $\overline{\rm MS}$ expression using $\bar\mu$. Authors often call the modified scale simply $\mu$. The label alone is therefore insufficient: state whether the integration measure contains factors such as $(e^{\gamma_E}/4\pi)^\epsilon$ or whether they are removed by the counterterm.

Minimal subtraction is a renormalization prescription, not a physical normalization condition. It fixes no mass or coupling directly from an observed pole, residue, or cross section. Those inputs must be supplied by a finite matching relation. Collins distinguishes this pole prescription from momentum and physical subtraction conditions and shows how it extends through subdivergences Collins 1984/2023, §§ 3.6 and 5.11, pp. 56–59 and 135–137.

The finite logarithm carries scale dependence

For the modified-minimal result,

$$\mu\frac{d}{d\mu} B_{\overline{\rm MS}}(p^2;\mu) = \frac{1}{8\pi^2}.$$

The renormalized loop is not by itself an observable. In a complete amplitude, the $\mu$ dependence of the renormalized coupling or local coefficient cancels this derivative through the retained order. The origin of the running is visible already here: removing a local $1/\epsilon$ pole leaves a logarithm of the arbitrary scale.

At fixed bare coupling,

$$0 = \mu\frac{d}{d\mu} \left[ \mu^{\kappa\epsilon}Z_g(g,\epsilon)g(\mu) \right].$$

The residues of the subtraction poles therefore determine the scale equation in a mass-independent scheme. The detailed derivation of beta functions and anomalous dimensions belongs to Beta Functions, Running Masses, and Field Anomalous Dimensions; here the invariant checkpoint is cancellation of $\mu$ in a complete prediction. ’t Hooft developed the pole/RG relation directly in dimensional regularization ’t Hooft 1973, pp. 455–468.

Scaleless zero does not mean no ultraviolet pole

Dimensional regularization assigns a scaleless integral such as

$$I_0 = \mu^{2\epsilon} \int\frac{d^d k}{(2\pi)^d} \frac{1}{(k^2)^2}$$

the value zero. Scale covariance makes this inevitable: there is no mass or external momentum from which a nonzero answer with the required dimension can be built. But the radial integral contains both endpoints,

$$\int_0^\infty dk\,k^{-1-2\epsilon}.$$

Split it at an arbitrary scale $\kappa$ and analytically continue the two pieces from their separate convergence domains:

$$\begin{aligned} \int_0^\kappa dk\,k^{-1-2\epsilon_{\rm IR}} &=-\frac{\kappa^{-2\epsilon_{\rm IR}}}{2\epsilon_{\rm IR}},\\ \int_\kappa^\infty dk\,k^{-1-2\epsilon_{\rm UV}} &=\frac{\kappa^{-2\epsilon_{\rm UV}}}{2\epsilon_{\rm UV}}. \end{aligned}$$

When the analytic parameters are identified, the ultraviolet and infrared pieces cancel. Symbolically,

$$0 \sim \frac{1}{\epsilon_{\rm UV}} - \frac{1}{\epsilon_{\rm IR}}.$$

The precise overall factor depends on the integral and signature. The lesson does not: a scaleless zero cannot establish the absence of a UV counterterm. Introduce an infrared mass or off-shell momentum, use a region label, or otherwise separate the endpoints before extracting ultraviolet data. The UV/IR Poles and the Renormalized-Amplitude Interface develops the amplitude-level bookkeeping.

Symmetry advantages and continuation limits

Analytic continuation retains loop-momentum shift invariance and covariant tensor reduction. These properties made the method particularly effective for non-Abelian gauge theories; in anomaly-free cases it can preserve the required Ward identities throughout the subtraction ’t Hooft and Veltman 1972, pp. 189–213.

That statement has boundaries:

• The metric trace becomes $g^\mu{}_{\mu}=d$, and every contraction must use the declared $d$-dimensional algebra.
• External polarization states and internal momenta may be continued in different named schemes; conversion terms are then required before amplitudes are compared.
• A matrix $\gamma_5$ with all familiar four-dimensional properties cannot be used without qualification in general $d$. Levi-Civita tensors and chiral traces need an explicit prescription and, often, finite symmetry restoration.
• Operators that vanish at $d=4$ can be nonzero for $d=4-2\epsilon$. These evanescent structures can multiply poles and leave finite effects; they must be retained when the operator sector requires them.
• Supersymmetric calculations may use dimensional reduction rather than ordinary dimensional regularization. The two are distinct schemes with separate consistency and translation obligations.

This page does not select a universal $\gamma_5$ prescription. The appropriate choice depends on the theory, operator basis, and target identity; named gauge-theory applications belong to Gauge Theories and the Standard Model, and closure is developed in Evanescent Operators, Finite Renormalization, and RG Closure.

Checks on the subtraction

The reusable validation fields specialize as follows for the massive bubble.

Field	Required entry
Regulated object	Euclidean $B(p^2;m^2)$ with $d=4-2\epsilon$, $m>0$, and nonexceptional $p^2$
Measure	$\mu^{2\epsilon}d^dk/(2\pi)^d$; state any modified-scale factor
Pole origin	Ultraviolet; the mass removes the infrared endpoint singularity
Subtraction	MS removes $1/\epsilon$; $\overline{\rm MS}$ removes $1/\bar\epsilon$
Finite input	A separate matching or renormalization condition fixes the corresponding local parameter
Finite result	The parameter integral of $-\ln(\Delta_x/\mu^2)/(16\pi^2)$ in $\overline{\rm MS}$
Scale check	$\mu\,dB_{\overline{\rm MS}}/d\mu=1/(8\pi^2)$, cancelled by coefficient running in a complete prediction
Removal	Subtract first, then take $\epsilon\to0$; no uncancelled pole remains
Independent check	Momentum subtraction or another regulator gives the same nonlocal $p^2$ dependence after a finite local map

This table makes explicit what a bare pole expression omits. In particular, “computed in dimensional regularization” does not specify MS versus $\overline{\rm MS}$, the meaning of $\mu$, the infrared prescription, or the finite parameter input.

A reproducible calculation should cover checking the expansion, subtraction convention, and finite map against the cutoff benchmark.

Common pitfalls

Dropping $4\pi$ and $\gamma_E$ before naming the scheme. Those constants are exactly the finite difference between MS and $\overline{\rm MS}$ in the unmodified measure. Absorb them only after stating which scale and counterterm convention results.

Calling $\mu$ a UV cutoff. Loop momenta remain integrated over their full analytically continued domain. The scale $\mu$ restores dimensions and labels renormalized coordinates; it does not exclude modes above $\mu$.

Subtracting before separating UV and IR poles. The same symbol $1/\epsilon$ can represent either endpoint. A UV counterterm may remove only the local ultraviolet pole; infrared cancellation or factorization is a different operation.

Reading absent power divergences as absent heavy-scale sensitivity. Analytic continuation does not make physical threshold corrections vanish. Match a heavy theory to a low-energy observable before drawing a naturalness conclusion.

Using four-dimensional identities inside a divergent $d$-dimensional expression. A factor of $d-4=-2\epsilon$ can multiply a $1/\epsilon$ pole and produce a finite term. Complete the declared $d$-dimensional algebra before taking the limit.

Exercises

1. Expand the master integral. Multiply $\Gamma(\epsilon)$, $(4\pi)^\epsilon$, and $(\mu^2/\Delta)^\epsilon$ through order $\epsilon^0$.

Solution

Use $\Gamma(\epsilon)=1/\epsilon-\gamma_E+\mathcal O(\epsilon)$. The two powers contribute $1+\epsilon[\ln4\pi+\ln(\mu^2/\Delta)]$. Multiplying by the pole promotes that order-$\epsilon$ bracket to a finite term, yielding $1/\epsilon-\gamma_E+\ln4\pi+\ln(\mu^2/\Delta)$.

2. Translate MS to modified MS. What finite term separates $B_{\rm MS}$ from $B_{\overline{\rm MS}}$ in the measure used here?

Solution

MS leaves $(-\gamma_E+\ln4\pi)/(16\pi^2)$ because it removes only $1/\epsilon$. Modified MS removes that constant together with the pole. The same translation can be expressed by $\bar\mu^2=4\pi e^{-\gamma_E}\mu^2$.

3. Diagnose a scaleless integral. Why is the statement $I_0=0$ insufficient for determining a UV counterterm?

Solution

The radial integral has an infrared and an ultraviolet endpoint with disjoint convergence domains. Dimensional continuation identifies their pole parameters and cancels them. A UV counterterm requires the ultraviolet part alone, obtained after adding an infrared scale, taking external legs off shell, or otherwise labeling the two regions.

Result and next step

With the convention

$$d=4-2\epsilon, \qquad \int_k=\mu^{2\epsilon}\int\frac{d^dk}{(2\pi)^d},$$

MS removes pure $1/\epsilon$ poles and $\overline{\rm MS}$ removes $1/\bar\epsilon=1/\epsilon-\gamma_E+\ln4\pi$. A massive nonexceptional scalar bubble then leaves a finite logarithm whose $\mu$ derivative is cancelled by the running of renormalized local data. Every pole must retain an ultraviolet or infrared label until that origin is proven.

Continue to Local Counterterms and Subdivergence Structure for the locality argument behind these pole subtractions. Continue to Renormalized Perturbation Theory and Counterterm Rules after the forest step to embed MS counterterms in a complete rule set. Return to QFT Regulator Families and Their Tradeoffs when dimensional continuation conflicts with the decisive symmetry or nonperturbative requirement.

References

• Collins, John C. 1984; open-access reissue 2023. Renormalization: An Introduction to Renormalization, the Renormalization Group and the Operator-Product Expansion. Cambridge Monographs on Mathematical Physics. Cambridge University Press. DOI and Open PDF.

• ’t Hooft, Gerard. 1973. “Dimensional Regularization and the Renormalization Group.” Nuclear Physics B 61: 455–468. DOI.

• ’t Hooft, Gerard, and Martinus Veltman. 1972. “Regularization and Renormalization of Gauge Fields.” Nuclear Physics B 44: 189–213. DOI.