Renormalization Conditions, Schemes, and Finite Parts

A renormalization scheme is a coordinate system on the same family of renormalized theories. The divergent local terms fix which counterterms are required; renormalization conditions fix their finite parts and thereby define the numerical mass, coupling, and field normalization called “renormalized.” Minimal subtraction, momentum subtraction, on-shell conditions, and physical input schemes generally assign different numbers to those coordinates.

They describe the same physics only after a finite parameter and field map is applied. Comparing an MS̄ coupling directly with a momentum-subtraction coupling at the same numerical value compares different theories. The valid check is to match the same inputs, translate the coordinates to the calculated order, and compare a non-input prediction; the difference should begin one order beyond the truncation.

Required background. Dimensional Regularization and Minimal Subtraction fixes the modified-MS pole convention used below.

Helpful background. The 1PI Effective Action and Mean-Field Equations supplies the inverse two-point and proper-vertex language. Local Counterterms and Subdivergence Structure explains why two local schemes can differ only by finite local terms.

Finite conditions define the renormalized coordinates

Let a regulated bare parameter be written in scheme $A$ as

$$g_0 = \mu^{\kappa\epsilon} Z_g^A(g_A,\epsilon)\,g_A.$$

The same bare theory can be parameterized in scheme $B$:

$$g_0 = \mu^{\kappa\epsilon} Z_g^B(g_B,\epsilon)\,g_B.$$

Equating the two expressions and removing the regulator gives a finite map,

$$g_B = g_A+c_1g_A^2+c_2g_A^3+\cdots.$$

Fields can also require a finite map,

$$\phi_B = \left( 1+r_1g_A+r_2g_A^2+\cdots \right)\phi_A.$$

The coefficients depend on the two scheme definitions. They are not additional observables. Once the map is fixed, an exact observable has the same value in both coordinate systems.

At finite order, suppose

$$\mathcal P = g_A+p_1^Ag_A^2+\mathcal O(g_A^3), \qquad g_B=g_A+c_1g_A^2+\mathcal O(g_A^3).$$

Inverting the map and substituting gives

$$\mathcal P = g_B+ \left( p_1^A-c_1 \right)g_B^2 +\mathcal O(g_B^3).$$

Thus $p_1^B=p_1^A-c_1$. The coefficient of a truncated expansion changes precisely enough to compensate the changed coupling. Scheme independence never means equality of the intermediate coefficients.

Collins formulates a change of renormalization prescription as a finite reparametrization of couplings, masses, and fields and proves equality of the corresponding physics Collins 1984/2023, § 7.1, pp. 169–176.

Four common choices

Scheme type	Defining condition	Main advantage	Main caution
MS or MS̄	Subtract only the declared dimensional poles, with or without the standard $-\gamma_E+\ln4\pi$ package	Mass independent; exposes renormalization-group structure	Parameters are not direct observables; heavy fields do not decouple automatically
Momentum subtraction	Fix inverse propagators and vertices at specified nonexceptional Euclidean momenta	Kinematic meaning is explicit; useful for nonperturbative comparisons	Generally mass and gauge-parameter dependent; exceptional points can create infrared problems
On shell	Put a stable-particle pole at its physical mass and normalize its residue; define charges from a stated physical limit	Inputs are closely tied to measured quantities	Infrared singularities, confinement, massless particles, and unstable states can obstruct naive conditions
Physical or observable based	Define parameters from a complete set of measured infrared-safe observables	Gauge invariant when the observables are	Often process specific; translations may contain large logarithms

MS and MS̄ differ only by a conventional finite rescaling of $\mu$, but writing “MS” while using the modified pole package creates a finite mismatch. Collins treats mass-shell oversubtractions and minimal subtraction separately, making clear that the difference is a finite prescription choice after locality is secured Collins 1984/2023, §§ 5.9–5.11, pp. 130–137.

On-shell two-point conditions

For a stable scalar with Minkowski inverse propagator

$$D_{\rm R}^{-1}(p^2) = p^2-M^2-\Sigma_{\rm R}(p^2),$$

unit pole residue is imposed by

$$\Sigma_{\rm R}(M^2)=0, \qquad \left. \frac{d\Sigma_{\rm R}}{dp^2} \right|_{p^2=M^2} =0.$$

These conditions fix the finite mass and field counterterms. For an unstable particle the invariant object is instead a complex pole; imposing a real-axis residue condition can be gauge dependent and physically misleading. Confining fields have no asymptotic one-particle pole at all.

Nonexceptional momentum subtraction

For a Euclidean scalar theory, a MOM scheme can impose

$$\begin{aligned} \Gamma_{\rm R}^{(2)}(p^2=Q_*^2) &= Q_*^2+m_{\rm MOM}^2, \\ \left. \frac{d\Gamma_{\rm R}^{(2)}}{dp^2} \right|_{p^2=Q_*^2} &=1, \\ \left. \Gamma_{\rm R}^{(4)} \right|_{\rm sym} &= \lambda_{\rm MOM}. \end{aligned}$$

At the symmetric four-point configuration,

$$P_s^2=P_t^2=P_u^2=Q_*^2>0.$$

Keeping every channel nonzero avoids an artificial infrared singularity in a massless limit. The numerical value of an off-shell MOM parameter can depend on the gauge and projector; only a translated physical prediction is a gauge-invariant checkpoint. Gauge-parameter dependence of off-shell counterterms and its cancellation in physical quantities is developed in Collins 1984/2023, § 12.4, pp. 309–314.

One-loop MS̄-to-MOM map in scalar φ⁴ theory

Use the Euclidean $\mathbb Z_2$-invariant scalar theory and the $d=4-2\epsilon$ convention of the previous pages. Factor the common $\mu^{2\epsilon}$ from the proper four-point coefficient. In MS̄,

$$\Gamma_{\overline{\rm MS}}^{(4)} = \lambda_{\overline{\rm MS}} + \frac{\lambda_{\overline{\rm MS}}^2}{32\pi^2} \sum_{X=s,t,u}F(P_X^2) +\mathcal O(\lambda^3),$$

where

$$F(P^2) \equiv \int_0^1dx\, \ln \frac{m^2+x(1-x)P^2}{\mu^2}.$$

To isolate the coupling map, take the mass to be matched by the same finite input condition in both descriptions. A one-loop difference in the mass coordinate inserted inside $F$ would first change the displayed coupling map at the next order.

Define $\lambda_{\rm MOM}$ by demanding that the proper vertex at the symmetric point equal the tree coefficient:

$$\left. \Gamma_{\rm MOM}^{(4)} \right|_{\rm sym} = \lambda_{\rm MOM}.$$

Evaluating the MS̄ expression at that point gives the finite map

$$\boxed{ \lambda_{\rm MOM} = \lambda_{\overline{\rm MS}} + \frac{3\lambda_{\overline{\rm MS}}^2}{32\pi^2} F(Q_*^2) +\mathcal O(\lambda^3) }.$$

Its inverse is

$$\lambda_{\overline{\rm MS}} = \lambda_{\rm MOM} - \frac{3\lambda_{\rm MOM}^2}{32\pi^2} F(Q_*^2) +\mathcal O(\lambda^3).$$

Substituting into the MS̄ vertex yields

$$\begin{aligned} \Gamma^{(4)} ={}& \lambda_{\rm MOM} \\ &+ \frac{\lambda_{\rm MOM}^2}{32\pi^2} \left[ \sum_{X=s,t,u}F(P_X^2) -3F(Q_*^2) \right] \\ &+ \mathcal O(\lambda^3). \end{aligned}$$

At the subtraction point the bracket vanishes, as the MOM definition requires. Away from that point, this is the same one-loop proper vertex as the MS̄ expression, written in different coordinates. The local finite term $3F(Q_*^2)$ moved into the definition of the coupling; the nonlocal differences $F(P_X^2)-F(Q_*^2)$ remain in the prediction.

After analytic continuation, the same equality holds for the one-loop scalar scattering amplitude. At this order scalar $\phi^4$ theory has no field-strength correction, so no additional external residue enters. Both schemes therefore predict the same channel logarithms and cuts through $\mathcal O(\lambda^2)$; their difference begins at $\mathcal O(\lambda^3)$.

What the bare-to-observable map requires

The figure shows where a finite scheme map belongs. Inspect the center: the two descriptions can have different counterterms and renormalized coordinates, but the lower path compares a non-input observable only after the same finite input data have been imposed.

Finite scheme translation in the bare-to-observable chain. The regulator and subtraction convention select auxiliary intermediate data; renormalization conditions define the finite coordinates; matched inputs and a non-input prediction test equivalence. The map is schematic and not to scale.

The scheme comparison should record:

Item	Scheme A	Scheme B	Equality test
Regulator convention	Same or explicitly translated	Same or explicitly translated	No hidden change in $\epsilon$, $\mu$, or normalization
Renormalized inputs	Declared observables or conditions	The same physical information	Equal input values after translation
Finite map	$g_A,m_A,Z_A$	$g_B(g_A),m_B(m_A,g_A),Z_B/Z_A$	Bare relations agree through the retained order
Symmetry identity	Restored with scheme-A finite terms	Restored with scheme-B finite terms	Same Ward or Slavnov–Taylor identity
Non-input quantity	$\mathcal P_A$	$\mathcal P_B$	$\mathcal P_A-\mathcal P_B=\mathcal O(g^{N+1})$ after an order-$N$ calculation
Residual variation	Change of $\mu$, subtraction point, or allowed finite terms	Corresponding translated change	Used as a diagnostic, not a universal probability distribution

For the scalar benchmark, the map above is exact through one loop at the stated kinematics. A reproducible calculation should cover evaluating $F(Q_*^2)$, applying the forward and inverse maps, and checking the $\mathcal O(\lambda^3)$ round-trip residual.

Scheme choice and perturbative truncation

An exact prediction is scheme independent, but a truncated result retains higher-order scheme dependence. This has three practical consequences.

First, apply a scheme map to the same order as the calculation. Using a two-loop amplitude with only a tree-level parameter identification leaves a spurious one-loop mismatch.

Second, choose scales and schemes that do not manufacture large coefficients. A MOM point far from all physical scales or a physical input containing a large hierarchy can move a large logarithm into the finite map rather than eliminate it.

Third, scheme variation is a useful stress test but not, by itself, a statistically calibrated uncertainty. A small variation can miss a large next coefficient, while an extreme finite redefinition can exaggerate the remainder. Combine it with scale variation, known asymptotics, power counting, and benchmark comparisons appropriate to the problem.

Heavy-particle thresholds add another qualification. A mass-independent scheme retains heavy fields in the beta functions until an effective theory is matched across the threshold. That is not a failure of MS̄; it is a signal that running and matching are separate operations, developed later in this volume.

Common pitfalls

Setting two scheme couplings to the same number. Their equality is not the matching condition. Use the finite map derived from a common bare theory or common physical inputs.

Changing the scheme in the loop term but not the tree term. Re-expand the entire truncated prediction, including masses, fields, and external residues.

Subtracting at exceptional momentum in a massless theory. Zero momentum can turn a ultraviolet definition into an infrared singular one. Use a nonexceptional Euclidean point or an explicitly infrared-safe prescription.

Treating an off-shell MOM coupling as gauge invariant. It can depend on gauge fixing and projectors. Translate it into a physical observable before making an invariant claim.

Using a real on-shell condition for an unstable state. The stable-pole assumptions fail. Use the complex pole and a treatment appropriate to unstable particles.

Reading scheme variation as a confidence interval. It samples selected higher-order terms but has no universal probabilistic interpretation.

Exercises

1. Invert the scalar map. Verify the inverse relation through $\mathcal O(\lambda^2)$.

Solution

Write $\lambda_{\rm MOM}=\lambda_{\overline{\rm MS}}+c\lambda_{\overline{\rm MS}}^2$, with $c=3F(Q_*^2)/(32\pi^2)$. Iterative inversion gives $\lambda_{\overline{\rm MS}}=\lambda_{\rm MOM}-c\lambda_{\rm MOM}^2+\mathcal O(\lambda^3)$. Substituting it into the forward map leaves $\lambda_{\rm MOM}+\mathcal O(\lambda^3)$.

2. Check the MOM condition. Set all three channel invariants to $Q_*^2$ in the MOM vertex.

Solution

The one-loop bracket becomes $3F(Q_*^2)-3F(Q_*^2)=0$, so $\Gamma^{(4)}|_{\rm sym}=\lambda_{\rm MOM}+\mathcal O(\lambda^3)$, exactly as defined.

3. Diagnose a comparison. Two calculations use the same numerical coupling, one in MS̄ and one in MOM, and differ at order $\lambda^2$. Is this scheme dependence of a physical prediction?

Solution

Not yet. The same number labels different theories in the two schemes. Apply the finite map or fit both couplings to the same input first. Any remaining difference should then start at order $\lambda^3$.

Finite-coordinate result

A scheme change moves finite local terms between parameters and coefficient functions:

$$\text{finite parameter map} + \text{re-expanded coefficient function} = \text{same prediction through the retained order}.$$

MS̄, MOM, on-shell, and physical schemes are therefore tools with different conditioning and bookkeeping properties, not competing physical laws.

Continue to Regulator Removal and Renormalized Predictions to combine finite matching with the removal limit, symmetry checks, and an explicit remainder. Continue later to threshold matching for the separate question of changing active degrees of freedom.

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.