Renormalons, OPE Ambiguities, and Power Corrections

Perturbative coefficients can grow like $n!$ even when every fixed order is ultraviolet renormalized. A renormalon is a Borel-plane singularity tied to the small- or large-momentum endpoint of loop integration. For an infrared renormalon on the positive Borel axis, the perturbative sum requires a prescription and acquires an ambiguity with the scaling of a power correction.

The ambiguity is not itself a prediction of a condensate, a particle, or a semiclassical configuration. In an operator product expansion, it instead states a cancellation contract: the prescription dependence assigned to a short-distance coefficient must cancel the prescription dependence assigned to the corresponding power-suppressed matrix element. This page derives that contract in a one-chain model and states exactly what can and cannot be inferred from it.

Required background. Running Couplings and Dimensional Transmutation supplies the invariant scale $\Lambda$. Large Logarithms and RG Improvement supplies one-loop running and controlled resummation. Asymptotic Scales, Remainders, Uniformity, and Optimal Truncation supplies asymptotic-series language. Free-Field OPE Preview supplies the operator-product expansion and its separation of short- and long-distance data.

Factorial growth and the Borel plane

Consider a dimensionless observable with formal expansion

$$R(a) \sim \sum_{n=0}^\infty r_n a^{n+1}, \qquad a>0.$$

Suppose its coefficients behave at large order as

$$r_n \sim K A^n\Gamma(n+1+b).$$

The ratio of successive terms eventually grows like $A a n$. Adding terms improves the approximation only until

$$n_\star \sim \frac{1}{|A|a}.$$

The smallest term is exponentially small, of order $e^{-1/(|A|a)}$ up to powers of $a$. Beyond $n_\star$, the partial sums move farther apart. This is optimal truncation, not convergence.

Define the Borel transform and its Laplace integral by

$$\mathcal B_R(t) \equiv \sum_{n=0}^\infty \frac{r_n}{n!}t^n, \qquad \mathcal S R(a) \equiv \int_0^\infty dt\, e^{-t/a}\mathcal B_R(t).$$

Expanding $\mathcal B_R$ inside the integral recovers the original formal series because

$$\int_0^\infty dt\, e^{-t/a}t^n = n!a^{n+1}.$$

If $\mathcal B_R(t)$ is analytic along the positive real axis and grows slowly enough, the integral defines a Borel sum. A singularity on that contour instead requires a deformation above or below it. For a simple pole written as

$$\mathcal B_R(t) \underset{t\to t_0}{\sim} \frac{K}{t_0-t}, \qquad t_0>0,$$

the two lateral prescriptions have imaginary parts

$$\operatorname{Im}\mathcal S_\pm R = \pm\pi K e^{-t_0/a}.$$

The sign labels which side of the pole the contour passes. The principal value is a useful real convention, but choosing it does not remove the need to define any same-order power term consistently. Beneke develops optimal truncation, Borel transformation, and the positive-axis ambiguity in Beneke 1999, § 2.1, pp. 5–7, Open PDF.

Factorial growth alone does not identify a renormalon. Large diagram multiplicities and instanton-related saddles can also generate large-order behavior. The term is reserved here for a Borel singularity whose origin can be traced to an endpoint of loop momentum.

A one-chain infrared model

Use the explicit asymptotically free convention

$$\frac{d a(\mu^2)}{d\ln\mu^2} = -\beta_0a^2(\mu^2), \qquad \beta_0>0.$$

At one loop,

$$a(k^2) = \frac{a_Q}{ 1+\beta_0a_Q\ln(k^2/Q^2) }, \qquad a_Q\equiv a(Q^2).$$

Now model the infrared part of a one-chain skeleton integral by

$$\boxed{ R_p(Q) \equiv \int_0^1 dx\, x^{p-1}a(xQ^2), \qquad x\equiv\frac{k^2}{Q^2}, \qquad p>0. }$$

The weight $x^{p-1}$ is the declared small-momentum behavior of the skeleton. Expanding the running coupling about $Q$ gives the formal series

$$a(xQ^2) = a_Q \sum_{n=0}^\infty \left[ \beta_0a_Q\ln\frac1x \right]^n.$$

This expansion is not uniform down to $x=0$ and the one-loop integrand itself has a pole at $x=\Lambda^2/Q^2$. Termwise integration is being used to extract the perturbative coefficients, not to define an exact nonperturbative integral.

The identity

$$\int_0^1 dx\, x^{p-1} \left(\ln\frac1x\right)^n = \frac{n!}{p^{n+1}}$$

follows by differentiating $\int_0^1dx\,x^{p-1}=1/p$ exactly $n$ times with respect to $p$. Therefore

$$\boxed{ R_p(Q) \sim \sum_{n=0}^\infty \frac{\beta_0^n n!}{p^{n+1}} a_Q^{n+1}. }$$

The factorial does not count diagrams in this model. It comes from the logarithmic moment of an increasingly endpoint-dominated loop integral. Its Borel transform is elementary:

$$\mathcal B_{R_p}(t) = \sum_{n=0}^\infty \frac{\beta_0^n}{p^{n+1}}t^n = \boxed{ \frac{1}{p-\beta_0t} }.$$

Thus the infrared singularity lies at

$$t_0=\frac{p}{\beta_0}>0.$$

Each lateral sum differs from the principal value by an imaginary ambiguity whose magnitude is

$$\left|\delta R_p\right| = \frac{\pi}{\beta_0} \exp\left[ -\frac{p}{\beta_0a_Q} \right].$$

Define the one-loop transmutation scale in this convention by

$$\Lambda^2 \equiv Q^2 \exp\left[ -\frac{1}{\beta_0a_Q} \right].$$

Then

$$\boxed{ \left|\delta R_p\right| = \frac{\pi}{\beta_0} \left( \frac{\Lambda^2}{Q^2} \right)^p. }$$

The pole position has become a power law. In the frequently used variable $u=\beta_0t$, the singularity is at $u=p$ and corresponds to an operator dimension

$$d=2p.$$

This calculation is the promised bubble-chain application in its cleanest form. A chain of renormalized vacuum-polarization insertions generates powers of a loop logarithm; the low-momentum moment converts those powers into $n!$. In a non-Abelian theory, replacing the fermionic bubble coefficient by the full $\beta_0$ requires a controlled large-flavor or large-$\beta_0$ argument and does not make the selected graphs equal to the full theory. Beneke derives the current-correlator bubble-chain example and emphasizes both its usefulness and its limits in Beneke 1999, § 2.2, pp. 8–13, Open PDF.

Optimal-truncation check

Take

$$p=2, \qquad \beta_0=1, \qquad a_Q=0.1.$$

The $n$th perturbative term is

$$T_n = \frac{n!}{2^{n+1}}(0.1)^{n+1}, \qquad \frac{T_{n+1}}{T_n} = \frac{n+1}{20}.$$

The terms decrease until $n=19$ and $20$, where adjacent terms are equal, then grow:

$n$	$\lvert T_n\rvert$	$\lvert T_{n+1}/T_n\rvert$
$0$	$5.000000000000\times10^{-2}$	$0.05$
$5$	$1.875000000000\times10^{-6}$	$0.30$
$10$	$1.771875000000\times10^{-8}$	$0.55$
$15$	$1.995352734375\times10^{-9}$	$0.80$
$19$	$1.160098079766\times10^{-9}$	$1.00$
$20$	$1.160098079766\times10^{-9}$	$1.05$
$21$	$1.218102983754\times10^{-9}$	$1.10$
$25$	$2.311350411673\times10^{-9}$	$1.30$

The power scale and lateral ambiguity are

$$\left( \frac{\Lambda^2}{Q^2} \right)^2 =e^{-20} =2.061153622439\times10^{-9},$$ $$\left|\delta R_2\right| = \pi e^{-20} =6.475305078173\times10^{-9}.$$

The minimal term and the ambiguity share the same exponential scaling, but their prefactors need not agree. A numerical implementation should reproduce $n_\star\simeq p/(\beta_0a_Q)-1=19$, the equality $T_{20}=T_{19}$, and the Borel pole at $t_0=2$. These are internal checks of the model, not uncertainty probabilities for a physical observable.

Infrared versus ultraviolet renormalons

The momentum endpoint fixes both the sign pattern and the interpretation in an asymptotically free convention:

Origin	Typical Borel location	Large-order sign	Immediate implication
Small loop momentum, $k\ll Q$	Positive real axis	Fixed sign	Obstructs the positive-axis Borel integral and exposes a long-distance power sensitivity
Large loop momentum, $k\gg Q$	Negative real axis	Alternating	Does not by itself obstruct the positive-axis contour; it is organized by short-distance counterterms and higher-dimension operators

The qualifications matter. A negative-axis ultraviolet singularity may dominate the magnitude of high-order coefficients even while a more distant positive-axis infrared singularity controls the ambiguity. Other positive-axis singularities can still obstruct Borel summation. Scheme changes can alter residues and the conventional normalization of $\Lambda$, while regular scheme transformations preserve the allowed power scaling once all quantities are translated.

The power is constrained by the small-momentum behavior and the available operator basis. In the standard massless vector-current OPE, for example, no gauge-invariant local scalar of dimension two exists. The would-be $u=1$ infrared renormalon is absent, while the dimension-four sector can support a $u=2$ ambiguity. This is an operator-selection check, not a general rule that every observable begins at $1/Q^4$ Beneke 1999, §§ 2.2–2.3, pp. 10–15, Open PDF.

The pole location does not determine a condensate’s value. Its residue is observable dependent, and a bubble-chain residue is not automatically the full-theory residue. Nor does the renormalon prove that a named matrix element is nonzero. It identifies a perturbative sensitivity and the dimension of data needed to make the factorized prediction prescription independent.

The OPE cancellation contract

For a dimensionless Euclidean short-distance observable, write the OPE schematically as

$$R(Q) = C_0(Q,\mu;\mathsf P) + \sum_d \frac{C_d(Q,\mu;\mathsf P)}{Q^d} \langle O_d(\mu)\rangle_{\mathsf P}.$$

Here $\mu$ is a factorization scale and $\mathsf P$ denotes the subtraction and Borel prescription. In a dimensional factorization scheme, $C_0$ samples all loop momenta and can carry an infrared-renormalon ambiguity. The renormalized higher-dimension matrix element has a compensating ultraviolet subtraction ambiguity from its mixing with lower-dimension operators.

At the first relevant power, prescription independence requires

$$\boxed{ \delta_{\mathsf P}C_0 + \frac{C_d}{Q^d} \delta_{\mathsf P}\langle O_d\rangle =0 }$$

through the declared accuracy. If $\delta_{\mathsf P}C_0\sim(\Lambda/Q)^d$, then

$$\delta_{\mathsf P}\langle O_d\rangle \sim \Lambda^d$$

with the normalization and sign fixed only after the operator, coefficient, and prescription conventions are specified. When several operators of the same dimension mix, the cancellation is vector valued and must be checked in the complete symmetry-allowed basis.

A hard factorization cutoff gives an equivalent but differently distributed description. Loop momenta below $\mu$ are removed from the Wilson coefficient and placed directly in matrix elements; the latter then contain power-like cutoff dependence. Dimensional factorization hides that separation inside prescription-dependent quantities. Either way, only the sum is physical. Beneke explains this cutoff-versus-dimensional-factorization translation and the cancellation of coefficient and condensate ambiguities in Beneke 1999, § 2.3, pp. 14–17, Open PDF.

Three consequences prevent overinterpretation:

1. The cancellation fixes prescription dependence, not the prescription-independent value of $\langle O_d\rangle$.
2. A renormalon analysis can miss power corrections protected from mixing with lower-dimension operators or generated by genuinely nonperturbative dynamics.
3. If no operator with the required quantum numbers and dimension exists, a purported Borel singularity must have zero residue, cancel in the observable, or signal that the assumed factorization problem was incomplete.

Coordinate-dependent data and invariant claims

The chapter’s comparison table places the renormalon row in the same scheme analysis used throughout this chapter.

Item	What may change	What survives a consistent translation	Required qualification or check
Renormalized $g^i$, masses, and field normalizations	Numerical values under finite scheme or basis changes	A prediction expressed in the same physical inputs	Translate every parameter and field factor through the retained order
Beta function away from a fixed point	Components and higher-order coefficients	The integral curves as geometric trajectories under a nonsingular coordinate map	Compare transformed vector fields, not coefficients at equal numerical coupling
Elementary-field anomalous dimension	Finite field rescaling; gauge parameter in a gauge theory	Scaling of a gauge-invariant observable after all factors are combined	Never identify a gauge-dependent elementary-field exponent with an observable
Exact fixed point	Coordinate location $g_\star^i$	Existence of the zero under a regular map	Exclude singular redefinitions and verify the fixed point lies in the method’s domain
Fixed-point stability data	Matrix representation and basis	Eigenvalues in a closed physical sector	Include operator mixing and redundant directions before diagonalizing
Transmuted scale	Its conventional normalization	Matched dimensionless ratios or predictions	State the scheme and reference condition defining the scale
Zero or singularity of a truncated beta function	Location and even apparent existence at insufficient order	Only the demonstrated breakdown of the stated approximation	Vary scheme/order and stop before couplings become large
Wilson coefficient versus power correction	Factorization scheme and, for an asymptotic series, summation prescription	Their consistently defined sum in an observable	Match the ambiguity of the perturbative term to the operator matrix element; developed on the renormalon page
Residual $\mu$ or scheme dependence	Numerical size at finite order	Vanishing in the exact consistently matched prediction	Treat the residual as a diagnostic, not a universal probability law

For this page, the invariant object is the fully matched $R(Q)$, not $C_0$, a condensate, a principal-value sum, or a truncated series in isolation. A prescription change is acceptable only when every term that shares its ambiguity is translated.

A reproducible diagnostic workflow

For a proposed renormalon claim:

1. declare the coupling and beta-function normalization, including whether the RG derivative is with respect to $\ln\mu$ or $\ln\mu^2$;
2. identify the momentum region responsible for the large logarithmic moments;
3. derive the large-$n$ coefficient behavior and Borel singularity rather than inferring it from a few low orders;
4. map $e^{-t_0/a(Q)}$ to a power of $\Lambda/Q$ in the same convention;
5. list the symmetry-allowed operators with the required dimension and quantum numbers;
6. state the factorization, subtraction, and Borel prescriptions for both coefficient and matrix element;
7. verify cancellation of the prescription dependence in their sum;
8. separate the structural power scaling from any model for the finite matrix element.

Failure at step 5 is especially informative: the observable cannot support the proposed isolated ambiguity in a valid local OPE. Failure at step 7 means the factorization formula or operator basis is incomplete.

Common pitfalls

Calling any factorial series a renormalon. The defining evidence is a small- or large-loop-momentum origin for a Borel singularity. Coefficient growth alone does not locate that origin.

Integrating through the one-loop Landau pole as if the result were exact. The running-coupling integral is a device for exposing endpoint sensitivity. Its pole and its divergent expansion require a prescription; neither is a nonperturbative definition of the theory.

Equating a Borel ambiguity with a measured power correction. The ambiguity fixes a scale and a cancellation requirement. It does not determine the real, prescription-independent part of a matrix element.

Assigning the ambiguity to only one OPE term. In dimensional schemes the coefficient and matrix element are individually conventional. Only the consistently prescribed sum can be compared with an observable.

Using a bubble chain as the full theory. It is a controlled model in a declared large-flavor or large-$\beta_0$ setting. Locations and scaling can be robust while residues and finite parts remain model dependent.

Exercises

Derive the Borel pole and power scaling of the one-chain model.

Solution

Expand

$$\frac{a_Q}{1-\beta_0a_Q\ln(1/x)} = a_Q\sum_{n=0}^\infty \left[ \beta_0a_Q\ln\frac1x \right]^n.$$

Using

$$\int_0^1dx\,x^{p-1}\ln^n\frac1x = \frac{n!}{p^{n+1}}$$

gives $r_n=\beta_0^nn!/p^{n+1}$. Therefore

$$\mathcal B_{R_p}(t) = \sum_{n=0}^\infty \frac{\beta_0^nt^n}{p^{n+1}} = \frac{1}{p-\beta_0t}.$$

The pole is at $t_0=p/\beta_0$. Since

$$e^{-t_0/a_Q} = e^{-p/(\beta_0a_Q)} = \left(\frac{\Lambda^2}{Q^2}\right)^p,$$

the ambiguity has the scaling of a dimension-$2p$ OPE contribution.

Show explicitly why a prescription-dependent Wilson coefficient is not an observable.

Solution

Let two prescriptions differ by

$$C_0^{(2)}-C_0^{(1)} = \Delta_d\left(\frac{\Lambda}{Q}\right)^d.$$

Define the matrix element in the second prescription by

$$\langle O_d\rangle_{(2)} = \langle O_d\rangle_{(1)} - \frac{\Delta_d\Lambda^d}{C_d}.$$

Then

$$\begin{aligned} C_0^{(2)} +\frac{C_d}{Q^d}\langle O_d\rangle_{(2)} &= C_0^{(1)} +\Delta_d\frac{\Lambda^d}{Q^d} \\ &\quad +\frac{C_d}{Q^d}\langle O_d\rangle_{(1)} -\Delta_d\frac{\Lambda^d}{Q^d} \\ &= C_0^{(1)} +\frac{C_d}{Q^d}\langle O_d\rangle_{(1)}. \end{aligned}$$

The split moved, while the prediction did not.

Where to continue

• Renormalons, OPE Ambiguities, and Transseries develops the nonperturbative and resurgent completion questions not decided here.
• Scheme Transformations and RG Invariants explains why $\Lambda$ normalizations and residues move while a consistently translated prediction survives.
• Dual Evolution of Operators and Wilson Coefficients supplies the matrix-valued evolution needed when several power operators mix.

References

• Beneke, Martin. “Renormalons.” Physics Reports 317 (1999): 1–142. DOI. Open PDF.