greatitself.
Continue
research29 min studyLesson 690 in Physics

Factorial growth exposes information beyond a finite loop expansion

Connect momentum-region logarithms to Borel singularities while separating asymptotic estimates from a nonperturbative definition.

Before you begin
On this page

The question for today

What you’ll understand

  • Derive factorial coefficients from a logarithmic momentum integral.
  • Locate simple infrared and ultraviolet Borel singularities in a declared toy convention.
  • Estimate optimal truncation and interpret prescription ambiguities cautiously.

Renormalized perturbative coefficients can grow factorially even after all ultraviolet divergences have been subtracted. Renormalons describe one mechanism linking this large-order growth to widely separated momentum regions. They are not the only cause of divergent perturbation theory, and a few low orders do not prove their asymptotic pattern. We will use a transparent integral model to derive factorial coefficients, identify Borel singularities, and explain what an ambiguity does and does not say about a physical prediction.

Large momentum logarithms can generate factorial coefficients

Take an asymptotically free illustrative coupling with dg/dlnμ=−bg², b positive, and a hard reference scale Q. Expanding a running coupling inside an infrared momentum integral generates powers of ln(Q/k). Let x=k/Q and weight the small-momentum region by x to the power p−1, with p positive. Integrating the nth logarithmic power gives n! divided by p to the power n+1. Thus each coefficient is finite but its magnitude eventually grows faster than any geometric sequence.

The expansion samples progressively smaller momentum as the order increases. At sufficiently small momentum, the perturbative running coupling itself is no longer trustworthy. The formal infrared integral with a one-loop Landau denominator must therefore not be presented as an unambiguous exact observable. It is a model for coefficients and their momentum sensitivity. Ultraviolet regions can generate a related factorial pattern with alternating signs in an asymptotically free convention. Other large-order mechanisms, including diagram proliferation and semiclassical saddles, can coexist with renormalons.

∫01xp−1[ln⁡(1/x)]ndx=n!pn+1,rnIR=bnn!pn+1,R(g)∼∑n=0∞rngn+1\int_0^1 x^{p-1}\left[\ln(1/x)\right]^n dx=\frac{n!}{p^{n+1}},\qquad r_n^{\mathrm{IR}}=\frac{b^n n!}{p^{n+1}},\qquad \mathcal R(g)\sim\sum_{n=0}^{\infty}r_ng^{n+1}
x=k/Q, p>0, b>0, and g=g(Q)>0 are dimensionless. The coefficients come from a formal low-momentum running-coupling expansion, not a definition of the exact integral through its strong-coupling pole. The beta derivative is μd/dμ.↔ Scroll wide equations horizontally.

The Borel plane packages the large-order pattern

Define the Borel transform by dividing the coefficient of g to the power n+1 by n factorial, and define its Laplace integral without an extra factor of 1/g. In this convention the infrared toy coefficients give a pole at positive t=p/b. Integrating along positive real t then requires a prescription around that singularity. Contours passing on opposite sides differ by a term proportional to exp(−p/(bg)). With the one-loop scale Λ=Q exp(−1/(bg(Q))), this exponential has the form of a power correction (Λ/Q) to the power p.

For the corresponding ultraviolet toy coefficients, alternating signs put the pole on the negative Borel axis and do not obstruct this positive-axis integral. The association of signs with infrared and ultraviolet regions here assumes the stated asymptotically free convention. In QCD, operator-product expansions relate certain infrared ambiguities to the definition of nonperturbative matrix elements. A complete observable is not intrinsically ambiguous: prescription dependence must cancel between consistently defined pieces. Renormalon reasoning alone generally does not determine the full numerical power correction or prove a unique nonperturbative completion.

B(t)=∑n=0∞rntnn!,R~(g)=∫0∞e−t/gB(t) dt,BIR=1p−bt,BUV=1p+bt\mathcal B(t)=\sum_{n=0}^{\infty}\frac{r_nt^n}{n!},\quad \widetilde{\mathcal R}(g)=\int_0^\infty e^{-t/g}\mathcal B(t)\,dt,\quad \mathcal B_{\mathrm{IR}}=\frac1{p-bt},\quad \mathcal B_{\mathrm{UV}}=\frac1{p+bt}
The Borel variable t and g are dimensionless. The infrared expression has a positive-axis pole; the ultraviolet expression has a negative-axis pole. Existence and growth conditions are needed even when no positive-axis singularity is present, and a Borel sum need not alone specify the full physical function.↔ Scroll wide equations horizontally.

Stop near the least term and distinguish that estimate from a theorem

For factorial coefficients, the ratio of successive term magnitudes eventually exceeds one at every fixed nonzero coupling. A partial sum can improve until its least term and worsen afterward. The least-term scale estimates the limitation of that asymptotic organization, subject to its large-order assumptions. Short-distance mass definitions can avoid a leading pole-mass renormalon in useful conversions, but that reorganizes sensitivity rather than removing every nonperturbative effect from physics. RG resummation of logarithms also does not automatically eliminate factorial large-order growth.

The optimization lab offers a separate convergence counterexample. Reset and set both controls to step size η=0.5 and initial position x₀=2; after one Step predict x₁=1. Reset and set η=2.5 and x₀=2; after one Step predict x₁=−3 and an unstable-step readout. More iterations in the second run grow the error. This is deterministic gradient descent, not a perturbative series or a Borel calculation. It illustrates why increased computational order needs a convergence or asymptotic argument rather than an assumption of automatic improvement.

Work it through

Locate the least terms of a factorial series

For the toy series with p=b=1 and g=0.20, the nth term is Tn=n!gⁿ⁺¹, starting at n=0. Compute T₃, T₄, and T₅, identify where the terms stop shrinking, and compare their size with exp(−1/g)≈0.00674.

  1. The terms are T₃=6×0.20⁴=0.00960 and T₄=24×0.20⁵=0.00768. The terms are still decreasing at this point.

  2. The ratio Tn₊₁/Tn=(n+1)g. At n=4 it equals one, so T₅=0.00768 too; the next ratio is 1.2 and T₆=0.009216. The least terms are the equal pair n=4 and n=5.

  3. The ambiguity scale exp(−5)≈0.00674 is of the same order as the least-term magnitude 0.00768, although not numerically identical. This is an asymptotic scale comparison rather than an exact error theorem.

The result

The terms stop shrinking at the equal minima T₄=T₅=0.00768, comparable in scale with exp(−1/g)≈0.00674.

The living notebook

Explore the model

Interactive lab

Interactive lab · optimization

Check that extra iterations need not improve an approximation

Reset and set BOTH controls to η=0.5 and x₀=2; one Step gives n=1 and x₁=1. Reset and set η=2.5 and x₀=2; one Step gives x₁=−3 and Unstable step. This deterministic optimization is a convergence analogy, not a perturbative-order or renormalon model.

Check that extra iterations need not improve an approximationGradient descent on f(x) = x²/2: xₙ = (1 − η)ⁿx₀. Playback advances two discrete iterations per second. Values outside |x| ≤ 4 are explicitly reported. Current readings: Position x0: 2; Objective x²/2: 2; Iteration behavior: Stable step. All parameters are adjustable with the labeled controls below.−4−202402.174.356.528.7Parameter xf(x) = x²/2n = 0
ObjectiveDiscrete iterate
Position x0
2
Objective x²/2
2
Iteration behavior
Stable step
n = 0 · t = 0 s

Gradient descent on f(x) = x²/2: xₙ = (1 − η)ⁿx₀. Playback advances two discrete iterations per second. Values outside |x| ≤ 4 are explicitly reported. Parameter changes pause playback and recompute the model at the current elapsed time. Reset restores both parameters and returns time to zero. Playback starts paused. Adjust either parameter, then choose Play or Step.

Look forAdditional computation improves an answer only within an appropriate convergence or asymptotic regime.

Pause & reason

Check your understanding

What does a positive-axis infrared Borel singularity imply in the stated toy convention?

Your turn

Try it yourself

Replace the toy coefficients by rn=(−1)ⁿn! with p=b=1. Find the Borel transform, locate its pole, and decide whether that pole obstructs the positive-axis Borel integral for g>0. Does the perturbative series become convergent?

Take a moment to make a prediction or work it out on paper. Open a hint when you need one.

Give me a hint

After dividing by n!, sum a geometric series in −t. Factorial magnitude growth is unchanged by alternating signs.

Compare with the solution

The Borel transform is 1/(1+t), with a pole at t=−1. That pole does not lie on the positive integration axis; the Laplace integral converges for positive g because the exponential controls large t. The original perturbative series still diverges factorially for every fixed nonzero g. An unobstructed toy Borel sum does not by itself prove the absence of other nonperturbative information in a field theory.

Keep the idea

Take it with you

  • Renormalization of UV poles does not guarantee a convergent perturbative series.
  • Borel singularities encode factorial patterns with convention-dependent locations.
  • Prescription dependence cancels in a complete definition, while power-correction magnitudes require additional information.

Read further

Original sources and deeper explanations to follow at your own pace.

Optional progress, saved in this browser when available. Revisit any lesson at any time.

Lesson not marked complete.