Infrared and Collinear Safety

Infrared and collinear (IRC) safety is a continuity requirement on the observable, not on an individual amplitude. A measurement function is IRC safe when adding a zero-energy particle or replacing one massless momentum by exactly collinear daughters does not change the measured value. This makes degenerate final states receive equal weights, allowing the real and virtual singularities associated with unresolved radiation to cancel order by order.

Required background. Measurement Functions and Inclusive Observables defines the family of weights $F_n$. Bloch–Nordsieck and KLN Cancellation supplies the inclusive cancellation mechanism and its assumptions.

The two unresolved limits

Write the hard configuration as $\{p\}_n$. As $\lambda\to0^+$, soft safety requires

$$F_{n+1}(\{p\}_n,\lambda q) \longrightarrow F_n(\{p\}_n).$$

Collinear safety requires, for two massless daughters with $p_a\to zp$ and $p_b\to(1-z)p$,

$$\begin{aligned} &F_{n+1}(\ldots,p_a,p_b,\ldots)\\ &\qquad\longrightarrow F_n(\ldots,p,\ldots), \qquad 0<z<1. \end{aligned}$$

The limiting equality must include any recombination rule and bin assignment. It is not enough that the numerical change be “usually small.” The singular phase-space measure probes arbitrarily close to the boundary, so a finite jump can multiply a divergent integral.

In a collinear region, the unresolved real-emission contribution has the schematic form

$$\mathrm d\sigma_R \sim \mathrm d\sigma_B\, \frac{\alpha_s}{2\pi} \mathcal P(z)\,\mathrm dz \frac{\mathrm d\theta^2}{\theta^2}.$$

up to process-dependent color, spin, and azimuthal structure. The soft enhancement is an endpoint of the appropriate splitting kernel; writing an additional independent $\mathrm dE/E$ factor would double-count that endpoint. The virtual correction contains the opposite unresolved poles. If $F_{n+1}-F_n$ vanishes sufficiently rapidly in both soft and collinear limits, the weighted difference is integrable after the common singular approximation is exposed. Catani and Seymour formulate the necessary jet-function limits in Catani and Seymour 1997, § 7, pp. 343–344.

Why the criterion enables cancellation

For an infrared-safe observable with final-state unresolved singularities, choose a local counterterm $\mathrm d\sigma_A$ that reproduces $\mathrm d\sigma_R$ in every singly unresolved limit and maps the real configuration to lower-multiplicity momenta. The NLO correction then has the exact add–subtract form

$$\begin{aligned} \delta\sigma_{\mathrm{NLO}} ={}&\int_{n+1} \left[\mathrm d\sigma_R F_{n+1} -\mathrm d\sigma_A F_n^{\mathrm{map}}\right]_{\epsilon=0}\\ &+\int_n \left[\mathrm d\sigma_V +\int_1 \mathrm d\sigma_A\right]_{\epsilon=0}F_n. \end{aligned}$$

$F_n^{\mathrm{map}}$ evaluates the lower-multiplicity measurement on a mapped configuration that approaches the unresolved parent momenta. The first integral is finite because both the matrix-element singularity and the measurement limit are matched locally; the integrated counterterm cancels the explicit virtual poles in the second integral. With massless incoming partons, the appropriate mass-factorization counterterm must also be included in the lower-multiplicity line. A subtraction method supplies the counterterm and its exact phase-space map.

The KLN theorem concerns sums over degenerate initial and final states under stated assumptions Kinoshita 1962, pp. 650–677 and Lee and Nauenberg 1964, pp. B1549–B1562. IRC safety is the observable-level condition that makes the relevant final-state degeneracy compatible with those sums. It does not prove that every perturbative coefficient is small, nor does it remove initial-state collinear factorization or nonperturbative corrections.

Examples and counterexamples

Energy-weighted angular observables. If

$$V=\frac1Q\sum_i E_i f(\widehat{\boldsymbol p}_i)$$

with bounded continuous $f$, a soft particle contributes $O(E_s/Q)$ and a collinear splitting preserves the energy sum in the common direction. The observable passes both elementary tests.

Thrust. For a final state in its center-of-mass frame,

$$T=\max_{|\boldsymbol n|=1} \frac{\sum_i|\boldsymbol p_i\cdot\boldsymbol n|} {\sum_i|\boldsymbol p_i|}.$$

A soft momentum perturbs numerator and denominator continuously. Exactly collinear daughters contribute the same projected and total momentum as their parent, so $T$ is IRC safe. Near $T\to1$, however, logarithms of $1-T$ become large; safety guarantees finiteness, not fixed-order convergence.

Sequential recombination jets. A jet algorithm is safe only if an unresolved particle is clustered or ignored without reorganizing hard jets discontinuously. The distance measure, recombination prescription, and tie-breaking at boundaries are all part of the measurement definition.

Particle multiplicity. $N_{n+1}=N_n+1$ even for an arbitrarily soft emission or collinear splitting. Bare massless-parton multiplicity is not IRC safe.

Leading-particle energy. A collinear splitting can change which daughter is leading and reduce the largest individual energy by a finite fraction. Without a fragmentation function or other nonperturbative definition, this is not a safe partonic observable.

The Sterman–Weinberg two-jet fraction is an early explicit example: energy outside two cones is allowed below a resolution fraction, so the observable is inclusive over sufficiently soft and collinear radiation Sterman and Weinberg 1977, pp. 1436–1439.

Boundaries, recoil, and non-global restrictions

A sharp cut does not automatically make an observable unsafe. A bin indicator can be safe if an unresolved emission changes the measured variable continuously; configurations exactly on the cut form a lower-dimensional boundary. Trouble occurs when an arbitrarily soft or collinear emission causes a finite reclassification over a region with nonzero singular weight.

Recoil can make this subtle. If an axis is defined by minimizing or maximizing a global functional, a soft emission may shift it. The shift must vanish quickly enough that the measured value approaches the lower-multiplicity result. Likewise, an observable that measures radiation only in part of phase space may remain IRC safe but develop non-global logarithms: finiteness and simple all-orders factorization are distinct questions.

IRC safety also does not eliminate hadronization. It suppresses sensitivity to arbitrarily low scales, often turning it into power corrections such as $(\Lambda/Q)^p$, but the exponent and coefficient depend on the observable. Near endpoints, those corrections can be enhanced and a purely fixed-order parton-level interpretation can fail.

A practical safety test

For every measurement implementation, test these operations analytically and numerically:

1. append a momentum $\lambda q$ and decrease $\lambda$ over several orders of magnitude;
2. use a specified momentum-conserving on-shell splitting map to resolve $p$ into two daughters, including a recoiling spectator when the map requires one, and send its transverse parameter $k_\perp\to0$;
3. repeat near cuts, recombination ties, axes, and flavor tags;
4. verify the limiting value and the rate of convergence, not only agreement at one tolerance.

These are static reproducibility checks. No runnable laboratory interface is currently available for this page.

Common pitfalls

Checking only the soft limit. Massless theories also have collinear singularities. Test generic $z$, including asymmetric but nonsoft splittings.

Confusing safety with small corrections. An IRC-safe observable can contain large Sudakov or non-global logarithms and substantial power corrections. Safety says the perturbative coefficient is finite after the prescribed cancellations.

Assuming every sharp cut is unsafe. What matters is the unresolved limit of the measured value. Analyze the cut boundary rather than replacing all bins by smooth weights without need.

Using a finite regulator result as proof. A mass, energy cutoff, or angular cutoff can hide an unsafe logarithm. Remove the regulator while keeping the observable definition fixed.

Check your understanding

Apply the two limits to the thrust numerator and denominator separately. Then apply them to particle multiplicity. Explain why the first observable has a well-defined lower-multiplicity limit while the second changes by a finite amount.

Where to continue

• Real–Virtual Cancellation and Subtraction turns the limiting equalities into finite numerical terms.
• Jets and Event-Shape Observables develops the main examples and their resolution scales.
• Sudakov Logarithms and Resummation explains why a finite observable can still require all-orders organization.

References

• Catani, Stefano, and Michael H. Seymour. “A General Algorithm for Calculating Jet Cross Sections in NLO QCD.” Nuclear Physics B 485 (1997): 291–419; erratum 510 (1998): 503–504. DOI. Open preprint.
• Kinoshita, Toichiro. “Mass Singularities of Feynman Amplitudes.” Journal of Mathematical Physics 3 (1962): 650–677. DOI.
• Lee, T. D., and Michael Nauenberg. “Degenerate Systems and Mass Singularities.” Physical Review 133 (1964): B1549–B1562. DOI.
• Sterman, George, and Steven Weinberg. “Jets from Quantum Chromodynamics.” Physical Review Letters 39 (1977): 1436–1439. DOI.