Anomaly and Generalized-Symmetry Constraints on Infrared Phases

An anomaly is an exact obstruction to realizing a declared symmetry with purely local counterterms. Because the obstruction is unchanged along renormalization-group flow, it rules out infrared phases that cannot reproduce it. It does not, by itself, choose among all phases that can. The practical task is therefore to specify the symmetry and its backgrounds precisely, compute the obstruction modulo allowed counterterms, and test every proposed infrared realization against that same data.

Required background. ’t Hooft anomaly matching supplies the obstruction and inflow language, breaking higher-form symmetry and diagnosing phases supplies the extended-operator diagnostics, and theta terms, periodicity, and vacuum sectors supplies the global counterterm data.

Helpful background. Theta dependence, CP, and model-dependent branches gives the branch interpretation used in the Yang–Mills example.

Shared comparison. The sector and periodicity comparison supplies the global-form and theta-period inputs, while the exact and rigorous status comparison separates the exact anomaly obstruction from model-dependent infrared inference and open phase-selection questions.

Background fields expose the obstruction

Let $G$ denote the full global symmetry, including any higher-form factors, and let $B$ denote its background fields. A symmetry transformation $g\in G$ can act on the partition function as

$$Z_X[g\!\cdot\!B] =Z_X[B]\exp\!\bigl(i\mathcal A_g[B]\bigr).$$

The phase $\mathcal A_g[B]$ is meaningful only modulo the variation of a local $d$-dimensional counterterm. An ’t Hooft anomaly is the equivalence class that remains after every counterterm allowed by locality, quantization, the global form of $G$, and the spacetime structure has been considered. Often it is represented by inflow from a $(d+1)$-dimensional invertible action.

Generalized symmetries enlarge both sides of this statement. A $p$-form symmetry couples to a $(p+1)$-form background, and its charged observables are extended rather than local. Gaiotto, Kapustin, Seiberg, and Willett formulate these symmetries and their anomalies systematically in Gaiotto et al. 2015, §§ 1, 3, and 7.

Three specifications are indispensable before drawing an infrared conclusion:

• the faithful symmetry group, including quotients and higher-group extensions;
• the admissible background bundles, defects, and spin or orientation structure;
• the set of quantized local counterterms regarded as changes of scheme.

Changing any of them can change the anomaly question. A Lie algebra or an action on perturbative local fields is not enough.

Matching is an exclusion test

Renormalization-group flow cannot change the anomaly class, so the infrared generating functional must have the same response:

$$\bigl[\mathcal A_{\mathrm{UV}}\bigr] =\bigl[\mathcal A_{\mathrm{IR}}\bigr].$$

The brackets are essential: equality is modulo allowed local counterterms. The main infrared mechanisms are:

Infrared realization	How the anomaly is reproduced	What must be checked
Spontaneous symmetry breaking	Distinct vacua are permuted; a low-energy order-parameter action or domain-wall inflow carries the response	The broken subgroup, vacuum multiplicity, defects, and residual anomaly
Gapless degrees of freedom	Massless fields or a conformal theory have anomalous currents or background dependence	The full anomaly coefficients and faithful symmetry action, not only a matching central charge
Topological order or a TQFT	Long-range entanglement and topological operators furnish the anomalous response	Genuine operator spectrum, symmetry fractionalization, and coupling to the same backgrounds
Symmetry extension or emergent gauge structure	The obstruction becomes trivial only after enlarging the infrared structure	Whether the extension is dynamical and whether the original symmetry is recovered faithfully
Boundary of a higher-dimensional bulk	Bulk inflow cancels the boundary variation	That the proposed system really includes the bulk; this is not a stand-alone $d$-dimensional realization

A unique, trivially gapped, symmetry-preserving vacuum has no nontrivial long-distance response and therefore cannot match a nonzero anomaly. Excluding that one option is powerful, but the table makes clear why the anomaly rarely selects a unique replacement.

The logic is summarized in the figure. Read it from the declared ultraviolet data to the obstruction, then across the parallel infrared alternatives; the branches are possibilities, not a dynamical decision tree.

Anomaly matching constrains the infrared without solving it. Every surviving option must reproduce the same counterterm-equivalence class under the stated global-form and spacetime assumptions; the schematic is not a claim that all options occur in one model.

CP and center symmetry at theta equal to pi

Consider four-dimensional pure $SU(N)$ Yang–Mills theory with its electric $\mathbb Z_N^{(1)}$ one-form symmetry. Coupling that symmetry to a two-form background $B$ requires quantized background counterterms. At $\theta=\pi$, CP reverses the theta term, while returning $-\pi$ to $\pi$ uses the $2\pi$ periodic identification. In a nontrivial $B$ background, that identification also shifts the quantized counterterm.

For even $N$, no allowed counterterm choice makes both CP and the one-form symmetry manifest at $\theta=\pi$: there is a mixed anomaly. For odd $N$, a counterterm can preserve CP at $\theta=\pi$, but it is incompatible with the natural CP-preserving choice at $\theta=0$; assuming a smooth symmetry-preserving interpolation gives a global inconsistency. The arithmetic and its dependence on the global data are derived in Gaiotto et al. 2017, §§ 2.2–2.4.

Under these assumptions, the familiar confining proposal with unbroken center symmetry and a unique trivial gapped CP-invariant vacuum at $\theta=\pi$ is excluded. Several outcomes remain:

• CP can break, giving degenerate vacua whose interchange realizes the obstruction;
• the theory can become gapless at the special theta value;
• a nontrivial topological sector can remain in the infrared;
• an intervening transition can invalidate an assumed smooth interpolation, especially in the odd-$N$ global-inconsistency case.

The often-drawn two-branch cusp is compatible with the first option, not derived uniquely from the anomaly. If CP breaks, anomaly inflow also constrains the degrees of freedom supported by a wall between the two vacua.

Rebuild the data when the theory changes

Adding matter can explicitly remove the one-form symmetry by screening Wilson lines. Changing $SU(N)$ to $SU(N)/\mathbb Z_k$ changes the genuine line operators, background sums, and discrete theta parameters. Working on non-spin manifolds can change counterterm quantization. Massless fermions can make the continuous theta angle redundant through an anomalous chiral rotation.

Any of these changes requires returning to the first step. It is invalid to retain the pure-$SU(N)$ anomaly conclusion after changing the very backgrounds that defined it.

The same caution applies to numerics and semiclassics. A finite-volume spectrum or a controlled compactification can provide decisive dynamical evidence for one matching option, but agreement with the anomaly is a consistency check, not proof that the same phase persists after every limit is taken.

Common pitfalls

Naming only the Lie algebra. Anomalies depend on the global symmetry, genuine extended operators, and admissible bundles.

Treating counterterm dependence as arbitrariness. Individual response phases can change by allowed counterterms; the nontrivial equivalence class cannot.

Concluding “therefore symmetry breaking.” Symmetry breaking is one matching mechanism. Gapless or topological infrared physics may also reproduce the obstruction.

Importing the even-$N$ statement to odd $N$. The former is a direct anomaly at $\theta=\pi$; the latter is naturally expressed as an incompatibility between counterterm choices at two theta values.

Exercises

Suppose a theory has a nontrivial anomaly for an unbroken symmetry $G$. Explain why a unique trivial gapped vacuum cannot match it, and identify the missing premise if one instead proposes two degenerate vacua.

Solution

A unique trivial gapped vacuum has a local infrared generating functional. Its background variation can therefore be removed by a local counterterm, so its anomaly class is zero. This contradicts the nonzero ultraviolet class. Two degenerate vacua can match the anomaly only if $G$ acts on the vacuum set and the associated walls or low-energy action reproduce the required background phase; degeneracy by itself is insufficient.

Let $\mathcal A_g[B]$ be shifted by the variation $C[g\!\cdot\!B]-C[B]$ of an allowed local counterterm $C[B]$. Show that whether the anomaly class is trivial is unchanged.

Solution

After adding $C$, the transformed partition function acquires

$$\mathcal A'_g[B] =\mathcal A_g[B]+C[g\!\cdot\!B]-C[B].$$

Thus $\mathcal A'$ and $\mathcal A$ are representatives of the same equivalence class. If one can be removed by an allowed counterterm, so can the other; if neither can, the obstruction remains.

References

• Gaiotto, Davide, Anton Kapustin, Nathan Seiberg, and Brian Willett. “Generalized Global Symmetries.” Journal of High Energy Physics 2015, no. 2 (2015): 172. arXiv. DOI.
• Gaiotto, Davide, Anton Kapustin, Zohar Komargodski, and Nathan Seiberg. “Theta, Time Reversal, and Temperature.” Journal of High Energy Physics 2017, no. 5 (2017): 091. arXiv. DOI.