The O(N) Model as a Strong-Coupling Laboratory

The two-dimensional $O(N)$ nonlinear sigma model is a clean laboratory for asymptotic freedom and dimensional transmutation because its microscopic action contains only a dimensionless coupling, yet for $N>2$ its long-distance correlators develop a finite scale. The statement has important boundaries: the one-loop coefficient vanishes at $N=2$, the leading large-$N$ saddle is not the exact finite-$N$ spectrum, and the absence of continuous symmetry breaking in $1+1$ dimensions does not by itself prove a mass gap.

Required background. Sigma-model target geometry and control fixes the normalization and dimension, while running couplings and dimensional transmutation explains how an RG-invariant scale replaces a dimensionless coupling. Helpful background. Vector models and auxiliary large-N saddles supplies the saddle-point and fluctuation expansion.

The constrained model and its declared regime

Work in two-dimensional Euclidean space with a unit vector

$$n^a(x)n^a(x)=1,\qquad a=1,\ldots,N,$$

and action

$$S_E[n] =\frac{1}{2g_0^2}\int\mathrm d^2x\, \partial_\mu n^a\partial_\mu n^a .$$

The global symmetry is $O(N)$, acting linearly on $n$. The target is $S^{N-1}$. We first take infinite volume and zero temperature, then remove an ultraviolet regulator along a continuum RG trajectory. Correlation lengths and pole positions below refer to connected correlators in that continuum theory.

At weak coupling the finite-$N$ beta function begins

$$\beta(g) \equiv\mu\frac{\mathrm dg}{\mathrm d\mu} =-\frac{N-2}{4\pi}g^3+O(g^5).$$

Thus $N>2$ is asymptotically free. Integrating the leading term gives an RG-invariant scale of the form

$$\Lambda_{\mathrm{RG}} =\mu\, \exp\!\left[-\frac{2\pi}{(N-2)g^2(\mu)}\right] \left[1+O(g^2)\right],$$

where the multiplicative constant depends on the renormalization scheme. The beta-function coefficient and its geometric origin were established in the early sigma-model analyses Polyakov 1975, pp. 79–81.

Running to strong coupling shows that perturbation theory loses control near $\Lambda_{\mathrm{RG}}$; it does not by itself identify a particle pole or prove the infrared phase. The large-$N$ saddle supplies a controlled bridge to a correlation length.

Leading large-N gap equation

Hold

$$t_0\equiv Ng_0^2$$

fixed as $N\to\infty$, and introduce the canonically normalized field $\varphi^a=n^a/g_0$. Its constraint is $\varphi^2=N/t_0$. A convenient Euclidean auxiliary-field representation is

$$S_E[\varphi,\lambda] =\frac12\int\mathrm d^2x\, \left[ (\partial_\mu\varphi^a)^2 +\lambda\left(\varphi^a\varphi^a-\frac{N}{t_0}\right) \right],$$

with the multiplier contour chosen so that the steepest-descent saddle has a real positive $\lambda=m^2$. Integrating the $N$ components of $\varphi$ gives

$$\frac{S_{\mathrm{eff}}}{N} =\frac12\operatorname{Tr}\ln(-\partial^2+\lambda) -\frac{1}{2t_0}\int\mathrm d^2x\,\lambda .$$

For a translation-invariant saddle $\lambda=m^2$, stationarity yields

$$\frac{1}{t_0} =\int_{\lvert p\rvert<\Lambda} \frac{\mathrm d^2p}{(2\pi)^2}\, \frac{1}{p^2+m^2} =\frac{1}{4\pi} \ln\!\left(1+\frac{\Lambda^2}{m^2}\right).$$

With this circular sharp cutoff,

$$m =\frac{\Lambda} {\sqrt{\exp(4\pi/t_0)-1}} \underset{t_0\ll1}{\sim} \Lambda\,e^{-2\pi/t_0}.$$

The exponential is dimensional transmutation. The prefactor in a different regulator is different until $t_0$ is matched to a specified renormalized scheme. The derivation and the large-$N$ organization are given in Marino 2015, § 6.2, pp. 194–200.

Why the saddle parameter is a physical leading mass here

At the saddle, the $\varphi$ two-point function is

$$\left\langle \varphi^a(p)\varphi^b(-p) \right\rangle_{\!N=\infty} =\frac{\delta^{ab}}{p^2+m^2}.$$

After analytic continuation, its pole is at $p^2=m^2$ in Lorentzian signature. Because $n^a$ is a physical local field carrying the global $O(N)$ vector quantum number, this establishes at leading order that

$$\xi^{-1}=m_{\mathrm{pole}}=m \qquad (N=\infty).$$

This equality is model- and order-specific. Fluctuations of $\lambda$ generate $1/N$ corrections to the pole, residue, and relation between $m$ and a chosen $\Lambda$ parameter. The auxiliary $\lambda$ field also has a nontrivial composite-channel propagator; its saddle value is not the mass of a “lambda particle.”

The chapter’s gap-equation and physical-mass analysis makes this translation test explicit across several models. The strong-coupling laboratory map and regime comparison show which parts of the argument transfer and which do not.

The finite-N statement and its evidence

For each fixed integer $N\ge3$, the continuum model is asymptotically free and has a finite correlation length. The exact integrable description supplies factorized scattering and mass ratios, while continuum lattice calculations test dimensionless ratios and the approach to scaling. These routes agree on the existence of a massive $O(N)$ vector multiplet, but they make different assumptions and control different errors; their comparison belongs to Strong-Coupling Phases and Cross-Method Evidence.

The large-$N$ saddle explains the mechanism and becomes systematically improvable in $1/N$. It is not a proof at every finite $N$, and its leading mass formula should not be substituted for an exact finite-$N$ mass–$\Lambda$ relation.

Why N = 2 is a different theory

For $N=2$, write

$$n=(\cos\theta,\sin\theta).$$

For smooth configurations the action becomes that of a compact free scalar,

$$S_E =\frac{1}{2g_0^2}\int\mathrm d^2x\,(\partial_\mu\theta)^2.$$

The perturbative coefficient proportional to $N-2$ vanishes. Vortices are singular from the perspective of the smooth continuum chart and require a UV prescription, such as a lattice model with compact variables and a vortex core. Their unbinding produces Berezinskii–Kosterlitz–Thouless behavior rather than the $O(N>2)$ asymptotically free mass-generation mechanism Kosterlitz and Thouless 1973, §§ 2–4.

At $N=1$, the target $S^0$ has two isolated points and no local tangent fluctuations, so the written continuum sigma-model kinetic term has no ordinary local dynamics. Neither endpoint can be treated by blindly substituting $N=1$ or $2$ into the $N>2$ conclusion.

Symmetry realization is not the gap equation

In $1+1$ dimensions at finite $N$, a continuous internal symmetry cannot be spontaneously broken under the standard locality and spectral assumptions. Therefore $\langle n^a\rangle=0$ in the infinite-volume vacuum. But “no continuous symmetry breaking” is weaker than “massive”: the $O(2)$ compact-boson phase gives a counterexample with algebraic correlations. A mass-gap claim needs the dynamics of the declared model, not only the symmetry theorem.

The order of limits matters. Taking $N\to\infty$ before the infrared limit can suppress fluctuations that restore a continuous symmetry at every finite $N$. For the $O(N)$ saddle above, the symmetric massive solution is compatible with the theorem, but this compatibility is a check rather than the source of the mass.

Common pitfalls

Using the gap integral without its regulator. The equality between $t_0$, $\Lambda$, and $m$ is regulator-dependent. Only a renormalized mass–scale relation in a named scheme, or a dimensionless observable ratio, can be compared across regulators.

Calling strong running a mass-gap proof. Perturbative RG locates the scale at which weak-coupling perturbation theory fails. The pole or exponential correlation length requires nonperturbative information.

Including O(2) in the N > 2 claim. The beta-function coefficient vanishes at $N=2$, and compact vortices control the BKT physics. This is a change of mechanism, not a small numerical correction.

Exercises

1. Evaluate the sharp-cutoff integral in the gap equation and solve exactly for $m/\Lambda$.

Solution

Using polar coordinates,

$$\int_{\lvert p\rvert<\Lambda} \frac{\mathrm d^2p}{(2\pi)^2}\frac{1}{p^2+m^2} =\frac{1}{2\pi}\int_0^\Lambda \frac{p\,\mathrm dp}{p^2+m^2} =\frac{1}{4\pi}\ln\!\left(1+\frac{\Lambda^2}{m^2}\right).$$

Equating this to $1/t_0$ gives

$$\frac{\Lambda^2}{m^2}=e^{4\pi/t_0}-1, \qquad \frac{m}{\Lambda} =\left(e^{4\pi/t_0}-1\right)^{-1/2}.$$

2. In a square box of side $L$ with periodic boundary conditions, replace the integral by a momentum sum. Explain qualitatively why the infinite-volume limit must be taken before identifying the pole mass.

Solution

The momenta become $p_\mu=2\pi k_\mu/L$, so

$$\frac1{t_0} =\frac1{L^2} \sum_{k\in\mathbb Z^2} \frac{1}{(2\pi k/L)^2+m_L^2},$$

with the same UV regulator imposed on the sum. The zero mode and the discrete spacing change the saddle when $mL$ is not large. A finite-volume energy level is not yet an infinite-volume pole; one must take $L/\xi\to\infty$ and control the finite-size corrections before extracting $m_{\mathrm{pole}}$.

Continue

Proceed to Gap Equations, Dimensional Transmutation, and Physical Mass to renormalize the saddle and distinguish several meanings of “mass.”

References

• Kosterlitz, J. M., and D. J. Thouless. “Ordering, Metastability and Phase Transitions in Two-Dimensional Systems.” Journal of Physics C: Solid State Physics 6 (1973): 1181–1203. DOI.
• Mariño, Marcos. Instantons and Large N: An Introduction to Non-Perturbative Methods in Quantum Field Theory. Cambridge University Press, 2015, § 6.2. DOI.
• Polyakov, A. M. “Interaction of Goldstone Particles in Two Dimensions. Applications to Ferromagnets and Massive Yang–Mills Fields.” Physics Letters B 59 (1975): 79–81. DOI.