Vector Models, Auxiliary Fields, and Large-N Saddles

An auxiliary field turns the singlet sector of an $O(N)$ vector theory into a saddle problem whose action is proportional to $N$. In the two-dimensional nonlinear sigma model, this gives a renormalized gap equation and the nonperturbative mass

$$m=\mu\exp\!\left[-\frac{2\pi}{\lambda_R(\mu)}\right].$$

The saddle is the leading term of a $1/N$ expansion, not the exact finite-$N$ vacuum; its meaning depends on the auxiliary-field contour, renormalization prescription, phase, and order of infrared and large-$N$ limits.

Required background. Large-N limits, normalizations, and orders of limits supplies vector-index counting and the fixed-coupling prescription. Gap equations, dimensional transmutation, and physical mass supplies the distinction between a regulator-dependent saddle parameter and a physical mass scale.

Helpful background. The one-particle-irreducible effective action supplies the effective-action interpretation of stationarity and inverse two-point kernels.

A normalized auxiliary-field representation

Work in Euclidean two-space with $N$ real fields $\phi_i$, $i=1,\ldots,N$. Begin with the conventional unit vector $n_i n_i=1$ and

$$S[n] = \frac{N}{2\lambda_0} \int\mathrm d^2x\,(\partial_\mu n_i)^2,$$

where the bare ’t Hooft-like vector coupling $\lambda_0$ is fixed as $N\to\infty$. After

$$\phi_i=\sqrt{\frac{N}{\lambda_0}}\,n_i,$$

the kinetic term is canonical and the constraint is $\phi_i\phi_i=N/\lambda_0$. A functional delta constraint may be written as

$$Z = \int\mathcal D\phi\,\mathcal D\sigma\, \exp\!\left\{ -\frac12\int\mathrm d^2x \left[ (\partial_\mu\phi_i)^2 +\sigma\left(\phi_i\phi_i-\frac{N}{\lambda_0}\right) \right] \right\}.$$

The $\sigma$ contour is initially parallel to the imaginary axis. This is essential: it implements the delta constraint and supplies the steepest-descent direction through the real positive saddle introduced below.

Integrating the $N$ Gaussian vector components gives

$$S_{\mathrm{eff}}[\sigma] = \frac N2 \operatorname{Tr}\log\!\left(\frac{-\partial^2+\sigma}{\mu^2}\right) -\frac{N}{2\lambda_0} \int\mathrm d^2x\,\sigma.$$

Every term is $O(N)$, so stationary configurations of $\sigma$ organize the singlet expansion. This determinant is not a classical elimination of $\phi_i$: it contains the leading quantum fluctuations of all $N$ vector components.

The renormalized translationally invariant saddle

For a homogeneous symmetric saddle $\sigma=m^2>0$, stationarity gives

$$\frac1{\lambda_0} = \int^\Lambda \frac{\mathrm d^2p}{(2\pi)^2} \frac1{p^2+m^2} = \frac1{4\pi} \log\!\frac{\Lambda^2+m^2}{m^2}.$$

At $\Lambda\gg m$, define the renormalized coupling by

$$\frac1{\lambda_R(\mu)} = \frac1{\lambda_0} -\frac1{4\pi}\log\!\frac{\Lambda^2}{\mu^2}.$$

Removing the cutoff at fixed $\lambda_R(\mu)$ yields

$$\frac1{\lambda_R(\mu)} = \frac1{4\pi}\log\!\frac{\mu^2}{m^2}, \qquad m = \mu\exp\!\left[-\frac{2\pi}{\lambda_R(\mu)}\right].$$

The same equation implies

$$\mu\frac{\mathrm d\lambda_R}{\mathrm d\mu} = -\frac{\lambda_R^2}{2\pi} +O\!\left(\frac1N\right).$$

Thus the dimensionless ultraviolet coupling is exchanged for a renormalization-group invariant mass. The leading $\phi_i$ propagator,

$$\langle\phi_i(p)\phi_j(-p)\rangle = \frac{\delta_{ij}}{p^2+m^2},$$

has a pole at $p^2=-m^2$. At $N=\infty$ this saddle parameter is therefore the vector mass. At finite $N$, the pole and its residue receive $1/N$ corrections and must be defined in a specified renormalization scheme.

This derivation, including the asymptotically free gap equation, is developed in Mariño 2015, §6.2, pp. 195–203 and reviewed for general vector models in Moshe and Zinn-Justin 2003, §§2.1–2.3, pp. 74–91.

Quadratic auxiliary fluctuations and 1/N control

Write a fluctuation along the steepest-descent contour as

$$\sigma(x) = m^2+\frac{i}{\sqrt N}s(x).$$

Expanding the trace logarithm, the linear term vanishes by the gap equation and

$$S_{\mathrm{eff}}^{(2)} = \frac14 \int\frac{\mathrm d^2p}{(2\pi)^2}\, s(p)\,\Pi(p)\,s(-p),$$

where

$$\begin{aligned} \Pi(p) &= \int\frac{\mathrm d^2q}{(2\pi)^2} \frac1{(q^2+m^2)\bigl((q+p)^2+m^2\bigr)}\\ &= \frac1{4\pi} \int_0^1\frac{\mathrm dx} {m^2+x(1-x)p^2}. \end{aligned}$$

For Euclidean $p^2\ge0$, $\Pi(p)>0$; in particular,

$$\Pi(0)=\frac1{4\pi m^2}.$$

The factor $i$ in the fluctuation is what turns the negative real-$\sigma$ Hessian into a positive quadratic form on the correct contour. The $s$ propagator is proportional to $\Pi^{-1}$, and each insertion of the physical $\sigma-m^2=i s/\sqrt N$ supplies $N^{-1/2}$. This produces a systematic singlet $1/N$ expansion.

The expansion is controlled only while the chosen saddle is isolated on its integration cycle and its relevant kernels stay invertible in the momentum region being probed. It can become nonuniform when:

• an infrared scale is sent to zero before the $1/N$ estimate is made;
• a finite-volume symmetric state mixes saddles that separate only after an infinite-volume limit;
• a critical auxiliary eigenvalue approaches zero;
• another saddle has an action difference of order $1/N$;
• composite-operator or coupling counterterms required at subleading order are omitted.

The auxiliary-field method reorganizes diagrams; it does not remove renormalization. Beyond leading order, the coupling, the $\phi_i\phi_i$ composite operator, pole positions, and residues must be renormalized consistently Moshe and Zinn-Justin 2003, §§2.4–2.6, pp. 91–110.

First application: mass generation without symmetry breaking

The homogeneous saddle has

$$\langle\phi_i\rangle=0, \qquad \langle\sigma\rangle=m^2>0.$$

It generates a mass while preserving $O(N)$. This is not a Higgs-like expectation value for one component. In two dimensions, the symmetric massive phase is also compatible with the absence of spontaneous breaking of a continuous internal symmetry at finite $N$.

The calculation demonstrates three separate statements:

1. dimensional transmutation fixes the leading large-$N$ scale $m$;
2. the positive quadratic kernel supplies local stability along the deformed auxiliary contour;
3. finite-$N$ observables require the $1/N$ expansion and the same renormalization conditions.

Shared calculation. The large-N counting and topology map shows why vector cactus graphs, rather than matrix-planar graphs, survive this limit. The large-N scaling comparison records the vector normalization and its infrared failure test.

Common pitfalls

Dropping the auxiliary contour. A real $\sigma$ fluctuation appears to have the wrong Hessian sign. The original delta-function contour and its steepest-descent deformation determine the correct fluctuation direction.

Calling the saddle exact. The determinant is leading because it sums $N$ vector components, but auxiliary fluctuations produce genuine $1/N$ corrections.

Equating a bare cutoff solution with a physical prediction. The cutoff dependence must be absorbed into $\lambda_R(\mu)$ before the transmuted scale can be compared between regulators.

Exercises

1. Derive the leading beta function from the renormalized gap equation.

Solution

Hold $m$ fixed and differentiate

$$\lambda_R^{-1} = \frac1{4\pi}\log\!\frac{\mu^2}{m^2}.$$

Then

$$-\frac{\beta(\lambda_R)}{\lambda_R^2} = \frac1{2\pi},$$

so $\beta(\lambda_R)=-\lambda_R^2/(2\pi)$ at leading order.

2. Evaluate $\Pi(0)$.

Solution

At zero momentum,

$$\Pi(0) = \int\frac{\mathrm d^2q}{(2\pi)^2} \frac1{(q^2+m^2)^2}.$$

Using polar coordinates and $u=q^2$ gives

$$\Pi(0) = \frac1{4\pi} \int_0^\infty\frac{\mathrm du}{(u+m^2)^2} = \frac1{4\pi m^2}.$$

3. Explain why the quadratic kernel does not by itself prove global dominance.

Solution

Positivity of $\Pi$ proves local stability along the selected steepest-descent contour. It does not compare the saddle action with other real or complex saddles, determine the contour’s intersection numbers, or exclude an $N$-dependent collision. Those are global questions.

References

• Mariño, M. (2015). Instantons and Large N: An Introduction to Non-Perturbative Methods in Quantum Field Theory. Cambridge University Press. doi:10.1017/CBO9781107705968.
• Moshe, M., and Zinn-Justin, J. (2003). “Quantum Field Theory in the Large N Limit: A Review.” Physics Reports 385, 69–228. doi:10.1016/S0370-1573(03)00263-1. Open PDF.