Large-N Factorization and Master-Field Claims

Large-$N$ factorization means that connected correlators of properly normalized invariant operators vanish in a specified state. For normalized single traces in a matrix-like theory,

$$\langle\widehat{\mathcal O}_1\widehat{\mathcal O}_2\rangle = \langle\widehat{\mathcal O}_1\rangle \langle\widehat{\mathcal O}_2\rangle +O(N^{-2}).$$

This conclusion requires a uniform genus expansion, a selected clustering phase, and a declared order of volume, source, and $N$ limits. It does not imply that every matrix becomes one number or that a unique ordinary classical field configuration exists.

Required background. Double-line counting and the topological expansion supplies the connected-boundary count.

Helpful background. Clustering, vacuum assumptions, and long-range correlations supplies the state-selection condition that excludes macroscopic mixtures.

Factorization as connected-correlator suppression

Let

$$\widehat{\mathcal O}_k(x) = \frac1N\operatorname{tr}M(x)^k$$

in a single-trace matrix or adjoint theory with action $N\operatorname{tr}V$. Assume:

1. the couplings and regulator are scaled as on the chapter’s normalization page;
2. expectation values are taken in one normalized state $\omega_N$ whose large-$N$ limit exists;
3. the state clusters in the infinite-volume limit, or a source has selected one pure phase before it is removed;
4. the external positions and momenta stay in a regime where the $1/N$ estimates are uniform;
5. no critical, infrared, or double-scaled enhancement compensates the nominal suppression.

For $r$ normalized trace insertions, ribbon counting gives

$$\left\langle \widehat{\mathcal O}_{k_1}\cdots \widehat{\mathcal O}_{k_r} \right\rangle_{\mathrm c} = O\!\left(N^{2-2r}\right)$$

at leading genus. In particular,

$$\operatorname{Var}(\widehat{\mathcal O}) = \left\langle\widehat{\mathcal O}^2\right\rangle -\left\langle\widehat{\mathcal O}\right\rangle^2 = O(N^{-2}).$$

The cumulant expansion then gives, for any fixed number of insertions,

$$\left\langle \prod_{a=1}^r\widehat{\mathcal O}_a \right\rangle = \prod_{a=1}^r \left\langle\widehat{\mathcal O}_a\right\rangle +O(N^{-2}),$$

where the first correction comes from one connected two-point block. This is the precise sense in which normalized invariant observables have vanishing relative fluctuations.

The $N^{-2}$ rate is matrix-like. In vector models, normalized singlet connected correlators are generally suppressed by powers of $1/N$ instead. “Factorization” does not fix a universal exponent until the index family and operator normalization are stated.

First application: the Gaussian matrix law is not one scalar

Consider

$$Z_N = \int\mathrm dM\, \exp\!\left(-\frac N2\operatorname{tr}M^2\right).$$

For the normalized moments

$$\widehat{\mathcal O}_{2k} = \frac1N\operatorname{tr}M^{2k},$$

the planar limit is the semicircle law Mariño 2015, §8.2, pp. 243–258. Its first moments are

$$\lim_{N\to\infty} \left\langle\widehat{\mathcal O}_2\right\rangle =1, \qquad \lim_{N\to\infty} \left\langle\widehat{\mathcal O}_4\right\rangle =2,$$

and

$$\left\langle \widehat{\mathcal O}_2\widehat{\mathcal O}_4 \right\rangle_{\mathrm c} = O(N^{-2}).$$

The invariant moments concentrate, but no ordinary number $m$ can reproduce them: $m^2=1$ would imply $m^4=1$, not $2$. The limiting object can instead be represented by the deterministic eigenvalue density

$$\rho_{\mathrm{sc}}(x) = \frac1{2\pi}\sqrt{4-x^2}, \qquad \lvert x\rvert\le2,$$

or by a semicircular element in a noncommutative probability space. Factorization therefore describes concentration of invariant data, not collapse of every microscopic degree of freedom to one classical value.

Why phase selection and clustering matter

Suppose two clustering phases have an invariant order parameter with limiting values $+v$ and $-v$. In the equal mixed state,

$$\langle\widehat{\mathcal O}\rangle_{\mathrm{mix}}=0, \qquad \langle\widehat{\mathcal O}^2\rangle_{\mathrm{mix}}\longrightarrow v^2.$$

The variance remains $O(1)$ even if each pure phase separately factorizes. A finite-volume symmetric state can realize exactly such a mixture. A valid factorization statement must therefore specify an order such as

$$\lim_{h\to0^+} \lim_{V\to\infty} \lim_{N\to\infty} \langle\widehat{\mathcal O}\rangle_{N,V,h},$$

or justify a different order. Here $h$ is a phase-selecting source, not a genus. If domain walls become light or correlation lengths diverge as $N$ grows, even fixed-phase connected estimates may cease to be uniform.

What “master field” can mean

A master field is best defined by the data it reproduces. Several inequivalent constructions occur:

• Concentrated invariant law. The limiting values of all normalized invariant moments define a deterministic linear functional.
• Collective classical variable. In a one-matrix model, an eigenvalue density or resolvent can encode all single-trace moments.
• Noncommutative probability. A tuple of noncommuting operators and a state can reproduce mixed planar moments; freeness replaces ordinary statistical independence.
• Stochastic or gauge-field representation. In special theories, the planar Schwinger–Dyson equations may be represented by stochastic variables or a gauge connection.

These representations need not be unique as ordinary configurations. Gauge-related representatives, different operator realizations with the same noncommutative law, or distinct constructions of the same invariant moments can be physically equivalent. Gopakumar and Gross give explicit noncommutative constructions and emphasize the Cuntz-algebra structure Gopakumar and Gross 1995, §§2–4, pp. 383–400. Douglas shows how a stochastic construction encodes the factorized Schwinger–Dyson equations Douglas 1995, §§2–4, pp. 118–124.

Existence of all separate moment limits is also not automatically enough: one needs positivity, consistency among products, and adequate control of the operator class to obtain a useful limiting state. Volume 16 develops operator-algebraic formulations; this page uses only the invariant-moment meaning.

Shared calculation. The large-N counting and topology map fixes the boundary powers behind connected suppression. The large-N scaling comparison states the operator and state hypotheses that must accompany factorization.

Common pitfalls

Applying factorization to unnormalized traces. The connected two-point function of $\operatorname{tr}M^k$ is $O(N^0)$, not $O(N^{-2})$. The suppression is relative to its $O(N)$ one-point normalization.

Using a symmetric mixture as though it were a pure phase. Macroscopic phase fluctuations survive at $N=\infty$. Select a clustering state and declare the source, volume, and $N$ limits.

Equating noncommutative classicality with one ordinary field. A limiting moment functional can be deterministic while its representing operators remain noncommuting and its eigenvalue law has finite width.

Exercises

1. Use cumulants to find the leading correction to a three-point product of normalized traces.

Solution

Write the moment as the sum over set partitions. The fully disconnected product is $O(1)$. Each partition with one connected pair and one one-point block is $O(N^{-2})$, while the connected three-point cumulant is $O(N^{-4})$. Hence

$$\langle O_1O_2O_3\rangle = \langle O_1\rangle\langle O_2\rangle\langle O_3\rangle +\sum_{\mathrm{pairs}} \langle O_iO_j\rangle_{\mathrm c}\langle O_k\rangle +O(N^{-4}).$$

2. Show directly that the equal mixture of two values $\pm v$ does not factorize.

Solution

The mixture has $\langle O\rangle=(v-v)/2=0$ and $\langle O^2\rangle=(v^2+v^2)/2=v^2$. Therefore

$$\langle O^2\rangle-\langle O\rangle^2=v^2,$$

which is not suppressed.

3. Why can the semicircle density be deterministic while an eigenvalue remains distributed?

Solution

The empirical measure $N^{-1}\sum_i\delta(x-\lambda_i)$ converges to a fixed density. This concentrates collective moments such as $N^{-1}\operatorname{tr}M^k$, but the limiting density has nonzero width. Determinism of the measure is not determinism of a sampled eigenvalue.

References

• Douglas, M. R. (1995). “Stochastic Master Fields.” Physics Letters B 344, 117–126. doi:10.1016/0370-2693(94)01547-P. Open PDF.
• Gopakumar, R., and Gross, D. J. (1995). “Mastering the Master Field.” Nuclear Physics B 451, 379–415. doi:10.1016/0550-3213(95)00340-X. Open PDF.
• Mariño, M. (2015). Instantons and Large N: An Introduction to Non-Perturbative Methods in Quantum Field Theory. Cambridge University Press. doi:10.1017/CBO9781107705968.