Double-Line Counting and the Topological Expansion

Double-line notation resolves a Hermitian-matrix propagator into two oriented index strands. With the overall-$N$ single-trace normalization below, a connected vacuum ribbon contraction has color factor

$$N^{V-E+F}=N^{2-2h}.$$

Here $h$ is the genus after capping the closed index boundaries. This is an exact identity for each resolved Hermitian-matrix contraction. For $SU(N)$ adjoint fields, the exact covariance also contains a traceless subtraction; the normalization page distinguishes it from the leading matrix color flow. Genus organizes perturbative coefficients; it does not by itself construct a worldsheet theory or prove a string dual.

Required background. Large-N limits, normalizations, and orders of limits supplies the overall-$N$ action, propagator, vertex, and normalized-trace conventions.

Helpful background. Matrix eigenvalue saddles and loop equations supplies a non-diagrammatic planar solution with which the leading graph expansion can be compared.

From matrix indices to a ribbon surface

Take a zero-dimensional Hermitian model for definiteness,

$$Z_N = \int\mathrm dM\, \exp\!\left[ -N\operatorname{tr} \left( \frac12M^2+\sum_{k\ge3}\frac{g_k}{k}M^k \right) \right],$$

with all $g_k$ fixed as $N\to\infty$. Use the formal Gaussian expansion, or choose a confining potential and the real Hermitian integration cycle when interpreting $Z_N$ as an actual integral. The Gaussian contraction is

$$\left\langle M^i{}_j M^k{}_l \right\rangle_0 = \frac1N\delta^i{}_l\delta^k{}_j.$$

It therefore contributes one factor $N^{-1}$ and preserves two separately traceable strands. A single-trace interaction vertex contributes $N$, while summing an index around each closed strand face contributes $N$. A connected vacuum graph $\mathcal G$ has

$$\mathcal A_{\mathcal G} \propto N^V N^{-E}N^F = N^{V-E+F}.$$

Replace every vertex by an oriented disk and every propagator by a ribbon glued to two disk edges. This gives an orientable surface with boundary, which retracts onto the original graph and has Euler characteristic $V-E$. Cap each of its $F$ closed index boundaries with a disk. The capped surface has a cell decomposition with $V$ zero-cells, $E$ one-cells and $F$ two-cells. Hence

$$V-E+F = \chi(\Sigma_h) = 2-2h.$$

The genus $h$ is a property of this capped surface, not of how a crossing happens to look in a two-dimensional drawing. Apparent crossings can be projection effects; it is the cyclic ordering and index gluing that determine the surface. Trace the strands or compute $V-E+F$.

The formal Wick expansion and its genus organization are developed in Di Francesco, Ginsparg and Zinn-Justin 1994, § 1.2, printed preprint pp. 7–10, PDF.

Boundaries and normalized correlators

An insertion $\operatorname{tr}M^k$ creates a marked boundary but supplies no interaction factor $N$. Keep these specified boundaries open while capping the closed index faces. With $r$ unnormalized trace insertions, a connected genus-$h$ contribution therefore scales as

$$\left\langle \operatorname{tr}M^{k_1}\cdots \operatorname{tr}M^{k_r} \right\rangle_{\mathrm c,h} = O\!\left(N^{2-2h-r}\right).$$

For normalized invariants

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

the explicit $N^{-r}$ gives

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

At genus zero, a one-point function is $O(1)$, a connected two-point function is $O(N^{-2})$, and a connected three-point function is $O(N^{-4})$. The disconnected two-point product remains $O(1)$.

Fundamental matter produces another kind of boundary. In the fixed-$N_f$ ’t Hooft limit, a closed fundamental loop carries $N_f$ but removes an adjoint face, so it is suppressed by $N_f/N$. In the Veneziano limit, $N_f/N$ is held fixed and arbitrarily many such boundaries can contribute at the same leading order. One must declare which limit is being taken before using a boundary count.

First application: two quartic vacuum contractions

Label the quartic vertex’s cyclic slots $1,2,3,4$. Its three Wick pairings are $(12)(34)$, $(14)(23)$ and $(13)(24)$: the first two are planar and the last is crossed. Representatives of the two topological types both have

$$V=1, \qquad E=2.$$

Tracing the strands in the planar pairing gives $F=3$, hence

$$\chi=1-2+3=2, \qquad \mathcal A_{\mathrm{planar}}=O(N^2).$$

The crossed index pairing has $F=1$, hence

$$\chi=1-2+1=0, \qquad \mathcal A_{\mathrm{crossed}}=O(N^0).$$

The second graph is genus one and is suppressed by $N^{-2}$. This conclusion depends on the action normalization. If the overall $N$ is deleted while the written field is held fixed, propagators and vertices become $O(1)$ and the two powers are instead $N^3$ and $N$. Topology has not changed; the relation between topology and the $N$ power has.

Shared calculation. The large-N counting and topology map displays these face counts and the normalization-changing counterexample. The large-N scaling comparison compares matrix boundaries with vector and tensor counting.

What the genus expansion does and does not establish

Let $Z_{N,0}$ be the Gaussian integral with the same measure and integration cycle, setting every $g_k=0$. The interaction-dependent free energy has the formal expansion

$$\log\frac{Z_N}{Z_{N,0}} \sim \sum_{h=0}^{\infty} N^{2-2h}F_h(\{g_k\})$$

Here $F_h$ sums connected interaction ribbon graphs of genus $h$ in the declared perturbative expansion; the coupling-independent Gaussian normalization has been divided out. Several stronger conclusions require additional input.

• The coefficients $F_h$ may themselves be asymptotic series in the couplings.
• The genus series need not converge, and sectors of order $e^{-cN}$ are invisible at every fixed genus.
• A continuum worldsheet requires a limit in which arbitrarily refined graphs acquire controlled weights.
• A string interpretation requires more than Euler counting: one needs a worldsheet measure, observables, consistency conditions, and a target-space dictionary.
• In a gauge theory, confinement, flux-tube dynamics, and the spectrum are dynamical facts, not consequences of planarity alone.

Thus $N^{2-2h-b}$ is string-like bookkeeping. Calling it a string dual without the missing dynamical construction reverses implication.

Common pitfalls

Counting line crossings instead of faces. The $N$ power follows from closed index strands. Redraw the ribbon graph topologically or enumerate its index cycles.

Forgetting the normalization of inserted operators. A trace boundary changes $\chi$; the explicit $1/N$ in a normalized trace changes the power once more.

Assuming the genus series converges. A well-ordered graph-by-graph $1/N$ count does not control large genus or exponentially small saddles.

Exercises

1. A connected vacuum ribbon graph has $V=4$, $E=6$, and $F=4$. Determine its genus.

Solution

Its Euler characteristic is

$$\chi=V-E+F=4-6+4=2.$$

Therefore $2-2h=2$ and $h=0$. It is planar and scales as $N^2$.

2. Determine the leading scaling of a connected four-point function of normalized single traces.

Solution

Set $h=0$ and $r=4$ in $N^{2-2h-2r}$. The result is $N^{-6}$.

3. Why does a fixed-$N_f$ fundamental loop cost one power of $N$?

Solution

Replacing an adjoint face by a fundamental boundary removes one color-index sum, changing $N^2$ to $N$ at fixed topology. The flavor loop supplies $N_f$, so the relative factor is $N_f/N$, which is $O(N^{-1})$ when $N_f$ is fixed.

References

• Di Francesco, P., Ginsparg, P., and Zinn-Justin, J. (1995). “2D Gravity and Random Matrices.” Physics Reports 254, 1–133. doi:10.1016/0370-1573(94)00084-G. Open PDF: arXiv hep-th/9306153v2 (1994).

Further reading

• Mariño, M. (2015). Instantons and Large N: An Introduction to Non-Perturbative Methods in Quantum Field Theory. Cambridge University Press. doi:10.1017/CBO9781107705968.
• ’t Hooft, G. (1974). “A Planar Diagram Theory for Strong Interactions.” Nuclear Physics B 72, 461–473. doi:10.1016/0550-3213(74)90154-0.