From Genus Counting to a Holographic String Regime

Double-line notation turns adjoint-matrix Feynman diagrams into ribbon graphs. With the ’t Hooft-scaled couplings held fixed, the power of $N$ is the Euler characteristic of the associated surface: every handle costs $N^{-2}$. After the normalization of operator insertions is tracked, this has the topology dependence of a perturbative closed-string expansion with a putative loop parameter $g_s=f(\lambda)/N$. This is strong organizational evidence for strings, but it is not yet a construction of a worldsheet theory, a weak-curvature limit, or a nonperturbative string completion.

Required background. Large-N Factorization and Classical Bulk Scaling fixes the operator normalization and large-$N$ limit.

Helpful background. Planar Gauge Dynamics and String-Like Organization develops the QFT counting. Tensor Large N and Melonic Dominance provides a non-planar contrast.

Read the page in three passes: first derive the matrix Euler characteristic, then translate trace boundaries into the chapter’s two normalization conventions, and finally repeat the count for vector and tensor interactions. The last pass identifies exactly which hypothesis licenses genus language.

Double lines and the Euler characteristic

Let $\Phi(x)$ be an $N\times N$ Hermitian matrix-valued field, let $K$ be its kinetic kernel, and let $\operatorname{Tr}$ denote the color trace. For a zero-dimensional matrix model, simply omit the spacetime integral and the $x$ dependence. Work first with $U(N)$ and single-trace interactions,

$$S =N\int\mathrm d^dx\, \operatorname{Tr}\!\left[ \frac12\Phi K\Phi +\sum_{k\geq3}\frac{\lambda_k}{k}\Phi^k \right],$$

with every ’t Hooft-scaled coupling $\lambda_k$ fixed as $N\to\infty$. A propagator contributes $N^{-1}$ and carries two oppositely oriented color strands. Each interaction vertex supplied by the action contributes $N$, while each closed strand face supplies one freely summed color index and hence another factor $N$.

For a connected vacuum graph $\mathcal G$, let $V_{\mathrm{int}}$ count action vertices, $E$ propagators, and $F$ closed index faces. Its color factor is

$$\mathcal A_{\mathcal G} \propto N^{F-E+V_{\mathrm{int}}}.$$

Replacing every action vertex by a disk and every propagator by a ribbon produces an orientable closed surface. The graph is a cell decomposition of that surface, so

$$F-E+V_{\mathrm{int}} =\chi =2-2g.$$

The genus $g$ belongs to the thickened index graph, not to a particular two-dimensional drawing. A line crossing can disappear under a redraw, whereas the closed index cycles and Euler characteristic cannot. The graph-by-graph derivation is ’t Hooft 1974, § 3, eqs. (3.2)–(3.7), printed pp. 465–467, Open PDF; Aharony et al. 2000, § 1.2, eqs. (1.4)–(1.8) and the paragraph following eq. (1.8), Open PDF review the same organization in the gauge/string setting.

For $SU(N)$, the adjoint completeness relation contains a traceless-projector term proportional to $-1/N$ in addition to the leading oriented double-line contraction. The simple $U(N)$ graph count is therefore not literally term-by-term exact for every $SU(N)$ contraction, although the same orientable genus expansion governs the leading large-$N$ sector and the projector terms supply subleading corrections.

Trace boundaries and chapter normalization

Now insert $b$ unnormalized single traces. A trace insertion is a cyclic marked vertex with no accompanying factor $N$ from the action. Excising a small disk around each marked vertex opens one boundary, and the color factor remains

$$N^{F-E+V_{\mathrm{int}}},$$

where $V_{\mathrm{int}}$ still counts only action vertices. For an orientable surface of genus $g$ with $b$ boundaries,

$$F-E+V_{\mathrm{int}}=2-2g-b.$$

Thus, at fixed operator lengths and for separated renormalized insertions, each genus contribution obeys

$$\left\langle X_{k_1}\cdots X_{k_b} \right\rangle_{\!\mathrm c,g} =\mathcal O\!\left(N^{2-2g-b}\right), \qquad X_k=[\operatorname{Tr}(\Phi^k)]_{\mathrm R}.$$

When the leading coefficient at that genus is nonzero, the displayed power is the actual scaling. A symmetry or selection rule can instead make that coefficient vanish.

Each trace factor in a multi-trace insertion creates its own combinatorial boundary. In the closed-string comparison these boundaries are capped and become punctures. They are not evidence for physical open-string boundaries or D-branes, whose existence requires additional dynamical and dictionary data.

Use the notation fixed on the prerequisite page:

$$\tau_k=\frac{X_k}{N}, \qquad \mathcal O_k =Z_k^{-1/2}\bigl(X_k-\langle X_k\rangle\bigr),$$

where $Z_k$ has a finite nonzero large-$N$ limit and is chosen so that the leading planar connected two-point coefficient of $\mathcal O_k$ is one. When several operators mix, replace $Z_k^{-1/2}$ by a whitening matrix with a finite nonsingular large-$N$ limit; subleading $1/N$ dependence is allowed.

For $b\geq2$, centering does not change a connected cumulant, and the genus-resolved powers are

$$\left\langle \tau_1\cdots\tau_b \right\rangle_{\!\mathrm c,g} =\mathcal O\!\left(N^{2-2g-2b}\right), \qquad \left\langle \mathcal O_1\cdots\mathcal O_b \right\rangle_{\!\mathrm c,g} =\mathcal O\!\left(N^{2-2g-b}\right).$$

Consequently, the planar connected two-point function of $\mathcal O$ is order one by normalization, while its planar connected $b$-point function is $\mathcal O(N^{2-b})$; the displayed power is attained when the leading coefficient is nonzero. This is the unit-two-point normalization used to compare boundary correlators with canonically normalized bulk fields.

A planar graph and its one-handle partner

Keep the large-N counting and topology map open beside this calculation. Its Volume VII convention calls the expectation-normalized trace $\widehat{\mathcal O}$; that symbol is $\tau$ on this page. In the matrix panel, compare the closed index faces while following the frozen counts $V_{\mathrm{int}}=1$, $E=2$, and $F=3$ versus $F=1$.

At first order in a quartic single-trace interaction, the planar pairing has

$$V_{\mathrm{int}}=1, \qquad E=2, \qquad F=3.$$

Therefore

$$\chi_{\mathrm{planar}} =3-2+1=2, \qquad \mathcal A_{\mathrm{planar}}\sim N^2.$$

The crossed index pairing has the same vertex and propagator counts but only one closed face:

$$\chi_{\mathrm{crossed}} =1-2+1=0, \qquad \mathcal A_{\mathrm{crossed}}\sim N^0.$$

The thickened crossed contraction is a torus, so it is suppressed by $N^{-2}$ relative to the sphere contraction. More generally, writing $W=\log Z$, which generates connected vacuum diagrams,

$$W(N,\lambda) \sim \sum_{g=0}^{\infty} N^{2-2g}W_g(\lambda).$$

The Euclidean free energy is $-W$, up to the appropriate inverse-temperature or spacetime-volume normalization. The displayed series is generally formal. At fixed genus, the coupling expansion can be analytic when renormalons are absent, whereas in Yang–Mills theory and QCD renormalons can still make it divergent. Even when each fixed-genus function $W_g(\lambda)$ is analytic, at fixed $\lambda$ the sequence $W_g(\lambda)$ typically grows as $(2g)!$ as $g\to\infty$, so the full $1/N$ expansion has zero radius of convergence Mariño 2015, § 10.2, printed pp. 304–305, especially eqs. (10.2.8)–(10.2.9).

For $b$ canonically normalized external closed-string states, a connected genus-$g$ string amplitude carries

$$g_s^{\,2g+b-2}.$$

A nonzero genus-$g$ contribution to a unit-two-point trace correlator has the generic power $N^{2-2g-b}$, so the $N$ dependence agrees when

$$g_s=\frac{f(\lambda)}{N},$$

with $f(\lambda)$ independent of $N$ in the fixed-coupling limit Aharony et al. 2000, § 1.2, especially eqs. (1.5)–(1.8), Open PDF. This identifies a candidate string-loop parameter, not its complete coupling dependence.

The same counting neither identifies nor suppresses $\alpha'$ corrections. When a separate parameter dictionary is known, curvature in string units can depend on the ’t Hooft coupling rather than on $N$. In the $AdS_5\times S^5$ example,

$$\frac{L^4}{\alpha'^2}\propto\lambda, \qquad g_s\propto\frac{\lambda}{N},$$

up to convention-dependent constants Aharony et al. 2000, § 3.1, eqs. (3.9)–(3.10), Open PDF. Large $N$ suppresses string loops at fixed $\lambda$, but a weakly curved supergravity regime requires additional control.

Adversarial vector and tensor counts

The matrix conclusion depends on two oriented strands and single-trace vertex normalization. A controlled large-$N$ limit can survive when those ingredients are changed, while genus organization does not.

For a concrete vector contrast, take $N$ real components in a massive, regulated continuum model,

$$S_{\mathrm v} =\int\mathrm d^dx\left[ \frac12(\partial\phi^a)^2 +\frac12m^2\phi^a\phi^a +\frac{\lambda}{4N}(\phi^a\phi^a)^2 \right],$$

with $m>0$, and hold $\lambda$ fixed. Alternatively, read the example as a zero-dimensional model after dropping the derivative term. Only color-index factors are counted below; momentum integrals and symmetry factors do not change their powers of $N$. A quartic vertex contributes $N^{-1}$, a propagator carries one Kronecker delta, and each closed vector-index loop contributes $N$. A graph with $V$ quartic vertices and $L$ vector-index loops therefore scales as

$$\mathcal A_{\mathrm v}\sim N^{L-V}.$$

At one vertex, pairing the two fields within each bilinear produces a connected double-bubble graph with two independent index sums:

$$(V,L)=(1,2) \qquad\Longrightarrow\qquad \mathcal A_{\mathrm{double\ bubble}}\sim N.$$

Cross-pairing the two bilinears forces their indices to agree and leaves only one independent sum:

$$(V,L)=(1,1) \qquad\Longrightarrow\qquad \mathcal A_{\mathrm{cross\ paired}}\sim N^0.$$

The relative suppression is $N^{-1}$, not the matrix handle factor $N^{-2}$. A vector propagator has no second strand, so $L-V$ is not the Euler characteristic of a ribbon surface and neither contraction has a matrix genus.

A rank-three colored tetrahedral tensor model supplies a second contrast. With a vertex scaled as $N^{-3/2}$, a graph with $F_{\mathrm{col}}$ colored strand faces and $V$ vertices carries

$$\mathcal A_{\mathrm t} \sim N^{F_{\mathrm{col}}-\frac32V}.$$

An elementary two-vertex melon adds three colored faces,

$$\Delta V=2, \qquad \Delta F_{\mathrm{col}}=3, \qquad \Delta\!\left(F_{\mathrm{col}}-\frac32V\right)=0,$$

so inserting this melon preserves the $N$ exponent of its host graph. Beginning with a degree-zero leading graph, repeated melonic insertions remain leading. Colored tensor graphs can be organized by jackets and Gurau degree, but not by one ribbon-surface genus Bonzom et al. 2011, §§ II–III, eqs. (2.7)–(2.11) and “The dominant order: the world of melons,” Open PDF.

The strongest surviving statement is therefore conditional: orientable adjoint-matrix contractions with the declared single-trace scaling admit a genus expansion. Vector and tensor theories can possess equally systematic large-$N$ expansions whose dominant graphs and possible bulk interpretations are different.

Evidence ceiling, variants, and handoff

Genus counting does not supply a two-dimensional conformal field theory, a worldsheet measure on moduli space, modular invariance, a target spacetime, or a physical string spectrum. It also does not prove convergence of the genus series or determine sectors of order $e^{-cN}$ that are invisible at every fixed genus. Those ingredients require additional dynamics.

The matrix expansion itself assumes $N\to\infty$ with the ’t Hooft-scaled couplings, the number and lengths of trace insertions, and the relevant species data fixed. Multi-trace vertices need their own declared $N$ scaling. A dynamical fundamental loop is distinct from an external trace boundary: at fixed $N_f$, each such loop carries a relative factor $N_f/N$; in the Veneziano limit $N_f/N$ is fixed, so the suppression is compensated Makeenko 2010, § IX.C–E, eqs. (108), (115), and (119)–(120), printed pp. 13–14, Open PDF; the original limit is Veneziano 1976, pp. 519–524.

Orthogonal and symplectic gauge groups permit nonorientable surfaces and crosscap corrections rather than only the orientable series above Cicuta 1982, pp. 87–92. These are valid large-$N$ organizations, but they are different topological expansions.

The immediate next article, Central Charge, Newton Coupling, and the Planck Scale, asks how boundary stress-tensor normalization fixes a bulk graviton kinetic scale. Genus counting supplies a loop-order analogy; it does not by itself determine the Planck hierarchy. For the longer construction path, continue to String, Brane, and Top-Down Constructions and then Nonperturbative String- and M-Theory Definition Proposals.

Common pitfalls

Counting crossings in the drawing. A visual crossing is not a topological invariant. Trace the two color strands and count their closed faces.

Forgetting the insertion normalization. An unnormalized trace opens a boundary, while an explicit factor of $1/N$ in $\tau$ changes the power once more. State which convention is being used before comparing correlators.

Equating genus bookkeeping with a string dual. Euler-characteristic weights provide a candidate loop organization. A worldsheet theory, target-space dictionary, scale hierarchy, and nonperturbative completion remain separate requirements.

Exercises

Two genera and two normalizations

Assume the relevant genus coefficients are nonzero. For three unit-two-point single-trace fluctuations $\mathcal O_i$, determine the $N$ scaling of the planar connected contribution and its first one-handle correction. Repeat for the centered expectation-normalized traces

$$\widetilde\tau_i =\tau_i-\langle\tau_i\rangle =\frac{Z_i^{1/2}}{N}\mathcal O_i,$$

where every $Z_i$ has a finite nonzero large-$N$ limit.

Solution — two genera and two normalizations

For the $\mathcal O_i$ convention, set $b=3$ in $N^{2-2g-b}$. The planar term is $N^{-1}$ and the genus-one term is $N^{-3}$.

Each $\widetilde\tau_i$ supplies one additional factor $N^{-1}$ up to an $N$-independent coefficient. Three insertions therefore give $N^{-4}$ at genus zero and $N^{-6}$ at genus one.

Vector counting without genus

For the vector action in the text, count the power of $N$ for the one-vertex double-bubble and cross-paired contractions. Which numerical difference shows that the second contraction is not obtained by adding a matrix handle?

Solution — vector loops, not handles

Each graph has $V=1$. The double-bubble has two independent index loops, so

$$N^{L-V}=N^{2-1}=N.$$

The cross-paired contraction has one independent loop, so

$$N^{L-V}=N^{1-1}=N^0.$$

The relative factor is $N^{-1}$. A matrix handle would cost $N^{-2}$, and the vector graph has no two-strand face decomposition from which to define such a handle.

Fundamental loops in two large-N limits

Compare an adjoint sphere graph with a planar graph of the same handle genus containing one dynamical fundamental boundary. Give the relative power at fixed $N_f$ and in the Veneziano limit $N_f/N=\nu$ fixed.

Solution — fixed-flavor and Veneziano limits

The adjoint sphere is $\mathcal O(N^2)$. Replacing one adjoint face by a fundamental boundary gives $\mathcal O(NN_f)$, so the relative factor is

$$\frac{NN_f}{N^2}=\frac{N_f}{N}.$$

At fixed $N_f$, this is suppressed by one power of $N$. In the Veneziano limit it approaches the fixed coefficient $\nu$, so diagrams with quark boundaries can contribute at leading order and must be resummed within that different expansion.

The chapter overview contains the structure diagram and validity and failure diagram. They are embedded there once so that their shared chapter-level context is not repeated on every article.

For the chapter-wide comparison of assumptions, counterevidence, falsifiers, and claim ceilings, see the claim-domain table.

References

• Aharony, Ofer; Gubser, Steven S.; Maldacena, Juan; Ooguri, Hirosi; and Oz, Yaron. “Large N Field Theories, String Theory and Gravity.” Physics Reports 323, 183–386 (2000). doi:10.1016/S0370-1573(99)00083-6. Open PDF.
• Bonzom, Valentin; Gurau, Razvan; Riello, Aldo; and Rivasseau, Vincent. “Critical Behavior of Colored Tensor Models in the Large N Limit.” Nuclear Physics B 853, 174–195 (2011). doi:10.1016/j.nuclphysb.2011.07.022. Open PDF.
• Cicuta, Giovanni M. “Topological Expansion for SO(N) and Sp(2N) Gauge Theories.” Lettere al Nuovo Cimento 35, 87–92 (1982). doi:10.1007/BF02754653.
• Makeenko, Yuri. “A Brief Introduction to Wilson Loops and Large N.” Physics of Atomic Nuclei 73, 878–894 (2010). doi:10.1134/S106377881005011X. Open PDF.
• Mariño, Marcos. Instantons and Large N: An Introduction to Non-Perturbative Methods in Quantum Field Theory. Cambridge: Cambridge University Press, 2015. doi:10.1017/CBO9781107705968.
• ’t Hooft, Gerard. “A Planar Diagram Theory for Strong Interactions.” Nuclear Physics B 72, 461–473 (1974). doi:10.1016/0550-3213(74)90154-0. Open PDF.
• Veneziano, Gabriele. “Some Aspects of a Unified Approach to Gauge, Dual and Gribov Theories.” Nuclear Physics B 117, 519–545 (1976). doi:10.1016/0550-3213(76)90412-0.