Nonperturbative Exponential Effects and Finite-N Sectors

Agreement to every algebraic order in $1/N$ does not determine the exact finite-$N$ theory. Contributions proportional to $e^{-aN}$, $e^{-bN^2}$, or an entropy-scale exponential are smaller than every power of $1/N$ at fixed kinematics, yet can encode branes, competing gravitational saddles, topology change, finite-rank identities, or spectral discreteness. Their scaling and interpretation depend on the theory and observable; there is no universal nonperturbative completion supplied by genus counting alone.

Required background. From Genus Counting to a Holographic String Regime fixes the perturbative handle expansion, and Corrections, Nonuniform Limits, and Failure Modes explains how exponentially small terms become relevant in scaled domains.

Helpful background. Euclidean Tunneling Saddles and Boundary Conditions develops saddle and contour dependence, while EFT Truncation Errors and Breakdown Diagnostics distinguishes asymptotic error estimates from exact remainders.

Beyond the perturbative genus series

For a normalized observable with a closed-string organization, write

$$\mathcal A_{\mathrm{pert}}(N) \sim \sum_{g=0}^{\infty}\frac{a_g}{N^{2g}}.$$

The symbol $\sim$ denotes an asymptotic expansion, not necessarily a convergent sum. Even if every coefficient $a_g$ is known, the exact observable may have a transseries structure

$$\mathcal A(N) = \sum_{g\ge0}\frac{a_g}{N^{2g}} +\sigma_1e^{-A_1N} \sum_{k\ge0}\frac{b_k}{N^k} +\sigma_2e^{-A_2N^2} \sum_{k\ge0}\frac{c_k}{N^k} +\cdots .$$

The actions $A_i$, powers multiplying each exponential, and parameters $\sigma_i$ are dynamical data. The genus coefficients can display large-order growth that signals possible exponential scales, but large-order analysis alone does not choose an integration contour, boundary condition, or transseries parameter.

In familiar matrix-like holographic examples, $g_s\sim N^{-1}$. A D-brane tension is proportional to $1/g_s$, so a Euclidean brane saddle can contribute at order $e^{-A N}$. A purely gravitational saddle has action proportional to $1/G_N$, often scaling like $N^2$, and can therefore contribute at order $e^{-B N^2}$. These are scaling guides, not universal identities: fluxes, charges, couplings, zero modes, and the normalization of the observable alter the exponent and prefactor. Polchinski’s identification of D-branes as carriers of Ramond–Ramond charge provides the string-theory basis for the $1/g_s$ action scale Polchinski 1995.

Four distinct finite-N mechanisms

It is useful to separate mechanisms that are often grouped under “nonperturbative.”

Brane sectors. Extended objects have tensions nonanalytic in $g_s$. Boundary operators whose quantum numbers grow with $N$, such as determinant or subdeterminant operators, can couple to these sectors at leading order even though fixed-length single-trace perturbation theory barely sees them.

Additional spacetime saddles. A Euclidean path integral may admit saddles with different topology or action. Whether they contribute depends on the contour and boundary conditions. A subdominant saddle can be exponentially suppressed in one observable and essential in another.

Finite-rank identities. At finite $N$, traces of sufficiently high powers of an $N\times N$ matrix are related by the Cayley–Hamilton theorem. The large-$N$ Fock-space picture therefore overcounts states once operator length or charge scales with $N$. This is exact finite-$N$ structure, not a small loop correction.

Level discreteness. In finite volume, exact energy levels generate late-time plateaus, fluctuations, and recurrences that a smooth saddle expansion can miss. Their magnitude may be entropy-suppressed while their relevance grows at entropy-dependent times.

These mechanisms can have similar exponential sizes without being physically equivalent. An $e^{-N^2}$ correction to a partition function from a competing saddle and an $e^{-S}$ correction to a late-time correlator must not be identified solely from scaling.

A D-brane-sensitive boundary observable

Consider the standard top-down example of $\mathcal N=4$ $SU(N)$ super-Yang–Mills theory and type-IIB strings on $AdS_5\times S^5$. Fixed-length, normalized single-trace correlators organize into a genus expansion. By contrast, the half-BPS operator

$$\mathcal O_{\det}=\det Z =\frac{1}{N!} \epsilon_{i_1\cdots i_N}\epsilon^{j_1\cdots j_N} Z^{i_1}{}_{j_1}\cdots Z^{i_N}{}_{j_N}$$

contains $N$ fields and creates a state with charge of order $N$. Determinants and subdeterminants furnish boundary descriptions of giant-graviton D3-branes; the operator–brane identification and its finite-$N$ cutoff were developed by Balasubramanian et al. 2002, building on the giant-graviton construction of McGreevy, Susskind, and Toumbas 2000.

This supplies the first application. Compare two measurements:

1. a four-point function of fixed-dimension single-trace operators, computed through several genera; and
2. a normalized two-point function or transition amplitude involving $\mathcal O_{\det}$ or a subdeterminant with charge $J\sim N$.

The first tests the coefficients $a_g$ in a fixed-charge sector. The second changes the scaling of the external state and directly probes finite-rank and brane physics. Perfect agreement of the perturbative genus series in measurement 1 does not predict measurement 2, because the limit $N\to\infty$ with $J$ fixed differs from the limit with $J/N$ fixed. A brane-scale exponential can also be sought through observables in which the brane is virtual, but then it competes with other nonperturbative saddles and must be isolated by charges, selection rules, or independent semiclassical actions.

Adversarial completions with identical power series

The ambiguity can be demonstrated without assuming a divergent series. Define, for $N>1$,

$$F_{\pm}(N) = \frac{1}{1+N^{-2}} \mathbin{\pm}e^{-N}.$$

Both functions have exactly the same all-orders algebraic asymptotic expansion,

$$F_{\pm}(N) \sim 1-\frac1{N^2}+\frac1{N^4}-\frac1{N^6}+\cdots ,$$

because for every integer $K>0$,

$$\lim_{N\to\infty}N^K \bigl[F_+(N)-F_-(N)\bigr]=0.$$

Nevertheless, $F_+-F_-=2e^{-N}$. An observable rescaled by $e^N$, or evaluated in a sector whose leading answer is itself $e^{-N}$, distinguishes the two completions at order one. This is the required adversarial test: equality of every $1/N$ coefficient proves perturbative equivalence, not uniqueness of the exact answer.

In an actual path integral, not every arbitrary exponential addition is allowed. Unitarity, symmetries, factorization, analyticity, boundary conditions, charge quantization, and a specified microscopic definition constrain the sectors. The example establishes the logical gap that those additional inputs must close.

Orders of limits and observable dependence

Three scaled limits recur:

$$N\to\infty\ \text{with}\ J\ \text{fixed},\qquad N\to\infty\ \text{with}\ \frac{J}{N}\ \text{fixed},\qquad N\to\infty\ \text{with}\ t\,e^{-S}\ \text{fixed}.$$

The first supports ordinary few-particle genus counting. The second retains branes and finite-rank constraints. The third retains entropy-suppressed late-time structure. None can be substituted for another. Likewise, an exponential that is invisible in a vacuum correlator may be leading in a charged sector or in the difference between two nearly degenerate saddles.

The fixed-theory versus ensemble distinction is also decisive. In Jackiw–Teitelboim gravity, the perturbative topology expansion is completed by a double-scaled random matrix integral in the formulation of Saad, Shenker, and Stanford 2019. That is a concrete, highly instructive completion in a particular model. It does not establish that a generic higher-dimensional holographic CFT is an ensemble, nor that its nonperturbative sectors have the same interpretation.

Evidence ceiling and handoffs

Evidence cutoff: 25 July 2026. Top-down dualities, protected brane operators, matrix models, and semiclassical saddles provide explicit examples of exponential and finite-$N$ physics. They establish that perturbative genus data can be incomplete. They do not provide a universal list of sectors, determine a unique contour from the asymptotic series, or prove a common completion for all holographic theories. Detailed brane origins belong to the string-regime chapter; definition proposals, topology-changing saddles, and ensemble realizations belong to their dedicated later chapters.

Exercises

1. Show that $e^{-aN}=o(N^{-K})$ for every fixed $a>0$ and integer $K>0$.

   Solution

   Consider $N^Ke^{-aN}$. Its logarithm is $K\log N-aN$, which tends to $-\infty$. Hence the product tends to zero, proving that the exponential is smaller than every fixed inverse power.

2. Why can the Cayley–Hamilton relations matter at leading order for $J\sim N$ while being invisible for every fixed operator length?

   Solution

   For an $N\times N$ matrix, independent trace relations first constrain sufficiently high powers and products. Any operator of fixed length remains below that threshold once $N$ is large. If the length grows proportionally to $N$, the threshold is reached in the scaling limit, so finite-rank relations alter the leading state counting.

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

• Balasubramanian, Vijay; Berkooz, Micha; Naqvi, Asad; and Strassler, Matthew J. “Giant Gravitons in Conformal Field Theory.” Journal of High Energy Physics 2002, 034 (2002). doi:10.1088/1126-6708/2002/04/034.
• McGreevy, John; Susskind, Leonard; and Toumbas, Nicolaos. “Invasion of the Giant Gravitons from Anti-de Sitter Space.” Journal of High Energy Physics 2000, 008 (2000). doi:10.1088/1126-6708/2000/06/008.
• Polchinski, Joseph. “Dirichlet-Branes and Ramond–Ramond Charges.” Physical Review Letters 75, 4724–4727 (1995). doi:10.1103/PhysRevLett.75.4724.
• Saad, Phil; Shenker, Stephen H.; and Stanford, Douglas. “JT Gravity as a Matrix Integral.” arXiv:1903.11115 [hep-th] (2019). arXiv:1903.11115.