betaFeedback

WorksPhysics and Astronomy

Only the disk: topological obstructions for warped AdS6×S2_6\times S^2 in Type IIB

Abstract

Supersymmetric AdS6×S2_6\times S^2 solutions of Type IIB supergravity are warped products over a Riemann surface ΣΣ. They are determined by two holomorphic differentials on ΣΣ, with five-branes at poles on its boundary. We derive restrictions on the genus and the number of boundary components of ΣΣ. Without seven-branes or orientifolds, the differentials are taken to be single-valued on ΣΣ. Regularity implies that the number of interior zeros of one differential equals the total rotation of the phase of the ratio of the two differentials along the boundary. It also bounds this rotation on each boundary component by the number of poles on that component. Combined with the known count of zeros on the double of ΣΣ, these two statements force ΣΣ to be a disk. With mutually local seven-branes and orientifolds as interior punctures, the same conclusion follows under additional assumptions on the global structure of the differentials and on their behavior at the punctures. The disk then admits at most one orientifold puncture. On a disk without punctures, the bound applied between successive poles also recovers the known cyclic ordering of the five-brane charges along the boundary.

Introduction

Type IIB supergravity admits solutions with sixteen supercharges and SO(2,5)×SO(3)SO(2,5) \times SO(3) isometry. These provide holographic duals for five-dimensional superconformal field theories. A large class of these theories is realized by intersections of (p,q)(p,q) five-branes, with or without additional seven-branes [1, 2, 3]. The local supersymmetry conditions and the S2S^2 fibration over a two-dimensional base were derived using pure spinors by Apruzzi et al. [4] and recovered through a Killing spinor analysis by Kim, Kim and Suh [5]. We refer to the three papers in the series Warped AdS6×S2_6 \times S^2 in Type IIB supergravity as DGKU I, DGU II and DGU III. In DGKU I, D’Hoker, Gutperle, Karch and Uhlemann obtained the general local solution in terms of holomorphic data [6].1 D’Hoker, Gutperle and Uhlemann constructed global solutions for five-brane webs [9], with the full construction in DGU II [10], and incorporated mutually local seven-branes in DGU III [11]. In these solutions the metric is

ds2=f62dsAdS62+f22dsS22+4ρ2∣dw∣2,(1)ds^2 = f_6^2 ds_{\mathrm{AdS}_6}^2 + f_2^2 ds_{S^2}^2 + 4\rho^2 \lvert dw\rvert^2, \tag*{(1)}

where ww is a local coordinate on a Riemann surface Σ\Sigma with boundary, and f6f_6, f2f_2 and ρ\rho are real functions on Σ\Sigma. The solutions realize the exceptional superalgebra F(4)F(4) and are specified by two locally holomorphic functions A±(w)\mathcal{A}_{\pm}(w). Poles of their differentials dA±=∂wA± dwd\mathcal{A}_{\pm} = \partial_w\mathcal{A}_{\pm}\,dw on ∂Σ\partial\Sigma correspond to external five-branes, with charges determined by the residues. Interior punctures carry the monodromies of seven-branes.

In the global solutions constructed in these works, Σ\Sigma is a disk. These disk solutions have enabled computations of free energies and entanglement entropies [12], operator spectra [13, 14], Wilson loops [15] and the Euclidean on-shell action [16]. The holographic sphere free energies and stress-tensor coefficients CTC_T are found to agree at large charges with localization results for the TNT_N and +N,M+_{N,M} theories [17]. Disk solutions have also been used to give evidence for a five-dimensional FF-theorem [18] and to study long quivers [19] and their mass deformations [20, 21]. Further developments incorporate orientifold planes [22] and holographic defects [23, 24], while Penrose limits probe the geometry near null geodesics [25].

DGU II formulated the regularity conditions for arbitrary genus and any nonzero number of boundary components [10 §6]. Their numerical search on the unpunctured annulus suggested an obstruction [10 §5.7], leaving open which topologies admit regular solutions. The question also bears on the comparison with four-dimensional class S\mathcal{S} theories. Their holographic duals involve M5-branes wrapped on a Riemann surface whose genus and complex structure can be varied [26]. We return to this comparison in § 7.

For the known disk solutions, Kaidi and Uhlemann identified the Schottky double Σ^\hat{\Sigma}, obtained by gluing Σ\Sigma to its mirror along the boundary, with the M-theory curve of the corresponding five-brane junction [27]. They checked this correspondence for several explicit families. If the identification extends to other topologies, a handle or an additional boundary component of Σ\Sigma would give an M-theory curve of positive genus. Kaidi and Uhlemann associated positive-genus curves with five-brane webs whose internal faces have finite size. They interpreted these configurations as mass deformations and hence expected only the disk to occur at the conformal fixed point. A finite face can also arise at fixed masses from a Coulomb branch expectation value, which likewise introduces a scale [2].

Our main result establishes the restriction to the disk directly from Type IIB regularity. In the monodromy-free class of DGU II, let Σ\Sigma have genus gg, ν\nu boundary components and LL boundary five-brane poles, with no interior punctures and single-valued meromorphic differentials dA±d\mathcal{A}_{\pm}. We prove that physical regularity, in the sense of § 2.2, requires

2g+ν<2,that is(g,ν)=(0,1): Σ is a disk.(2)2g+\nu<2,\qquad\text{that is}\qquad(g,\nu)=(0,1):\ \Sigma\text{ is a disk}. \tag*{(2)}

In particular, the annulus is excluded for arbitrary charges and distributions of the poles between its two boundaries.

For surfaces with mutually local seven-brane punctures, each carrying a parabolic monodromy, Theorem 5.6 again gives (g,ν)=(0,1)(g,\nu)=(0,1). Mutual locality restricts the seven-brane charges to a common direction. Supersymmetric brane webs also admit mutually non-local seven-branes [3], which we do not consider in this paper. We assume that the DGU III construction holds globally, with no independent duality holonomy, that is, no duality transformation beyond those generated by the punctures, and that the monodromy-invariant combination of dA±d\mathcal{A}_{\pm} in (4) is nonzero at each puncture. The theorem excludes the annulus ansatz with localized D7-branes of Lozano, Macpherson and Montero [28 §3] for all choices of its parameters. On continuation around an orientifold puncture, the seven-brane transformation of the two differentials is accompanied by an overall minus sign. With this sign included in the global construction and the local asymptotics of § 5.3, Theorem 5.8 again gives a disk, with at most one orientifold puncture.

The proof in the monodromy-free case compares two statements about the ratio λ=dA+/dA−\lambda=d\mathcal{A}_{+}/d\mathcal{A}_{-}, which has unit modulus on ∂Σ\partial\Sigma. Let NN be the number of zeros of dA+d\mathcal{A}_{+} in the interior of Σ\Sigma, counted with multiplicity, and let 2πdi2\pi d_i be the increase of arg⁡λ\arg\lambda along the boundary component Γi\Gamma_i, which carries LiL_i of the poles. Interior regularity excludes zeros of dA−d\mathcal{A}_{-}, so N=∑idiN=\sum_i d_i by the argument principle. Positivity and superharmonicity of a real function G\mathcal{G} that enters the metric, together with the pole asymptotics, bound the winding on each component. The count of zeros and poles of dA+d\mathcal{A}_{+} on the Schottky double fixes NN [10 §6.4, (6.23)]. The two results are

di<Li,N=L+4g+2ν−4.(3)d_i < L_i,\qquad N=L+4g+2\nu-4. \tag*{(3)}

Summing the first over the components gives N<LN<L, and the second then gives (2).

Around a seven-brane puncture, monodromy can mix dA+d\mathcal{A}_{+} and dA−d\mathcal{A}_{-}, so their ratio λ\lambda need not return to its original value. For mutually local seven-branes, the common charge direction selects the differential

D dw=(α ∂wA+−αˉ ∂wˉA−) dw,(4)D\,dw=(\alpha\,\partial_w\mathcal{A}_{+}-\bar{\alpha}\,\partial_{\bar{w}}\mathcal{A}_{-})\,dw, \tag*{(4)}

where α\alpha is a constant of unit modulus fixed up to sign. The global assumptions stated above make D dwD\,dw single-valued. Interior regularity excludes zeros at regular points, and the assumption D≠0D\ne0 at the punctures is the condition that each puncture carries a seven-brane source. The zeros and poles of D dwD\,dw on the double can therefore be counted along the boundary, where the phase rotation of λ\lambda still obeys the bounds used in the monodromy-free proof. Comparing this count with the bounds again forces Σ\Sigma to be a disk. With orientifold punctures DD is defined only up to a sign, and the count is applied to its square.

Sections 2–4 give the regularity conditions, boundary estimates and proof in the monodromy-free case. § 5 treats seven-branes and orientifolds. In § 6 we also recover the charge ordering of Fluder and Uhlemann [18] from the boundary estimates. We conclude with a discussion in § 7. The counting steps of the three theorems and several of the local statements have been checked in Lean 4; the scope is described in the AI statement at the end of the paper.

The Type IIB AdS6 solutions

We collect the properties of the Type IIB solutions used in the proof [6, 10, 11]. The Riemann surface Σ\Sigma has genus gg, ν\nu boundary components and a total of LL five-brane poles on its boundary. In the monodromy-free case, NN denotes the number of zeros of dA+d\mathcal{A}_{+} in the interior of Σ\Sigma, counted with multiplicity.

Local solutions

The SO(2,5)×SO(3)SO(2,5) \times SO(3) isometry of (1) fixes the bosonic ansatz: the five-form vanishes, the complex three-form is F(3)=dC∧vol⁡S2F_{(3)} = dC \wedge\operatorname{vol}_{S^2}, and f6f_{6}, f2f_{2}, ρ\rho, CC and the axion-dilaton BB are functions on Σ\Sigma alone. The BPS equations were solved completely in DGKU I [6]: the general local solution is determined by two locally holomorphic functions A±\mathcal{A}_{\pm} on Σ\Sigma through the two real combinations

κ2=∣∂wA−∣2−∣∂wA+∣2,G=∣A+∣2−∣A−∣2+B+B‾,(5)\kappa^{2} = \lvert\partial_{w}\mathcal{A}_{-}\rvert^{2} - \lvert\partial_{w}\mathcal{A}_{+}\rvert^{2}, \qquad\mathcal{G} = \lvert\mathcal{A}_{+}\rvert^{2} - \lvert\mathcal{A}_{-}\rvert^{2} + B + \overline{B}, \tag*{(5)}

where the locally holomorphic B\mathcal{B} is defined up to an additive constant by

∂wB=A+∂wA−−A−∂wA+.(6)\partial_{w}\mathcal{B} = \mathcal{A}_{+}\partial_{w}\mathcal{A}_{-} - \mathcal{A}_{-}\partial_{w}\mathcal{A}_{+}. \tag*{(6)}

Under a holomorphic change of coordinate, κ2\kappa^{2} transforms so that i2κ2dw∧dw‾\frac{i}{2}\kappa^{2}dw \wedge d\overline{w} is invariant; G\mathcal{G} itself is a scalar. In Einstein frame the metric factors are

f62=c62κ2(1+R)ρ2(1−R),f22=c62κ2(1−R)9ρ2(1+R),ρ2=(R+R2)1/2(κ2)3/2∣∂wG∣(1−R)3/2,(7)f_{6}^{2} = \frac{c_{6}^{2}\kappa^{2}(1+R)}{\rho^{2}(1-R)}, \qquad f_{2}^{2} = \frac{c_{6}^{2}\kappa^{2}(1-R)}{9\rho^{2}(1+R)}, \qquad\rho^{2} = \frac{(R+R^{2})^{1/2}(\kappa^{2})^{3/2}}{\lvert\partial_{w}\mathcal{G}\rvert(1-R)^{3/2}}, \tag*{(7)}

with c6c_{6} a constant and RR determined by

R+1R=2+6κ2G∣∂wG∣2,(8)R + \frac{1}{R} = 2 + \frac{6\kappa^{2}\mathcal{G}}{\lvert\partial_{w}\mathcal{G}\rvert^{2}}, \tag*{(8)}

on the branch 0≤R<10 \le R < 1 that positivity of the metric factors selects [10 §2.3]. In the interior, R=0R=0 occurs exactly at the critical points of G\mathcal{G}, where the right-hand side of (8) diverges. The metric factors remain finite and positive there. With u:=(1−R)/(1+R)u := (1-R)/(1+R) and c62=1c_{6}^{2}=1, the normalization at a five-brane pole [10 (3.39)–(3.40)], they are

f62=6G u−1/2,f22=196G u3/2,ρ2=κ26Gu−1/2.(9)f_{6}^{2} = \sqrt{6\mathcal{G}}\,u^{-1/2}, \qquad f_{2}^{2} = \frac{1}{9}\sqrt{6\mathcal{G}}\,u^{3/2}, \qquad\rho^{2} = \frac{\kappa^{2}}{\sqrt{6\mathcal{G}}}u^{-1/2}. \tag*{(9)}

Two further properties will be used throughout: G\mathcal{G} is superharmonic, and the duality group acts linearly on the differentials. Differentiating (5) and noting that B+B‾B+\overline{B} is harmonic gives

∂w∂w‾G=−κ2,(10)\partial_{w}\partial_{\overline{w}}\mathcal{G} = -\kappa^{2}, \tag*{(10)}

which is [10 (2.14)], so G\mathcal{G} is superharmonic: its flat Laplacian Δ=∂x2+∂y2=4∂w∂w‾\Delta= \partial_{x}^{2}+\partial_{y}^{2}=4\partial_{w}\partial_{\overline{w}} is negative wherever κ2>0\kappa^{2}>0, with w=x+iyw=x+iy. A non-constant superharmonic function cannot attain its minimum in the interior. This is the minimum principle used below.

The classical Type IIB S-duality group acts on the differentials by SU(1,1)SU(1,1),

\begin{pmatrix} \partial_w\mathcal{A}_+ \\ \partial_w\mathcal{A}_- \end{pmatrix} \longmapsto M \begin{pmatrix} \partial_w\mathcal{A}_+ \\ \partial_w\mathcal{A}_- \end{pmatrix}, \qquad M=\begin{pmatrix} a & b \\ \bar{b} & \bar{a} \end{pmatrix}, \qquad|a|^2-|b|^2=1 \tag*{(11)}\end{pmatrix}

under which κ2\kappa^2 and G\mathcal{G} are invariant. A monodromy is such an MM attached to a closed circuit: continuing (∂wA+,∂wA−)(\partial_w\mathcal{A}_+,\partial_w\mathcal{A}_-) around a puncture brings it back multiplied by MM, and we write MjM_j for the monodromy of the puncture wjw_j. The duality group is also described as SL(2,R)SL(2,\mathbb{R}), acting on τ=χ+ie−2ϕ\tau=\chi+i e^{-2\phi}, with χ\chi the axion and ϕ\phi the dilaton, by τ↦(aτ+b)/(cτ+d)\tau\mapsto(a\tau+b)/(c\tau+d) with real a,b,c,da,b,c,d and ad−bc=1ad-bc=1. The Cayley transform B=(1+iτ)/(1−iτ)B=(1+i\tau)/(1-i\tau) maps the upper half-plane to the unit disk and turns this action into one of SU(1,1)SU(1,1) on BB. In the conventions of [6 §2.2, §5.1], the matrix MM in (11) is then given by

a=12[(a+d)+i(b−c)],b=12[(a−d)−i(b+c)].a=\frac{1}{2}\left[(a+d)+i(b-c)\right], \qquad b=\frac{1}{2}\left[(a-d)-i(b+c)\right].

Equivalently, the combinations ∂wA++∂wA−\partial_w\mathcal{A}_+ + \partial_w\mathcal{A}_- and i(∂wA+−∂wA−)i(\partial_w\mathcal{A}_+ - \partial_w\mathcal{A}_-) form a doublet on which the real matrix with entries a,b,c,da,b,c,d acts. We use SU(1,1)SU(1,1) for the holomorphic differentials and SL(2,R)SL(2,\mathbb{R}) for the brane charges.

Global regularity and five-brane charges

We require the metric factors to be real and positive in the interior of Σ\Sigma, so κ2>0\kappa^2>0 and G>0\mathcal{G}>0 there. We consider surfaces whose completion is compact after filling in the prescribed punctures and boundary poles. Such a surface must have a boundary: otherwise, integrating the Poisson equation (10) and applying Stokes’ theorem gives zero on one side and a strictly negative integral on the other side. The same argument applies in the presence of the prescribed seven-brane and orientifold punctures, since the flux of G\mathcal{G} through small circles around them vanishes.2 We therefore denote the boundary components by Γi\Gamma_i, i=1,…,νi=1,\ldots,\nu, with ν≥1\nu\geq1, using the notation of DGU II.

Along a regular boundary arc, the S2S^2 must shrink smoothly while the AdS6\mathrm{AdS}_6 radius remains finite. Regularity therefore requires the holomorphic data to satisfy

κ2=G=0on ∂Σ.(12)\kappa^2=\mathcal{G}=0\quad\text{on }\partial\Sigma. \tag*{(12)}

In a local coordinate w=x+iyw=x+iy that places the boundary on the real axis and Σ\Sigma in y>0y>0, the differentials of the DGU II class are taken to obey the reflection condition

∂wA−=−∂wA+‾on ∂Σ.(13)\partial_w\mathcal{A}_-=-\overline{\partial_w\mathcal{A}_+}\quad\text{on }\partial\Sigma. \tag*{(13)}

These conditions apply on the open boundary arcs. Throughout, physical regularity allows the prescribed localized brane singularities and requires the supergravity fields to be regular away from the sources.

The reflection condition (13) ensures only that κ2=0\kappa^2=0 and that G\mathcal{G} is constant on each regular boundary arc, as follows from the identities in § 3.1. Once these constants are required to agree across the five-brane poles, there is one constant on each Γi\Gamma_i, not yet set to zero. The common additive constant in G\mathcal{G} can be chosen to set its value on Γ1\Gamma_1 to zero. The requirement G=0\mathcal{G}=0 in (12) then imposes ν−1\nu-1 further conditions, one for each remaining boundary component. As shown in § 3.1, G\mathcal{G} also tends to zero at each five-brane pole from the interior. At the prescribed interior punctures G\mathcal{G} is bounded and extends across them as a superharmonic function. The minimum principle and (10) then give G>0\mathcal{G}>0 in the regular interior, so G\mathcal{G} attains its global minimum on every boundary component, as used in Theorem 3.2.

Away from the prescribed interior punctures, regularity excludes poles of ∂wA±\partial_w\mathcal{A}_{\pm} in the interior. At the candidate interior pole analyzed in DGU II, A−\mathcal{A}_{-} diverges logarithmically while A+\mathcal{A}_{+} remains finite, so G∼−∣A−∣2→−∞\mathcal{G}\sim-\lvert\mathcal{A}_{-}\rvert^2\to-\infty, in conflict with G>0\mathcal{G}>0 [10 App. A, (A.8)]. The same conclusion holds at any interior pole. Since κ2>0\kappa^2>0, the ratio λ=∂wA+/∂wA−\lambda=\partial_w\mathcal{A}_{+}/\partial_w\mathcal{A}_{-} is holomorphic with ∣λ∣<1\lvert\lambda\rvert<1 in a punctured neighborhood, so it extends to the pole, where ∣λ∣<1\lvert\lambda\rvert<1 by the maximum principle. Thus ∂wA−\partial_w\mathcal{A}_{-} is singular there and ∂wA+=λ∂wA−\partial_w\mathcal{A}_{+}=\lambda\partial_w\mathcal{A}_{-}. To leading order A+≃λA−\mathcal{A}_{+}\simeq\lambda\mathcal{A}_{-} with λ\lambda evaluated at the pole, while B\mathcal{B} is subleading, so G≃−(1−∣λ∣2)∣A−∣2→−∞\mathcal{G}\simeq-(1-\lvert\lambda\rvert^2)\lvert\mathcal{A}_{-}\rvert^2\to-\infty. The five-brane poles therefore lie on ∂Σ\partial\Sigma. We denote their total number by LL, their positions by pℓp_\ell, which we call the five-brane points, and their residues by Z±ℓZ^\ell_{\pm}, with ℓ=1,…,L\ell=1,\ldots,L; LiL_i counts the poles on Γi\Gamma_i, so L=∑iLiL=\sum_i L_i, as illustrated in Figure 1(a). The reflection condition (13) implies

A surface with two boundary components and local coordinates near a boundary arc

Figure 1. (a) A surface with two boundary components, five-brane poles (dots), and a seven-brane puncture (×). The dashed line is a branch cut. The holomorphic data jump across it, while κ2\kappa^2 and G\mathcal{G} remain invariant. The cut direction and its position relative to the poles enter the construction [11 §3.6, §4]; cuts are chosen to meet neither poles nor one another [11 §3.3]. Positive boundary orientation keeps Σ\Sigma on the left. (b) Local coordinates near a boundary arc, with Σ\Sigma in the upper half-plane and the inward normal pointing along +y+y. For disk topology, the upper half-plane also provides a global conformal representation.

Z−ℓ‾=−Z+ℓ,(14)\overline{Z^\ell_{-}}=-Z^\ell_{+}, \tag*{(14)}

regardless of the topology or the presence of punctures. Although derived in a real boundary coordinate, this relation holds in any holomorphic coordinate, since Z±ℓZ^\ell_{\pm} are the coordinate-invariant residues of the one-forms dA±\mathrm{d}\mathcal{A}_{\pm}. We abbreviate Zℓ=Z+ℓZ_\ell=Z^\ell_{+}. At each boundary pole the Type IIB supergravity fields have the near-horizon asymptotics of a (p,q)(p,q) five-brane source. The charge of the corresponding external five-brane stack is determined by the residue ZℓZ_\ell [10 §3.9–3.10]. A regular solution has at least one pole on each boundary component. In the monodromy-free case this follows from the boundary winding argument in § 4, and with interior punctures from Theorem 5.5 under the assumptions stated there.

In the monodromy-free class, the charges also satisfy

∑ℓ=1LZ±ℓ=0(single-valued differentials, no punctures),(15)\sum_{\ell=1}^{L} Z^\ell_{\pm}=0 \qquad\text{(single-valued differentials, no punctures),} \tag*{(15)}

by the residue theorem on the double. The Schottky double Σ^\widehat{\Sigma} is formed by gluing Σ\Sigma to its mirror, equipped with the conjugate complex structure, along their common boundary. Its genus is g^=2g+ν−1\widehat{g}=2g+\nu-1, where gg is the genus of Σ\Sigma. Schwarz reflection using (13) then extends the single-valued differentials meromorphically across this boundary, and their total residues on the resulting closed surface must vanish. Each boundary pole is a single point on the double and retains its residue Z±ℓZ_{\pm}^{\ell}, so it is counted only once. The sum includes poles on all boundary components and does not impose charge conservation separately on each one.3 With seven-brane punctures, the external five-brane charges need not sum to zero by themselves; charge conservation also involves the seven-brane sector, with charges compared in a common duality frame [11].

For an unpunctured disk the solution is explicit. In the upper half-plane representation, the holomorphic functions are exactly A±(w)=A±0+∑ℓZ±ℓln⁡(w−pℓ)\mathcal{A}_{\pm}(w)=\mathcal{A}_{\pm}^{0}+\sum_{\ell}Z_{\pm}^{\ell}\ln(w-p_{\ell}), where pℓ∈Rp_{\ell}\in\mathbb{R} and A±0\mathcal{A}_{\pm}^{0} are complex integration constants. The double is a sphere, so there are no additional globally holomorphic one-forms. The boundary values of G\mathcal{G} must match across the five-brane poles [10 §3.6]. Writing A0:=12(A+0−A‾−0)\mathcal{A}^{0}:=\frac{1}{2}(\mathcal{A}_{+}^{0}-\overline{\mathcal{A}}_{-}^{0}), this gives the conditions

A0Z−ℓ+A‾0Z+ℓ+∑m≠ℓZ[m,ℓ]ln⁡∣pm−pℓ∣=0,Z[m,ℓ]:=Z+mZ−ℓ−Z+ℓZ−m,(16)\mathcal{A}^{0}Z_{-}^{\ell}+\overline{\mathcal{A}}^{0}Z_{+}^{\ell}+\sum_{m\ne\ell}Z^{[m,\ell]}\ln|p_{m}-p_{\ell}|=0,\qquad Z^{[m,\ell]}:=Z_{+}^{m}Z_{-}^{\ell}-Z_{+}^{\ell}Z_{-}^{m}, \tag*{(16)}

for ℓ=1,…,L\ell=1,\ldots,L [10 (3.25)]. DGU II uses the gauge A−0=−A‾+0\mathcal{A}_{-}^{0}=-\overline{\mathcal{A}}_{+}^{0}, in which A0=A+0\mathcal{A}^{0}=\mathcal{A}_{+}^{0}; the other combination of integration constants changes only the gauge of the two-form potential.4

The expression above is written in terms of the poles and residues, and does not display the zeros of the differentials. The one-form dA+d\mathcal{A}_{+} has LL simple poles on the sphere and is regular at infinity by (15), so it has L−2L-2 zeros sns_n. DGU II take the poles and these zeros as the parameters [10 §3.1],

∂wA+=ω0λ0∏n=1L−2(w−sn)∏ℓ=1L(w−pℓ),∂wA−=ω0λ‾0∏n=1L−2(w−s‾n)∏ℓ=1L(w−pℓ),(17)\partial_w\mathcal{A}_{+}=\omega_{0}\lambda_{0}\frac{\prod_{n=1}^{L-2}(w-s_n)}{\prod_{\ell=1}^{L}(w-p_{\ell})},\qquad\partial_w\mathcal{A}_{-}=\omega_{0}\overline{\lambda}_{0}\frac{\prod_{n=1}^{L-2}(w-\overline{s}_n)}{\prod_{\ell=1}^{L}(w-p_{\ell})}, \tag*{(17)}

with ∣λ0∣=1|\lambda_{0}|=1 and ω‾0=−ω0\overline{\omega}_{0}=-\omega_{0}, which solves (13). The residues Z±ℓZ_{\pm}^{\ell} are then determined by the poles and zeros. Since ∣w−s‾n∣>∣w−sn∣|w-\overline{s}_n|>|w-s_n| when ww and sns_n both lie in the upper half-plane, placing all the zeros sns_n in the interior gives κ2>0\kappa^{2}>0 there [10 §3.2]. At least one zero is needed: for L=2L=2 the differentials (17) have equal modulus and κ2\kappa^{2} vanishes identically, so a solution without punctures has L≥3L\ge3 [10 §3.3]. Since G\mathcal{G} vanishes on the boundary arcs and tends to zero at the poles, G>0\mathcal{G}>0 then follows from the minimum principle. On the disk the conditions (16), imposed on the parameters of (17), are therefore sufficient as well as necessary for a regular solution.

This freedom is special to the disk. On a surface of any other topology the double has positive genus, and the number of zeros of dA+d\mathcal{A}_{+} on the double exceeds L−2L-2 by twice that genus. Positivity of κ2\kappa^{2} still requires all of them to lie in Σ\Sigma and none in its mirror. Theorems below show that this is incompatible with the boundary behavior of λ\lambda.

Charge conservation and the antisymmetry of Z[m,ℓ]Z^{[m,\ell]} imply that the sum of the LL equations (16) vanishes identically, so at most L−1L-1 of them are independent. The parametrization (17) has 3L−23L-2 real parameters, equivalent to the LL pole positions and the LL residues ZℓZ_\ell, which obey (15). The group PSL(2,R)PSL(2,\mathbb{R}) moves the poles and leaves the residues unchanged, so L−3L-3 cross-ratios of the poles remain. Together with the complex constant A0A^0 these are L−1L-1 real unknowns for the L−1L-1 equations. If the equations are independent, the solutions for given residues are isolated, and the 2L−22L-2 real parameters that remain are the independent charges of the conformal five-brane web [10 (3.28), §3.8]. Existence and uniqueness are discussed in § 6.

The topology argument uses the local boundary properties established in § 3; its extension to punctured surfaces requires the additional global assumptions introduced next.

Seven-branes and orientifolds

Following DGU III [11], we restrict the seven-brane punctures at w=wjw=w_j to be mutually local: their charges share a common [p,q][p,q] direction, and their monodromies commute. For coprime integers p,qp,q, a single seven-brane has monodromy M[p,q]=(1−pqp2−q21+pq)∈SL(2,Z)M_{[p,q]}=\begin{pmatrix}1-pq & p^2 \\ -q^2 & 1+pq\end{pmatrix}\in SL(2,\mathbb{Z}). A puncture wjw_j contains mjm_j branes of this same charge, where mjm_j is a positive integer, and has monodromy Mj=M[p,q]mjM_j=M_{[p,q]}^{m_j}. Since (M[p,q]−1)2=0(M_{[p,q]}-1)^2=0, this gives Mj=1+mj(M[p,q]−1)M_j=1+m_j(M_{[p,q]}-1).

We use the notation mj=nj2m_j=n_j^2 of DGU III, with njn_j real, and call nj2n_j^2 the strength of the puncture. In the matrix parametrization of DGU III, M[p,q]mj=M[njp,njq]M_{[p,q]}^{m_j}=M_{[n_jp,n_jq]} [11 (3.1),(3.15)]. The pair [njp,njq][n_jp,n_jq] parametrizes the monodromy matrix of the stack, not the charge of an individual brane. Each constituent has charge [p,q][p,q], and nj2n_j^2 can be any positive integer.5 A single duality transformation sends the common charge direction to [1,0][1,0]. In this D7-brane frame, all punctures have monodromy Mj=Tnj2M_j=T^{n_j^2}, where T=(1101)T=\begin{pmatrix}1&1\\0&1\end{pmatrix}. The transformation leaves κ2\kappa^2, G\mathcal{G} and the boundary regularity conditions unchanged. Each puncture thus carries a parabolic monodromy, ∣tr⁡Mj∣=2|\operatorname{tr}M_j|=2, which preserves a charge direction. Composite seven-branes with elliptic monodromy, ∣tr⁡Mj∣<2|\operatorname{tr}M_j|<2, such as those of type E6,7,8E_{6,7,8} and H0,1,2H_{0,1,2} [29], preserve none and are not included, even when only one puncture is present.

DGU III constructs disk solutions in the D7-brane frame by adding monodromy to a pair of seed functions. The probe limit, brane-web interpretation and sphere partition functions of these solutions were studied in [30], and further solutions were obtained in [31]. In the SU(1,1)SU(1,1) basis of (11), the D7-brane monodromy Tnj2T^{n_j^2} acts on the pair (dA+,dA−)(d\mathcal{A}_+,d\mathcal{A}_-) by the matrix

1+inj22(1−11−1)1+\frac{in_j^2}{2}\begin{pmatrix}1&-1\\1&-1\end{pmatrix}

Thus the monodromy shifts both differentials by inj22(dA+−dA−)\frac{in_j^2}{2}(d\mathcal{A}_+-d\mathcal{A}_-), leaving their difference invariant. A function ff that shifts by inj2/2in_j^2/2 on a counterclockwise circuit of wjw_j implements this monodromy. We write the disk construction of DGU III [11 (3.8), (3.11)] in terms of the Dirichlet Green’s function as

∂wA±=∂wA±seed+f(∂wA+seed−∂wA−seed),Re⁡f=−14π∑jnj2g(w,wj).(18)\partial_w\mathcal{A}_{\pm}=\partial_w\mathcal{A}_{\pm}^{\mathrm{seed}}+f\left(\partial_w\mathcal{A}_{+}^{\mathrm{seed}}-\partial_w\mathcal{A}_{-}^{\mathrm{seed}}\right),\qquad\operatorname{Re}f=-\frac{1}{4\pi}\sum_j n_j^2 g(w,w_j). \tag*{(18)}

Here gg is the Dirichlet Green’s function on Σ\Sigma, normalized by Δwg(w,wj)=−2πδ(2)(w−wj)\Delta_w g(w,w_j)=-2\pi\delta^{(2)}(w-w_j). It vanishes on ∂Σ\partial\Sigma and behaves as ln⁡(1/∣w−wj∣)\ln(1/\lvert w-w_j\rvert) near wjw_j. The functions A±seed(w)\mathcal{A}_{\pm}^{\mathrm{seed}}(w) are the monodromy-free seed functions of DGU III, written there as A±(0)(w)\mathcal{A}_{\pm}^{(0)}(w). On the disk they have the same constant-plus-logarithm structure as in § 2.2. The term containing ff introduces the seven-brane monodromies and also changes the boundary residues. Writing Z±,seedℓZ_{\pm,\mathrm{seed}}^\ell for the seed residues, the full residues Z±ℓZ_\pm^\ell obey, in a branch where ff is regular near pℓp_\ell,

Z±ℓ=Z±,seedℓ+f(pℓ)(Z+,seedℓ−Z−,seedℓ).Z_\pm^\ell=Z_{\pm,\mathrm{seed}}^\ell+f(p_\ell)\left(Z_{+,\mathrm{seed}}^\ell-Z_{-,\mathrm{seed}}^\ell\right).

Thus even on the disk, prescribed external five-brane charges constrain this combination of seed and puncture data, rather than the seed residues alone.

Lozano, Macpherson and Montero adapted the disk construction of DGU III to the annulus. For a given annulus modulus and puncture data, they expressed the monodromy function ff in terms of theta functions [28 (3.1)]. They left open whether the parameters of this ansatz could be chosen to satisfy the regularity conditions on both boundary components simultaneously [28 §§3, 9]. In their ansatz the seed differentials are single-valued, and the shift f(w+1)−f(w)f(w+1)-f(w) accounts for the full monodromy of dA±d\mathcal{A}_{\pm} around the annulus [28 (A.10), (A.13)–(A.14)].

On a surface of arbitrary topology we assume the same global construction (18), with one pair of single-valued meromorphic seed differentials satisfying (13) and having poles only at the boundary five-brane points. The Dirichlet Green’s function fixes Re⁡f\operatorname{Re}f, so the Cauchy–Riemann equations determine the change in Im⁡f\operatorname{Im}f along any closed curve in the punctured interior. We assume that the resulting period accounts for the full duality transformation of dA±d\mathcal{A}_{\pm} along the curve. This is the assumption of no independent duality holonomy used in Theorem 5.3.

In the D7-brane frame, the seven-brane monodromies leave D dw=dA+−dA−D\,dw=d\mathcal{A}_{+}-d\mathcal{A}_{-} invariant, corresponding to α=1\alpha=1 in (4). By (18) the terms containing ff cancel in this difference, so D=∂wA+seed−∂wA−seedD=\partial_w\mathcal{A}_{+}^{\mathrm{seed}}-\partial_w\mathcal{A}_{-}^{\mathrm{seed}} is fixed by the seed alone and is finite at the punctures wjw_j. At a five-brane pole pℓp_\ell its residue is Z+ℓ−Z−ℓ=2Re⁡Z+ℓZ_+^\ell-Z_-^\ell=2\operatorname{Re}Z_+^\ell by (14), and is proportional to the NS5 charge [11 (5.1)]. With ξ=w−wj\xi=w-w_j, the behavior of κ2\kappa^2 near a seven-brane puncture is

κ2=−nj24π∣D(wj)∣2ln⁡∣ξ∣2+O(1),(19)\kappa^2=-\frac{n_j^2}{4\pi}\lvert D(w_j)\rvert^2\ln\lvert\xi\rvert^2+\mathcal{O}(1), \tag*{(19)}

The coefficient cc of DGU III [11 (3.64)] equals −D(wj)-D(w_j) in our convention. Define the seed quantity κ(0)2:=∣∂wA−seed∣2−∣∂wA+seed∣2\kappa_{(0)}^2:=\lvert\partial_w\mathcal{A}_{-}^{\mathrm{seed}}\rvert^2-\lvert\partial_w\mathcal{A}_{+}^{\mathrm{seed}}\rvert^2. If κ(0)2>0\kappa_{(0)}^2>0 at wjw_j, as in the disk construction of DGU III [11 (3.6)], the seed differentials have unequal magnitudes there, so their difference D(wj)D(w_j) is nonzero. On a general surface we impose D(wj)≠0D(w_j)\ne0 as a local source condition. This excludes little. By (18),

κ2=κ(0)2−2Re⁡f∣D∣2,κ(0)2=−2Re⁡(∂wA+seed‾D)+∣D∣2.(20)\kappa^2=\kappa_{(0)}^2-2\operatorname{Re}f\lvert D\rvert^2,\qquad\kappa_{(0)}^2=-2\operatorname{Re}\left(\overline{\partial_w\mathcal{A}_{+}^{\mathrm{seed}}}D\right)+\lvert D\rvert^2. \tag*{(20)}

If DD vanishes at wjw_j while ∂wA+seed\partial_w A_+^{\mathrm{seed}} does not, the first term of κ(0)2\kappa^2_{(0)} dominates near wjw_j and is negative in some directions, so regularity already fails. The condition therefore excludes a possibly regular solution only if both seed differentials vanish at the puncture. In that case κ2\kappa^2 has no logarithmic divergence, the expansion of RR in [11 (3.68)] does not apply, and the local geometry is not that of the seven-brane of DGU III. We do not consider this case. For D(wj)≠0D(w_j) \ne0 the local analysis gives G=Gj+O(∣ξ∣)\mathcal{G} = G_j + O(|\xi|) with a finite constant Gj>0G_j > 0.

The identity (20) also shows that the bound L≥3L \ge3 of § 2.2 is special to solutions without punctures. For a seed with L=2L = 2, κ(0)2\kappa^2_{(0)} vanishes identically and DD has no zeros unless it vanishes identically. Then κ2=−2Re⁡f∣D∣2\kappa^2 = -2 \operatorname{Re} f |D|^2 is positive in the interior, because Re⁡f<0\operatorname{Re} f < 0 there. These are the two-pole solutions of Chaney and Uhlemann [31 §3.1].

An orientifold puncture carries monodromy −Tk-T^k, with k=nj2k = n_j^2 [22]. The extra minus sign reverses both seed differentials on continuation around the puncture, so D dwD\,\mathrm{d}w also changes sign. This holds even for k=0k = 0, when the monodromy is the central element −1-1. We give the local expression in § 5.3.

Boundary regularity and phase rotation

In this section we bound the rotation of the phase of the ratio λ=∂wA+/∂wA−\lambda= \partial_w A_+/\partial_w A_- of DGU II [10 (2.17)] along a boundary component Γi\Gamma_i. Positivity of κ2\kappa^2 fixes the direction in which this phase rotates, while positivity and superharmonicity of G\mathcal{G} fix the sign of its inward normal derivative. Combining these properties with the pole asymptotics gives the bound di<Lid_i < L_i of (3).

Boundary phase and pole asymptotics

We first establish monotonicity of the boundary phase, continuity of G\mathcal{G} at the five-brane poles, and an identity for its inward normal derivative.

On every component Γi\Gamma_i of ∂Σ\partial\Sigma, the reflection condition (13) gives

∣λ∣=1,arg⁡λ=π+2arg⁡∂wA+,(21)|\lambda| = 1,\qquad\arg\lambda= \pi+ 2\arg\partial_w A_+, \tag*{(21)}

where phases are chosen continuously along each arc between consecutive poles. We show below that ∂wA+\partial_w A_+ does not vanish on these arcs. We traverse the boundary positively, with Σ\Sigma on the left; in a local upper-half-plane coordinate this means increasing xx, with ∂y\partial_y pointing inward. Figure 1 fixes these conventions.

Since κ2>0\kappa^2 > 0 in the interior, ∂wA−\partial_w A_- has no zeros there and each local branch of λ\lambda is holomorphic with ∣λ∣<1|\lambda| < 1. If either ∂wA+\partial_w A_+ or ∂wA−\partial_w A_- vanished at a point on an open boundary arc, (13) would give ∂wA+=∂wA−=0\partial_w A_+ = \partial_w A_- = 0 there. Differentiating (5) would then give ∂wG=0\partial_w \mathcal{G} = 0 at that local upper-half-plane coordinate, a small disk tangent to the real axis from above at the boundary point provides the interior ball required by Hopf’s lemma. Writing log⁡λ=log⁡∣λ∣+iψ\log\lambda= \log|\lambda| + i\psi, the Cauchy–Riemann equations give ∂xψ=−∂ylog⁡∣λ∣\partial_x\psi= -\partial_y\log|\lambda|. Take ss to be positively oriented boundary arclength in any smooth conformal metric on Σ\Sigma, and ∂n\partial_n the inward unit normal in the same metric. The two coordinate derivatives acquire the same positive factor, giving

∂sψ=−∂nlog⁡∣λ∣>0.(22)\partial_s\psi= -\partial_n \log|\lambda| > 0. \tag*{(22)}

Since log⁡∣λ∣\log|\lambda| is constant along the boundary, ∂slog⁡λ=i∂sψ\partial_s\log\lambda= i\partial_s\psi there, that is

∣∂sλ∣=∂sψ>0,(23)|\partial_s\lambda| = \partial_s\psi> 0, \tag*{(23)}

so λ′:=∂wλ\lambda' := \partial_w\lambda has no zero on a regular boundary arc.6 In § 5.2 we use λ′≠0\lambda' \ne0 to show that the boundary zeros of the monodromy-invariant one-form D dwD\,dw in (4) are simple.

We record the increase of arg⁡λ\arg\lambda along Γi\Gamma_i, traversed positively and measured in units of 2π2\pi, as

di:=12πΔΓiarg⁡λ>0.(24)d_i := \frac{1}{2\pi}\Delta_{\Gamma_i}\arg\lambda> 0. \tag*{(24)}

In the monodromy-free case, λ\lambda is single-valued. Since κ2>0\kappa^2 > 0, its denominator ∂wA−\partial_w\mathcal{A}_- never vanishes in the interior, so λ\lambda has no interior poles and its zeros are precisely those of ∂wA+\partial_w\mathcal{A}_+, with the same multiplicities. At each boundary pole the simple poles in the ratio cancel, leaving a finite nonzero value. The argument principle applies on Σ\Sigma of any genus. Indeed, dλ/λd\lambda/\lambda is a globally defined one-form, closed away from the zeros of λ\lambda. Removing small disks around these zeros and applying Stokes’ theorem gives

∑idi=12πi∮∂Σdλλ=∑λ(ζ)=0ord⁡ζ(λ),\sum_i d_i=\frac{1}{2\pi i}\oint_{\partial\Sigma}\frac{d\lambda}{\lambda}=\sum_{\lambda(\zeta)=0}\operatorname{ord}_{\zeta}(\lambda),

since the integral around each small positively oriented circle is 2πi2\pi i times the order of the zero. Each did_i is the winding of λ\lambda along Γi\Gamma_i with its induced boundary orientation, and the sum over all components gives the total number NN of interior zeros of dA+d\mathcal{A}_+, counted with multiplicity. We summarize these statements.

Lemma 3.1 (Boundary winding). di>0d_i>0 on every positively traversed boundary component. If λ\lambda returns to itself along Γi\Gamma_i, then did_i is a positive integer. In the monodromy-free class, ∑idi\sum_i d_i equals the number of interior zeros of ∂wA+\partial_w\mathcal{A}_+, counted with multiplicity.

With punctures, continuation around Γi\Gamma_i multiplies the differentials by a monodromy matrix MΓiM_{\Gamma_i}. The continuation takes the initial value λin\lambda_{\mathrm{in}} to (aλin+b)/(bˉλin+aˉ)(a\lambda_{\mathrm{in}}+b)/(\bar{b}\lambda_{\mathrm{in}}+\bar{a}), with aa, bb the entries of MΓiM_{\Gamma_i} as in (11). In a fixed duality frame the same transformation acts at each starting point, but its angular displacement on the unit circle can vary with λin\lambda_{\mathrm{in}}. Thus did_i can depend on the starting point and need not be an integer. On a component containing poles we start at any five-brane pole, so that the circuit consists of complete arcs from one pole to the next. The global count of § 5.2 is independent of this choice. We give the dependence of did_i on a general base point in Theorem 5.4.

To determine the limit of GG at a five-brane pole pℓp_\ell, we use the local expansion of DGU II [10 §3.6]. With the residue relation (14), the holomorphic data are

A+=Zℓlog⁡(w−pℓ)+a++O(w−pℓ),A−=−Z‾ℓlog⁡(w−pℓ)+a−+O(w−pℓ).(25)\mathcal{A}_+=\mathcal{Z}_\ell\log(w-p_\ell)+a_++\mathcal{O}(w-p_\ell),\qquad\mathcal{A}_-=-\overline{\mathcal{Z}}_\ell\log(w-p_\ell)+a_-+\mathcal{O}(w-p_\ell). \tag*{(25)}

Here a±a_\pm are the finite parts at the chosen pole. The logarithm-squared terms cancel in ∣A+∣2−∣A−∣2|\mathcal{A}_+|^2-|\mathcal{A}_-|^2. In ∂wB\partial_w\mathcal{B} of (6), the terms proportional to log⁡(w−pℓ)/(w−pℓ)\log(w-p_\ell)/(w-p_\ell) also cancel, leaving a simple pole of residue −(a+Zℓ+a−Z‾ℓ)-(a_+\mathcal{Z}_\ell+a_-\overline{\mathcal{Z}}_\ell). Let GℓRG_\ell^{\mathrm{R}} denote the limit of GG along the boundary arc to the right of pℓp_\ell. Substitution in (5) gives

G(w)=GℓR−4Cℓarg⁡(w−pℓ)+O(∣w−pℓ∣log⁡∣w−pℓ∣),Cℓ=Im⁡(a‾+Zℓ)+Im⁡(a‾−Z‾ℓ).(26)G(w)=G_\ell^{\mathrm{R}}-4C_\ell\arg(w-p_\ell)+\mathcal{O}(|w-p_\ell|\log|w-p_\ell|), \qquad C_\ell=\operatorname{Im}(\overline{a}_+\mathcal{Z}_\ell)+\operatorname{Im}(\overline{a}_-\overline{\mathcal{Z}}_\ell). \tag*{(26)}

On the boundary arcs to the right and left of pℓp_\ell, the argument is 00 and π\pi, respectively. Since G=0G=0 on both arcs, the right limit gives GℓR=0G_\ell^{\mathrm{R}}=0 and the left limit then gives Cℓ=0C_\ell=0. Thus only the O(rlog⁡r)\mathcal{O}(r\log r) term remains, and G→0G\to0 also as ww approaches pℓp_\ell from the interior, with r=∣w−pℓ∣→0r=|w-p_\ell|\to0.

To express the inward normal derivative of GG in terms of the holomorphic data, define on each boundary component

A^:=12(A+−A‾−).(27)\widehat{\mathcal{A}}:=\frac{1}{2}\left(\mathcal{A}_+-\overline{\mathcal{A}}_-\right). \tag*{(27)}

It is finite on each open arc between successive poles and diverges logarithmically at the poles. Differentiating (5) and using (13) to eliminate ∂wA−\partial_w\mathcal{A}_- gives, on a regular boundary arc,

∂wG=(A‾+−A−)∂wA+−(A+−A‾−)∂wA‾+=2A^‾ ∂wA+−2A^ ∂wA‾+,(28)\partial_wG=\left(\overline{\mathcal{A}}_+-\mathcal{A}_-\right)\partial_w\mathcal{A}_+-\left(\mathcal{A}_+-\overline{\mathcal{A}}_-\right)\overline{\partial_w\mathcal{A}}_+=2\overline{\widehat{\mathcal{A}}}\,\partial_w\mathcal{A}_+-2\widehat{\mathcal{A}}\,\overline{\partial_w\mathcal{A}}_+, \tag*{(28)}

that is

∂yG=−8Im⁡(A^‾ ∂wA+),(29)\partial_y \mathcal{G}=-8\operatorname{Im}\left(\overline{\widehat{\mathcal{A}}}\,\partial_w\mathcal{A}_+\right), \tag*{(29)}

with ∂y\partial_y along the inward normal. On each regular boundary arc, (13) also gives ∂xA^=∂wA+\partial_x\widehat{\mathcal{A}}=\partial_w\mathcal{A}_+.

The rotation bound

As above, let ss denote positively oriented boundary arclength. Write A^˙=dA^/ds\dot{\widehat{\mathcal{A}}}=d\widehat{\mathcal{A}}/ds and set P(s)=A^‾A^˙P(s)=\overline{\widehat{\mathcal{A}}}\dot{\widehat{\mathcal{A}}} at regular boundary points. In a local upper-half-plane coordinate w=x+iyw=x+iy, A^˙=(dx/ds) ∂wA+\dot{\widehat{\mathcal{A}}}=(dx/ds)\,\partial_w\mathcal{A}_+ on the boundary. Since dx/ds>0dx/ds>0, these two quantities have the same phase. On each open arc II between successive poles, let δI=ΔIarg⁡A^˙\delta_I=\Delta_I\arg\dot{\widehat{\mathcal{A}}}. By (21), the phase of λ\lambda increases by 2δI2\delta_I on II. At a pole the common simple poles of ∂wA±\partial_w\mathcal{A}_\pm cancel in their ratio, so λ\lambda extends through it with nonzero value. Thus ∑IδI=πdi\sum_I\delta_I=\pi d_i for a circuit starting at a pole.

Lemma 3.2 (Rotation bound). Suppose the value of G\mathcal{G} along Γi\Gamma_i is its global minimum on Σ‾\overline{\Sigma}, and let Li≥1L_i\geq1. In the absence of monodromy the base point for did_i is arbitrary. With monodromy, start at any pole so that the circuit consists of complete arcs between successive poles. Then δI<π\delta_I<\pi on each arc II, and

∑I⊂ΓiΔIarg⁡A^=π(di−Li)<0,sodi<Li.(30)\sum_{I\subset\Gamma_i}\Delta_I\arg\widehat{\mathcal{A}}=\pi(d_i-L_i)<0,\qquad\text{so}\qquad d_i<L_i. \tag*{(30)}

Proof. By (10), G\mathcal{G} is strictly superharmonic in the regular interior. Since G\mathcal{G} takes its minimum value along Γi\Gamma_i, at each regular point of this boundary component Hopf’s lemma gives ∂yG>0\partial_y\mathcal{G}>0 in the local upper-half-plane coordinate. By (29),

Im⁡P=−18dxds ∂yG<0.(31)\operatorname{Im}P=-\frac{1}{8}\frac{dx}{ds}\,\partial_y\mathcal{G}<0. \tag*{(31)}

In particular A^\widehat{\mathcal{A}} is nonzero on each open arc, and d(arg⁡A^)/ds=Im⁡P/∣A^∣2<0d(\arg\widehat{\mathcal{A}})/ds=\operatorname{Im}P/|\widehat{\mathcal{A}}|^2<0 there.

Consider one arc II traversed from pole to pole. Near either endpoint pp of residue ZZ, write xpx_p for its position in the local boundary coordinate xx. By (25) and (27), the two logarithms in A^\widehat{\mathcal{A}} combine to give A^=Zlog⁡∣x−xp∣+O(1)\widehat{\mathcal{A}}=Z\log|x-x_p|+O(1), while ∂xA^=Z/(x−xp)+O(1)\partial_x\widehat{\mathcal{A}}=Z/(x-x_p)+O(1). Hence

P∼dxds∣Z∣2log⁡∣x−xp∣x−xp.P\sim\frac{dx}{ds}|Z|^2\frac{\log|x-x_p|}{x-x_p}.

The logarithm is negative near either pole. The denominator is positive at the start of II and negative at the end. Since PP stays in the open lower half-plane, its argument in (−π,0)(-\pi,0) therefore tends to −π-\pi at the start and to 00 at the end. Thus ΔIarg⁡P=π\Delta_I\arg P=\pi. Since conjugation reverses the phase, P=A^‾A^˙P=\overline{\widehat{\mathcal{A}}}\dot{\widehat{\mathcal{A}}} gives, for phases followed continuously along II,

π=ΔIarg⁡P=ΔIarg⁡A^˙−ΔIarg⁡A^=δI−ΔIarg⁡A^.\pi=\Delta_I\arg P=\Delta_I\arg\dot{\widehat{\mathcal{A}}}-\Delta_I\arg\widehat{\mathcal{A}}=\delta_I-\Delta_I\arg\widehat{\mathcal{A}}.

Thus ΔIarg⁡A^=δI−π\Delta_I\arg\widehat{\mathcal{A}}=\delta_I-\pi, which is negative because d(arg⁡A^)/ds<0d(\arg\widehat{\mathcal{A}})/ds<0 on II. Summing over the LiL_i arcs gives (30).

For a physically regular solution, every boundary component attains the minimum of G\mathcal{G}, so the rotation bound applies to every component containing poles. The following estimate covers components without poles. It does not require A^\widehat{\mathcal{A}} to be single-valued along the component.

Lemma 3.3 (Pole-free boundary bound). Suppose the value of G\mathcal{G} along Γi\Gamma_i is its global minimum on Σ‾\overline{\Sigma}. If Li=0L_i=0, then di<1d_i<1.

Proof. Continue A^\widehat{\mathcal{A}} along the whole component. It is finite, and P(s)P(s) stays in the open lower half-plane by (31). Hence a continuous branch of arg⁡P\arg P takes values in (−π,0)(-\pi,0) and Δarg⁡P<π\Delta\arg P<\pi, even if the continued values do not close. Also Δarg⁡A^<0\Delta\arg\widehat{\mathcal{A}}<0 by (31). By (21),

πdi=Δarg⁡(dA^/ds)=Δarg⁡P+Δarg⁡A^<π.\pi d_i=\Delta\arg(d\widehat{\mathcal{A}}/ds)=\Delta\arg P+\Delta\arg\widehat{\mathcal{A}}<\pi.

Disk topology without monodromy

We now combine the boundary bounds of § 3 with the count of zeros on the double.

Theorem 4.1 (Only the disk). Let Σ\Sigma have genus gg and ν≥1\nu\geq1 boundary components, with five-brane poles only on ∂Σ\partial\Sigma, no interior punctures, and single-valued meromorphic differentials dA±d\mathcal{A}_{\pm} satisfying the boundary reflection condition (13) of DGU II. If the solution is physically regular in the sense of § 2.2, then

2g+ν<2,that is(g,ν)=(0,1): Σ is a disk.(32)2g+\nu<2,\qquad\text{that is}\qquad(g,\nu)=(0,1):\ \Sigma\text{ is a disk}. \tag*{(32)}

Proof. To recover the count of DGU II [10 §6.4, (6.23)], consider the meromorphic one-form dA+d\mathcal{A}_{+} on the double Σ^\widehat{\Sigma}. Its only poles are the L^\widehat{L} simple poles on the gluing boundary. Reflection identifies the zeros of dA+d\mathcal{A}_{+} in the mirror interior with those of dA−d\mathcal{A}_{-} in the physical interior, where κ2>0\kappa^{2}>0 excludes them. On a regular boundary arc a zero would be common to both differentials, which Hopf’s lemma excludes as in § 3.1. The total zero order on Σ^\widehat{\Sigma} is therefore exactly NN, the number of zeros of dA+d\mathcal{A}_{+} in the interior of Σ\Sigma. Since dA+d\mathcal{A}_{+} is a meromorphic one-form on the double, its total zero order minus its total pole order is the degree of the canonical bundle, deg⁡KΣ^=2g^−2\deg K_{\widehat{\Sigma}}=2\widehat{g}-2, with g^=2g+ν−1\widehat{g}=2g+\nu-1. Hence

N−L=2g^−2,N=L+4g+2ν−4.(33)N-L=2\widehat{g}-2,\qquad N=L+4g+2\nu-4. \tag*{(33)}

We now show that the boundary estimates imply N<LN<L. By Theorem 3.1, N=∑idiN=\sum_i d_i and each did_i is a positive integer. A pole-free boundary component would have di<1d_i<1 by Theorem 3.3, so every component carries poles. Since G=0\mathcal{G}=0 on every boundary component and G>0\mathcal{G}>0 inside, Theorem 3.2 gives di<Lid_i<L_i on each. Summing yields

N=∑idi<∑iLi=L.N=\sum_i d_i<\sum_i L_i=L.

Substitution in (33) gives 2g+ν<22g+\nu<2, which permits only (g,ν)=(0,1)(g,\nu)=(0,1).

On the disk, d1=L−2>0d_1 = L - 2 > 0 recovers the bound L≥3L \ge3 of DGU II [10 §3.3]. In the brane web, L=2L = 2 would mean two external stacks with opposite charges by charge conservation. They would lie along a single straight line and meet in no junction.

The annulus studied in DGU II [10 §5.7] shows the two ways in which the proof excludes a surface. Suppose first that all poles lie on one boundary component. On the other component, Theorem 3.3 gives di<1d_i < 1, whereas did_i is a positive integer. If both components carry poles, the rotation bound gives d1+d2<L1+L2=Ld_1 + d_2 < L_1 + L_2 = L, while the global count (33) requires d1+d2=Ld_1 + d_2 = L.

Monodromy and orientifold punctures

The boundary rotation bounds of § 3.2 still hold in the presence of punctures. In § 4, λ\lambda was single-valued, so the argument principle identified the total boundary winding ∑idi\sum_i d_i with the number of interior zeros of dA+d\mathcal{A}_+. Monodromy can mix dA+d\mathcal{A}_+ and dA−d\mathcal{A}_-, making λ\lambda multivalued. The argument principle then no longer gives this identification, and the did_i need not be integers. We instead count, on each Γi\Gamma_i, the zeros of D dwD\,dw and the five-brane poles at which its residue vanishes, with D dwD\,dw as in (4). These are precisely the boundary poles of the ratio V=α∂wA+/D\mathcal{V} = \alpha\partial_w\mathcal{A}_+/D studied below. Their number ziz_i satisfies zi≤diz_i \le d_i under the hypotheses below. The degree formula on the Schottky double fixes ∑izi\sum_i z_i when applied to D dwD\,dw for seven-branes or to D2 dw⊗dwD^2\,dw \otimes dw in the presence of orientifolds.

The invariant differential

Consider first mutually local seven-branes. In the common D7-brane frame of § 2.3, monodromy shifts both differentials by the same one-form, so their difference is single-valued. In a general duality frame we denote the corresponding monodromy-invariant one-form by D dwD\,dw. It extends meromorphically to the double, so its zeros and poles obey the same degree formula used in § 4. We write DD for the coefficient of D dwD\,dw in a local coordinate and refer to D dwD\,dw as the invariant differential. Statements that DD is real use a coordinate in which the boundary is real.

Lemma 5.1 (Invariant differential). There is a complex constant α\alpha, with ∣α∣=1|\alpha| = 1, such that

D:=α ∂wA+−αˉ ∂wˉA−(34)D := \alpha\,\partial_w\mathcal{A}_+ - \bar{\alpha}\,\partial_{\bar w}\mathcal{A}_- \tag*{(34)}

is invariant under each seven-brane monodromy MjM_j. In the common D7-brane frame these matrices are powers TnjT^{n_j} of the same generator, with strengths nj2>0n_j^2 > 0 that may vary among punctures. The common invariant direction determines α\alpha up to sign when at least one such puncture is present. On an unpunctured surface in the monodromy-free class, any unit α\alpha may be chosen. Under the global and local assumptions of § 2.3, D dwD\,dw has the following properties.

  1. D dwD\,dw is single-valued and meromorphic on Σ\Sigma, regular at the punctures, with poles only at the five-brane points, where it has at worst a simple pole, of residue 2Re⁡(αZℓ)∈R2\operatorname{Re}(\alpha Z_\ell) \in\mathbb{R}.

The residue may vanish, and DD is then holomorphic at that point; we write L0L_0 for the number of five-brane points at which it does, so that DD has exactly L−L0L-L_0 poles.

  1. DD is real along ∂Σ\partial\Sigma, so D dwD\,\mathrm{d}w extends to a meromorphic one-form D^\widehat{D} on the Schottky double Σ^\widehat{\Sigma}.

  1. DD has no zero in the interior of Σ\Sigma, nor in its mirror.

  1. Define V:=α ∂wA+/D\mathcal{V}:=\alpha\,\partial_w A_+/D. At regular points where D≠0D\ne0, we have

κ2=∣D∣2(1−2Re⁡V),λ=α‾α⋅VV−1,and on ∂ΣRe⁡V=12.(35)\kappa^2=|D|^2(1-2\operatorname{Re}\mathcal{V}),\qquad\lambda=\frac{\overline{\alpha}}{\alpha}\cdot\frac{\mathcal{V}}{\mathcal{V}-1},\qquad\text{and on }\partial\Sigma\quad\operatorname{Re}\mathcal{V}=\frac{1}{2}. \tag*{(35)}

Thus interior positivity is equivalent to Re⁡V<12\operatorname{Re}\mathcal{V}<\frac{1}{2}. The Cayley transform in (35) maps the boundary line Re⁡V=12\operatorname{Re}\mathcal{V}=\frac{1}{2} to the unit circle in the λ\lambda-plane.

Proof. Invariance of DD means that (α,−α‾)(\alpha,-\overline{\alpha}) is a left eigenvector of MjM_j with eigenvalue 1. A non-trivial parabolic element of trace +2+2 has aj=1+iβja_j=1+i\beta_j, with βj∈R∖{0}\beta_j\in\mathbb{R}\setminus\{0\}, and ∣bj∣=∣βj∣|b_j|=|\beta_j| by the SU(1,1)SU(1,1) constraint. The two components of the eigenvector equation are conjugate to one another and reduce to

b‾j=iβjαα‾.(36)\overline{b}_j=i\beta_j\frac{\alpha}{\overline{\alpha}}. \tag*{(36)}

Since ∣b‾j/(iβj)∣=1\left|\overline{b}_j/(i\beta_j)\right|=1, we can choose α=eiϑ\alpha=e^{i\vartheta} with e2iϑ=b‾j/(iβj)e^{2i\vartheta}=\overline{b}_j/(i\beta_j). This determines α\alpha up to sign. Commuting parabolic elements share their invariant direction, so the same α\alpha works for every puncture. Changing its sign changes DD by a sign but leaves ∣D∣2|D|^2 and V\mathcal{V} unchanged.

(1) The absence of independent duality holonomy makes DD single-valued. In the D7-brane frame, the logarithmic terms in (18) cancel, leaving D=α(∂wA+seed−∂wA−seed)D=\alpha(\partial_w A_+^{\mathrm{seed}}-\partial_w A_-^{\mathrm{seed}}), regular at punctures. At a five-brane pole, (14) gives the residue 2Re⁡(αZℓ)2\operatorname{Re}(\alpha Z_\ell); if it vanishes, the pole is removable.

(2) On ∂Σ\partial\Sigma the reflection condition (13) gives D=α ∂wA++α‾ ∂wA+‾=2Re⁡(α ∂wA+)D=\alpha\,\partial_w A_++\overline{\alpha}\,\overline{\partial_w A_+}=2\operatorname{Re}(\alpha\,\partial_w A_+), which is real. Schwarz reflection therefore extends its coefficient to the mirror by D(w‾)=D(w)‾D(\overline{w})=\overline{D(w)}, giving the one-form on the double.

(3) If D=0D=0 then ∣∂wA+∣=∣∂wA−∣|\partial_w A_+|=|\partial_w A_-|, because ∣α∣=∣α‾∣|\alpha|=|\overline{\alpha}|, hence κ2=0\kappa^2=0, which interior positivity excludes away from the punctures. At a seven-brane puncture wjw_j, we assume the logarithmic divergence of κ2\kappa^2 specified in § 2.3. Its coefficient in (19) is proportional to nj2∣D(wj)∣2n_j^2|D(w_j)|^2, so this source condition requires D(wj)≠0D(w_j)\ne0. Reflection gives the same conclusion on the mirror.

(4) Substituting ∂wA+=α−1VD\partial_w A_+=\alpha^{-1}\mathcal{V}D and ∂wA−=α‾−1(V−1)D\partial_w A_-=\overline{\alpha}^{-1}(\mathcal{V}-1)D into the definitions of λ\lambda and κ2\kappa^2 gives (35).

Five-brane points at which the residue of DD vanishes must be retained in the count: in the D7-brane frame the residue of DD is proportional to the NS5 charge, so a pure D5 pole is regular for DD but is a pole of V\mathcal{V} [11 below (3.71)].

Counting the boundary poles

We count the boundary poles of V\mathcal{V} using the degree of D^\hat{D}, that is, its total zero order minus its total pole order.

Lemma 5.2 (Boundary pole count). Let ziz_i denote the number of poles of V\mathcal{V} on Γi\Gamma_i. On the regular boundary arcs these poles occur precisely at the zeros of DD. At a five-brane pole, V\mathcal{V} has a pole precisely when the residue of D dwD\,dw vanishes. In that case DD is finite and nonzero. All poles of V\mathcal{V} on Γi\Gamma_i are simple. Summing both contributions over the boundary components gives

∑i=1νzi=(L−L0)+2g^−2+L0⏟five-brane poles withRes⁡pℓ(Ddw)=0=L+2g^−2=L+4g+2ν−4.(37)\sum_{i=1}^{\nu} z_i = (L-L_0)+2\hat{g}-2+\underbrace{L_0}_{\substack{\text{five-brane poles with}\\ \operatorname{Res}_{p_\ell}(D\,dw)=0}} = L+2\hat{g}-2=L+4g+2\nu-4. \tag*{(37)}

Proof. By § 3, neither of ∂wA±\partial_w\mathcal{A}_{\pm} vanishes on an open boundary arc. Thus a zero of DD there gives a pole of V\mathcal{V} of the same order. At a five-brane point with Re⁡(αZℓ)≠0\operatorname{Re}(\alpha Z_\ell)\ne0, the poles in numerator and denominator cancel and V→αZℓ/[2Re⁡(αZℓ)]\mathcal{V}\to\alpha Z_\ell/[2\operatorname{Re}(\alpha Z_\ell)]. At the remaining L0L_0 points, the numerator still has a simple pole while DD is holomorphic.

All these poles are simple, so that their number equals their total order. This follows from (23). Dividing (34) by α∂wA−\alpha\partial_w\mathcal{A}_{-} gives

λ−αˉα=Dα∂wA−,(38)\lambda-\frac{\bar{\alpha}}{\alpha}=\frac{D}{\alpha\partial_w\mathcal{A}_{-}}, \tag*{(38)}

with ∂wA−\partial_w\mathcal{A}_{-} nonzero on every regular boundary arc. At a boundary zero x0x_0 of DD of order mm away from the poles, (38) gives λ−αˉ/α∼c(w−x0)m\lambda-\bar{\alpha}/\alpha\sim c(w-x_0)^m and λ′∼mc(w−x0)m−1\lambda'\sim mc(w-x_0)^{m-1}, which (23) forbids unless m=1m=1. At a five-brane point λ\lambda is still holomorphic, with λ(pℓ)=−Zℓ/Zˉℓ\lambda(p_\ell)=-Z_\ell/\bar{Z}_\ell of unit modulus. Applying Hopf’s lemma to the extended log⁡∣λ∣\log|\lambda| gives λ′(pℓ)≠0\lambda'(p_\ell)\ne0 at the pole as well. When the residue of D dwD\,dw vanishes, (38) gives λ(pℓ)=αˉ/α\lambda(p_\ell)=\bar{\alpha}/\alpha. Since ∂wA−∼−Zℓ/(w−pℓ)\partial_w\mathcal{A}_{-}\sim-Z_\ell/(w-p_\ell), taking the limit in D=α∂wA−(λ−αˉ/α)D=\alpha\partial_w\mathcal{A}_{-}(\lambda-\bar{\alpha}/\alpha) gives D(pℓ)=αZℓλ′(pℓ)≠0D(p_\ell)=\alpha Z_\ell\lambda'(p_\ell)\ne0. Thus each of the L0L_0 points contributes precisely one simple pole of V\mathcal{V}.

The same degree formula used in § 4 gives L−L0+2g^−2L-L_0+2\hat{g}-2 zeros of D^\hat{D}, counted with multiplicity. They all lie on the boundary by Theorem 5.1 and are simple by the preceding argument. Each boundary point is counted once on the double. Adding the L0L_0 five-brane points with vanishing residue, which are poles of V\mathcal{V} but not of DD, gives (37).

For a disk, g=0g=0 and ν=1\nu=1, so (37) gives z1=L−2z_1=L-2 for the number of poles of V\mathcal{V} on its single boundary component. For the two-pole disks of Chaney and Uhlemann this boundary count vanishes regardless of the number of interior seven-brane punctures [31 §§3.4–3.5]. For example, the solution denoted by +N,M,k,j+N,M,k,j in their notation has two poles and two punctures at distinct interior points [31 §4.4, Fig. 7].

Orientifold punctures and the sign of DD

An orientifold seven-plane has monodromy MO7±=−T±4M_{\mathrm{O7}^{\pm}}=-T^{\pm4}. More generally, −Tk-T^k describes an O7−\mathrm{O7}^{-} with k+4k+4 D7-branes or, for k≥1k\ge1, an O7+\mathrm{O7}^{+} with k−4k-4 D7-branes [22 §II.1]. We restrict to k=nj2≥0k=n_j^2\geq0, which excludes an O7−O7^{-} with fewer than four D7-branes. For k>0k>0, this monodromy commutes with non-trivial seven-brane monodromies only when they lie in the same parabolic subgroup. For k=0k=0, it is the central element −1-1, corresponding to an O7−O7^{-} with four D7-branes and no logarithmic axion monodromy.

In the common D7-brane frame, continuation around the puncture acts as [22 §II.5]

A±⟼−(A±+ik2(A+−A−)),(39)A_{\pm}\longmapsto-\left(A_{\pm}+\frac{ik}{2}(A_{+}-A_{-})\right), \tag*{(39)}

which in the notation of (11) is M=−PM=-P, with PP the trace-+2+2 parabolic element of Theorem 5.1 that a seven-brane puncture of strength kk carries. The left eigenvector of PP with eigenvalue 11 is (α,−α)(\alpha,-\alpha) with α\alpha real, so that we choose α=1\alpha=1 and write D=∂wA+−∂wA−D=\partial_wA_{+}-\partial_wA_{-}, as in [22]. For M=−PM=-P the eigenvalue is −1-1:

D⟼−D.(40)D\longmapsto-D. \tag*{(40)}

For k=0k=0, the central element −1-1 imposes no constraint on α\alpha, but still sends DD to −D-D. No nonzero linear combination is fixed by this monodromy. Thus DD is double-valued: its local branches differ only by a sign, with D↦−DD\mapsto-D around each orientifold puncture. Equivalently, D dwD\,dw takes values in a flat Z2\mathbb{Z}_2 bundle. The single-valuedness in Theorem 5.1(1) fails, while D2 dw⊗dwD^2\,dw\otimes dw remains single-valued.

Uhlemann presented closed-form solutions with one orientifold puncture, obtained as a quotient of disk solutions with two seven-brane punctures [22 §II.4]. In a local coordinate ξ\xi vanishing at the puncture, their differentials behave as [22 §II.5]

∂ξA±=1−ξ[a±(ξ)+k4πln⁡(−ξ)(a+(ξ)−a−(ξ))],k=nj2≥0,(41)\partial_{\xi}A_{\pm}=\frac{1}{\sqrt{-\xi}}\left[a_{\pm}(\xi)+\frac{k}{4\pi}\ln(-\xi)\left(a_{+}(\xi)-a_{-}(\xi)\right)\right],\qquad k=n_j^2\geq0, \tag*{(41)}

with a±a_{\pm} holomorphic at ξ=0\xi=0; they are written ∂uA±(0)\partial_uA_{\pm}^{(0)} there. This behavior was proposed in [22] as the starting point for solutions with several orientifold punctures. Following this proposal, we require (41) at every orientifold puncture, independently of the topology of Σ\Sigma and the number of punctures. We also assume that C:=a+(0)−a−(0)≠0C:=a_{+}(0)-a_{-}(0)\ne0. The sign change of both differentials comes from the common factor 1/−ξ1/\sqrt{-\xi}, and a constant phase in the choice of branch is absorbed in a±a_{\pm}. The logarithmic terms cancel in the difference, so

D=∂ξ(A+−A−)=a+−a−−ξ→ξ→0C−ξ.(42)D=\partial_{\xi}(A_{+}-A_{-})=\frac{a_{+}-a_{-}}{\sqrt{-\xi}}\xrightarrow{\xi\to0}\frac{C}{\sqrt{-\xi}}. \tag*{(42)}

We say that DD has a half-order pole. The change −ξ→−−ξ\sqrt{-\xi}\to-\sqrt{-\xi} on a circuit is the sign change (40). Expanding the difference of squares in (5) gives

κ2=1∣ξ∣[∣a−∣2−∣a+∣2+k2πln⁡1∣ξ∣∣a+−a−∣2].(43)\kappa^2=\frac{1}{|\xi|}\left[|a_{-}|^2-|a_{+}|^2+\frac{k}{2\pi}\ln\frac{1}{|\xi|}|a_{+}-a_{-}|^2\right]. \tag*{(43)}

If k>0k>0 the last term dominates and

κ2=k∣C∣22πln⁡(1/∣ξ∣)∣ξ∣+…,(44)\kappa^2=\frac{k|C|^2}{2\pi}\frac{\ln(1/|\xi|)}{|\xi|}+\ldots, \tag*{(44)}

which is positive and divergent, as in the seven-brane case (19). For k=0k=0 the logarithm is absent, and κ2=(∣a−(0)∣2−∣a+(0)∣2)/∣ξ∣+O(1)\kappa^2=(|a_-(0)|^2-|a_+(0)|^2)/|\xi|+O(1), which is positive when ∣a−(0)∣>∣a+(0)∣|a_-(0)|>|a_+(0)|. For k<0k<0 the logarithmic term is negative and divergent, so the restriction to k≥0k\geq0 made at the beginning of this subsection is required by regularity, as noted in [22 §II.1]. Likewise, (19) is negative near a seven-brane puncture with nj2<0n_j^2<0 and D(wj)≠0D(w_j)\ne0.

The condition C≠0C\ne0 is an assumption of the orientifold theorem (Theorem 5.8). If C=0C=0, then a+(0)=a−(0)a_+(0)=a_-(0), and there are two cases. If the common value is nonzero, regularity already fails locally: either a+=a−a_+=a_- identically and κ2\kappa^2 vanishes, or a−−a+=bξm+⋯a_--a_+=b\xi^m+\cdots with b≠0b\ne0 and m≥1m\geq1. In the second case the bracket in (43) is 2Re⁡(a+(0)bξm)+⋯2\operatorname{Re}(a_+(0)b\xi^m)+\cdots, which is negative for some directions of ξ\xi, however small ∣ξ∣|\xi| is. The assumption therefore excludes a possibly regular solution only if a+(0)=a−(0)=0a_+(0)=a_-(0)=0. In that case DD vanishes at the puncture and has no half-order pole, so the count of § 5.6 would change. We do not consider this case.

Boundary winding in the presence of monodromy

In the common D7-brane frame, α=1\alpha=1 and D=∂wA+−∂wA−D=\partial_w\mathcal{A}_+-\partial_w\mathcal{A}_-. By § 2.3, continuation counterclockwise around a seven-brane puncture wjw_j sends ∂wA±↦∂wA±+inj22D\partial_w\mathcal{A}_\pm\mapsto\partial_w\mathcal{A}_\pm+\frac{\mathrm{i}n_j^2}{2}D, and by (39) an orientifold puncture adds an overall minus sign. In both cases the ratio V=∂wA+/D\mathcal{V}=\partial_w\mathcal{A}_+/D changes by

V⟼V+inj22.(45)\mathcal{V}\longmapsto\mathcal{V}+\frac{\mathrm{i}n_j^2}{2}. \tag*{(45)}

The sign change cancels in V\mathcal{V}, and likewise in λ\lambda. The imaginary shift remains and can make did_i non-integral. At an orientifold puncture, (41) gives V\mathcal{V} the same logarithm nj24πln⁡(−ξ)\frac{n_j^2}{4\pi}\ln(-\xi) as ff has at a seven-brane puncture. The sum over jj in (18) therefore runs over the punctures of both types, and an orientifold puncture contributes with its nj2n_j^2 in the same way as a seven-brane puncture. The next lemma gives the shift accumulated along a boundary component.

Lemma 5.3 (Shift along a boundary component). Assume the global construction of § 2.3. Continued once around Γi\Gamma_i in the positive direction, V\mathcal{V} returns to V+isi\mathcal{V}+\mathrm{i}s_i. Let ωi\omega_i be the harmonic measure of Γi\Gamma_i in Σ\Sigma — the harmonic function on Σ\Sigma with boundary value 11 on Γi\Gamma_i and 00 on the other components, so that ∑iωi≡1\sum_i\omega_i\equiv1. Then

si=12∑jnj2ωi(wj)≥0,∑isi=12∑jnj2.(46)s_i=\frac{1}{2}\sum_j n_j^2\omega_i(w_j)\geq0,\qquad\sum_i s_i=\frac{1}{2}\sum_j n_j^2. \tag*{(46)}

The coefficients nj2n_j^2 are non-negative for both types of puncture. Since harmonic measure is strictly positive in the interior, si>0s_i>0 whenever at least one puncture has positive strength.

Proof. In the D7-brane frame the ff-terms in ∂wA±\partial_w\mathcal{A}_\pm are identical and cancel in DD, so D=∂wA+seed−∂wA−seedD=\partial_w\mathcal{A}_+^{\mathrm{seed}}-\partial_w\mathcal{A}_-^{\mathrm{seed}} and

V=f+∂wA+seedD.\mathcal{V}=f+\frac{\partial_w\mathcal{A}_+^{\mathrm{seed}}}{D}.

By the global assumption of § 2.3 the second term is single-valued: the seed is single-valued, or, in the orientifold case, changes sign together with DD. The change of V\mathcal{V} around Γi\Gamma_i is therefore that of ff. Since Re⁡f\operatorname{Re} f is single-valued, only Im⁡f\operatorname{Im} f changes, and the change of ff is isiis_i with si=∮ΓidIm⁡fs_i=\oint_{\Gamma_i}d\operatorname{Im} f. By the Cauchy–Riemann equations, dIm⁡f=−∂nRe⁡f dℓd\operatorname{Im} f=-\partial_n\operatorname{Re} f\,d\ell along the boundary, with ∂n\partial_n the inward normal derivative, and (18) gives

si=14π∑jnj2∮Γi∂ng(⋅,wj) dℓ=12∑jnj2ωi(wj),s_i=\frac{1}{4\pi}\sum_j n_j^2\oint_{\Gamma_i}\partial_n g(\cdot,w_j)\,d\ell=\frac{1}{2}\sum_j n_j^2\omega_i(w_j),

using the standard relation 12π∮Γi∂ng(⋅,wj) dℓ=ωi(wj)\frac{1}{2\pi}\oint_{\Gamma_i}\partial_n g(\cdot,w_j)\,d\ell=\omega_i(w_j) between the Green’s function and harmonic measure.

The total ∑isi=12∑jnj2\sum_i s_i=\frac{1}{2}\sum_j n_j^2 is the sum of the local shifts (45), as it must be: V\mathcal{V} is multi-valued, but dVd\mathcal{V} is a single-valued holomorphic one-form in the punctured interior, and the sum of the boundary components, pushed slightly into the interior, is homologous there to the sum of small circles around the punctures. The global assumption enters only in the distribution of this total among the boundary components. Without it, Im⁡(V−f)\operatorname{Im}(\mathcal{V}-f) could have a nonzero period along each Γi\Gamma_i, which would add to sis_i in (46). These periods sum to zero but need not have a fixed sign, so si≥0s_i\geq0 would no longer follow. This limitation is noted in § 7.

Lemma 5.4 (The fractional part). Let ziz_i be as in Theorem 5.2, and let t=Im⁡Vt=\operatorname{Im}\mathcal{V}. For a base point at which V\mathcal{V} is finite, with t0t_0 the value of tt there,

di=zi+ϵi,ϵi=1π[arctan⁡(2(t0+si))−arctan⁡(2t0)]∈[0,1),(47)d_i=z_i+\epsilon_i,\qquad\epsilon_i=\frac{1}{\pi}\left[\arctan\left(2(t_0+s_i)\right)-\arctan(2t_0)\right]\in[0,1), \tag*{(47)}

and for a base point at a pole of V\mathcal{V}, one has di=zid_i=z_i. A five-brane point is of this type only if the residue of DD vanishes there. Otherwise V\mathcal{V} is finite and t0=Im⁡(αZℓ)/[2Re⁡(αZℓ)]t_0=\operatorname{Im}(\alpha Z_\ell)/[2\operatorname{Re}(\alpha Z_\ell)]. In all cases di≥zid_i\geq z_i. At a finite base point equality holds precisely when si=0s_i=0, equivalently when every puncture has zero strength.

Proof. By (35) and Re⁡V=12\operatorname{Re}\mathcal{V}=\frac{1}{2} on ∂Σ\partial\Sigma, arg⁡λ=2arctan⁡(2t)−π+arg⁡(α‾/α)\arg\lambda=2\arctan(2t)-\pi+\arg(\overline{\alpha}/\alpha), so arg⁡λ\arg\lambda is a function of tt alone and πdi=ΔΓiΘ\pi d_i=\Delta_{\Gamma_i}\Theta. Here Θ\Theta is obtained from arctan⁡(2t)\arctan(2t) by adding multiples of π\pi so that it remains continuous when tt passes through infinity. By (22), tt is strictly increasing between its poles, and it passes through ∞\infty exactly at the poles of V\mathcal{V}, all of which are simple by Theorem 5.2; each passage advances Θ\Theta by π\pi, contributing πzi\pi z_i. The endpoints differ by the monodromy, tend=t0+sit_{\mathrm{end}}=t_0+s_i, contributing the bracket, which lies in [0,π)[0,\pi) because si≥0s_i\geq0 by Theorem 5.3 and vanishes exactly when si=0s_i=0. For a base point at a pole of V\mathcal{V} the circuit runs from t=−∞t=-\infty to t=+∞t=+\infty, so zi−1z_i-1 passages occur inside it and the endpoints contribute π\pi: again πzi\pi z_i, the monodromy being invisible between the two ends of the real line.

Lemma 5.5 (No pole-free component). Assume the global construction of § 2.3, with DD single-valued up to a sign. In a regular solution every boundary component carries a five-brane pole: Li≥1L_i\geq1 for all ii.

Proof. Suppose Li=0L_i=0. By Theorem 3.3, di<1d_i<1. Since ziz_i is a non-negative integer and zi≤diz_i\leq d_i by Theorem 5.4, it vanishes. Hence DD, which is real on Γi\Gamma_i, has no zero there by

Theorem 5.2. It cannot change sign on a circuit of Γi\Gamma_i, so it returns to itself, and the integral of D dwD\,dw along Γi\Gamma_i is real and nonzero.

We show that this integral must be imaginary. Continuation around Γi\Gamma_i acts as A±↦(MA)±+a±A_\pm\mapsto(MA)_\pm+a_\pm, with MM the monodromy and a±a_\pm constants. The linear part leaves G\mathcal{G} invariant. By (5) and (6), the constants a±a_\pm change G\mathcal{G} by

2Re⁡[(a‾+−a−)(A++A‾−)]+const.2\operatorname{Re}\left[(\overline{a}_+ - a_-)(A_+ + \overline{A}_-)\right] + \mathrm{const}.

If a−≠a‾+a_- \ne\overline{a}_+ this is not constant, since otherwise λ\lambda would be a constant of unit modulus and κ2\kappa^2 would vanish. Since G\mathcal{G} is single-valued, a−=a‾+a_-=\overline{a}_+. Since DD returns to itself, MM leaves αA+−α‾A−\alpha A_+ - \overline{\alpha}A_- invariant, so the integral of D dw=d(αA+−α‾A−)D\,dw=d(\alpha A_+ - \overline{\alpha}A_-) along Γi\Gamma_i is αa+−α‾a−=2iIm⁡(αa+)\alpha a_+ - \overline{\alpha}a_- = 2i\operatorname{Im}(\alpha a_+). This contradicts the first paragraph.

Topology constraint with seven-branes

We combine the count of Theorem 5.2 with the bounds of § 3.2 and Theorem 5.4.

Theorem 5.6 (Only the disk, with seven-branes). Let Σ\Sigma have genus gg and ν≥1\nu\ge1 boundary components, with five-brane poles only on ∂Σ\partial\Sigma, and any number of punctures carrying mutually commuting parabolic monodromies of trace +2+2. Assume the global construction (18) of § 2.3, with no independent duality holonomy, and D(wj)≠0D(w_j) \ne0 at each puncture. If the solution is physically regular in the sense of § 2.2, then

2g+ν<2,that is(g,ν)=(0,1): Σ is a disk.(48)2g+\nu< 2,\qquad\text{that is}\qquad(g,\nu)=(0,1):\ \Sigma\text{ is a disk}. \tag*{(48)}

Proof. The function G\mathcal{G} vanishes on ∂Σ\partial\Sigma and is positive in the interior, including at the punctures, so the hypothesis of Theorem 3.2 holds on every component. By Theorem 5.5 every component carries a pole. Take the base point of did_i at a five-brane pole. Then Theorem 3.2 gives di<Lid_i<L_i and Theorem 5.4 gives zi≤diz_i\le d_i. Since ziz_i is an integer, zi≤Li−1z_i\le L_i-1. Summing over the ν\nu components and using (37),

L+4g+2ν−4=∑izi≤L−ν.L+4g+2\nu-4=\sum_i z_i\le L-\nu.

Thus 4g+3ν≤44g+3\nu\le4, and Σ\Sigma is a disk.

Without punctures si=0s_i=0 and di=zid_i=z_i, so this proof reduces to that of Theorem 4.1, with the boundary poles of V\mathcal{V} counted in place of the interior zeros of λ\lambda.

Topology constraint with orientifolds

With orientifolds DD is defined only up to a sign, and we count with D2D^2 instead.

Lemma 5.7 (The count with orientifolds). Assume (41) with C≠0C\ne0 at each orientifold puncture. Let I−I_- be the number of such punctures, and let ziz_i be as in Theorem 5.2. Then

∑i=1νzi=L+I−+4g+2ν−4.(49)\sum_{i=1}^{\nu}z_i=L+I_-+4g+2\nu-4. \tag*{(49)}

Proof. DD is defined only up to sign, but Q:=D2 dw⊗dwQ := D^2\,\mathrm{d}w\otimes\mathrm{d}w is single-valued and meromorphic, and real on ∂Σ\partial\Sigma because DD is, so QQ doubles to a meromorphic quadratic differential Q^\widehat{Q} on Σ^\widehat{\Sigma}. A quadratic differential has twice the degree of a one-form, so its number of zeros minus its number of poles, counted with orders, is 4g^−44\widehat{g}-4. Its poles are double at each of the L−L0L-L_0 five-brane points of non-vanishing residue, which lie on the fixed locus, and simple at each orientifold puncture and at its mirror image by (42); a seven-brane puncture contributes none, and neither does one of the L0L_0 five-brane points at which DD is holomorphic. Its zeros: none in the interior or its mirror, since D=0D=0 forces κ2=0\kappa^2=0; and each zero of DD on ∂Σ\partial\Sigma is simple by Theorem 5.2, hence a double zero of QQ. So

2(∑izi−L0)−2(L−L0)−2I−=4g^−42\left(\sum_i z_i-L_0\right)-2(L-L_0)-2I_- = 4\widehat{g}-4

which gives (49).

Theorem 5.8 (Only the disk, with orientifolds). Let Σ\Sigma have genus gg and ν≥1\nu\geq1 boundary components, with five-brane poles only on ∂Σ\partial\Sigma. Allow puncture monodromies Mj=σjTkjM_j=\sigma_j T^{k_j} in a common D7-brane frame, where σj=±1\sigma_j=\pm1 and kj=nj2≥0k_j=n_j^2\geq0. This includes the central monodromy −1-1 when σj=−1\sigma_j=-1 and kj=0k_j=0. Assume the global construction (18) of § 2.3, with DD single-valued up to a sign, D↦−DD\mapsto-D around each orientifold puncture, and no independent duality holonomy. Assume D(wj)≠0D(w_j)\ne0 at each seven-brane puncture, and (41) with C≠0C\ne0 at each orientifold puncture. If the solution is physically regular in the sense of § 2.2, then

(g,ν)=(0,1)andI−≤1:Σ is a disk with at most one orientifold puncture.(50)(g,\nu)=(0,1)\qquad\text{and}\qquad I_-\leq1:\quad\Sigma\text{ is a disk with at most one orientifold puncture.} \tag*{(50)}

Proof. Theorems 5.3 to 5.5 hold for punctures of either type, so zi≤Li−1z_i\leq L_i-1 on every component, as in the proof of Theorem 5.6. The count (49) gives

L+I−+4g+2ν−4=∑izi≤L−ν,that isI−+4g+3ν−4≤0.L+I_-+4g+2\nu-4=\sum_i z_i\leq L-\nu,\qquad\text{that is}\qquad I_-+4g+3\nu-4\leq0.

Since ν≥1\nu\geq1, this requires (g,ν)=(0,1)(g,\nu)=(0,1) and I−≤1I_-\leq1.

The counts alone would allow solutions without five-brane poles. For L=0L=0, every ziz_i vanishes by Theorems 3.3 and 5.4, and (49) gives I−+4g+2ν−4=0I_-+4g+2\nu-4=0, satisfied by an annulus without orientifolds or a disk with two orientifolds. Both are excluded by Theorem 5.5.

The fractional winding is realized by the regular one-pole disk solutions of [22 §III.2], denoted IN,j±I^{\pm}_{N,j} there. Their NN is a five-brane charge, and their jj is the exponent k=n12k=n_1^2 of the monodromy −Tk-T^k. With one orientifold, (49) gives z1=0z_1=0, so 0<d1=ϵ1<10<d_1=\epsilon_1<1 for k>0k>0. On the disk Theorem 5.3 gives s1=k/2s_1=k/2. The residue Z+Z_+ of these solutions is real, so t0=0t_0=0 for a base point at the five-brane pole, and (47) gives

d1=1πarctan⁡k,d_1=\frac{1}{\pi}\arctan k,

which lies between 0 and 12\frac{1}{2} and is smaller than L1=1L_1=1. One five-brane pole thus suffices in the presence of an orientifold, whereas Theorem 4.1 requires at least three without punctures and (37) requires at least two with seven-branes alone.

The external legs of the web

Fluder and Uhlemann [18 §III.1] observed that regularity enforces agreement between the order of the boundary poles and the order of the external five-branes around the junction. We recover this ordering from Theorem 3.2. On an unpunctured disk, label the poles in positive cyclic order along the boundary, with indices understood modulo LL, and let IℓI_\ell be the open boundary arc from pℓp_\ell to pℓ+1p_{\ell+1} in this orientation. Write θℓ=Arg⁡(Zℓ+1/Zℓ)∈(−π,π]\theta_\ell=\operatorname{Arg}(Z_{\ell+1}/Z_\ell)\in(-\pi,\pi], with Arg⁡\operatorname{Arg} the principal value, for the rotation of the residue from one pole to the next along the boundary. The direction of ZℓZ_\ell in the complex plane represents the direction of the five-brane charge. Figure 2 illustrates the two orientations for the four-stack web discussed below.

Charge ordering for the +N,M web

Figure 2. Charge ordering for the +N,M+N,M web. Left: positive traversal of the disk boundary takes p1p_1 to p2p_2 counterclockwise. The angular spacing of the poles depends on the choice of disk coordinate. Right: the charge directions are drawn with unit length and rotated together to place the D5 stacks horizontally. They turn clockwise from Z1Z_1 to Z2Z_2. For +N,M+N,M each step has θℓ=−π/2\theta_\ell=-\pi/2. In general the angles may differ, but their sum is −2π-2\pi.

Corollary 6.1 (External charge ordering). For a regular disk solution with LL poles and no punctures,

θℓ∈(−π,0)for every ℓ,∑ℓ=1Lθℓ=−2π.(51)\theta_\ell\in(-\pi,0)\quad\text{for every }\ell,\qquad\sum_{\ell=1}^{L}\theta_\ell=-2\pi. \tag*{(51)}

The five-brane charge directions follow the cyclic order of the poles and make one complete clockwise turn.

Proof. The disk has a single boundary component, on which G\mathcal{G} takes its minimum, so Theorem 3.2 gives δIℓ<π\delta_{I_\ell}<\pi, while (21) and (22) give δIℓ=12ΔIℓarg⁡λ>0\delta_{I_\ell}=\frac{1}{2}\Delta_{I_\ell}\arg\lambda>0. Recall that δIℓ\delta_{I_\ell} is the change of arg⁡∂wA+\arg\partial_w\mathcal{A}_+ along IℓI_\ell in an upper-half-plane coordinate w=x+iyw=x+iy. Near each end of IℓI_\ell the differential is dominated by the pole there. Just after pℓp_\ell we have ∂wA+≃Zℓ/(x−pℓ)\partial_w\mathcal{A}_+\simeq Z_\ell/(x-p_\ell) with x−pℓ>0x-p_\ell>0, so the phase starts at arg⁡Zℓ\arg Z_\ell. Just before pℓ+1p_{\ell+1} we have ∂wA+≃Zℓ+1/(x−pℓ+1)\partial_w\mathcal{A}_+\simeq Z_{\ell+1}/(x-p_{\ell+1}) with x−pℓ+1<0x-p_{\ell+1}<0, so the phase ends at arg⁡Zℓ+1+π\arg Z_{\ell+1}+\pi. Hence δIℓ≡θℓ+π(mod2π)\delta_{I_\ell}\equiv\theta_\ell+\pi\pmod{2\pi}. Since θℓ+π∈(0,2π]\theta_\ell+\pi\in(0,2\pi] and δIℓ∈(0,π)\delta_{I_\ell}\in(0,\pi), the two are equal, and θℓ∈(−π,0)\theta_\ell\in(-\pi,0). Finally, ∑ℓδIℓ=πd1\sum_\ell\delta_{I_\ell}=\pi d_1 by Section 3.2, and d1=N=L−2d_1=N=L-2 by Theorem 3.1 and (33). Therefore ∑ℓθℓ=π(L−2)−πL=−2π\sum_\ell\theta_\ell=\pi(L-2)-\pi L=-2\pi.

Since the charge directions turn strictly clockwise through a total of 2π2\pi, distinct boundary poles cannot carry five-brane charges in the same direction, and consecutive legs cannot be anti-parallel. A single pole may carry the total charge of several coincident five-branes. Non-consecutive legs may be anti-parallel, as in the +N,M+_{N,M} web of Figure 2 [14 §2.2], where each anti-parallel pair is separated by one other stack in the cyclic order.

As counted in Section 2.2, a regular disk solution with prescribed charges ZℓZ_\ell amounts to a solution of (16) for the cross-ratios of the poles and the constant A0\mathcal{A}^{0}, and such solutions are generically isolated. In addition κ2>0\kappa^2>0 requires the L−2L-2 zeros of ∂wA+=∑ℓZℓ/(w−pℓ)\partial_w\mathcal{A}_+=\sum_\ell Z_\ell/(w-p_\ell), the points sns_n of (17), to lie in the interior. The cyclic ordering of Theorem 6.1 is necessary for such a solution. Whether it is sufficient, and whether the solution is unique, remains open. The equations have been solved numerically for up to ten poles in [18], which also reports roots that violate κ2>0\kappa^2>0.

Discussion

In the monodromy-free class of DGU II, a regular solution with boundary five-brane poles requires Σ\Sigma to be a disk. This excludes the unpunctured annulus of DGU II [10 §5.7] as well as higher genus, even for a surface with a single boundary component. This proves, within the DGU II class, the disk restriction anticipated by Kaidi and Uhlemann from the M-theory description of five-brane webs [27 §3]. For webs with a regular supergravity dual in this class, Theorem 4.1 justifies the choice of a disk as the base of the S2S^2 fibration in the symmetry TFT derivation of Apruzzi et al. [34 §5.1].

For mutually local seven-branes, Theorem 5.6 excludes both positive genus and multiple boundary components under the global and local assumptions of § 2.3. In particular, no choice of parameters in the annulus ansatz of Lozano et al. [28 §3] yields a regular solution with standard localized D7-brane sources. Allowing an overall minus sign around orientifolds in the global construction, and imposing the non-degenerate local asymptotics of § 5.3, we find that Σ\Sigma is again a disk, with at most one orientifold puncture. In both cases every boundary component carries a five-brane pole (Theorem 5.5), so seven-branes and orientifolds alone cannot source a regular solution. More general puncture asymptotics and independent duality holonomy remain outside this analysis.

The restriction to localized sources matters for the topology question. The explicit AdS6\mathrm{AdS}_6 annulus of Lozano et al. has smeared sources on both boundary components [28 §5]. A continuous source distribution along a boundary precludes the decomposition into smooth arcs between isolated poles used to bound the phase rotation. Whether smeared sources admit surfaces of positive genus or with more than two boundary components remains open. A physical role for smearing also arises in the M-theory AdS5\mathrm{AdS}_5 duals proposed by Bah et al. [35, 36] and generalized by Couzens et al. [37], where a smeared M5-brane source realizes the irregular puncture of a four-dimensional Argyres–Douglas theory.

Theorems 5.65.6 and 5.85.8 rely on a charge direction preserved by every puncture monodromy. Such a direction exists for the parabolic monodromies of mutually local seven-branes and, up to a sign, for orientifolds. The latter include the D4D_4-type puncture with monodromy −1-1 produced by a Z2\mathbb{Z}_2 quotient [29 §2.1], which is the k=0k=0 orientifold of Section 5.3 [22]. Combinations of mutually non-local seven-branes, which realize exceptional flavor symmetries in five-brane webs [3], need not have such a direction. They can enter in two ways. First, a single puncture can carry their combined monodromy. In the S-fold solutions of Apruzzi et al. [29 §3], a puncture of type E6,7,8E_{6,7,8} or H0,1,2H_{0,1,2} sits at the center of a disk. Its monodromy is elliptic and has no eigenvalue ±1\pm1, so it preserves no charge direction, even up to a sign. Second, distinct punctures can carry non-commuting monodromies. The construction of such solutions was raised as an open problem in DGU III [11 §6]. In both cases the invariant differential D dwD\,dw is unavailable, and its zero and pole count on the double cannot be applied. Our argument therefore does not determine whether regularity still forces Σ\Sigma to be a disk in this broader class, although the known examples of the first kind are disks. We leave for future work the construction of supergravity solutions with several non-commuting monodromies, and the question whether Σ\Sigma is restricted to a disk in that generality.

The theorems can be restated in terms of the four-dimensional internal space M4M_4, the S2S^2 fibration over Σ\Sigma that degenerates on ∂Σ\partial\Sigma. The fiber is simply connected, so b1(M4)=b1(Σ)b_1(M_4)=b_1(\Sigma). For the disk, M4M_4 is a four-sphere with the points over the five-brane poles removed, and the statement that Σ\Sigma is a disk is the vanishing of b1(M4)b_1(M_4). After reduction on AdS6\mathrm{AdS}_6, the Type IIB equations on M4M_4 are those of gravity coupled to a nonlinear sigma model with target SL(3,R)/SO(2,1)SL(3,\mathbb{R})/SO(2,1) and a potential [5, 38]. The puncture monodromies act there as symmetries, with no reference to a preserved charge direction. Whether b1(M4)=0b_1(M_4)=0 follows from a curvature bound on M4M_4, by a Bochner-type argument, is an open question.

One motivation for the topology question comes from the comparison with four-dimensional class S\mathcal{S} theories. Fenati, Montero and Valenzuela [26 §4] considered increasing the genus gSg_S of the Riemann surface wrapped by the M5-branes, at fixed large M5-brane number. In the AdS5\mathrm{AdS}_5 dual the internal space is an S4S^4 fibered over the hyperbolic surface, whose area grows as gS−1g_S-1 at fixed curvature radius. The surface itself supplies no light tower, since the low-lying spectrum of its Laplacian is in general set by the curvature scale rather than by the area [26 §4]. The central charge gives (ℓAdSMPl,5)3∝gS−1(\ell_{\mathrm{AdS}}M_{\mathrm{Pl},5})^3\propto g_S-1, with ℓAdS\ell_{\mathrm{AdS}} the AdS radius and MPl,5M_{\mathrm{Pl},5} the five-dimensional Planck mass. The KK tower associated with the S4S^4 fiber therefore becomes light in five-dimensional Planck units as (gS−1)−1/3(g_S-1)^{-1/3}, while its mass scale in AdS units stays finite. This supplies the light tower expected from the AdS distance conjecture [39]. In the Type IIB classes covered by our theorems, the disk restriction excludes this large-genus limit altogether.

On the disk one can instead increase the number LL of boundary poles. Treating the charges as continuous, consider the rotationally symmetric family with poles at pℓ=e2πiℓ/Lp_\ell=e^{2\pi i\ell/L} in the unit-disk coordinate ζ\zeta and residues Zℓ=q e−2πiℓ/L\mathcal{Z}_\ell=q\,e^{-2\pi i\ell/L} with q>0q>0, so the charge directions turn clockwise as the poles run counterclockwise. The stress-tensor coefficient CT∝∫Σκ2g d2wC_T\propto\int_\Sigma\kappa^2 g\,d^2w [13 (4.11)] grows as L4L^4 at fixed qq, but tends to a finite limit at fixed LqLq, the sum of the charge magnitudes. Adding poles at fixed total charge therefore does not make CTC_T grow without bound. For quantized charges, each external stack contains at least one five-brane, so the total brane count bounds LL. In class S\mathcal{S}, the genus can grow without bound at fixed M5-brane count.

At fixed genus, the complex structure moduli of the class S\mathcal{S} surface parametrize exactly marginal couplings of the four-dimensional theory [26 §4]. By contrast, five-dimensional SCFTs admit no supersymmetry-preserving exactly marginal deformations [40]. On the disk this is consistent with the expectation that regularity fixes the cross-ratios of the boundary poles once the five-brane charges are specified [18 §III.1]. An annulus whose modulus was fixed by the charges would be equally consistent with the absence of a conformal manifold. Our topology theorem excludes even such an isolated solution in the monodromy-free class.

On an unpunctured disk, a common rescaling of the charges at fixed pole positions multiplies κ2\kappa^2 and G\mathcal{G} by the same factor. The spin-two spectrum in AdS units is therefore unchanged [13 (3.12)]. As the charges are scaled up, ℓAdSMPl,6→∞\ell_{\mathrm{AdS}}M_{\mathrm{Pl},6} \to\infty and each fixed massive spin-two KK level becomes light in six-dimensional Planck units. In AdS units, varying the charges or the number of poles cannot close the spin-two gap. A mode of S2S^2 angular momentum J≥1J \ge1 obeys m2ℓAdS2≥3J(3J+5)m^2\ell_{\mathrm{AdS}}^2 \ge3J(3J+5), with equality for a universal mode at each JJ [13 (3.14), (3.17)]. For J=0J = 0 this bound gives only m2≥0m^2 \ge0, and the gap comes instead from the F(4)F(4) multiplet classification [41 Table 22]: a nonconserved spin-two conformal primary has Δ≥6\Delta\ge6, hence m2ℓAdS2=Δ(Δ−5)≥6m^2\ell_{\mathrm{AdS}}^2 = \Delta(\Delta- 5) \ge6. Thus the lightest massive spin-two mode on every regular unpunctured disk satisfies 6≤mgap,22ℓAdS2≤246 \le m^2_{\mathrm{gap},2}\ell_{\mathrm{AdS}}^2 \le24, where the upper bound is the universal mode at J=1J = 1. The gap is held at the AdS scale. The upper bound excludes conventional scale separation, in which the gap grows without bound in AdS units, proposed for instance in [42]. The lower bound excludes the inverse scale separation found, e.g., in AdS4×S3×S3\mathrm{AdS}_4 \times S^3 \times S^3 solutions of the non-supersymmetric O(16)×O(16)O(16) \times O(16) heterotic string. There, for unequal fluxes, one S3S^3 becomes parametrically larger than the AdS radius [43], so that the gap set by its radius closes in AdS units.

In the AdS5\mathrm{AdS}_5 duals of class S\mathcal{S} theories, on the other hand, the spin-two gap is not held at the AdS scale. The difference does not come from the supergravity bounds. These duals are solutions of eleven-dimensional supergravity in which the SU(2)R×U(1)rSU(2)_R \times U(1)_r R-symmetry acts as the isometry of an S2S^2 and a circle, and Chen, Gutperle and Uhlemann found a lower bound on the spin-two masses that is fixed by the two R-charges and saturated by universal modes [44 (2.40)]. For R-symmetry singlets it reduces to m2≥0m^2 \ge0, just as the AdS6\mathrm{AdS}_6 bound 3J(3J+5)3J(3J + 5) does at J=0J = 0. In neither case does the supergravity bound say how light a massive singlet can be. That is decided by the superconformal algebra and by the internal geometry.

Consider the simplest class S\mathcal{S} case, M5-branes wrapped on a compact hyperbolic surface. The spin-two masses follow from a separation of variables [45 (22)]. Modes that are constant on the S4S^4 fiber are R-symmetry singlets and have m2ℓAdS2=2λSm^2\ell_{\mathrm{AdS}}^2 = 2\lambda_S, with λS\lambda_S an eigenvalue of the Laplacian of the surface in the metric of curvature −1-1. The lightest massive mode of this kind is set by the first nonzero eigenvalue. As noted above, this eigenvalue is in general set by the curvature scale. It depends on the complex structure, however, and becomes parametrically small when closed geodesics that separate the surface are pinched [46]. In the four-dimensional theory, this is the limit in which the theory separates into decoupled sectors, each with its own stress tensor. The four-dimensional N=2\mathcal{N}=2 algebra allows the gap to close, since a nonconserved spin-two operator only needs Δ>4\Delta> 4, that is m2>0m^2 > 0, whereas the F(4)F(4) algebra requires Δ≥6\Delta\ge6 [41].

Taken together, these comparisons point to a rigidity of the AdS6\mathrm{AdS}_{6} solutions dual to five-dimensional SCFTs, in contrast with the AdS5\mathrm{AdS}_{5} duals of class SS theories. There the genus of the surface can grow without bound, and its complex structure spans a conformal manifold along which the spin-two gap can close. In the monodromy-free AdS6\mathrm{AdS}_{6} class, regularity fixes the topology, the dual theories admit no exactly marginal deformations, and the F(4)F(4) superconformal symmetry holds the spin-two gap at the AdS scale. The external five-brane charges are then expected to be the only data left to vary. It would be interesting to learn whether smeared sources relax any part of this rigidity.

Acknowledgments: The author thanks Hyojoong Kim and Minwoo Suh for discussions and for an earlier collaboration on AdS6\mathrm{AdS}_{6} solutions of Type IIB supergravity, and Se-Jin Kim for suggesting to use Lean to check the proofs. This work is supported by the National Research Foundation of Korea under grants RS-2025-00518906, RS-2025-25441170, and RS-2025-25457100.

AI statement: The author used Claude Code (Anthropic) and Codex (OpenAI) to help check derivations, search the literature, prepare figures and revise the manuscript. Parts of the argument were formalized in Lean 4 with Mathlib, with the code written by Claude Code: the counting steps of Theorems 4.1, 5.6 and 5.8, the pointwise identities, the non-vanishing of λ′\lambda' at boundary points, and the rotation bound of Theorem 3.2. The remaining analytic inputs, including Hopf’s lemma for G\mathcal{G} and the degree formula on the double, enter there as hypotheses. The Lean files are included as ancillary files. The author verified the results and proofs and takes responsibility for the final text.

References

  1. [1]O. Aharony and A. Hanany, Branes, superpotentials and superconformal fixed points, Nucl. Phys. B 504 (1997) 239–271, [hep-th/9704170].DOI
  2. [2]O. Aharony, A. Hanany, and B. Kol, Webs of (p,q) five-branes, five-dimensional field theories and grid diagrams, JHEP 01 (1998) 002, [hep-th/9710116].
  3. [3]O. DeWolfe, A. Hanany, A. Iqbal, and E. Katz, Five-branes, seven-branes and five-dimensional E(n) field theories, JHEP 03 (1999) 006, [hep-th/9902179].DOI
  4. [4]F. Apruzzi, M. Fazzi, A. Passias, D. Rosa, and A. Tomasiello, AdS₆ solutions of type II supergravity, JHEP 11 (2014) 099, [arXiv:1406.0852].DOI
  5. [5]H. Kim, N. Kim, and M. Suh, Supersymmetric AdS₆ Solutions of Type IIB Supergravity, Eur. Phys. J. C 75 (2015), no. 10 484, [arXiv:1506.05480].DOI
  6. [6]E. D’Hoker, M. Gutperle, A. Karch, and C. F. Uhlemann, Warped AdS₆ × S² in Type IIB supergravity I: Local solutions, JHEP 08 (2016) 046, [arXiv:1606.01254].
  7. [7]E. D’Hoker, M. Gutperle, and C. F. Uhlemann, Exact solutions to complex Type IIB supergravity for complex superalgebra F(4) and its real forms, JHEP 07 (2026) 148, [arXiv:2508.14959].DOI
  8. [8]S. Komatsu, A. Martina, J. Pedendones, A. Vuignier, and X. Zhao, Einstein gravity from a matrix integral. Part I, JHEP 12 (2025) 029, [arXiv:2410.18173].arxiv.org/abs/2410.18173
  9. [9]E. D’Hoker, M. Gutperle, and C. F. Uhlemann, Holographic duals for five-dimensional superconformal quantum field theories, Phys. Rev. Lett. 118 (2017), no. 10 101601, [arXiv:1611.09411].DOI
  10. [10]E. D’Hoker, M. Gutperle, and C. F. Uhlemann, Warped AdS₆ × S² in Type IIB supergravity II: Global solutions and five-brane webs, JHEP 05 (2017) 131, [arXiv:1703.08186].DOI
  11. [11]E. D’Hoker, M. Gutperle, and C. F. Uhlemann, Warped AdS₆ × S² in Type IIB supergravity III: Global solutions with seven-branes, JHEP 11 (2017) 200, [arXiv:1706.00433].DOI
  12. [12]M. Gutperle, C. Marasinou, A. Trivella, and C. F. Uhlemann, Entanglement entropy vs. free energy in IIB supergravity duals for 5d SCFTs, JHEP 09 (2017) 125, [arXiv:1705.01561].DOI
  13. [13]M. Gutperle, C. F. Uhlemann, and O. Varela, Massive spin 2 excitations in AdS₆ × S² warped spacetimes, JHEP 07 (2018) 091, [arXiv:1805.11914].
  14. [14]O. Bergman, D. Rodríguez-Gómez, and C. F. Uhlemann, Testing AdS₆/CFT₅ in Type IIB with stringy operators, JHEP 08 (2018) 127, [arXiv:1806.07898].
  15. [15]C. F. Uhlemann, Wilson loops in 5d long quiver gauge theories, JHEP 09 (2020) 145, [arXiv:2006.01142].DOI
  16. [16]S. Adhikari, J. Hong, C. Joung, G. Lee, and S. Roychowdhury, Boundary-Value Problem in Type IIB Supergravity and Holography, arXiv:2609.03126.
  17. [17]M. Fluder and C. F. Uhlemann, Precision test of AdS₆/CFT₅ in Type IIB, Phys. Rev. Lett. 121 (2018) 171603, [arXiv:1806.08374].
  18. [18]M. Fluder and C. F. Uhlemann, Evidence for a 5d F-theorem, JHEP 02 (2021) 192, [arXiv:2011.00006].DOI
  19. [19]C. F. Uhlemann, Exact results for 5d SCFTs of long quiver type, JHEP 11 (2019) 072, [arXiv:1909.01369].DOI
  20. [20]M. Akhond, A. Legramandi, C. Nunez, L. Santilli, and L. Schepers, Massive flows in AdS₆/CFT₅, Phys. Lett. B 840 (2023) 137899, [arXiv:2211.09824].
  21. [21]M. Akhond, A. Legramandi, C. Nunez, L. Santilli, and L. Schepers, Matrix models and holography: mass deformations of long quiver theories in 5d and 3d, SciPost Phys. 15 (2023) 086, [arXiv:2211.13240].DOI
  22. [22]C. F. Uhlemann, AdS₆/CFT₅ with O7-planes, JHEP 04 (2020) 113, [arXiv:1912.09716].
  23. [23]M. Gutperle and C. F. Uhlemann, Surface defects in holographic 5d SCFTs, JHEP 04 (2021) 134, [arXiv:2012.14547].DOI
  24. [24]L. Santilli and C. F. Uhlemann, 3d defects in 5d: RG flows and defect F-maximization, JHEP 06 (2023) 136, [arXiv:2305.01004].DOI
  25. [25]M. Gutperle and N. Klein, A Penrose limit for type IIB AdS₆ solutions, JHEP 07 (2021) 073, [arXiv:2105.10824].DOI
  26. [26]G. Fenati, M. Montero, and I. Valenzuela, Stress-Testing Swampland Bounds with Class S Theories, arXiv:2608.28333.
  27. [27]J. Kaidi and C. F. Uhlemann, M-theory curves from warped AdS₆ in Type IIB, JHEP 11 (2018) 175, [arXiv:1809.10162].DOI
  28. [28]Y. Lozano, N. T. Macpherson, and J. Montero, AdS₆ T-duals and type IIB AdS₆ × S² geometries with 7-branes, JHEP 01 (2019) 116, [arXiv:1810.08093].
  29. [29]F. Apruzzi, O. Bergman, H.-C. Kim, and C. F. Uhlemann, Generalized quotients and holographic duals for 5d S-fold SCFTs, JHEP 04 (2023) 027, [arXiv:2211.13243].DOI
  30. [30]M. Gutperle, A. Trivella, and C. F. Uhlemann, Type IIB 7-branes in warped AdS₆: partition functions, brane webs and probe limit, JHEP 04 (2018) 135, [arXiv:1802.07274].DOI
  31. [31]A. Chaney and C. F. Uhlemann, On minimal Type IIB AdS₆ solutions with commuting γ-branes, JHEP 12 (2018) 110, [arXiv:1810.10592].
  32. [34]F. Apruzzi, F. Bonetti, I. García Etxebarria, S. S. Hosseini, and S. Schäfer-Nameki, Symmetry TFTs from String Theory, Commun. Math. Phys. 402 (2023) 895–949, [arXiv:2112.02092].DOI
  33. [35]I. Bah, F. Bonetti, R. Minasian, and E. Nardoni, Holographic Duals of Arg yres-Douglas Theories, Phys. Rev. Lett. 127 (2021) 211601, [arXiv:2105.11567].DOI
  34. [36]I. Bah, F. Bonetti, R. Minasian, and E. Nardoni, M5-brane Sources, Holography, and Argyres-Douglas Theories, JHEP 11 (2021) 140, [arXiv:2106.01322].DOI
  35. [37]C. Couzens, H. Kim, N. Kim, and Y. Lee, Holographic duals of M5-branes on an irregularly punctured sphere, JHEP 07 (2022) 102, [arXiv:2204.13537].DOI
  36. [38]H. Kim and N. Kim, Comments on the symmetry of AdS₆ solutions in string/M-theory and Killing spinor equations, Phys. Lett. B 760 (2016) 780–787, [arXiv:1604.07987].DOI
  37. [39]D. Lüst, E. Palti, and C. Vafa, AdS and the Swampland, Phys. Lett. B 797 (2019) 134867, [arXiv:1906.05225].DOI
  38. [40]C. Córdova, T. T. Dumitrescu, and K. Intriligator, Deformations of superconformal theories, JHEP 11 (2016) 135, [arXiv:1602.01217].DOI
  39. [41]C. Córdova, T. T. Dumitrescu, and K. Intriligator, Multiplets of Superconformal Symmetry in Diverse Dimensions, JHEP 03 (2019) 163, [arXiv:1612.00809].arxiv.org/abs/1612.00809
  40. [42]O. DeWolfe, A. Giryavets, S. Kachru, and W. Taylor, Type IIA moduli stabilization, JHEP 07 (2005) 066, [hep-th/0505160].DOI
  41. [43]I. Basile, D. Robbins, and H. Saleem, Non-supersymmetric heterotic strings on AdS₄ × S³ × S³, SciPost Phys. 21 (2026) 028, [arXiv:2604.18692].
  42. [44]K. Chen, M. Gutperle, and C. F. Uhlemann, Spin 2 operators in holographic 4d 𝒩 = 2 SCFTs, JHEP 06 (2019) 139, [arXiv:1903.07109].DOI
  43. [45]R. Bhattacharya, A. Katyal, and O. Varela, Class S superconformal indices from maximal supergravity, Phys. Rev. Lett. 134 (2025) 181601, [arXiv:2411.16837].arxiv.org/abs/2411.16837
  44. [46]R. Schoen, S. Wolpert, and S.-T. Yau, Geometric bounds on the low eigenvalues of a compact surface, in Geometry of the Laplace operator (Univ. Hawaii, Honolulu, 1979), vol. XXXVI of Proc. Sympos. Pure Math., pp. 279–285. Amer. Math. Soc., Providence, RI, 1980.DOI

Paper details

vlnc:W_849tgatg
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 3

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Claude Code, Codex

View statement
Looking for related papers…
Contents