Surface Defects and Codimension-Two Monodromy

In four dimensions, a surface defect is a codimension-two insertion. Its definition requires more than an embedded surface or a holonomy label: one must specify the normal orientation, the meridional boundary condition, any fields and action localized on the support, allowed counterterms, endpoint and junction data, and the class of support deformations under which the operator is claimed to be invariant. In general dimension the support has dimension $d-2$ and need not literally be a surface.

Monodromy is therefore a useful transverse label, not a complete quantum operator. It can remain fixed while the defect retains metric, shape, renormalization-scale, or localized-field dependence. A defect is topological only after correlation functions are shown to be invariant under the declared deformations; conformal covariance or a quantized label does not establish that stronger property.

Required background. Support, Codimension, and Operator Data supplies embeddings, transverse links, normal-first orientation, endpoints, junctions, and the distinction between geometric and topological defects.

Helpful background. Disorder Operators and Singular Boundary Conditions supplies the domain-changing path-integral definition, meridional monodromy, and defect-local counterterms. Local Potentials and Global Gauge Configurations supplies patching, holonomy, compact-flux normalization, and global-group constraints.

Monodromy lives on the normal circle bundle

Let

$$\iota:\Sigma^{d-2}\hookrightarrow M^d$$

be a smooth interior embedding in an oriented Euclidean spacetime. Give $\Sigma$ an orientation and use the normal-first convention

$$o(TM)|_\Sigma=o(N\Sigma)\wedge o(T\Sigma).$$

This orients the rank-two normal plane and hence the positive meridian $\ell_{x,\epsilon}\simeq S^1$ about each $x\in\Sigma$. If $\mathcal N_\epsilon(\Sigma)$ is a small tubular neighborhood, its boundary

$$Y_\epsilon=\partial\mathcal N_\epsilon(\Sigma)=S(N\Sigma)$$

is the normal circle bundle. It need not be the product $\Sigma\times S^1$. A polar angle $\vartheta$ is therefore only a local coordinate unless the normal bundle has been trivialized; specifying a framing is stronger than merely orienting the normal plane.

For a compact connected gauge group $G$, use the inherited coupling-absorbed Hermitian connection $\mathfrak a$, transforming as

$$\mathfrak a^h =h\mathfrak a h^{-1}-i(\mathrm dh)h^{-1}.$$

The gauge-invariant transverse datum is a prescribed conjugacy class of small-meridian holonomy,

$$M_x =\lim_{\epsilon\to0} \operatorname{Pexp}\!\left(i\oint_{\ell_{x,\epsilon}}\mathfrak a\right), \qquad [M_x]\subset G.$$

For the standard defect with a fixed label, this conjugacy class is locally constant along a connected smooth stratum. After choosing a maximal-torus representative in a local normal trivialization, one may write

$$\mathfrak a\sim\alpha\,\mathrm d\vartheta+O(1), \qquad M=e^{2\pi i\alpha}.$$

The parameter is identified by the affine Weyl action

$$\alpha\sim w(\alpha)+B, \qquad w\in W, \qquad B\in\Lambda_{\rm cochar}(G), \qquad e^{2\pi iB}=\mathbf1_G.$$

Thus the actual global group, not only its Lie algebra, fixes the allowed shifts and bundle sectors. A pure holonomy condition preserves the centralizer $H=Z_G(M)$ along the defect. A stronger singular representative can preserve a smaller subgroup, so $Z_G(M)$ should not automatically be identified with the stabilizer of every additional singular field.

Reversing the orientation of $\Sigma$ at fixed ambient orientation reverses the normal orientation and the positive meridian:

$$[M]\longmapsto[M^{-1}], \qquad \alpha\longmapsto-\alpha.$$

The two labels need not lie in the same conjugacy or affine-Weyl orbit. Gukov and Witten derive the local singularity, Weyl and cocharacter identifications, and the global normal-bundle restrictions in a specific twisted supersymmetric gauge theory Gukov and Witten 2008, § 2.1, arXiv v2, pp. 4–8, eqs. (2.2)–(2.10), and § 3.5, p. 58, eqs. (3.58)–(3.59), Open PDF. Their anti-Hermitian variables $A_{\rm src}=-i\mathfrak a$ and $\alpha_{\rm src}=-i\alpha$ give $e^{-2\pi\alpha_{\rm src}}=e^{2\pi i\alpha}$ in the convention used here. Their supersymmetric completion is an example, not a universal definition of surface defects.

A singular boundary condition needs quantum completion

Excise the tubular neighborhood and set

$$M_\epsilon=M\setminus\mathcal N_\epsilon(\Sigma).$$

Let $\mathfrak F_{[M]}(M_\epsilon)$ be the bulk fields whose restriction to $Y_\epsilon$ obeys the prescribed holonomy and bundle conditions. If $\varphi$ denotes fields localized on $\Sigma$, a schematic renormalized insertion is

$$\begin{aligned} \left\langle \mathcal S^{\rm ren}(\Sigma;\mu)\,\mathcal X \right\rangle ={}& \lim_{\epsilon\to0}\frac{1}{Z_0} \int_{\mathfrak F_{[M]}(M_\epsilon)} \mathcal D\Phi\,\mathcal D\varphi \\ &\times \exp\!\left[ -S_{\rm bulk}[\Phi] -S_\Sigma[\varphi,\Phi_\Sigma^{\rm reg}] -S_{\rm ct}(\epsilon,\mu) \right]\mathcal X. \end{aligned}$$

Here $Z_0$ is the vacuum normalization without the defect, $\mathcal X$ denotes other insertions away from the regulator, and $\mu$ is the renormalization scale. The symbol $\Phi_\Sigma^{\rm reg}$ denotes the regular or renormalized limiting data supplied to the localized theory; only smooth background fields have a literal pullback $\iota^*\Phi$. The formula fixes the displayed sign convention for $S_{\rm ct}$; it is not a claim that every defect admits the same regulator or counterterms.

The localized theory must couple consistently to the residual $H$-bundle and to the pullbacks of the bulk background fields. Its gauge and gravitational anomalies must cancel internally or through declared inflow. Allowed local terms can include a surface tension, intrinsic-curvature, extrinsic-curvature, and normal-bundle terms, couplings to restricted bulk fields, as organized for smooth defects in Billò et al. 2016, § 5.1, arXiv v2, pp. 27–28, especially eqs. (5.1)–(5.9), Open PDF. When endpoints or junctions are part of the support, locality permits additional terms on those lower strata. Finite choices can distinguish operator definitions rather than merely change notation.

The two broad constructions—prescribing singular transverse data and coupling a lower-dimensional theory to the bulk—are reviewed in Gukov, arXiv:1412.7127v1, §§ 1.1–1.2, especially eqs. (1.8) and (1.10)–(1.14), Open PDF. That review focuses on supersymmetric gauge theories. In particular, its familiar $(\alpha,\beta,\gamma,\eta)$ package is model-specific and is not assumed on this page.

Independent data that can enter a codimension-two defect

Construction	Defining datum	Deformation ceiling
Monodromy condition	Holonomy conjugacy class around a positive meridian	The label can be transported locally; topology is not established
Singular boundary condition	Asymptotic fields, global bundle sector, and regulator prescription	The answer can depend on its quantum completion
Defect-local QFT	Localized fields, action, couplings, and anomaly cancellation	Usually retains metric and shape dependence
Topological coupling	A quantized exponentiated integral on a closed support	Deformation invariance needs source and boundary checks
Conformal defect	A preserved conformal subgroup and defect operator data	Conformal covariance is not arbitrary isotopy invariance
Topological defect	Correlation functions invariant under the declared deformations	The property must be proved, not inferred from the label

Compact U(1) separates holonomy from curvature

Consider a closed oriented surface $\Sigma$ in an oriented Euclidean four-manifold, and first work in a trivialized tubular patch. Write the faithfully normalized compact connection as

$$a:=gA_{U(1)}, \qquad a\longmapsto a+\mathrm d\lambda, \qquad \lambda\sim\lambda+2\pi, \qquad f=\mathrm da\quad\text{locally}.$$

The monodromy defect $\mathcal S_\alpha(\Sigma)$ imposes

$$a=\alpha\,\mathrm d\vartheta+a_{\rm reg}, \qquad \operatorname{Hol}_\ell(a) =\exp\!\left(i\oint_\ell a\right) =e^{2\pi i\alpha}, \qquad \alpha\in\mathbb R/\mathbb Z.$$

A large gauge transformation $e^{ik\vartheta}$ shifts $\alpha\mapsto\alpha+k$. A charge-$n$ probe is insensitive to this shift because $n\in\mathbb Z$. For a disjoint oriented line $C$ and surface $\Sigma$ in a region where the integer linking number is defined,

$$\begin{aligned} W_n(C)[a_{\rm sing}+a_{\rm reg}] ={}& \exp\!\left( 2\pi i n\alpha\,\operatorname{Lk}(C,\Sigma) \right) \\ &\times W_n(C)[a_{\rm reg}], \qquad W_n(C)=\exp\!\left(i n\oint_C a\right). \end{aligned}$$

This is a background or integrand identity for the declared singular field, not a universal factorization theorem for arbitrary interacting correlators. It checks the transverse monodromy label but says nothing by itself about the localized action or the shape dependence of the complete defect.

Define the Poincaré-dual current by

$$\int_M\delta_\Sigma\wedge\omega=\int_\Sigma\omega$$

for compactly supported test two-forms $\omega$. With the chosen orientation, choose a real lift $\widetilde\alpha\in\mathbb R$ of the holonomy class $\alpha\in\mathbb R/\mathbb Z$. The local distributional shorthand is then

$$f=2\pi\widetilde\alpha\,\delta_\Sigma+f_{\rm reg},$$

because integrating over a positively oriented transverse disk gives $\int_{D^2}f=\oint_{\partial D^2}a=2\pi\widetilde\alpha$. Changing the lift by an integer changes the chosen distributional extension across $\Sigma$ but not the holonomy on the excised domain. This is why excision and holonomy, not the delta-current formula, define the operator. One must not square the delta-current in a Maxwell action. The form $\widetilde\alpha\,\mathrm d\vartheta$ is flat on the punctured disk, so there is no universal classical Maxwell power divergence associated with this pure holonomy term; core and quantum counterterms remain model-dependent. The local $\alpha\,\mathrm d\vartheta$ representative, delta-current shorthand, integral shifts, and normal-bundle extension caveats are developed in Gukov and Witten 2008, § 2.1, arXiv v2, pp. 5–8, especially eqs. (2.2)–(2.5) and (2.8)–(2.10), Open PDF, with the convention translation given above.

There is also an independent topological coupling. Suppose the regular residual $U(1)$ bundle restricts smoothly to the closed surface and set

$$\begin{aligned} m_\Sigma &=\frac{1}{2\pi}\int_\Sigma f_\Sigma\in\mathbb Z, \qquad \eta\in\mathbb R/\mathbb Z, \\ V_\eta(\Sigma) &=\exp\!\left(2\pi i\eta m_\Sigma\right) =\exp\!\left(i\eta\int_\Sigma f_\Sigma\right). \end{aligned}$$

Here $f_\Sigma=\iota^*f_{\rm reg}$ is the tangential residual curvature after the prescribed transverse singularity has been removed; the raw pullback of $\delta_\Sigma$ is not defined. If $\Sigma'-\Sigma=\partial B$ and the regular curvature extends smoothly across the interpolating three-chain,

$$\frac{V_\eta(\Sigma')}{V_\eta(\Sigma)} =\exp\!\left(i\eta\int_B\mathrm df_{\rm reg}\right).$$

The factor is deformation invariant only when no magnetic source crosses $B$, the supports are closed, and the boundary and global bundle data permit the interpolation. A metric-independent factor does not make the full monodromy defect topological. With the label convention above,

$$\mathcal S_{\alpha,\eta}(\overline\Sigma) =\mathcal S_{-\alpha,-\eta}(\Sigma).$$

The character-valued surface parameter, its normalization, and its closed-support qualification appear in Gukov and Witten 2008, §§ 2.3–2.4, arXiv v2, pp. 11–12 and 18, especially eqs. (2.13)–(2.14) and (2.33), with footnote 11, Open PDF. The regularized separation between transverse singular curvature and tangential residual curvature is essential outside that model-specific setting.

Surface junctions remember ordered meridians

Suppose codimension-two sheets meet along a codimension-three stratum $K$, which is a line in four dimensions. A small oriented $S^2$ linking $K$ is punctured by the incident sheets. Choose a base point and a distinguished ordered system of based meridian generators $\gamma_i$—equivalently, a cut or arc system—for the punctured sphere. With incidence signs $\epsilon_i=\pm1$, choose the generators so their defining ordered relation is $\gamma_1^{\epsilon_1}\cdots\gamma_r^{\epsilon_r}=1$, and set $M_i=\operatorname{Hol}(\gamma_i)$. An ordinary flat junction must obey

$$M_1^{\epsilon_1}M_2^{\epsilon_2}\cdots M_r^{\epsilon_r} =\mathbf1_G.$$

There is no intrinsic cyclic ordering of arbitrary punctures on $S^2$. Changing the distinguished generators acts by conjugations and Hurwitz moves. Conjugacy classes therefore cannot simply be multiplied: representatives, relative conjugations, base paths, and a junction operator or intertwiner are extra data. In the Abelian case the relation reduces to

$$\sum_i\epsilon_i\alpha_i=0\pmod{\mathbb Z}.$$

For two incoming sheets and one outgoing sheet this is $\alpha+\beta-\gamma=0$ modulo integers. It is a necessary incidence condition, not a proof that the junction exists, has a chosen normalization, or satisfies coherent associativity. At a spacetime boundary the meridians can become relative cycles, so explicit absorption and boundary conditions replace the closed-link argument.

A finite one-form symmetry gives a topological surface network

The preceding constructions describe generic surface defects. A special, clean topological subclass arises in an oriented four-dimensional QFT with an exact, non-anomalous, invertible group-like $\mathbb Z_N$ one-form symmetry. Let $U_a(\Sigma)$, $a\in\mathbb Z_N$, be a closed oriented symmetry surface, and let $L_r(C)$ be a closed line of one-form charge $r\in\mathbb Z_N$. For disjoint supports in a region where integer linking is defined, with all other insertions $\mathcal X$ outside the swept region,

$$\begin{aligned} &\left\langle U_a(\Sigma)L_r(C)\mathcal X \right\rangle \\ &\qquad= \exp\!\left[ \frac{2\pi i}{N}\,a r\, \operatorname{Lk}(\Sigma,C) \right] \left\langle L_r(C)\mathcal X\right\rangle. \end{aligned}$$

Fix the convention that a positively oriented unit link between $U_1$ and $L_1$ gives $e^{2\pi i/N}$. Orientation and fusion then give

$$\begin{aligned} U_a(\overline\Sigma)&=U_{-a}(\Sigma), \qquad L_r(\overline C)=L_{-r}(C), \\ U_a\otimes U_b&\simeq U_{a+b\;\mathrm{mod}\;N}. \end{aligned}$$

If two incoming sheets and one outgoing sheet meet along an oriented line $K$, their incidence data are

$$J_{a,b}^{\ c}(K): U_a\otimes U_b\longrightarrow U_c, \qquad a+b-c=0\pmod N.$$

The congruence is necessary but does not construct, normalize, or prove the coherence of the junction. A nonzero residual label requires separately declared line or junction data. An anomaly need not erase every topological surface, but it can twist the junction data and obstruct the untwisted group-like network used here.

For example, take $N=6$, $r=2$, $a=4$, and a positive unit link. The phase is

$$\exp\!\left(\frac{2\pi i}{6}\,4\cdot2\right) =\exp\!\left(\frac{2\pi i}{3}\right).$$

The proposed junction $U_4\otimes U_5\to U_3$ passes the incidence test because $4+5-3=6\equiv0\pmod6$.

The group-like surface action and linking relation are given in Gaiotto, Kapustin, Seiberg, and Willett 2015, § 3, arXiv v2, pp. 11–13, eqs. (3.1)–(3.4), Open PDF and Bhardwaj et al. 2024, § 2.2.2, arXiv v2, pp. 18–20, especially eqs. (2.63) and (2.69)–(2.70), Open PDF. The trivalent incidence law is the additive $\mathbb Z_N$ specialization of the finite-group network law with the outgoing orientation reversed Gaiotto, Kapustin, Seiberg, and Willett 2015, § 2, arXiv v2, pp. 6–8, especially eq. (2.2) and the junction paragraph preceding eq. (2.6), Open PDF.

Topological and conformal are different qualifications

A topological defect has correlation functions invariant under a declared class of isotopies that avoid other insertions, boundaries, and forbidden singular configurations. A conformal defect instead preserves a conformal subgroup and carries defect-local operator data. It can remain sensitive to its shape.

For a flat defect, choose local coordinates $x=(x_\parallel,x_\perp)$ and define the sign of the displacement operator $\mathcal D^i$ by the transverse Ward identity

$$\partial_\mu T^{\mu i}(x) =\delta^{(2)}(x_\perp)\, \mathcal D^i(x_\parallel)+\cdots.$$

Here $\delta^{(2)}(x_\perp)$ is a scalar normal delta density, distinct from the two-form current $\delta_\Sigma$ used above. The displacement insertion measures the response to changing the embedding. A nontrivial $\mathcal D^i$ is compatible with conformal symmetry and shows that the defect is not invariant under arbitrary shape deformations. Topological behavior requires the displacement response to vanish or become trivial in the declared separated correlators, together with the relevant anomaly, boundary, framing, and junction checks. The conformal-defect Ward identities and displacement operator are developed in Billò et al. 2016, § 5.1, arXiv v2, pp. 27–31, eqs. (5.1)–(5.25), Open PDF.

Detailed conformal data belong to Conformal Boundaries and Defects and The Displacement Operator and Defect Ward Identities. Supersymmetric completions belong to BPS Boundaries, Surface Defects, Interfaces, and Fusion, and brane realizations belong to Extended Operators, Defects, and Brane Charges.

What the local surface data do not determine

A complete surface-defect specification follows this order:

1. state the ambient theory, actual global group, allowed bundles, and tangential and boundary structures;
2. give the embedded or stratified support, normal orientation, and any framing;
3. prescribe the meridional holonomy or stronger singular boundary condition, including its global extension;
4. supply the residual bundle, localized fields, action, anomalies, and counterterms;
5. declare endpoints, attached operators, junctions, and their incidence and coherence data; and
6. state which deformations are allowed and whether geometric, conformal, or topological dependence has actually been established.

Local monodromy does not classify global normal-bundle extensions, defect CFT spectra, supersymmetric completions, brane constructions, or coherent fusion categories. Surface networks and their lower strata continue in Fusion, Junctions, and Endpoints, while the geometry of their phases continues in Linking, Braiding, and Framing. The symmetry interpretation of the finite network belongs to Higher-Form Symmetry from Operators and Linking.

The next page, Boundaries, Interfaces, and Domain Walls, changes the support geometry: an interface is locally two-sided, whereas a physical boundary is one-sided and changes the field domain.

Common pitfalls

A holonomy label is the whole defect. It specifies one transverse boundary condition. Localized fields, global extension, counterterms, anomalies, and junction data can change the quantum operator without changing that holonomy.

A local polar angle defines the defect globally. The normal circle bundle can be twisted. Use fiberwise meridians and compatible patch data unless a normal framing has actually been chosen.

A quantized or metric-independent label makes the defect topological. Quantization can make one coupling well defined while other parts remain shape- or scale-dependent. Topological status requires a deformation test for the complete insertion.

Conjugacy classes multiply at a junction. A non-Abelian junction also needs representatives, ordered meridians, relative conjugations, and a junction operator. The product relation is only an incidence condition.

A conformal defect is topological. A conformal defect can have a nonzero displacement operator and nontrivial shape response. These are different qualifications.

Check your understanding

1. Reverse the orientation of a compact-$U(1)$ surface defect labeled by $(\alpha,\eta)$. What happens to its meridional holonomy and topological factor?
2. For $\alpha=2/5$, a charge-$3$ Wilson line, and linking number two, compute the monodromy phase.
3. Let $\Sigma'-\Sigma=\partial B$. Derive the change of $V_\eta$ and state what happens when a magnetic worldline of charge $m$ crosses $B$.
4. In a $\mathbb Z_6$ network, test the proposed junction $U_4\otimes U_5\to U_3$. What does the arithmetic leave unproved?
5. Explain why a planar conformal surface defect with a nonzero displacement operator is not topological.

Solutions

Orientation reversal reverses the positive meridian, so $\alpha\mapsto-\alpha$ and the holonomy is inverted. It also reverses the surface integral, so $\eta\mapsto-\eta$ in the fixed-orientation label convention:

$$\mathcal S_{\alpha,\eta}(\overline\Sigma) =\mathcal S_{-\alpha,-\eta}(\Sigma).$$

The Wilson-line phase is

$$\exp\!\left(2\pi i\cdot3\cdot\frac25\cdot2\right) =\exp\!\left(\frac{4\pi i}{5}\right).$$

Stokes’ theorem gives

$$\frac{V_\eta(\Sigma')}{V_\eta(\Sigma)} =\exp\!\left(i\eta\int_B\mathrm df_{\rm reg}\right).$$

It is one when the Bianchi identity holds on $B$. If a magnetic worldline of charge $m$ crosses the chain with positive incidence, then $\int_B\mathrm df=2\pi m$ and the factor is $e^{2\pi i\eta m}$; the deformation has crossed another operator and is not an allowed topological isotopy of the separated configuration.

The junction passes the incidence test because $4+5-3=6\equiv0\pmod6$. This does not construct a junction field, fix its normalization, or prove associativity and anomaly-free coherence.

The displacement operator measures the variation under changing the embedding. A nonzero insertion therefore gives nontrivial shape response, even though the planar defect can preserve a conformal subgroup.

References

• Bhardwaj, Lakshya, Lea E. Bottini, Ludovic Fraser-Taliente, Liam Gladden, Dewi S. W. Gould, Arthur Platschorre, and Hannah Tillim. “Lectures on Generalized Symmetries.” Physics Reports 1051 (2024): 1–87. DOI. Open PDF, arXiv v2.
• Billò, Marco, Vasco Gonçalves, Edoardo Lauria, and Marco Meineri. “Defects in Conformal Field Theory.” Journal of High Energy Physics 2016, no. 4 (2016): 091. DOI. Open PDF, arXiv v2.
• Gaiotto, Davide, Anton Kapustin, Nathan Seiberg, and Brian Willett. “Generalized Global Symmetries.” Journal of High Energy Physics 2015, no. 2 (2015): 172. DOI. Open PDF, arXiv v2.
• Gukov, Sergei. “Surface Operators.” In New Dualities of Supersymmetric Gauge Theories, edited by Jörg Teschner, 223–259. Cham: Springer, 2016. DOI. Open PDF, arXiv v1.
• Gukov, Sergei, and Edward Witten. “Gauge Theory, Ramification, and the Geometric Langlands Program.” Current Developments in Mathematics 2006 (2008): 35–180. DOI. Open PDF, arXiv v2.