Boundaries, Interfaces, and Domain Walls

A codimension-one coupling is consistent only after its geometry, orientation, bulk theories, admitted fields and variations, localized action and degrees of freedom, matching conditions, operator endpoints, counterterms, and allowed deformations have all been specified. Its total first variation must have no uncontrolled surface term, and every symmetry claimed to survive must preserve the field domain and satisfy the wall flux and anomaly balance. These tests are necessary for a local quantum interface, but they do not by themselves make it transparent, invertible, conformal, or topological.

Required background. Support, Codimension, and Operator Data supplies the distinction between a one-sided boundary and a two-sided interface, together with orientation, attachment, and renormalization data. Boundaries, Flux, and Boundary Ward Identities supplies outward-normal flux balance, localized wall currents, and the test that a transformation preserves fixed boundary data.

Helpful background. Edge Modes, Subregions, and Factorization distinguishes a fiducial cut from a physical wall and explains why gluing is matching followed by reduction; an auxiliary edge frame is not automatically a dynamical wall field.

Boundary, interface, and wall are different codimension-one data

Let an oriented Lorentzian spacetime be divided by a smooth non-null interior hypersurface $W$. Locally,

$$M=M_-\cup_W M_+ .$$

Choose the unit coorientation $n$ to point from $M_-$ to $M_+$. It is the outward normal of the cut-open region $M_-$, while $-n$ is the outward normal of $M_+$. Thus the same interface term has opposite outward-normal signs on its two sides. Reversing the coorientation exchanges the ordered theories; it produces the orientation-reversed or adjoint wall, not automatically an inverse wall. A timelike physical wall has spacelike $n$; null supports need different measure and normal data and are outside the derivation below.

A physical boundary $B\subset\partial M$ instead has a one-sided collar $[0,\epsilon)\times B$ and only one bulk trace. A domain wall is more specific than either term: it is a field configuration or phase in which couplings or order parameters interpolate between vacua or phases. Its finite-thickness profile can have tension and localized modes. Replacing it by a sharp interface is an effective limit that must retain the induced wall action and counterterms. This dynamical, vacuum-interpolating use of “domain wall” is illustrated in Shifman 2022, § 2.1.1, pp. 41–44; it should not be silently extended to every prescribed interface.

Four codimension-one constructions require different defining data

Construction	Local geometry	Defining data and main test	Does not imply
Physical boundary	One-sided collar	Boundary condition or boundary theory; cancel the one bulk surface variation	Reflection, conformality, or a trivial boundary theory
Interface	Two-sided interior hypersurface with ordered sides	Matching map and wall action; cancel the two oriented surface variations	Transparency, invertibility, or topological fusion
Domain wall	Finite-thickness interpolation or its thin limit	Underlying profile, localized modes, and effective matching data	That every interface has a solitonic realization
Fiducial cut	Artificial division of one system	Matching and reduction must reconstruct the uncut theory	A material wall or independent boundary dynamics

The table separates geometry from operator type. A boundary may support a full quantum field theory, an interface may be reflective, and a symmetry wall may be topological; none of those conclusions follows from codimension alone.

A well-posed variation fixes the matching equations

For fields $\Phi_\pm$ in the two regions and wall fields $\varphi$, write

$$S_{\mathrm{tot}} =S_-[\Phi_-]+S_+[\Phi_+] +S_W[\varphi,i_-^*\Phi_-,i_+^*\Phi_+] .$$

After integrating the bulk variations by parts, the complete first variation has the schematic form

$$\begin{aligned} \delta S_{\mathrm{tot}} ={}&\int_{M_-}\mathcal E_-\,\delta\Phi_- +\int_{M_+}\mathcal E_+\,\delta\Phi_+ \\ &+\int_W\left( \Theta_-^{(+n)}+\Theta_+^{(-n)}+\delta\mathcal L_W \right), \end{aligned}$$

with additional lower-stratum terms when $W$ itself has a boundary or junction. The total integrated wall-and-corner variation must vanish for every variation tangent to the declared matching and source data, after any intended wall equations of motion are imposed. Locally, the wall density may vanish or be a tangential exact term $\mathrm d_W C$, provided its induced corner or lower-stratum contribution is retained and cancelled. This is the differentiability test developed in Boundaries, Variations, and Well-Posed Actions and in Harlow and Wu 2020, § 2.2, pp. 11–13, especially eqs. (2.15)–(2.20), and § 3.3, pp. 26–27, eqs. (3.18)–(3.24), Open PDF.

A scalar illustrates the orientation sign. Take

$$S_s=\int_{M_s}\!\sqrt{-g}\, \left(\frac{Z_s}{2}\,\partial_\mu\phi_s\partial^\mu\phi_s -V_s(\phi_s)\right), \qquad s=\pm ,$$

match the traces to a common field $\phi_W$, and first set $S_W=0$. The wall variation is

$$\int_W\!\sqrt{|h|}\, \left(Z_-n^\mu\partial_\mu\phi_- -Z_+n^\mu\partial_\mu\phi_+\right)\delta\phi_W .$$

Free $\delta\phi_W$ therefore makes the common-normal canonical momentum continuous. Adding $S_W=-\int_W\sqrt{|h|}\,U(\phi_W)$ changes the equation to

$$Z_-n^\mu\partial_\mu\phi_- -Z_+n^\mu\partial_\mu\phi_+ -U'(\phi_W)=0 .$$

On a one-sided boundary the same calculation has only $\Theta_{\mathrm{bulk}}^{(n_{\mathrm{out}})}+\delta\mathcal L_B$. Four notions must remain distinct: an imposed boundary condition restricts the field domain; a natural boundary equation follows from free variation; an external source is varied as controlled background data; and a dynamical boundary theory has its own fields and equations. Making the action differentiable also does not prove existence, uniqueness, stability, or causal well-posedness of the associated initial-boundary value problem.

Locality and symmetry impose independent checks

A wall interaction must be local along $W$ after its regulator is removed, with allowed wall-local counterterms and lower-stratum terms included. A preserved subgroup $H$ must preserve the two bulk couplings and backgrounds, the admitted fields and variations, the matching conditions, and the wall action. In the common-normal convention, the wall Ward identity has the form

$$D_Ak^A-n_\mu j_-^\mu+n_\mu j_+^\mu +\mathcal B_W+\mathcal A_W=0 .$$

Here $k^A$ is the wall current, $\mathcal B_W$ records explicit wall breaking or controlled source response, and $\mathcal A_W$ denotes a quantum anomaly term. If all three are absent, the normal bulk charge flux is continuous: $n\mathbin{\cdot}j_-=n\mathbin{\cdot}j_+$. This is charge balance, not proof that waves or operators cross without reflection.

There are two anomaly checks, in different dimensions. First, an absolute symmetry-preserving interface between $d$-dimensional theories folds to $\mathcal T_-\otimes\overline{\mathcal T_+}$. For perturbative anomalies, assume an even-dimensional unitary QFT and a unitary Lorentz-invariant boundary. For a finite group-cohomology anomaly, assume an unbroken subgroup $H$ with the compatible wall-network coherence used to define the background. In these proved domains, the folded anomaly difference must trivialize. Equivalently, after the declared symmetry identification and admissible local counterterms,

$$\left[\mathcal A_-\big|_H\right] =\left[\mathcal A_+\big|_H\right] .$$

These hypotheses and the folded-interface statement are given in Thorngren and Wang 2021, §§ 2.2–2.3, pp. 8–14; § 3.4, p. 26; and § 6, p. 37, Open PDF. Beyond those anomaly classes this page requires a separate test of the full anomaly theory; it does not assert a dimension-independent vanishing theorem.

The orientation-reversed factor contributes the difference $[\mathcal A_-]-[\mathcal A_+]$ in the folded description. An intrinsic $(d-1)$-dimensional wall theory cannot by itself erase a nontrivial $(d+1)$-dimensional anomaly theory. Second, when a $d$-dimensional response term jumps across phases, its localized inflow may cancel the separate $(d-1)$-dimensional anomaly of wall degrees of freedom:

$$[\mathcal A_W]+[\mathcal A_{\mathrm{inflow}}]=0 .$$

Gauge transformations kept as redundancies require complete cancellation. For a fixed global-symmetry background, a relative bulk–wall system may carry nonzero ‘t Hooft data, but anomaly compatibility remains necessary rather than a theorem that the wall exists. The local descent and inflow mechanism used in the second check is developed on Anomaly Polynomials and Inflow.

Tangential stress has an analogous balance: a jump in normal stress flux must be absorbed by wall stress and forces. Broken normal translations produce a displacement or shape response. Consequently, a differentiable and symmetry-compatible interface can still reflect energy.

Folding carries both theories to one boundary

Locally, a separating interface between $\mathcal T_-$ and $\mathcal T_+$ can be reflected across its collar and regarded as a boundary condition for

$$\mathcal T_-\otimes\overline{\mathcal T_+} .$$

The bar carries the reversed spacetime orientation and the corresponding conjugation of tangential, spin, background, and anomaly data. Both original bulk traces remain independent boundary data until the folded boundary condition relates them. This translation is useful for calculations, but it does not establish locality, unitarity, invertibility, topologicality, or a finite collision limit. Nonseparating walls and orientation-reversing, antiunitary, Spin, or Pin gluing require their own global treatment. Bachas and Brunner 2008, § 1, corrected arXiv v3, pp. 2–3, Open PDF gives the folding construction in the controlled setting of two-dimensional conformal interfaces.

A compact U(1) coupling wall reflects and transmits

Consider compact Maxwell theory on an oriented four-dimensional Lorentzian spacetime, divided by a planar wall $W$ at $z=0$. Let $n$ point from $M_-$ to $M_+$, take real positive couplings $e_\pm$, no theta term, no wall charge, and one faithfully normalized compact connection $a$ across the wall:

$$S=-\frac14\sum_{s=\pm}\frac1{e_s^2} \int_{M_s}\!\sqrt{-g}\,f_{s\,\mu\nu}f_s^{\mu\nu}, \qquad f_s=\mathrm da_s\ \text{locally} .$$

The compact normalization means $W_q(C)=\exp(i q\oint_C a)$ with $q\in\mathbb Z$ and $(2\pi)^{-1}\int_\Sigma f\in\mathbb Z$ for every closed oriented two-cycle $\Sigma$. Saying that $a$ is one connection means that its bundle transition data are matched across $W$; it does not mean that one gauge potential exists globally.

The wall part of the Maxwell variation is

$$\int_W\!\sqrt{|h|}\, \left( \frac1{e_+^2}n_\mu f_+^{\mu A} -\frac1{e_-^2}n_\mu f_-^{\mu A} \right)\delta a_A .$$

Free tangential variation therefore gives

$$\frac1{e_-^2}n_\mu f_-^{\mu A} =\frac1{e_+^2}n_\mu f_+^{\mu A} .$$

The common compact connection also gives $i_-^*f_-=i_+^*f_+$ when no magnetic surface source is present. These are transmission conditions, not automatic transparency. They are the direct coupling-wall analogue of the Abelian boundary and wall variations in Kapustin and Tikhonov 2009, §§ 2.1–2.2, pp. 2–9, especially eqs. (1), (2), (5), and (6), Open PDF; their Euclidean normalization is translated here to the site’s Lorentzian convention.

For a normally incident transverse mode with the same light speed on both sides, set

$$\begin{aligned} a_x^-&=e^{-i\omega t}\left(e^{ikz}+r\,e^{-ikz}\right), & z&<0,\\ a_x^+&=e^{-i\omega t}\,\tau\,e^{ikz}, & z&>0 . \end{aligned}$$

Trace and flux matching give

$$1+r=\tau, \qquad \frac{1-r}{e_-^2}=\frac{\tau}{e_+^2},$$

and hence

$$r_{-\to+}=\frac{e_+^2-e_-^2}{e_+^2+e_-^2}, \qquad \tau_{-\to+}=\frac{2e_+^2}{e_+^2+e_-^2} .$$

With energy flux normalized by the action,

$$R=|r|^2, \qquad T=\frac{e_-^2}{e_+^2}|\tau|^2, \qquad R+T=1 .$$

Thus the identity-coupling case $e_-=e_+$ is transparent, whereas an unequal coupling wall is generally reflective. Reversing the coorientation exchanges $e_-$ and $e_+$: the reflection amplitude changes sign, while the transmitted power is unchanged.

A closed Wilson line crosses this wall with the same integer label only because the same compact group and charge lattice were declared on both sides. An open segment ending on $W$ is not made gauge invariant by flux matching; it requires a wall operator of compensating charge or another explicit endpoint rule. With independent connections $a_\pm$, one must replace trace equality by a compact-group gluing map or transition field.

A finite symmetry wall carries a junction network

Now take an oriented three-dimensional QFT with an exact, non-anomalous, group-like zero-form symmetry $\mathbb Z_N$. Its cooriented topological symmetry walls are surfaces $U_\alpha(W)$, $\alpha\in\mathbb Z_N$, with

$$U_\alpha(\overline W)=U_{-\alpha}(W), \qquad U_\alpha\otimes U_\beta\simeq U_{\alpha+\beta\bmod N} .$$

Two incoming surfaces and one outgoing surface can meet on an oriented line junction

$$J_{\alpha,\beta}^{\ \gamma}(L): U_\alpha\otimes U_\beta\longrightarrow U_\gamma, \qquad \alpha+\beta-\gamma=0\pmod N .$$

The congruence is an incidence condition, not a construction or normalization of $J_{\alpha,\beta}^{\ \gamma}$ and not a proof of associativity. A consistent network also needs its lower-stratum junction operators and coherence data. This is the additive $\mathbb Z_N$ specialization of the finite-symmetry wall network in Gaiotto, Kapustin, Seiberg, and Willett 2015, § 2, pp. 5–8, especially eq. (2.2) and the junction discussion before eq. (2.6), Open PDF.

For example, in $\mathbb Z_6$ the incidence condition permits $U_4\otimes U_5\to U_3$ because $4+5-3=6$, but that arithmetic alone does not supply the line operator or its channel-changing point junctions. At a physical boundary, $U_\alpha$ may terminate only on declared boundary-line data. If the wall maps a boundary condition $b$ to a distinct condition $\alpha\mathbin{\cdot}b$, it relates two boundary theories rather than acting as an internal symmetry of $b$.

A vacuum domain wall should not be identified with this topological symmetry wall. It is generally tensionful and shape-sensitive even when a symmetry permutes the vacua that label its two sides.

Composition requires a controlled collision limit

Suppose $W_{12}:\mathcal T_1\to\mathcal T_2$ and $W_{23}:\mathcal T_2\to\mathcal T_3$ are separated by a strip of $\mathcal T_2$ of width $\ell$. A composite exists only after their common orientation and tangential structures, global background data, matching polarizations, symmetry and anomaly classes, junctions, and collision counterterms are compatible, and after a renormalized $\ell\to0$ limit has been shown to exist. Gapless strip modes or relevant wall couplings can make that limit singular or scheme dependent. Even conformal-interface fusion can be singular and generate a nontrivial RG flow, as emphasized in Bachas and Brunner 2008, § 1, corrected arXiv v3, p. 2, Open PDF.

The following qualifications are independent:

• Transparent means the relevant stress, symmetry, and operator data cross without reflection or loss in the declared sector.
• Invertible means there exists an oppositely directed wall whose two compositions give the appropriate identity walls, with all junction data included. An adjoint or orientation reverse is not automatically that inverse.
• Topological means correlators are invariant under allowed wall deformations that avoid other insertions and boundaries.
• Conformal means the wall preserves a conformal subgroup; it may still have nontrivial displacement data and reflection.

Formal juxtaposition therefore is not yet fusion, and an orientation-reversed wall is not automatically an inverse.

What the compatibility test does not classify

The tests above do not classify analytic boundary-value problems, boundary RG flows, solitonic wall profiles, conformal interface spectra, holographic realizations, relative topological boundaries, or categorical composition. The respective continuations are Kinks and Domain Walls, Conformal Boundaries and Defects, Interfaces, Folding, and Fusion, Holographic RG Flows and Domain-Wall Geometries, and Operators, Boundaries, and Relative Topological Theories.

For theorem-level boundary gauge theory and functorial formulations, see Perturbative Boundary Gauge Theory: Scope and Open Problems, Boundaries, Defects, and Extended Operators in TQFT, and Boundary Conditions and Interfaces in Functorial Field Theory. The immediate next step is Fusion, Junctions, and Endpoints, where collision limits and lower-stratum operators become the main objects.

Common pitfalls

Treating a physical boundary as an interface with nothing. A boundary is one-sided and changes the field domain. Replacing its missing side by a trivial theory is an additional construction, not its definition.

Checking only the bulk equations. Integration by parts can leave an uncancelled wall term even when both bulk equations hold. The complete first variation, including wall and corner terms, decides differentiability.

Calling flux continuity transparency. Continuity of charge or canonical momentum is a conservation statement. The Maxwell coupling wall conserves flux while reflecting whenever $e_-\ne e_+$.

Assuming a bulk symmetry survives the wall. The transformation must also preserve the matching domain, wall action, localized measure, sources, and anomaly data. A wall can preserve only a subgroup or map one boundary condition to another.

Using folding to discard orientation. Folding reverses one theory and therefore conjugates its anomaly and tangential data. It is a change of description, not a proof of consistency or invertibility.

Inferring a junction from charge conservation. The signed label sum is only a selection rule. The junction operator, its normalization, and its coherence under re-resolution remain independent data.

Check your understanding

1. Opposite normals and the scalar jump

With the common unit normal $n$ pointing from $M_-$ to $M_+$, recover the scalar interface equation after adding $S_W=-\int_W\sqrt{|h|}\,U(\phi_W)$.

Solution

The outward normals of the cut regions are $+n$ and $-n$. Their surface variations therefore add as

$$\left(Z_-n\mathbin{\cdot}\partial\phi_- -Z_+n\mathbin{\cdot}\partial\phi_+ -U'(\phi_W)\right)\delta\phi_W .$$

Free $\delta\phi_W$ sets the coefficient to zero. This establishes variational differentiability, not analytic existence or uniqueness.

2. Reflection at the Maxwell coupling wall

Use $1+r=\tau$ and $e_-^{-2}(1-r)=e_+^{-2}\tau$ to derive $r$, $\tau$, and $R+T$.

Solution

Solving the two linear equations gives

$$r=\frac{e_+^2-e_-^2}{e_+^2+e_-^2}, \qquad \tau=\frac{2e_+^2}{e_+^2+e_-^2} .$$

Then

$$R+T =\frac{(e_+^2-e_-^2)^2+4e_-^2e_+^2} {(e_+^2+e_-^2)^2} =1 .$$

The wall conserves power but is transparent only when $e_-=e_+$.

3. A boundary symmetry or a map between boundary conditions?

Suppose a symmetry wall $U_\alpha$ sends boundary condition $b$ to $\alpha\mathbin{\cdot}b\ne b$. Does $\alpha$ act as an internal symmetry of the boundary theory on $b$?

Solution

No. The wall relates two different boundary conditions. An internal action on $b$ additionally requires an allowed boundary-line endpoint identifying $\alpha\mathbin{\cdot}b$ with $b$, together with compatible anomaly and junction data.

4. A finite-wall junction

In $\mathbb Z_6$, test the proposed two-in/one-out junction $U_4\otimes U_5\to U_3$. What remains unproved?

Solution

The signed incidence is $4+5-3=6=0\pmod 6$, so the selection rule passes. It does not construct the line junction, fix its normalization, or prove the coherence of different junction resolutions.

5. Folding is not inversion

Name three kinds of data changed or retained when folding $\mathcal T_-\lvert\mathcal T_+$ to a boundary of $\mathcal T_-\otimes\overline{\mathcal T_+}$.

Solution

The orientation and anomaly sign of $\mathcal T_+$ reverse; its spin, tangential, and background data must be conjugated consistently; and both original trace fields remain until the folded boundary condition relates them. None of this supplies an inverse wall or a finite fusion limit.

References

• Bachas, Constantin, and Ilka Brunner. “Fusion of Conformal Interfaces.” Journal of High Energy Physics 2008, no. 2 (2008): 085. DOI. Open PDF, arXiv v3
• 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
• Harlow, Daniel, and Jie-qiang Wu. “Covariant Phase Space with Boundaries.” Journal of High Energy Physics 2020, no. 10 (2020): 146. DOI. Open PDF, arXiv v4
• Kapustin, Anton, and Mikhail Tikhonov. “Abelian Duality, Walls and Boundary Conditions in Diverse Dimensions.” Journal of High Energy Physics 2009, no. 11 (2009): 006. DOI. Open PDF, arXiv v1
• Shifman, Mikhail. Advanced Topics in Quantum Field Theory: A Lecture Course. 2nd ed. Cambridge: Cambridge University Press, 2022. DOI
• Thorngren, Ryan, and Yifan Wang. “Anomalous Symmetries End at the Boundary.” Journal of High Energy Physics 2021, no. 9 (2021): 017. DOI. Open PDF, arXiv v1