BF Couplings and Discrete Topological Data

A compact BF coupling pairs two higher-form gauge connections of complementary degree. Its local equations set their curvatures to zero, while the sum over compact global sectors retains finite holonomy. At integer level $N$, the resulting electric and magnetic operators carry labels in $\mathbb Z_N$, and linked complementary supports acquire an $N$th-root-of-unity phase. That conclusion uses compactness, the integral flux lattice, and the global path-integral sum; it does not follow from the local density $B\wedge\mathrm dA$ alone.

This page starts on a closed smooth oriented $D$-manifold and uses the Euclidean weight $e^{-S_E}$. It then opens a boundary and threads the same global-definition test through four-dimensional Higgs theory, three-dimensional Abelian Chern–Simons theory, compact BF theory, and finite gauge theory. Fixed fields define a response phase; integrating the same fields defines a dynamical theory and is a logically separate choice.

Required background. When Is a Topological Term Well Defined? supplies the exponentiated-action and period tests used to quantize the BF coefficient. Differential Forms, Integration, and Stokes’ Theorem supplies the degree counting, orientation, and boundary formula.

Helpful background. Gauging a Higher-Form Symmetry separates a local coupling from the global sum over gauge fields. Linking, Braiding, and Framing fixes the complementary-support linking convention. Chains, Homology, Cohomology, and Exact Sequences supplies the finite-coefficient and torsion information that ordinary differential forms miss.

Compact BF theory pairs two higher-form connections

Let $M^D$ be closed, smooth, and oriented, and choose $1\leq p\leq D-2$. Let $A_p$ be a compact $U(1)$ $p$-form connection and $B_{D-p-1}$ a compact $U(1)$ $(D-p-1)$-form connection. Their curvatures have $2\pi$-integral periods. In a simultaneous local trivialization, the Euclidean BF action is

$$S_E^{\mathrm{loc}}[A,B] = \frac{iN}{2\pi} \int_M B_{D-p-1}\wedge\mathrm dA_p, \qquad N\in\mathbb Z.$$

The degrees add correctly:

$$\deg B+(\deg A+1)=(D-p-1)+(p+1)=D.$$

For the finite theory below take $N\geq1$; the sign of a nonzero $N$ can be reversed by $B\mapsto-B$. Locally the gauge transformations are

$$A\longmapsto A+\mathrm d\alpha_{p-1}, \qquad B\longmapsto B+\mathrm d\beta_{D-p-2}.$$

Large transformations and nontrivial bundles are not exhausted by these exact shifts. A compact higher connection is described by compatible patch, overlap, and integral cocycle data. Intrinsically one may package the fields as differential cohomology classes

$$\check A\in\widehat H^{p+1}(M;\mathbb Z), \qquad \check B\in\widehat H^{D-p}(M;\mathbb Z),$$

and define the phase through their differential-cohomology product. With the ordering chosen to reproduce $B\wedge\mathrm dA$ locally,

$$e^{-S_E[A,B]} = \exp\!\left[ -2\pi iN\int_M\check B\smile\check A \right].$$

The integral here is valued in $\mathbb R/\mathbb Z$. It retains flat and torsion data even when every de Rham curvature vanishes. The ordinary form integral is therefore useful local shorthand, not the complete global definition. Kapustin and Seiberg give an explicit compact-field construction and its Deligne–Beilinson completion in Kapustin and Seiberg 2014, § 3, arXiv v2, printed pp. 9–12, eqs. (3.1)–(3.13), PDF.

No metric appears in the BF phase. That establishes metric independence of this term, not by itself a complete topological quantum field theory. The field role still matters:

• if $A$ and $B$ are fixed backgrounds, the phase is an invertible response;
• if one field is summed, it imposes a compact Fourier constraint on the other; and
• if both are summed, the result is a dynamical compact BF theory.

Integer level turns flatness into finite holonomy

The coefficient lattice follows from the exponentiated action. Let $B\mapsto B+2\pi\beta$, where $\beta$ is a closed integral $(D-p-1)$-form, and let $\mathrm dA/(2\pi)$ represent an integral curvature class. Then

$$\Delta S_E = \frac{iN}{2\pi} \int_M 2\pi\beta\wedge\mathrm dA = 2\pi iN \int_M\beta\wedge\frac{\mathrm dA}{2\pi}.$$

The last integral is an integer. Hence $e^{-S_E}$ is invariant for $N\in\mathbb Z$. When the allowed compact sectors realize a unit pairing, integrality is also necessary. A restricted charge or flux lattice can change the allowed coefficient lattice, so the bare assertion “$N$ is an integer” always includes the standard compact normalization used here. The large-gauge calculation is made explicit in Kapustin and Seiberg 2014, § 3, arXiv v2, printed p. 11, eq. (3.6), PDF.

Varying the local action on a closed manifold gives, away from operator insertions,

$$\mathrm dA=0, \qquad \mathrm dB=0.$$

These are curvature equations. They do not imply that either compact connection is gauge-equivalent to zero. A flat connection can have nontrivial holonomy around a noncontractible cycle, including torsion holonomy invisible to a real differential form.

The compact path integral sharpens the statement. Summing one field performs a finite Fourier projection, so the surviving holonomies obey

$$\frac{N}{2\pi}\oint A\in\mathbb Z, \qquad \frac{N}{2\pi}\oint B\in\mathbb Z$$

on the corresponding cycles. Thus their values are $N$th roots of unity, and the finite labels lie in $\mathbb Z_N$. Gauge-equivalence classes of a flat $p$-form $\mathbb Z_N$ connection are organized by

$$H^p(M;\mathbb Z_N).$$

Brennan and Hong derive the Fourier projection, the finite holonomy, and the operator labels from the compact BF sum in Brennan and Hong 2023, § 3, arXiv v2, printed pp. 32–34, eqs. (3.22)–(3.35), PDF. Replacing either compact field by an unconstrained real form destroys the integral sector sum and does not produce a $\mathbb Z_N$ gauge theory.

Linked Wilson operators measure the mixed pairing

Let $C_p$ and $\Sigma_{D-p-1}$ be disjoint closed oriented supports. Their dimensions add to $D-1$, so they can link in $D$ dimensions. For the ordinary integer linking number below, assume that $\Sigma$ bounds over the relevant coefficients, equivalently that a global $\eta_\Sigma$ can be chosen; this is automatic for the local examples on a sphere. If the correlator is written as a ratio, also work in a sector where the two one-point normalizations are nonzero. General manifolds instead require the torsion or differential-cohomology pairing developed in the next section. Define

$$W_e(C) := \exp\!\left(ie\int_C A\right), \qquad V_m(\Sigma) := \exp\!\left(im\int_\Sigma B\right),$$

with $e,m\in\mathbb Z$. Fix the orientation convention by choosing a form $\eta_\Sigma$ with $\mathrm d\eta_\Sigma=\delta_\Sigma$ and setting

$$\operatorname{Lk}(C,\Sigma) := \int_C\eta_\Sigma.$$

In the Euclidean convention of this page, integrating over $B$ in the presence of $V_m(\Sigma)$ fixes the sourced curvature of $A$. Substitution into $W_e(C)$ gives the mixed topological factor

$$\frac{\left\langle W_e(C)V_m(\Sigma)\right\rangle} {\left\langle W_e(C)\right\rangle \left\langle V_m(\Sigma)\right\rangle} = \exp\!\left[ \frac{2\pi i\,em}{N} \operatorname{Lk}(C,\Sigma) \right].$$

The sign is fixed by the displayed $e^{-S_E}$ convention and the definition of linking above. Reversing the orientation of either support changes the linking number’s sign and complex-conjugates the phase. The charges are defined modulo $N$: shifting $e$ or $m$ by $N$ leaves every mixed phase unchanged. At $N=1$ the finite linking data are trivial.

In four dimensions with $p=1$, $W_e$ is a line and $V_m$ is a surface. In three dimensions with $p=1$, both supports are lines. The same formula then becomes mutual line braiding. The compact normalization and positive linking phase agree with Banks and Seiberg 2011, § 2, arXiv v2, printed p. 8, eqs. (2.12)–(2.13), PDF and Brennan and Hong 2023, § 3, arXiv v2, printed pp. 33–34, eqs. (3.29)–(3.35), PDF.

Torsion requires global gauge data

The finite cohomology group above contains more than reductions of smooth de Rham periods. The universal-coefficient sequence gives

$$0\longrightarrow \operatorname{Ext}_{\mathbb Z}^{1} \!\left(H_{p-1}(M;\mathbb Z),\mathbb Z_N\right) \longrightarrow H^p(M;\mathbb Z_N) \longrightarrow \operatorname{Hom}_{\mathbb Z} \!\left(H_p(M;\mathbb Z),\mathbb Z_N\right) \longrightarrow0.$$

The sequence splits noncanonically. Its left term records finite-coefficient classes not visible from homomorphisms on $p$-cycles alone. If $H_p(M)$ contains a cyclic direct summand $\mathbb Z_\ell$, then that summand contributes

$$\operatorname{Hom}(\mathbb Z_\ell,\mathbb Z_N) \cong \mathbb Z_{\gcd(\ell,N)}.$$

Thus a cyclic direct summand of order $\ell$ contributes $\gcd(\ell,N)$ compatible finite holonomies—not automatically $\ell$ and not automatically $N$. For an order-$\ell$ cycle embedded in a larger homology group, the available values form the image of the restriction map to its cyclic subgroup and can be smaller. Equivalently, if $\ell\gamma=\partial\Gamma$, a globally defined torsion operator must combine the integral over $\Gamma$ with holonomy data on $\gamma$; the local form integral over $\Gamma$ alone is not gauge invariant. Kapustin and Seiberg construct this operator and show that its order divides $\gcd(N,\ell)$ in Kapustin and Seiberg 2014, § 3, arXiv v2, printed p. 12, eqs. (3.11)–(3.13), PDF.

This is the precise sense in which BF theory can be locally flat and globally nontrivial. Curvature probes the free real part of cohomology; compact cocycles, holonomies, and linking pairings retain the finite part.

First application: a charge-N Higgs phase becomes compact BF theory

Take a four-dimensional Euclidean theory on a closed oriented manifold $X^4$ with a compact $U(1)$ connection $A$, normalized by

$$\frac{1}{2\pi}\int_{\Sigma_2}\mathrm dA\in\mathbb Z.$$

Let a charge-$N$ scalar condense while no lower-charge field condenses. At fixed radial mode write

$$\Phi=v e^{i\varphi}, \qquad \varphi\sim\varphi+2\pi, \qquad \varphi\longmapsto\varphi+N\lambda, \qquad A\longmapsto A+\mathrm d\lambda.$$

The phase stiffness contains

$$S_{\mathrm{phase}} = \frac{v^2}{2} \int_{X^4} (\mathrm d\varphi-NA)\wedge *\,(\mathrm d\varphi-NA).$$

Dualizing the periodic scalar gives a compact two-form connection $B$. Up to metric-dependent kinetic terms, the dual Euclidean action contains

$$S_E^{\mathrm{dual}} = S_{\mathrm{kin}}[A,B] + \frac{iN}{2\pi} \int_{X^4}B\wedge\mathrm dA.$$

The kinetic terms are suppressed deep in the gapped infrared, leaving the compact BF phase. Periodicity of $\varphi$ is what makes $B$ compact; the local dualization without the winding-sector sum would not recover the finite global theory. The charge-$N$ Higgs derivation and its normalization are worked out in Banks and Seiberg 2011, § 2, arXiv v2, printed pp. 6–8, eqs. (2.3)–(2.13), PDF and Brennan and Hong 2023, § 3, arXiv v2, printed pp. 35–36, eqs. (3.36)–(3.45), PDF.

The surviving operators are the Wilson line $W_q(C)$ and the vortex surface $V_m(\Sigma)$. For a once-linked unit pair at $N=3$,

$$\operatorname{Lk}(C,\Sigma)=1 \quad\Longrightarrow\quad \frac{\langle W_1(C)V_1(\Sigma)\rangle} {\langle W_1(C)\rangle\langle V_1(\Sigma)\rangle} =e^{2\pi i/3}.$$

Reversing either support gives $e^{-2\pi i/3}$, and three units of either charge have trivial mutual phase. This is a direct infrared measurement of the residual $\mathbb Z_3$ data.

The threaded comparison keeps field role and global data visible

The four rows below apply the same questions—what is global, what is summed, and what is quantized—without identifying theories of different dimension or different field content. The Chern–Simons and finite-gauge rows are bounded comparators, not derivations from the four-dimensional Higgs model.

One global-definition test across the threaded Abelian Chern–Simons, compact BF, and finite-gauge comparison; the rows are not equivalent theories

Model and field role	Global datum	Quantized input	Operator check	What does not follow
4d compact BF from the Higgs phase; both fields dynamical	Compact one- and two-form connections, including torsion sectors	Integer charge and BF level N	Line–surface phase is an Nth root of unity	The local flatness equations alone do not reconstruct the compact sector sum
3d compact Abelian BF; both one-form fields dynamical	An integral off-diagonal K-matrix and compact bundles	K has off-diagonal entries N	Mutual line phase is an Nth root of unity	Twists and boundary data are not fixed by the untwisted BF coefficient
3d Abelian Chern–Simons; background or dynamical connection declared separately	A differential characteristic class and its spin or oriented refinement	The allowed level lattice, not a theta-period identification	Line–line braiding and self-linking require framing data	A quantized local coefficient does not decide invertibility or the field role
Finite Dijkgraaf–Witten gauge theory; finite bundles summed	A finite gauge group, flat bundles, measure, and cocycle class	A discrete cohomology class rather than a continuous BF level	Bundle holonomies and cocycle-weighted amplitudes	A generic finite gauge theory need not admit this Abelian BF presentation

One bounded state-space check makes the finite sector count concrete. In an untwisted $A$-polarization on a closed spatial manifold $Y$, a basis is labeled by the finite flat sectors,

$$\dim\mathcal H(Y) = \left\lvert H^p(Y;\mathbb Z_N)\right\rvert.$$

Consequently the three-dimensional $p=1$ theory has $\dim\mathcal H(T^2)=N^2$, while the four-dimensional $p=1$ theory has $\dim\mathcal H(T^3)=N^3$. The conjugate $B$ holonomies act as finite clock and shift operators on this space; they are not a second independent set of basis labels to be multiplied in again. On a lens space $Y=L(\ell,q)$, the same formula gives $\dim\mathcal H(Y)=\gcd(N,\ell)$, matching the torsion calculation above. The flat-bundle state construction appears in Dijkgraaf and Witten 1990, § 6.3, printed pp. 416–417, eqs. (6.15)–(6.17); the canonical four-dimensional BF count is given in Bergeron, Semenoff, and Szabo 1995, § 4.1, arXiv v1, printed p. 18, PDF.

In three dimensions, untwisted compact BF theory can be written in the Abelian $K$-matrix form

$$K_{\mathrm{BF}} = \begin{pmatrix} 0&N\\ N&0 \end{pmatrix}, \qquad \det K_{\mathrm{BF}}=-N^2.$$

The genus-$g$ Abelian Chern–Simons state-space count is $\lvert\det K\rvert^g$, so on a two-torus

$$\dim\mathcal H(T^2)=N^2.$$

For $N>1$ this immediately rules out invertibility of the dynamical theory. Belov and Moore give the determinant count in Belov and Moore 2005, § 5.3, printed p. 26, immediately after eq. (5.17), PDF. A bosonic Abelian twist may be encoded by

$$K_r = \begin{pmatrix} 2r&N\\ N&0 \end{pmatrix}, \qquad r\in\mathbb Z_N.$$

Its determinant remains $-N^2$, but its spins and braiding data change. In the compact-field convention of Kapustin and Seiberg, their diagonal coefficient is $p_{\mathrm{KS}}=2r$ for an ordinary oriented bosonic Dijkgraaf–Witten class; odd $p_{\mathrm{KS}}$ instead requires spin data. Thus $N$ fixes the untwisted mixed pairing but does not classify every finite topological term. The continuum action and coefficient identification are given in Kapustin and Seiberg 2014, § 5, arXiv v2, printed pp. 20–21, eqs. (5.1)–(5.3), PDF.

For $r=0$, compact three-dimensional BF theory realizes the untwisted $\mathbb Z_N$ finite gauge theory after the global sector sum and measure are matched. This is a statement about the completed quantum theory, not an identity inferred from a local Euler–Lagrange equation. Dijkgraaf and Witten construct the finite-bundle sum and cocycle weights in Dijkgraaf and Witten 1990, § 6.2, printed pp. 415–416, eqs. (6.8)–(6.10).

Boundaries require additional choices

Now let $M$ have a boundary. Under the local $B$-field gauge transformation $B\mapsto B+\mathrm d\Lambda$,

$$\Delta_\Lambda S_E = \frac{iN}{2\pi} \int_M\mathrm d\Lambda\wedge\mathrm dA = \frac{iN}{2\pi} \int_{\partial M}\Lambda\wedge\mathrm dA,$$

where the final sign uses the induced boundary orientation. By contrast, $A\mapsto A+\mathrm d\alpha$ leaves the displayed local curvature $\mathrm dA$ unchanged. Global gauge invariance still belongs to the full differential-cocycle definition.

There is a second, independent boundary issue. An arbitrary variation of $A$ produces a boundary term proportional to

$$\int_{\partial M}B\wedge\delta A,$$

with the sign fixed by $D$ and $p$. A well-posed theory therefore needs a choice such as fixed boundary pullback of $A$, a complementary polarization, restricted boundary gauge transformations, a boundary counterterm that swaps the polarization, or additional boundary fields whose variation cancels the bulk term. The local BF density selects none of these choices by itself. Kapustin and Seiberg display one concrete completion in Kapustin and Seiberg 2014, § 6, arXiv v2, printed pp. 25–26, eqs. (6.19)–(6.25), PDF; it is an example, not a universal boundary theory.

What the BF coupling establishes—and what it leaves open

The compact BF term establishes a robust set of statements within the declared normalization:

• the integer level is required by large compact gauge transformations;
• the local equations impose curvature flatness;
• the compact global sum reduces continuous holonomy to finite $\mathbb Z_N$ data;
• complementary Wilson supports measure the mixed linking pairing; and
• torsion sectors require global cocycles rather than de Rham forms alone.

It does not, by itself, determine every datum of the completed theory. A finite cocycle twist, a spin refinement, a boundary polarization, kinetic deformations away from the infrared, the path-integral normalization, and the full state-space and gluing assignments are additional inputs. Nor does coefficient quantization imply invertibility: the fixed-field BF response is invertible as a phase, while the dynamical $N>1$ BF theory has multi-dimensional state spaces.

The next BF Theory as a Topological Gauge Theory treatment will develop the state spaces, gluing maps, observables, and exact topological data. Finite Gauge Theory and Dijkgraaf–Witten Twists will separate untwisted BF presentations from general cocycle-twisted finite gauge theories. The already developed Chern–Simons Actions and Level Quantization treatment supplies the self-pairing, spin, boundary, and framing side of the comparison.

Common pitfalls

Flat does not mean trivial. The equations $\mathrm dA=\mathrm dB=0$ remove local curvature. They leave holonomy and torsion sectors, which are precisely where the finite gauge data live.

A noncompact BF density does not define a finite gauge theory. The $\mathbb Z_N$ conclusion uses compact higher connections, integral fluxes, and the sum over global sectors. Dropping those ingredients changes the theory even though the local Euler–Lagrange equations look the same.

An integer coefficient does not make a dynamical theory invertible. A fixed BF phase has a stacking inverse, but the dynamical three-dimensional $N>1$ theory already has $N^2$ states on $T^2$.

Boundary variation does not select a unique edge theory. It identifies data that must be completed. Boundary conditions, polarizations, counterterms, and boundary fields are distinct completions with different physics.

Untwisted BF is not every finite gauge theory. In the cyclic Abelian $K$-matrix family above, a cocycle twist changes spins and braiding while $\lvert\det K\rvert=N^2$ and the torus-state count stays fixed. Twists for a generic finite group need not preserve that count, and non-Abelian theories require data not captured by the displayed Abelian form.

Check your understanding

1. Degree check. In $D=7$, let $A$ be a two-form connection. What is the degree of $B$, and what dimensions do the two complementary Wilson supports have?

Solution

Here $p=2$, so $B$ has degree $D-p-1=4$. The $A$ operator is supported on a two-cycle and the $B$ operator on a four-cycle. Their dimensions add to six, which is $D-1$, so disjoint supports can link in seven dimensions.

2. Large-gauge check. Suppose $\int_M\beta\wedge\mathrm dA/(2\pi)=3$. What happens to the Euclidean weight under $B\mapsto B+2\pi\beta$?

Solution

The action changes by $\Delta S_E=6\pi iN$. Therefore $e^{-S_E}\mapsto e^{-S_E}e^{-6\pi iN}=e^{-S_E}$ for integer $N$.

3. Linking check. For $N=5$, $e=2$, $m=3$, and $\operatorname{Lk}(C,\Sigma)=-1$, compute the mixed phase.

Solution

The phase is $\exp[-12\pi i/5]=\exp[-2\pi i/5]$. Reducing the charges modulo five gives the same result. Reversing either support changes it to $\exp[2\pi i/5]$.

4. Torsion check. Suppose $H_p(M)$ has a cyclic direct summand $\mathbb Z_{12}$ generated by $\gamma$. How many $\mathbb Z_8$ holonomies can this summand support?

Solution

$\operatorname{Hom}(\mathbb Z_{12},\mathbb Z_8)$ is $\mathbb Z_{\gcd(12,8)}=\mathbb Z_4$. There are four compatible holonomies. Ordinary de Rham periods do not detect them.

5. Field-role check. Why can the same BF phase describe an invertible response in one use and a noninvertible dynamical theory in another?

Solution

With $A$ and $B$ fixed, each background configuration is assigned one nonzero phase, whose stacking inverse is its complex conjugate. When both fields are integrated, the theory acquires nontrivial sums, operators, and state spaces; in three dimensions $\dim\mathcal H(T^2)=N^2$, so $N>1$ cannot be invertible.

6. Boundary check. What must be added to the statement “$B\wedge\mathrm dA$ is gauge invariant” when $M$ has a boundary?

Solution

Under $B\mapsto B+\mathrm d\Lambda$, the action changes by $\frac{iN}{2\pi}\int_{\partial M}\Lambda\wedge\mathrm dA$. One must specify restricted boundary transformations, a boundary condition or polarization, a compensating counterterm, or boundary degrees of freedom. The arbitrary variation also has a separate $B\wedge\delta A$ boundary term.

References

• Banks, Tom, and Nathan Seiberg. “Symmetries and Strings in Field Theory and Gravity.” Physical Review D 83 (2011), 084019. DOI. Open PDF, arXiv:1011.5120v2.
• Belov, Dmitriy M., and Gregory W. Moore. “Classification of Abelian Spin Chern–Simons Theories.” arXiv:hep-th/0505235v1 (2005). Stable record.
• Bergeron, M., G. W. Semenoff, and R. J. Szabo. “Canonical BF-Type Topological Field Theory and Fractional Statistics of Strings.” Nuclear Physics B 437 (1995), 695–722. DOI. Open PDF, arXiv:hep-th/9407020v1.
• Brennan, T. Daniel, and Sungwoo Hong. “Introduction to Generalized Global Symmetries in QFT and Particle Physics.” arXiv:2306.00912v2 (2023), revised 2 June 2023. Stable record.
• Dijkgraaf, Robbert, and Edward Witten. “Topological Gauge Theories and Group Cohomology.” Communications in Mathematical Physics 129 (1990), 393–429. DOI.
• Kapustin, Anton, and Nathan Seiberg. “Coupling a QFT to a TQFT and Duality.” Journal of High Energy Physics 2014, article 001 (2014). DOI. Open PDF, arXiv:1401.0740v2.