Connections, flux, and holonomy

A magnetic monopole exposes a limitation of the phrase “the gauge potential.” Its field strength can be smooth on a surrounding sphere even though no single smooth potential covers that sphere. The potential must instead be given in patches, with a rule for comparing charged fields on their overlap. That rule contains physical information.

We will construct the patches explicitly, obtain the quantized flux, and then study a different phenomenon: a flat connection whose effect survives around a circle. Basic Abelian gauge theory is sufficient preparation. These are gauge-geometric calculations used in three- and four-dimensional field theories; the two-sphere below is an integration surface, not a claim about a two-dimensional supersymmetry algebra. No supersymmetry is assumed in this lesson.

Fix the group before the charge unit

Take the compact gauge group $U(1)$ and write a gauge transformation as $g(x)=e^{i\Lambda(x)}$. The angle $\Lambda$ is defined modulo $2\pi$. Our real connection one-form $a=a_\mu dx^\mu$ acts on a field of charge $q$ by

$$D\psi=(d-iqa)\psi,\qquad \psi'=e^{iq\Lambda}\psi,\qquad a'=a+d\Lambda,\qquad f=da.$$

The curvature two-form $f$ is gauge invariant. Substitution gives $D'\psi'=e^{iq\Lambda}D\psi$: the derivative of the phase cancels the change of the potential.

Because $\Lambda$ and $\Lambda+2\pi$ describe the same group element, an allowed representation must have $q\in\mathbb Z$. We permit a charge-one Wilson probe, whether or not a dynamical charge-one particle occurs in the action. This fixes the normalization of the connection throughout the calculation.

A connection tells us how to compare fields at nearby points. Along a path $\gamma$, a parallel field satisfies $D\psi=0$, and hence

$$\psi(\gamma(1))= \exp\!\left(iq\int_\gamma a\right)\psi(\gamma(0))$$

in one trivialization, meaning one local choice of basis for the charged field. For a closed path, the resulting phase is the holonomy $W_q(\gamma)=\exp(iq\oint_\gamma a)$. If a path crosses patches, their transition phases must also be included. Holonomy compares the final and initial field in the same fiber, the one-dimensional complex vector space attached to the starting point.

Two regular potentials for a monopole

Use a sphere of radius $\ell$, with polar coordinates $0\leq\theta\leq\pi$ and $\varphi\sim\varphi+2\pi$. Fix the orientation $\sin\theta\,d\theta\wedge d\varphi$. On a patch excluding the south pole and another excluding the north pole, define

$$a_N=\frac m2(1-\cos\theta)\,d\varphi,\qquad a_S=-\frac m2(1+\cos\theta)\,d\varphi.$$

For now $m$ is a real parameter. Each potential gives

$$f=da_N=da_S =\frac m2\sin\theta\,d\theta\wedge d\varphi,\qquad \int_{S^2}f=2\pi m.$$

The apparent coordinate problem at a pole deserves a check. Embed the sphere as $x^2+y^2+z^2=\ell^2$, with $z=\ell\cos\theta$. Since $d\varphi=(-y\,dx+x\,dy)/(x^2+y^2)$ away from the axis,

$$a_N=\frac{m(-y\,dx+x\,dy)}{2\ell(\ell+z)},\qquad a_S=-\frac{m(-y\,dx+x\,dy)}{2\ell(\ell-z)}.$$

The first expression is smooth near $z=+\ell$; the second is smooth near $z=-\ell$. Their excluded poles are covered by the other patch. There is no singular field strength on the sphere.

On the overlap,

$$a_N-a_S=m\,d\varphi=d\Lambda_{NS},\qquad \Lambda_{NS}=m\varphi$$

locally. Thus a charged field must obey

$$\psi_N=e^{iqm\varphi}\psi_S.$$

The potentials and local fields together define a bundle with connection: local descriptions glued by gauge transformations. The underlying $U(1)$ transition function is $g_{NS}=e^{im\varphi}$. It must have the same value when the azimuth advances by $2\pi$. Therefore

$$e^{2\pi i m}=1,\qquad \boxed{m\in\mathbb Z,\qquad \frac1{2\pi}\int_{S^2}f=m.}$$

The integer is the first Chern number in our normalization. For the charge-$q$ associated line bundle, the corresponding integer is $qm$. The transition phase need not have a globally single-valued real logarithm; it is the group-valued function that must be single-valued.

This also explains why one smooth global $a$ cannot describe $m\ne0$. If it existed, Stokes’ theorem on the closed sphere would give $\int_{S^2}da=0$. With patches, apply Stokes separately to the two hemispheres. Their equatorial boundary orientations are opposite, so

$$\int_{S^2}f =\oint_{\mathrm{equator}}a_N-\oint_{\mathrm{equator}}a_S =\oint_{\mathrm{equator}}m\,d\varphi =2\pi m.$$

The nonzero answer comes from the overlap transition, not from a singularity in the physical curvature.

Holonomy and the choice of spanning cap

Consider a latitude at fixed $\theta$, traversed in increasing $\varphi$. Both local potentials describe the loop when $0<\theta<\pi$:

$$W_q^{(N)} =e^{\,i\pi qm(1-\cos\theta)},\qquad W_q^{(S)} =e^{-i\pi qm(1+\cos\theta)}.$$

Their ratio is $e^{2\pi iqm}=1$. The line integral of a potential can change by $2\pi$ times an integer under a large gauge transformation, but the allowed charged probe has an unambiguous phase.

There is a second derivation. The north cap has flux $\pi m(1-\cos\theta)$. The south cap, with the orientation whose boundary is the same increasing-$\varphi$ loop, has flux $-\pi m(1+\cos\theta)$. Their fluxes differ by the whole sphere’s flux. Quantization makes their exponentials equal:

$$W_q=\exp\!\left(iq\int_{\mathrm{cap}}f\right).$$

Thus Stokes’ theorem for a Wilson phase is compatible with more than one spanning surface. It does not require the unexponentiated flux through those surfaces to agree.

For $m=q=1$ and $\theta=\pi/2$, both expressions give $W_q=-1$. Shrinking the loop to the north pole gives $W_q\to1$, although the southern line integral approaches $-2\pi$. The integer ambiguity is precisely what a phase ignores.

A flat connection can still matter

Now take a circle with dimensionless coordinate $\chi\sim\chi+2\pi$ and metric $ds^2=\ell^2d\chi^2$. In a periodic trivialization let

$$a=h\,d\chi,\qquad h\in\mathbb R.$$

This connection has $f=0$. Nevertheless,

$$W_q=e^{2\pi iqh}.$$

Locally one could choose $\Lambda=-h\chi$ to remove the potential. That is a globally allowed $U(1)$ gauge transformation only if $e^{-2\pi ih}=1$. More generally $g=e^{in\chi}$, $n\in\mathbb Z$, changes $h$ to $h+n$, so inequivalent flat connections are labeled by

$$h\in\mathbb R/\mathbb Z.$$

Curvature measures transport around sufficiently small contractible loops. A circle that winds around a noncontractible cycle carries additional information. This is why $f=0$ does not imply that a connection is globally gauge equivalent to zero.

The effect can be measured in a charged scalar’s spectrum. For periodic fields, the modes are $\psi_n=e^{in\chi}$ with $n\in\mathbb Z$. The positive covariant Laplacian is

$$-\frac1{\ell^2}D_\chi^2\psi_n =\frac{(n-qh)^2}{\ell^2}\psi_n,\qquad D_\chi=\partial_\chi-iqh.$$

Under $h\to h+k$, the mode transforms as $\psi_n\to\psi_{n+qk}$. Since $qk$ is integral, this simply relabels the same spectrum. A periodic parallel section exists precisely when $qh\in\mathbb Z$, for then $\psi=e^{iqh\chi}$ is single-valued and has zero eigenvalue.

This condition is representation dependent. At $h=1/2$, a charge-two field has a parallel mode but a charge-one field does not. The charge-one holonomy still distinguishes this connection from $h=0$. Detecting trivial transport with one representation does not prove that every allowed representation sees trivial transport.

The normalization also keeps units transparent. A component along a length coordinate has mass dimension one, but $h$ and the angular component $a_\chi$ are dimensionless. In an orthonormal frame $a_{\hat\chi}=h/\ell$, and the eigenvalues have dimension two.

Global data beyond the local action

Suppose the only dynamical matter has charge two. Its overlap condition alone would test $e^{4\pi im}=1$, apparently permitting half-integral $m$. In the $U(1)$ theory specified at the beginning, that conclusion is too weak: the charge-one Wilson line is also allowed, and the principal transition function itself must be single-valued. The flux remains integral.

One can instead define a theory with a different set of global identifications and allowed probes. Although a quotient of $U(1)$ by a finite subgroup is again abstractly a circle, its connection normalization and its identification of allowed charges have changed. One must translate those data together. The matter fields appearing in a Lagrangian do not by themselves fix all of them.

A non-Abelian example makes the distinction concrete. Let $\tau_3=\operatorname{diag}(1,-1)$ and consider

$$g(\varphi)=\exp\!\left(\frac{i\varphi\tau_3}{2}\right).$$

At $\varphi=2\pi$ this matrix is $-I$. It is therefore not a closed transition function around the equator for gauge group $SU(2)$. Its image in $SO(3)=SU(2)/\{I,-I\}$ is closed: conjugation by $-I$ acts trivially on an adjoint vector. A fundamental doublet would detect the sign and is not a representation of $SO(3)$. The two groups share a Lie algebra, but not the same bundles and Wilson probes.

This distinction is developed in Aharony, Seiberg, and Tachikawa, section 1.1. In particular, allowed Wilson lines are associated with representations of the chosen group, including representations absent from the dynamical matter content.

In three Euclidean dimensions, an $S^2$ can surround a point at which a monopole boundary condition is imposed. In four dimensions, it can be transverse to a magnetic worldline. If the flux is nonzero, a smooth closed curvature cannot extend across the enclosed three-ball: Stokes’ theorem would again force zero flux. This geometric statement identifies the needed defect or excluded region. It does not establish that a particular field configuration has finite action.

Translate a paper’s charge unit

Factors of two in fluxes often encode a different electric-charge unit. Seiberg and Witten’s compactification discussion, section 3, printed page 21 uses an Abelian potential $A_{\mathrm{SW}}$ for which the smallest probe charge is $1/2$. The monopole flux is $4\pi$, and the Wilson-line integral is identified modulo $4\pi$.

To use our unit-charge convention, set

$$A_{\mathrm{SW}}=2a,\qquad q_{\mathrm{SW}}=\frac q2.$$

Then $q_{\mathrm{SW}}\oint A_{\mathrm{SW}}=q\oint a$, and a $4\pi$ flux becomes a $2\pi$ flux for $a$. On the paper’s circle $x^4\sim x^4+2\pi R$, it writes $A_{\mathrm{SW},4}=b/(\pi R)$ with $b$ of period $2\pi$. Thus

$$\oint A_{\mathrm{SW}}=2b,\qquad \oint a=b,\qquad h=\frac b{2\pi}.$$

This is a connection and charge normalization, not a choice of electric-magnetic period basis. A later period calculation needs its own explicit basis convention.

Periodic scalars obtained by dualizing a three-dimensional photon are likewise tied to charge quantization; Aharony et al., section 2.2 explains that setting. The patch calculation here supplies the elementary global information that such a dual description must preserve.

Exercises

1. Two caps, one measured phase

Take $m=q=1$ and the latitude $\theta=\pi/3$, oriented toward increasing $\varphi$. Calculate its Wilson phase using both potentials. Then explain what fails if one keeps the same unit-charge theory but chooses $m=1/2$.

First hint

The two line integrals differ by $2\pi m$, independently of the latitude.

Second hint

For $\theta=\pi/3$, $\cos\theta=1/2$. Compare the exponentials, not just the real integrals.

Solution

The north and south integrals are $\pi/2$ and $-3\pi/2$, respectively. Both give $W_1=i$. Their difference is $2\pi$, the flux through the sphere.

For $m=1/2$, the two proposed phases are $e^{i\pi/4}$ and $e^{-3i\pi/4}$, whose ratio is $-1$. They cannot describe the same unit-charge Wilson observable. Equivalently, $g_{NS}=e^{i\varphi/2}$ changes sign after one circuit and is not a well-defined $U(1)$ transition function. The problem is global; differentiating either local potential still gives a perfectly regular local curvature.

2. Which field detects the flat connection?

At $h=1/2$ on a circle of radius $\ell$, find the lowest covariant-Laplacian eigenvalue for charges one and two, including its mode numbers. Can the charge-two result justify setting the connection to zero in the original theory?

First hint

Minimize $(n-q/2)^2$ over integer $n$.

Second hint

Also compare the charge-one Wilson phase with its value at $h=0$.

Solution

For $q=1$, the closest integers to $1/2$ are $n=0$ and $n=1$. Both modes have eigenvalue $1/(4\ell^2)$, and there is no parallel mode. For $q=2$, $n=1$ gives zero, with $\psi=e^{i\chi}$ solving $D_\chi\psi=0$.

The charge-two holonomy is $1$, but the charge-one holonomy is $-1$, whereas it is $1$ at $h=0$. A gauge transformation removing $h=1/2$ would require $g=e^{-i\chi/2}$, which is not periodic. The charge-two field alone is insensitive to this distinction; it does not remove the allowed charge-one probe from the theory.

3. The same local algebra, a different transition

For $g(\varphi)=\exp(i\varphi\tau_3/2)$, compare its action on a fundamental doublet $u$ and on an adjoint matrix $X=X^a\tau_a/2$ after one circuit. Explain why an adjoint-only Lagrangian does not tell you whether the global group is $SU(2)$ or $SO(3)$. Give a closed $SU(2)$ transition obtained by changing the exponent.

First hint

The doublet transforms as $u\mapsto gu$, whereas $X\mapsto gXg^{-1}$.

Second hint

A central sign survives in the first action but cancels in the second. Doubling the exponent changes its endpoint.

Solution

At $2\pi$, the doublet transforms to $-u$, while the adjoint returns to $X$. Therefore the proposed transition is closed as an $SO(3)$ transformation but not as an $SU(2)$ transformation. A fundamental Wilson probe can distinguish the two global choices even when every dynamical field is adjoint.

The loop $g_2(\varphi)=\exp(i\varphi\tau_3)$ ends at $I$ and is a valid closed $SU(2)$ transition. This verifies closure only; it does not assert that it defines a topologically nontrivial principal $SU(2)$ bundle. Indeed, closed loops in $SU(2)$ can be contracted. Global classification and the choice of an Abelian reduction are additional questions.

The useful habit is to carry three items together: the local covariant derivative, the patch transition functions, and the allowed charged probes. Curvature, quantized flux, and holonomy then become compatible parts of one gauge field rather than competing definitions.

For the corresponding question about spin frames, Spinors as bundle sections separates a component sign from the return map of parallel transport. A charged spinor must satisfy both its spin-frame and gauge transition laws.

Circle bundles and the Hopf fibration attaches a geometric circle to the monopole’s phase data. Its normalized connection is smooth on the total three-sphere even though its base potentials require two patches.