[X_i, X_j] = curvature that makes the geometry
sub-Riemannian. It is the mathematical spine behind
Part 1.
Connection: a rule for carrying things along a path
A connection answers a question that has no automatic answer on a curved or twisted space: as I move along a path, how do I carry a quantity so that it stays “the same”? Parallel transport of a vector on a sphere; the phase of a quantum state as parameters change; the direction “north” as you walk the globe. In each case the rule is encoded by a connection one-form $\alpha$ — a recipe that, given your velocity, tells you the compensating change to apply.
For a charged particle the connection is the vector potential $\mathbf A$, and the quantity carried is a phase. Transporting along a step $d\boldsymbol\ell$ accrues phase proportional to $\mathbf A\cdot d\boldsymbol\ell$.
Holonomy: what a loop leaves behind
Transport all the way around a closed loop and you generally do not come back to where you started — the carried quantity has shifted. That leftover is the holonomy. On a sphere, parallel transport around a triangle rotates a vector by the enclosed solid angle. For the charged particle, going around a loop $\gamma$ accrues the magnetic flux
\[\varphi \;=\; \oint_\gamma \mathbf A\cdot d\boldsymbol\ell \;=\; \iint_S \mathbf B\cdot d\mathbf S,\]by Stokes’ theorem — the flux threading the enclosed surface. Holonomy is not a defect; it is a measurement. The loop reads out the field it encloses.
Curvature: holonomy in the small
Shrink the loop to an infinitesimal parallelogram spanned by two directions, and the holonomy per unit area is the curvature $F = d\alpha$, the exterior derivative of the connection form. For magnetism, $F_{ij} = \partial_i A_j - \partial_j A_i$, i.e.
\[\mathbf B = \nabla\times\mathbf A .\]Curvature is the local density of holonomy: how much phase a tiny loop catches per unit area. Where the curvature (field) vanishes, tiny loops catch nothing to leading order — the point that drives the vanishing-order $k$ in the law $Q = d + k + 2$.
The dictionary
| Geometry | Magnetism | Rotation (Coriolis) | Quantum (Berry) |
|---|---|---|---|
| connection $\alpha$ | vector potential $\mathbf A$ | angular-velocity potential | Berry connection $\langle\psi\rvert\,d\,\lvert\psi\rangle$ |
| holonomy $\varphi$ | magnetic flux | Coriolis circulation | Berry phase |
| curvature $F$ | field $\mathbf B$ | vorticity $2\boldsymbol\omega$ | Berry curvature |
The series works the magnetic column in full and points at the others (Part 1’s selection table). The mathematics is identical; only the physical name of the field changes.
The flux lift, and why the bracket is the curvature
Here is the step that turns a connection into a sub-Riemannian geometry. Adjoin the holonomy as an extra coordinate: state $= (x_1,\dots,x_d, \varphi)$, with $\varphi$ constrained to track the connection, $d\varphi = \sum_i A_i\,dx_i$. The allowed motions — the ones respecting that constraint — are the horizontal frame
\[X_i = \partial_{x_i} + A_i\,\partial_\varphi .\](Standing hypothesis, used everywhere and worth saying once: the potential depends on position only, $A_i = A_i(x)$, never on the fibre coordinate $\varphi$ — that is what makes the bracket below close on $\partial_\varphi$ and, in D4, what conserves $w = p_\varphi$. This “adjoin the holonomy as a coordinate” construction is itself classical: it is the central extension of Kostant–Souriau prequantization and the magnetic/isoperimetric model of Montgomery’s book — the series builds on it, it did not invent it.)
Their Lie bracket is a direct computation:
\[[X_i, X_j] = (\partial_{x_i} A_j - \partial_{x_j} A_i)\,\partial_\varphi = F_{ij}\,\partial_\varphi .\]The bracket of two horizontal moves is the curvature, pointing in the forbidden holonomy direction. So wherever the curvature is nonzero the distribution is bracket-generating (Chow’s condition holds, Appendix D2) and the geometry is genuinely sub-Riemannian; wherever it vanishes, you must go to higher brackets, and the geometry degenerates (Martinet and beyond). Every quantitative result in the series — the law, the null jump, the caustic gradient — is downstream of this one identity (a genealogy, not a claim that the identity does the work by itself).
Gauge freedom, and what survives it
The connection is not unique: $\mathbf A \to \mathbf A + \nabla\chi$ leaves the curvature $\mathbf B$ unchanged (a gauge transformation), and correspondingly shifts the holonomy coordinate $\varphi \to \varphi + \chi$. Anything physical must be gauge-invariant. The curvature is; so are the growth vector and the conjugate time (the series checks the latter numerically — symmetric and Landau gauges give identical caustics). But the coordinate $\varphi$ is not gauge-invariant for open paths, which is why the measurements must be done in a gauge adapted to the base point ($\mathbf A(q_0)=0$) — a subtlety that, ignored, silently corrupts the growth-vector estimate. Gauge care is not optional bookkeeping; it is the difference between measuring geometry and measuring an artefact.
References
- M. Nakahara (2003). Geometry, Topology and Physics, 2nd ed. IOP. (Connections, holonomy, curvature.)
- M. V. Berry (1984). “Quantal phase factors accompanying adiabatic changes.” Proc. R. Soc. Lond. A 392, 45–57.
- R. Montgomery (2002). A Tour of Subriemannian Geometries. AMS. (The bracket-generating condition and holonomy.)
- R. Montgomery (1995). “Hearing the zero locus of a magnetic field.” Comm. Math. Phys. 168, 651–675. (The magnetic lift itself, a decade earlier and in this exact setting: the extra bracket at a nondegenerate zero and the zero locus as a strictly abnormal minimizer — the closest prior work; companion article §6.)