research27 min studyLesson 953 in Mathematics

Connections as horizontal distributions

Derive connection forms, their affine overlap law, and the parallel-transport equation from a horizontal splitting.

On this page

What you’ll understand

  • Relate horizontal distributions to principal connection forms.
  • Derive the local connection law and covariant derivative.
  • Solve transport with a specified sign and initial condition.

A bundle describes which fibers belong over nearby points, but it does not identify vectors in different fibers. A connection supplies an infinitesimal comparison rule. On a principal bundle this is a horizontal complement to the directions generated by the group action. On an associated vector bundle it becomes a covariant derivative. We keep the same right-action and column-vector conventions, so the signs of both descriptions can be checked against one transport equation. The base remains a smooth paracompact manifold, and all paths used for transport are piecewise smooth on compact parameter intervals.

A horizontal complement

The vertical space at a point of PP is the kernel of the differential of the projection. A principal connection chooses a smooth complement HpH_p that is carried to HpgH_{pg} by right translation. Equivalently, it is a Lie-algebra-valued one-form ω\omega that returns ξ\xi on the fundamental vector generated by pexp⁡(tξ)p\exp(t\xi) and transforms equivariantly under right multiplication. Its kernel is precisely the horizontal distribution. The normalization identifies vertical motion with the correct Lie-algebra element.

Pulling ω\omega back along a section gives a local potential Ai=si∗ωA_i=s_i^*\omega. Moving from sis_i to sj=sigijs_j=s_i g_{ij} introduces both conjugation and the derivative of the changing frame. Thus a connection form does not transform like a tensor. Dropping the derivative term would make ordinary derivatives of sections disagree on overlaps. A global principal form always exists for a connection, but a single base-valued matrix potential requires a global trivialization.

ω(ξ#)=ξ,Rg∗ω=Ad⁡(g−1)ω,Aj=gij−1Aigij+gij−1dgij.\omega(\xi^\#)=\xi,\quad R_g^*\omega=\operatorname{Ad}(g^{-1})\omega,\quad A_j=g_{ij}^{-1}A_i g_{ij}+g_{ij}^{-1}dg_{ij}.
The first two identities characterize a principal right connection; the third is its local expression under a moving section.↔ Scroll wide equations horizontally.

The derivative downstairs

For the standard matrix representation, define the local covariant derivative of a component column by ∇vi=dvi+Aivi\nabla v_i=dv_i+A_i v_i. Since vi=gijvjv_i=g_{ij}v_j, the connection overlap law makes ∇vi=gij∇vj\nabla v_i=g_{ij}\nabla v_j. For a general representation, replace AiA_i by its image under the differential representation. This compatibility, together with the Leibniz rule, makes the derivative intrinsic even though its separate terms depend on the local frame. It is a differential operator, not ordinary fiberwise multiplication.

Two connections differ by an adjoint-bundle-valued one-form: their inhomogeneous derivative terms cancel. Consequently the space of connections on a fixed bundle is affine. It has no preferred zero unless extra data select one. Convex combinations with coefficients summing to one remain connections, and a subordinate partition of unity glues local choices. These existence statements use paracompactness. They should not be confused with a claim that a globally flat connection exists on every bundle; flatness is a further nonlinear condition.

Horizontal lifts and transport

Along a path in one trivializing chart, write a principal lift as p(t)=s(γ(t))g(t)p(t)=s(\gamma(t))g(t). Horizontality is equivalent to the matrix equation below. The same equation transports associated columns through v(t)=U(t)v(0)v(t)=U(t)v(0). Principal connections admit transport over each finite piecewise smooth path; local ordinary differential equations patch uniquely, and the group structure prevents a finite-path escape of the principal lift. This conclusion is special to the compatible principal connection, not an unrestricted completeness assertion for every Ehresmann connection.

For a constant commuting matrix potential, transport is an ordinary exponential. For a varying noncommuting potential, writing the exponential of the integrated matrix generally fails; time ordering records the successive multiplication. A change of frame changes endpoint matrices, while the underlying map between fibers remains the same. This is a useful practical test of a formula: a purported transport law must both solve its differential equation and transform correctly at the two endpoints.

U˙(t)=−Aγ(t)(γ˙(t))U(t),U(0)=I,Uj(t)=g(γ(t))−1Ui(t)g(γ(0)).\dot U(t)=-A_{\gamma(t)}(\dot\gamma(t))U(t),\quad U(0)=I,\qquad U_j(t)=g(\gamma(t))^{-1}U_i(t)g(\gamma(0)).
The minus sign comes from requiring dv+Av to vanish; endpoint frame changes act on opposite sides.↔ Scroll wide equations horizontally.

Transport in a rotating line frame

On the trivial complex line over an interval, let A=2i dtA=2i\,dt and transport from t=0t=0 to t=π/4t=\pi/4. Then change the local frame by g(t)=e−itg(t)=e^{-it}.

  1. The equation U′=−2iUU'=-2iU with U(0)=1U(0)=1 gives U(t)=e−2itU(t)=e^{-2it}. At the specified endpoint it is −i-i.

  2. The changed potential is A′=A+g−1dg=2i dt−i dt=i dtA'=A+g^{-1}dg=2i\,dt-i\,dt=i\,dt, because this group is abelian.

  3. Call the new coordinate transport V(t)V(t). It solves V˙=−iV\dot V=-iV with V(0)=1V(0)=1, so V(t)=e−itV(t)=e^{-it}; at π/4\pi/4 it equals e−iπ/4e^{-i\pi/4}. The endpoint rule gives the same number: eite−2itg(0)=e−ite^{it}e^{-2it}g(0)=e^{-it}.

The result

The coordinate matrices differ because their endpoint frames differ. They represent the same intrinsic transport map, and both satisfy the covariant equation with their own local potential.

Explore the model

Interactive lab

Interactive lab · vectors

Coordinate rotation without changing length

Pause/Reset and set angle 45 degrees and magnitude 2. Predict projections (√2,√2). Reset and set both controls to angle −45 degrees and magnitude 2: the projections become (√2,−√2), with the same length. This compares coordinate phase directions; the instrument does not integrate a connection or show a horizontal lift. Changing controls retains elapsed time.

Coordinate rotation without changing lengthPlayback traces a fixed vector and its projections. Angle and magnitude set the geometry; the moving marker does not change them. Current readings: x component: 1.393; y component: 0.975; Magnitude: 1.7. All parameters are adjustable with the labeled controls below.−7.46−3.7303.737.46−3.5−1.7501.753.5xyvvₓ = 1.39
Vectorx / y projectionsTrace marker
x component
1.393
y component
0.975
Magnitude
1.7
t = 0 s

Playback traces a fixed vector and its projections. Angle and magnitude set the geometry; the moving marker does not change them. Parameter changes pause playback and recompute the model at the current elapsed time. Reset restores both parameters and returns time to zero. Playback starts paused. Adjust either parameter, then choose Play or Step.

Look forA changing direction of components must be interpreted together with the changing frame.

Check your understanding

For ∇=d+A\nabla=d+A on columns, what equation defines parallel transport UU?

Try it yourself

On a trivial real rank-two bundle over [0,1][0,1], take A=(0300)dtA=\begin{pmatrix}0&3\\0&0\end{pmatrix}dt. Find transport to time one and its action on (1,2)T(1,2)^T. Explain why its determinant is one.

Take a moment to make a prediction or work it out on paper. Open a hint when you need one.

Give me a hint

The constant coefficient matrix squares to zero, so its exponential terminates after the linear term.

Compare with the solution

Writing the coefficient as NN gives N2=0N^2=0. Thus U(t)=e−tN=I−tNU(t)=e^{-tN}=I-tN and U(1)=(1−301)U(1)=\begin{pmatrix}1&-3\\0&1\end{pmatrix}. The transported column is (−5,2)T(-5,2)^T. Direct differentiation verifies U˙=−NU\dot U=-NU, and U(0)=IU(0)=I. Its triangular diagonal entries are one, so its determinant is one; equivalently the trace of NN vanishes. This particular connection preserves area but does not preserve the standard Euclidean norm.

Take it with you

  • A principal connection is an equivariant horizontal splitting.
  • Local potentials have an inhomogeneous overlap law.
  • Parallel transport solves a signed matrix ODE with endpoint covariance.

Read further

Original sources and deeper explanations to follow at your own pace.

Optional progress, saved in this browser when available. Revisit any lesson at any time.

Lesson not marked complete.