jet bundles

References

Summary

Definitions

Local sections of a bundle

Let π : E → M be a fiber bundle. A local section of the bundle is a pair (U, f), where U ⊆ M is an open set and f : U → E is a section of the bundle: π ∘ f = idU

Global versus local

Global sections are the case where U = M. Such sections don't always exist, which is why local section are useful.

Derivatives of a section

Let f : M → E be a section of a fiber bundle. Take any open set around a point p ∈ M , U(p) and pick local coordinates xi. Since we have a fiber bundle, the local coordinates on E will look like (xi, yα) , where yα are the coordinates of the typical fiber F.

In local coordinates, the section has the form f(x) = (x, ϕ(x)) where ϕ : U → F. Then since F has local coordinates, we can view ϕ in local coordinates as ϕα(x1, …, xn), where each ϕα is now a map from ℝn → ℝ. We can thus take its derivative, which we define to be the partial derivative $$ \partial_{i} \phi^{\alpha} = \dfrac{\partial \phi^{\alpha} }{\partial x^{i}} $$

Notes

This derivative is coordinate dependent — it's very physics-y. There's no way to say that a section is constant — it can look like it has zero derivative in one trivialization, but then have non-zero derivative in another.

Connections solve this problem — they essentially pick a certain section to be equal to zero (like picking a certain section to be light) and then forcing everything else to mold around that.

Consider the bundle TM → M. We take the first jet space J1(TM). The elements of this jet space correspond uniquely to horizontal subspaces (which are used to define Ehresmann connections). So a section Γ : TM → J1(TM) is a choice of horizontal subspace at each point in the tangent bundle.

In local coordinates on the tangent bundle (xμ, vλ), we can express the section Γ(xμ, vλ) = (xμ, vλ, Γμνλ(x)vν)

Gauge theory: The local coordinates are also the same as picking a gauge

Jet equivalence on a fiber bundle

Let π : E → M be a smooth fiber bundle, and let Γ(p) be the space of local sections whose domains contain p ∈ M.

Two smooth sections f, g ∈ Γ(p) are k-equivalent if all of their partial derivatives up to order k are equal at p. In particular, f(p) = g(p).

Jets: The equivalence class of local sections under the jet equivalence is called a k-jet, and is denoted jpk. Given f ∈ Γ(p), we denote the equivalence class of this element by jpkf.

Jet bundle

Let π : E → M be a fiber bundle, and let p ∈ M. The kth jet manifold of π is the set Jk(π) = {jpkf ∣ p ∈ M, f ∈ Γ(p)}

Prolongation

Let π : E → M be a fiber bundle and let f : M → E be a section. The kth jet prolongation of f jkf : M → Jk(π) is defined by

(jkf)(p) = jpkf = (p, f(p), ∂if(p), ∂ijf(p), …, ) where in the last equation we wrote the jet in the induced local coordinates of the jet space.

Intuition

The coordinates on the jet bundle are just formal symbols. There's no fixed underlying function on the jet spaces that has its derivatives being taken. Instead, for any fixed section of the original bundle E → M, you perform prolongation in order to turn it into an actual function living on the jet space.

Equivalence classes: Because the points on the jet bundle are, by definition, equivalence classes of sections and their derivatives, we have a built in way of moving a given section onto the jet space.

Functions on jet space: If we have some F : Jk(π) → ℝn representing a differential equation, we can plug in any section f : M → E by doing the precomposition F ∘ j−k(f) : M → Jk(π) → ℝn

Now if we have a submanifold given by the zero set of a function on the jet space F, a solution to the differential equation is a section f such that its prolongation lies entirely within said submanifold: F ∘ jpk(f) = 0,  ∀p ∈ M

Problem: What happens if we have a solution to the differential equation that doesn't come from a section of the original bundle? How do we just isolate the proper solutions? This is where contact forms come in

Prolongation of vector field

Consider a vector field on E Q : E → TE

Contact forms

Let Ω1(Jk(π)) be the one-forms of jet space. The space of contact forms is the subspace {θ ∈ Ω1(Jk(π)) ∣ θ(djkf) = 0,  ∀f ∈ Γ(M)}

where djkf : TM → TJk(π) is the pushforward (differential) (not the exterior derivative) of the prolongation

Cartan distribution: Let p ∈ Jk(π). The kernels of all contact forms carve out a subspace of the tangent space called the Cartan distribution

Local coordinates: The local coordinates on the jet space induce a basis of basic contact one-forms θI for all multi-indices I of order  ≤ k − 1. θIα = duIα − uIjαdxj where α corresponds to the coordinates on the bundle E, and Ij means appending j to the multi-index I.

Intuition

Since the jet space is a smooth manifold, we can define differential forms on it: they're just sections of the exterior powers of the cotangent bundle: ⋀ℓTp*Jk(π) → Jk(π)

Local coords: θ = du − ujdxj. The condition θ = 0 on all prolongations, means that df = ∂ifdxi for all sections f. In other words, the coordinate uj on the jet space actually acts like the derivative on all sections f : M → E. Similarly, the other contact forms assure that the other coordinates of the jet space act like derivatives of sections of the original bundle.

Consequence: Now, it turns out that things which satisfy the differential equation carved out by some submanifold of the jet space will be actual prolongations if and only if all of the contact forms vanish on them.

Horizontal and vertical differentials

Because we have θIα = duIα − uIjαdxj, we can write duIα = θIα + uIjαdxj.

So the ordinary exterior derivative on the jet space breaks down into the sum of the contact forms and the exterior derivatives on the coordinate part of the jet space.

Writing the definition of the ordinary exterior derivative for a smooth function on the jet space

df = ∂ifdxi + ∑|I| ≥ 0∂uIαfduIα

we can substitute in duIα = θIα + uIjαdxj to find

df = (∂if + ∑|I| ≥ 0uIiα)dxi + ∑|I| ≥ 0∂uIαfθI

Horizontal: We define the horizontal differential as dHf = (∂if + ∑|I| ≥ 0uIiα)dxi

Vertical: We define the vertical differential as dVf = ∑|I| ≥ 0∂uIαfθI

Higher degrees: We can seee how to extend these formulas to higher degrees by simply taking the ordinary exterior derivative on higher degrees and then collecting terms appropriately.

dθIα = dxi ∧ θIiα

A basis element of Ωp, q has the form fdxi1 ∧ … ∧ dxip ∧ θI1α1 ∧ … ∧ θIqαq

The action of dV is zero on everything except f, on which it introduces an extra factor of θ, sending q → q + 1

The action of dH acts on f and on θ, introducing an extra dx factor, sending p → p + 1

Variational bicomplex

Ω•(J∞E) is bigraded by wedge products of the underlying coordinates of the manifold M and wedge products of the contact forms.

Types of objects

Horizontal degree: Only goes up to the dimension of the manifold M. This degree corresponds to the dimension of the submanifold you're integrating over.

Vertical degree: Can be arbitrarily large. This degree corresponds to the number of infinitesimal variations

  1. Horizontal

    Lagrangians: These live in Ωn, 0, taking the form Ldx1 ∧ … ∧ dxn, where L : J∞E → ℝn

    Conserved currents: Noether's theorem lives on Ωn − 1, 0

  2. Vertical

    We interpret elements of Ω0, q as taking in infinitesimal variations δiu and splitting out real numbers.

    Euler-Lagrange equations: E ∈ Ωn, 1 takes in δu and spits out $\dfrac{\delta L}{\delta u} \delta u d^{n} x$

Source form

Source forms are elements of Ωn, 1 that have no contact forms of shape θI for |I| > 0.

Euler-Lagrange operator

Given a Langragian Ldx1…dxn, the vertical differential has a unique decomposition into

Theorems

Examples

Ordinary smooth functions

We think of smooth maps f : M → ℝn as sections of the trivial bundle M × ℝn → M.

The jet bundles Jk(M × ℝn) have the data of coordinates in M, coordinates in ℝn, and all partial derivatives up to order k.

Example: M = ℝ , n = 1. The second jet space J2(ℝ2) then has dimension 4, with coordinates (x, u, ux, uxx). Differential equations then define surfaces, like uxx + sin (u) = 0

One dimensional EL equations

Consider a Lagrangian L : J1(E) → ℝ. The actual Lagrangian density is then ℒ = L(x, u, ux)dx ∈ Ω1, 0.

In order to take the first variation, we apply the vertical differential.

$$ d_{V} \mathcal{L} = ( \dfrac{\partial L}{\partial u} \theta + \dfrac{\partial L}{\partial u_{x}} \theta_{x}) \wedge dx $$

Now we define $\eta = \dfrac{\partial L}{\partial u_{x}} \theta_{x} \in \Omega^{0,1} $

dH(θIα) = dxi ∧ θIiα dHf = (∂if + ∑|I| ≥ 0uIiα)dxi

We have $d_{H} \eta = (\partial_{x} \dfrac{\partial L}{\partial u_{x}})dx \wedge \theta + \dfrac{\partial L}{\partial u_{x}} dx \wedge \theta_{x} $

We get that $$ d_{V} \mathcal{L} = (\dfrac{\partial L}{\partial u} - \partial_{x} \dfrac{\partial L}{\partial u_{x}} ) \theta \wedge dx - d_{H} \eta $$

We define $E(\mathcal{L}) = (\dfrac{\partial L}{\partial u} - \partial_{x} \dfrac{\partial L}{\partial u_{x}} ) \theta \wedge dx $

In general: For any Lagrangian, we have dVℒ = E(ℒ) + dHη

Noether's theorem: $\eta = \dfrac{\partial L}{\partial u_{x}} \theta_{x} \in \Omega^{0,1} $

Noether's theorems

An infinitesimal symmetry is a prolonged vector field from E.