Manifolds modeled on locally convex spaces

3 Tangent bundles

The tangent bundle and the tangent map of a manifold modeled on a locally convex space, the double tangent bundle with its canonical flip, and curves with their velocities [ GN26 , § 3.3 ] , [ Lan99 , Ch. III ] . As in Chapter 2, no completeness is assumed.

3.1 The tangent bundle and tangent maps

For charts \(i\), \(j\) of the atlas of \(M\) and \(x\) in both domains, the coordinate change of tangent vectors is \(v \mapsto \mathrm{d}(j \circ i^{-1})(i(x))\, v\). The tangent bundle \(TM\) is the fiber bundle with fiber \(E\) obtained by gluing the trivial bundles over the chart domains along these coordinate changes. For a chart \(i\), the map \(TM \supseteq TM|_{\operatorname {dom} i} \to E \times E\) that sends \((x, v)\) to \((i(x), v_i)\), with \(v_i\) the coordinates of \(v\) in the chart \(i\), is the chart of \(TM\) over \(i\).

In Lean, \(TM\) is the total space of a Mathlib FiberBundleCore indexed by the charts of the atlas, whose index at \(x\) is the preferred chart \(c_x\). Thus a point of \(TM\) is a pair \((x, v)\) with \(x \in M\) and \(v \in E\), and \(v\) is the tangent vector written in the coordinates of the preferred chart at \(x\). In any other chart \(i\), its coordinates are \(\mathrm{d}(i \circ c_x^{-1})(c_x(x))\, v\). This is a choice of representative, not a restriction: every construction below (tangent maps, local representatives of vector fields, Lie brackets, derivations, velocities) is defined in the preferred chart and then shown to transform correctly in every chart of the atlas. The coordinate changes satisfy the cocycle condition by the chain rule (LocallyConvex.Manifolds.tangentCoordChange_comp), and they are continuous in \((x, v)\) since the transition maps are \(C^{1}\).

The charts of \(TM\) over the charts of \(M\) form an atlas modeled on \(E \times E\) whose transition maps are

\[ (y, w) \longmapsto \bigl(\tau (y), \mathrm{d}\tau (y)\, w\bigr), \qquad \tau = j \circ i^{-1}. \]

In particular they are smooth, and \(TM\) is a manifold modeled on \(E \times E\).

Proof ▶

The formula is the cocycle property of the coordinate changes. The map \((y, w) \mapsto \mathrm{d}\tau (y)\, w\) is smooth because \(\tau \) is, by the recursive definition of \(C^{k}\), and pairs of smooth maps are smooth. The inverse transition has the same form with \(\tau ^{-1}\).

Definition 3.3 The tangent map
✓

Let \(f \colon M \to N\). The tangent map \(Tf \colon TM \to TN\) sends \((x, v)\) to

\[ \bigl(f(x),\ \mathrm{d}(c'_{f(x)} \circ f \circ c_x^{-1})(c_x(x))\, v\bigr), \]

where \(c_x\) and \(c'_{f(x)}\) are the preferred charts at \(x\) and \(f(x)\).

The definition makes sense for every \(f\), since \(\mathrm{d}\) is a total function in Lean; it is meaningful when \(f\) is \(C^{1}\).

Let \(f \colon M \to N\) be \(C^{1}\). The derivatives of the local representatives of \(f\) transform under changes of charts by the derivatives of the transition maps. Consequently, in the charts of \(TM\) and \(TN\) over charts \(i\) of \(M\) and \(j\) of \(N\), the tangent map reads

\[ (y, w) \longmapsto \bigl(h(y), \mathrm{d}h(y)\, w\bigr), \qquad h = j \circ f \circ i^{-1}. \]

Moreover \(T\mathrm{id}= \mathrm{id}\) and, for \(C^{1}\) maps \(f\) and \(g\), \(T(g \circ f) = Tg \circ Tf\).

Proof ▶

Near a point, the local representative of \(f\) in one pair of charts agrees with the composition of its local representative in another pair with transition maps; the chain rule for maps that agree with a composition near the point gives the transformation laws. The formula in charts follows by composing these laws with the coordinate changes of \(TM\) and \(TN\), and functoriality is the chain rule in the preferred charts.

Theorem 3.5 Smoothness of the tangent map
✓

If \(f \colon M \to N\) is \(C^{k+1}\), then \(Tf \colon TM \to TN\) is \(C^{k}\).

Proof ▶

In the charts of the tangent bundles \(Tf\) is \((y, w) \mapsto (h(y), \mathrm{d}h(y)\, w)\), and \(\mathrm{d}h\) is \(C^{k}\) on its domain because \(h\) is \(C^{k+1}\).

3.2 The double tangent bundle

The double tangent bundle is \(TTM = T(TM)\), the tangent bundle of the manifold \(TM\) (Theorem 3.2). We follow [ Lar26 , Cap. 5 ] and [ Lan99 , § IV.3 ] . A chart \(i\) of \(M\) gives a chart of \(TM\) and then a chart of \(TTM\), in which points are written \(((x, v), (u, w))\): \((x, v)\) is the base point in \(TM\), and \((u, w)\) the tangent vector at it.

Let \(\tau = j \circ i^{-1}\). In the charts of \(TTM\) over \(i\) and \(j\), the transition map is

\[ ((x, v), (u, w)) \longmapsto \bigl((\tau x, \mathrm{d}\tau _x v),\ (\mathrm{d}\tau _x u,\ \mathrm{d}^2\tau _x(u, v) + \mathrm{d}\tau _x w)\bigr). \]
Proof ▶

The transition maps of \(TTM\) are those of the tangent bundle of \(TM\), namely the transition map \(\sigma (x, v) = (\tau x, \mathrm{d}\tau _x v)\) of \(TM\) together with its derivative. Differentiating \(\sigma \) in the direction \((u, w)\) gives \((\mathrm{d}\tau _x u, \mathrm{d}^2\tau _x(u, v) + \mathrm{d}\tau _x w)\), since \(\mathrm{d}\tau _x\) is linear in \(v\).

The swap \(((x, v), (u, w)) \mapsto ((x, u), (v, w))\) commutes with the transition maps of \(TTM\). Hence the map \(\kappa \colon TTM \to TTM\) defined as the swap in the preferred chart reads as the swap in every chart of \(TTM\) over a chart of \(M\). It covers the identity of \(M\), and \(\kappa \circ \kappa = \mathrm{id}\).

Proof ▶

By Theorem 3.6, the claim amounts to \(\mathrm{d}^2\tau _x(u, v) = \mathrm{d}^2\tau _x(v, u)\), the symmetry of the second differential. The rest is immediate from the coordinate expression.

The smoothness of \(\kappa \) is not stated in the library.

3.3 Curves and velocities

Curves are maps \(\alpha \colon \mathbb {R}\to M\). We treat them, like functions, through their local representatives, without regarding \(\mathbb {R}\) as a manifold.

Definition 3.8 \(C^{k}\) curves and lifts
✓

A curve \(\alpha \colon \mathbb {R}\to M\) is \(C^{k}\) if it is continuous and \(i \circ \alpha \) is \(C^{k}\) on the open set \(\alpha ^{-1}(\operatorname {dom} i)\) for every chart \(i\) of the atlas. A lift of \(\alpha \) is a curve \(\beta \colon \mathbb {R}\to TM\) with \(\beta (t) \in T_{\alpha (t)} M\); its fiber coordinate in the chart \(i\) is the second component \(\mu _i(t)\) of the chart of \(TM\) over \(i\) at \(\beta (t)\). The lift is \(C^{k}\) if every \(\mu _i\) is \(C^{k}\) on \(\alpha ^{-1}(\operatorname {dom} i)\).

Definition 3.9 Velocity
✓
#

The velocity of a curve \(\alpha \) is the lift \(\dot\alpha (t) = (\alpha (t), (c_{\alpha (t)} \circ \alpha )'(t))\), the derivative of the local representative in the preferred chart at \(\alpha (t)\).

Let \(\alpha \) be a \(C^{1}\) curve. For charts \(i\), \(j\) around \(\alpha (t)\) we have \((j \circ \alpha )'(t) = \mathrm{d}(j \circ i^{-1})(i(\alpha (t)))\, (i \circ \alpha )'(t)\); hence the fiber coordinate of \(\dot\alpha (t)\) in any chart \(j\) is \((j \circ \alpha )'(t)\). The velocity is a continuous curve in \(TM\), and if \(\alpha \) is \(C^{k+1}\), then \(\dot\alpha \) is a \(C^{k}\) lift.

Proof ▶

Near \(t\), \(j \circ \alpha = (j \circ i^{-1}) \circ (i \circ \alpha )\), and we apply the chain rule. The second statement is the first one with \(i\) the preferred chart. In the chart of \(TM\) over \(i\), \(\dot\alpha \) reads \(s \mapsto (i(\alpha (s)), (i \circ \alpha )'(s))\), which is continuous (respectively \(C^{k}\) in the second component); the charts of \(TM\) are homeomorphisms onto their images.

Proposition 3.11 Germs and the chain rule for velocities
✓

The velocity \(\dot\alpha (t)\) only depends on the germ of \(\alpha \) at \(t\). If \(p \colon N \to M\) is \(C^{1}\) and \(\Gamma \) is a \(C^{1}\) curve in \(N\), then \(Tp\, (\dot\Gamma (t))\) is the velocity of \(p \circ \Gamma \) at \(t\).

Proof ▶

Directional derivatives only depend on germs. For the chain rule, near \(t\) we write \(c' \circ p \circ \Gamma = (c' \circ p \circ c^{-1}) \circ (c \circ \Gamma )\) in the preferred charts \(c\) at \(\Gamma (t)\) and \(c'\) at \(p(\Gamma (t))\), and apply the chain rule.

Remark 3.12
#

The continuity of \(\dot\alpha \) in \(TM\) is what makes lengths of curves meaningful for Finsler metrics (see Definition 13.4); the chain rule for velocities is used for quotient metrics, where curves in \(N\) are pushed down along a projection \(p\). On the other hand, we do not claim the existence of integral curves of vector fields, or of curves with prescribed velocity: this would require existence theorems for ordinary differential equations, which fail in general locally convex spaces, and in Banach spaces need completeness.