Michal–Bastiani calculus on locally convex spaces

4 Weak integrals

Integration enters Michal–Bastiani calculus in two places: through the mean value theorem, which expresses \(f(x + h) - f(x)\) as an integral of the differential along the segment, and through integrals depending on parameters, which are a basic tool to produce smooth maps. Without a norm there is no Bochner integral, and without completeness Riemann sums need not converge. The integral used in infinite-dimensional Lie theory is therefore defined by duality: a vector is the integral of a curve if every continuous linear functional sees it as such.

Weak integrals defined in this way are unique, by Hahn–Banach, but they need not exist; existence is the subject of the next chapter (Chapter 5). Everything in the present chapter holds in an arbitrary locally convex space, with the existence of the relevant integrals as an explicit hypothesis. We first establish the basic properties of the integral and the change of variables formula, then the first part of the fundamental theorem of calculus, which needs no hypothesis on \(F\) at all, and finally the continuity and differentiability of integrals depending on parameters. Our reference is [ GN26 , § 1.1 ] .

4.1 Definition and basic properties

Definition 4.1 Weak integral
✓
#

Let \(\gamma \colon \mathbb {R}\to F\) and \(a, b \in \mathbb {R}\). A vector \(y \in F\) is a weak integral of \(\gamma \) over \([a, b]\) if

\[ \varphi (y) = \int _a^b \varphi \bigl(\gamma (t)\bigr)\, dt \]

for every continuous linear functional \(\varphi \) on \(F\).

Lemma 4.2 Uniqueness
✓

Weak integrals are unique.

Proof ▶

Two weak integrals have the same image under every continuous linear functional, and these separate points by Hahn–Banach.

We write \(\int _a^b \gamma \) for the weak integral when it exists.

Remark 4.3
#

In Lean, the function that returns the weak integral is total: its value is \(0\) when the integral does not exist. It is only used where existence is known.

For continuous curves, weak integrals are linear in the integrand and additive in the interval: \(\int _a^b \gamma + \int _b^c \gamma = \int _a^c \gamma \). The constant curve \(v\) has integral \((b - a)v\). If \(L \colon F \to G\) is continuous and linear, then \(L\bigl(\int _a^b \gamma \bigr)\) is the weak integral of \(L \circ \gamma \).

Proof ▶

Apply a functional and use the corresponding rules for real integrals. For \(L\), it suffices to note that \(\varphi \circ L\) is a continuous linear functional on \(F\) whenever \(\varphi \) is one on \(G\). This is [ GN26 , 1.1.4 and 1.1.8 ] .

The next lemma replaces the estimate \(\| \int _a^b \gamma \| \leq (b - a) \sup \| \gamma \| \) of the normed case. Since there is no norm, the size of a vector is measured by the closed convex neighborhoods of \(0\) that contain it.

Lemma 4.5 Integrals of small curves are small
✓

Let \(a \leq b\) and let \(\gamma \) be continuous on \([a, b]\) with weak integral \(y\). If \(W \subseteq F\) is closed, convex, contains \(0\) and contains \(\gamma ([a, b])\), then \(y \in (b - a + 1)\, W\).

Proof ▶

If not, Hahn–Banach separates \(y\) from the closed convex set \((b - a + 1)W\) by a functional \(\varphi \). Bounding \(\int _a^b \varphi (\gamma )\) by \((b - a)\sup _W \varphi \) gives a contradiction. The constant \(b - a + 1\) instead of \(b - a\) avoids a separate treatment of the case \(a = b\). Compare [ GN26 , 1.1.9 ] .

Remark 4.6
#

When \(F\) is normed and \(W\) is the closed ball of radius \(r\) centered at \(0\), the lemma says that \(\| y\| \leq (b - a + 1)\, r\) whenever \(\| \gamma (t)\| \leq r\) on \([a, b]\). Up to the harmless constant, this is the classical estimate. For the applications in this document (continuity and convergence statements) only the shape of the bound matters, not the constant.

4.2 Change of variables

Let \(\phi \) be a real \(C^{1}\) function on \([a, b]\) and \(\gamma \) continuous on \(\phi ([a, b])\). Then \(y\) is a weak integral of \(t \mapsto \phi '(t)\, \gamma (\phi (t))\) over \([a, b]\) if and only if it is a weak integral of \(\gamma \) over \([\phi (a), \phi (b)]\).

Note that the statement is an equivalence: the existence of either integral implies the existence of the other. The case used most often is the affine one, \(\phi (t) = a + t(b - a)\) on \([0, 1]\), which shows that \(\int _a^b \gamma \) exists if and only if \(\int _0^1 (b - a)\, \gamma (a + t(b - a))\, dt\) does, and that both coincide.

Proof ▶

Apply a functional and use the change of variables formula for real integrals.

4.3 The fundamental theorem of calculus, first part

Let \(\gamma \) be differentiable on \([a, b]\), with continuous derivative \(\gamma '\). Then \(\gamma (b) - \gamma (a)\) is the weak integral of \(\gamma '\) over \([a, b]\).

Proof ▶

For every functional \(\varphi \), the real function \(\varphi \circ \gamma \) has continuous derivative \(\varphi \circ \gamma '\), and it suffices to apply the real fundamental theorem of calculus. This is [ GN26 , 1.1.5 ] .

We remark that no completeness of \(F\) is needed: the integral exists because \(\gamma (b) - \gamma (a)\) is a candidate. In other words, derivatives of \(C^{1}\) curves are always integrable; the difficulty of the existence problem lies with curves that are not known to be derivatives.

Corollary 4.9 Mean value theorem
✓

If \(f\) is \(C^{1}\) on a set containing the segment \([x, x + h]\), then \(f(x + h) - f(x) = \int _0^1 \mathrm{d}f(x + th)(h)\, dt\), the weak integral existing in \(F\).

Proof ▶

This is Proposition 2.10, read as a statement about weak integrals. See [ GN26 , 1.2.6 ] .

4.4 Integrals depending on parameters

We now let the integrand depend on a parameter \(x\) and study the regularity of \(x \mapsto \int _a^b f(x, t)\, dt\). As in the classical case, the arguments rest on the compactness of \([a, b]\), which turns pointwise closeness into uniform closeness in \(t\); the uniform estimate is then transferred to the integrals by Lemma 4.5.

Proposition 4.10 Continuity
✓

Let \(X\) be a topological space, \(U \subseteq X\) and \(T \subseteq \mathbb {R}\) open with \([a, b] \subseteq T\), and let \(f \colon U \times T \to F\) be continuous. If the weak integrals \(I(x) = \int _a^b f(x, t)\, dt\) exist for \(x \in U\), then \(I\) is continuous on \(U\).

Proof ▶

By compactness of \([a, b]\) (the tube lemma), for \(x'\) close to \(x\) the curve \(f(x', \cdot ) - f(x, \cdot )\) takes values in a given closed convex neighborhood \(W\) of \(0\) on \([a, b]\). By Lemma 4.5, \(I(x') - I(x) \in (b - a + 1)W\). Since the closed convex neighborhoods of \(0\) form a basis, this proves the claim. This is [ GN26 , 1.1.11 ] .

Theorem 4.11 Differentiation under the integral sign
✓

Let \(f \colon U \times T \to F\) be continuous, with partial derivatives \(D(x,t)(h) = \mathrm{d}f\bigl((x,t), (h, 0)\bigr)\) jointly continuous on \((U \times T) \times E\). Suppose that the weak integrals \(I(x) = \int _a^b f(x, t)\, dt\) exist for \(x \in U\), and that \(J = \int _a^b D(x, t)(h)\, dt\) exists. Then \(I\) has derivative \(J\) at \(x\) in the direction \(h\).

In other words, \(\mathrm{d}I(x)(h) = \int _a^b D(x, t)(h)\, dt\), as in the classical case, provided that both sides make sense. The existence of the integrals is part of the hypotheses; in a space with a suitable completeness property it is automatic (Chapter 5).

Proof ▶

For \(s \neq 0\), the difference quotient \(\frac{1}{s}\bigl(I(x + sh) - I(x)\bigr)\) is the weak integral of \(t \mapsto \frac{1}{s}\bigl(f(x + sh, t) - f(x, t)\bigr)\). By the mean value inclusion and the tube lemma, for small \(s\) this curve is uniformly close to \(t \mapsto D(x, t)(h)\) on \([a, b]\), and Lemma 4.5 gives the convergence. This is [ GN26 , 1.3.15 and 1.3.23 ] .