= Solution
A <symplectic manifold> is a smooth even-dimensional manifold $X^{2n}$ with a closed, nondegenerate <differential two-form> $\omega$. A <Lagrangian submanifold> $L\subset X$ has dimension $n$ and $\omega|_{TL}=0$; equivalently $T_xL=(T_xL)^\omega$ at every point.
On $T^*M$, let $\pi:T^*M\to M$ and define the <canonical one-form on a cotangent bundle> by $\theta_{(q,p)}(v)=p(d\pi(v))$. In local coordinates $\theta=\sum_i p_i\,dq_i$, so our sign convention gives
$$
\omega_0=-d\theta=\sum_i dq_i\wedge dp_i.
$$
The intrinsic definition of $\theta$ makes this form globally defined. It is closed because $d^2=0$, and its displayed coordinate matrix is invertible. The <zero section> pulls $\theta$, and therefore $\omega_0$, back to zero and has half the total dimension. It is a <Lagrangian submanifold>.
We prove the <Weinstein neighborhood theorem>, rather than merely invoking it. Along a <Lagrangian submanifold> $L\subset(X,\omega)$, choose a <compatible almost complex structure> $J$; its existence is proved by the metric construction in Question 5. Then $JL$ at the tangent-bundle level means $J(TL)$, and
$$
TX|_L=TL\oplus J(TL).
$$
Indeed, $Jv\in TL$ with $v\in TL$ would imply $0=\omega(v,Jv)$, forcing $v=0$. Also $\omega(Ju,Jv)=\omega(u,v)=0$ for $u,v\in TL$, so $J(TL)$ is a <Lagrangian complement>. Identify a normal vector $n\in J(TL)$ with the covector $\eta(v)=\omega(v,n)$. This is a vector-bundle <isomorphism> with $T^*L$, and under it
$$
\omega((v,\eta),(w,\zeta))=\zeta(v)-\eta(w),
$$
which is exactly the form $\omega_0$ along the <zero section>. The <tubular neighborhood theorem> therefore supplies a smooth map $F$ from a neighborhood of that section to a neighborhood of $L$, restricting to the identity and with this derivative along $L$.
Write $\omega_1=F^*\omega$ and $\delta=\omega_1-\omega_0$. This <closed differential form> vanishes as an entire <bilinear form> along the <zero section>. The <relative Poincaré lemma> gives $\delta=d\beta$ with $\beta$ zero along the section. Explicitly, if $r_s(q,p)=(q,sp)$ and $R=\sum p_i\partial_{p_i}$, use
$$
\beta=\int_0^1\frac1s r_s^*(\iota_R\delta)\,ds.
$$
The homotopy identity gives $d\beta=\delta-r_0^*\delta=\delta$; the vanishing at the section ensures smoothness of this integral. The forms $\omega_t=\omega_0+t\delta$ are nondegenerate on a sufficiently small neighborhood of the section for all $0\leq t\leq1$. Solve uniquely
$$
\iota_{Y_t}\omega_t=-\beta.
$$
The <vector field> $Y_t$ vanishes on the section. After another neighborhood shrink its <flow> $\phi_t$ exists up to time one and fixes the section. For noncompact $L$, the sizes of these neighborhoods may vary with the base point; no <compactness> of $L$ is required for the local conclusion. By <Cartan's magic formula>,
$$
\frac d{dt}\phi_t^*\omega_t
=\phi_t^*(\delta+d\iota_{Y_t}\omega_t)=0.
$$
Thus $(F\circ\phi_1)^*\omega=\omega_0$. This establishes the universal neighborhood model, with the <Lagrangian submanifold> identified with the <zero section>.
For the section $s_\alpha(q)=(q,\alpha_q)$, the canonical <differential one-form> satisfies $s_\alpha^*\theta=\alpha$. Hence
$$
\boxed{s_\alpha^*\omega_0=-d\alpha,\qquad\Gamma_\alpha\text{ is Lagrangian}\iff d\alpha=0.}
$$
Its dimension is already half that of $T^*M$, so vanishing of this pullback is sufficient as well as necessary.
Suppose now that a smooth open six-dimensional region were a <fiber bundle> with <Lagrangian submanifolds> diffeomorphic to $S^3$ as its fibres. Fix one compact fibre $L$. Smooth local triviality and <compactness> imply that the other sufficiently nearby fibres are $C^1$-close embeddings of $L$. In the <Weinstein neighborhood>, projection to $L$ restricts on each such fibre to a <local diffeomorphism> close to the identity. It has degree one, and <compactness> makes it a <covering map>, so it is a <diffeomorphism>. Each nearby fibre is therefore a graph of a <closed differential one-form> $\alpha$ on $S^3$. Since $H^1(S^3;\mathbb R)=0$, this is $df$ for a smooth real function $f$. A maximum of $f$ exists by <compactness>, and there $df=0$. The graph meets the original <zero section>, contradicting disjointness of different fibres. \b[No such Lagrangian three-sphere fibration exists.] More generally this is the <compact first-cohomology obstruction to a Lagrangian fibration>.
Back to article page
