Classical and microlocal analysis of the x-ray transform on Anosov manifolds (Q2037473)

The paper completes the microlocal analysis of the x-ray transform for Anosov geodesic flows on Riemannian manifolds that started with [\textit{C. Guillarmou}, J. Differ. Geom. 105, No. 2, 177--208 (2017; Zbl 1372.37059)] and continued with [\textit{C. Guillarmou} and \textit{T. Lefeuvre}, Ann. Math. (2) 190, No. 1, 321--344 (2019; Zbl 1506.53054)]. The geometric setting is the unit tangent bundle \({\mathcal M}\) of a smooth closed Riemannian manifold \((M,g)\). Let \(X\) be the geodesic vector field on \({\mathcal M}\) and assume that the flow \(\varphi_t\) generated by \(X\) is transitive and Anosov (the latter is guaranteed when \(M\) has negative sectional curvature). Let \({\mathcal G}\) denote the collection of closed orbits of the Anosov flow and \(f\in C^0({\mathcal M})\). The x-ray transform \(If:{\mathcal G}\to {\mathbb R}\) is defined by \[ If (\gamma) = \frac{1}{\ell(\gamma)} \int_0^{\ell(\gamma)} f(\varphi_t x)\ dt, \] where \(x\in \gamma\) and \(\ell(\gamma)\) is the length of \(\gamma\). The Hölder regularity Livshits theorem [\textit{A. N. Livshits}, Math. USSR, Izv. 6, 1278--1301 (1974; Zbl 0273.58013)] characterizes those functions \(f\) that are in the kernel of the x-ray transform as coboundaries: for \(0<\alpha\leq 1\), if \(f\in C^\alpha({\mathcal M})\) is such that \(If = 0\), then there exists \(u\in C^\alpha({\mathcal M})\) with \(u\), differentiable in the flow direction, such that \(f = Xu\) with \(u\) unique up to an additive constant. There is a smooth version of the Livshits theorem established by \textit{R. De la Llave} et al. [Ann. Math. (2) 123, 537--611 (1986; Zbl 0603.58016)]. \textit{C. Guillarmou} [J. Differ. Geom. 105, No. 2, 177--208 (2017; Zbl 1372.37059)] proved a version of the Livshits theorem in Sobolev regularity in 2017, which implies that smooth version. The positive Livshits theorem for Anosov flows was proved by \textit{A. O. Lopes} and \textit{Ph. Thieullen} [Ergodic Theory Dyn. Syst. 25, No. 2, 605--628 (2005; Zbl 1078.37021)] in 2005. To state this positive version, a metric on \({\mathcal M}\) is needed. Define \(d:{\mathcal M}\times{\mathcal M}\to {\mathbb R}\) to be the Riemannian distance on \({\mathcal M}\) inherited from the Riemannian metric \(g\) on \(M\). The \(\alpha\)-Hölder norm of \(f\) is defined as \[ \Vert f\Vert_{C^\alpha} = \sup_{x\in {\mathcal M}} \vert f(x)\vert + \sup_{x,y\in{\mathcal M},x\ne y} \frac{ \vert f(x)-f(y)}{d(x,y)^\alpha}. \] The positive Livshits theorem established by \textit{A. O. Lopes} and \textit{Ph. Thieullen} is: for \(0<\alpha \leq 1\), there exist \(0<\beta\leq \alpha\) and \(C>0\) such that for all \(f\in C^\alpha({\mathcal M})\) there exists \(u\in C^\beta({\mathcal M})\), differentiable in the flow direction with \(Xu\in C^\beta({\mathcal M})\), and \(h\in C^\beta({\mathcal M})\) with \(h\geq 0\) such that \(f = Xu + h + m(f)\) where \(m(f) = \inf_{\gamma\in {\mathcal G}} If(\gamma)\), and \(\Vert Xu\Vert_{C^\beta} \leq C \Vert f\Vert_{C^\alpha}\). This paper establishes a finite approximate version of the Livshits theorem: for \(0<\alpha\leq 1\) there exist \(0<\beta\leq \alpha\) and \(\tau,C>0\) with the following properties. Let \(\varepsilon>0\). Consider a function \(f\in C^\alpha({\mathcal M})\) with \(\Vert f\Vert_{C^\alpha}\leq 1\) such that \(\vert If(\gamma)\vert \leq \epsilon\) for all \(\ell(\gamma) \leq \varepsilon^{-1/2}\). Then there exist \(u\in C^\beta({\mathcal M})\), differentiable in the flow direction with \(Xu\in C^\beta({\mathcal M})\), and \(h\in C^\beta({\mathcal M})\) such that \(f = Xu + h\) and \(\Vert u\Vert_{C^\beta} \leq C\) and \(\Vert h\Vert_{C^\beta} \leq C \varepsilon^\tau\). The constants \(\beta\), \(C\), and \(\tau\) do depend on the Anosov flow, but in a locally uniform manner. To apply the theorem the \(L^\infty\)-norm \[ \Vert If\Vert_{l^\infty} = \sup_{\gamma\in {\mathcal G}} \vert If(\gamma)\vert \] is used. In particular the assumptions of the theorem hold if \(\Vert If\Vert_{l_\infty} \leq \varepsilon\) for \(f\in C^\alpha{\mathcal M}\) satisfying \(\Vert f\Vert_{C^\alpha} \leq 1\), which leads to the global control \[ \Vert If\Vert_{l^\infty} \leq C \varepsilon^\tau. \] The x-ray transform extends to symmetric tensors. For \(f\in C^\infty(M,\otimes_S^m T^*M)\), a symmetric tensor, the pull-back \(\pi_m^* f(x,v) = f_x(v,v,\dots,v)\) is a smooth real-valued function on \({\mathcal M} = SM\), and the x-ray transform \(I_m\) of \(f\) is defined by \(I_mf = I \circ \pi_m^*f\). An \(f\in C^\infty(M,\otimes_S^m T^*M)\) has a unique decomposition as \(f=Dp + h\) for \(p\in C^\infty(M,\otimes_S^{m-1} T^*M)\) and \(h\in C^\infty(M\to \otimes_S^m T^*M)\) satisfying \(D^*h = 0\), where \(D=\sigma\circ\nabla\) is the derivation of symmetric tensors for \(\sigma\) the operator of symmetrization of tensors and \(\nabla\) the Levi-Civita connection, and \(D^*\) is the formal adjoint of \(D\). The part \(Dp\) of the decomposition is called the potential part and satisfies \(I_m(Dp) = 0\). The part \(h\) of the decomposition is called a solenoidal tensor. The map \(I_m\) is solenoidal injective, or in short \(s\)-injective, if the restriction of \(I_m\) to \[ C^\infty_{\mathrm{sol}}(M,\otimes_S^m T^*M) = C^\infty(M,\otimes_S^m T^*M) \cap \operatorname{ker}(D^*) \] is injective. It is conjectured that \(I_m\) is \(s\)-injective for all Riemannian manifolds whose geodesic flow is Anosov, in any dimension and without any assumption on the curvature. If the sectional curvatures are nonpositive, then \(I_m\) is \(s\)-injective. Without any assumptions on the sectional curvatures, the map \(I_m\) is \(s\)-injective when \(m=0,1\) and \(M\) is a surface. For higher order tensors without any conditions on the curvature, the conjecture remains wide open. This paper provides the following results when \(I_m\) is assumed to be \(s\)-injective. The first is a stability result and is as follows. Assume \(I_m\) is \(s\)-injective. Then for all \(0<\beta<\alpha<1\) there exists \(\theta_1 = \theta_1(\alpha,\beta)>0\) and \(C=C(\alpha,\beta)>0\) such that if \(f\in C^\alpha(M,\otimes_S^m T^*M)\cap \operatorname{ker}(D^*)\) is a solenoidal tensor such that \(\Vert f\Vert_{C^\alpha}\leq 1\), then \(\Vert f\Vert_{C^\beta} \leq C \Vert I_mf\Vert_{l^\infty}^{\theta_1}\). The proof of this result depends on the properties of the generalized x-ray transform \(\Pi_m\), properties such as \(\Pi_m\) is a pseudo-differential operator of order \(-1\) that is elliptic on solenoidal symmetric tensors, and \(s\)-injectivity of \(I_m\) implies the injectivity of \(\Pi_m\) on the Sobolev spaces of symmetric tensors each intersected with \(\operatorname{ker}(D^*)\). The second result combines previous theorems and is as follows. Assume \(I_m\) is \(s\)-injective. Then for all \(0<\beta<\alpha<1\) there exists \(\theta_2 = \theta_2(\alpha,\beta)>0\) and \(C=C(\alpha,\beta)>0\) such that for any \(L>0\) large enough, if \(f\in C^\alpha(M,\otimes_S^m T^*M)\cap \operatorname{ker}(D^*)\) is solenoidal and satisfies \(\Vert f\Vert_{C^\alpha}\leq 1\) and \(I_m f(\gamma) = 0\) for all closed geodesics \(\gamma \in {\mathcal G}\) such that \(\ell(\gamma)\leq L\), then \(\Vert f\Vert_{C^\beta} \leq CL^{-\theta_2}\).