What the local Cauchy data set actually records
The measurement is not a boundary map. On a smooth closed connected Riemannian manifold of dimension at least two, the fractional Laplacian is defined by the spectral theorem for each exponent between zero and one, and the fractional Schrödinger operator pairs that nonlocal power with a smooth complex-valued potential. The Cauchy data set on a nonempty open set collects the pairs consisting of a smooth solution restricted to that set together with its fractional Laplacian restricted to the same set, where the equation is imposed only on the complement of the observation set.
Two features of this definition drive the result. First, the equation is required to hold off the observation set, so the inaccessible region is where the operator is actually constrained. Second, the data record the fractional Laplacian of the solution, not the full operator applied to it. That distinction matters because the potential is invisible in the second component: adding any smooth compactly supported function to the potential on the observation set leaves the Cauchy data set unchanged. The paper states this explicitly, which is why the theorem assumes the two potentials already agree on the observation set.
Why the potential on the observation set must be assumed
The paper proves that the values of the potential on the observation set cannot be recovered from this data set. For every smooth compactly supported function on that set, adding it to the potential produces exactly the same Cauchy data. So the assumption that the two potentials coincide on the observation set is not a technical convenience; it is necessary for a global potential-recovery statement from this data.
The paper also records an alternative measurement that removes this assumption. If one instead measures the solution restricted to the observation set together with the full fractional Schrödinger operator applied to the solution, again with the equation imposed off the observation set, then equality of those data determines the entire potential, including its values on the observation set, as well as the manifold and metric. The reason is that adding a common compactly supported function on the observation set now shifts the second component by that function times the solution, and the leading-symbol argument recovers the metric on the observation set, after which the difference of the two potentials is smoothing there and hence vanishes. The two data sets answer different questions, and the paper keeps them separate.
Recovering the local resolvent from fractional data
The proof does not attack the nonlocal problem directly. It first removes the kernels of the two fractional Schrödinger operators by adding a common real-valued compactly supported perturbation on the observation set, which leaves both Cauchy data sets unchanged and makes the resulting operators invertible. Equality of the Cauchy data then gives equality of the local source-to-solution maps, and their leading symbols determine the metrics on the observation set.
The central step is a comparison principle. Once the metrics agree on the observation set, two smooth functions on the respective manifolds whose fractional Cauchy pairs match there have identical heat evolutions on that set for every positive time. The functions need not vanish on the observation set or be supported there. The proof combines finite speed of propagation for the wave equation with the Gaussian representation of the heat semigroup to get exponential smallness of the difference of the heat evolutions and all its time derivatives, locally uniformly on the observation set, for short times. A moment argument then extends this to all positive times.
Applying this comparison principle to the local source-to-solution maps and taking a Laplace transform in time yields equality on the observation set of the resolvents of the Laplace–Beltrami operators composed with those maps. A bilinear transposition reverses the order of the two inverse operators, which is what allows complex-valued potentials. A second application of the comparison principle, combined with an ordinary differential equation in time and the heat semigroup representation of the fractional Laplacian, eliminates the source-to-solution factor and leaves equality of the local resolvents of the Laplacians themselves.
From local resolvents to heat kernels and the metric
Equality of the local resolvents for every positive spectral parameter, together with the heat semigroup representation of the resolvent and uniqueness of the Laplace transform for bounded continuous functions, gives equality of the local heat kernels on the observation set for all positive times. This is a statement about the heat kernel as a function of two points in the observation set, not merely about heat evolutions of individual functions.
The paper then invokes a geometric determination result for heat kernels to obtain a smooth diffeomorphism between the two manifolds that restricts to the identity on the observation set and pulls the second metric back to the first. The observation set may be any nonempty open subset, and it need not be connected or have a regular boundary. The metric on the observation set is recovered as part of the argument rather than assumed known in advance.
Recovering the potential off the observation set
After identifying the manifolds by the recovered isometry, the paper uses fractional unique continuation to show that solutions with matching Cauchy pairs coincide. A single solution corresponding to a nonzero source supported in the observation set then determines the potential on the inaccessible region. The set where that solution is nonzero is dense in the complement of the observation set, so the difference of the two potentials vanishes there, and continuity extends the equality to the whole manifold. The known values of the potential on the observation set complete the recovery.
The theorem also covers the degenerate case where the observation set is dense in either manifold. Then the equation off the observation set is vacuous and the Cauchy data set is independent of the potential, but the recovery of the manifold and metric still follows, and the potential equality then holds everywhere by continuity. The paper notes that the case of primary interest is the one where the complement of the observation set is nonempty, so that the equation is imposed on a genuinely inaccessible region.
How this differs from the local Calderón problem
The problem is a nonlocal analogue of the geometric anisotropic Calderón problem, where one seeks to determine a Riemannian metric from boundary Cauchy data of harmonic functions. That local problem remains open for general smooth metrics in dimensions three and higher, with known resolutions covering real-analytic settings and, for a conformal factor, suitable transversally anisotropic geometries. The fractional setting behaves differently because local measurements retain information about global spectral data.
The paper's own earlier work obtained the first simultaneous recovery of manifold, metric, and potential in this setting, but under additional assumptions. The present proof takes a different route: it recovers the local action of the resolvent of the Laplace–Beltrami operator for every positive spectral parameter, giving a direct path from fractional Cauchy data to geometric determination. The paper also notes that an additive potential outside the fractional power is harder than recovering fractional powers of a local operator that already contains lower order terms, because the fractional power and the potential do not commute in general.
What the result does and does not settle
The theorem is a uniqueness statement, not a stability statement. It says that equality of the Cauchy data forces the two manifolds, metrics, and potentials to be related by an isometry fixing the observation set, and conversely that such an isometry preserves the data. It does not quantify how much the recovered objects can change when the data change by a small amount, and the paper does not claim such a bound.
The scope is also specific. The manifolds are smooth, closed, and connected, of dimension at least two, and the potential is smooth and complex-valued. The observation set is any nonempty open subset, with no connectedness or boundary regularity required. The simultaneous recovery problem for the fractional Schrödinger operator with external data in Euclidean space is treated in a forthcoming paper rather than here. Within these boundaries, the sharpest practical distinction the paper draws is between the two measurement protocols: the data set that records the fractional Laplacian of the solution leaves the potential on the observation set undetermined and therefore requires the potentials to agree there, while the data set that records the full fractional Schrödinger operator applied to the solution determines the entire potential, including its values on the observation set, without that assumption. For a reader deciding what to measure, that choice of second component is the decisive one.
Sources used in this answer
Fractional Calderón problem with a potential
Local Cauchy data for a fractional Schrödinger equation on a smooth closed connected Riemannian manifold of dimension at least two determine the manifold, its metric, and a smooth complex-valued potential up to an isometry fixing the observation set, for any nonempty open observation set and with no further geometric assumptions.