An inverse source problem in optical molecular imaging

Similar documents
Microlocal Methods in X-ray Tomography

A SHARP STABILITY ESTIMATE IN TENSOR TOMOGRAPHY

Recent progress on the explicit inversion of geodesic X-ray transforms

Microlocal analysis and inverse problems Lecture 4 : Uniqueness results in admissible geometries

Numerical Methods for geodesic X-ray transforms and applications to open theoretical questions

Recovery of anisotropic metrics from travel times

Inversions of ray transforms on simple surfaces

MICROLOCAL ANALYSIS METHODS

Elliptic Regularity. Throughout we assume all vector bundles are smooth bundles with metrics over a Riemannian manifold X n.

c 2003 International Press Vol. 10, No. 1, pp , March

arxiv: v2 [math.ap] 27 Jul 2016

Short note on compact operators - Monday 24 th March, Sylvester Eriksson-Bique

THE GEODESIC RAY TRANSFORM ON RIEMANNIAN SURFACES WITH CONJUGATE POINTS

Double Layer Potentials on Polygons and Pseudodifferential Operators on Lie Groupoids

The oblique derivative problem for general elliptic systems in Lipschitz domains

BOUNDARY RIGIDITY AND STABILITY FOR GENERIC SIMPLE METRICS

LECTURE NOTES ON GEOMETRIC OPTICS

Index theory on manifolds with corners: Generalized Gauss-Bonnet formulas

An inverse scattering problem in random media

On stable inversion of the attenuated Radon transform with half data Jan Boman. We shall consider weighted Radon transforms of the form

THERMOACOUSTIC TOMOGRAPHY ARISING IN BRAIN IMAGING

Photo-Acoustic imaging in layered media

Boundary problems for fractional Laplacians

Lecture 2: Some basic principles of the b-calculus

Chapter One. The Calderón-Zygmund Theory I: Ellipticity

UNIVERSITY OF MANITOBA

The Helmholtz Equation

THE BOUNDARY RIGIDITY PROBLEM IN THE PRESENCE OF A MAGNETIC FIELD

A new class of pseudodifferential operators with mixed homogenities

Local smoothing and Strichartz estimates for manifolds with degenerate hyperbolic trapping

Microlocal Analysis : a short introduction

Lecture 9 February 2, 2016

Exact Solutions of the Einstein Equations

Algebras of singular integral operators with kernels controlled by multiple norms

Travel Time Tomography and Tensor Tomography, I

Fast and accurate methods for the discretization of singular integral operators given on surfaces

Fractional Index Theory

On a class of pseudodifferential operators with mixed homogeneities

We denote the space of distributions on Ω by D ( Ω) 2.

Stability estimates in stationary inverse transport

Continuous dependence estimates for the ergodic problem with an application to homogenization

Zeta Functions and Regularized Determinants for Elliptic Operators. Elmar Schrohe Institut für Analysis

Introduction to Index Theory. Elmar Schrohe Institut für Analysis

Lecture 8: Boundary Integral Equations

Inverse problems for hyperbolic PDEs

Hyperbolic inverse problems and exact controllability

Notes for Elliptic operators

Overview I. Approaches to Reconstruction in Photoacoustic Tomography. Bell and the Photoacoustic Effect I. Overview II

INSTANTON MODULI AND COMPACTIFICATION MATTHEW MAHOWALD

HARMONIC ANALYSIS. Date:

Integral Representation Formula, Boundary Integral Operators and Calderón projection

Walsh Diffusions. Andrey Sarantsev. March 27, University of California, Santa Barbara. Andrey Sarantsev University of Washington, Seattle 1 / 1

The first order quasi-linear PDEs

The X-ray transform for a non-abelian connection in two dimensions

Deforming conformal metrics with negative Bakry-Émery Ricci Tensor on manifolds with boundary

AN INVERSE PROBLEM FOR THE WAVE EQUATION WITH A TIME DEPENDENT COEFFICIENT

MICROLOCAL APPROACH TO TENSOR TOMOGRAPHY AND BOUNDARY AND LENS RIGIDITY

YERNAT M. ASSYLBEKOV AND PLAMEN STEFANOV

u xx + u yy = 0. (5.1)

Inverse Scattering with Partial data on Asymptotically Hyperbolic Manifolds

A local estimate from Radon transform and stability of Inverse EIT with partial data

THE WEIGHTED DOPPLER TRANSFORM

Incoming and disappearaing solutions of Maxwell s equations. Université Bordeaux I

Fractional order operators on bounded domains

Inverse Transport Problems (Generalized) Stability Estimates. Guillaume Bal

The Poisson boundary of certain Cartan-Hadamard manifolds of unbounded curvature

Inverse problems Total Variation Regularization Mark van Kraaij Casa seminar 23 May 2007 Technische Universiteit Eindh ove n University of Technology

Reconstructing inclusions from Electrostatic Data

Analysis in weighted spaces : preliminary version

Some Aspects of Solutions of Partial Differential Equations

Hamburger Beiträge zur Angewandten Mathematik

Green s Functions and Distributions

Fourier Transform & Sobolev Spaces

THE DIRICHLET PROBLEM WITH BM O BOUNDARY DATA AND ALMOST-REAL COEFFICIENTS

9 Radon-Nikodym theorem and conditioning

Optical Tomography on Simple Riemannian Surfaces

Fast algorithms for integral formulations of steady-state radiative transfer equation

Bielefeld Course on Nonlinear Waves - June 29, Department of Mathematics University of North Carolina, Chapel Hill. Solitons on Manifolds

EXPOSITORY NOTES ON DISTRIBUTION THEORY, FALL 2018

In this chapter we study elliptical PDEs. That is, PDEs of the form. 2 u = lots,

Singular integrals on NA groups

On the Interior Boundary-Value Problem for the Stationary Povzner Equation with Hard and Soft Interactions

Existence, stability and instability for Einstein-scalar field Lichnerowicz equations by Emmanuel Hebey

Adiabatic limits and eigenvalues

Elliptic Operators with Unbounded Coefficients

Optical Tomography for Variable Refractive Index with Angularly Averaged Measurements

Can There Be a General Theory of Fourier Integral Operators?

Lecture Notes on PDEs

RECENT PROGRESS ON THE BOUNDARY RIGIDITY PROBLEM

Micro-local analysis in Fourier Lebesgue and modulation spaces.

A review: The Laplacian and the d Alembertian. j=1

Mathematical Tripos Part III Michaelmas 2017 Distribution Theory & Applications, Example sheet 1 (answers) Dr A.C.L. Ashton

On m-accretive Schrödinger operators in L p -spaces on manifolds of bounded geometry

Travelling bubbles in a moving boundary problem of Hele-Shaw type

LECTURE 3 Functional spaces on manifolds

On uniqueness in the inverse conductivity problem with local data

Determinant of the Schrödinger Operator on a Metric Graph

Regularizations of Singular Integral Operators (joint work with C. Liaw)

THE BOUNDARY RIGIDITY PROBLEM IN THE PRESENCE OF A MAGNETIC FIELD

CHAPTER 2. Laplace s equation

Transcription:

An inverse source problem in optical molecular imaging Plamen Stefanov 1 Gunther Uhlmann 2 1 2 University of Washington

Formulation Direct Problem Singular Operators Inverse Problem Proof Conclusion Figure: The first and the last picture of this talk

Formulation Formulation of the problem The radiative transport equation in Ω is given by Z θ xu(x, θ)+σ(x, θ)u(x, θ) k(x, θ, θ )u(x, θ ) dθ = f (x), u SΩ = 0, S n 1 where σ is the absorption and k is the collision kernel. The source term f is assumed to depend on x only. Here, SΩ consists of x Ω and θ pointing inwards. The boundary measurements are modeled by Xf (x, θ) = u +SΩ, (x, θ) +SΩ, where u(x, θ) is a solution of the transport equation, and +SΩ denotes the points x Ω with direction θ pointing outwards.

Formulation Direct Problem Given f (and σ, k), find Xf. Inverse Problem Given Xf (and σ, k), find f. Clearly, it is a linear problem. Let σ = k = 0 first. Then X is just the X-ray transform: Xf (x, θ) = If (x, θ) := Z 0 τ (x,θ) f (x + tθ) dt, (x, θ) +SΩ (σ = k = 0),

Formulation Let k = 0 only. Then we get a weighted X-ray transform: Z Xf (x, θ) = I σf (x, θ) := E(x + tθ, θ)f (x + tθ) dt, (x, θ) +SΩ (k = 0), where E(x, θ) = exp Z 0 «σ(x + sθ, θ) ds. If σ = σ(x), then we get the attenuated X-ray transform, that we know how to invert. Without assuming that any one is zero, Bal and Tamasan proved injectivity when k = k(x, θ θ ), and k is small enough in a suitable norm. The main idea there is to treat k as a perturbation; then X is a perturbation of the attenuated X-ray transform. Also, results by Sharafutdinov on Riemannian manifolds, smallness conditions on the curvature k and σ. Our goal is to consider this problem for general (σ, k).

The Direct Problem: Generic Solvability The direct problem first We need assumptions, even for solvability of the direct problem! Assuming k 1 is enough. Also, R k(, θ, )dθ < σ suffices. Those conditions prevent a nuclear explosion, i.e., the corresponding time-dependent dynamics is bounded. They are not necessary conditions though. Theorem 1 (a) The direct problem is uniquely solvable for a dense open set of pairs (σ, k) in C 2, even for f = f (x, θ). (b) X : L 2 (Ω S n 1 ) L 2 ( +SΩ, dσ). Here, dσ = ν θds xdθ.

Sketch of the Proof Set T 0 = θ x, T 1 = T 0 + σ, T = T 0 + σ K, where σ and K are viewed as operators corresponding to the σ and to the k terms in the transport equation. Then T 0, T 1, T can be defined as closed operators acting on functions f satisfying f SΩ = 0. Then Z 0 [T 1 1 f ](x, θ) = exp Z 0 s «σ(x + τθ, θ) dτ f (x + sθ, θ) ds. Why did not we start with T 1 0? It turns out that T is a kind of relatively compact perturbation of T 1 but not of T 0. So, we have to think of σ as a part of the principal symbol but k is of low order. We now apply T 1 1 to both sides of Tu = f, u SΩ = 0. to get u = T 1 1 (Ku + f ) = (Id T 1 1 K)u = T 1 1 f. Therefore, if we can invert Id T 1 1 K, we can solve the direct problem.

Sketch of the Proof This is also equivalent to the existence of (Id KT 1 1 ) 1. Set [Jf ](x, θ) := f (x), Then (with f = f (x) independent of θ), and R +h = h +SΩ u = T 1 Jf = (Id T 1 1 K) 1 T 1 1 Jf, Xf = R +T 1 1 (Id KT 1 1 ) 1 Jf. Lemma 2 The operator KT 1 1 J : L 2 (Ω) L 2 (Ω S n 1 ) is compact. Proof. It is an integral operator of the form Z Σ x, x y, x y k x, θ, x y x y x y f (y) dy, x y n 1 where Ω Z 0 «Σ(x, s, θ ) = exp σ(x + τθ, θ ) dτ. s

Calderón-Zygmund Theorem Let K be an integral operator with kernel k(x, y) = φ(x, θ)r n, where θ = (x y)/ x y, r = x y, i.e., formally, [Kf ](x) = Z φ x, x y x y f (y) dy, x y n and the characteristic φ has mean value 0 as a function of θ, for any x. Then K is well defined, where the integral has to be understood in the principle value sense. Moreover, K L 2 (Ω) L 2 C sup φ(x, ) L 2 (S n 1 ), x The characteristic φ does not need to have zero mean value in θ but then the integral has to be considered as a convolution in distribution sense. The latter is well defined because the Fourier transform of the kernel w.r.t. the variable z = rθ is homogeneous of order 0, thus bounded.

Compactness of weakly singular operators Compactness of weakly singular operators Also, if B is an operator with a weakly singular kernel ψ(x, θ)r n+1, then xb is an integral operator with singular kernel x[β(x, θ)r n+1 ]. The latter, up to a weakly singular operator, has a singular kernel of the type φr n, and the integration is again understood in the principle value sense, see the next paragraph. In particular, the zero mean value condition is automatically satisfied. Informally speaking, operators with kernels of the kind x y n+1 map L s (Ω) into H 1, and are therefore compact as operators from L 2 (Ω) to itself. In particular, this proves the lemma.

Another operator is compact! If f = f (x, θ), then Z Σ x, x y, x y k x, θ, x y [KT 1 x y x y 1 f ](x, θ) = f x y n 1 y, «x y dy. x y No enough integrations to make it compact! To invert Id KT 1 1 for generic k s, we want to apply the Analytic Fredholm Theorem. But KT 1 1 is not compact. Its square however is! Lemma 3 The operator KT 1 1 K : L 2 (Ω S n 1 ) L 2 (Ω S n 1 ) is compact. Sketch of the Proof: ZZΩ Sn 1 α x, y, x y, x y, θ, [KT 1 x y θ 1 Kf ](x, θ) = f (y, θ ) dy dθ x y n 1 Looks like a weakly singular operator but not exactly.

Generalized Calderón-Zygmund type of theorem This calls for a generalization of the Calderón-Zygmund arguments. Lemma 4 Let A be the operator [Af ](x) = Z α x, y, x y, x y x y f (y) dy x y n 1 with α(x, y, r, θ) compactly supported in x, y. Then (a) If α C 2, then A : L 2 H 1 is continuous with a norm not exceeding C α C 2. (b) Let α(x, y, r, θ) = α (x, y, r, θ)φ(θ). Then A L 2 H 1 C α C 2 φ H 1 (S n 1 ). Now we can prove the previous lemma.

End of the Proof of Thm 1 End of the Proof of Thm 1 Instead of trying to invert Id KT 1 1, we will invert its square (and then write P 1 = P(P 2 ) 1 ). Set A(λ) = Id 1 2 1 `λkt1. By the Analytic Fredholm Theorem, for any fixed (σ, k), A(λ) is a meromorphic family. Thm 1 easily follows from this.

Inverse Problem The Inverse Problem Fix Ω 1 Ω. Define X 1 as X but in Ω 1. Theorem 5 For (σ, k) in an open and dense set of pairs in C 2 ( Ω S n 1 ) C 2 Ωx S n 1 θ ; C n+1 (S n 1 θ ), including (0, 0), the direct problem is solvable in Ω 1, and (a) the map X 1 is injective on L 2 (Ω), (b) the following stability estimate holds f L 2 (Ω) C X 1 X 1f H 1 (Ω 1 ), f L 2 (Ω), with a constant C > 0 locally uniform in (σ, k).

Sketch of the Proof Sketch of the Proof Denote X 1 = X again. Outline If k = 0, then X X = I σ I σ, that is an elliptic ΨDO, and the theorem is true (Frigyik, Uhlmann & S). k is responsible for a relative compact perturbation: X X = I σ I σ + L, with L : L 2 H 1 smoothing, i.e., L : L 2 H 2. Apply a parametrix Q of I σ I σ to X X : QX X = QI σ I σ + QL = Id + K with K one degree smoothing (and therefore compact). We get a Fredholm problem: invert Id + K, with K = K σ,k. For σ analytic, Id + K σ,0 is invertible. Consider now Id + K σ,λk for λ [0, 1]. Still invertible for generic λ s.

Sketch of the Proof Why is the theorem true for k = 0? Then Iσ I σ has a weakly singular kernel: Z [Iσ α(x, y) I σf ](x) = f (y) dy x y n 1 and is also an elliptic ΨDO of order 1. It is in the analytic class for σ analytic. If A is an elliptic ΨDO on Ω 1, if supp f Ω, and if Af = 0, then f is analytic in Ω 1. It also has compact support. Therefore, f = 0. Therefore, I σ I σ is injective on L 2 (Ω) for σ analytic. Apply a parametrix becomes a Fredholm problem. Therefore, stable under perturbations. Thus, I σ I σ is injective for an open dense set of σ s. The stability estimate also follows from the Fredholm Theory.

L : L 2 H 2 Proof of L : L 2 H 2 We write X = I σ + L, 1 L := R +T 1 1 KT 1 1 Id KT 1 1 J. Then X X = I σ I σ + L, L := I σ L + L I σ + L L. First (and most important) step: Can we prove that I σ L : L 2 H 2? A model operator A : h(x, θ) [Af ](x): [Ah](x) = ZZ α x, w, y, x w, w y, x w, w y x w w y w y h y, h (y) dy dw x w n 1 w y n 1 w y A : L 2 H 2? Not clear. Not so clear even when h is independent of θ...

L : L 2 H 2 [Ah](x) = ZZ α x, w, y, x w, w y, x w, w y x w w y x w h y, h (y) dy dw. x w n 1 w y n 1 w y If h = h(x) is independent of θ, and if α is of product type: «α x, w, y, x w, w y, x w, w y x w w y = α 1 x, w, x w, x w α x w 2 w, y, w y, w y, w y then we have a composition of two operators, each one of order 1, and we are done by our generalized C-Z theorem. This makes the statement believable. α is of product type, if, for example, k(x, θ, θ ) = Θ(θ)κ(x, θ ). In general, this is not true, but nothing prevents us from studying (infinite) linear combinations of terms like this.

L : L 2 H 2 We can always write k in the form k(x, θ, θ ) = X Θ j (θ)κ j (x, θ ). (1) j=1 For example, choose Θ j to be the spherical harmonics. Then the series converges in L 2 but for our analysis we need convergence in C 1 (almost). This explains the need for higher regularity in the θ variable. Now, each term in (1) contributes a compact operator; we have uniform convergence; so we are done. Well, we still have to deal with the fact that h depends on θ as well...

Conclusion If we want to recover the singularities of f only, up to order 1, then it is enough to apply to X Xf a parametrix of I σ I σf, that is explicit and depends on σ only. To actually find f from Xf = h, we still apply the explicit parametrix Q above, and get a Fredholm problem (Id + K)f = QX h, where Q and K are computable, and h is given. It seems strange that k plays no role in the recovery of the singularities of f (up to order 1). In reality, Z σ = σ true absorption + k(, θ, ) dθ. Therefore, k is involved in that step.