Ferkans — Interactive Telecom Tutor

Why Distributions and Sobolev Spaces?

Classical calculus requires functions to be differentiable. But many objects central to RF imaging are not:

A point scatterer at $\mathbf{r}_0$ has scene function $c(\mathbf{r}) = \gamma\,\delta(\mathbf{r} - \mathbf{r}_0)$ — a Dirac delta, which is not a function in any $L^p$ sense.
The Heaviside step models sharp edges in SAR urban scenes.
The Green's function of the Helmholtz operator has a non-integrable singularity at the source point.

Distributions extend classical calculus to handle these rigorously. Sobolev spaces provide the function-space language for PDEs and give us a way to encode prior smoothness assumptions in reconstructions.

This section provides the rigorous foundation for Chapter 5's electromagnetic scattering theory.

Historical Note: Laurent Schwartz and the Theory of Distributions

1944–1951

Laurent Schwartz (1915–2002) introduced the theory of distributions (also called "generalised functions") in 1944–45, publishing the foundational monograph Théorie des distributions in 1950–51. The Fields Medal in 1950 recognised this as the deepest advance in mathematical analysis since Lebesgue's integration theory.

Schwartz's key insight: instead of defining the Dirac delta as a "function" that is zero everywhere except at one point with infinite height, define it as a functional — a rule for assigning a number to every smooth function. This functional viewpoint unifies the delta function, its derivatives, and all classical functions under one framework.

Paul Dirac had introduced the delta function in physics in 1927 (in The Principles of Quantum Mechanics), but without mathematical rigour. Schwartz's theory validated decades of physicist's calculations and provided the rigorous setting for quantum field theory, the wave equation, and — directly relevant here — the Green's function of the Helmholtz operator.

Definition:
Test Functions and the Space $\mathcal{D}(\Omega)$

The space of test functions is

$\mathcal{D}(\Omega) = C_c^\infty(\Omega),$

the set of all infinitely differentiable functions $\varphi: \Omega \to \mathbb{C}$ with compact support contained in $\Omega$ (i.e., $\varphi$ vanishes outside some bounded closed subset of $\Omega$ ).

A sequence $\varphi_k \to \varphi$ in $\mathcal{D}(\Omega)$ if: (a) all $\varphi_k$ are supported in a common compact set, and (b) all derivatives converge uniformly: $D^\alpha \varphi_k \to D^\alpha\varphi$ uniformly for every multi-index $\alpha$ .

Definition:
Distributions — $\mathcal{D}'(\Omega)$

A distribution (or generalised function) $T$ on $\Omega$ is a continuous linear functional on $\mathcal{D}(\Omega)$ :

$T: \mathcal{D}(\Omega) \to \mathbb{C}, \quad \varphi \mapsto \langle T, \varphi \rangle \triangleq T(\varphi).$

The space of all distributions is $\mathcal{D}'(\Omega)$ .

Every locally integrable function $f \in L^1_{\mathrm{loc}}(\Omega)$ defines a regular distribution via

$\langle T_f, \varphi \rangle = \int_\Omega f(\mathbf{r})\, \varphi(\mathbf{r})\,d\mathbf{r}.$

Distributions that do not arise this way (e.g., the Dirac delta) are called singular distributions.

Example: The Dirac Delta as a Distribution

Show that the Dirac delta $\delta_{\mathbf{r}_0}$ is a distribution, verify it is not a regular distribution (i.e., not an $L^1$ function), and explain its role in RF imaging.

Solution

Definition as a functional

Define $\delta_{\mathbf{r}_0}$ by its action on test functions:

$\langle \delta_{\mathbf{r}_0}, \varphi \rangle = \varphi(\mathbf{r}_0).$

This is clearly linear. Continuity follows because $\varphi_k \to \varphi$ in $\mathcal{D}(\Omega)$ implies $\varphi_k(\mathbf{r}_0) \to \varphi(\mathbf{r}_0)$ (pointwise evaluation is continuous under uniform convergence). Hence $\delta_{\mathbf{r}_0} \in \mathcal{D}'(\Omega)$ .

Not a regular distribution

Suppose $\delta_{\mathbf{r}_0} = T_f$ for some $f \in L^1_{\mathrm{loc}}(\Omega)$ . Then for any $\varphi \in \mathcal{D}(\Omega)$ with $\varphi(\mathbf{r}_0) = 0$ :

$\int_\Omega f(\mathbf{r})\,\varphi(\mathbf{r})\,d\mathbf{r} = 0.$

By the fundamental lemma of the calculus of variations, this implies $f = 0$ a.e., contradicting $\langle T_f, \varphi \rangle = \varphi(\mathbf{r}_0)$ for $\varphi(\mathbf{r}_0) \neq 0$ . So no such $f$ exists.

Imaging interpretation

A point scatterer at $\mathbf{r}_0$ with complex reflectivity $\gamma$ has

$c(\mathbf{r}) = \gamma\,\delta(\mathbf{r} - \mathbf{r}_0).$

This is the simplest nontrivial imaging scene. In a discretised model it becomes a single nonzero voxel — a perfectly sparse signal. The Dirac delta also appears in the Green's function: $\Delta G + \kappa^2 G = -\delta$ (Helmholtz equation), making distributions indispensable for scattering theory.

Definition:
Weak Derivative

The weak derivative (or distributional derivative) of a distribution $T \in \mathcal{D}'(\Omega)$ in the direction of multi-index $\alpha = (\alpha_1, \ldots, \alpha_d)$ is defined by

$\langle D^\alpha T, \varphi \rangle = (-1)^{|\alpha|}\,\langle T, D^\alpha\varphi \rangle \quad \forall\,\varphi \in \mathcal{D}(\Omega).$

Every distribution has weak derivatives of all orders.

The sign $(-1)^{|\alpha|}$ comes from integration by parts — the derivative is transferred to the smooth test function $\varphi$ , and boundary terms vanish because $\varphi$ has compact support.

Key example: The Heaviside function $H(x) = \mathbf{1}_{x > 0}$ has weak derivative $H' = \delta_0$ (the Dirac delta at 0). This formalises the physicist's statement " $d/dx\,H(x) = \delta(x)$ " — the sharp edge of an urban building's SAR return has a distributional derivative that is a delta function.

Definition:
Sobolev Spaces $H^s(\Omega)$

For $s \in \mathbb{N}_0$ , the Sobolev space $H^s(\Omega) \triangleq W^{s,2}(\Omega)$ consists of $L^2(\Omega)$ functions whose weak derivatives up to order $s$ are all in $L^2(\Omega)$ :

$H^s(\Omega) = \bigl\{f \in L^2(\Omega) : D^\alpha f \in L^2(\Omega)\;\text{for all}\;|\alpha| \leq s\bigr\},$

equipped with the inner product

$\langle f, g \rangle_{H^s} = \sum_{|\alpha| \leq s} \langle D^\alpha f, D^\alpha g \rangle_{L^2}.$

For non-integer $s$ , $H^s$ is defined via the Fourier transform: $f \in H^s(\mathbb{R}^d)$ iff $(1 + |\xi|^2)^{s/2}\hat{f} \in L^2(\mathbb{R}^d)$ .

Hierarchy: $H^2(\Omega) \subset H^1(\Omega) \subset L^2(\Omega) = H^0(\Omega)$ . Higher-order Sobolev spaces contain smoother functions. Reconstruction in $H^s$ with $s > 0$ is a form of regularisation: it penalises rapid spatial variation and produces smoother images than plain $L^2$ regularisation.

For negative $s$ , $H^{-s}(\Omega)$ is the dual space of $H^s(\Omega)$ and contains "rougher" objects like distributions. The Green's function of the Helmholtz equation is in $H^{-1}$ near the singularity.

Definition:
Green's Function of the Helmholtz Operator

The Helmholtz operator in $d$ dimensions is $\mathcal{L} = \Delta + \kappa^2$ where $\kappa = 2\pi/\lambda$ is the wavenumber.

The free-space Green's function $G(\mathbf{r}, \mathbf{r}')$ satisfies

$(\Delta_{\mathbf{r}} + \kappa^2)\,G(\mathbf{r}, \mathbf{r}') = -\delta(\mathbf{r} - \mathbf{r}')$

subject to the Sommerfeld radiation condition (outgoing waves at infinity). In 3-D:

$G(\mathbf{r}, \mathbf{r}') = \frac{e^{j\kappa\|\mathbf{r} - \mathbf{r}'\|}}{4\pi\,\|\mathbf{r} - \mathbf{r}'\|}.$

The Dirac delta on the right-hand side requires the distributional framework to make rigorous sense. The Green's function is not in $L^2$ near $\mathbf{r} = \mathbf{r}'$ (it has a $1/r$ singularity in 3-D), but it is in $H^{-1}_{\mathrm{loc}}$ .

The physical meaning: $G(\mathbf{r}, \mathbf{r}')$ is the field at $\mathbf{r}$ produced by a unit point source at $\mathbf{r}'$ . The factor $e^{j\kappa r}$ is the outgoing spherical wave (the Sommerfeld condition selects the physical solution). The $1/(4\pi r)$ factor is the spherical spreading loss. This is the kernel of the forward operator in the Born approximation (Ch~5).

Example: The Green's Function as the Kernel of the Forward Operator

Using the Green's function $G(\mathbf{r}, \mathbf{r}')$ of the Helmholtz equation, write the Born-approximation forward operator that maps the contrast function $\chi(\mathbf{r}) = \kappa^2[\epsilon_r(\mathbf{r}) - 1]$ to the scattered field. Show it is an integral operator and identify its kernel.

Solution

Born approximation

Under the Born approximation (linearisation assuming weak scattering, see Ch~5), the scattered field at receiver position $\mathbf{r}_{\mathrm{rx}}$ due to an incident plane wave $E^{\mathrm{inc}}$ is

$E^{\mathrm{sc}}(\mathbf{r}_{\mathrm{rx}}) = \int_{\Omega} G(\mathbf{r}_{\mathrm{rx}}, \mathbf{r})\, \chi(\mathbf{r})\, E^{\mathrm{inc}}(\mathbf{r})\,d\mathbf{r}.$

Identifying the forward operator

This is an integral operator $\mathcal{A}$ with kernel

$K(\mathbf{r}_{\mathrm{rx}}, \mathbf{r}) = G(\mathbf{r}_{\mathrm{rx}}, \mathbf{r})\, E^{\mathrm{inc}}(\mathbf{r}).$

The input is $\chi \in L^2(\Omega)$ and the output is $E^{\mathrm{sc}}$ evaluated at a discrete set of receiver positions. Since $G$ is square-integrable away from the singularity and $E^{\mathrm{inc}}$ is smooth and bounded, the Hilbert–Schmidt condition is satisfied — the forward operator is compact.

Why This Matters: Distributions in Scattering Theory

Distributions appear at three points in the RF imaging forward model:

Point scatterers. An isolated scatterer at $\mathbf{r}_0$ has $c(\mathbf{r}) = \gamma\,\delta(\mathbf{r} - \mathbf{r}_0)$ — a singular distribution. The discretised imaging model approximates this by a delta on the voxel grid.
The Green's function. The Helmholtz Green's function $G(\mathbf{r}, \mathbf{r}')$ satisfies $(\Delta + \kappa^2)G = -\delta$ in the distributional sense. Without distributions, this equation has no meaning — yet it is the kernel of the forward operator in every Born-approximation imaging model.
Sobolev regularity of scenes. Assuming $c \in H^s$ with $s > 0$ is a smoothness prior. It makes the inverse problem less ill-posed by restricting the feasible set to smoother functions, improving reconstruction stability (at the cost of resolution).

See full treatment in Chapter 5

Quick Check

The Dirac delta $\delta_{\mathbf{r}_0}$ is best described as ___.

A function in $L^1(\Omega)$

A function in $L^2(\Omega)$

A continuous linear functional on $\mathcal{D}(\Omega)$

An element of $H^1(\Omega)$

Correction:

A continuous linear functional on

\mathcal{D}(\Omega)

The Dirac delta is a distribution — a continuous linear functional on the space of test functions $\mathcal{D}(\Omega)$ , defined by $\langle \delta_{\mathbf{r}_0}, \varphi \rangle = \varphi(\mathbf{r}_0)$ . It is not a function in any $L^p$ space.

Quick Check

Assuming the scene reflectivity $c \in H^1(\Omega)$ (instead of $L^2$ ) as a prior for reconstruction ___.

Has no effect on the reconstruction

Makes the reconstruction less smooth

Penalises large spatial gradients, producing smoother images

Promotes sparse reconstructions

Correction:

Penalises large spatial gradients, producing smoother images

$H^1$ regularisation adds the term $\|\nablac\|_{L^2}^2$ to the objective. This penalises large spatial gradients, biasing the reconstruction toward smooth scenes. It is intermediate between $L^2$ (Tikhonov) and $H^2$ (biharmonic, even smoother).

Distribution (generalised function)

A continuous linear functional $T: \mathcal{D}(\Omega) \to \mathbb{C}$ on the space of test functions. Distributions include all locally integrable functions, the Dirac delta, and their derivatives of all orders. The theory allows differentiation of non-smooth and singular objects.

Sobolev space $H^s$

The Hilbert space $H^s(\Omega)$ of $L^2$ functions whose weak derivatives up to order $s$ are in $L^2$ . Equipped with the $H^s$ inner product. For $s > 0$ this is a proper subspace of $L^2$ containing smoother functions; for $s < 0$ it is a space of distributions.

Green's function

The fundamental solution $G(\mathbf{r}, \mathbf{r}')$ of a differential operator $\mathcal{L}$ , satisfying $\mathcal{L}G = -\delta$ in the distributional sense. For the Helmholtz operator in 3-D: $G = e^{j\kappa r}/(4\pi r)$ . The kernel of the Born-approximation forward operator.

Key Takeaway

Three core messages of this section:

(1) Distributions extend classical functions to handle point scatterers (Dirac delta), sharp edges (Heaviside), and the singularities of Green's functions — all of which appear naturally in RF imaging.

(2) The Green's function of the Helmholtz operator ( $G = e^{j\kappa r}/(4\pi r)$ in free space) is the kernel of the Born-approximation forward operator. The equation $(\Delta + \kappa^2)G = -\delta$ is only meaningful in the distributional sense.

(3) Sobolev spaces $H^s$ provide a quantitative measure of smoothness. Reconstruction in $H^s$ encodes a smoothness prior — increasing $s$ produces smoother images at the cost of reduced ability to recover sharp features. This is the function-space analogue of Tikhonov regularisation with a derivative penalty.

Distributions, Sobolev Spaces, and Green's Functions

Why Distributions and Sobolev Spaces?

Historical Note: Laurent Schwartz and the Theory of Distributions

Definition: Test Functions and the Space D(Ω)\mathcal{D}(\Omega)D(Ω)

Definition: Distributions — D′(Ω)\mathcal{D}'(\Omega)D′(Ω)

Example: The Dirac Delta as a Distribution

Definition as a functional

Not a regular distribution

Imaging interpretation

Definition: Weak Derivative

Definition: Sobolev Spaces Hs(Ω)H^s(\Omega)Hs(Ω)

Definition: Green's Function of the Helmholtz Operator

Example: The Green's Function as the Kernel of the Forward Operator

Born approximation

Identifying the forward operator

Why This Matters: Distributions in Scattering Theory

Quick Check

Quick Check

Distribution (generalised function)

Sobolev space HsH^sHs

Green's function

Key Takeaway

Definition:
Test Functions and the Space $\mathcal{D}(\Omega)$

Definition:
Distributions — $\mathcal{D}'(\Omega)$

Definition:
Weak Derivative

Definition:
Sobolev Spaces $H^s(\Omega)$

Definition:
Green's Function of the Helmholtz Operator

Sobolev space $H^s$