# Inversion of Radon transform by incomplete data: specific case

Let $R[f](p,t)$ denote the Radon transform of smooth function $f(x) \colon \mathbb{R}^n \to \mathbb{R}$ with compact support in $\mathbb{R}^n_+$: $$R[f](p,t) = \int\limits_{x \cdot p = t} f(x) dx.$$ In field of mathematical economics $p$ and $t$ may have meaning of prices and then be positive. Hence there arises a problem of recovery of function $f(x)$ via it's Radon transforms $R[f](p,t)$ with $p \in \mathbb{R}^n_+$ and $t > 0$. I would like to know if there are some results on this topic.

I was asked about economical interpretation of the Radon transform. A part of economical models consists of models of production. The simplest model of production is so called Leontieff model. Say, in some sector one produce some product and use $n$ other resourses. To produce $1$ unit of our product we have to use $x_1$ units of first resourse, $x_2$ units of second e.t.c., in other words, if we have $(u_1,\ldots,u_n)$ resourses we can produce $\min \left( \frac{u_1}{x_{1}},\ldots,\frac{u_n}{x_n}\right)$ units of our production. This function is called Leontieff production function. In real models there exists an effect of substitution of production factors at the microlevel and we have to modify the Leontieff production function by the appropriate neoclassic production function. But we will still talk about Leontieff model. Let $p_0$ be a price of a unit of our production and $p=(p_1,\ldots,p_n)$ be a vector of prices of resourses, $\mu$ be a measure, that describes a distribution of available production powers over production technologies (any technology is given by a vector of resourses to produce a unit vector of our production, but we can use many technologies in production, see Hauthekker-Johaneson model) Then our profit function is a function $$\Pi(p,p_0) = \int (p_0 - p \cdot x)_{+} \mu(dx)$$ and it gives us a maximal profit possible (it is a result of maximisation over distributions $\mu$ allowed by existing production powers). It's second derivative is exactly the Radon transform of measure $\mu$.

-
Could you expand on the relation of the Radon transform and economics? (Out of curiosity.) –  Dirk Oct 20 '12 at 18:31
Of course, I modified a message. –  Nimza Oct 21 '12 at 9:19
This is well studied in the mathematics of medical imaging, you might look around there. The Radon transform is a smoothing operator with the degree of smoothing being exactly $1/2$ of a derivative, so inversion is mildly ill-posed. –  Nick Alger Oct 21 '12 at 10:21
Thank you Nick Alger, but can you be more exact please? I listened to a course on computer tomography (with applictions in medicine) and in this course we were speaking about reconstruction of function by incomplete data of another character: we know only integrals over any linear manifold, that doesn't intersect some convex body (Cormack-type theorems (1963-1964)). –  Nimza Oct 21 '12 at 13:16
The Radon transform smooths $H^s \mapsto H^{s+1/2}$, with eigenvalues $\lambda_i \rightarrow 0$ corresponding to increasingly oscillatory eigenvectors. If you add white noise to a radon transformed image then try to invert it, the $i'th$ components of the noise will be amplified by $1/\lambda_i$, which can be arbitrarily large. To overcome this one usually needs regularization or a prior, depending on whether you take a deterministic or probabilistic approach to the problem. The $1/2$ power smoothing is pretty weak compared to other problems. I don't know any more specifics. –  Nick Alger Oct 21 '12 at 17:01

The first one contains a detailed exposition of the recovery of $f$ in the special case $n=2$ (chapter 5):