4
$\begingroup$

Let $\alpha$ be a positive constant, $\mu$ be a Borel nonnegative measure in $\mathbb{R}^n_+$. We can define a transform $$ \tilde{L}\[\mu\](p) = \int\limits_{\mathbb{R}^n_+} e^{-(p_1 x_1 + \ldots + p_n x_n)^{\alpha}} \mu(dx). $$ This is a simple generalization of the Laplace transform for which $\alpha = 1$. Is there something in literature on this transform? I'm interested in questions of inversion, characterisation e.t.c.

$\endgroup$

2 Answers 2

1
$\begingroup$

In one dimension, if $\mu$ has a density $f$ with respect to the Lebesgue measure, $$ (\tilde L f)(p)=\int_0^{+\infty} e^{-p^\alpha y}f(y^{\frac{1}{\alpha}})\frac{1}{\alpha} y^{\frac{1}{\alpha}-1} dy=(Lg_\alpha)(p^\alpha), $$ that is the ordinary Laplace transform of $g_\alpha$, $g_\alpha(y)=f(y^{\frac{1}{\alpha}})\frac{1}{\alpha} y^{\frac{1}{\alpha}-1} $ evaluated at $p^\alpha$. The extension to a class of Borel measure is certainly possible.

The multidimensional case is not so clear.

$\endgroup$
1
$\begingroup$

Here is some hope. (For simplicity, when referring to the unit sphere, I really mean only the part of the unit sphere contained in the first quadrant $\mathbb{R}^n_+$.)

If $\mu$ is the Dirac delta concentrated at a point $\vec{x}$ then $L[\mu](\vec{p})= e^{-(\vec{p}\cdot\vec{x})^\alpha}$, which shows that $\mu$ is determined by this transform. More generally suppose $\mu$ is a linear combination of $\delta$-s

$$\mu=\sum_{k=1}^N c_k \delta(\vec{x}-\vec{x}_k). $$

In this case we have

$$L[\mu](\vec{p})=\sum_{k=1}^N c_k e^{-(\vec{p}\cdot\vec{x}_k)^\alpha}=:f(\vec{p}). $$

Consider the set $P$ of $\vec{p}$-s of length $1$ so that the numbers $\vec{p}\cdot\vec{x}_i$ are pairwise disjoint. $P$ is open and dense in the unit sphere because it's the complement in the unit sphere is the union of the great spheres $\vec{p}\cdot \vec{x}_i=\vec{p}\cdot\vec{x}_j $, $i\neq j$.

The set $P$ is made up of several connected components, chambers, $P_1,\dotsc, P_\nu$. For any chamber $P_s$, there exists $i(s)=1,\dotsc, N$ such that

$$\vec{p}\cdot \vec{x}_{i(s)} < \vec{p}\cdot \vec{x}_j,\;\;\forall j\neq i(s),\;\;\forall \vec{p}\in P_s. $$

Then as $t\to \infty$ we have

$$\frac{\log f(t\vec{p})}{t^\alpha} \sim -(\vec{p}\cdot\vec{x}_{i(s)})^\alpha. $$

This determines $\vec{p}\cdot\vec{x}_{i(s)}$ for any $p\in P_s$. This determines the point $\vec{x}_{i(s)}$.

The constant $c_{i(s)}$ is then determined from the equality

$$c_{i(s)}= \lim_{t\to\infty} e^{t^\alpha(\vec{p}\cdot\vec{x}_{i(s)})^\alpha} f(t\vec{p}),\;\; \vec{p}\in P_s.$$

The function

$$ g(\vec{p})= f(\vec{p})-c_s e^{-(\vec{p}\cdot \vec{x}_{i(s)})^\alpha} $$

is the generalized Laplace transform of a linear combination of $\delta$-s concentrated at $(N-1)$ points.

Thus, the generalized Laplace transform is injective when restricted to the vector space spanned by $\delta$-s. $\newcommand{\bR}{\mathbb{R}}$ $\newcommand{\vp}{\vec{p}}$ $\newcommand{\vx}{\vec{x}}$ $\newcommand{\vy}{\vec{y}}$

Remark 1 Let me denote by $Q_x$ the quadrant $\mathbb{R}^n_+$ in the $x$-space and by $Q_p$ the same quadrant but viewed in the $p$-space. Consider the kernel $K_\alpha:Q_p\times Q_x\to\bR$

$$ K_\alpha(\vec{p},\vec{x}) = e^{-(\vp\cdot {\vx})^\alpha}. $$

and its dual $K_\alpha^*:Q_y\times Q_p\to\bR$

$$ K_\alpha^*(\vec{y},\vp) = e^{-(\vec{y}\cdot\vp)^\alpha}. $$

The kernel $K_\alpha$ defines the generalized Laplace transform $L_\alpha$ when applied to functions on $Q_x$ that decay sufficiently fast. The convolution of the $K_\alpha^*$ with $K_\alpha$ is the kernel

$$ N_\alpha(\vy,\vx)=K_\alpha^* \ast K_\alpha(\vy,\vx)=\int_{Q_p} e^{-(\vy\cdot\vp)^\alpha-(\vp\cdot\vx)^\alpha} d\vp. $$

Note that

$$ N_\alpha(\vx,\vy)=N_\alpha(\vy,\vx)>0, \;\; \forall \vx,\vy\in Q$$

and if $f, g:Q\to\bR$ decay sufficiently fast at $\infty$ we have

$$\int_Q\int_Q N_\alpha(\vx,\vy) f(\vx)g(\vy) d\vx d\vy = (L_\alpha[f],L_\alpha[g])_{L^2(Q_p)}. $$

The integral operator $T_\alpha$ defined by $N_\alpha$ is symmetric and nonnegative, i.e.,

$$ (T_\alpha f,f)\geq 0,\;\; \forall f\in L^2(Q). $$

To check the injectivity of $L_\alpha$ one needs to show that

$$ (T_\alpha f,f)=0,\;\; f\in L^2(Q) \Rightarrow f=0. $$

Comment There seems to be problems with MathJax because I get distorted display and I cannot find my TeX errors. I'll post as is and Edit later.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.