Consider the following family of measures on $(\mathbb R,\mathcal B(\mathbb R))$: $$ K_x(A) = \begin{cases} \int\limits_A \frac{1}{|x|\sqrt{2\pi}}\mathrm e^{-y^2/2x^2}\,dy&,\text{ if }x\neq 0, \\ I_{A}(0)&,\text{ if }x = 0. \end{cases} $$ where $I_A(t)$ is an indicator (characteristic) function of the set $A$. I wonder if it is possible to find a measure $\mu$ such that $\mu>>K_x$ for any $x$ (i.e. if $\mu(A) = 0$ then $K_x(A) = 0$) and $$ \xi(x,y):=\frac{\mathrm dK_x}{\mathrm d\mu} $$ is a continuous function of $(x,y)$.

Of course, we cannot take $\mu$ to be Lebesgue measure since $K_0(\{0\}) = 1$ and the only problem is in this point. Still I am not sure that there are no such measures at all, but I didn't fund an example or prove that there are no an example.


A continuous density does not exist.

From the relation $\int_A \xi(0,y)\, \mu(dy)=\delta_0(A)=0$ for any Borel set $A$ that doesn't include the origin, we find that $\xi(0,y)=0$ $\mu$-almost everywhere on ${\mathbb R}\setminus\{0\}$.

Since $K_1\ll\mu$, this implies $\xi(0,y)=0$ $K_1$-almost everywhere on ${\mathbb R}\setminus\{0\}$. But $K_1$ has full support on ${\mathbb R}$ so, by continuity, we get $\xi(0,y)=0$ for all $y\in {\mathbb R}$.

But this contradicts $1=\int_{\mathbb R} \xi(0,y)\, \mu(dy)$.

  • $\begingroup$ thank you, that's a nice proof. $\endgroup$ – Ilya Aug 10 '11 at 15:49
  • $\begingroup$ Glad to help out. $\endgroup$ – user940 Aug 10 '11 at 18:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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