2
$\begingroup$

Let $f_V$ be a pdf of random variable $V$ where \begin{align} V=U+Z \end{align} and where $U$ and $Z$ are independent and $Z$ is Gaussian. Moreover, suppose that $|U| \le A$.

Can we find the upper bound on the number of modes of $V$?

Note that the the pdf $V$ is given by \begin{align} f_V(v)=E\left[\frac{1}{\sqrt{2\pi}}e^{-\frac{(v-U)^2}{2}}\right] \end{align}

I am actually not interested in a very tight bound. I more interested in the methods that can be used to produce such bounds.

$\endgroup$

1 Answer 1

2
$\begingroup$

$\newcommand{\al}{\alpha} \newcommand{\de}{\delta} \newcommand{\De}{\Delta} \newcommand{\ep}{\varepsilon} \newcommand{\ga}{\gamma} \newcommand{\Ga}{\Gamma} \newcommand{\la}{\lambda} \newcommand{\Si}{\Sigma} \newcommand{\thh}{\theta} \newcommand{\om}{\omega} \newcommand{\R}{\mathbb{R}} \newcommand{\Z}{\mathbb{Z}} \newcommand{\F}{\mathcal{F}} \newcommand{\E}{\operatorname{\mathsf E}} \renewcommand{\P}{\operatorname{\mathsf P}} \newcommand{\ii}[1]{\operatorname{\mathsf I}\{#1\}} \newcommand{\tf}{\widetilde{f}}$

The pdf $f_V$ is a mixture of the family $\F_A:=(f_a)_{a\in[-A,A]}$ of the normal pdfs $f_a$ with mean $a$ and variance $1$. By Theorem 6 on p. 2129 of Kemperman, for each natural $s$, the pdf $f_V$ will have at most $s$ modes (or, more precisely, modal intervals) iff any mixture of any $2s$ members of the family $\F_A$ has at most $s$ modal intervals.

This theorem is a generalization of Theorem 4 on p. 2128 of Kemperman, which implies that all mixtures of a family of pdfs are unimodal iff any mixture of any two members of the family is unimodal.

Further, by Remark 1 on p. 2133 of Kemperman, our pdf $f_V$ will be unimodal for all admissible $U$ iff for all $a,b,x$ such that $-A\le a<x<b\le A$ we have \begin{equation} f'_a(x)f''_b(x)\ge f'_b(x)f''_a(x), \end{equation} which can be rewritten as $(a-x)(b-x)\ge-1$ for all such $a,b,x$, which is equivalent to $A\le1$.

The case of more modes than $1$ appears much more difficult analytically.

$\endgroup$

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.