Suppose I have a sequence for which I can prove convergence in distribution of some infinite subsequence, e.g. let $\{X_n: n \geq 1\}$ be the original sequence and suppose I can show that

$$ X_{n_k} \to X \textrm{ in distribution as } k \to \infty$$ for a subsequence $n_k$ satisfying $n_k \to \infty$ as $k \to \infty$.

Question: What is the weakest set of conditions on the entire sequence I can impose in order to be able to extend the convergence of the subsequence back up to the whole subsequence level?

I would imagine this would be true if we could bound the deviation of the elements between the elements of the subsequence, but I am just spitballing. For example, something like: for any $k$, $$ \sup_{n_k \leq n \leq n_{k+1}} \left| X_n - X_{n_k} \right| \to 0 \textrm{ in probability as } k \to \infty $$


$\def\dto{\xrightarrow{\mathrm{d}}}\def\Pto{\xrightarrow{P}}\def\peq{\mathrel{\phantom{=}}{}}$A stronger proposition can be proved under your condition.

Suppose $\{X_t \mid t > 0\}$ is a sequence of random variables and $\{X_{t_n} \mid n \in \mathbb{N}_+\}$ is a subsequence such that $t_n \nearrow ∞$. If $X_{t_n} \dto X\ (n → ∞)$ and $\sup\limits_{t_n < t < t_{n + 1}} |X_t - X_{t_n}| \Pto 0\ (n → ∞)$, then $X_t \dto X\ (t → ∞)$.

Proof: Define $W_n = \sup\limits_{t_n < t < t_{n + 1}} |X_t - X_{t_n}|$, $C = \{x \in \mathbb{R} \mid F_X \text{ is continuous at }x\}$. Note that $\mathbb{R} \setminus C$ is at most countable, thus $C$ is dense in $\mathbb{R}$.

Consider fixed $x_0 \in C$. Take two sequences $\{y_m\}, \{z_m\} \subseteq C$ such that $y_m \nearrow x_0\ (m → ∞)$, $z_m \searrow x_0\ (m → ∞)$. Because for any $m \in \mathbb{N}_+$ and $t \geqslant t_1$, suppose $t_n \leqslant t < t_{n + 1}$, then\begin{align*} P(X_t \leqslant x_0) &= P(X_t \leqslant x_0) - P(X_{t_n} \leqslant y_m) + P(X_{t_n} \leqslant y_m)\\ &\geqslant -P(X_t > x_0,\ X_{t_n} \leqslant y_m) + P(X_{t_n} \leqslant y_m)\\ &\geqslant -P(|X_t - X_{t_n}| > x_0 - y_m) + P(X_{t_n} \leqslant y_m)\\ &\geqslant -P(W_n > x_0 - y_m) + P(X_{t_n} \leqslant y_m). \end{align*} Since $t_n \nearrow ∞$ and $W_n \Pto 0$, make $t → ∞$ to get$$ \varliminf_{t → ∞} P(X_t \leqslant x_0) \geqslant -\lim_{n → ∞} P(W_n > x_0 - y_m) + \lim_{n → ∞} P(X_{t_n} \leqslant y_m) = P(X \leqslant y_m), $$ then make $m → ∞$ to get$$ \varliminf_{t → ∞} P(X_t \leqslant x_0) \geqslant P(X \leqslant x_0). $$ Analogously,$$ \varlimsup_{t → ∞} P(X_t \leqslant x_0) \leqslant P(X \leqslant x_0), $$ thus $\lim\limits_{n → ∞} P(X_t \leqslant x_0) = P(X \leqslant x_0)$. Therefore, $X_t \dto X\ (t → ∞)$.


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.