In class today, we learned about the familiar $L^p$ martingale convergence theorem: For $p >1,$ if $X_n$ is a martingale with $\sup \mathbb{E}|X_n|^p <\infty$, then $X_n \rightarrow X$ a.s. and in $L^p$.

Comparing this result with many other convergence theorems, we find that is requires X to be a martingale instead of just a submartingale. I assume this must have some reason, i.e. this result does not hold for submartingales.

Intuitively, I would assume that this is caused by something related to a restriction to the nonnegativity of the submartingale, i.e. the result will not hold when $X_n$ is a negative submartingale. Is this true? Or does the theorem also hold for submartingales?

If it does not hold, could you give me a counterexample?


1 Answer 1


The $L^p$ martingale convergence theorem holds also true for non-negative submartingales. The proof relies on Doob's maximal inequality:

Let $(X_j)_{j \in \mathbb{N}}$ be a non-negative submartingale (or a martingale). Then $X_n^* := \sup_{j \leq n} |X_j|$ satisfies $$\|X_n^*\|_p \leq \frac{p}{p-1} \|X_n\|_p$$ for any $p>1$. Moreover, for $X_{\infty}^* := \sup_{j \geq 1} |X_j|$ we have $$\|X_{\infty}^*\|_p \leq \frac{p}{p-1} \sup_{j \geq 1} \|X_j\|_p. \tag{1}$$

For a proof see e.g. René Schilling: Measures, Integrals and Martingales, Theorem 19.12.

So let's prove the convergence theorem for non-negative submartingales:

Let $(X_j)_{j \in \mathbb{N}}$ be a non-negative submartingale which is bounded in $L^p$ for some $p>1$, i.e.

$$\sup_{j \geq 1} \|X_j\|_p <\infty \tag{2}.$$

Then $(X_j)_{j \in \mathbb{N}}$ is in particular bounded in $L^1$ and therefore, by a standard convergence theorem for submartingales, we have $X_j \to X$ almost surely for some random variable $X$. As $|X| \leq X_{\infty}^*$, we find by $(1)$ and $(2)$

$$|X-X_j| \leq 2 X_{\infty}^* \in L^p.$$

Consequently, the dominated convergence theorem proves

$$\|X-X_j\|_p \to 0 \qquad \text{as $j \to \infty$.}$$

  • $\begingroup$ Nice, good to see that my intuition was right (i.e. only nonnegative submartingales also converge in Lp). Could you give me some counterexample for a negative submartingale where this convergence fails? $\endgroup$
    – Olorun
    Sep 15, 2015 at 23:24
  • 2
    $\begingroup$ @Olorun Beware that to negate that a submartingale is nonnegative is not to assert that it is negative. :-) $\endgroup$
    – Did
    Sep 16, 2015 at 6:56
  • $\begingroup$ @Did That was indeed confusing to me, just realized that the negation of a non-negative random variable just means that it can negative value with probability strictly greater than 0. $\endgroup$
    – Olorun
    Sep 16, 2015 at 7:05
  • $\begingroup$ @Olorun Exactly. $\endgroup$
    – Did
    Sep 16, 2015 at 7:22
  • 1
    $\begingroup$ @saz Not at the moment. Such a submartingale would converge in $L^r$ for every $r<p$ (for a reference, see theorem 6.12 there). $\endgroup$
    – Did
    Sep 16, 2015 at 12:22

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.