The fact that the variance of the sum of independent random variables is the sum of their variances allows one to have a good understanding of how well-concentrated each term $X_i$ in a sum of $n$ independent random variables $S= X_1+X_2+\dots+X_n$ has to be in order for $S$ to be concentraded around the mean. It seems that an analogous fact should be true for the sum taken modulo $1$. More concretely, I believe that the following should be true, but I cannot seem to find a proof or a reference. Help finding either of these would be highly appreciated.

Lemma (tentative). Suppose that $X_1, X_2, ..., X_n$ are (jointly) independent real random variables with $\mathbb{E} X_i = 0$ for each $i \leq n$, and that there exists a real number $\gamma$ such that $$ \mathbb{E} [ \lVert X_1 + X_2 + \dots + X_n - \gamma \rVert_{\mathbb{R}/\mathbb{Z}}^2] \leq \varepsilon $$ for some small $\varepsilon > 0$. Then $$ \mathbb{E} [ \lVert X_1\rVert_{\mathbb{R}/\mathbb{Z}}^2 ] + \mathbb{E} [ \lVert X_2\rVert_{\mathbb{R}/\mathbb{Z}} ^2 ] + \dots + \mathbb{E} [ \lVert X_n\rVert_{\mathbb{R}/\mathbb{Z}}^2 ] \leq C \varepsilon,$$ where $C$ is an absolute constant. Here, $\lVert x\rVert_{\mathbb{R}/\mathbb{Z}} $ denotes the distance of $x$ from the nearest integer.


  1. The lemma would be true if $\lVert x\rVert_{\mathbb{R}/\mathbb{Z}} $ were replaced with $\lvert x\rvert_{\mathbb{R}/\mathbb{Z}} $; indeed this is just the additivity of variance.

  2. It is not hard to check that the lamma is true if $X_i$ have normal distribution or if $X_i$ have Bernoulli distribution with vales $0$ and $1/2$.

  3. On the intuitive level, if we imagine that the distribution of $X_1+\dots+X_n$ is roughly bell-shaped with a single maximum, the only way for $\mathbb{E} [ \lVert X_1 + X_2 + \dots + X_n - \gamma \rVert_{\mathbb{R}/\mathbb{Z}}^2]$ to be small is if $X_1 + X_2 + \dots + X_n$ is very strongly concentrated around that maximum. In particular the ``wrap-around'' issues should not play much role and the above lemma should be roughly equivalent to the version with absolute value in place of the distance from $\mathbb{Z}$.


On the one hand, the proof is very cheap. Let $Z_j=e^{2\pi iX_j}$. $X=\sum_j X_j$, $Z=e^{2\pi i X}$. Note that $\operatorname{Var}_{\mathbb R/\mathbb Z}X\approx 1-|EZ|$ and similarly for $X_j$ and $Z_j$. Now just use the identity $EZ=\prod_j EZ_j$ to conclude.

On the other hand, finding the reference may be a highly non-trivial task, so I leave it to somebody else :-).

P.S. What can be really concluded here is that there exist $\gamma_j$ summing to $\gamma$ with $\sum_jE\|X_j-\gamma_j\|_{\mathbb R/\mathbb Z}^2\le C\varepsilon$. The condition $EX_j=0$ does not allow one to conclude from here that we can take $\gamma_j=0$: take $X_j$ symmetric with values about $\pm \frac 12$ for a counterexample.

| cite | improve this answer | |

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.