You are currently browsing the tag archive for the ‘large deviation inequality’ tag.
This is however not the end of the matter; there are many variants, refinements, and generalisations of the central limit theorem, and the purpose of this set of notes is to present a small sample of these variants.
First of all, the above theorem does not quantify the rate of convergence in (1). We have already addressed this issue to some extent with the Berry-Esséen theorem, which roughly speaking gives a convergence rate of uniformly in if we assume that has finite third moment. However there are still some quantitative versions of (1) which are not addressed by the Berry-Esséen theorem. For instance one may be interested in bounding the large deviation probabilities
in the setting where grows with . Chebyshev’s inequality gives an upper bound of for this quantity, but one can often do much better than this in practice. For instance, the central limit theorem (1) suggests that this probability should be bounded by something like ; however, this theorem only kicks in when is very large compared with . For instance, if one uses the Berry-Esséen theorem, one would need as large as or so to reach the desired bound of , even under the assumption of finite third moment. Basically, the issue is that convergence-in-distribution results, such as the central limit theorem, only really control the typical behaviour of statistics in ; they are much less effective at controlling the very rare outlier events in which the statistic strays far from its typical behaviour. Fortunately, there are large deviation inequalities (or concentration of measure inequalities) that do provide exponential type bounds for quantities such as (2), which are valid for both small and large values of . A basic example of this is the Chernoff bound that made an appearance in Exercise 47 of Notes 4; here we give some further basic inequalities of this type, including versions of the Bennett and Hoeffding inequalities.
where is now bounded (but can grow with ). The central limit theorem predicts that this quantity should be roughly , but even if one is able to invoke the Berry-Esséen theorem, one cannot quite see this main term because it is dominated by the error term in Berry-Esséen. There is good reason for this: if for instance takes integer values, then also takes integer values, and can vanish when is less than and is slightly larger than an integer. However, this turns out to essentially be the only obstruction; if does not lie in a lattice such as , then we can establish a local limit theorem controlling (3), and when does take values in a lattice like , there is a discrete local limit theorem that controls probabilities such as . Both of these limit theorems will be proven by the Fourier-analytic method used in the previous set of notes.
We also discuss other limit theorems in which the limiting distribution is something other than the normal distribution. Perhaps the most common example of these theorems is the Poisson limit theorems, in which one sums a large number of indicator variables (or approximate indicator variables), each of which is rarely non-zero, but which collectively add up to a random variable of medium-sized mean. In this case, it turns out that the limiting distribution should be a Poisson random variable; this again is an easy application of the Fourier method. Finally, we briefly discuss limit theorems for other stable laws than the normal distribution, which are suitable for summing random variables of infinite variance, such as the Cauchy distribution.
Finally, we mention a very important class of generalisations to the CLT (and to the variants of the CLT discussed in this post), in which the hypothesis of joint independence between the variables is relaxed, for instance one could assume only that the form a martingale. Many (though not all) of the proofs of the CLT extend to these more general settings, and this turns out to be important for many applications in which one does not expect joint independence. However, we will not discuss these generalisations in this course, as they are better suited for subsequent courses in this series when the theory of martingales, conditional expectation, and related tools are developed.
In the theory of discrete random matrices (e.g. matrices whose entries are random signs ), one often encounters the problem of understanding the distribution of the random variable , where is an -dimensional random sign vector (so is uniformly distributed in the discrete cube ), and is some -dimensional subspace of for some .
It is not hard to compute the second moment of this random variable. Indeed, if denotes the orthogonal projection matrix from to the orthogonal complement of , then one observes that
In fact, one has sharp concentration around this value, in the sense that with high probability. More precisely, we have
for some absolute constants .
In fact the constants are very civilised; for large one can basically take and , for instance. This type of concentration, particularly for subspaces of moderately large codimension , is fundamental to much of my work on random matrices with Van Vu, starting with our first paper (in which this proposition first appears). (For subspaces of small codimension (such as hyperplanes) one has to use other tools to get good results, such as inverse Littlewood-Offord theory or the Berry-Esséen central limit theorem, but that is another story.)
Proposition 1 is an easy consequence of the second moment computation and Talagrand’s inequality, which among other things provides a sharp concentration result for convex Lipschitz functions on the cube ; since is indeed a convex Lipschitz function, this inequality can be applied immediately. The proof of Talagrand’s inequality is short and can be found in several textbooks (e.g. Alon and Spencer), but I thought I would reproduce the argument here (specialised to the convex case), mostly to force myself to learn the proof properly. Note the concentration of obtained by Talagrand’s inequality is much stronger than what one would get from more elementary tools such as Azuma’s inequality or McDiarmid’s inequality, which would only give concentration of about or so (which is in fact trivial, since the cube has diameter ); the point is that Talagrand’s inequality is very effective at exploiting the convexity of the problem, as well as the Lipschitz nature of the function in all directions, whereas Azuma’s inequality can only easily take advantage of the Lipschitz nature of the function in coordinate directions. On the other hand, Azuma’s inequality works just as well if the metric is replaced with the larger metric, and one can conclude that the distance between and concentrates around its median to a width , which is a more non-trivial fact than the concentration bound given by that inequality. (The computation of the median of the distance is more complicated than for the distance, though, and depends on the orientation of .)
Remark 1 If one makes the coordinates of iid Gaussian variables rather than random signs, then Proposition 1 is much easier to prove; the probability distribution of a Gaussian vector is rotation-invariant, so one can rotate to be, say, , at which point is clearly the sum of independent squares of Gaussians (i.e. a chi-square distribution), and the claim follows from direct computation (or one can use the Chernoff inequality). The gaussian counterpart of Talagrand’s inequality is more classical, being essentially due to Lévy, and will also be discussed later in this post.