In this post, we present two important results of the field of Gaussian Processes: the Slepian and Gordon Theorems. For both of them, we illustrate their power giving specific applications to random matrix theory. More precisely, we derive bounds on the expected operator norm of gaussian matrices.
1. Slepian’s Theorem and application
Theorem (Slepian Fernique 71′)
Let and be two centered gaussian vectors such that
Then,
Theorem (Slepian 65′)
We keep assumptions and notations of the Slepian-Fernique Theorem and we ask further that
Then it holds
Remarks:
The assumptions of the Slepian’s Theorem, namely and are equivalent to the following condition
Stated otherwise, the covariance matrices and of and are entrywise comparable with the coefficient of greater or equal to those of with the same variances.
Note that since
the conclusion of the Slepian’s Theorem is stronger compared to the one of the Slepian-Fernique’s Theorem.
Consequence
Let us consider a random variable variables taking values in with entries . Then, denoting the operator norm, it holds
where for all and for all .
Moreover,
i.e.
Proof.
Let us recall that
Hence it holds
where for all ,
Let us first consider finite sets and . Then we get for any and any ,
where in the last equality, we used the fact that and have a -norm equal to . Hence we proved that ,
Now, let us introduce for any the random variable
with . Then,
leading to
We can then apply the Slepian Fernique theorem to obtain that
To conclude the proof, we need to show that the previous computations derived with finite sets and are enough to get a bound on the operator norm (for which the supremum run on both and ). For this, we need to introduce the notion of covering sets.
Definition
Let us consider and . A set is called an -net (covering) of , if
Then, the following Lemma proves that it is sufficient to work with supremum on finite sets rather than the whole continuous spaces and . By taking the limit , the Lemma and the previous computations directly give the stated bound for the expected operator norm of a gaussian matrix .
Lemma
There exists an -net for of size
For an -net,
2. Gordon’s Theorem and application
Gordon’s Theorem
Let and be two centered gaussian processes such that
Then
Remarks
If we apply Gordon’s inequality for and , we get . Hence Gordon’s inequality contains Slepian’s inequality by taking the second index set to be a singleton set.
Consequence
For a given matrix , if (meaning that the linear map is injective), then the map
is an isomorphism with inverse denoted . The operator norm of is given by
Hence, the previous expression allows us to understand that the Gordon Theorem can be a convenient tool to upper bound the expected value of the inverse of the operator norm of the inverse of a gaussian matrix . Following an approach similar to the previous section, one can prove that
with
Remark
Let us finally point out that the approaches used in the proofs can allow to get a uniform bound for a random quadratic form given by
where and are arbitrary bounded sets. Such inequality can be given by the Chevet’s inequality which states that given a matrix whose entries are independent, mean zero, sub-gaussian random variables, it holds
where is an absolute constant and where is the maximum of the 2-Orlicz norms of the entries of . and are respectively the gaussian width and the radius of the set and are defined as follows
with . and are defined analogously.
We refer to Section 8.7 of the book High-Dimensional Probability from R.Vershynin for details.