# A Short Note on Concentration Inequalities for Random Vectors with SubGaussian Norm

In this note, we derive concentration inequalities for random vectors with subGaussian norm (a generalization of both subGaussian random vectors and norm bounded random vectors), which are tight up to logarithmic factors.

• 49 publications
• 49 publications
• 57 publications
• 60 publications
• 258 publications
02/16/2021

### Concentration of measure and generalized product of random vectors with an application to Hanson-Wright-like inequalities

Starting from concentration of measure hypotheses on m random vectors Z_...
10/04/2019

### Introduction to Concentration Inequalities

In this report, we aim to exemplify concentration inequalities and provi...
12/01/2021

### Invariance principle of random projection for the norm

Johnson-Lindenstrauss guarantees certain topological structure is preser...
01/07/2015

### An Introduction to Matrix Concentration Inequalities

In recent years, random matrices have come to play a major role in compu...
03/16/2019

### On-Line Balancing of Random Inputs

We consider an online vector balancing game where vectors v_t, chosen un...
02/08/2019

### Bernstein Concentration Inequalities for Tensors via Einstein Products

A generalization of the Bernstein matrix concentration inequality to ran...
02/28/2020

### Information Geometry of smooth densities on the Gaussian space: Poincaré inequalities

We derive bounds for the Orlicz norm of the deviation of a random variab...

## 1 Introduction

Concentration (large deviation) inequalities are one of the most important subjects of study in probability theory. A class of distributions for which sharp concentration inequalities have been developed is the class of subGaussian distributions.

###### Definition 1.

is subGaussian, if there exists so that:

 Eeθ(X−EX)≤eθ2σ22,∀θ∈R.
###### Definition 2.

A random vector is subGaussian, if there exists so that:

 Ee⟨v,X−EX⟩≤e∥v∥2σ22,∀v∈Rd.

The concentration bounds of subGaussian random vectors/variables depends on the parameter – smaller the better the concentration bounds. While subGaussian distributions arise naturally in several applications, there are settings where the random vectors have nice concentration properties but the subGaussian parameter is very large (so that applying concentration bounds for general subGaussian random vectors gives loose bounds). In this short note, we consider a related but different class of distributions, called norm-subGaussian random vectors and establish tighter concentration bounds for them.

Organization: In Section 2, we introduce norm subGaussian random vectors and some of their properties and we prove our main results in Section 3. We conclude in Section 4.

## 2 Norm SubGaussian Random Vector

The norm subGaussian random vector is defined as follows.

###### Definition 3.

A random vector is norm-subGaussian (or ), if so that:

 P(∥X−EX∥≥t)≤2e−t22σ2,∀t∈R.

Norm subGaussian includes both subGaussian (with a smaller parameter) and bounded norm random vectors as special cases.

###### Lemma 1.

There exists absolute constant so that following random vectors are all .

1. A bounded random vector so that .

2. A random vector , where and random variable is -subGaussian.

3. A random vector that is -subGaussian.

###### Proof.

The fact that the first two random vectors are immediately follows from the arguements in scalar version counterparts. For the third random vector, WLOG, assume . Let be a -cover of unit sphere (thus ). By property of subGaussian random vector, we know for each fixed :

 P(⟨vi,X⟩≥t)≤e−dt22σ2

Then let , since is a -cover, there always exists a so that in cover and . Therefore, we have:

 ∥X∥= ⟨v(X),X⟩=⟨vj(X),X⟩+⟨v(X)−vj(X),X⟩ ≤ ⟨vj(X),X⟩+∥X∥/2

Rearranging gives . Finally, the covering number of -cover over can be upper bounded by . Therefore, by union bound:

 P(∥X∥≥t)≤P(⟨vj(X),X⟩≥t/2)≤P(∃i,⟨vi,X⟩≥t/2)≤4de−dt28σ2

Now we are ready to check the second claim of Lemma 1, when , we have,

 P(∥X∥≥t)≤1≤2e−t216σ2

when , we let where , then:

 P(∥X∥≥t)≤4de−dt28σ2=e−ds8σ2≤e−s16σ2=2e−t216σ2

In sum, this proves that is . ∎

The following lemma gives equivalent characterizations of norm subGaussian in terms of moments and moment generating function (MGF).

###### Lemma 2 (Properties of norm-subGaussian).

For random vector , following statements are equivalent up to absolute constant difference in .

1. Tails: .

2. Moments: for any .

3. Super-exponential moment: .

###### Proof.

Note is a 1-dimensional random variable. This lemma directly follows from the equivalent properties of -dimensional subGaussian, for instance, Lemma 5.5 in (Vershynin, 2010). ∎

The following lemma says that if a random vector is , then its norm squared is subexponential and its projection on any direction a is subGaussian random variable.

###### Lemma 3.

There is an absolute constant so that if random vector is zero-mean , then is -subExponential, and for any fixed unit vector , is -subGaussian.

The undesirable thing about the MGF characterization in Lemma 2 is that even if is a zero mean random vector, is not zero mean, so it is difficult to directly work with MGF of . Instead, we first convert the random vector to a matrix and characterize the MGF of .

###### Lemma 4 (MGF Characterization).

There is an absolute constant , if random vector is zero-mean , then let

 Y:=(0X⊤X0)∈R(d+1)×(d+1)

we have for any .

###### Proof.

Note

is a rank-2 matrix whose eigenvalues are

, and for any . On the other hand, we also have for any . Therefore, by Lemma 2, there exists constant , for any :

 EeθY=I+∞∑p=1θ2pEY2pi(2p)!⪯(1+∞∑p=1θ2pE∥X∥2p(2p)!)I⪯(1+∞∑p=1(c⋅θ2σ2p)p(2p)!)I⪯ec⋅θ2σ2I

where in the last inequality we used the fact that , this finishes the proof. ∎

## 3 Vector Martingales with SubGaussian Norm

In this section, we will prove our main result (Lemma 6, Corollaries 7 and 8) giving concentration bounds for norm subGaussian random vectors. The main tool we use is Lieb’s concavity theorem.

###### Theorem 5 (Tropp (2012)).

Let be a fixed symmetric matrix, and let be a random symmetric matrix. Then,

 Etr(exp(A+Y))≤trexp(A+log(EeY))

We will prove our concentration result for norm subGaussian random vectors in a general setting where the subGaussian parameter for the vector can itself be a random variable.

###### Condition 4.

Let random vectors , and corresponding filtrations for satisfy that is zero-mean with . i.e.,

###### Lemma 6.

There exists an absolute constant such that if satisfy condition 4, then for any fixed , , with probability at least :

 ∥n∑i=1Xi∥≤c⋅θn∑i=1σ2i+1θlog2dδ
###### Proof.

According to Lemma 4, there exists an absolute constant so that holds for any . Therefore, we have:

 Etrexp(−c⋅θ2n∑i=1σ2iI+θn∑i=1Yi)=E{E[trexp(−c⋅θ2n∑i=1σ2iI+θn∑i=1Yi)|Fn−1]} (1)≤ Etrexp(−c⋅θ2n∑i=1σ2iI+θn−1∑i=1Yi+logE[eθYn|Fn−1])(2)≤Etrexp(−c⋅θ2n−1∑i=1σ2iI+θn−1∑i=1Yi) ≤ …≤trexp(0I)=d

where step (1) is due to Theorem 5, and step (2) used the fact that if matrix , then

. On the other hand, since identity matrix commutes with any matrix, we know:

 exp(−c⋅θ2n∑i=1σ2iI+θn∑i=1Yi)=exp(−c⋅θ2n∑i=1σ2i)⋅exp(θn∑i=1Yi)

Therefore, for any , , by Markov’s inequality, we have:

 P[∥n∑i=1Xi∥≥c⋅θn∑i=1σ2i+t/θ](1)=P[∥n∑i=1Yi∥≥c⋅θn∑i=1σ2i+t/θ] (2)= ≤

where step (1) is because is a rank-2 matrix whose eigenvalues are ; step (2) is due to all preconditions are symmetric with respect to 0. Finally, setting RHS equal to , we finish the proof. ∎

###### Corollary 7 (Hoeffding type inequality for norm-subGaussian).

There exists an absolute constant such that if satisfy condition 4 with fixed , then for any , with probability at least :

 ∥n∑i=1Xi∥≤c⋅ ⎷n∑i=1σ2ilog2dδ
###### Proof.

Since now are fixed which are not random, we can pick in Lemma 6 as a function of . Indeed, pick finishes the proof. ∎

###### Corollary 8.

There exists an absolute constant such that if satisfy condition 4, then for any fixed , and , with probability at least :

 either n∑i=1σ2i≥Bor∥n∑i=1Xi∥≤c⋅ ⎷max{n∑i=1σ2i,b}⋅(log2dδ+loglogBb)
###### Proof.

For simplicity, denote log factor By Lemma 6, we know for any fixed , with probability , we have:

 ∥n∑i=1Xi∥≤c⋅θn∑i=1σ2i+ιθ

Construct two sets of and , where and with last element , . It is easy to see . By union bound, we have with probability :

 ∥n∑i=1Xi∥≤minj∈[s][c⋅θjn∑i=1σ2i+ιθj]

Consider following two cases: (1) . Then, there exists such that :

 ∥n∑i=1Xi∥≤c⋅θjn∑i=1σ2i+ιθj=c⋅√ιψjn∑i=1σ2i+ι⋅√ψjι≤(2c+1) ⎷n∑i=1σ2i⋅ι

(2) . In this case we know and:

 ∥n∑i=1Xi∥≤c⋅θ1n∑i=1σ2i+ιθ1=c⋅√ιbn∑i=1σ2i+ι⋅√bι≤(2c+1)√b⋅ι

Combining two cases we finish the proof.

## 4 Conclusion

In this short note, we introduced the notion of norm subGaussian random vectors, which include subGaussian random vectors and bounded random vectors as special cases. While it is true that , applying concentration bounds for would yield bounds which have at least linear dependence on . In contrast, the bounds we develop (in Lemma 6 and Corollaries 7 and 8) have only logarithmic dependence on . It is not clear if this logarithmic dependence is tight – totally eliminating this dependence is an interesting open problem.

## Acknowledgements

We thank Gabor Lugosi and Nilesh Tripuraneni for helpful discussions.