Fourth recitation
Contents
- This chapter focuses on continuous distributions, multivariate normal vectors, change-of-variables formulas, GOE, and several integral estimates.
- For multivariate problems, the main objects are densities, linear transformations, covariance matrices, and orthogonal invariance.
- Jensen's inequality, tail integral formulas, and normal tail bounds will be used later for laws of large numbers and concentration inequalities.
Tip. In normal distribution problems, zero covariance implies independence only when the variables come from a linear Gaussian structure.
Exercise 3.1
For continuous densities and convolution problems, handle normalization, substitutions, Jacobians, and independence separately. Check the support before doing each integral.
Find the normalizing constants for the following densities:
- , ;
- , .
Proof
From ,
Compute the integral. Let . Then , and
Thus
- Since , direct evaluation gives .
Find the normalizing constant for
Proof
Let and . Then
Therefore
Hence
Thus .
Let be independent exponential random variables with parameter . Find the density of .
Proof
Directly,
- Let be independent, and suppose each is uniformly distributed on .
(1) Find the density of .
(2) Prove that is also uniformly distributed on .
Proof
(1) We have
so the density is .
(2) Since ,
It is enough to prove that has the exponential distribution with parameter . Let
By the previous problem,
Also,
It remains to prove that has the exponential distribution with parameter .
Use the change of variables , . Then
Thus
By the density transformation formula,
Therefore
Exercise 3.2
There are three common ways to compute moments: direct integration, recursion or integration by parts, and generating functions. It is useful to compare the moments of the normal law and the semicircle law.
Let have the standard normal distribution, and let have the standard Wigner semicircle law. Find all their moments.
Proof
Both random variables are symmetric, so we only need the moments of order .
For the semicircle law,
From real analysis,
Substitution gives the final answer
The joint density of is
Find the constant , the conditional density , and the conditional expectation .
Proof
Thus .
Similarly,
Therefore
- Let be continuously differentiable on , and assume that both and are bounded. Prove that
Proof
It is enough to consider .
Conversely, suppose satisfies
Consider the Stein equation
where . For a fixed , this is an ordinary differential equation in and has a unique solution. Take . Substituting and taking expectations gives
Thus .
Let be i.i.d. random variables with finite variance. Set
Find .
Proof
Let be a nonnegative random variable. Prove that for every ,
Proof
For i.i.d. random variables and , prove:
(1) and are uncorrelated, but need not be independent.
(2) If , then and are independent.
Proof
(1)
Also,
Thus and are uncorrelated. If are independent and both have the Bernoulli distribution , then and are not independent.
(2)
Therefore and are independent.
Exercise 3.3
Linear transformations of multivariate normal vectors are most cleanly described by covariance matrices. After a linear transformation, first compute the mean and covariance.
Let have a bivariate standard normal distribution. Find:
(1) the joint density and marginal densities of and ;
(2) and .
Proof
Set and , and suppose
For a bivariate standard normal vector, write
Let
Then
By Theorem 3.3.3 in the notes,
A direct calculation gives
Thus are independent, , and .
(2) Since are independent,
- Let have the multivariate normal distribution , where is positive definite. Prove that
where are real numbers. Also find when are not all zero.
Proof
(1) Linear combinations of the components of a multivariate normal vector are again jointly normal. Hence is bivariate normal. Clearly,
Moreover,
(2) For normal variables, uncorrelatedness is equivalent to independence. Thus
Hence is independent of . Therefore
By linearity of conditional expectation,
Substituting the covariances gives
Let be i.i.d. random variables, and let . Find .
Proof
By assumption, are i.i.d., with and . Also,
First compute the covariance:
By independence, when , while . Hence
Also,
and
Thus
Let be a fixed unit vector in with . Let , where is the identity matrix. Let be the square of the length of the projection of onto the line spanned by . Find the density of .
Proof
The squared projection length is
Since , by an orthogonal transformation we may assume that is the first coordinate vector . This is allowed because is spherically symmetric. Let be the transformed vector. Then , and
Thus
Let and . Then and are independent, and
If and are independent, then
Here and , so
Its density is
Equivalently,
and it is elsewhere.
Exercise 3.4
Complex Gaussian and GOE problems often use invariance. Find the symmetry first; it is usually cleaner than computing a density directly.
Let . Prove that
Proof
Let be independent.
Find the density of .
Use Exercise 3.4.1 to compute for positive integers .
Proof
(1) Since ,
Let
Then
The Jacobian determinant of this transformation is . Hence
Integrating out gives the marginal density of :
Let . Then , and the integral does not depend on :
Let and . Then , so
Thus
Since are independent, . Let
Then
By Exercise 3.4.1 with ,
Therefore
- (Moments of GOE) Let have the GOE distribution, and set . Compute the first six moments , .
Proof
By symmetry, only the even moments need to be considered. In random matrix notation, , where is the usual trace and is the matrix dimension. By Wick's formula,
For ,
For ,
Wick's formula gives three pairings:
- : the term is nonzero only when . If , the contribution is ; if or , the contribution is ; if and , the contribution is . The total contribution is
-
: this is the crossing pairing. It is nonzero only when , or when . The total contribution is .
-
: this is the same as the first type, and contributes .
After summing,
For , cyclic symmetry gives five cases, represented by the pairings
These cases contain pairings respectively. For the first representative, :
if and only if , or ;
if and only if , or ;
if and only if , or .
There are eight cases, each contributing . For instance, choosing the second relation in all three places gives and free indices , so there are possibilities. Continuing in the same way gives
The other four representative pairings contribute
The total is
M. Ledoux, in "A recursion formula for the moments of the Gaussian orthogonal ensemble", gives a five-term recursion for the exact GOE moments. Write
Then
Suppose are i.i.d. random variables, and let
Construct the symmetric matrix
Prove that has the GOE distribution.
Proof
Since , its entries are
Because are i.i.d. , we can compute the distribution of each entry of .
For diagonal entries,
Since ,
For off-diagonal entries with ,
Since and are independent and both have distribution , their sum has distribution . Multiplying by gives
There are diagonal entries and off-diagonal entries, for a total of independent entries.
The joint density is the product of the densities of these independent entries:
Since
we get
(Orthogonal invariance of GOE) Let have the GOE distribution. Prove that for every orthogonal matrix , the matrix also has the GOE distribution.
Proof
Let . Its joint density is
where
Set
because is orthogonal.
First, is still symmetric:
The trace is invariant:
Thus the exponential part of the density is unchanged:
For the Jacobian, is a linear transformation, so there exists a matrix such that
where stacks the entries of a matrix into an -dimensional vector. A linear map is orthogonal exactly when it preserves the Euclidean norm. Here
Thus is orthogonal, and the absolute value of the Jacobian determinant is .
The density after the change of variables is therefore
Thus also has the GOE distribution.
Exercise 4.1
This section prepares tail integrals, Jensen's inequality, and moment bounds. Later we will turn them into probability bounds.
For a nonnegative random variable , prove that
Proof
For a nonnegative random variable ,
Also,
Hence
The other side is similar.
(Jensen's inequality) A function is called convex if for every there exists such that
A convex function is called strictly convex if is strictly increasing in .
- Prove that if is convex and has an expectation, then
- Prove that if is strictly convex and , then is a constant with probability .
Proof
(1) Take . By convexity,
(2) Equality can hold only when a.e. Hence is a constant with probability .
Let be a nonnegative random variable. Prove that for every ,
Proof
Fix .
- If , prove that
- If
prove that for every . Does necessarily hold? Give a reason or a counterexample.
Proof
(1) We have
Also, , and is integrable. By the dominated convergence theorem,
This proves the claim.
(2) For every , choose such that for all ,
By the tail integral formula,
Thus .
However, need not hold. For example, take
The tail integral formula shows that diverges.
Exercise 3.5
The normal tail probability has order . Integration by parts is the main tool here.
For , prove the standard normal tail estimate
Proof
The upper bound follows from
Thus
For the lower bound, define
We have , , and
For , the upper bound already proved gives . Hence , which is the desired lower bound.
- Define functions , , by and . Prove that is a degree polynomial with leading term , and that
Also prove
Solution
By definition,
where
We compute and . By induction,
It follows that is monic of degree .
For orthogonality, assume . Then
Integrating by parts times, with boundary terms equal to , gives
If , then ; if , orthogonality follows by the same integration-by-parts argument. Since , the stated relation follows.
Finally, by Taylor expansion,
But
Substituting and cancelling gives the generating function.
- For positive integers , compute the correlation coefficient .
Solution
From the generating function in (1),
First,
Comparing coefficients gives and for .
Next consider the joint generating function:
Since is standard bivariate normal, , so
Hence
On the other hand, expanding the generating functions gives
Since
comparing the coefficients of gives
Together with for and
we get
- Let and be nonconstant polynomials. Prove that
Solution
Expand and in Hermite polynomials:
The constant terms do not affect covariance. By (2),
Also,
It remains to prove
By Cauchy-Schwarz and for and ,
Hence , where .
End-of-chapter check
- The original problems and solutions in this chapter come from the corresponding TeX source files.
- You can first read only the problem boxes, write down the main identities, and then open the proof or solution.
- If a conclusion uses independence, countable additivity, a change-of-variables formula, or a moment condition, it is worth marking that point explicitly.