Math Problem Statement

Let y be a k x 1 random vector with E[y] = μ and Var[y] = V. Let A be a k x k matrix of real numbers. Show that E[y' A y] = tr(AV) + μ' A μ.

Solution

To prove the given equation, let's analyze the expectation of the quadratic form involving the random vector y\mathbf{y} and matrix A\mathbf{A}. We are given:

  • y\mathbf{y} is a k×1k \times 1 random vector.
  • E[y]=μ\mathbb{E}[\mathbf{y}] = \boldsymbol{\mu} and Var[y]=V\text{Var}[\mathbf{y}] = \mathbf{V}.
  • A\mathbf{A} is a k×kk \times k matrix.

We need to show that: E[yAy]=tr(AV)+μAμ.\mathbb{E}[\mathbf{y}' \mathbf{A} \mathbf{y}] = \text{tr}(\mathbf{A} \mathbf{V}) + \boldsymbol{\mu}' \mathbf{A} \boldsymbol{\mu}.

Step-by-Step Proof

  1. Express y\mathbf{y} in terms of its mean and deviation: Let y=μ+z\mathbf{y} = \boldsymbol{\mu} + \mathbf{z}, where z\mathbf{z} is the deviation vector such that E[z]=0\mathbb{E}[\mathbf{z}] = \mathbf{0} and Var[y]=V=E[zz]\text{Var}[\mathbf{y}] = \mathbf{V} = \mathbb{E}[\mathbf{z} \mathbf{z}'].

  2. Expand the quadratic form: Substituting y=μ+z\mathbf{y} = \boldsymbol{\mu} + \mathbf{z} into E[yAy]\mathbb{E}[\mathbf{y}' \mathbf{A} \mathbf{y}]: E[(μ+z)A(μ+z)].\mathbb{E}[(\boldsymbol{\mu} + \mathbf{z})' \mathbf{A} (\boldsymbol{\mu} + \mathbf{z})].

    Expanding this, we get: E[μAμ+μAz+zAμ+zAz].\mathbb{E}[\boldsymbol{\mu}' \mathbf{A} \boldsymbol{\mu} + \boldsymbol{\mu}' \mathbf{A} \mathbf{z} + \mathbf{z}' \mathbf{A} \boldsymbol{\mu} + \mathbf{z}' \mathbf{A} \mathbf{z}].

  3. Simplify each term:

    • μAμ\boldsymbol{\mu}' \mathbf{A} \boldsymbol{\mu} is deterministic, so: E[μAμ]=μAμ.\mathbb{E}[\boldsymbol{\mu}' \mathbf{A} \boldsymbol{\mu}] = \boldsymbol{\mu}' \mathbf{A} \boldsymbol{\mu}.

    • For E[μAz]\mathbb{E}[\boldsymbol{\mu}' \mathbf{A} \mathbf{z}]: μA\boldsymbol{\mu}' \mathbf{A} is a constant vector, and E[z]=0\mathbb{E}[\mathbf{z}] = \mathbf{0}, so: [ \mathbb{E}[\boldsymbol{\mu}' \mathbf{A} \mathbf{z}] = \boldsymbol{\mu}' \mathbf{A} \mathbb{E}[\mathbf{z}] = \boldsymbol{\mu}' \mathbf{A} \mathbf{0} = 0. ]

    • Similarly, E[zAμ]=E[zA]μ=0\mathbb{E}[\mathbf{z}' \mathbf{A} \boldsymbol{\mu}] = \mathbb{E}[\mathbf{z}' \mathbf{A}] \boldsymbol{\mu} = 0.

    • For the term E[zAz]\mathbb{E}[\mathbf{z}' \mathbf{A} \mathbf{z}], use the trace property: [ \mathbb{E}[\mathbf{z}' \mathbf{A} \mathbf{z}] = \text{tr}(\mathbf{A} \mathbb{E}[\mathbf{z} \mathbf{z}']) = \text{tr}(\mathbf{A} \mathbf{V}). ]

  4. Combine the results: Adding up the non-zero terms, we get: E[yAy]=μAμ+tr(AV).\mathbb{E}[\mathbf{y}' \mathbf{A} \mathbf{y}] = \boldsymbol{\mu}' \mathbf{A} \boldsymbol{\mu} + \text{tr}(\mathbf{A} \mathbf{V}).

Thus, we have shown that: E[yAy]=tr(AV)+μAμ.\mathbb{E}[\mathbf{y}' \mathbf{A} \mathbf{y}] = \text{tr}(\mathbf{A} \mathbf{V}) + \boldsymbol{\mu}' \mathbf{A} \boldsymbol{\mu}.

Do you have any questions or need further details?

Here are 5 related questions for further exploration:

  1. How does the trace property help in simplifying the expectation of quadratic forms?
  2. What happens if A\mathbf{A} is not symmetric in this derivation?
  3. How is this result used in multivariate normal distributions?
  4. What would the expression look like if A\mathbf{A} was a diagonal matrix?
  5. How does the variance matrix V\mathbf{V} affect the outcome of E[yAy]\mathbb{E}[\mathbf{y}' \mathbf{A} \mathbf{y}]?

Tip: When working with expectations of quadratic forms, breaking down the random variable into its mean and deviation can simplify the calculations.

Ask a new question for Free

By Image

Drop file here or Click Here to upload

Math Problem Analysis

Mathematical Concepts

Linear Algebra
Expectation of Quadratic Forms
Multivariate Statistics

Formulas

E[y' A y] = tr(A V) + μ' A μ
Var[y] = V

Theorems

Expectation of Quadratic Forms

Suitable Grade Level

Graduate level (or advanced undergraduate in probability and linear algebra)