2025 USA-NA-AIO Round 1, Problem 1, Part 6

In the remaining parts of this problem, for all non-coding tasks, when you need to use those eigenvalues,

\color{red}{\textbf{keep them as $\lambda_0$ and $\lambda_1$. No need to apply their formulae.}}

Part 6 (10 points, non-coding task)

Since matrix \mathbf{A} is symmetric, let us do the spectral decomposition of it.

That is, you should write \mathbf{A} in the following form:

\mathbf{A} = \mathbf{Q} \mathbf{\Lambda} \mathbf{Q}^\top ,

where

  • \mathbf{Q} \in \Bbb R^{2 \times 2} is an orthonormal matrix.

  • \mathbf{\Lambda} = \begin{bmatrix} \lambda_0 & 0 \\ 0 & \lambda_1 \end{bmatrix} \in \Bbb R^{2 \times 2} is a diagnomal matrix with \lambda_0 > \lambda_1

In this task, you need to compute \mathbf{Q}. Reasoning is required.

\color{green}{\text{### WRITE YOUR SOLUTION HERE ###}}

To compute an eigenvector \mathbf{x}_i assocaited with the eigenvalue \lambda_i for i \in \left\{ 0, 1 \right\}, we solve the following equation:

\begin{bmatrix} 1 - \lambda_i & 1 \\ 1 & - \lambda_i \end{bmatrix} \begin{bmatrix} x_{i0} \\ x_{i1} \end{bmatrix} = \begin{bmatrix} 0 \\ 0 \end{bmatrix} .

Since the determintant of the 2-by-2 matrix on the L.H.S. is 0, the rank of this matrix is 1. Thus, it is sufficient to only solve the following equation:

\begin{bmatrix} 1 & - \lambda_i \end{bmatrix} \begin{bmatrix} x_{i0} \\ x_{i1} \end{bmatrix} = \begin{bmatrix} 0 \\ 0 \end{bmatrix} .

One solution is

\mathbf{x}_i = \begin{bmatrix} x_{i0} \\ x_{i1} \end{bmatrix} = \begin{bmatrix} \lambda_i \\ 1 \end{bmatrix} .

Thus, the $i$th orthonormal vector is

\begin{align*} \mathbf{q}_i & = \frac{\mathbf{x}_i}{|| \mathbf{x}_i ||_2} \\ & = \frac{1}{\sqrt{1 + \lambda_i^2}} \begin{bmatrix} \lambda_i \\ 1 \end{bmatrix} . \end{align*}

Therefore,

\begin{align*} \mathbf{Q} = \boxed{\begin{bmatrix} \frac{\lambda_0}{\sqrt{1 + \lambda_0^2}} & \frac{\lambda_1}{\sqrt{1 + \lambda_1^2}} \\ \frac{1}{\sqrt{1 + \lambda_0^2}} & \frac{1}{\sqrt{1 + \lambda_1^2}} \end{bmatrix} }. \end{align*}

\color{red}{\text{""" END OF THIS PART """}}