Practical examples of numerical instability in QR factorization

📚 Quick Study Guide

• 📐 QR factorization decomposes a matrix $A$ into an orthogonal matrix $Q$ and an upper triangular matrix $R$, such that $A = QR$.
• 🔢 Numerical instability arises when small errors in the input data or during computation become amplified, leading to inaccurate results.
• 💣 Common sources of instability include ill-conditioned matrices (matrices with a high condition number) and the accumulation of rounding errors during floating-point arithmetic.
• 🧪 Algorithms like Gram-Schmidt, Householder reflections, and Givens rotations can be used for QR factorization, each with varying degrees of numerical stability. Householder reflections are generally the most stable.
• 📉 Loss of orthogonality in $Q$ is a typical symptom of numerical instability. Ideally, $Q^TQ = I$ (identity matrix), but due to numerical errors, this might not hold exactly.
• 💡 Condition number, defined as $cond(A) = ||A|| * ||A^{-1}||$, quantifies the sensitivity of the solution of a linear system to errors in the data. Higher condition numbers imply greater sensitivity and potential for instability.

Practice Quiz

1. Which of the following is most susceptible to numerical instability when performing QR factorization?
   1. A. Householder reflections
   2. B. Modified Gram-Schmidt
   3. C. Givens rotations
   4. D. Classical Gram-Schmidt
2. What is a common symptom of numerical instability in QR factorization?
   1. A. Perfect orthogonality of Q ($Q^TQ = I$)
   2. B. Exact upper triangularity of R
   3. C. Loss of orthogonality in Q
   4. D. Exact preservation of the matrix norm
3. What does a high condition number of a matrix A indicate?
   1. A. The matrix is well-conditioned and stable.
   2. B. The matrix is likely to produce stable QR factorization.
   3. C. The matrix is highly sensitive to errors, potentially leading to instability.
   4. D. The matrix has orthogonal columns.
4. Which of these methods is generally considered the most numerically stable for QR factorization?
   1. A. Classical Gram-Schmidt
   2. B. Modified Gram-Schmidt
   3. C. Householder reflections
   4. D. Cholesky decomposition
5. In floating-point arithmetic, what is a major cause of numerical instability?
   1. A. Exact representation of all real numbers
   2. B. Accumulation of rounding errors
   3. C. Perfect matrix inversion
   4. D. Infinite precision
6. If you observe that $A \neq QR$ after performing QR factorization, what could this indicate?
   1. A. The QR factorization was performed correctly.
   2. B. Numerical instability has occurred.
   3. C. The matrix A is singular.
   4. D. The matrix A is orthogonal.
7. Which operation during Gram-Schmidt is most prone to exacerbating numerical instability?
   1. A. Vector normalization
   2. B. Vector addition
   3. C. Dot product (inner product)
   4. D. Scalar multiplication