Science SchoolAdministration

E-TRADE TOGETHER GLOBAL ACADEMY

Numerical Linear Algebra

Level 6 · Numerical Methods · Prerequisites: Courses 2.4 and 6.1 · Connects to Simulation, Control, AI and Robotics

ACADEMICALLY VERIFIED · MATHEMATICAL FOUNDATIONS FORMAT PARITY PASS · MOBILE QA ENVIRONMENT BLOCKED

5 canonical classes · 2 laboratories · 4 module assessments · computational project · final

Administrator review mode

No payment, enrollment or prerequisite restriction. Review content is available only to authorized administrators.

UI/UX authority: current Mathematical Foundations implementation. Rendered-content verification requires format parity.

MODULE 1

Conditioning & iterative solution

Connect problem sensitivity to practical convergence evidence for linear-system solvers.

Dependencies: Courses 2.4 and 6.1; preceding canonical classes where applicable.

6.2.1 · v1 · 90 minutes

Matrix conditioning

Why it matters

Conditioning determines how strongly data perturbations can change the exact solution of a linear system, separating problem sensitivity from algorithm behavior.

Prerequisites

Courses 2.4 and 6.1; preceding canonical classes where applicable.

Concept and explanation

For nonsingular A, κ(A)=||A||||A⁻¹|| measures worst-case relative sensitivity. A large κ warns that small relative data changes may produce large solution changes; it does not by itself prove every computed answer is inaccurate.

Key terms

condition number; norm; residual; forward error; backward error; perturbation; scaling; singularity

Notation / representation

r=b−Ax̂. In a compatible norm, relative forward error is bounded approximately by conditioning times relative backward error under stated assumptions.

Numerical method / algorithm

Inspect structure and scale → choose norm → estimate κ → solve → compute residual → perturb data → compare solution sensitivity → report credible digits.

Mathematical development

For A=diag(1,ε) in the 2-norm, κ₂(A)=1/|ε| when 0<|ε|≤1. The second component of Ax=b amplifies relative uncertainty as ε becomes small.

Learning objectives

  • Estimate and interpret matrix condition numbers in a declared norm.
  • Distinguish residual, forward error, backward error and scaling effects.

Worked examples

Worked example 1

Problem · matrix/data · goal: Analyze A=diag(1,0.001) with b=(1,0.001).

Assumptions · method · algorithm · computation: κ₂(A)=1000 and x=(1,1). Perturb b₂ to 0.001001 while A is fixed.

Result · residual/error · validation · interpretation · limitations: x₂ becomes 1.001: a 0.1% change in b₂ causes a 0.1% component change; κ is a worst-case bound, not a prediction that every perturbation is maximally amplified.

Worked example 2

Problem · matrix/data · goal: A computed x̂ has ||r||₂=10⁻¹⁰ but κ₂(A)≈10⁸.

Assumptions · method · algorithm · computation: Normalize the residual and combine it with a condition estimate before judging forward accuracy.

Result · residual/error · validation · interpretation · limitations: A tiny residual supports a small backward error, but the large condition number permits much larger forward error; report both.

Common mistake and counterexample

Incorrect: a small residual guarantees an accurate solution. An ill-conditioned system can have a tiny residual and substantial forward error.

Guided practice

  1. Rebuild Worked Example 1 with matrix structure, dimensions, parameters and an iteration or factorization record.
  2. Change one input, predict sensitivity, recompute, measure residual/error and validate independently.

A/B/C/D practice

A · Foundation

Compute condition numbers for diagonal matrices in two norms.

B · Application

Assess scaling in a synthetic sensor-calibration system.

C · Reasoning and error detection

Diagnose a residual-only accuracy claim.

D · Challenge / transfer

Design perturbation evidence and a credible-digit statement.

Knowledge check and summary

  1. State the matrix structure, dimensions and assumptions.
  2. Record the algorithm, tolerance, iteration limit and stopping evidence.
  3. Compute a residual and interpret it together with conditioning.
  4. Validate with a direct method, known eigenpair, perturbation or independent implementation.

Mastery criterion: 4/4 correct with residual, conditioning/stability distinction and independent validation. Correct each miss and complete a fresh equivalent check.

Related laboratory

Next class: Iterative solvers