Large-Scale Systems

Numerical Analysis · Axiom Academy

REAL WORLD Solving Massive Linear Systems From weather prediction to PageRank: linear algebra at scale In textbooks, we solve 3×3 systems by hand. In the real world, engineers and scientists routinely solve systems with millions or billions of unknowns. At these scales, even O(n²) algorithms are impractical. The key insight: most real-world matrices are sparse —they have mostly zero entries. When engineers design bridges, buildings, or aircraft, they use finite element analysis (FEA) to predict stress and deformation. FEA leads to: Ku = f where K is the stiffness matrix, u is displacement, and f is force. K is typically: Symmetric positive definite (SPD) → Use Cholesky! Size: millions of unknowns for complex structures Google's original search algorithm ranks web pages by solving a massive eigenvector problem—equivalent to solving linear systems! With billions of web pages, direct methods are impossible. PageRank uses power iteration (similar to our iterative methods) because: Matrix is extremely sparse (average page links to ~10 others) Only need matrix-vector products, not matrix storage Converges fast due to spectral gap Weather prediction discretizes the atmosphere into a 3D grid and solves partial differential equations at each time step. Scale matters: Real problems have millions to billions of unknowns Sparsity is key: Most entries are zero—exploit this! Iterative methods dominate: Direct methods can't scale

This is the written version of the interactive lesson above. See the full Numerical Analysis course.