EVOPEDIA

← all systems · see it in The System

Linear algebra (the practice of solving, transforming and taking apart linear systems with vectors and matrices, and the structures it works with)

Linear algebra is the branch of mathematics concerning linear equations such as a 1 x 1 + ⋯ + a n x n = b, linear maps such as the following: ↦ a 1 x 1 + ⋯ + a n x n, and their representations in vector spaces and through matrices. Linear algebra is central to almost all areas of mathematics.

Properties published scienceour proposal

Sourced fields, and our own readings.

aliasesLinear algebra
born atelimination on counting-rod arrays in the Nine Chapters (parts about 150 BC; C-1198); vector spaces from Grassmann (1844) and Peano (1888); modern form in the first half of the 20th century
born precisioncenturies
boundaryInside: the practice of writing simultaneous linear conditions as vectors and matrices and solving, transforming, factoring and decomposing them (by hand, by human computers with bracket notation and tables, by machines running BLAS and LAPACK), the three modes in which people describe and think it (geometric arrows and planes, arithmetic n-tuples and matrices, structural vector spaces), and the structures it works with (vector spaces, linear maps, bases and dimension, kernels and images, determinants, eigenvalues and eigenvectors, inner products, duality; modules and tensors as extensions). Outside: the algebra, arithmetic and geometry it draws on, the sciences, engineering, statistics, search engines, graphics and machine learning that run on its results, the people and machines that carry it, and the writing that holds its arrays.
nested levelA branch of mathematics (part inside a whole), shelved by the algebra walk as a kind of algebra (C-1189); runs on algebra's and arithmetic's operations (their L6) and on writing; carried by people (rung 11), by computers (rung 12) and inside trained neural networks (rung 13); the structures as a role at rung 0.
scalesocial

Measures (18) published science

characteristic clock: hand solution of normal equations by a professional human computer (Doolittle, 1878)41 equations per week equations/week
characteristic clock: 18 equations with desk calculators, four mathematicians including Turing (1946)about 2 weeks
characteristic clock: multiplying punch (1930s-1940s), 10 simultaneous equationsabout 4 hours
characteristic clock: PageRank for 26 million Web pages on a medium size workstation (1998)a few hours
arithmetic operations to solve n equations by eliminationabout 2n^3/3 operations
operations for n = 18 by elimination (Grcar's check)approximately 3,888 operations
matrix multiplication exponent, upper bound (2023)omega <= 2.371552 (from 2.371866) exponent
multiplications for 4x4 matrices over Z2: AlphaTensor vs Strassen two-level47 vs 49 multiplications
non-equivalent factorizations found departing from Strassen-square (one size)more than 14,000 factorizations
matrix multiplication tensors improved over state of the art (n, m, p <= 12)more than 70 tensors
growth factor bound, complete pivoting (2023-2025)n^(0.2079 ln n + 0.91), improving Wilkinson's 1961 2n^(0.25 ln n + 0.5) ratio
Nine Chapters, chapter eight: problems solved by elimination18 problems, 2 to 5 equations
first LINPACK benchmark report (1979)23 computers, matrix of size 100
LINPACK 100 benchmark workroughly 2/3 million floating-point operations
TOP500 list500 most powerful systems, compiled twice a year since 1993
HPL look-ahead depth with best gain1 or 2 panels
PageRank damping factor (usual setting)0.85 probability
AlphaTensor speed-up benchmark size8,192 x 8,192 (median over 200 runs; s.d. < 0.4 percentage points) matrix size

Open contentions (5) open contention

Where sources or readings disagree. Both stay on the map.

Partners and parts published science

Built from / builds (9)

Sits inside / holds (6)

Trades with (valence) (25)

expressed by (12)

floor filled by (21)

represents (1)

Kinds (6)

Notes published science

Layers

The ways it takes in its world, where they meet, and how it acts back, each from L1 (where it meets the world) up.

a linear system: quantities combined by adding and scaling, written as vectors and matrices, and solved, transformed or taken apart (reader's view)

  1. L1: The problem in the world: several unknowns tied by linear conditions (survey angles, orbits, circuits, the Web's links, a scene to draw, a dataset), written as arrays on paper or numbers in memory
  2. L2: Taking it in: arrays read row by row and column by column, arrows and planes seen, data loaded from memory (one load and store per add/multiply pair unless blocks are reused)
  3. L3: Codes (the ports): geometric, arithmetic and structural modes; notations (rod arrays, Gauss's brackets, the single-letter matrix); for machines, the BLAS interface
  4. L4: Operating: elimination, multiplication, change of basis, factorization, iteration
  5. L5: Control: method choice, pivoting, conditioning, rounding-error bounds, convergence, checking
  6. L6: The structure: vector spaces, linear maps, bases and dimension, kernels and images, determinants, eigenvalues and eigenvectors, inner products, duality
  7. L7: Use: geometry, physics, engineering, scientific computation, statistics, search, graphics, machine learning

the boardroom (shared floors)

  1. L3: Codes (the ports): the three modes of description and thought: geometric (arrows, lines, planes, transformations), arithmetic (n-tuples, matrices, rank, solutions of systems) and structural (vector spaces, subspaces, dimension, operators, kernels); notations from counting-rod arrays and Gauss's brackets to a single letter for a matrix
  2. L4: Operating: row operations and elimination (about 2n^3/3 operations), back-substitution, matrix multiplication, change of basis, factorizations (LU, QR, SVD, eigendecomposition), iteration (power iteration, Krylov methods); by hand, by human computers, by machine kernels
  3. L5: Control: choosing the method (Cramer only for n = 2 or 3; elimination; stable decompositions over the normal equations), pivoting against small divisors, conditioning, growth-factor bounds on rounding error, convergence checks, duplicated hand calculation
  4. L6: The structure: vector spaces over a field, linear maps, subspaces, span, bases and dimension, kernel and image, determinants, eigenvalues and eigenvectors, inner products and duality; modules, tensors, Hilbert spaces as extensions; in expert mathematicians, carried by a bilateral prefrontal-parietal-temporal network shared with number and space (shown for algebra and other domains, not linear algebra by name)
  5. L7: Linear algebra put to use: geometry; physics (quantum states and operators, vibration modes, principal axes); engineering (fluids, power systems); scientific computation and the ranking of supercomputers; statistics (least squares, PCA); Web search (PageRank); computer graphics; machine learning (attention as matrix products)

way in: Data: observations and records arranged as arrays (survey angles, orbits, datasets, the Web's links)

  1. L1: Observations with errors, more than needed: supernumerary survey angles, astronomical positions; large datasets; links between Web pages
  2. L2: The observations written as an overdetermined system (more equations than unknowns); a dataset as a matrix; the Web as a link matrix

way in: Sight: arrows, lines, planes and transformations drawn (figures, animations, graphics)

  1. L1: Figures: planes whose intersection is the common solution of a system; vectors as arrows; a linear transformation rotating, stretching or shearing them; students reading eigen-equations as pictures

way in: Sight: vectors and matrices written as arrays (rows, columns, entries)

  1. L1: Arrays on a surface: a system's coefficient matrix and its right-member vector; in the Nine Chapters, rectangular arrays of counting rods

way in: The machine route: numbers in memory taken through BLAS/LAPACK kernels on processors

  1. L1: Floating-point numbers in memory that cannot represent all reals exactly
  2. L2: Loads and stores: for every add/multiply pair a load and store of the elements (the data-movement bottleneck)
  3. L3: Kernels: Level 1-3 BLAS; Level 3 does O(n^3) work on blocks so data is reused; GPUs built with a matrix structure

way in: The world's linear structure: phenomena that add and scale (vibrations, fields, quantum states), and nonlinear ones near a point

  1. L1: Physical phenomena for which linear structures are natural models; most are modelled by partial differential equations
  2. L2: Linearization: a first-order approximation (the derivative at a point gives the line that models local behaviour)
  3. L3: Discretization: space cut into small interacting cells, giving very large matrices

way in: Thought: an equation such as Ax = 2x held and read in the mind (relationally or operationally, as a picture or as numbers)

  1. L1: Resources a student brings to an eigen-equation: recognizing it, reading the equal sign relationally or operationally, knowing a relation can be a solution (where these resources come from is contested: C-413)

way in: Words: definitions, lessons and problems read or heard (textbooks, lectures, the lesson Newton supplied)

  1. L1: The lesson in words: how to solve simultaneous equations, written for learners; definitions of vector space and linear map

way out: Programming the machine: calls to BLAS/LAPACK, run on processors built for matrices

  1. L1: Processors with matrix structure (GPUs) executing the arithmetic
  2. L3: Kernel calls: Level 3 BLAS on blocks
  3. L5: Choosing the routine and its parameters (pivoting, block size, look-ahead depth)

way out: Rendering: transformations project a 3D scene onto a 2D screen

  1. L1: The image on the screen
  2. L4: Rotation and projection matrices applied to the object
  3. L7: A scene to be shown

way out: Speech and explanation: saying how the equation reads and what its solutions are

  1. L1: Spoken explanation in an interview (recorded on video and in transcripts)
  2. L6: The student's reasoning about the eigen-equation

way out: Writing each reduced array (the new rows after every elimination step)

  1. L1: Marks on paper: by hand, human computers wrote whole systems out (41 normal equations in a week)
  2. L4: The new rows: multiples of the leading equation subtracted from those below
  3. L6: The rule: row operations keep the solutions; the reduced form is unique

Who fills each layer

Sources (28) published science

Algebra (Wikipedia, via Wikimedia Enterprise snapshot 2026-11-01)
Linear algebra (Wikipedia, via Wikimedia Enterprise snapshot 2026-11-01)
Grcar JF. How ordinary elimination became Gaussian elimination. Historia Mathematica 38 (2011); arXiv:0907.2397. doi:10.1016/j.hm.2010.06.003 (behind its links)
Eigenvalues and eigenvectors (Wikipedia, via Wikimedia Enterprise) (behind its links)
Gaussian elimination (Wikipedia, via Wikimedia Enterprise) (behind its links)
Fawzi A, Balog M, et al. (2022) Discovering faster matrix multiplication algorithms with reinforcement learning. Nature 610:47-53. doi:10.1038/s41586-022-05172-4 (behind its links)
Vector space (Wikipedia, via Wikimedia Enterprise) (behind its links)
Matrix (mathematics) (Wikipedia, via Wikimedia Enterprise) (behind its links)
Numerical linear algebra (Wikipedia, via Wikimedia Enterprise) (behind its links)
Brin S, Page L. The anatomy of a large-scale hypertextual Web search engine. Computer Networks and ISDN Systems 30 (1998) (behind its links)
Vassilevska Williams V, Xu Y, Xu Z, Zhou R. New bounds for matrix multiplication: from alpha to omega. arXiv:2307.07970 (2023) (behind its links)
Bisain A, Edelman A, Urschel J (2023) A New Upper Bound For the Growth Factor in Gaussian Elimination with Complete Pivoting. arXiv:2312.00994 (behind its links)
Dongarra JJ, Luszczek P, Petitet A. The LINPACK Benchmark: past, present and future. Concurrency and Computation: Practice and Experience 15 (2003); author copy dated December 2001 (behind its links)
Donevska-Todorova A. Three modes of description and thinking of linear algebra concepts at upper secondary education. Beitraege zum Mathematikunterricht 2014, pp. 305-308 (WTM-Verlag) (behind its links)
Wawro M, Christensen W, Watson K. Student reasoning about eigenvectors and eigenvalues from a resources perspective. PERC 2019 (talk abstract) (behind its links)
Vaswani et al. 2017, Attention Is All You Need (arXiv:1706.03762) (behind its links)
Jolliffe IT, Cadima J. Principal component analysis: a review and recent developments. Phil Trans R Soc A 374:20150202 (2016). doi:10.1098/rsta.2015.0202 (behind its links)
Vector calculus (Wikipedia, via Wikimedia Enterprise snapshot 2026-11-01) (behind its links)
Quantum mechanics (encyclopedia, Wikimedia Enterprise Snapshot 2026-10-01) (behind its links)
Hamiltonian mechanics (encyclopedia, Wikimedia Enterprise snapshot 2026-11-01, version 1381752488) (behind its links)
Boudot, Gaudry, Guillevic, Heninger, Thome and Zimmermann, Comparing the difficulty of factorization and discrete logarithm: a 240-digit experiment, CRYPTO 2020 (arXiv:2006.06197) (behind its links)
Ponte JP (1992) The history of the concept of function and some educational implications. The Mathematics Educator 3(2) (behind its links)
Complex number (Wikipedia, via Wikimedia Enterprise snapshot 2026-11-01) (behind its links)
NIST, FIPS 203: Module-Lattice-Based Key-Encapsulation Mechanism Standard, August 2024 (doi:10.6028/NIST.FIPS.203) (behind its links)
Discrete Fourier transform (Wikipedia, via Wikimedia Enterprise snapshot 2026-11-01) (behind its links)
Axiom of choice (Wikipedia, via Wikimedia Enterprise snapshot 2026-11-01) (behind its links)
Graph theory (Wikipedia, via Wikimedia Enterprise snapshot 2026-11-01) (behind its links)
Wawro M, Watson K, Christensen W. Student reasoning about eigenvectors and eigenvalues from a resources perspective. PERC 2018 (poster abstract) (behind its links)