Eigenvalues & Eigenvectors
See how a 2×2 matrix transforms space — eigenvectors stay collinear, eigenvalues reveal the stretching factor
📐 Eigenvalues & Eigenvectors
A vector v is an eigenvector of matrix
A if the transformation only scales it — it does
not rotate:
A·v = λ·v —
where scalar λ is the corresponding
eigenvalue.
To find eigenvalues, solve the
characteristic equation:
det(A − λI) = 0 ⟹ λ² − tr(A)·λ + det(A) = 0
giving λ = (tr ± √(tr² − 4·det)) / 2
Geometrically: the unit circle maps to an ellipse whose semi-axes point along the eigenvectors with lengths equal to |λ|. Real distinct eigenvalues → stretching/compression along two axes. Complex eigenvalues → rotation + spiral. Repeated eigenvalues → uniform scaling or shear.
Eigenvalues are central to PCA, quantum mechanics (energy levels), graph theory (spectral graph theory), and stability analysis of ODEs.
About Eigenvalues & Eigenvectors — 2D Linear Transformation Visualizer
This simulation lets you explore how a 2x2 matrix transforms 2D space by adjusting its four entries interactively. The unit circle deforms into an ellipse, basis vectors bend, yet special directions called eigenvectors remain perfectly collinear under the transformation — only stretched or flipped by their corresponding eigenvalue. Eigenvalues are found by solving the characteristic equation det(A - lambdaI) = 0, which for a 2x2 matrix reduces to the quadratic lambda^2 - trace(A)*lambda + det(A) = 0.
Eigenanalysis is one of the most powerful tools in applied mathematics, underpinning Principal Component Analysis (PCA) in machine learning, the energy levels of quantum systems, Google's PageRank algorithm, and the structural stability calculations engineers use when designing bridges and control systems.
Frequently Asked Questions
What is an eigenvector?
An eigenvector of a matrix A is a nonzero vector v that satisfies A*v = lambda*v, meaning the matrix only scales the vector — it does not rotate or change its direction. The scalar lambda is called the corresponding eigenvalue and measures how much the vector is stretched (|lambda| > 1), compressed (|lambda| < 1), or flipped (lambda < 0) by the transformation.
How do I use this simulation?
Use the four sliders labeled a, b, c, d to set the entries of the 2x2 matrix A = [[a, b], [c, d]]. The left panel shows original space and the right panel shows how the matrix deforms it: the unit circle becomes an ellipse, and the yellow and cyan dashed arrows mark the eigenvectors. Toggle the checkboxes to show or hide the transformed grid, the unit circle ellipse, the eigenvector lines, and the smooth identity-to-A animation. Use the preset buttons to jump to classic matrix types such as scaling, shear, reflection, rotation, and symmetric matrices.
What does it mean when the simulation shows complex eigenvalues?
When the discriminant of the characteristic equation (trace^2 - 4*det) is negative, the two eigenvalues form a complex conjugate pair of the form alpha +/- beta*i. Complex eigenvalues always correspond to a rotation combined with scaling: the transformation spirals rather than stretching along fixed directions, so no real eigenvectors exist. The simulation labels the matrix type as "Stable spiral", "Unstable spiral", or "Center" depending on whether the real part alpha is negative, positive, or zero.
What is the characteristic polynomial and how are eigenvalues computed?
The eigenvalues of a 2x2 matrix A are the roots of its characteristic polynomial det(A - lambda*I) = 0. Expanding the determinant gives the quadratic lambda^2 - trace(A)*lambda + det(A) = 0, where trace = a + d and det = a*d - b*c. Applying the quadratic formula yields lambda = (trace +/- sqrt(trace^2 - 4*det)) / 2. If the discriminant trace^2 - 4*det is positive, there are two distinct real eigenvalues; if zero, a repeated real eigenvalue; if negative, a pair of complex conjugate eigenvalues.
How are eigenvalues used in Principal Component Analysis (PCA)?
In PCA, you compute the covariance matrix of your data set and then find its eigenvalues and eigenvectors. The eigenvectors point in the directions of greatest variance (the principal components), and the corresponding eigenvalues tell you how much variance is captured along each direction. By keeping only the eigenvectors with the largest eigenvalues, you can project high-dimensional data onto a lower-dimensional subspace while preserving as much information as possible. This technique drives dimensionality reduction in machine learning, image compression, and exploratory data analysis.
Is it true that all matrices have eigenvectors?
Over the real numbers, a matrix may have no real eigenvectors — for example, a pure rotation matrix has only complex eigenvalues. However, over the complex numbers, every n x n matrix has exactly n eigenvalues (counting multiplicity) by the Fundamental Theorem of Algebra. Additionally, a matrix may have fewer than n linearly independent eigenvectors even when all eigenvalues are real: this happens when an eigenvalue is repeated and the matrix is not diagonalizable (a defective matrix). The simulation illustrates this with the shear preset, which has a repeated eigenvalue lambda = 1 but only one independent eigenvector direction.
Who discovered eigenvalues and when?
The concept emerged gradually through the 18th and 19th centuries. Leonhard Euler studied principal axes of rotation in 1743, and Joseph-Louis Lagrange developed related ideas in his work on quadratic forms around 1759. Augustin-Louis Cauchy was the first to prove the spectral theorem for symmetric matrices in 1829, establishing that all eigenvalues are real. The terms "eigenvalue" (Eigenwert) and "eigenvector" (Eigenvektor) were coined in German by David Hilbert in the early 20th century; "eigen" means "own" or "characteristic" in German.
What other simulations are related to eigenvalues and linear transformations?
Eigenvalues connect deeply to many areas visualized in interactive simulations: spectral graph theory links them to network connectivity and random walks; differential equations use them to classify fixed points as nodes, saddles, or spirals (exactly the types shown in this simulation's stats bar); quantum mechanics uses them to find energy levels of wave equations; and 3D computer graphics rely on eigendecomposition for principal-axis rotation and inertia tensor calculations. The Fractal Explorer and Number Spirals simulations on this site also explore mathematical structures that arise from iterated linear and nonlinear maps.
How are eigenvalues used in engineering and technology?
Structural engineers compute the eigenvalues of stiffness matrices to find the natural vibration frequencies of bridges, aircraft wings, and buildings — resonance occurs when external forces match these frequencies, which engineers must design around. Control system engineers use eigenvalues of state-space matrices to determine whether a system is stable (all eigenvalues with negative real parts), marginally stable, or unstable. In signal processing, the Discrete Fourier Transform is essentially an eigenvector decomposition of the cyclic shift matrix, and in Google's PageRank, the dominant eigenvector of the web link matrix determines the importance ranking of every page.
What are current research frontiers involving eigenvalues?
Random matrix theory studies the statistical distribution of eigenvalues for very large matrices with random entries, with applications in nuclear physics, wireless communications, and financial modeling. Researchers in quantum computing seek efficient algorithms for estimating eigenvalues of exponentially large Hamiltonian matrices, since this "quantum phase estimation" task would offer an exponential speed-up over classical computers. In deep learning, the eigenvalue spectrum of weight matrices and Hessians is actively studied to understand training dynamics, generalization, and the geometry of loss landscapes.