
To understand eigenvalues and eigenvectors, stop treating Av = λv as a formula to memorize and read it as a transformation sketch. A matrix A moves vectors. An eigenvector v is a nonzero vector whose image Av stays on the same line through the origin; the eigenvalue λ tells you how much that vector is stretched, squeezed, reversed, or collapsed.
This article is for students who can perform matrix multiplication but still find the idea mysterious. The most useful study order is sketch, predict, calculate, check. Geometry gives the symbols a job, and the algebra confirms whether your picture is right.
A square matrix represents a linear transformation. Most vectors change both length and direction when the transformation acts. Eigenvectors are the special nonzero vectors that do not get pushed off their original line. In Av = λv, the left side applies the transformation, while the right side says the result is only a scalar multiple of the starting vector.
The value of λ describes what happens along that eigenvector direction. If λ is greater than 1, the vector grows. If 0 < λ < 1, it shrinks. A negative λ reverses its direction as well as scaling it, and λ = 0 sends it to the zero vector. The transformed arrow may point backward, but it still lies on the same line.
Georgia Tech's Interactive Linear Algebra expresses the geometry precisely: v and Av are collinear with the origin. That wording is better than saying an eigenvector “does not move.” It can change dramatically in length or orientation; what it preserves is its one-dimensional span.
The determinant method can produce correct numbers without explaining them. A sketch makes you ask the conceptual question first: which directions does this transformation leave on their own lines? For a horizontal stretch and vertical squeeze, the coordinate axes are natural candidates. For a reflection, the mirror line and the perpendicular direction are candidates.
The sketch is not a substitute for proof. It is a prediction tool and an error detector. If your algebra claims that a diagonal stretch sends (1, 0) to a tilted vector, the picture tells you immediately that a multiplication or transcription error has occurred.
Repeat the routine from a blank page rather than rereading a completed solution. Your goal is to retrieve the meaning of each step: preserved line, scale factor, characteristic equation, and direct verification.

Let A = [[2, 0], [0, 0.5]]. Sketch a unit square. The matrix doubles every x-coordinate and halves every y-coordinate, so the square becomes a rectangle that is twice as wide and half as tall. Before calculating, predict that the x-axis and y-axis directions will remain aligned.
Now calculate. The characteristic equation is det(A − λI) = (2 − λ)(0.5 − λ) = 0, so the eigenvalues are 2 and 0.5. For λ = 2, solving (A − 2I)v = 0 gives vectors of the form (x, 0), with x ≠ 0. For λ = 0.5, the solutions have the form (0, y), with y ≠ 0.
Finally, check one representative from each direction. A(1, 0) = (2, 0) = 2(1, 0), so (1, 0) is an eigenvector with eigenvalue 2. A(0, 1) = (0, 0.5) = 0.5(0, 1), so (0, 1) is an eigenvector with eigenvalue 0.5. The arithmetic matches the transformation sketch.
Notice that (1, 0) and (4, 0) do not represent different preserved directions. Every nonzero multiple on the x-axis belongs to the same eigenspace and has eigenvalue 2. An eigenvector is an arrow; an eigenspace collects the zero vector and all eigenvectors associated with one eigenvalue.
![For A = [[2, 0], [0, 0.5]], the x-axis direction is scaled by 2 and the y-axis direction is scaled by 0.5.](/_next/image?url=https%3A%2F%2Fraw.githubusercontent.com%2Fpioscelina%2Fsnitch%2Fmain%2Fblog%2Fstudy-eigenvalues-eigenvectors-transformation-sketches%2Finline-two-by-two-matrix-transformation-05707ad451a7.png&w=1920&q=75)
Software can calculate eigenpairs quickly, but it does not replace interpretation. NumPy, for example, returns eigenvectors as columns paired with its eigenvalue array. You still need to ask what each direction means in the transformation and remember that numerical round-off may blur an exact relationship.
It triples the vector's length and reverses its direction. The result remains on the same line through the origin, so v is an eigenvector with eigenvalue −3.
A(1, 1) = (2, 0.5). There is no single scalar λ that turns both coordinates of (1, 1) into those values, so the result is not collinear with the original vector.
Calculate the candidate eigenvalue and eigenspace, then verify Av = λv by direct multiplication. Prediction builds intuition; verification establishes the result.
MIT OpenCourseWare: Lecture 21 — Eigenvalues and eigenvectors
Georgia Institute of Technology, Interactive Linear Algebra: Eigenvalues and Eigenvectors
ノート、クイズ、ポッドキャスト、フラッシュカード、チャット — アップロード1回で全部。
最初のノートを無料で試す