What Are Eigenvalues and Eigenvectors?
In linear algebra, when a linear transformation represented by a square matrix A acts on a non-zero vector v, the resulting vector Av typically changes both its length and direction. However, for certain special vectors, the transformation only stretches, shrinks, or reverses the vector without changing its span. These special vectors are called eigenvectors, and the corresponding scaling factors are called eigenvalues (denoted by the Greek letter lambda, λ).
The fundamental eigenvalue equation is:
Av = λv or equivalently (A - λI)v = 0
Where I is the identity matrix of the same dimension as A, and v ≠ 0.
The 3-Step Method to Solve for Eigenvalues and Eigenvectors
Step 1: Form and Solve the Characteristic Equation
For the non-trivial solution (A - λI)v = 0 to exist, the matrix (A - λI) must be singular (non-invertible). Therefore, its determinant must equal zero:
det(A - λI) = 0
Evaluating this determinant yields the characteristic polynomial. Solving this polynomial equation for λ gives the eigenvalues.
Step 2: Find the Eigenspace for Each Eigenvalue
For each computed eigenvalue λ, substitute it back into the homogeneous system:
(A - λI)v = 0
Row reduce the augmented matrix [A - λI | 0] to Reduced Row Echelon Form (RREF) to find the basis vectors of the null space (the eigenvectors).
Step 3: Normalize or Verify the Eigenvectors
Multiply Av and compare it directly to λv to confirm that both sides produce the exact same vector output.
Complete 2x2 Worked Example
Let matrix A be:
A = [ 4 2 ]
[ 1 3 ]
Characteristic Polynomial:
det(A - λI) = det([ 4-λ 2 ]
[ 1 3-λ ])
= (4 - λ)(3 - λ) - (2)(1)
= 12 - 7λ + λ² - 2
= λ² - 7λ + 10 = 0
Factoring the Quadratic:
(λ - 5)(λ - 2) = 0 ==> λ₁ = 5, λ₂ = 2
Finding Eigenvector for λ₁ = 5:
(A - 5I)v = [ -1 2 ] [ x ] = [ 0 ]
[ 1 -2 ] [ y ] [ 0 ]
==> -x + 2y = 0 ==> x = 2y
Eigenvector v₁ = [ 2, 1 ]^T
Finding Eigenvector for λ₂ = 2:
(A - 2I)v = [ 2 2 ] [ x ] = [ 0 ]
[ 1 1 ] [ y ] [ 0 ]
==> x + y = 0 ==> x = -y
Eigenvector v₂ = [ 1, -1 ]^T
Real-World Applications
- Google PageRank: The stationary distribution of web page relevance is the principal eigenvector (λ = 1) of the Google transition matrix.
- Quantum Mechanics: Energy states of quantum particles correspond to eigenvalues of the Hamiltonian operator.
- Data Science & PCA: Principal Component Analysis uses eigenvalues of the covariance matrix to find directions of maximum variance for dimensionality reduction.