Visual Tools
Calculators
Tables
Mathematical Keyboard
Converters
Other Tools

How to Find a Basis for an Eigenspace

The eigenspace EλE_\lambda is the null space of A−λIA - \lambda I: every eigenvector for λ\lambda, plus the zero vector. A basis comes from the parametric solution, one vector per free variable, and the number of free variables is the geometric multiplicity.

Recognize the task

The question says eigenspace, basis of eigenvectors, geometric multiplicity, or "all eigenvectors for λ\lambda", usually for a repeated eigenvalue: λ=1\lambda = 1 of (211121112)\begin{pmatrix} 2 & 1 & 1 \\ 1 & 2 & 1 \\ 1 & 1 & 2 \end{pmatrix}. Test: if A−λIA - \lambda I reduces to fewer than n−1n - 1 nonzero rows, the eigenspace has dimension 22 or more and one eigenvector is not enough.
Eigenspaces are defined in the Eigenspaces section; the dimension count is rank–nullity: dim⁡Eλ=n−rank⁡(A−λI)\dim E_\lambda = n - \operatorname{rank}(A - \lambda I).

Steps

StepOn A=(211121112)A = \begin{pmatrix} 2 & 1 & 1 \\ 1 & 2 & 1 \\ 1 & 1 & 2 \end{pmatrix}, λ=1\lambda = 1
1Form A−λIA - \lambda I. Use an eigenvalue you have already found; here p(λ)=−(λ−1)2(λ−4)p(\lambda) = -(\lambda - 1)^2(\lambda - 4).A−I=(111111111)A - I = \begin{pmatrix} 1 & 1 & 1 \\ 1 & 1 & 1 \\ 1 & 1 & 1 \end{pmatrix}
2Row reduce to reduced echelon form. Use reduced row echelon form and count the pivots.(111000000)\begin{pmatrix} 1 & 1 & 1 \\ 0 & 0 & 0 \\ 0 & 0 & 0 \end{pmatrix}: rank 11
3Name the free variables. Each non-pivot column is a free variable; their number is n−rank⁡n - \operatorname{rank}, the geometric multiplicity.v2=sv_2 = s, v3=tv_3 = t: dim⁡E1=3−1=2\dim E_1 = 3 - 1 = 2
4Write the parametric vector form. Solve for the pivot variables, then split the solution by parameter.v=s(−1,1,0)+t(−1,0,1)\mathbf{v} = s(-1, 1, 0) + t(-1, 0, 1)
5Read off the basis and check. The vectors multiplying the parameters form the basis; they are independent automatically. Check each with Av=λvA\mathbf{v} = \lambda\mathbf{v}.E1=span⁡{(−1,1,0),(−1,0,1)}E_1 = \operatorname{span}\{(-1, 1, 0), (-1, 0, 1)\} ✓

Worked examples

Example 1 · a simple eigenvalue

dimension 1
λ=4\lambda = 4 for the same AA: A−4I=(−2111−2111−2)A - 4I = \begin{pmatrix} -2 & 1 & 1 \\ 1 & -2 & 1 \\ 1 & 1 & -2 \end{pmatrix}
reduces to v1=v3v_1 = v_3, v2=v3v_2 = v_3; one free variable
rank 22
E4=span⁡{(1,1,1)}E_4 = \operatorname{span}\{(1, 1, 1)\}
check: A(1,1,1)=(4,4,4)A(1, 1, 1) = (4, 4, 4)

Example 2 · too few vectors

defective
A=(3103)A = \begin{pmatrix} 3 & 1 \\ 0 & 3 \end{pmatrix}, λ=3\lambda = 3 twice: A−3I=(0100)A - 3I = \begin{pmatrix} 0 & 1 \\ 0 & 0 \end{pmatrix}
algebraic multiplicity 22
v2=0v_2 = 0, v1v_1 free
rank 11, one free variable
E3=span⁡{(1,0)}E_3 = \operatorname{span}\{(1, 0)\}, dimension 1<21 < 2
geometric multiplicity falls short

Example 3 · a plane of eigenvectors

3×3
A=(200020103)A = \begin{pmatrix} 2 & 0 & 0 \\ 0 & 2 & 0 \\ 1 & 0 & 3 \end{pmatrix}, λ=2\lambda = 2 (double)
lower triangular: 2,2,32, 2, 3
A−2IA - 2I has one nonzero row: v1+v3=0v_1 + v_3 = 0
rank 11
v2=sv_2 = s, v3=tv_3 = t: v=s(0,1,0)+t(−1,0,1)\mathbf{v} = s(0, 1, 0) + t(-1, 0, 1)
pivot variable v1=−tv_1 = -t
E2=span⁡{(0,1,0),(−1,0,1)}E_2 = \operatorname{span}\{(0, 1, 0), (-1, 0, 1)\}, dimension 22
matches multiplicity 22

See it happen

For a simple eigenvalue the eigenspace is a line through the origin. A repeated eigenvalue can give a whole plane of eigenvectors: for λ=1\lambda = 1 above, every vector in the plane v1+v2+v3=0v_1 + v_2 + v_3 = 0 is left unchanged by AA. Or it can give only a line, as for (3103)\begin{pmatrix} 3 & 1 \\ 0 & 3 \end{pmatrix}, where only the xx-axis survives. A basis is any pair of directions that spans the plane, or one direction along the line. Compare the cases in the eigenvalues and eigenvectors visualizer.

Common mistakes

One vector for a two-dimensional eigenspace

✗ E1=span⁡{(−1,1,0)}E_1 = \operatorname{span}\{(-1, 1, 0)\}
✓ E1=span⁡{(−1,1,0),(−1,0,1)}E_1 = \operatorname{span}\{(-1, 1, 0), (-1, 0, 1)\}: one vector per free variable

Basis from the rows

✗ basis {(1,1,1)}\{(1, 1, 1)\}, taken from the nonzero row of A−IA - I
✓ the row gives the equation v1+v2+v3=0v_1 + v_2 + v_3 = 0; the basis comes from its solutions
(1,1,1)(1, 1, 1) is in fact the eigenvector for λ=4\lambda = 4, a different eigenspace.

Assuming dimension equals multiplicity

✗ λ=3\lambda = 3 is a double root of (3103)\begin{pmatrix} 3 & 1 \\ 0 & 3 \end{pmatrix}, so dim⁡E3=2\dim E_3 = 2
✓ only 1≤dim⁡Eλ≤ma(λ)1 \le \dim E_\lambda \le m_a(\lambda) is guaranteed (the multiplicity inequality); here 2−1=12 - 1 = 1

Putting the zero vector in the basis

✗ E3E_3 has basis {0,(1,0)}\{\mathbf{0}, (1, 0)\}
✓ basis {(1,0)}\{(1, 0)\}; a set containing 0\mathbf{0} is never independent