When a matrix stretches an eigenvector without knocking it off its line, it stretches every other vector on that same line, and any combination of vectors stretched by that same factor, right along with it.
1
What an Eigenspace Is
The eigenspace for an eigenvalue λ is the entire subspace of vectors that get stretched by λ, together with the zero vector at the origin. Depending on the matrix, an eigenspace can be a 1D line, a 2D plane, or even all of 3D space.
What an Eigenspace Is
Subtracting λ from the diagonal of A shifts the transformation so that every vector stretched by λ under A gets sent to the origin under A−λI:
(A−λI)v=0⟹eigenspace for λEλ=null space of A−λI
To describe an eigenspace, we find a basis for the null space of A−λI using its free variables.
Whenever a vector v lies on the eigenspace line through the origin, Av stays on that same line stretched by λ, while (A−λI)v lands at the origin.
When a vector sits on the eigenspace line, the matrix just stretches it along that line, and shifting the matrix squashes it to zero at the origin.
2
Algebraic and Geometric Multiplicity
An eigenvalue has two numbers attached to it: its algebraic multiplicity is the number of times λ repeats as a root in the factored characteristic polynomial, and its geometric multiplicity is the dimension of its eigenspace. To see why those two numbers don't always match, let's compare two 2×2 matrices that differ only by a 1 in the top-right corner:
Subtract λ from the diagonal and set the determinant to zeroThe 1 in the top-right corner of Matrix 2 gets multiplied by the 0 in the bottom-left, so both matrices give the exact same characteristic equation and the same repeated root λ=3:
Find the free variables of Matrix 1 minus 3ISubtracting 3 from the diagonal makes every entry of Matrix 1 zero, which puts no restriction on x or y and leaves both variables free:
[0000][xy]=[00]⟹0x+0y=0⟹[xy]=x[10]+y[01]
Find the free variables of Matrix 2 minus 3IThe 1 in the top-right corner stays after subtracting 3 from the diagonal and forces y=0, while the all-zero first column leaves x as the only free variable:
[0010][xy]=[00]⟹0x+1y=0⟹y=0⟹[x0]=x[10]
Both matrices have λ=3 with algebraic multiplicity 2, but Matrix 1 stretches the entire 2D plane with basis {[10],[01]} and geometric multiplicity 2, whereas the sideways shear in Matrix 2 tilts every non-horizontal vector off its line and shrinks the eigenspace to the x-axis with basis {[10]} and geometric multiplicity 1.
Try comparing a pure stretch with a stretch plus a shear:
A pure stretch keeps every direction on its own line, while adding any sideways shear pushes every direction off its line except the horizontal axis.
3
Eigenspaces of a Triangular Matrix
Now let's move to 3×3 matrices, starting with an upper-triangular matrix where the zeros below the main diagonal let us read the eigenvalues straight off the diagonal. We want to find every eigenvalue, its algebraic and geometric multiplicities, and a basis for each eigenspace of
A=200120305
Eigenspaces of a Triangular Matrix
A=200120305
λ=?Eλ=?
Text solution
Multiply the diagonal entries to factor the characteristic equationBecause A has all zeros below the main diagonal, the determinant of A−λI is the product of its diagonal entries, which gives λ=2 with algebraic multiplicity 2 and λ=5 with algebraic multiplicity 1:
(2−λ)(2−λ)(5−λ)=(2−λ)2(5−λ)1=0
Subtract 2 from the diagonal and solve (A−2I)v=0The bottom row forces z=0 and the top row then forces y=0, while the all-zero first column leaves x as our single free variable:
Subtract 5 from the diagonal and solve (A−5I)v=0The middle row forces y=0, which simplifies the top row so that x must equal the single free variable z:
For λ=2, the off-diagonal 1 shrinks the eigenspace from a plane to a 1D line with basis ⎩⎨⎧100⎭⎬⎫ and geometric multiplicity 1; for λ=5, the eigenspace is a 1D line with basis ⎩⎨⎧101⎭⎬⎫ and geometric multiplicity 1.
Every eigenvalue has at least one line of eigenvectors, and the dimension of its eigenspace can never be larger than the number of times that eigenvalue repeats as a root:
1≤Geometric Multiplicity≤Algebraic Multiplicity
4
Row Reducing to Find Tilted Eigenspaces
When a matrix is full of nonzero entries and its eigenspaces are tilted in 3D space, we can't read the free variables straight off A−λI. Instead, we row reduce A−λI to find the pivots and free variables. Let's find the multiplicities and a basis for each eigenspace of
Subtract 2 from the diagonal and row reduce B−2IAdding Row 1 to Row 2 and subtracting Row 1 from Row 3 makes both bottom rows all zeros in one step, leaving one pivot in the first column (x) and two free variables (y and z):
Substitute for the pivot variable and split by the free variablesPlugging x=y−2z into the vector and separating the y-part from the z-part gives us one basis vector for each free variable:
xyz=y−2zyz=y110+z−201
Subtract 6 from the diagonal and row reduce B−6ISwapping the third row to the top puts a 1 in the top-left corner to make the rest of the first column zero, and reducing the remaining rows leaves z as the single free variable:
Write the pivot variables in terms of the free variable zThe two nonzero rows of the reduced matrix give both x and y in terms of z:
{x−z=0⟹x=zy+z=0⟹y=−z⟹xyz=z−zz=z1−11
For λ=2, both the algebraic and geometric multiplicities are 2, and the basis ⎩⎨⎧110,−201⎭⎬⎫ spans a tilted 2D plane through the origin where every vector stretches by 2; for λ=6, both multiplicities are 1, and the basis ⎩⎨⎧1−11⎭⎬⎫ spans a 1D line stretched by 6.
If row reducing A−λI ever leaves a pivot in every column and no free variables at all, stop and check your arithmetic: because geometric multiplicity is at least 1, an eigenvalue must always give at least one row of zeros.
Every linear combination y110+z−201 of the two basis vectors stays on its own line inside that tilted plane and stretches by a factor of 2.
Every combination of the two basis vectors in this tilted plane stays on its own line and stretches by 2.
Key concepts
To find the multiplicities and eigenspace for an eigenvalue λ, work from the factored characteristic equation to the null space of A−λI:
Algebraic Multiplicity sets the maximum dimension
The power on (λ0−λ) in the factored characteristic equation is the Algebraic Multiplicity, and the eigenspace dimension has to obey 1≤Geometric Multiplicity≤Algebraic Multiplicity. In (2−λ)2(5−λ)1=0, the root λ=2 can have up to a 2D plane of eigenvectors, whereas λ=5 has power 1 and is guaranteed to have a 1D line.
Geometric Multiplicity counts the free variables in (A−λI)v=0
The eigenspace Eλ is the null space of A−λI, and each free variable gives one basis vector.
When you can read the variables without row reducing: If A−λI is already triangular or has an all-zero column, read the equations straight from the rows. In [0010][xy]=[00], the top row gives y=0 and the all-zero first column leaves x free right away, giving [x0]=x[10].
When to row reduce: When A−λI has nonzero entries across its rows like B−2I=1−11−11−12−22, row reduce to find the pivots, write the pivot variables in terms of the free variables (x=y−2z), and split the vector by the free variables so each one gives you a basis vector: y−2zyz=y110+z−201.
5
Practice
1
A 4×4 matrix has the factored characteristic equation (4−λ)3(1−λ)=0. For each eigenvalue, find its algebraic multiplicity and list all possible values for its geometric multiplicity.
The exponent on each factor gives the largest possible dimension of its eigenspace; what is the smallest dimension an eigenspace for an eigenvalue can ever have?
Read the powers off (4−λ)3(1−λ)1=0 and use 1≤Geometric Multiplicity≤Algebraic Multiplicity. Be careful not to include 0 in the list (which would give 0,1,2,3 for λ=4) or assume the geometric multiplicity must equal 3. Every eigenvalue has at least one line of eigenvectors, so the possible values always run from 1 up to the algebraic multiplicity.
For λ=4, the algebraic multiplicity is 3 and the geometric multiplicity can be 1,2, or 3. For λ=1, the algebraic multiplicity is 1 and the geometric multiplicity must be 1.
2
For the matrix A=[4−204], find its eigenvalue, the algebraic multiplicity, a basis for the eigenspace, and the geometric multiplicity.
Subtract λ from the diagonal to find the repeated root, then look at which variable is forced to 0 by the bottom row of A−λI and which column is all zeros.
Because A is triangular with zeros above the diagonal, its characteristic equation is (4−λ)2=0, giving λ=4 with algebraic multiplicity 2. Subtract 4 from the diagonal:
A−4I=[0−200],[0−200][xy]=[00]⟹−2x=0⟹x=0
Be careful not to write [10] by treating x as free the way it was for Matrix 2: here the −2 is in the first column and forces x=0, while the all-zero second column leaves y free, giving [0y]=y[01].
λ=4 has algebraic multiplicity 2, eigenspace basis {[01]}, and geometric multiplicity 1.
3
Find every eigenvalue, its algebraic and geometric multiplicities, and a basis for each eigenspace of the upper-triangular matrix A=300030201.
After subtracting each eigenvalue from the diagonal, check how many columns of A−λI are all zeros before writing the row equations.
From the diagonal entries, the characteristic equation is (3−λ)2(1−λ)=0.
For λ=3 (algebraic multiplicity 2), subtract 3 from the diagonal:
A−3I=00000020−2⟹2z=0⟹z=0
The first and second columns are both all zeros, leaving x and y free:
xyz=xy0=x100+y010
For λ=1 (algebraic multiplicity 1), subtract 1 from the diagonal:
A−1I=200020200⟹{2y=0⟹y=02x+2z=0⟹x=−z
When solving 2x+2z=0, remember to flip the sign when subtracting 2z from both sides (dropping the minus sign gives 101 instead of −101). With z free, xyz=−z0z=z−101.
For λ=3, algebraic multiplicity =2, geometric multiplicity =2, and basis =⎩⎨⎧100,010⎭⎬⎫. For λ=1, algebraic multiplicity =1, geometric multiplicity =1, and basis =⎩⎨⎧−101⎭⎬⎫.
4
The matrix B=22−125−2−1−22 has characteristic equation (1−λ)2(7−λ)=0. Find the algebraic multiplicity, geometric multiplicity, and a basis for the eigenspace for each eigenvalue.
For λ=1, compare the three rows of B−I to see how many rows are multiples of Row 1; for λ=7, row reduce B−7I until only the free variable z has no pivot.
For λ=1 (algebraic multiplicity 2), subtract 1 from the diagonal and make the bottom two rows zero:
Don't plug in y=1 and z=1 together to get a single vector −111 (which loses a basis vector and gives a geometric multiplicity of 1 instead of 2); split the vector into its y-part and z-part:
xyz=−2y+zyz=y−210+z101
For λ=7 (algebraic multiplicity 1), subtract 7 from the diagonal and row reduce:
For λ=1, algebraic multiplicity =2, geometric multiplicity =2, and basis =⎩⎨⎧−210,101⎭⎬⎫. For λ=7, algebraic multiplicity =1, geometric multiplicity =1, and basis =⎩⎨⎧−1−21⎭⎬⎫.
5
Consider C=5003206k2. For the eigenvalue λ=2, find the value of k for which the geometric multiplicity is 2, and find a basis for the eigenspace both for that value of k and when k=1.
Subtract 2 from the diagonal to form C−2I, and look at the second row to see how k determines whether z is free or forced to 0.
Subtract 2 from the diagonal of C:
C−2I=3003006k0
The middle row says kz=0. For the geometric multiplicity to be 2, we need two free variables, which happens only when k=0 so that the middle row becomes all zeros.
When k=0, don't jump straight to 010 and 001 by ignoring the top row. Even with two zero rows at the bottom, the top row 3x+3y+6z=0 still relates x to the free variables y and z:
x⟹xyz=−y−2z=−y−2zyz=y−110+z−201
When k=1, the middle row 1z=0 forces z=0, which simplifies the top row to x=−y with only y free:
xyz=−yy0=y−110
The geometric multiplicity is 2 when k=0, which leaves both y and z free and gives the basis ⎩⎨⎧−110,−201⎭⎬⎫. When k=1,z=0 leaves only y free, so the geometric multiplicity is 1 and a basis is ⎩⎨⎧−110⎭⎬⎫.