Diagonalisation of matrices

Diagonalisation of matrices

Spec C11 · AQA A-Level Further Maths: Revision notes

Studying this course? Keep your progress in one place.

Key facts

  • A diagonal matrix is raised to a power by raising each diagonal entry to that power.
  • With real eigenvalues, M=UDU−1M=UDU^{-1}: the columns of UU are eigenvectors and DD holds the matching eigenvalues in the same order.
  • Mn=UDnU−1M^n=UD^nU^{-1}: only DD is raised to the power.
  • Check MU=UDMU=UD, which avoids finding U−1U^{-1}.
  • A matrix with no real eigenvalues cannot be diagonalised with real matrices.

Diagonal matrices

To raise a diagonal matrix to a power, raise each diagonal entry to that power.

A diagonal matrix has zeros everywhere except on the leading diagonal.

(d100d2)n=(d1n00d2n).\begin{pmatrix} d_1 & 0 \\ 0 & d_2 \end{pmatrix}^n=\begin{pmatrix} d_1^n & 0 \\ 0 & d_2^n \end{pmatrix}.

The same holds for 3×33\times3. Diagonalisation uses this: it rewrites a matrix so the hard work is done on a diagonal matrix instead.

OA1B1C1A2B2C2A3B3C3
D = diag(2, 3) stretches the unit square to 2 by 3; D² stretches it to 4 by 9. Each axis is scaled by its own diagonal entry raised to the power.
  • (d100d2)n\begin{pmatrix} d_1 & 0 \\ 0 & d_2 \end{pmatrix}^n(d1n00d2n)\begin{pmatrix} d_1^n & 0 \\ 0 & d_2^n \end{pmatrix}

Evaluate (2003)4\begin{pmatrix} 2 & 0 \\ 0 & 3 \end{pmatrix}^4.

Writing M=UDU−1M=UDU^{-1}

Put the eigenvectors in the columns of UU and the matching eigenvalues, in the same order, on the diagonal of DD.

If MM has real eigenvalues λ1,λ2,…\lambda_1,\lambda_2,\dots with eigenvectors v1,v2,…\mathbf{v}_1,\mathbf{v}_2,\dots, then

M=UDU−1.M=UDU^{-1}.

This is diagonalisation. The kkth column of UU must go with the kkth diagonal entry of DD. Check by confirming U−1MU=DU^{-1}MU=D.

Mv₁ = 2v₁v₁Mv₂ = 5v₂v₂OV1W1V2W2
For M = (3 1; 2 4): eigenvectors (1, -1) and (1, 2) with eigenvalues 2 and 5. M stretches each along its own line.
  • MMUDU−1UDU^{-1}
  • Columns of UUeigenvectors
  • Diagonal of DDmatching eigenvalues, same order

Worked example

Diagonalise M=(3124)M=\begin{pmatrix} 3 & 1 \\ 2 & 4 \end{pmatrix}.

If the first column of UU is the eigenvector for λ=5\lambda=5 and the second is the eigenvector for λ=2\lambda=2, what is DD?

The method

Find the eigenvalues, then the eigenvectors, build UU and DD in matching order, and invert UU.

The steps always work, and the order of the eigenvalues is your choice.

Any non-zero multiples of the eigenvectors work, and any order works, provided UU and DD agree. So UU and DD are not unique.

Check UDU−1UDU^{-1} by testing one entry of MM, or check MU=UDMU=UD.

  1. 1

    Find the eigenvalues

    solve det⁡(M−λI)=0\det(M-\lambda I)=0

  2. 2

    Find the eigenvectors

    solve (M−λI)v=0(M-\lambda I)\mathbf{v}=\mathbf{0} for each

  3. 3

    Build UU and DD

    eigenvectors as columns; eigenvalues in the same order

  4. 4

    Find U−1U^{-1}

    for 2×22\times2, 1ad−bc(d−b−ca)\frac{1}{ad-bc}\begin{pmatrix} d & -b \\ -c & a \end{pmatrix}

Diagonalising a matrix

Why is MU=UDMU=UD a useful check?

Powers of a matrix

The inner U−1UU^{-1}U pairs cancel, so Mn=UDnU−1M^n=UD^nU^{-1} and only the diagonal entries are raised to the power.

M2=UDU−1UDU−1=UD2U−1,Mn=UDnU−1.M^2=UDU^{-1}UDU^{-1}=UD^2U^{-1},\qquad M^n=UD^nU^{-1}.

This gives a formula for any nn. For M=(3124)M=\begin{pmatrix} 3 & 1 \\ 2 & 4 \end{pmatrix}:

Mn=13(2×2n+5n5n−2n2×5n−2×2n2n+2×5n).M^n=\frac13\begin{pmatrix} 2\times2^n+5^n & 5^n-2^n \\ 2\times5^n-2\times2^n & 2^n+2\times5^n \end{pmatrix}.

The eigenvalues of MnM^n are λn\lambda^n, and the eigenvectors stay the same. Check with n=1n=1 (you get MM) and n=0n=0 (you get II).

0.511.522.5320406080100120xy(2, 4)(2, 25)2ⁿ5ⁿ
The diagonal entries of Dⁿ are 2ⁿ and 5ⁿ. The larger eigenvalue soon dominates the powers of M.
  • MnM^nUDnU−1UD^nU^{-1}
  • DnD^ndiag(λ1n,λ2n)\mathrm{diag}(\lambda_1^n,\lambda_2^n)

Use the formula for MnM^n above with n=0n=0. What matrix do you get?

3x3 and real eigenvalues

The method is identical for 3×33\times3, but it needs real eigenvalues: a matrix with none cannot be diagonalised with real matrices.

For a 3×33\times3 matrix use a 3×33\times3 matrix UU of eigenvectors and a diagonal DD with three eigenvalues.

The method needs real eigenvalues. The 90∘90^\circ rotation (0−110)\begin{pmatrix} 0 & -1 \\ 1 & 0 \end{pmatrix} has characteristic equation λ2+1=0\lambda^2+1=0, so no real eigenvalues and no real UDU−1UDU^{-1}.

(1, 0)image (0, 1)OAB
The 90° rotation turns every vector off its own line: (1, 0) goes to (0, 1). There is no real eigenvector, so no real diagonalisation.

Real eigenvalues

  • Diagonalise as UDU−1UDU^{-1}
  • Real UU and DD
  • Powers by UDnU−1UD^nU^{-1}

No real eigenvalues

  • e.g. 90∘90^\circ rotation
  • λ2+1=0\lambda^2+1=0
  • No real eigenvectors, cannot diagonalise with real matrices

Worked example

Find B5B^5 for B=(100031013)B=\begin{pmatrix} 1 & 0 & 0 \\ 0 & 3 & 1 \\ 0 & 1 & 3 \end{pmatrix}.

Can the 90∘90^\circ rotation (0−110)\begin{pmatrix} 0 & -1 \\ 1 & 0 \end{pmatrix} be diagonalised using real matrices?

Try an exam question

A=(4123)\mathbf{A}=\begin{pmatrix} 4 & 1 \\ 2 & 3 \end{pmatrix} has eigenvalues 22 and 55 with eigenvectors (1−2)\begin{pmatrix} 1 \\ -2 \end{pmatrix} and (11)\begin{pmatrix} 1 \\ 1 \end{pmatrix} respectively. Write A\mathbf{A} in the form UDU−1\mathbf{U}\mathbf{D}\mathbf{U}^{-1} and hence find A3\mathbf{A}^3.

[5 marks]

That's the notes covered.

Carry on to the next subtopic.

Try a question

Which of these pairs of matrices UU and DD satisfies M=UDU−1M=UDU^{-1}?

1 mark, marked against the mark scheme

Revision notes