Diagonalisation and the Cayley-Hamilton theorem

Diagonalisation and the Cayley-Hamilton theorem

Spec 3.2–3 · Edexcel A-Level Further Maths: Revision notes

Studying this course? Keep your progress in one place.

Key facts

  • With eigenvectors as the columns of P\mathbf{P}: P−1AP=D\mathbf{P}^{-1}\mathbf{A}\mathbf{P}=\mathbf{D}, the diagonal matrix of eigenvalues in the same order.
  • Powers: An=PDnP−1\mathbf{A}^n=\mathbf{P}\mathbf{D}^n\mathbf{P}^{-1}.
  • A symmetric matrix has real eigenvalues and perpendicular eigenvectors, so QTSQ=D\mathbf{Q}^{\mathrm{T}}\mathbf{S}\mathbf{Q}=\mathbf{D} with Q\mathbf{Q} orthogonal.
  • A matrix is not diagonalisable without two independent eigenvectors.
  • Cayley-Hamilton: A2−tA+dI=0\mathbf{A}^2-t\mathbf{A}+d\mathbf{I}=\mathbf{0}, with tt the trace and dd the determinant.

Diagonalising a matrix

Put the eigenvectors in the columns of P\mathbf{P} and the eigenvalues, in the same order, on the diagonal of D\mathbf{D}.

If A\mathbf{A} has eigenvalues λ1,λ2\lambda_1,\lambda_2 with independent eigenvectors v1,v2\mathbf{v}_1,\mathbf{v}_2, form P\mathbf{P} with the eigenvectors as columns. Then P−1AP=D\mathbf{P}^{-1}\mathbf{A}\mathbf{P}=\mathbf{D}, a diagonal matrix.

The order of the eigenvalues in D\mathbf{D} must match the order of the eigenvectors in P\mathbf{P}. Swapping columns of P\mathbf{P} swaps the entries of D\mathbf{D}.

Check AP=PD\mathbf{A}\mathbf{P}=\mathbf{P}\mathbf{D} to avoid finding P−1\mathbf{P}^{-1}. A matrix is diagonalisable when this works.

v₁Av₁ = 3v₁v₂Av₂ = 2v₂OV1W1V2W2
Example A = (3 1; 0 2): eigenvectors v₁ = (1, 0) and v₂ = (1, −1) keep their direction; A scales them by 3 and 2.
  • P\mathbf{P}eigenvectors as columns
  • P−1AP\mathbf{P}^{-1}\mathbf{A}\mathbf{P}D=(λ100λ2)\mathbf{D}=\begin{pmatrix}\lambda_1&0\\0&\lambda_2\end{pmatrix}

Worked example

Diagonalise A=(4123)\mathbf{A}=\begin{pmatrix}4&1\\2&3\end{pmatrix}.

A\mathbf{A} has eigenvalue 33 with eigenvector u\mathbf{u} and eigenvalue 77 with eigenvector w\mathbf{w}. If P=(w    u)\mathbf{P}=(\mathbf{w}\;\;\mathbf{u}), what is D\mathbf{D}?

Powers of a matrix

Since A=PDP−1\mathbf{A}=\mathbf{P}\mathbf{D}\mathbf{P}^{-1}, the inner factors cancel and An=PDnP−1\mathbf{A}^n=\mathbf{P}\mathbf{D}^n\mathbf{P}^{-1}.

Since A=PDP−1\mathbf{A}=\mathbf{P}\mathbf{D}\mathbf{P}^{-1}, each P−1P\mathbf{P}^{-1}\mathbf{P} in a product cancels, leaving An=PDnP−1\mathbf{A}^n=\mathbf{P}\mathbf{D}^n\mathbf{P}^{-1}.

Powering a diagonal matrix just powers each entry: Dn=(λ1n00λ2n)\mathbf{D}^n=\begin{pmatrix}\lambda_1^n&0\\0&\lambda_2^n\end{pmatrix}.

Always check n=1n=1 returns the original matrix.

0204060801234nλⁿ
  • λ = 2
  • λ = 3
The diagonal entries of Dⁿ are λⁿ: each eigenvalue is powered separately (here λ = 2 and λ = 3).
  • An\mathbf{A}^nPDnP−1\mathbf{P}\mathbf{D}^n\mathbf{P}^{-1}
  • Dn\mathbf{D}^ndiag(λ1n,λ2n)\mathrm{diag}(\lambda_1^n,\lambda_2^n)

Worked example

Find Mn\mathbf{M}^n for M=(2103)\mathbf{M}=\begin{pmatrix}2&1\\0&3\end{pmatrix}.

If A=PDP−1\mathbf{A}=\mathbf{P}\mathbf{D}\mathbf{P}^{-1}, what is A3\mathbf{A}^3?

Symmetric matrices

A symmetric matrix has real eigenvalues and perpendicular eigenvectors, so it is diagonalised by an orthogonal matrix.

A real symmetric matrix (S=ST\mathbf{S}=\mathbf{S}^{\mathrm{T}}) always has real eigenvalues, and eigenvectors for distinct eigenvalues are perpendicular.

If the eigenvectors are normalised, Q\mathbf{Q} (with them as columns) is an orthogonal matrix: QTQ=I\mathbf{Q}^{\mathrm{T}}\mathbf{Q}=\mathbf{I}, so Q−1=QT\mathbf{Q}^{-1}=\mathbf{Q}^{\mathrm{T}}. Then QTSQ=D\mathbf{Q}^{\mathrm{T}}\mathbf{S}\mathbf{Q}=\mathbf{D}.

Perpendicular columns are not enough: each must also have length 11.

λ = 6λ = 1OAB
Eigenvectors (2, 1) and (1, −2) of a symmetric matrix meet at a right angle.

Worked example

S=(5222)\mathbf{S}=\begin{pmatrix}5&2\\2&2\end{pmatrix} has eigenvalues 66 and 11. Find Q\mathbf{Q} with QTSQ=D\mathbf{Q}^{\mathrm{T}}\mathbf{S}\mathbf{Q}=\mathbf{D}.

A symmetric matrix has an eigenvector (34)\begin{pmatrix}3\\4\end{pmatrix} for one eigenvalue. Which could be an eigenvector for a different eigenvalue?

When diagonalisation fails

A 2×22\times2 matrix needs two independent eigenvectors; without them P\mathbf{P} is not invertible.

A 2×22\times2 matrix with a repeated eigenvalue that has only one eigenvector direction cannot be diagonalised, since P\mathbf{P} would not be invertible.

For example (3−111)\begin{pmatrix}3&-1\\1&1\end{pmatrix} has λ=2\lambda=2 twice and only the eigenvector (11)\begin{pmatrix}1\\1\end{pmatrix}.

With complex eigenvalues a real P\mathbf{P} does not exist, though a complex one does. Always check the two independent eigenvectors are not multiples; det⁡P≠0\det\mathbf{P}\neq0 is a quick test.

(1, 1)A(1, 1) = (2, 2)(1, 0)A(1, 0) = (3, 1)OEFUW
A = (3 −1; 1 1): A(1, 1) = (2, 2) stays on its line, but A(1, 0) = (3, 1) leaves it, so there is only one eigenvector direction.

Diagonalisable

  • Two distinct real eigenvalues
  • Symmetric matrices
  • Repeated eigenvalue with two eigenvectors, e.g. 3I3\mathbf{I}

Not diagonalisable

  • Repeated eigenvalue, one direction, e.g. (2102)\begin{pmatrix}2&1\\0&2\end{pmatrix}
  • Complex eigenvalues (no real P)

(2102)\begin{pmatrix}2&1\\0&2\end{pmatrix} has eigenvalue 22 twice and only one eigenvector direction. Is it diagonalisable?

The Cayley-Hamilton theorem

Every square matrix satisfies its own characteristic equation.

The Cayley-Hamilton theorem: for a 2×22\times2 matrix with characteristic equation λ2−tλ+d=0\lambda^2-t\lambda+d=0 (trace tt, determinant dd),

A2−tA+dI=0.\mathbf{A}^2-t\mathbf{A}+d\mathbf{I}=\mathbf{0}.

The constant term becomes dId\mathbf{I}, not just dd. This reduces any power of A\mathbf{A} to a combination of A\mathbf{A} and I\mathbf{I}.

The same holds for 3×33\times3 matrices, with a cubic.

−11234567−4−224681012xy(1, 0)(5, 0)(3, −4)λ² − 6λ + 5
The characteristic polynomial of C = (2 1; 3 4) has roots 1 and 5, its eigenvalues. Cayley-Hamilton: put C in place of λ and get 0.
  • λ2−tλ+d=0\lambda^2-t\lambda+d=0A2−tA+dI=0\mathbf{A}^2-t\mathbf{A}+d\mathbf{I}=\mathbf{0}

Worked example

Use Cayley-Hamilton to find C2\mathbf{C}^2 for C=(2134)\mathbf{C}=\begin{pmatrix}2&1\\3&4\end{pmatrix}.

A matrix has characteristic equation λ2−5λ+6=0\lambda^2-5\lambda+6=0. Which expresses A2\mathbf{A}^2?

Inverses and powers from Cayley-Hamilton

Rearrange the relation to A×(…)=kI\mathbf{A}\times(\ldots)=k\mathbf{I} for an inverse, or multiply by A\mathbf{A} for higher powers.

Rearrange C2−6C+5I=0\mathbf{C}^2-6\mathbf{C}+5\mathbf{I}=\mathbf{0} as C(6I−C)=5I\mathbf{C}(6\mathbf{I}-\mathbf{C})=5\mathbf{I}. Then C−1=15(6I−C)=15(4−1−32)\mathbf{C}^{-1}=\frac15(6\mathbf{I}-\mathbf{C})=\frac15\begin{pmatrix}4&-1\\-3&2\end{pmatrix}.

For higher powers multiply the relation by C\mathbf{C} and substitute repeatedly: C3=6C2−5C=31C−30I\mathbf{C}^3=6\mathbf{C}^2-5\mathbf{C}=31\mathbf{C}-30\mathbf{I}.

These give the same results as diagonalisation, so use whichever the question hints at.

  1. 1

    Start from the relation

    C² − 6C + 5I = 0

  2. 2

    Factor out C

    C(6I − C) = 5I

  3. 3

    Divide by 5 for the inverse

    C⁻¹ = (6I − C)/5

  4. 4

    Multiply by C for powers

    C³ = 6C² − 5C = 31C − 30I

Using C² − 6C + 5I = 0 for C = (2 1; 3 4).
  • A−1\mathbf{A}^{-1}1d(tI−A)\dfrac{1}{d}\left(t\mathbf{I}-\mathbf{A}\right)

A matrix satisfies A2−4A+3I=0\mathbf{A}^2-4\mathbf{A}+3\mathbf{I}=\mathbf{0}. What is A−1\mathbf{A}^{-1}?

Try an exam question

The matrix A=(1232)\mathbf{A}=\begin{pmatrix}1&2\\3&2\end{pmatrix}. Use the Cayley-Hamilton theorem to find A−1\mathbf{A}^{-1}.

[5 marks]

That's the notes covered.

Carry on to the next subtopic.

Try a question

Which is P−1AP\mathbf{P}^{-1}\mathbf{A}\mathbf{P}?

1 mark, marked against the mark scheme

Revision notes