Calculus &
Linear Algebra II

Chapter 16

16 Diagonalisation

By the end of this section, you should be able to answer the following questions:

  • How do you find a matrix $P$ which diagonalises a given matrix $A$?
  • How do you determine if $A$ is diagonalisable?
  • What are two applications of diagonalisation?



16 Diagonalisation

A square matrix $A$ is diagonalisable if there is a non-singular matrix $P$ such that $P^{-1}AP$ is a diagonal matrix. Here we consider the question: given a matrix, is it diagonalisable? If so, how do we find $P$?

The secret to constructing such a $P$ is to let the columns of $P$ be the eigenvectors of $A$. We immediately have that $AP=PD,$ where $D$ is a diagonal matrix with eigenvalues on the diagonal. We know from section 15.1 that $P$ is invertible if and only if the columns of $P$ are linearly independent.



16 Diagonalisation

Hence, we have the following result:

The $n\times n$ matrix $A$ is diagonalisable if and only if $A$ has $n$ linearly independent eigenvectors.

Is the matrix

$A=\left(\begin{array}{rrr}-3&1&0\\1&-2&1\\0&1&-3\end{array}\right)$

diagonalisable?





16 Diagonalisation

We know this matrix has eigenvalues $-3, -1, -4,$ with corresponding eigenvectors

$\v_1=\mat{r}{1\\0\\-1},\;$ $\;\v_2=\mat{c}{1\\2\\1},\;$ $\;\v_3=\mat{r}{1\\-1\\1}$.

We form the matrix

$\ds P=\left(\begin{array}{rrr}1&1&1\\0&\; 2&-1\\-1&1&1\end{array}\right) $

whose columns are these eigenvectors.

$A=\left(\begin{array}{rrr}-3&1&0\\1&-2&1\\0&1&-3\end{array}\right)$


16 Diagonalisation

$\;A=\left(\begin{array}{rrr}-3&1&0\\1&-2&1\\0&1&-3\end{array}\right)\,\;$ and $\;\ds P=\left(\begin{array}{rrr}1&1&1\\0&\; 2&-1\\-1&1&1\end{array}\right) $

We can check directly that $P^{-1}AP=D,$ where $$\left(\begin{array}{rrr}-3&0&0\\0&-1&0\\0&0&-4\end{array}\right)=D$$ is a diagonal matrix with the eigenvalues on the diagonal.





16 Diagonalisation

So steps for diagonalising an $n\times n$ matrix $A$:

  1. Find eigenvalues $\lambda_1, \lambda_2, \cdots,$ and the corresponding eigenvectors $\v_1, \v_2, \cdots,$ of $A.$
  2. Check if $A$ has $n$ linearly independent eigenvectors.
  3. If no, $A$ is not diagonalizable.
    If yes, $A$ is diagonalizable. In this case, form matrix $P=(\v_1|\v_2|\cdots|\v_n)$. Then $P^{-1}$ exists and $P$ diagonalizes $A,$ i.e., $P^{-1}AP=diag\{\lambda_1, \cdots, \lambda_n\}.$


16.1 Similar matrices

Two matrices $A$ and $B$ are similar if there is a non-singular matrix $P$ such that $B=P^{-1}AP.$

The statements "$A$ is diagonalisable" and "$A$ is similar to a diagonal matrix" are equivalent.

16.1.1 Theorem (similar matrices)

Similar matrices have the same eigenvalues.




16.1 Similar matrices

16.1.1 Theorem (similar matrices)

Similar matrices have the same eigenvalues.

In fact, if $B=P^{-1}AP$ and $\v$ is an eigenvector of $A$ corresponding to eigenvalue $\lambda,$ then $P^{-1}\v$ is an eigenvector of $B$ corresponding to eigenvalue $\lambda$. This is because

$B\left(P^{-1}\v\right)$ $=$ $\left(P^{-1}AP\right)P^{-1}\v$
$=$ $ P^{-1}\left(A\v\right)$
$=$ $P^{-1}\left(\lambda \v\right)$
$=$ $\lambda \left(P^{-1}\v\right)$

16.2 A closer look at the diagonal matrix

Let the matrix $A$ be $n\times n$ with $n$ linearly independent eigenvectors $\v_1,\dots, \v_n$ corresponding to eigenvalues $\lambda_1,\dots,\lambda_n.$ Let $$ P=(\v_1| \dots |\v_n) $$ be the $n\times n$ matrix whose columns are the eigenvectors. Then $$ P^{-1}AP=\left(\begin{array}{cccc} \lambda_1&0&\cdots&0\\0&\lambda_2&\cdots&0\\\vdots&\vdots&\ddots&\vdots\\ 0&0&\cdots&\lambda_n\end{array}\right), $$ the diagonal matrix with the eigenvalues down the main diagonal.


16.2 A closer look at the diagonal matrix

Let the matrix $A$ be $n\times n$ with $n$ linearly independent eigenvectors $\v_1,\dots, \v_n$ corresponding to eigenvalues $\lambda_1,\dots,\lambda_n.$ Let $$ P=(\v_1| \dots |\v_n) $$ be the $n\times n$ matrix whose columns are the eigenvectors. Then $$ P^{-1}AP=\left(\begin{array}{cccc} \lambda_1&0&\cdots&0\\0&\lambda_2&\cdots&0\\\vdots&\vdots&\ddots&\vdots\\ 0&0&\cdots&\lambda_n\end{array}\right), $$ the diagonal matrix with the eigenvalues down the main diagonal.

The important point here is the order in which the eigenvalues appear. They correspond to the order in which the associated eigenvectors appear in the columns of $P.$


16.3 Diagonalisability

We know that an $n\times n$ matrix $A$ is diagonalisable if and only if $A$ has $n$ linearly independent eigenvectors.

Now say $\lambda_1,\dots,\lambda_m$ are distinct eigenvalues of $A$, with corresponding eigenvectors $\v_1,\dots,\v_m.$ Then we have also seen that $\v_1,\dots,\v_m$ are linearly independent.

Hence if $A$ is $n\times n$ with $n$ distinct eigenvalues, then $A$ is diagonalisable.

The question remains, if $A$ has fewer than $n$ distinct eigenvalues, how do we know if $A$ is diagonalisable?



16.3.1 Example

Let $A=\left(\begin{array}{ccc}2&1&3\\0&1&0\\0&0&1\end{array}\right)$ and $B=\left(\begin{array}{ccc}2&1&3\\0&1&1\\0&0&1\end{array}\right)$.

Easy to see the characteristic equation of both $A$ and $B$ is $$(2-\lambda)(1-\lambda)^2=0,$$ so $\lambda=2,1,1.$



16.3.1 Example

Let $A=\left(\begin{array}{ccc}2&1&3\\0&1&0\\0&0&1\end{array}\right)$ and $B=\left(\begin{array}{ccc}2&1&3\\0&1&1\\0&0&1\end{array}\right).$ Characteristic equation of both $A$ and $B$ is $(2-\lambda)(1-\lambda)^2=0,$ so $\lambda=2,1,1.$

Note that both matrices have two ($n\lt 3$) distinct eigenvalues.

For $A:$ We have that for $\lambda=2$

$\ds \left(A-\lambda I\right)\mathbf x $ $= \left( \begin{array}{rrr} 0 & 1 & 3 \\ 0 & -1 &0 \\ 0 & 0 & -1 \\ \end{array} \right) \left( \begin{array}{c} x_1 \\ x_2 \\ x_3 \\ \end{array} \right) $ $ =\left( \begin{array}{c} 0 \\ 0 \\ 0 \\ \end{array} \right) $

$\Ra$ $x_3 = x_2 = 0,$ which means that $x_1$ is free variable.

Thus $ \left( \begin{array}{c} x_1 \\ x_2 \\ x_3 \\ \end{array} \right) =\left( \begin{array}{c} x_1 \\ 0 \\ 0 \\ \end{array} \right) $ $ =x_1 \left( \begin{array}{c} 1 \\ 0 \\ 0 \\ \end{array} \right). $ The eigenvector is $\v_1 = \left( \begin{array}{c} 1 \\ 0 \\ 0 \\ \end{array} \right). $


16.3.1 Example

For $A:$ Now for $\lambda=1$

$\ds \left(A-\lambda I\right)\mathbf x= \left( \begin{array}{ccc} 1 & 1 & 3 \\ 0 & 0 &0 \\ 0 & 0 & 0 \\ \end{array} \right) \left( \begin{array}{c} x_1 \\ x_2 \\ x_3 \\ \end{array} \right) =\left( \begin{array}{c} 0 \\ 0 \\ 0 \\ \end{array} \right) $ $\;\Ra\;$ $\;x_1+x_2+3x_3=0.$

So $\;x_1=-x_2-3x_3$. Thus $\; \left( \begin{array}{c} x_1 \\ x_2 \\ x_3 \\ \end{array} \right) =\left( \begin{array}{c} -x_2-3x_3 \\ x_2 \\ x_3 \\ \end{array} \right) $ $ =x_2 \left( \begin{array}{c} -1 \\ 1 \\ 0 \\ \end{array} \right) + x_3 \left( \begin{array}{c} -3 \\ 0 \\ 1 \\ \end{array} \right). $

The eigenvectors are $\v_2 = \left( \begin{array}{c} -1 \\ 1 \\ 0 \\ \end{array} \right) $ and $\v_3 = \left( \begin{array}{c} -3 \\ 0 \\ 1 \\ \end{array} \right). $


16.3.1 Example

$A=\left(\begin{array}{ccc}2&1&3\\0&1&0\\0&0&1\end{array}\right)$

Therefore $A$ is diagonalisable, since $A$ has three linearly independent vectors.

Eigenvalue Eigenvector
$\lambda=2$ $\left( \begin{array}{r} 1 \\ 0 \\ 0 \\ \end{array} \right)$
$\lambda=1$ $\left( \begin{array}{r} -1 \\ 1 \\ 0 \\ \end{array} \right)$
$\lambda=1$ $\left( \begin{array}{r} -3 \\ 0 \\ 1 \\ \end{array} \right)$

16.3.1 Example

Let $A=\left(\begin{array}{ccc}2&1&3\\0&1&0\\0&0&1\end{array}\right)$ and $B=\left(\begin{array}{ccc}2&1&3\\0&1&1\\0&0&1\end{array}\right).$ Characteristic equation of both $A$ and $B$ is $(2-\lambda)(1-\lambda)^2=0,$ so $\lambda=2,1,1.$

Now for $B:$ We have that for $\lambda=2$

$\ds \left(A-\lambda I\right)\mathbf x $ $= \left( \begin{array}{rrr} 0 & 1 & 3 \\ 0 & -1 & 1 \\ 0 & 0 & -1 \\ \end{array} \right) \left( \begin{array}{c} x_1 \\ x_2 \\ x_3 \\ \end{array} \right) $ $ =\left( \begin{array}{c} 0 \\ 0 \\ 0 \\ \end{array} \right) $

$\Ra$ $x_2 = x_3 = 0,$ which means that $x_1$ is free variable.

Thus $ \left( \begin{array}{c} x_1 \\ x_2 \\ x_3 \\ \end{array} \right) =\left( \begin{array}{c} x_1 \\ 0 \\ 0 \\ \end{array} \right) $ $ =x_1 \left( \begin{array}{c} 1 \\ 0 \\ 0 \\ \end{array} \right). $ The eigenvector is $\v_1 = \left( \begin{array}{c} 1 \\ 0 \\ 0 \\ \end{array} \right). $



16.3.1 Example

Now for $B:$ On the other hand, for $\lambda=1$

$\ds \left(A-\lambda I\right)\mathbf x $ $= \left( \begin{array}{ccc} 1 & 1 & 3 \\ 0 & 0 & 1 \\ 0 & 0 & 0 \\ \end{array} \right) \left( \begin{array}{c} x_1 \\ x_2 \\ x_3 \\ \end{array} \right) $ $ =\left( \begin{array}{c} 0 \\ 0 \\ 0 \\ \end{array} \right) $

$\Ra$ $x_3 = 0,$ and then $x_1+x_2+ 3x_3=0.$

Thus $ \left( \begin{array}{c} x_1 \\ x_2 \\ x_3 \\ \end{array} \right) =\left( \begin{array}{r} x_1 \\ -x_1 \\ 0 \\ \end{array} \right) $ $ =x_1 \left( \begin{array}{r} 1 \\ -1 \\ 0 \\ \end{array} \right). $ The eigenvector is $\v_2 = \left( \begin{array}{r} 1 \\ -1 \\ 0 \\ \end{array} \right). $



16.3.1 Example

$B=\left(\begin{array}{ccc}2&1&3\\0&1&1\\0&0&1\end{array}\right).$

Therefore $B$ is NOT diagonalisable, since $B$ has only two linearly independent vectors.

Eigenvalue Eigenvector
$\lambda=2$ $\left( \begin{array}{r} 1 \\ 0 \\ 0 \\ \end{array} \right)$
$\lambda=1$ $\left( \begin{array}{r} 1 \\ -1 \\ 0 \\ \end{array} \right)$


16.4 Algebraic and geometric multiplicity

If we are only interested in finding out whether or not a matrix is diagonalisable, then we need to know the dimension of each eigenspace. There is one theorem (which we will not prove!) that states:

If $\lambda_i$ is an eigenvalue, then the dimension of the corresponding eigenspace cannot be greater than the number of times $(\lambda-\lambda_i)$ appears as a factor in the characteristic polynomial.




16.4 Algebraic and geometric multiplicity

We often use the following terminology:

  • The geometric multiplicity of the eigenvalue $\lambda_i$ is the dimension of the eigenspace corresponding to $\lambda_i$.
  • The algebraic multiplicity of the eigenvalue $\lambda_i$ is the number of times $(\lambda-\lambda_i)$ appears as a factor in the characteristic polynomial.

The main result is the following:

A square matrix is diagonalisable if and only if the geometric and algebraic multiplicities are equal for every eigenvalue.


16.5 Applications of diagonalisability

16.5.1 Systems of differential equations

For a system of coupled differential equations which can be written in matrix form as $$ \dot{\mathbf x}=A{\mathbf x} $$ (where ${\mathbf x}=(x_1,\dots,x_n)^T,$ $\dot{\mathbf x}=(\dot x_1,\dots,\dot x_n)^T$), if $A$ can be diagonalised, say $P^{-1}AP=D$ with $D$ diagonal, then make the substitution ${\mathbf x}=P{\mathbf y}.$ This yields $$ \dot{\mathbf y}=D{\mathbf y} $$ which is easily solved.


16.5 Applications of diagonalisability

16.5.1 Systems of differential equations

For a system of coupled differential equations $ \dot{\mathbf x}=A{\mathbf x}, $ if $A$ can be diagonalised, say $P^{-1}AP=D$ with $D$ diagonal, then make the substitution ${\mathbf x}=P{\mathbf y}.$ This yields $ \dot{\mathbf y}=D{\mathbf y} $ which is easily solved.

For example, consider the coupled system of ODEs: \[ \left\{ \begin{array}{rcrcr} \dot x_1 & = & x_1 & + & 2x_2\\ \dot x_2 & = & 2x_1 & + & x_2\\ \end{array} \right. \] Write $ \left( \begin{array}{r} \dot x_1 \\ \dot x_2 \\ \end{array} \right) = \left( \begin{array}{cc} 1 & 2\\ 2 & 1\\ \end{array} \right) \left( \begin{array}{r} x_1 \\ x_2 \\ \end{array} \right)\; $ or $\; \dot{\mathbf x} = A \mathbf x.$ $\quad (\star)$

Is $A$ diagonalizable? 🤔 The answer is "Yes". $\;\Ra \; P^{-1}A P = D.$



16.5 Applications of diagonalisability

16.5.1 Systems of differential equations

For a system of coupled differential equations $ \dot{\mathbf x}=A{\mathbf x}, $ if $A$ can be diagonalised, say $P^{-1}AP=D$ with $D$ diagonal, then make the substitution ${\mathbf x}=P{\mathbf y}.$ This yields $ \dot{\mathbf y}=D{\mathbf y} $ which is easily solved.

Write $ \left( \begin{array}{r} \dot x_1 \\ \dot x_2 \\ \end{array} \right) = \left( \begin{array}{cc} 1 & 2\\ 2 & 1\\ \end{array} \right) \left( \begin{array}{r} x_1 \\ x_2 \\ \end{array} \right)\; $ or $\; \dot{\mathbf x} = A \mathbf x.$ $\quad (\star)$

Is $A$ diagonalizable? 🤔 The answer is "Yes". $\;\Ra \; P^{-1}A P = D.$

From $\,(\star)\,$ we have

$P^{-1}\dot{\mathbf x}$ $= P^{-1}A \mathbf x$ $= P^{-1}A I {\mathbf x}$ $= P^{-1}A P P^{-1}{\mathbf x}$ $= D P^{-1}{\mathbf x}.$

Thus $ \quad \ds \frac{d}{dt}\left(P^{-1}\mathbf x \right)= D \left(P^{-1}\mathbf x\right)\qquad (\star \star) $



16.5 Applications of diagonalisability

16.5.1 Systems of differential equations

For a system of coupled differential equations $ \dot{\mathbf x}=A{\mathbf x}, $ if $A$ can be diagonalised, say $P^{-1}AP=D$ with $D$ diagonal, then make the substitution ${\mathbf x}=P{\mathbf y}.$ This yields $ \dot{\mathbf y}=D{\mathbf y} $ which is easily solved.

$ \dfrac{d}{dt}\left(P^{-1}\mathbf x \right) = D \left(P^{-1}\mathbf x\right)\qquad (\star \star) $

Introducing the expression $ \mathbf y(t) = \left( \begin{array}{r} y_1(t) \\ y_2(t) \\ \end{array} \right) = P^{-1}\mathbf x,\, $ implies

$\dot{\mathbf y} = D \mathbf y\;$ (from $\;(\star \star)$)

$\;\Ra\; \left( \begin{array}{r} \dot y_1(t) \\ \dot y_2(t) \\ \end{array} \right) = \left( \begin{array}{cc} \lambda_1 & 0 \\ 0 & \lambda_2\\ \end{array} \right) \left( \begin{array}{r} y_1(t) \\ y_2(t) \\ \end{array} \right) $

$\Ra\; \dot y_1 = \lambda_1y_1\;$ and $\; \dot y_2 = \lambda_2y_2.\qquad $



16.5 Applications of diagonalisability

16.5.1 Systems of differential equations

For a system of coupled differential equations $ \dot{\mathbf x}=A{\mathbf x}, $ if $A$ can be diagonalised, say $P^{-1}AP=D$ with $D$ diagonal, then make the substitution ${\mathbf x}=P{\mathbf y}.$ This yields $ \dot{\mathbf y}=D{\mathbf y} $ which is easily solved.

$\dot{\mathbf y} = D \mathbf y\;$ (from $\;(\star \star)$)

$\;\Ra\; \left( \begin{array}{r} \dot y_1(t) \\ \dot y_2(t) \\ \end{array} \right) = \left( \begin{array}{cc} \lambda_1 & 0 \\ 0 & \lambda_2\\ \end{array} \right) \left( \begin{array}{r} y_1(t) \\ y_2(t) \\ \end{array} \right) $

$\Ra\; \dot y_1 = \lambda_1y_1\;$ and $\; \dot y_2 = \lambda_2y_2.\qquad $

Finally, solve for $\,y_1$ and $y_2,\,$ then substitute into $\,\mathbf x = P\mathbf y\,$ to get $\,x_1, x_2.$



16.5.2 Matrix powers

If $A$ is diagonalisable, say $P^{-1}AP=D$ with $D$ diagonal, then $$ A^n=PD^nP^{-1}. $$ This gives an easy way to calculate $A^n$.

From $A= PDP^{-1}$ we have that

$A^n$ $=\underbrace{PDP^{-1} P D P^{-1} \cdots P D P^{-1}}_{n-\text{times}}$ $=P D^n P^{-1},$

where $\;D^n $ $ = \left( \begin{array}{ccc} \lambda_1 & \ldots & 0 \\ \vdots & \ddots & \vdots \\ 0 & \ldots & \lambda_n\\ \end{array} \right)^n$ $= \left( \begin{array}{ccc} \lambda_1^n & \ldots & 0 \\ \vdots & \ddots & \vdots \\ 0 & \ldots & \lambda_n^n\\ \end{array} \right).$



16.5.3 Linear systems

What can we say about the linear system (corresponding to a square matrix $A$)

$$ A{\mathbf x} = {\mathbf b}, $$

if we know how to diagonalise $A$?

Multiply $A{\mathbf x} = {\mathbf b}$ by $P^{-1}$ from left on both sides. Then we get

$P^{-1}A \mathbf x $ $= P^{-1}A I\mathbf x$ $= P^{-1}A P P^{-1}\mathbf x$ $= P^{-1} \mathbf b.\;$

Then $\;D\left(P^{-1} \mathbf x\right) = P^{-1} \mathbf b.$

Set $\mathbf y = P^{-1}\mathbf x,$ then $D\mathbf y = P^{-1} \mathbf b.$ 👈 This is a decoupled system! Finally, solve for $\mathbf y $ and then $\mathbf x = P \mathbf y.$



Credits