Comprehensive Guide on Algebraic and Geometric Multiplicity
Start your free 7-days trial now!
Algebraic multiplicity
The algebraic multiplicity of an eigenvalue $\lambda$ is the number of times $\lambda$ appears as a root of the characteristic polynomiallink. We sometimes denote the algebraic multiplicity of $\lambda$ as $\mathrm{AM}(\lambda)$.
Finding the algebraic multiplicity (1)
Consider the following matrix:
Find the algebraic multiplicity of every eigenvalue of $\boldsymbol{A}$.
Solution. The characteristic polynomial of $\boldsymbol{A}$ is:
The eigenvalue $\lambda=5$ repeats twice so the algebraic multiplicity of $\lambda$ is $2$.
Finding the algebraic multiplicity (2)
Consider the following matrix:
Find the algebraic multiplicity of every eigenvalue of $\boldsymbol{A}$.
Solution. The characteristic polynomial of $\boldsymbol{A}$ is:
The eigenvalues of $\boldsymbol{A}$ are $\lambda_1=3$ and $\lambda_2=4$, and they both occur once. Therefore, the algebraic multiplicity of $\lambda_1$ is $1$ and that of $\lambda_2$ is also $1$.
Geometric multiplicity
Let $\boldsymbol{A}$ be a square matrix. The geometric multiplicity of an eigenvalue $\lambda$ of $\boldsymbol{A}$ is the dimension of the eigenspace of $\lambda$. In other words, the geometric multiplicity is the nullitylink of $\boldsymbol{A}-\lambda\boldsymbol{I}$. We sometimes denote the geometric multiplicity of $\lambda$ as $\mathrm{GM}(\lambda)$.
Finding the geometric multiplicity
Consider the following matrix:
Find the geometric multiplicity of every eigenvalue of $\boldsymbol{A}$.
Solution. The characteristic polynomial of $\boldsymbol{A}$ is:
Therefore, the eigenvalues of $\boldsymbol{A}$ are $\lambda_1=3$ and $\lambda_2=4$. Let's now find the geometric multiplicity of $\lambda_1$ and $\lambda_2$.
The null spacelink of $\boldsymbol{A}-\lambda_1\boldsymbol{I}$ is the set of vectors $\boldsymbol{x}$ such that $(\boldsymbol{A}-\lambda_1\boldsymbol{I})\boldsymbol{x}=\boldsymbol{0}$. In matrix form, this is:
We have that $x_1$ is a free variablelink and $x_2$ is a basic variablelink. Let $x_1=t$ where $t$ is some scalar. The solution of the homogeneous system is:
Therefore, the basislink for the null space of $\boldsymbol{A}-\lambda_1\boldsymbol{I}$ is:
The nullity is the dimensionlink of the null space of $\boldsymbol{A}-\lambda_1\boldsymbol{I}$. In other words, the nullity is equal to the number of basis vectors for the null space of $\boldsymbol{A}-\lambda_1\boldsymbol{I}$. Since the basis of the null space consists of a single vector, we have that the nullity of $\boldsymbol{A}-\lambda_1\boldsymbol{I}$ is $1$. Therefore, the geometric multiplicity of $\lambda_1$ is $1$.
Similarly, let's now find the geometric multiplicity of $\lambda_2=4$. The matrix form of $\boldsymbol{A}-\lambda_2\boldsymbol{I}$ is:
Here, $x_1$ is a basic variable while $x_2$ is a free variable. Therefore, we let $x_2=t$ where $t\in\mathbb{R}$. The first row gives us:
Therefore, the solution of the homogenous system is:
Therefore, the basis for the null space of $\boldsymbol{A}-\lambda_2\boldsymbol{I}$ is:
Because the nullity of $\boldsymbol{A}-\lambda_2\boldsymbol{I}$ is $1$, the geometric multiplicity of $\lambda_2$ is $1$.
Finding algebraic and geometric multiplicity
Consider the following matrix:
Find the algebraic and geometric multiplicity of every eigenvalue of $\boldsymbol{A}$.
Solution. The characteristic polynomial of $\boldsymbol{A}$ is:
The eigenvalue of $\boldsymbol{A}$ is $\lambda=5$. Since $\lambda=5$ repeats twice, we have that the algebraic multiplicity of $\boldsymbol{A}$ is $2$. Let's now find the geometric multiplicity of $\lambda$. The matrix form of $\boldsymbol{A}-\lambda\boldsymbol{I}$ is:
We have that $x_1$ and $x_2$ are both free variables. Let $x_1=r$ and $x_2=t$ where $r$ and $t$ are some scalars. The solution can be expressed as:
The eigenspace $\mathcal{E}$ of $\lambda=5$ is:
The basis $\mathcal{B}$ of $\mathcal{E}$ is:
Since $\mathcal{B}$ has $2$ basis vectors, the geometric multiplicity of $\lambda$ is $2$. Let's summarize the results below:
Eigenvalue | Algebraic multiplicity | Geometric multiplicity |
---|---|---|
$\lambda=5$ | $2$ | $2$ |
We will later prove that the algebraic multiplicity is always at least as large as the geometric multiplicity.
Lower bound of the geometric multiplicity of an eigenvalue
Let $\boldsymbol{A}$ be an $n\times{n}$ matrix and $\lambda$ be an eigenvalue of $\boldsymbol{A}$. The geometric multiplicity of $\lambda$ is equal to or greater than $1$.
Proof. Recall the characteristic equationlink:
By theoremlink, this means that the matrix $\boldsymbol{A}-\lambda\boldsymbol{I}$ is not invertiblelink. By theoremlink, we have that the homogeneous system $(\boldsymbol{A}-\lambda\boldsymbol{I})\boldsymbol{x}=\boldsymbol{0}$ has at least one non-trivial solution $\boldsymbol{x}$. This means that the null space of $\boldsymbol{A}-\lambda\boldsymbol{I}$ contains at least one eigenvector for each eigenvalue $\lambda$.
By theoremlink, we know that scalar multiples of the eigenvector are also eigenvectors. Therefore, the eigenspace contains a set of all linear combinations of at least one eigenvector. We conclude that the eigenspace of every eigenvalue must have a dimension of at least $1$.
This completes the proof.
Upper bound of the algebraic multiplicity of an eigenvalue
Let $\boldsymbol{A}$ be an $n\times{n}$ matrix and $\lambda$ be an eigenvalue of $\boldsymbol{A}$. The algebraic multiplicity of $\lambda$ is a most $n$.
Proof. Let $\boldsymbol{A}$ be an $n\times{n}$ matrix. By theoremlink, we know that the degree of the characteristic polynomial of $\boldsymbol{A}$ is $n$. This means that an eigenvalue can occur as a root at most $n$ times. In other words, the algebraic multiplicity of an eigenvalue is at most $n$. This completes the proof.
Sum of algebraic multiplicities is equal to the size of the associated matrix
Let $\boldsymbol{A}$ be an $n\times{n}$ matrix. If $\lambda_1$, $\lambda_2$, $\cdots$, $\lambda_k$ are the eigenvalues of $\boldsymbol{A}$, then the sum of the corresponding algebraic multiplicities is equal to $n$, that is:
Proof. By theoremlink, if $\boldsymbol{A}$ is an $n\times{n}$ matrix, then the characteristic polynomial of $\boldsymbol{A}$ will have degree $n$. A polynomial with degree $n$ has exactly $n$ roots (including duplicates). Since the roots of the characteristic polynomial correspond to the eigenvalues of $\boldsymbol{A}$, we conclude that the sum of the algebraic multiplicities of the eigenvalues must be equal to $n$. This completes the proof.
Sum of geometric multiplicities of a square matrix with linearly independent columns is greater than its size
Let $\boldsymbol{A}$ be an $n\times{n}$ matrix with $n$ linearly independent eigenvectors. If $\lambda_1$, $\lambda_2$, $\cdots$, $\lambda_k$ are the eigenvalues of $\boldsymbol{A}$, then the sum of the corresponding geometric multiplicities is equal to or greater than $n$, that is:
Note that this is only an intermediate result and we will later prove that the $\ge$ can be replaced with an equality.
Proof. Suppose $\boldsymbol{A}$ has the following $n$ linearly independent eigenvectors:
However, some of these eigenvectors may be associated with the same eigenvalue. For instance, suppose $\boldsymbol{v}_1$ and $\boldsymbol{v}_2$ are associated with the same eigenvalue $\lambda=4$. Since $\boldsymbol{v}_1$ and $\boldsymbol{v}_2$ are linearly independent, the number of basis vectors for the eigenspacelink of $\lambda=4$ must be at least $2$. The reason we say "at least" here is that the number of basis vectors for $\lambda=4$ may potentially be greater than $2$. For instance, consider the following case:
The eigenspace for the eigenvalue may be spanned by the following basis vectors:
Here, because there are $3$ basis vectors, the corresponding geometric multiplicity is $3$. Note that we will later show that such a case is impossible.
By repeatedly applying this logic to the rest of the eigenvectors in \eqref{eq:lwwwN9VKQ6EzRnoSxiM}, we conclude that the sum of geometric multiplicities is greater than or equal to $n$. This completes the proof.
Algebraic multiplicity is greater than or equal to the geometric multiplicity
Let $\boldsymbol{A}$ be a square matrix and $\lambda$ be an eigenvalue of $\boldsymbol{A}$. The algebraic multiplicity of $\lambda$ is greater than or equal to the geometric multiplicity of $\lambda$, that is:
Proof. Suppose we have an $n\times{n}$ matrix $\boldsymbol{A}$ and an invertible $n\times{n}$ matrix $\boldsymbol{P}$. By theoremlink, we know that similarlink matrices have the same characteristic polynomial:
Our goal is to derive an expression for the right-hand side of \eqref{eq:IHhrbeAVsRoFdBeXaWV} that includes a term representing the geometric multiplicity of an eigenvalue $\lambda_i$.
To simplify notations, let $k$ denote the geometric multiplicity of eigenvalue $\lambda_i$ and let $\{\boldsymbol{x}_1,\boldsymbol{x}_2, \cdots,\boldsymbol{x}_k\}$ be the basis for the corresponding eigenspace $\mathcal{E}_i$. We define an $n\times{n}$ invertible matrix $\boldsymbol{P}$ whose first $k$ columns are these basis vectors:
Note that there exists such an invertible matrix $\boldsymbol{P}$ because the first $k$ columns are linearly independent by definition of basis vectors. We can ensure $\boldsymbol{P}$ is invertible by coming up with a set of vectors $x_{k+1}$, $x_{k+2}$, $\cdots$, $x_{n}$ that is also linearly independent. By theoremlink, an $n\times{n}$ matrix with $n$ linearly independent columns is invertible. This guarantees the existence of an invertible matrix $\boldsymbol{P}$.
We now express $\boldsymbol{P}$ as a block matrixlink:
Where:
$\boldsymbol{B}$ is an $n\times{k}$ sub-matrix whose columns are filled with the eigenbasis vectors.
$\boldsymbol{C}$ is an $n\times(n-k)$ sub-matrix whose columns are $\boldsymbol{x}_{k+1}$, $\boldsymbol{x}_{k+2}$, $\cdots$, $\boldsymbol{x}_{n}$.
Similarly, let's represent the inverse matrix $\boldsymbol{P}^{-1}$ using block format:
Where:
$\boldsymbol{D}$ is a $k\times{n}$ sub-matrix.
$\boldsymbol{E}$ is a $(n-k)\times{n}$ sub-matrix.
By theoremlink, the matrix product $\boldsymbol{P}^{-1}\boldsymbol{P}$ can be expressed as:
Where:
$\boldsymbol{DB}$ is an $k\times{k}$ sub-matrix.
$\boldsymbol{DC}$ is an $k\times(n-k)$ sub-matrix
$\boldsymbol{EB}$ is an $(n-k)\times{k}$ sub-matrix.
$\boldsymbol{EC}$ is an $(n-k)\times(n-k)$ sub-matrix.
By definitionlink, a matrix multiplied by its inverse matrix is equal to the identity matrix, that is, $\boldsymbol{P}^{-1}\boldsymbol{P}=\boldsymbol{I}_n$. Therefore:
$\boldsymbol{DB}=\boldsymbol{I}_k$.
$\boldsymbol{EC}=\boldsymbol{I}_{n-k}$.
$\boldsymbol{DC}$ is an $k\times(n-k)$ sub-matrix whose entries are all zeroes.
$\boldsymbol{EB}$ is an $(n-k)\times{k}$ sub-matrix whose entries are all zeroes.
Now, consider $\boldsymbol{P}^{-1}\boldsymbol{AP}$. By theoremlink and theoremlink, $\boldsymbol{P}^{-1}\boldsymbol{AP}$ can be expressed as:
Because $\lambda_i$ is an eigenvalue of $\boldsymbol{A}$ and $\boldsymbol{x}_i$ is a corresponding eigenvector, the following equation holds true by definition:
We now multiply both sides by $\boldsymbol{D}$. Note that this is possible because:
$\boldsymbol{D}$ has shape $k\times{n}$.
$\boldsymbol{A}$ has shape $n\times{n}$.
$\boldsymbol{x}_i$ has shape $n\times{1}$.
Some algebraic manipulation yields:
This means that:
Similarly, we have that $\boldsymbol{EA}=\lambda_i\boldsymbol{E}$. Now we go back to \eqref{eq:jBeRUvHtoFRDerjz6Sl} and perform the following substitutions:
Subtracting both sides by $\lambda\boldsymbol{I}$ where $\lambda$ is any eigenvalue of $\boldsymbol{A}$ yields:
We now take the determinant of both sides. By theoremlink, we have that:
Next, by theoremlink, we get:
We substitute this into \eqref{eq:IHhrbeAVsRoFdBeXaWV} to finally get:
Because $(\lambda_i-\lambda)$ appears $k$ times, the algebraic multiplicity of $\lambda_i$ is at least $k$. This completes the proof.
Relationship between diagonalization, algebraic and geometric multiplicity
Let $\boldsymbol{A}$ be an $n\times{n}$ matrix. The following statements are equivalent:
$\boldsymbol{A}$ is a diagonalizable matrixlink.
the sum of the geometric multiplicities of the eigenvalues of $\boldsymbol{A}$ is equal to $n$.
the geometric multiplicity and the algebraic multiplicity of every eigenvalue are equal.
Proof. We will prove $(1)\implies(2)\implies(3)\implies(1)$. We start by assuming $(1)$ that the $n\times{n}$ matrix $\boldsymbol{A}$ is diagonalizable. By theoremlink, since $\boldsymbol{A}$ is diagonalizable, $\boldsymbol{A}$ has $n$ linearly independentlink eigenvectors $\boldsymbol{v}_1$, $\boldsymbol{v}_2$, $\cdots$, $\boldsymbol{v}_n$.
Suppose $\boldsymbol{A}$ has $t$ distinct eigenvalues and let the corresponding geometric multiplicities be $k_1$, $k_2$, $\cdots$, $k_t$ and the algebraic multiplicities be $m_1$, $m_2$, $\cdots$, $m_t$.
By theoremlink, the sum of the geometric multiplicities must be greater than or equal to $n$, that is:
By theoremlink, the sum of the algebraic multiplicities is equal to $n$, that is:
Substituting \eqref{eq:x90dgy3rmNu24WhyvLP} into \eqref{eq:TQhsAouxJRmwjztMdpm} gives:
Now, from theoremlink, we have that the algebraic multiplicity of an eigenvalue is greater than or equal to the geometric multiplicity of the same eigenvalue, that is:
Summing all of these inequalities gives:
Combining \eqref{eq:oo3zHU2Uji1nxu4Z8Dl} and \eqref{eq:SoPWDF64YY5gRAlrCBa} gives:
This means that:
Combining with \eqref{eq:x90dgy3rmNu24WhyvLP} gives:
This proves $(1)\implies(2)$.
We now prove $(2)\implies(3)$. Assume the sum of the geometric multiplicities is $n$. By theoremlink, the sum of the algebraic multiplicities is $n$, which means:
Again, by theoremlink, we have that:
Combining \eqref{eq:vCICCrzPdLZnYtErI9X} and \eqref{eq:DRURNqiWpvCY2V0roHb} implies:
This means that for every eigenvalue $\lambda_i$ for $i=1,2,\cdots,t$, the corresponding geometric multiplicity and algebraic multiplicity are equal. This proves $(2)\implies(3)$.
We now prove $(3)\implies(1)$. Suppose $\lambda_1$, $\lambda_2$, $\cdots$, $\lambda_t$ are distinct eigenvalues of an $n\times{n}$ matrix $\boldsymbol{A}$ where $t\le{n}$. Denote the corresponding eigenspace as $\mathcal{E}_1$, $\mathcal{E}_2$, $\cdots$, $\mathcal{E}_t$ and the corresponding basis as $\mathcal{B}_1$, $\mathcal{B}_2$, $\cdots$, $\mathcal{B}_t$ respectively.
We assume $(3)$ that the geometric multiplicity and algebraic multiplicity for every eigenvalue are equal. By theoremlink, because the sum of the algebraic multiplicities is equal to $n$, the sum of geometric multiplicities is also equal to $n$.
By theoremlink, because the basis $\mathcal{B}_1$, $\mathcal{B}_2$, $\cdots$, $\mathcal{B}_t$ correspond to distinct eigenvalues $\lambda_1$, $\lambda_2$, $\cdots$, $\lambda_t$, the union of the basis $\mathcal{B}_1\cup \mathcal{B}_2\cup \cdots\cup \mathcal{B}_t$ is linearly independent. This union can be expressed as:
Where $\boldsymbol{w}_i$ for $i=1,2,\cdots,n$ is a basis vector. The reason why the union contains $n$ vectors is that the sum of the geometric multiplicities is equal to $n$, which means that the total number of basis vectors when combined into a single set must be $n$.
We've now established that $\boldsymbol{A}$ has $n$ linearly independent eigenvectors. By theoremlink, we know that an $n\times{n}$ matrix with $n$ linearly independent eigenvectors is diagonalizable. Therefore, $\boldsymbol{A}$ is diagonalizable.
This completes the proof.
Showing that a matrix diagonalizable using algebraic and geometric multiplicity
Show that the following matrix is diagonalizable:
Solution. We have shown in examplelink that:
Eigenvalue | Algebraic multiplicity | Geometric multiplicity |
---|---|---|
$\lambda=5$ | $2$ | $2$ |
By theorem, we can refer to either of the following fact to prove that $\boldsymbol{A}$ is diagonalizable:
the geometric multiplicity of the eigenvalue is n.
the algebraic multiplicity and the geometric multiplicity are equal.