First, a theorem:

**Theorem O**. Let *A* be an *n* by *n* matrix. If the *n* eigenvalues of *A* are distinct, then the corresponding eigenvectors are linearly independent.

*Proof*. The proof of this theorem will be presented explicitly for *n* = 2; the proof in the general case can be constructed based on the same method. Therefore, let *A* be 2 by 2, and denote its eigenvalues by λ _{1} and λ _{2} and the corresponding eigenvectors by **v** _{1} and **v** _{2} (so that *A* **v** _{1} = λ _{1} **v** _{1} and *A* **v** _{2} = λ _{2} **v** _{2}). The goal is to prove that if λ _{1} ≠ λ _{2}, then **v** _{1} and **v** _{2} are linearly independent. Assume that

**v** _{1} and **v** _{2} that gives the zero vector; the goal is to show that the above equation implies that *c* _{1} and *c* _{2} must be zero. First, multiply both sides of (*) by the matrix *A*:

Next, use the fact that *A* **v** _{1} = λ **v** _{1} and *A* **v** _{2} = λ **v** _{2} to write

Now, multiply both sides of (*) by λ _{2} and subtract the resulting equation, c _{1};λ _{2} **v** _{1} + c _{2}λ _{2} **v** _{2} = **0** > from (**):

Since the eigenvalues are distinct, λ _{1} − λ _{2} ≠ 0, and since **v** _{1} ≠ **0** ( **v** _{1} is an eigenvector), this last equation implies that *c* _{1} = 0. Multiplying both sides of (*) by λ _{1} and subtracting the resulting equation from (**) leads to c _{2}(λ _{2} − λ _{1}) **v** _{2} = **0** and then, by the same reasoning, to the conclusion that *c* _{2} = 0 also.

Using the same notation as in the proof of Theorem O, assume that *A* is a 2 by 2 matrix with distinct eigenvalues and form the matrix

*A*. Now consider the product *AV*; since *A* **v** _{1} = λ _{1} **v** _{1} and *A* **v** _{2} = λ _{2} **v** _{2},

This last matrix can be expressed as the following product:

If *A* denotes the diagonal matrix whose entries are the eigenvalues of *A*,

*AV* = *VA*. If **v** _{1} and **v** _{2} are linearly independent, then the matrix *V* is invertible. Form the matrix *V* ^{−1} and left multiply both sides of the equation *AV* = *VA* by *V* ^{−1}:

(Although this calculation has been shown for *n* = 2, it clearly can be applied to an *n* by *n* matrix of any size.) This process of forming the product *V* ^{−1} *AV*, resulting in the diagonal matrix *A* of its eigenvalues, is known as the **diagonalization** of the matrix *A*, and the matrix of eigenvectors, *V*, is said to **diagonalize** *A. The key to diagonalizing an n by n matrix A is the ability to form the n by n eigenvector matrix V and its inverse; this requires a full set of n linearly independent eigenvectors*. A sufficient (but not necessary) condition that will guarantee that this requirement is fulfilled is provided by Theorem O: if the *n* by *n* matrix *A* has *n distinct* eigenvalues.

One useful application of diagonalization is to provide a simple way to express integer powers of the matrix *A*. If *A* can be diagonalized, then *V* ^{−1} *AV* = *A*, which implies

When expressed in this form, it is easy to form integer powers of *A*. For example, if *k* is a positive integer, then

The power *A ^{k} *is trivial to compute: If λ

_{1}, λ

_{2}, …, λ

_{n }are the entries of the diagonal matrix

*A*, then

*A*is diagonal with entries λ

^{k}_{1}

^{k },λ

_{2}

^{k },…, λ

_{n}

^{k }. Therefore,

**Example 1**: Compute *A* ^{10} for the matrix

This is the matrix of Example 1. Its eigenvalues are λ _{1} = −1 and λ _{2} = −2, with corresponding eigenvectors **v** _{1} = (1, 1) ^{T} and **v** _{2} = (2, 3) ^{T}. Since these eigenvectors are linearly independent (which was to be expected, since the eigenvalues are distinct), the eigenvector matrix *V* has an inverse,

Thus, *A* can be diagonalized, and the diagonal matrix *A* = *V* ^{−1} *AV* is

Therefore,

Although an *n* by *n* matrix with *n* distinct eigenvalues is guaranteed to be diagonalizable, an *n* by *n* matrix that does not have *n* distinct eigenvalues may still be diagonalizable. If the eigenspace corresponding to each *k*‐fold root λ of the characteristic equation is *k* dimensional, then the matrix will be diagonalizable. In other words, diagonalization is guaranteed if the *geometric* multiplicity of each eigenvalue (that is, the dimension of its corresponding eigenspace) matches its *algebraic* multiplicity (that is, its multiplicity as a root of the characteristic equation). Here's an illustration of this result. The 3 by 3 matrix

_{1} = −1 and λ _{2} = 3. The algebraic multiplicity of the eigenvalue λ _{1} = −1 is one, and its corresponding eigenspace, *E* _{−1}( *B*), is one dimensional. Furthermore, the algebraic multiplicity of the eigenvalue λ _{2} = **3** is two, and its corresponding eigenspace, *E* _{3}( *B*), is two dimensional. Therefore, the geometric multiplicities of the eigenvalues of *B* match their algebraic multiplicities. The conclusion, then, is that although the 3 by 3 matrix *B* does not have 3 distinct eigenvalues, it is nevertheless diagonalizable.

Here's the verification: Since {(1, 0, −1) ^{T}} is a basis for the 1‐dimensional eigenspace corresponding to the eigenvalue λ _{1} = −1, and {(0, 1, 0) ^{T}, (1, 0, 1) ^{T}} is a basis for the 2‐dimensional eigenspace corresponding to the eigenvalue λ _{2} = 3, the matrix of eigenvectors reads

Since the key to the diagonalization of the original matrix *B* is the invertibility of this matrix, *V*, evaluate det *V* and check that it is nonzero. Because det *V* = 2, the matrix *V is* invertible,

*B* is indeed diagonalizable:

**Example 2**: Diagonalize the matrix

First, find the eigenvalues; since

*A* is diagonalizable. Verify that an eigenvector corresponding to λ = 1 is **v** _{1} = (1, 1) ^{T}, and an eigenvector corresponding to λ = 5 is **v** _{2} = (1, −3) ^{T}. Therefore, the diagonalizing matrix is

Another application of diagonalization is in the construction of simple representative matrices for linear operators. Let *A* be the matrix defined above and consider the linear operator on **R** ^{2} given by *T*( **x**) = *A* **x**. In terms of the nonstandard basis *B* = { **v** _{1} = (1, 1) ^{T}, **v** _{2} = (1, −3) ^{T}} for **R** ^{2}, the matrix of *T* relative to *B* is *A*.