代码拉取完成，页面将自动刷新
<?xml version="1.0" encoding="UTF-8"?>
<!--********************************************************************
Copyright 2017 Georgia Institute of Technology
Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free Documentation License, Version 1.3 or
any later version published by the Free Software Foundation. A copy of
the license is included in gfdl.xml.
*********************************************************************-->
<restrict-version versions="1554 default">
<section xml:id="similarity">
<title>Similarity</title>
<objectives>
<ol>
<li>Learn to interpret similar matrices geoemetrically.</li>
<li>Understand the relationship between the eigenvalues, eigenvectors, and characteristic polynomials of similar matrices.</li>
<li><em>Recipe:</em> compute <m>Ax</m> in terms of <m>B,C</m> for <m>A=CBC\inv</m>.</li>
<li><em>Picture:</em> the geometry of similar matrices.</li>
<li><em>Vocabulary word:</em> <term>similarity</term>.</li>
</ol>
</objectives>
<introduction>
<p>
Some matrices are easy to understand. For instance, a diagonal matrix
<me>
D = \mat{2 0 ; 0 1/2}
</me>
just scales the coordinates of a vector: <m>D{x\choose y} = {2x\choose y/2}</m>. The purpose of most of the rest of this chapter is to understand complicated-looking matrices by analyzing to what extent they <q>behave like</q> simple matrices. For instance, the matrix
<me>
A = \frac 1{10}\mat{11 6; 9 14}
</me>
has eigenvalues <m>2</m> and <m>1/2</m>, with corresponding eigenvectors <m>v_1 = {2/3\choose 1}</m> and <m>v_2 = {-1\choose 1}</m>. Notice that
<me>
\spalignsysdelims..\syseq{
D(xe_1+ye_2) = xDe_1+yDe_2 = 2xe_1-\frac12ye_2;
A(xv_1+yv_2) = xAv_1+yAv_2\hfil, = 2xv_1-\frac12yv_2\rlap.}
</me>
Using <m>v_1,v_2</m> instead of the usual coordinates makes <m>A</m> <q>behave</q> like a diagonal matrix.
</p>
<figure>
<caption>The matrices <m>A</m> and <m>D</m> behave similarly. Click <q>multiply</q> to multiply the colored points by <m>D</m> on the left and <m>A</m> on the right. (We will see in <xref ref="diagonalization"/> why the points follow hyperbolic paths.)</caption>
<mathbox source="demos/dynamics2.html?mat=2,0:0,1/2&v1=2/3,1&v2=-1,1&y=1,5" height="600px"/>
</figure>
<p>
The other case of particular importance will be matrices that <q>behave</q> like a rotation matrix: indeed, this will be crucial for understanding <xref ref="complex-eigenvalues"/> geometrically. See this <xref ref="similarity-to-rotation"/>.
</p>
<p>
In this section, we study in detail the situation when two matrices behave similarly with respect to different coordinate systems. In <xref ref="diagonalization"/> and <xref ref="complex-eigenvalues"/>, we will show how to use eigenvalues and eigenvectors to find a simpler matrix that behaves like a given matrix.
</p>
</introduction>
<subsection>
<title>Similar Matrices</title>
<p>
We begin with the algebraic definition of similarity.
</p>
<definition>
<idx><h>Similarity</h><h>definition of</h></idx>
<idx><h>Matrix</h><h>similar</h><see>Similarity</see></idx>
<statement>
<p>
Two <m>n\times n</m> matrices <m>A</m> and <m>B</m> are <term>similar</term> if there exists an invertible <m>n\times n</m> matrix <m>C</m> such that <m>A = CBC\inv</m>.
</p>
</statement>
</definition>
<example xml:id="similarity-eg1">
<p>
The matrices
<me>
\mat{-12 15; -10 13} \sptxt{and} \mat{3 0; 0 -2}
</me>
are similar because
<me>
\mat{-12 15; -10 13} = \mat{-2 3; 1 -1} \mat{3 0; 0 -2} \mat{-2 3; 1 -1}\inv,
</me>
as the reader can verify.
</p>
</example>
<example>
<p>
The matrices
<me>
\mat{3 0 ; 0 -2} \sptxt{and} \mat{1 0; 0 1}
</me>
are not similar. Indeed, the second matrix is the identity matrix <m>I_2</m>, so if <m>C</m> is any invertible <m>2\times 2</m> matrix, then
<me>
CI_2C\inv = CC\inv = I_2 \neq \mat{3 0; 0 -2}.
</me>
</p>
</example>
<p>
<idx><h>Similarity</h><h>identity matrix</h></idx>
<idx><h>Identity matrix</h><h>similarity</h></idx>
As in the above example, one can show that <m>I_n</m> is the only matrix that is similar to <m>I_n</m>, and likewise for any scalar multiple of <m>I_n</m>.
</p>
<bluebox>
<p>
Similarity is unrelated to row equivalence. Any invertible matrix is row equivalent to <m>I_n</m>, but <m>I_n</m> is the only matrix similar to <m>I_n</m>. For instance,
<me>\mat{2 1; 0 2} \sptxt{and} \mat{1 0; 0 1}</me>
are row equivalent but not similar.
</p>
</bluebox>
<p>
As suggested by its name, similarity is what is called an <em>equivalence relation</em>. This means that it satisfies the following properties.
</p>
<proposition xml:id="similarity-eq-reln">
<idx><h>Similarity</h><h>equivalence relation</h></idx>
<statement>
<p>
Let <m>A,B</m>, and <m>C</m> be <m>n\times n</m> matrices.
<ol>
<li>
<alert>Reflexivity:</alert> <m>A</m> is similar to itself.
</li>
<li>
<alert>Symmetry:</alert> if <m>A</m> is similar to <m>B</m>, then <m>B</m> is similar to <m>A</m>.
</li>
<li>
<alert>Transitivity:</alert> if <m>A</m> is similar to <m>B</m> and <m>B</m> is similar to <m>C</m>, then <m>A</m> is similar to <m>C</m>.
</li>
</ol>
</p>
</statement>
<proof>
<p>
<ol>
<li>
Taking <m>C = I_n = I_n\inv</m>, we have <m>A = I_n A I_n\inv</m>.
</li>
<li>
Suppose that <m>A = CBC\inv.</m> Multiplying both sides on the left by <m>C\inv</m> and on the right by <m>C</m> gives
<me>
C\inv A C = C\inv(CBC\inv)C = B.
</me>
Since <m>(C\inv)\inv = C</m>, we have <m>B = C\inv A (C\inv)\inv</m>, so that <m>B</m> is similar to <m>A</m>.
</li>
<li>
Suppose that <m>A = DBD\inv</m> and <m>B = ECE\inv</m>. Subsituting for <m>B</m> and remembering that <m>(DE)\inv = E\inv D\inv</m>, we have
<me>
A = D(ECE\inv)D\inv = (DE)C(DE)\inv,
</me>
which shows that <m>A</m> is similar to <m>C</m>.
</li>
</ol>
</p>
</proof>
</proposition>
<example>
<p>
The matrices
<me>\mat{-12 15; -10 13} \sptxt{and} \mat{3 0; 0 -2}</me>
are similar, as we saw in this <xref ref="similarity-eg1"/>. Likewise, the matrices
<me>\mat{-12 15; -10 13} \sptxt{and} \mat{-12 5; -30 13}</me>
are similar because
<me>
\mat{-12 5; -30 13} = \mat{2 -1; 2 1}\mat{-12 15; -10 13}\mat{2 -1; 2 1}\inv.
</me>
It follows that
<me>\mat{-12 5; -30 13} \sptxt{and} \mat{3 0; 0 -2}</me>
are similar to each other.
</p>
</example>
<p>
We conclude with an observation about similarity and powers of matrices.
</p>
<fact xml:id="similarity-powers">
<idx><h>Similarity</h><h>and powers</h></idx>
<idx><h>Matrix multiplication</h><h>powers</h><h>and similarity</h></idx>
<statement>
<p>
Let <m>A = CBC\inv</m>. Then for any <m>n\geq 1</m>, we have
<me>A^n = CB^n C\inv.</me>
</p>
</statement>
<proof visible="true">
<p>
First note that
<me>
A^2 = AA = (CBC\inv)(CBC\inv) = CB(C\inv C)BC\inv = CBI_nBC\inv = CB^2C\inv.
</me>
Next we have
<me>
A^3 = A^2A = (CB^2C\inv)(CBC\inv) = CB^2(C\inv C)BC\inv = CB^3C\inv.
</me>
The pattern is clear.
</p>
</proof>
</fact>
<example>
<statement>
<p>
Compute <m>A^{100}</m>, where
<me>A = \mat{5 13; -2 -5} = \mat{-2 3; 1 -1}\mat{0 -1; 1 0}\mat{-2 3; 1 -1}\inv.</me>
</p>
</statement>
<solution>
<p>
By the <xref ref="similarity-powers"/>, we have
<me>
A^{100} = \mat{-2 3; 1 -1}\mat{0 -1; 1 0}^{100}\mat{-2 3; 1 -1}\inv.
</me>
The matrix <m>\smallmat 0{-1}10</m> is a counterclockwise rotation by <m>90^\circ</m>. If we rotate by <m>90^\circ</m> four times, then we end up where we started. Hence rotating by <m>90^\circ</m> one hundred times is the identity transformation, so
<me>
A^{100} = \mat{-2 3; 1 -1}\mat{1 0; 0 1}\mat{-2 3; 1 -1}\inv
= \mat{1 0; 0 1}.
</me>
</p>
</solution>
</example>
</subsection>
<subsection>
<title>Geometry of Similar Matrices</title>
<idx><h>Similarity</h><h>geometry of</h></idx>
<p>
Similarity is a very interesting construction when viewed geometrically. We will see that, roughly, <em>similar matrices do the same thing in different coordinate systems.</em> The reader might want to review <m>\cB</m>-coordinates and nonstandard coordinate grids in <xref ref="bases-as-coord-systems"/> before reading this subsection.
</p>
<p>
By conditions 4 and 5 of the <xref ref="imt-2"/>, an <m>n\times n</m> matrix <m>C</m> is invertible if and only if its columns <m>v_1,v_2,\ldots,v_n</m> form a basis for <m>\R^n</m>. This means we can speak of the <m>\cB</m>-coordinates of a vector in <m>\R^n</m>, where <m>\cB</m> is the basis of columns of <m>C</m>. Recall that
<me>
[x]_\cB = \vec{c_1 c_2 \vdots, c_n} \sptxt{means}
x = c_1v_1 + c_2v_2 + \cdots + c_nv_n
= \mat{| | ,, |; v_1 v_2 \cdots, v_n; | | ,, |}\vec{c_1 c_2 \vdots, c_n}.
</me>
Since <m>C</m> is the matrix with columns <m>v_1,v_2,\ldots,v_n</m>, this says that <m>x = C[x]_\cB</m>. Multiplying both sides by <m>C\inv</m> gives <m>[x]_\cB = C\inv x.</m> To summarize:
</p>
<bluebox>
<idx><h><m>\cB</m>-coordinates</h><h>change of basis matrix</h></idx>
<p>
Let <m>C</m> be an invertible <m>n\times n</m> matrix with columns <m>v_1,v_2,\ldots,v_n</m>, and let <m>\cB = \{v_1,v_2,\ldots,v_n\}</m>, a basis for <m>\R^n</m>. Then for any <m>x</m> in <m>\R^n</m>, we have
<me>
C[x]_\cB = x \sptxt{and} C\inv x = [x]_\cB.
</me>
This says that <em><m>C</m> changes from the <m>\cB</m>-coordinates to the usual coordinates</em>, and <em><m>C\inv</m> changes from the usual coordinates to the <m>\cB</m>-coordinates.</em>
</p>
</bluebox>
<p>
Suppose that <m>A = CBC\inv.</m> The above observation gives us another way of computing <m>Ax</m> for a vector <m>x</m> in <m>\R^n</m>. Recall that
<m>CBC\inv x = C(B(C\inv x)),</m>
so that multiplying <m>CBC\inv</m> by <m>x</m> means <em>first</em> multiplying by <m>C\inv</m>, <em>then</em> by <m>B</m>, <em>then</em> by <m>C</m>. See this <xref ref="matrix-mult-order-ops"/>.
</p>
<bluebox>
<title>Recipe: Computing <m>Ax</m> in terms of <m>B</m></title>
<idx><h>Similarity</h><h>action on a vector</h></idx>
<p>
Suppose that <m>A = CBC\inv,</m> where <m>C</m> is an invertible matrix with columns <m>v_1,v_2,\ldots,v_n.</m> Let <m>\cB = \{v_1,v_2,\ldots,v_n\}</m>, a basis for <m>\R^n</m>. Let <m>x</m> be a vector in <m>\R^n</m>. To compute <m>Ax</m>, one does the following:
<ol>
<li>
Multiply <m>x</m> by <m>C\inv</m>, which changes to the <m>\cB</m>-coordinates: <m>[x]_\cB = C\inv x</m>.
</li>
<li>
Multiply this by <m>B</m>: <m>B[x]_\cB = BC\inv x</m>.
</li>
<li>
Interpreting this vector as a <m>\cB</m>-coordinate vector, we multiply it by <m>C</m> to change back to the usual coordinates:
<m>Ax = CBC\inv x = CB[x]_\cB.</m>
</li>
</ol>
<latex-code>
\begin{tikzpicture}[scale=.8, thin border nodes]
\draw[help lines] (-3,-3) grid (3,3);
\node at (0, 3.5) {$\cB$-coordinates};
\draw[seq-green, vector] (0, 0) to["{$[x]_\cB$}" above right] (-2, 1);
\draw[seq-red, vector] (0, 0) to["{$B[x]_\cB$}"] (-2, -1);
\draw[->, shorten=2mm, thick] (8, 1) to[bend right]
node[above=1mm] {multiply by $C\inv$} (3, 1);
\draw[->, shorten=2mm, thick] (3, -1) to[bend right]
node[below=1mm] {multiply by $C$} (8, -1);
\point at (0, 0);
\begin{scope}[xshift=11cm]
\draw[help lines] (-3,-3) grid (3,3);
\node at (0, 3.5) {usual coordinates};
\draw[seq-green, vector] (0, 0) to["$x$" below right] (1, 2);
\draw[seq-red, vector] (0, 0) to["$Ax$"] (-1, 2);
\point at (0, 0);
\end{scope}
\end{tikzpicture}
</latex-code>
</p>
<p>
To summarize: if <m>A = CBC\inv</m>, then <m>A</m> and <m>B</m> do the same thing, only in different coordinate systems.
</p>
</bluebox>
<p>
The following example is the heart of this section.
</p>
<specialcase xml:id="similarity-worked-eg">
<idx><h>Similarity</h><h>worked example</h></idx>
<p>
Consider the matrices
<me>
A = \mat{1/2 3/2; 3/2 1/2} \qquad
B = \mat{2 0; 0 -1} \qquad
C = \mat{1 1; 1 -1}.
</me>
One can verify that <m>A = CBC\inv</m>: see this <xref ref="diagonal-eg-22"/>. Let <m>v_1 = {1\choose 1}</m> and <m>v_2 = {1\choose -1}</m>, the columns of <m>C</m>, and let <m>\cB = \{v_1,v_2\}</m>, a basis of <m>\R^2</m>.
</p>
<p>
The matrix <m>B</m> is diagonal: it scales the <m>x</m>-direction by a factor of <m>2</m> and the <m>y</m>-direction by a factor of <m>-1</m>.
<latex-code>
\def\theo{\includegraphics[width=6cm]{theo11.jpg}}
\begin{tikzpicture}[scale=.8, thin border nodes]
\node[transform shape] at (0, 0) {\theo};
\draw[seq-violet, opacity=1, vector] (0,0) -- (1,0) node[right] {$e_1$};
\draw[seq-blue, opacity=1, vector] (0,0) -- (0,1) node[above] {$e_2$};
\point at (0, 0);
\begin{scope}[xshift=10cm, xscale=2, yscale=-1]
\clip (-1.5, -3) rectangle (1.5, 3);
\node[transform shape] at (0, 0) {\theo};
\draw[seq-violet, opacity=1, vector] (0,0) -- (1, 0) node[right] {$Be_1$};
\draw[seq-blue, opacity=1, vector] (0,0) -- (0, 1) node[below] {$Be_2$};
\point at (0, 0);
\end{scope}
\draw (-3, -3) rectangle (3, 3);
\draw[xshift=10cm] (-3, -3) rectangle (3, 3);
\draw[->, shorten=2mm, thick] (3, 1) to[bend left, "$B$" above=.5mm] (7, 1);
\end{tikzpicture}
</latex-code>
To compute <m>Ax</m>, first we multiply by <m>C\inv</m> to find the <m>\cB</m>-coordinates of <m>x</m>, then we multiply by <m>B</m>, then we multiply by <m>C</m> again. For instance, let <m>\textcolor{seq-green}x = {0\choose -2}</m>.
<ol>
<li>
We see from the <m>\cB</m>-coordinate grid below that
<m>\textcolor{seq-green}x = -\textcolor{seq-violet}{v_1} + \textcolor{seq-blue}{v_2}</m>. Therefore, <m>C\inv\textcolor{seq-green}x = \textcolor{seq-green}{[x]_\cB} = {-1\choose 1}.</m>
</li>
<li>
Multiplying by <m>B</m> scales the coordinates: <m>\textcolor{seq-red}{B[x]_\cB} = {-2\choose -1}</m>.
</li>
<li>
Interpreting <m>{-2\choose -1}</m> as a <m>\cB</m>-coordinate vector, we multiply by <m>C</m> to get
<me>
\textcolor{seq-red}{Ax} = C\vec{-2 -1} = -2\textcolor{seq-violet}{v_1} - \textcolor{seq-blue}{v_2} = \vec{-3 -1}.
</me>
Of course, this vector lies at <m>(-2, -1)</m> on the <m>\cB</m>-coordinate grid.
</li>
</ol>
<latex-code>
\begin{tikzpicture}[scale=.85, thin border nodes]
\draw[help lines] (-3,-3) grid (3,3);
\node at (0, 3.5) {$\cB$-coordinates};
\draw[seq-violet, vector, opacity=.5] (0,0) -- (1,0);
\draw[seq-blue, vector, opacity=.5] (0,0) -- (0,1);
\draw[seq-green, vector] (0, 0) to["{$[x]_\cB$}" {above left, pos=1}] (-1, 1);
\draw[seq-red, vector] (0, 0) to["{$B[x]_\cB$}"] (-2, -1);
\draw[->, shorten=2mm, thick] (9, 1) to[bend right]
node[above=1mm] {multiply by $C\inv$} (3, 1);
\node[align=center] (A) at (6, 0)
{scale $x$ by $2$\\scale $y$ by $-1$};
\draw[->, shorten >=2mm, thick] (A) -- (3, 0);
\draw[->, shorten=2mm, thick] (3, -1) to[bend right]
node[below=1mm] {multiply by $C$} (9, -1);
\point at (0, 0);
\begin{scope}[xshift=12cm]
\node at (0, 3.5) {usual coordinates};
\draw[help lines] (-3, -3) rectangle (3, 3);
\path[clip] (-3, -3) rectangle (3, 3);
\begin{scope}[cm={(1,1,1,-1,(0,0))}, scale=.7]
\draw[help lines] (-5,-5) grid (5,5);
\draw[seq-violet, vector, opacity=.5] (0,0) -- (1,0);
\draw[seq-blue, vector, opacity=.5] (0,0) -- (0,1);
\draw[seq-green, vector] (0, 0) to["$x$" left] (-1, 1);
\draw[seq-red, vector] (0, 0) to["$Ax$" above left] (-2, -1);
\end{scope}
\point at (0, 0);
\end{scope}
\end{tikzpicture}
</latex-code>
Now let <m>\textcolor{seq-green}x = \frac 12{5\choose -3}</m>.
<ol>
<li>
We see from the <m>\cB</m>-coordinate grid that
<m>\textcolor{seq-green}x = \frac 12\textcolor{seq-violet}{v_1} + 2\textcolor{seq-blue}{v_2}</m>. Therefore, <m>C\inv\textcolor{seq-green}x = \textcolor{seq-green}{[x]_\cB} = {1/2\choose 2}.</m>
</li>
<li>
Multiplying by <m>B</m> scales the coordinates: <m>\textcolor{seq-red}{B[x]_\cB} = {1\choose -2}</m>.
</li>
<li>
Interpreting <m>{1\choose -2}</m> as a <m>\cB</m>-coordinate vector, we multiply by <m>C</m> to get
<me>
\textcolor{seq-red}{Ax} = C\vec{1 -2} = \textcolor{seq-violet}{v_1} - 2\textcolor{seq-blue}{v_2} = \vec{-1 3}.
</me>
This vector lies at <m>(1, -2)</m> on the <m>\cB</m>-coordinate grid.
</li>
</ol>
<latex-code>
\begin{tikzpicture}[scale=.85, thin border nodes]
\draw[help lines] (-3,-3) grid (3,3);
\node at (0, 3.5) {$\cB$-coordinates};
\draw[seq-violet, vector, opacity=.5] (0,0) -- (1,0);
\draw[seq-blue, vector, opacity=.5] (0,0) -- (0,1);
\draw[seq-green, vector] (0, 0) to["{$[x]_\cB$}" {above left, pos=1}] (1/2, 2);
\draw[seq-red, vector] (0, 0) to["{$B[x]_\cB$}" pos=.7] (1, -2);
\draw[->, shorten=2mm, thick] (9, 1) to[bend right]
node[above=1mm] {multiply by $C\inv$} (3, 1);
\node[align=center] (A) at (6, 0)
{scale $x$ by $2$\\scale $y$ by $-1$};
\draw[->, shorten >=2mm, thick] (A) -- (3, 0);
\draw[->, shorten=2mm, thick] (3, -1) to[bend right]
node[below=1mm] {multiply by $C$} (9, -1);
\point at (0, 0);
\begin{scope}[xshift=12cm]
\node at (0, 3.5) {usual coordinates};
\draw[help lines] (-3, -3) rectangle (3, 3);
\path[clip] (-3, -3) rectangle (3, 3);
\begin{scope}[cm={(1,1,1,-1,(0,0))}, scale=.7]
\draw[help lines] (-5,-5) grid (5,5);
\draw[seq-violet, vector, opacity=.5] (0,0) -- (1,0);
\draw[seq-blue, vector, opacity=.5] (0,0) -- (0,1);
\draw[seq-green, vector] (0, 0) to["$x$" {above right, pos=.7}] (1/2, 2);
\draw[seq-red, vector] (0, 0) to["$Ax$" left] (1, -2);
\end{scope}
\point at (0, 0);
\end{scope}
\end{tikzpicture}
</latex-code>
To summarize:
<ul>
<li>
<m>B</m> scales the <m>e_1</m>-direction by <m>2</m> and the <m>e_2</m>-direction by <m>-1</m>.
</li>
<li>
<m>A</m> scales the <m>v_1</m>-direction by <m>2</m> and the <m>v_2</m>-direction by <m>-1</m>.
</li>
</ul>
<latex-code>
\def\theo{\includegraphics[width=6cm]{theo11.jpg}}
\begin{tikzpicture}[scale=.8, thin border nodes]
\node[transform shape] at (0, 0) {\theo};
\draw[seq-violet, opacity=1, vector] (0,0) -- (1,0) node[right] {$e_1$};
\draw[seq-blue, opacity=1, vector] (0,0) -- (0,1) node[above] {$e_2$};
\point at (0, 0);
\begin{scope}[xshift=9cm, xscale=2, yscale=-1]
\clip (-1.5, -3) rectangle (1.5, 3);
\node[transform shape] at (0, 0) {\theo};
\draw[seq-violet, opacity=1, vector] (0,0) -- (1, 0) node[right] {$Be_1$};
\draw[seq-blue, opacity=1, vector] (0,0) -- (0, 1) node[below] {$Be_2$};
\point at (0, 0);
\end{scope}
\begin{scope}[yshift=-9cm]
\clip (-3, -3) rectangle (3, 3);
\begin{scope}[cm={(1,1,1,-1,(0,0))}]
\node[transform shape] at (0, 0) {\theo};
\draw[seq-violet, opacity=1, vector] (0,0) -- (1,0) node[right] {$v_1$};
\draw[seq-blue, opacity=1, vector] (0,0) -- (0,1) node[right] {$v_2$};
\end{scope}
\point at (0, 0);
\end{scope}
\begin{scope}[yshift=-9cm, xshift=9cm]
\clip (-3, -3) rectangle (3, 3);
\begin{scope}[cm={(1,1,1,-1,(0,0))}]
\begin{scope}[xscale=2, yscale=-1]
\node[transform shape] at (0, 0) {\theo};
\draw[seq-violet, opacity=1, vector] (0,0) -- (1,0) node[right] {$Av_1$};
\draw[seq-blue, opacity=1, vector] (0,0) -- (0,1) node[left] {$Av_2$};
\end{scope}
\end{scope}
\point at (0, 0);
\end{scope}
\draw (-3, -3) rectangle (3, 3);
\draw[xshift=9cm] (-3, -3) rectangle (3, 3);
\draw[yshift=-9cm] (-3, -3) rectangle (3, 3);
\draw[xshift=9cm, yshift=-9cm] (-3, -3) rectangle (3, 3);
\draw[->, shorten=2mm, thick] (3, 1) to[bend left, "$B$" above=.5mm] (6, 1);
\draw[->, shorten=2mm, thick, yshift=-9cm]
(3, -1) to[bend right, "$A$" above=.5mm] (6, -1);
\draw[->, shorten=2mm, thick] (-1, -6) to[bend left, "$C\inv$" right=.5mm] (-1, -3);
\draw[->, shorten=2mm, thick, xshift=9cm]
(1, -3) to[bend left, "$C$" left=.5mm] (1, -6);
\end{tikzpicture}
</latex-code>
</p>
<figure>
<caption>The geometric relationship between the similar matrices <m>A</m> and <m>B</m> acting on <m>\R^2</m>. Click and drag the heads of <m>x</m> and <m>[x]_\cB</m>. Study this picture until you can reliably predict where the other three vectors will be after moving one of them: this is the essence of the geometry of similar matrices.</caption>
<mathbox source="demos/similarity.html?B=2,0:0,-1&C=1,1:1,-1&closed" height="600px"/>
</figure>
</specialcase>
<example hide-type="true" xml:id="similarity-another-matrix">
<title>Interactive: Another matrix similar to <m>B</m></title>
<p>
Consider the matrices
<me>
A' = \frac 15\mat{-8 -9; 6 13} \qquad
B = \mat{2 0; 0 -1} \qquad
C' = \frac 12\mat{-1 -3; 2 1}.
</me>
Then <m>A' = C'B(C')\inv</m>, as one can verify. Let <m>v_1'=\frac 12{-1\choose 2}</m> and <m>v_2' = \frac 12{-3\choose 1}</m>, the columns of <m>C'</m>, and let <m>\cB' = \{v_1',v_2'\}</m>. Then <m>A'</m> does the same thing as <m>B</m>, as in the previous <xref ref="similarity-worked-eg"/>, except <m>A'</m> uses the <m>\cB'</m>-coordinate system. In other words:
<ul>
<li>
<m>B</m> scales the <m>e_1</m>-direction by <m>2</m> and the <m>e_2</m>-direction by <m>-1</m>.
</li>
<li>
<m>A'</m> scales the <m>v_1'</m>-direction by <m>2</m> and the <m>v_2'</m>-direction by <m>-1</m>.
</li>
</ul>
<latex-code>
\def\theo{\includegraphics[width=6cm]{theo6.jpg}}
\begin{tikzpicture}[scale=.8, thin border nodes]
\node[transform shape] at (0, 0) {\theo};
\draw[seq-violet, opacity=1, vector] (0,0) -- (1,0) node[right] {$e_1$};
\draw[seq-blue, opacity=1, vector] (0,0) -- (0,1) node[above] {$e_2$};
\point at (0, 0);
\begin{scope}[xshift=9cm, xscale=2, yscale=-1]
\clip (-1.5, -3) rectangle (1.5, 3);
\node[transform shape] at (0, 0) {\theo};
\draw[seq-violet, opacity=1, vector] (0,0) -- (1, 0) node[right] {$Be_1$};
\draw[seq-blue, opacity=1, vector] (0,0) -- (0, 1) node[below] {$Be_2$};
\point at (0, 0);
\end{scope}
\begin{scope}[yshift=-9cm]
\clip (-3, -3) rectangle (3, 3);
\begin{scope}[cm={(-1/2,1,-3/2,1/2,(0,0))}]
\node[transform shape] at (0, 0) {\theo};
\draw[seq-violet, opacity=1, vector] (0,0) -- (1,0) node[above] {$v_1'$};
\draw[seq-blue, opacity=1, vector] (0,0) -- (0,1) node[left] {$v_2'$};
\end{scope}
\point at (0, 0);
\end{scope}
\begin{scope}[yshift=-9cm, xshift=9cm]
\clip (-3, -3) rectangle (3, 3);
\begin{scope}[cm={(-1/2,1,-3/2,1/2,(0,0))}]
\begin{scope}[xscale=2, yscale=-1]
\node[transform shape] at (0, 0) {\theo};
\draw[seq-violet, opacity=1, vector] (0,0) -- (1,0) node[above=-1mm] {$A'v_1'$};
\draw[seq-blue, opacity=1, vector] (0,0) -- (0,1) node[right] {$A'v_2'$};
\end{scope}
\end{scope}
\point at (0, 0);
\end{scope}
\draw (-3, -3) rectangle (3, 3);
\draw[xshift=9cm] (-3, -3) rectangle (3, 3);
\draw[yshift=-9cm] (-3, -3) rectangle (3, 3);
\draw[xshift=9cm, yshift=-9cm] (-3, -3) rectangle (3, 3);
\draw[->, shorten=2mm, thick] (3, 1) to[bend left, "$B$" above=.5mm] (6, 1);
\draw[->, shorten=2mm, thick, yshift=-9cm]
(3, -1) to[bend right, "$A'$" above=.5mm] (6, -1);
\draw[->, shorten=2mm, thick] (-1, -6) to[bend left, "$(C')\inv$" right=.5mm] (-1, -3);
\draw[->, shorten=2mm, thick, xshift=9cm]
(1, -3) to[bend left, "$C'$" left=.5mm] (1, -6);
\end{tikzpicture}
</latex-code>
</p>
<figure>
<caption>The geometric relationship between the similar matrices <m>A'</m> and <m>B</m> acting on <m>\R^2</m>. Click and drag the heads of <m>x</m> and <m>[x]_{\cB'}</m>.</caption>
<mathbox source="demos/similarity.html?B=2,0:0,-1&C=-1/2,-3/2:1,1/2&closed" height="600px"/>
</figure>
</example>
<example xml:id="similarity-eg-rotation">
<title>A matrix similar to a rotation matrix</title>
<p>
Consider the matrices
<me>
A = \frac 16\mat{7 -17; 5 -7} \qquad
B = \mat{0 -1; 1 0} \qquad
C = \mat{2 -1/2; 1 1/2}.
</me>
One can verify that <m>A = CBC\inv.</m> Let <m>v_1 = {2\choose 1}</m> and <m>v_2 = \frac 12{-1\choose 1}</m>, the columns of <m>C</m>, and let <m>\cB = \{v_1,v_2\}</m>, a basis of <m>\R^2</m>.
</p>
<p>
The matrix <m>B</m> rotates the plane counterclockwise by <m>90^\circ</m>.
<latex-code>
\def\theo{\includegraphics[width=6cm]{theo10.jpg}}
\begin{tikzpicture}[scale=.8, thin border nodes]
\node[transform shape] at (0, 0) {\theo};
\draw[seq-violet, opacity=1, vector] (0,0) -- (1,0) node[right] {$e_1$};
\draw[seq-blue, opacity=1, vector] (0,0) -- (0,1) node[above] {$e_2$};
\point at (0, 0);
\begin{scope}[xshift=10cm]
\clip (-3, -3) rectangle (3, 3);
\begin{scope}[rotate=90]
\node[transform shape] at (0, 0) {\theo};
\draw[seq-violet, opacity=1, vector] (0,0) -- (1, 0) node[above] {$Be_1$};
\draw[seq-blue, opacity=1, vector] (0,0) -- (0, 1) node[left] {$Be_2$};
\end{scope}
\point at (0, 0);
\end{scope}
\draw (-3, -3) rectangle (3, 3);
\draw[xshift=10cm] (-3, -3) rectangle (3, 3);
\draw[->, shorten=2mm, thick] (3, 1) to[bend left, "$B$" above=.5mm] (7, 1);
\end{tikzpicture}
</latex-code>
To compute <m>Ax</m>, first we multiply by <m>C\inv</m> to find the <m>\cB</m>-coordinates of <m>x</m>, then we multiply by <m>B</m>, then we multiply by <m>C</m> again. For instance, let <m>\textcolor{seq-green}x = \frac 32{1\choose 1}</m>.
<ol>
<li>
We see from the <m>\cB</m>-coordinate grid below that
<m>\textcolor{seq-green}x = \textcolor{seq-violet}{v_1} + \textcolor{seq-blue}{v_2}</m>. Therefore, <m>C\inv\textcolor{seq-green}x = \textcolor{seq-green}{[x]_\cB} = {1\choose 1}.</m>
</li>
<li>
Multiplying by <m>B</m> rotates by <m>90^\circ</m>: <m>\textcolor{seq-red}{B[x]_\cB} = {-1\choose 1}</m>.
</li>
<li>
Interpreting <m>{-1\choose 1}</m> as a <m>\cB</m>-coordinate vector, we multiply by <m>C</m> to get
<me>
\textcolor{seq-red}{Ax} = C\vec{-1 1} = -\textcolor{seq-violet}{v_1} + \textcolor{seq-blue}{v_2} = \frac 12\vec{-5 -1}.
</me>
Of course, this vector lies at <m>(-1, 1)</m> on the <m>\cB</m>-coordinate grid.
</li>
</ol>
<latex-code>
\begin{tikzpicture}[scale=.75, thin border nodes]
\node at (0, 3.5) {$\cB$-coordinates};
\begin{scope}[scale=3/2, arrows={[line width=.6pt]}]
\draw[help lines] (-2,-2) grid (2, 2);
\draw[help lines, seq-yellow] (0, 0) circle[radius=1];
\draw[seq-violet, vector, opacity=.5] (0,0) -- (1,0);
\draw[seq-blue, vector, opacity=.5] (0,0) -- (0,1);
\draw[|->, shorten=1.5mm, dashed]
(1,1) arc[radius=sqrt(2), start angle=45, delta angle=90];
\draw[seq-green, vector] (0, 0) -- (1, 1) node[above right] {$[x]_\cB$};
\draw[seq-red, vector] (0, 0) -- (-1, 1) node[above left] {$B[x]_\cB$};
\end{scope}
\draw[->, shorten=2mm, thick] (9, 1) to[bend right]
node[above=1mm] {multiply by $C\inv$} (3, 1);
\node (A) at (6, 0) {rotate by $90^\circ$};
\draw[->, shorten >=2mm, thick] (A) -- (3, 0);
\draw[->, shorten=2mm, thick] (3, -1) to[bend right]
node[below=1mm] {multiply by $C$} (9, -1);
\point at (0, 0);
\begin{scope}[xshift=12cm]
\node at (0, 3.5) {usual coordinates};
\draw[help lines] (-3, -3) rectangle (3, 3);
\path[clip] (-3, -3) rectangle (3, 3);
\begin{scope}[cm={(2,1,-1/2,1/2,(0,0))}, arrows={[line width=.6pt]}]
\draw[help lines] (-10,-10) grid (10, 10);
\draw[help lines, seq-yellow] (0, 0) circle[radius=1];
\draw[seq-violet, vector, opacity=.5] (0,0) -- (1,0);
\draw[seq-blue, vector, opacity=.5] (0,0) -- (0,1);
\draw[|->, dashed, shorten=1.5mm]
(1,1) arc[radius=sqrt(2), start angle=45, delta angle=90];
\draw[seq-green, vector] (0, 0) -- (1, 1) node[above right] {$x$};
\draw[seq-red, vector] (0, 0) -- (-1, 1) node[below=1mm, xshift=1mm] {$Ax$};
\end{scope}
\point at (0, 0);
\end{scope}
\end{tikzpicture}
</latex-code>
Now let <m>\textcolor{seq-green}x = \frac 12{-1\choose -2}</m>.
<ol>
<li>
We see from the <m>\cB</m>-coordinate grid that
<m>\textcolor{seq-green}x = -\frac 12\textcolor{seq-violet}{v_1} - \textcolor{seq-blue}{v_2}</m>. Therefore, <m>C\inv\textcolor{seq-green}x = \textcolor{seq-green}{[x]_\cB} = {-1/2\choose -1}.</m>
</li>
<li>
Multiplying by <m>B</m> rotates by <m>90^\circ</m>: <m>\textcolor{seq-red}{B[x]_\cB} = {1\choose -1/2}</m>.
</li>
<li>
Interpreting <m>{1\choose -1/2}</m> as a <m>\cB</m>-coordinate vector, we multiply by <m>C</m> to get
<me>
\textcolor{seq-red}{Ax} = C\vec{1 -1/2} = \textcolor{seq-violet}{v_1} - \frac 12\textcolor{seq-blue}{v_2} = \frac 14\vec{9 3}.
</me>
This vector lies at <m>(1, -\frac 12)</m> on the <m>\cB</m>-coordinate grid.
</li>
</ol>
<latex-code>
\begin{tikzpicture}[scale=.75, thin border nodes]
\node at (0, 3.5) {$\cB$-coordinates};
\begin{scope}[scale=3/2, arrows={[line width=.6pt]}]
\draw[help lines] (-2,-2) grid (2, 2);
\draw[help lines, seq-yellow] (0, 0) circle[radius=1];
\draw[seq-violet, vector, opacity=.5] (0,0) -- (1,0);
\draw[seq-blue, vector, opacity=.5] (0,0) -- (0,1);
\draw[|->, shorten=1.5mm, dashed]
(-1/2,-1) arc[radius=sqrt(5/4), start angle={atan2(-1,-1/2)}, delta angle=90];
\draw[seq-green, vector] (0, 0) -- (-1/2, -1) node[below left=-1mm] {$[x]_\cB$};
\draw[seq-red, vector] (0, 0) -- (1, -1/2) node[right] {$B[x]_\cB$};
\end{scope}
\draw[->, shorten=2mm, thick] (9, 1) to[bend right]
node[above=1mm] {multiply by $C\inv$} (3, 1);
\node (A) at (6, 0) {rotate by $90^\circ$};
\draw[->, shorten >=2mm, thick] (A) -- (3, 0);
\draw[->, shorten=2mm, thick] (3, -1) to[bend right]
node[below=1mm] {multiply by $C$} (9, -1);
\point at (0, 0);
\begin{scope}[xshift=12cm]
\node at (0, 3.5) {usual coordinates};
\draw[help lines] (-3, -3) rectangle (3, 3);
\path[clip] (-3, -3) rectangle (3, 3);
\begin{scope}[cm={(2,1,-1/2,1/2,(0,0))}, arrows={[line width=.6pt]}]
\draw[help lines] (-10,-10) grid (10, 10);
\draw[help lines, seq-yellow] (0, 0) circle[radius=1];
\draw[seq-violet, vector, opacity=.5] (0,0) -- (1,0);
\draw[seq-blue, vector, opacity=.5] (0,0) -- (0,1);
\draw[|->, shorten=1.5mm, dashed]
(-1/2,-1) arc[radius=sqrt(5/4), start angle={atan2(-1,-1/2)}, delta angle=90];
\draw[seq-green, vector] (0, 0) -- (-1/2,-1) node[below] {$x$};
\draw[seq-red, vector] (0, 0) -- (1, -1/2) node[right] {$Ax$};
\end{scope}
\point at (0, 0);
\end{scope}
\end{tikzpicture}
</latex-code>
To summarize:
<ul>
<li>
<m>B</m> rotates counterclockwise around the circle centered at the origin and passing through <m>e_1</m> and <m>e_2</m>.
</li>
<li>
<m>A</m> rotates counterclockwise around the ellipse centered at the origin and passing through <m>v_1</m> and <m>v_2</m>.
</li>
</ul>
<latex-code>
\def\theo{\includegraphics[width=6cm]{theo10.jpg}}
\begin{tikzpicture}[scale=.75, thin border nodes]
\node[transform shape] at (0, 0) {\theo};
\draw[seq-yellow, opacity=.5] (0,0) circle[radius=1];
\draw[seq-violet, opacity=1, vector] (0,0) -- (1,0) node[right] {$e_1$};
\draw[seq-blue, opacity=1, vector] (0,0) -- (0,1) node[above] {$e_2$};
\point at (0, 0);
\begin{scope}[xshift=9cm, rotate=90]
\clip (-3, -3) rectangle (3, 3);
\node[transform shape] at (0, 0) {\theo};
\draw[seq-yellow, opacity=.5] (0,0) circle[radius=1];
\draw[seq-violet, opacity=1, vector] (0,0) -- (1, 0) node[right] {$Be_1$};
\draw[seq-blue, opacity=1, vector] (0,0) -- (0, 1) node[below] {$Be_2$};
\point at (0, 0);
\end{scope}
\begin{scope}[yshift=-9cm]
\clip (-3, -3) rectangle (3, 3);
\begin{scope}[cm={(2,1,-1/2,1/2,(0,0))}]
\node[transform shape] at (0, 0) {\theo};
\draw[seq-yellow, opacity=.5] (0,0) circle[radius=1];
\draw[seq-violet, opacity=1, vector] (0,0) -- (1,0) node[above] {$v_1$};
\draw[seq-blue, opacity=1, vector] (0,0) -- (0,1) node[left] {$v_2$};
\end{scope}
\point at (0, 0);
\end{scope}
\begin{scope}[yshift=-9cm, xshift=9cm]
\clip (-3, -3) rectangle (3, 3);
\begin{scope}[cm={(2,1,-1/2,1/2,(0,0))}, rotate=90]
\node[transform shape] at (0, 0) {\theo};
\draw[seq-yellow, opacity=.5] (0,0) circle[radius=1];
\draw[seq-violet, opacity=1, vector] (0,0) -- (1,0) node[above] {$Av_1$};
\draw[seq-blue, opacity=1, vector] (0,0) -- (0,1) node[below] {$Av_2$};
\end{scope}
\point at (0, 0);
\end{scope}
\draw (-3, -3) rectangle (3, 3);
\draw[xshift=9cm] (-3, -3) rectangle (3, 3);
\draw[yshift=-9cm] (-3, -3) rectangle (3, 3);
\draw[xshift=9cm, yshift=-9cm] (-3, -3) rectangle (3, 3);
\draw[->, shorten=2mm, thick] (3, 1) to[bend left, "$B$" above=.5mm] (6, 1);
\draw[->, shorten=2mm, thick, yshift=-9cm]
(3, -1) to[bend right, "$A'$" above=.5mm] (6, -1);
\draw[->, shorten=2mm, thick] (-1, -6) to[bend left, "$C\inv$" right=.5mm] (-1, -3);
\draw[->, shorten=2mm, thick, xshift=9cm]
(1, -3) to[bend left, "$C$" left=.5mm] (1, -6);
\end{tikzpicture}
</latex-code>
</p>
<figure>
<caption>The geometric relationship between the similar matrices <m>A</m> and <m>B</m> acting on <m>\R^2</m>. Click and drag the heads of <m>x</m> and <m>[x]_\cB</m>.</caption>
<mathbox source="demos/similarity.html?B=0,-1:1,0&C=2,-1/2:1,1/2&snap=off" height="600px"/>
</figure>
</example>
<p>
To summarize and generalize the previous example:
</p>
<bluebox xml:id="similarity-to-rotation">
<title>A Matrix Similar to a Rotation Matrix</title>
<p>
Let
<me>
B = \mat{\cos\theta,-\sin\theta; \sin\theta,\cos\theta}
\qquad
C = \mat{| |; v_1 v_2; | |}
\qquad
A = CBC\inv,
</me>
where <m>C</m> is assumed invertible. Then:
<ul>
<li>
<m>B</m> rotates the plane by an angle of <m>\theta</m> around the circle centered at the origin and passing through <m>e_1</m> and <m>e_2</m>, in the direction from <m>e_1</m> to <m>e_2</m>.
</li>
<li>
<m>A</m> rotates the plane by an angle of <m>\theta</m> around the ellipse centered at the origin and passing through <m>v_1</m> and <m>v_2</m>, in the direction from <m>v_1</m> to <m>v_2</m>.
</li>
</ul>
<latex-code>
\def\theo{\includegraphics[width=6cm]{theo10.jpg}}
\begin{tikzpicture}[scale=.75, thin border nodes]
\node[transform shape] at (0, 0) {\theo};
\draw[seq-yellow, opacity=.5] (0,0) circle[radius=1];
\draw[seq-violet, opacity=1, vector] (0,0) -- (1,0) node[right] {$e_1$};
\draw[seq-blue, opacity=1, vector] (0,0) -- (0,1) node[above] {$e_2$};
\point at (0, 0);
\begin{scope}[xshift=9cm, rotate=90]
\clip (-3, -3) rectangle (3, 3);
\node[transform shape] at (0, 0) {\theo};
\draw[seq-yellow, opacity=.5] (0,0) circle[radius=1];
\draw[seq-violet, opacity=1, vector] (0,0) -- (1, 0) node[right] {$Be_1$};
\draw[seq-blue, opacity=1, vector] (0,0) -- (0, 1) node[below] {$Be_2$};
\point at (0, 0);
\end{scope}
\begin{scope}[yshift=-9cm]
\clip (-3, -3) rectangle (3, 3);
\begin{scope}[cm={(2,1,-1/2,1/2,(0,0))}]
\node[transform shape] at (0, 0) {\theo};
\draw[seq-yellow, opacity=.5] (0,0) circle[radius=1];
\draw[seq-violet, opacity=1, vector] (0,0) -- (1,0) node[above] {$v_1$};
\draw[seq-blue, opacity=1, vector] (0,0) -- (0,1) node[left] {$v_2$};
\end{scope}
\point at (0, 0);
\end{scope}
\begin{scope}[yshift=-9cm, xshift=9cm]
\clip (-3, -3) rectangle (3, 3);
\begin{scope}[cm={(2,1,-1/2,1/2,(0,0))}, rotate=90]
\node[transform shape] at (0, 0) {\theo};
\draw[seq-yellow, opacity=.5] (0,0) circle[radius=1];
\draw[seq-violet, opacity=1, vector] (0,0) -- (1,0) node[above] {$Av_1$};
\draw[seq-blue, opacity=1, vector] (0,0) -- (0,1) node[below] {$Av_2$};
\end{scope}
\point at (0, 0);
\end{scope}
\draw (-3, -3) rectangle (3, 3);
\draw[xshift=9cm] (-3, -3) rectangle (3, 3);
\draw[yshift=-9cm] (-3, -3) rectangle (3, 3);
\draw[xshift=9cm, yshift=-9cm] (-3, -3) rectangle (3, 3);
\draw[->, shorten=2mm, thick] (3, 1) to[bend left, "$B$" above=.5mm] (6, 1);
\draw[->, shorten=2mm, thick, yshift=-9cm]
(3, -1) to[bend right, "$A'$" above=.5mm] (6, -1);
\draw[->, shorten=2mm, thick] (-1, -6) to[bend left, "$C\inv$" right=.5mm] (-1, -3);
\draw[->, shorten=2mm, thick, xshift=9cm]
(1, -3) to[bend left, "$C$" left=.5mm] (1, -6);
\end{tikzpicture}
</latex-code>
</p>
</bluebox>
<example hide-type="true" xml:id="similarity-33-eg">
<title>Interactive: Similar <m>3\times 3</m> matrices</title>
<p>
Consider the matrices
<me>
A = \mat{-1 0 0; -1 0 2; -1 1 1} \qquad
B = \mat{-1 0 0; 0 -1 0; 0 0 2} \qquad
C = \mat{-1 1 0; 1 1 1; -1 0 1}.
</me>
Then <m>A = CBC\inv</m>, as one can verify. Let <m>v_1,v_2,v_3</m> be the columns of <m>C</m>, and let <m>\cB = \{v_1,v_2,v_3\}</m>, a basis of <m>\R^3</m>. Then <m>A</m> does the same thing as <m>B</m>, except <m>A</m> uses the <m>\cB</m>-coordinate system. In other words:
<ul>
<li>
<m>B</m> scales the <m>e_1,e_2</m>-plane by <m>-1</m> and the <m>e_3</m>-direction by <m>2</m>.
</li>
<li>
<m>A</m> scales the <m>v_1,v_2</m>-plane by <m>-1</m> and the <m>v_3</m>-direction by <m>2</m>.
</li>
</ul>
</p>
<figure>
<caption>The geometric relationship between the similar matrices <m>A</m> and <m>B</m> acting on <m>\R^3</m>. Click and drag the heads of <m>x</m> and <m>[x]_\cB</m>.</caption>
<mathbox source="demos/similarity.html?B=-1,0,0:0,-1,0:0,0,2&C=-1,1,0:1,1,1:-1,0,1&closed" height="600px"/>
</figure>
</example>
</subsection>
<subsection>
<title>Eigenvalues of Similar Matrices</title>
<p>
Since similar matrices behave in the same way with respect to different coordinate systems, we should expect their eigenvalues and eigenvectors to be closely related.
</p>
<theorem type-name="Fact" xml:id="similarity-charpoly">
<idx><h>Similarity</h><h>and the characteristic polynomial</h></idx>
<idx><h>Characteristic polynomial</h><h>of similar matrices</h></idx>
<statement>
<p>Similar matrices have the same characteristic polynomial.</p>
</statement>
<proof visible="true">
<p>
Suppose that <m>A = CBC\inv</m>, where <m>A,B,C</m> are <m>n\times n</m> matrices. We calculate
<me>
\begin{split}
A - \lambda I_n
\amp= CBC\inv - \lambda CC\inv
= CBC\inv - C\lambda C\inv \\
\amp= CBC\inv - C\lambda I_n C\inv
= C(B-\lambda I_n)C\inv.
\end{split}
</me>
Therefore,
<me>
\det(A-\lambda I_n) = \det(C(B-\lambda I_n)C\inv)
= \det(C)\det(B-\lambda I_n)\det(C)\inv
= \det(B-\lambda I_n).
</me>
Here we have used the <xref ref="det-defn-mult-prop">multiplicativity property</xref> and its <xref ref="det-defn-power-prop">corollary</xref>.
</p>
</proof>
</theorem>
<p>
Since the eigenvalues of a matrix are the roots of its characteristic polynomial, we have shown:
</p>
<bluebox>
<idx><h>Similarity</h><h>and eigenvalues</h></idx>
<idx><h>Eigenvalue</h><h>of similar matrices</h></idx>
<p>
Similar matrices have the same eigenvalues.
</p>
</bluebox>
<p>
<idx><h>Similarity</h><h>and the trace</h></idx>
<idx><h>Similarity</h><h>and the determinant</h></idx>
<idx><h>Determinant</h><h>of similar matrices</h></idx>
<idx><h>Matrix</h><h>trace of</h><h>similar matrices</h></idx>
By this <xref ref="charpoly-is-poly"/>, similar matrices also have the same trace and determinant.
</p>
<note>
<p>
The converse of the <xref ref="similarity-charpoly"/> is false. Indeed, the matrices
<me>\mat{1 1; 0 1} \sptxt{and} \mat{1 0 ; 0 1}</me>
both have characteristic polynomial <m>f(\lambda) = (\lambda-1)^2</m>, but they are not similar, because the only matrix that is similar to <m>I_2</m> is <m>I_2</m> itself.
</p>
</note>
<p>
Given that similar matrices have the same eigenvalues, one might guess that they have the same eigen<em>vectors</em> as well. Upon reflection, this is not what one should expect: indeed, the eigenvectors should only match up after changing from one coordinate system to another. This is the content of the next fact, remembering that <m>C</m> and <m>C\inv</m> change between the usual coordinates and the <m>\cB</m>-coordinates.
</p>
<theorem type-name="Fact" xml:id="similarity-evecs">
<idx><h>Eigenvector</h><h>of similar matrices</h></idx>
<idx><h>Similarity</h><h>and eigenvectors</h></idx>
<statement>
<p>
Suppose that <m>A = CBC\inv</m>. Then
<me>
\begin{split}
v \text{ is an eigenvector of } A
\amp\quad\implies\quad C\inv v \text{ is an eigenvector of } B \\
v \text{ is an eigenvector of } B
\amp\quad\implies\quad C v \text{ is an eigenvector of } A. \\
\end{split}
</me>
The eigenvalues of <m>v</m> / <m>C\inv v</m> or <m>v</m> / <m>Cv</m> are the same.
</p>
</statement>
<proof visible="true">
<p>
Suppose that <m>v</m> is an eigenvector of <m>A</m> with eigenvalue <m>\lambda</m>, so that <m>Av = \lambda v</m>. Then
<me>B(C\inv v) = C\inv(CBC\inv v) = C\inv(Av) = C\inv\lambda v = \lambda(C\inv v),</me>
so that <m>C\inv v</m> is an eigenvector of <m>B</m> with eigenvalue <m>\lambda</m>. Likewise if <m>v</m> is an eigenvector of <m>B</m> with eigenvalue <m>\lambda</m>, then <m>Bv = \lambda v</m>, and we have
<me>A(Cv) = (CBC\inv)Cv = CBv = C(\lambda v) = \lambda(Cv),</me>
so that <m>Cv</m> is an eigenvalue of <m>A</m> with eigenvalue <m>\lambda</m>.
</p>
</proof>
</theorem>
<bluebox>
<idx><h>Eigenspace</h><h>of similar matrices</h></idx>
<idx><h>Similarity</h><h>and eigenspaces</h></idx>
<p>
If <m>A = CBC\inv</m>, then <m>C\inv</m> takes the <m>\lambda</m>-eigenspace of <m>A</m> to the <m>\lambda</m>-eigenspace of <m>B</m>, and <m>C</m> takes the <m>\lambda</m>-eigenspace of <m>B</m> to the <m>\lambda</m>-eigenspace of <m>A</m>.
</p>
</bluebox>
<specialcase>
<p>
We continue with the above <xref ref="similarity-worked-eg"/>: let
<me>
A = \mat{1/2 3/2; 3/2 1/2} \qquad
B = \mat{2 0; 0 -1} \qquad
C = \mat{1 1; 1 -1},
</me>
so <m>A = CBC\inv.</m> Let <m>v_1 = {1\choose 1}</m> and <m>v_2 = {1\choose -1}</m>, the columns of <m>C</m>. Recall that:
<ul>
<li>
<m>B</m> scales the <m>e_1</m>-direction by <m>2</m> and the <m>e_2</m>-direction by <m>-1</m>.
</li>
<li>
<m>A</m> scales the <m>v_1</m>-direction by <m>2</m> and the <m>v_2</m>-direction by <m>-1</m>.
</li>
</ul>
This means that the <m>x</m>-axis is the <m>2</m>-eigenspace of <m>B</m>, and the <m>y</m>-axis is the <m>-1</m>-eigenspace of <m>B</m>; likewise, the <q><m>v_1</m>-axis</q> is the <m>2</m>-eigenspace of <m>A</m>, and the <q><m>v_2</m>-axis</q> is the <m>-1</m>-eigenspace of <m>A</m>. This is consistent with the <xref ref="similarity-evecs"/>, as multiplication by <m>C</m> changes <m>e_1</m> into <m>Ce_1 = v_1</m> and <m>e_2</m> into <m>Ce_2 = v_2</m>.
<latex-code>
\def\theo{\includegraphics[width=6cm]{theo11.jpg}}
\begin{tikzpicture}[scale=.8, thin border nodes]
\node[transform shape] at (0, 0) {\theo};
\draw[seq-violet, opacity=1] (-3,0) -- (3,0)
node[below,pos=.75,font=\small] {$2$-eigenspace};
\draw[seq-blue, opacity=1] (0,-3) -- (0,3)
node[above,pos=.25, rotate=90, font=\small] {$-1$-eigenspace};
\point at (0, 0);
\begin{scope}[xshift=9cm]
\clip (-3, -3) rectangle (3, 3);
\begin{scope}[cm={(1,1,1,-1,(0,0))}]
\node[transform shape] at (0, 0) {\theo};
\draw[seq-violet, opacity=1] (-3,0) -- (3,0)
node[above,pos=.75, rotate=45, font=\small] {$2$-eigenspace};
\draw[seq-blue, opacity=1] (0,-3) -- (0,3)
node[above,pos=.25, rotate=-45, font=\small] {$-1$-eigenspace};
\end{scope}
\point at (0, 0);
\end{scope}
\draw (-3, -3) rectangle (3, 3);
\draw[xshift=9cm] (-3, -3) rectangle (3, 3);
\draw[->, shorten=2mm, thick] (3, 1) to[bend left, "$C$" above=.5mm] (6, 1);
\end{tikzpicture}
</latex-code>
</p>
<figure>
<caption>The eigenspaces of <m>A</m> are the lines through <m>v_1</m> and <m>v_2</m>. These are the images under <m>C</m> of the coordinate axes, which are the eigenspaces of <m>B</m>.</caption>
<mathbox source="demos/eigenspace.html?mat=1/2,3/2:3/2,1/2&nomult" height="600px"/>
</figure>
</specialcase>
<example hide-type="true">
<title>Interactive: Another matrix similar to <m>B</m></title>
<p>
Continuing with this <xref ref="similarity-another-matrix"/>, let
<me>
A' = \frac 15\mat{-8 -9; 6 13} \qquad
B = \mat{2 0; 0 -1} \qquad
C' = \frac 12\mat{-1 -3; 2 1},
</me>
so <m>A' = C'B(C')\inv</m>. Let <m>v_1'=\frac 12{-1\choose 2}</m> and <m>v_2' = \frac 12{-3\choose 1}</m>, the columns of <m>C'</m>. Then:
<ul>
<li>
<m>B</m> scales the <m>e_1</m>-direction by <m>2</m> and the <m>e_2</m>-direction by <m>-1</m>.
</li>
<li>
<m>A'</m> scales the <m>v_1'</m>-direction by <m>2</m> and the <m>v_2'</m>-direction by <m>-1</m>.
</li>
</ul>
As before, the <m>x</m>-axis is the <m>2</m>-eigenspace of <m>B</m>, and the <m>y</m>-axis is the <m>-1</m>-eigenspace of <m>B</m>; likewise, the <q><m>v_1'</m>-axis</q> is the <m>2</m>-eigenspace of <m>A'</m>, and the <q><m>v_2'</m>-axis</q> is the <m>-1</m>-eigenspace of <m>A'</m>. This is consistent with the <xref ref="similarity-evecs"/>, as multiplication by <m>C'</m> changes <m>e_1</m> into <m>C'e_1 = v_1'</m> and <m>e_2</m> into <m>C'e_2 = v_2'</m>.
<latex-code>
\def\theo{\includegraphics[width=6cm]{theo6.jpg}}
\begin{tikzpicture}[scale=.8, thin border nodes]
\node[transform shape] at (0, 0) {\theo};
\draw[seq-violet, opacity=1] (-3,0) -- (3,0)
node[below,pos=.75,font=\small] {$2$-eigenspace};
\draw[seq-blue, opacity=1] (0,-3) -- (0,3)
node[above,pos=.25, rotate=90, font=\small] {$-1$-eigenspace};
\point at (0, 0);
\begin{scope}[xshift=9cm]
\clip (-3, -3) rectangle (3, 3);
\begin{scope}[cm={(-1/2,1,-3/2,1/2,(0,0))}]
\node[transform shape] at (0, 0) {\theo};
\draw[seq-violet, opacity=1] (-3,0) -- (3,0)
node[above,pos=.72, rotate={atan2(-1,1/2)}, font=\small] {$2$-eigenspace};
\draw[seq-blue, opacity=1] (0,-3) -- (0,3)
node[below,pos=.315, rotate={atan2(-1/2,3/2)}, font=\small] {$-1$-eigenspace};
\end{scope}
\point at (0, 0);
\end{scope}
\draw (-3, -3) rectangle (3, 3);
\draw[xshift=9cm] (-3, -3) rectangle (3, 3);
\draw[->, shorten=2mm, thick] (3, 1) to[bend left, "$C$" above=.5mm] (6, 1);
\end{tikzpicture}
</latex-code>
</p>
<figure>
<caption>The eigenspaces of <m>A'</m> are the lines through <m>v_1'</m> and <m>v_2'</m>. These are the images under <m>C'</m> of the coordinate axes, which are the eigenspaces of <m>B</m>.</caption>
<mathbox source="demos/eigenspace.html?mat=-8/5,-9/5:6/5,13/5&nomult" height="600px"/>
</figure>
</example>
<example hide-type="true">
<title>Interactive: Similar <m>3\times 3</m> matrices</title>
<p>
Continuing with this <xref ref="similarity-33-eg"/>, let
<me>
A = \mat{-1 0 0; -1 0 2; -1 1 1} \qquad
B = \mat{-1 0 0; 0 -1 0; 0 0 2} \qquad
C = \mat{-1 1 0; 1 1 1; -1 0 1},
</me>
so <m>A = CBC\inv</m>. Let <m>v_1,v_2,v_3</m> be the columns of <m>C</m>. Then:
<ul>
<li>
<m>B</m> scales the <m>e_1,e_2</m>-plane by <m>-1</m> and the <m>e_3</m>-direction by <m>2</m>.
</li>
<li>
<m>A</m> scales the <m>v_1,v_2</m>-plane by <m>-1</m> and the <m>v_3</m>-direction by <m>2</m>.
</li>
</ul>
In other words, the <m>xy</m>-plane is the <m>-1</m>-eigenspace of <m>B</m>, and the <m>z</m>-axis is the <m>2</m>-eigenspace of <m>B</m>; likewise, the <q><m>v_1,v_2</m>-plane</q> is the <m>-1</m>-eigenspace of <m>A</m>, and the <q><m>v_3</m>-axis</q> is the <m>2</m>-eigenspace of <m>A</m>. This is consistent with the <xref ref="similarity-evecs"/>, as multiplication by <m>C</m> changes <m>e_1</m> into <m>Ce_1 = v_1</m>, <m>e_2</m> into <m>Ce_2 = v_2</m>, and <m>e_3</m> into <m>Ce_3 = v_3</m>.
</p>
<figure>
<caption>The <m>-1</m>-eigenspace of <m>A</m> is the green plane, and the <m>2</m>-eigenspace of <m>A</m> is the violet line. These are the images under <m>C</m> of the <m>xy</m>-plane and the <m>z</m>-axis, respectively, which are the eigenspaces of <m>B</m>.</caption>
<mathbox source="demos/eigenspace.html?mat=-1,0,0:-1,0,2:-1,1,1&nomult" height="600px"/>
</figure>
</example>
</subsection>
</section>
</restrict-version>
此处可能存在不合适展示的内容，页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容，可点击提交进行申诉，我们将尽快为您处理。
马建仓 AI 助手