Eigenvalues
Eigenvalues
Sheldon Axler
[email protected]
This document contains Chapter 5 of the future fourth edition of Linear Algebra
Done Right. Notice the nice results that follow from the increasing use of the
minimal polynomial. Suggestions for improvements are most welcome.
The fourth edition of Linear Algebra Done Right will be an Open Access book,
which means that the electronic version will be legally free to the world. The
print version will be published by Springer and will be reasonably priced.
If you will teach an appropriate class beginning in January 2023, please consider
using the fourth edition of Linear Algebra Done Right as your textbook. Instructors
can contact me for an electronic version of the book that can be freely distributed
to your students (the electronic version will not be publicly available until the
print version is published). Your class testing can help improve this book.
Because this is not necessarily the final version of Chapter 5, please do not post
this document elsewhere on the web, although it is fine to link to it.
𝑛 𝑛
1 + √5
1 1 − √5
𝐹𝑛 = [( ) −( )]
√5 2 2
Cover equation: Formula for 𝑛th Fibonacci number. Exercise 16 in Section 5D
derives this formula by diagonalizing an appropriate operator.
About the Author
Sheldon Axler was valedictorian of his high school in Miami, Florida. He received
his AB from Princeton University with highest honors, followed by a PhD in
Mathematics from the University of California at Berkeley.
As a postdoctoral Moore Instructor at MIT, Axler received a university-wide
teaching award. He was then an assistant professor, associate professor, and
professor at Michigan State University, where he received the first J. Sutherland
Frame Teaching Award and the Distinguished Faculty Award.
Axler received the Lester R. Ford Award for expository writing from the Math-
ematical Association of America in 1996, for a paper that eventually expanded into
this book. In addition to publishing numerous research papers, he is the author of
six mathematics textbooks, ranging from freshman to graduate level. Previous
editions of this book have been adopted as a textbook at over 350 universities and
colleges and have been translated into three languages.
Axler has served as Editor-in-Chief of the Mathematical Intelligencer and
Associate Editor of the American Mathematical Monthly. He has been a member
of the Council of the American Mathematical Society and a member of the Board
of Trustees of the Mathematical Sciences Research Institute. He has also served
on the editorial board of Springer’s series Undergraduate Texts in Mathemat-
ics, Graduate Texts in Mathematics, Universitext, and Springer Monographs in
Mathematics.
He is a Fellow of the American Mathematical Society and has been a recipient
of numerous grants from the National Science Foundation.
Axler joined San Francisco State University as Chair of the Mathematics
Department in 1997. He served as Dean of the College of Science & Engineering
from 2002 to 2015, when he returned to a regular faculty appointment as a
professor in the Mathematics Department.
v
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Contents
Acknowledgments xvii
Chapter 1
Vector Spaces 1
1A 𝐑𝑛 and 𝐂𝑛 2
Complex Numbers 2
Lists 4
𝐅𝑛 6
Digression on Fields 10
Exercises 1A 10
1B Definition of Vector Space 12
Exercises 1B 16
1C Subspaces 18
Sums of Subspaces 19
Direct Sums 21
Exercises 1C 24
Chapter 2
Finite-Dimensional Vector Spaces 27
2A Span and Linear Independence 28
Linear Combinations and Span 28
Linear Independence 31
Exercises 2A 37
vi
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Contents vii
2B Bases 39
Exercises 2B 42
2C Dimension 44
Exercises 2C 48
Chapter 3
Linear Maps 51
3A Vector Space of Linear Maps 52
Definition and Examples of Linear Maps 52
Algebraic Operations on ℒ(𝑉, 𝑊) 55
Exercises 3A 57
3B Null Spaces and Ranges 59
Null Space and Injectivity 59
Range and Surjectivity 61
Fundamental Theorem of Linear Maps 62
Exercises 3B 66
3C Matrices 69
Representing a Linear Map by a Matrix 69
Addition and Scalar Multiplication of Matrices 71
Matrix Multiplication 72
Exercises 3C 76
3D Invertibility and Isomorphic Vector Spaces 78
Invertible Linear Maps 78
Isomorphic Vector Spaces 82
Linear Maps Thought of as Matrix Multiplication 84
Exercises 3D 86
3E Products and Quotients of Vector Spaces 88
Products of Vector Spaces 88
Products and Direct Sums 89
Quotients of Vector Spaces 90
Exercises 3E 95
3F Duality 97
Dual Space and Dual Map 97
Null Space and Range of Dual of Linear Map 100
Matrix of Dual of Linear Map 105
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
viii Contents
Chapter 4
Polynomials 113
Uniqueness of Coefficients for Polynomials 116
Division Algorithm for Polynomials 116
Zeros of Polynomials 118
Factorization of Polynomials over 𝐂 119
Factorization of Polynomials over 𝐑 123
Exercises 4 125
Chapter 5
Eigenvalues and Eigenvectors 128
5A Invariant Subspaces 129
Eigenvalues 129
Polynomials Applied to Operators 133
Exercises 5A 135
5B Eigenvalues and the Minimal Polynomial 139
Existence of Eigenvalues 139
Minimal Polynomial 140
Exercises 5B 145
5C Upper-Triangular Matrices 148
Exercises 5C 155
5D Diagonalizable Operators 156
Exercises 5D 163
5E Commuting Operators 166
Exercises 5E 170
Chapter 6
Inner Product Spaces 172
6A Inner Products and Norms 173
Inner Products 173
Norms 177
Exercises 6A 182
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Contents ix
Chapter 7
Operators on Inner Product Spaces 217
7A Self-Adjoint and Normal Operators 218
Adjoints 218
Self-Adjoint Operators 223
Normal Operators 225
Exercises 7A 229
7B Spectral Theorem 232
Real Spectral Theorem 232
Complex Spectral Theorem 235
Exercises 7B 236
7C Positive Operators 239
Exercises 7C 243
7D Isometries, Unitary Operators, and 𝑄𝑅 Factorization 245
Isometries 245
Unitary Operators 247
𝑄𝑅 Factorization 250
Exercises 7D 253
7E Singular Value Decomposition 255
Singular Values 255
SVD for Linear Maps and for Matrices 258
Exercises 7E 263
7F Consequences of the Singular Value Decomposition 265
Norms of Linear Maps 265
Approximation by Linear Maps with Lower-Dimensional Range 268
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
x Contents
Chapter 8
Operators on Complex Vector Spaces 275
8A Generalized Eigenvectors and Nilpotent Operators 276
Null Spaces of Powers of an Operator 276
Generalized Eigenvectors 278
Nilpotent Operators 283
Exercises 8A 285
8B Decomposition of an Operator 288
Description of Operators on Complex Vector Spaces 288
Multiplicity of an Eigenvalue 289
Block Diagonal Matrices 291
Square Roots of Operators 293
Exercises 8B 295
8C The Characteristic Polynomial 297
Cayley–Hamilton Theorem 297
Exercises 8C 298
8D Jordan Form 301
Exercises 8D 304
Chapter 9
Operators on Real Vector Spaces 306
9A Complexification 307
Complexification of a Vector Space 307
Complexification of an Operator 308
Minimal Polynomial of Complexification 310
Eigenvalues of Complexification 310
Characteristic Polynomial of Complexification 313
Exercises 9A 315
9B Operators on Real Inner Product Spaces 318
Normal Operators on Real Inner Product Spaces 318
Unitary Operators on Real Inner Product Spaces 322
Exercises 9B 324
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Contents xi
Chapter 10
Trace and Determinant 326
10A Trace 327
Change of Basis 327
Trace: A Connection Between Operators and Matrices 330
Exercises 10A 335
10B Determinant 337
Determinant of an Operator 337
Determinant of a Matrix 339
Sign of Determinant 349
Volume Via Determinants 351
Exercises 10B 359
Chapter 11
Multilinear Algebra and Tensors 362
11A Multilinear Algebra 363
11B Tensors 364
Index 367
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Major changes for the fourth edition:
• Singular value decomposition now done for linear maps from an inner product
space to another (possibly different) inner product space, rather than only deal-
ing with linear operators from an inner product space to itself (see Section 7E).
• Polar decomposition now proved from singular value decomposition, rather
than in the opposite order; this has led to cleaner proofs of both the singu-
lar value decomposition (see Section 7E) and the polar decomposition (see
Section 7F).
• New subsection on norms of linear maps on finite-dimensional inner product
spaces, using the singular value decomposition to avoid even mentioning
supremum in the definition of the norm of a linear map (see Section 7F).
• New subsection on approximation by linear maps with lower-dimensional
range (see Section 7F).
Chapter 5
Eigenvalues and Eigenvectors
Linear maps from one vector space to another vector space were the objects of
study in Chapter 3. Now we begin our investigation of operators, which are linear
maps from a vector space to itself. Their study constitutes the most important
part of linear algebra.
To learn about an operator, we might try restricting it to a smaller subspace.
Asking for that restriction to be an operator will lead us to the notion of invariant
subspaces. Each one-dimensional invariant subspace arises from a vector that
the operator maps into a scalar multiple of the vector. This path will lead us to
eigenvectors and eigenvalues.
We will then prove one of the most important results in linear algebra: every
operator on a finite-dimensional, nonzero, complex vector space has an eigenvalue.
This result will allow us to show that for each operator on a finite-dimensional
complex vector space, there is a basis of the vector space with respect to which
the matrix of the operator has at least almost half its entries equal to 0.
Note that in this chapter we assume that 𝑉 is finite-dimensional.
• 𝐅 denotes 𝐑 or 𝐂 .
• 𝑉 and 𝑊 denote vector spaces over 𝐅 , and 𝑉 is finite-dimensional.
Hans-Peter Postel CC BY
5A Invariant Subspaces
Eigenvalues
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
130 Chapter 5 Eigenvalues and Eigenvectors
𝑈 = {𝜆𝑣 ∶ 𝜆 ∈ 𝐅} = span(𝑣).
𝑇𝑣 = 𝜆𝑣.
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5A Invariant Subspaces 131
Proof Conditions (a) and (b) are equivalent because the equation 𝑇𝑣 = 𝜆𝑣
is equivalent to the equation (𝑇 − 𝜆𝐼)𝑣 = 0. Conditions (b), (c), and (d) are
equivalent by 3.53.
Substituting the value for 𝑤 given by the second equation into the first equation
gives
−𝑧 = 𝜆2 𝑧.
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
132 Chapter 5 Eigenvalues and Eigenvectors
Now 𝑧 cannot equal 0 [otherwise 5.10 implies that 𝑤 = 0; we are looking for
solutions to 5.10 where (𝑤, 𝑧) is not the 0 vector], so the equation above leads
to the equation −1 = 𝜆2. The solutions to this equation are 𝜆 = 𝑖 and 𝜆 = −𝑖.
You should be able to verify easily that 𝑖 and −𝑖 are eigenvalues of 𝑇. Indeed,
the eigenvectors corresponding to the eigenvalue 𝑖 are the vectors of the form
(𝑤, −𝑤𝑖), with 𝑤 ∈ 𝐂 and 𝑤 ≠ 0, and the eigenvectors corresponding to the
eigenvalue −𝑖 are the vectors of the form (𝑤, 𝑤𝑖), with 𝑤 ∈ 𝐂 and 𝑤 ≠ 0.
5.14 operator cannot have more eigenvalues than dimension of vector space
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5A Invariant Subspaces 133
5.15 notation: 𝑇 𝑚
𝑝(𝑧) = 𝑎0 + 𝑎1 𝑧 + 𝑎2 𝑧2 + ⋯ + 𝑎𝑚 𝑧𝑚
𝑝(𝑇) = 𝑎0 𝐼 + 𝑎1 𝑇 + 𝑎2 𝑇 2 + ⋯ + 𝑎𝑚 𝑇 𝑚.
This is a new use of the symbol 𝑝 because we are applying 𝑝 to operators, not
just elements of 𝐅 . The idea here is that to evaluate 𝑝(𝑇), we simply replace 𝑧 with
𝑇 in the expression defining 𝑝. Note that the constant term 𝑎0 in 𝑝(𝑧) becomes the
operator 𝑎0 𝐼 (which is a reasonable choice because 𝑎0 = 𝑎0 𝑧0 and thus we should
replace 𝑎0 with 𝑎0 𝑇 0 , which equals 𝑎0 𝐼).
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
134 Chapter 5 Eigenvalues and Eigenvectors
If we fix an operator 𝑇 ∈ ℒ(𝑊), then the function from 𝒫(𝐅) to ℒ(𝑉) given
by 𝑝 ↦ 𝑝(𝑇) is linear, as you should verify.
(𝑝𝑞)(𝑧) = 𝑝(𝑧)𝑞(𝑧)
for 𝑧 ∈ 𝐅 .
Part (b) of the next result states that the order does not matter when taking
products of polynomials of a single operator.
Proof
(a) Suppose 𝑝(𝑧) = ∑𝑚 𝑎 𝑧𝑗 and 𝑞(𝑧) = ∑𝑛𝑘 = 0 𝑏𝑘 𝑧𝑘 for 𝑧 ∈ 𝐅 . Then
𝑗=0 𝑗
𝑚 𝑛
(𝑝𝑞)(𝑧) = ∑ ∑ 𝑎𝑗 𝑏𝑘 𝑧𝑗 + 𝑘 .
𝑗 = 0 𝑘=0
Thus
𝑚 𝑛
(𝑝𝑞)(𝑇) = ∑ ∑ 𝑎𝑗 𝑏𝑘 𝑇 𝑗 + 𝑘
𝑗 = 0 𝑘=0
𝑚 𝑛
= ( ∑ 𝑎𝑗 𝑇 𝑗 )( ∑ 𝑏𝑘 𝑇 𝑘 )
𝑗=0 𝑘=0
= 𝑝(𝑇)𝑞(𝑇).
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5A Invariant Subspaces 135
We observed earlier that if 𝑇 ∈ ℒ(𝑉), then null 𝑇 and range 𝑇 are invariant
under 𝑇 [see 5.4]. Now we show that the null space and the range of each
polynomial of 𝑇 are also invariant under 𝑇.
Suppose 𝑇 ∈ ℒ(𝑉) and 𝑝 ∈ 𝒫(𝐅). Then null 𝑝(𝑇) and range 𝑝(𝑇) are
invariant under 𝑇.
𝑇𝑢 = 𝑇(𝑝(𝑇)𝑣) = 𝑝(𝑇)(𝑇𝑣).
Exercises 5A
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
136 Chapter 5 Eigenvalues and Eigenvectors
8 Define 𝑇 ∈ ℒ(𝐅3 ) by
(𝑇𝑝)(𝑥) = 𝑥𝑝′(𝑥)
where ℳ(𝑇)𝑗, 𝑘 denotes the entry in row 𝑗, column 𝑘 of the matrix of 𝑇 with
respect to the basis 𝑣1 , …, 𝑣𝑛 .
For a different bound on the absolute value of an eigenvalue of an operator 𝑇,
see Exercise 18 in Section 6A.
𝑇(𝑧1 , 𝑧2 , … ) = (0, 𝑧1 , 𝑧2 , … )
has no eigenvalues.
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5A Invariant Subspaces 137
20 Suppose 𝑇 ∈ ℒ(𝑊) and there exist nonzero vectors 𝑣 and 𝑤 in 𝑊 such that
𝑇𝑣 = 3𝑤 and 𝑇𝑤 = 3𝑣.
Prove that 3 or −3 is an eigenvalue of 𝑇.
21 Suppose 𝑆, 𝑇 ∈ ℒ(𝑉). Prove that 𝑆𝑇 and 𝑇𝑆 have the same eigenvalues.
22 Suppose 𝑅, 𝑆, 𝑇 ∈ ℒ(𝑉). Prove or give counterexample: 𝑅𝑆𝑇 and 𝑅𝑇𝑆
have the same eigenvalues.
23 Suppose 𝐴 is an 𝑛-by-𝑛 matrix with entries in 𝐅 . Define 𝑇 ∈ ℒ(𝐅𝑛 ) by
𝑇𝑥 = 𝐴𝑥, where elements of 𝐅𝑛 are thought of as 𝑛-by-1 column vectors.
(a) Suppose the sum of the entries in each row of 𝐴 equals 1. Prove that 1
is an eigenvalue of 𝑇.
(b) Suppose the sum of the entries in each column of 𝐴 equals 1. Prove that
1 is an eigenvalue of 𝑇.
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
138 Chapter 5 Eigenvalues and Eigenvectors
for 𝑣 ∈ 𝑉.
(a) Show that the definition of 𝑇/𝑈 makes sense (which requires using the
condition that 𝑈 is invariant under 𝑇) and show that 𝑇/𝑈 is an operator
on 𝑉/𝑈.
(b) Show that 𝑇/(range 𝑇) = 0.
(c) Show that 𝑇/(null 𝑇) is injective if and only if null 𝑇 ∩ range 𝑇 = {0}.
(d) Show that each eigenvalue of 𝑇/𝑈 is an eigenvalue of 𝑇.
37 Define 𝑇 ∈ ℒ(𝐅𝑛 ) by
𝑇(𝑥1 , 𝑥2 , 𝑥3 , …, 𝑥𝑛 ) = (𝑥1 , 2𝑥2 , 3𝑥3 , …, 𝑛𝑥𝑛 ).
(a) Find all eigenvalues and eigenvectors of 𝑇.
(b) Find all invariant subspaces of 𝑇.
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5B Eigenvalues and the Minimal Polynomial 139
Proof Suppose 𝑉 is a complex vector space with dimension 𝑛 > 0 and 𝑇 ∈ ℒ(𝑉).
Choose 𝑣 ∈ 𝑉 with 𝑣 ≠ 0. Then
𝑣, 𝑇𝑣, 𝑇 2 𝑣, …, 𝑇 𝑛 𝑣
is not linearly independent, because 𝑉 has dimension 𝑛 and this list has length
𝑛 + 1. Hence some linear combination (with not all the coefficients equal to 0) of
the vectors above equals 0. Thus there there exists a nonconstant polynomial 𝑝 of
smallest degree such that
𝑝(𝑇)𝑣 = 0.
By the first version of the fundamental theorem of algebra (see 4.12), there
exists 𝜆 ∈ 𝐂 such that 𝑝(𝜆) = 0. By 4.10, there exists a polynomial 𝑞 ∈ 𝒫(𝐂)
such that
𝑝(𝑧) = (𝑧 − 𝜆)𝑞(𝑧)
for every 𝑧 ∈ 𝐂 . Thus 5.19 implies that
0 = 𝑝(𝑇)𝑣 = (𝑇 − 𝜆𝐼)(𝑞(𝑇)𝑣).
Because 𝑞 has smaller degree than 𝑝, we know that 𝑞(𝑇)𝑣 ≠ 0. Thus the equation
above implies that 𝜆 is an eigenvalue of 𝑇 with eigenvector 𝑞(𝑇)𝑣.
The proof above makes crucial use of the fundamental theorem of algebra.
The comment following Exercise 10 helps explain why the fundamental theorem
of algebra is so tightly connected to the result above.
The hypothesis in the result above that 𝐅 = 𝐂 cannot be replaced with the
hypothesis that 𝐅 = 𝐑 , as shown by Example 5.9. The next example shows that
the finite-dimensional hypothesis in the result above also cannot be deleted.
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
140 Chapter 5 Eigenvalues and Eigenvectors
Minimal Polynomial
In this subsection we introduce an important polynomial associated with each
operator. We begin with the following definition.
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5B Eigenvalues and the Minimal Polynomial ≈ 100√2
The paragraph above shows that there is a monic polynomial with degree at
most dim 𝑉 that when applied to 𝑇 gives the 0 operator. Thus there is a monic
polynomial with smallest degree with this property, completing the existence part
of this result.
Let 𝑝 ∈ 𝒫𝑛 (𝐅) be a monic polynomial of smallest degree such that 𝑝(𝑇) = 0.
To prove the uniqueness part of the result, suppose 𝑟 ∈ 𝒫(𝐅) is a monic poly-
nomial with the same degree as 𝑝 and 𝑟(𝑇) = 0. Then (𝑝 − 𝑟)(𝑇) = 0 and also
deg(𝑝 − 𝑟) < deg 𝑝. If 𝑝 − 𝑟 were not equal to 0, then we could divide 𝑝 − 𝑟 by
the coefficient of the highest order term in 𝑝 − 𝑟 to get a monic polynomial (with
smaller degree than 𝑝) that when applied to 𝑇 gives the 0 operator. Thus we
conclude that 𝑝 − 𝑟 = 0, and hence 𝑟 = 𝑝, as desired.
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
142 Chapter 5 Eigenvalues and Eigenvectors
Thus 3𝑒1 − 6𝑇𝑒1 = −𝑇 5 𝑒1 . The list 𝑒1 , 𝑇𝑒1 , 𝑇 2 𝑒1 , 𝑇 3 𝑒1 , 𝑇 4 𝑒1 , which equals the list
𝑒1 , 𝑒2 , 𝑒3 , 𝑒4 , 𝑒5 , is linearly independent, so no other linear combination of this list
equals −𝑇 5 𝑒1 . Hence the minimal polynomial of 𝑇 is 3 − 6𝑧 + 𝑧5 .
Suppose 𝑇 ∈ ℒ(𝑉).
(a) The zeros of the minimal polynomial of 𝑇 are the eigenvalues of 𝑇.
(b) If 𝑉 is a complex vector space, then the minimal polynomial of 𝑇 has the
form
(𝑧 − 𝜆1 )⋯(𝑧 − 𝜆𝑚 ),
where 𝜆1 , …, 𝜆𝑚 is a list of all the eigenvalues of 𝑇, possibly with
repetitions.
𝑝(𝑧) = (𝑧 − 𝜆)𝑞(𝑧),
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5B Eigenvalues and the Minimal Polynomial 143
A nonzero polynomial has at most as many distinct zeros as its degree (see
4.11). Thus part (a) of the previous result, along with the result that the minimal
polynomial of an operator on 𝑉 has degree at most dim 𝑉, gives an alternative
proof of 5.14, which states that an operator on 𝑉 has at most dim 𝑉 distinct
eigenvalues.
Every monic polynomial is the minimal polynomial of some operator, as
shown by Exercise 10, which generalizes Example 5.28. Thus 5.29(a) shows that
finding exact expressions for the eigenvalues of an operator is equivalent to the
problem of finding exact expressions for the zeros of a polynomial (and thus is
not possible for some operators).
The matrix of 𝑇 with respect to the standard basis of 𝐂5 is the 5-by-5 matrix in
Example 5.28. As we showed that example, the minimal polynomial of 𝑇 is the
polynomial
3 − 6𝑧 + 𝑧5 .
No zero of the polynomial above can be expressed using rational numbers,
roots of rational numbers, and the usual rules of arithmetic (a proof of this would
take us considerably beyond linear algebra). Because the zeros of the polynomial
above are the eigenvalues of 𝑇 [by 5.29(a)], we cannot find an exact expression
for any eigenvalue of 𝑇 in any familiar form.
Numeric techniques, which we will not discuss here, show that the zeros of the
polynomial above, and thus the eigenvalues of 𝑇, are approximately the following
five complex numbers:
Note that the two nonreal zeros of this polynomial are complex conjugates of
each other, as we expect for a polynomial with real coefficients (see 4.14).
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
144 Chapter 5 Eigenvalues and Eigenvectors
The next result completely characterizes the polynomials that when applied to
an operator give the 0 operator.
as desired.
To prove the other direction, now suppose 𝑞(𝑇) = 0. By the division algorithm
for polynomials (4.7), there exist polynomials 𝑠, 𝑟 ∈ 𝒫(𝐅) such that
5.32 𝑞 = 𝑝𝑠 + 𝑟
The next result shows that the constant term of the minimal polynomial of an
operator determines whether the operator is invertible.
An operator 𝑇 ∈ ℒ(𝑉) is not invertible if and only if the constant term of the
minimal polynomial of 𝑇 is 0.
where the first equivalence holds by 5.7, the second equivalence holds by 5.29(a),
and the last equivalence holds because the constant term of 𝑝 equals 𝑝(0).
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5B Eigenvalues and the Minimal Polynomial 145
Exercises 5B
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
146 Chapter 5 Eigenvalues and Eigenvectors
𝑎0 + 𝑎1 𝑧 + … + 𝑎𝑛 − 1 𝑧𝑛 − 1 + 𝑧𝑛 .
The matrix above is called the companion matrix of the polynomial above.
This exercise shows that every monic polynomial is the minimal polynomial
of some operator. Hence a formula or an algorithm that could produce
exact eigenvalues for each operator on each 𝐅𝑛 could then produce exact
zeros for each polynomial [by 5.29(a)]. Thus there is no such formula or
algorithm. However, the efficient numeric methods for obtaining very good
approximations for the eigenvalues of an operator are often used to find
very good approximations for the zeros of a polynomial by considering the
companion matrix of the polynomial.
11 Suppose 𝑉 is a complex vector space with dim 𝑉 > 0 and 𝑇 ∈ ℒ(𝑉). Define
a function 𝑓 ∶ 𝐂 → 𝐑 by
ℰ = {𝑞(𝑇) ∶ 𝑞 ∈ 𝒫(𝐅)},
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5B Eigenvalues and the Minimal Polynomial 147
span(𝑣, 𝑇𝑣, …, 𝑇 𝑛 − 1 𝑣)
is invariant under 𝑇.
17 Suppose 𝑇 ∈ ℒ(𝑉) and 𝑈 is a subspace of 𝑉 that is invariant under 𝑇. Prove
that the minimal polynomial of 𝑇 is a polynomial multiple of the minimal
polynomial of 𝑇|𝑈 .
18 Suppose 𝑇 ∈ ℒ(𝐅4 ) is such that the eigenvalues of 𝑇 are 3, 5, 8. Prove that
(𝑇 − 3𝐼)2 (𝑇 − 5𝐼)2 (𝑇 − 8𝐼)2 = 0.
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
148 Chapter 5 Eigenvalues and Eigenvectors
5C Upper-Triangular Matrices
In Chapter 3 we discussed the matrix of a linear map from one vector space to
another vector space. That matrix depended on a choice of a basis of each of the
two vector spaces. Now that we are studying operators, which map a vector space
to itself, the emphasis is on using only one basis.
If the basis is not clear from the context, then the notation ℳ(𝑇, (𝑣1 , …, 𝑣𝑛 ))
is used.
Note that the matrices of operators are square arrays, rather than the more
general rectangular arrays that we considered earlier for linear maps.
If 𝑇 is an operator on 𝐅𝑛 and no ba-
The 𝑘 th column of the matrix ℳ(𝑇) is
sis is specified, assume that the basis in formed from the coefficients used to
question is the standard one (where the write 𝑇𝑣 as a linear combination of
𝑘 th basis vector is 1 in the 𝑘 th slot and 0
𝑘
the basis 𝑣1 , …, 𝑣𝑛 .
in all the other slots). You can then think
of the 𝑘 th column of ℳ(𝑇) as 𝑇 applied to the 𝑘 th basis vector, where we identify
𝑛-by-1 column vectors with elements of 𝐅𝑛 .
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5C Upper-Triangular Matrices 149
𝜆
⎛
⎜ ⎞
⎟
⎜
⎜ 0 ∗ ⎟
⎟
⎜
⎜ ⎟
⎟ ;
⎜ ⋮ ⎟
⎝ 0 ⎠
here the ∗ denotes the entries in all the columns other than the first column.
To prove this, let 𝜆 be an eigenvalue of 𝑇 (one exists by 5.21) and let 𝑣 be a
corresponding eigenvector. Extend 𝑣 to a basis of 𝑉. Then the matrix of 𝑇 with
respect to this basis has the form above.
Soon we will see that we can choose a basis of 𝑉 with respect to which the
matrix of 𝑇 has even more 0’s.
The diagonal of a square matrix consists of the entries on the line from the
upper left corner to the bottom right corner.
A square matrix is called upper triangular if all the entries below the diagonal
equal 0.
⎛
𝜆1 ∗ ⎞
⎜ ⎟
⎜
⎜ ⋱ ⎟;
⎟
⎝ 0 𝜆𝑛 ⎠
the 0 in the matrix above indicates that We often use ∗ to denote matrix entries
all entries below the diagonal in this that we do not know or that are irrele-
𝑛-by-𝑛 matrix equal 0. Upper-triangular vant to the questions being discussed.
matrices can be considered reasonably
simple—for 𝑛 large, at least almost half the entries in an 𝑛-by-𝑛 upper-triangular
matrix are 0.
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
150 Chapter 5 Eigenvalues and Eigenvectors
Proof First suppose (a) holds. To prove that (b) holds, suppose 𝑘 ∈ {1, …, 𝑛}. If
𝑗 ∈ {1, …, 𝑛}, then
𝑇𝑣𝑗 ∈ span(𝑣1 , …, 𝑣𝑗 )
because the matrix of 𝑇 with respect to 𝑣1 , …, 𝑣𝑛 is upper triangular. Because
span(𝑣1 , …𝑣𝑗 ) ⊂ span(𝑣1 , …𝑣𝑘 ) if 𝑗 ≤ 𝑘, we see that
𝑇𝑣𝑗 ∈ span(𝑣1 , …𝑣𝑘 )
for each 𝑗 ∈ {1, …, 𝑘}. Thus span(𝑣1 , …, 𝑣𝑘 ) is invariant under 𝑇, completing the
proof that (a) implies (b).
Now suppose (b) holds, so span(𝑣1 , …, 𝑣𝑘 ) is invariant under 𝑇 for each
𝑘 = 1, …, 𝑛. In particular, 𝑇𝑣𝑘 ∈ span(𝑣1 , …, 𝑣𝑘 ) for each 𝑘 = 1, …, 𝑛. Thus
(b) implies (c).
Now suppose (c) holds, so 𝑇𝑣𝑘 ∈ span(𝑣1 , …, 𝑣𝑘 ) for each 𝑘 = 1, …, 𝑛. This
means that when writing each 𝑇𝑣𝑘 as a linear combination of the basis vectors
𝑣1 , …, 𝑣𝑛 , we need to use only the vectors 𝑣1 , …, 𝑣𝑘 . Hence all entries under the
diagonal of ℳ(𝑇) are 0. In other words, ℳ(𝑇) is an upper-triangular matrix,
completing the proof that (c) implies (a).
We have shown that (a) ⟹ (b) ⟹ (c) ⟹ (a), which shows that (a), (b),
and (c) are equivalent.
The next result tells us that if 𝑇 ∈ ℒ(𝑉) and with respect to some basis of 𝑉
we have
𝜆
⎛ 1
∗ ⎞
ℳ(𝑇) = ⎜ ⎜
⎜ ⋱ ⎟
⎟,
⎟
⎝ 0 𝜆𝑛 ⎠
then 𝑇 satisfies a simple equation depending on 𝜆1 , …, 𝜆𝑛 .
Suppose 𝑇 ∈ ℒ(𝑉) and 𝑉 has a basis with respect to which 𝑇 has an upper-
triangular matrix with diagonal entries 𝜆1 , …, 𝜆𝑛 . Then
(𝑇 − 𝜆1 𝐼)⋯(𝑇 − 𝜆𝑛 𝐼) = 0.
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5C Upper-Triangular Matrices 151
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
152 Chapter 5 Eigenvalues and Eigenvectors
The next example illustrates the result following the example: an operator has
an upper-triangular matrix with respect to some basis if and only if the minimal
polynomial of the operator is the product of degree 1 polynomials.
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5C Upper-Triangular Matrices 153
Proof First suppose 𝑇 has an upper-triangular matrix with respect to some basis
of 𝑉. Let 𝛼1 , …, 𝛼𝑛 denote the diagonal entries of that matrix. Define a polynomial
𝑞 ∈ 𝒫(𝐅) by
𝑞(𝑧) = (𝑧 − 𝛼1 )⋯(𝑧 − 𝛼𝑛 ).
Then 𝑞(𝑇) = 0, by 5.39. Hence 𝑞 is a polynomial multiple of the minimal polyno-
mial of 𝑇, by 5.31. Thus the minimal polynomial of 𝑇 equals (𝑧 − 𝜆1 )⋯(𝑧 − 𝜆𝑚 )
for some 𝜆1 , …, 𝜆𝑚 ∈ 𝐅 with {𝜆1 , …, 𝜆𝑚 } ⊂ {𝛼1 , …, 𝛼𝑛 }.
To prove the implication in the other direction, now suppose the minimal
polynomial of 𝑇 equals (𝑧 − 𝜆1 )⋯(𝑧 − 𝜆𝑚 ) for some 𝜆1 , …, 𝜆𝑚 ∈ 𝐅 . We will use
induction on 𝑚. To get started, if 𝑚 = 1 then 𝑧 − 𝜆1 is the minimal polynomial of
𝑇, which implies that 𝑇 = 𝜆1 𝐼, which implies that the matrix of 𝑇 (with respect
to any basis of 𝑉) is upper-triangular.
Now suppose 𝑚 > 1 and the desired result holds for 𝑚 − 1. Let
𝑈 = range(𝑇 − 𝜆𝑚 𝐼).
From 5.44 and 5.45, we conclude (using 5.38) that 𝑇 has an upper-triangular
matrix with respect to the basis 𝑢1 , …, 𝑢𝑀 , 𝑣1 , …, 𝑣𝑁 of 𝑉, as desired.
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
154 Chapter 5 Eigenvalues and Eigenvectors
The set of numbers {𝜆1 , …, 𝜆𝑚 } from the previous result equals the set eigen-
values of 𝑇 (because the set of zeros of the minimal polynomial of 𝑇 equals the
set eigenvalues of 𝑇, by 5.29), although the list 𝜆1 , …, 𝜆𝑚 in the previous result
may contain repetitions.
In Chapter 8 we will improve even the wonderful result of part (a) below; see
8.36 and 8.54.
Suppose 𝑇 ∈ ℒ(𝑉).
(a) If 𝐅 = 𝐂 , then 𝑇 has an upper-triangular matrix with respect to some
basis of 𝑉.
(b) If 𝐅 = 𝐑 , then 𝑇 has an upper-triangular matrix with respect to some basis
of 𝑉 if and only if every zero of the minimal polynomial of 𝑇, thought of
as a polynomial with complex coefficients, is real.
Proof The desired result follows immediately from 5.43 and the second version
of the fundamental theorem of algebra (see 4.13).
For an extension of the result above See Exercise 18 in Section 9A for addi-
to two operators 𝑆 and 𝑇 such that tional necessary and sufficient condi-
tions for an operator on a real vector
𝑆𝑇 = 𝑇𝑆,
space to have an upper-triangular ma-
see 5.74. Also, for an extension to more trix with respect to some basis.
than two operators, see Exercise 8(b) in
Section 5E.
Caution: If an operator 𝑇 ∈ ℒ(𝑉) has a upper-triangular matrix with respect
to some basis 𝑣1 , …, 𝑣𝑛 of 𝑉, then the eigenvalues of 𝑇 are exactly the entries on
the diagonal of ℳ(𝑇), as shown by 5.40, and furthermore 𝑣1 is an eigenvector of
𝑇. However, 𝑣2 , …, 𝑣𝑛 need not be eigenvectors of 𝑇. Indeed, a basis vector 𝑣𝑘 is
an eigenvector of 𝑇 if and only if all the entries in the 𝑘 th column of the matrix of
𝑇 are 0, except possibly the 𝑘 th entry.
You may recall from a previous The row echelon form of the matrix
course that every matrix of numbers can of an operator provides no informa-
be changed to a matrix in what is called tion about the eigenvalues of the oper-
row echelon form. If starting with a ator. In contrast, an upper-triangular
square matrix, the matrix in row echelon matrix with respect to some basis pro-
form will be an upper-triangular matrix. vides complete information about the
Do not confuse this upper-triangular ma- eigenvalues of the operator. However,
trix with the upper-triangular matrix of there is no method for computing ex-
an operator with respect to some basis actly such an upper-triangular matrix,
whose existence is proclaimed by 5.46(a) even though 5.46(a) guarantees its ex-
if 𝐅 = 𝐂 —there is no connection be- istence if 𝐅 = 𝐂 .
tween these upper-triangular matrices.
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5C Upper-Triangular Matrices 155
Exercises 5C
1 (a) Prove or give counterexample: If 𝑇 ∈ ℒ(𝑉) and 𝑇 has an upper-
triangular matrix with respect to some basis of 𝑉, then 𝑇 2 has an upper-
triangular matrix with respect to some basis of 𝑉.
(b) Prove or give counterexample: If 𝑇 ∈ ℒ(𝑉) and 𝑇 2 has an upper-
triangular matrix with respect to some basis of 𝑉, then 𝑇 has an upper-
triangular matrix with respect to some basis of 𝑉.
2 Suppose 𝐴 and 𝐵 are upper-triangular matrices of the same size. Show that
𝐴 + 𝐵 and 𝐴𝐵 are upper-triangular matrices.
The result in this exercise is used in the proof of 5.75.
8 Suppose that 𝑇 ∈ ℒ(𝑉) and there exists a nonzero vector 𝑣 ∈ 𝑉 such that
𝑇 2 𝑣 + 2𝑇𝑣 = −2𝑣.
(a) Prove that if 𝐅 = 𝐑 , then there does not exist a basis of 𝑉 with respect
to which 𝑇 has an upper-triangular matrix.
(b) Prove that if 𝐅 = 𝐂 and 𝐴 is an upper-triangular matrix that equals
the matrix of 𝑇 with respect to some basis of 𝑉, then −1 + 𝑖 or −1 − 𝑖
appears on the diagonal of 𝐴.
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
156 Chapter 5 Eigenvalues and Eigenvectors
5D Diagonalizable Operators
because 𝑇(1, 4) = (69, 276) = 69(1, 4) and 𝑇(7, 5) = (322, 230) = 46(7, 5).
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5D Diagonalizable Operators 157
For 𝜆 ∈ 𝐅 , we will find it convenient to have a name and a notation for the set
of vectors that an operator 𝑇 maps to 𝜆 times the vector.
𝐸(𝜆1 , 𝑇) + ⋯ + 𝐸(𝜆𝑚 , 𝑇)
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
158 Chapter 5 Eigenvalues and Eigenvectors
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5D Diagonalizable Operators 159
Thus conditions (b), (c), and (d) of 5.54 are easily seen to fail (of course,
because these conditions are equivalent, it is sufficient to check that only one of
them fails). Thus condition (a) of 5.54 also fails, and hence 𝑇 is not diagonalizable,
regardless of whether 𝐅 = 𝐑 or 𝐅 = 𝐂 .
The next result shows that if an operator has as many distinct eigenvalues as
the dimension of its domain, then the operator is diagonalizable.
In later chapters we will find additional conditions that imply that certain
operators are diagonalizable. For example, see the real spectral theorem (7.29)
and the complex spectral theorem (7.31).
The result above gives a sufficient condition for an operator to be diagano-
lizable. However, this condition is not necessary. For example, the operator 𝑇
on 𝐅3 defined by 𝑇(𝑥, 𝑦, 𝑧) = (6𝑥, 6𝑦, 7𝑧) has only 2 eigenvalues (6 and 7) and
dim 𝐅3 = 3, but 𝑇 is diagonalizable (by the standard basis of 𝐅3 ).
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
160 Chapter 5 Eigenvalues and Eigenvectors
The next example illustrates the im- For a spectacular application of these
portance of diagonalization, which can techniques, see Exercise 16, which
be used to compute high powers of an shows how to use diagonalization to
operator, taking advantage of the equa- find an exact formula for the 𝑛th term
tion 𝑇 𝑘 𝑣 = 𝜆𝑘 𝑣 if 𝑣 is an eigenvector of of the Fibonacci sequence.
𝑇 with eigenvalue 𝜆.
𝑇(𝑥, 𝑦, 𝑧) = 𝜆(𝑥, 𝑦, 𝑧)
for 𝜆 = 2, then for 𝜆 = 5, and then for 𝜆 = 8. These simple equations are easy
to solve: for 𝜆 = 2 we have the eigenvector (1, 0, 0); for 𝜆 = 5 we have the
eigenvector (1, 3, 0); for 𝜆 = 8 we have the eigenvector (1, 6, 6).
Thus (1, 0, 0), (1, 3, 0), (1, 6, 6) is a basis of 𝐅3 consisting of eigenvectors of 𝑇,
and with respect to this basis the matrix of 𝑇 is the diagonal matrix
2 0 0
⎛
⎜ ⎞
⎟
⎜
⎜ 0 5 0 ⎟
⎟.
⎝ 0 0 8 ⎠
To compute 𝑇 100 (0, 0, 1), for example, write (0, 0, 1) as a linear combination
of our basis of eigenvectors:
𝑇 100 (0, 0, 1) = 61 (𝑇 100 (1, 0, 0)) − 13 (𝑇 100 (1, 3, 0)) + 16 (𝑇 100 (1, 6, 6))
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5D Diagonalizable Operators 161
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
162 Chapter 5 Eigenvalues and Eigenvectors
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5D Diagonalizable Operators 163
Exercises 5D
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
164 Chapter 5 Eigenvalues and Eigenvectors
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5D Diagonalizable Operators 165
𝐹0 = 0, 𝐹1 = 1, and 𝐹𝑛 = 𝐹𝑛 − 2 + 𝐹𝑛 − 1 for 𝑛 ≥ 2.
Each 𝐹𝑛 is a nonnegative integer, even though the right side of the formula
in part (d) does not look like an integer.
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
166 Chapter 5 Eigenvalues and Eigenvectors
5E Commuting Operators
For example, if 𝑇 is an operator and 𝑝, 𝑞 ∈ 𝒫(𝐅), then 𝑝(𝑇) and 𝑞(𝑇) commute
[see 5.19(b)].
As another example, if 𝐼 is the identity operator on 𝑊, then 𝐼 commutes with
every operator on 𝑊.
5.67 𝑝= ∑ 𝑎𝑗, 𝑘 𝑥𝑗 𝑦𝑘 ,
𝑗+𝑘 ≤ 𝑚
where the indices 𝑗 and 𝑘 are allowed to take on only nonnegative integer values
such that 𝑗 + 𝑘 ≤ 𝑚, each 𝑎𝑗, 𝑘 ∈ 𝐑 , and 𝑥𝑗 𝑦𝑘 denotes the function on 𝐑2 defined
by (𝑥, 𝑦) ↦ 𝑥𝑗 𝑦𝑘.
Define operators 𝐷𝑥 , 𝐷𝑦 ∈ 𝒫𝑚2 by
𝜕𝑝 𝜕𝑝
𝐷𝑥 𝑝 = = ∑ 𝑗𝑎 𝑥𝑗 − 1 𝑦𝑘 and 𝐷𝑦 𝑝 = = ∑ 𝑘𝑎 𝑥𝑗 𝑦𝑘 − 1 ,
𝜕𝑥 𝑗 + 𝑘 ≤ 𝑚 𝑗, 𝑘 𝜕𝑦 𝑗 + 𝑘 ≤ 𝑚 𝑗, 𝑘
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5E Commuting Operators 167
The next result shows that two operators commute if and only if their matrices
(with respect to the same basis) commute.
Proof We have
𝑆 and 𝑇 commute ⟺ 𝑆𝑇 = 𝑇𝑆
⟺ ℳ(𝑆𝑇) = ℳ(𝑇𝑆)
⟺ ℳ(𝑆)ℳ(𝑇) = ℳ(𝑇)ℳ(𝑆)
⟺ ℳ(𝑆) and ℳ(𝑇) commute,
as desired.
The next result shows that if two operators commute, then every eigenspace
for one operator is invariant under the other operator. This result, which we will
use several times, is one of the main reasons why a pair of commuting operators
behaves better than a pair of operators that does not commute.
Two diagonalizable operators on the same vector space have diagonal matrices
with respect to the same basis if and only if the two operators commute.
Proof First suppose 𝑆, 𝑇 ∈ ℒ(𝑉) have diagonal matrices with respect to the
same basis. The product of two diagonal matrices of the same size is the diagonal
matrix obtained by multiplying the corresponding elements of the two diagonals.
Thus any two diagonal matrices of the same size commute. Thus 𝑆 and 𝑇 commute,
by 5.68.
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
168 Chapter 5 Eigenvalues and Eigenvectors
To prove the implication in the other direction, now suppose that 𝑆, 𝑇 ∈ ℒ(𝑉)
are diagonalizale operators that commute. Let 𝜆1 , …, 𝜆𝑚 denote the distinct
eigenvalues of 𝑆. Because 𝑆 is diagonalizable, part (c) of 5.54 shows that
5.71 𝑉 = 𝐸(𝜆1 , 𝑆) ⊕ ⋯ ⊕ 𝐸(𝜆𝑚 , 𝑆).
For each 𝑘 = 1, …, 𝑚, the subspace 𝐸(𝜆𝑘 , 𝑆) is invariant under 𝑇 (by 5.69).
Because 𝑇 is diagonalizable, 5.64 implies that 𝑇|𝐸( 𝜆𝑘, 𝑆) is diagonalizable for
each 𝑘. Hence for each 𝑘 = 1, …, 𝑚, there is a basis of 𝐸(𝜆𝑘 , 𝑆) consisting of
eigenvectors of 𝑇. Putting these bases together gives a basis of 𝑉 (because of
5.71), with each vector in this basis being an eigenvector of both 𝑆 and 𝑇. Thus 𝑆
and 𝑇 both have diagonal matrices with respect to this basis, as desired.
See Exercise 2 for an extension of the result above to more than two operators.
Suppose 𝑉 is a finite-dimensional, nonzero, complex vector space. Then every
operator on 𝑉 has an eigenvector (see 5.21). The next result shows that if two
operators on 𝑉 commute, then there is a vector in 𝑉 that is an eigenvector for both
operators (but the two commuting operators might not have a common eigenvalue).
For an extension of the next result to more than two operators, see Exercise 8(a).
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5E Commuting Operators 169
The next result extends 5.46(a) [the existence of a basis that gives an upper-
triangular matrix] to two commuting operators.
Proof Let 𝑛 = dim 𝑉. We will use induction on 𝑛. The desired result holds if
𝑛 = 1 because all 1-by-1 matrices are upper triangular. Now suppose 𝑛 > 1 and
the desired result holds for all complex vector spaces whose dimension is 𝑛 − 1.
Let 𝑣1 be any common eigenvector of 𝑆 and 𝑇 (using 5.72). Hence 𝑆𝑣1 ∈
span(𝑣1 ) and 𝑇𝑣1 ∈ span(𝑣1 ). Let 𝑊 be a subspace of 𝑉 such that
𝑉 = span(𝑣1 ) ⊕ 𝑊;
see 2.33 for the existence of 𝑊. Define a linear map 𝑃 ∶ 𝑉 → 𝑊 by
𝑃(𝑎𝑣1 + 𝑤) = 𝑤
for 𝑎 ∈ 𝐂 and 𝑤 ∈ 𝑊. Define 𝑆̂, 𝑇̂ ∈ ℒ(𝑊) by
̂ = 𝑃(𝑇𝑤) and 𝑇𝑤
𝑆𝑤 ̂ = 𝑃(𝑇𝑤)
Exercise 8(b) extends the result above to more than two operators.
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
170 Chapter 5 Eigenvalues and Eigenvectors
Proof There is a basis of 𝑉 with respect to which both 𝑆 and 𝑇 have upper-
triangular matrices (by 5.74). With respect to that basis,
ℳ(𝑆 + 𝑇) = ℳ(𝑆) + ℳ(𝑇) and ℳ(𝑆𝑇) = ℳ(𝑆)ℳ(𝑇),
as stated in 3.33 and 3.40.
The definition of matrix addition shows that each entry on the diagonal of
ℳ(𝑆 + 𝑇) equals the sum of the corresponding entries on the diagonals of ℳ(𝑆)
and ℳ(𝑇). Similarly, because ℳ(𝑆) and ℳ(𝑇) are upper-triangular matrices,
the definition of matrix multiplication shows that each entry on the diagonal of
ℳ(𝑆𝑇) equals the product of the corresponding entries on the diagonals of ℳ(𝑆)
and ℳ(𝑇). Furthermore, ℳ(𝑆 + 𝑇) and ℳ(𝑆𝑇) are upper-triangular matrices
(see Exercise 2 in Section 5B).
Every entry on the diagonal of ℳ(𝑆) is an eigenvalue of 𝑆, and every entry
on the diagonal of ℳ(𝑇) is an eigenvalue of 𝑇 (by 5.40). Every eigenvalue
of 𝑆 + 𝑇 is on the diagonal of ℳ(𝑆 + 𝑇), and every eigenvalue of 𝑆𝑇 is on
the diagonal of ℳ(𝑆𝑇) (these assertions follow from by 5.40). Putting all this
together, we conclude that every eigenvalue of 𝑆 + 𝑇 is an eigenvalue of 𝑆 plus
an eigenvalue of 𝑇, and every eigenvalue of 𝑆𝑇 is an eigenvalue of 𝑆 times an
eigenvalue of 𝑇.
Exercises 5E
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler
Section 5E Commuting Operators 171
Linear Algebra Done Right, preliminary version of fourth edition, by Sheldon Axler