![Page 1: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/1.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Constructing and (some) classification of integermatrices with integer eigenvalues
Chris Towse* and Eric Campbell
Scripps CollegePomona College
January 6, 2017
![Page 2: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/2.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
The question
An example
Solve a linear system:x′(t) = Ax
Answer:
x = C1v1eλ1t + C2v2e
λ2t + · · ·+ Cnvneλnt
... or something similar.
![Page 3: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/3.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
The question
An example
Solve a linear system:x′(t) = Ax
Answer:
x = C1v1eλ1t + C2v2e
λ2t
+ · · ·+ Cnvneλnt
... or something similar.
![Page 4: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/4.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
The question
An example
Solve a linear system:x′(t) = Ax
Answer:
x = C1v1eλ1t + C2v2e
λ2t + · · ·+ Cnvneλnt
... or something similar.
![Page 5: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/5.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
The question
An example
An example:
A =
2 2 11 3 11 2 2
Eigenvalues are λ = 1, 1, 5.
So the characteristic polynomial is χA(x) = x3 − 7x2 + 11x − 5.
Reasonable?
![Page 6: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/6.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
The question
An example
An example:
A =
2 2 11 3 11 2 2
Eigenvalues are λ = 1, 1, 5.
So the characteristic polynomial is χA(x) = x3 − 7x2 + 11x − 5.
Reasonable?
![Page 7: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/7.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
The question
An example
An example:
A =
2 2 11 3 11 2 2
Eigenvalues are λ = 1, 1, 5.
So the characteristic polynomial is χA(x) = x3 − 7x2 + 11x − 5.
Reasonable?
![Page 8: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/8.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
The question
An example
An example:
A =
2 2 11 3 11 2 2
Eigenvalues are λ = 1, 1, 5.
So the characteristic polynomial is χA(x) = x3 − 7x2 + 11x − 5.
Reasonable?
![Page 9: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/9.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Set-up
GOAL: Construct a matrix A with relatively small integer entriesand with relatively small integer eigenvalues. (IMIE)
Such an A will be a good example.
Note: Martin and Wong, ”Almost all integer matrices have nointeger eigenvalues”
![Page 10: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/10.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Set-up
GOAL: Construct a matrix A with relatively small integer entriesand with relatively small integer eigenvalues. (IMIE)
Such an A will be a good example.Note: Martin and Wong, ”Almost all integer matrices have nointeger eigenvalues”
![Page 11: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/11.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Set-up
Recall/notation:
A = PDP−1
where
P =
| | · · · |~u1 ~u2 · · · ~un| | · · · |
has columns which form a basis of eigenvectors for A andD = 〈λ1, λ2, . . . , λn〉 is the diagonal matrix with correspondingeigenvalues on the diagonal.
![Page 12: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/12.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Set-up
Our example:
A =
2 2 11 3 11 2 2
D = 〈1, 1, 5〉
And in this case,
P =
−2 −1 11 0 10 1 1
A = PDP−1.
![Page 13: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/13.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
First approach
Failure
Idea: With the same P, try PDP−1 for D = 〈1, 3, 5〉.
Then
P
1 0 00 3 00 0 5
P−1 =
5/2 3 −1/21 3 1
1/2 1 7/2
![Page 14: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/14.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
First approach
Failure
Idea: With the same P, try PDP−1 for D = 〈1, 3, 5〉.Then
P
1 0 00 3 00 0 5
P−1 =
5/2 3 −1/21 3 1
1/2 1 7/2
![Page 15: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/15.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
First approach
Failure
The problem is that
P−1 = (1/4)
−1 2 −1−1 −2 31 2 1
=1
detPPadj
Or, really, that detP = 4.
![Page 16: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/16.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
First approach
Failure
The problem is that
P−1 = (1/4)
−1 2 −1−1 −2 31 2 1
=1
detPPadj
Or, really, that detP = 4.
![Page 17: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/17.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
First approach
Failure
Solutions:
1. Choose P with detP = ±1. (New problem!)
2. If k = detP, choose eigenvalues that are all multiples of k .(But probably avoid 0.)But actually ...
![Page 18: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/18.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
First approach
Failure
Solutions:
1. Choose P with detP = ±1. (New problem!)
2. If k = detP, choose eigenvalues that are all multiples of k .(But probably avoid 0.)But actually ...
![Page 19: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/19.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
First approach
Failure
Solutions:
1. Choose P with detP = ±1. (New problem!)
2. If k = detP, choose eigenvalues that are all multiples of k .(But probably avoid 0.)
But actually ...
![Page 20: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/20.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
First approach
Failure
Solutions:
1. Choose P with detP = ±1. (New problem!)
2. If k = detP, choose eigenvalues that are all multiples of k .(But probably avoid 0.)But actually ...
![Page 21: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/21.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
First approach
Improvement: modulo k
Proposition (Eigenvalues congruent to b modulo k)
Let P be an n × n invertible matrix with k = detP 6= 0. Supposeevery λi ≡ b mod k. (So all the λi ’s are congruent to eachother.) Then A = P〈λ1, . . . , λn〉P−1 is integral.
For example,
P〈1,−3, 5〉P−1 =
1 0 41 3 12 4 −1
which has χ(x) = x3 − 3x2 − 13x + 15 = (x − 1)(x + 3)(x − 5).
Good?
![Page 22: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/22.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
First approach
Improvement: modulo k
Proposition (Eigenvalues congruent to b modulo k)
Let P be an n × n invertible matrix with k = detP 6= 0. Supposeevery λi ≡ b mod k. (So all the λi ’s are congruent to eachother.) Then A = P〈λ1, . . . , λn〉P−1 is integral.
For example,
P〈1,−3, 5〉P−1 =
1 0 41 3 12 4 −1
which has χ(x) = x3 − 3x2 − 13x + 15 = (x − 1)(x + 3)(x − 5).
Good?
![Page 23: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/23.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
First approach
Improvement: modulo k
Proposition (Eigenvalues congruent to b modulo k)
Let P be an n × n invertible matrix with k = detP 6= 0. Supposeevery λi ≡ b mod k. (So all the λi ’s are congruent to eachother.) Then A = P〈λ1, . . . , λn〉P−1 is integral.
For example,
P〈1,−3, 5〉P−1 =
1 0 41 3 12 4 −1
which has χ(x) = x3 − 3x2 − 13x + 15 = (x − 1)(x + 3)(x − 5).
Good?
![Page 24: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/24.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
First approach
Improvement: modulo k
Proposition (Eigenvalues congruent to b modulo k)
Let P be an n × n invertible matrix with k = detP 6= 0. Supposeevery λi ≡ b mod k. (So all the λi ’s are congruent to eachother.) Then A = P〈λ1, . . . , λn〉P−1 is integral.
For example,
P〈1,−3, 5〉P−1 =
1 0 41 3 12 4 −1
which has χ(x) = x3 − 3x2 − 13x + 15 = (x − 1)(x + 3)(x − 5).
Good?
![Page 25: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/25.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
First approach
Improvement: modulo k
Eric Campbell (Pomona) created a web app
http://ericthewry.github.io/integer matrices/
Input: eigenvectors (really P)Select good eigenvaluesOutput: IMIE and its characteristic polynomial
![Page 26: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/26.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
First approach
Improvement: modulo k
Eric Campbell (Pomona) created a web app
http://ericthewry.github.io/integer matrices/
Input: eigenvectors (really P)Select good eigenvaluesOutput: IMIE and its characteristic polynomial
![Page 27: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/27.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
First approach
Improvement: modulo k
Eric Campbell (Pomona) created a web app
http://ericthewry.github.io/integer matrices/
Input: eigenvectors (really P)Select good eigenvaluesOutput: IMIE and its characteristic polynomial
![Page 28: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/28.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
![Page 29: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/29.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A useful property
Proposition
If X is invertible than XY and YX have the same characteristicpolynomials.
Idea [Renaud, 1983]: Take X ,Y integer matrices. If YX is upper(or lower) triangular then the eigenvalues are the diagonalelements, hence integers. So XY will be a good example.
If X and Y are n × n, we are imposing n(n − 1)/2 othogonalityconditions on the rows of Y and columns of X .
![Page 30: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/30.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A useful property
Proposition
If X is invertible than XY and YX have the same characteristicpolynomials.
Idea [Renaud, 1983]: Take X ,Y integer matrices. If YX is upper(or lower) triangular then the eigenvalues are the diagonalelements, hence integers. So XY will be a good example.
If X and Y are n × n, we are imposing n(n − 1)/2 othogonalityconditions on the rows of Y and columns of X .
![Page 31: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/31.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A useful property
Proposition
If X is invertible than XY and YX have the same characteristicpolynomials.
Idea [Renaud, 1983]: Take X ,Y integer matrices. If YX is upper(or lower) triangular then the eigenvalues are the diagonalelements, hence integers. So XY will be a good example.
If X and Y are n × n, we are imposing n(n − 1)/2 othogonalityconditions on the rows of Y and columns of X .
![Page 32: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/32.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A useful property
Let X =
(1 2−3 2
)and Y =
(1 23 1
).
Then XY =
(7 43 −4
)has the same eigenvalues as
YX =
(−5 ∗0 8
), namely λ = −5, 8.
Starting with X , the only condition on Y is that the second row ofY must be orthogonal to the first row of X .
![Page 33: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/33.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A useful property
Let X =
(1 2−3 2
)and Y =
(1 23 1
).
Then XY =
(7 43 −4
)has the same eigenvalues as
YX =
(−5 ∗0 8
), namely λ = −5, 8.
Starting with X , the only condition on Y is that the second row ofY must be orthogonal to the first row of X .
![Page 34: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/34.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A better version
Better fact:
Theorem (Folk Theorem)
Let U be an r × s matrix and V be an s × r matrix, where r ≤ s.Then χUV (x) = x s−rχVU(x).
Proof idea (due to Horn and Johnson):(UV 0V 0
)∼(
0 0V VU
)via
(Ir U0 Is
).
![Page 35: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/35.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A better version
Better fact:
Theorem (Folk Theorem)
Let U be an r × s matrix and V be an s × r matrix, where r ≤ s.Then χUV (x) = x s−rχVU(x).
Proof idea (due to Horn and Johnson):(UV 0V 0
)∼(
0 0V VU
)
via
(Ir U0 Is
).
![Page 36: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/36.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A better version
Better fact:
Theorem (Folk Theorem)
Let U be an r × s matrix and V be an s × r matrix, where r ≤ s.Then χUV (x) = x s−rχVU(x).
Proof idea (due to Horn and Johnson):(UV 0V 0
)∼(
0 0V VU
)via
(Ir U0 Is
).
![Page 37: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/37.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A better version
In other words, UV and VU have nearly the same characteristicpolynomials and nearly the same eigenvalues.
The only difference: 0 is a eigenvalue of the larger matrix, repeatedas necessary.
For now, let us take U to be n × (n − 1) and V to be (n − 1)× n.
![Page 38: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/38.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A better version
An example. Begin with
U =
1 2 −30 1 21 2 10 −2 2
V =
1 −2 1 1−1 −1 1 11 2 −1 1
![Page 39: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/39.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A better version
Then
VU =
2 0 −40 −3 40 0 2
with eigenvalues λ = 2,−3, 2and
UV =
−4 −10 6 01 3 −1 30 −2 2 44 6 −4 0
.
So UV is an integer matrix with eigenvalues λ = 2,−3, 2 andλ = 0.
![Page 40: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/40.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A better version
Then
VU =
2 0 −40 −3 40 0 2
with eigenvalues λ = 2,−3, 2
and
UV =
−4 −10 6 01 3 −1 30 −2 2 44 6 −4 0
.
So UV is an integer matrix with eigenvalues λ = 2,−3, 2 andλ = 0.
![Page 41: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/41.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A better version
Then
VU =
2 0 −40 −3 40 0 2
with eigenvalues λ = 2,−3, 2and
UV =
−4 −10 6 01 3 −1 30 −2 2 44 6 −4 0
.
So UV is an integer matrix with eigenvalues λ = 2,−3, 2
andλ = 0.
![Page 42: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/42.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A better version
Then
VU =
2 0 −40 −3 40 0 2
with eigenvalues λ = 2,−3, 2and
UV =
−4 −10 6 01 3 −1 30 −2 2 44 6 −4 0
.
So UV is an integer matrix with eigenvalues λ = 2,−3, 2 andλ = 0.
![Page 43: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/43.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A better version
Writing
U =
| | |~u1 ~u2 ~u3
| | |
V =
− ~v1 −− ~v2 −− ~v3 −
We needed ~v2 · ~u1 = ~v3 · ~u1 = ~v3 · ~u2 = 0.Then the eigenvalues of UV are the three dot products ~vi · ~ui and 0.
![Page 44: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/44.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A better version
Writing
U =
| | |~u1 ~u2 ~u3
| | |
V =
− ~v1 −− ~v2 −− ~v3 −
We needed ~v2 · ~u1 = ~v3 · ~u1 = ~v3 · ~u2 = 0.
Then the eigenvalues of UV are the three dot products ~vi · ~ui and 0.
![Page 45: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/45.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A better version
Writing
U =
| | |~u1 ~u2 ~u3
| | |
V =
− ~v1 −− ~v2 −− ~v3 −
We needed ~v2 · ~u1 = ~v3 · ~u1 = ~v3 · ~u2 = 0.Then the eigenvalues of UV are the three dot products ~vi · ~ui and 0.
![Page 46: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/46.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A better version
Note: we now only need an (n − 1)× (n − 1) matrix to betriangular. There are now (n − 1)(n − 2)/2 orthogonalityconditions.
This means there are no conditions to be checked in order toconstruct a 2× 2 example.
![Page 47: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/47.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A better version
Note: we now only need an (n − 1)× (n − 1) matrix to betriangular. There are now (n − 1)(n − 2)/2 orthogonalityconditions.
This means there are no conditions to be checked in order toconstruct a 2× 2 example.
![Page 48: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/48.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A better version
Any (u1
u2
)(v1 v2
)=
(u1v1 u1v2
u2v1 u2v2
)is an integer matrix with eigenvalues 0 and u1v1 + u2v2.
E.g. (21
)(1 −4
)=
(2 −81 −4
)λ = −2, 0
![Page 49: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/49.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A better version
Any (u1
u2
)(v1 v2
)=
(u1v1 u1v2
u2v1 u2v2
)is an integer matrix with eigenvalues 0 and u1v1 + u2v2.E.g. (
21
)(1 −4
)=
(2 −81 −4
)λ = −2, 0
![Page 50: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/50.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A better version
Disadvantages:• We always get 0 as an eigenvalue.• What are the eigenvectors?Advantage:• This is (nearly) a complete characterization for these matrices.
![Page 51: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/51.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A shift by b
Great tool No. 2:
Proposition (Shift Proposition)
If A has eigenvalues λ1, . . . , λn.
Then A + bIn has eigenvalues b + λ1, . . . , b + λn .
Why does this work? Note that A and A + bI have the sameeigenvectors.
![Page 52: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/52.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A shift by b
Great tool No. 2:
Proposition (Shift Proposition)
If A has eigenvalues λ1, . . . , λn.
Then A + bIn has eigenvalues b + λ1, . . . , b + λn .
Why does this work? Note that A and A + bI have the sameeigenvectors.
![Page 53: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/53.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A shift by b
Great tool No. 2:
Proposition (Shift Proposition)
If A has eigenvalues λ1, . . . , λn.
Then A + bIn has eigenvalues b + λ1, . . . , b + λn .
Why does this work? Note that A and A + bI have the sameeigenvectors.
![Page 54: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/54.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A shift by b
Great tool No. 2:
Proposition (Shift Proposition)
If A has eigenvalues λ1, . . . , λn.
Then A + bIn has eigenvalues b + λ1, . . . , b + λn .
Why does this work? Note that A and A + bI have the sameeigenvectors.
![Page 55: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/55.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A shift by b
Example:
VU =
2 0 −40 −3 40 0 2
so UV =
−4 −10 6 01 3 −1 30 −2 2 44 6 −4 0
.
has eigenvalues 2,−3, 2, 0.
So
A = UV − I =
−5 −10 6 01 2 −1 30 −2 1 44 6 −4 −1
has eigenvalues 1,−4, 1,−1.
![Page 56: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/56.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A shift by b
Example:
VU =
2 0 −40 −3 40 0 2
so UV =
−4 −10 6 01 3 −1 30 −2 2 44 6 −4 0
.
has eigenvalues 2,−3, 2, 0.So
A = UV − I =
−5 −10 6 01 2 −1 30 −2 1 44 6 −4 −1
has eigenvalues 1,−4, 1,−1.
![Page 57: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/57.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A shift by b
Earlier we had
UV =
(21
)(1 −4
)=
(2 −81 −4
)with λ = −2, 0
So
UV + I =
(3 −81 −3
)has eigenvalues λ = −1, 1.
![Page 58: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/58.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A shift by b
Earlier we had
UV =
(21
)(1 −4
)=
(2 −81 −4
)with λ = −2, 0So
UV + I =
(3 −81 −3
)has eigenvalues λ = −1, 1.
![Page 59: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/59.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
A different approach
A shift by b
Note: This gives a quick proof to our Eigenvalues Congruent to bmodulo k Proposition.
![Page 60: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/60.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
![Page 61: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/61.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
This is actually a classification of all IMIEs.
Theorem (Renaud)
Every n × n integer matrix with integer eigenvalues can be writtenas a UV + bIn where VU is a triangular (n − 1)× (n − 1) matrix.
Note: U is n × (n − 1) and V is (n − 1)× n.
![Page 62: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/62.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
This is actually a classification of all IMIEs.
Theorem (Renaud)
Every n × n integer matrix with integer eigenvalues can be writtenas a UV + bIn where VU is a triangular (n − 1)× (n − 1) matrix.
Note: U is n × (n − 1) and V is (n − 1)× n.
![Page 63: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/63.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
Is there any way to take further advantage of the Folk Theorem?
What if U is n × r and V is r × n?
In particular, if U is n × 1 and V is 1× n then VU is triviallytriangular.
![Page 64: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/64.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
Is there any way to take further advantage of the Folk Theorem?
What if U is n × r and V is r × n?
In particular, if U is n × 1 and V is 1× n then VU is triviallytriangular.
![Page 65: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/65.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
Is there any way to take further advantage of the Folk Theorem?
What if U is n × r and V is r × n?
In particular, if U is n × 1 and V is 1× n then VU is triviallytriangular.
![Page 66: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/66.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
Take U =
111
and V =(1 2 1
).
Then VU = (4) and UV must be a 3× 3 integer matrix witheigenvalues 4, 0, 0.
In fact UV + I3 =
2 2 11 3 11 2 2
= A, our original example.
![Page 67: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/67.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
Take U =
111
and V =(1 2 1
).
Then VU = (4) and UV must be a 3× 3 integer matrix witheigenvalues 4, 0, 0.
In fact UV + I3 =
2 2 11 3 11 2 2
= A, our original example.
![Page 68: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/68.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
![Page 69: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/69.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
What about the eigenvectors of UV ?
Suppose VU is not just triangular, but diagonal.
Then the (n − 1) columns of U are eigenvectors corresponding tothe λ1, . . . , λn−1.Any vector, ~w in the null space of V is an eigenvectorcorresponding to b.
![Page 70: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/70.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
What about the eigenvectors of UV ?Suppose VU is not just triangular, but diagonal.
Then the (n − 1) columns of U are eigenvectors corresponding tothe λ1, . . . , λn−1.Any vector, ~w in the null space of V is an eigenvectorcorresponding to b.
![Page 71: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/71.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
What about the eigenvectors of UV ?Suppose VU is not just triangular, but diagonal.
Then the (n − 1) columns of U are eigenvectors corresponding tothe λ1, . . . , λn−1.
Any vector, ~w in the null space of V is an eigenvectorcorresponding to b.
![Page 72: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/72.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
What about the eigenvectors of UV ?Suppose VU is not just triangular, but diagonal.
Then the (n − 1) columns of U are eigenvectors corresponding tothe λ1, . . . , λn−1.Any vector, ~w in the null space of V is an eigenvectorcorresponding to b.
![Page 73: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/73.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
Example. U =
−3 −21 01 2
and V =
(1 −2 10 1 −1
).
Then VU =
(−4 00 −2
), diagonal.
So the columns of U are eigenvectors (corresponding toλ = −4,−2, resp.) of
UV =
−3 2 11 −2 11 2 −3
.
~w =
111
has V ~w = ~0. So it is an eigenvector for UV
corresponding to λ = 0.
![Page 74: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/74.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
Example. U =
−3 −21 01 2
and V =
(1 −2 10 1 −1
).
Then VU =
(−4 00 −2
), diagonal.
So the columns of U are eigenvectors (corresponding toλ = −4,−2, resp.) of
UV =
−3 2 11 −2 11 2 −3
.
~w =
111
has V ~w = ~0. So it is an eigenvector for UV
corresponding to λ = 0.
![Page 75: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/75.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
Example. U =
−3 −21 01 2
and V =
(1 −2 10 1 −1
).
Then VU =
(−4 00 −2
), diagonal.
So the columns of U are eigenvectors (corresponding toλ = −4,−2, resp.) of
UV =
−3 2 11 −2 11 2 −3
.
~w =
111
has V ~w = ~0. So it is an eigenvector for UV
corresponding to λ = 0.
![Page 76: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/76.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
![Page 77: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/77.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
Recall: A = PDP−1 approach.
Idea 1. Make sure P has determinant ±1.
How to construct such P?
![Page 78: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/78.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
One technique [Ortega, 1984] uses:Take ~u, ~v ∈ Rn. Then P = In + ~u ⊗ ~v has detP = 1 + ~u · ~v .
So choose ~u ⊥ ~v . Then P = In + ~u ⊗ ~v has detP = 1and P−1 = In − ~u ⊗ ~v .
This is just a special case of what we have been doing with theFolk Theorem and the Shift Propostion.
![Page 79: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/79.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
One technique [Ortega, 1984] uses:Take ~u, ~v ∈ Rn. Then P = In + ~u ⊗ ~v has detP = 1 + ~u · ~v .
So choose ~u ⊥ ~v . Then P = In + ~u ⊗ ~v has detP = 1
and P−1 = In − ~u ⊗ ~v .
This is just a special case of what we have been doing with theFolk Theorem and the Shift Propostion.
![Page 80: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/80.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
One technique [Ortega, 1984] uses:Take ~u, ~v ∈ Rn. Then P = In + ~u ⊗ ~v has detP = 1 + ~u · ~v .
So choose ~u ⊥ ~v . Then P = In + ~u ⊗ ~v has detP = 1and P−1 = In − ~u ⊗ ~v .
This is just a special case of what we have been doing with theFolk Theorem and the Shift Propostion.
![Page 81: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/81.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
One technique [Ortega, 1984] uses:Take ~u, ~v ∈ Rn. Then P = In + ~u ⊗ ~v has detP = 1 + ~u · ~v .
So choose ~u ⊥ ~v . Then P = In + ~u ⊗ ~v has detP = 1and P−1 = In − ~u ⊗ ~v .
This is just a special case of what we have been doing with theFolk Theorem and the Shift Propostion.
![Page 82: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/82.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
Example. Try ~u =
124
and ~v =
−2−11
. (Check: ~u · ~v = 0.)
Then
~u ⊗ ~v =
−2 −1 1−4 −2 2−8 −4 4
Get
P = I + ~u ⊗ ~v =
−1 −1 1−4 −1 2−8 −4 5
.
and
P−1 = I − ~u ⊗ ~v =
3 1 −14 3 −28 4 −3
.
![Page 83: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/83.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
Example. Try ~u =
124
and ~v =
−2−11
. (Check: ~u · ~v = 0.)
Then
~u ⊗ ~v =
−2 −1 1−4 −2 2−8 −4 4
Get
P = I + ~u ⊗ ~v =
−1 −1 1−4 −1 2−8 −4 5
.
and
P−1 = I − ~u ⊗ ~v =
3 1 −14 3 −28 4 −3
.
![Page 84: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/84.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
Example. Try ~u =
124
and ~v =
−2−11
. (Check: ~u · ~v = 0.)
Then
~u ⊗ ~v =
−2 −1 1−4 −2 2−8 −4 4
Get
P = I + ~u ⊗ ~v =
−1 −1 1−4 −1 2−8 −4 5
.
and
P−1 = I − ~u ⊗ ~v =
3 1 −14 3 −28 4 −3
.
![Page 85: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/85.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
Example. Try ~u =
124
and ~v =
−2−11
. (Check: ~u · ~v = 0.)
Then
~u ⊗ ~v =
−2 −1 1−4 −2 2−8 −4 4
Get
P = I + ~u ⊗ ~v =
−1 −1 1−4 −1 2−8 −4 5
.
and
P−1 = I − ~u ⊗ ~v =
3 1 −14 3 −28 4 −3
.
![Page 86: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/86.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
So Ortega uses this to say that any A = PDP−1 will be an IMIE.
But in fact, P itself is an IMIE. It has the single eigenvalue 1.
P is clearly non-diagonalizable.−1 −1 1−4 −1 2−8 −4 5
∼1 0 0
0 1 10 0 1
.
![Page 87: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/87.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
So Ortega uses this to say that any A = PDP−1 will be an IMIE.
But in fact, P itself is an IMIE. It has the single eigenvalue 1.
P is clearly non-diagonalizable.−1 −1 1−4 −1 2−8 −4 5
∼1 0 0
0 1 10 0 1
.
![Page 88: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/88.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
Ortega’s full result is that if ~u · ~v = β then P = In + ~u ⊗ ~v hasdetP = 1 + β and (if β 6= −1) then P−1 = In − 1
1+β ~u ⊗ ~v .
The case β = −2 is interesting.
![Page 89: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/89.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
Earlier we had
UV =
(21
)(1 −4
)=
(2 −81 −4
)with λ = −2, 0
So
Q = UV + I =
(3 −81 −3
)has eigenvalues λ = −1, 1.In fact, this Q is coming from Ortega’s construction. So Q−1 = Q.
![Page 90: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/90.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
Earlier we had
UV =
(21
)(1 −4
)=
(2 −81 −4
)with λ = −2, 0So
Q = UV + I =
(3 −81 −3
)has eigenvalues λ = −1, 1.In fact, this Q is coming from Ortega’s construction. So Q−1 = Q.
![Page 91: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/91.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Classification Results
Further refinements
![Page 92: Constructing and (some) classification of integer matrices ... · A = PDP 1 where P = 0 @ j j j u~ 1 ~u 2 u~ n j j j 1 A has columns which form a basis of eigenvectors for A and](https://reader034.vdocuments.mx/reader034/viewer/2022042309/5ed6841cff0e593c0b63fd4e/html5/thumbnails/92.jpg)
Constructing and (some) classification of integer matrices with integer eigenvalues
Christopher Towse and Eric Campbell. Constructing integermatrices with integer eigenvalues. The Math. Scientist 41(2):45–52, 2016.
W. P. Galvin., The Teaching of Mathematics: Matrices with“Custom-Built” Eigenspaces. Amer. Math. Monthly, 91(5):308–309, 1984.
Greg Martin and Erick Wong. Almost all integer matrices haveno integer eigenvalues.Amer. Math. Monthly, 116(7):588-597, 2009.
James M. Ortega. Comment on: “Matrices with integerentries and integer eigenvalues”. Amer. Math. Monthly, 92(7):526, 1985.
J.-C. Renaud.Matrices with integer entries and integereigenvalues. Amer. Math. Monthly, 90(3): 202–203, 1983.