Linear Algebra II
Exercise Sheet no. 2
SS 2011
Prof. Dr. Otto April 20, 2011
Dr. Le Roux Dr. Linshaw
Exercise 1 (Further properties of eigenvalues/eigenspaces) Throughout this exercise,Ais a square matrix with entries inR.
(a) Prove or disprove thatAandAt have the same eigenvalues. (At is the transpose ofA, see lecture notes for Linear Algebra I.)
(b) Prove or disprove thatAandAthave the same eigenspaces.
(c) AssumeAis regular and letvbe an eigenvector ofAwith eigenvalueλ. Show thatvis also an eigenvector ofA−1 with eigenvalue λ1.
(d) Letvbe an eigenvector of the matrixAwith eigenvalueλand lets be a scalar. Show thatvis an eigenvector of A−sEwith eigenvalueλ−s.
Solution:
a) The claim is correct: ifλis an eigenvalue ofA, then det(A−λE) =0. Since the determinant is invariant under transposition, also det(A−λE)t =det(A−λE) =0. Making the following computation:(A−λE)t=At−(λE)t= At−λE, we see that det(At−λE) =0. Thereforeλis also an eigenvalue ofAt. The other direction follows by symmetry ((At)t=A).
b) This claim is incorrect, as can be seen by considering the following counterexample. WhenA:=
0 1 0 0
, the only eigenvalue ofAis 0. Its eigenspace isV0=span(e1) ={α
1 0
:α∈R}. 0 is also the only eigenvalue ofAt, but its eigenspace isV00=span(e2) ={α
0 1
:α∈R}.
c) Letλbe an eigenvalue ofAandvbe one of its eigenvectors. ThenAv=λvholds. SinceAis invertible,A−1exists.
By multiplying the previous equation byA−1from the left, we obtainv=λA−1v. Now we multiply by λ1 and get A−1v= 1λv(λ6=0asAis regular). Thusvis an eigenvector ofA−1with eigenvalue λ1.
d) CombiningAv=λvandsEv=svwe get:Av−sEv=λv−sv. This can be written as(A−sE)v= (λ−s)v, therefore vis an eigenvector ofA−sEwith eigenvalueλ−s.
Exercise 2 (Application of diagonalisation: Fibonacci Numbers, Golden Mean)
Recall that the sequence f0,f1,f2, . . .of Fibonacci numbers is inductively defined as follows (cf Exercise H6.4 from LA I):
f0 = 0, f1 = 1, fk+2 = fk+1+fk.
(a) We defineuk= fk+1
fk
∈R2. Find a matrixAsuch thatuk+1=Aukfor allk∈N.
1
(b) What are the eigenvalues ofA? Give an explicit formula for fk.
Hint: Use the eigenvaluesλ1andλ2as abbreviations as long as possible.
(c) Compute the limita=limk→∞ fk+1f
k .
The limit is called the Golden Mean, it divides a line segment of length1into two parts a and 1−a such that
1 a =1−aa. Solution:
a) uk+1= fk+2
fk+1
=
fk+1+fk fk+1
= 1 1
1 0 fk+1
fk
= 1 1
1 0
uk.Therefore,
A= 1 1
1 0
.
As a consequence, for allk∈N,
uk=Aku0=Ak 1
0
.
b) det(A−λE) =λ2−λ−1= (λ−1+2p5)(λ−1−2p5) =0.
Thus, the eigenvalues ofAare:
λ1= 1+p 5
2 and λ2=1−p 5 2 . Corresponding eigenvectors arev1=
λ1
1
forλ1andv2= λ2
1
forλ2. It follows thatA=S DS−1, withS=
λ1 λ2
1 1
and D=
λ1 0 0 λ2
. The conjugation map M 7→S M S−1is an automorphism of the ring ofn×nmatrices (and in particular preserves sums and products of matrices), so we haveAk= (S DS−1)k=S DkS−1. SinceS−1=p15
1 −λ2
−1 λ1
,
Ak = S DkS−1= 1 p5
λ1 λ2
1 1
λ1k 0 0 λk2
1 −λ2
−1 λ1
= 1 p5
λk1+1 λk2+1 λk1 λ2k
1 −λ2
−1 λ1
= 1 p5
λk1+1−λ2k+1 −λ1k+1λ2+λk2+1λ1
λ1k−λ2k −λk1λ2+λk2λ1
. It follows that
fk+1 fk
=uk=Ak 1
0
= 1 p5
λk+11 −λk+12 λk1−λk2
,
hence
fk= 1
p5(λ1k−λ2k) = 1 p5
1+p 5 2
k
−
1−p 5 2
k! .
c) We have that
fk+1
fk = λk+11λk−λ2k+1 1−λ2k = λλk+11k
1 ·
1−λ k+1 2 λk+1
1 1−λ k2 λk 1
=λ1·1−(
λ2λ1)k+1 1−(λ2λ1)k and
λ2 λ1
=
1−p 5 1+p
5
=pp5−15+1<1.
It follows that
k→∞lim fk+1
fk =λ1·
1−limk→∞λ
2 λ1
k+1
1−limk→∞λ
2 λ1
k =λ1.
2
Exercise 3 (Euclidean algorithm and recursive functions)
The greatest common divisord=gcd(a,b)of two non-zero natural numbersaandbis a natural number characterised by the following property:
• d|bandd|a.
• Ifr|aandr|b, thenr|d
The Euclidean algorithm is a procedure for determining the greatest common divisor of two numbersaandb.
Step0: Swap a and b ifa<b.
Euclid (a,b): IFb=0THEN returna, ELSE return Euclid(b,amodb).
[After initialising: d1:=min{a,b},d0:=max{a,b}, we divide with remainder in each step: dk−1=qkdk+dk+1with 0≤dk+1<dk, and ends ifdk+1=0. We getdk=gcd(a,b).]
(a) Prove that the gcd is well defined, that is, uniquely characterised by the above properties.
(b) Prove that gcd(a,b) =gcd(b,amodb)for0<b.
(c) Assume thatb<aand that(a,b)→E (a0,b0)→E (a00,b00)are two steps in the Euclidean algorithm. Prove thata00< a2 andb00<2b.
(d) Deduce that Euclid(a,b)returns gcd(a,b).
Solution:
a) 1divides bothaandbso the set of common divisors is non-empty. Letdandd0satisfy the second condition. Then d|d0andd0|d, sod=d0.
b) Supposea=q b+rwhere0≤r<b, sor=amodb. Sincee|aande|biffe|r=a−q bande|b. Letd=gcd(a,b) andd0=gcd(b,r). We haved|d0andd0|d, sod=d0.
c) Ifb≤ a2, then b0< b≤ a2; if a2 <b, thenq=1so b0 =a−b< a2. So in every case, b0< a2. Similarly b00< a20. Moreovera0=banda00=b0, soa00<a2 andb00< b2.
d) Every step(a,b)→E (a0,b0)of the algorithm preserves the gcd, in the sense that gcd(a,b) =gcd(a0,b0). Moreover the algorithm terminates due to part (c), so we obtain(gcd(a,b), 0)after finitely many steps.
Exercise 4 (Diagonalization and recursive sequences)
Letak be the sequence of real numbers defined recursively as follows: a0=0,a1=1, andak+2= 12(ak+1+ak). In other words, each term in the sequence is the average of the two previous terms.
(a) As we did with the Fibonacci sequence, we want to study this sequence using diagonalization of matrices. For k≥0, let
uk= ak+1
ak
.
Using the equationsak+2= 12(ak+1+ak)andak+1=ak+1, find a2×2matrixAsuch thatuk+1=Auk. (b) Find the eigenvalues and eigenvectors ofA, and find a matrixSand a diagonal matrixDwithD=S−1AS.
(c) Find a formula forak, and calculatelimk→∞akif it exists.
Solution:
a) A=
1
2 1 2
1 0
.
b) The eigenvalues ofAare1and−12. Forλ=1, the eigenspaceV1is spanned by
1 1
. Forλ=−12, the eigenspaceV−1/2is spanned by 1
−2
. So S=
1 1 1 −2
andD=
1 0 0 −12
.
3
c) First, we haveuk=Aku0=S DkS−1u0, whereu0= 0
1
, andakis just the second component of the vectoruk. We calculate
S−1=
2/3 1/3 1/3 −1/3
, S−1u0=
1/3
−1/3
, Dk=
1 0 0 (−12)k
.
It follows easily thatak=13+23 −12k
. It is clear from this formula thatlimk→∞ak=13.
4