14.2. THE SPECTRAL RADIUS 391

Recall the following symbols and their meaning. limsupn→∞ an, liminfn→∞ an. Theyare respectively the largest and smallest limit points of the sequence {an} where ±∞ isallowed in the case where the sequence is unbounded. They are also defined as

lim supn→∞

an ≡ limn→∞

(sup{ak : k ≥ n}) ,

lim infn→∞

an ≡ limn→∞

(inf{ak : k ≥ n}) .

Thus, the limit of the sequence exists if and only if these are both equal to the same realnumber.

Lemma 14.2.2 Let J be a p× p Jordan block

J =

λ 1 0

λ. . .. . . 1

0 λ

Then limn→∞ ||Jn||1/n = |λ |

Proof: The norm on matrices can be any norm. It could be operator norm for example.If λ = 0, there is nothing to show because Jp = 0 and so the limit is obviously 0. Therefore,assume λ ̸= 0.

Jn =p

∑i=0

(ni

)Ni

λn−i

Then

∥Jn∥ ≤p

∑i=0

(ni

)∥∥Ni∥∥ |λ |n−i = |λ |n +C |λ |np

∑i=1

(ni

)|λ |−i (14.4)

≤ |λ |n (1+Cnp)≤ |λ |n C̃np (14.5)

where the C depends on ∑pi=1 |λ |

−i and the∥∥Ni∥∥. Therefore,

lim supn→∞

∥Jn∥1/n ≤ |λ | lim supn→∞

(C̃np)1/n

= |λ |

Next let x be an eigenvector for λ such that ∥x∥ = 1, the norm being whatever norm isdesired. Then Jx= λx. It follows that Jnx= λ

nx. Thus ∥Jn∥≥ ∥Jnx∥= |λ |n ∥x∥= |λ |n .It follows that liminfn→∞ ∥Jn∥1/n ≥ |λ | . Therefore,

|λ | ≤ lim infn→∞∥Jn∥1/n ≤ lim sup

n→∞

∥Jn∥1/n ≤ |λ |

which shows that limn→∞ ∥Jn∥1/n = |λ |. The same conclusion holds for any other norm.Indeed, if |||·||| were another norm, there are constants δ ,∆ such that

δ1/n ∥Jn∥1/n ≤ |||Jn|||1/n ≤ ∆

1/n ∥Jn∥1/n

Then since limn→∞ δ1/n = limn→∞ ∆1/n = 1, the squeezing theorem from calculus implies

that limn→∞ |||Jn|||1/n = ρ . ■

14.2. THE SPECTRAL RADIUS 391Recall the following symbols and their meaning. limsup,_,.,@n, liminf,.. dn. Theyare respectively the largest and smallest limit points of the sequence {a,} where +e isallowed in the case where the sequence is unbounded. They are also defined aslim supa, = _ lim (sup{a,:k >n}),n—oo noolim inf dn = Jim, (inf {a,:k >n}).Thus, the limit of the sequence exists if and only if these are both equal to the same realnumber.Lemma 14.2.2 Let J be a p x p Jordan blockA 1 0AJ=10 AThen lity sx \|J"||!/" = |A|Proof: The norm on matrices can be any norm. It could be operator norm for example.If A = 0, there is nothing to show because J? = 0 and so the limit is obviously 0. Therefore,assume A + 0.P n iniy= N alAeThen‘ n i n-t n ne n -iI'l se, PIN = lap eel yet) ia (14.4)i=0 i=l< |Al"(1+0n?) < |A|"Cn? (14.5)where the C depends on Y?._, |A|~! and the ||N’||. Therefore,lim sup ||J"||!/" < |A| lim sup (Gn?) " =|A|n—-roo n—yooNext let x be an eigenvector for A such that ||a|| = 1, the norm being whatever norm isdesired. Then Ja =A. It follows that "a2 = A"a. Thus ||J"|| > |[J"a|| = |A|” ||a|] =|A".It follows that lim inf, 42. ||J/”||'/" > |A|. Therefore,|A| <lim inf |JJ"\|!/" < lim sup |JJ"|!/" < |a|noo n—yoowhich shows that limy—+.o ||J”|| Vn _ |A|. The same conclusion holds for any other norm.Indeed, if |||-||| were another norm, there are constants 6,A such thatayy < |jaryyl" <a yarnThen since limy_s.. sin Limyyc0 Al/ ” = |, the squeezing theorem from calculus impliesthat limyseo |||J"|||!/" =p. i