TeoriaOperatorów Kraków,22-28 Seplcmber 2003
pp. 35-42
PRODUCTS OF n-ROOTS OF THE IDENTITY
PIOTR BUDZYŃSKI
Abstract. We present recent results concerning decomposition of operators into products of finite number of n-roots of the identity. We show that every invertible bounded linear operator on a complex infinite-dimensional Hilbert space is a product of five n-roots of the identity for every n > 2. In the case when n = 2 seven factors suffice.
1. Prelminaries
Assume that all operators appearing in this paper act on an infinite-dimensional separable Hilbert space. All results are given only in this settings. Generalizations to the nonseparablecase will be obvious for the reader.
Definition 1.1. Let n €N. Operator A € S(W) is called an n-root of the identity (n-root) if:
By root we mean 2-root.
Definition 1.2. Let T be an operator acting on a Hilbert space H. We call this operator:
(i) normal iff TT* = T*T, (ii) unitary iff TT* = T*T = I, (iii) self-adjoint iff T* = T
(iv) compact iff the set T({f€ H : ||/|| < 1}) iscompact in H.
Corollary 1.3. Let A € B(Tt) bea product offinite number of n-roots of the identity.
If operator B g B(7C) is similar to A (i.e. B = SAS-1 for some invertible S) then itselfis the product ofthe same numberofn-roots.
Proof. Assume that there exist operator S e B(H,IC) and n-roots of the identity TUT2,...,7fc € B(H) such that B = SAS-1 i A = Tj.. .Tk. Then we have:
and every operator STiS \ i = l,...,fc, is n-root. If S is unitary operator and B = {ST1S-l)...{STkS~1)
□
3G PIOTR BUDZYŃSKI
Definition 1.4. Banach algebra is (complex) unital algebra A equipped with a norm satisfying followingconditions:
(0 ll^yll < lklll|y||,for every x,y eA (ii) ||e|| = 1, (e is neutral element of>1), (iii) (.4,|| • ||) is Banach space.
Spectrum of element a of algebra Ais defined by :
<7.4 (a) = {A G C : (Ae— a) is not invertible in .4}.
From the spectral theorem weget:
Corollary 1.5. Let N G be a normal operator . For every natural n > 2 there exist closed subspacesTCi,...,ICn C H, with the same dimension and operators Nj G B(JCj) (j = 1,... ,n) such that N = Ni ffi... ® Nn.
2. Roots
Theorem 2.1. Everyunitary operator on H is a product of fourself-adjoint roots of the identity.
Proof. Decompose space 7Y intoorthogonal sum Tt = ©■¿Hn of infinite-dimensional subspaces, reducingoperator U. Find unitary operator S acting on H satisfying:
SHn = Hn+1. (2.1)
Ifwe define T = S*U, then obviously:
U = ST and
THn = S*UHn = S*Hn=Hn-1. (2.2) Since the adjoint of every operator T satisfying (2.2) satisfies condition (2.1), it is enough to show that operatorSis aproduct of two self-adjoint roots ofthe identity.
Define unitary operators P, Q on Tl by:
P\h„ =51-2n|«„ (2.3)
and
Q\nn =S~2n\-Hn. (2.4)
Simple calculations show that P, Q are roots ofthe identity andS = PQ. □ Example 2.2. On every Hilbert space H there exists a unitaryoperatorU that is not a product of three roots. Consider a G C : a3 = 1, 1 anddefine U = al.
Such operator commutes with any other operator, hence ifX, Y, Z are roots of the identity suchthat U= XYZthen:
[74 = U3U = UXUYUZ = U(XU)Y(UZ) = U(YZ)Y(XY) =
= Y(UZY){XY) = Y(X)(XY) = Y2 = I.
what is contradiction.
Lemma 2.3. If B G B(H) is an invertible operator, then B © B 1 is a product of two roots.
Proof. Observe that:
(B 0 \ = / 0 BWO A
b-1) v#-1 oJv oj’
□ Lemma 2.4. If B G B(TT) is an invertible operator then B © 1-n is aproduct of four roots.
Proof. Since 7Y is infinite-dimensionalthen:
B®I-h=B®I®I®I®---
= (B © B_1 ©B®B_1 ©•••)(/© B © B_1 © B® B~! © ■ ••).
From the previous lemma we get B © B_1 = TS, where T2 = S2 = I and S* = S therefore equalities:
B © In = (TS © TS ©•••)(/© TS © TS © • ••)
= (T ® T ® •••)(S ® S ffi •••)(/©T © T©••• )(Z © S ® S©•••),
finish the proof. □
Theorem 2.5. IfN G Bffi) is an invertible, normal operator then N is a product of six roots.
Proof. We can decompose N into orthogonal sum:
N = M ffi N2 “B © C e B(K. © K).
We have
BffiC = (C“1 ©C) (CB®I).
Apply both lemmas. □
Using polar decompositionweget:
Theorem 2.6. Every invertible operator A G B(W) is a product ofsixroots. Moreover four of them are self-adjoint operator.
3. n-ROOTS
Definition 3.1. Aninvertible operator T acting ona Hilbertspace H is commuta
tor if there exist invertibleoperators A and B on H suchthat:
T = ABA~1B~1.
We arenowto present and provelemma which we will frequentlyuseinthissection.
38 PIOTR BUDZYŃSKI
Theorem 3.2. LetA3,... ,An G B(TL), n > 2, be such that aproduct Ai ••• ATl is a commutator. Then there exist Ti,T2,T3 G B(TC) such that:
Ai ffi... ffi An = TiT2T3 and
fJ-tTl _ rjvn _ _ Moreover operator T3 can be chosen unitary.
Proof. Let Ai... An = UVU 1V '. Define unitary operator W by nx n operator matrix:
0 0
I0 0 I
0\
0
w =
0 0 0 I
V
0 0Now put
Zi = ([/-'AMj® A^Af'UVAi ffi 7®...ffi/ffiAf’V“1),
Z2 = (VAiA2 ® A71 Af^Ai ffiAf1 AiA2A3 ® A4 ffi ... ® An), and
T^ZrW, T2 = Z2W, T3 = W~2. Elementary calculationsshow that T" = I, j = 1,2,3. Moreover
Ai ffi A3ffi A4 ffi ... ffiAnffi A2 = T\T2T3.
Since operators Ai ffi... ffi An and Ai ffiA3ffi A4 ffi... ffiAn ffiA2 are unitary equivalent the proof isfinished (n-root73 is unitary so is Tb = UTsl/-1, if U is unitary). □ Before wegettothe specific application of above lemma we briefly discuss technique of decomposinginvertiblenormal operator into a product of TV-roots. By Corollary 1.5 any normaloperator N on an infinite-dimensional Hilbert space can be split into the orthogonal sum of arbitrarily many normal operators acting on infinite-dimensional subspaces:
N = Mffi ... ffi Nn.
Onecan immediately notice connection withthe lemma— knowingthat the product Ni ••• Nn is a commutator we follow theproofand construct proper n-roots. Wesee that the main difficulty is containedin the question: which properties ofsummands Niy..., Nn imply that the product N3 ■ ■■Nn is a commutator? It turns out that the answer is based on exact analysis ofessential spectrum <7ess(N) of operator N and behavior of N onreducing subspaces.
Theorem 3.3. If T is an invertible operator on B(TL) and K, is closed, infinite dimensional subspaceofH which reduces T tounitary operatorthen Tis commutator.
A simple demonstration of mentioned techniqueoffers:
Theorem 3.4. Let N = M ffi U & B(Lt®Lt) be an invertible operator such that jVj
is normal and U isunitary. Then operator N is a product ofthree n-rootsof identity, for every n >2.
Proof. We can decompose N into thesum:
N = N] ffi U2 ffiU3 ffi... ffi U2n and find unitary operator V so:
VNV~r =M ffi U2 ffi U3 ffi ...ffi U2n € B(W2n), where Ui € B(Tt), i= 2,..., 2n, are unitary. Operator:
T = (M ffi U2)(U3ffi U4)... (U2n_iffi U2n)
= (MC/3 . . .U2n-1) ffi(U2U4 ... U2n)
has infinite-dimensional subspace reducing to unitaryoperator,hence is commutator
by previous theorem. □
Corollary 3.5. Every unitary operator is aproduct of three unitary n-roots, for every natural n > 2.
This resultis best possible as shows following:
Example 3.6. Let a € C : |a| = 1 and an / 1.
Unitary operator al cannot be a product oftwo n-roots of identity. Suppose that X, Y are such roots. From the equalities a = XY and Xn = Yn = I we get that X and Y are commuting (X = aYn~!'j so:
on = (XY)n = XnYn = I, what is obvious contradiction.
We said that one of the possible ways to answer the question if the product TVi• • ■ Nn is to analyze essential spectrum of N. This notion is strictly connected with Calkin algebra.
Definition 3.7. Calkin algebra (over Hilbertspace Lt) is quotient Banach algebra defined:
Ac(Li) = B(Lt)/C(Lt), where C(Li) is two-sidedideal ofcompact operatorson Lt.
Definition 3.8. Let A be an operatoron Li. Essential spectrum of A is the set:
<Tess(j4) = &Ac ([-^]~)-
Theorem 3.9. Let N € B(Li) be an invertible normal operator. If there exist a, fl € Oess{N) such that |a| < |/3| then for every natural n > 2 operator N is aproduct of
three n-roots ofidentity.
Wedon’tgivethe proof, mainly because it is to long,howeverwementionfollowing theorem of Brown and Pearcy which plays crucial role in it.
40 PIOTRBUDZYŃSKI
Theorem 3.10. If invertible operators F,G, acting on a Hilbert space H, are not of the form A + К, where К is compact operator, then orthogonal sum F ф G is commutator.
It is also used in the next one.
Theorem 3.11. Let N = Nф (a + /<) G фH) be an invertible and normal operator such that: cress(N) C [z G C : |z| = r} for some r > 0 and К is compact operator. Ifr=l or there exists /3 G cress(N) such that/3 a then N is a product of threen-roots, for every naturaln > 2.
Proof. Considerfirst the case when r= 1. If a isthe only point inessential spectrum cre.S5(N), then N = a+ К with some compact operator К and |ct| = 1. This comes from spectral mapping theorem and:
аЛс([^ - a[/]~) = {0}.
Use spectral theorem to get:
N ~ (a + K4) Ф ... ф (a +Kn) € B(fCu).
a product (o + Ki)... (a+ Kn) = an +K' is commutatorby Theorem3.10,sinceK' is compact. Now it suffices to use Theorem 3.2 to prove this case.
Assume now that there exists (3 a contained in essential spectrum Decompose:
N ~Ni ф (a + K2) Ф W3 Ф (a + K4) © (a +K5)ф ... ф (a+K4n),
where N-¡,N3 are normal, K2,K4,... ,K4n are compact, each of them acts on an infinite-dimensional Hilbert space /С and:
/3 £ &ess(Ni) О СГе.чз(А^з).
Put:
Ti =N\(a + Кз)(а+Jfg)... (a + К4п_з)
ф(а + K2)(a + K§)(a + /Сю)... (a+ К4п_2)
=(a"_1M +/ф©(ап+*7),
Т2 =N3(a + К7)(а + Кп)... (a + K4n-i) ф (а + /С4)(а + Kg,) (а+ К42)... (а + К4п)
={ап~1 N3 + К'2) ®(ап + K'f).
Observethat neither T\ nor T2cannot beof the form X+K for some compact operator K. If we suppose that then computing essentialspectrum:
{A}=aess(T1)D{an,an-1^},
similarly for T2. It is obvious contradiction to the fact that (3 / a. Thus from Theorem 3.10 follows that operator T\ ф T2 is commutator. Theorem 3.2 gives the
way offinding adequate n-roots. □
The last twotheorems refer to the case when essential spectrumcontains at least two points. Now it’s time to consider the simplest situation: essential spectrum is just asingleton.
Corollary 3.12. Let N = a + K G B(?Z), where K is compact and |o| / 1 be normal, invertible operator. Then for every naturaln > 2 operator N is aproduct of four n-roots and at least two ofthem are unitary operator.
Proof. Notice first that a 0 since any compact operator is not invertible. Let at G C : wn = 1. Standardprocedure gives:
N~ (a + Ko) ® ... © (a+ G B(/Cn).
Now define:
M =Z © w_1Z © w_2Z© ... © w_n+1Z,
N2 =(a + Ko) ©u>(a + ZCi)®u>2(a + K2) ® ... © wn 1 (a+ Kn_i).
One can easily check that N1 is n-root of the identity. As for N2, it comes from Lemma ?? that one is a product of three n-roots (one of which is unitary). To see this compute essential spectrum:
Ccs.^M’)= {o,cua,iv2a,...,wn_1o} C {zG C : |z| = |o|}
and observe that card(cre.,s(X2))> 2. MoreoveryViX2 = (a + Z<o)®- • .©(a+ZG-i) = N which finally impliesthatN is aproductof four n-roots of theidentityand at least
twoofthem are unitary. □
Example 3.13. Every invertible normal operator N = a+ K G B(H), where |a| / 1 and K is compact cannot be a product of three 3-roots ofthe identity. Supposethat there exist three 3-rootsX, Y, Z such that a + K = XYZ. It implies that:
aZ~r+ KZ-1 =XY, and then:
a3+ K' = (XY)3,
where K' is compactoperator. Observe that for any pair ofoperators X, Ywe have formula:
(
xy) 3 = [[x,
y],
yx2 ]
yx3
y2 ,
where [X, Y] denotes commutator XYX~1Y~1. Immediately we get that if X3 = Y3 = I then:
(XY)3 = [[X,Y],YX2],
thus a3 + K' = (XY)3 is a commutatorwhat is impossible (x) .
Theorem 3.14. Let N =rU ©FG B(H®J-) be aninvertible normal operator where U is unitary, 1 / r > 0 and Hilbert space F isfinite dimensional. If there are no isolated points in spectrumcr(U) then N is a product ofthreen-roots for every natural n > 2.
From Theorems?? follows general theorem for normaloperators:
(') If T is commutator and T = X + K, where K is compact operator, then |A| = 1.
42 PIOTR BUDZYŃSKI
Theorem 3.15. Let N be an invertible, normal operator and n > 2 be natural. Iffor some a € C, such that |a| 1, and some compact operator K we have N — a+ K thenN is aproduct offourn-roots of the identity and atleast two of them are unitary.
In opposite case we can decompose N into a product of threen-roots, one ofthemis unitary.
Use the polar decomposition toobtain:
Theorem 3.16. If T is an invertible operator and n > 2 is natural number then T is a product of five n-roots of the identity. At least three of them are unitary.