This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
- = =
This =>
T(IX)=O
¥
CIt
. .. (3)
E V=> T=O".
In either case 1'= O. . Theorem 29. Let T be a linear trans/ormation on Hilbert space V(F). Show that
T-O <:>
Let
11>=<0(2), {J)=(O, r;>=o.
V ex, (J e V.
To prove T=O. In particular (I) => (T(2o), T(at»=O, where .. T(Ot)=O
V
at
e
. .. (1)
~=T(Gt)
Y
• T=41
Theorem 30. Suppose T is self-adjoint operator on Hilbert space V(K).
393
HILBERT SPAC8
Then S*TS is self adjoint for el'ery linear operator Son V. Further if S is invertible and S*TS is self-adjoint, then T is also selfadjoint. [Kanpur M.Sc. F. Linear Algebra 1976J Proof. Let T and S be arbitrary linear operators on a Hilbert space V(K). Step (i). Let l' be self-adjoint so that T = T*. To prove that S*TS is self-adjoint (S· TS)* = S*T* S, *= S*P"S= S*TS (S'1',S)*=S*TS. From this the required result follows. Step (ii). Let·S be invertible so that S-1 and (S*)-1 exist. S iTS is self-adjoint. => (S* n)* = s* TS. ~ S*T'*'S';'~=S*TS ~-S*'T*S=S*TS
S*T*(SS-I)=S*T(5S-}) (S*)-1 S*T*I=(S*)-1 S*TI ~ IT*I=ITI ;> T*=" T::> l' is self-adjoint. Theorem 31. Every linear operaTor T 011 a complex Hilbert space is uniquely expressible as 1'= T} + iT2 , where Tl alld T2 are Hermitian (self adjoint) operators alld i= v( -1) fKanpur 1991;86] ::>
::>
Proof. space.
Let T be a linear operator on a complex Hilbert
Evidently
-1'*)
1
T~=~i (1'-1'*), we
get
T=i (T+T*)+;
Taking T)=1 (1'+T*),
Gi
(T
T~, T 1 +iT2 • Remains to prove that Tl and T2 are Hermitian operators
on V. T/I<=1 (T+T-.\O)*=' (T*+T)~' (1'+1'*)=1'1' I 1 I T 2*=-i{ (T--T*)*=-2i (T*-T)=2; (T-P)=1'z• :. Tl·~Tl' T'1,*=T2This => T) and T2 are Hermitian. Uniqueness. If possible, let l' =P+ iQ be anol her representation where P*=P, Q*=Q. Then r*=(p+iQ)*=P"'+'iQ*=P-iQ
394
HlLBEIlT
N
T=P+iQ l:)Ti-l'*=p T
ow T"*'=P-iQf
2
~PACE
T*' =Q
'2;
~ T 1 =P, T2 =Q· Hence the representation T= T1 + iT2 is unique. Theorem 32. A necessary and sufficient. condition that Q linear operator T on a (omplex Hilbert Space V (unitary space) be selfadjoint (or Hermitian) is that
V(F). (i)
Let T be self-adjoint so that T=T*. To prove that
or
(T{CIt), (I.)=
ThIs
~
(I)
. .. (1)
.. (T(<<), a)=(T«(I.), «)=(<<, T(a» =(T*(at), «) => (T(<<), Q[)=
(T(CIt), a)-
.. (T(IX)-T-!«(a), cr)=O ~ «(T~T*)
(a), «)=0
=> T-T.==O T=T*. Theorem 33. Let U be a linear operator on a Hilbert space V( K).
Then the following are equh·alent.
U*=U-1, UU*=U*V==I, (3)
(I) (I)
0:>
(2) ~ (3) ~ (4) •
~l).
~ (1).
U·=U-1". VU=*UU- 1, U.U=U-1U
HILBERT SPACE:.
395
=> UU*=I. U*U=! => UU*=U*U=I. (2) => (3).
UU*=", U·V=I => < (U*V) (u), v>=
.: I(u) = "
~
=-
(1)
II V(u) Ii=il u!!
~ ii U(U) 1:2=11 u !12 => ::0- «U*U) (U), U>==(U), => «U*V-/) (U), U> =0 =>
U*U-I=O
u":;
By Th. 21.
~ U*U=I+O=1 ~ U*UU-l=IU-1 => U*/=U-1 .. U*=U-1. Corollary.
U is unitary ¢o II U(u -VI II = II u-v II ..;:. distance between the vectors u and V =distance between U(/I) and U(v) <* U is an isometry.
Ex. 10. Suppose T is unitary (orthogonal operator on a Hilbert shace Vand W is a T-im·ariant. Then _.how that W.l. Is also T·invariant. Solution. Suppose T is a unitary operator on a Hilbert space V(F) so that TT*=T*T=I. Let W be T-invariallt so that We v any I\' E W => T~ 1\') E W. To prove that Wi jc; T-invariant. it is enouguh to show that any WE Wi=> T(lr) e Wi.
Let
It'
E
Wand u E Wi be arbitrary; then E W.
T~u), T(II'»
=
:'196
HILBERT SPACE
Thus, < T~u). wl > = 0 ¥ wl E W. This * T(u) 1. W => T(u) E W J.. :. any u E WJ. => T(u) e WJ.
Proved.
Theorem 34. (a) let T be a linear operator on a complex lIilbert space V. Theil T is normal - 11 T*(u) 11= II T(u) II vuE V. [Deihl 1980; Bundelkhand 81; Meeurt 80J Proved. Suppose 7 is a linear operator on a complex Hilbert
space V. (i) T is normal => TT*= T*T. «rr*) (u), u)=«1 *T)(u), u) .. (T(T*(u», u)=(1fr(T(u)J, u) => (T*(u), T*(u»==(T(u), T**(u» => " T*(u) lill=(T(u), T(u»=I: T.(u) 1111 => II T*(u) 11=11 T(u) II· (U) Again 11"'(u) Ii.eli T(II) II. II T*(u) 112=11 T(u) III
=> (T"(u), T*(u»=(T(u), T(u» => «TT*) (u), II)=«T*T) (u), u) u>=O. Also V is wtnplex Hilbert space => 1T*- T*T=O (Refer Theorem ':J.7) => TT*=T*T => «TT*~ T"'T) (u),
Proved, T is normal Theort:m 34. (b). If T is a normal orerafor on Hi/bert space H, thm II TlII= 1/ T 11 2 • Proof Let U E /l be arbitrary. Then T is a normal operator => I: Tu I/=I! T*u /I .. (1), by Theorem 34a. II 12 11== SIJP iii 1'u !I : 1 Ii I ~ J} =sup (i T(Tu) 1/ : I II I! ~ I}. using (J) "..:sup {II T*(TII) Ii : iI u ii ~ I} =t;UP {I! (T*7) (u' Ii : i! Ii I! ~ I} = :! T*T 1i=I!T III, by Theorem 25. Theorem 35. Suppose T is a linear operator on a Hilbert space. Then T Is normal iff its real and imaginary parts commute. [Delhi 1981; Kanpur M.Sc. F. Linear Algebra 67, 77J Proof. We know that every linear opetator T on a Hilbert space is expressible as T=A+iB, where A=i (T+T*), =>
B=~ (T--T*)
are self-adjoint operators so that
A*=-·~.·f
B=B*.
397
HILBERT SPACE
T*=(A+iB}*=A*+TB*~=A iB. TT*,,,-(A+iB) (A-iB)~"'A2+ B2 +j (BA·-AB), T*T=(A.-iB) (A+iB)c--=A2+B2+i (AB-Bo4). T i" normal ~ TT*~-~ T*T --. A2 +-B2+i (BA -AB)'==A2+B~+i (AB-BA) e> BA - AB:-·=AB-BA ¢ 2AB=2BA <=> AB=BA <=> A and B commute ¢¢> real and im:lginary part') of I commute. Ex 11. Show that the set of all unitary operators on a Hilbert space V(K) is group under the composition of operators. [Kanpur M.Sc. F. 1985, 82; Meerut 69, 70] Solution. Let G denote the set of all unitary operators on a Hilbert space V(K). Let T, S, FE G be arbitrary so that T, S, F
are unitary operators on V(K). Consequently
, Tr*=T*T=I, SS*=$"*S=I. Firstly we shall show that TS is a unitary operator. (TS) (TS)*=(TS) (S*TI(:)=T(SS*) T*'=TIT* =TT*-=/ (TS)· (TS)=(S*T*) (TS)==S* (T*T) S=S*}S =S*S=/ (TS)* (TS)=(TS) (TS)*=I. This
~
1"S is a unitary operator. E G.
=> TS
(i) T. S E G -> TS E G. (if) 3 identity element I (identity operator)
Tl=IT=T. 1/*= 1=1*/ (iii) Every element T E G has its inverse For TT*=T*T=I ~ T*=T-1. Also T-I (T-1)*=T* (T*r 1
-= G S.t.
For
=1.
Similarly
(T-1)* (T-l)=i. Thus (T-1) (T-1)*=I=(T-1)*, (T-1) This::;. T-1 is a unitary operator ::;. T-l E G. (iv) (TS) F= T(SF). For [(TS) FJ (u)=(TS) [FlU)), uE V =
T lS(F(u)}]
j1
E G.
398
HILBERT SPACE
= T [(SF)(/I)] =[T(SF)](u) [(TS)F] (u)=[T(SF)](u) ¥ u e V. This => (TS) F=T(SF). The results (i), (ii), (iii) and (iv) taken together imply that G is a group w.r.t. operation of product of operators. E~. 12. 1fT is an arbitrary linear operator on afinite dimensional complex Hilbert space and if a and b are complex numbers such that, a 1=1 b 1=1, then aT+bT* is normal. {Meerut 1971; Kanpur 88] Solution. Let I a 1= I b I= I so that laj2=/bI 2 =1 or aa= bjj = 1 (aT+bT")*={aT* +bT*)*=aT""'+ bT (aT+bT*)*{a r +bT*)=(aT""'+bT) (aT+bT*) =aaT*T+ab (T*)2+baT2+bbTT* =T*T+ab (T*)2+abT2+TT* ... (1) (aT+hT*) (aT+bT*)*=(aT+bT*) (aT*+bT) =aaTI*+abT2+ba (T*)2+bbT*T =TT*+abT2+ab {T*)2+T*T =T·q+ab (T"",)2+al)TI + TT.. ...(2) Equating (1) to (2), we get (aT +bT*) (aT+hP)*=(aT +bT*)* (aT+bT*). This ~ aT +hT'iI< is a normal operator. Ex:. 13. Let T he a normal operator on a Hilbert space and fa polynomial with complex coefficients. Then the operator f(T) is normal. [Kanpur 1987J Letf=ao+a1x+a2x 3 + ... +a",x"', then /(T)=ool+a 1T+a 2 TJ+ ... +o",T"', (T*)A=T*T* ... T* (k factors) =(TT... T k factors)* =(Tk)*. Hence (T*)k=(Tk)* [f(T)]*=(aol+a1 T+ ... + a", T"')* =dol+d1T*+ ... +d", (1"")* [J(T)]*=dll+o1T*+ ... +a... (T*)",. Tis norm:!1 ~ TT*= T*T => T(T*f=T[T*(T*,)t-l]=(TT*) (T*YH =(T*'T) (T*) ..-1
Solution
or
... (1)
... (2)
.. (3)
399
HILBERT SPACE
=T*[T(T*)·-IJ= ...
= TX<[Tiit1 (T*)k-S] ~
T(T'*)"=(Ti'/<)2
[T(T;\~)a-2]
=(T:Jt:)t T T(T*)"=(T*" T. Similarly we can show that T"T*= T:Jt:7". By virtue of (3) and (4). it is easy to show that ~
.. (4)
[f(T)) [f(T»)*=[f(T)]* [f(1')]· f( T) is a normal operator. DeC. Projection If P is a project:on with range space M and null space N, then This
=-
M={x E H; Px=x}, N={x E H; Px=O}. Here we also say that P is a projeJtion on M along N. Also H=M~N.
A projection P on a Hilbert space H is called perpendicular projection if M and N are orthogonal. In this case N .L M and so N=Ml or M==;Nl.. Thus a perpendicular projection P on M is a projl!ction on AI along Ml. rn the theory of Hilbert space, we shall consider only tho~e projections which are perpendicular. Hence we omit the repeated use of the word perpendicular. Thus by a projection P on n, we shall always mean a perpfndicular projection on H. An operator P on a Hilbert space H is a projection iff P2=P,
p=p* Also Px=x ¥ x E M. Important result. (i) Px =X \rI x E M (ii) II Px II ~ Ii x \I (iii) pa=p*=p (iv) (Px, x)= II Px III. Theorem 36. P is a projection on a closed linear subspace M
on H ~ I-P is a projecTion on Mi. Proof. Let P be projection on a closed linear subspace M of a Hilhert space H. Aim. I-P is a projection on Mi. We have pZ=P*=P, Px=x >.,f x E M. (1_ p)I=/S+P2_IP-P/=J+P_P_1' =I-P (/-P)*=I*-P*=l-P Let N be range space of-I :-P.
400
HILBIlRT SPACE
any x EN.¢> (1-- P) (x)=x ¢> J(x) -P(x)=x <=> x-P(x)=x ~ --P(x)=O ~ P(x)=O ~ x is in the null space of P. ~ x E MJ.
.•.
N=MJ. •
Finally, we have proved that (l-P)2=I-P, (/-P)*=l-P, (I_·P) (x)=x ¥ x E Ml.. These facts prove that 1- P is a projection on MJ.. Conversely. Suppose that J-P is projection on MJ., where M is a closed linear subspace of H so that MJ.J.=M. It follows from the first part that l-(l-P) is a projection on (M.l) 1 =MJ.J. i e., P is a projection on M. Problem 14. If P is a projection on a closed linear subspace Mof H, then x E M ~ P(x)=x 4> " P(x) /1=/1 x II Solution. P is a projeetion on M ~ p2=P*=P I. Let Px=x where x E H. x E H ;... Px E M as M is range space => x E M as Px=x. Thus Px=x ~ x E M Conversely suppose that x E M: To prove Px=x. or
Let Px=y, then PPx=Py or Py=P 2x=Px Px=Py or Px-Py=O or P(x-y)=O This
~
x-y is in null space of P x-y E MJ. => 3 z E Ml. S.t. x--y=z ~ x=y+z. Px EM:$> Y E M as Px=y Now x=y+z with y E M, z E Ml. where H=MffiMJ.. Also x=x+O where x E M (given) and 0 E MJ.. Thus x+O=x=y+z Also x has unique representation. Comparing both sides, z=O, x=y. But y E M and so x E M. Thus x EM=> Px=x Finally x E M <:> Px=x. II. Aim. Px=x <:> " Px 1/=11 x" Obviously Px=x => /I Px !/=Il X" ...(1) ~
401
HILBERT SPACE
Conversely suppose that II Px
li=1/ x 1/.
••• (2)
To prove that Px=x. Evidently II Px+(I-P) x 11=11 x II. ...(3) But Px E M and (I-P) (x) E Ml.. This ~ Px .l (I-P) (x). By Pyttlagoras theorem. I Px+(l-P) (x) 111=11 Px 111+11 (I-P) (x) III using (2) and (3), II x 112=11 x 111+ ,,(/-P) (x) liZ or II (I-P) (x) !I =0 or "x-P(x) li=O or x-P(x}=O or Px=x. . .. (4) Finally x=Px <:> II Px 11= II x II, by (I) and (4). Theorem 37. (Projection Theorem). q P is a projection on a Hi/bert space H with range space M and null space N, then M .L N <:> P is self adjoint [Meerut 1973; Kanpur 84, 81] and in this case N=Ml.. Proof. Let P be a projection on a Hilbert space H, i.e., P is a projection on M along N so that H=MffiN, and z E H is uniquely expressible as z=x+y with x E M, YEN. Also Pz= P(x+y) =x. I Let M I. N, then (x, y)=O=(y, x). Now (Pz, 'z)=(x, z)=(x, x+y)=(x, x)+<x, y) =(x, x)+O=I! x Ill. . .. (1) Also (P*z, z)=(z, Pz)=(z, x)==(x+y, x) =(x, x)+(y, x)=(x, x)+O=il x II·. ... (2) Equating (I) to (2), (Pz, z)=(P*z, z) or «P -P*) z, z)=O yo z e H. This 0 P-P*=O ~ p=p* 0 P is self adjoint. U. Let P be self adjoint. To prove M .l N. x E M~ Px=x, By def. of M. y E No Py=O, by def. of N. (x, y)=(Px, y)=(x, p·~y)=<x, Py)
HILBERT SPACE
402
=(x, 0)=0 (X, y)=O x E M, YEN.
:.
Also This
=> M .L N.
... (1) Let M.l N. To prove N= Ml. x E N ~ x..LM ~ x E M.l. any . .. (2) /. Ne Mi. . .. (3) Suppos e (I) is not true, then N~M.1.. By (2) and (3), N is a proper subset of Mi. ~ N is a closed subspac~ of Hilbert space M.l. {For N is null space ~ N is closed] zo-:cO E Ml. S.t. zo...LN. ~ MJ.. => zoA..M. Also zo.lN. E Zo Now
III.
a
Hence we can conclu de zo.lH as H=M( JjN. This is possibl e only if zo=O. A contrad ication .
N=M.1.. Problem 15. If T is an arbitra ry linear operator 011 afinite dimensional comple x Hilbert space and if a al1d b are complex numbers such that I a 1=1 b I, /hen aT+b. * is normal. (Meeru t 1971]' Solution. Let I a / :-~ I b / so that I a 12~1 b /2 or ad=bli . (a r +bT-iF)* =dT*+ b(T*)* =dT*+ bT (aT+b T*)* (aT+b T* =(dr* +iiT) (aT+b T*) ... (1) =daT* T+dh (T*)2 +b'1r2 +bbTT * liT) (dT*+ T*) ~aT+bT*) (aT+b T*)*= (aT+b =aiiTr *+abT 2+bd (T*)2+ bbT*T . Using ad = bfj, in first and last term, (aT+b T*) (aT+bT *)*=bb TT-'F+ afjT2+ bd(T*) 2 . .. (2) +adr* T. get we (2), to Equati ng (I) (aT+b T*) (aT+b P')*=( aT-t-b T*)* (aT+bT~). Thi'l ~ aT+br *' is a normal opera.t or. Proble m 16. A closed linear subspace M of a Hi/bert space H is invariant under an operator T <:> MJ. is invariant under T". Solutio n I. Let M be invaria nt under T. To prove that M.l. Hence
is invaria nt under T'*'. Let
XE MJ..,
yE M, then (x, Y) =0.
HIL8ERT SPACE
403
By def. of adjoint, (T*'x, y)=(x, Ty)=(x, Yl)=O where Yl=TyEM. Thus (Tx:'F, y)=O. Also yEM. Hence this declares that T*'xeMJ.. Thu,; xEMl. ~ T*xEMJ.. :. MJ. is T*-invariant.
II. Let M be a closed linear subspace of H so that MJ.J.=M. Let Ml. be' invariant under T*. Aim. M is invariant under T. It follows from the first part that (Ml.)J. is invariant under (T*)* i.e. Ml.l. is invariant under T** i.e., M is invariant under T. Theorem 38. If P is a p:rojection on closed linear subspace M of a HUbert space Hand T is a linear operator on H. then , (Meerut 1987. 86) M reduces T <:> TP=PT. Proof. M is a closed subspace of H => H=M(f)Ml.. Suppose P is a projection on M along Ml., Pz=x, Px=x, Py=O then ... (1) where any zE H is uniquely expressible as z=x+y with xEM, yEMl.. I. Let M reduce T so that M and Ml. both are invariant under T. To prove TP=PT. By assumption, Tx E M, TyE MJ.. . .• (2) (PT) (z)=(PT) {x+y)=P [Tx+Ty]. Take TX=X1EM. Ty=y,EMl.. (PT) (z)=P(x1+Yt)=:=PX1+PYl=X1+O=X1 ••• (3), by (1) .. (4) (TP) (z) = T(fz) = Txd.x 1 by (1). Equating (3) to (4), (PT) (z)=( TP) (z). This => PT= TP. II. Conversely let TP=PT. To prove M reduces T, i.e., to prove M and Ml. both are invariant under T.
404
HILBERT SPACE
P (Tx)=(PT) (x)=(TP) (x)=T(Px)=Tx, by (I) P (Tx)=Tx, this::;. Tx E M P (Ty)=(PT) (y)=(TP) (y)=T(Py)=T(O)=O, by (I). P (Ty)=O, this TYEMJ. ••
xEM => TxEM
YEMJ. => TyEMJ..
Hence M reduces T. Theorem 39. Let El and E2 be perpendicular projections on U1 a71d U2 respectively, where U 1 and Ua are closed subspaces of a Hi/bert space V(K).· Then sholl' that El and £2 are orthogonal <=> U1 and U2 are orthogonal. E 1E 2 =0 or E2 E1 =0 <=> U}.1U2• [Kanpur M.Sc. 81, 78} Proof. Suppose E} and E2 are perpendicular projections on U1 and U2 respectively which are subspaces of a Hilbert space V so that E}~=El=El* E22=E2=E~*
'
"IE VI L 1:.~(U2) ="2 ~ "2 E U2 Let E1 and Ea be orthogonal so that E1(U1)
Step (i)
}
=U1
<=>
r
E1E~=O, E2 E1=0.
.. (I) ... (2) ...(3)
Let "IE VI' U2 E U 2 be arbitrary. To prove that VI and V,. are orthogonal, it is enough to show that ".,)=0.
<"1'
(u1o
ul E VI ~ E1(U1)=U1 }, by (2). u.E V 2 => E 2 ("-4)=U 2 u2 )=(E1(U 1 ), £2(U2 )=«E2 * E 1 ) (u 1 ), u2 ) =«E2E1) (U1), u2), by (I).
=(0 (u 1 ), u~), by !3) =(0, u2 )=0. Step (ii) Let VI and V 2 be orthogonal so that vd.U2 => U1 =U.J. ... (4) To prove that E1 and E2 are orthogonal, it is enough to show
that
H1LBER T SPACE
40S
Let U 1 E VI be arbitrary, then UtE U2 .L, by (4) "lEUI :::> £1(U1)=U1 :::> E1(u1)eUt =U2.L => £1(u 1)EU2.L => E1(U1)¢U2 For U2 nU2J.={0}. => £2(£I(U1»=0 For u¢U2 => E2(u)=0 => E 2E 1(U1 )=0 ¥ ul E U1
=> E2E 1 =0 => (£2£.)*=(0)*=0 :::> El*E2*=0,. E1 £2=O, by (1). Pr,oved. Theorem 40. If P is a projection on a closed linear subspace M of H, then M is invariant under an operator T ~ TP=PTP. Proof. Let P be the projection on a closed subspace M of H so that I'X=X ¥ xEM ... (1) ze H =>' PZE M. and ... (2) I. Let M be invariant under an operator T, then xEM => TxeM ... (3) Aim. TP=PTP ZEH:::> PzEM => T(PZ)EM, by (3) => r(pz)=pz ... (4) Further (PTP)(z)=aP [r(pz)J=P(Pz), by (4) _p2Z =PZ as PI=P -T(Pz), again by (4). =(TP) ,z) :. PTP=TP. PTP=TP n. Let .. ,(5) To prove that M is invariant under T. Let xeM, then Px=x. Now (5) gives (PTP) (x)=(TP) (x) or (PT) (Px)=T(Px) or (PT) (x)= Tx or p(Tx)e::. Tx This :> TXEM, according to (I). Thus xe M => TXE M This proves that Mis T-invariant. Theorem 41. If Pl' P 2• .. , P.,. are projections on closed linear sub,\p~lces M 1 , M 2 ,··., Mn of H, thell
406
HILBERT SPACE II
p= ~ P, is projection ~ P,'s are pairwise orthogonal
,-,
and in this case P is a projection on M = M J + M2 + ... + M ". lKanpur 1981, 78J Proof. Let P, be projection on closed linear subspaces of M, of H, then ... (1) P,I=P,=P,· and Pi (X,)=x, ¥ x, E Mt
Also let P*=(
or I.
P=PI +P2 + ... +P,,= "~ P, i_I
~ P,=P '£ _1 p,)* = '=£1 Pt*= .-1
P*=P Let P,'s be pairwise orthogonal so that
p,p,=O for i::pj Ai.... P is a projection. Then
P'==(
£ P,
=~
... (3)
)1 =~
P.2+2
~
i::pj P.+O=P, by (1) and (3)
,i-l
i
PiP,
Finally, J'2=P, p=p* This pr~es that P is projection. 11. Let P be projection. To prove that PIs are pairwise orthogonal. If T is any projection and zE H, then (Tz, z)=(Tlz, z)=(TTz, z)=(Tz, T*z) =(Tz, Tz)=1i Tz iiThus T is projection ~
nx 112=\1 P,(x) IP' ~ 1_1 ~" or
... (2)
... (4)
II P,(x) 11 2 = E" (PI x, x), by (4)
'_I
==(P1 x, x)+ ... +
407
HILBf:RT SPACE
From this we conclude that the sign of equality must hold throu~hout the above calculation. Hence to
II P,x
.
W= 1-1 ~
II P,(x) 1/2
~
II PJ(x) ill=O ¥ j s.t. j=l=i if j::p.i => x E Null space of P1 => x € M,L ifj¢:i => x J..MJ ¥ j s.t. j#i where x € M, This proves that M,.1. MJ for i:j:i
This
= Pj{X)=O
This ~ p,P,=O for i#}. Ill. Finally we have to prove that to
P is projection on M = ~ M, J-1
For this we have to prove that R(P)=M. any x E R(P) => Px=x
~ ( ~ P,
) (x)=x
x=P1X+P 2 X+ .. +p"x But PIX e Ml • PzX e Ma etc. Therefore x E M1+M.+ ... +M,,=M Thus x E R(P) => x € M o
.'!. R(P) C M ... {S) We have proved that II py 11=11 x II ¥ X € M, and each M, is contained in the range of P and so M is also contained in the range of P.
:. Me R(P) By (5) and (6). R(P)=M This completes the proof.
Problem 17.
... (6)
If P and Q are prujections on closed linear sub-
SpUCf:S M wnd N of H, 'pnJl'e that PQ is projection ... PQ=QP
n N. [Kanpur 1981, 78; Meerut 88.87] Solution. Suppose P and Q are projections on closed linear subspaces M and N respectively. Then P2=P=P*, Q2=Q=Q*. ...(1) In this case, show that PQ is projection on M
408
HILBERT SPAce
Px=x ¥ x E M Ilnd Qx=x ¥ x E N ... (2) Step I. Let PQ be a projection. Aim. PQ=QP. By assumption, (PQ)*=PQ or Q~p·=PQ Using (I), QP=PQ. Step II. Let PQ=QP ... (3) To prove that PQ is projection. (PQ)2:a(PQ) (PQ) = (PQ) (QP), by (3) =p(Q2) P=PQP, by (I) =P(QP)=P(PQ), by (3) =P"Q=-PQ, by (I) or (PQ)I=PQ ... (4) (PQ)*=Q*P*=QP, by (I) =PQ, by (3) or (PQ)*=PQ .. (5) By (4) and (5), PQ is projection. Step III. To show that PQ is projection on Mn-N For this we must show that R(PQ)=MnN and x EM n N ~ x € M, x € N ~ Px=-x, Qx=x => (PQ) (x)==P(Qx)=Px=x ~ (PQ) (x)=x ~ x € R(PQ) :. M nNe R(PQ) ... (6) • J Agam any y € R(I-Q) =-> (PQ) (y)=y ... (7) => P((PQ) (y)]==Py ~ (P2Q) (y)=Py =-> (PQ) (y)=Py, by (I) Usjn~ (7), we get Py=y, this => Y E M Using PQ=QP in (7), (QP) (y)=y ... (8) This ~ Q[(QP)y]=Qy ~ (Q2P) (y)=Qy ~ (QP) (y)=Qy ~ Qy=y, by (8) ~ Y € N. Thus y E R(PQ) =-> y E M, YEN ~ y E M n N :. R(PQ) C M n N ... (9) By (6) and (9), R(PQ)=M n N
This proves the required result. Problem 18. .If P is projection on (i) P is positive operator, i.e., P (ii) O~P~I (iii) II Px II E; II x II ¥ X e H. (iv) II P II ~ 1.
Solution. P is H.
proj~ction
on a
Q
Hilbert apace H, then 0
~
subspa~e
M of a Hilbert space
409
HILBERT SPACfl
=> PX=p "". and To prove (i)
X
p2=P*=P Let x E H
E M
... (1) ... (2)
=> PXJ...(I-P) (x)
By Pythagoras theorem, " Px+(I-P) (x) 1!2= II Px 1i1+1i (I· x) (x) 1111 or I x 1i 2 =1t Px 112+1i x--Px !l'! ~ " Px j:~ or II x 112 ~ II Px liZ or II Px II ~11 x II To prO\'e (i\') By def. " P ii-sup { II Px II: /I x II <;; I} But II Px II ~ !! x /I , by (iii) part II P II ~ su p { II x II : II x \I ..; I} ~ I or II P II ~ 1. Problem 19. If Pond Q are projcctiultJ on closed linear subspaces M and N of a Hilbert space H, thell 5how that tile folluwing are equivalent: (I) P ~ Q. (Ii) II P,'{ II ~ II Qx II 'V x e H (iii) MeN. (ill) QP=P. (v) fQ=P. [Meerut 1980J Solution. In order to prove the Problem, we shall prove that (i) => (ii), (ii) => (iii), (iii) => (iv) , (iv)~ (V), (v) => 0)
or
410
HILBERT SPACE ~
(ii) P ~ Q ~ Px ~ Qx ~ < Px, x > <; < Qx, x > => \I Px 112 II QX 11 2, by (I) => II Px II ~ /I QX " (ii) '"" (iii) '>.f x EM==- Px=x ~ II Px 112=/1 X 112 -::> II X 1/=/1 Px II EO;; II QX H=> II X II II QX II ... (2) But II QX" E; " X II ... (3) by Problem 18. Bv (2) and (3), II QX 1:=1/ X II This => Qx=x, by Problem 14. =>xEN 1 hus any ~ EM=> X E N : . MeN. (iii) => (iv). It is given that MeN. Px E M=range space of P. But MeN. :. Px E N and so Q(Px) E N. This preves that Q(Px)=Px. For yEN => Qy=y or (QP) (x)=Px ¥ X E H :. QP=P. (iv) ~ (v). It is given that QP=P. This ::;> (QP)*=P* => P'~Q*=P* => PQ= P. For P and Q are projections. (v) => (i). It is given that PQ=P. Let x E H. Then by (I),
<
<
P"
<
<
411
HILBERT SPACE
(Q -P)*=Q*-P*=Q-P (Q_P)2=(Q _ P) (Q_P)=Q2+p2_PQ_QP =Q+P-PQ-QP=Q+P-P-P=Q-~
For P ~ Q => PQ=P, QP=P, by problem 19. Finally, (Q_P)2=Q_P, (Q-P)*=Q--P. This proves that Q-P is a projection. Second Part. To show that Q- P is a projection on NnM.1.. For this we must prove that R (Q-p)=NnM.1. and x E R (Q-P) _.. (Q-P) (x)=x ... (1) => Q (Q--P) (X)= Qx "'" (QI_QP) (x)=Qx. But Qll=Q; QP=PQ=P => (Q-P) (x)=Qx => x=Qx, by (1) => xERtQ) => xEN Again by (I), P(Q-P) (x)=Px => (PQ_P2) (x)=Px => (P-P) (x)=Px :;> O=Px "'" x E null space of P => x C M.1.. Finally, x E R (Q-P) .. x E N, x E M.1. => x E NnM.1.. :. R (Q-P) C NnM.1.. . .. (2) Again any y E NnMJ.. "'" YEN, Y E M.1. .. Qy=y, Py=O .. (Q-P) (y)=y-O=y => y e R (Q-P). :. NnM.1. C R (Q-P). ...(3) By (2) and (3), NnM.1.=R(Q-P). This completes the proof. Theorem 42. If N is a normal operator on a Hilbert space H, then II N 1/ 2 = II N2 II. lMeerut 19801 Proof. Let x E H be arbitrary. N is normal':> II Nx 1/=11 N*x II ... (1), by TIl 26. Replacing x by Nx, we get II NNx 1/=/1 N*Nx II or II N2x II = II N* Nx II .. (2) . By def. II NI II =sup t II N2x II : II x II ~ I} =sup { \I N*Nx II: II x II .;;; I}, by (2) = IIN*NII or II NI II = II N* N n using the fact that II N 111= II N*N II is true for any operator N, we get
II NI \I
= II Nlil.
Pro"ed:.
412
HILB£;(;. T SPACE
Exercises 1. (a) Define a Hilbert space. Prove the following in a Hilbert space H : (i) I <x, y> I .;;. II x II • H y II Oi) XII- X. Yn-+)' => (X n , YII) ~ (x, y) where <XII>,
<
i
(b) Define normal and unitary operators on a Hilbert space H. Let T be a normal operator on a Hilbert !>pace il and p(x) a polvnomial with complex coefficients. Prove that p[TJ is a normal operator. [Kanpur 1987] 4. Define a positive operator r on a Hilbert space H. Show that it hHs on unIque positive sequare root. lKan, ur 19871 S (a) Define a normed linear spaces (NLS). Prove the following5 in NLS (N, II II). (i) / [I x II - II Y ii/ ~ II x-J II (ii) lf {Xn} be a sequence in N, such tllat "',,-X, then II Xn II ... II x {Iii) If {'''n}, {Y .. } are sequences in Nand {IX n }. {(3.. } are scal~r sequences such that x n- x, J,,'-~Y and Cl" .... CIt. Bn-B. then ex"x,,+(;")'n-:- x+r~.r. [Kanpur 1989) (b) Prove that the linear spa:e C lO, 11 ot all contilluoUS real functions defined 011 the dosed unit interval [0, 1] is a normed linear space (NLS) w.r.t. the norm oran element/defined by
.
11/11
=
but not a Banach space.
f: 11(t) I
dt, [Kanpur 19861
HILBERT SPACE
413
6. (a) Define a Hilbert space. If B is a Banach space whose norms obeys the parallelogram law and if an inner product is delin d on B by
4(x,
y)=iJ
x+y
l/a_11 x-y :il+i" x+;y 111-; II x-Iy II!
then prove that B is a Hilbert space. (b) If x and yare any two elements of a Hilbert space H, prove that
I (x, y) I ~ II x II . II y
"
Explain when does equality hold. [Kanpur 1986] 1. (a> Show that an operator T on a Hilbert space H can uniquely he expressed as T= T1+iTjj where Tl and T2 are self adjoint operators H. (b) If T is an operator on a Hilbert space H for which (Tx. x)=O for all x, then show that T=O. (c) Prove that every Hilbert space i<; reflexive. [Kaopur 1986J 8. (a) Define a Hilbert space. Prove the following in Hilbert space H: (i) I (x, y) I < "x II • II y II (ii> Xn-X, Yn-y ~ (x .. , Yn)-(X, y) where !xnl, {Y .. } are sequences in H and x, Y E H. (b) Let H be a Hilbert space and {e,,} be an orthonormal set in H. Prove that the following conditions are all equivalent to one another: (i) fed is complete (ii) x..L {ei} x=O (iii) x E H => x=l; (x, e,:) e, (iv) x E H => II x i I=l; I (x, e,) 12 LAaopur 1985] 9. (a> Define self adjoint operator on a Hilbert space. Prove that the self adjoint oper!ltors in 8{H) from a closed real linear subsp!lce of B(H) which c.:mtains the identity transformation. (b) Define normal and unitary operators on a Hilbett space H. Prov~ that unitary opera.tors on H form a group. (Kanpur 1985] 10. Define a Hilbert space H and prove that a closed convex subset C of H contains a unique vector of smallest norm. [Kaopur 1984) 11. (a) If M is a closed linear subspace of a Hilbert space H, then prove that H=M(£)Ml., stating all the theorems that are used in the proof. (b) State and prove Bessel's inequality. [Kaaput 1984] U. (a> Define the adjoint operator r-~ r* on B (H) and prove that
=-
414
HILBERT SPACE
(Tt T2 ):'«=T2*' Tt '1: II T* II 17 Til (iii) II T* Til = II Till. (i)
(ii)
=
(b) If P is a projection on a Hilbert space H with range M and null space N then prove that M J. N <=> P is self adjoint. [Kanpor 19K4] 13. (a) S:ate and prove the closed graph theorem. (b) Define the conjugate T* of an operator T on a normed linear space N and prove that the mapping T-+T*' is a norm pre[Kanpor 1983] serving of B (.TIl) into B (N*'). 14. (a) Define a Hilbert space and prove that a closed convex subset C of a Hilbert space H contains a unique vector of smallest norm. (b) If M is a proper closed linear subspace of a Hilbert space H. then prove that there exists a non-zero vector Zo in H s.t. zoLM lKaDpar 1983] IS. (a) Let H be a Hilbert space and/be an arbitrary functional in H*. Prove that there exists a unique vector yin H such /(X)=(x. y) Y x E H. (b) Define the adjoint operation T-T~ on B (H) and prove that it is one-one onto as a mapping of B (H) into itself. [Kanpar 1983] 16. (a) If {ej} is an orthonormal set in a Hilbert space B, then show that 1:: I {x. e,} 12 ~ II X 1!2 for every vector x in H. (b) Prove that a Hilbert space H is separable <:> every orthonormal set in H is countable. [Kanpor M.Sc. F. 198Z] 17. (a) Lei: B be an ordered b:lsis of a Hilbert space H. Describle how using this fixed basis B a matrix [T] can be assigned to each operator Ton H. Such that: (i) the mapping T-+[T] is a one-one mapping of the set of all operators on H onto the set of all matrices. (ii) algebraic operations can be defined on the set of all matrices in such a manner that the mapping T-+[T] preserves the algebraic structure of B(B). (b) Show that the dimension of B( H} is fll, where n=dim H. [Kanpor M.Sc. F. 1981] 18. Prove that a closed convex subset lof a Hilbert space contains a unique element of smallest norm. [Kanpur M.Sc. F. 1981]
HILBERT SPACE
415
J9. (a) Prove that P lo P20 •• P n arc projections on closed linear sub;paces M i , M 2 , .••. , M .. of a Hilbert space H, then P=PI+Pa+ ... + 'n isa projection ifand only if P,'S are pairwise orthogonal. When this is the ca'>' , show further that P IS the projection on M = MI + M2+'" + M ... (b) If P and Q are projections on closed linear subspaces M and N of a Hilbert space H, prove that PQ is a projection if and onlv if PQ=QP. In this case, show that PQ is a projection on M n N. [Kanpur M.Sc. F. 1981, 78] 20. (a) If P is projection on H with range M and null space N, then show that MJ.. N <> P is self adjoint and in this case N=ML. (b) Show that the unitary operators on H form a group. [Kanpur M.Sc. F. 1982] 21. Prove that a Hilbert space H is separable if and only if every orthonormal set in H is countable. [Kanpor M.Sc. F. 19R11 (a) If M is a proper closed linear subspace of a Hilbert space H, prove that there exists a non-zero vector Zo in H, such that zoJ.M. (b) Prove that if M and N are closed linear subspaces of a Hilbert space H such that M .lN, then M +N is also closed. [Kaupur M.Sc. 1980, 78] 22. (a) Let H be a Hilbert space. Prove that the self adjoint operators in B(H) form a closed linear subspace of 8(H). (b) If an operator T on a Hilbert space /J is such that (Tx, x)=O for all x in H, show that T=O. [Kanpur M.Sc: F. 1980, 79J "0
20 Finite Dimensional Spectral Theory 20'1. Der. Let T be a linear operator on Hilbert space H(K). Any scalar I. E K is called an eign ~alue of T if 3 non-zero vector v E H s.t. TV=AV. Such a vector v is called eign vector of T corresponding to the eigen value ". The following have the same meaning: proper vector, eigen vector, characteristic vector, latent vector. . Similarly the following have the same meaning: latent value, elgn value, proper value, characteristic root, spectral value. The set of eign values of an operator T is called the spectrum of T and is denoted by aCT). Let M denote the set of all veetors of H. s.t. Tv='\v i.e. M={v E H: Tv="v} Then M is called eign space of~. [Kanpur 1988, 85] 20'2. Spectral Solution. Let T be a linear operator on a Hilbert space H. If:l distinct ~ and non-zero pairwise orthogonal complex numbers projections PI' Pz, "', P .. S.t .
"10 "I' .. ..
... (1 )
and
;=1
Then the expression (I) is called spectral solution for T. 20'3. Invariant. Let T be a linear operator T on a Hilbert space H. A closed subset M of H is said to be T-invarlant or invariant j under T if any x EM=> Tx E M. 20'4. Reducibility. Let T be linear operator on a Hilbert ~pace H. If M is a closed subspace of H, then T is said to be reduced by M if M and MJ. both are invariant under T. If M is invariant under T, then we also say sometimes that T is reduced by M or M reduces T. Theorem I. Let A be an eigen value 0/ an operator T on a
FINdE DIMeNSIONAL SPECTRAL THEORY
417
Hilbert Space H. LeI M be the eigen space of A, then M is a closed subspace of H. Proof. We have M={x E H: Tx=Ax}. Let a, b be scalars. x, y E M ~ Tx=Ax, Ty=Ay => T (ax+by)=aTx+bTy as T is linear ==aAx+bAy=A (ax+by) => T (ax+by)=A (ax+by). ax+by E H ~ ax+by EM. This proves that M is a subspace M={x E H: Tx=Ax=Mx}={x E H: (T-M)(x)=O} =Null space of linear map T-M =c1osed space. For null space of linear map is always closed. Hence the result. Theorem 2. The eigen space of a linear operator T on a Hilbert space H is invariant under T. Proof. Let M be an eigen space of T corresponding to eigen Then M ={x E H : Tx=Ax} ;\ E K, x EM=> Ax E M. by def. of subspace ~ Tx EM, by def. of M .. x E M~ Tx E M. Hence M is invariant under T. Theorem 3. The eigen values of a self adjoint operator are a/l real. , [Kanpur M.Sc. 1974, 71] Proof. Let S be a self adjoint operator on a Hilbert space H (K) Let Tx=;\x, then A is an eigen value of T relative to eigen vector x. To prove A is real. ;\ <x, x)=
I
418
FINITE DIMEN5lONAL SPECTRAL THEORY
The prove T is self adjoint (Refer Theorem 3) ,\ (x, Y)=(AX, y)=( Tx, y) =(x, T*y) =(x, Ty) =(x, I-'y)
or
=p (x, Y)=/L (x, y)
(A-I') (x, y) =0
But Hence Tbeorem 5. Then
'\:F 1-'. (x, y)=O or x.1y. LeI T be a normal operator on a Hilhert space H.
Tu=,\u ~ T*U=AU, u E H [Kanpur M.Sc. F. Linear Algebra 1978, 76] Or Every vector of a normal operator T is also characteristic vector of T*. [Meerut 1988J Proof. Let T be a normal operator on a Hilbert space H, then TT* = T*T. The result will be proved in two steps. (i) Let x E H and a scalar II TtIC 1I1=(Tx, Tx) =(K, T* (Tx,)=('(, (T*T) (x» = (x, (Tf*) (x» (T*x, T*x)=11 T*x 112 This ~ \I Tx 11=11 T·x II. •.. (1) (ji) To prove T -aI is normal (T-al)*= T*-(a/)* =T*-aI·=T*-iiJ (T-al)(T-al)*={T-al) (T*-tii) =TT*.:aTI-aIT*+auI2
-a r
aT* + aul (T-:lI)*(T-all=(T*-aJ) (T-a/) = IT*
.. (2)
=T*T-aT*I-uIT+aaI2 =T*T-aT*-ar+aii, Equating (2) to (3) and noting Tr* -=T*T, we get (T-al) (T-a/)*=(T-aI,* (T-aI). This ~ T -aI is normal. (iii) Now we can replace T by T-AI in (I). Hence II (T -AI) (u) II = II (T -Al)* (u) II or
II Tu -AU II
= IIlT*-M I (u)
or
II Tu-,\u II
=
II T*u-fu II
II
... (3)
FINITE DIMENSIONAL SPECTRAL THEORY
419
This declares that II Tu-Au \I =0 <=> • T*II-AU H =0
Tu =A.u ~ T*u=Xu. Theorem 6 If T is a normal operator on a Hilbert space H. then eigen veclors of T belonging to distinct eigen values are [Kanpur 1988; Meerut 88] orthogonal. er
Proof. Suppose T is a normal operator on a Hilbert space H(K). Let
Tx=oU, Ty=p.y To prove
where p.-:FA.
xl.y
Tx=).x, Ty=p.y => T*x=Ax, T*y=j;)., by Theorem 5. ). (~, y)=(Ax, y)=(Tx, y)=(x, T*y)=('IC, p-y) =p. (x, y)
or (.\- p.) (x, y) =0 giving y)=O, this => xl.y. Remark. The above problem can also be expressed as follows: If Ml and M. are eigen spaces of a normal operator T relative to two ei!!en values ).10 ).2, then Ml and M. are orthogonal. Theorem 7. Supp',se T is a normal operator on a Hilbert space H(K). Let M be an eigen space of .\, where)' is an eigen value of T. Then M and Ml. both are invariant under T. [Kanpur M.Sc. F. Topology 1977, Linear Algebra 78] Proof. We have M={x E H : Tx=).x} ). E K, x EM=> Tx=).x.'\x E M as M is a subspace => Tx E M :. M is invariant under T. T is normal => Tr*-T*T T [T* x] =(17'*) (x) -(T*T)(x)-T*(Tx) = T*(.\)C) =).T*x .or TlT*x]=A (T*x).
<x
t
420
FIN1TE DI\fENSIONAl SPECTRAL THEORY
By def. of, M, this => T*x E M. Thus ¥ x EM=> , *x E M. This proves M is invariant under T*. Now M is invariant under T* => M 1. is invariant under (T*)* = T (Refer remark written just below this theorem) ~
M1. is invariant under T.
Hence the result. Remark. LeI M be invariant under any linear operator T so that
x EM=> Tx EM. To prove M1. is invariant undt-r T*. Let x E M1., Y E M. then (x. y)=O. By def. of adjoin{. (T*x. ?)=(x. Ty)=(x. YJ)=O ¥
where
Yl=Ty E M [For x E M1., YI ~ M => (x. YJ)=O]
Thus
(T*(x), y)=O
Also Y EM. Hence tbis declares that T* x E M L. Thus
•
M.l.
X
EM1. => T*x E M1..
ts T*-invariant.
Tbeorem 8. (Spectral Theorem for normal operator). Let T be an operator on a finite dimensional Hilbert space H. Let "1> be distinct eigen values of T, let MiJ M 2 • .... Mil be their eigen spaces; and let Pl. Pt , .... Pn be the projections 011 these eigen spaces. Then the following statemen!s are equivalent to one another:
"a..... ""
M/s are pairwise orthogonal and .'pan H. (ii) P;'s are pairwise orthogonal, (/)
n
n
I=E. PI, T=E "IPI, I-I
(iii) Tis numal.
Proof.
1_1
[Meerut 1988 ; Kanpur M.Sc. F. 83. 79. 77!
(i) => (ii). i e .• suppose (i) is true.
To prove (ii). By assumption.
H=M 1(£; M 2 (fJ ... (fJMn •
421
FINITE DIMENSIONAL SPECTRAL THEORY
Hence any x E H is uniquely expressible as n
X= E XI with XI E M, n-l
¥
I.
P,'S are projections on M,'s and Mt's are pairwise orthogonal, ~
P;'s are pairwi!>e orthogonal
... (1)
=> P;Pj=O=PjPI for l::I=j. P; is projection on M; => PI(X)=XI
... (2)
n
This => 1= ~ PI
... (3)
I-I
Tx=T (
i
XI )' =
I-I
i
TXI as T is linear
'-1
n
~
=
XI E MI
A;xl as
=>
TXI=}.lx,
.. (4)
From (I), l3) and (4), the required result follows. To prove (ii) => (iii), i.e., suppose (ii) is true. To prove (iii).
f AIP,r
Tt=[
=
i: ~I PI*= 1: -"I PI
as PI is projection => P,=P,*=P;I,
7' A;P) (
Tr*=( (ii)
:.
f
fx,P/)-f
... (5)
AlxjPIP,.
=> Pt's are pairwise orthogonal => P,P,=O for ;::I=j. TT*=E '\/5:; P;P,=E 1'\1 \2 P,2=E I '\; II Pi by (5).
Thus
;
;
TT*=E t
I Ai \2 Pi'
I
422
FINITE DIMENSIONAL SPE'TltAL tHEORY
T*T=E I A, \' PI'
,
Similarly
Hence TT*=T*T, showing thereby by T is normal. To prove (iii) => (i), i e., suppose T is normal To prove (i). T is a normal operator on Hilbert space Hand M/s are eigen spaces of T. => M/s are pairwise orthogonal, (by Th. 6). ...(6) • P,'S are pairwise orthogonal projections on M/s, (Refer Theorem 39 of Chapter 19) From (6), the first part of the r~sult is established. Write \ Then M is a closed subspace of H associated with projection n
"
P= 2' P, s.t. P,P,=P;P,=O for i¥=j.
'-1
.
T is a normal operator on H, M, is eigen space of T => M, reduces T 101 i, by Theorem 7. Also P, is a projection on M, • TP,= PIT of i, by Theorem 9.
TP=T ~ P,=E I
or
,
TP,=~ P,T=(~ ,
I
P,) T=PT
TP=PT, also T
IS operator on M. Hence M reduces T, by Theorem 38, Chapter 19. Hence M and MJ. both are invariant under T. Let T' be the restriction of T to MJ.. Then T' : ML-+Ml. S.t.
T'(x)=T(x)
¥
x E MJ..
... (1)
Let x be an eigen vector of T' relative to eigen value A, then T' (X)=Ax. ...(2) But T' is defined on MJ., (2) => x E M_L => T(x)=T' (x), by (I) using (2), we get T (X)=Ax => x is an eigen vector of T. :. Every eigen vector of T' is also an eigen vector of T. But T has no eigeo vector in MJ. as all the eigeo vectors of T are in M. Also M n MJ.={Ol. Consequently T' : ¥J.~M.J. is an operator with no eigen :. M~={O}. vectors and so no eigeo values.
FINITE DIlY.ENSlO AL SPECtRAL lHEORY
This
Q
=>
423
H=M as M(£;Ml.:::aH M,'s span H.
" Also H=M(£;Ml.=M= EM,. /-1
This Qo M,'s span H. Problem J. Let T be a normal operator on a finite dimensional Hilbert space and let p be a polynomial. Show that if A is an eigen value ot T, then p~A) is an eigen vahle of p(n. [Meerut J987] Solution. Let T be a no, mal operator on a finite, dimensional Hilbert space H. Let Tu=Au, i.e., A is an eigen value of T. I. To prove that PlT) is a normal operator. Prove this as in Ex. ) 3, page 398. II. To prove that peA) is an eigen value of p(T). p(x) =ao +a,x +' a.x l + ... +amxm PlT)=Qol+a1T+al T1 + ... + a,,,Tm .. (5) Tu=Au => T'u= T. Tu=T.(Au)=A(Tu)=A.Au => T 2u=A2 u Generalising tbis, ~e get T"u=A"u. In view of this (5) gives plT)u=aolu +a1Tu+ollT2u+ ... +arnT"'u = 0ou+a1Au+ a.;\'u+ +arnAmu =(oll+alA+a.;\2+ ... +amAm)u _p(A)u or, p( T)u=p(A)u. This => p(;\) is an eigen value of p(T). Problem 2. Let T be a normal operator on a Hilbert space
H. Let {A1' A2 • .... An} be the set of eigen values of T. Suppose M1-={x E H: ,Tx=A,x}. Prove that M,'s span H. [Meerut 1986] Hint. Here prove (iii) => (i) of Theorem 8. Problem 3. Let T be 0 linear operator on a jinlte-dimen.siona/ complex Hilbert .~pace V(K). Then T is normal iff the adjoint T* is a polynomial in T. [Meerut 1982, 79) Solution. Step (i). Suppose T is normal. To prove tbat 1* is polynomial in T. .
424
FINIfE DIMENSIONAL &PECTRAL THfOHY
Since T is normal and hence T is expressible as the spectral n
solution namely T= 1: c/Ej where c/ E K. i-I
and E/ is perpendicular pairwise orthogonal projections so that E j *=Ej =E/2. T is normal => TT*=T*T
f;
T*==(
CjE/
/-1
)* =
f; CiEi*=
i-I
i: CiEI
i-I
n
T* =
or
}; CiEi'
/~1
We know that each E/ is a polynomial in T. Consequently T* is a polynomial in T. Step (ii). Let T* be a polynomial in T so that T*=/(T}. To prove that T is normal.
TT*=T/(T}, T*T=!
... (1)
be a spectral resolution of T, where P;'s are pairwise non-zere orthogonal projections S.t. ..,(2)
and "/s are distinct complex numbers Step t. W ~ shall show that '\;'s are eigen values of T. There exists a non-zero vector x in the range of Pi S.t. PiX =X. TX=(
,f;
J-I
AiP,) X=
f
1-1
>'JPjPI (x)
using the fact that PiP 1= 0 if i ¥.j, we get TX=>.;P;2 (X}=-.\i PiX as Pil=p/ =.\/x or Tx=.\x
425
FINITE DIMENSIONAL SP CfRAL THEORY
This proves that A/
Step II.
'S an eigen \'alue of T
Write A ={Ah A2 ,
••••
¥ i.
A,,).
Aim. All the eigen values of T are precisely members of T. If possible, let A be an arbitrary eigen value cf T s.t. Tx=Ax.
Then
or
Tx=A lex)
(
"
£ AiPI )
i-I
X=A (
i
i-I
P; ) x
II
E AiPI(X)=A E PI(x)
or
I-I
i-I
};" A; p)p/ (X)=A E" PjP; (x)
or
'-1
1-1
or
>.) Pl (X)=A P/ (x)
or or
(A}-A) x=O.
A,P, X=AP, (x) ••
But
or
AjX='\X
x~O.
l-ience every eigen value of T is a member of A.
>',=A.
Thus we have proved that in the spectral resolution (I) of T, the scalars J..;'s are dIstinct eigen values of T.
Step III. Remains to prove the uniqueness of the solution (i). til
Let T= E
.-1
I)(.;QI ••• (3)
be another spectral resolution of T, where
ext's are distinct eigen values of T and Qt's are pairwise orthogonal projections.
We can also write (3) as
= E AI 2P,"= 1: A,2P; f
,
or
T2= 1:" A,2P,'
This:;>
TI= 1:
'-I n
1_1
A/l
Pi
... (5) ... (6)
426
FINn E DIMENSIONAL SPiCTRAL THEORY
Let get) by any polynomial \\ ith complex coffident in the complex variable I. Taking Irnear combination, we get n
g(T)= E g ('\i) PI /=1
~
p; is a polynomial s.t. Pi ('\,)= {
if i=j if i =1)
.. (7)
Taking p; in place of g,
Thus p,(T)=Pj • If we take po (t)= -<1-'\1) (I - '\2) ..• (I+'\n) I
(,\/-,\d (,\;-,\2) ••• (,\/_oA n )
then p; sat isfies (7). If we apply this discussion for Q/s then we shall get Q/ ""' p,(T) for each i. Therefore p/= Q; for each i. Hence the two spectral resolutions of T are the same. PlOblem 4. If T is an operator on a finite dimensional Hilbert .Ipace If, then prove the following statements. (0) V is singular ¢> 0 E (J (T), lKanpur 1984,81, 7S] i.e. TOis s ngular (/fO is the spectrum ofT.
(b) (c)
If T is non-~ingular, then A E (J (T) ::;.. ,\-1 E
(J
E (T-l)
[Kanpur 1984]
If A is r.onosingular, then (] (ATA-1)= (J(T)
If A E (] (Tl, and if p is any polynomial, then p (A) E a [p I T»). [Meerut 1982, 80] Hence (J (T) denote~ tire spectrum of T, i.e. the set of all eigen values of T. Solution la)o T is a sipgular operator ~ 3 x E H S.t. Tx=O where x.;iO ¢> Tx=Ox ¢> 0 is eigen value of T (d)
¢>
(b)
0E
(J
(T).
T is non-singular, " E
(J
(T)
:::. ,\ is eigen value of non-singular operator T ~ A=FO eigen value of non·~ingular T
427
FIN I 'E DIMf:!I"SIONAL SPECTR \L THEORY
~
3 x E If S.t x#O and Tx=Ax X=T-l (Ax)=AT-l (x) :;.. A-I X=T-l (x) => A-I E a (T-l).
~
Let S=ATA-], Then S-Al= ATA-I-A1AA-I=ATA-I_A(AlA·-I) =A [T--AT] A-I det(S-.\l)=det A det (T-Al).det A--I =det A. det A-I. det (T - AI) =det l.det (T -AI) as det A det A-l=det AA-l = I det (T- AI) or det (S~,u)=det (T-A1) :. det (S-,u)==-O ,'", det (T-Al)=O (c)
.. (1)
But det (S-M) -=- 0 ," A E a (S). :.
By (1), A E a (S) <=> A E a (T)
or or
a (8)=a (T) a (ATA-l)=a (T).
(d) (1)
Let ,\ E a (T), then 3 non-zero vector x S.t. Tx=Ax T 2x= T (Ax)=ATx=A AX=,\2x ~ 1'2x =A'x ~ T (Px)=T (AlIX)=.VTx => T3 X =A 2 . Ax ~ T3 X =A 3X.
.•. (2)
:;;10
Finally, TX=AX, T2 X=A 2 )(., T8X=AIX. Generalising, we get TnX=A ..X. Let p (1)=aO+alt-~ a2(2+ ... + 0 ,,1" be a polynomial in t of degree n \\-here an¥O. Then p tTl x=(ool +a1 T+aaP+ ... +anP) x ==aOx+a 1 Ax+a2 A2x+ ... +a.A n x = (au+olA+02A2+ ... +anA") x=p (A) x or p (T) xc=p (A),X with r,,=AX. This proves that
,\ E
(J
(T) => P (A) E
(J
[p (T»).
Problem 5. Let T be any operator 011 a fillite dimem;onal Hilbert space Hand N 0 normal operator on H. Shmv that if T commutes with N. then T alw commutes with N*. Solution Let T commute with normal operator N, then NN*=N*N ... (J) and TN",-NT .. (2) To prove tbat TN*=N*T I. We claim TN'''=NmT ... (3)
428
FINITB DIMENSIONAL SPECTRAL THlORY
The result is true for k = 1, by (2). Let (J) be true for k =m, TN'" = Nn'T. ...(4) Now TNtn+l=(TN"') N=(N"'T) N, by (4) =Nm (TN, =Nm(NT) = Nnl-tlT or TNm+l=Nm+l T which proves that the result is true for n=m+ 1 if it is true for n=m. Hence tbe result (3) is true by Induction. Let p (t)=a O+a l t+a.t 2+ ... +ant" be aby polynomial with complex coefficients. II. Tp (N)=T [ao/+alN+azNI+ ... +anNn] =a"T + alTN +a~TN2+ ... +anTNn =aolT +alNT+a2 N2T+ ... +anNnT, by (3) =(oo/+a l N+a 2N2+ ... + a"N") T or Tp (N)=p (N) T. ...(5) then
III.
Let N = 1:" Ai Pi be the I>pectral resolutIOn of the normal ;-J
operator N, where PI is projection and so that p/2=PI*=p/.
Then N* = (I' A/PI)· = 1: )./p/* =1: I
or
~/P/
I
N*=}; ~jPI'
... (6)
i
But each operator Pi is a polynomial in N. Hence, by virtue of (5), N* commutes with T. Problem 6 Let T be a normal operator on a finite dimens;onal Hilbert space H with lipearllnl AI. A2, ... , Am. Then prove the/ollowing statements (a) Tis Sl:'lj adjoint ~. each '\; is real, (b) T is positive "c- each ~ 0, (c) T is unitary «> I A; I= I ¥ i. c[Kanpur 1989, 91; Meerut 82J
"I
Ii
Solution.
Let T= I' "iP,
... (1)
1=1
be the spectral resolution of T, where AI. "', An are eigen values of. l' and PI, P 2 , ,Pn are non-zero pairwise orthogonal projections. This,* P?=P;*=P; ¥ i and P;P,=O for i~j. Here we also ha\e I'" P;=T. 1-1
...(2)
4
FINITE DIMENSIONAL SPECTRAL THEORY
T*=(E A;P;}*=E
).IP'"
-.c=E )./p;
/
... (
T"'=E ~Pi (a> let T be self adjoint, then T = T"'. Putting the values from (I) and (3) in (4), we get E (i;P;) = E AlP"
or
... (
i
or
P;=O.
E ().;-A/) i
or or or
Multiplying both sides by P j and nothing that p,p) ='1 ¥ i except i=j, we get C~/-Ai) P/'=O (~J-'\~) P/=O ("i,-A;) Pj=O or '3:,-A j =O ,\j= 1./.
This ==- Aj is real ¥ j. Conversely Suppose that A; is real l.tJ i, then A/= )./. . •. (5 Aim. T is self a~joint. By (3) and (5), T'" =E AiP/, By (I) and (6), T"'=T, showing thereby T is self adjoint. (b) Let x E H be arbitrary.
=2-' E .\/ i
i
= E A/ /' j
• {Pil But P; Pj=PjP/= 0 Usin~
< p/
<
x, Pj
X
> ... (7.
Pi'" p/ x, x>
for &' lor
this in (7), we get < Tx. x >=E A/ I
i =- j "
':1=J
• (Pj =PJ ).
< P,2
X, X
>
But
,
Let T be positive, then < TJI, x> ;;:: Aim. AI ~ 0 ¥i. By (9) and (10), E A. 1 PIX il 2 ~ O.
o.
,
Let us suppose that x is in the range of Ph then p.x=x, Plx=O ¥ i except i= 1.
. .. (9) .•• (10) ••• (11)
4JO
FINITE DlMLN310NAL SPECTRAL THEORY
Then (II) gives Al ~ O. Similarly we can prove A2 ~ O. Gelleralising, we get Ai ~ 0 ¥ i. Conversely suppose that Ai ;;;:. 0 ¥ i. Aim. T is positive. By (9) and (12),
~
... (12)
O.
TT· = (.r Ai P;) (E)./ PJ) i
= E
I' }
or
AI ~'J PI Pj=E Ai ii P I2 I
TT*=E I AI 12 PI.
... (13)
I
T is unitary
-¢>
TT*=I
-¢>
E I
~
I '\1 \2 P I =
E Pi>
by (2) aDd (13) ..
E (IA/12-1) PI =,0. I
Multiplyi"g both sides by PI and noting that PiP} =0 ¥ i except i =--j. we get (I
",\2_]) PJ=O
I A} \2-1=0 {:> IAj 1=]
vj. Exercises 1. (a) Define eigen value. eigen vector, eigen space and spectral resolution of an operator T on a Hilbert space H. Prove that. if A be an eigen value of an operator T, on a finite dimensional vector space H. then I A I ~ II Til. [Kanpur 1988J 2. (a I Define a Banach Algebra. Show that the set G of all regular elements in a Blnach Algebra is an open set. Further show that the mapping x-*x- 1 is continuous on G. [Kanpur 1989J (b) For an element x in a Banach Algebra A, prove that a (x) is non-empty. [Kanpur 1988] 3. Define spectrum and spectral resolution of an operator T on a non-zero Hilbert space. Prove that eigen values of an operator T constitutes a non-empty finite subset of the complex plane and the number of points in thi~ set does not exceed the [Kanpur 1987] dimension of the space. 4. (a) Define a Banach Algebra A. Also define regular and singular points in A. Prove that every elem:nt x for which II x . 1 II < ] is regular and its inverse is given by the formula -¢>
00
x- 1 =1+E (I-x)". ....1
[Kanpur 1989]
FI"ITE
DIMEN~JON"L
SPECTRAL THEORY
431
(b) Define spectral radius rex) of an element x of a Banach Algebra. Prove that
lim
n-+oo \I x' I lin . [Kaopur 1987] 5. (a) Define eigen vector, eigen value, eigen space, spectrum and spectral resolution of an operator T on a non-zero Hilbert splce H. Prove that jf A be an eigen. value of a operator T on a finite dimensional Hilbert space H, then I A I ~ II Til. (b) Let T be a normal operator 00 a Hilbert space H with spectrum {At. A2 , ••• , Am}. Use the spectral resoJution of T to prove th ~ following statements. (i) T is ~elf adjoint -= each Al is real. (ii) T is positive ¢> Ai ;?- 0 for each i, (iii) T is unitary I A/ I= I for each i. lKaopur 1986] 6. (a) Define eigen value, eigen vector, eigen space and spectral resolution of an operator T on a Hilbert space H. (b) Prove that the spectral resolution of a normal operator on a finite dimensional non-zero Hilbert space is unique. [Kanpur 1985] 7. (a) Prove that two matrices in A. are silT ilar';:> they are the matrics of a single operator H relative to different bases. (b) D~fine the spectrum (} (T) of an operator T on Hand prove that 0) T is singular.;:> 0 E a (T). (ii) If T is non-singular, then A E a (T) ¢> >,-1 E a (TI). {Kaopur 198'] 8. (a) Define regular and singular elements of a Banach Algebra A Prove that every element x for which II X -I II < I is reg'Jlar and the inverse of such an element is giver) by the formula r (.x) =
-=
X-I
"" = 1 + E (I-x)n. n-l
(b\ For an element x in the Banach Algebra A, prove that a (xl is non·empty. [Kanpur 1984] [Kanpur 198J] 9. State and prove the spectral Theorem. 10. (a) If S denotes the set of singular element of a Banach Algebra and Z denotes the set of all topologkal divisors of zero then prove that (i) Z C S. (ii) the boundary of S is a subset of Z. (b) Prove the following: (i) If r is an element of R then 1 -r is regular.
432
FINITE DIMENSIONAL SPECTRAL THEORY
(ii) If t -xr is regular then 1 -rx is also regular. . [Kaopur 1983] 11. Let G and S denote the set of regular elements and the set of singular elements respectively in a Banach Algebra A. Prove the following: (a) G is an open set. (b) The mapping X~X-I of G into Gis homcomorpliism of G onto itself. (c) The boundary of S is a subset of the set of all topological divisiors of zero. [Kanpur M.Sc. F. 19821 12.