Commutator Theory On Hilbert Space

Can. J. Math., Vol. XXXIX, No. 5, 1987, pp.
1235-1280
COMMUTATOR THEORY ON HILBERT SPACE

DEREK W. ROBINSON
1. Introduction. Commutator theory has its origins in constructive

quantum field theory. It was initially developed by Glirnm and Jaffe [7] as
a method to establish self-adjointness of quantum fields and model
Hamiltonians. But it has subsequently proved useful for a variety of other
problems in field theory [17] [15] [8] [3], quantum mechanics [5], and Lie
group theory [6]. Despite all these detailed applications no attempt
appears to have been made to systematically develop the theory although
reviews have been given in [22] and [9]. The primary aim of the present
paper is to partially correct this situation. The secondary aim is to apply
the theory to the analysis of first and second order partial differential
operators associated with a Lie group.
The basic ideas of commutator theory and perturbation theory are very
similar. One attempts to derive information about a complex system
by comparison with a simpler reference system. But the nature of the
comparison differs greatly from one theory to the other. Perturbation
theory applies when the difference between the two systems is small but
commutator theory only requires the complex system to be relatively
smooth with respect to the reference system. No small parameters enter in
the latter theory and hence it could be referred to as singular perturbation
theory [10], but this latter term is used in many different contexts [20].
In order to give a more precise comparison suppose H is a given
self-adjoint operator on a Hilbert space ft, with domain D(H) and
C°°-elements ft^ = Dn^lD(Hn% and further suppose that K is a
symmetric operator from ft^ into ft. Then the simplest theorem of
Kato-Rellich perturbation theory [12] [22] states that K is essentially
self-adjoint whenever
|| (K- H)a\\ ^k\\Ha\\ + /|M|, a e ft^,
for some k e [0, 1) and / ^ 0, i.e., the difference between K and H is

small in comparison with H. In contrast the Glimm-Jaffe commutator
theorem states that K is essentially self-adjoint whenever Kfy^ Q D(H)
and the commutator (ad H)(K) = HK — KH satisfies
e
(1.1) || (ad H)(K)a\\ ^ k{\\Ha\\ + lx\\a\l " &«»
Received October 8, 1986 and in revised form December 8, 1986.

1235
Downloaded from https://www.cambridge.org/core. 26 Aug 2021 at 15:22:23, subject to the Cambridge Core terms of use.
1236 DEREK W. ROBINSON
for some kvlx i^ 0. In fact most versions of the commutator theorem also
require H ^ 0 but this is not necessary [4]. If H ^ 0 essential self-
adjointness of K even follows from the weaker commutator bound
| (a, (ad H)(K)b) | ^ k\\\Hl/2a\\ • \\Hl/2b\\ + l\\\a\\ • \\b\\9
a,b^ li^.
The next refinement of perturbation theory occurs if K is positive, or lower
semi-bounded. Then self-adjointness properties can be obtained from
weaker perturbative estimates. One of the principal results of Section 2
establishes that analogous refinements arise in commutator theory. If
K ^ 0 then it is essentially self-adjoint whenever Kfy^ Q rj2 and
(1.2) \\(zdH)2(K)a\\ ^ k2\\H2a\\ + / 2 l|a||, a e ï^.
If, in addition, H ^ 0 then it suffices that the form bound
| (a, (ad H)2(K)b) \ ^ k'2\\Ha\\ • \\Hb\\ + l'2\\a\\ • I N , a, b <= ^
is valid. A partial result of this kind was given earlier by Driessler and
Summers [4].
Thus, in summary, there are four basic self-adjointness results of
commutator theory classified by stability properties, expressed by
positivity of H or K, and smoothness properties, expressed by commutator
estimates. In rough terms, the stronger the stability the weaker the
smoothness required for self-adjointness. In each case one can also derive
invariance properties of the unitary group St = exp{itK} or, if K ^ 0, the
contraction semigroup Tt = exp{— tK}. For example, if (1.1) is satisfied
then ija = D( \H\a) is invariant under S for all a <= [0, 1] and the
restriction of S to fya is continuous in the graph norm. Alternatively, if
K ^ 0 and (1.2) is satisfied then ï)a is invariant under Tfor all a e [0, 2],
and the restriction of T to £)a is continuous in the graph norm. Details are
described in Section 2. Earlier results on invariance occur in [8] [3] [6] [15]
and continuity properties in [16].
A simple illustration of the self-adjointness results is given by taking
Ï) = L (R) and H = id/ dx the self-adjoint generator of the unitary group
of translations. Then if A denotes the operator of multiplication by a
real-valued function with bounded second derivative it follows that the
first-order differential operator K = HA + AH is a symmetric operator
from 1)^ into rj. Moreover condition (1.1) is satisfied, with kx = 2||/l r || 00
and lx = II^T'H^, and hence Aîs essentially self-adjoint. Alternatively if A
is positive with a bounded third derivative and B is positive with a
bounded second derivative then the second-order differential operator
K = HAH + B is a positive symmetric operator from fj^ into fj which
satisfies (1.2). Therefore K is self-adjoint. These examples illustrate two
important points. First the differentiability, i.e., the smoothness, of the
coefficients is necessary for the boundedness of the commutators
COMMUTATOR THEORY 1237
(ad H)(K), (ad H)2{K), on rj^. Second the order of the bound, i.e., the
order in //, reflects the order of the differential operator. Since the O(H)
bound cannot be relaxed in the Glimm-Jaffe theorem (see [22] Example 1
in Section X.5) this theorem is restricted to first-order differential
operators, unless one makes a different choice of H. In contrast the
theorem for K ^ 0 applies to second-order differential operators because
the commutator bound is 0(H ). Again this order cannot be increased (see
[4] Section 4).
It is natural to try and extend the foregoing discussion of differential
operators on L 2 (R) to partial differential operators on L2(Rd). This then
leads to the consideration of operators associated with a unitary
representation of R^ or a general Lie group G. Commutator theorems for
such operators are derived in Section 3. They generalize results of Poulsen
[21] for operators commuting with the group action. The extension of the
results of Section 2 is quite straightforward since the differential structure
of the representation is completely determined by a Laplace operator. This
Laplacian then plays the role of the self-adjoint operator H. We note that
many of the quantum-mechanical applications of commutator theory have
been based on the choice of H as the Hamiltonian of the harmonic
oscillator, acting on L (R ). But this operator is exactly the Laplacian of
the Heisenberg group, acting in the Schrôdinger representation on L 2 (R^).
Hence these examples are naturally described by the results of
Section 3.
In Sections 4 and 5 we establish commutation theorems. Basically we
prove that if Kx and K2 are two symmetric operators each satisfying the
criteria of one of the commutator theorems of Section 2, and if Kx formally
commutes with K2, then the self-adjoint extensions Kx and K2 commute.
Finally, in Section 6, we elaborate on the application of our results to
general partial differential operators associated with a Lie group.
Although this paper only considers commutator theory on Hilbert space
we note that the Glimm-Jaffe theorem, as described above, is also valid on
Banach space [1] [23]. It would be of great interest to find a Banach space
analogue of the double commutator theorem.
2. Commutator theorems. In this section we develop the theory of com-

mutator theorems on Hilbert space. There are three classes of theorem, for
dissipative, symmetric, and positive, operators, and there are also three
types of result, generator properties, invariance, and boundedness and
continuity properties. In each successive class the results are more
detailed.
Throughout the section H denotes a self-adjoint operator on the Hilbert
space ï) with domain D(H). For each a ^ 0 we set
ha = D((i + HY2)
equipped with the norm
IWL = II (/ + H2)a/2a\\.
In particular ï) = ï)0 a n d I I ' l l == 11 "I lo- N e x t for a < 0 we define rja to be the
completion of fj with respect to the norm
a G Ï, -> || fl || a = || (/ + # 2 )" / 2 f l ||.
It follows that each rja, a G R, is a Hilbert space, the operators
(/ + H2yn define unitary maps from t)a into ha-^ a n d the dual fj* of fja
is identifiable as fy_a. Moreover i)a D ï)^ if a = /?. Finally we define ^
by
The space f^ is a Fréchet space with respect to the topology defined by

the family of norms { ||-||a; a e R}.
If H = 0 it is also convenient to introduce a different family of
norms
I K = ll(/ + #)a*1l, a e ^
on the spaces rja. It is easy to see that |HI« is equivalent to ||-||a. Moreover
(/ -f Hf defines a map from \)a into £)a_^ which is unitary with respect to
the norm ||-||;.
a. Dissipative operators. The operator K on the Hilbert space t) is defined
to be dissipative if
Re(tf, Ka) ^ 0, a G £>(#),
or, equivalently, if
| | ( / + €*>i|| ^ | H | , Ae D(K\
for all c > 0. Each densely defined dissipative operator is closable, and its
closure is also dissipative.
Now we consider dissipative operators K from fj^ into f) and their
commutators (ad H)(K) with H. These commutators are not necessarily
defined as operators but they can be interpreted as sequilinear forms over
rjoo X rjoo, and we will consistently make this interpretation. Thus, with a
slight abuse of notation, we define the sequilinear form (ad H)(K) by
û ^ e f l o o X Ô o o ^ (*> ( a d H)(K)°) = (#a, ^ » ) - (A, KHb\

THEOREM 2.1. L^/ K be a dissipative operator from ^ />2/6> £) tfftd
suppose that
(2.1) | (a, (ad # X W I ^ *>H« • ll*lli-«, a. * e ^

Tor som^ a G [0, 1 ) and a ka = 0.
// follows that the closure K of K generates a strongly continuous

contraction semigroup.
Proof. It follows from the Hille-Yosida theorem that K generates a
contraction semigroup if, and only if, the range R(I + eK) of / + eK is
dense in ft for small e > 0. Now assume this is false. Then for each e > 0
there is a non-zero a e {) such that
(a, (/ + eK)b) = 0, ft e i)ôo-
0
Next introduce the self-adjoint contraction semigroup S = {S r }ô

where
St = Qxp{-tH2}.
Then Sti) Q f^, by spectral theory. Therefore
(a, (I + eK)S2ta) = 0
for all / > 0. Consequently
(2.2) \\Sta\\2 = - c Re(a, KS2ta)
= -e Re(Sta, KSta) + e Re(a, (ad St)(K)Sta)
S e Re(a, (ad S,)(#)S,a).
Now consider the everywhere defined operators
Ct = (ad St)(K)Sr t > 0.
We will prove that these operators are bounded and
c - sup{ ||C,||; 0 < t < 1} < +oo,
Then from (2.1) one deduces that
\\a\\2 = lim \\Sta\\2 =§ cclMI2
which gives an inconsistency if ec < 1. Thus it remains to establish the

boundedness properties of the Ct.
Let c, d e t)^ then
(2.3) (c, (ad S,)(/OV) = - ' / I * ( V . ( ad H2){K)S,a„s)d)
= ~t j\ds{ (HStsc, (ad H)(K)S,{2_s)d)
+ (Slsc, (ad H){K)HSt(2_s)d) }.

Therefore
| (C, (ad S,)(tf)V) I = V / l * { IIVHl+« • H^(2-,^lll-a

+ IIVHa ' WS,{2-s)d\\2-a}.
Now for 0 < s ^ 1, and ft ^ 0, one has

\\Ssc\\p = II (/ + H2f2e~sH2c\\ â CfiS-
p/2
\\c\\
where
Co = sup (1 + \f/2e'\
by spectral theory. Consequently

| (c, (ad St)(K)Std) | ^ *J|c|| • ||J|| { q + « q _ « / 1 + « + cac2.Ja}
for 0 < / < 1 where
Ia= j\dss-«i\2-s)-^«<\
But since 0 ^ a < 1 both integrals Ia9 and / ] + c r are finite. Thus Ct is
bounded and ||CJ| is uniformly bounded for 0 < / < 1. This establishes
that AT is a generator.
Remark 2.2. This proof follows an argument of Nelson [17] who proved
a similar result under the additional hypothesis H ^ 0.
Next we consider properties of the semigroup generated by K, in
particular boundedness, continuity, and invariance properties, which can
be derived if a = 0 in (2.1).
THEOREM 2.3. Let K be a dissipâtive operator from f)^ into fj. Suppose
(2.4) | (a, (ad H){K)b) | S fc||<z|| • ||ft||„ a, b e b ^
Let Tt = exp{— tK) denote the contraction semigroup generated by K.
It follows that ATb^ Q h, and
1. Ç&, ç b„ ? ^ 0,
A
2. 117^11, ^ e' ||a||„ a e b,
3. r | h « IHIpconf/nwotw.
Proof. By definition
| (//a, AZ») | ë | (a, 7076) | + | (a, (ad H)(K)b) \
S ||fl||{ ||tf/ft|| + * | | i | | , } , Û, b G boo
and hence Kb e 7>(77) = b,.
Next we prove that AT + kl is ||-||,-dissipative. For this first note that
||a||, = ||(7 + iH)a||. Then
|| (/ + e(K + kl) ) fl ||, = || (/ + / / / ) ( / + e(K + kl) )a\\
g ||(7 + e(K + kl)(I + iH)a\\
- e|| (ad 77)(À>||
â (1 + c^lMI, - cArllflll, = Hall,

where we have used dissipativity of K, and (2.4).
Finally we prove that R(I + e(K + kl) ) is 11-111 -dense in \)x for small
€ > 0. Then it follows from the Hille-Yosida theorem that the |H| r closure
of K 4- kl generates a IH^-continuous, |H|j-contractive, semigroup ont),.
The stated properties of T are an immediate consequence. We use an
argument devised in [1]. Suppose there is an / e fyf such that
(2.5) / ( (/ + e(K + kl) )a) = 0, a e ^
]
Now R = (I -f iH)~ is a bounded map from rj into t)j and hence the
adjoint R* defines a bounded map from Ï)* into I). But to prove / = 0 it
suffices to prove that R*f = 0, because the range of R* equalsfy.It follows
from (2.5), however, that
(/**/, (/ + eK)a) = -ek(R*f a) + c/( (ad R)(K)a)
= -ek(R*f, a) + /€(/**/, (ad H)(K)Ra)
where we have used ATf^ Q t)j to define (ad H)(K) on r ^ . Thus
| (**/, (/ + £*)fl| ê 2€*||/**/|| ' || (/ + €*)û||.
Now R(I + eK) is dense in fy and consequently

\\R*f\\ ë 2^||JR*/||.
Thus if € < 1/2/c one must have R*f = 0.
b. Symmetric operators. If K is SL symmetric operator on fy then z±/A' are
both dissipative. Hence Theorem 2.1 gives criteria for ±iK to generate
contraction semigroups. But ±iK generate such semigroups if, and only if,
K is self-adjoint. Thus Theorem 2.1 immediately yields a result on
essential self-adjointness of a symmetric operator. But this result can be
strengthened by symmetry.
OBSERVATION 2.4. Let s be a symmetric sesqui linear form over f^ X t)^
and suppose
\s(a,b)\ fS c\\a\\a- \\b\\p a,b e l ^ ,
for some a, jS e R. Then
\s(a, b) | ^ c\\a\\a+y • \\b\\p_r a, b e li^,
for all y G [0, iS - «] z/£ ^ a, or y e [0 - a, 0] / / £ ^ a.
This is established in two steps. First by symmetry one has
\s(a,b)\ ^ c\\a\\o- \\b\\a
and then by interpolation theory (see, for example, [22] appendix to

Section 1X.4) the bound extends to all pairs of intermediate indices with
sum a 4 ft.
The observation immediately implies that if K is symmetric then
Theorem 2.1 can be extended to the value a = 1, by symmetry. It also
implies that if one has a family of conditions such as
\s(a,b)\ ^ c\\a\\a- \\b\\„_a, a,b e l)oo,
with a e [0, n] then the weakest condition is for a = nil and the
strongest is for a = 0, or a = n. The foregoing discussion of dissipative
operators was based on commutation conditions with index n = 1 and
we next discuss generalizations for symmetric operators. Subsequently we
show that similar results can be obtained for positive symmetric operators
from commutation conditions with index n = 2.
First we consider the weak conditions with index n = 1 then Theorems
2.1 and 2.3 can be generalized if 77 ^ 0. Second we consider the strong
conditions with index n = 1 and show that then the generalized invariance
properties can be derived without positivity of 77.
THEOREM 2.5. Let K be a symmetric operator from i)^ into rj and
suppose that
(2.6) | (a, (ad H)(K)b) | Si *|M| 1 / 2 • ||*|| 1/2 , a, b G f)oo
J or some k ^ 0.
It follows that K is essentially self-adjoint.
If, in addition, 77 ^ 0 then Kl)^ Q fj 1/2 . Moreover, for each a e [0, 1/2]
the unitary group Vt = exp{itK} has the following properties.
1. VX = l)a, t G R,
Mak/2
2. Il^li; ^ e \\a\\'a, t G R,a G t,a,
3. V\{] is \\-\\'a-continuous.
Proof The essential self-adjointness follows from Theorem 2.1, with
a = 1/2, applied to zb/'AT. Now we consider the additional properties
with H ^ 0. Recall that if 77 ^ 0 the norms |H|« are defined by
IkC = ll(/ + H)aa\\, a e i)a.
It then follows that \\a\\a ^ \\a\\fa, for a ^ 0.
To prove Ki)^ Q t) 1/2 we observe that if a, Z?e l ^ then
| ( (/ + H)V2a, Kb)\ ë | ( (7 4 H)U2a, (ad H)(K)b) \
l/2
4 I ( (7 4 H)~ a, K(I 4 H)b) \
^ k\\a\\ • \\b\\\/2 4 |W| • \\K(I 4 H)b\\.
1/2
But since I) is a core for (7 4 77) it follows that
Kb e D((I + H)V1).
In order to establish the properties of the unitary group V it is necessary
to have an estimate for commutators of fractional powers.
LEMMA 2.6. Let K be a closable operator from fy^ into fy. Suppose
H ^ 0 and
(2.7) | (a, (ad J/)(tf)6) | â *|M|' 1/2 • ||ft||'1/2, Û, ft G r ^ .
Then, for each a e [0, 1],
(2.8) | (a, (ad (/ + Hf)(K)b) | â a*||*||; /2 • ||ft||; /2 , a, ft e ^
Proof. Set L = I + H. We use the integral representation
/*oo
a
L = ca J0 d\X~aL(I + AL) _ 1
where
oo
~l /
c
o ^J A
W - a / i _i_ \ \ - l
" ( i + xy
/
Then
(a, (ad La)(K)b)
foo
= ca J Q d\A~ a ( (/ + \ L ) _ , a , (ad / / ) ( ^ ) ( / 4- \L) _ 1 ft).

Consequently
| (fl, (ad La)(K)b) |
/*00
^ kca J dXX~a\\L]/2(I + \L)_1Û|| • ||L 1 / 2 (/ + XL)"1/?!

Therefore by the Cauchy-Schwarz inequality one obtains
| (a, (ad L a )(^)ft) | â kMa(a)Ma(b)
where
/*oo
A4(Û) 2
= « j o ^ - a | | L 1 / 2 ( / + AJL)- ] «j| 2 .
c
But if £ denotes the spectral measure of H then

oo /*oo
-2
/ a
j d(a,E(x)a)x J0 dXX~ (\ + Xx)
oo Ax>
2
/
, J(fl, £(x)a)x a J 0 dPP'a(\ + p)-
Too
a 2
a J ] </(*, £(x)tf)x = a( ||a||; / 2 )
This gives the desired bounds.
We note in passing that a slight adaptation of this argument gives

bounds
| (*, (ad(7 + H)«)(K)b) \ ^ kaj\\a\\fi • IIÎU—^, fl, b e ^
with /c a ^ < -hoo whenever |8, a - j8 e [0, 1/2).
Now we are prepared to prove the invariance properties in
Theorem 2.5.
First note that since K is symmetric it is closable, and hence closable as
an operator from the Fréchet space ^ into the Hilbert space f). Therefore,
by the closed graph theorem, there is an integer p ^ 1 and a c = 0
such that
(2.9) \\Ka\\ ^ cp\\a\\'p = cp\\ (7 + HYa\\, a e ^

Consequently for each e > 0 there is a c (c) such that
| | # ( / + eHYM\ ^ cp(e)\\a\\, a e ft.
Now introduce the family of bounded self-adjoint operators
(2.10) K€ = (7 + eHypK(I 4- c i / ) " ' .
It follows that *Teft ç ft^.
Second remark that the norm
Ml; = II (/ + / W l , « G *>«,
satisfies the bounds
(2.11) Ml ^ Hall; ^ 2 a / 2 ||a|| a .

Therefore (2.6) implies (2.7) and hence (2.8) is valid for a e [0, 1].
Third, consider the unitary groups V€ generated by the K€. One
has V*\)a ç, fta for all t e R and a e [0, 1] by power series expansion.
Then since
one has V'§a = t)a. But
y ( l l ^ l i ; ) 2 = y ( (/ + " W f l , (/ + HfVâ)
at at
= /((/ + eH)~pVfa,
(ad(7 + H)2a)(K)(I + eH)~pVfa)
â a*(||(/ + £i/)-^;fl||;)2
=S ^ ( | | F > | | ; ) 2
where the first inequality uses (2.6), (2.11), and Lemma 2.6, and is valid
for 0 ^ a ^ 1/2. Now integrating this differential inequality gives
(2.12) \\Vlta\\'a =§ eâkn\\a\\'a, a e bm
and it is important that this bound is uniform in e.
Next if a e ^ then
|| (Kt - K)a\\ ^ \\K((I + Œ)» - I)a\\
+ | | ( ( 7 + eHy - I)Ka\\
â cp\$(I + £//)"" - I)a\\p
+ ||( (7 + eHy - I)Ka\\
-» 0
£-K)
where we have used || (/ + eH) ]\\ ^ 1 and (2.9). Since ^ is a core of AT

it follows immediately that V\ converges strongly to Vn for each t e R,
and the convergence is uniform for / in compact subsets of R (see, for
example, [2] Section 3.1.3).
Now if E denotes the spectral measure of (/ + H) then for a e ^ one
deduces from (2.12) that
foo
(eW/4\\a\\\/2)2 a j N d(Vfa, E(X)Vâ)\
foo
= N]-2a(\\(i- EN)v;a\o2
where EN denotes the projection onto the subspace for which (I + H) =
NI. Thus if 0 S a < 1/2
(2.i3) H ( K ; - K / V H ; g o + N)a\\EN(v; - K,Vll'
+ 2eW/4\\a\\'U2/Nu2-a
and
(2.14) || ( F / - l) f l ||; ^ (1 + N)a\\EN(Vf - /)a||
/4
+ (e^ + l)\\a\\\/2/Nu2-a.
It immediately follows from (2.13) and (2.14) that Ve is ||-|^-convergent
to F a s e —> 0. Thus J^l)a ç fy^ and J^l)a = t)a by the group property. But
then it follows from (2.12), (2.13) and (2.14), that the restriction of F to $a
is ||-| ^-continuous and satisfies the bounds
\\V,a\\'a ^ eWak/2\\a\\'a, a e fta,
for 0 ^ a < 1/2. Next it follows from these bounds and monotone
convergence that Vti)l/2 = £)1/2 and
\\Vta\\\/2 g eWk/4\\a\\\/2, a e t, 1/2 .
Finally, since Kfy^ ç f), /2 and
(V,- I)a = i f\dsVJCa, a e !)«,,

one has
| | ( ^ - / ) a | | ' 1 / 2 ^ kkW/4||^ll'1/2, Û e t,oo-

But ^ is ||-|r i/2 -dense in fj 1/2 and hence this establishes that Vt is
IH|'1/2-continuous.
Remark 2.7. The original Glimm-Jaffe commutator theorem [7] proved
essential self-adjointness from a commutator condition of index one under
the assumption H ^ 0. Other versions based on similar assumptions were
subsequently given in [17] [5] [15] [16]. An invariance property of f)1/2 was
first proved by Faris and Lavine [5] by a differential inequality method
similar to the above. But the modified method using the K€ approximation
technique was introduced by Glimm and Jaffe [8] [9] for the proof of
invariance and commutation properties. Similar applications were sub-
sequently given by Driessler and Frôhlich [3] [6]. The invariance of
the fractional spaces fya and the ||-|^-continuity of V on ï)a appear to
be new.
Our second result on symmetric operators is based on the stronger form
of the commutator condition with index one and does not require H = 0.
The essential self-adjointness statement is essentially Theorem 1 of [4].
THEOREM 2.8. Let K be a symmetric operator from rj^ into I) and
suppose that
(2.15) | (a, (ad H)(K)b) \ ^ k\\a\\ • ||*||„ a, b e Ï, oo
for some k = 0.
It follows that K is essentially self adjoint.
Moreover, Kfy^ Q ïjj and the unitary group Vt = exp{itK] has the
following properties for each a e [0, 1]:
1. VX = ï,a, / e R,
2. ||â|| a g é>"|a*IMIa, t R, a e i,a,
3. VL is \\'\\a-continuous.
Proof. The essential self-adjointness follows from Theorem 2.1, with
a = 0, applied to ±iK. Next note that
(2.16) | (a, (ad H2)(K)b) \

^ | (Ha, (ad H)(K)b) | + | ( (ad H)(K)a, Hb) \
^k(\\Ha\\' \\b\\x + \\a\\x • \\Hb\\) ^ 2k\\a\\x - \\b\\x.
Since the C°°-elements of H and H2 are the same it follows that the
hypotheses of Theorem 2.5 are fulfilled with respect to the positive
self-adjoint operator H . Hence the statements about V follow essentially
from Theorem 2.5 when one notes that the fractional space rja of H
corresponds to the space fya/2 of H . The estimates on Vta can be
improved, however, because it is no longer necessary to introduce the
equivalent norm ||-||^. For example one has the following:
LEMMA 2.9. Let K be a symmetric operator from h^ into f) and
suppose that
| (a, (ad H2)(K)b\ ^ 2k\\a\\x • ||*|li, *, * e r^
then for each a G [0, 1]
| (a, (ad(7 + H2)a)(K)b) | ^ 2ak\\a\\a - \\b\\„ a, b G ft^.
The proof is identical to that of Lemma 2.6.
Next repeating the derivation of inequality (2.12), but using (2.16) and
Lemma 2.9 one deduces that
| | K ; < S eMak\\a\\a, a e t,a.
The proof then continues as in Theorem 2.5.
Finally we note that the index one in the commutator bounds (2.6) and
(2.15) is necessary for the essential self-adjointness of K. Example 1 in
Section X.5 of [22] gives counterexamples to any relaxation of this
condition, even with H ^ 0. Nevertheless self-adjointness follows even
from a mixed estimate of index one, i.e., an estimate of the type
//
| (a, (ad H)(K)b) | S 2 £,IMI<i+,e,)/2 " I N o - f t ) ^ a-b e 'W
/= 1
with Pi G [0, 1]. The invariance properties are, however, more delicate. If
one has an intermediate bound
| (a, (ad H)(K)b) | ^ k\\a\\{X+p)/2 • ||i|| ( 1 _^ ) / 2 , a, b G l ^ ,
for j8 e [0, 1] then this estimate together with symmetry gives
| (*, (ad H2){K)b) | g 2*11*11,_y • ||fe||1+y, a, b G ^
where y = (1 — |/?| )/2. Therefore by a slight variation of the proof of
Lemma 2.6 one finds
| (a, (ad(7 + H2f)(K)b) \ ^ 2kay\\a\\a • ||è|| a , a, b G ^
valid for a e [0, 1 — |y| ). Hence one has the invariance of l)a under Vt for
each a e [0, (1 4- |/?| )/2>.
c. Positive operators. The results of the previous subsection can be
extended if the symmetric operator K is positive. First it is only necessary
to have a bound on the double commutator (ad Hy(K) and second it
suffices that this bound has index two. It is the latter point which is of
greatest interest in applications.
Again there are two results based on a weak bound of index two and a
strong bound, respectively.
THEOREM 2.10. Let K be a positive operator from fy^ into h and
suppose that
(2.17) | (a, (adH)2(K)b)\ Si *|M|, -11*11,
for some k = 0.
//, in addition, 7 / ^ 0 then Kb^ Q f)j and for each a e [0, 1] the
contraction semigroup Tt — exp{ — tK} has the following properties.
1. TX ç l>a, t ^ 0,
2. \\Tta\\'a ^^k/2\\a\Ya9 t ^ 0, a e t,a,
3. T\fy is \\'\\a-continuous.
Proof Since K ^ 0 it suffices, for essential self-adjointness, to prove
that R(I 4- eK) is dense in £) for sufficiently small € > 0. But if there is an
Û G () orthogonal to the range of / + eK then by the calculation used to
derive (2.2) one has
(2.18) ||S>|| 2 = - c Re(S,0, KSta) + e Re(a, (ad St)(K)Sta)
^ (c/2){ (a, (ad St)(K)Sta) + ( (ad Sf)(K)Sta, a) }
where St = exp{ — tH }. Now consider the operators
C2J = (ad St)2(K)
defined on i)^. We will prove that the C2t have bounded closures
C2j and
co = sup{ \\C2J|; 0 < / < 1/2} < +oo.
Then one has
(a, (ad St)(K)Sta) + ( (ad S, )(#)£, Û, Û) -(a, C2ta)
by continuity, and hence
\\a\\2 = l i m l l V H 2 = (^/2)co||^||2
by (2.18). But if eco < 2 this implies a = 0. Hence K is essentially

self-adjoint.
Next we establish the boundedness properties of the C2r
If c, d <= f)^, then
(c, (ad ^) 2 (iO</)
= '2 / I Jr
/ I *(5/<r+^ (ad H2)\K)St{2_r_s)d)
and hence by linear algebra, and (2.17), one finds
(2.19) | (c, (ad sfiKyd) |
= ^ /I Jr fo ds J 0 (^)ll i/2 "%^)ll. • \\HmSl(2_r_s)d\U
= k'2 fo dr
fo dS
^ ( £ )IIS,(,+ „ll3-„ • HS„2-r-- 5 ) l l l + w
But
HS.cllp SÉ c ^ / 2 | k | |
for 0 < s ^ 1 and /? i^ 0, where
^ = sup{ (1 + \f/2e~\ X ^ 0}
by spectral theory. Substituting these bounds into (2.19) then gives
| (c, (ad S,) 2 (/OJ) I ^ Îkll ' ll^ll
for some k' ^ 0 and all 0 < / ^ 1/2. The boundedness properties of
C2t are then apparent.
Next we establish that Kfy^ Q r^ when 7 / ^ 0 . This requires an
estimate on the commutator (ad H)(K), which does not rely on the as-
sumption 7 / ^ 0 .
LEMMA 2.11. Let K be an operator from fy^ into rj and suppose that
\{a,Kb)\ â k0\\a\\ao • 1 1 % ,
| (a, (ad Z / ) 2 ^ ) I ^ *2II«H«2 • 1 1 % . a fc e

' &«»
for some kt = 0 and a,-, /?, = 0. Tftew
| (a, (ad H)(K)b) | â ^ I M L , • 1 1 % , a, b e ^
â/ =
/or some /cj = 0 where ax = a0 V a 2 ^ A A) ^ &•
Proof. First for each / > 0 define 7^(iT) as an operator from fj^ into ï)
by setting
2;(À> = UtKU_ta, ab e ^
where Ut = exp{itH}. Then
\{a,Tt(K)b)\ ^ ko\\U_ta\\ao-\\U_rb\\0o
a
= fcoMLo ' Uplift' >h G
6oo-
1
Next for s , ? > 0 define sesquilinear forms
Fst(a, b;K) = 2 ^/r^-(tf, Ç,(W

PÔ Pl-
over h^ X rj^. Then one has the bounds
\FSJ(a, b;K)\£ k0\\a\\ao • H ^ .
But one calculates by rearrangement of convergent power series that
Fst(a, b\K) = 2 t^-(a, Dpt(K)b)
where we have set Dt = (I — Tt)/t. Hence

dp
—pFsM* b> K) = ( - 0 % ( f l 9 b; Dpt(K) ).
Now
(a, Z>?(#)6) = - ^ j \ dr fQ ds(U^r__sa, (ad H)\K)U_r_sb\

o
Hence
| (fl, D^(K)b) | â A:2||fl||Œ2 • ||M| ft , a, 6 e t ^ .
Consequently
\Fs/a9 b; D2tK) ) | ^ * 2 || fl || a2 • ||fc||ft, a, fee I)oo.
Therefore one deduces from the Taylor series expansion
Fs/a, b; K) = (a, #ft) - s(a, Dt(K)b)
+ / ^ Jr(r - *)^,(a, fe; /)?(*) )

that
*| (A, Dt(K)b) | ^ 2^ 0 |k|| a() • ||/>||A) + (k2s2/2) \\a\\a2 • ||è|| f t
for all a, b ^ l)^ and 5 > 0. Finally
I (a, (ad H)(K)b) | = lim | (a, Dt(K)b) \
*-K) +
and hence
| (a, (ad H)(K)b) | ^ kx\\a\\ax ' 11*11/*,

with a, = a 0 V a 2 , /3j = /?0 V /?2 and
fcj = (2k0/s) + (^2/2)
where 5 > 0.
Now we return to the proof that Ki)^ Q t)x when H = 0.
First ^ must satisfy the bound (2.9). Thus
I (a, Kb) I Si cp\\a\\ • \\b\\'p, a,b e &TC.
Second by the assumption of the theorem
I (a, (ad Hf{K)b) | ë *||a||', • \\b\\\, a, b e t1oo.
Hence by Lemma 2.11 there is a £, such that
| (a, (ad # ) ( * ) * ) | S ^llall', • ||*||;, a, b e !>„.
Then one has
| ( (/ + i/)a, A6) | ë | ( (/ + / / ) " V (ad(/ + H?)(K)b) |
+ | ( ( / + i f ) - ' a , # ( / + f/) 2 fc)|
ïi | ( ( / + i f ) " ' a , (ad if) 2 (iO^)|
+ 2| ( (/ + i i ) _ 1 a , (ad H)(K)(I + H)b) \
+ | ( ( / + H)~]a, K(I + Hfb) |
^k\\a\\- \\b\\\ + 2kM\- \\b\\'p + i
+ cp\\a\\ • \\b\\'p+2
for all a, b e ï) œ . But since 1)^ is a core for H it follows that
Kb e / ) ( / + H), i.e., tfï^ ç 6 l .
Next we discuss the properties of the semigroup T. For this we need the
double commutator analogue of Lemma 2.6.
LEMMA 2.12. Let K be a closable operator from i)^ into I). Suppose
H ^ 0 and
| (a, (ad Hf(K)b) \ ^ k\\a\\\ • ||ft||'„ a, Z> G Ï)OO-
Then, for each a G [0, 1]
| (a, (ad (/ + H)a)\K)b) | ^ < A | | < • ||fc||;9 a, * e ^
Proof. The proof is an easy extension of the proof of Lemma 2.6.
Next we reintroduce the approximants
Ke = (I + tH)~pK(I + €//) _ / ? ,
€ > 0, used in the proofs of Theorems 2.5 and 2.8. We assume p ^ 1 and
hence Kfi^ Q i)x. Note that since K ^ 0 one has K€ ^ 0. Now we prove
that K€ + col is 11 -11^-dissipative for a G [0, 1] for co sufficiently large.
LEMMA 2.13. Let K be a positive operator from rj^ into f). Suppose
H ^ 0 and
| (a, (ad (/ + Hff{K)b) \ ^ c o | | < • ||fc||;, a, 6 e r ^ ,
p
/or 5om^ a G [0, 1]. 7 / ^ e = (/ + eHy K(I + c i / ) " ' wifA e > 0 and
/> = 1 ^/ze« ^Te + (<o/2)7 w |H|^-Jm//?^zve.
Proof. Let 6 e fy^ and take « e f)J = ^ _ a such that
*(*) = IWl-«- \\b\\'a.
Now (7 -f H)~a defines a unitary map from fy* into fj* and (7 -f / / ) " is a
unitary map from £)a into fj. Thus
c = (/ + Hyaa e ï, and d = (I + H)% e f).
But since
(c J ) = fl(6) = || fl ||'_ a • ||6||; = Hell • \\d\\
it follows that
c = d = (7 + 7 7 ) ^ G ^
Now with c€ = (7 -f eH)~pc one has
Re a(Ay>) - Re( (7 + 77)ac, Ke(I + if )~ a c)
= Re(c, # e c) + Re(c, (ad(7 + Hf)(Ke)(I + i/)~ac)
^ (l/2)(c e , (7 + 7/)~ a (ad(7 + Hff(K)(I + //)"acc)
^ — (co/2) | k j | 2 S -(co/2)||c|| 2
= -(co/2)inr_a- I N ; ,
i.e., ^ e + (co/2)7is ||||^-dissipative.
Now we are in a position to prove the second half of Theorem 2.10. We
proceed as in the proof of Theorem 2.5 and introduce
K€ = (7 + eH)~pK(I + eH)~p
with p ^ 1 large enough that Ke is bounded. Then K€ i? 0 and it is easily
verified by series expansion that the associated contraction semigroups
rt = exp{-//Q
map l)a into l)a for each a e [0, 1]. But it follows from (2.17), Lemma 2.12,
and Lemma 2.13, that Ke + (a2k/2)I is |H|^-dissipative for a <= [0, 1].
Hence
\\Tta\\'a ^ e'a2k/2\\a\\'a, t ^ 0, a e f,œ

for all a e [0, 1]. The remainder of the proof is identical to the proof of
Theorem 2.5.
Remark 2.14. The semigroup T on rj is a bounded holomorphic
semigroup of angle IT/2 and it follows by interpolation theory that the
restriction of T to each I)a, a e [0, 1], is a bounded holomorphic
semigroup.
Remark 2.15. It is unclear whether the hypotheses of Theorem 2.10 and
the condition H ^ 0 imply that rja, a e [0, 1], is invariant under the
unitary group Vt = exp{itK}.
Next we have a version of Theorem 2.10 based on the stronger
commutation condition of index 2 which does not require H ^ 0.
THEOREM 2.16. Let K be a positive operator from ^ into rj and
suppose that
(2.20) | (a, (ad H?(K)b) | Si k\\a\\ • \\b\\2, a, b e ^
/or so/we /: = 0.
// follows that K is essentially self adjoint and Kfy^ ç rj2. Moreover,
for each a e [0, 2] z7ze contraction semigroup Tt = exp{ — /A^} /JOS z7z£
following properties'.
1. 7;£)a ç t)a, , ^ 0,
ta2k/2
2. ||2; fl || a ^ e \\a\\a, ^ 0 , « G ^
7
3. T ^ w 11 -11^continuous:
Proof It follows from (2.20) by Observation 2.4 that
| (fl, (ad H)\K)b) | ^ fclMI^ • l l i l l ^ , a, ft G ï ) ^
for /? e [0, 2], In particular Kis essentially self-adjoint, by Theorem 2.10.
But applying this bound with ft = 0, 1, 2 one also has
(2.21) | (a, (ad // 2 ) 2 (*)ft) | ^ 4*||a|| 2 • ||6|| 2 , A, b e r ^ .
Hence one can now apply Theorem 2.10 with respect to the positive
operator H2 to obtain the properties of T. Alternatively, one can repeat
the arguments used to prove the theorem with H replaced by H . For
example it follows from (2.21) by an extension of the proof of Lemma
2.6 that
| (a, (ad (/ + H2)a/1)\K)b) \ ^ a2k\\a\\a - \\b\\a, a, b e ï>a,
for all a e [0, 2]. Hence by the proof of Lemma 2.13 one concludes that
K€ + (a2k/2)I is ||-||a-dissipative for a <E [0, 2]. This gives the bound on
||rjtf|| a for the approximating semigroups
and the properties of T\^ are deduced by the same arguments used in the
proof of Theorem 2.5.
Finally since K satisfies (2.9) it follows from (2.20) and Lemma 2.11 that
there is a kx ^ 0 such that
| (a, (ad H){K)b) \ ^ kx\\a\\ ' I ^ W *> b G
*W
Then one deduces that Kfy^ Q fj2 by the estimate
| (H2a, Kb) | ^ | (a, (ad H)2(K)b) | 4- 2| (a, (ad H)(K)Hb) \
4- | (a, iO/2Z>) |
which establishes continuity of a e ^ œ H (H2ay Kb), for each Z> e fy^.
d. Multiple commutators. The preceding results on generators and
invariance properties were based on estimates for first- and second-order
commutators. Bounds on higher order commutators lead to improved
results of invariance [1] [6]. The simplest result is for dissipative
operators.
THEOREM 2.17. Let K be a dissipative operator from fj^ into rj and
suppose that
| (a, (ad HY(K)b) \ ^ kp\\a\\ • \\b\\p9 a, b e ^ p = 1, 2, . . . , m.
Then Kfy^ Q fjm and the contraction semigroup T generated by K satisfies
the following properties, for p = 1, 2, . . . , m\
1. Tt \ C Ï , / ^ 0,
2. ||7;a|| ^ exp{? 2 (^)*,}lMU * ^ 0, a e &„,
3. T\^ is 11-11 -continuous.

Proof. The property ^ r j ^ C f) follows by continuity from the
identity
(Hpa9 Kb) = (a, KHpb)
+ 2 M(A, (adi/)^)^"^), a, b e ^
The rest of the proof is an extension of the argument used to prove

Theorem 2.3. First one proves |||| -dissipativity of K + col for large co = 0
as follows. One has
(2.22) || (/ + iH)p(I 4- €(^ 4- co/) )a\\
â || (/ 4- €(* 4- co/))(J 4- z # ) ^ | | - £ | | (ad (/ + iH)P)(K)a\\.
But then
(ad (/ 4- iH)p)(K)a = 2 P )(ad iH)\K)(I + iH)p~qa

q=\ W/
and hence
(2.23) || (ad (/ + iHY)(K)a\\ ^ 2 (P\kq\\a\\p.

q=\ W/
Now (2.22) and (2.23) together with dissipativity of ^ establish that
K -f <o/ is |H| -dissipative whenever
a
- I (:K
Next one argues that R(I + e(K + coi) ) is ||-|| -dense in fj for small
c > 0. For this suppose there is an / e f^f such that
/ ( ( / + €(* + co/)) fl ) = 0, a G ï,^.
Now setting R = (I + iH)~p one has !?£/* G ÏJ and
(*;/, (/ + eK)a) = - C ( O ( / Î ; / 9 A)
+ €(*;/)( (ad (/ + iH)p)(K)Rpa).

Then using (2.23) and dissipativity of K one finds
| (R*pf, (I + tK)a) |
Ê « o | | * ; / | | • \\a\\ + £ (2 ( ^ ) l l * ; / l l • IMII
3É e(u + i ^ ) ^ ) | | i ? ; / | | • || (/ + e^) a ||.
Since R(I + eK) is dense in Ï) it follows that R*f = 0 for sufficiently small
€ > 0.
Finally one concludes that the ||-|| -closure of K generates a semigroup
with the three properties stated in the theorem. An easy argument then
shows that this semigroup is the restriction of T to t) .
If K is symmetric similar conclusions can be reached for the unitary
group generated by iK but the combinatorics are more complicated.
THEOREM 2.18. Let K be a symmetric operator from ^ into rj and
suppose that
(2.24) \(a,(aàHf(K)b)\^k\\a\\-\\b\\ a, b e h^P = 1, 2,. . ., 2m.
Then Kfy^ Q fj2w and for each a e [0, 2m] the unitary group V generated
by iK has the following properties:
1. VX = t)„, / G R,
2. ||^||0 ^ e^lWU, f e R , a e t,a,
/or some k ~ 0,
Proof The proof is based upon the observation that

| (*, (ad Hf(K)b) | ë *„| W|^ • \\b\\p_q, q e [0, />],
by Observation 2.4. Then it follows by rearrangement that
| (a, (ad (/ + H2)m)(K)b) \ ^ k\\a\\m • ||fc||w, a, 6 e l^
for a suitable k ^ 0, and consequently, by Lemma 2.6,
|a, (ad (/ + H2f){K)b) \ S *||a|| a • ||ft||a, a, b, e t)oo
for a e [0, m\. The rest of the proof then follows the proofs of Theorems
2.5 and 2.10.
Finally one has similar results for positive symmetric operators.
THEOREM 2.19. Let K be a positive symmetric operator from l)^ into I) and
suppose that
| (a, (ad H)P(K)b) | â kp\\a\\ • 11*11,, a, b e l)co,p = 2, 3 , . . ., 2m.
77ze« the semigroup T generated by K has the following properties for each
a <E [0, 2m]:
i. ?;t,a c K, t ^ o,
a2A
2. ||7;a|| a â e' ||fl|| a , r ê 0, a e t,a,
Proof. Now one uses the multiple commutator conditions, and

Observation 2.4, to deduce that
| (a, (ad(/ + H2T)\K)b) | ^ £|M|W • ||i||w, fl, è e i^,
for a suitable k = 0. Then the proof follows the line of reasoning used to
prove Theorems 2.10 and 2.16.
3. Lie groups. In this section we extend the previous results to operators

approximately invariant under the unitary action of a Lie group.
Let U be a strongly continuous unitary representation of the Lie
group G on the Hilbert space JriP and for each positive integer n let J^n
denote the C"-elements of U, i.e, those a e 3tf for which the function
g G G H U(g)a G Jtfis «-times continuously differentiate on G. Further-

more let J^ denote the C°°-elements of U. Then
Next if g is the Lie algebra of G there is a representation d£/ of a on J^

such that the representative dU(X) of X G g is the infinitesimal generator
of the one-parameter group / 1—> U(etX). The representation dU extends
uniquely to a representation of the universal enveloping algebra of the
complexification of g and it can be used to give several operator-theoretic
characterizations of the subspaces J^n.
First if X b . . . , Xd is an arbitrary basis of g then
Jfn = n D(dU(X{r • • • dU(Xd)^)
\a\=n
where a = (a l5 . . ., ad), at = 0, and \a\ = a] -f . . . + aci (see, for
example, [11] Proposition 1.1). Moreover if p0(a) = \\a\\ and
P » = max \\dU(X.). . . dU(X)a\l n g 1,
then J ^ is a Banach space with respect to the norm

n
IkC = 2 P„»
ro=0
(see [11] Corollary 1.1). In addition J^ is a Fréchet space with respect to

the family of norms { ||-||^; n = 0}.
An alternative description of the J^n can be given in terms of the
Laplacian A determined by the basis Xx, . . . , Xd. The operator A is
defined as the closure of the sum
-^t\(dU(X,)f.
It is both self-adjoint and positive. Then it follows from [18] that
j £ = D( (I + A)"/2)
and the norm j|-||^ onJ%t is equivalent to the norm
a H* |M|„ = || (7 + A)" /2 a ||
(see [11] Proposition 1.3). Hence for a ^ Owe introduce the fractional
spaces
Jtfa = D((I + A) a/2 )
with the norm
IWI„ = II ( / + A) a / 2 a||
and if a < 0 we define Jâ to be the completion of Jtf with respect to

the norm
IMIa = | | ( / + A)a/2*||.
Again each J ^ is a Hilbert space with respect to the norm ||-||a and the
operators (/ 4- A)^ /2 define unitary maps from ^ i n t o ^ _ y 8 . Moreover
the dual ^ * of 3Pa is identifiable as Jf_a.
Now it is straightforward to use this characterization of the differenti-
a t e structure of p0f G, U) in terms of the Laplacian to extend the results
of Section 2. Again we consider the three classes of dissipative, symmetric,
and positive, operators.
a. Dissipative operators. First one has a direct analogue of Theorem
2.1.
THEOREM 3.1. Let K be a dissipative operator from J^ into Jtf and
suppose that
(3.1) | (a, (ad dU(X,) )(K)b) \ ^ ka\\a\\a ' \\b\\\-a, a, b & J^
for some a e [0, 1 ) and a ka = 0.
It follows that the closure K of K generates a strongly continuous
contraction semigroup.
Proof It follows immediately from (3.1) that
| (*, (ad A)(K)b) | ^ fc>||1+tt • 11*11, _ a + ka\\a\\a ' \\b\\2-a

for all a, b e j ^ . Now the proof is the same as the proof of Theorem 2.1
but with St = e x p { - / A } .
Next we consider the limiting case a = 0 in Theorem 3.1.
U a = 0 in (3.1) then it follows by continuity, as in the proof of
Theorem 2.3, that KJ^ Q J%. It further follows by the proof of this
latter theorem that K 4- ul is |HI',-dissipative for co ^ k0. But it is not
evident that the semigroup Tt = exp{ — tK} leaves J^ invariant. This
follows, however, if one has appropriate higher-order commutator
estimates.
THEOREM 3.2. Let K be a dissipative operator from J^ into JF and
suppose that
(3.2) | (a, (ad dU(e • X) f(K)b) | ^ cp\\a\\ • \\b\\p, a, b e ^ > ,
for all e e R with \e\ = 1, and all p = 1, 2, . . . , m where m = 2 and
c = 0. Then the contraction semigroup T generated by K satisfies the follow-
ing properties for p = 1, 2, . . . , m\
i. Ttjep ç jep
2. there exists w ^ 0 such that

\\Tta\\'p tk e<V|M|;, / è 0, a e j £
Proof. The proof follows the reasoning used in Theorems 2.3 and 2.17.
We will just outline the argument.
First it follows easily from (3.2) that KJ^ Q J^m. Then setting
Ht = dUiX^ one has for a <= J^
\\Hlx . . . Hlp(I + c(# + co/) )a\\ ^ (1 + ceo) ||/^. . . . fl^||
-€||(ad//Zi...^)(^)^||
by dissipativity of K. Then by (3.2), and linear algebra there is an cop ^ 0
such that
\\(zdHh...Hip)(K)a\\ âp\\a\\'p.
Thus if co ^ co it follows that
| | ( / + £ ( * + co/) )a\\'p ^ \\a\\'p9
i.e., K + co/ is ||||^-dissipative. Next one proves that R(I + e(AT + co/) ) is
||-||;-densein^.
Suppose there is an / e j% such that
/ ( ( / + € ( * + « / ) ) £ ! ) = 0, C7 G ^ .
Then if p is even set
*„ = (/ + A)""72,
and if p is odd set
*, = (/ + A)~(/? + 1)/2.
Thus ^ / 7 ^ = ^ , or J ^ + 1 . In both cases to prove / = 0 it suffices to
prove R*f = 0 because the range of Rp is ||-|| -dense in 3tiPp. But
the proof that R*f = 0 is similar to the proof of the comparable property
in Theorem 2.17.
One concludes from these observations that the ||-||'-closure of K
generates a semigroup with the properties stated in the theorem, and this
semigroup must be the restriction of T to 3^
b. Symmetric operators. Again there are two results based on the weak
and strong form of the commutator bounds of index one.
THEOREM 3.3. Let K be a symmetric operator from Jtf^ into J^ and
suppose that
(3.3) | (a, (ad dU(X,) )(K)b) | ^ k\\a\\U2 • \\b\\V2, a, b e J&,,
for some k ^ 0, and all i = 1, 2, . . . , d.

This follows by applying Theorem 3.1, with a = 1/2, to ±iK.
THEOREM 3.4. Let K be a symmetric operator from J^ into J^ and
suppose that
(3.4) | (*, (ad dU(Xt) )(K)b) | ^ k\\a\\ • \\b\\l9 a, b <= jg^,
for some k ^ 0 and all i = 1, 2, . . . , d. Then the unitary group
Vt = cxp{itK} has the following properties for each a e [0, 1];
1. K = / „ , ( e R,
2. \\Via\\a ë eMakd\\a\\a, t& R,a e ^
3. F | j ^ is | HI ^continuous.
Proof Set i^. = dUiX;). The proof is based on the observation that
| (a, (ad A)(tf)*) |
d
^ 2 { I (^-fl, (ad H$K)V> I + I ((ad HtXK)a9 Htb) \ }
d
^ 2 k\\Hfl\\ • \\b\\x + fcHall, • \\Htb\\
i= \
^ 2kd\\a\U • 11*11,.
Now the result follows from Theorem 2.5 with H replaced by A (see in
addition the proof of Theorem 2.10).
c. Positive operators. The extension of the results of Section 2c is a little
less straightforward because for non-abelian G one also needs bounds on
single commutators.
THEOREM 3.5. Let K be a positive operator from J^ into J4? and
suppose that
(3.5) | (a, (ad dUiX^b) \ ^ kx\\a\\\ • ||6|| l5
(3.6) | (a, (ad dU(Xe) )(ad dU(X;) )(K)b) \ ^ k2\\a\\x ' ll*lli. ^ ^ 4
for some kx, k2 = 0 for all i,j = 1, 2, . . . , d.
Proof. This follows from the proof of the first statement in Theorem
2.12 once one replaces S by the semigroup St = exp{—/A}. But for this
one needs an estimate on (ad A)2(iQ. Now setting Ht = dU(X)
ci ci
(ad A)2(K) = 2 2 (ad i^)(ad HJ)(K)

i«W=i
</ ci
= 2 2 {//,(ad //z)(/*;(ad /$•)(*) + (ad / $ ) ( * )^-)
+ (ad ^ ) ( ^ ( a d /$)(*) + (ad H^KW^}.

Now using the structure relations for g one can re-express the summand as
a linear combination of terms / / m ( a d H)n(K)H ~m where n = 2 and
m = 0, 1, 2, or n = 1 and m = \. Here / / represents an Hi9 H a product
/ ^ . , (ad H)(K) represents (ad Ht)(K) etc. Hence from (3.5) and (3.6) one
obtains a bound
| (a, (ad A)2(K)b) | = C0IWI3 • \\b\\} + Cx\\a\\i ' Mi
+ C2|Wl! • ||*||3.
This suffices for the argument used to prove Theorem 2.12.
Remark 3.6. If G is abelian then the only terms occurring in the
calculation of (ad A)2(K) are of the form / / m ( a d H)2(K)H2~m. No
terms //(ad H)(K)H occur. Hence the bound (3.5) is unnecessary for
abelian G.
Alternatively if one has a bound
\(a,Kb)\ ^ f c 0 N l i -llftlli, a,b e j ^ ,
then (3.5) follows from (3.6) by Lemma 2.11.
COROLLARY 3.7. Let K be a positive operator from 3^ into J^ and
suppose that
\(a9Kb)\ â*olMli -Il*lli>

| (a, (ad dU(Xt) )(ad dU{Xj) )(K)b) | = k2\\a\\x • \\b\\l9 a, b e Jt^,
for some k0, k2 = 0 ««J a// /,./' = 1, 2, . . . , d.
Finally one has generalization of Theorem 2.16.
THEOREM 3.8. Let K be a positive operator from J^ into 3? and
suppose that
(3.7) | (a, (ad dU(Xt) )(K)b) | ^ /qlMI, • \\b\\x,
(3.8) | (*, (ad dU{Xt) )(ad dU(Xj) )(K)b) \ = k2\\a\\ • ||fe||2, a, b e ^ ,
/or 5ome /c1? &2 = 0 Û « J tf// i9j = 1, 2, . . . , J. Then for each a e [0, 2] f/ze
contraction semigroup Tt = exp{ — tK) has the following properties;
1. T&QJtZ tÔ,
2. there is an to {independent of a) such that
\\Tta\\a ^ e,a\a\\a t ^ 0, a ^ Jfa
3. T\jp> is \\-\\ ^continuous.
Proof. The proof is based upon the reasoning used to establish Theorem
2.16 but with H replaced by A. Again it is necessary to have appropriate
estimates on (ad A) (K). But by Observation 2.4 and (3.8) one has
(3.9) | (a, (ad dU(Xt) )(ad dlT^Xj) )(K)b) \
^ k2\\a\\fi • \\b\\2_fi, a.bejg^,
for all 0 e [0, 2]. Now estimating | (a, (ad kf(K)b) \ as in the proof
of Theorem 3.5 but using (3.7) and (3.9), with ft = 0, 1, 2, one finds
a bound
| (a, (ad Af(K)b) \ ^ k\\a\\2 • ||i|| 2 fljie^
for some k ^ 0. Now one can apply the arguments used in the proof of
Theorem 2.10 with H replaced by A.
Remark 3.9. If G is abelian (3.7) is unnecessary (see Remark 3.6).
COROLLARY 3.10. Let K be a positive operator from 3%^ into 3tf* and
suppose that
(3.9) \(a,Kb)\^k0\\a\\r\\b\\],
| (a, (ad dU{X,) )(ad dU(Xj) )(K)b) | â k2\\a\\ • \\b\\2, a, b e 3^
for some k0, k2 = 0 and all z, j = 1, 2, . . . , d. Then the conclusions of
Theorem 3.8 are valid.
The corollary follows because (3.7) follows from (3.9), (3.10), and
Observation 2.4.
4. Commutation theorems. In this section we return to the context

of Section 2 and examine pairs of symmetric operators Kx, K2 satisfying
the criteria of essential self-adjointness given in Theorems 2.5, 2.8, and
2.16. Our aim is to show that if Kx and K2 formally commute, i.e.,
if (ad K})(K2) = 0 as a form on rj^ X t)^, then Kx and K2 commute,
i.e., all bounded functions of Kx and K2 commute. But to establish this
result we need to impose form bounds
| (a, Ktb) | ^ k\\a\\p • \\b\\q9 a,b e ^
on the Kt. Therefore it is natural to slightly reformulate our basic
hypotheses in terms of forms. Then, following [8], we can use the form
bounds combined with commutator bounds to prove the existence of

operators. This is particularly convenient in applications. Thus we now
assume that k, kx, k2 etc. are sesquilinear forms over 3 X 3 where
3 Q f)^ is a core of H and H3 Q 3. Then to emphasize that we
are dealing with forms we now explicitly define the commutators as
forms, e.g.,
(ad H)(k)(a, b) = k(Ha9 b) - k(a, Hb), a, b ^ 3.
In general a sesquilinear form does not determine an operator, unless it is
closable, but Glimm and Jaffe [8] established that it does if one has a weak
commutator bound of index one. We will demonstrate a similar result
using a double commutator bound of index two. But first we give a slight
generalization of the Glimm-Jaffe theorem.
PROPOSITION 4.1. Let H ^ 0 and consider a sesquilinear form k over
3X3 satisfying
(4.1) | (ad H)(k)(a, b) | g kx\\a\\\n ' ll*lli/2> a, b <= fy

for some kx i^ 0. Then for each a ^ 0 the following conditions are
equivalent:
1. \k(a,b)\ ^k'0\\a\\a- \\b\\a, a,b e 0 ,
2. \k(a, b) | â k%\\a\\ • \\b\\2a, a,b^3.
Moreover, if these conditions are satisfied then k determines a unique linear
operator K such that D(K) = \)2a and
k(a, b) = (a9 Kb), a, b e 3.
Proof. Both Conditions 1 and 2 can be extended to f)^ X fj^ by
continuity. Then setting R = (I + H)~ Condition 1 can be reformu-
lated as
\k(Raa9 Rab) | ^ fc'olMI • ||*||, M M M
and Condition 2 as
\k(a9 R2ab) | ^ k^\\a\\ • \\b\l a9b G ^
But if a = n + ft with n integer and 0 ^ /? < 1 then
Jfc(*aa, £ a 6 ) - *(a, i?2afc) = 2 (ad //)(/c)(« a ~ m + l a, 7?a + '"Z>)

m= 1
+ (ad(/ 4- Hf)(k)(Râ9 R2ab)

where the sum is absent if n = 0, i.e., if a < 1. Hence it follows from (4.1)
and Lemma 2.6 that
1264 D E R E K W. R O B I N S O N
n
\k(Raa, R%) - k(a, R2ab) | S k{ 2 \\Ra~m+Xa\\U2
m= \
+ k\\\Rta\\p/2 • | | # 2 ^ / 2
^(nkx + ^ ) | W | - IN
for some k\ â 0. Therefore 1 $=> 2.
Finally if Condition 2 is satisfied then there exists a bounded operator
Ka such that
(a, Kab) = k(a, R2ab) for all a, b e ^
Then one can define ^ on !) 2a by # = Ka(I + / / ) 2 a .
Next we establish an analogue of this result based on a weak double
commutator bound of index two.
PROPOSITION 4.2. Let H = 0tf/?<iconsider a symmetric sesquilinear form
k over 3 X 3) satisfying
(4.2) | (ad H)\k)(a, b) \ ^ k2\\a\\x • \\b\\x, a, b e 0,
/or som^ /c2 = 0. Then for each a e [0, 1] the following conditions are
equivalent
1. \k(a,b)\ ^k'0\\a\\a- \\b\\a a,b^S)
2. \k(a, b) | â ÎWI • \\b\\2a a, b, G 2.
Moreover, if these conditions are satisfied then k determines a unique linear
operator K such that D(K) = ))2a and k(a, b) = (a, Kb), a, b G 3, and
(4.3) | (a, (ad (/ + H)*)(K)b) \ ^ kx\\a\\ ' ll*ll2«> a, b <= @9
Jor some kx = 0.
Proof 2 => 1. Since k is symmetric this follows from Observation 2.4
without use of (4.2)
1 => 2. First we prove a weaker form of (4.3)
(4.4) | (ad (/ + Hf)2(k)(a, b) | â k\\\a\\a • \\b\\a a, b e 3>,
by exploiting Condition 1 and (4.2).
Since a G [0, 1] it follows from Lemma 2.12 and (4.2) that
(4.5) | (ad (/ + Hf)\k)(a, b) \ ^ a2k2\\a\\'a • \\b\\'a a, b G 3.
Next we repeat the proof of Lemma 2.11 with Ut replaced by
Ut" = exp{/7(/ + Hf)
and using the form k in place of the operator K. Thus for s, t > 0 we
define sesquilinear forms
Fst(a, b;k)= 2 e~s/t(^fk(U%ta, U%tb)
where it is understood that k has been extended to ï)a X f)a. Then from
Condition 1 one has
\FJa, b;k)\^ k^\\a\\a • \\b\\a.
But using (4.5) one estimates as in the proof of Lemma 2.11 that
d2 2
—2^,,(a, b; k) ^ a k2\\a\\'a • WbW'a a,b e \)a.
as
Therefore one deduces from the Taylor series expansion of Fst that
\k(Ulta, Ua_tb) - k(a, b) \/t ^ k\\\a\\a • \\b\\a, a, b e i)a
for a suitable k\ i£ 0 and all t â 0. Then (4.4) follows by taking the limit
f-*0.
Next with R = (I + / / ) " ' one has
(4.6) \k(a, R2ab) - k(Raa, Rab) | â | (ad (7 + H)a)(k)(Raa, R2ab) \
â k\\\a\\ • \\b\\
by (4.4). This establishes that 1 <=> 2.
But replacing k by (ad (I + H)a)(k) in (4.6) and using (4.5) one
concludes that (4.3) and (4.4) are equivalent.
Finally if Condition 2 is satisfied then there is a bounded operator K
such that
(a, Kab) = k(a, R2ab) for all a, b e ^
and one can define K on ï)2a by AT = Ka(I + 7/) 2a .
One can immediately deduce versions of Propositions 4.1 and 4.2
without the assumption 7 / ^ 0 .
COROLLARY 4.3. Let k by a symmetric sesquilinear form over 3) X 3) and
suppose that
(4.7) | (ad H)(k)(a, b) \ ^ kx\\a\\ ' H*Hi> *> b e %
for some kx i^ 0. Then for each a = 0 the following conditions are
equivalent',
1. \k(a,b)\ ^k'0\\a\\a- \\b\\a a,b<E2>,
2. \k(a, b) | â kS\\a\\ • \\b\\2a a,b&®.
Moreover, if these conditions are satisfied then k determines a unique
operator K.
Proof. It follows from (4.7) that

| (ad H2)(k)(a, b) | ^ 2kx\\a\\\ " INi-
Therefore the hypotheses of Proposition 4.1 are fulfilled with respect to
the positive operator H . (The norm ||-||a with respect to H corresponds
to the norm | | | | a / 2 with respect to H2). The result follows immediately.
COROLLARY 4.4. Let k be a symmetric sesquilinear form over 2 X Q) and
suppose that
(4.8) | (ad H)\k)(a, b) \ ^ k2\\a\\ • ||*||2, a, b e @,
for some k = 0. Then for each a e [0, 2] the following conditions are
equivalent]
1. \k(a,b)\ ^ ^ o l M l a ' ll*IL a,be@9
2. \k(a, b) | ^ k%\\a\\ • ||*||2a a, b e 9.
Moreover, if these conditions are satisfied then k determines an operator
K and
(4.9) | (a, (ad (/ + H2fn)(K)b) | â *,||<i|| • ||6|| 2a> a, b e %
for some kx = 0.
Proof It follows from (4.8) and Observation 2.4 that
| (ad H)\k)(a, b) | Si k2\\a\\p • \\b\\2_p, a, b e ®,
for p = 0, 1, 2. Therefore
| (ad H2)\k)(a, b) | Ê 4k2\\a\\2 • \\b\\2, a, b e 2,
and the corollary follows from Proposition 4.2, with H replaced by H .
After these preliminaries on form bounds we now state and prove the
commutation theorems for operators satisfying the criteria of the theorems
in Section 2. Again Glimm and Jaffe [8] [9] were the first to prove results
of this nature but subsequent generalizations of their results were given by
Driessler and Frôhlich [3] [6]. These papers all deal with the commutation
of two operators satisfying the conditions of Theorem 2.8, but a partial
result was given by Driessler and Summers [4] for one operator satisfy-
ing the conditions of Theorem 2.8 and the second satisfying the condi-
tions of Theorem 2.16. The next theorem gives a complete result for all
such operators.
THEOREM 4.5. Let Kx, K2 be symmetric operators from fj^ into fy and
suppose they each satisfy one of the following two conditions',
| \(a,Kb)\ =i *olM| 1/2 • ||é|| 1 / 2
| | (a, (ad H)(K)b) | S * , | H | • ||6||, a, b e f,œ,
{
K ^ 0
\(a,Kb)\ ^k0\\a\\x -IHIi
| (a, (ad Hf(K)b) \ ^ k2\\a\\ • \\b\\2 a9 b Œ ^
Further assume that
(4.9) (Kxa, K2b) = (K2a, Kxb) a, b <E bJoo,
It follows that the self-adjoint closures Kx, and K2, commute.
Proof. The proof divides into three cases. Case 1 ; Kx and K2 both satisfy
Condition 1. Case 2; Kx and K2 both satisfy Condition 2. Case 3; Kx
satisfies Condition 1 and K2 satisfies Condition 2.
Case 1. First it follows from Corollary 4.3 that
\(a9Ktb)\ ^k't\\a\\- \\b\\l9 a,b e r ^ ,
for some k% â 0. Thus ï>i ç D(Kt) and
\\Ktb\\ ^ kS\\b\\l9 b e fcb
by continuity. Therefore one deduces from (4.9) that
(Kxa9 K2b) = (K2a9 Kxb\ a, b e ft,.
Next let
^ = expfz^}, j = 1, 2,
then F^rjj = ï)| by Theorem 2.8. Hence if a, b e i)x one has
(MK>,2 - VJVxs)b)
= / / * ^ ( a , VI(K,V] ~ v)Kxyv)_J>)
J Jv
= /I " /o < (K2V2_vvl,fi, K.vlX-ub)
-(K,V2_vVlua, K2VI vVl_ub)}
= 0.
Therefore the unitary groups V and F commute and hence all bounded
functions of Ki and K2 commute (see, for example, [22] Chapter VIII).
Case 2. Now it follows from Corollary 4.4 that
\(a,K,b)\ ÂtflMI- ||ft||2, a, * e b ^ ,
for some A# ^ 0. Thus b 2 £ £(•£;) and
||*,ft|| ^ *tf||ft||2, ft e ft2,
by continuity. Therefore one deduces from (4.9) that
(K}a, K2b) = (K2a, Kxb\ a, b e fj2.

Next let
T\ = exp{-*#,•}, i = 1,2,
then 7^2 ç fy2 by Theorem 2.16. Hence one can compute as in Case 1 and
conclude that the semigroups T 1 and T1 commute. But then all bounded
functions of Kx and K2 commute.
Case 3. It follows as before that rj2 ç ^ c D(KX\ fj2 c Z)(# 2 ), and
\\Kxb\\ g ^Hftll, â kïQ\\b\\2,
\\K2b\\ ^ *£||fc||2,
for some /CQ, A:Q = 0 and all b £= rj2. Thus one again deduces from
(4.9) that
{Kxa, K2b) = (£ 2 a, tf,6), a, fc e t)2.
Next for a > 0 define ATa by
Ka = - J dtU,KU_,
where L£ = exp{/Y//} and K = Kx. Then a simple computation establishes

that A^ satisfies Condition 1.
In particular Ka is self-adjoint, by Theorem 2.8, and rj2 Q fyx C D(Ka)
by Corollary 4.3 and the previous reasoning. But one also has
(ad /0 2 (tf o ) = (ia) _ , {£/ a (ad H)(K)U_a - (ad / / ) ( # ) }.
Therefore
| (a, (ad H)2(Ka)b) \ ^ (2/a)kx\\a\\ ' ll*lln ^ * ^ ^W
Thus if K" - exp{z/^ a } then F , % - I)2 by Theorem 2.18. But it also
follows from Theorem 2.8 that Vf))x = \)x and
(4.10) H^flH, êm'\MU, a e t,,.
Next setting Tt = exp{ — tK2} one calculates as in Case 1 above that
(a, (VfT, - T,V:)b)
= l II du
/ I *<• (R2TvVa_ua, KaTt_vV:_ub)
- (KaTvVlua, K2T,__vV^tlb) }
for all a, b e fy2. But if Ka is replaced by K the integrand is zero, by
commutation of K and K2. Therefore
(4.11) \{a,(V?Tt- TtV«)b)\
= /o du i'o dv^ WK2%V-A • II (*« " K)T,_vVf-J>W

+ || (Ka - K)TvVa_ua\\ • \\K2T,_vVf_J>\\ }.
Now
||^ 2 r v F« u a|| 2 S (VLua, Tv/2K2Tv/2Vtua) \\K2TV\\
by Condition 2 and spectral theory, where

cx = sup \e~ .
Moreover
||rv/2^„a||, g / ^ W i i , ^ ^'V^'iMl,
by Theorem 2.16 and (4.10). Thus combination of these estimates
establishes there is a c s£ 0 such that
(4.12) \\K2TvVlua\\ =i cv~' / 2 ||«||,, a e b2,
for all | ti| s i . a n d O g v S i .
Next note that
f\t\
\\(UtKU.t - K)a\\ ^ J 0 ds\\(adH)(K)U„sa\\
e
=§ \t\kM\u « ftoo
by Condition 1. Therefore
|| (Ka - K)a\\ ë a " 1 f*Q dt\\ (UtKU_t - K)a\\

â (a/2)*,||a||„ a G b^.
Consequently
|| (tfa - K)TvVa_ua\\ ^ (a/2)kx\\TvVa_ua\\,
and arguing as above one concludes there is a c' = 0 such that
(4.13) || ( £ a - K)TvVa_ua\\ ^ ad\\a\\X9 a e f)2,
for all |w| ^ 1 andO ^ v ^ 1.
Combination of (4.11), (4.12), and (4.13), allows us to conclude that
(4.14) | (fl, (VfTt - TtV«)b) | ^ 4acc1Wl! • ||6||,
for all fl,i£^2 and M ^ 1, 0 S f ë 1.
Finally for a e f^
|| (*„ - A > | | ^ a " 1 / " d>{ || (I/, - /)tfa|| + ||tf(I/_ f - I)a\\ }
^ a~l f°Qdt{ || (Ut - I)Ka\\ + k^\\ (£/_, ~ I)a\\x}

where we have used Condition 1 and Corollary 4.3. Thus
lim || (Ka - K)a\\ = 0, A e 6M
a-K)
and hence
Vf -> ^ = e x p { / ^ } as a -> 0
(see, for example, [2] Section 3.1.3). Therefore
(a, (VsTt - TtVs)b) = 0, a, b e fc2
for |s| ^ 1 and 0 ^ / ^ 1, by (4.14). But it then follows from boundedness,
density, and the group property, that V and T commute. Thus Kx and K2
commute.
If H ^ 0 and one considers operators K]9 K2 satisfying weak
commutator bounds of the type occurring in Theorem 2.5 and 2.10
the commutation properties are less evident. But one has the following
partial analogue of Theorem 4.5.
THEOREM 4.6. Let H ^ 0 and consider two symmetric operators Kx, K2
from f^QQ into rj satisfying
j|(a,^)|^/c 0 IMIi/2-|N./2
I $a, (adH)(K,)b)\ S /c,|M| 1/2 • ||è|| 1 / 2 , a, Z> e {,«,.
Further assume that
(4.16) (Kxa, K2b) = (tf2a, * , * ) , a, be ^
It follows that the self-adjoint closures Kx, and K2, commute.
Proof First it follows from Proposition 4.1 that
\(a9Kéb)\ ^ kS\\a\\- \\b\\x, a,b e r ^ ,
for some k$ ^ 0. Thus $x Q D(Kt) and
(4.17) \\Ktb\\ â ^'H*lli. b e ï,„
by continuity. Thus one deduces from (4.16) that
(4.18) (Kxa, K2b) = (K2a, Kxb\ a, b e &,.
Next let
^ = exp{itKj}9j = 1,2.
Then V\\)x/2 = fj 1/2 , by Theorem 2.5 and
||K/fl|| 1/2 ^ exp{cok| } |M| 1 / 2 , a e ï) 1/2
where co = /c 1 /4. The rest of the proof is divided into two parts.
OBSERVATION 4.7. For a, b e ^ arcd t e R
(A, K/Â26) - (K2a, V)b).
Proof. For each € > 0 define
Ke = (/ + eH)~]K2(I + cif)" 1 .
Then it follows straightforwardly with the aid of (4.17) that
lim \\K€a - K2a\\ = 0, a e r^.
£-K)
Next if ^f, 6 G ï)j then
(a, K , 1 ^ ) - ( i ^ , F > ) = 1 / ^ ds{Kxvlsa, KtV]_sb)

x
-(KeV _sa,Kxvlsb)).
But one also has
( £ , ( / + <LH)~xvlsa, K2(I + tH)~xVxt_sb)
= (K2(I + cH)~xvLsa, 7?,(7 + cH)~lV^sb)
by (4.18). Therefore
(4.19) (a, VJKJb) - (Kfi, V]b)
'' = 1 f0 ds{ ( (ad (/ + Œ)-x){K,)Vlsa, K2(I + eH)~xvlsb)
- (K2(I + eH)-]V]_sa, (ad (/ + tH)~x){K{)V^jb) }.

Now consider the first term in the integrand. One has
I ((ad(7 + eHy^K^vl.a, K2(I + tH)-xv}_Jb) I
^ c| ( ( / + €//)~'(ad H)(KX)(I + eH)~]vLsa,
K2(I + Œ)-Wlsb)\
]
êk.Wil + iH)- Vlsa\\i/2
•II (7 + Œ)~XK2(1 + £ / / ) " ' V}_jb\\uï

where we have used (4.15). But
||(7 + c 7 7 ) ^ l F l s a | | l / 2 ë \Wlsa\\U2 fk e^\\a\\\n
by Theorem 2.5. On the other hand
«II (/ + eH)-]K2(I + eH)-]V^sb\\u2
1* cU2\\K2(I + €H)-XV}_M
ê]/\\\(I + eHr]Vl_sb\\i
where the last estimate uses (4.15) and Proposition 4.1. But if E denotes
the spectral measure of H then
c||(/ + eHylV}_sb\\2\
^ Jf°°
0
i„A l
d(V U „,^T,\ ^
t _sb, E(x)V;_sb)(\ +, x)e(\
„wi
+, x)(l
,.n _.-!
+t ex)
Jo
-*€(1
Now x —* 6(1 ++ x)(l x ) - isisunifc
x)(\ ++ eex)~ uniformly bounded on [0, oo) and tends
pointwise to zero as e —-> 0. Moreover
foo 2
Jo0 ' rf(K,'_A £ ( x ) F , L ^ ) ( l + x) = ( I I K ^ H ' , ^ ) < + œ
by Theorem 2.6. Therefore
]imel/2\\(I + eH)-lv}_sb\\\ =°
by the Lebesgue dominated convergence theorem. This establishes that the

first term in the integrand of (4.19) is uniformly bounded on [0, t] and
tends pointwise to zero as e —» 0. The second term has similar properties,
by symmetry and interchange of s and t — s. Therefore
lim { {a, V*KJb) - (K(a, v}b) } = 0
by another application of the Lebesgue dominated convergence theorem.

But since
\\K€a - K2a\\ -> 0 and \\Keb - K2b\\ -> 0
this establishes the observation.
OBSERVATION 4.8. For all sy t e R
(ad VIXVJ) = 0.
Proof. For each 8 > 0 define K8 by
where £/, = exp{/ri/}. Then

(ad / / ) ( * , ) = (i8)-\usK2U„s - K2)
and hence
|| (ad H)(K8)a\\ ^ 2(k0/8) \\a\\x, a e ^
by (4.15) and Proposition 4.1. Thus if
Ff = exp{i7A,}
then F, ï)j = fjls / ^ R, by Theorem 2.8. But one also has
| (a, (ad H)(Ks)b) \ ^ k}\\a\\\/2 ' H*Hi/2> *> b^ ôo

by (4.15). Hence V%/2 = î)1/2, t e R, and
(4.20) \\V?a\\']/2 ^ eMa\\a\\\/29 f e R, a e Ï, 1/2 ,
by Theorem 2.5. Moreover
(a, (tfa - K2)b) = S"1 / ' A / ^ dsi(a, C/,(ad H)(K2)U_sb)

and consequently
(4.21) | ( (a, (tf8 - tf2)&) | ë ( M / 2 ) IWIi/2 • II^Hi/2-
In particular
| (a, (Vf - Vf)b) \=\j'Q Ma, Vs(Ks - K2)Vf_sb)
^ ( k ^ f'0ds\\V2_sa\\u2 • \WLb\\]/2
S {kx8/2)e^\\a\\\n • ||6||'1/2
by another application of Theorem 2.5. Hence
lim (a, (F,2 - Vf)b) = 0
for all (3, b e fy1/2 and then, by continuity, for all a, b e f). Thus
(A, (ad Vxs\v})b) = lim (a, (ad F ^ X ^ ) .
But if a, b e f^ then
(a, (ad F s ')(J/>) = i J'0 du{ (KsVS_ua, v}v?_ub)
- (Vs_ua, VxsK&vlub) }
= i f0 du{ (Ks ~ K2)VS_ua, VJvf_J>)
- {Vtua, Vl(Ks - K2)Vf_ub)}

by Observation 4.8, where it is important that V \)x = i)x. Finally one
estimates that
\(a,(adV^(vf)b)\^(ki8/2)j'0 du{ \\vS_ua\\xn\\vlvlub\\xn
+ \\vl,V*-A\vï\\vlMxa)
a
^ kx8é*^Ha\\\n-\\b\\\/2> > è G *> 1/2
by (4.20), (4.21), and Theorem 2.5. Therefore

(a, (ad Vxs)(V})b) = lim (a, (ad VJ)(vf)b) = 0
for all a, b e rj 1/2 and then, by continuity, for all ay b <= E).
Since the unitary groups F 1 , F 2 commute the operators Kx, K2
commute and the proof is complete.
Remark 4.9. Although no proof of Theorem 4.6 seems to appear in the
literature the result is referred to in notes added in proof to [3] and [6] and
is attributed to Nelson.
5. Lie groups and commutation. Combination of the results of Sections 3

and 4 immediately give a commutation theorem for self-adjoint operators
associated with a unitary representation of a Lie group.
THEOREM 5.1. Let Kx, K2 be symmetric operators from Ji^ into Jf? and
suppose they each satisfy one of the following two conditions'.
I \(a,Kb)\ ^ /colMli/2 ' ll*lli/2>

1. < | (*, (ad dU{Xt) )(K)b) | ^ k2\\a\\ • \\b\\x,
\ i = 1, . . . , J, a, b <= Jtf^
K g 0
2. \(a,Kb)\ ^k0\\a\\i '\\b\U,

< | (a, (ad dU(Xt) )(ad dU(Xj) )(K)b) \ ^ k2\\a\\ • ||*||2,
ij = l,...,d,a9b G J^.
Further assume that
(Kxa, K2b) = (K2a, Kxb\ ay b e 3^.
It follows that the self adjoint closures Kx, and K2, commute.
Proof. It follows from Theorem 3.4 and Corollary 3.10 that Kx and K2
are self-adjoint. But if K satisfies Condition 1 then it also satisfies
Condition 1 of Theorem 4.5 with H = A, the Laplacian associated with
the representation. This is verified in the proof of Theorem 3.4.
Alternatively if K satisfies Condition 2 then it satisfies Condition 2 of
Theorem 4.5 with H = A. This follows because K must satisfy the
hypotheses of Theorem 3.8, as a consequence of Observation 2.4 and
Lemma 2.11. Then the bound on (ad A) (K) is verified in the proof of
Theorem 3.8.
Now Theorem 5.1 is a direct corollary of Theorem 4.5.
6. Partial differential operators. As an illustration of the foregoing

theory we discuss various applications to general partial differential
operators.
Let U be a strongly continuous unitary representation of the Lie group
G on the Hilbert space JF and adopt the notation of Section 3. Moreover
set Ht = dU{Xi). Next let 38 denote the C*-algebra of a bounded operator
on ^f Then G acts as a a-weakly continuous group of *-automorphisms of
38 through the action
rg(A) = U(g)AU(gy\ Aea,geG.
Consequently for each Xt there exists a a-weakly continuous one-
parameter group
t\-* U(etX<)AU(e~tX<)
of *-automorphisms of 38. Let 8X denote the generator of this subgroup.
Then 8X is a a-weak densely defined, a-weakly - a-weakly closed,
derivation with
8X(A)* = -8X(A*).
Moreover A e D(8X) if, and only if, (ad Ht)(A) is a bounded sesqui-
linear form over D(Ht) X D(Ht) or, equivalently, AD(Ht) Q D(H{) and
(ad Ht)(A) has a bounded closure (see, for example, [2] Proposition
3.2.55). If A e D(8X) then 8X(A) is equal to the closure of (ad Ht)(A).
One can associate with the 8X a (^-structure on 38 similar to the
structure on 3tf defined in Section 3 with the Ht. For example, we set
l d
\a\=n
and
n
U\\n = 2 Pm(A)
m= 0
where p0(A) = \\A\\ and

pn(A) = max \\8X, . . . 8 ^ . ^ | | .
Next we examine partial differential operators associated with the

system (3% G, U). By this we mean polynomials in the generators Ht with
coefficients in 38. Nelson and Stinespring [19] established essential self-
adjointness results for elliptic operators with constant coefficients and
similar results for second-order elliptic operators on Banach space. The
Banach space results were then extended to all orders, and significantly
generalized by Langlands [13] [14], but again for operators with constant
coefficients. The subsequent discussion is more restrictive in that it only
applies to first- and second-order operators, but more general insofar as it
allows coefficients in 38. We begin with first-order operators.
If Ai = Af e 38, i = 1,. . . , d then one can define a symmetric

sesquilinear form over 3fé{ X 3fé[ by
K = i 2 ( ^ . + 41$).
7= 1
But if ^ G ^ j then AT defines a symmetric operator on J^ because

d
K= / 2 (Z4,./Ç. + Sx(Aj) )
7
y= l
and one has an estimate
(6.1) \\Ka\\ ^k\\a\\l9 a G 3%.
Next if Ai G ^ 2 then a straightforward calculation, using the structure
relations of g, establishes that
| (a, (ad Ht\K)b) | ê /qMI • ||6||„ ^ J E ^
for some ^, ^ 0 and all / = 1, . . . , d. Hence K is essentially self-adjoint
o n J ^ by Theorem 3.3. But it then follows from (6.1) that Kis essentially
self-adjoint on any ||-|| r dense subspace of 3tf{. Note that it also follows
from Theorem 3.3 that the unitary group Vt = exp{itK} leaves the
subspacesJ^, a G [0, 1], invariant and V\^> is |||| a -continuous. Finally it
follows by the methods of Section 2d that if Al G 38n + x then V leaves
3â, a G [0, «], invariant. Smoothness properties of the coefficients are
reflected by smoothness properties of V.
Next we examine second-order partial differential operators. Let
At- = A* At = Af, A = A* be elements of 38 then
d d
K = 2 H^Hj + i 2 (AJHJ + HjAj) + /I
ij=\ 7=1
is a symmetric sesquilinear form over ^ f X ^f. Moreover if ^ / ; , ,4, G 38X

then iT defines a symmetric operator on 3^ satisfying estimates
(6.2) \\Ka\\ ^ kQ\\a\\2, a G ^
| (a, ( a d Ht)(K)b) \ fà kx\\a\\\ ' H*Hi> *, * e Jéf,
But if Atp Aj G ^ 2 and A & 3&x one has a further estimate
| (fl, (ad ^.)(ad J p * ) * ) I ^ * 2 IMI, • ||i||„ a, 6 e j«f.
Thus the commutator estimates of Theorem 3.5 are verified. Consequently
if K is lower semi-bounded then it is essentially self-adjoint on 3^ by
Theorem 3.5. Strong ellipticity is a simple sufficient condition for
semi-boundedness.
The operator K is defined to be strongly elliptic if there is an e > 0

such that
(Ay ~ eUifSyl)
is positive definite in the sense
d
2 (ai9 (Ay - C I K I I V H O ^ 0, at e JÉf
'\/=i
Note that if ^- = 0 then this condition reduces to positive-definiteness of

(A-) which clearly ensures that
K ^ A -Mil/-
The general case is a small perturbation of this special case.
Set X( = \\AjW2 and adopt the convention that
AA/\ = 0 if Al = 0.
Then for each 8 > 0 one has
J d
8 2 (/Ça, ^ / / / 0 = 8 2 (flj*, (Ay - eX&jDHja)
+ 2 II (ïV^y?- ± (i/VS\My)ûii
7= 1
+ / 2 (a, ( / ^ + ^-)fl)
7=1
d
- 2 IH,«||2/(c8X,.).
7=1
Therefore one has the perturbation estimate

d
2 (a, ( / ^ + AMW
d d
^ 8 2 (HfrAyHfi) + (eSy] 2 M/i|| 2 A;
/,y = l y= l
and in particular
(a,Ka) ^ -(d/e)\\a\\2 + (A, yla) S -((*// € ) + IUH ) IW|2.
Thus K is lower semi-bounded. Hence K is essentially self-adjoint on J^
or, by (6.2), essentially self-adjoint on any ||-||2-dense subspace of J^2.

Finally if A{- e ^ 3 , and Ai9 A e <%2 one has a further estimate
| (a, (ad ^.)(ad Hj)(K)b) \ ta k'2\\a\\ ' ll*ll2, a, b e 3f2
and hence the semigroup Tt = exp{ — tK) mapsJ^ i n t o ^ , for a e [0, 2],
and T\<#> is ||-|| a -continuous by Theorem 3.8.
To conclude we discuss the more specific setting of partial differential
operators on Rd. Thus we choose J^ = L2(Rd), set p. = id/dxj the self-
adjoint operator of differentiation, defined through multiplication on the
Fourier transform, and set #, the operator of multiplication by x,-. Now if
y J J
G = R and U denotes the action of G on Jf 'by translations we can choose

Xj such that K = p. Next if A(q) e ^ denotes the operator of
multiplication by a bounded function
x e Rd hÂ(x)
then v4(g) G ^ if, and only if, x I—» A(X) is Lipschitz, i.e.,
\A(x) ~A(y)\ ^c\x -yl \x - y\ ^ 1,
for some c ^ 0. Moreover v4(#) G ^ + 1 if x I—» ^4(-x) is /t-times
continuously differentiable and the «-th order derivatives are Lipschitz.
Thus the above considerations give self-adjointness and smoothness
results for first and second order partial differential operators with
differentiable coefficients, e.g.
d
(6.3) K = 2 PiA^pj +A(q)
is essentially self-adjoint on C^(Rd) if AJq) e ^ 2 , (AJx) ) is positive-

definite for each x G R , and A(q) e <%. (No differentiability of A is
necessary because it is bounded.) Moreover if A •• e ^ 3 and ^ e ^ then
the semigroup Tt = exp{ — tK) maps the subspaces Jâ into J^, for
a G [0, 2], and T\^ is |H| a -continuous. Note that it also follows from
Theorems 3.5 and 3.8 that these conclusions are still valid if one adds to K
a positive operator
K = 2 capa
with constant coefficients ca G C .

One drawback of such results is that they require boundedness of the
coefficients but such restrictions can be removed by the choice of other
groups. For example, there is a unitary action of the Heisenberg group on
L (R ) given by the Schrôdinger representation. Then we can choose a
basis Xi9 i = 0, 1, . . ., 2d of g such that H0 = I9 H} = pj9 Hd+j = qp
j = 1, . . ., d. But for this action J^ is equal to the Schwartz space
£f(R ) (see, for example, [21] Example 5.1) and hence (6.3) is defined as a
form o v e r ^ Q X Ji^ whenever Atj(q) and A(q) are polynomially bounded
in q. Moreover the hypotheses of Theorem 3.5 are satisfied whenever A-
satisfies the previous hypotheses but A ^ 0, x H-> A(X) is once-
differentiable, and
\VA(x) | ^ cx\x\ + c2 for some c]9 c2 = 0.
Hence one obtains results for A which are 0(x ) at infinity.
Finally we give an improvement of an example discussed by Driessler
and Summers [4]. Let Kbe a positive fourth-order polynomial in the/?, q
Then it follows that K is essentially self-adjoint on S^(RJ), and the
contraction semigroup Tt = exp{—tK) maps each 3tPa into itself, for
a ^ 0, and T\#> is |||| a -continuous. The self-adjointness statement follows
easily from Theorem 2.6 by choosing
d
H = 2 (pj + q]) + /
/= l
to be the Laplacian of the Schrôdinger representation, i.e., the harmonic

oscillator Hamiltonian. Then it follows from the Heisenberg commutation
relations that (ad H)P(K) is also a fourth-order polynomial in the pt, qf for
each/? = 1, 2, . . . . Therefore
|| (ad HY(K)a\\ â kp\\a\\2, a G <?(Rd) = J&,
by Nelson's estimates [18] used in Section 3 for a general Lie group (see
[11] Proposition 1.3). Now the statements concerning K follow from
Theorems 2.16, and 2.19.
Acknowledgements. This work began as a direct result of an invitation
by John Conway to lecture at the Bloomington meeting on 'C*-algebras
and Single Operators'. In the course of this meeting I had a useful
conversation with Larry Brown. A major part of the work was
accomplished whilst I was visiting the University of Toronto at the
invitation of George Elliott. During this visit I had a number of useful
discussions with Ola Bratteli and Palle J0rgensen.
REFERENCES
1. C. J. K. Batty a n d D . W. R o b i n s o n , Commutators and generators, C a n b e r r a p r e p r i n t

(1985)
2. O. Bratteli and D . W. R o b i n s o n , Operator algebras and quantum statistical mechanics I
(Sp ringer- Verlag, 1979).
3. W. Driessler a n d J. Frôhlich, The reconstruction of local observable algebras from the
Euclidean Green's functions of relativistic quantum field theory, A n n . L'Inst. H e n r i
P o i n c a r é 2 7 ( 1 9 7 7 ) 221-236.
4. W. Driessler and S. J. Summers, On commutators and self-adjointness, Letters in M a t h .
Phys. 7 ( 1 9 8 3 ) , 319-326.
1280 D E R E K W. R O B I N S O N
5. W. Faris a n d R. Lavine, Commutators and self-adjointness of Hamiltonian operators.

C o m m u n . M a t h . Phys. 35 (1974), 39-48.
6. J. Frôhlich, Application of commutator theorems to the integration of representations of Lie
algebras and commutation relations, C o m m u n . M a t h . Phys. 54 (1977), 135-150.
7. J. G l i m m a n d A. Jaffe, The (\<p )2 quantum field theory without cutoffs. IV. Perturbations
of the Hamiltonian, J. M a t h . Phys. 13 (1972), 1568-1584.
8. Functional integral methods in quantum field theory in New developments in
quantum mechanics and statistical mechanics (Plenum, 1977).
9. Quantum physics. A functional integral point of view (Springer-Verlag, 1981).
10. Singular perturbations of self adjoint operators, C o m m u n . Pure Appl. M a t h . 22
(1969), 401-414.
11. R. G o o d m a n , One parameter groups generated by operators in an enveloping algebra,
J. F u n c . Anal. 6 (1970), 218-236.
12. T. K a t o , Perturbation theory for linear operators (Springer-Verlag, 1966).
13. R. P. L a n g l a n d s , Some holomorphic semigroups, Proc. N a t . Acad. Sci. (U.S.A.) 46 (1960),
361-363.
14. Semi-groups and representations of Lie groups, Thesis, Yale University (1960)
unpublished.
15. O. M c B r y a n , Local generators for the Lorentz group in the P(<P)2 model, N u o v o C i m e n t o A
7 5 ( 1 9 7 3 ) , 654-662.
16. Self-adjointness of relatively bounded quadratic forms and operators, J. Func. A n a l .
7 9 ( 1 9 7 5 ) , 97-103.
17. E. N e l s o n , Time ordered operator products of sharp time quadratic forms, J. F u n c t . Anal.
II (1972), 211-219.
18. Analytic vectors, A n n . M a t h . 70 (1959), 572-615.
19. E. N e l s o n and W. F . Stinespring, Representation of elliptic operators in an enveloping
algebra, Amer. J. M a t h . 81 (1959), 547-560.
20. R. E. O'Malley, Introduction to singular perturbations (Academic Press, 1974).
21. N . S. Poulsen, On C00-vectors and intertwining bilinear forms for representations of Lie
groups, J. F u n c . A n a l . 9 (1972), 87-120.
22. M. R e e d a n d B. Simon, Methods of modern mathematical physics. 1. Functional analysis.
II. Fourier analysis, self-adjointness (Academic Press, 1972 a n d 1975).
23. D. W. R o b i n s o n , Commutators and generators II, C a n b e r r a p r e p r i n t (1985).
Australian National University,

Canberra, Australia

Commutator Theory On Hilbert Space

Uploaded by

Copyright:

Available Formats

Commutator Theory On Hilbert Space

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Commutator Theory On Hilbert Space

Uploaded by

Copyright:

Available Formats

Can. J. Math., Vol. XXXIX, No. 5, 1987, pp.

COMMUTATOR THEORY ON HILBERT SPACE

1. Introduction. Commutator theory has its origins in constructive

|| (K- H)a\\ ^k\\Ha\\ + /|M|, a e ft^,

for some k e [0, 1) and / ^ 0, i.e., the difference between K and H is

Received October 8, 1986 and in revised form December 8, 1986.

2. Commutator theorems. In this section we develop the theory of com-

equipped with the norm

The space f^ is a Fréchet space with respect to the topology defined by

û ^ e f l o o X Ô o o ^ (*> ( a d H)(K)°) = (#a, ^ » ) - (A, KHb\

(2.1) | (a, (ad # X W I ^ *>H« • ll*lli-«, a. * e ^

// follows that the closure K of K generates a strongly continuous

Next introduce the self-adjoint contraction semigroup S = {S r }^o

which gives an inconsistency if ec < 1. Thus it remains to establish the

+ (Slsc, (ad H){K)HSt(2_s)d) }.

| (C, (ad S,)(tf)V) I = V / l * { IIVHl+« • H^(2-,^lll-a

Now for 0 < s ^ 1, and ft ^ 0, one has

by spectral theory. Consequently

â (1 + c^lMI, - cArllflll, = Hall,

Now R(I + eK) is dense in fy and consequently

and then by interpolation theory (see, for example, [22] appendix to

= ca J Q d\A~ a ( (/ + \ L ) _ , a , (ad / / ) ( ^ ) ( / 4- \L) _ 1 ft).

^ kca J dXX~a\\L]/2(I + \L)_1Û|| • ||L 1 / 2 (/ + XL)"1/?!

But if £ denotes the spectral measure of H then

We note in passing that a slight adaptation of this argument gives

(2.9) \\Ka\\ ^ cp\\a\\'p = cp\\ (7 + HYa\\, a e ^

(2.11) Ml ^ Hall; ^ 2 a / 2 ||a|| a .

one has V'§a = t)a. But

where we have used || (/ + eH) ]\\ ^ 1 and (2.9). Since ^ is a core of AT

(V,- I)a = i f\dsVJCa, a e !)«,,

| | ( ^ - / ) a | | ' 1 / 2 ^ kkW/4||^ll'1/2, Û e t,oo-

(2.16) | (a, (ad H2)(K)b) \

by (2.18). But if eco < 2 this implies a = 0. Hence K is essentially

= ^ /I Jr fo ds J 0 (^)ll i/2 "%^)ll. • \\HmSl(2_r_s)d\U

| (a, (ad Z / ) 2 ^ ) I ^ *2II«H«2 • 1 1 % . a fc e

where Ut = exp{itH}. Then

Fst(a, b;K) = 2 ^/r^-(tf, Ç,(W

Fst(a, b\K) = 2 t^-(a, Dpt(K)b)

where we have set Dt = (I — Tt)/t. Hence

(a, Z>?(#)6) = - ^ j \ dr fQ ds(U^r__sa, (ad H)\K)U_r_sb\

+ / ^ Jr(r - *)^,(a, fe; /)?(*) )

| (a, (ad H)(K)b) | ^ kx\\a\\ax ' 11*11/*,

\\Tta\\'a ^ e'a2k/2\\a\\'a, t ^ 0, a e f,œ

2. ||7;a|| ^ exp{? 2 (^)*,}lMU * ^ 0, a e &„,

3. T\^ is 11-11 -continuous.

The rest of the proof is an extension of the argument used to prove

(ad (/ 4- iH)p)(K)a = 2 P )(ad iH)\K)(I + iH)p~qa

(2.23) || (ad (/ + iHY)(K)a\\ ^ 2 (P\kq\\a\\p.

+ €(*;/)( (ad (/ + iH)p)(K)Rpa).

3É e(u + i ^ ) ^ ) | | i ? ; / | | • || (/ + e^) a ||.

Proof The proof is based upon the observation that

Proof. Now one uses the multiple commutator conditions, and

3. Lie groups. In this section we extend the previous results to operators

g G G H U(g)a G Jtfis «-times continuously differentiate on G. Further-

Next if g is the Lie algebra of G there is a representation d£/ of a on J^

P » = max \\dU(X.). . . dU(X)a\l n g 1,

then J ^ is a Banach space with respect to the norm

(see [11] Corollary 1.1). In addition J^ is a Fréchet space with respect to

(2.1) | (a, (ad # X W I ^ >H« • lllli-«, a. * e ^

+ / ^ Jr(r - )^,(a, fe; /)?() )

| (a, (ad H)(K)b) | ^ kx\\a\\ax ' 1111/,

2. ||7;a|| ^ exp{? 2 (^),}lMU ^ 0, a e &„,

| (, (ad A)(K)b) | ^ fc>||1+tt • 1111, _ a + ka\\a\\a ' \\b\\2-a

\(a9Kb)\ âolMli -Illli>

Jfc(aa, £ a 6 ) - (a, i?2afc) = 2 (ad //)(/c)(« a ~ m + l a, 7?a + '"Z>)

| (a, (ad H)(Ks)b) \ ^ k}\\a\\\/2 ' HHi/2> > b^ ^oo