Beruflich Dokumente
Kultur Dokumente
Managing Editor:
M. HAZEWINKEL
Centre for Mathematics and Computer Science, Amsterdam, The Netherlands
Volume 437
Logarithms and
Antilogarithms
An Algebraic Analysis Approach
by
Danuta Przeworska-Rolewicz
Institute of Mathematics of the Polish Academy of Sciences,
Warsaw, Poland
PREFACE. vii
1. Preliminaries. Introduction to Algebraic Analysis. 1
2. Basic equation. Logarithms and antilogarithms. 11
3. Logarithms and antilogarithms of higher order. 39
4. Logarithms and antilogarithms of operators having either finite
nullity or finite deficiency. 52
5. Reduction theorems. 63
6. Multiplicative case. 82
7. Leibniz algebras. 92
8. Linear equations in Leibniz algebras. 113
9. Trigonometric mappings and elements. 136
10. Semigroup properties of solutions to linear equations 149
11. Operator e hD . 156
12. Power mappings. Polylogarithmic functions. Nonlinear equations. 167
13. Smooth logarithms and antilogarithms. 187
14. Riemann-Hilbert type problems in Leibniz algebras. 201
15. Periodic problems. 216
16. Equations with multiplicative involutions of order N. 236
17. Remarks on the fractional calculus 256
APPENDIX. Functional shifts. By Z. Binderman. 259
AI. Functions of a right invertible operator. 259
A2. Functional shifts. 264
A3. Isomorphisms of spaces of functional shifts. 271
A4. Functional shifts induced by operators of complex
differentiation. 285
A5. Euler-Maclaurin type formulae. 296
A6. Differential and integral properties. 304
REFERENCES 324
SUBJECT INDEX 334
AUTHORS INDEX 341
LIST OF SYMBOLS 343
v
Leonhard Euler in his paper De 1a controverse en-
tre Mrs. Leibniz et Bernoulli sur 1es logarithmes des
nombres negatifs et imaginaires considered the rule
d(log x) = dx / x. He rejected an earlier suggestion of
Leibniz that this rule is only valid for positive real val-
ues of x with the following observation:
"( ... ) Car, comme ce calcul roule sur les quantites vari-
ables, c. a d. sur des quantites considerees en general,
s'il n'etait pas vrai generalement qu'il fOt d . lx = dx/x,
quelque quantite qu'on donne a x, soit possible positive
ou negative, ou meme imaginaire, on ne pourrait jamais se
servir de cette regie, la verite du calcul differentiel etant
fondeesur la generalite des regles qu'il renferme." *)
Memoirs de l'Academie des Sciences de Berlin 5(1749),
139-171, in OPERA, (1) 17,195-232; cf. Fraser F[I]
*) "( ... ) For, at this calculus con cernes variable quantities consid-
ered in general, if it were not generally true that d· lx = dx/x,
whatever value we give to X, either positive, negative, or even imag-
inary, we would never be able to make use of this rule, the truth of
differential calculus being founded on the generality of the rules it
contains."
vi
PREFACE
vii
viii Preface
it easy to understand for the reader acquainted with basic notions of Functional
Analysis and Linear Algebra. The only exception is few times used notion of locally
pseudo-convex and locally bounded spaces. This theory can be found in the book
of ROLEWICZ R[l). Another possibility is to restrict the corresponding results to
Banach spaces. Clearly, these results remain true, but much less interesting (cf.
the following diagram).
All proofs are closed with •. Definitions, examples, notes and remarks are closed
with D.
I hope that this book will be used not only by mathematicians, but also by these
non-mathematicians who are looking for direct methods in order to solve equations
and problems obtained in their works.
It would be nice to say about this book words, which were used by Mercator in
his Logarithmo-technia as a subtitle:
Acknowledgment
A big part of this book has been written during my several visits in Institut fUr
Statistik und Mathematische Wirtschaftstheorie at Universitat Karlsruhe. I am
and I always will be very indebted to Professor Diethard Pallaschke and Authorities
of this University for their efforts in order to organize my life there in the possibly
best way. In particular, I appreciate very much that the preprint, a preliminary
version of this book, has been printed at Universitat Karlsruhe.
My gratitude should be extended to my husband, Professor Stefan Rolewicz, for
his permanent assistance.
The term" Algebraic Analysis" was initially used by LAGRANGE L[1] in 1797 in the
title of his book
Theorie des Fonctions Analytiques contenant Les Principes du Calcul Differentiel,
degages de toute consideration d'infiniment petits, d'evanouissans de limites et de
fluxions, et reduit a I'analyse algebriques de quantites finies
in order to point out that most of results have been obtained by algebraic opera-
tions on analytic quantities. In that general and common sense this name, adopted
also by Cauchy, was used in 19th and in 20th century (d. FRASER F[IJ, JAHNKE
J[1J, also the author PR[26J, PR[28).
The main idea of Algebraic Analysis in its present, more strict, sense derives from
the fact that the differential operator D = ft
is right invertible in several functions
spaces.
Foundations of Algebraic Analysis are following (d. PR[8]).
As usually, denote by N, Z, Q, R, C the sets of all positive integers, integers, ratio-
nals, reals and complexes, respectively. Write No = NU{O}, Il4 = {r E R: r ~ O}.
Let X be a linear space (in general, without any topology) over a field IF of scalars
of the characteristic zero. In the sequel we shall write 1 for the unit in IF and 1.
for ~ . 1 (n EN). We also write n
• L(X) is the set of all linear operators with domains and ranges in X;
• dom A is the domain of an A E L(X);
• ker A = {x E dom A : Ax = O} is the kernel of an A E L(X);
• Lo(X) = {A E L(X) : dom A = X}.
An operator D E R(X) is said to be right invertible if there is an operator R E
Lo(X) such that RX C dom D and DR = I, where I denotes the identity operator.
The operator R is called a right inverse of D. By R(X) we denote the set of all
right invertible operators in L(X). Let D E R(X). Let RD C Lo(X) be the set of
all right inverses for D, i.e. DR = I whenever R E RD. We have
dom D = RX EB ker D, independently of the choice of an R E RD.
Elements of ker D are said to be constants, since by definition, Dz = 0 if and only
if z E ker D. The kernel of D is said to be the space of constants. We should point
out that, in general, constants are different than scalars, since they are elements
of the space X. For all R,R' E RD,X E X we have Rx - R'x E ker D, i.e. a
difference of two primitives of x is a constant. If two right inverses commute each
with another, then they are equal. Let
FD = {F E Lo(X) : F2 = F; FX = ker D and 3RE'RD FR = O}.
1
D. Przeworska-Rolewicz, Logarithms and Antilogarithms
© Kluwer Academic Publishers 1998
2 Chapter 1
If two initial operators commute each with another, then they are equal. Thus
this theory is essentially noncommutative. An operator F is initial for D if and
only if there is an R E 'RD such that
(1.1) F =I - RD on dom D.
Even more. Write'R D = {Ry}-yEf. Then, by (1.1), we conclude that 'RD induces
in a unique way the family F D = {F-y }-YEr of the corresponding initial operators
defined by means of the equality
Some results can be proved also for left invertible operators. Write
• A(X) is the set of all left invertible operators belonging to Lo(X) (we assume
that dom T = X for T E A(X));
• LT = {S E L(X) ST = I} is the set of all left inverses to aTE A(X);
• gT = {G E L(X) G2 = G, kerG = TX and :3 SG = O} is the set of all
SECT
co-initial operators for aTE A(X);
• I(X) = R(X) n A(X) is the set of invertible operators.
Clearly, if ker D -I {O}, then the operator D is right invertible, but not invertible.
Here the invertibility of an operator A E L(X) means that the equation Ax = y
has a unique solution for every y E X. If D E I(X) then FD = gD = {O} and
RD = LD = {D- 1 }.
The above determined sets are, in a sense, dual. Namely, if D E R(X) and FE F D
corresponds to an R E RD, i.e. FR = 0 then R E A(X) and F E gD corresponds
to DE LD.
With these facts one can obtain Calculus and solutions to linear equations (under
appropriate assumptions on resolving equations). If the field IF is algebraically
closed then solutions of linear equations with scalar coefficients can be calculated
by a decomposition of a rational function into vulgar fractions (similarly, as in
Operational Calculus). If the space under consideration is an algebra (a linear
ring), one can obtain some more results. Also there is a reach theory of shifts and
periodic problems. Some of them will be discussed later.
Recall then that any solution of the equation
Dx = y, Y E X, D E R(X),
is of the form x = Ry + z, where R E RD and Z E ker D is arbitrary. This form is
independent of the choice of R. Indeed, if R' E RD and R' -I R then for all x E X
we have R'X - Rx E ker D. So that, a change of R implies only a change of the
constant z.
The equation
Tx = y, yET X, T E A(X)
has a unique solution x = Sy which is independent of the choice of S E LT.
Indeed, by definition, x = STx = Sy and kerT = {O}.
Write
N
Q(D) =L Qk Dk , where D E R(X), Qo, ... ,QN-l E L(X), QN = I.
k=O
_ ))
(Fax J.l"'---,1"""":'-'--'--
(t = a
m(s)x(s)ds
::c for X E X, a E [0,1]' m E X,
Ia m(s)ds
AFa + (1 - A)Fb, AFa + (1 - A)Fb, for bE [0,1], A E IF.
They are also initial operators, because they are projections onto the space of
constants. In several particular cases the corresponding right inverses have eigen-
values.
Introduction to Algebraic Analysis 5
If we consider the space C[0,1] over C then the only continuous Volterra right
inverses are f:,a E [0,1]. However, if IF =
~, then one can find a Volterra right
inverse, which is not of that form (cf. the author and ROLEWICZ PRR[3], PRR[4]).
The Taylor formula applied to a function x E dom DN = GN[O, 1] gives a classical
Taylor formula with a remainder in an integral form. In order to obtain remainders
either in the Lagrange form or in the Cauchy form it is enough to apply the
Darboux property of continuous functions by two different estimations, without
any theorem about the intermediate value, like the Rolle theorem or Lagrange
theorem.
Observe that X = G[O, 1] is a commutative algebra (a linear ring) with respect to
the usual pointwise multiplication. The operator D satisfies the so-called Leibniz
condition:
D(xy) = xDy + yDx far X,y E dom D = G1 [O, 1]
(cf. Formula (1.6)).
We point out that the restriction to the interval [0,1] is not essential. The same
considerations are valid also for any interval [a, b] and for half-axis or the real line.
o
Example 1.2. Let X = G(O), where 0 = [0,1] x [0, 1], and (Dx)(t, s) = :tx(t, s).
we conclude that dim ker D = +00. The induced family of initial operators is
defined as follows: (Fax)(t, s) = x(a, s) for all t, s E [0,1]. 0
Example 1.3. (Difference operators). Let X = (s)JF be the space of all sequences
{x n }, Xn E IF, n E N. Let D{x n } = {X n+l - x n } for all {x n } E X. Then the
operator Rm defined by means of the equality:
ifn:::; m;
(m E N)
if n ~ m + 1;
is a Volterra right inverse of D, i.e. I - ARm is invertible for every A E IF. Each of
the induced initial operators Fm maps any element {x n } into a constant sequence.
Namely, if we write en = 1 for all n E N, then we get Fm{x n } = xm{e n } (m EN).
The kernel of D consists of constant sequences. Thus dim ker D = 1. Using these
definitions, we can solve, for instance, difference equations of the form
N
In general, write
(1.10)
(1.11)
fVlV2(X,y)
= fVl (x, y) + DIfv. (x, y) + +cv l CV2 [(DIX)D2Y + (D2X)DIY]'
Corollary 1.1. Suppose that all assumptions of Proposition 1.3 are satisfied. If
DID2 = D2DI and D E A(X) then the non-Leibniz component for the superpo-
sition can be written in a symmetrized form
(1.12)
(1.13) Dn(xy) = t
k=O
(~) (Dkx)Dn-ky far x, Y E dom D n (n EN),
i.e.
x, y E dom D n (n E N).
Introduction to Algebraic Analysis 9
Recall that in the case of a right invertible D E A(X) the algebra X is said to be
aD-algebra.
Example 1.5. Suppose that X and D are defined as in Example 1.2. Then
X with the pointwise multiplication is a commutative Leibniz algebra, since the
operator D satisfies the Leibniz condition. 0
Example 1.6. Suppose that X and D are defined as in Example 1.3. Then X
with the coordinatewise multiplication is a commutative algebra. It is easy to
verify that the product rule for D is the following:
are said to be quasi-Leibniz algebras (d. DUDEK D[l], who considered quasi-
Leibniz D-algebras, also PR[8]). 0
Example 1.7. Let X and D be defined as in Example 1.1. Then X with the
pointwise multiplication is a commutative algebra. Let n E N be arbitrarily fixed.
Consider the operator Dn. Then CDn = 1 and fDn are defined by Formulae (1.14).
Clearly, D does not satisfy the Leibniz condition for n f. 1, hence X is not a
Leibniz algebra with respect to Dn (n 2:: 2). 0
Example 1.8. Let X be defined as in Example 1.3 and let Dp{x n } = {xn+l -
px n }, where pElF is fixed. By similar arguments as those used in Example 1.3, we
prove that D E R(X). X with the coordinatewise multiplication is a commutative
algebra. It is not difficult to verify that the product rule for D is
M(X) = {A: dom A c X and A(xy) = A(x)A(y) for X,y E dom A}j
10 Chapter 1
I if j = kj
{
Ojk = 0 if j i- k.
CHAPTER 2
We begin with
Definition 2.1. Suppose that D E A(X). Let Or> 0 1 : dom D --+ 2dom D be
multifunctions defined as follows:
(2.1) Oru = {x E dom D : Du = uDx} for u E dom D.
Proof. Let z, z' E ker D. Then Dz = Dz' = 0, zDz' = 0, (Dz')z = 0. Thus z and
z' satisfy right and left basic equations. •
Proposition 2.2. Let D E R(X)nA(X). Then
domO;1 :J{x=R(u- 1 Du)+z: UEI(X)ndomD; zEkerD; RERD}=f:.0;
domO / 1 :J{x=R[(Du)u- 1]+z: uEI(X)ndomD; zEkerD; RERD}=f:.0.
11
D. Przeworska-Rolewicz, Logarithms and Antilogarithms
© Kluwer Academic Publishers 1998
12 Chapter 2
A similar proof for the left basic equation. If D E A(X) then e E dom O. •
Proposition 2.3. Let u E I(X) n dom D. If D E A(X) and either x E Oru,
y E nru or x E OIU, Y E OIU, then x - y E ker D. If D E A(X) and x E Ou,
Y E nu, then x - y E ker D.
Proof. If x, y satisfy the right basic equation then, by definition, Dx = u- l Du,
Dy = u- l Du. Hence D(x - y) = o. A similar proof if x satisfies the left basic
equation. If X is a commutative algebra then Or = 0 1• Again we conclude that
x - y E kerD. •
Proposition 2.4. For any Y C X we write
which implies
Thus ZU E J(X) n dom Or. A similar proof shows that uz E J(X) n dom 0 / .
(iv) follows immediately from (iii) if we apply the commutativity of X. •
Suppose that (u r , x r ) E graph Or, (UI, XI) E graph OJ, Ln LI are selectors of On
0/, respectively, and E r , EI are selectors of 0;:-1,0,1, respectively. By definitions,
Lru r E dom 0;:-1, Erx r E dom On Llul E dom 0,1, E1xI E dom 0 1 and the
following equations are satisfied:
Lr = LI = L, Er = EI = E, (L,E) E G[O).
Clearly, by definition, for all (Lr,Er) E G[Or], (ur,x r ) E graph Or, (Ll,EI ) E
G[Oz], (Ul, xz) E graph Oz we have
Moreover,
(ii) A similar proof in the commutative case. (cf. Definition 7.3, Theorem 7.1 and
Corollary 7.8).
(iii) If Lr E M(X) then
d 1
DLg = -In t = _ = g-1 (far t :j:. 0).
dt t
o
Proposition 2.8. Let X E Lgr(D) (X E Lgl(D), X E Lg(D), respectively).
(i) H(Lr,Er), (L~,E~) E G[Or] and (Ur,X r ) E graph Or, Ur E leX), then
Basic equation 17
Proof. (i) Let y = L~ur. Then y E Orur. By definition and Proposition 2.3, we
have Lru r - L~ur = Xr - Y = z E ker D. This implies that Erxr = Ur = E~L~ur =
E~y = E~(xr + z). The same proof for (ii) and (iii). •
Proposition 2.9. Let X E Lgr(D) (X E Lg1(D), X E Lg(D), respectively).
If (Lr,Er ) E G[Or], (unxr) E graph On L~ur = Lrur + z, E~(xr + z) = Erxr
for a z E kerD then (L~,E~) E G[Or]' If (L/,Ed E G[Od, (UI,X,) E graph 0 1,
L;ul = Llul + Z, EI(xl + z) = Elxl for a z E ker D then (L;,Ef) E G[Od. If
(L, E) E G[O], (u, x) E graph 0, L'u = Lu + z, E'(x + z) = Ex for a z E ker D
then (L',E') E G[O].
Proof. By our assumptions, Dz = 0, Xr = Lrur . Let y = Erx r . Hence
L
m-l m-l
L~ = (I - RjFDj)L r , E~(I - L Rj FDj) = Er;
i=O j=O
18 Chapter 2
m-l m-l
(Erx) (ErY)
(2.10) = Er{CDY + R[CD(Ery)-l(Dx)(Ery)
+ (Ery)-l(Erx)-l fD(Erx, Ery))}.
If X E Lgl(D) then there are R E 'RD and (LI,EI) E GR,dnd such that for
u, v E leX) n dom nl and x = Lu, y = Lv we have
(EIX) (ElY)
(2.12) = EI{CDX + R[CD(EIX)-l(Dy)(EIX)
+ fD(EIX, ElY)(Ely)-l(EIX)-l]}.
If X E Lg(D) then there are R E 'RD and (L, E) E G R,l [n] such that for u, v E
leX) n dom n and x = Lu, y = Lv we have
(Ex)(Ey)
(2.10') = E{CDX + R[CDDy
+ fD(Ex, Ey)(Ey)-l(Ex)-l]}.
= uv = ErLr(uv)
= Er{cDLru + R[CDV-1(DLru)v + V-1U- 1fD(U, v))}
= Er{CDX + R[CDY + (Ery)-l(Dx)(Ery)
+ (Ery)-l (ErX)-l fD(Erx, Ery))}.
Similar proofs for left logarithms and antilogarithms. Let X E Lg(D). Then, by
the commutativity and (2.9) (or (2.11)), v-1(DLu)v = v-1vDLu = DLu. This
implies (2.9') and (2.10'). •
20 Chapter 2
(I - 2cD)DL r e,
(2.13) fD(e, e) = { (1 - 2cD)DL1e,
(1 - 2cD)DLe,
respectively. Moreover,
Proof. Since De = De 2 = cv[eDe + (De)e + fD(e, e)) = 2c DDe + fD(e, e), we find
(2.15) (1- 2cD)De = fD(e,e).
Now, applying the proof of Theorem 2.2 in the case u = v = e, we get Formulae
(2.13) and (2.14). This, and (2.15) together imply for CD i- ~ that DLre =
DLle = DLe = I_leD fD(e, e) = De. •
Corollary 2.7. If X E Lgr(D) then there are R E RD and (Lr,Er ) E GR,dOr)
such that
(2.16) Lru = cDLre + R[cDDLru + u- I fD(U, e)) for u E J(X) n dom Or.
If X E Lg1(D) then there are R E RD and (Ll,Et) E GR,dOd such that
(2.18) (1 - cD)Lru = cDLre + R[u- I fD(U, e)) for u E I(X) n dom Or.
If X E Lg 1(D) then there are R E R D and (LI' EI) E G R,dOd such that
°
Proof. Let F be an initial operator for D corresponding to R. Since (Ln Er) E
GR,l[Or), we have FLru = for U E I(X)ndom Or. Since F = I -RD on dom D,
by Formula (2.18) we get for U E I(X) n dom Or
1 1
Lre = 1
-
2
CD
RiD(e, e) + F Lre = 1
-
2
CD
RiD(e, e).
Theorem 2.3. Let X E Lgr(D) and let (Lr,Er ) E GR,dOr] for an R E RD.
Then the mapping
is constant if and only if CD = 1. Let X E Lg1(D) and let (LI, E I ) E GR,dOd for
an R E RD. Then the mapping
1
Lru = --(cDLre + Ra) for all U E leX) n dom Or,
I-CD
If CD -=I- 0 then Dx = xDe for all x Edam D. On the other hand, if x = e, then we
=
have -cDDe cDeDe = =
cDDe, i.e. 2cDDe O. Since CD -=I- 0, we find De O.=
This implies Dx = xDe = 0 for all x E dom D, which means that D is a trivial
operator. This contradicts our assumption. Hence CD O. = •
Theorem 2.4. Suppose that either X E Lgr(D) or X E Lg1(D) or Lg(D) and
CD = O. Let a E IF\ {O}. If a -=I- 1 then au f/. dom Or whenever u E dom Or (au f/.
dom Ot whenever u E dom Ol, au f/. dom 0 whenever u E dom 0, respectively).
In other words: If CD = 0 then intersections of the domains of the multifunctions
Or, 0 1, 0 with lines passing by zero consist of one point only.
Proof. By Corollary 2.5, there are (Lr' Er) E GR,t[Or] for an R E RD. Suppose
that a E IF \ {O} and u E dom Or. By our assumption, au E dom Or. Since
CD = 0, Formula (2.18) implies that Lru = RgD,r(u). Hence
Lr(au)
= RgD,r(au) = R[(au)-l fD(au, e)]
= R[a-1au-1fD(u,e)] = R[u-1fD(u,e)] = RgDAu) = Lru.
Hence u = ErLru = ErLr(au) = au which implies a = 1. Similar proofs for 01
andbO. •
Corollary 2.8. Suppose that all assumptions of Theorem 2.4 are satisfied. Then
-e f/. dom On -e f/. dom 0 1, -e f/. dom O.
Proposition 2.13. Let CD -=I- O. Let X E Lgr(D) and let (Lr, Er) E GR,l[O] for
an R E RD. Then u 2 E leX) n dom Or if and only if u E leX) n dom Or. If it is
the case then
(2.22)
(2.23) wu+uw =v
with v = ci/u[DLru2 - u- 2fD(U, u)] has a solution w = Lru. Let X E Lg1(D)
and let (LI,Et) E GR,l[O] for an R E RD. Then u 2 E J(X) n dom Ol if and only
if u E leX) n dom Ol. If it is the case then
(2.24)
n-2 n-2
Lu n = (CD L ct + 2cv)Lu + R[u- n L uk fD(U, u n- 1- k )].
k=O k=O
Theorem 2.5. Let a E IF\ {O}. Let CD =1= O. If X E Lgr(D), (Lr,Er ) E GR,dOr]
for an R E RD and u E J(X) n dom Or then au E J(X) n dom Or and
Proof. Suppose that a E IF \ {O} and u E J(X) n dom Or. Since e E dom Or, we
conclude that ae E dom Or. Indeed, by definition,
which implies Lr(ae) = e + z', where z' E ker D. Again by definition, DLre =
eD Lre = De, which implies Lre = e + zit, where zit E ker D. This, and Formulae
(2.9), (2.16) together imply that
= Lr[(-e)(-e)]
= cDL r ( -e) + R[CD( _e)-l DLr( -e)( -e) + (_e)-2 fD( -e, -e)]
= cDL r ( -e) + cDRDL r ( -e) + RfD(e, e) = 2CDL r (-e) + RfD(e, e)
= 2c DL r (-e) + (1- 2cD)L re,
which implies 2CDL r ( -e) = 2cDL re. Since CD =I 0, we get Lr( -e) = Lre. Suppose
that U E leX) n dom Or. By Formula (2.9),
Lr( -U)
= Lr[u(-e)]
=cDL r ( -e) + R[CD( -e)-l(DLru)( -e) + (_e)-lu- 1fD(U, -e)]
= cDLr(-e)+ cDRDLru + R[u- 1 fD(u,e)]
=cD[Lru + Lre] + RgD,r(U).
Similar proofs for left logarithms and logarithms.
Corollary 2.11. Suppose that IF = <C and CD =I O. If X E Lgr(D), (Lr' Er) E
•
GR,dOr] for an R E RD and u E leX) n dom Or, then iu E leX) n dom Or and
Since CD =I- 0, we get the required formula for Lr(iu). Similar proofs for left
logarithms and logarithms. •
Write for A E L(X)
which implies DErx = (I - R xJ- 1[(FErx)x~. Hence Erx is of the required form.
Further arguments are the same as in the proof of Proposition 2.14. •
As it was mentioned, the set WA(X) defined by Formula (2.27) has not been
considered in the commutative case. The following fact (used in the proof of
Proposition 2.16) has not been stated before:
Proposition 2.17. Suppose that D E R(X)nA(X), F is an initial operator for
D corresponding to an R E RD and a E WR(X), Then Dv = va if and only if
v = R(I - Ra+-)-l[(Fv)a] + Fv.
Proof. Necessity. Dv = va = (Fv+RDv)a, which implies (Fv)a = Dv-(RDv)a =
(I - Ra+-)Dv. By our assumption, the operator J - Ra+- is invertible. Hence
v = RDv + Fv = R(J - Ra+-)-l[(Fv)a] + Fv.
Sufficiency is proved by checking.
Proposition 2.18. Suppose that X is a D-algebra with CD "11. Let the mappings
•
gD,T) gD,I, gD defined by Formulae (2.20), (2.21), (2.20'), respectively, be not
constant. Let the mappings A r , AI, A be defined by the following formulae:
1
Aru = -l--[cDe + RgD,r(u)]
-CD
for u E J(X) n dom Or, where R E RDi
1
AIU = -1--[cDe
-CD
+ RgD,I(U)] for U E J(X) n dom 0 1, where R E RDi
1
Au = -1--[cDe
-CD
+ RgD(U)] for U E J(X) n dom 0, where R E RD.
If Ar is invertible then there are (Lr' Er) E GR,dOr] such that Lr = A r , Er = A;l,
i.e. X E Lgr(D). If Al is invertible then there are (Lt,EI) E GR,d0tl such
that Ll = AI, El = A,l, i.e. X E LgJ(D). If A is invertible then there are
(L,E) E GR,l[Od such that L = A, E = A-1, i.e. X E Lg(D).
Basic equation 29
1
DAru = --[cDDe + gD,r(U))
1-CD
far u E J(X) n dom nr .
Hence x = Aru E dom n;1 and Du = uDx. This implies that there are (LT' Er) E
G[nr) such that x = Lru, u = Erx, i.e. Lr = Ar, Er = A;I. By definition,
(Lr. Er) is I-normalized. Similar proofs for Al and A. •
Example 2.2. Let X be a commutative D-algebra with the product rule
fD(X, y) = D(xy) - xDy - (Dx)y = xya + bxy - x(ya + by) - (xa + bx)y =
Basic equation 31
°
(2.3». Similarly, as in (i), suppose that X has unit e and a is an involution. Then
lx, a[= if and only if x = [y, al, y EX. This result is dual to the previous one in
(i) and also can be generalized for almost involutions (d. PR[G]).
(iii) Let X have unit e. Let Dx = xa + bx, where a, b are involutions. Then
x E ker D, i.e. xa + xb = 0, if and only if x = ya - by, y E X. A necessary and
sufficient condition for the equation xa + bx = y to have a solution for ayE X
is that ya - by = 0. If it is the case, then x = !ya + za + bz, where z E X is
arbitrary.
Indeed, suppose that x E ker D, i.e. xa = -bx. Write y = !xa. Since a2 = e = b2 ,
we get ya = !xa 2 = !x and by = !b(xa) = -!b2 x = -~x. Thus x = ya - by.
Sufficiency is proved by checking. Suppose that xa + bx = Y EX. By our
assumptions,
°
Thus the condition ya - by = is sufficient for the equation xa + bx = y to
have solutions. Moreover, !ya is a particular solution of this equation. Clearly
za + bz E ker D for any z EX, which implies the required form of the general
solution to the equation Dx = y.
A duality of the above statements is easy to see if we let: D+x = xa - bx for
x EX. Then they may be rewritten as follows:
32 Chapter 2
provided that the necessary and sufficient condition u-1(Du)a - bu- 1Du = 0 is
satisfied. Thus
by TONG (cf. To[l]) in the algebra of all bounded linear operators over a Hilbert
space. ApOSTOL and FIALKOV (cf. AF[l]) have defined in a Banach algebra A
operators of multiplication MA,B mapping A into itself as follows: MA,B(X) =
AX B for X E A, where A, B E A are given. Every finite sum of such operators:
S = L~l MAi,Bi is said to be an elementary operator. In their cited paper there
are studied structural properties of elementary operators. 0
L
00
L
00
i.e. the operator D is invertible on the subspace X (a, b). Moreover, selectors of
nr and nl are of the form
L
00
L (_1)nb-(n+1)(uau- 1 + b)an
00
Example 2.5. Suppose that X is a non-commutative algebra (over IF) with unit
e. Let a E X be nilpotent of order N i.e. aN = 0 and a k ¥- 0 for k = 1, ... , N - 1.
Let Dx = xa + bx, where bE I(X). Then Dx = y, Y E X, if and only if
N-l
X =L (_l)nb-(n+1)ya n .
n=O
34 Chapter 2
Dx = xa + bx
N-I
= L (_l)n[b-(n+I)ya n+I + b-nya n ]
n=O
N N-I
= - L (_l)nb- n ya n + L
(_l)nb- n ya n
n=l n=O
N-I
- (_l)Nb- N ya N + L
(_l)n[b- n ya n - b-nya n ] + y
n=l
=y.
Observe that, due to the assumption that a is nilpotent, we obtain a result similar
to that given in Theorem 2.6 without any topological assumption about the space
X. It is so, because the series in (2.30) has only a finite number of terms. We
conclude that the operator D is invertible on the whole space X. In order to find
formulae for selectors of Or and OJ it is enough to take in Formulae (2.32) finite
sums with the first N terms. 0
Example 2.6. Denote by X = (s)JR the space of all real sequences {Xn}nEN with
the coordinatewise addition and multiplication by scalars. Let the multiplication
in X be defined as a convolution, i.e. let the product rule be of the form
n
X * Y = {Xn} * {Yn} = {L XkYn+l-d for x, Y E X.
k=l
Then X is a commutative algebra with unit e = {8In } and without zero divisors
(cf. for instance PR[8], Example 6.1.3a). Consider the operator D defined by the
Basic equation 35
x * Dy +y * Dx
n n
= {L:: xk(n + 1- k)YnH-d + {L kXkYnH-k}
k=l k=l
n
= {L [(n + 1 - k) + klxkYnH-d
k=l
n
= {(n + 1) L XkYn+l-d = D(x * y) + x * y.
k=l
Hence CD = 1 and the non-Leibniz component is fD(x, y) = X * y. The basic
equation Du = u * Dx has a unique solution x provided UI ::j; 0 and this solution
determines the logarithmic mapping, i.e. we have
1
L
n 1
Lu =x = {-
n
kUkVnH-k}, where VI = - and
k=l
Xl
Consider now the space the same space X with the coordinatewise multiplication.
Then X is a commutative algebra with unit e = {en}, where en = 1 for all
n E N and without zero divisors. Let D'{x n } = {Xn+l - x n } for x E X. Clearly,
D' E R(X) and the operator R' defined as follows: R'x = Y, where Yl = 0,
Yn = I:;';;;~ Xk for n 2: 2, is a right inverse of D'. Moreover,
ker D' = {z = {zn} EX: V zn
nElIi
= C E lR}.
The product rule is of the form
n
Xl = C E lR, xn=L
k=O
o
Example 2.7. Suppose that X E Lg(D) and (L,E) E G[Ol. Consider the
Cartesian product
X = 'X- -x..... .x. -X,' where x = (Xl, ... ,xn ) E X for Xl, ... ,xn E X,
n-times
36 Chapter 2
----
Ul, "',U n E J(X). Write also
H(u,x) E graph 0 1 then (Au, Ax) E graph 01 and LIAu = ALIU, EIAx = AElx.
Proof. Let (u, x) E graph 0 1. Then, by definition, Du = (Dx)u, x = Llu, U = Elx.
Let x = Ax. Then, by (2.33),
O~u = {x E dom D': D'u = uD'x} for u E dom D', where D' = d- l D.
Basic equation 37
Then (Au, Ax) E graph D~ whenever (u,x) E graph D~ and there are (L~,E~) E
G[D~]such that L~Au = ALru, E~Ax = AErx.
Proof. By definition, dom D' = dom D. Let (u,x) E graph Dr. Then Du = uDx,
x = Lru, u = Erx. Let x' = Ax = ALru. Then, by (2.33),
D' Au = d- l DAu
= d-ldADu = ADu = A(uDx) = (Au)ADx
= (Au)d-ldADx = (Au)d- l DAx = (Au)D'Ax = (Au)D'x',
i.e. there is no non-trivial initial operators. Left invertible operators can be studied
in a similar manner. In both cases of left invertible and invertible operators the
multifunctions Dr, Dl and D are single-valued. Moreover, if D is left invertible
then dom D = X. Hence the assumption that the domain of D is a sub algebra of
X is automatically satisfied. Nevertheless, we shall consider later some properties
of logarithmic and antilogarithmic mappings induced by left invertible operators.
CHAPTER 3
The equations
(3.3)
are said to be the right and left basic equations, respectively. Clearly,
If D E A(X) then Or,n = Ol,n for all n E N and we write Or,n = Ol,n = On. If it
is the case, then right and left basic equations do coincide. If it does not lead to
any misunderstanding then we shall omit the index n = 1, i.e. we shall write (as
before), Or,1 = Or, !i 1,1 = 0 1, 0 = 0 1 . 0
Clearly, the multifunctions now defined have similar properties as these introduced
before and we shall not repeat all considerations of Section 2.1. We have only to put
D n instead of D. Recall that a power of a right invertible (left invertible, invertible)
operator again is right invertible (left invertible, invertible, respectively).
By definitions,
Let n E N. Suppose that (un x r ) E graph Or,n, (UI, xL) E graph Ol,n, Lr,n, LI,n
are selectors of Or,n, Ol,n, respectively, and Er,n, EI,n are selectors of O;'~, O~~,
respectively. By definitions, Lr,nur E dom O;,~, Er,nxr E dom Or,n, LI,nUI E
dom O~~, E1,nXI E dom !il,n and the following equations are satisfied:
39
D. Przeworska-Rolewicz, Logarithms and Antilogarithms
© Kluwer Academic Publishers 1998
40 Chapter 3
(3.7)
where W E ker Dn. If w = 0 then y is of the form (3.9). Suppose that w f:. O. By our
assumption, F Diy = cDF Di Lr,nu = 0 for j = 0, ... , n - 1. Put y' = Rn+1 Dn+1y
(cf. Proposition 2.10). Then FDiy' = FRn+1-iD n+1 y = 0 for j = O,l, ... ,n, i.e.
y' is of the form (3.9). We still have to prove that y = Lr,n+l U, where L r,n+1 is an
invertible selector of Or,n+1. The basic equation Dn+1u = uDn+1y together with
the conditions F Diy = 0 for j = 0,1, ... , n can be written as follows:
+L
n
y = Rn+1(u- 1 Dn+1u) RiFDiy = R n+1(u- 1 D n+1 u ).
i=O
Hence for every u E J(X) n dom Or,n+l there is a unique y E dom 0;';+1. This
implies that there exists an invertible selector E r,n+1 of 0;';+1 and L r,n+1 =
-1
E r,n+l·
In order to complete our inductive proof, we have to show Formula (3.9) for n = 2.
This is proved in the same manner, as the implication n --t n + 1. We have only
to apply (3.9) for n = 1.
Similar proofs for nth left logarithms and antilogarithms and for nth logarithms
and antilogarithms. •
In a similar manner as Theorem 2.2 we prove
Theorem 3.2. Let 2 ~ n E N. If X E Lgr(D) then X E Lgr(Dn) and there are
R E RD and (Lr,n, Er,n) E G R,n[Or,n) such that for u, v E J(X) n dom Or,n and
x = Lr,nu, y = Lr,nv
(3.11)
(3.12)
= Et,n{ci)x + Rn[ci)(Et,nx)-l(Dny)(Et,nx)
+ fj;') (EI,nX, EI,ny)(EI,ny)-l(EI,nX)-l)}.
Higher order 43
If X E Lg(D) then X E Lg(Dn) and there are R E RD and (Ln, En) E GR,n[On)
such that for u, v E leX) n dom On and X = Lnu, y = Lnv
(3.11')
Corollary 3.1. Right (left) logarithms and antilogarithms of order n (if they
exist) are uniquely determined up to a D-polynomial of degree n -1, i.e. elements
ofker Dn (cf. Corollary 2.4).
Corollary 3.2. Let n E N. If X E Lgr(D) then there are R E RD and
(Lr,n, Er,n) E GRn,dOr,n] such that
(3.13)
Lr,nu = cDLr,ne + Rn[CDDnLr,nu + u- 1 f1n )(u, e)] for U E leX) n dom Or,n,
i.e.
(3.14)
i.e.
(3.16)
If X E Lg(D) then there are R E RD and (Ln, En) E G Rn ,1 [On] such that
i.e.
(3.14')
(3.17)
(3.18)
44 Chapter 3
If X E Lg(D) then there are R E RD and (Ln' En) E G Rn,1 [Or,n] such that
(3.17')
o
Open Question 3.1. Does there exist a D-algebra X with unit which is a Leibniz
Dn-algebra for an n ~ 2 ? 0
Theorem 3.3. Let n E N. Let x E leX). If X E Lgr(Dn), (Lr,n, Er,n) E G[Or,n]'
a E X and Rna E dom O;:;~ for an R E RD, then
(3.21)
Then
(3.26)
Proof. We shall prove only Formulae (3.23) and (3.24). The proofs of Formulae
(3.25), {3.26} are similar. Let then n E N. By our assumptions, for U Edam Or,nn
dom Ol,n we find
n-1 n-1
(3.31) II En(Rjx) = En(II Rjx) = E 1x.
j=O j=O
Lemma 3.1. Let X be a D-algebra with CD ¥- O. Then for all u E I(X) n dom D
•
r
= c /u- 1[De - cD(Du)u- 1 - fD(U,U- 1)]
-l[D e - CDU -lDU - f D(-1
= CD U ,U)]-1
U .
If X E Lgl(D), (L1,n, El,n) E G[OI,n], (n = 1,2), and U E J(X) n dom 01,2, then
for an R E RD
(3.33)
L 1,1(iu) = cDL1,2U + R2{(DL1,lu)[De - cDDL1,lU - iD(U,U- 1)] + fD(Du,u- 1n·
If X E Lg(D), (Ln' En) E G[On], (n = 1,2), and u E J(X) n dom O2 , then for an
RERD
Proof. Let u E I(X) n dom Or,2. Then, by definitions and Lemma 3.1,
D 2 L r ,1(iu)
= D[DL r,l (iu)D(iu)] = D[-iu- 1iDu] = D(u- 1Du)
=CD[U- 1D 2u + (Du- 1)Du] + fD(U-l, u)
=cDLr,2U + CDCv1u- 1[De - CDU- 1Du - fD(U- 1,u)]u- 1Du
+ fD(U- 1, Du)
=cb[Lr,2U + [De - cDDLr,lU - fD(U- 1,u)]DLr,lU + fD(U- 1, Du).
48 Chapter 3
By Corollary 3.1, we get (3.32) for an R E RD. Similar proofs for (3.33) and
(3.32'). •
Corollary 3.6. Suppose that all assumptions of Theorem 3.5 are satisfied and X
is a Leibniz algebra. Then
Ll(iu) = L 2u - R2(DLIU)2,
respectively.
Example 3.1. Let X E Lg(D) be a Leibniz D-algebra. By (3.1), if (Ln, En) E
GR,n[On] for an R E RD and u E dom On (n E N) then
o
Example 3.2. Let IF = ~, X = C(~) and D = ft.
Then X with pointwise
multiplication is a commutative Leibniz D-algebra. By Theorem 3.1, X E Lg(Dn)
for all n E N. By Corollary 3.1, logarithms and antilogarithms of order n are
uniquely determined up to a polynomial of degree n -1. Let u(t) = eAt. Note that
here and in the sequel we shall denote by "e" the basis of an exponential function,
i.e. the number
00 1
e=2:-
n=O n!
in order to distinguish between this number and the unit e of the algebra under
consideration. Fix n E N. Let (Ln' En) E G[f!n]. Since Dnu = ).nu, by definition,
we have DnLnu = u-1Dnu = ).n. This implies that
o
Example 3.3. Denote by C(Oj IF) the space of all functions continuous on a set
o and with values in IF (considered as a linear space over IF). In this example 0
will be a closed interval in R
Fix c: > O. Let X = C([c:,7r - c:];~) and let D = ft.
Then X with pointwise
multiplication is a commutative Leibniz D-algebra belonging to Lg(D). By similar
arguments to those used in Example 3.2, for u(t) = sin t, we find
1 . tj
Lju = -~(arcsin u)1 = ~ (j = 2k, kEN),
J. J.
Lju = J E:
t (t - s)j
.,
J.
In sin sds (j = 2k + 1, kEN).
Ljv = I t
-7r/2+E:
(t
.,
J.
s)j
In cos sds (j=2k+l, kEN).
i arccos v = Ll (v + i ~) E Y'.
o
Example 3.4. Let X be a noncommutative algebra (over JR or C) with unit e
and a complete linear metric space. Similarly as in Theorem 2.6, Formula (2.30),
write for a E X, bE J(X):
= {y EX: L
00
(3.35) D 2 x (n+l)(_I)nb-(n+2)ya n ,
n=O
i.e. the operator D2 is invertible on the subspace X 2(a, b). Moreover, selectors of
Or,2 and 01,2 are of the form
=L
00
=L
00
= xa 2 + b2x + 2bxa
=L
00
L
00
L
00
Let x = LT,2U for such u that a2 + u- 1b2u + 2u- 1bua E X 2(a, b). Then
Example 3.6. Let X and D be defined as in Example 3.5 and let a, b be idem-
potents, i.e. a2 = a, b2 = b. By an easy induction, we get
Indeed,
L -L 2n-l(L) (2n-2)(2n-2n+l)
T,n U - T,lU - ~ T,lU a+ 2n a.
A formula for left logarithms is much more complicated. A similar result for left
logarithms can be obtained if a = e. D
CHAPTER 4
Let A E L(X). Recall that the nullity and the deficiency of A are
respectively (cf. the author and ROLEWICZ PRR[I]). Clearly, the index KA is
finite if and only if there is a K E JC(X) such that A = T + K, where T E I(X).
Theorem 4.1. Suppose that D E R(X), K E ,qX) and D' = D + K. Let
X E Lgr(D), (Lr,Er) E G[OT]' Write
52
D. Przeworska-Rolewicz, Logarithms and Antilogarithms
© Kluwer Academic Publishers 1998
Operators with finite nullity and/or deficiency 53
Then the mapping L' is a selector of 0'. If L' is invertible and E' = (L,)-l then
(L',E') E G[O'] and
Proof Let X E Lg1(D), (Ll, El) E G[Od and let D' = D + K. Let u E J(X) n
dom 0 1 and let x = L/u. Then E/x E leX). Let x' = x+R[(Ku)u-1-Kx]. Then
and
D'u = Du + Ku = (Dx)u + Ku = [(D'x) - Kx]u + Ku =
= D'{x + R[(Ku)u- 1 - Kx]}u = (D'x')u.
Hence Dx' = DL;u which proves that Li is a selector of 0;. If L; is invertible,
E; = (LD- 1 and x = LIU then E1x = u = E;Liu, which implies (4.4). Similar
proofs for selectors of Or and O. _
Definition 4.1. By Lgr#(D) (LgI#(D), respectively) we denote the class of
algebras with the operator DE A(X) such that D E A(X) and Or = Lr (0 1 = L l ,
respectively) is invertible. If D E A(X) then Lgr#(D) = LgI#(D). This class is
denoted by Lg#(D). D
In the same manner as Theorem 4.1 we obtain
54 Chapter 4
Then the mapping L' is a selector ofn'. If L' is invertible and E' = (L,)-l then
(L', E') E G[n'] and
where <Pl, ... , <Pn belong to a conjugate space X' (i.e. a total space of linear func-
tionals over X what means that <p(x) = 0 for all <P E X' and an x E X implies
x = 0) and Xl, ... ,xn E X are linearly independent. Then the operator 1- RK
(respectively, 1- SKY is invertible if and only if
where CPI, ... ,CPn belong to a conjugate space X' (cf. Lemma 4.1), Xl, ... ,Xn E X
are linearly independent. If there is an R E RD (respectively, an S E .cD, a
T E I(X)) such that
(respectively, ~(S) =f:. 0, ~(T) =f:. 0) then there exist invertible selectors ofO~, OJ
and 0', which are of the form (4.1) (respectively, of the form (4.1'), (4.3), (4.5),
(4.5'), (4.7)).
Proof. Let (u, x) E graph Or. By our assumptions, we have X = Lru, i.e. x' = L~u
and
x' = L~u = Lru + R[u- l Ku - KLru] = x + R(u- l Ku - Kx).
This implies (J - RK)x = x' - R(u- l Ku) = w. Putting V = x in Lemma 4.1 we
conclude that for every x' there exists a unique u such that L~u = x' (x is already
fixed). This means that L~ is invertible. The same arguments are used in the case
of left logarithms, logarithms and left invertible operators. •
As in Chapter 2, we get
Proposition 4.1. Suppose that D E A(X) n A(X). Then
dom Or:::> {x = S(u-IDu): u E J(X) ndom D;S E .cD} =f:. 0;
dom 01 :::> {x = S[(Du)u- l ] : u E J(X) n dom D; S E .cD} =f:. 0
56 Chapter 4
(4.9)
(4.10)
In the case of left invertible and invertible operators some pathologies may appear,
as it is shown by
Theorem 4.3. Let A E A(X). Consider the following conditions:
(i) X is a Leibniz algebra, i.e. CA = 1 and fA = 0;
(ii) X has a unit e;
(iii) ker A = {O}, i.e. A is left invertible.
Then these 3 conditions cannot hold simUltaneously, i.e. if ker A = {O} then either
X is not a Leibniz algebra or X has not a unit.
Proof. Suppose that X is a Leibniz algebra with unit e and ker A = {O}. Then
=
Ae Ae2 = = = =
2eAe 2Ae which implies Ae O. Thus e 0, a contradiction. •
Example 4.1. Suppose that X = G[O, T] and
D = :t with dom D == {x E G'[O, T] : x(O) = O}.
It is easy to verify that X (with respect to the usual addition, multiplication by
scalars and multiplication of elements) is a Leibniz algebra without unit and that
the operator D is invertible and its inverse is D-1 = J;. 0
This example can be generalized in the following manner:
Operators with finite nullity and/or deficiency 57
Proposition 4.4. Suppose that X E Lg#(D). Then the logarithm and antiloga-
rithm of zero are not defined.
Proof. Suppose that L E I(X). If x = L(O) then x should satisfied the equality
o= DO = ODx. Thus L(O) is not well determined. By definition, we have
DE(O) = E(O)D(O) = O. Since ker D = {O}, we conclude that E(O) = O. This
implies that L(O) = LE(O) = 0, a contradiction (cf. Corollary 4.3). •
Proposition 4.5. Suppose that X E Lg~(D), S E LD, e E DX and 9 = Se.
Then 9 E dom L if and only if 9 E J(X). In that case, 9 is not a zero divisor.
Proof. Since e E DX and Se = 9 = SDg, we get Dg = e. The basic equation
implies that e = Dg = gDLg = (DLg)g. Clearly, 9 is not a zero divisor (cf.
Corollary 2.2). •
Theorem 4.4. If X E Lg# (D) and (L, E) E G[L] then for u, v E J(X) n dom L
and x = Lu, y = Lv we have, independently of the choice of an S E LD,
L(uv) = SDL(uv)
= SD{CD(Lu + Lv) + S[U-1V- 1fD(U,v)]}
= cD(Lu + Lv) + S[u-1v- 1fD(U, v)]
(cf. the proof of Theorem 2.2). By Proposition 4.4, Formula (4.12) is independent
of the choice of a left inverse S, and Formula (4.13) is an immediate consequence
of the definition and (4.12). •
Corollary 4.4. Suppose that CD :/: 0, X E Lg#(D) and (L,E) E G[L]. Then
n dom L if and only if u E J(X) n dom L. In that case
u 2 E J(X)
Proof. Formula (4.14) follows immediately from our assumptions and Theorem
4.4 (cf. Proposition 2.11). The remaining part of the proof is similar to that of
Theorem 2.3. (cf. also Theorem 4.1). •
Theorems 4.1, 4.3 and Proposition 4.4 together imply (cf. Proposition 2.12) the
following
Proposition 4.6. Suppose that X E Lg~(D) and (L, E) E G[L]. Then
(i) (1- 2cD)Le = SfD(e,e) , where S E CD;
(ii) (1 - 2cD)DLe = (1 - 2cD)De = fD(e, e);
(iii) CD = 1/2 implies fD(e, e) = 0;
(iv) CD ::j: 1/2 implies Le = e .
Similarly to Theorems 2.4, 2.5 and Corollaries 2.8-2.11 we obtain
Proposition 4.7. Suppose that X E Lg#(D) and CD = O. Let a E IF \ {OJ.
Let u E dom L. If a ::j: 1 then au ¢ dom L. In particular, if X E Lg~(D then
-e ¢ dom L.
Proposition 4.8. Suppose that X E Lg~(D) and CD ::j: O. Let a E IF \ {OJ. Let
u E dom L. Then au E dom L and
Proof. Let a E IF\ {OJ. Let u E dom L. Theorems 4.4 and 4.5 together imply that
L(au)
= L(u· ae) = cD[Lu + L(ae)] + Sa-1u- 1fD(U, ae)
= cD(Lu + L(ae)] + a-1au-1SfD(U, e) = cD[Lu + L(ae)] + SgD(U)
= Lu - cDLe - SgD(U) + cDL(ae) + S9D(U) = Lu + cD[L(ae) - Le] .
Proposition 4.9. Suppose that D E A(X) n A(X), X has unit e and CD ::j: 1. Let
•
the mapping A with dom A :J dom n be defined by the following formula:
1
Au = -l--[cDe
- CD
+ SgD(U)] for u E J(X) n dom A, where S E CD,
and the mapping gD(U) = u- 1 fD(U, e) defined by Formula (2.20') is, by Theorem
4.3, not constant. If A is invertible then (A, A- 1) E G[n], i.e. X E Lg(D).
Proof By definitions, for all u E J(X) n dom A we have
1 1
SDA = Au = -l--[cDSDe + SgD(U)] = S--[cDe + gD(U)],
- CD 1- CD
60 Chapter 4
which implies
1
DAu = -l--[De + gD(U)]
-CD
(Up to an s E ker S which can be assumed to be zero). Further arguments are the
same as in the proof of Proposition 2.18. •
In the same way, as in Chapter 3, we introduce right and left logarithms and
antilogarithms of the higher order. Instead of Theorem 3.1 and Corollaries 3.1,
3.3 we have the following
Theorem 4.6. Let the IDultifunctions Or,n, Ol,n, On (n E N) be defined as
in Definition 3.1. Let (Lr,n, Er,n) E G[Lr,nj, (Ll,n, EI,n) E G[LI,nj, (Ln, En) E
G[Ln]. Then dom Lr,n+1 C dom Lr,n, dom L1,n+1 C dom L1,n, dom Ln+1 c
dom L n , respectively (n EN). If X E Lgr#(D) then X E Lgr#(Dn), if X E
Lgl#(D) then X E Lg1#(Dn), if X E Lg#(D) then X E Lg#(Dn) for all n E N.
Moreover, if (Lr,l, Er,l) E G[Lr,lj, (L1,1> El,d E G[LI,lj, (L1' Ed E G[Llj, then
(Lr,n, Er,n) E G[Lr,nj, (Ll,n, El,n) E G[L1,nj, (Ln, En) E G[Ln] for every n E N
and for (ur,xr), (vr,Yr) E graph Lr,n, (Ul,Xc), (Vt,Yl) E graph Ll,n, (u,x), (v,y) E
graph Ln we have
(Er,nx)(Er,nY)
(4.17) = Er,n{CD(X + y)
+ sn[(Er,nx)-l(Er,ny)-l fi;')(Er,n x , Er,nY)]} ,
L1,nU
(4.18) = cDL1,n-lU + sn[cD(DL1,lU)D n- 1LI,n-lU
+ u- l fD(U, D n- 1L1,n-lU)] (n ~ 2),
(El,nx)(E1,nY)
(4.20) = El,n{CD(X + y)
+ sn[(E1,nX)-1(El,ny)-1 f1n)(El,nx ,E1,nY)]},
Operators with finite nullity and/or deficiency 61
Lnu
(4.15') = cDLn-lu + sn[cD(DL1u)D n- 1Ln-1U
+U-l!D(U,Dn-1Ln_lU)] (n ~ 2),
(4.16')
(Enx)(EnY)
(4.17)
= En{CD(X + y) + sn[(Enx)-l(Eny)-l !1n)(Enx, EnY)]},
independently of the choice of S E CD, i.e. Lr,n, Er,n, LI,n, EI,n, L n, En (n E
N are uniquely determined. If X E Lgr#(D) (X E Lg1#(D), X E Lg#(D),
respectively), then
(4.21)
(4.22)
(4.21')
respectively.
Lemma 4.2. Suppose that D E A(X) n A(X). If DE ML(X) and S E CD then
S E ML(X).
Proof. Let D E ML(X). Let x,y E X, S E CD and let u = Dx, v = Dy. Then
x = Su, y = Sv and S(uv) = S[(Dx)(Dy)] = SD(xy) = xy = (Su)(Sv). •
By similar arguments to those used in the proof of Theorem 3.4, applying Lemma
4.2, we obtain
Theorem 4.7. Suppose that X E Lg#(D) and D E ML(X). Then X E
Lg#(Dn) for all n E N, (Ln,En) E G[Ln] n ML(X) and for (u,x) E graph L
andnEN
n-l
n-l n-l
(4.24) II En(Sjx) = En(II sjx) = E1x
j=O j=O
REDUCTION THEOREMS
63
D. Przeworska-Rolewicz, Logarithms and Antilogarithms
© Kluwer Academic Publishers 1998
64 Chapter 5
DL(uv)
= (uv)-l D(uv)
= u-1v-1[C(uDv + vDu) + d(Du)Dv + auvJ
=c(v- 1Dv + u- 1 Du) + d(u- 1Du)(v-1Dv) + au-1uv-1v
=c(DLu + DLv) + d(DLu)DLv + a
and
H(uv) = dDL(uv) +c
= d[c(DLu + DLv) + d(DLu)(DLv) + a] + c
= d 2 (DLu)DLv + cd(DLu + DLv) + ade + c
= (dDLu + c) (dDLu + c) + (ad - c2 + c)
= (Hu)(Hv) + h.
1
D(xy) = 2(xDy + yDx) for x,y E dom D and H = I.
Reduction theorems 65
1
D(xy) = {--I}(Dx)Dy, where x = {x n }, y = {Yn} E X.
n+
for n = 1;
where Xn = { 0 Un for n 2: 2;
Un-l
1II
1 for n = 1;
X2 for n = 2;
where u~ = n
Xj for n 2: 3;
j=2
Rp E RD, where
if n = 1;
if n = 2;
otherwise.
Moreover,
ker Dp = {z = r{po ...Pn-d : r E IF},
e = {en}, where en = 1 for n E N, and
c
°
Proof. If Le = then He = dDLe + c = c. This, and (5.1) together imply that
= He = He 2 = (He)(He) + h = c2 + (ad - c2 + c) = ad + c. Hence ad = 0. •
Proposition 5.2. Suppose that X E AD(a,c,d). If DO = dD + cI then X E
ADo(h,O,e), where h = ad - c2 + c.
Proof. By definition, dD = DO - cI and for x, y E domD. Hence
= dD(xy) + cxy
= d[c(xDy + yDx) + d(Dx)Dy + axy) + cxy
= c(xdDy + ydDx) + (dDx)dDy + (ad + c)xy
= c[x(DOy - cy) + y(DOx - ex)) + (DOx - cx)(DOy - cy) + (ad + c)xy
= c(xDOy + yDOx - xDOy - yDOx) + (DOx)DOy
+ (ad + c - 2c2 + e2 )xy
= (DOx)(DOy) + hxy.
If (LO, EO) E G[OO] then X E Lg(D) and (L, E) E G R,l [0], where
(e - 2c)De = O.
Suppose that e E ker D and (L, E) E G[n]. Then for every x E [dom n \ ker D] n
J(X) we have Dx = D(xe) = c(xDe + eDx) = cxDe + cDx, i.e. (e - c)Dx = 0
and DLx = (e - c)x-1xDx = 0, which implies DLx = O. Hence Lx = z E ker D
and x = Ez E ker D, which contradicts our assumption that x (j. ker D.
If e - 2c is not a zero divisor then, by (5.2), we get De = 0, i.e. e E ker D. Again
we conclude that invertible selectors of n do not exist.
Suppose then that c = ~. Let (L, E) E G[n]. Let u E dom n and let x = Lu.
Then u = Ex. Since here CD = ~ and fD = 0, we get
1 1 1 1
Lu = L(ue) = L[(Ex)(Ee)] = LE['2(x + Le)] = '2 (x + Le) = '2 Lu + '2Le.
Hence Lu = Le for all u, i.e. L is a constant mapping. This contradicts our
assumption that L is invertible. •
Example 5.4. Suppose that X = C([O, T]; q. Consider the so-called Pommiez
operator:
x(t) - x(O)
if t ~ 0;
(Dx)(t) = { t
x'(O) if t = O.
(cf. LINCHUK Li[l]).
Note that dom D = {x E C[O, T] : 3x'(0)}. The operator D is right invertible and
has a right inverse defined by (Rx)(t) = tx(t) for x E X. An initial operator F
corresponding to R is (Fx)(t) = x(O). Clearly, X is a D-algebra with respect to the
pointwise multiplication. This algebra has a unit e E ker D, namely the function
e(t) == 1. We shall show that in the case of the Pommiez operator invertible
selectors of n do not exist.
Indeed, let x, y E dom D. It is easy to verify that
1
D(xy) = '2[(x + Fx)Dy + (y + Fy)Dx] for x, y E dom D.
D'(xy) = gxy - D(xy) = gxy - xDy - yDx for x,y Edam D'.
Thus CD' = -1, fD' = gxy and X E AD,(g,-e,O), Le. here C = -e =fi e and
2c - e = -3e E I(X). We therefore conclude that X ft Lg(D'). In the classical
case, when X = C[O, T], D = ft,
R = J~, g(t) = (Re)(t) == t, we find D'x = tx-x'
for x E C I [0, T]. Clearly, the operator R' = I - ~t J~ is invertible. The operator
D' is the so-called the Hermite derivative (cf. ANTOSIK and MIKUSINSKI AM[I]).
o
Proposition 5.6. Suppose that X E AD(a, c, 0) and 2c - e E I(X). Let b =
a(2c - e)-I. Suppose that the operator I + Rb is invertible for an R E RD. Then
(i) DO = D + bE R(X), RO = (I + Rb)-I R E RDo and X E AD(O, c, 0);
(ii) X E Lg(D) only if c = e.
Proof. If the operator 1+ Rb is invertible for an R E RD then DO = D + b =
D(I + Rb)-I E R(X) and RO = (I + Rb)-I R E RDo. Moreover, for x, y E
dom DO = dom D we have
N
P(D) =L Pk Dk , Po,···,PN EX, PN = e.
k=O
Set
70 Chapter 5
N
= u- 1P(D)u = u- 1 L Pk Dku
k=O
N N
= L Pk U - 1 Dku = Pou- 1 + L Pk Dk Lku,
k=O k=1
which implies that x is of the form (5.3), i.e. x = £Ou. If the mapping LO is
invertible and EO = (LO)-1 then (£0, EO) E 0[0°]. Let F be an initial operator
for D corresponding to R. Clearly, since F R = 0, we have F Di LOu = 0 (j =
0,1, ... ,N -1). •
Note 5.2. If PN =j:. e, however PN E leX), then by substitution: p~ = pi/Pk we
obtain a polynomial
N
PI(D) = L p~Dk
k=O
satisfying the conditions of Theorem 5.2. o
Theorem 5.3. Suppose that X E Lg(D), (Ln, En) E O[nn] and
N
(5.4.) Q(D)=L Qk Dk , QkEL(X) (k=O,l, ... ,N-l), QN=I.
k=O
Set
DO = Q(D), no: dom DO = dom DN ---t 2dom DO,
Let Ok be defined as in Theorem 3.1 and let (Lk, Ek) E G[OkJ [or k = 1,2, ... , N.
Suppose that the operator
=L
N
(5.5) Q(I, R) QkRN-k
k=O
lEthe mapping LO is invertible then X E Lg(DO) and (LO, EO) E GR,N[ooJ, where
EO = (Lo)-l.
Proof By our assumptions, we conclude that DO = Q(D) E R(X) and RO =
RN[Q(I, R)J-1. Let LO be defined by Formula (5.6). Since, by our definitions,
Q(D)R N = Q(I, R), we find, for all u E I(X) n dom 0° and for x = Uu that
N
(5.7) Q(D) = L DkQk'
k=O
N
(5.8) Q(I, R) =L RN-kQk
k=O
is invertible for an R E RD, (LN' EN) E G R,N[ON]. Write for u E dom 0° n I(X)
72 Chapter 5
If the mapping £0 is invertible then X E Lg(DO) and (LO, EO) E C R,N[nO], where
EO = (£0)-1.
Proof. By our assumptions, we have DO = Q(D) E R(X) and W = [Q(I, R)]-l RN
(cf. PR[8], Section 3). The similarity property:
(cf. PR[8], Corollary 3.2.1) implies that for x = LOu, where £0 is defined by
Formula (5.9), we get
u- 1 Q(D)u
= u- 1 DN Q(I, R)u
= u- 1 (Q(I, R)u)DN LN(Q(I, R)u).
Further arguments are the same as in the proofs of Theorem 5.2 and 5.3. •
Note 5.3. We can obtain similar results (under appropriate assumptions) for
operators of the form
All these operators are right invertible (cf. PR[8], Section 3). NGUYEN VAN MAU
(cf. N[2)) proved that the resolving operators Q(I, R) and Q(I, R) are simultane-
ously invertible (left invertible, right invertible) or not. In the same paper he has
considered operators of a more general form
M N
Q[D] = L L D m AmnD n , with Amn E L(X)
m=O n=O
which are again right invertible (under appropriate assumptions). Thus, apply-
ing his results, one can construct logarithmic mappings for the operator Q[D]
(provided that they exist) in a similar manner as in Theorems 5.3 and 5.4. 0
Proposition 5.7. Suppose that X E Lg(D') n Lg(D"), both superpositions
D' D", D" D' exist and D'D" = D" D' = D on dom D = dom D' n dom D". Let
(L', E') E C[n'], (L", E") E C[n"], where
n': dom D' ~ 2dom D', n'u = {x E dom D': D'u = uD'x}, u E dom D',
Lu = CD' L"u + R[cD' (D' L'u)D" L"u + u- 1 fD' (u, D" L"u)],
DLu= Dx
=u- 1 Du = u- 1 Du = u- 1 D' D"u
= u- 1 D'(uu- 1 D"u) = u- 1 D'(uD" L"u)
= u- 1 {CD' [(D'u)D" L"U + uD' D" L"u] + fD' (u, D" L"u)}
= CD,[(u- 1 D'u)D"L"u+ DL"u] + u- 1 fD'(U,
D"L"u)
= CD' [(D' L'u)D" L"u + DL"u] + u- fD' (u, D" L"u),
1
Proof. By our assumption, CD' = 1. This, and Proposition 5.7(ii), (iv) now implies
that X E Lg(D) and (L, E) E C[O]. For all x, y E dom D we have
D(xy) = D'D"(xy)
= D'(xD"y + yD"X)
= (D'x)D"y + xD'D"y + (D'y)D"x + yD'D"x
= xDy + yDx + (D'x)D"y + (D"x)D'y.
Hence for u, v E J(X) n dom 0 we get
DL(uv)
= (uv)-1 D(uv)
= (uv)-I[uDv + vDu + (D'u)D"v + (D"u)D'v]
= V-I Dv + u- 1Du + (u- 1D'u)(v- 1D"V) + (u- 1D"u)(V- 1D'v))
= DLu + DLv + (D' L'u)D" L"V + (D" L"u)D' L'v
= DLu + DLv + d(u, v).
Further arguments are the same as in the proof of Theorem 2.2 (if we apply the
commutativity of the algebra X). •
Example 5.6. Suppose that E = [0, a] x [O,b] E IR!, X = C(E;IR), D' = and
D" = %s' Then X is a Leibniz D'-algebra and D"-algebra with respect to the
:t
pointwise multiplication, i.e. CD' = CD" = 1 and fD' = fD" = 0. This algebra
has a unit e. Namely, e(t, s) == 1. Moreover, e E ker D' n ker D". Note that both,
ker D' and ker D" are infinite dimensional. The operator
EP
D = atas'
as a superposition of right invertible operators, is again right invertible. It is
well-known that
a2 x a2 x
atas = asat'
provided that both these derivatives are continuous. Several examples of right
inverses to the operator D can be found in the book PR[8]. The operator D
satisfies all conditions of Corollary 5.3, 0
Proposition 5.8. Suppose that X E Lg(D(j)) (j = 1, .. " n) and the operator
n
n n
D =L D(j) with dom D = dom D(j)
j=1 j=l
Reduction theorems 75
Then
(i) there is a selector L of 0 of the form
n
Lu = L L(j)u far u E J(X) n dam 0;
j=1
DLu = u- 1 Du
n n
= u- 1 L D(j)u = L u- 1 D(j)u
j=1 j=1
n n
=L D(j)L(j)u = D L L(j)u,
j=1 j=1
To do this, one can apply the construction of right inverses for D given by VIR-
SIK V[I] (in a more general case, over JR). This construction, however, is too
complicated to be given here as an example. D
Proposition 5.9. Suppose that X E Lg(D(j)) (j = 1, ... , n) and the operator
n
n n
D= L Drj) with dom D = dom D(j)
j=1 j=l
76 Chapter 5
Then
(i) there is a selector L of 0 of the form
n
Lu = L {CD(j) [L(j)u + Rfj) (D(j)L(j)u)2] + u- 1 fD(j) (u, D(j)L(j)u)}
j=l
DLu = u- 1 Du
n n
-_ u- 1 'L..- 2 U
(j) = 'L..-
" D2 U " u- 1 D (j)
j=l j=l
n
= D L {cD(j)[L(j)u
j=l
n
D(xy) = xDy + yDx + L (D(j)x)D(j)Y for x, y E dom D.
j=l
Clearly, X is a Leibniz D(j)-algebra with unit e(t) == 1 with respect to the pointwise
multiplication. The Laplace operator
is right invertible in X (cf. PR[8]) and satisfies all condition of Corollary 5.4. 0
Example 5.8. Let D E A(X). Denote by ND(a,b,c,d) the class of algebras X
with the following product rule for D:
y c = {y E Y: V xy = yx}.
xEY
If h = ad+ e - e2 = °
then DO = dD + eI E ML(X), i.e. X E NDo(O,O,e). 0
Consider now left invertible operators.
Definition 5.2. Denote by BD(a,e,d) the class of algebras X with D E A(X)
such that the product rule is given by the same formula as in Definition 5.1:
° °
Proposition 5.11. Let X E B'b(a, e, d). Suppose that X E Lg#(D) and (L, E) E
G[L]. IE Le = then ad = (ef. Corollary 5.1).
Proposition 5.12. Suppose that X E B'b(a, e, d), dE I(X), DO = dD + eI and
a = ed-1(c - e). Then
(i) D E ML(X), i.e. X E B~o(O,O,e);
(ii) if the operator dI + Se is invertible for an S E CD then DO E A(X) and
SO = d(dI + Sc)-lSd- 1 E CDO;
Reduction theorems 79
(iii) if e = o:e, d = (3e, where 0:,(3 E IF \ {O} and -fjo E vIFS, then DO E A(X) n
ML(X) and SO = (0:1 + (3S)-lS E CDo.
Proof (cf. the proof of Proposition 5.2, Corollary 5.2 and Proposition 5.3). We
only have to check that SO E £ DO. Indeed,
Proposition 5.13. Suppose that all assumptions of Proposition 5.12 are satisfied.
•
Let DO = dD + eI,
N
P(D) =L Pk Dk , Po, ···,PN E X, PN = e.
k=O
Let
DO = P(D), nO: dom DO = dom DN --t 2dom DO,
nOu = {y E dom DO: DOu = uDOy} for u E dom DO.
Let nk be defined as in Definition 3.1 and let (Lk, Ek) E G[Lk] for k = 1,2, ... , N.
Suppose that the operator
N
P(I, S) = L Pk RN - k
k=O
is invertible for an S E CD and
N
LOu = SN[P(I, S)]-l(pou- l + L PkDLku) for u E dom LN n I(X).
k=l
If the mapping L O is invertible then X E Lg# (DO) and (LO, EO) E G[LO], where
EO = (LO)-l.
80 Chapter 5
N N
P(D) =L PkSN-kDN-kDk =L PkSN-kDN = P(I,S)D N .
k=O k=O
Further considerations are similar to those in the proof of Theorem 5.2, if we take
into account the fact that ker D = {o}. •
Theorem 5.7. Suppose that X E Lg#(D), (Ln, En) E G[Lnl and
N
Q(D) =L QkD\Qk E L(X) (k = 0,1, ... , N - 1), QN = I.
k=O
Set
DO = Q(D), nO: dom DO = dom DN ~ 2dom DO,
Let nk be defined as in Definition 3.1 and let (Lk' E k ) E G[Lkl for k = 1,2, ... , N.
Suppose that the operator
N
Q(I, S) = L QkSN-k
k=O
N
(5.13) Q(D) =L DkQk,
k=O
Let Dk be defined as in Definition 3.1 and let (Lk, E k ) E G[Lk) for k = 1,2, ... , N.
Suppose that the operator
N
Q(I,S) = L SN-kQk
k=O
since all these operators are left invertible under assumptions of Theorems 5.7 and
5.8, respectively (cf. Note 5.3). 0
To conclude Sections 2-5, note that the existence (or non-existence) of right and
left logarithmic and antilogarithmic mappings and logarithmic and antilogarithmic
mappings is a property of an algebraic structure together with an operator and
not a property of the operator only.
In the next sections we shall consider particular properties in algebras with mul-
tiplicative operator D and in Leibniz algebras and formulae of solutions to linear
equations.
CHAPTER 6
MULTIPLICATIVE CASE
We shall see that in the case of algebras with multiplicative operators they may
appear some pathologies. One of them has been shown by Theorem 2.4. Now we
shall study these operators in commutative algebras.
To begin with, we shall prove the following
Lemma 6.1. Suppose that X is a commutative algebra over IF with unit e,
A E M(X), A"f 0, dom A has no zero divisors and e E dom A. Then
(i) Ae = e;
(ii) Ax E leX) and Ax- 1 = (AX)-l for x E leX) n dom A;
(iii) Axn = (Ax)n for all x E dom A, n E N;
(iv) Ax E lz(dom A) and Ax 1 / Z = (Ax?/z for x E 12(dom A).
Proof. (i) Since Ae = Ae z = (Ae)(Ae), we find (Ae)(Ae - e) = O. Since dom A
has no zero divisors, we have either Ae = 0 or Ae = e. Suppose then that Ae = o.
Then for all x E dom A we have Ax = A(xe) = (Ax)(Ae) = 0, which contradicts
our assumption that A "f o. Thus Ae = e.
(ii) Let x E I(X)ndom A. Then e = Ae = A(xx- I ), which implies that Ax E leX)
and (AX)-I = AX-I.
(iii) is proved by an easy induction.
(iv) Let x E 12(dom A). Then there is a y such that y2 = x, i.e. y = XI/2. Hence
(Ay)2 = Ay2 = Ax, which implies that Ax E 12(dom A) and AX I / 2 = (AX)I/Z . •
Proposition 6.1. Suppose that X is a D-algebra with unit e E dom D, D E
M L(X) and dom D has no zero divisors. Then
(i) De = e;
(ii) Dx E leX) and Dx- I = (DX)-I for x E leX) n dom D;
(iii) Dxn = (Dx)-n for xE dom D, n E N;
(iv) Dx E 12(dom D) and DX I / 2 = (DX)I/2 for x E 12(dom D);
(v) ker D n leX) = 0 and (dom D)(ker D) C ker D.
Proof. Assertions (i)-(iv) are immediate consequences of Lemma 6.!.
(v) Suppose that z E ker D n leX). By (i), De = e. Since Dz = 0, we get
e = De = D(zz-l) = (Dz)Dz- 1 = 0, a contradiction. Let x E dom D, z E ker D
82
D. Przeworska-Rolewicz, Logarithms and Antilogarithms
© Kluwer Academic Publishers 1998
Multiplicative case 83
o
Proposition 6.3. Suppose that all assumptions of Definition 6.1 are satisfied,
E is multiplicative with exponent). > 0, x E dom 0- 1 and a E R+ \ {O}. If
ax E dom 0- 1 then a = 1.
Proof. Let a E R+ \ {O}. Let x, ax E dom 0- 1 . By Proposition 6.2(i),
e E dom 0- 1 . Suppose that ae E dom 0- 1 . By Proposition 6.2(v), E(ax) =
E(ae)Ex. By Definition 6.1, ae = LE(ae) = L(a"e). By Theorem 2.5, a A = 1,
i.e. ). = O.This contradicts our assumptions. Hence a = 1. •
84 Chapter 6
Note 6.1. Theorem 2.4 and Proposition 6.3 together show that few results of the
author's papers PR[16] and PR[18] are trivial. 0
Theorem 6.1. Suppose that Condition [M) holds and (L, E) E G[O). Let
y = {x E dom 0- 1 : u = Ex E J(X)}.
Then
Ex = E(ex)
= (Ee)(Ex) = E{R[e- 1(Ex)-1(De)(DEx)]}
= ER[e(Ex)-1e(Ex)Dx) = E(RDx) = E(x - Fx),
F[(Rx)(Ry)) = (FRx)(FRy) = O.
(iv) Let X,y Edam D. Then x = Ru + Z1, y = Rv + Z2, where u, v E X, Z1 = Fx,
Z2 = Fy E ker D and
we get
Dx = D[E(Ra)E(g + z)]
= [DE(Ra)][DE(g + z)] = E(Ra)aE(g + z)e
= aE(Ra)E(g + z) + ax.
Proposition 6.7. Suppose that all assumptions of Proposition 6.6 are satisfied
•
and a, Ra E I(X). If x E ker(D - a) then x is of the form (6.2).
Proof. Proposition 6.2 and our assumptions together imply that E(Ra) E I(X)
and [E(Ra)]-1 = E[(Ra)-1]. Write u = x[E(Ra)]-1. Then x = E(Ra)u and
aE(Ra)(Du - u)
=aE(Ra)Du - aE(Ra)u = [DE(Ra)]Du - aE(Ra)u
= D[E(Ra)u] - aE(Ra)u = Dx - ax = O.
Theorem 6.2. Suppose that all assumptions of Corollary 6.2 are satisfied and
•
1 E vFR, i.e. the operator I - R is invertible (ef. (2.28)). Then all solutions of the
equation
(6.4) Dx = ax + y, yEX
Multiplicative case 87
(6.6)
Then
Hence Xl is a particular solution of Equation (6.4). This, and Corollary 6.2 to-
gether imply that all solutions of that equation are of the form (6.5)-(6.6). •
An immediate consequence of this theorem and Proposition 6.8 is
Corollary 6.3. Let either IF = C or IF = JR. Suppose that Condition [M] holds,
(L, E) E G[n]; E is multiplicative with an exponent A > 0, R E RD and 9 = Re E
leX) n dom n- l . If 1 E vlF'R (i.e. if 1 is a regular value of R; cf. (2.28)), then all
solutions of the equation
(6.10) Dnx = ax + y, y E X, (n E N)
88 Chapter 6
Le.
n-l
Z= L Rk Zk, ZO, ... , Zn-l are arbitrary,
k=O
Proof. Let z E ker D. Let z' = Lz, i.e. z = Ez'. By Proposition 2.5, z' E ker D.
Since D E M L(X), by Proposition 2.6, L, E E M(X) and
(6.12) Dx = ax + y, y E dom L
independently of the choice of S.
Proof. Let x be a solution of Equation (6.12) and let v = xi)lx. Recall that
Xo = E(Sa) E ker(D - a). Since D E ML(X), by our assumptions, we find
Proposition 6.12. Suppose that Condition [MJ# holds, (L, E) E G[LJ and
e E dom L-1. Then
Proposition 6.14. Suppose that Condition [Ml# holds and (L, E) E G[L). Then
elements E( -e), L( -e) do not exist, in particular, -e ¢ dom L n dam L -1 (d.
Proposition 6.4).
Proposition 6.15. Suppose that IF = C, Condition [Ml# holds and (L, E) E
G[L). Then elements E(ie), L(ie) do not exist, in particular, ie ¢ dam Lndom L- 1
(d. Corollary 6.1).
90 Chapter 6
Theorem 6.3. Suppose that Condition [Ml# holds, (Ln, En) E G[Lnl for n E N,
a E J(X) and there is an S E CD such that sne E dom L;;l, sna E J(X)ndom L;;l
and 1 E V]Fsn. Then the equation
Dnx = ax + y, y E X, (n E N)
i.e. the first of Formulae (6.15). A similar proof for left logarithms.
Theorem 6.4. Suppose that D E M(X)n A(X), X E Lgr(D) n Lgl(D) has the
•
unit e, (LnEr) E G[Or], (Ll,El ) E G[Od and U,V E leX) n dom Dr n dom Ol.
Then
(6.16)
In particular,
(6.17)
= D[Ll(V-1u)]D[Lr(uv)]
- D[v-1(LlU)]D[(Lru)v]
= D[v-1(LIU)]VV- 1D[(Lru)v]
- D[v-1(Llu)(Lru)v]
= D[v-1(L1U)]D[(Lru)v]
- D[V-l(LIU)(Lru)v] =
= D[v-1(L1u)(Lru)v]
- V-I (L1U)(Lru)v] = 0,
LEIBNIZ ALGEBRAS
We begin with
Definition 7.1. Suppose that DE A(X). Then X is said to be a Leibniz algebra
if the Leibniz condition
92
D. Przeworska-Rolewicz, Logarithms and Antilogarithms
© Kluwer Academic Publishers 1998
Leibniz algebras 93
L(iu) = ~Lu
2
= ~L(-u)
2
for all u E leX) ndom O.
94 Chapter 7
In particular, L(ie) = o.
Proof. By our assumptions, we have CD = 1, fD = 0 and Lre = 0, L/e = 0,
Le = O. This, and Corollaries 2.10,2.11, 7.5 together imply our statement (cf. the
proof of Corollary 2.11). •
Proposition 7.1. Let X be an algebra with unit e. If X E Lr(D) then there are
(Lr,E r ) E G[Or) such that
Proof. (cf. the proof of Theorem 2.2). Let u, v E I(X) n dom Or. Then Du =
uDLru, Dv = vDLrv and
(7.6)
This immediately implies Formulae (7.5). If we add both sides of these formulae
and divide the sum by 2, then we get Formula (7.6). •
Proposition 7.2. Suppose that X is an algebra with unit e and D E R(X). If
X E Lr(D) then there are (Ln Er) E G[Or], Z E ker D, REnD such that
(7.8)
Proof. Suppose that D E R(X) and (Ln Er) E G[Or]. Let u, v E leX) n dom Or.
Then x = Lru, y = Lrv E dom 0;:-1. Formula (7.5) implies that
= ErLr(uv)
= Er{Lrv + R[v- 1 (DL ru)v + z]}
= Er{y + R[(Ery)-l (Dx)(Ery)] + Z},
for some Z E ker D and REnD.
Proposition 7.3. Suppose that D E R(X). If X E Lr(D)n Ll(D) has unit e,
•
then there are (Lr, Er) E G[Or], (LI' Ed E G[Ot) such that for an REnD and for
all u, v E leX) n dom Or n dom 0 1
(7.10)
Proof. Let D E R(X). Our conclusion follows immediately from Proposition 7.1',
if we take into account that (Lr, E r ), (LI' EI) are I-normalized by R. •
Definition 7.3. Suppose that X E Lr(D) has unit e and there are (Lr,Er ) E
G[Or] such that Equation (7.3) is satisfied. Then Lr is said to be of the right
almost exponential type (shortly: RAE-type). Suppose that X E LI(D) has unit
96 Chapter 7
e and there is (LI, El) E G[Ot] such that Equation (7.4) is satisfied. Then LI is
said to be of the left almost exponential type (shortly: LAE-type) (cf. Definition
2.3). 0
Theorem 7.1. Suppose that D E A(X), X has unit e and (Lr,Er ) E G[Or]
((L1,E1) E G[Od, respectively). Then Lr is of the RAE-type (LAE-type, respec-
tively) if and only if X is a Leibniz algebra.
Proof. Sufficiency has been proved by Proposition 7.1. Suppose now that Lr is of
the RAE-type, i.e. Equation (7.3) holds. Then for all u, v E I(X) n dom Or we
have
(7.11)
Leibniz algebras 97
Proof. By Corollary 2.1, dom Or n dom 0 1 =1= 0. Let U E leX) n dom D. Let
Aru = R[u- 1Du], A1u = R[(Du)u- 1], where REnD is arbitrarily fixed. Clearly,
Ar and Al are selectors of Or, 0 1, respectively. Indeed, DAru = u- 1 Du, DAlu =
(Du)u- l . We therefore conclude that u E dom Or n dom 01.
Suppose then that u E I (X) ndom Or ndom 01. Let v = u -1. Since X is a Leibniz
algebra, we have D(uv) = =
De = O. By Corollary 7.4, Lre O. By Theorem 7.1,
Lr is of the RAE-type, Ll is of the LAE-type. By Equation (7.3),
0= DLre
= DLr(uv) = DLrv + v-1(DLru)v = DLru- 1 + u(DLru)u- 1
= DLru- 1 + (Du)u- 1 = DLru- 1 + DLlu = D(Lru- l + L1u),
(7.11')
Proof. Let x E dom 0;:-1, -x E dom 0,1 and let u = Erx, v = El( -x). Hence
Lru = x, LIV = -x. This implies Lru + LIV = x - x = 0, i.e. LIV = -Lru. By
Theorem 7.2, we find L/v = -Lru = LIU- 1. Hence u E leX) n dom Or n dom 01
and
Proof. Let u, v E dom Or n dom 0/. Let x = L/u, Y = Lrv. Then, by definitions,
(7.13)
(7.14)
Proof. Suppose that D E R(X) and (Lr,E r ) E G[Or]. Let U E J(X) n dom Or
and let z E ker D. Then x = Lru, z' = Lrz E dom 0;1 and z' E ker D. Formula
(7.7) implies that
where XO = e. Then
= (Dnx)x + xn Dx = In_l(X)X n- l x + xn Dx
=In-l (x)x n + xn Dx
n-l
=L xk(Dx)x-kxn + xn(Dx)x- n+n
k=O
n
= L xk(Dx)x-kx n = In(x)x n .
k=O
Corollary 7.12. Suppose that D E A(X) and X is a Leibniz algebra with unit
e. Then for all n E N and x E J(X) ndom D such that xDx - (Dx)x = 0 we have
DLru n = u- n Dun
=u- nu n- 1r n_1(u) = u- 1r n_1(u)
n-1
= u- 1 L u-k(Du)u k
k=O
n-1
= L U- kU- 1(Du)u k
k=O
n
=L u-k(DLru)u k = r n-1(u,DL ru).
k=O
Suppose that D is closed. Proposition 7.6 implies a necessary condition for the
series
=L
00
(Ds)(x) an+lx n
n=O
to be convergent. Namely, if
then the series Ds(x) is not convergent. Clearly, this condition is reduced to the
condition
lim nan::J 0,
n-too
if X is a commutative algebra. o
Definition 7.4. Suppose that either IF = R or IF = <C and D E A(X) with unit e
is a complete linear metric space. Write x O = e and
n
= L;'
00
n
L;'
00
respectively.
Proof. Let x E X D • Our assumptions and Corollary 7.12 together imply that
Dxn = nx n- l Dx. Hence
n n-l n-l xn
L D; =De+ L ~Dx= L L
00 00 00 00
Proof. (i) By our assumptions, Dg = DRe = e. Hence In(g) and rn(g) defined by
Formulae (7.15) have the form
n n
In(g) =L gk(Dg)g-k =L gkg-k = (n + 1)e; rn(g) = (n + 1)e (n EN).
k=O k=O
This, and Formulae (7.16) together imply that Dgn = In_l(g)gn-l = gn-lrn_l(g)
= ngn-l (n EN).
(ii) By (i), we get
n n n
DP(g) =D L Pkl = poDe + L PkDl =L Pk(k + 1)gk-l = P'(g).
k=O k=l k=l
Hence P(g) and DP(g) commute each with another, since both are polynomials
in 9 with scalar coefficients. This implies that P(g) E XD. Now (ii) and (7.23)
together imply (7.25). •
Leibniz algebras 103
Proof. (i) Let x E dom n-l. By Theorem 7.3 and Proposition 2.5, we have
e = E(O) = E(x - x) = (Ex)E( -x). Hence both, Ex and E( -x) belong to leX)
and (EX)-l = E( -x).
(vi) Let u, v E dom 0 and let x = Lu, y = Lv. Then u, v E leX) and u = Ex,
v = Ey. Since L is of the exponential type, by (i),
L(uv- 1 ) = Lu + L(v- 1 )
=Lu + L(Ey)-l = Lu + LE( -y)
= Lu - y = Lu - Lv.
L(!!.e)
q
= L[(pe)(qe)-l] = L(pe) - L(qe), L(pqe) = L(pe)q = qL(pe)
for positive p, q E Q. There are positive integers M and N such that q = ~ and
= II
nN mM
N prj
nj' M = II pSk
mk'
j=l k=l
Leibniz algebras 105
where nj, mk, rj, Sk E N, nj -:j; mk (j = 1, ... , nNj k = 1, ... , mM). This implies
that
nN mM
(7.27)
106 Chapter 7
(d. the author and VON TROTHA PRT[l], also PR[B], Section 6.2). By Corollary
7.B, L is of the exponential type. Since 9 E J(X), we have 9 E dom n. Thus we
get
gn 1 1
L(Rnz) = L(z,) = Lg n + Lz + L(,) = nLg + Lz + L(,e).
n. n. n.
We prove (i) also by induction. For n = 1 (i) is already proved. Suppose now (i)
to be true for an arbitrarily fixed n E N. Then
Now we prove (ii) by induction. Let n = 1. We have shown in Example 2.1 that
DLg- l = g-l. Suppose (ii) to be true for an arbitrarily fixed n E N. Then
Leibniz algebras 107
where eY is defined by the series (7.16). Here this series is convergent for all y E X.
A component of a set n c X is a connected subset of n which is not contained
in a larger connected subset of n. Denote by 91 this component of leX) which
contains the unit e (equipped with the norm topology induced by the topology of
X). It is well-known that in commutative Banach algebras with unit exp(X) = 91.
In particular, exp(X) is closed in leX). A path from a to b in X is a continuous
function I : [0,1] ---+ X such that 1(0) = a and 1(1) = b. Then the following
theorem holds:
Theorem 7.10. (cf. DI BUCCHIANICO DB[I], DB[2]) Let X be a commutative
Banach algebra (over JR. orC) with unit e. Then x E exp(X) ifand only if x E leX)
and there is a path I in leX) from ae to x for an a E C \ {o}.
Unfortunately, this theorem is false when X is not commutative An example is the
Banach algebra X = B(H) of all bounded linear operators over a Hilbert space H.
108 Chapter 7
(cf. DB[l], Remark 1.4, also DB[2], Chapter 3). Then J(X) is pathwise connected
but need not to be equal to exp(X).
Another result concerning a connection between the invertibility of an element
and existence of its logarithmic derivative can be found in in a paper of BART,
ERHARDT and SILBERMANN BES[l].
Observe that these results are, in a sense, related to Theorems 2.1 and 7.4.
We shall see now that some algebras "close" in a sense to Leibniz algebras must
be Leibniz. We shall give here without the proof the following
Theorem 7.11. (cf. SMERL S[l]). Let X be an infinite dimensional Banach
space and let X = B(X) be the algebra of all bounded linear operators defined
on E. Let y be a standard subalgebra of X, i.e. Y contains the ideal of all finite
dimensional operators. Let a function 'P : Il4 -t Il4 be such that
lim 'P(t) = o.
t-++oo t
Let V E L(X) with dom V = Y have the property:
(7.30) IIV(ST) - SVT - (VS)TII < 'P(IISII . IITII) far all S, T E y.
Then there is an A E X such that
Theorem 7.12. (cf. Smerl S[2]). Let X, Y be Banach spaces and let dim X =
00. Let X, Y be standard operator subalgebras of B(X), B(Y), respectively (cf.
Theorem 7.11). Let € > O. Suppose that 'P : X -t Y is a bijective mapping such
that
II'P(ST) - 'P(S)'P(T) II ~ € far all S,T EX.
Then 'P is of the form 'P(T) = ST S-1 for T EX, where S : X -t Y is either a
bounded linear bijective or a conjugate bounded linear bijective operator.
Corollary 7.17. Suppose that all assumptions of Theorem 7.11 are satisfied. Ten
X = B(X) is a Leibniz algebra.
Proof. If 'P = 0 then the Leibniz condition is automatically satisfied. Suppose
then that Condition (7.30) is satisfied with 'P t= O. By (7.31), we have
V(ST)
= ST A - AST = ST A - SAT + SAT - AST
=S(T A - AT) + (SA - AS)T
=S(VT) + (VS)T
for all S, T E Y = dom V. •
Leibniz algebras 109
Example 7.1. Let either IF = IR or IF = C. C(Oj IF) will stand for all functions
defined on a set 0 C IR and with values in IF. Denote by An(Oj IF) the set of
all square matrices x = (aii)i,j=l, ... ,n of dimension n with entries xii E C(OjIF)
(n E N). For the sake of brevity, we shall consider the case when n = 2, 0 = [0, 1j.
Indeed, in this case one can see everything, what we want to show. Let then
X = A 2 ([O, 1jj IF). Write
We shall write in the sequel: X~j = -itXij. Clearly, D E R(X), D E A(X) and
F is an initial operator for D corresponding to REnD, D satisfies the Leibniz
condition and the matrix e = (bij )i,j=1,2 is the unit of X. The kernel of D consists
of all scalar matrices of dimension 2.
Suppose that U = (uijki=1,2 E I(X). Clearly, not all entries vanish simultane-
ously and det U is invertible. Assume (again for the sake of brevity) that det U = e.
Then we have
and
(up to a constant, i.e. scalar matrix). In this case we may apply Theorem 7.6.
(iii) U21 = U12 =f. 0, U22 = O:U11, 0: E IF \ {O, I}, U11 =f. 0.
Matrices of this form do not commute each with another. Nevertheless, with the
denotation used in Point (ii), writing J~ a(s)b(s) ds = (ab)(t), we get
L rU -_ ( In ab
" o:(ab) + In aba) , L lTU -_ ( ~In ab" o:(ab) + In ~)
.
In a; aba In a;
~
(ab) + In % o:(ab) + In
X11 = a, X12 = b E I(X), X21 = zlb, X22 = a+ Z2b, Zl, Z2 E I(X) n ker D.
°
Sufficiency is proved by checking. In order to prove the necessity, observe that the
condition xDx - (Dx)x = for x E X implies the following equations for entries:
(7.32)
(7.33)
(7.34)
Leibniz algebras 111
Similarly, the second of Formulae (7.16) applied to D(xnx- n ) implies the second
of Formulae (7.35).
Let n E N, X E J(X) n dom D and let u = x-no Then u E J(X). Even more,
u E In(X). Formulae (7.35) can be then rewritten in the following manner:
Example 1.4. The space X = C[O, T] with the usual pointwise multiplication
and with the operator D = -it
is a commutative Leibniz D-algebra with unit and
with the logarithmic and antilogarithmic mappings defined in the standard way:
Lu = In u and Ex = eX, provided (u, x) E C[n]. The same space X with the
multiplication defined by the convolution:
and with the operator D is a commutative algebra without the unit. However, in
this case instead of the Leibniz condition for the product rule we have the Duhamel
condition D(x * y)(t) = (x * Dy)(t) + x(O)y(t) - y(O)x(t) for x, y E X which, by
the commutativity yields the product rule: D(x * y) = ~(x * Dy + Y * Dx) (cf. for
instance PR[8J, also Examples 2.3(v), 5.1(iii». Since CD = ~, we conclude that
logarithmic and antilogarithmic mappings do not exist.
Consider now the space X with the convolution as the multiplication and with the
operator M of the multiplication by argument: (Mx)(t) = tx(t) for x E X. It is
easy to verify that M satisfies the Leibniz condition, i.e. M(x * y) = x * My +
y * M x for x, y EX. The basic equation for logarithmic mappings induced by the
convolution * is M u = u * M x, where x = Lu. This equation can be written as
One can show that under appropriate conditions imposed on u the last equation
has a unique solution. This solution x determines the logarithmic mapping, for
x = Lu by definition. Note that all algebras mentioned in this example have zero
divisors. 0
Example 1.5. Suppose that X E Lg(D) and (L, E) E C[n. Then, in general,
neither L nor E is linear. Indeed, E = L -1 is linear if and only if L is linear. On
the other hand, if CD = 0 then, by Theorem 2.4, L is not homogeneous, hence not
linear. Suppose now that X is a Leibniz algebra and L is linear. Then, by Corollary
7.8, L is of the exponential type and we have Ex+Ey = E(x+y) = (Ex)(Ey) for
x, y E dom n. Let y = e. Then Ey = E(O) = e and Ex + e = Ex + E(O) = E(x +
0) = (Ex)e = Ex. This implies e = 0, a contradiction. Since E is not additive, E
is not linear. Thus L is also not linear. In several other cases X ¢ Lg(D) (cf. for
instance, Theorem 5.1, Proposition 5.6). The following question arise: Do exist
X E Lg(D) with linear logaritmic and antilogarithmic mappings? Certainly, such
algebras X (if they exist) cannot belong to AD(a, c, d). 0
Example 7.6. Suppose that D E R(X) and X E Lg(D) has unit e. By Corollary
7.8 and Theorem 7.4, the following conditions are equivalent: (i) X is a Leibniz
algebra; (ii) logarithms are of the exponential type; (iii) a logarithm of u E X
exists if and only if u is invertible. These 3 equivalent conditions are fundamental
for the classical calculus of logarithms (and antilogarithms). 0
CHAPTERS
The main purpose of the present chapter is to find solutions of linear equations with
an operator D by means of already known properties of right and left logarithms
and antilogarithms.
Theorem 8.1. Suppose that X E LI(D) has unit e, D E R(X), (L/, Ez) E G[Oz),
a E X and Ra E dom nIl for an R E RD. Let x E leX). Then x E ker(D - a) if
and only if
Proof By our assumptions and Theorem 7.3, [Er (-Ra)J-1 = E/(Ra). This, and
Theorem 8.1 together imply (8.2). •
Theorem 8.2. Suppose that X E Lr(D) has unit e, D E R(X), (Lr,Er ) E G[Or],
a E X and Ra E dom 0;1 for an R E RD. Let x E leX). Then x E ker(D - afo-)
if and only if
113
D. Przeworska-Rolewicz, Logarithms and Antilogarithms
© Kluwer Academic Publishers 1998
114 Chapter 8
Dx = D[zEr(Ra)]
= (Dz)Er(Ra)zDEr(Ra) = zE1(Ra)(DRa)
= zE/(Ra)a = xa.
Proof. Our assumptions and Theorems 7.3,8.2 together imply (8.4) (cf. the proof
of Corollary 8.1). •
Theorem 8.3. Suppose that X E Lr(D)nL1(D) has unit e, D E R(X), (LI,EI) E
G[n!], (Lr, Er) E G[nrJ, a E X and Ra E dom nil n dom n;:-l for an R E 'RD.
Then the following identities hold for all x E dom D:
D[xEI( -Ra)]
= (Dx)EI( -Ra) + xDEI( -Ra)
= (Dx)EI(-Ra) + x(-DRa)EI(-Ra) = (Dx - xa)EI(-Ra).
Equations in Leibniz algebras 115
Again by Theorem 7.3, we get for y E dom n;:-l n dom nil and Rl = R,
(D - a)Ra y
= Er(Ra)D{EI( -Ra)Er (Ra)RdEI (-Ra)y])
= Er(Ra)DRdEI(-Ra)y] = Er(Ra)E/(-Ra)y = y,
i.e.
D[xEI(Ra)] = DRdyE/(-Ra)] = yEI(-Ra).
Hence (D - a+-)x = yE1( -Ra)Er(Ra) = y, which proves that D - a+- E R(X)
and that Ra+- is it right inverse independently of the choice of Rl E 'RD. •
Theorem 8.3 shall be used in order to determine general solutions of non-
homogeneous equations. Namely, we have
Theorem 8.4. Suppose that all assumptions of Theorem 8.3 are satisfied. Then
the equation
Theorem 8.6. Suppose that X E Lr(D)n L,(D) has unit e, DE R(X), (Lr' Er) E
G[Orl, (L" E,) E G[O,l, a, b E X and Ra, Rb E dom 0;:-1 ndom 0,1 for an REnD.
Then
(i) the equation
Proof. (i) Write Y1 = (D - b)x. Then, instead of Equation (8.11), we have the
equation (D - a)Y1 = y. By Theorem 8.4,
Proof. By our assumptions, the element DRa = a commutes with Ra. This, and
Theorems 8.4 and 7.6 immediately imply (8.15). •
Similarly, we have
Corollary 8.4. Suppose that X E Lr(D)n L/(D) with unit e E dom 0;1 n
=
dom 011 is a complete linear metric space, either IF IR or IF C, D E R(X) is =
closed, (LnEr) E G[Or]' (L1,Et) E G[Od, a E X and Ra Edam 0;1 ndom 011 n
XD for an R E RD. Then Equation (8.9)
(8.17)
n-l
Z = L zkRke E ker D n and Zo, ••. , Zn-l E ker D are arbitrary.
k=O
(8.18)
k=O
n-1
= EI(Ag)(L zkRkHe + zn)
k=O
n
= EI(Ag)(L zjRje + zn)
j=1
n
= EI(Ag)(L zjRje).
j=O
if and only if X is the direct sum of the principal spaces of the operator Sh
corresponding to the eigenvalues t1, ... , tN:
Equations in Leibniz algebras 119
and PI, ... , PN are disjoint projectors giving the partition of unit:
N
PjPk = 8jkPk (j, k = 1, ... , N), L Pj = I.
j=l
If rl = ... = r N, i.e. if the characteristic roots are single, then these projectors are
of the form
II
N
t - tk .
-- (J = 1, ... ,N)
tj - tk
k=l,k#j
n
(8.19) P(t) = II (t - tjt j , rj ~ 1, rl + ... + rn = N,
j=l
n rj-l n rj-l
Proof. By the Leibniz formula, we have for u, v Edam Dn, A E IF, n E N':
=L
k=O
(~) (_l)n-k An- k Dk(uv)
=L
k=O
n
(~) (_l)n-k An- k ~
k
G) (Dk-ju)Djv
•
Proposition B.2. Suppose that all assumptions of Proposition 8.1 are satisfied.
Then the equation
(8.21)
(D - u)nXI
= (D - u)n{Et(.~g)Rn[Er(-Ag)y)}
= EI(Ag)Er(-Ag)y
=y.
Theorem 8.8. Suppose that all assumptions of Theorem 8.7 are satisfied. Then
•
the equation
(8.25)
n
= II (D - tjlriXI
j=l
n-l
= [II (D - Uri](D - urn {Et{tng)Rrn[Er(-tng)y)}
j=l
n-l n-2
= [II (D - tj1ri]Yn_1 = [II (D - tjlri](D - tj1rn-1Yn_1
j=l j=l
n-2
= [11 (D - tj 1r i ]Yn-2 = ... = (D - t1lr1Yl = Yo = y.
j=l
•
122 Chapter 8
Recall that X is a complete linear metric locally pseudo convex space over IF
(IF = ~ or IF = C) if is equipped with the topology induced by a sequence of
Pn-homogeneous F -norms:
(cf. ROLEWICZ R[l]). In particular, complete linear metric spaces which are
either locally convex or locally bounded, hence also Banach spaces, are locally
pseudoconvex.
Definition 8.1. X is said to be a complete m-pseudoconvex algebra if it is an
algebra and a complete locally pseudo convex space with the topology induced
by a sequence {II . lin} of submultiplicative Pn-homogeneous F-norms, i.e. such
pseudonorms that
o
Corollary 8.5. Suppose that eitherlF = ~ orlF = C, X E Lr(D)nLI(D) with unit
e E dom 0;:-1 n dom Oil is a complete m-pseudoconvex algebra and (Lr,Er) E
G[Or], (L I, Ed E G[Od. Let D be closed. Let
XD = {x E CD(X) : xDx = (Dx)x}
(cf. Formula (7.22) in Theorem 7.6). Let 9 = Re and let >.g E dom 0;:-1 ndom Oil
for an R E RD and a >. E IF. Let the initial operator F corresponding to R be
multiplicative. Then >.g E XD, >. E 1JFR and
(8.26)
L
n=O
Xo =L (L etjg)zjkRke =L (L zjkRke)e-tjg,
j=O k=O j=O k=O
= etngRrn(e-tngYn_d, Yo = Y,
Xl
Corollary 8.9. Suppose that all assumptions of Proposition B.1 are satisfied.
Then for all n E N Equation (B.21) has the general solution either of the form
(B.22) (as in Proposition B.1) or of the form
In the same manner, as Theorem 8.6 follows from Proposition 8.1, the following
theorem follows from Corollary 8.8:
Theorem 8.9. Suppose that all assumptions of Theorem B.7 are satisfied. Then
Equation (B.23) has the general solution either oEthe form (B.24) x = Xl +xo, where
Xo E ker P(D) and Xl is defined by Formulae (B.25), or of the form X = x~ + Xo,
where
(8.33)
In general, even in Leibniz algebras, we cannot use any analogue of the results
already obtained in order to represent solutions of linear equations of higher order
as a product of a constant and an antilogarithm, as we did it until now. It is so,
because any power Dn, n ~ 2, of an operator D satisfying the Leibniz condition
does not satisfy this condition. Thus we cannot solve non-homogeneous equations
in a similar manner, as it was done for equations of the first order. We will show
the main difficulty for n = 2.
IE X E L/(D} then there are (L,,2, E , ,2) E G[n /,2) such that
(E, ,2 X)(E,,2Y)
(8.35) = E,,2n{X + R2[(E,,2X)(D2y)(E,,2X)-1
+ 2(E,,2y)-I(E,,2X)-I(Dx)(E,,2X)(Dy)E,,2Y + z]},
(8.39) Dnx = xa + y, y E X,
(8.40) L .
n-l ( n 1) (Djv)Dn-I-jxo = y.
j=O J +
Suppose that X E LI(Dn), (Ll,n, El,n) E G[f!l,n], a E X, and Rna E dom f!~~ for
an R E RD. Then x is a solution of the equation
(8.41) Dnx = ax + y, y E X,
(8.42) L ..
n-l ( n ) (Dn-I-jxo)Djv = y.
j=O J +1
126 Chapter 8
y = DnXI - aXI
= Dn(xoRv) - aXoRv
= t
k=O
(~) (Dn-kxo)D kRv - aXoRv
= t
k=l
(~) (Dn-kxo)Dk-lv + aXoRv - aXoRv
=L
n-l
j=O
C: 1) (Dn-I-jxo)Djv,
Note 8.2. Clearly, one can find solutions of Equation (8.40) if the operator
I - Rna is invertible (left invertible, right invertible, respectively, cf. PR[8], also
N[2]) , i.e. if a belongs to the set V.nn (X) defined by Formula (2.25). Similarly,
one can find solutions of Equation (8.39) if a belongs to the set W Rn (X) defined
by Formula (2.27). As it was shown by Propositions 2.14-2.17, these assumptions
assure the existence of the corresponding right and left antilogarithms. We have
only to remember that, by definition, a right (left) antilogarithm of order n for
the operator D is a right (left) antilogarithm of order 1 for the operator Dn. 0
n-l
(8.43) Dn(Lr,nu + Ll,nu-l) =- L (~) (D n- kLr,n_kU)D kLl,kU- I .
k=O
= Dn(uv) = t
k=O
(~) (Dn-ku)Dkv
Du = '\u = u'\e = u'\DRe = uD(,\g), which implies that'\g E dom n- 1 and there
are (L, E) E G[n] such that e A = u = E(,\g). •
Recall that elements of the form eAz are called exponentials for D (cf. PR[8]).
Indeed, these elements belong to ker(D - '\1), hence they are eigenvectors of D.
Corollary 8.11. Suppose that Condition [L] holds, (L,E) E G[n], R E RD,
,\ E VF R and F is an initial operator for D corresponding to R. Then
If F is almost averaging, i.e. F(zx) = zFx for all x E dom D, z E ker D, then
(8.47) Dx - ax = y, y E dom n
is of the form
Proof. Either Theorem 8.4 or Theorem 8.5, the commutativity of the algebra
X and Theorem 8.11 together immediately imply (8.48). Since E(-Ra)y =
E(-Ra)ELy = E(Ly - Ra), we obtain from (8.48) the form (8.49). •
Observe that solutions given by Formula (8.49) in the classical case of the operator
D= ft
are of the same form as those obtained by the constants variation method.
Equations in Leibniz algebras 129
Corollary 8.12. Suppose that all assumptions of Theorem 8.11 are satisfied and
Ra E J(X). If FE ML(X) is an initial operator for D corresponding to R then
the initial value problem for Equation (8.47) with the condition
x
(8.51) = E(Ra)R[E(-Ra)y] +xOE(Ra)FE(-Ra)
= E(Ra)[RE(Ly - Ra) + XO FE( -Ra)].
= F{E(Ra)R[E(-Ra)y] + zE(Ra)}
= [FE(Ra)]FR[E(-Ra)y] + F[zE(Ra)]
= (Fz)FE(Ra) = zFE(Ra).
By our assumption and Proposition 7.7.1(i), E(Ra) E J(X) and [E(Ra)]-l
E(-Ra). Since F E ML(X), we find
WLOKA SW[l] for more general questions concerning the factorization problem;
also Examples 12.3, 12.4). 0
Theorem 8.13. Let Condition [L] hold. Suppose that (L, E) E G[OJ, at, ... , an E
X and there is an R E 'RD such that Raj E dom 0- 1 (j = 1, ... , n). Then the
general solution of the equation
is of the form
(8.53)
where
Yo = y, Yj = E(Raj)[RE(LYj_l - Raj)] + Zj (j = 1, ... , n)
ZI, ... , Zn E ker D are arbitrary.
n+l
(8.54) [II (D - aj)]x = y, where Ran+l E dom 0-1, y Edam D.
j=1
Write
(8.55)
This proves that Equation (8.54) has all solutions of the required form. •
Proposition 8.1 immediately implies the following
Proposition 8.7. Suppose that [L] holds, (L,E) E G[OJ, there are>. E IF and
R E 'RD such that >.g = >.Re E dom 0- 1 and n E N is arbitrarily fixed. Then
Xo E ker(D - >-.J)n if and only if
(8.56)
n-l
Xo = zE(>.g), with Z = L zkRke E ker D n and arbitrary Zo, ... , Zn-1 E ker D.
k=O
Equations in Leibniz algebras 131
n rj-l
Note 8.4. If IF = lR, it may appear pairs of conjugate imaginary roots: i>.., -i>..,
where>.. E IR \ {O}. To every such pair there corresponds the operator D2 + >..2 I.
This operator, particular properties oflogarithms and antilogarithms when IF = C
132 Chapter 8
and their relation to the real case will be considered in next sections. We should
point out that properties of right invertible operators in the complex case have been
recently studied in a series of papers by BINDERMAN (cf. Bin[1]-[19], BPR[1]-[2],
also Appendix). 0
Example 8.2. Suppose that Condition [L] holds and (L,E) E GR,dO] for an
R EnD. We recall that the Wronskian of the system {Xl. ... , XN} C dom D N -1
is the determinant
. 1
WD = WD(XI. ... , XN) = det(DJ- Xk)j,k=l, ... ,N
and that this determinant have the same properties as in the classical case, i.e.
we can prove the Wronski theorems for linear equations in D with coefficients
belonging to X (cf. PR[8], Section 6.3).
A system {X1, ... ,XN} E dom D N - 1 is said to be fundamental if and only if
W D(X1, ... ,XN) E leX) (cf. PR[8], Definition 6.3.3). If {Xl. .... XN} is a funda-
mental system of solutions to linear equation a(D)x = 0, where
N
a(D) = L akDk, and ao, ... ,aN E X, aN = e,
k=O
(8.64)
Formula (8.64) implies that DLWD = WD' l DWD = (-1)N aN_ 1. Hence we get
Dx = D[E(Sa)v]
= D[E(Sa)]v + E(Sa)Dv = [(DE(Sa]v + E(Sa)Dv
= E(Sa)(av + Dv) = E(Sa)(av + Dv) = ax + E(Sa)Dv,
which implies that x is a solution of Equation (8.66) only if E(Sa)Dv = y. If there
is a v E dom D which is a solution of the equation E(Sa)Dv = y, then a solution
of Equation (8.66) is x = E(Sa)v. The uniqueness of that solution follows from
the left invertibility of D. •
Note 8.5. In Proposition 8.8 X is a Leibniz algebra and DE A(X). We therefore
conclude that X has no unit (cf. Theorem 4.1). This is a reason, why we cannot
assume that E(Sa) is invertible. If we assume that -Sa E dom E then we also
conclude that E(Sa) is invertible, which again leads to a contradiction. In the
case of algebras with right invertible operators the invertibility of elements E(Ra),
where R E R D , follows from Theorem 7.3. This shows us that we cannot expect
too much results for left invertible and invertible operators, since they have trivial
kernels. 0
We shall consider now problems of a different kind.
Proposition 8.10. Suppose that X E Lgr(D), A E Lo(X), there is an R E RD
such that the operator I - RA is invertible and F is an initial operator for D
corresponding to R. Write D A = D - A and
Then
(i) DA E R(X), RA = (I - RA)-IR ERDA and FA = (I - RA)-IF is an initial
operator for D A corresponding to RA.
(ii) Selectors ofOA,r are ofthe form
LA,rU
= RA(DLru - u- 1 Au) + (I - RA)-l Z
= (I - RA)-l[R(DLru - u- 1 Au + z]
= (1 - RA)-l[RDLru - R(u- 1 Au) + z]
= (I - RA)-l[Lru - FLru - R(u- 1 Au) + z].
FALA,r
= (I - RA)-l F[Lru - FLru - R(u- 1 Au) + z]
= (I - RA)-l[FLru - F2L ru - FR(u- 1 Au) + Fz]
= (I - RA)-l Z.
In this chapter we shall consider particular properties of left and right logarithmic
and antilogarithmic mappings in algebras over C. In order to do it, we shall make
use of the following condition:
[Cln IF = C, n E N is arbitrarily fixed, X E Lgr(Dn) n Lgl(Dn),
and En = dom 0;,; n dom ot:"~ is symmetric, i.e. -x E En whenever x E En.
Clearly, if X E Lg(Dn) then En = dom 0;;-1.
Observe that, by Theorem 2.4, if either dom Or, 1 or dom 0 1,1 is symmetric, then
CD=j: O. By an easy induction we obtain the same conclusion for all n E N, since
CDn = cD.
Definition 9.1. Suppose that Condition [Clt holds and (Lr,Er ) E G[Or],
(L/, Ed E G[od. For ix E E1 we write
136
D. Przeworska-Rolewicz, Logarithms and Antilogarithms
© Kluwer Academic Publishers 1998
Trigonometric mappings 137
(CX)2 + (SX)2
= ~[(EI(iX) + Er( -ixW + 4~2 [(El(ix) - Er( -ixW
= ~[EI(iX)Er(-iX) + Er(-ix)EI(ix)].
Corollary 9.1. Suppose that all assumptions of Proposition 9.1 are satisfied.
•
Then the mappings C', S' defined as follows: C'x = C(x + z), S'x = S(x + z) for
ix E Eo, z E ker D also satisfy assertions (i)-(iv) of that Proposition.
Proof Let x' = x + z. Then ix' E Eo and Dx' = Dx. This, and Proposition 2.9
together imply the conclusion. •
Proposition 9.2. Suppose that Condition [Ch holds and (L, E) E G[O]. Then
for all ix E dom 0- 1 we have
(9.7)
(9.8)
138 Chapter 9
= :i
:i
DSx [DE(ix) - DE( -ix)]
= i:i[E(iX) + E( -ix)]Dx
= ~[E(iX) + E( -ix)]Dx = (Cx)Dxj
(DCx)2 + (DSx)2 = [(-SX)Dx]2 + [(CX)Dx]2 = [(CX)2 + (Sx)2](Dx)2 j
D 2Cx = D[-(Sx)Dx]
= -cD[(DSx)Dx + (Sx)D 2x] -iD(SX, Dx)
= -cD[(Cx)(Dx)2 + (Sx)D x]- iD(SX, DX}j
2
D 2Sx = D[(Cx}Dx]
= cv[-(SX)(Dx}2 + (Cx}D 2x] + iD(Cx,Dx)
= -cD[(Sx)(Dx}2 - (Cx)D 2x] + iD(Cx,Dx).
Corollary 9.2. Suppose that all assumptions of Proposition 9.1 are satisfied and
•
X is a Leibniz D-aIgebra with unit e. Then the Trigonometric Identity holds, i.e.
Proposition 9.3. Suppose that Condition [elt holds, (L, E) E C[O] and the
trigonometric identity (9.9) holds. Then
(i) E(ix), E( -ix) E J(X) and E( -ix) = [E(ix)]-1 for all ix E dom 0- 1 ;
(ii) E(O) = e, hence Le = 0;
(iii) e E ker D;
(iv) if X is an almost Leibniz D-algebra, i.e. if !D(X,Z) =0 for x E dom D,
z E kerD, then CD = 1;
Le = L[( -e)( -e) = 2CDL( -e) + R[( _e)-2 !D( -e, -e) = R!D(e, e).
Thus
= Le 2 = 2cDLe + R!D(e, e) = 2cDLe + Le
Le
which implies 2cDLe = O. Since, by our assumption, CD =J. 0, we get Le = 0, which
implies e = ELe = E(O) = -e, a contradiction. Thus z = e and Le = LE(O) = O.
This also proves (iii), since e = z E ker D.
(iv) Suppose that X is an almost Leibniz algebra. We already have proved that e E
ker D. This means that for all u E J(X)ndom 0 we have 9D(U) = u- 1!D(u,e) = 0,
i.e. the mapping 9D is constant. Since, by our assumption, X E Lg(D), Theorem
2.3 implies that CD = 1.
(v) Suppose that CD = 1. Since Le = 0, for all u E J(X) n dom n we have
!D(u,e) = uDRu- 1!D(u,e)
= uDRgD(U) = uD(1- cD)Lu
= (1 - cD)uDLu = (1- cD)Du.
This implies that 9D = O. Note that, by Theorem 2.3, the mapping 9D is constant,
but not necessarily zero. •
140 Chapter 9
Proposition 9.3 shows that commutative algebras with the trigonometric identity
are very" similar" to Leibniz algebras.
Open Question 9.1. Do exist non-Leibniz algebras with the trigonometric iden-
tity (9.9) ?
Corollary 9.3. Suppose that Condition [Clt holds, (L, E) E C[O], R E RD,
9 = Re and A E C. If Aig E dom 0- 1 then
Corollary 9.4. Suppose that all assumptions of Corollary 9.3 are satisfied and
•
X is a Leibniz D-algebra. If Aig E dom 0- 1 then
Corollary 9.5. Suppose that all assumptions of Corollary 9.3 are satisfied and
•
X is a Leibniz D-algebra. If Aig E dom 0- 1 then
DC()..g)
Proposition 9.7. Suppose that all assumptions of Proposition 9.6 are satisfied.
Then U E ker(D 2 +)..2 J) if and only if there are (L r ,2, E r,2) E G[Or.2], (Ll. 2, EI. 2) E
G[0/.2] such that
Suppose that v E ker(D2 + ),,2 J). Then zv, vz E ker(D 2 + ),,2 J) for all z E ker D.
Indeed, by (9.12'),
This, Propositions 9.5 and 9.6 and similar arguments as those in the proof of
Proposition 9.7 imply the first 4 equalities (9.12). The 5th and 6th equalities
(9.12) are proved by Proposition 9.7. •
Corollary 9.6. Suppose that all assumptions of Theorem 9.1 are satisfied. Then
all solutions of the equation
(9.13)
So that, if uv =f. vu then (u + iv) (u - iv) =f. u 2 + v 2 • Also we cannot use the
fact that the mapping C is even and the mapping S is odd (cf. also Proposition
9.1(iii)), which is here necessary. 0
Corollary 9.7. Suppose that Condition [Ch holds and X is a commutative
Leibniz D-algebra. Suppose that A E C, R E RD, g = Re and Aig E dom E l . If
(Ll,El ) E G[Oll and (L 2 ,E2 ) E G[02l then
(9.15)
Note 9.2. The assumption that Ai,-Ai E vcR ensures that Aig, -Aig E
dom 0- 1 = E l . In this case, -A 2 E vCR2 (cf. the following Example 9.1.)
o
Example 9.1. We shall show connections of the trigonometric mappings and
elements with the trigonometric operators and elements induced by a right inverse
of aD E R(X) in linear spaces (cf. PR[8], Sections 2.3 and 6.2). Suppose then
that Condition [Ch holds, R E RD and A E vcR. Let g = Re. Then there are
(L, E) E G[Ol such that
Indeed, elements of the form u = e>.z = (I - AR)-l z are well-defined for all z E
ker D and called exponentials for D since they are eigenvectors for D corresponding
to the eigenvalue A: (D - AI)U = D(I - AR)u = Dz = 0 (cf. PR[8]). Moreover,
144 Chapter 9
(9.17)
are said to be cosine and sine operators, respectively. Let z E ker D. Then c,\z
and s,\z are said to be cosine and sine elements, respectively. It is not difficult to
verify that
Thus
c,\z= C(Ag); s,\z = S(Ag).
We therefore conclude that c,\ and s,\ have all properties listed in Proposition
9.1. Clearly, proofs in the book PR[8] are different, since they follows just from
definitions. Also the assumption made there was much stronger. Namely, we have
assumed that R is a Volterra operator, i.e. I - AR is invertible for all scalars A. 0
Example 9.2. As in Definition 9.1 suppose that Condition [Clt holds and
(Lr,E r ) E G[Or]' (Ll,E1) E G[Od. For ix E El write
o
Note 9.3. We could define 0* = 0 and 101 = 0, which would require some little
modification in definitions of C(X) and JR(X). However, since 0 f/. dom n, this
modification is useless. 0
Proposition 9.8 Suppose that all assumptions of Definition 9.2 are satisfied and
(L, E) E GR,dn) for an R E RD. If w E C(X) then there is an x such that
ix E dom n- 1 and
(9.20) Lw = L I w I + ix.
I I I
Proof. Let u ' = u w -1 and v' = v Wi-I. Clearly, {U I )2 + (V ' )2 = e and
w' = u ' + iv' E dom n. Let x = -iL{u' + iV'), i.e. x is defined as in (9.19). By
definition,
Lw = L[ w I I E(ix») = I
cD[L wi + LE(ix») + R[ I w 1-1 E{ix)-I) =
= cD{L w I I + ix) + R[ I w I E( -ix»).
Putting CD = 1, !D = 0, we get (9.20) with x defined by (9.19).
Example 9.3. Suppose that all assumptions of Definition 9.2 are satisfied. Let
•
X be a D-algebra over JR with unit e. Write
y = X + iX = {a + ib: a, bE X}.
Clearly, Y is a D-algebra over C with unit e and with the addition, multiplication
by scalars and multiplication of elements defined as follows: for a, b, c, d EX, A,
fLEJR
(a + ib) + (c + id) = (a + b) + i(c + d); (AifL)(a + ib) = (Aa - fLb) + i{Ab + fLa);
146 Chapter 9
Proposition 9.9. Suppose that all assumptions of Definition 9.2 are satisfied,
X is a complete linear metric space over C, D is closed, (L, E) E G R,l [0) for an
R E RD, W E C(X) and x = argw. Ifx E [D(X) then also
(cf. Example 12.2(xii)). Since X is a commutative algebra, this and Theorem 7.6
together imply that E(.\e) = eAe. Let w = 27ri and let k E Z be arbitrarily fixed.
Suppose that x = argw E [D(X). Again by Theorem 7.6,
L
00
(9.21) cnE(n)..ig)
n=-oo
is convergent, then we say that it is the Fourier series and that the function
L
00
L
00
L
When
(9.23) x()..)E(n)..ig)d)" = Cn for all n E Z ?
Are Conditions (9.23) necessary and sufficient for the series (9.21) to be conver-
gent?
L L
When
(9.25) x()")C()..g)d)" = an (n E No), x()")S()..g)d)" = bn (n E N) ?
Are Conditions (9.25) necessary and sufficient for the series (9.24) to be conver-
gent?
When either {E(nAig)}nEZ = X or {ao,C(nAg),S(nAg): n E N} = X? 0
CHAPTER 10
is a commutative group with unit e with respect to the addition in IF, i.e. for any
x)." xp. E Xo(D) there are x)." Xp. E Xo(D) such that x).,xp. = XMp., (X).,)-l = X_).,
whenever A, J.L E IF. If D E A(X) then the set
is a commutative semigroup with respect to the addition in X, i.e. for any Xa, Xb E
X(D) there is an Xa+b E X(D) such that XaXb = Xa+b whenever a, b EX.
Proof. First we prove that X(D) is a commutative semigroup. Suppose that
a, bE X and Xa, Xb E X(D). Since X is a commutative algebra, by definitions and
the Leibniz condition, we find
which implies that XaXb = Xa+b' The proof for Xo(D) is the same, since elements
Ae, J.Le commute with x)., and xl" Clearly, e = Xo E ker D is the unit of Xo(D).
Indeed, for an arbitrarily fixed x)., we have x).,e = X).,Xo = X)"+o = x)., = Xo+)., =
XOX)., = ex).,. Hence X).,X_)., = X_).,X)., = Xo = e, which implies (X).,)-l = X_).,. We
therefore conclude that Xo(D) is a commutative group with unit. •
In the sequel, whenever we shall consider the set X(D) and some induced sets, we
shall admit that D E A(X).
Proposition 10.2. Suppose that DE A(X) is a Leibniz algebra with unit e and
F is an initial operator for D. Then the set
is a commutative group with unit e with respect to the addition in IF and the set
149
D. Przeworska-Rolewicz, Logarithms and Antilogarithms
© Kluwer Academic Publishers 1998
150 Chapter 10
Proof. For the reasons used in the proof of Proposition 10.1, we prove only that
X(D; F) is a commutative group with unit. Let D E A(X). By the Leibniz
condition, we have De = O. Hence Fe = e. Let a = O. Then Xo E ker D which
implies Xo = Fxo = e, by our assumption that Xo E X(D; F). Let a E X be
arbitrarily fixed. By Proposition 10.1, XaX-a = X-aXa = Xa-a = Xo = e, which
implies X-a = (xa)-l. •
Corollary 10.1. Suppose that D E A(X) is a Leibniz algebra with unit e, D E
R(X) and F, F' are initial operators for D. Let the sets Xo(D; F), X(D; F) be
defined by Formulae (10.2), (10.2'), respectively. If F' is multiplicative then the
sets
F'Xo(D; F) = =
{u,x F'x,x : X,x E Xo(D; F)},
which implies the required formula (cf. Formulae (2.25), (2.26) and (2.28)). A
similar proof for X,x. •
Let F'D be a set of multiplicative initial operators for D, i.e.
(fO C f).
Theorem 10.1. Suppose that D E A(X) is a Leibniz algebra with unit e, D E (X)
and F'D = {F-yhHo C FD n ML(X). Write
(10.3)
(10.3')
Semigroup properties 151
(with the convention admitted in Proposition 10.1). Then z8>.(p) E ker(D - >.I),
z8 a(P) E ker(D - a) for an arbitrary z E ker D and Fp[z8>.(p)] = z. In particular,
Fp8>.(P) = Fp8 a(p) = e.
Proof. Let D E A(X) and let x = z8 a(p), where z E ker D. Our assumptions and
the Leibniz condition together imply that
Dx = D[z8 a(p)]
= zD8 a(p) = zD[(Fpx-a)xa] = z(Fpx_a)Dxa
= z(Fpx_a)axa = aZ8 a(p) = ax.
Hence x E ker(D - a). Since Fp is multiplicative, by Theorem 8.11, we find
In particular, Fp8 a(p) = e. A similar prooffor 8>.(p) follows from Proposition 8.2
for n = 1. •
Theorem 10.2. Suppose that all assumptions of Theorem 10.1 are satisfied.
Write for p, q E ro, x>. E Xo(D; Fp), Xa E X(D; Fp)
(10.4)
(10.4')
=(Fqxa)(Fpx-a) = (Fpxa)-l(FqX_a)-l
= [(Fpxa)(Fpx_a)t1 = [8 a(p, qt 1;
152 Chapter 10
=(Fpxa)(Frx-a)(Frxa)(FqX-a)
=(Fpxa)[Fr(xax-a)(Fqx-a)
=(Fpxa)Fre(Fqx-a) = (Fpxa)e(Fqx_a)
= (Fpxa)(Fqx_a) = 0 a(P, q);
0 a(p, q)0b(P, q)
= (Fpxa)(Fqx-a)(FpXb)(Fqx_b)
= [Fp(XaXb)][Fq(x-aX-b)]
= (Fpxa+b)(Fqx-(a+b)) = 0 a+b(p,q).
A similar prooffor 0;,(P,q). •
Recall that for D E A(X) by definition, Erx = EIX = Ex for x E dom 0- 1 =
dom 0;:-1 = dom Oil.
Theorem 10.3. Suppose that X E Lr(D)n LI(D) has unit e, (Lr,Er) E G[Or]
and (LI,E I) E G[Ozl. Let El = dom 0;:-1 n dom Oil (cf. Section 9, Condition
[C]n). Let R E RD and let 9 = Re. Write
(10,5)
are commutative semigroup with unit e with respect to the addition in IF and X,
respecti vely.
Proposition 8.2 and Theorems 10.1 and 10.3 together imply
Corollary 10.3. Suppose that X E Lr(D)n LI(D) has unit e, (Lr,Er ) E G[Or],
(LI,Et) E G[Ot] and F1 = {F,},Ho c FD n ML(X). Let Fp E F1 correspond
to an Rp E RD (p E rO). Let gp = Rpe. Then
(10.10)
(10.14)
(10.14)
(10.15)
or, respectively,
(10.16)
are particular solutions of Equations (10.11) and (10.12) satisfying the conditions
Fpx' = 0, Fpx" = 0, respectively. By our assumptions,
The following example show that classical fundamental solutions for an ordinary
differential equation of order one do coincide with fundamental solutions in the
sense of Definition 10.1. Namely, we have
Example 10.1. Let X = e[O, Tj (over ~) with the pointwise multiplication. Let
D = ft,
Rp = J;,
(Fpx)(t) = x(p) for x E X, P E [O,Tj. Let a E X be given.
Write
B(t,p)
t
°
By definition, O(t,p) - O(s,p) = B(t, s) and O(p,p) = for p, t, s E [0, Tj. Hence
8 a(p)(t) = {[Fpe-RpajeRpa}(t) = e(J(t,p). Thus, by Theorem lOA, the equation
(10.16) x' - ax = y
has a unique solution x satisfying the initial condition x(p) = xo, namely,
OPERATOR ehD
Recall some definitions and theorems which will be used in our subsequent con-
siderations.
By V(X) we denote the set of all Volterra operators belonging to L(X), i.e. the set
of all operators A E L(X) such that 1- AA is invertible for all scalars A. Clearly,
A E V(X) if and only if VIFA = IF \ {O} (d. Formula (2.28)).
Let X be a Banach space. Denote by QN(X) the set of all quasinilpotent operators
belonging to L(X), i.e. the set of all bounded operators A E Lo(X) such that
lim
n-+oo
v'IIAnxll = 0 for x E X.
The set of all operators w-almost quasinilpotent on the set 0 will be denoted by
AQN(O;w). If w = VJFA then we say that A is almost quasinilpotent on O. The
set of all almost quasinilpotent operators on 0 will be denoted by AQN(O). 0
Lemma 11.1. Write
Then
(11.3) (t # 1; n E N)
and for n > mEN
sm,n(t)
(11.4)
= -(1 - t)-(m+l) [1 - t
k=m+l
(~) (1 - t)ktn-k]
+~ (n: k)t nH - m .
156
D. Przeworska-Rolewicz, Logarithms and Antilogarithms
© Kluwer Academic Publishers 1998
Operator ehD 157
s~~~ = v! nfv
)=0
(i: v) ti far all v E N.
Let xn(t) = 1 - t n for n E N and let y(t) = (1 - t)-I. Then, again by induction,
we obtain
y(k)(t) = k!(l- t)-(k+1) = k!yk+l(t) far kEN;
= ~ (7)x~k)(t)y(m-k)(t)
=_ f
k=O
(m) n! tn-k(m _ k)!ym-k+1(t)
k (n - k)!
= _~ m!n!(m - k)! tn-k(1- t)-(m-k+1)
f::o k!(m - k)!(n - k)!
= -m!(l - t)-(m+1) f
k=O
(~) (1 - t)ktn-k
t (~)tn-k(l_ t)k]
k=m+1
n
= -m!(l - t)-(m+1)[(l_ t + t)n - L
t
k=m+1
= ~ e:m)tk
t
k=O
= nfm (k+m) k
k=O m t + k=n-m
1
= _s(m)
m! O,n
+"~
)=0
m
1 d
= -
m
- nL:- t k +
1
L: m
(n + k) t n+k-m
m!dt m m
k=O k=O
= ~ d m
1-
m! dtm 1- t L....-
tn ~ (n m+ k) t n +k - m
k=O
=-(I-t)-(m+l)[I- t
k=m+l
(~)(I_t)ktn-kl+'f (n~k)tn+k-m,
k=O
which proves Formula (11.4). •
Theorem 11.1. (cf. PR[13]). Let n be a subspace of a complete linear metric
space X over IF. If A E L(X), n c dom A and 0 =J we vFA, then the following
conditions are equivalent:
(i) A is w-almost quasinilpotent on n;
(ii) for every A E w, x E n the series 2:::;"=0 An Anx is convergent and
L
00
lim An Anx =
n-+oo
o.
The arbitrariness of A E w, x E f! implies that A E AQN(f!iW).
(i)-+(iii) Since A E AQN(f!iW), Formula (11.4) implies that for all m ~ 2, n > m,
A E w, x E f!
Sm-l,n(AA)x
= (I - AA)-m[x - t
k=m
(~) (J - AA)k An- k An-kx]
Doo = n
kENo
dom Dk, where dom DO = Xi
or
Sh(I - AR)-l F = eAh(I - AR)-l F for A E Vf'R,
then Sh are said to be true shifts. The family {ShhEA(R) is a semigroup (or group)
with respect to the superposition of operators as a structure operation. 0
The following theorems will be given here without proofs, since the theory of
functional shifts presented in Appendix is more general. However, in this Chapter
we shall make use of properties of true shifts for their connections with old classical
theorems.
Theorem 11.2. (cf. PR[13]). Suppose that all conditions of Definition 11.2
are satisfied, {ShhEA(R) is a strongly continuous semigroup (group) of true shifts
and either P(R) = X or E(R) = X. Then D is an infinitesimal generator for
{ShhEA(R), hence dom D = X and ShD = DSh on dom D. Moreover, the
canonical mapping /'i, defined as
Theorem 11.3. (cf. PR[13]). Suppose that all conditions of Definition 11.2 are
satisfied and {ShhEA(IR) is a family of true shifts. Then for all h E A(JR) and
x E AR(D) the series
is convergent,
(11.8)
Definition 11.3. Suppose that X is a linear space over IF, D E R(X) and ker D :f:
{O}, F is an initial operator for D corresponding to an R E 'RD, A(JR) = ll4 or lR.
Let {ShhEA(IR) be a family of linear operators satisfying the condition
1
..1(h) = hD.hR, where hE A(JR) \ {O}
A(t) =- t = '""'
~ n!
-, B(t) Bnn! for t E JR,
n=l n=O
where Bn (n E No) are Bernoulli numbers (i.e. coefficients of the Taylor expansion
of the function
162 Chapter 11
1 1 1 1
B 2n+l = 0 for n EN, Bo = 1, Bl = -2' B2 = 6' B4 = - 30' B6 = 42' ...
and so on; all Bn are rational numbers). Then
ehD 1
(11.10) .J(h) = A(hD) = D- on AR(D),
Moreover,
= Bo.J(h) + L ~~ hn-l~hDn-l
00
= {x EX: L
00
and
L
00
where
An
= {x E dom A: L
00
An
=L
00
An
L -,
00
(11.16)
Clearly,
(11.17)
Theorem 11.6 (d. PR[15]) Suppose that all assumptions of Theorem 11.2 are
satisfied. Then on AR(D) n L~h (X)
1
= "hln(I + ~h) = L
00 ~n
(11.18) D (-It nhx for hE A(R) \ {O}.
n=l
is convergent for all x E 0, A E VJFA. This, and Formulae (11.13), (11.14) together
imply that 0 C L-AA(X) and that Formula (11.19) holds. •
Note 11.1. A ZELAZKO'S example (d. for instance, R[l]) shows that out of
locally pseudoconvex spaces convergence of series appearing in Theorem 11.7 may
hold in a finite number of points only. 0
164 Chapter 11
Corollary 11.1. Suppose that all assumptions of Theorem 11. 7 are satisfied. If
y = In(I - "\A)x E E-AA(X) for an x E E and a..\ E 'Uf'A then
(11.20)
Proof. Let II~II ~ a < 1. By our assumptions and Theorem 11.6, for x E AR(D)
we find
II In (I + ~h)xll
= II f:
n=l
(_I)n ~~ xii
n
~ f:
n=l
II~hlln Ilxli <
n
co n
< L
n=l
~llxli
n
= In (a + 1)lIxll·
Proposition 11.2. Suppose that all assumptions of Theorem 11.2 are satisfied
•
and X is a Banach space. Write for a, c > 0
II~hxll =
II(Sh - I)xll = II(ehD - I)xll
Ihl n n
= II L -, xII ~ cL
00 hnDn 00
_,a_llxll
n. n.
n=l n=l
Example 11.1 (d. PR[lO)) Suppose that all assumptions of Theorem 11.2 are
satisfied and dim ker D = 1. Then there is a 0, 10hl < Ihl, such that
Formula (11.21) shows that for true shifts a theorem about the intermediate value
holds (provided that the kernel of D is one-dimensional). 0
Results of this section can be essentially generalized if we take into account the so-
called functional shifts induced by analytic functions belonging to H(O). Here the
generating function is et . The theory of functional shifts was recently developed
by BINDERMAN and is presented in Appendix. Various families of these shifts have
the structure of a commutative algebra (with respect to the superposition, as the
structure operation). So that, under appropriate assumptions, we may prove that
the function e hD is an antilogarithm induced by the operator V = d~ and the
corresponding logarithm is hD.
Example 11.2. Suppose that X = Co(l~) is the space of all functions bounded
and uniformly continuous for t E JR and D = ft.
Note that X with the pointwise
multiplication is also an algebra.
The operator D is not right invertible in X. Indeed, constants are bounded func-
tions and a primitive of a constant is not a bounded function: J;
cds = ct. Hence
the operator R = J~ does not preserve the space X. However, in the space X
there is defined a strongly continuous semigroup {Sh} hEIR defined by means of the
Gauss kernels:
(ShX)(t) = {
(27rh)1/2 1+
-00
00
e-(t-s)2/ 2h x (s)ds for h > 0,
(h E JR).
x(t) for h = 0,
The infinitesimal generator for this semigroup is the operator
Rn
eARz = 2: .x
00
n _, z for all >. E C, z E ker D.
n=O n.
The operator eAR is invertible and its inverse is e- AR . Theorem 11.2 implies that
tk
(Rkz)",(t) = k!z for kEN, z E ker D,
(eARz)(t) = z 2:
00 (>.t)n
(n!)2
for all >. E C, z E ker D.
n=O
o
CHAPTER 12
We begin with
Definition 12.1. Let X E Lgr(D) n Lg,(D). Write for A E IF:
The mappings E r,).., El,).. and E).. are said to be of the power type with exponent
A or, if it does not lead to any misunderstanding, shortly, power mappings. 0
Note 12.1. We have seen that, without any additional assumptions, just by
definitions, left and right logarithms and antilogarithms of elements qe, where e
is the unit of X and q E Q, are well-defined (provided that CD =f. 0, d. Theorem
2.3). The questions connected with extensions of left and right logarithms and
antilogarithms to lR and <C in Leibniz algebras have been discussed in details in
Chapter 7 (d. Theorems 7.7, 7.8 and 7.9, Proposition 7.8 and Corollary 7.14.
o
Proposition 12.1. Suppose that either X E Lgr(D) n Lg,(D) or X E Lg(D),
(Lr,Er ) E G[DrJ, (Ll,EI) E G[Dd, (L,E) E G[D] and the mappings E r,).., El,)..,
E).. are defined by Formulae (12.1), (12.1'), respectively. Then for all A, /-l E IF
(i) Er,)..(dom Dr) C dom Dr, El,)..(dom Dl ) C dom Dl, E)..(dom D) C dom D;
(ii) LrEr,).. = ALr; LIEl,).. = ALl, LE).. = AL;
(iii)Er,)..Er,/J = Er,)../J' EI,)..EI,/J = EI,)..I" E)..E/J = E)..I"
i.e. the mappings E r,)..,
El,).., E).. are multiplicative functions of the parameter A.
Proof Let u E dom Dr, A E IF. By definitions,
Theorem 12.1. Suppose that X E Lr(D)n Ll(D) and (Lr, Er) E G[Or],
(Ll' E I) E G[Od. Let El = dom Orndom 0 1• Then for all A E IF and U E J(X)n El
(i) EI,AU- 1 = (Er,AU)-l;
(ii) if X E L(D), (L, E) E G[O], u E dom 0, A E IF then EAu- 1 = (EAu)-l;
(iii) if D E R(X) then Er,A' EI,A' EA E M(X).
Proof. (i) Theorems 7.2 and 7.3 together imply that
(ii) follows immediately from (i), since X is commutative and, by Theorem 7.4,
uEX.
Er,A(UV) = Er[ALr(uv)]
= Er{ALrv + AR[v-1(DLru)v]}
= Er{Ay + AR[(Ery)-l(Dx)Ery]}
= Er(AX)Er(AY)
= Er(ALru)Er(ALrv)
= Er,A(U)Er,A(V).
Similar proofs for EI,A and EA.
Proposition 12.2. Suppose that X E L(D) and (L, E) E G[O]. Then for all A,
•
J.L E IF and u E dom 0
(12.2)
If J.L = -A then we get (EAu)(E_Au) = E(O· Lu) = E(O) = e. This implies that
EAu E J(X) and (EAu)-l = E_Au. •
Observe that Proposition 12.2 does not hold in the non-commutative case. In
general, we have
Power mappings 169
= Er{CDALrv
+ R[CDA(Er,>.v)-lu-l(Du)(Er>.v)
+ (Er,>.V)-l (Er,>.u)fD(Er,>.u, Er,>.v)) + zr};
= Ec{CDALIU
+ R[CDA(EI,>.u)-1(Dv)v- 1 (EI>.u)
fD(E I,>.u,EI,>.v)(EI,>.v)-l(EI,>.U)) + zz}.
The proof follows immediately from Definition 12.1 and Theorem 2.2.
Corollary 12.1. Suppose that all assumptions of Proposition 12.3 are satisfied
and CD = O. Then the mappings E r,>., E l ,>. are not defined for A =P 1. If A = 1
then Er,l = EI,l = Jldom 0·
Proof Let u E dom Or. Then x = Lru E dom 0- 1 . Let A E IF. By Theorem
2.4, if AX E dom 0;1 then A = 1. Hence E r,>. is not defined for A =P 1. If A = 1
then we have Er,>.u = Er(ALru) = ErLru = u. The arbitrariness of U E dom Or
implies that E r,l = Jldom 0 •. A similar prooffor El,>.. •
Corollary 12.1 implies that for multiplicative D the mappings E r,>., E l ,>., E>. are
not defined (cf. Note 12.1).
Clearly, we can extend Definition 12.1 to left invertible operators. For the sake of
brevity, we shall consider only the commutative case, i.e. the class Lg#(D). We
get
Proposition 12.4. Suppose that X E Lg#(D}, (L, E) E G[L) and the mapping
E>. is defined by (12.1'). Let A E IF. Then E>.(dom L) C dom L and LE>. = AL
(cf. Proposition 12.1).
Proposition 12.5. Suppose that Condition [L)# holds and (L, E) E G[L). Let
A E IF. Then E>. E M(X) and DE>.u = A(E>'_lU)Du for u E dom L (cf. Proposi-
tion 12.2).
In general, we have
Proposition 12.6. Suppose that X E Lg#(D) and (L, E) E G[L). If A E IF and
u, v E dom L, E>.u, E>.v E J(X) then
Corollary 12.2. Suppose that all assumptions of Proposition 12.6 are satisfied
and CD = O. Then the mapping E>. is not defined for A =I- 1. If"\ = 1 then
E1 = Jldom !1 (cf. Corollary 12.1).
Definition 12.2. Suppose that either IF = JR or IF = C, X is a complete m-
pseudoconvex Leibniz algebra with unit e, either X E Lg(D) or X E Lg#(D),
e E dom 0- 1 and (L, E) E G[O] (Recall that for D E A(X) we have 0 = L).
Write
(12.3) Eb(X) = {u E dom L: "\Lu E ED(X) for (L,E) E G[OJ, ..\ E IF}
(12.5)
u>' = un = U· .... u.
~
n-times
Proof. Since X is a Leibniz algebra, we have Dxn = nxn- 1Dx for all x E dom D,
n E N. (cf. Corollary 7.12). Since D is closed, by Theorem 7.6, De x = eX Dx.
Hence eX = Ex E dom 0 and LeX = LEx = x.
(i) Suppose that AU E ED(X) for a ..\ E IF, u E dom O. Then, by Proposition
12.2, u>' = E>. = E("\Lu) = e)..Lu and Du>' = De)..Lu = e>.Lu D("\Lu) = u>' D(..\Lu).
Hence u>' = e)..Lu E dom 0 and Lu>' = ALu.
(ii) If u E J(X) n EbE(X) then Du>' = DE>.u = "\(E>'_lU)Du = ..\u>'-1 Du.
(iii) Let ..\ = n E N. Since X is a Leibniz algebra, by (i), u>' = E>.u = E(ALu) =
E(nLu) = ELu n = un. •
Example 12.1. Let X = C(JR) with the pointwise multiplication and with the
topology induced by the uniform convergence on closed intervals. Let D = ft,
x E X, x(t) > 0 for t E JR. Then x>' = e>.lnx, Dx>' = "\X>.-1 Dx.
Consider the space X equipped with the topology induced by the uniform conver-
gence on compact sets and assume that D is closed. It is easy to verify that in
Power mappings 171
this case true shifts are of the form (S~x)(t) = x(rt), where h = In r, r E ll4 (cf.
Definition 11.2, Theorem 11.3). 0
In order to establish the relationship between the number e and the unit e of an
algebra under consideration it is very useful the following
Definition 12.2'. Suppose that X E Lg(D) and (L,E) E G[O]. Write
and
x Y = E(yLx) whenever (x,y) E Y(O).
By definition, Lx Y = yLx. Indeed, Lx Y = LE(yLx) = yLx. Let u = x Y for
(x, y) E Y and let y E I(X). Then
D(xY) = DE(yLx)
= E(yLx)D(yLx) = xY[(Dy)Lx + yDLx]
= xY[(Dy)Lx + yx- 1 Dx].
n=l
tn
P
=
{
lt S S -ld S
L'Zp-l () for p ~ 2,
ItI < 1 (p E No).
o
o
Definition 12.4. Suppose that X E Lg(D) is a complete m-pseudoconvex algebra
over IR or C. Let p E No be arbitrarily fixed. Write
00 un
(12.8) £~)(X) = {u E dom 0: '"'
~ nJ
--:- are convergent (j = O,l, ... ,p)}.
n=l
o
Proposition 12.8. Suppose that X E L(D) is a complete m-pseudoconvex algebra
over IR or C, (L, E) E G[Ol, D is closed and pEN. If u E £~) (X) then
(12.10)
00 Dun
Dx = L - = (e - u)-l Du = -(e - u)D(e - u) = -DL(u - e),
n=l
n
174 Chapter 12
which implies (12.9). Since Li1u = x = -L(e - u), we find e - u = E(-x) and
u = e - E( -x). Similarly, for p 2:: 2 we get
D n n-l un-1
L n~ = L =L
00 00 00
= n:p n P - 1 Du
n=l n=l n=l
n
(L
00
Corollary 12.3. Let pEN be arbitrarily fixed. Suppose that all assumptions of
•
Proposition 12.8 are satisfied, 9 = Re E cg) (X) for an R E RD and the initial
operator F corresponding to R is continuous. Then
(12.11)
Note 12.2. Observe that both, Formula (12.10) and the second of Formulae
•
(12.11) are generalizations of (12.7). 0
Proposition 12.9. Suppose that X is a complete m-pseudoconvex algebra over
~ or ee, Condition [M] holds, (L, E) E G[O] and D is closed. Then
(12.12)
1
L
00
(12.14) Lipg = 9 nP ' where 9 = Re, R E RD (p E No).
n=1
00
1 00
1 00
1
Lipg = R[e"
~nP
-J + FLipg = (Re) "~nP
- = g~nP
"
n=1 n=1 n=1
The functions Lip have been introduced and studied from the first beginning of
•
Mathematical Analysis, since they have several interesting properties, some of an
importance in applications to physics and engineering in the last few decades.
The results already obtained can be applied in order to solve some non-linear
equations. For instance, we have
Proposition 12.10. Suppose that X E Lgr(D) n LgI(D) is a Leibniz algebra,
(Lr,Er) E G[Or], (Ll,El) E G[Od, a,y E X, a E «:11\ {O} and Ra E dom Or n
dom 0 1 for an R E RD. Then
(i) x is a solution of the equation
(iii) ify = 0 then all solutions of Equation (12.15) are of the form x = -O'.- 1L , (z-
O'.Ra) , where z E ker D is arbitrary;
(iv) ify = 0 then all solutions of Equation (12.17) are of the form x = _0'.-1 Lr(z-
oRa), where z E ker D is arbitrary;
(v) if -a E vQRy and Z E ker D then all solutions of Equation (12.16) are of the
form u = LI(I + O'.Ry)-1(z - O'.Ra);
(vi) if -a E vtQRy+- and Z E ker D then all solutions of Equation (12.17) are of
the form x = -O'.Lr(I + O'.Ry+-)-1(z - O'.Ra);
(vii) if -1/0'. is an eigenvalue of the operator Ry then a necessary and sufficient
condition for the resolving equation (12.16) to have solutions is that there is a
Z E ker D such that z - O'.Ra E (I + O'.Ry)dom Or. If it is the case, then all
solutions of (12.16) are of the form u = (z - O'.Ry) (-1) + U-1/a, where by V( -1) we
denote an element of the inverse image of an element v by the mapping I + O'.Ry,
U-1/a E ker(l + O'.Ry) is arbitrary, hence Equation (12.15) has all solutions of the
form x = -0'.-1 L1u.
(viii) if -1/0'. is an eigenvalue of the operator RYt- then a necessary and sufficient
condition for the resolving equation (12.18) to have solutions is that there is a
z E ker D such that z - O'.Ra E (I + O'.Ry+-)dom 0 1. If it is the case, then all
solutions of (12.18) are of the form u = (z - O'.Ra)(_l) + U-1/a, where we denote
by v( -1) an element of the inverse image of an element v by the mapping I +O'.RYt-,
U-1/a E ker(I + O'.Ry+-) is arbitrary, hence Equation (12.17) has all solutions of
the form x = _0'.-1 Lru.
Proof (i) Let x be a solution of Equation (12.15). Set u = E,(-O'.x). Then
u E dom Or. By Theorem 7.3, u E I(X) and u- 1 = [E,( -O'.x)]-1 = Er(O'.x).
Hence
(Du)u- 1 = D[EI(O'.X)]Er(-O'.x)]
= -O'.(Dx)[E,(O'.x)]Er(-O'.x)
= -O'.Dx = -O'.(au- 1 + y)
= O'.(a + yu)u-l,
i.e. Du = O'.(a + yu). This implies that (I + Ry)u = -O'.Ra + z, where Z E ker D
is arbitrary, i.e. u should satisfy the resolving equation (12.16). Conversely, if u
is a solution of Equation (12.16) for a Z E ker D, then x = _0'.-1 LIU is a solution
of Equation (12.15). A similar proof for (ii).
(iii) Suppose that y = O. Then from the resolving equation (12.16) we find u =
z - O'.Ra, where Z E ker D is arbitrary. Hence x = -0'.-1 L1U = _0'.-1 LI(Z - O'.Ra).
A similar proof for (iv).
(v) Suppose that y =I 0, however -a E vtQRy, i.e. the operator I +O'.Ry is invertible.
Then from the resolving equation (12.16) we get u = (I +O'.Ry)-1(z-O'.Ra), where
Z E ker D is arbitrary. A similar prooffor (vi).
Power mappings 177
Proof. Let v = yE_Otx. Then y = vEOtx and Equation (12.19) can be written
in the form: Dx = (a + v)EOt. This, and Proposition 12.11 together imply that
x = E 1!(1_Ot)[(1-a)R(a+v)+z], where z = Fx E ker D. Thus x satisfies Equation
(12.19). On the other hand, if x satisfies (12.20), then it is easy to check that x is
a solution of Equation (12.19). •
Proposition 12.13. Suppose that either IF' = R or IF' = C, X is a complete m-
pseudoconvex algebra with unit e, Condition [L] holds, e E dom n- 1 , (L, E) E
G[O], D is closed, a E X and (l-a)Ra E £b(X) for an R E RD and an a E 1F'\{1}.
Then the general solution of the equation
(12.21)
(12.21')
DE[(e - b)Lx]
= E[(e - b)Lx]D[(e - b)Lx] = E[(e - b)Lx](e - b)DLx
= E[(e - b)Lx](e - b)x- 1Dx = E[(e - b)Lx](e - b)x-1axb
= E[(e - b)Lx](e - b)ax b- e = (e - b)aE[(e - b)Lx]E[(b - e)Lx]
= (e - b)aE[(e - b + b - e) Lx] = (e - b)aE(O) = (e - b)ae
=(e-b)a.
Power mappings 179
(12.22) Du = au 2 +y,
Dv = Du+Dw'
=
= au 2 + y + Dw' a( v - W')2 + y + Dw'
= av 2 - 2avw' + aw,2 + y + Dw',
which implies Dw' = 2avw' + aw'2. Suppose that w' E leX) and w = W,-I. Then
we get Dw = DW,-1 = _W,-2 Dw' = -(2avw,-1 + a) = -2avw - a which implies
D[l + 2R(av)] = -a. Hence w is of the form (12.23). •
In particular, if a E ker D, y = _a- 1 z 2 , z E ker D, Equation (12.22) can be
reduced by the substitution v = -yu-1 to the equation D 2v = v 2 - z2 solvable in
an explicit way. Namely, we have
Proposition 12.15. Suppose that Condition [L] holds, (L, E) E G[O), R E RD,
9 = Re and a E leX) n ker D. Then the special generalized Riccati equation
(12.24)
L(x - a)(x + a)-l = L[(x - a)(x + a)-I] = 2Ra + z' = 2ag + z',
Power mappings 181
(12.25)
belonging to J(X) n dom 0 are of the form x = z'ER[z - R(d - e)t 1, where
z,z' E ker D.
Proof. Let u = Lx, i.e. x = Eu and
Dx = DEu = (Eu)Du = xDu; D 2x = (Dx)Du + xD 2u = X[(DU)2 + D 2u].
182 Chapter 12
N
Q(D) =L Qk Dk , Qk E L(X) (k = 0,1, ... , N - 1), QN = I.
k=O
Suppose, moreover, that the operator Q(I, R) = I:f=o QkRN-k is invertible for
an R E RD. Then x is a solution of the equation
The proof follows immediately from the form of the general solution to the equation
Q(D)x = y (cf. Theorem 5.3). •
Similar results can be obtained for operators of the form:
(cf. in particular, Theorems 5.3 and 5.4). Equations with operators of a more
general form can be solved by applications ofresults of NGUYEN VAN MAU N[2]).
Proposition 12.18. Suppose that Condition [L] holds, (L, E) E G[n], a E dom n
and f is a mapping of dom n into itself. Then a necessary and sufficient condition
for the equation
Proof. We are looking for solutions x E dom n- l . Let u = Ex. Then u E dom n.
By definition, Du = uDx = u[f(aEx) + y] = u[J(au) + y], which implies u =
Ru[f(ay) + y] + z, where Z E ker D is arbitrary. Clearly, if this equation has a
solution u E dom n then x = Lu is a solution of Equation (12.26). Conversely, if x
is a solution of Equation (12.26) then u = Ex E dom n satisfies Equation (12.27) .
(12.28) Dx = aEf(x) + y, y E X,
has solutions Xl = REI /(a-1) (0'.-1 )..-lg) + Z; X2 = RZI + zo, where z, Zl, Zo E ker D
are arbitrary. If (L, E) E GR,l [flj then Xl = REl /(cr-1) (0'.-1 )..-lg).
Proof. Similarly, as in the proof of Proposition 12.20 we obtain from Equation
(12.31) the equation [)"O'.(Dx)cr-l + g]D 2x = O. The equation D 2x = 0 has all
solutions of the form X2 = RZI + Zo, where zo, Zl E ker D. Suppose now that
)"O'.(Dx)cr-l + 9 = O. Then (Dx)cr-l = _O'.-l)..-lg. Hence
f(z) __ az+b, h
were a, b, c, d E II....
tr'
an d a d - b..J.
c r O.
cz+d
o
We shall see that this property can be generalized. Namely, we have
Theorem 12.3. Suppose that Condition [Lj holds and (L, E) E G[fl]. Let
generalized Schwarzians Sn (n E N) be defined as follows:
Let n ~ 2. Then
1
u = (DX)-1 D 2x = --1 (zo - g)nn(n - l)(zo - g)-n-l = n(zo _ g)-I.
n-
Thus u 2 = n 2(zo - g)-2 and Du = -n(zo - g)-2( -DRe) = n(zo _ g)-2 = ~u2
which implies that x E kerSn . Let now n = 1. Then x = -L(zo - g) and
DLv = v- 1Dv = u
=n(zo - g)-1 = -nDL(zo - g)
= D[-nL(zo - g)] = DL(zo _ g)-n,
which implies Lv = L(zo - g)-n + LZI = L[ZI(ZO - g)-n], where ZI E ker D is
arbitrary. Condition [L] implies that
1 1
Dx = v = ZI(ZO - g)-n = zID[--(zo
n-l
- g)-n+l] = D[--ZI(ZO - g)-n+l],
n-l
186 Chapter 12
which implies
1
x = --1Z1 (Zo - 9 )-n+l + Z2, where Z2 E ker D is arbitrary.
n-
(12.35)
(12.36)
n-l
where Z = L Rkzk' ZI, .. ·,Zn-l E kerD are arbitrary,
k=O
has a solution U E dom On. If U is a solution of Equation (12.36) for a Z E ker D,
then x = -a- 1 Lnu is a solution of Equation (12.35).
Proof Let U = En(ax). Then u E dom On and x = a-I Lnu. By definition,
In Chapter 11 we have introduced the space Doc of smooth elements and its
subspaces S = P(R) of D-polynomials, E(R) of exponentials and AR(D) of D-
analytic elements. In addition, we will introduce other subspaces of Doc. Namely,
let D E R(X), ker D -:j:. {O} and let F be an initial operator for D corresponding
to an R. We shall consider the following subspaces of Doc:
• the space of singular elements
QR(D) = {x E Doc: V FDnx = O} = {x E Doc: V RnDnx = x};
nENo nEN
and, if X is a complete linear metric space and either IF = lR or IF = C,
• the space of D-paraanalytic elements
i.e. such elements that the nth terms in their Taylor expansions are tending to
zero (cf. PR[9]). Note that the inclusion AR(D) ffi QR(D) c PAR(D) is essential
(cf. ROGULSKI Rog[l], also PR[9]).
Theorem 13.1. Suppose that X E Lgr(D) (X E LgI, X E Lg(D)) is a complete
linear metric space over IF (lR or IF = C) , (Lr,Er ) E G[Or] (Ll,Et) E G[Ot],
(L,E) E G[OJ, respectively), D E R(X), there is an R E RD n AQN(ker D), F
is an initial operator for D corresponding to Rand 9 = Re. If >..g E dom Or
(>..g E dom Ol, >..g E dom 0, respectively) for>.. E VFR, then
Er(>"g), El(>"g), E(>..g) E ker(Dn - >..n I) far all n EN,
187
D. Przeworska-Rolewicz, Logarithms and Antilogarithms
© Kluwer Academic Publishers 1998
188 Chapter 13
Since>. E VJFR, we conclude that Er(>.g) = (I - >'R)-l Zr. By our assumption the
series 2:~=o >.n Rn z is convergent for all z E ker D and>' E VJFR. This implies
00
n=O
n=O
= (L
00
Moreover, Er(>.g) = =
0 if and only if FhEr(>.g) 0 for all h E A(JR), El(>.g) 0 if=
and only if FhE/(>.g) = 0 for all hE A(JR), E(>.g) = 0 if and only if FhE(>.g) = 0
for all h E A(JR), where Fh = FSh.
Proof By Theorems 11.3 and 13.1, for all h E A(JR)
Thus
FhEr(>.g) = FShEr(>.g) = eAhFEr(>.g) = eAh(I - >.R)Er(>.g).
This and the invertibility of I - >'R together imply that Er(>.g) = 0 if and only if
FhEr(>.g) = 0 for all h E A(JR). Similar proofs for El(>.g) and E(>.g). •
Note 13.2. Clearly, the operators Fh = FSh (h E A(JR)) belong to :Fv, since they
are projections onto ker D and the corresponding right inverses are Rh = R - FhR
Smooth logarithms and antilogarithms 189
(h E A(~». One can prove that VFRh = VFR for all h E A(JR) (cf. BPR[2j, also
Appendix). 0
Theorem 13.2. Suppose that X E Lr(D) (X E LI(D), X E L(D») is a complete
linear metric space over IF (where IF = ~ or IF = C) and (Lr,Er ) E GR,l[Orj
(Ll,EI ) E GR,dOd, (L,E) E GR,dO], respectively) for a continuous R E 'RD. If
u E J(X) n AR(D) and u- l E AR(D) then Lru, L1u, Lu E AR(D), respectively.
Proof. Ifu,v E AR(D) then Du,Ru, DV,Rv E AR{D) (cf. PR[9]). By the Leibniz
formula, uv E AR(D). Let u E J(X) n AR(D) and let u- l E AR{D). Then
DLru = u- l Du, DLlu = (Du)u- l , DLu = u- l Du E AR(D). Let F be an initial
operator for D corresponding to R. By our assumptions, F Lru =
0, F Lzu 0, =
FLu = 0, respectively. This implies
(13.1)
L
n=O
:!n
00 n 00 1
(13.2) Fe x = FL..J
" ~ = Fe + "
n! L..J
-Fxn
n!
= e.
n=O n=l
FDn+le x = FDn(De X )
t
k=O
= (~) (FDn-kex)FDk+lx = O.
k=O
Theorem 13.5. Suppose that all assumptions of Theorem 13.3 are satisfied,
FI =F F is a multiplicative initial operator for D corresponding to an Ri E nD
and 9 = Re E leX).
(i) If
(13.3)
(13.4)
(RIe)n
RiZ = z--,
n.
- for Z E ker D, n E N.
R~FIDnLrg
lim RrFIDng
n~oo
= DRI lim DnLrg
n~oo
=D lim R~+lFIDn+lLrg
n~oo
= O.
Hence g-1 E PARI (D). The same conclusion is obtained by means of the left
logarithm L , .
(iii) Since the series (13.4) is convergent, by (ii), the series
L
00 00
Hence, by (iii), g-1 = DLrg = DL,g EARl (D). Since the product of two elements
belonging to ARI (D) again belongs to ARI (D) (cf. PR[9]) , by an easy induction
we prove that g-n = (g-l)n EARl (D). •
Immediate consequences of Theorem 13.5 are
Corollary 13.3. Suppose that all assumptions of Theorem 13.5 are satisfied, Rl
is continuous,
N
P(t) = L Pk tk E JF[t]
k=O
and the series (13.5) is convergent. Then P(g-l) = g-np1(g) E ARI(D), where
N = degP(t) and
N
Pt(t) =L PN_k tk .
k=O
Corollary 13.4. Suppose that all assumptions of Theorem 13.5 are satisfied and
P(t) E JF[t]. Then P(Lrg), P(L,g) EARl (D).
Theorem 7.6 immediately implies for commutative algebras with left invertible
operators the following
Proposition 13.1. Suppose that X is a complete m-pseudoconvex algebra over
JR, Condition [L]# holds, (L, G) E G[L] and D is closed. Then De" = e" Dx for
x E £D(X) (cf. also Theorem 13.4).
However, we have to remember that spaces S, AR(D) and QR(D) are well-defined
for right invertible operators with ker D ¥ {O}.
Again we shall see that in the mUltiplicative case we have a different situation.
Smooth logarithms and antilogarithms 193
n=O
n!
Note 13.5. Observe that e'" is not an antilogarithm for a multiplicative D. If,
•
in particular, D E A(D), X is a complete m-pseudoconvex algebra over IF, where
either IF = IR or IF = C, Condition [M] holds, (L, E) E G[O] and D is closed, then
De'" = e D'" whenever x, Dx E CD(X). 0
Proposition 13.3. Suppose that all assumptions of Proposition 13.2 are satisfied
and L is continuous. Then
(i) Lu E CD(X) and eLu = LeU = u for u E CD(X) n dom OJ
(ii) Du = e Du - u ifu, Du, Du - u E CD(X).
Proof By Proposition 2.6, Land E are multiplicative.
(i) By definition, if u E CD(X) n dom 0 then x = Lu E dom 0- 1 and
00 (L)n 00 L nOOn
U = Ex = eX = e Lu = ""'
L....t
_u_
n!
= ""'
L....t
= L ""'
~
n!
~
L....t n!
= Leu.
n=O n=O n=O
Example 13.1. Suppose that all assumptions of Proposition 13.3 are satisfied, F
•
is an initial operator for D corresponding to an R E RD and Z E ker D n CD(X).
We have De z = e Dz = eO = e and De Rz = e DRz = e Z • Hence
o
194 Chapter 13
o
Proposition 13.4. Suppose that Condition [M] holds and (L,E) E G[O]. Then
(i) Dn Ex = (Ex)r~(x) for all x Edam Dn n dam 0- 1 , n E N;
(ii) Ex E Doo for x E Doo n dam 0- 1 .
Proof. (i) by induction. For n = 1, by definition, DEx = (Ex)Dx = (Ex)rp(x).
Observe that r~(x) is well-defined for x Edam Dn. Suppose (i) to be true for an
arbitrarily fixed n E N. Let x E dom D n +1 n dom 0- 1 . Then
D n+1 Ex = D[(Ex)r~(x)]
n
= (DEx)[Dr*,(x)] = (Ex)(Dx)D II Dix
i=1
n n+1
= (Ex) (Dx) II Di+ 1 x = (Ex) II Dkx
j=1 k=1
= (Ex)r*'+1(x).
Corollary 13.5. Suppose that all assumptions of Proposition 13.5 are satisfied
•
and X is a complete linear metric space over R If FEx = 0 then Ex AR(D). rt
Proof. It is an immediate consequence of Proposition 13.5, since AR(D)nQR(D) =
{O} (d. PR[9]) and Ex i= o. •
Proposition 13.6. Suppose that Condition [M] holds, (L, E) E G[O], F E
ML(X) is an initial operator for D corresponding to an R E RD and u E Doo n
dom O. Suppose that FLu = Fu = O. Then Lu E QR(D) if and only if u E
QR(D).
Smooth logarithms and antilogarithms 195
Proof. Let u E Doondom r! and let x = Lu. Then u = Ex. Since D,F E ML(X)
and Du = uDLu, we get for all n E N
FDnu = F[(Du)n)
(13.7)
= (FDu)n = [F(uDLu)t
= (Fu)n(FDLu)n = (Fu)n F(DLu)n
= (Fu)nFDnLu.
Since, by our assumption, ker D has no zero divisors, we find F Dn Lu = 0 for all
n E N, which implies Lu E QR(D). •
Proposition 13.7. Suppose that X is a complete linear metric space over JR,
Condition [M] holds, F, Fl E M L(X) are initial operators for D corresponding to
R,R1 E RD, Fl f:- F, 9 = Re E leX) and (L,E) E G[r!). If
(13.8)
then
(i) Lg EARl (D);
(ii) g-1 EARl (D), provided that Rl is continuous.
Proof. (i) Observe that Dgn = Dg-n = e for n E N. Indeed, for n E N we have
Dg n = (Dg)n = en = e and e = De = D(gng-n) = (Dgn)(Dg-n) = Dg-n.
Hence for n ~ 2 we have Dn Lg = Dn-l g l = e. We therefore obtain
and
n=1
= R1 FIg- I + RlFle + ... + Rf' FIe
= RIg- 1 - Rie + Rie - Rre + ... + Rf' e - Rf'+le
= RIg- 1 - Ri"+1e.
196 Chapter 13
is convergent and
00
(ii) follows from (i) in a similar manner, as (iv) from (iii) in the proof of Proposition
13.3. •
Example 13.2. Let X, e, D, R be defined as in Example 5.3. Then D E M L(X)
and Condition [M] holds. Let U = {un}. If Un =J 0 for n E N then L{ Un} = {x n },
where
Xn = ~ for n ~ 2 and Xl E IF \ {O} is arbitrary,
Un-I
is a logarithm of u. Observe that all Xn =J O. It is easy to verify that E{x n } =
{Un} = UI bn(x)}, where
for n = 1;
otherwise;
(13.9)
(13.10)
the operators DO, RO, FO are well-defined on XO and map XO into itself.
We assume that the topology in XO is determined by the coordinatewise conver-
gence induced by the topology in X. It is not difficult to verify that DO E R(XO),
FO is an initial operator for DO corresponding to Volterra right inverse RO of DO
and ARo(DO) = Xo.
The following question arises: when does a sequence {zn} C ker D determine a D-
analytic element, i.e. when does there exist an x E AR(D) such that Zn = FDn x
for all n E No?
The answer is given by
Theorem 13.6. Suppose that X is a complete linear metric space, F is continuous
initial operator for a closed D E R(X) corresponding to a continuous R E 'RD.
Write
00
(13.11)
L..J RnZn+k+l is convergent}.
V '"'
ZD ,R = {{Zn} C ker D: kENo
n=O
Then for every {Zn} E ZD,R there is a unique x E AR(D) such that
(13.12)
is convergent. Put
=L
00
x Rnzn+ 1 •
n=O
Then for all kEN we find
L
00
Dkx = Dk Rnzn+l
n=O
and
00 k-l 00 00
=L
00
=L
00
= Rn L L =L
00 00 00
(13.14)
i.e. q, is a D-R structure preserving isomorphism (d. VON TROTHA T[l]). Sup-
pose, moreover, that dim ker D = 1. It means that ker D = lin {d} for an ar-
bitrarily fixed d E ker D \ {O}. We thus may identify ker D with the field IF of
scalars by the mapping: Z = ad --+ a, where a E IF, Z E ker D. SO that we
can introduce in XO = ARo(D) the coordinatewise multiplication of elements and
Smooth logarithms and antilogarithms 199
Zn
where Vn = - - for n 2': 2, Vl E IF \ {O} is arbitrary,
Zn-l
The function ax is said to be a symbol function of the element x (cf. PR[lO]). Our
assumptions imply that
d
aDx(t) = dtax(t) for x E dom D,
for x E X, t, h E A(JR).
Suppose that X is a commutative algebra and the symbol function is multiplica-
tive, i.e. axy = axay for x, y EX. It is so, for instance, when F and Sh
(h E A(JR) are all multiplicative. Then the basic equation can be rewritten as
follows: a~ = aua~ for u,x n dom D, u E leX). Thus aLu = ax = In au + Inc,
c E IF is arbitrary.
Note that the notion of a symbol has been introduced firstly by S. G. MICHLIN in
1948 for multidimensional singular integral equations. 0
Example 13.5. Suppose that X E Lg(D)) is a complete linear metric space over
C, (L, E) E G[n], D E R(X), there is an R E RD n AQN(ker D), F is an initial
operator for D corresponding to R, 9 = Re, >"9 E dom n for>.. E vcR and {ShhEIR
is a family of true shifts induced by R (cf. Definition 11.2). Write
n m
(13.16) WeD) =L LakjDkS_hj,
k=O j=O
200 Chapter 13
(13.20) W(D)x = o.
These >"s, which satisfy the equation W" (>.) = 0, are said to be characteristic roots
of the quasipolynomial W"(t). Observe that, in general, the quasipolynomial has
an infinite number of roots, Le. Equation (13.20) has infinite number of solutions
being antilogarithms.
On the other hand, if X is a complete linear metric locally convex space then, by
Theorem 11.2, we get
In the case, when g, G belong to the space HJ.L(f) of functions satisfying the Holder
condition on f with an exponent J.L, 0 < J.L < 1:
HJ.L(f) = {x E C(r) : Ix(t) - x(s)1 S Cit - slJ.L} (C E 114),
the singular integral operator 8 defined by (**) is an involution in the space
X = HJ.L(f): 8 2 = I on X. Thus there are disjoint projectors p+ and P- giving
the partition of unit and such that cI>+ = P+x, cI>- = P-x for an x E X (cf. PR[3] ,
also PR[6]). This x is to be found if we apply logarithms to the homogeneous
problem (*) (i.e. with 9 = 0). Then the non-homogeneous problem is solved by
a use of the already found solution to the homogeneous problem.
This problem can be formulated and solved in the same manner if f is a finite set
of closed regular arcs with a properly defined orientation of that system. 0
This is the way which we will follow in order to find solutions to a generalized
Riemann-Hilbert problem. Namely, we shall consider algebras with logarithms
induced by a linear operator D and with an involution 8. In commutative alge-
bras solutions will be obtained in a similar manner as in the classical case. In
noncommutative algebras we shall need some additional assumptions and some
modifications of considerations used in the commutative case. Recall
201
D. Przeworska-Rolewicz, Logarithms and Antilogarithms
© Kluwer Academic Publishers 1998
202 Chapter 14
Theorem 14.1. (cf. PR[3J; also properties of algebraic operators in Chapter 8).
Let X be a linear space over an algebraically closed field IF with an involution
S E Lo(X) (i.e. S2 = I, S f. 1). Then p+ = 1(1 + S) and P- = ~(I - S) are
disjoint projectors giving partition of unit: (pf)Z = p±; p+ p- = p- p+ = 0
and p+ + p- = I. Moreover, p+ - p- = Sand Sp± = p±S = ±P±. So that,
if we write X± = p±X, x± = P±x for x EX, then we have
This implies
Corollary 14.1. Let all assumptions of Theorem 14.1 be satisfied. If u E X,
u± = P±u and
(14.1) where v E X,
Xo = Xo+ + Xo-
= ELxci + ELxci = Ey+ + Ey-
= E(P+y) + E(P-y) = E(P+ SLa) + E(P- SLa)
= E(P+ La) + E( -P- La).
204 Chapter 14
Since X is a Leibniz algebra, we find E( -P- La)E(P- La) = E(P- La-P- La) =
E(O) = e. Thus, by Lemma 14.1, we find xi) = E( -P- La) = [E(P- La)]-l.
Hence xi) E J(X). Similarly, xt = E(P+ La) = [E( -p+ La)]-l. Thus xci E J(X) .
xoE(P- La)
= E(P+ La)E(P- La) + E(P- La)E( -P- La)
= E[(P+ + P-)La] + E[P-(La - La)]
= ELa + EP-O = a + E(O) = a + e.
Since E(P- La) E J(X) and [E(P- La)]-l = E( -P- La), we get the required
formula. •
Corollary 14.3. Suppose that all assumptions of Theorem 14.2 are satisfied and
a E J2 (dom n). Then the solution to the problem (14.3) can be written in the
form: Xo = (a~ + a-~)E(SLa~).
Proof. Since X is a Leibniz algebra, our assumptions and Corollary 14.2 together
imply that
y+ = (+)-1
Xo Xl+ = (-)-1
Xo a -l( aX l- + b) = (-)-1
Xo Xl- + (Xo-)-1 a -lb = y - + Yo,
where
i.e.
We also have
Thus
Xl =
+ + Xl-
Xl
Corollary 14.5. Suppose that Condition (A) is satisfied. Then the involution S
•
is not multiplicative.
n
Proof. Suppose that the involution S is multiplicative.Let x, y E X dom ben
arbitrarily fixed. By Lemma 14.2, x-y- E X-. By Proposition 14.1, x-y- E X+.
Since X = X+ EBX-, we have X+ nX- = {OJ. Thus x-y- = O. The arbitrariness
of x and y implies x- = y- = 0 and X- = {OJ. Thus S = I, which contradicts
our assumption that S is an involution, i.e. S2 = I, however S :I I. Hence S is
not multiplicative. •
Corollary 14.5 is important, since there are several examples of functional-
differential equations with a multiplicative involution which can be also solved by
means of algebraic methods (cf. PR[3], PR[4], PR[5], PR[6], also next chapters).
In order to study the noncommutative case, instead of Condition (A) we shall
admit either the condition
(A)I Let IF be algebraically closed. Let D E A(X). Suppose that X is a Leibniz
algebra with unit e and with an involution S E Lo(X) such that X E Lg1(D),
uv E X± whenever u,v E X± and DS = SD on dom D, (LI,EI ) E G[nd and
(14.7)
or the condition
(A)r Let IF be algebraically closed. Let D E A(X). Suppose that X is a Leibniz
algebra with unit e and with an involution S E Lo(X) such that X E Lgr(D),
uv E X± whenever u,v E X± and DS = SD on dom D, (Lr,Er ) E G[nr] and
(14.8)
In particular, (14.7) implies that X± C dom nl, (14.8) implies that X± C dom nr •
Note 14.1. Let either Condition (A)I or Condition (A)r be satisfied. The assumed
condition that uv E X± whenever u, v E X± in the commutative case is proved
by Lemma 14.2. It is so, because in that case logarithms are of the exponential
type. Without this property, we cannot prove a lemma corresponding to Lemma
14.2.
We should point out also that here, in the noncommutative algebras, we shall
need essentially the property that D and S commute each with another. Thus
Condition (A) is not a particular case of Conditions (A)l and (A)r. 0
Riemann-Hilbert problems 207
Theorem 14.4. Suppose that either Condition (A)I or Condition (A)r holds and
D E R(X). Then the Riemann-Hilbert problem (14.9) has a solution if and only
if there a U such that
(14.11)
(14.13)
then the Riemann-Hilbert problem (14.9) has a solution of the form (14.11), where
Proof. Let Condition (A)l be satisfied and let a E leX) n dom 0 1• Let (14.16)
hold, i.e. aX- a-I C X-. Let v- = au- a-I. Then (14.10) can be rewritten
in the following manner: u+ - v- = u+ - au-a- 1 = (Da)a- 1 = DLla. By our
assumptions,
u+ = P+DL/a = DP+Lla = DL/P+a;
v- = -P- DLa = -DP- L/a = -DL/P-a.
Thus au-a- 1 = v- = -DP- L,a = -DL/P-a, so that u- = -a-1(DL/P-a)a
and u = u+ + u- = DL1P+a - a-1(DL/P-a)a. A similar proof if Condition
(A)r is satisfied, a E leX) n dom Or and the second of Formulae (14.17) holds.
Riemann-Hilbert problems 209
u = DLIP+a + a-I[DLr(P-a)-lja
(u = a(DLIP+a)a- 1 + DLr(p-a)-I, respectively).
Theorem 14.6. Suppose that all assumptions of Theorem 14.5 are satisfied,
a:l ±e and Xo is a solution of the homogeneous problem (14.9). Then the nonho-
mogeneous Riemann-Hilbert problem (14.18) has a solution of the form
(14.19)
Example 14.2. Suppose that either Condition (A) or Condition (A}r or Condi-
tion (A}l is satisfied. Consider the equation:
(14.22)
(14.24)
·
Proo.f B your assump t lOns, Xo+ = axo-,+
,xo = ax o,- . L et y ± = L' Xo± an d Iet
y = y+ + y-. In the same manner as in the proof of Theorem 2.2, we find
Riemann-Hilbert problems 211
o
Sy = y+ - y- = L'xt - L'x = L'a. Thus y = SL'a. Clearly, by definition,
v E ker Dm. Since X is a Leibniz algebra, we get
~~± = E'y±
= E'(P±SL'a) = E'(±P±L'a) = E'[±P±(La+ v)]
= E'(±P±La± P±v) = E'(±p±v)E(±P±La)= E'(±v±)x~.
x~ = E'(v+)xt + E'(v-)xo
=E'(v+)~(xo + SXo) + E'(v-)~(xo - SXo)
=w+ (xo + Sxo) + w- (xo - SXo)
1
= "2[(w + Sw)(xo + SXo) + (w - Sw)(xo - Sxo)]
=wXo + (Sw)Sxo.
This proves Formula (14.23). We find
Then
m-l
(14.26) E'(v±) = II E'(vj).
j=O
If (L, E) is m-normalized by R then Yj = Xo (j = 0,1, ... , m - 1). In particular, if
F La = 0 then Yo = Xo. If (L, E) is not p-normalized for any p :s: m, then solutions
Yo, ... , Ym-l are linearly independent.
Proof By our assumptions, F Dj La = 0 for j = 0,1, ... , m - 1. Thus
Yj± -- E'(Vj±) Xo± -- E'(P±FDjLa ) Xo± -- E'(O) Xo± -- exo± -- Xo± , i.e. Yj = Xo·
In order to prove Formula (14.26), observe that X is a Leibniz algebra. Thus
E'(v±) = E'(P±v)
m-l m-l
= E'( - p± L RjFDjLa) = E'(p± L Vj)
j=o j=o
m-l m-l
LjXj - LiXi
= Lj[(a + e)Ej ( -P- Lja)]- Ld(a + e)Ei( -P- Lia)]
= Lj(a + e) - Li(a + e) - P-(Lja - Lia)] = aij'
Riemann-Hilbert problems 213
(14.29) aji = -aij; Aij E leX) and Aji = AijI (j # i; i, j = 0,1, ... , n)
(14.30) aij =aOj -aOi; Aij = AOjAo/ (j 'Ii; i,j = 1, ... ,n)
m-I m-I
(14.31) L ai,HI = aOm ; II Ai,i+l = Aom if 1 ::; m ::; n ;
i=O i=O
Prool Let i,j (j # i; i,j = O,I, ... ,n) be fixed. The first of equalities (14.29),
(14.30) follow immediately from the definitions (14.27), (14.28). Then Aji =
Ei(aji) = E i ( -aij) = [E(aij]-I = AijI and
Formulae (14.27) and (14.30) together imply (by an easy induction) Formulae
(14.31). •
Corollary 14.12. Suppose that X E Lg(D), D E R(X), F is an initial operator
for D corresponding to an R E RD, (Lo,Eo) E GR,I[O], (Li,Ei ) E G[O] (i =
1, ... , n) (n EN), a + e E leX) and Xi are solutions of the homogeneous Riemann-
Hilbert problem (14.9) induced by (Li,Ei ) (i = 0,1, ... ,n), respectively. Then
(14.32)
definition, AOj = Eo(aoj), Since aOj E ker D, (Lo, Eo) E GR,dO], and Lja - Loa
is a constant, we find
aOj = Faoj
= F[Lj(a + e) - Lo(a + e) - P-(Lja - Loa)]
= FLj(a + e) - P-(Lja - Loa).
Since Loxj - Ljxj is a constant, say Lozil it follows that
(14.33) (He)(t)
1
= 211" 10
r21T s- t
cot -2- x (s)ds,
where the singular integral He is understood as the Cauchy principal value (cf.
Example 14.1.). It is known that 1+ H; = K on HIL[O, 211"], where
10r
21T
1
(Kx)(t) = 211" x(s)ds, and dim K = 1.
.
(Hu)(e tt ) = -27r1 1211" . S -
u(e tS ) cot --ds
t
0 2
H=
- {OiHc for t = 1;
otherwise;
PERIODIC PROBLEMS
(15.1)
such that
(15.4) XSN = X(l) ffi '" ffi X(N) , where X(j) = PjXSN (j = 1, ... , N)
(15.6)
216
D. Przeworska-Rolewicz, Logarithms and Antilogarithms
© Kluwer Academic Publishers 1998
Periodic problems 217
and let E(j) =PjE w =PjENr. where Pj are determined by Formula (15.6). If
U = E( - ~:i jg)v, where vEEr, i.e. S-rv = v (j = 1, ... , N), then U E E(j), i.e. if
S-r u = €Ju.
Proof. Let j (j = 1, ... , N) be fixed. By our assumptions and Proposition 15.1,
since W = Nr, we get
S_ru = S-r[E( -.\jg)v)
= [S_rE( - 27ri jg)J(S-r V )
W
-_ e21rijr/wE( - -27ri.)
Jg v
W
27ri.)
_ 21rijr/Nr E( --Jgv
-e
W
=e21rij/N u = €ju.
In order to prove the necessity of the condition given in Theorem 15.1 we need the
•
assumption that X is a Leibniz algebra. Namely, we have
Theorem 15.2 Suppose that X E L(D) with unit e is a complete linear metric
space over C, (L, E) E c[n], D E R(X) is closed, {ShhEIR is a family of true
shifts induced by an R E RD n AQN(ker D), S_r is multiplicative for an r E JR,
w = Nr (N EN), 9 = Re and 2:ig E dom n- 1 whenever 2:i E vcR. Let
E(j) = PjE w = PjENr' where Pj and ENr are determined by Formulae (15.6)
and (15.7), respectively. Then u = E(_21rijg)v, w,
where vEEr, i.e. S-rv = v
(j = 1, ... , N), only if u E E(j), i.e. S_ru = €Ju.
Proof Since X is Leibniz algebra, we have (Ex)-1 = E( -x) whenever x E
dom n- 1 . Suppose that u E E(j), j = 1, ... ,N is fixed. Then S-ru = €ju. Write
v = Ee: i jg)u. By Proposition 15.1,
Note 15.1. In the classical case of the space X = C(~) over C, when D = ft,
(ShX)(t) = x(t + h) for x E X, t,h E JR, we find that Ew , w = Nr, is the space of
of w-periodic functions. The operator S-r is an involution of order N on Ew for
(S~rx)(t) = x(t - Nr) = x(t - w) = x(t). Thus any w-periodic function x E X is
of the form
L
N
X= e21rijt/wVj where VI, ... , VN are r-periodic functions
j=l
(cf. PRR[2], also PR[7]). Functions of the form eAtv, where>. E C, V is a periodic
function, and their linear combinations are said to be exponential-periodic func-
tions. Elements of the form E(>.g)v, where>. E C, V is a periodic element, and
their linear combinations are said to be exponential-periodic elements. Recall that
in Corollary 15.1 X is a Leibniz algebra, hence antilogarithms are exponentials.
For an arbitrary MEN and a true shift Sh we denote by
XEP(h; >'1, ... , >'M)
= lin {E(>'jg)Vj : Vj E Xs h ; >'j E C; >'m ::f >'j + 2~k, m ::f j;
k EZ (m,j = 1, ... ,M)}
a space of exponential-periodic elements. By Corollary 15.1, x E Ew if and
only if x E XEp(h; - 2;i, ... , - 21r~N) (Sh = S-r). More details about spaces
of exponential-periodic functions and their applications can be found in PR[6],
PR[7]. 0
Theorem 15.1 and Corollary 15.1 will be used in order to generalize the Floquet
theorem. The classical Floquet theorem says that every linear ordinary differential
equation with periodic coefficients has at least one non-zero exponential-periodic
solution, i.e. a solution of the form eAtu(t), where u ::f 0 is a periodic function and
>. E C is properly chosen (cf. ARscoTT Ar[l], INCE In[l]).
Lemma 15.1. Suppose that X is a linear space over IF, D E R(X), ker D ::f {O},
F is an initial operator for D corresponding to an REnD and {ShhElR is a family
of R-shifts, i.e. a family of linear operators such that
k hk - j
(15.8) So = I, ShRkF = L, (k _ .)!RjF for kENo·
)=0 J
Periodic problems 219
(15.9)
(15.10)
(15.11) D hR oh = I on E h, ROD
h h= I h ,z.e. D-
on E(l). h 1 = ROh'
Thus we get
R~DhX
1 1
= (I + Fh + hRFh)R(D - hFh)x
1 1
= (I + Fh + hRFh)R(Dx - hFhX)
1 1
= (I + Fh + hRFh)(RDx - hRFhX)
1 1
= (I + FS-h + hRFh)(x - Fx - hRFhX)
1 1
= x - Fx - hRFhX + FS-hX - FS_hFx - hFS-hRFhX+
1 1 1
+ hRFS-hFx - hRFS-hFhX - h2RFS-hRFhX
1 2 1
= x - Fx - -RFhX + Fx - F x - -F(RFx - hFx)
h h
1 2121 12
=x-Fx--FRFx+F x--RF x--RfRFx+-RF x
h h h2 h
1 1
= x - Fx + Fx - h + h RFx = x.
1
Dy = D(R + FhR + hRFhR)X
1
=DRx + DFS_hRx + hDRFShRx
1
=x + hFS-hRx.
Moreover,
S-hY
1
=S-h(I + Fh + hRFh)Rx
1
=S_hRx + S_hFS_hRx + hS-hRFS-hRx
1
= S_hRx + FS_hRx + h(RFS-hRx - hFS_hRx) =
1
= S-hRx + FS_hRx + hRFS-hRx - FS_hRx
1
=S-hRx + hRFS-hRx.
Periodic problems 221
Hence
u = Y - S-hY =
1
o - (S_h X - TtRFS-hRx
Rhx )
1 1
= Rx + FS_hRx + TtRx + TtRFS-hRx - S_hRx
1
+ TtRFS-hRx
= Rx - S-hRx + FS_hRx,
Du+D(y-S_hY) = DRx-DS_hRx+DFS_hRx = x = S_hDRx = X-S_hX = 0,
which implies u E ker D. On the other hand,
Definition 15.2. Suppose that all conditions of Definition 15.1 are satisfied. Let
mEN. An operator A E Lo(X) is said to be Sm-periodic if smA = Asm. 0
Clearly, if D E A(X), Ax = ax for an a E X and S is multiplicative, then
sm(Ax) = Sm(ax) = (Sma)Smx. Thus A is Sm-periodic if and only if a E XSTn
(m EN).
Lemma 15.2. Suppose that all assumptions of Theorem 15.2 are satisfied. Write
K
(15.12) Q(D) =L Qk Dk , Qk E Lo(X), S-rQk = QkS-r (k = 0, 1, .. , K),
k=O
i.e. Qk are S_r-periodic. Suppose, moreover, that v E dom DK is an S_r-periodic
element and u = E(Ag)V E kerQ(D). Then
K K
= L QkDkS':.ru = L QkS':.r Dku
k=O k=O
K K
=L S':.rQkDku = S':.r L Dku = S':.rQ(D)u = 0.
k=O k=O
222 Chapter 15
o= S~rQ(D)u
= Q(D)S~ru = Q(D)S~r[E(>'g)vl
= Q(D)[S~rE(>'g)lS~rv = Q(D)[e-AnrE(>.g)v
= e-AnrQ(D)[E(>.g)v] = e-AnrQ(D)u.
Thus for n =N we find eW =e- ArN = 1, i.e. >. is of the required form. •
Lemma 15.3. Suppose that all assumptions of Lemma 15.2 are satisfied. Let
>. = 2:i
j, where j E Z is arbitrary. Then Q(D + AI)v = O. Conversely, if
Q(D + >.I)v = 0, then u = E(>.g)v E ker Q(D).
Proof Necessity. Since DE(>.g) = >'E(>.g), we have
[Q(D + >.I)v]E(>.g)
K
= L Qkf(D + AI)kv]E(>.g)
k=O
K
= L [E(>.g)v]
k=O
K
=L QkDku = Q(D)u = O.
k=O
(15.13)
K
(15.14) -
Q(t, s) = '~
"' -
Qkt k K-k
s , -
Q(t) = Q(t,
-
1).
k=O
Periodic problems 223
If the operator Q(I, R~) is invertible in the space Er then the equation
(15.15) Q(D)x =0
has exponential-periodic solutions which are of the form
h S-rV = \ 27ri.
(15.16) x = E()..g)v, were V, /\ = -J (j E Z),
w
K k-1
(15.17) V = (R~)K[Q(I, R~)r1 L Qk L r-k+iz·
J'
k=O j=O
Q(D + M)
m=O
K
= L QkDk = Q(D).
k=O
Observe that Qo, ... , QK, as linear combinations of S_r-periodic operators, are
again S_r-periodic. We can rewrite Equation (15.18) in the following way:
(15.19) Q(D)v = O.
Since the operator Q(I, R~) is invertible in the space Er , we conclude that Equation
(15.19) has S_r-periodic solutions of the form (15.17). •
In order to reduce an equation with an involution of order N, in particular, an
equation with a true shift in the space of periodic elements, we need
224 Chapter 15
Lemma 15.4. (cf. PR[7]) Suppose that X is a linear space over C, DE R(X),
S E Lo(X) commutes with D: SD = DS on dom D, the operators Qkm E Lo(X)
are S-periodic, i.e. SQkm = QkmS (k = 0,1, ... , N - 1; N ~ 2; m = 0,1, ... , M).
Write
N-l M
Qm(S) = L Qkm Sk , Q(D,S) = L Dm+Ml (Ml E No))·
k=O m=O
j=1
and projectors PI, .... , PN are defined by (15.6).
Proof. Since the operators D and Pj (j = 1, ... , N) commute, the decomposition
(15.4) implies that on XSN for j = 1,2, ... , N
Q(D,S)Pj
M M
= L Qm(S)Dm+Ml Pj =L Q(S)PjDm+Ml
m=O m=O
M M
=L (L L (L
~ ~
QkmSkpj)Dm+Ml = Qkmcjkpj)Dm+Ml
m=O k=O m=O k=O
M M
= L Q(cj)PjDm+Ml =L Q(cj)Dm+MIPj = Q(D,cj)Pj .
m=O m=O
Since
N
L Pj =1 on XSN,
j=1
we conclude that on XSN
N N N
Q(D,S) = Q(D,S) L Pj =L Q(D,S)Pj =L Q(D,cj)Pj .
j=1 j=1 j=1
Proposition 15.2. (cf. PR[7]) Suppose that all assumptions of Lemma 15.4 are
•
satisfied. Then the equation
and XU) are defined by the decomposition (15.4), Xj,Yj are defined by Formulae
(15.5).
Proof The definitions of the operators Pj and elements xi> Yj (j = 1, ... , N)
together imply that x = Xl + ... +XN, Y = Y1 + .. ·+YN and that this decomposition is
uniquely determined. Since the operators Qkm are S-periodic, we have SQkmXj =
QkmSXj = c:jQkmXj for j = 1, ... , N, k = 0,1, ... , N - 1, m = 0,1, ... , M. This
implies that the each operator Q(D, c: j ) preserves the space Xj (j = 1, ... , N).
Indeed, if Xj E Xj (j = 1, ... , N) then
M N-1 M N-1
Q(D,c:j)xj =L L QkmDm+M,pjx = Pj L L QkmDm+M,X E Xj.
m=O k=O m=O k=O
Q(D, S)x
N N
= Q(D, S) L PjX = L Q(D, S)Pj
j=l j=l
N N
= L Q(D,c:j)Pjx =L Q(D,c:j)xj.
j=l j=l
Since
N N
y= L PjY= L Yj,
j=l j=l
the space XSN is the direct sum of spaces X(j) and Xj,Yj E X(j) (j = 1, ... ,N),
we conclude that Equation (15.21) is equivalent to the system (15.22) of N inde-
pendent equations. •
Theorem 15.4. (cf. PR[7]) Suppose that all assumptions of Lemma 15.4 are
satisfied. If each of Equations (15.22) has a solution Xj E Xj (j = 1, ... , N) then
Equation (15.21) has a solution x = Xl + .... + XN E XSN. Conversely, if Equation
(15.21) has an SN -periodic solution X then the jth Equation (15.22) has a solution
Xj = PjX E Xj.
Proof If Equation (15.21) has an SN-periodic solution then, by definition, X E
XSN. Hence SN x = x. Thus x = Xl + ... +XN, where Xj = PjX E Xj (j = 1, ... , N).
Proposition 15.2 implies that Equation (15.21) is equivalent to the system (15.22).
Thus the jth equation (15.22) has a solution Xj E Xj. Conversely, suppose that
the jth equation (15.22) has a solution Xj E Xj (j = 1, ... , N). Since PJ = Pj ,
Pj ::j:. Pk for j ::j:. k and Xj = PjX for an X E X(j, k = 1, ... , N) (cf. PR[6]), writing
x = Xl + ... + X N we find
N N
PjX = Pj L Xk =L PjPk X = PjX = Xj (j = 1, ... , N).
k=l k=l
226 Chapter 15
Theorem 15.4 has several applications. We may use this theorem when S = Sh is a
true shift (under appropriate additional assumptions). Some of these applications
are quite far from classical differential-difference equations (cf. PR[6], PR[7]).
For instance, this method can be used for a reduction of a stochastic differential-
difference equation to a stochastic differential equation in order to find its periodic
solutions (cf. WILKOWSKI Wi[l]). In a similar manner one can consider the
operator
M
Q(D,S) = DMl 2: Qm(S)Dm.
m=O
(15.23)
(15.24) - = -
D Xj 27rij
-Xj
- + G-j (-Xl, ... ,XN
-) (j = 1, ... ,N),
W
N 2 ..
(15.26) X = '"' 7rZJ_
~ E(-g)xj,
j=O W
where (Xl, ... , XN) is a h-periodic solution of the system (15.24), where h = -r.
Periodic problems 227
Proof. Acting on both sides of Equation (15.23) by projectors PI, ... , PN and
applying Theorem 15.2 we can rewrite this equation as a system of N equations
without shifts. Since, by Theorem 15.2, Xl, ... , XN are h-periodic functions, where
h = -r, for v = 1, ... , N we have
Theorem 15.5 can be use also in order to study linear equations with shifts, in
•
particular, differential-difference equations with periodic coefficients. In a similar
manner one can consider other periodic problems (cf. PRR[2], also PR[6], Section
V.5). Note that in Theorem 15.5 we obtain a system of equations which, in general,
are not independent.
Consider again the space of exponential-periodic elements
(15.27)
N
pet) = II (t - tj), where tj = eA;h (j = 1, ... , N)
j=l
Proof. Suppose that x = l:f=o E(Ajg)Vj E XEP(h; Aj, ... , AN) is arbitrarily fixed.
Recall that Vj E Er (j = 0, ... , N). By our assumptions,
N N
=L Sh[E(Ajg)Vj] =L [ShE(Ajg)](ShVj)
j=l j=l
N N
=L eA;hE(Ajg)Vj =L tjE(Ajg)Vj.
j=l ;=1
228 Chapter 15
N N
S~x = S~ 2: E(Ajg)Vj = 2: tJE(Ajg)Vj (k EN).
j=1 j=1
N N N
= L PkS~X =L Pk L tJE(Ajg)Vj
j=1 k=O j=1
N N N
= 2: (2: PktJ)E(Ajg)Vj = 2: P(tj)E(Ajg)Vj = O.
j=1 k=O j=1
The arbitrariness of x implies that P(Sh) = 0 on XEP(h; AI, ... , AN). The roots
of the polynomial P(t) are single because, by our assumptions, Am :j:. Aj + 2""ik
whenever m :j:. j, k E Z, which implies that tm :j:. tj whenever m :j:. j (j,m =
1, ... ,N). •
An immediate consequence of Theorem 15.6 is
Corollary 15.2. Suppose that all assumptions of Theorem 15.6 are satisfied.
Then X = XEP(h; AI, ... , AN) is a direct sum of the eigenspaces of the operator
Sh corresponding to the eigenvalues tl, ... , tN:
(15.29)
(15.30)
Periodic problems 229
(cf. Lemma 15.1) and Formula (15.8). Then for all h E JR, z E ker D, k, n E N, we
have
k
(15.31) ShnRkz =~
L..J h(n)
k,m Rmz ,were
h f or J. = 0, 1, ... , k
m=O
k
(15.33) hk(n:t- l )
,1
=~
L....J
hen) h(l).
k,m m,j'
hk,k
(nH) -- 1 'f n >
t _ 1.
m=l
Proof, by induction. Let h E JR, z E ker D and kEN be arbitrarily fixed. For
n = 1 (15.31) follows just from the definition (15.30). Suppose then (15.31) holds
for an arbitrarily fixed n E No By (15.32) and (15.33), we get
m=O m=O
M
(15.35)
(0) (nH) _ ~ (n) hen) (nH) - 1
PM,m =Pm, PM,j - L..J PM,m m,j' PM,M - ,
m=j
where h~!j are defined by Formulae (15.32), (15.33). Then pt~M = 1 for MEN,
n E No and
Proof, by induction. Let hER and z E ker D be arbitrarily fixed. Lemma 15.5
and Formulae (15.35) together imply that
M M M M
SI: P(R)z =L PM,jRj z =L [L Pmh~!jlRj z =L p<f),jRj z = p(l) (R)z.
j=O j=O m=j j=O
Suppose Formula (15.36) to be true for an arbitrarily fixed n > 1. By our assump-
tions and Lemma 15.5,
SI:P(R)z
M M
= SI: L Pm Rmz =L PmSI:Rmz
m=O m=O
M M M M
where
M
PM,j =L Pmh~!j (j = 0,1, ... , M)
m=j
and PM,M = PMht~M = PM = 1. This, and Formulae (15.35) together imply that
j=O m=j
M
= L pt~2) Rjz = P(n+2)(R)z.
j=O
Proposition 15.4. Suppose that all assumptions of Lemma 15.1 are satisfied.
Then for every h E ~ and kENo the operator Sh is algebraic on the space
ker Dk+l with the characteristic polynomial P(t) = (t - l)k+l, i.e.
Proof. Let h E ~ and kEN be arbitrarily fixed. Since Sh is an R-shift, for all
z E ker D we have Sz = z, i.e. (Sh - I)z = o. Indeed, ShZ = ShRoz = ROz = z.
Clearly, ker Dk+l = lin {Rkz : kENo}. Let p(O)(t) = t-1, i.e. M = 1, Pl,O = -1,
Pl,l = 1. By Lemma 15.5,
m=O m=O
k-l k-l
= (Sh - I) L h1~~Rmz = L h1~~(Sh - I)Rmz
m=O m=O
(k-2) /l-
- "h(l) "h(l) RV
- L..t k,/l-+l L..t 1'+1 ,v z
1'=0 v=o
k-2 k-2
= "["
L..t L..t h(l)
JL+l,v ]RVz .
k,JL+l h(l)
Writing
we get
(15.37)
v=O
k-j
(Sh - I)j = "L..t h,u-l)
k,v R z and
V
(Sh - I)k+l Rk z = (Sh - I) [(Sh - I)k Rk z] = (Sh - I)h~~;;l) z = h1~;;1) (Sh - I)z = o.
•
232 Chapter 15
Corollary 15.3. Let D E R(X), ker D ::j:. {o} and let F be an initial operator
for D corresponding to an R E RD. Let X E A(D). Let {ShhEIR C Lo(X) be a
family of multiplicative R-shifts (cf. Lemma 15.5). Then
with the characteristic polynomial P(t) = (t N _1)k+l and with the characteristic
roots c: j = e21rij/N (j = 0,1, ... , N), each of multiplicity k + 1.
Proof. Let k, N E N, z E ker D, h E JR, v E XSN be arbitrarily fixed. Then
h
Sf:(vRkz) = (Sf:v)(Sf:Rk z ) = vSf:Rkz. Thus (Sf: - I)(vRkz) = vSf:Rkz-
vRk z = v(sf: - I)Rkz. By an easy induction we prove that (sf: - I)m(vRkz) =
v(Sf: - I)m Rk z for mEN. This, and Proposition 15.4 together imply that
(15.38)
M
= {u = L vmRmzm: Zm E ker D, S~mvm = Vm (m = 0,1, ... , M)}.
m=O
Then
m=O
(tnm - I)M+1,
i.e. Sh is an algebraic operator on the space Xpp(h; no, ... , nM) with the character-
isticpolynomialP(t). The characteristic roots arec:t,. = e~ (m = 0,1, ... ,M; j =
O,I, ... ,nm - 1), each of the multiplicity M + 1. Thus Xpp(hjno, ... ,nM) is
a direct sum of principal spaces X jm such that (Sh - c:t,.I)M+1 = on X jm
(j=O,I, ... ,n m -l; m=O,I, ... ,M).
°
Elements of spaces Xpp(h; no, ... , nM) are said to be D-polynomial-periodic ele-
ments.
Periodic problems 233
Theorem 15.7. Suppose that X E L(D) has the unit e and is a complete linear
metric space over C, (L, E) E G[O], D E R(X) is closed, {ShhER is a family
of multiplicative true shifts induced by an R E RD n AQN(ker D), 9 = Re,
M, K j , N j E N and Ajg E dom 0- 1 whenever Aj E vcR (Aj 1= Ak + 27ril/h if
j 1= k; I E Z; j, k = 0,1, ... , M). Then Sh is an algebraic operator on the space of
D-polynomial-exponential-periodic elements:
M Kj
= {x = L (L VjkRkZjk)E(Ajg): Zjk E kerD, Vjk E X s :/' k = O,I, ... ,Kj}
j=O k=O
with the characteristic polynomial
II (t Nj -
M
(15.41) P(t) = t JNj )K j+l, h
were tj = e Ajh (.J = 0 , 1, ... , M) .
j=O
sf: (VRK z)
= sf: w = (sf: v)[sf: E(Ag)]sf: RK z
= ve ANh E(Ag)Sf: RK Z = eANhvE(Ag)Sf: RK Z
= eANhvSf: RK z.
Thus
(sf: - e ANh I)w = eANhvE(Ag)(Sf: - I}RK z,
and, by Proposition 15.4, we get
(15.43)
Suppose now that u is an arbitrary element of XPEP(h; Aj; Kj; Nj; M), i.e. u is
of the form
M Kj
U = L (L
VjkRkZjk)E()..jg),
j=O k=O
234 Chapter 15
where h, M, Aj, K j , N j , Vjk, Zjk are described in Formula (15.40). The assumption
that Aj :f:. Ak + 27f'il / h whenever j :f:. k, l E Z, asserts us that antilogarithms
E(Ajg) are pairwise different and linearly independent. It is so because, by our
=
assumption that Aj E vcR (j 0,1, ... , M), we have E(Ajg) =
e)..;zj, Zj E ker D
and exponentials e)..;zj are linearly independent (cf. PR[8)). Let pet) be defined
by Formula (15.41). Formula (15.43) implies that
M M K,
= [II (8:; - tf;)K;+l] L (L vlk Rkzlk)E(A19)
j=O 1=0 k=O
M K, M
=L {L [II (8:; - tf;)K;+l]vlkRkzlk}E(Alg) = O.
1=0 k=O j=O
Corollary 15.5. Suppose that all assumptions of Theorem 15.7 are satisfied.
•
Then a principal space corresponding to the root 1 of the multiplicity k + 1 is
lin {Rjz: Z E kerD, j = O,I, ... ,k}.
We can solve now equations with shifts in the spaces Xpp(h; no, ... , nM) and
XPEP(h; Aj, Kj; Nj; M) in the same manner as we did it in the space of periodic
elements and in XEP(h; Aj, ... , AM) (cf. for instance, Proposition 15.2, Theorem
15.4).
Note 15.2. Theorems 15.6, 15.7 and Corollary 15.2 are proved for Leibniz alge-
bras. A modified proof could be used for quasi Leibniz algebras, i.e. commutative
algebras with the product rule D(xy) = xDy+yDx+d(Dx)(Dy) for x, y E dom D,
where d:f:. 0 is a scalar, i.e. if X E AD(O,e,de) (cf. Example 5.1 and Corollary
5.4). We should only to remember that in this case logarithms (provided that they
exist) are not of of the exponential type, but satisfy the following equation
Thus
WeD) =L LakjDkS_hj,
k=O j=O
By (13.19, E(>..g) E ker WeD) if and only if W"(>..) = o. It means that in order to
determine solutions of Equation (13.20)
W(D)x = 0,
which are of the form E(>..g), it is enough to find characteristic roots, i.e. zeros of
the quasipolynomial W"(t). According to Definition 9.2, assume that Condition
[eh is satisfied, i.e. -x E dom 0- 1 whenever x E dom 0- 1 and the trigonometric
identity holds. Let w = u + iv E dom 0, w* = u - iv and let
then
N
(16.2) sna(S) = L c:jn[sna(c:j)]Pj for all n E N,
j=1
where PI, ... , PN are disjoint projectors giving the partition of unity and such that
SPj = PjS = c:j Pj (j = 1, ... , N) (cf. Definition 15.1).
Proof. Let S be a multiplicative involution of order N 2': 2. By an easy induction
we prove that
snpj = c:jnPj (j = 1, ... ,N) and sn E ML(X) for all n E N.
By Formula (15.3),
N
a(S) = a(S) L Pj
j=1
N N N-I
= L a(S)Pj =L L akS k Pj
j=1 j=1 k=O
N N-I N
=LL akc:jkPj =L
j=1 k=O j=1
Hence
N
(16.3) a(S) = L a(c:i)pj .
j=1
236
D. Przeworska-Rolewicz, Logarithms and Antilogarithms
© Kluwer Academic Publishers 1998
Multiplicative involutions 237
N N
= sn L a(ei)Pi =L sn[a(ei)Pi] =
i=1 j=1
N N
=L [sna(ej))snpi = L ein[sna(ein))Pj.
j=1 j=1
Proposition 16.2. Suppose that all assumptions of Proposition 16.1 are satisfied
•
and X has the unit e E X N = PN X. Then Se = e, Sx E J(X) whenever x E J(X)
and (SX)-l = Sx- 1.
Proof By definition, Se = eN e = e21ri e = e. Suppose that x E J(X). Then
(Sx)(Sx- 1) = S(xx- 1) = Se = e. Hence Sx E J(X) and (Sx)-l = Sx- 1. •
Theorem 16.1. Suppose that X is a commutative algebra over C, S E M L(X)
is an involution of order N 2:: 2, e = e21ri / N and a(t) is defined by (16.1). Then
the equation
(16.4) a(S)x = y, y E X,
(16.6)
a(e)
eSa(e)
=
Proof Acting on both sides of Equation (16.4) by the operators S, ... ,SN-1 and
applying Proposition 16.1, we get the system (16.5). Clearly, the determinant of
that system is of the form (16.6). •
Note 16.1. Suppose that all assumptions of Theorem 16.1 are satisfied. If there
is an index j (1 ~ j ~ N) such that a(e i ) = 0, then ~(a) = O. In other words, if
~(a) ¥- 0 then a(ei ) ¥- 0 for j = 1, ... , N. 0
An immediate consequence of Theorem 16.1 is
238 Chapter 16
Corollary 16.1. Suppose that all assumptions of Theorem 16.1 are satisfied. If
~(a)E leX) then Equation (16.4) has a unique solution of the form
N
(16.7) X= L Xj, where
j=l
N N-l
Xj = [~(a)tl L L (_l)i+k+l~jk(a)Sky (j = 1, ... ,N)
j=l k=O
and by ~jk(a) is denoted the minor determinant obtained by cancelling the jth
row and the (k+1)th column in ~(a) (j = 1, ... , N; k = 0, ... , N - 1).
Corollary 16.2. Suppose that all assumptions of Theorem 16.1 are satisfied. If
a(c i ) = 0 for j :j:. m and a(c m ) E leX) (j, m = 1, ... , N) then Equation (16.4) is
reduced to the equation
(16.8)
A necessary and sufficient condition for Equation (16.8) to have a solution is that
(16.9)
N
(16.10) X = [a(cm)tly + L Xi, where Xi are arbitrary.
i=l,i#m
N
(16.12) ~(a) = VN(c) IT a(c j ),
j=l
Multiplicative involutions 239
hence Equation (16.4) has a unique solution if and only if a(c: i ) 1:- 0 for j = 1, ... , N.
This solution is of the form (16.7).
Corollary 16.4. Suppose that all assumptions of Theorem 16.1 are satisfied and
Sa(c: i ) - a(c:i)S = 0 (j = 1, ... , N). Then ~(a) is of the form (16.12).
Proof By our assumptions,
~(a)
Proof. All assumptions of Corollary 16.4 are satisfied in the quotient algebra XI J .
Corollary 16.6. Suppose that all assumptions of Theorem 16.1 are satisfied and
•
a(c: j ) E Xi (j = 1, ... , N). Then Equation (16.4) is equivalent to the system
We therefore have obtained the system (16.13). If a(c: j ) E I(X) for j = 1, ... , N,
then we find Xj = PjX = [a(c: j )t 1Yi> which yields Formula (16.14). •
Note 16.2. The condition a(c: j ) E I(X) (j = 1, ... , N) is the classical one for
equations with involutions of order N (without the assumption that they are mul-
tiplicative). The same property have solutions given by Formula (16.14). 0
We shall consider now algebras belonging to Lg(D) and multiplicative operators
commuting with D.
240 Chapter 16
(16.16)
N-l
(16.17) A(S) =L akESk, where aO, .. ·,aN-l EX.
k=O
Let a(S) and ~(a) be defined by Formulae (16.1), (16.6), respectively. Let
~jk(a) be the minor determinant obtained by cancelling the jth row and the
(k+1)th column in ~(a) (j = 1, ... , N; k = 0, ... , N - 1). If ~(a) E I(X) and
[~(a)J-l ~jk(a)Sky E dom 0 (j = 1, ... , N; k = 0, ... , N - 1), then the equation
(16.18) A(S)x = y, Y E X,
Multiplicative involutions 241
N-1 N-1
A(S) =L akES k =L akS kE = a(S)E.
k=O k=O
By Corollary 16.1,
N N-1
Ex = [~(a)l-1 L L (_1)i+k+1 ~jk(a)Sky,
j=1 k=O
which implies (16.19), since x = LEx.
Example 16.1. Suppose that all assumptions of Proposition 16.3 are satisfied
•
and X is a Leibniz algebra with unit e. Then
N
SEx = [E(II ej Pjx)]1 IN.
j=1
Since X is a Leibniz algebra, by Formulae (16.3), (16.15) and Proposition 7.7(iii),
we get
N N
SEx = ESx = E(~ L ejpjx) = [E( II ejpjx)] liN.
j=l j=l
Let N = 2, i.e. S2 =I and e = -1. Then S = p+ - P- , I = p+ + P-. Hence
which implies
o
Example 16.2. Suppose that all assumptions of Theorem 16.2 are satisfied and
N = 2. Then e = -1 and
~(a) = I-Sa(-I)
a(-I) a(1) I
Sa(l) = a(-I)Sa(l) + a(I)Sa(-I) E J(X).
(a) h(t) = -t on IR
It should be pointed out that in examples (a)-(e) there is only one fixed-point a
of the function h, i.e. a point a such that h(a) = a. Namely, a = 0 in Examples
(a) and (c), a = 1 in Examples (b) and (d), a = c/2 in Example (e).
The following function satisfies the Carleman condition of order m:
Lemma 16.3. (cf. PR[4)) Let h be a continuous Carleman function on IR. Then
•
(i) h is decreasing;
(ii) the derivative h' exists almost everywhere and h(t) = J: h'(s)ds;
(iii) h has a unique fixed-point ah;
(iv) if h is odd then h(t) = -to
Multiplicative involutions 245
Proof. (i) By Lemma 16.1, h is one-to-one, hence strictly monotone. Suppose that
h is increasing. Since, by definition, h(t) t
t, there is a to such that h(to) ::j:. to.
Then either h(to) > to or h(to) < to. The first inequality implies that ho =
h[h(to)) > h(to) > to, which contradicts our assumption. A similar proof the
second inequality. Hence h is decreasing.
(ii) Since h is continuous and monotone, its derivative exists almost everywhere
and h(t) = J: h'(s)ds, as it is well-known from Calculus of one real variable.
(iii) Our assumptions and (i) immediately imply that
(16.24) lim
t-++oo
h(t) = -00, lim
t-+-oo
h(t) = +00.
Since h is continuous and decreasing on ~ and Formulae (16.24) hold, there exists
a unique point ah such that h(ah) - ah = O.
(iv) Suppose now that h is odd, i.e. h( -t) = -h(t) for t E R Then h(O) = -h(O),
which implies h(O) = O. Thus 0 is the unique fixed-point of h. Suppose that
h(t) ::j:. -to If h(t) > -t then t = h[h(t)] < h( -t) = -h(t), since h is a decreasing
function. This implies that -t > h(t) > -t, a contradiction. For the inequality
h(t) < -t we get a contradiction in a similar manner. Hence h(t) = -to •
An immediate consequence of Lemmas 16.2 and 16.3 is
Lemma 16.4. (d. PR[4]) Every continuous Carleman function on ~ has the
graph symmetric with the respect to the line x = t.
In the sequel we will denote the superposition of two functions x and y by x 0 y,
i.e. by definition,
(x 0 y)(t) = x[y(t)],
provided that this superposition exist.
Theorem 16.5. (d. PR[4]) Every continuous Carleman function on ~ with
a fixed-point a is the superposition of a homeomorphism H of ~ onto itself, a
reflection and of the inverse homeomorphism H-I,
for h. •
Corollary 16.11. (cf. PR[4]) Every continuous Carleman function on an open
arc L is of the form: h(t) = H-l[-H(t)], where H = hI 0 h2 0 h3, h3 is a homeo-
morphism L onto JR, h2 (s) = s - a, a is the fixed-point of the function h3 0 h 0 h3 1 ,
and hI is a homeomorphism oflR onto itself such that hI (0) = O. Conversely, every
function of this form is a Carleman function on L with the fixed-point h3 l (a).
Proof Let h3 be a homeomorphism of the arc L onto JR and let s = h3(t). Then
the function
(16.27)
h(b)
= h3 1 0 h21 0 h I 1 { - h1 0 h2 0 h3[h3 1(a)]}
= hI1 0 h21 0 h3 1[-(h1 0 h2)(a)]
=h3 1 h21 h11[-h 1(0)] = (h3 1
0 0 0 h21 0 hll)(O)
=(h3 1 0 h21)(0) = h3 1(a) = b.
Hence b is the fixed-point of h.
•
Multiplicative involutions 247
11" b+a
x(t) = tan b _ a (t - -2-)
is a homeomorphism of the open interval (a, b) onto JR.. The function x(t) = In tis
a homeomorphism of the positive open half-axis onto JR.. The function x(t) = -t
is a homeomorphism of (-00,0) onto (0, +(0).
By definition, each open arc on a real or complex plane( as homeomorphic with an
open interval) is homeomorphic with JR.. 0
Theorem 16.6. (cf. PR[4]) Continuous functions satisfying Carleman condition
of order m > 2 on JR. do not exist.
Proof. Suppose that there exists a function h such that h(t) ~ t and
t < s {:} h(t) > h(s) {:} ... {:} hm-1(t) < hm-1(s) {:} t = hm(t) > hm(s) = s,
a contradiction. If h is increasing then there is a to such that h(to) > to and from
the identity (16.27) we find
again a contradiction. Thus m cannot be odd. Let m = 2k, k > 1. Since (16.28)
holds, hk is a Carleman function and therefore decreasing. •
Corollary 16.12. (cf. PR[4]) Continuous functions satisfying the Carleman
condition of order m > 2 on an open arc L do not exist.
Proof. Suppose that hm(t) == t on L for an m > 2. Let H be a homeomorphism
of L onto JR. of the type described in Corollary 16.12. We define on JR. the function
h* = H 0 h 0 H- 1 of s = H(t) E JR., tEL. Then
h; = h* 0 h* = (H 0 h 0 H- 1 ) 0 (H 0 h 0 H- 1) = H 0 h 0 h 0 H- 1 = H 0 h2 0 H- 1,
Proof. (i) Let H be a diffeomorphism of (a, b) onto R Since both derivatives H'
and (H-l)' exist and are continuous by assumption, it follows that h' also exists
and is continuous.
(ii) Since H' i- 0, we have
h'(t)
Example 16.7. Suppose that all assumptions of Example 16.5 are satisfied and
N = 2, i.e. e = -1. Then the operator D and S are anticommuting (cf. Example
16.6). Consider the equation
, _I
~(a) -
a(l)
-Sa(l) Sa(-I) 1-
a( -1) - a(I)Sa(-I) + a(-I)Sa(l) -
- ~(a),
o
Example 16.8. Let 0 C C be a domain invariant under the rotation through
the angle 2;/, i.e. eO = 0, where e = e27ri / N (N EN). Consider the algebra
It
X = C(O), the operator D = and the operator S of rotation through the angle
2;/: (Sx)(t) = X(et) for x E X. Clearly, S is a (multiplicative) permuting operator
for N 2: 2, since (DSx)(t) = [x(et)l' = eX'(et) = e(SDx)(t) (cf. Example 16.5). 0
Consider now equations of the Carleman type.
Theorem 16.7. Suppose that X E Lg(D) and S is a Carleman operator of
order N 2:: 2 on X, i.e. S E ML(X) is an involution of order N 2:: 2 on X and
DS = dSD on dom D, where d E leX) \ {e}. Let
N-I
a(S) = L akS k , where ao, ... , aN-l EX.
k=O
(16.30) Dx = a(S)x + y, y EX
N-I
(16.32) where aj(S) =L aj,kSk and
k=O
(dS)jak-o if k ~ N -1;
aj,k ={ j J
(dS) ak-j-N if k 2:: Nj
(j,k == O, ... ,N -1).
250 Chapter 16
Acting on both sides of Equation (16.30) by operators dS, ... ,(dS)N-l and using
Equalities (16.34) we obtain the system (16.31). Conversely, if we put in (16.31)
j = 0 then we get Equation (16.30). •
An immediate consequence of this Theorem is
Corollary 16.3. Suppose that all assumptions of Theorem 16.7 are satisfied and
aj,k (j, k = 0, ... , N - 1) are determined by Formulae (16.32). If the determinant
of the system (16.31) is invertible, then Equation (16.30) has a unique solution,
which is of the form
N-l
(16.36) X = [~(a)rl L (_1)k+1~k(a)(dS)ky,
k=O
o
Theorem 16.8. Suppose that X E Lg(D), S is a Carleman operator on X, i.e.
S E ML(X), S2 = I and DS = dSD on dom D, where d E I(X) \ {O}, and
there is an initial operator F corresponding to an R E RD such that FS = F. If
x E dom D2 is a solution of the equation
(16.37) Dx = Sx + y, y E dom D
Multiplicative involutions 251
(16.38)
D 2x = D(Dx)
= D(Sx + y) = DSx + Dy = dSDx + Dy = dS(Sx + y) + Dy
=dS 2x + dSy + Dy = dx + dSy + Dy,
i.e. x is a solution of Equation (16.38).
Sufficiency. Suppose that x is a solution of Equation (16.38) satisfying Condition
(16.39) and the operator I - R 2d is invertible. We write u = Dx - Sx - y and
we show that u = 0, i.e. Equation (16.37) is satisfied. Observe that u satisfies the
equation Du + dSu = 0 with the condition Fu = O. Indeed,
Du = D(Dx - Sx - y) =
D 2x - DSx - Dy = D 2x - dSDx - Dy
=dx + dSy - dSDx = dS(Sx +y - Dx) = -dSu,
but this fact was not used in an explicit way. Particular properties of functions
considered there assure that Equation (16.37) is equivalent to Equation (16.38)
together with Condition (16.39). 0
An immediate consequence of Theorem 16.8 is
Corollary 16.14. Suppose that all assumptions of Theorem 16.8 are satisfied,
(L 2, E 2) E G[02]' 1- R 2d is invertible, R 2d E dom 0- 1 and y = O. Then
M
(16.41) A(D) =L akDk, where ao, ... , aM E X,
k=O
M
(16.42) A(D) = L (Sak)(Dd- 1 )kd.
k=O
M
(16.46) A(I,R) = L RM-kak
k=O
Multiplicative involutions 253
(16.47)
Indeed, since S is multiplicative and Formulae (16.47) hold, for all x E dom DM
we have
dSa(D)x
M M
=dS L ak Dkx = d(Sao)Sx + L {Sak)dSDkx
k=O k=l
M
=d(Sao)Sx + L (Sak)(Dd- 1 )kdSx = A(D)dSx.
k=l
dx + dSy = dS(Sx + y) =
dSa(D)x = A(D)Sx = A(D)[A(D)x - y]
= A(D)dA(D)x + dx - A(D)y,
which implies Equation (16.44).
Sufficiency. Suppose that x satisfies Equation (16.44) and Conditions (16.45). By
(16.48), we can rewrite Equation (16.44) as follows:
M-l
A(I,R)DMv = A(D)v - 2: RkFDkv = A(D)v = 0.
k=O
M-l
V = 2: Rk F Dkv + RM DM V = 0.
k=O
Corollary 16.15. Suppose that all assumptions of Theorem 16.9 are satisfied and
•
Sak = ak (k = 0,1, ... , M). Then A(D) = A(Dd- 1).
Proof. By definitions (16.41), (16.42) and our assumption,
M M
A(D) =L (Sak)(Dd-1)k =L ak(Dd-1)k = A(Dd- 1).
k=O k=O
Corollary 16.16. Suppose that all assumptions of Theorem 16.9 are satisfied and
•
Sak = akS (k = 0,1, ... , M). Then SA (D) = A(D) and A(D) = dA(D)Sd- l .
Proof. By definitions (16.41, (16.42) and our assumption,
M M M M
SA(D) =S L ak Dk =L (Sak)SD k = L ak S2 Dk =L ak Dk = A(D).
k=O k=O k=O k=O
Multiplicative involutions 255
Using Theorem 16.9 (Theorem 16.8, respectively) and Theorem 16.3, we may solve
equations of the form
Write
M-l
XM =L Rkxk.
k=O
Clearly, XM E ker DM. Suppose that the operator A(I, R) is defined by Formula
(16.46). Formula (16.51) implies that
L
M-l
H(x) = A(D)x = A(I, R)DM x + A(D) Rk F Dkx = A(I, R)DM x + XM.
k=O
(16.54)
In this section we will show how one can introduce elements of the fractional
calculus (cf. for instance, KIRYAKOVA Kir[l]), by means of antilogarithms. We
begin with
Proposition 17.1. Suppose that X is a Leibniz D-algebra with unit e and F is
a multiplicative initial operator for D corresponding to an R E RD and 9 = Re.
Then
(17.1) D m 9n = (n _n!m)!g n-m f ar a II n, m E ~T
l~O, n ~ m,
Note 17.1. Suppose that all assumptions of Proposition 17.1 are satisfied and
•
9 E J(X). Then, by Proposition 7.9(iii), Fg- 1 does not exists, since F E M L(X).
o
Proposition 17.2. Suppose that IF = JR, X is a complete linear metric space,
X E L(D) and has the unit e E dom 0, (L, E) E G[OJ, F is a multiplicative initial
operator for D corresponding to an R E RD, 9 = Re E J(X) and aLg E dom 0- 1
for every a E R Then for all a E 1I4
(i) gO = Eog = E(aLg);
(ii) Dgo = ag o - 1;
(iii) 9 E £b(X) implies gO = eoLg ;
(iv) Dmgo = r(~(~t~~)go-m for all mEN, where r is the Euler function;
(v) g-O = Eog = E( -aLg) = [E(ag)]-1 = (gO)-1.
Proof. Recall again that 9 E J(X) if and only if 9 E dom O. Assertions (i)-(iii)
and (v) are immediate consequences of our definitions and assumptions. Point (iv)
is proved by an easy induction. •
Definition 17.1. Suppose that IF = JR, X is a complete linear metric space,
X E L(D) and has the unit e E dom 0, (L, E) E G[O], F is a multiplicative initial
256
D. Przeworska-Rolewicz, Logarithms and Antilogarithms
© Kluwer Academic Publishers 1998
Fractional calculus 257
D
Corollary 17.1. Suppose that all conditions of Definition 17.1 are satisfied. Then
Write
(17.6)
Write
Corollary 17.3. Suppose that all conditions of Definition 17.1 are satisfied and
the set Po (g) is defined by (17.7). Then
(17.8)
Note 17.2. By Corollaries 17.2, 17.3, the operator D is invertible on the sets
peg), Po(g), respectively, so that ker D = {O}. On the other hand, we know that
in a Leibniz D-algebra with unit we have ker D =j:. {O} (cf. Theorem 4.3). It is
so, since the unit e does not belong to these sets. Indeed, gO = e and we consider
integers (rational numbers, respectively) different than zero. If F is an initial
operator for D corresponding to Rand {3 = 1 then we get for all a E lR:
Fg<> = (1 - RD)g<>
_ <> _ R r(a + 1) <>-1 _ <> _ f(a + 1) R <>-1
- 9 f(a + 1 - l)g - 9 f(a) 9
_ <> _ f(a + 1 r(a) <>-1+1 _ <> _ <> _ 0
-g f(a) f(a-l+l+l)g -g 9 - .
258 Chapter 17
t (t s)k-1
(17.9) (RkX)(t) = Jto (; _ I)! x(s)ds for x EX (k EN).
This formula motivated at least one way for introducing the fractional calculus.
Namely, the so-called Riemann-Liouville fractional integral of order a is defined
as follows:
[t (t S )C>.-1
(17.20) (RC>.x)(t) = Jto f(a) x(s)ds for x E X (Re a E 114)·
The integral (17.20) coincides with the integral (17.19) for a = kEN. We admit
RO = I. Thus we may write
In particular, DC>. Ra. x = x for Re a E 114 and all x EX, i.e. DC>. RC>. = I. Putting
a = 0, we get for x E dom Dk = Ck[O, 1]
o
APPENDIX
FUNCTIONAL SHIFTS
ZBIGNIEW BINDERMAN
We shall preserve the denotations admitted in the preceding chapters. Let H(n)
be the space of all functions analytic on a set n ~ C. In our subsequent considera-
tions, the field of scalars is always IF = Co Denote by lK( aj r) disks on the complex
plane, i.e.
lK(ajr) = {h E C: Ih - al < r}, 0 < r ~ +00.
For the sake of brevity, we shall write JK,. = lK(O, r).
In this section lK will stand either for the unit disk lKl = lK(Oj 1) or for the complex
plane C.
Note that, by our assumption, there exists always a ..\ E lK such that
L
00
Sr(D) = n
nENo
S}n) (D).
259
260 Functional shifts
=L
00
Y akhk Dk+nx =
k=O
m m
L L L L
00 00
Hence
for n > mj
(A1.2)
for m ~ n,
where z~ = E;=~n-k aphPzk+n+p E ker D (k = 0,1, ... ,m - n). This implies that
the series E~o akhk Dk+nx is convergent for all n E No, i.e. x E Sr(D).
(ii) Let>. E lK and let x E ker(D - >'1), h E lK be arbitrarily fixed. By our
assumption, Dx = >.x. The equalities Dm x = >.mx for all m E No (obtained by
an easy induction) together imply for all n E No
00 00 00
Write
•
00
Lemma ALl. Suppose that all assumptions of Proposition Al.I are satisfied
and D is closed. Let h E l!{ be arbitrarily fixed. Then
L ak hk Dkx = L
00 00
L L
00 00
L
00
L
00
Lemma A1.2. Suppose that all assumptions of Proposition AI.I are satisfied,
R E RD and A E vcR. Then A is an eigenvalue of the operator D and the
corresponding eigenvectors are
L
00
This implies that the scalar series E~=o lanlPlhlnplIDnxll is convergent for all
h E OCl (for all h E ee, respectively). Since X is a complete linear metric space
with an invariant metric J.t(x, y) defined by the equality: J.t(x, y) = Ilx - yll for all
x, y EX, we conclude that the series E~=o anh nDnx is absolutely convergent for
allhEK •
Note ALl. Suppose that all assumptions of Proposition A1.2 are satisfied. Then
We still assume that X is a complete linear metric space and f E H(lK), where
lK = lKr (0 < r :::; 00). Suppose that D E R(X). We shall study families
T"K = {T"hhEIK C Lo, where the operators T"h coincide with the operator-
valued functions f(hD) on the set S,(D).
Clearly, by our definitions, Proposition Al.I and Corollary Al.3, we have
Definition A2.1. Let X be a complete linear metric space and let f E H(lK)
have the expansion
L
00
where the operator f(hD) and the set S,(D) are defined by Formulae (Al.4),
(Al.l), respectively. 0
Remark A2.1. By the definition of the set S, the sum E~=o anh n Dnx for
h E lK, xES has only a finite number of members different than zero. Thus some
further results for S are true if we assume only that X is a linear space (without
any topology). 0
In the sequel we shall assume (for the sake of brevity only) that either lK = C or
lK is the unit disk lKl .
Remark A2.2. Some further results for the sets ker(D - >.J) (>. E lK) and E are
true if X is only a linear topological space.
o
Let G be a subset of an open set 0 C C, let 9 E H(O) and let DE R(X), R E RD.
A family Tg,G = {Tg,hhEG C Lo(X) was named by
• PRZEWORSKA-RoLEWICZ a family of R-shifts (D-shifts on the set E(R) for
R E RD n V(X» if g(h) = eh and either G = lR or G = Il4 (cf. PR[7], PR[8]);
• BINDERMAN a family of complex R-shifts for an entire function 9 (cf. Bin[l]),
functional shifts for an analytic function 9 (cf. Bin[5]).
Appendix 265
Families of R-shifts and D-shifts were defined in two different ways. In general,
they do not coincide. If D-shifts and R-shifts coincide, then in PR[13] are called
true shifts (cf. also Definition 11.2). Note that the definition of a family of R-shifts
is independent of the choice of a right inverse R (cf. Bin[5]).
Shifts for a right invertible operator D induced by a function analytic in a ring and
a right inverse of D are studied in Bin[12], Bin[13] and Bin[15]. Shifts induced by a
member of a subalgebra of the algebra of all functions continuous on the unit circle
are considered in Bin[16]. For the entire collection, see also Bin[2]-Bin[4], Bin[6]-
Bin[ll], Bin[14], BPR[l], BPR[2], PRUDNIKOV Pru[l], PR[14], PR[15], PR[23].
It will be shown that, if Tg,h have behaviours as the operator-valued function ehD
=
(h E IG) has on a set containing either S P(R) or E(R), then Tg,G is a family of
R-shifts on P(R) and a family of D-shifts on E(R).
An immediate consequence of the definition is
Proposition A2.1. Suppose that X is complete linear metric space, D E R(X),
ker D =f:. {O} and Tf,'{t;. = {T"hhElK is a family of functional shifts induced by a
function f E H (lK). Then
(i) the operator T"h ofthe functional variable f E H (lK) and the complex variable
hE lK, as an operator acting in linear space Lo(S,(D)), is linear, i.e.
(ii) for all h E lK the operators T"h are uniquely determined on the set S,(D);
(iii) if S,(D) = X and T"h are continuous for h E lK then T"h are uniquely
determined on the whole space X .
Theorem A2.1. Suppose that X is complete linear metric space, D E R(X),
ker D =f:. {O}, F is an initial operator for D corresponding to an R E RD and
TO{ = {T"hhElK is a family of functional shifts induced by a function f E H(lK).
Then
(i) for all h E lK, kENo and Z E ker D
k k
(A2.3) T"hRk F =L ajh j Rk-j F, i.e. T"hRk Z =L ajh j Rk-j Z;
j=O j=O
k
S"hR~z = L ajhjR~-jz for all hE lK, kENo, z E kerD, where Rl E RD,
j=O
then
266 Functional shifts
00
=L
j=o
k 00
=L ajhjRk-jz+ L
j=o j=k+1
k k
j
=L ajh Rk-j z =L ajhj Rk-j Fx.
j=O j=O
=L
m
Proposition A2.2. Suppose that all assumptions of Proposition A2.1 are satis-
fied. Then for all h E I[( the operators T"h commute on the set S with the operator
D.
Proof Let h E 1[(, kEN, z E ker D be arbitrarily fixed. By Theorem A2.l,
k
=D L ajh j Rk-j z - T"hR k- l Z
j=O
k-l k-l
- " aJ·hjRk-j-l z - ~
- ~ " aJ·hjRk-l-j z -- 0 .
j=O j=O
and the operators T"h commute on ker(D - >..1) with the operator D.
Proof Formula (A2.4) is an immediate consequence of (Al.3). Let x E ker(D->..I)
be arbitrarily fixed. Formula (A2.4) together with the equality Dx = >..x imply
for all h E I[(
DT"hX - Tj,h Dx
= D[f(>"h)x] - T"h(>"X) =
f(>"h)Dx - >"T"hX
= f(>"h)>"x - >..f(>"h)x = O.
Proposition A2.5. Suppose that all assumptions of Theorem A2.1 are satisfied
and the operator I - AR is invertible for a A E OC \ {O}. Then for all h E OC
(i) T"he).,F = f(Ah)e).,F, where e)., = (I - AR)-l j
(ii) The operators T"h commute on the set E).,(R) (defined in Corollary A1.3(i))
with Dj
(iii) FT"he).,Z = f(Ah)z for all z E ker D.
Proof. (i) and (ii) follow from Lemma A1.2 and Proposition A2.4.
In order to prove (iii), assume that z E ker D and h E OC are arbitrarily fixed. By
(i) and properties of an initial operator for D, we get
(FT"he).,]Z = F[J(Ah)e).,z]
= f(Ah)Fe).,z = f(Ah)(I - RD)e).,z
= f(Ah)[e).,z - R(Ae).,z)]
= f(Ah)(I - AR)(I - AR)-l z = f(Ah)z.
=T/,h L L
00 00
L =L L
00 00 00 00
=Lk=O
akhkDkx -
k=l
akh k
n=O
DkRnFDnx
k-l
L L
00
= f(hD)x - ak hk Dk-nFDnx
k=l n=O
k
=
00
f(hD)x - L ak hk
k-l m=l
L
DmFDk-mx = f(hD)x.
(ii) Fix x E AR(D). By the continuity of R, we have Dkx E AR(D) for all kENo
(cf. PR[9], Theorem 3.2). This, and Theorem A2.1(i) together imply for all h E OC
=T/,h L =L L
00 00 00
=L L L
00 00 00 00
=L
DO
ak hk Dkx = f(hD)x.
k=O
Proposition A2.6. Suppose that all assumptions of Theorem A2.2 are satisfied.
•
If F is a continuous initial operator for D corresponding to a continuous R E RD
then T/,h map AR(D) into itself for all h E K
Proof. Let x E AR(D) and h E OC be arbitrarily fixed. The continuity of the
operators Rand F implies that D is closed (cf. PR[9]). By Theorem A2.2 (either
(i) or (ii)) , T/,hx =
f(hD)x. Since Dk E AR(D) for all kENo (cf. PR[9]) , by
270 Functional shifts
definition,
n=O
=L L
00 00
RnFD n akhkDkx
n=O k=O
=L L
00 00
akhkRnFDn(Dkx)
n=O k=O
00 00
k=O n=O
L
00
= ak hk Dkx = T"h X •
k=O
•
Appendix 271
It is well-known that the set H(OC) is a commutative algebra with the following
operations:
(A3.2)
where f, 9 E H(OC), a E Ci
Oi) the algebras H(OC) and Ts(OC) are isomorphic and T : f -t T/,Kls is an algebra
isomorphism from H(OC) onto Ts(OC).
Proof. (i) It is enough to show that the multiplication of two families of functional
shifts for D is well defined. Suppose then that we are given families T/,K, Tg,K,
where f,g E H(OC) are arbitrarily fixed and
=L
00
=L
00
k
=T"h(Tg,hR k z) = T"h '~
"'
bk-jh k -3. R3. z
j=O
k k j
=L bk_jhk-jT"hRjz =L bk_jhk- j L aj_mhj-mRmz
j=O j=O m=O
k j k k
= L L bk_jaj_mhk-m Rmz =L [L aj_mbk_jjh k- mRmz
j=O m=O m=O j=m
k
= '"'
~ Ck-m hk-mRmz,
m=O
where n
C n= L apbn_p (n E No).
p=O
Now, by the Cauchy theorem about the multiplication of two series, we conclude
that w = fg E H(OC) and
L
00
w(h) = cnhn.
n=O
(ii) the neutral elements and units of the commutative algebras H(OC) and Ts(lK)
are
O(h) == 0 on lK, ORkz = 0, l(h) == 1 on lK, IRkz = Rkz,
respectively (k E No, z E ker D, R E RD).
Observe that
(T/,hTg,h)X
(A3.3) =T/,h(Tg,hX) = T/,h[g(Ah)x] = g(Ah)T/,hX
= g(Ah)f(Ah)x = T/g,hX.
The proof of (ii) is going along the same lines as the proof of Theorem A3.2(ii) . •
Similarly, we prove
Theorem A3.3. Suppose that all assumptions of Theorem A3.1 are satisfied. Let
OC = C and let Y = E = E(R). Then
(i) the set Ty (OC) is a commutative algebra with the operations defined by Formulae
(A3.2);
(ii) ifE =f. ker D then the algebras H(OC) and Ty(OC) are isomorphic and T : f -+
T/,Kly is an algebra isomorphism from H(OC) onto Ty(OC).
Write for an operator D E R(X)
(A3.4) SK(D) = n
/EH(K)
Sr(D).
where Sr(D) is defined by Formula (Al.I). Proposition Al.I and Corollary Al.3
together imply that
S c SK(D),
00 m 00 m
=L (L am_nbn)hmDmx =L L am_nbnhmDmx
m=O n=O m=O n=O
00 00 00 00
(ii) the algebras H(OC) and Y are isomorphic and T : f ~ Y is an algebra isomor-
phism from H (OC) onto Y.
Proof. (i) It is enough to show that the multiplication of two families of continuous
functional shifts for D restricted to AR(D) is well-defined. Suppose then that we
are given families Tn". , Tg,K E TC(OC), where f, 9 E H(OC) are arbitrarily fixed and
=L =L
00 00
= bn_khn-kT"hRk F Dnx
n=O k=O
n k
=L L L
00
ak_jbn_khn-jRjFDnx
n=O k=O j=O
n n
=L L (L
00
ak_jbn_i)hn-iRiFDnx
n=O j=O k=j
n
L L
00
= cn_jRi FDnx,
n=O j=O
where
n
Cn = L apb n _ p (n EN).
p=O
Since w = fg E H(OC) and
=L
00
L L
00 00
Proposition A3.3. Suppose that all assumptions of Theorem A3.1 are satisfied
and the function J satisfies Equation (A3.6) for all hI, hz E 11(. Then T/,lKis is a
commutative semigroup for II( = 11(1 and an Abelian group for II( = C with respect
to the addition of operators, i.e.
Proposition A3.4. Suppose that all assumptions of Theorem A3.1 are satisfied
and the function J satisfies Equation (A3. 7) for all hI, hz E 11(. If II( = C then
T/,lKis is an Abelian group with respect to the superposition of operators,i.e.
Proposition A3.5. Suppose that all assumptions of Theorem A3.1 are satisfied
and the function J satisfies Equation (A3.8) for all hI, hz E K Then T!,I{is is a
commutative semigroup for II( = 11(1 and an Abelian group for II( = C with respect
to the superposition of operators, i.e.
Appendix 277
Remark A3.1. Suppose that all assumptions of Theorem A3.I are satisfied.
Propositions similar to Propositions A3.2-A3.5 for the family T'.Kls
hold for the
families T'.Kly,where
Proposition A3.6. (d. BPR[I]) Let D E R(X) and let ker D -:j:. {o}. Let T,.o,
TI'.o be families of functional shifts for the operator D induced by the functions
r
f, E H(IK), respectively. Then
(i) if ho Ene IK is a limit point of n then
I
lim
03h-+ho
ho _ h(T'.ho - T,.h)X = TI'.hoDx = DTI'.hoX for all xES U EA;
Definition A3.1. (cf. BPR[I]) Let 0 Ene IK be the limit point of the set n.
A family To = {ThhEO C Lo(X) has the limit property on a set Y C X if for
every x E Y there exists lim -hI (Th - To)x. If it is the case, then the operator A
h-+O
defined as
(A3.9) Ax = h-+O
lim -hI (Th - To)x for x EY = dom A
is said to be an infinitesimal generator for the family To. o
Note that the infinitesimal operator A (if exists) is well-defined. Clearly, if X
is a locally convex complete linear metric space, then any strongly continuous
semigroup To = {ThhEO of linear operators is a family with the limit property
and its infinitesimal generator satisfies Condition (A3.9) (d. HILLE and PHILLIPS
HP[I]).
Proposition A3.6 leads us to the following
Corollary A3.1. (d. BPR[I]) Suppose that T,.o is a family of functional shifts
for D induced by a function f E H(IK) and 0 E n is a limit point of n. Then
278 Functional shifts
this family has the limit property on the set SUEA and j'(O)D is its infinitesimal
generator.
Theorem A3.6. Suppose that OC = c, n c OC is the interior of a spinal semi-
module (i.e. an additive semigroup which contains a ray from the origin and an
open set intersected by this ray, cf. HILLE and PHILLIPS HP[l]) containing the
positive real axis, D E R(X), ker D =f:. {O} and T/,n = {T/,hhEO is a family of
functional shifts for D induced by a function f E H(C), f ~ O. Then T/,n is
a commutative semigroup with respect to the superposition of operators on the
set Y = S U EA (Y = Sn(D) if D is closed, Y = AR(D) if R is continuous,
respectively), i.e.
if and only if there is an a E C such that f (h) = eoh . If it is the case then
i.e. the family Tj,K has the limit property on the set S U EA, its infinitesimal
generator is f(O)(D + A) and T;,K acts in the following manner:
Tj,hx = (f(hD) - f(O)](I + RA)x + f(O)x for all xES U EA, hE 11.
If, in addition, f (0) # 0 then the family TJ,f!{ = TIl f' ,oT;,f!{ has the limit property
on S U EA and its infinitesimal generator is D + A. Moreover,
1
T7,h X = reO) ([f(hD) - f(O)](1 + RA)x + f(O)x} for xES U EA, hE lK.
Proof. Suppose that the family Tj,o is defined by Formula (A3.12). Then T;,o =
Tj,o. Let xES U EA be arbitrarily fixed. By Proposition A3.6, we get
1 (Tj,h
lim -h I I
- Tj,o )
x
h-tO
1
= lim -h[Tj,h - Tj,o
h-tO
+ (Tj,h - Tj,o)RA]x
T;,h X
= Tj,h(I + RA)x - Tj,oRAx = f(hD)(1 + RA)x - f(O)RAx
= [f(hD) - f(O)](1 + RA)x + f(O)x.
The conclusion if f (0) # 0 is obvious. •
Note A3.2. Suppose that D E R(X), R E RD, A E Lo(X) and the operator
1 +RA is invertible. Recall that DO = D+A E R(X) and RO = (I +RA)-I R E RD
(cf. PR[8]). Recall also that the sum of a closed operator and a continuous operator
is again closed. So that, if D is closed and A is continuous, then DO is closed. 0
280 Functional shifts
More details about the limit property, perturbations of functional shifts and ex-
amples can be found in BPR[l I.
Theorems A3.1, A3.2, A3.3, A3.4, A3.5, respectively, immediately imply the fol-
lowing
Theorem 18.2. Suppose that all assumptions of Theorem A3.1 are satisfied. Let
Theorem 18.3. Suppose that all assumptions of Theorem A3.1 are satisfied,
Y is determined by Formulae (18.1) and f, 1/ f E H(lK). Then Y E Lg(D) and
(L, E) E G[OI, where
DLT x = T x = DT x for x E Y,
J,K f' / J,K In J,K
L
00
ajhjDiRnDnx
j=O
n-l
=L L L
00 00
L L
n-l k
= f(hD)x - ajhi Rk-i FDkx.
k=O i=O
Appendix 281
Theorem A3.9. (cf. BPR[2]) Suppose that D E R(X), ker D :f; {O}, F is an
initial operator for D corresponding to an R E RD, Tj,K = {Tj,hhEK, Tl/j,K =
{Tll/,hhEK are families of functional shifts for D induced by functions f, 1/ f E
H(][{) , respectively, and Fh,Rh (h E OC) are defined by Formulae (A3.13). Let
ker(D -,xI) :f; {O} and let,xOC C OC (,x E C). Suppose, moreover, that f(h) :f; 0 for
hE oc. Then
L
00
On the other hand, again by (A3.14) and the obvious equality Rx = ,x-I (x - Fx),
we obtain
RhX = Rx - FhRx
= Rx - a Dl FT/,hRx = Rx - a Dl FP-l[f(,xh)x - aoFx]}
= ,x-I (x - Fx) - aolp-l[f(,xh)Fx - aoF2 x]}
o
= ,x-IX - ,x-lFx - ,x-la l f(,xh)Fx + ,x-I Fx
= ,x-I [x - a Dl f(,xh)Fx].
282 Functional shifts
L L
n-l k
=>..-nf(>'h)U-1(>'h)>..-1(x-Fx)- >.k bj hi R k+1-iFx
k=O j=O
o
- a f-1(>"h)>.-1 F[f(>'h)x - Tf,hFxj
1
n-l k
+a oL
1 >..kF L bjhjTf,hRk+1-jFx}
k=O j=O
k
L L
n-l
= >.-(n+l) f(>.h)(f-l(>.h)x - rl(>'h)Fx - >.k+l bjh j R k+1-j Fx
k=O j=O
- a o Fx + a o rl(>'h)aoF x
1 1 2
n-l k k+l-j
+ a o L >.k+l L bjhj F L
1 aphP Rk+l-j-p Fx]
k=O j=O p=o
k
L L
n-l
=>.-(n+1) f(>.h)[f-l(>.h)x - f-l(>.h)Fx - >.k+l bjh j R k+1-j Fx
k=O j=O
n-l k
- ao l Fx + rl(>'h)Fx + a o1 L >.k+l L bjhjak+1_jhk+l-j Fx]
k=O j=O
n-l k
=>.-(n+1) f(>.h)(f-l(>.h)x - a o Fx - L
1 >.k+l L bjh j R k +1-j Fx
k=O j=O
n-l
- ao1ao L >.k+ 1bk+1hk+1 Fx].
k=O
Appendix 283
L L
n-I k+I
= A-(n+1) I(Ah)[rI(Ah)x - a l Fx -o Ak+1 bjh j R k+1-i Fx]
k=O j=O
n-I
= I(Ah)A-(n+I) [J-I(Ah)x - o
a l Fx - L Ak+1TII/,hRk+1 Fx]
k=O
n-I
= I(Ah)A-(n+1)[!-l(Ah)x - boFx - L Ak+1TI I/,h Rk +1 Fx]
k=O
n
= I(Ah)A-(n+1)[r I (Ah)x - TII/,h Fx - L APTII/,hRPFx]
p=l
n
= I(Ah)A-(n+1) [rI(Ah)x - L AkTII/,hRkFx]
k=O
Theorem A3.10. (cf. BPR[2j) Suppose that D E R(X), ker D :j; {O} and
R E nDnAQN(ker D;w). Let T/,¥:. = {T/,hhE¥:., T II /,¥:. = {TII/,hhEK be families
of functional shifts for D induced by the functions 1,1/1 E H(OC), respectively. If
the operator TII/,h is continuous for some hE OC then Rh E nD nAQN(ker D;w).
Proof By our assumptions, for A E w and Z E ker D
lim AnRhz
n-too
= lim Anl(O)TII/hRnz=/(O) lim TII/h(AnRnz)
n-t(X) ! n-too'
The continuity of the operators Rh (h E OC) on ker D follows from Formula (A3.I6).
Suppose now that A E wand there is an h E OC such that the operator I - ARh
is not invertible on ker D. It means that there is a z E ker D \ {O} such that
z E ker (I - ARh), i.e. z - ARhZ = O. Hence we get
The list of similar results for other subsets of S/(D) can be easily prolonged.
Appendix 285
The purpose of the present section is to give formulae for functional shifts induced
by an operator D of generalized complex differentiation on the linear span S =
P(R) of all D-monomials. The choice of functions under consideration depends
on algebraic summation formulae. Results obtained here will permit us to derive
explicit forms of functional shifts in much more complicated cases. This section
contains results of Bin[5]'[8]'[9J.
Let X = H(OC), where OC is either the unit disk OCl or C. Recall that H(OC)
is a complete linear metric space with the topology of uniform convergence on
compact sets. Denote by Xl C X the set of all polynomials defined on OC and with
coefficients in C, i.e. Xl = crt], (t E OC). Clearly, Xl is a linear space over C with
the usual pointwise addition and multiplication by scalars. Let U = {Uk} C C be a
sequence such that Uk f:. 0 for all kENo. Define linear operators D and R, called
operators of generalized complex differentiation and integration, respectively, as
follows:
The operators D and R E Lo(Xt} are uniquely determined on Xl and dim ker D =
1, codim RX I = 1 (cf. FAGE and NAGNIBIDA FN[l]). The kernel of D consists of
all constant functions on][{, Since there is a one-to-one correspondence between
all constant functions belonging to X and complex numbers, we can identify here
constants functions with scalars if it does not lead to any misunderstanding. We
can prove (cf. Bin[6]) that
(i) D E R(Xt}, R E RD;
(ii) (Fx)(t) = [(J - RD)xJ(t) = x(O) for t E OC, x E Xl;
(iii) S = P(R) = Xl and Xl = P(R) = X.
In the sequel we shall study only such operators D and R for which the condition
1·1m
n--+oo
(-!£n =
n --
Un+l
1
x(t) - x(O)
for t::j: OJ
(Dx)(t) = {
t
(A4.2) (Rx)(t) = tx(t)
x(t) ~ x(O) It=o = x'(O) for t = OJ
The operator D is called the Pommiez operator (cf. Example 5.4). In this case
00 00 (nl(o) 00 (nl(o)
" (RnFDnx)(t)
L...J
="
L...J
_x-Rnl
n!
="
L...J
_x_tn
n!
= x(t) (t ElK).
n=O n=O n=O
This implies that AR(D) = Doo = X. If]l{ = IKl then it is easy to verify that the
operator J - )"R is invertible for all ).. E IKl and
= <C then for every).. E <C \ {O} the operator is not invertible in X. Indeed,
If ]I{
consider the equation (J - )"R)x = y with yet) == 1. Then a solution of this
equation is x(t) = 1/(1- )..t). Clearly, x is not holomorphic at the point t = 1/)".
Moreover, if ]I{ = ]l{l then ker(D - AI) ::j: {O} if and only if A E ]l{l and
Hence
Appendix 287
Let OC = OC1 and let f(h) = 1-1 h' Let x E X, h E OC1 be arbitrarily fixed. By
(A4.3),
00
n=O
=~ hn ~ t k x(k) (0) =~
x(k) (0) ~ hntk-n
~ tn ~ k! k!
~ ~
n=O k=n k=O n=O
1L
tk+l - hk+1 x(k) (0)
L
00
h-t ~ fortoJh;
_ k=O
T
- 00 (k)(O)
(k+1)h k fort=h.
k=O
Hence
tx(t) - hx(h) for t oJ h'
L
00 {
hn(Dnx)(t) = t- h '
n=O x(h) + hx'(h) for t = h.
This proves that S,(D) = X. Moreover, this shows that the family TKI
{ThhEKt> where
tx(t) - hx(h)
for t oJ hj
(nx)(t) = { t- h
x(h) + hx'(h) for t = hj
is the unique family of functional shifts for D induced by the function f(h) =
h ~ l' The operators Th (h E OCt} are continuous (cf. Bin[5]). 0
Proposition A4.1. Let OC = OC1 and let f E X = H(OC): f(h) = h _1 l' Let
T"K = {T"hhEOC be a family of functional shifts for the Pommiez operator D
(A4.2) induced by f. Then
(A4.6)
Let Tj,K = {Tj,hhEK be a family of functional shifts for the Pommiez operator D
defined in (A4.2) induced by the function f. Then for all x E X, hE OC
=L
00
n=l
1
00
- ~ k(k+1)h kx (k)(O)
~ 2 k! for t = h.
k=l
Appendix 289
n=l
h ~ 2 x(k) (0) k-l x(k) (0) k
= (h _ t)2 L..J [h (k _1)!h + t-k!-t ]
f:
k=l
and for t = h E IK
k=l
h2
= hx'(h) + -x"(h).
2
This proves that SI (D) = X. Moreover, this also shows that for all x EX, h E IK
+L
00
ao(Ix)(t) nhn(Dnx)(t) =
n=l
(A4.7) D = i, R= {t.
dt 10
In this case
(A4.8)
290 Functional shifts
and every element xES can be written (in a unique manner) in the form
n n
(A4.9) x(t) =L bkt k =L ck Rk1 , (n E No),
k=O k=O
L
00
o
Proposition A4.3 and Remark A4.2 together imply
Corollary A4.1. LetOC = C. Let Tj,K = {Tj,hhEK be a family offunctionalshifts
induced by the exponential function. Let an operator T of the form (A4.1O) be an
isomorphism of the space H(OC). Then there are 13, hE C such that T = f3Tj,h'
Proposition A4.4. Let f E X = H(OC): f(h) = ao + heh, where ao E C is
arbitrarily fixed. Let Tj,K = {Tj,hhEK be a family of functional shifts for the
operator D = -itinduced by f. Then for all xES, h E OC
hk
=L = ao + L
00 00
Let h E ][{ and xES of the form (A4.9) be arbitrarily fixed. By (A4.8), for t E ][{
we have
n n n k
=T/,h L ckR k l = L ckT/,hRkl = L Ck L a j h j R k- j l
k=O k=O k=O j=O
n k
= L [L ajhj Rk-jl + aoRkl] + aoCo
k=1 j=1
n n k k'
-_ aoCo + ao '~
" Ck R k 1 + '~
" bk '~
" (. _ l)!(k
. _ .)! h j t k-j
k=1 k=1 3=1 J J
n k N
= aox(t) + (; bk ~ G)ihjtk-i = aox(t) + {; bkkh(h + t)k-1,
(A4.12)
hk
=L = ao + a1 h + L
00 00
Let h E ][{ and xES of the form (A4.9) be arbitrarily fixed. By (A4.8), for t E ][{
292 Functional shifts
we have
n k
=L Ck L ajh j R k- j 1
k=O j=O
1 k n 1
= L Ck L ajh j R k- j l +L Ck L ajh j R k- j l
k=O j=O k=2 j=O
n k
+ L Ck L aj h j R k- j 1
k=2 j=2
n
=boao +b1(aot+alh) + L bk(aot k +alhktk- l )
k=2
~ ~ k! 'k'
+ L...J bk L...J (' _ 2)!(k _ ,)!h3 t -3
k=2 j=l J J
n n
= ao L bk tk + alh L bkktk- 1
t t
k=O k=l
+ bk (~)j(j - 1)h t j k- j
k=2 j=2 J
L
n
I(h) = {;
00
ak hk = ao + t;
00 khk
(k _ 1)! lor all hE lK.
Appendix 293
Let h E OC and xES of the form (A4.9) be arbitrarily fixed. By (A4.8), for t E OC
we have
n k
= L Ck L ajh j R k- j l
k=O j=O
1 k n k
= L Ck L ajh j R k- j l +L Ck L ajh j R k- j 1
k=O j=O k=2 j=l
n
= boao + b1(aot + h) + ao L bkt k
k=2
n k .k'
+ "L..t bk "L..t (. _ l)!(k
J . _ .)! h j tk- j
k=2 j=l J J
By the formula
t
k=O
(~) k 2akbn- k = nb(a + nb)(a + b)n-2 for all a, bEe, 1 < n E N,
for t E OC we get
n
= aox(t) + hx'(O) + t L kbk(h + kt)(h + t)k-2
k=2
n
= aox(t) + hx'(O) + t L kbk(h - t)k-l
k=2
n
+ t2 L bkk(k - l)(h + t)k-2
k=2
= aox(t) + hx'(O) + t[x'(t + h) - x'(O)] + t 2x"(t + h),
which implies (A4.13).
Example A4.3. Take the sequence a = {ad, where ao = 1, ak = (k~l)! for all
•
kEN. Then the operators D, R E Lo(X) determined by Formulae (A4.1) have
the form
(A4.14) (Dx)(t) = x'(t)+x'(O)- x(t) ~ x(O) , where x(t) ~ x(O) It=o = x'(O),
294 Functional shifts
1 for k = 0
(A4.15) Rk1 ={ tk
(k _ 1)! for k E Nj
and every element xES can be written (in a unique way) in the form
n n
(A4.16) x(t) =L bktk = L ck Rk1 , Co = bo, Ck = (k-1)!bk' bk E C (k EN).
k=O k=O
o
Proposition A4.7. Let OC = C and let X = H(OC). Let T/,'ft{ = {T/,hhE'ft{ C
Lo(X) be a family of functional shifts for the operator D determined by Formula
(A4.14) induced by the exponential function f(h) = e h • Then for all XES,
h, t E OC
It is obvious that
ih
4
(T/,hX)(t) = { -h h
i[x(O) - x(-2h) - 2hx'(O)] +~ x(s) ~ x(O) ds for t = -h
2 _ x(t + h) - x(t - h) _ ~ x(t + h) - x(t - h) - 2x(O)
(T x)(t) = _ t
/,h t2 _ h2 2 t 2 - h2 2 +
+ (h x(s) - x(O) ds otherwise.
l-h s
o
Appendix 295
x(t) - x(O) ,
where, as before, t It=o = x (0).
In a similar way as Proposition A4.6, we prove
Proposition A4.10. Let OC = C. Let f E X = H(OC) : f(h) = ao + (h + h2)e h ,
where ao E C is arbitrarily fixed. Let Tj,K = {Tj,hhEK C Lo(X) be a family of
functional shifts for the operator D determined by Formula (A4.14) induced by f.
Then for all XES, h, t E OC
t 2 (h-t)[x(h+t)-x(0)y t 3 I/(h )
+ h+t h+t + h+t X +t for t f. -h.
296 Functional shifts
The purpose of the present section is to give some analogues of the Euler-Maclaurin
Summation Formula (cf. MILLER Mil[l], PR[I]j also Chapter 11) induced by
functional shifts. We shall give examples of applications of formulae obtained for
few elementary functions (cf. Bin[14]).
Assume that X is a complete linear metric space, OC = OC(O, r), 0 < r :S: 00 and a
function f E H (OC) has the expansion
=L
00
Suppose that D E R(X), ker D ::J {O} and TO.. = {T/,hhEIK is a family of func-
tional shifts induced by the function f E H(OC). Consider the Pommiez type
operators defined as follows:
L
00
This implies
(A5.5) P E Lo[H(OC)].
Note that the operator (Pg)(hD) for the function g(h) = eh is called the Bernoulli
operator for D (cf. PR[14]' also Chapter 11).
Using Proposition A2.2, it is easy to prove
Proposition A5.I. Suppose that DE R(X), ker D ::J 0 and T/,K = {T/,hhEIK is
a family of functional shifts for D induced by a function f E H(OC). Then
L
00
all forn=Oi
(A5.7) b - {
L
n-1
n - -all bk a n -k+1 for n E N.
k=O
Theorem A3.1 implies that the operators (P f) (hD) (h E 1K) are invertible on S
and [(P f)(hD)]-l = g(hD). Taking into account the expansion of the function g,
by Formulae (A5.4) and (A5.9), we get
[L HL L
00 00 00
where
L
00
We are now ready to give the main theorem of this section, which is an immediate
consequence of Theorem A5.2 and Formula (A5.6).
Theorem A5.3. Suppose that all assumptions of Theorem A5.1 are satisfied.
Then the following Euler-Maclaurin type formulae hold on the set S:
+L
00
+L
00
1= (PeS)ls=hD + f:
n=l
Bn h:~l (e hD - I)D n - 1 for hE lK \ {O},
where, as before, B 2n are Bernoulli numbers. In this case Formula (A5.1O.b) has
the form
(b) Let JK. = lK.r /2 and let f be the hyperbolic tangentfunction. Then for s E JK.\ {O}
00 22n(22n-l _ 1)
tanh(s) =~ (2n)! B2ns2n-l,
Appendix 299
where B 2n are Bernoulli numbers. In this case Formula (A5.1O.b) has the form
n=1
()'
2n.
-
IB2n lh 2n ,
where B 2n are Bernoulli numbers. In this case Formula (A5.1O.b) has the form
(d) Let OC = OC7T and let f be hyperbolic sine function. Then for h E lK. \ {O} we
have (Psinh)(t) = t- 1 sinht and,
t 00 2(2 2n - 1 - 1)
g(t) = sinht = 1- ~ (2n)! B 2n t 2n ,
where B 2n are Bernoulli numbers. In this case Formula (A5.10.b) has the form
o
Remark A5.1. Suppose that all assumptions of Proposition A5.1 are satisfied
and f(h) i- 0 for hE K Then, by Theorem A3.1,
L
00
Co = ao-1 , far n E N.
o
Assume that either lK. = C or lK. is the unit disk lK.1 . In the sequel we will show
that the Euler-Maclaurin type Formulae (A5.1O) hold for the sets
300 Functional shifts
• Soc(D) = n S';(D),
gEH(OC)
• AR(D), R E RD.
Let DE R(X) and let ker(D - >.1) =f. {O} for h E lK\ {O}. We rewrite the Pommiez
type operator s,p defined by Formula (A5.2) as follows
Clearly, Formula (A5.12) implies that the operator D commutes with s,p on the
set Tker(D-'xI)(lK) (d. Theorem A3.2).
A result similar to Proposition A5.1 is
Proposition A5.2. Suppose that D E R(X), ker D =f. {O} and TI,oc = {TI,hhEOC
is a family of functional shifts for D induced by a function J E H(lK). Suppose,
moreover, that ker(D - >.1) =f. {O} for a A E lK. Then
= (Pf)(hD)D = [L = h- 1 L
00 00
Proof. Let x E Spf(D) and hE lK\ {O} be arbitrarily fixed. By (A1.l) and (Al.4),
(Pf)(hD)x
L =L
00 00
= ak+1hkDkX ak+lhkDkDRx
k=O k=O
= h-1[f(hD)Rx - aoRx] = h-1[f(hD) - f(O)I]Rx = I.l3Tf,h.
Clearly, Formula (A5.l4) holds also for h = O.
Observe that, by the last equality, Rx E Sf(D) whenever x E SPf(D). Conversely,
x E Spf(D) whenever Rx E Sf(D). •
A consequence of Lemma A5.l is
Proposition A5.3. Suppose that all assumptions of Lemma A5.1 are satisfied.
Then Equality (A5.14) holds on the set Soc.(D) determined by Formula (A3.4).
Clearly, by definitions and Corollary A1.3, we have
Applying Theorem A3.2, in a similar way as Theorem A5.l (cf. Theorem A5.2),
we prove
Theorem A5.4. Suppose that D E R(X), ker D ¥- {O} and Tf,oc. = {Tf,hhEK
is a family of functional shifts for D induced by a function f E H (lK). Let Y =
ker(D - >d) ¥- {O} for a A E lK. Let f(1)(O) ¥- 0 and f(h) ¥- f(O) for h E
Iy
lK \ {O}. Then Tpf,oc. is invertible in the algebra Ty(lK) = T(lK) and its inverse is
(Tpf,oc.)-l = Tg,oc. E Ty(lK), where the function 9 is determined by Formula (A5.8).
Theorem A5.4 and Proposition A5.2 together imply
Theorem A5.5. Suppose that all assumptions of Theorem A5.4 are satisfied.
Then Formulae (A5.1O.a), (A5.lO.b) hold on ker(D - >d).
Remark A5.2. Clearly, by Theorem A3.2, if all assumptions of Proposition A5.2
are satisfied and f(h) ¥- 0 for h E lK, then Formula (A5.11) holds on ker(D - AI).
o
Corollary A5.1. (cf. PR[8]) Suppose that all assumptions of Proposition A5.2
are satisfied, R E RD and A E lK n vcR. Then
Corollary A5.l implies that Theorems A5.4, A5.5 and Remark A5.2 hold also for
E).,(R). If J[( = C then Theorems A5.4, A5.5 and Remark A5.2 hold for the set
E= E(R). Clearly, if D has a Volterra right inverse R then Corollary A5.l holds
for all A E C (cf. PR[8]).
Definition (A3.4) of the set Soc.(D), Formula (A5.5) and Proposition A2.5 together
imply
302 Functional shifts
Proposition A5.4. Suppose that all assumptions of Theorem A2.1 satisfied and
the operator D is closed. Then Formula (A5.13) holds on the set SK(D).
Applying Theorem A3.4, in a similar way as Theorem A5.I (cf. Theorem A5.2),
we prove
Theorem AS.6. Suppose that X is complete linear metric space, D E R(X) is
closed, ker D -:j:. {O}, F is an initial operator for D corresponding to an R E RD,
T/,K = {T/,hhEK is a family offunctiona1 shifts induced by a function f E H(IK),
f(1)(O) -:j:. 0 and f(h) -:j:. f(O) for h E IK \ {O}. Let Y = SK(D). Then Tp/,K is
invertible in the algebra Ty and (Tp/,K)-l = Tg,K E T y , where the function 9 is
defined by (A5.B).
Theorem A5.6 and Proposition A5.4 together imply
Theorem AS.7. Suppose that all assumptions of Theorem A5.6 are satisfied.
Then Formulae (A5.lO.a), (A5.lO.b) hold on the set SK(D).
Remark A5.3. Suppose that all assumptions of Theorem A2.I are satisfied, the
operator D is closed and f(h) -:j:. 0 for hE lK. By Theorem A3.4, Formula (A5.H)
holds on SK(D). 0
Proposition A5.5. Suppose that X is complete linear metric space, D E R(X),
ker D -:j:. {O}, R E RD is continuous and T/,K = {T/,hhEK is a family of continuous
functional shifts for D induced by a function f E H (1K). Then
Proof. Fix x E AR(D) and set y = Dx. The continuity of R E RD implies that
AR(D) c SPIeD), D[AR(D») c AR(D). The first inclusion follows from Theorem
A2.2 and Formula (A5.5), the second one is proved in PR[9], Theorem 3.2. This,
and Formula (A5.4) together imply that for h E IK \ {O} we get
[(Pf)(hD)D)x = (Pf)(hD)y
= L: = L:
00 00
n=l
=h-1[f(hD) - f(O)I]x = ~T/,hX.
For h = 0 Formula (A5.I5) follows from the definition.
Theorem A5.8. Suppose that all assumptions of Proposition A5.5 are satisfied,
•
f(1)(O) ¥ 0 and f(h) -:j:. f(O) for h E IK \ {O}. Then Th,K is invertible in the
algebra TAR (D) and [Th,K]-l = T;,K E TAR(D)' where the function 9 is defined by
(A5.B).
Theorem A5.8 and Proposition A5.5 together imply
Appendix 303
Theorem A5.9. Suppose that all assumptions of Proposition A5.5 are satisfied,
f(l)(O) f=. 0 and f(h) f=. f(O) for h E lK. \ {OJ. Then Formulae (A5.1O.a) and
(A5.1O.b) hold on AR(D).
Remark A5.4. Suppose that all assumptions of Proposition A5.5 are satisfied
and f(h) f=. 0 for h E lK.. By Theorem A5.S, Formula (A5.11) holds on AR(D). 0
304 Functional shifts
Write
o form=O;
8 {~m
(8h P) (h) = jXj(h - a)j-1 for m ~ 1;
The operators
8 8- 1
8h'
8h
are said to be the first formal derivative and the formal integral of P with respect to
h, respectively. If these operators are applied n times, then the resulting operators
are denoted by
8h n '
8h n '
and are called the nth formal derivative and the nth formal integral, respectively.
Let OC = OC(a;r) (0 < r::; 00). Write for j E H(OC), hE OC
=L
00
L
00
L
00
If it does not lead to any misunderstanding, then we shall call extended functional
shifts briefly functional shifts. 0
Remark A6.I. By definition of the set S, the sums in Definition A6.1 consists
only of a finite number of terms different than zero. 0
In a similar way as Theorem A2.1, we prove
Theorem A6.I. Suppose that all assumptions of Definition A6.1 are satisfied,
T;:i (p E Z) is a family of extended functional shifts for Dinduced by f
and is F
an initial operator for D corresponding to an R E RD. Then
(i) for all h E lK, kENo, n E N
L
k
aj(h - a)i Rk-j F for p = OJ
j=O
o for p = n > kj
k
L j(j - 1) ... (j - n + l)aj(h - a)j-n Rk-j F for p = n ~ kj
j=n
k
" aj (h _ a)i+n Rk-j F for p = -nj
f::o (j + l)(j + 2) ... (j + n)
306 Functional shifts
L
k
aj(h - a)jRk-jz for P = OJ
j=O
o for p = n > k;
L
k
j(j - 1) ... (j - n + l)aj(h - a)j-n Rk-j z for p = n $ kj
j=n
L
k
aj (h - a)i+n Rk-j z for p = -nj
j=O (j + l)(j + 2) ... (j + n)
w;;i
(iii) if a family = {W;;~hEK c L(X), p E Z, satisfies for an Rl E RD and
for all h E ][{, kENo, z E ker D, n E N on the set S the conditions
L
k
aj(h - a)j R~-j z for p = 0;
j=O
o for p = n > kj
L
{p)Rkz - k
W f,h 1 -
j(j - 1) ... (j - n + l)aj(h - a)j-n R~-j z for p =n $ kj
j=n
k
" aj (h _ a)i+nRk-jz £or p
~ (j
j=O
+ 1)(j + 2) ... (j + n) 1 = - n',
then
W{p) -
f,h -
T{P)
f,h
on S far all h E ][{, p E Z.
Remark A6.2. Observe that the definition of the family (p E Z) depends T;;i
essentially on the disk][(' For example, let f(h) = h (h E q. Then the function
has the Taylor expansion
hz if Tf,h E Tf,K';
Tf,h Rz = { Rz + (h - l)z 1'f T f,h E T f,K" .
o
An immediate consequence of the definition are
Proposition A6.I. Suppose that all assumptions of Theorem A6.1 are satisfied.
Then
Appendix 307
(i) the operators Tj:;:), mE Z, h E OC, are uniquely determined on the set 8;
(i) T~7~1'9,h = >'T):;:) + /LT;:;:) for all I,g E H(OC), >',/L E C, hE OC;
(ii) for 1 E H(OC), hE OC, and n E N
T(-n) _ 8- 1 T(-n+1) _ 8- n T .
',h - 8h ',h - 8h n ',h,
(iii) the operators T):;:) commute with D on the set 8 for all h E II{.
We admit here and in the sequel that all assumptions of Definition A6.1 are satis-
fied and TjZ:?,K (m,p E Z) is a family of extended functional shifts for D induced
by I.
Corollary A6.1. Let n E N. Then for all hE OC we have on the set 8
T(m)
,(n-1) ,h for mEN;
T,(m) - {
(n) -1 h - -m
[J 1, T(m) _ l(n-1)( ) (h - a) I
for -m E No.
,(n-1),h a (-m)!
o for mEN;
T (m)
1,h =
{ (h - a)-m
(-m)! I for -m E No·
Formula A6.1 and Proposition A6.2(i) together imply that on 8 for all h E OC we
have
T (-n)R k
f,h Z
~ (h - a)i+n k- .
=L...J a· R Jz
j=O J (j + 1) ... (j + n)
k+n ).
= "" J. n (h - a J RHn-j
~ a - (j _ n + l)(j _ n + 2) ... j Z
J=n
= [Tf(_n),hRnjRkz.
Lemma A6.1 and the equality [j(-m)j(-n) = f(-m-n) for m, n E No together imply
•
Proposition A6.3. Let m, n E No. Then
T f(-m),h
(-n)
= Tf(-m-n),h R n S
on, h E lK, R E RD are arbitrary.
(-m-l) _ [T(-m)
T f(n),h (n - I)! (h )mljR S
- f(-n-l),h - m! an-l - a on .
Proof. Observe that f(q)(a) = q!a q and Ti"q) = (q!)-l(h - a)Q[ on S for all
q E No, where, as before, l(h) == 1 on lK. Fi~ h E lK and R E RD. Lemma A6.2
and Corollary A6.1 together imply that on S
[ T(n+l) R]Rkz
f,h
= TthH ) RkH z
k+1
= L j(j - 1) ... (j - n)aj(h - a)j-n-1 Rk+1-j z
j=n+1
L
k
= (j + l)j ... (j + 1 - n)aH1 (h - a)j-n Rk-j z
j=n
=E
k
j(j - l) ... (j - n + l)[(j + l)ajH](h - a)j-n Rk-jz
j=n
(n) Rk
T !(I),h
= Z.
[T(1) R]Rkz
j,h
k+1
= TX2Rk+1 z =E jaj(h - a)j-1 RkH-j z
j=1
k
= '
~ " ' k-JZ=Tj(I),hR
(j+l)aH1(h-aFR ' k z.
j=O
Suppose Formula (A6.3) to be true for an arbitrarily fixed n ;::: 1. Then, by (A6.4),
on S
T j(n+I),h = 7'
.L(j(l)](n),h = T(n)
j(l),h
Rn
= T(nH)RRn
!,h = T(n+1)Rn
j,h
+ 1.
Lemma A6.3 and the equality [j(m)](n) = j(m+n) for n E Nand m E Z together
•
imply
Proposition A6.5. Let n E N, mE Z, R E 'RD. Then for all hE lK
(A6.5)
Proof. Let h E 1K, R E 'RD be arbitrarily fixed. Propositions A6.2(iii) and A6.5
together imply that on S for q, mE No and p E Z we have
=T(m+ q) Dm+qRm+q
j(p-q),h
- Dm+qT(m+ q)
- j(p-q),h
-- Dq+mTj(p-q+m+q) ,h
-- Dq+mTj(p+m),h'
Theorem A6.3. Let q E No, m,p E Z. Then for all hE IK Formula (A6.5) holds
•
on S. If, in addition, q + m ~ 0 then for all h ElK Formula (A6.6) holds on S.
Proof. Fix h E IK and R E 'RD. Propositions A6.2(iii) and A6.3 together imply
that
D qT(m)
j(p),h = DqTj(p+m),h R- m = T j(p+m),h DqR- m .
Again by Proposition A6.(iii) and by Theorem A6.2, for q ~ -m we get
q) T(m+ q)
T j(p+m),h D qR-m = T j(p+m),h R-m-q = T(m+
f(p+m),h = j(p-q),h'
(-m-l) - T(-m)
DTj(-p),h (p - 1)1 (h )m]
- j(-p-l),h - ml ap-l - a .
o
Proposition A6.6. Let n E N. Then for all h E IK
Appendix 311
Proof Fix h E IK, R E RD. Proposition A6.2(ii), (iii) and Theorem A6.3 together
imply that
Now we compute derivatives and integrals for these functional shifts. The equality
TPZ = 0 on S for h E IK and Theorem A6.2 together imply that on S for all h E IK
we' have
(A6.7)
(A6.9) (n EN),
312 Functional shifts
(A6.1O)
o
Let an operator T(rn) mapping the space X into the set of operators defined on OC
and with values in X be defined by means of the equality:
T ,(rn) x = {T(rn)}
',h X hEIK
j or x E X .
(A6.11)
(iii) if OC1 = IK(b; rI) ~ lK then T,(x; h) = TjO) (x; h) has the Taylor expansion
~ (h - b)k (k)
T,(x;h)=L...J k! T, (x;b) jarhEOC1,xES.
k=O
Lemma A6.4. Let X be a complete linear metric space and let X' be the space
of all linear functionals defined on X. Let D E R(X), ker D =I {O}. Let kENo,
z E ker D, x' E X', R E RD be arbitrarily fixed. Then x' [T,(Rkz; .)] as a complex-
valued function is holomorphic in IK and
Proof. Let h E ][{, R E RD be arbitrarily fixed. Let {(n} be a sequence such that
h + (n E lK for all n E N and lim (n = O. Write t = h - a, tn = t + (n' By
n-too
Theorem A6.1(ii), we have for k ~ 1
Appendix 313
k k
= l!..m
n~oo
(;:;-1 [x' (2: ajt~Rk-j 2: ajtj Rk-j z)]
Z -
j=O j=O
k k
= J~~ (;:;-1[2: ajt~x'(Rk-jz) - 2: ajtjx'(Rk-jz)]
j=O j=O
k
= J~~ (;:;-1 [2: aj(t~ - tj)x'(Rk-j z)]
j=l
Proposition A6.9. Suppose that all assumptions of Lemma A6.4 are satisfied
•
and x E 5, x' E X' are arbitrarily fixed. Then x'[T, (x; .)] as a complex-valued
function is holomorphic in ][{ and
by Lemma A6.4,
d~ x'[T,(xj h)]
d
2:
n
= dh X' [T,(R k Zk)]
k=O
= t d:
k==O
x' [T,(Rkzk j h)] = t
k=O
x'[DT,(l)(Rkzk)]
n
= x' [DT,(1) (2: RkZkjh)) =x'[DT,(l)(xjh)].
k=O
Lemma A6.5. Suppose that all assumptions of Lemma A6.4 are satisfied. Then
•
the mappingT,(Rk zj ') is strongly holomorphic in][{, i.e. for every hE ][( the limit
T,(Rk Zj h) - T,(Rk Zjw )
lim
w-th h- w
exists and
(A6.12) d k k
dh T,(R Zj h) = DT,(l)(R Zj h) far all h E ][{.
Proof Let h E ][{, R E RD be arbitrarily fixed. Let {(n} be a sequence such that
h + (n E ][{ for all n E N and lim (n = O. Write
n--+oo
(h + (n - a)j - (h - a)j .
tn,j = (n aj (J = 0,1, ... ,kj n EN).
k
- 2: aj(h-a)iRk-jz]
j=O
= n--+oo
lim "t
k
L.J n ,)
·Rk-j Z
j==O
=" [lim
k
L.J n-too
tn ,)·]Rk-j Z
j==O
k
= 2: jaj(h - a)j-1 R k - 1-j Z
j=1
= DT,(1) (Rkzj h).
Appendix 315
Formula (A6.12) is true also for k = 0 (cf. the proof of Lemma A6.4). •
Lemma A6.5 implies (cf. the proof of Proposition A6.9)
Proposition A6.10. Suppose that all assumptions of Lemma A6.4 are satisfied
and xES is arbitrarily fixed. Then the mapping Tf(x;·) is strongly holomorphic
in OC and
d
dh Tf (x; h) = DTf (1) (x; h) for all h E K
l g«()d(.
Since the space X* is total, i.e. X* separates points of X (cf. Lemma 4.1), the
element y is uniquely determined and called the integral of 9 over r. 0
Proposition A6.11 Let r c ][ be an oriented arc with the initial point band
the final point c. Let X a complete linear metric locally convex space. Then for
every xES the integral fr T,(x; h)dh exists. If x E Xn then
where
(c - a)k+l - (b - a)k+l
Ck = ak k +1 (k = 0,1, ... , n).
Proof. Let
n
X =L Rk Zk E Xn C S, x· E X*
k=O
be arbitrarily fixed. By definition and Proposition A6.9, we have
r x*[T,(x; h)]dh
lr
n n
=L ckx*(Dkx) = x*(L ck Dkx ).
k=O k=O
o
Appendix 317
1
IndrCw) = -2'
ITt
i
r
d(
-(-.
- w
IndrCw) 1. { ~ =0
= -2 far all w t/. IK,
ITt lr (- w
£ £T~~;~)
then
Tf(x; ()d( =0 and Tf(x; h) = 2~i d(
Theorem A6.6 can be extended for the whole space. For instance, we have
Corollary A6.2. Suppose that all assumptions of Theorem A6.6 are satisfied.
IES = X and {AhhEK C Lo(X) is a family of uniformly continuous operators
I
such that Ah s = Tf,h for h E IK, then
We shall consider now an indefinite integral of the mapping TJ(x;,) for xES.
Theorem A6.7. Suppose that all assumptions of Theorem A6.6 are satisfied. For
all XES, R E RD
TjC-I) (Rx; h)
(h - a)k
=L
n+1 k
ak-l k D Rx
k=l
=L
n+l
ak-l (h - a)kk-l Dk-1X
k=l
(h _ a)k+l
L
n
= ak k 1 Dkx
k=O +
= lh Tj(x; ()d( (h E OC).
The equality T;-l) (x; h) = Tj(-l)(Rx;h) for hE II{ follows from Lemma A6.1. •
Propositions A6.6 and Proposition A6.1O together show relations between the
formal derivative, the strong derivative and the operator D. Namely, we have
Corollary A6.3. Suppose that all assumptions of Theorem A6.6 are satisfied.
For all h E OC
J d
c5h Tj,h = dh Tj,h = DTjCI) ,h on S.
Propositions A6.7 and Theorem A6.7 together show relations between the formal
integral, the integral and a right inverse of D.
Corollary A6.4. Suppose that all assumptions of Theorem A6.6 are satisfied.
For all h E OC
Extensions of Corollaries A6.3 and A6.4 to the whole space X can be obtained in
a similar way as Corollary A6.2.
Similarly, as Theorems 18.2 and 18.3, we get
Corollary 18.1. Let j, g E H (OC). Then the Leibniz condition holds in the
following sense:
Corollary 18.2. Suppose that all assumptions of Theorem A6.6 are satisfied. Let
j, 1/ j E H(K). Then for all h E OC
{} d
c5h 1inj,h = dh 1inj,h = DT(lnf)(I),h = DT/'Ij,h on S,
Appendix 319
i.e.
(n)T(m) - Dn+mT
T J,h g,h - f(n)g(n),h if m,n E Ill;
= {TJ(P),hTg(q)'hDP+q if p + q ~ 0;
if p+q > 0;
(q)T(P)
T J,h g,h -- T J(q),h R-qTg(p),h DP'f "T "T
Z P E 1'1, -q E 1'1;
(P)T(-q)
T J,h g,h -- T f(-p),h RPTg(-q),h Rq'f
Z p,q E "T
1'1.
o
Example A6.2. Let the families fO, Co, so, ch o , shn be defined as in Example
A6.1 and let R E RD be arbitrarily fixed. For all xES we have
d I (fh - I)x
d h ch h-Ox = lim h = Dcox = Dlx = x,
- h-too
d d d d
dh Shxlh=O = Dx, dh ShhXlh=O = Dx, dh Chxlh=o = 0, dh ChhXlh=o = 0
(cf. Corollary A6.2, Formula (A6.9)). By Corollary A6.3 and Formula (A6.1O),
for all xES we have
o
Now we shall study canonical mappings (cf. PR[7]) generated by a family of
extended functional shifts.
Definition A6.3. Suppose that F is an initial operator for D E R(X) corre-
sponding to an R E RD. A mapping I\;j~~) of X into the set of operators defined
320 Functional shifts
if q = 0;
o
In the sequel we admit that all assumptions of Definition A6.3 are satisfied. From
this definition it follows immediately
Proposition A6.l2. Let p,q E Z. Then the function ",j~~)(x;.) of a complex
variable h E IK and with values in X defined in Definition A6.3 has the following
properties:
(i) for all X,y E X, >',/.L E C
",(q) (x' h)
>'f+l'g'
= )..",(q)
f(x' ' 'h)'+'1J",(q)
9
(x' h)
, •
Proposition A6.l3. The restriction of the canonical mapping ",j~~) to the set S
of all D-polynomials has the following properties:
OJ for all I E H (IK) and xES
=L
00
n n
= L ak(h-a)kFDkx= L adh-a)kFDkx
k=O k=m
n n
=L ak(h-a)kFDk(L Rjzj )
k=m j=O
n n
= L ak(h - a)k F( L Rj-k Zj )
k=m j=k
n n
= L ak(h-a)k(L FRj-k Zj )
k=m j=k
n n
= L ak(h - a)k FZk =L ak(h - a)kzk'
k=m k=m
where am =f. 0, Zj =f. 0 (j = m, m + 1, ... , n). This shows that am+l = ... = an = 0
for n > m implies that /'i,f(x; h) = 0 (h E JK) for x =f. O. We therefore conclude
that 0 =f. x rt. ker "'I if and only if ak =f. 0 for all n E No.
Assertion (iv) follows immediately from (iii) and Proposition A6.12(i). •
Theorem A6.B. Let XES, x' E X' be arbitrarily fixed. Then x' [/'i,}m) (x; .)], as
a complex-valued function, is holomorphic in JK and
Theorem A6.9. The canonical mapping ",}m) (x; .) (m E No) is strongly holo-
morphic in JK and for all h E JK, xES
where
(c - a)k+l - (b - a)k+l
Ck = ak k +1 (k E No).
Indr(w) 1
= -2'
1rZ
1 =
r
-(d(
-
- w
0 for all w ~ K
Then
(n)( . ) _ n!
Kf x, h - 21ri r
1 «( _ Kf(X;()
h)n+l d( (n E No
)
=L
00 (h - b)k k
"'f(x; h) k! "'~ ) (x; b) where hE OC1 , xES.
k=O
324
References 325
Bin[9] Some properties of operators of complex differentiation and shifts. Sci. Bull.
Lodz. Technical Univ. Math., 24(1993),5-18.
Bin[10] On periodic solutions of equations with right invertible operators induced
by functional shifts. Demonstratio Math., 26(1993), 535-543.
Bin[ll] A unified approach to shifts induced by right invertible operators. Math.
Nachrichten, 161(1993), 239-252.
Bin[12] On some functional shifts induced by operators of complex differentiation.
Opuscula Math., 14(1993), 45-57.
Bin[13] Some remarks on sequential shifts induced by right invertible operators.
Functiones et Approximatio, 22(1993), 71-84.
Bin[14] Applications of sequential shifts to an interpolation problem. Collect.
Math., 44(1993), 47-57.
Bin[15] Some fundamental properties of functional R-shifts for right invertible
operators. Commentationes Math., 33(1993), 9-22.
Bin[16] A note of functional R-shifts for right invertible operators. Discussiones
Math., 14(1994),63-76.
Bin[17] On summation formulas induced by functional shifts of right invertible
operators. Demonstratio Math., 28(1995), 301-314.
Bin[18] On isomorphisms of spaces of functional R-shifts for right invertible oper-
ators. In: Different Aspects of Differentiability. Proc. Intern. Conf. Warszawa,
September 1993. Ed. D. Przeworska-Rolewicz. Dissertationes Math. 340,
Warszawa, 1995, 23-36.
Bin[19] Remarks on functional R-shifts for right invertible operators. In: Proc.
Intern. Conf. Different Aspects of Differentiability II. Warszawa, September 1995.
Integral Transforms and Special Functions, 1-2, 4(1996), 39-48.
Z. BINDERMAN and D. PRZEWORSKA-RoLEWICZ
BPR[l] The limit property and perturbations of functional shifts. Demonstratio
Math., 28(1995), 835-85l.
BPR[2] Almost quasinilpotent right inverses. In: Proc. Intern. Conf. Different
Aspects of Differentiability II. Warszawa, September 1995. Integral Transforms
and Special Functions, 1-2, 4(1996), 23-38.
T. CARLEMAN
Ca[l] Sur la resolution de certaines equations integrales. Arkiv Math. Astron.
Fys., 16(1922), 19.
Ca[2] Sur la theorie des equations integrales et leurs applications. Verhandl. In-
ternation. Math. Kongress, 1. Zurich, 1932, 138-15l.
A.L. CAUCHY
C[1] COUTS d' analyse de l'Ecole Poly technique. Ire partie, Analyse algebrique.
Paris, 1821; Oeuvres, (2)3, Paris, 1897.
J. B. CONWAY and B. B. MOREL
CM[l] Roots and logarithms of bounded operators on Hilbert space. J. Funct.
Analysis, 70(1987), 171-193.
C. DAVIS
Da[l] Explicit functional calculus. Algebra and Appl., 6(1973), 193-199.
326 References
A. DI BUCCHIANICO
DB[l] Banach algebras, logarithms, and polynomials of convolution type. J. Math.
Anal. Appl., 156(1991), 253-273.
DB[2] Polynomials of Convolution Type. Ph.D. Thesis, Groningen, 1991.
I. H. DIMOVSKI
Di[l] Convolutional Calculus. Kluwer Acad. Publishers, Dordrecht, 1990 (1st ed.
Publish. House of the Bulg. Acad. Sci., Sofia, 1982)
V. A. DITKIN and A. P. PRUDNIKOV
DP[l] 'fransformations Integrales et CaIcul Operationnel. Mir, Moscou, 1982;
Russian ed. 1972.
PINGPING DONG
Do[l] Minorations de combinaisons lineaires de logarithmes p-adiques de nombres
algebriques. Dissertationes Math. 343, Warszawa, 1995.
Z. DUDEK
D[l] On decomposition of quasi-Leibniz D-algebras. Demonstratio Math., 14
(1981),745-757.
D[2] On multiplicative operators in commutative algebras. Ibidem, 23(1990), 921-
928.
M. K. FAGE and N. I. NAGNIBIDA
FN[l] An Equivalence Problem of Ordinary Linear Differential Operators. (In
Russian), Nauka, Novosibirsk, 1987.
C. G. FRASER F[l] The calculus of algebraic analysis: some observations on math-
ematical analysis in the 18th century. Arch. Hist. Exact. Sci., 39(1989), 317-335.
H. GANGL
G[l] Families of functional equations for polylogarithms. Preprint 95-101, Max-
Planck Institut flir Mathematik, Bonn, 1995.
I. M. GELFAND and R. MACPHERSON
G MP [1] Geometry of Grassmannians and a generalization of the dilogarithm. Ad-
vances in Math., 44(1982), 279-312.
I. G RADSTEIN and I. RYZHIK
GR[l] Tables of Integrals, Sums, Series and Products. (In Russian), Nauka,
Moskva, 1971.
W. HAACK and W. WENDLAND
HW[l] Partial and Pfaflian Differential Equations. Pergamon Press, Oxford-New
York, 1972.
M. HANAMURA and R. MACPHERSON
HMP[l] Geometric construction of polylogarithms, II. In: Functional Analysis on
the Eve of 21th Century. In Honor of the Eightieth Birthday of 1. M. Gelfand.
Eds. S. Gindikin, J. Lepowsky and R. Wilson. Vol. II (Progress in Mathematics
132). Birkhauser, Boston-Basel-Berlin, 1996, 215-282.
References 327
M. G. HAPLANOV
H[l] Linear Transformations of analytic spaces (In Russian). Dokl. AN SSSR,
80(1951), 21-24.
E. HILLE and R. S. PHILLIPS
Functional Analysis and Semigroups. AMS Colloquium Publications, v. 31, Prov-
idence, R.I., 1957 (reprinted 1985).
E. L. INCE
In[l) Ordinary Differential Equations. (Longmans, Green 1926), London, 1927.
H.N.JAHNKE
J[l) Algebraic analysis in Germany, 1780-1840: some mathematical and philosoph-
ical issues. Historia Math., 20 (1993), no. 3, 265-284.
I. KAPLAN SKY
K[l) An Introduction to Differential Algebra. Hermann, Paris, 1957.
D. KARAYANNAKIS
Ka[l) Logarithms of isometries and their spectral decompositions in Hardy spaces.
Ann. Mat. Pura Appl., 165(1993), 1-2l.
W. Z. KARWOWSKI and D. PRZEWORSKA-RoLEWICZ
KPR[l) Green operator for linear boundary value problems with a right invertible
operator DN. Math. Nachrichten, 152(1991), 21-34.
KPR[2) General linear boundary value problems for polynomials in a right invert-
ible operator. Demonstratio Math., 25(1992), 325-340.
W. KIERAT and K. SKORNIK
KS[l) A remark on solutions of Laguerre differential equations. In: Different
Aspects of Differentiability. Proc. Intern. Conf. Warszawa, September 1993. Ed.
D. Przeworska-Rolewicz. Dissertationes Math. 340. Warszawa, 1995, 137-141.
V. KIRYAKOVA
Kir[l] Generalized Fractional Calculus and Applications. Pitman Research Notes
in Mathematics 301. Longman, Harlow-London, 1994.
I. I. KOLODNER
Kol[l) On exp (tA) with A satisfying a polynomial. J. Math. Anal. Appl.,
52(1975), 514-524.
H. KORNACKI
Kor[l) Non-Leibniz components in non-commutative algebras with unit. Demon-
stratio Math., 18(1986), 483-497.
M. KUCZMA
Ku[l) Functional Equations in a Single Variable. PWN-Polish Scientific Publish-
ers, Warszawa, 1968.
Ku[2) An Introduction to the Theory of Functional Equations and Inequalities.
Cauchy's Equation and Jensen's Equation. PWN-Polish Scientific Publishers and
the Silesian University, Warszawa-Krak6w-Katowice, 1985.
328 References
J.P. LAGRANGE
L[l] TMorie des Fonctions Analytiques con tenant Les Principes du Calcul Differ-
en tiel, degages de toute consideration d'infiniment petits, d'evanouissans de limites
et de fluxions, et reduit a l'analyse algebriques de quantites finies. 2nd revised
and enlarged ed., Mme Ve Courcier, Imprimeur-Librairie pour les Mathematiques,
Paris, 1813 (1st ed. 1797).
H. LAUSCH, D. PALLASCHKE and S. ROLEWICZ
LPaR[l] Universal derivations in spaces of functions with values in non-commuta-
tive Banach algebras. In: Proc. Intern. Conf. Different Aspects of Differentiability
II. Warszawa, September 1995. Integral Thansforms and Special Functions, 1-2,
4(1996), 115-120.
H. LAUSCH and D. PRZEWORSKA-RoLEWICZ
LPR[l] Pseudocategories, paraalgebras and linear operators. Math. Nachrichten,
138(1988), 67-87.
LPR[2] Some functional equations appearing in Algebraic Analysis. Opuscula
Math., 6(1990), 111-122.
S. J. LEE and M. Z. NASHED
LN[l] Algebraic and topological selections of multivalued linear relations. Ann.
Scuola Norm. Sup. Pisa Cl. Sci., (4) 17(1990),111-126.
L. LEVIN
Lev[l] Structural Properties of Polylogarithms. AMS monographs, Providence,
R.I., 1991.
N. LINCHUK
Li[l] Representation of commutants of the Pommiez operator and their applica-
tions. (In Russian). Mat. Zametki, 44(1988), 794-802.
D. E. LOEB
L[l] The iterated logarithmic algebra. Advances in Mathematics, 86(1991), 155-
234.
L[2] The iterated logarithmic algebra. II. Sheffer Sequences. J. Math. Anal.
Appl., 156(1991), 172-183.
D. E. LOEB and G.-C. ROTA
LRo[l] Formal power series of logarithmic type. Advances in Mathematics,
75(1989), 1-118.
G. LUMER and M. ROSENBLUM
LR[l] Linear operator equations. Proc. Amer. Math. Soc., 10(1959), 32-4l.
E. MEISTER
Me[l] Randwertaufgaben der Funktionentheorie. B. G. Teubner, Stuttgart, 1983.
N. MERCATOR (or: N. KAUFMANN)
Mr[l] Logarithmo-technia. Londini, 1668.
References 329
M. MICHALSKI
M[l] Derivatives of noninteger order and their applications. Dissertationes Math.
328, Warszawa, 1993.
S. G. MICHLIN
Mi[l] Integral Equations and their Applications. (in Russian), 2nd ed. OGIZ,
Moskva-Leningrad, 1949.
J. B. MILLER
Mil[l) The standard summation operator, the Euler-MacLaurin sum formula and
the Laplace transformation. J. Australian Math. Soc., 39(1985),376-390.
J. NAPIER (or: NEPER)
Np[1] Mirif1ci logarithmorum canonis descriptio. Lugduni, 1614.
NGUYEN VAN MAU
N[l] Generalized algebraic elements and singular integral equations with trans-
formed argument. Warsaw Univ. of Technology Publications. Warszawa, 1989.
N[2] Boundary value problems and controllability of linear systems with right
invertible operators. Dissertationes Math. 316, Warszawa, 1992.
W. POGORZELSKI
P[1) Integral Equations and their Applications. (in Polish; 4 volumes, Warsaw,
1953-70) English Ed. (in one volume vol. 1-3), Pergamon Press and PWN-Polish
Scientific Publishers, Oxford-Warszawa, 1966.
S. D. POISSON
Poi(1) Memoires sur les equations aux differences melees. Journal de l'EcoIe PoIy-
technique, 6, Cahier 13 (1806), 126-147.
T. POULSEN and H. STETKJER
PS(1) Functional equations on abelian groups with involutions. Matematisk Insti-
tut, Aarhus Univ. Preprint Ser. 1995, N° 16. .
A. PRINGSHEIM and G. FABER
PF[l] II C 1. Algebraische Analysis. In: Encyklopiidie der Mathematischen Wis-
senschaften mit Einschluss ihrer Anwendungen. Zweiter Band Analysis. Eds. H.
Burkhardt, M. Wirtinger, R. Fricke, R. Hilb. Teubner, Leipzig, 1909-1921.
A. P. PRUDNIKOV
Pru[l) On the continuation of the ideas of Heaviside and Mikusinski in Opera-
tional Calculus. In: Different Aspects of Differentiability. Proc. Intern. Conf.
Warszawa, September 1993. Ed. D. Przeworska-Rolewicz. Dissertationes Math.
340, Warszawa, 1995, 237-287.
A. P. PRUDNIKOV, Yu. A. BRICHKOV and O. I. MAL'CEV
PBM[l] Integrals and Series. Complementary Chapters. (In Russian). Nauka,
Moskva, 1986.
D. PRZEWORSKA-RoLEWICZ
PR[l) Probleme non-lineaire d'Hilbert pour Ie systeme infini de fonctions. Annales
Polan. Math., 5(1958), 293-301.
330 References
PR[2] Sur l'integrale de Cauchy pour un arc ferme it l'infini. Annales Polon. Math.,
8(1960),155-171.
PR[3] Sur les equations involutives et leurs applications. Studia Math., 20(1960),
95-117.
PR[4] Generalized linear equations of Carleman type. Bull. Acad. Polon. Sci.,
20(1972), 635-639.
PR[5] A characterization of commutators with Hilbert transform. Studia Math.,
44(1972), 27-30.
PR[6] Equations with Transformed Argument. An Algebraic Approach. PWN-
Polish Scientific Publishers and Elsevier Scientific Publish. Compo Warszawa-
Amsterdam, 1973.
PR[7] Shifts and Periodicity for Right Invertible Operators. Research Notes
in Mathematics, 43. Pitman Advanced Publishing Program, Boston-London-
Melbourne, 1980.
PR[8] Algebraic Analysis. PWN-Polish Scientific Publishers and D.Reidel. War-
szawa-Dordrecht, 1988.
PR[9] Spaces of D-paraanalytic elements. Dissertationes Math. 302, Warszawa,
1990.
PR[10] Advantages of one-dimensional kernels. Math. Nachrichten, 149 (1990),
133-147.
PR[11] Property (c) and interpolation formulae induced by right invertible oper-
ators. Demonstratio Math., 21(1988), 1023-1044.
PR[12] Commutators with right invertible operators. J. Math. Anal. Appl.,
158(1991),414-426.
PR[13] True shifts. J. Math. Anal. Appl., 170(1992), 27-48.
PR[14] Generalized Bernoulli operator and Euler-MacLaurin formula. In: Ad-
vances in Optimization. Proc. The 6-th French-German Conf. on Optimization,
Lambrecht(Landau), FRG, June, 1991. Lectures Notes in Economics and Math.
Systems, 382. Springer-Verlag, Berlin-Heidelberg-New York, 1992, 355-368.
PR[15] The operator exp(hD) and its inverse formula. Demonstratio Math.,
26(1993), 545-552.
PR[16] D-algebras with logarithms. Math. Nachrichten, 161(1993), 321-344.
PR[17] Remarks on commutators with operators having either finite nullity or
finite deficiency. Linear Algebra and its Applications, 208/209(1994), 199-207.
PR[18] Logarithmic and antilogarithmic mappings. Dissertationes Math. 337,
Warszawa, 1994.
PR[19] Consequences of Leibniz condition. In: Different Aspects of Differentia-
bility. Proc. Intern. Conf. Warszawa, September 1993. Ed. D. Przeworska-
Rolewicz. Dissertationes Math. 340, Warszawa, 1995, 289-300.
PR[20]. Introducing to Different Aspects of Differentiability. In: Different As-
pects of Differentiability. Ed. D.Przeworska-Rolewicz. Dissertationes Math. 340.
Warszawa, 1995,9-12.
PR[21] Some applications of logarithmic and antilogarithmic mappings to nonlin-
ear equations in Leibniz D-algebras. Workshop on Transform Methods and Special
Functions. Sofia, August 12-17, 1994. Eds. P. Rusev, 1. Dimovski, V. Kiryakova.
SCTP, Singapore, 1995,238-252.
References 331
P. SMERL
S[I] The functional equation of multiplicative derivation is superstable on standard
operator algebras. Integral Equations Operator Theory, 18(1994), 118-122.
S[2] Isomorphism of standard operator algebras. Proc. Amer. Math. Soc.,
123(1995), 1851-1855.
J. H. M. STEENBRINK
Sb[I]Logarithmic embeddings of varieties with normal crossings and mixed Hodge
structures. Math. Ann., 1, 301(1995), 105-118.
H. STETKiER
St[l] D' Alembert's and Wilson's functional equations for vector and 2 x 2 matrix
valued functions. Matematisk Institut, Aarhus Univ. Preprint Ser. 1996, NO 28.
G.!. TARGONSKI
T[I] Seminar on Functional Operators and Equations. Lecture Notes in Math. 33,
Springer Verlag, Berlin, 1967.
L. VERDE-STAR
VS[I] An algebraic approach to convolutions and transform methods. Preprint,
Dept. of Math., Universidad Aut6noma Metropolitana, Mexico City, 1996; Ad-
vances in Applied Mathematics (to appear).
G. VIRSIK
V[I] Right inverses of vector fields. J. Austral. Math. Soc., 58(1995),411-420.
E. WEGERT
Weg[l] Nonlinear Boundary Value Problems for Holomorphic Functions and Sin-
gular Integral Equations. Akademie Verlag, Berlin, 1992.
Weg[2] Nonlinear Riemann-Hilbert problems with unbounded restriction curves.
Math. Nachrichten, 170(1994),307-313.
W. L. WENDLAND
Wen[l] Elliptic Systems in the Plane. Pitman, London-San Francisco-Melbourne,
1979.
J. WILDESHAUS
Wh[l] Polylogarithmic extensions on mixed Shimura varietes. Schriftenreihe des
Math. Inst. der Universitat Munster, 3. Serie, 12, Munster, 1994.
Wh[2] Realization of polylogarithms. Lectures Notes in Math. 1650, Springer
Verlag, Berlin, 1997.
J. WILKOWSKI
Wi[l] On periodic solutions of differential-difference equations with stochastic co-
efficients. Bull. Acad. Polon. Sci., 21(1973), 253-256.
A. WLODARSKA-DYMITRUK
WD[l] Polynomial-periodic solutions of differential-difference equations. Demon-
stratio Math., 8(1975), 44-65.
H. WYSOCKI
Wy[l] A matrix operational calculus. Acta Math. Hungar., 64(3)(1994), 285-303.
References 333
K. YOSIDA
Yl1] Functional Analysis. Springer Verlag, Berlin-Gottingen-Heidelberg, 1965.
Yu SUN TONG
Toll] Higher-order generalized derivations. Chinese Ann. Math., Ser. A, 6(1985),
566-572; An English summary in Chinese Ann. Math., Ser. B, 6(1985), no. 4,
493.
D. ZAGIER
Z[l] The Bloch-Wigner-Ramakrishnan polylogarithm function. Math. Ann.,
286(1990), 613-624.
SUBJECT INDEX
334
Subject index 335
co dimension 52 coordinatewise
coefficients addition 34
of D-analytic element 196 convergence 65
co-initial operator 3 multiplication 19
common multiple 286 by scalars 34
commutant 78 coset 31
cosine
commutative
element 136
algebra viii, 5
mapping 136
group 149
cotangent Hilbert transform 214
semigroup 149 Cramer system 55
commutator 31 cyclic group 243
complete linear metric space ix, 32
complex D-algebra 9
D-analytic element 160
differentiation 285
D-invariant space 192
extension 146
D-paraanalytic element 187
integration 285
D-polynomial 41
complex of degree n 41
number 1 D-polynomial-exponential element 232
R-shift 264 D-polynomial-exponential-periodic
component of a set 107 element 233
condition (A) 203 D-shift 264
condition (A)l 206 D-R invariant space 197
condition (A)r 206 D-R structure
condition [Cln 136 preserving isomorphism 198
condition [L] 101 deficiency 52
condition [L]# 107 de Moivre formula 49, 136
condition [M] 83 determinant 54
Di Bucchianico theorem 107
condition [M]# 89
diffeomorphism 248
conjugate
difference operator 5
bounded linear operators 108 differential-difference equation 200
harmonic function 215 dimension of kernel 4
imaginary roots 131 direct sum 1
space 54 disjoint projectors 118
constant 1 disk 259
mapping 22 domain
function 4 of multifunction 11
sequence 5 of operator 1
constants variation method 128 with boundary
continuous of Liapunov type 77
functional 287 Duhamel condition 30, 112
right inverse 5 eigenspace 118
convolution 4, 34, 114 eigenvalue 104
336 Subject index
initial left
condition 16 almost exponential type 96
operator 2 antilogarithm 13
point 316 of higher order 40
of order n 40
value problem 4
antilogarithmic mapping 13
integer 1 of higher order 40
integral of order n 40
equation 255 basic equation 11
over r 316 inverse of operator 3
remainder 5 invertible operator 3
integrable over r function 316 logarithm 13
intermediate value of higher order 40
of order n 40
of function 5
logarithmic
of shift 165 mapping 13
invariant metric 263 of higher order 40
inverse of order n 40
element 10 Leibniz
function 37 algebra 6
mapping 13 condition 5, 6
multifunction 11 D-algebra 9
formula 8
operator 56
Liapunov type boundary 77
invertible limit
element 10 point of sequence 277
mapping 13 property of operator family 277
operator 3 linear
selector viii, 13 equation with scalar coefficients 4
involution 31, 201 functional 54
metric space 32
of order N 216
operator 1
isomorphic algebras 271
ordinary differential equation 4
isomorphism 198 with periodic coefficients 218
ring 3
kernel of operator 1
space 1
Kronecker symbol 10
linearly independent
LAE-type 96 elements 54
functionals 54
Lagrange remainder 5
Liouville formula 132
Lagrange-Poisson formula 161 locally
Laguerre differential operator bounded space ix, 122
of the second order 155 convex space ix, 122
Laplace operator 77 pseudoconvex space ix, 122
338 Subject index
power right
function 170 almost exponential type 95
mapping 168 antilogarithm 13
type 168 antilogarithmic mapping 13
primitive 1 basic equation 11
principal space 118 inverse 1
invertible operator 17
product rule 6
logarithm 13
projection 2
of order n 40
projector 118
logarithmic mapping 13
proper
of order n 40
non-Leibniz component 7 Rolewicz theorem 262
two-sided ideal 31 Rolle theorem 5
quasinilpotent operator 156 root
of element 10
quasipolynomial 200
of quasipolynomial 218
quasi-Leibniz
rotation 249
algebra 39
D-algebra 9 *-modulus 146
quotient SN -periodic element 205
algebra 31 Sm-periodic operator 221
space 52 scalar
coefficient 4
RAE-type 95 matrix 109
R-shift 218 Schwarz derivative 184
range of operator 1 Schwarzian 184
rational selector viii, 13
function 3 semigroup 149
number 1 separation of points 160
real number 1 shift 160
rectifiable arc 317 similarity property 72
recurrence 7 simple Duhamel algebra 64
sine
reflection of argument 243
element 136
regular value of operator 27
mapping 136
remainder 5
single characteristic root 118
resolving
single-valued multifunction 38
equation 175 singular
operator 72 element 187
restriction of operator 274 integral
Riccati equation 179 equation 199
Riemann-Hilbert problem 201 operator 6, 201
Riemann-Liouville fractional Smerl theorem 108
integral 258 smooth element 159
340 Subject index
space symmetric
of linear functionals 54 domain of multifunction 11, 136
of all linear continuous form 8
functionals 287 matrix 109
of analytic functions 165 Targonski theorem 63
of coefficients of Taylor
D-analytic elements 196 expansion 161, 312
of constants 1 formula 2
of D-analytic elements 160 Taylor-Gontcharov formula 2
of D-paraanalytic elements 187 total space 54
of D-polynomials 159 transformation of argument 37
of D-polynomial-periodic triangle matrix 110
elements 232 trigonometric
of D-polynomial-exponential- element 136
periodic elements 233 identity 138
of exponentials 160 mapping 1364
of exponential-periodic true shift 160
elements 218 two-sided ideal 31
functions 218
uniform convergence
of periodic elements 216 on compact sets 170
of singular elements 187 on closed intervals 170
of smooth elements 159 uniformly continuous operators 317
special generalized unit 1
Clair aut equation 183 disk 259
Riccati equation 180 of algebra 15
square matrix 109
standard Vandermonde determinant 238
subalgebra 108 Volterra
operator sub algebra 108 operator 4
right inverse 4
stochastic differential-difference
continuous right inverse 5
equation 226
vulgar fractions 3
structure operation 160
strong differentiation 318 well-posed initial value problem 4
strongly continuous Wronski theorems 132
function 322 Wronskian 132
group 160 zero 1
semigroup 160 divisor 16
holomorphic mapping 314 element 1
sub algebra 38 of quasipolynomial 235
submultiplicative F-norm 122
summation
formulae 285
operator 162
superposition 8
symbol function 199
AUTHORS INDEX
341
342 Authors index
Przeworska-Rolewicz viii, 1, 5, 6, 8,
9, 16, 29, 31, 32, 34, 52, 66
72, 74, 77, 90, 92, 106, 112,
118, 122, 128, 132, 138, 143,
144, 147, 150, 154, 155, 156,
158, 160, 161, 162, 163, 165,
181, 187, 189, 190, 191, 196,
199, 200, 201, 202, 206, 210,
214, 216, 218, 224, 225, 226,
227, 228, 235, 243, 244, 245,
246, 247, 251, 255, 262, 264,
265, 269, 278, 279, 280, 281,
283, 284, 290, 296, 298, 301,
319,325,327,328,329,330,
331
Rizhik 298, 326
Rogulski 187, 331
Rolewicz vii, ix, 5, 52, 122, 163,
218, 262, 328, 330, 331
Roman 331
Rosenblum 32, 328
Rota 328, 331
Silbermann 108, 324
Sk6rnik 129, 155, 181, 327, 331
Smer! 108, 332
Steenhrink 332
Stetkrer 148, 329, 332
Targonski 63, 332
Tong 32,333
von Trotha 106, 122, 198, 331
Ulam vi
Verde-Star 332
Virsik 75, 332
Wegert 201, 215, 332
Wendland 201,326,332
Wildeshaus 332
Wilkowski 226, 332
Wloka 130, 181, 331
Wlodarska-Dymitruk 228, 332
Wysocki 155, 332
Yosida 165, 333
Yu Sun Tong 32, 333
Zagier 173, 333
Zelazko 163
LIST OF SYMBOLS
d .
<IT IX,
1 A(X) 6
C 1 IV(x, y) 6
IF ix, 1 CD 6
N 1 Ibn) 7
Q 1 Aa 9
IR 1 M(X) 10
/l 1 ML(X) 10
No 1 leX) 10
1141 In(Y) 10
L(X) ix, 1 x l / n 10
dom A 1 a+- 10
ker A 1 IF[tJ 10
Lo(X) 1 Oij 10
R(X) 1 n1 11
RD 1 nr 11
EB 1 n viii, 11
FD 1 n-I 1 11
A(X) 3 n;1 11
iT 3 n- 1 11
QT 3 graph n1 11
L(X) 3 graph nr 11
Q(D) 3 graph n 11
Q(I,R) 3 n1(y) 12
VeX) 4 nr(y) 12
C[O,IJ 4 Ll 13
J: 4 Lr 13
dim ker D 4 El 13
C(n) 5 Er 13
CN[O, 1] 5 L 13
C([O, I] X [0,1]) 5 E 13
C'(n) 5 G[nd 13
ft 5 G[nr] 13
(s)JF 5 G[n] 13
A(X) 6 cent 15,48
343
344 List of symbols
LgI(D) 15 En 40
Lgr(D) 15 G[Or,n] 40
Lg(D) 15 G[OI,n] 40
C(lR+) 16 G[On] 40
lnt 16
lin 40
GR,m[Od 18
GRn ,rrOr,n] 41
GR,m[Or] 18
GRn,rrOI,n] 41
GR,m[O] 18
gD,I(U) 22 Ar,n 45
gD,r(U) 22 A1,n 45
gD(U) 22 An 45
Vl(X) 27 C(lR) 48
Vl(X) 27 e 48
WA(X) 27 C(O;lF) 48
Aa+- 27 C([e,7r - e; lR) 49
vlFA 27
sint 49
C[a,b] 29
arcsinu 49
HD(X,y) 29
cot t 49
GD,r(X,y) 30
C[-7r/2+e,7r/2-e;lR) 49
G D,I(X, y) 30
[a, b] 31 cost 49
la, b[ 31 arccosv 49
[a] 31 tant 49
D+ 31 C([e,7r - e]; C) 49
MA,B(X) 33 C([-7r/2 + e,7r/2 - e];C) 49
X(a,b) 33 X 2 (a,b) 49
x *y 34 K(X) 52
Or,n 39 co dim 52
Ol,n 39
<lA 52
0-r,n1 39
(3A 52
0-
l,n 39
1
II:A 52
On 39
Lgr # 53
Lr,n 40
Er,n 40 LgI# 53
Ll,n 40 Lg# 53
El,n 40 X' 54
Ln 40 6.(R) 54
List of symbols 345
C'[O,T] 56 XD 101
8(S) 57 [L] 103
Lg# 57 [L]# 107
G(L) 58 exp(X) 107
AD(a,c,d) 63
{h 107
Hu 64
B(H) 107
h = ad - c? + c 64
B(X) 108
P(D) 69
An(OjlF) 109
P(I,R) 70
Q(D) 70 detu 109
Q(I,R) 71 R a 114
Q(D) 71 R a +- 114
Q(R, J) 71 II· lin 122
Q[D] 72 R~n 123
IP
8t8s 74 R~,n 123
2:j=o aj(t)%t 75 €>.z 127
8(u,v) 76 WD 132
C(O, JR) 77
WD(Xl, ... ,XN) 132
80 77
det(Dj-1Xk)j,k=1, ... ,N 132
A 77
flA,r 133
ND(a,b,c,d) 77
yc 78 DA 133
ND(a, c, d) 78 RA 133
BD(a, c, d) 78 FA 133
B!];(a, c, d) 78 LA,r 133
[M] 83 EA,r 133
[M]# 89 OA,1 134
Lr(D) 92 LA,1 134
Ll(D) 92 EA,l 134
L(D) 92
[Cln 136
In(x,y) 99
En 136
In(x) 99
Cx 136
rn(x, y) 99
Sx 136
rn(x) 99
s(x) 100 c>. 144
po 196 R~ 219
ZD,R 197 E~l) 219
.p 198
Q 222
ax 199
Q(t,s) 222
W(D) 199
WA(t) 199 Q(t) 222
an 201 Q(D,S) 224
n 201 h(n)
k,m
229
n+ 201 Xpp(hjno, ... ,nM) 232
n- 201 XPEP(hj Ajj Kjj Njj M) 233
.p+ 201
x y 245
.p- 201 0
Di3 256
fr ~<:) dr 201
HIL(f) 201 Ri3 256
p± 201 f(x) 256
X± 202 P(g) 257
x± 202 Po (g) 227
(A) 203 gilL 258
(A)l 206 OC(ajr) 259
(A)r 206
OCr 259
"'a211
OC 259
Aij 212
S(k) (D) 259
aij 212
Aj 213 S,(D) 259
1
2tr
ttr ()scot
0 X
s-t
- 2 ds 214 S,(D) 259
He 214 f(hD) 260
H 214 E)..(R) 261
if 214 E 261
XSN 216 X1(D) 262
Pj 216
p,(x,y) 263
e 216
T,,'K. = {T"hhE'K. 264
Xj 216
Enr 217
Tg,G = {Tg,hhEG 264
A(D) 268
E(j) 217
XEP(hj Al, .... , AM) 218 T(OC) 271
Ph 219 Ts(OC) 271
Dh 219 O(h) 272
348 List of symbols
l(h) 272
Soc (D) 273
TC(IK) 274
Tf,ociy 274
Xl = qt] 285
X* 287
x* 288
sinh t 296
cosht 296
P 296
I.l3 296
tanht 298
ih 304
6- 1
MI 304
6n
6h n 304
6- n
6h n 304
fen) 304
f(-n) 304
A. van der Burgh and J. Simonis (eds.): Topics in Engineering Mathematics. 1992,
266 pp. ISBN 0-7923-2005-3
F. Neuman: Global Properties of Linear Ordinary Differential Equations. 1992,
320 pp. ISBN 0-7923-1269-4
A. Dvurecenskij: Gleason's Theorem and its Applications. 1992,334 pp.
ISBN 0-7923-1990-7
D.S. Mitrinovic, J.E. Pecaric and A.M. Fink: Classical and New Inequalities in
Analysis. 1992,740 pp. ISBN 0-7923-2064-6
H.M. Hapaev: Averaging in Stability Theory. 1992,280 pp. ISBN 0-7923-1581-2
S. Gindinkin and L.R. Volevich: The Method of Newton's Polyhedron in the
Theory ofPDE's. 1992,276 pp. ISBN 0-7923-2037-9
Yu.A. Mitropolsky, A.M. Samoilenko and D.I. Martinyuk: Systems of Evolution
Equations with Periodic and Quasiperiodic Coefficients. 1992,280 pp.
ISBN 0-7923-2054-9
I.T. Kiguradze and T.A. Chanturia: Asymptotic Properties of Solutions of Non-
autonomous Ordinary Differential Equations. 1992,332 pp. ISBN 0-7923-2059-X
V.L. Kocic and G. Ladas: Global Behavior of Nonlinear Difference Equations of
Higher Order with Applications. 1993, 228 pp. ISBN 0-7923-2286-X
S. Levendorskii: Degenerate Elliptic Equations. 1993,445 pp.
ISBN 0-7923-2305-X
D. Mitrinovic and J.D. Keckic: The Cauchy Method of Residues, Volume 2. Theory
and Applications. 1993, 202 pp. ISBN 0-7923-2311-8
R.P. Agarwal and P.J.Y Wong: Error Inequalities in Polynomial Interpolation and
Their Applications. 1993,376 pp. ISBN 0-7923-2337-8
A.G. Butkovskiy and L.M. Pustyl'nikov (eds.): Characteristics of Distributed-
Parameter Systems. 1993,386 pp. ISBN 0-7923-2499-4
B. Stemin and V. Shatalov: Differential Equations on Complex Manifolds. 1994,
504 pp. ISBN 0-7923-2710-1
S.B. Yakubovich and Y.F. Luchko: The Hypergeometric Approach to Integral
Transforms and Convolutions. 1994,324 pp. ISBN 0-7923-2856-6
C. Gu, X. Ding and C.-c. Yang: Partial Differential Equations in China. 1994, 181
pp. ISBN 0-7923-2857-4
V.G. Kravchenko and G.S. Litvinchuk: Introduction to the Theory of Singular
Integral Operators with Shift. 1994,288 pp. ISBN 0-7923-2864-7
A. Cuyt (ed.): Nonlinear Numerical Methods and Rational Approximation II. 1994,
446 pp. ISBN 0-7923-2967-8
Other Mathematics and Its Applications titles of interest:
A.J. Jerri: Linear Difference Equations with Discrete Transform Methods. 1996,
462 pp. ISBN 0-7923-3940-1
I. Novikov and E. Semenov: Haar Series and Linear Operators. 1997,234 pp.
ISBN 0-7923-4006-X
L. Zhizhiashvili: Trigonometric Fourier Series and Their Conjugates. 1996,
312 pp. ISBN 0-7923-4088-4
R.G. Buschman: Integral Transformation, Operational Calculus, and Generalized
Functions. 1996,246 pp. ISBN 0-7923-4183-X
V. Lakshmikantham, S. Sivasundaram and B. Kaymakcalan: Dynamic Systems on
Measure Chains. 1996,296 pp. ISBN 0-7923-4116-3
D. Guo, V. Lakshmikantham and X. Liu: Nonlinear Integral Equations in Abstract
Spaces. 1996,350 pp. ISBN 0-7923-4144-9
Y. Roitberg: Elliptic Boundary Value Problems in the Spaces of Distributions.
1996,427 pp. ISBN 0-7923-4303-4
Y. Komatu: Distortion Theorems in Relation to Linear Integral Operators. 1996,
313 pp. ISBN 0-7923-4304-2
A.G. Chentsov: Asymptotic Attainability. 1997,336 pp. ISBN 0-7923-4302-6
S.T. Zavalishchin and A.N. Sesekin: Dynamic Impulse Systems. Theory and
Applications. 1997,268 pp. ISBN 0-7923-4394-8
U. Elias: Oscillation Theory of Two-Term Differential Equations. 1997,226 pp.
ISBN 0-7923-4447-2
D. O'Regan: Existence Theory for Nonlinear Ordinary Differential Equations.
1997,204 pp. ISBN 0-7923-4511-8
Yu. Mitropolskii, G. Khoma and M. Gromyak: Asymptotic Methods for Investigat-
ing Quasiwave Equations of Hyperbolic Type. 1997,418 pp. ISBN 0-7923-4529-0
R.P. Agarwal and P.J.Y. Wong: Advanced Topics in Difference Equations. 1997,
518 pp. ISBN 0-7923-4521-5
N.N. Tarkhanov: The Analysis of Solutions of Elliptic Equations. 1997,406 pp.
ISBN 0-7923-4531-2
B. Rieean and T. Neubrunn: Integral, Measure, and Ordering. 1997,376 pp.
ISBN 0-7923-4566-5
N.L. Gol'dman: Inverse Stefan Problems. 1997,258 pp. ISBN 0-7923-4588-6
S. Singh, B. Watson and P. Srivastava: Fixed Point Theory and Best Approxima-
tion: The KKM-map Principle. 1997,230 pp. ISBN 0-7923-4758-7
A. Pankov: G-Convergence and Homogenization of Nonlinear Partial Differential
Operators. 1997,263 pp. ISBN 0-7923-4720-X
Other Mathematics and Its Applications titles of interest: