Fattorini H.O., Kerber A. The Cauchy Problem

Подождите немного. Документ загружается.

2.5. Miscellaneous Comments

taking E to be a space of functions in X and defining an evolution operator

in E by

(S(t, s) f)(x) = f (I (t, s; x)),

which, in the time-invariant case, will be given by

(S(t) f)(x) = f (4'(t; x)). (2.5.19)

An obvious bonus is that S will always be a linear operator; naturally S

satisfies the semigroup equations (2.1.2) in the time-invariant case (and the

equations (7.1.7) an (7.1.8) in general). For a temporally homogeneous

Markov process we have

fu(T(t, x)) = f

,q, d))x(dq),

(2.5.20)

which suggests the idea of "cancelling out" the

11111

measure and studying

directly the semigroup

(S(t)u)() =

rl;

(2.5.21)

in a space of s-measurable functions. Continuity of is assured by

additional conditions on the transition probability (see Yosida [1978: 1, p.

398]).

In actual practice, not even f(x(t)) may be an observable quantity;

what an instrument reads is some time average like

T f

Tf(4)

(t; x)) dt,

which, if the system "tends to a steady state" sufficiently fast will be a good

approximation to its limit as T - oo. This is one among the motivations for

the study of diverse "limit averages" of a semigroup S(.) of linear bounded

operators, for instance, the CBsaro limit

lim

1 fTS(t)udt

T- oo T 0

or the Abel limit

lim A f

e-XIS(t)udt

A -oo

in various topologies. This sort of problems comprise what is called ergodic

theory. In its early stages it dealt with "concrete" semigroups of the form

(2.5.19) (see Hopf [1937:

11) and with discrete rather than continuous

averages. The first works on the abstract theory (Von Neumann [1932: 2],

Riesz [1938: 1], Visser [1938: 1], Yosida [1938: 1], and Kakutani [1938: 1])

also refer to the limit of the discrete average

I n-1

lim - E Sju,

noo n j=0

Properly Posed Cauchy Problems: General Theory

where S is an operator whose powers Si are uniformly bounded in the norm

of (E). The theory for the continuous case was initiated by Wiener [1939:

1]. Some results in ergodic theory deal with general semigroups in Banach

spaces: others (for instance, those involving pointwise convergence) with

semigroups in certain function spaces. For accounts of ergodic theory and

bibliography see Yosida [1978:

1]; other treatments are found in Hille-

Phillips [1957: 1] and Dunford-Schwartz [1958: 1].

The study of Markov processes from the point of view of semigroup

theory was initiated by Hille [1950: 1] and Yosida [1949: 1], [1949: 2]. A

systematic investigation of the subject was conducted shortly after by Feller

([1952:

11, [1953:

11, [1954: 1], and subsequent papers). Short and very

readable expositions can be found in Feller [1971: 1] and Yosida [1978: 1].

Other references are Dynkin [1965: 1], [1965: 2], Gihman-Skorohod [1975:

1], [1975: 2], [1979: 1], Ito-McKean [1974: 1], Stroock-Varadhan [1979: 1],

and Mandl [1968: 1]. We limit ourselves to note that one of the central

problems is the computation of the operator A in the equation (2.5.12), (or,

rather, in the equation corresponding to the semigroup (2.5.21)), called the

Fokker-Planck equation of the process, and the reconstruction of the

transition probabilities from A. Under adequate assumptions on a (2 a

subset of o m or more generally a manifold) and on the transition probabili-

ties, the Fokker-Planck equation is a second-order parabolic differential

equation, and the actual trajectories of the process are selected on the basis

of "boundary conditions," sometimes of nonstandard type.

For results on the relation between 4' and the semigroup defined by

(2.5.19) see Neuberger [1971/72: 1], [1973: 2], and Durrett [1977: 11.

previous subsection would seem to indicate that every conceivable physical

process that satisfies the major premise of Huygens' principle must neces-

sarily be described by a first-order equation of the form (2.5.12), at least if

the necessary conditions of time invariance, linearity, and continuity are

satisfied. Thus, even if the process stems from an equation of higher order in

time like

u(")(t)+AIu("-')(t)+ + A,,u(t) = 0 (2.5.22)

(where A,,

... ,

A, are, say, closed and densely defined operators in a Banach

space E), a reduction of (2.5.22) to a first-order equation (conserving

whatever existence, uniqueness, and continuous dependence properties it

may have) should be possible. This is, in fact, so in examples like the wave

and vibrating plate equation, where energy considerations dictate the reduc-

tion. An abstract treatment of these ideas was given by Weiss [1967: 1].

A study of the equation (2.5.22) in a different vein was carried out by

the author [1970: 2]. The Cauchy problem for (2.5.22) is declared to be well

posed in an interval 0 < t < T if solutions exist in the interval for initial data

2.5. Miscellaneous Comments

u(0), u'(0),...,u(n-1)(0) in a dense subspace D of E and depend continu-

ously on them uniformly on compact intervals in the sense that

IIu(s)II

(0<s<t<T)

j=0

with as Section 1.2. This point of view leads to some curious proper-

ties: for instance, the Cauchy problem may be well posed in a finite interval

without being well posed in t > 0 or, even if it is well posed in t > 0, the

solutions of (2.5.22) may increase faster than any exponential (in fact,

arbitrarily fast) at infinity. Admittedly, this may only indicate that the

above definition of well-posed problem is too inclusive. A more restricted

definition was attempted in the author's [1981: 1] for the case n = 2; it

implies the existence of a strongly continuous semigroup S

in a suitable

product space that propagates the solutions of (2.5.22) in the sense that

C5 (t)(u(0), u'(0)) = (u(t), u'(t)), and generalizes an earlier result of Kisynski

[1970: 1] where the equation is u"(t) = Au(t).

Necessary and sufficient conditions in order that the Cauchy prob-

lem for (2.5.22) be well posed seem to be unknown, except in the particular

case

u(n)(t) = Au(t). (2.5.23)

If n > 3, the Cauchy problem for (2.5.23) is well posed if and only if A is

everywhere defined and bounded. This was proved by the author [1969: 2]

(where the case in which E is a linear topological space is also dealt with)

and independently by Chazarain [1968: 1], [1971: 1] with a different proof

that works for more general equations. The case n = 2 is of more interest:

here the equation is

u"(t)=Au(t),

(2.5.24)

and the substitution u(t) - u(- t) maps solutions into solutions, thus the

Cauchy problem is well posed in (- no, oo) if it is well posed in t >_ 0. In the

spirit of Section 1.2 we can define two propagators and by

C(t)u = uo(t), S(t)u = u1(t), where

(resp.

is the solution of

(2.5.24) satisfying u(0) = u, u'(0) = 0 (resp. u(0) = 0, u'(0) = u), and extend

them to all of E by continuity. is strongly continuous and satisfies the

cosine functional equation(s)

C(O)=I, 2C(s)C(t) = C(s + t)+C(s - t) (2.5.25)

for all s, t, whereas S obeys the formula

S(t)u= f0 tC(s)uds, (2.5.26)

hence is (E)-continuous and satisfies with the sine functional equa-

tion(s)

S(0)=0, 2S(s)C(t)=S(s+t)+S(s-t).

100 Properly Posed Cauchy Problems: General Theory

Functions satisfying (2.5.25) with values in a Banach algebra were

considered by Kurepa [1962: 1]; among other results, it is shown that

measurability implies continuity and that C(t) = cos(tA) (the cosine defined

by its power series) for some A E 8. The theory of strongly continuous

cosine functions or cosine operator functions was initiated by Sova [1966: 2],

who among other things proved that there exist K, t such that

IIC(t)II,<Ke'*1

(-oo<t<oo),

(2.5.27)

defined the infinitesimal generator of (roughly, the second derivative at

the origin) and proved a generation theorem of Hille-Yosida type (which

was independently discovered by Da Prato-Giusti [1967: 3]). Other results

are due to the author [1969: 2] and [1969: 3], where the setting is (in part)

that of linear topological spaces; in particular, a "measurability implies

continuity" theorem of the type of Example 2.3.16 (which implies that the

concepts of well posed and uniformly well posed problem coincide for the

second-order equation) and an analogue of Lemma 2.2.3 for second order

equations. We also point out a relation proved in [1969: 2] between the

second order equation (2.5.24) and the first-order equation (2.1.1) that turns

out to be extremely useful in singular perturbation problems: if the Cauchy

problem for (2.5.24) is properly posed in the sense outlined above, then the

Cauchy problem for (2.1.1) is as well properly posed; the propagator V(t)

of (2.1.1) is explicitly given by the "abstract Poisson formula"

V(t)u= 1 1OOe-s2,"C(s)uds, (2.5.28)

art

which incidentally shows that V can be extended to a (E )-valued analytic

function in the half plane Re > 0 (in fact, A E Q (ir/2 - ); see the

forthcoming Chapter 4). The fact that well posedness of (2.5.24) implies the

corresponding property of (2.1.1) was proved independently by Goldstein

[1970: 4] in a different way.

Given a strongly continuous group U(-), it is immediate that

C(t)=Z(U(t)+U(-t))

(2.5.29)

is a strongly continuous cosine function whose infinitesimal generator is

A=B 2, B the infinitesimal generator of U(.). The central problem in the

theory of cosine functions is whether a representation of the form (2.5.29)

(analogous to the equality cos at =

(e 'at +e- `at )) exists for any cosine

function. The answer is in the negative; it suffices to take C(t) = cos(tA),

where A is a bounded operator having no (bounded or unbounded) square

root.2 Other counterexamples were given by Kisynsky [1971: 1], [1972: 1],

where C is uniformly bounded in norm: see the author [1981: 1] for

2In fact, it can be easily shown that if (2.5.29) holds then A = B2, where A is the

infinitesimal generator of and B is the infinitesimal generator of U(.).

2.5. Miscellaneous Comments

101

additional information. It was proved by the author [1969: 3] that in certain

spaces the representation (2.5.29) must hold if

is uniformly bounded;

the construction of U is based upon the Hilbert transform formula

U(t)u=C(t)+2it r

C(S)

(2.5.30)

the (singular) integral understood as the limit for e - 0 of the integrals in

It - s I >, E. The (E)-valued function can be shown to be a strongly

continuous group in spaces where the Hilbert transform operator, defined

for E-valued functions, has desirable continuity properties, such as E = LP,

1 < p < oo (in particular in Hilbert spaces). In the same spaces, a cosine

function that is not uniformly bounded but merely satisfies (2.5.27) may not

be decomposable in the form (2.5.29), but it can be proved (see again the

author [1969: 1]) that if w is the constant in (2.5.27), b > co and C,(.) the

cosine function generated by Ab = A - b2I, then the decomposition holds

for Cb ( ); in other words, Ab posseses a square root Ab 2 that generates a

group Ub(.). It follows that the second-order equation (2.5.23) can be

reduced to the first-order system

u;(t)= (Ae 2+ibl)u2(t), u2(t)= (A'e 2-ibI)ui(t)

in the product space E x E, thus in a way the theory circles back into the

reduction-of-order idea outlined at the beginning of this subsection. We

indicate at the end additional bibliography on cosine functions and on the

equation (2.5.24), pointing out by the way that some of the ideas used in the

treatment of the equations (2.5.23) and (2.5.24) originate in early work of

Hille [1954: 1], [1957: 1] (see also Hille-Phillips [1957: 1, Ch. 23]), where it is

assumed at the outset that A = B" and the Cauchy problem is understood in

a different sense.

The general equation (2.5.22) has received a great deal of attention

during the last three decades, especially in the second-order case

u"(t)+ Bu'(t)+Au(t) = 0.

Most of the literature deals with sufficient conditions on A, B that allow

reduction to a first-order system and application of semigroup theory. In the

references that follow, some works treat as well the case where the coeffi-

cients A, B depend on t; we include as well material on the equation (2.5.24)

and on cosine functions. See Atahodzaev [ 1976: 11, Balaev [ 1976: 11, Buche

[1971: 1], [1975: 1], Carasso [1971: 1], Carroll [1969: 2], [1979: 2] (which

contain additional bibliography), Daleckil-Fadeeva [1972: 1], Dubinskii

[1968: 1], [1969:

11, [1971:

1], Efendieva [1965: 11, Gaymov [1971: 11,

[ 1972: 1 ], Giusti [ 1967: 11, Goldstein [ 1969: 1], [1969: 3], [1969: 4], [1970: 11,

[ 1970: 2], [1970: 4], [1971: 11, [1972: 1], [1974: 2], [1975/76: 2], Goldstein-

Sandefur [1976: 11, Golicev [1974: 1], Hersh [1970: 1], Jakubov [1964: 2],

[ 1966: 11, [1966: 2], [1966: 4], [1966: 5], [1967: 3], [1970: 11, [1973: 1 ], Jurcuk

102

Properly Posed Cauchy Problems: General Theory

[ 1974: 1], [1976: 11, [1976: 2], Kolupanova [ 1972: 11, Kononenko [ 1974: 11,

Krasnosel'skiff-Krein-Sobolevskii [1956: 1], [1957: 1], Krein [1967: 1] (with

more references), Kurepa [1973:

1], [1976:

1], Ladyzenskaya [1955:

1],

[1956:

11, [1958:

1], Lions [1957: 1], Lions-Raviart [1966: 11, Lomovicev-

Jurcuk [ 1976: 1 ], Mamedov [ 1960: 11, [1964: 11, [1964: 2], [1965: 1], [1966:

11, Mamil [ 1965: 1], [1965: 2], [1966: 11, [1967: 11, [1967: 2], Mamii-Mirzov

[1971: 1], Masuda [1967: 1], Maz'j a-Plamenevskii [ 1971: 1], Melamed [1964:

11, [ 1965: 1], [1969: 11, [197 1: 1 ], Nagy [ 1974: 1], [1976: 11, Obrecht [ 1975:

1 ], Pogorelenko-Sobolevskii [ 1967: 1], [1967: 2], [1967: 3], [1970: 1], [1972:

1 ], Radyno-Jurcuk [ 1976: 11, Raskin [ 1973: 1], [1976: 1 ], Raskin-Sobolevskii

[ 1967: 11, [1968: 11, [1968: 2], [1969: 11, Russell [ 1975: 1 ], Sandefur [ 1977: 11,

Sobolevskii [1962: 1], Sova [1968: 11, [1969: 1], [1970: 11, Straughan [1976:

1], Travis [1976: 1], Tsutsumi [1971: 1], [1972: 11, Veliev [1972: 11, Veliev-

Mamedov [1973: 1], [1974: 1], Yosida [1956: 1], [1957: 2], and the author

[1971: 2]. In some of these works the equations in question are time-depen-

dent and/or semilinear (see (k)) or the setting is, at least in part, more

general than that of properly posed Cauchy problems (see (e) and also

Chapter 6).

(d) Semigroup and Cosine Function Theory in Linear Topological

Spaces. Consider the translation group

S(t)u(x) = u(x + t)

in the linear topological space E consisting of all functions u continuous in

- oo < t < oo, endowed with the family of seminorms

Ilulln = Sup

I u(x)I

1xI <n

n =1, 2,... . Obviously,

is strongly continuous in E, but it is easily seen

that, in general, the integral

oo e-x`S(t)udt

(2.5.31)

fails to exist in any reasonable sense whatever the value of A, thus blocking

any generalization of formula (2.1.10). That this generalization fails is not

surprising, since the infinitesimal generator of S (in an obvious sense) is

A = d/dx, with domain consisting of all u continuously differentiable in

- oo < t < oo and the resolvent set of A is empty (the equation Au - u' = v

always has infinitely many solutions in E). This example makes clear that

the main difficulty in generalizing semigroup theory to linear topological

spaces is the loss of (2.1.3) (which, in the present setting must be translated

into the requirement that (e-`S(t); I >- 0) be equicontinuous for some co).

If this is assumed, the Banach space results generalize readily to a locally

convex quasicomplete space. This extension was suggested by Schwartz

[1958: 2] and independently carried out by Miyadera [1959: 2] for Frechet

2.5. Miscellaneous Comments

103

spaces; the general version is due to Komatsu [1964: 1] and Yosida in the

first edition (appeared in 1965) of his treatise [1978: 1]. We refer the reader

to that book for details.

The general case is tougher. The only assumption is that S is strongly

continuous and that (S(t); 0 < t < 1) is equicontinuous; this is a conse-

quence of the strong continuity if the space E is barreled, as the uniform

boundedness theorem (Kothe [1966: 1]) shows. Generation theorems have

been found by T. Komura [1968: 1], Ouchi [1973: 1], Dembart [1974: 11,

and Babalola [1974: 1], and can be roughly described as follows. In the

approach of Komura the integral (2.5.31) is given a sense through the theory

of Fourier-Laplace transforms due to Gelfand and Silov [1958: 1] where

(generalized) functions of arbitrary growth at infinity can be transformed:

the generalized resolvent R(A; A) then appears as a vector-valued analytic

functional rather than a function, and is gotten at through "approximate

resolvents"

obtained cutting off the domain of integration in

(2.5.31). The methods of Dembart, Ouchi, and Babalola are somewhat

related; the generalized resolvent is not used and its place is taken by

a sequence of approximate resolvents constructed in the way indicated

above. These ideas are related to the asymptotic resolvent of Walbroeck

[1964: 1].

The theory of cosine functions admits a similar extension to linear

topological spaces. When {e-"I`IC(t); - oo < t < oo) is assumed equicon-

tinuous for w sufficiently large, the treatment (due to the author [1969: 2]

for barreled, quasi-complete E) uses the same ideas as in the Banach space

case. The "reduction to first order" outlined in (c) can be carried out if an

additional assumption is satisfied; one of its forms is the requirement that

the function

G(t) = f0 Ilogs(S(s + t)+S(s - t)) ds

(S defined by (2.5.26)) satisfy (a) G(t)E c D(A), and (b) AG(.) is a

strongly continuous function of t for all t (as pointed out by Nagy [1976: 1],

however, this assumption fails to hold even in some Banach spaces). The

case where only (S(t); 0 < t < 1) is assumed to be equicontinuous has been

examined by Konishi [1971/72: 1] who obtained generation theorems (in

the style of those of T. Komura for semigroups) and many additional

results.

We include below several references on semigroups, cosine functions,

and linear differential equations in linear topological spaces; some of these

belong to the subjects treated in Chapters 4 and 5 and will be also found

there. See Chevalier [1969:

1], [1970: 1], [1975:

1], Dembart [1973:

1],

V. V. Ivanov [1973:

11,

[1974:

11,

[1974: 2], Kononenko [1974:

11,

Lovelady [1974/75: 1], Mate [1962: 2], Milstein [1975: 1], Ouchi [1973: 1],

Povolockii-Masjagin, [1971:

1],

Sarymsakov-Murtazaev [1970:

11,

Singbal-Vedak [ 1965: 1], [1972: 1 ], Terkelsen [ 1969: 1], [1973: 1 ], Tillmann

104

Properly Posed Cauchy Problems: General Theory

[1960: 1], Ujishima [1971: 1], [1972: 1], Vainerman-Vuvunikjan [1974: 1],

Vuvunikjan [1971: 1], Watanabe [1972: 1], [1973: 1], and Yosida [1963: 1].

(e) The equation

u'(t) = Au(t) (2.5.32)

(as well as higher-order and time-dependent versions) has been extensively

studied under hypotheses sometimes considerably weaker than (or not

comparable to) those for a properly posed Cauchy problem. In some cases

the objective is to prove existence and/or uniqueness results for solutions in

t > 0 or other intervals, as in Ljubi6 [1966: 3]. We include a small sample of

results.

2.5.1 Example.

Let A be a densely defined operator such that R(X) exists

in a logarithmic region

A=A(a,,6)=(1\: ReX>,(3+aloglAI)

with a >, 0 (see Section 8.4, especially Figure 8.4.1) and assume that

IIR(A)II <C(1+ I1`I)m

(X,=-A)

for some m > 0. Then (a) for every uo E D(A°°) = n D(A") there exists a

genuine solution of (2.5.32) in t > 0 with u(0) = uo; (b) genuine solutions of

the initial value problem are unique.

Uniqueness holds in fact in a strong version: every solution of

(2.5.32) in a finite interval 0 < t < T with u(0) = 0 is identically zero in

0 < t < T. This is best proved with the "convolution inverse" methods of

Chapter 8 (or using the next example). Solutions are constructed by inverse

Laplace transform of R(A) in the style of formula (2.1.12); however,

integration must be performed over the boundary r of A and further terms

in formula (3.7) are necessary:

(t)

AJuo +

2,ri

R(A)ApuodX.

J=o

This expression defines a solution in 0 < t < tp = (p - m - 2)/a with initial

value u(0) = uo. Thus (making use of uniqueness at each step) we can define

a solution `by pieces" in t > 0.

We note that under the hypotheses above D(A°°) is dense in E, thus

the set of initial vectors is dense in E. See Ljubic [1966: 3] and Beals [1972:

2] for far more general statements in this direction. Uniqueness can be also

established in much greater generality:

2.5.2 Example. Assume that R(X) exists for A > to and that

IIR(A)II = O(e'll) as A + oo

for some r > 0. Then if

is a solution of (2.5.32) in 0 < t < T with

u(0)=0, we have u(t)=0 in 0 <t <T- r.

2.5. Miscellaneous Comments

105

To prove this result we note that

(XI - A)f e_ 'u(t) dt = - e-XTu(T)

hence

f e-'`tu(t) dt

O(e-X(T-r))

as A -* 00,

which implies that u(t) = 0 in 0 < t < T - r by (a slight variant of) the

Paley-Wiener theorem.

In other lines of treatment (for instance Agmon-Nirenberg [1963: 1],

[1967: 1]) existence is not in question and only properties of solutions such

as asymptotic behavior at infinity of theorems of Phragmen-Lindelof type

are of interest. Many of the results have connections with the abstract

Cauchy problem, however. We shall not treat in this work the equation

(2.5.32) in this vein and we limit ourselves to indicate some bibliography.

Besides the papers above, information can be found in the treatises of

Carroll [1969: 2], [1979: 2], Ladas-Lakshmikantham [1972: 2], and Zaidman

[1979: 1], together with references on the subject. Other works are Aliev

[1974: 1], Arminjon [1970: 1], [1970: 2], Beals [1972: 1], [1972: 2], Djedour

[1972: 11, Goldstein-Lubin [1974: 1], Ladas-Lakshmikantham [1971:

1],

[ 1972: 1 ], Levine [ 1970: 1], [1970: 2], [1972: 1], [1973: 1], [1973: 2], [1975: 11,

Levine-Payne [1975: 1], Malik [1967: 11, [1972: 1], [1973: 1], [1975: 1], and

Pazy [1967: 1].

An important area of study concerns almost periodic solutions of

(2.5.32) and of time-dependent analogues; most of the results are generali-

zations to the infinite-dimensional setting of theorems on almost periodicity

of solutions of ordinary differential equations due to Bohl, Bohr and

Neugebauer, Favard and others (see Favard [1933: 1] for results and

references). The setting is sometimes that of well-posed Cauchy problems,

other times a more general one. See the treatises of Amerio-Prouse [1971: 1]

and Zaidman [1978: 1] for expositions of the theory and bibliographical

references. Additional works are Zikov [1965: 1], [1965: 2], [1966: 11, [1967:

1], [1970:

1], [1970: 2], [1970: 3], [1971:

1], [1972: 1], [1975: 1], Zikov-

Levitan [1977: 1], Levitan [1966: 1], Zaki [1974:

1], Rao [1973/74: 1],

[1974/75:

11, [1975:

1], [1975: 2], [1977: 1], Rao-Hengartner [1974:

1],

[ 1974: 2], Aliev [ 1975: 1 ], Bolis Basit-Zend [ 1972: 1 ], Biroli [ 1971: 11, [1972:

1], [1974: 1], [1974: 2], Lovicar [1975: 1], Mignaevskil [1971: 1], [1972: 1.],

Moseenkov [1977: 1], Welch [1971: 1], and the author [1970: 4]. For almost

periodic functions in Banach spaces see Corduneanu [1961: 11.

(f) Transmutation. In a somewhat vague description, a transmuta-

tion formula transforms a solution of an abstract differential equation into a

solution of another abstract differential equation. Two rather trivial trans-

mutation formulas are u(t) - e°tu(t) (transforming a solution of u'(t) =

106 Properly Posed Cauchy Problems: General Theory

Au(t) into a solution of u'(t) = (A + aI) u (t)) and the "identity formula"

u(t) - u(t), which transforms a twice differentiable solution of u'(t) = Au(t)

such that u(t) E D(A2) into a solution of the second order equation

u"(t) = A2u(t). We list below a number of transmutation formulas. In all

the following examples, A is a closed operator in the Banach space E and

solutions are understood in the strong sense as in Section 1.2.

2.5.3 Example.

Let u(.) be a solution of

u"(t)=Au(t) (t>, 0) (2.5.33)

such that u(O) = u, u'(0) = 0, and

11u(t)II 5 Ce'

(t > 0). (2.5.34)

Then

v(t) =

1 J e-SZ/anu(s)

art

is a solution of

v'(t) = Av(t) (t >, 0) (2.5.35)

with v(0) = u, satisfying an estimate of the type of (2.5.34) with w2 instead

of w. This formula is an abstract version of the classical Poisson formula for

solution of the heat equation u, = u,,x, where E is a space of functions in

(- oo, oo) (say, L2) and u(t)(x) = u(x + t). In its abstract version it can be

traced back (in a different context) to Romanoff [1947: 1] and it was used

by the author in [1969: 2] to relate the Cauchy problem for (2.5.34) and

(2.5.35) (see (2.5.28) and surrounding comments). See also the author [1983:

3] for an extension to linear topological spaces.

2.5.4 Example. Let

be an arbitrary solution of (2.5.33), a > - 1.

Define

v(t)=

r(a+3/2) f1

u(tn)(l-712drl (2 5 36).

,cr(a+l)

Then v(.) is a solution of

2(a+ 1)

' = A

2 5 37

(t)+

v (t) v(t)

(tv )

. . )

(

with initial conditions v(0) = u(0), v'(0) = 0. In case A is the m-dimensional

Laplacian, formula (2.5.36) transforms the wave equation (2.5.33) into the

Darboux equation (2.5.37) and can be used to compute an explicit solution

of the wave equation (see Courant-Hilbert [1962: 1, p. 700]).

2.5.5 Example.

Let u(.) be a solution of (2.5.33) with u'(0) = 0. Define

v(t)=

f'JO(c(t2-s2)1/2)u(s)ds,

(2.5.38)