I nternat. J. Math. & Mah. Sci. (1982) 31-40 Vol. 5 No. 31 SEMIGROUP STRUCTURE UNDERLYING EVOLUTIONS G. EDGAR PARKER Department of Mathematics Pan American University 78539 U.S.A. Edinburg, Texas (Received December 4, 1979) ABSTRACT. A member of a class of evolution systems is defined by averaBing a one- parameter family of invertible transformations G with a semigroup G(t)T(t-s)G(s) resulting evolution system, U(t,s) strong continuity, and in case G -1, T The preserves continuity and is a linear family, may have an identifiable generator and resolvent both of which are constructed from T. Occurrences of the class of evolutions are given to show possible applications. KEY WORDS AND PHRASES. Evolon system, semigroup 1980 MATHEMATICS SUBJECT CLASSIFICATION CODES. i. of transformations, resolvent. Primary 47H15, Secondary 34G05. INTRODUCTION. Given a subset C system of continuous transformations on U(t,s):. i. ii. t "_; s _> U(t,s) U(s,s) 0 H, a strongly continuous evolution of a Banach space so that if t _> s _> is a family of functions C r _> 0 then is a continuous function from C into C, I, the identity on C, (DI iii. U(t,s)U(s,r) iv. if x E U(t,r), and C, then {(p, U(p,s)x) p _> s is continuous. U G.E. PARKER 32 Such systems arise as a result of having unique solutions to the evolution equation y’(t) A(t)(y(t)), y(s) x, where x is a point of C and if t O, A(t) is a function from a subset of C into H. The study of evolution systems has in several papers been made through the study of semigroups (see for instance Kato[5]). A strongly continuous semigroup of continuous transformations on C is a family T functions from C into C so that if t and s are in i. T(t+s) ii. T(0) { T(t): t > 0 } of continuous [0, ), then T(t)(T(s)), I, and (D2 iii. if x is an element of C, is continuous. By the relationship U(t,s) {(p,T(p)(x)) p > 0 } T(t-s), each semigroup is a specialized evolution One approach to the study of evolution systems has been to study semi- system. groups and try to generalize to evolution systems properties the semigroups display. A second approach (see Ball [i], or Neveu [8]) has been to embed U into a semigroup T on [0, =)xC by the action T(t)(s,x) (t+s,U(t+s,s)x) and study T. Goldstein ([4]), however, has shown that even if U has linear, non-expansive maps and analytic trajectories, T may not even have Lipschitz maps and hence may fail to satisfy the hypothesis which is typically used in theorems about semigroups. This paper gives an alternative approach to studying evolution systems through semigroups. Here the evolution systems is not studied by analogy but rather by identifying a class of evolution systems for which semigroups are actual structural components. Results concern the structure of such evolution systems as they relate to the existing theory for evolution systems. Examples are given that illustrate under what conditions these evolution systems occur. 2. DEFINITIONS and THEOREMS. In addition to the notation established in the introduction, the following notation will be used. L(H,H) will denote the normed vector space of continuous linear transformations on H with, for f in L(H,H), fl glb {m: if p H, Ilf (p) I __< m II ell }. SEMIGROUP STRUCTURE UNDERLYING EVOLUTIONS R(H,H) L(H,H) whose elements have will denote the subset of inverses that are in L(H,H). will be denoted by j. 33 The identity function on the numbers The domain (range) of a function f will be Df(Rf) denoted Suppose that G is a family of functions so that DEFINITION: >_ 0, G(t) C-> C. The statement that G is strongly continuous if t at the point x of C means that g continuous, {(t,G(t)(x)) x t > 0} is will be called the trajectory of G from x. g X Suppose that U is a strongly continuous evolution DEFINITION: system of continuous tranformations on C. The statement that A is the infinitesimal enerator for U means that A is so that if -i lim h t > {(x,y) O, A(t) (U(t+h,t) (x) x g {A(t):t > O} C and y x) }. h/O Theorem i establishes a way of combining a semigroup with a one parameter family so that the result exhibits evolution system structure. Conditions are given under which the frm of the infinitesimal generator can be guaranteed. THEOREM i. Suppose that T is a strongly continuous semi- group of continuous transformations on C with infinitesimal generator B and G is a strongly continuous so that if t > 0, G(t) is family of functions on C Lipschitz and has a Lipschitz inverse and if a and b are numbers so that a < b, on C; k is the Lipschitz norm of define U(t,s) G(p)} G(t)T(t-s)G(s) {k: a< p < b and is a bounded set. For t > s > O, -i Then U is a strongly continuous evolution system of continuous transformations on C. G(s) is g Suppose, in addition, that C=H, x R(H,H) for each s, and in the domain of B and t is a number so that the trajectory of G from g H, G(t)-ix G(t)-ix is 34 G.E. PARKER Then, if A denotes the infinitesimal generator differentiable at t. G(t)BG(t) and A(t)x for U, x is in the domain of A(t) -i (x)+gy’(t) G(t)-Ix. where y To see that U is an evolution system of continuous -i is transformations on C, note that (i) U(t,s) --G(t)T(t-s)G(s) PROOF: the composition of continuous functions each mapping C into (ii) U(s,s)= G(s)T(s-s)G(s) C; -i I; and (iii) U(t,s)U(s,r) -I U(t,r). G(t)T(t-r)G(r) G(t)T(t-s)G(s)-l(G(s)T(s-r)G(r)- For strong continuity, suppose that s {t n} n=l llG(tn)T(tn-S)G(s)-l(x) -G(t n )T(t-s)G(s) -i G(t)T(t-s)G(s) I G(t )T(t-s)G(s) n continuous at I (x) G(t)T(t-s)G(s)-l(x) I + I converges to t in [s, ). -i < + IG( t n I -I (x) _> 0, U(t ,s)xn < and U(t,s)xll llG(tn)T(tn -s)G(s) G(t )T(t-s)G(s) -i n llr(t x g C, -I (x) (x) -s)G(s)-l(x)-r(t-s)G(s)-l(x) Lip -i (x) -G(t)T(t-s)G(s) G(s)-l(x), {IG(tn) I} (x) I T is strongly is a bounded number n=l sequence, and G is strongly continuous at T(t-s)G(s) -I (x). Thus each term of the final inequality can be made arbitrarily small and each trajectory of U is continuous. With G satisfying the additional assumptions, -i an element of H so that y G(t) -i consider lira h (U (t+h t) (x) -x) h/0 x) lim suppose x is (x) is in the domain of B and -i lim h h*0 (G(t+h)T(h)G(t) -i (x) h-l(G(t+h)T(h)G(t)-l(x) G(t+h)G(t)-l(x)) + h-l(G(t+h)G(t)-l(x) G(t)G(t)-l(x)) lim G(t+h) (h-l(T(h)-I)(G(t) -I h*0 (x))) + lim h -i h/0 (g (t+h) Y g (t)) G(t)BG(t) Y h/0 Hence x is in the domain of A(t) and A(t) (x) -i (x) + g ’(t). Y -i G(t)BG(t) (x) + g ’(t). Y Theorem 2 provides a partial converse to Theorem i. thrust of the theorem is that a non-linear problem may, The through 35 SEMIGROUP STRUCTURE UNDERLYING EVOLUTIONS the structure from Theorem l, be exchanged for a linear problem. Suppose that {A(t)} THEOREM 2: >0 A(t)(v(t)), v(s) unique solution to v’(t) x in the domain of A(s). G with domain [0, is so that there is a t x for each point Suppose also that there is a function and range in R(H,H) and a strongly continuous semigroup T of continuous transformations on H with infinitesimal generator B so that if t > 0 and =E H, i. ii. if t g [0, then T(t)(x)g D then G(t) (D) D B D if x iii. and x [0, oo) Yx G(t) (x) y (t) iv. B then {IG(r) if t > s > 0, B A(t) H is defined by y x (t) s < r G(t)B(x) A(t)G(t) (x) _< t} is and a bounded set. Then if U is the evolutionsystem generated by A and t -i U(t,s) G(t)T(t-s)G(s) Suppose that t > s and that x e PROOF: there is a unique function v so that v(s) f f x G(t)(D B) ability (g ’(t) Y T(t-s)G(s) llh + r(t-s) is a subset of D -i (G(t+h)r(t+h-s)G(s) G(t)BT(t-s)G(s)-l(x)II (x)) -I (x) and B From assumption iii. on the strong differenti- and denoting T(t-s)G(s) llh x (t)) makes sense since by assumption DA(t) can be written x. (x).f (s) I and if t > s, R of G, -i -l I II (x)) that A(t)(f A(s) Consider h A(t) (f x (t)) (G(t+h)r(t+h-s)G(s) A(t)G(t)r(t-s)G(s) (x)) (t)) G(t)T(t-s)G(s) x is in D G(t)T(t-s)G(s) )+ H defined by f (t) x I h-l( f x (t+h) Notice A(t) (v(t)) and Also, U(t,s)(x) is defined to be v(t). x. [0, x v’ (t) By assumption, DA(s). (G(t+h) (T(t-s)G(s) -i (x)) I G(t+h)(Br(t-s)G(s)-l(x)) x by y -i (x) the above norm -i G(t)r(t-s)G(s) (x))- (N-I(T(h) (T(t-s)G(s) (x)) I + -i _< fIG(t+h) G(t+h)(B(r(t-s)G(s) -i -i G(t+h) (B(T(t-s)G(s) -i (x))) g’ G(t)Br(t-s)G(s)-l(x))II The first term converges to 0 by the differentiability of T at (x)) (t) II G.E. PARKER 36 -i T(t-s)G(s) IG(r) (x) and the boundedness of [t, t+m]} r the second converges to 0 by the strong differentiability of G at -i T(t-s)G(s) (x) and the third converges to 0 by the strong contin-i uity of G at BT(t-s)G(s) (x) (t) Thus f eA(t) (f X (t)) X and by the uniqueness of solutions U(t,s)(x)=v(t)=f G(t)r(t-s)G(s)-l(x) X (s)=x f X (t) In earlier work on nonlinear evolution systems (see for instance Crandall [3]) the assumption of accretiveness on the argu- ments of A provides existence of resolvents for A from which product formulas for the evolution system can be constructed. Theorem 3 shows that if the linear differentiability of the hypothesis of Theorem 2 is strengthened, then the resolvent structure for the underlying semigroup is carried forward to the evolution system. THEOREM 3: in [0, )/ R(H,H) L(H,H) and that T is a strongly uous is Suppose that G: continuous is differentiable semigroup of contin- transformations on H with generator B so that If t>0, a Lipschitz transformation on H, {(I that the Lipschitz norms for Define set. U(t,s) to be if 0 < m < k, -i and that there is a number M so tB) -i G(t)T(t-s)G(s) infinitesimal generator for U. (l-tB) < t 0 -i < M} is a bounded and let A denote the Then if t > O, there is k> 0 so that mA(t)) -I is a Lipschitz transformation with (I domain H and Lipschitz norm no greater than (IG(t) (I IG(t)-ll mB)-i IG’ (t) IG( t)-I / (I mlG(t) I(I mB)-i ). PROOF Initially note that the differentiability assumed for Theorem 3 is in the norm topology of L(H,H), t > O, G’(t) is in L(H,H). written as that is, if Hence, from Theorem i, A(t) can be T(t)BG(t) -I + G’ (t)G(t) -I Pick x from H. For I mA(t) 37 SEMIGROUP STRUCTURE UNDERLYING EVOLUTIONS to be invertible with domain H there must be exactly one point y so that Fm: (I mA(t))(y) Fm(p) H-H by For m> 0, define a function x. mB)-l(mG(t)-iG ’(t)G(t)-l(p) G(t)(l -I c>0, let M denote an upper bound for {(I -nB) II Fro(P) Thus II Fm(q) m < min{(IG(t) IG(t)-llmlG’(t) )-i M G(t)-l(x)). Given __< c}. mB)-ll IG(t)-ll21G’(t)l I( I mlG(t) __< n + P c} implies that F m q is a strict contraction and has a unique fixed point y. -mB)-l(mG(t)-lG ’(t)G(t)-l(y)+(t)-l(x)), it mg)G(t)-l(y) mG(t)-lG ’(t)G(t)-l(y)+(t)-l(x), (I G(t)-l(y) m(BG(t)-l(y) + G(t)-lG (t)G(t)-l(y)) G(t)-l(x), and G(t)(I Since y y m((G(t)BG(t) G’(t)G(t)-l)(y)) (I -mA(t)) Thus y For II(I + (I Yl mA(t)) -i (x -i 3 x. mA(t)) I) (I (x). -i (x l) and mA(t)) -i (x Y2 (I 2) mA(t)) G(t)(l -i mB)-lllG(t) -I 2 IG,(t) )) x x i (x2) mB) (mG(t)-iG ’(t)G(t)-l(yl) + G(t)-l(xl )) G(t)(l mB) (mG(t)-lG ’(t)G(t)-l(y 2) + G(t)-l(x2))II < IG(t)] I(l (mlG(t)-llm]G’(t)l flY I- Y2]I + IG(t)-ll x I x211 )" Thus -ii)/( I I( I mA(t))(y) that is, (I x the calculations shows that any such point must be a fixed point of Reversing Fm. -1 follows that -i mB) -i mIG(t) 2 OCCURRENCES. There appear to be several possibilities for the application of the evolution structure identified in the preceding theorems. Example 1. In [7], Neuberger studied semigroups of Lipschitz transformations so that if T is one of them, then there is c > 0 so that if x and y are elements of H, and t > 0, then G(t) e -ct T(t)(x) T(t)(y)II < e ct x y By taking I, Neuberger’s special evolution system which produced product formulas for the semigroup can be seen to be an instance of the theory. Example 2. Suppose that T is a semigroup of continuous transformations on H with infinitesimal generator B and G is a group of continuous linear trans- G.E. PARKER 38 formations on H with generator C. with generator A so that B Define U by PROOF: Then there is an evolution system U on H + C C A(O). G(t)-iT(t-s)G(s) -I. U(t,s) to Theorem i is satisfied and B(x) + lim h -i (G(h)(x) A(0)(x) G(O)BG(0) A(0)(x) x) -i If x e D (x) + the hypothesis B+C (0). gG(O)_l(x Hence B(x) + C(x). h+0 Thus, even though such sums may fail to guarantee the generation of semigroups (see Chernoff [2]), they must always support the generation of globally defined evolutions. Generation theories for evolution systems typically require that be dense in C and that (I mA(t)) -i be Lipschitz with domain C. D s < t A (s) These restrictions need not apply for the theory discussed in Theorem i to be viable. Example 3. from [0, oo) Let H be the Banach space of bounded uniformly continuous functions into R with for f e (A(s) (f))(x) fll H, I lub f’(x) if x 0 if x By taking T to be the semigroup generated by and G defined by (G(t) (f)) (x) U, defined by U(t,s) I G(t)T(t-s)G(s) {If(x) _< x > 0 }. Define A(s) by s > s {(f,g) f s H, g s H, and g f’} f(x) if x > t 2f(x) -i f(t) if x < t has generator A. Notice that the two conditions on the domain of A(s) are provided separately by the differentiability of T and G respectively. 0 < a < bDA(a) Here U(t,s) is defined globally although is nowhere dense in any neighborhood and I mA(t) does not have range H. Even when existing theories cover a given evolution system, the structures studied here may provide alternative computation possibilities. Example 4. y’(t) Consider the ordinary differential equation A/((Bt + C)y(t)) +-Dy(t)/(Bt + C), y(s) x, D/B _I_ 2 A typical solution might involve solving for y 2 in the associated linear equation. However, by letting G solve the linear term, a semigroup generator can be found -I by purely algebraic means from which U(t,s) can be built to G(t)T(t-s)G(s) 39 SEMIGROUP STRUCTURE UNDERLYING EVOLUTIONS Here the forms in which the two methods give the solve the original equation. solution are computationally different. Example 5. Suppose that A is a linear group generator and that C is a continuous function from H into the compact subset K of H. Consider, for x e H, the problem y’(t) (A + C)(y(t)), y(0) x which can be solved by the method of convergence of approximate solutions as employed by Martin[6]. Using the approach suggested in the theorems, let G denote the group gener- ated by A and look for a function f from [0, d] into H so that y {(t,G(t)(f(t))) 0 < t < d} solves the problem. G(t)-iCG(t)(f(t)). a solution it must be true that f’(t) X C([0, m),H), For such a function to be Letting this in turn translates into a fixed point problem for the function F defined by, for k e X, (F(k))(t) x + CG(j) is a compact map on [O,d] xH. and G(j) lim h->O h -i G(t+h)(h G(j) (G(t+h)(f(t+h)) -i CG(j)k 0 -i (f(t+h) f(t) G(t)G(t)-iCG(t)(f(t)). G(t)(f(t)) lim h/0 If f is a fixed point for F, then h-l(G(h)G(t)(f(t)) AG(t)(f(t)) + G(t)(f’(t)) Hence y’(t) G(t)(f(t)) + AG(t)(f(t)) + (A + C)(y(t)) and y(0) x. Notice that the burden for solution is shifted from evaluating the convergence of approximate solutions to finding a fixed point associated with a compact map. 4. 0UESTIONS. In Example 4, when G is produced by solving the linear problem -Du(t)/(Bt+c), finding the appropriate associated semigroup generator u’(t) depends on finding a function In this case f(p) f so that A/((Bt A/p works. + C)x) cD/Bf(c-D/B(Bt + C)x). Any systematic application for the evolution system in this paper must include an analysis of the conditions under which a family {B(t)} can be algebraically resolved to the form {G(t)fG(t) -i }t < 0" t <0 Theorem i identifies a closed form for the evolution generator in case G is a linear family. Can other forms of the generator for G(t)T(t-s)G(s) identified when G is not linear? -I be G.E. PARKER 40 REFERECES I. Ball, John M. "Measurability and Continuity Conditions for Nonlinear Evolutionary Processes." .P.rp.ceedinFs AMS, 53(1976), 353-358. 2. Chernoff, Paul R. "Product Formulas, Nonlinear Semigroups, and Addition of Unbounded Operators." Memoirs AMS, no. 140(1974). 3. Crandall, M. G., and Pazy, A. "Nonlinear Evolution Equations in Banach Spaces." Israel J. Math., 11(1972), 57-94. 4. Goldstein, J. A. voor Wiskunde "An Example of a Nonlinear Semigroup." Nieuw Archief (3), XXIII(1974), 170-174. 5. Kato, T. 6. Martin, R. H., Jr. Banach Spaces. 7. Neuberger, J. W. "Nonlinear Semigroups and Evolution Equations." Japan, 19(1967), 508 ff. J. Math. So___c. Nonlinear Operators and Differential Equations in Wiley, New York, 1976. "Product Integral Formulae for Nonlinear Expansive Semigroups and Non-Expansive Evolution Systems." J. of Mat_h. and Mech., 19(1969), 403-409. 8. Neveu, J. "The’orie des semi-grousde Markov", Univ. Calif. Publ. Statistics 2(1958), 319-394.