Feller semigroups Shiu-Tang Li Finished: April 19, 2013 Last updated: November 2, 2013 1 Feller semigroups Definitions 1.1 Let E be a LCCB space, which means second countable locally compact Hausdorff space. A complex function f is said to vanish at infinity if for every > 0, there exists a compact set K ⊂ E s.t. |f (x)| < for all x ∈ K c . The class of all continuous function f on E which vanish at infinity is called C0 (E). Notice that C0 (E) is a Banach space. Definition 1.2 Let X be a Banach space. We define B(X) to be the collection of all bounded linear maps from X to X. Definition 1.3 Let X be a Banach space. A semigroup {S(t)}t≥0 is a collection of mappings from [0, ∞) to B(X), which satisfies (i) S(t1 + t2 ) = S(t1 ) ◦ S(t2 ) for all t1 , t2 ≥ 0. (ii) S(0) = IX . If in addition we have limt↓0 S(t)x = x for every x ∈ X, then {S(t)}t≥0 is called a c0 -semigroup. If, furthermore, we have limt↓0 kS(t) − IX kB(X) = 0, then {S(t)}t≥0 is called a uniformly continuous semigroup. Definition 1.4 A Feller semigroup {S(t)}t≥0 is a c0 -semigroup on the Banach space C0 (E). Theorem 1.5 Let {Pt }t≥0 be transition function on (E, E ). Then {Pt }t≥0 is a Feller semigroup if and only if (i) Pt (C0 (E)) ⊂ C0 (E) for all t ≥ 0. (ii) For all f ∈ C0 (E), for all x ∈ E, limt↓0 (Pt f )(x) = f (x). 1 Proof. If {Pt }t≥0 is Feller, then all the properties above obviously hold. Conversely, assume that we have all the properties above for {Pt }t≥0 . We have for all x ∈ E, f ∈ C0 (E), f (x) = lim(Pt f )(x) = lim((Pt P0 )f )(x) = lim(Pt (P0 f ))(x) = (P0 f )(x), t↓0 t↓0 t↓0 so P0 is the identity map. Now the goal is to prove (Pt f )(x) → f (x) uniformly in x ∈ E when t ↓ 0. We divide the proof into several steps. Step 1. (Defining resolvents) Pn·2n −1Since (t, x) 7→ (Pt f )(x) is right continuf )(x), which is (R+ × ous in t, we define gn (t, x) := j=0 1{ jn ≤t< j+1 } (P j+1 2 2n 2n E, B(R+ ) ⊗ E ) measurable, and gn (t, x) → (Pt f )(x) for all (t, x) ∈ (R+ × E). It follows that the map (t, x) 7→ (Pt f )(x) is also (R+ × E, B(R+ ) ⊗ E ) measurable. R∞ Fix f ∈ C0 (E). We define the map Up f : x 7→ 0 e−pt (Pt f )(x) dt, p > 0. Pn·2n −1 −p· jn Since the map is just the pointwise limit of x 7→ 21n · j=0 e 2 (P jn f )(x), 2 which are all continuous maps, we have (Up f )(x) is measurable for all x ∈ E. Step 2. (A key property for Feller resolvents) Let T be the topology of (E, E ) s.t. E = σ(T ). Assume that (E, E ) is not compact. Then, by one-point compactifiction, we may define E ∗ = E ∪ {∞}, with the new topology T ∗ := T ∪ {U ⊂ E ∗ : E ∗ \ U compact in (E, T )}. Since E is LCCB and hence locally compact Hausdorff, (E ∗ , T ∗ ) is a compact Hausdorff space [See Richard Bass, real analysis for graduate students, Sec 20.9]. Besides, since (E, T ) is second countable, every base of (E, T ) has a countable subfamily which is still a base for (E, T ), and we may use this fact to construct a countable base for (E ∗ , T ∗ ), which shows (E ∗ , T ∗ ) is also second countable. In addition, every function f : E → R belongs to C0 (E) if and only if the extension f˜ : E ∗ → R, f˜(∞) = 0, f˜(x) = f (x) for x ∈ E is a continuous function on (E ∗ , T ∗ ). We claim that (Up f )(x) ∈ C0 (E) for all f ∈ C0 (E). First, we fix g f ∈ C0 (E). Since Pt f ∈ C0 (E), we may denote its extension by P t f , which ∗ belongs to C(E ) by the above arguments. We also denote the extension R ∞ −pt g g g of Up f by U (P p f , and its not hard to see that Up f = 0 e t f )(x) dt. To ∗ g see Up f ∈ C(E ), by the sequence lemma for first countable spaces, it suf2 fices to check that for any xn → x ∈ E ∗ , (Up f )(xn ) → (Up f )(x) ((E ∗ , T ∗ ) is second countable and hence first countable). But this is simply a fact from dominated convergence theorem; note that since kf kC0 (E) = M for some g M > 0, we have kPt f kC0 (E) ≤ M for all t ≥ 0, and hence kP t f kC(E ∗ ) ≤ M for all t ≥ 0. Now, again by the arguments in the preceding paragraph, Up f ∈ C0 (E) under the assumption that (E, E ) is not compact. When (E, E ) is compact, it’s much easier to prove Up f ∈ C0 (E), by using the same ideas as above, but avoiding all the compactification troubles. Step 3. (More properties for Feller resolvents) We claim that (pUp f )(x) → f (x) for all x ∈ E when p → ∞. Indeed, we first fix x ∈ E, for any > 0, we may pick δ = δ(x) > 0 so that |(Pt f )(x) − f (x)| R< whenever ∞ 0 ≤ t < δ. By dominated convergence theorem we have limp→∞ δ pe−pt (Pt f )(x) dt = Rδ Rδ 0. Since we also have 0 pe−pt |(Pt f )(x) − f (x)| dt ≤ 0 pe−pt dt ≤ and R δ −pt pe f (x) dt = f (x)(1 − epδ ), the claim follows. 0 Next, we prove the resolvent equation Up f − Uq f = (q − p)Up Uq f = (q − p)Uq Up f , for all f ∈ C0 (E). To see this, Z ∞ (Up Uq f )(x) = e−pt (Pt Uq f )(x) dt Z ∞ Z0 ∞ Z −pt e−qs (Ps f )(y) ds dt = e Pt (x, dy) 0 Z0 ∞ Z E∞ = e−pt e−qs (Ps+t f )(x) ds dt Z0 ∞ Z ∞ 0 = e−pt e−q(r−t) (Pr f )(x) dr dt Z0 ∞ t Z r e−pt e−q(r−t) dt dr (Pr f )(x) = 0 Z0 ∞ 1 e−pr − e−qr dr = (Up f − Uq f ). = (Pr f )(x) q−p q−p 0 From the resolvent equation, we know that for any p > 0, f ∈ C0 (E), Up f = Uq (f + (q − p)Up f ) ∈ Uq (C0 (E)). This showsR Up (C0 (E)) is indepen∞ dent of p > 0. Besides, for all x ∈ E, |(pUp f )(x)| ≤ 0 pe−pt kPt f kC0 (E) dt = kPt f kC0 (E) , so kpUp f kC0 (E) ≤ kPt f kC0 (E) ≤ kf kC0 (E) . Step 4. (Up (C0 (E)) is dense in C0 (E)) Assume that Up (C0 (E)) is not dense in C0 (E). Thus, Up (C0 (E)) is a proper closed linear subspace of C0 (E). Now we recall an application of Hahn-Banach theorem [Rudin, real 3 and complex analysis, Theorem 5.19]: Let M be a linear subspace of a normed linear space X, and let x0 ∈ X. Then x0 is in the closure M of M if and only if there is no bounded linear functional f on X s.t. f (x) = 0 for all x ∈ M but f (x0 ) 6= 0. Therefore, there is some bounded linear functional Λ : C0 (E) → R and some g ∈ C0 (E) \ Up (C0 (E)), so that Λ(f ) = 0 for all f ∈ Up (C0 (E)), but Λ(g) 6= 0. Next, we state a version of C0 -Riesz representation theorem [Lawrence Baggett, functional analysis, Theorem 1.4], as follows: Let X be a second countable locally compact Hausdorff space, and let φ be a bounded linear functional on C0 (X). Then φ is the difference φ1 − φ2 of two positive linear functionals φ1 and φ2 , Rwhence there exists a unique finite signed Borel measure µ such that φ(f ) = X f dµ for all f ∈ C0 (X). R We hence have 0 = pΛ(Up g) = Λ(pUp g) = E (pUp g)(x) dµ(x) for all p > 0, where µ is a finite signed Borel measure, Λ, g is defined in preceding paragraphs. Since R kpUp gkC0 (E) ≤ kgkC0 (E) , by bounded convergence theorem we have 0 = E g(x) dµ(x) = Λ(g) 6= 0, a contradiction. Therefore, Up (C0 (E)) is dense in C0 (E). Step 5 (The last approximation). For any f ∈ C0 (E), x ∈ E, we have |(Pt Up f )(x) − (Up f )(x)| Z ∞ Z ∞ Z −ps e−ps (Ps f )(x) ds| e (Ps f )(y) ds − =| Pt (x, dy) ZE∞ Z 0 Z0 ∞ =| e−ps Pt (x, dy)(Ps f )(y) ds − e−ps (Ps f )(x) ds| 0 E 0 Z ∞ Z ∞ −ps =| e (Ps+t f )(x) ds − e−ps (Ps f )(x) ds| 0 Z ∞ Z0 ∞ =|ept e−ps (Ps f )(x) ds − e−ps (Ps f )(x) ds| Zt ∞ Z0 ∞ Z t −ps −ps pt ≤|e e (Ps f )(x) ds − e (Ps f )(x) ds| + | e−ps (Ps f )(x) ds| t 0 Z ∞ t ≤|ept − 1| · kf kC0 (E) · | e−ps ds| + tkf kC0 (E) . t 4 As a result, k(Pt Up f ) − (Up f )kC0 (E) ≤ |ept − 1| · kf kC0 (E) + tkf kC0 (E) → 0, as t ↓ 0. Now, give > 0, and arbitrary g ∈ C0 (E), there exists some h ∈ C0 (E) and p > 0 so that kg − Up hkC0 (E) < 3 . Therefore, kPt g − gkC0 (E) ≤ kPt Up h − Pt gkC0 (E) + kPt Up h − Up hkC0 (E) + kg − Up hkC0 (E) ≤ 2kg − Up hkC0 (E) + kPt Up h − Up hkC0 (E) 2 < + kPt Up h − Up hkC0 (E) 3 < , where the last inequality holds for all t > 0 small enough. 5