18.034, Honors Differential Equations Prof. Jason Starr Lecture 3: Existence + Uniqueness, Part I Feb 9, 2004 1. Terminology: Let S be a subset of the plane IR2. An interior point of S is a point (t0,y0) for which there exits a, b>0 (depending on S and (t0,y0)) such that is contained in S. The notation b) y b, -(y x a) t a, -(t 0 0 0 0 b) y b, -(y x a) t a, -(t 0 0 0 0 is shorthand for the set of all pairs (t,y) such that b y y b -y and a t t a -t 0 0 0 0 . A limit point of S is a point (t,y) that is the limit of a convergent sequence every term of which is in S. Each point (t,y) of S is a point of S since it is the limit of the constant sequence : for all 0,1,2, i yi) (ti, y) (t, ) y , (t i i . 0,1,2, i . The interior of S is the collection of all interior points of S. The closure of S is the set of all limit points of S. According to your textbook, a region is a set that is equal to its interior. This is different than the def’n I gave in lecture (and which many mathematicians use), that a region is a set R that is contained in the closure of its interior. According to this terminology, an “open region” is a set that is equal to its interior, and a “closed region” is a set that is equal to the closure of its interior. However, I will follow the notation in the book from here on. Let R be a region (= “open region”) in IR2 and let be a R D bounded, closed region. In other words, (1) D is a subset of R; (2) there exists N>0 such that D N] [-N, N] [-N, D , (3) D is the closure of its interior. Also, since it doesn’t follow automatically from the above, we usually also demand that D is not empty. Let f(t,y) be a real-valued function defined on R. Let (t0,y0) be an interior point of D. Def’n: A solution of the IVP y(t)) f(t, y(t) 0 0 y ) y(t defined in D is a pair consisting of an interval y(t)) (I, b) (a, I and a real-valued function y(t) such that (0) t0 is in b) (a, (1) the graph of y(t) is contained in the interior of I (2) is differentiable y(t) (3) for all t in , b) (a, y(t)) f(t, (t) y' (4) 0 0 y ) y(t A solution is called a y) (I, maximal solution (or a maximally-extended solution) if for every solution , the interval I ) y , (I 1 1 1 is contained in I1 and y1 is the restriction of y to I1. 2. The main theorem is this.f(t,y) Thm [=Thms A.1.1, A.2.3, A.3.1] Let R be a region in the plane, let D R be a bounded, closed region, let be defined on R, and let (t R D y) f(t, 0,y0) be an interior point of 18.034, Honors Differential Equations Lecture 3 Prof. Jason Starr Page 1 of 5 D. If is continuous on R, and if the partial derivative ) , ( y t f yf is everywhere defined and continuous on R, then there exists a maximal solution (I,y) of the IVP: y) f(t, y' 0 0 y ) y(t Moreover, denoting , and exists---call them & y(b)---and , are b) (a, I y(t) lima t y(t) limb t y(a) y(a)) (a, y(b)) (b, boundary points of D i.e points of D that are not interior points. This statement is a bit complicated. What it really says is this: (1) Existence: There exists an interval contains t b) (a, 0 and a solution y(t) of the IVP defined on . b) (a, (2) Uniqueness: If y1(t) is any solution of y(t) defined on , then y b) (a, 1(t)=y(t) for all t in (a, . b) (3) Maximal extendability: A formal consequence of (1) & (2) is that there exists a solution where (a,b) is the largest possible interval on which any solution is defined. (4) Endpoints of y: The endpoints of the graph of y(t) lie on the boundary of D. Picture 18.034, Honors Differential Equations Lecture 3 Prof. Jason Starr Page 2 of 5 : R y The proof of the theorem proceeds in several stages. Most of the proof is carried out after replacing D by a small rectangle containing (t0,y0) (at the end, it will be shown how to deduce the theorem from this case). 3. Shrinking Di Lipschitz condition. Because (t0,y0) is an interior point of D, there exists such that . Because D is closed, also the closed rectangle is contained in D. 0 b a, D b) + y b, -(y a) + t a, -(t 0 0 0 0 b) + y b, -(y a) + t a, -(t 0 0 0 0 The maximum principle states that any continuous function on a bounded, closed region attains a (finite) maximum at some point in the closed region. It also attains its minimum at some point. As a consequence, there exists M>0 such that for all (t,y) in , b) + y b, -(y a) + t a, -(t 0 0 0 0 ( , ) f t y M .F D graph of y(t). Notice the endpoints of the graph are on the boundary of D. a b tSimilarly, because yf is defined and continuous, there exists L>0, such that for all (t,y) in b) + y b, -(y a) + t a, -(t 0 0 0 0 , L y t yf ) , ( . This last inequality has a useful consequence. Def’n: A function f defined on a rectangle b] + y b, -[y a] + t a, -[t 0 0 0 0 is a Lipschitz with respect to y, with Lipschitz constant L, if for every (t,y1) and (t,y2) in the rectangle, 2 1 2 1 ( , ) ( , ) f t y f t y L y y . The function f(t,y) is Lipschitz with respect to y, with Lipschitz constant L. To see this, note that by the mean value theorem, ) ).( , ( ) , ( ) , ( 1 2 3 1 2 y y y t yf y t f y t f for some y3 in (y1,y2). Therefore, since L y t yf ) , ( 3 , we have 2 1 2 ( , ) ( , ) f t y f t y L y y1 for all (t,y1) and (t,y2) in the rectangle. Let c>0 be any positive number less than min 1 , ,2 b aM L . Observe that each of 1 , ,2 b aM L is positive, so also c is positive. The rectangle we work with is b] y b, -[y X C] t , [t 0 0 0 0 . 4. Metric spaces; Contraction mappings. There is a useful language, sometimes not introduced until 18.100, of metric spaces. The idea is to identify and make abstract all properties of Euclidean space that can be stated and proved using only the distance function. Def’n: A metric space is a pair (X, ) consisting of a set X and a real-valued function , defined for all pairs p,q of elements of X, and that satisfies: q) (p, (0) for all p,q , 0 q) (p, (1) iff (= “if and only if”) p=q 0 q) (p, (2) for all p,q, p) (q, q) (p, (i.e. is “symmetric”) (3) for all p,q and r, r) (q, q) (p, r) (p, (this is called the “triangle inequality”). Main example: X= IRn and ((x Eucl 1,….,xn), (y1,…..,yn))= 2 2 2 2 2 1 1 ( ) x (y .. ) x (y ) x y n n the usual Euclidean distance. Also, for X and subset of IRn , the restriction of to pairs in X defines a metric space. Eucl 2 very interesting examples. (1) Let [a,b] be any interval in IR. Let C([a,b], IR ) denote the set of all real-valued continuous functions y(t) defined on [a,b]. The distance is defined to be ) y , (y 2 1 maximum value of ) ( ) ( 1 2 t y t y on [a,b]. By the maximum principle, this exists and is finite. If , then for all t, 0 ) y , (y 2 1 0 ) ( ) ( 1 2 t y t y , i.e. . So if (t) y (t) y 1 2 0 ) y , (y 1 2 , then y2= y1. Symmetry of is obvious. Exercise (not to be handed in): Check that (y1,y2) satisfies the triangle inequality. 18.034, Honors Differential Equations Lecture 3 Prof. Jason Starr Page 3 of 5 (2) A slight variation of the last example is the one x we are interested in: b) , (y B 0 is defined to be the set of all real-valued continuous functions y(t) on [t0, t0+c] such that for all t, b ) ( 0 y t y . In other words, the set of continuous functions whose graph lies in the rectangle . The distance function is the same as in the previous example. b] y b, -[y x c] t , [t 0 0 0 0 Def’n: Let be a metric space. ) (X, A contraction mapping on (X,d) is a mapping T defined on X and taking values in X such that for every pair q p p,q of elements in X, . Let p) (q, T(p)) (T(q), 1 o be a real number. The mapping T is an -contraction mapping if for all p,q, we have p) (q, T(p)) (T(q), . Given a mapping T defined on X and taking values in X, a fixed point is an element p in X such that , i.e. p is p T(p) fixed by T. Many important theorems in mathematics are descriptions of fixed points of certain mappings. One of the simplest theorems is the “contraction mapping fixed point thm”. Thm [Contraction mapping fixed point thm, part I]: Let T be a contraction mapping on ) (X, . There is at most 1 fixed point of T. Pf: If T has no fixed points, the theorem is vacuously true. Thus suppose there exists a fixed point p. Let q be any point other than p. Then T(q)) (T(p), T(q)) (p, (b/c ) T(p) p < q) (p, (b/c T is a contraction mapping). Since ,in particular q) (p, T(q)) (p, q q T ) ( . So p is the unique fixed point. The second half of this theorem will assert that there exists a fixed point of T if T is an -contraction mapping for some 1 0 and if X “has no holes”. 5. The integral operator. Let y(t) be a continuous function defined on [t0, t0+c] whose graph is contained in R. Define z = T(y) to be the function defined on [t0, t0+c] by . t t ds s y s f 0 )) ( , ( y z(t) 0 By the fundamental theorem of calculus, z(t) is differentiable and y(t)) f(t, (t) z' . In particular, z(t) is continuous. Suppose that the graph of z is in a] y a, -[y x c] t , [t 0 0 0 0 . Then for every t, t t ds s y s f y t z 0 )) ( , ( 0 ds s y s f t t 0 )) ( , ( t t s M 0 c M ) t -M(t 0 . Because we chose c< Mb , b c M 0 y t z . Thus z(t) is again an element of b) , (y B 0 . 18.034, Honors Differential Equations Lecture 3 Prof. Jason Starr Page 4 of 5 So the rule that associates to each y(t) the function z is a mapping T from b) , (y B 0 to itself. Moreover, for any 2 elements y1, y2, for every t, t t ds s y s f s y s f t z t z 0 )) ( , ( )) ( , ( 1 2 1 2 ds s y s y L t t 0 ) ( ) ( . 1 2 (b/c of the Lipschitz condition satisfied by f) c ds y y L t t ) y , (y L ) t -(t ) y , (y L max 1 2 0 1 2 1 2 0 Because we chose L c 21 , we conclude ) , ( 21 1 2 1 2 y y t z t z for all t. Therefore y1) (y2, 21 T(y1)) (T(y2), . By part 1 of the contraction mapping fixed point theorem, there is at most one fixed point of T. 6. Tying up loose ends. Thm: Let R be a region in the plane, let f be a function on R, and let (t0,y0) be a point of R. Let I=(a,b) be an interval containing t0, and let y1,y2 be differentiable functions on (a,b) whose graphs are contained in R, both of which solve the IVP y) f(t, y' 0 0 y ) y(t If f and yf are continuous on R, then 2 1 y y . Proof: We prove that the restrictions of y1 and y2 to [t0,b] are equal. The proof that the restrictions to [a,t0] are equal is almost the same. If 1 2 y y on [t0,b], then there exits a largest number t1 with such that the restrictions of y b t t 1 0 1,y2 to [t0,t1] are equal. After replacing t0 by t1 and y0 by , we may suppose that y ) (t y (t) y 1 2 1 1 & y2 are sol’ns and for all >0, there exists t with such that t t t 0 1 0 (t) y (t) y 1 2 . Let b] y b, -[y x a] t a, -[t 0 0 0 0 be a rectangle contained in R. Let M and L be as above. Because y2 and y1 are continuous, there exists c