18.101 Supplementary Notes

Math 18.101 Supplementary Notes 1 Introduction The change of variables formula in multi-variable calculus asserts that if U and V are open subsets of Rn and f : U → V a C 1 diffeomorphism then, for every continuous function, φ : V → R the integral � φ(y) dy V exists if and only if the integral � U φ ◦ f (x)| det Df (x)| dx exists, and if these integrals exist they are equal. A proof of this can be found in Chapter 4 of Munkres’ book. These notes contain an alternative proof of this result. This proof is due to Peter Lax. Our version of his proof in the notes below makes use of the theory of differential forms; but, as Lax shows in the papers cited at the end of this section (which we strongly recommend as collateral reading for this course), references to differential forms can be avoided, and his proof can be couched entirely in the language of elementary multivariable calculus. The virtue of Lax’s proof is that it allows one to prove a version of the change of variables theorem for other mappings besides diffeomorphisms, and involves a topological invariant, the degree of a mapping, which is itself quite interesting. Some properties of this invariant, and some topological applications of the change of variables formula will be discussed in §6 of these notes. Remark: The proof we are about to describe is somewhat simpler and more transparent if we assume that f is a C 2 diffeomorphism. We’ll henceforth make this assumption. Bibliography 1. P. Lax, Change of variables in multiple integrals I, Amer. Math. Monthly 106 (1999), 497-501. 2. ——, Change of variables in multiple integrals II, Amer. Math. Monthly 108 (2001), 115-119. 2 1 Sard’s theorem Let U be an open subset of Rn and f : U → Rn a C 1 mapping. A point p ∈ U is a critical point of f if the derivative of f at p, Df (p) : Rn → Rn , is not bijective. Let Cf be the set of critical points of f . This set can be quite large. For instance if f is the constant mapping, p ∈ U → c, then Df (p) = 0 for all p ∈ U; so Cf = U. Sard’s theorem asserts, however, that no matter how big Cf is, its image with respect to f is “small”. (For instance this is certainly the case in the example above where f (Cf ) = {c}). Theorem 1.1. The image, f (Cf ), of the critical set of f is of measure zero in Rn . Let Q ⊂ U be a cube of width ℓ. To prove Sard’s theorem we will first prove Lemma 1.2. Given ǫ > 0 there exists a δ > 0 such that |f (y) − f (x) − (Df )(x)(x − y)| ≤ ǫ|x − y | (1.1) for x, y ∈ Q and |x − y | ≤ δ. Proof. Df is a continuous function and Q is compact, so for every ǫ > 0, there exists a δ > 0 such that ǫ (1.2) |Df (x) − Df (y)| ≤ n for x, y ∈ Q and |x − y | < δ. Moreover, by the mean value theorem, for every pair of points x, y ∈ Q there exists a point z on the line joining x to y for which fi (y) − fi (x) = Dfi (z)(y − x) . (1.3) Hence if |x − y | < δ, fi (y) − fi (x) − Dfi (x)(y − x) = (Dfi (z) − Dfi (x))(x − y) . Thus |fi (y) − fi (x) − Dfi(x)(y − x)| ≤ n|Df (z) − Df (x)| |y − x| ≤ ǫ|y − x| . We will now prove that the set f (Cf ∩ Q) is of measure zero. First of all note that, by (1.3), |f (y) − f (x)| ≤ c|x − y | 3 (1.4) for x, y ∈ Q where c is the supremum over Q of n|Df |. Now divide Q into N n subcubes each of width, ℓ/N < δ. If R is one of these subcubes and R intersects Cf in a point x0 then by (1.1), |f (x) − f (x0 ) − Df (x0 )(x − x0 )| ≤ ǫ|x − x0 | (1.5) for all x ∈ R. Let y0 = f (x0 ) and let A = Df (x0 ). Then by (1.4) and (1.5), f (R) is contained in the intersection of the set |y − y0 − A(y − y0 )| ≤ ǫ Nℓ (1.6) |y − y0 | < c Nℓ . (1.7) and the set By assumption, A is not bijective; so its image is contained in a subspace W of Rn of dimension (n − 1). Let v be a unit vector perpendicular to W . If y − y0 satisfies √ (1.6), its Euclidean distance to W is less than nℓ/Nǫ; i.e., √ (1.8) |�y − y0 , v�| ≤ n Nℓ ǫ . Let w be the orthogonal projection of y − y0 onto W . Then y − y0 = av + w with a = �y − y0 , v�. Hence if ǫ < 1, (1.7) and (1.8) imply √ √ �w� ≤ |a| + �y − y0 � ≤ n Nℓ (c + ǫ) ≤ 2 n Nℓ c . This shows that the image of √ ℓ B is an (n − 1)√ Rℓis contained in a cylinder whose base dimensional ball of radius 2 n N c and whose height, by (1.8) is 2 n N ǫ. Moreover, √ B in turn is contained in an (n − 1)-dimensional cube whose sides are of length 4 n Nℓ c; so we finally conclude: If R ∩ Cf is non-empty f (R) is contained in a rectangular solid of volume √ 2ǫ (4c)n−1 ( nℓ)n n . N n Since there are at most N of these rectangles, f (Q ∩ Cf ) is contained in a finite union of rectangular solids whose total volume is less than 2c1 ǫ, where √ c1 = (4c)n−1 ( nℓ)n . Since c and ℓ don’t depend on ǫ, and since ǫ can be made arbitrarily small, it follows from this that f (Q ∩ Cf ) is of measure zero. (See exercise 5.) To conclude the proof of Sard’s theorem let Qi , i = 1, 2, 3, . . . be a covering of U by cubes. Then ∞ f (Cf ) = f (Qi ∩ Cf ) i=1 and since each of the sets on the right is of measure zero, f (Cf ) is of measure zero. 4 Exercises for §1 1. (a) Let f : R → R be the map: f (x) = (x2 − 1)2 . What is the set of critical points of f ? What is its image? (b) Same question for the map f (x) = sin x + x. (c) Same question for the map f (x) = � 0, x≤0 . − x1 e , x>0 2. Let f : Rn → Rn be an affine map, i.e., a map of the form f (x) = Ax + y0 where A : Rn → Rn is a linear map. Prove Sard’s theorem for f . 3. Let ρ : R → R be a C ∞ function which is supported in the interval (− 21 , 12 ) and has a maximum at the origin. Let r1 , r2 , r3 , . . ., be an enumeration of the rational numbers, and let f : R → R be the map f (x) = ∞ � i=1 ri ρ(x − i) . Show that f is a C ∞ map and show that the image of Cf is dense in R. (The moral of this example: Sard’s theorem says that the image of Cf is of measure zero, but the closure of this image can be quite large.) 4. Let S be a bounded subset of Rn . Show that if S is rectifiable it can be covered by a finite number of rectangles of total volume, νS + ǫ, where νS is the volume of S. Hint: Let Q be a rectangle containing S in its interior. Recall that � νS = 1S . Q Now let P be a partition of Q with U(1S , P ) ≤ νS + ǫ . 5. Let S be a bounded subset of Rn . Suppose that for every ǫ > 0 there exists a rectifiable set, A, with A ⊃ S and ν(A) < ǫ. Show that S is of measure zero. 6. Let U be an open subset of Rn and let f : U → Rn be a C 1 mapping. Let q be an element of Rn not in f (Cf ). Show that if p ∈ f −1 (q) there exists a neighborhood, U0 , of p with the property that p is the only point in U0 that gets mapped onto q. Hint: The inverse function theorem. 5 7. Let U be an open subset of Rn and f : U → Rn+k , k ≥ 1, a C 1 mapping. Prove that f (U) is of measure zero. Hint: Let π : Rn+k → Rn be the projection, π(x1 , . . . , xn+k ) = (x1 , . . . , xn ), and let V = π −1 (U). Show that the mapping, f ◦ π : V → Rn+k , has the same image as f and that Cf ◦π = V . 6 2 The Poincare lemma for rectangles Let ν be a k-form on Rn . We define the support of ν to be the closure of the set {x ∈ Rn , νx = � 0} and we say that ν is compactly supported if supp ν is compact. Let ω = f dx1 ∧ . . . ∧ dxn be a compactly supported n-form with f continuous. We define the integral of ω over Rn � ω Rn to be the usual integral of f over Rn � f dx . Rn (Since f is continuous and compactly supported this integral is well-defined.) Now let Q be the rectangle [a1 , b1 ] × · · · × [an , bn ] and let U = Int Q. The Poincare lemma for rectangles asserts: Theorem 2.1. Let ω be a compactly supported n-form of class C r , r ≥ 1, with supp ω ⊆ U. Then the following assertions are equivalent: � (a) ω = 0. (b) There exists a compactly supported (n−1)-form, µ, of class C r with supp µ ⊆ U satisfying dµ = ω. We will first prove that (b)⇒(a). Let µ= n � i=1 �i ∧ . . . ∧ dxn , fi dx1 ∧ . . . ∧ dx (the “hat” over the dxi meaning that dxi has to be omitted from the wedge product). Then n � ∂fi dx1 ∧ . . . ∧ dxn , dµ = (−1)i−1 ∂x i i=1 and to show that the integral of dµ is zero it suffices to show that each of the integrals � ∂f dx (2.1)i Rn ∂xi 7 is zero. By Fubini we can compute (2.1)i by first integrating with respect to the variable xi and then with respect to the remaining variables. But � � �xi =bi ∂f =0 dxi = f (x) �� ∂xi xi =ai since fi is supported on U. We will prove that (a) ⇒ (b) by proving inductively the following assertion. Lemma 2.2. If ω is supported on U and its integral is zero there exists an (n − 1)form, µ, with supp µ ⊆ U such that the form ω − dµ = f dx1 ∧ . . . ∧ dxn (2.2) has the property � f (x1 , . . . , xk−1 , xk , . . . , xn ) dxk · · · dxn = 0 . (2.3)k Note that for k = n + 1 (2.3)k is just the assertion that ω − dµ = 0, i.e., ω = dµ, and for k = 1, (2.3)k is just the assertion (a). Let’s prove that (2.3)k ⇒ (2.3)k+1. Proof. Since ω − dµ is supported on U its support is contained in a rectangle [c1 , d1] × · · · × [cn , dn ] with ai < ci < di < bi . Let g(x1 , . . . , xk ) = � f (x1 , . . . , xk , xk+1 , . . . , xn ) dxk+1 . . . dxn . (2.4) This function is compactly supported on the rectangle Q′ = [c1 , d1 ] × · · · × [ck , dk ] since the integrand on the right is zero when (x1 , . . . , xk ) lies outside Q′ . Moreover, since f is of class C r , this integral is of class C r as a function of x1 , . . . , xk . (See exercise 3 below.) Let � xk h(x1 , . . . , xk ) = g(x1 , . . . , xk−1 , s) ds ak for ak ≤ xk ≤ bk . By (2.3)k and (2.4), h(x1 , . . . , xk ) is zero if ak ≤ xk ≤ ck or if dl ≤ xk ≤ bk ; so h is compactly supported on Q′ , is of class C r and satisfies ∂ h=g ∂xk 8 (2.5) by the fundamental theorem of calculus. Now let ρ = ρ(xk+1 , . . . , xn ) be a compactly supported function of class C ∞ with support on the rectangle Q′′ = [ck+1 , dk+1] × · · · × [cn , dn ] and with the property � ρ = 1. (2.6) Q′′ Let �k ∧ . . . ∧ dxn . ν = (−1)k−1 h(x1 , . . . , xk )ρ(xk+1 , . . . , xn ) dx1 ∧ . . . ∧ dx (2.7) Then by (2.5) and (2.7) dν = g(x1 , . . . , xk )ρ(xk+1 , . . . , xn ) dx1 ∧ . . . ∧ dxn and hence by (2.2) and (2.3)k ω − d(µ + ν) = f (x1 , . . . , xn ) − g(x1 , . . . , xk )ρ(xk+1 , . . . , xn ) and by (2.6) and (2.4) � f (x1 , . . . , xn ) − g(x1 , . . . , xk )ρ(xk+1 , . . . , xn ) dxk+1 . . . dxn = g(x1 , . . . , xk ) − g(x1 , . . . , xk ) = 0 , so ω − d(µ + ν) satisfies (2.3)k . Hence, by induction if ω satisfies the hypothesis (a) of Theorem 2.1 it satisfies (2.3)k for all k and in particular satisfies (2.3)n+1 which just says that ω − dµ = 0. Exercises for §2. (1) Let f : R → R be a compactly supported function of class C r with support on the interval (a, b). Show that the following are equivalent. �b (a) a f (x) dx = 0. (b) There exists a function, g : R → R of class C r+1 with support on (a, b) dg = f. with dx Hint: Show that the function � x g(x) = f (s) ds a is compactly supported. 9 (2) Let f = f (x, y) be a compactly supported function on Rk ×Rℓ with the property that the partial derivatives ∂f (x, y) , i = 1, . . . , k , ∂xi and are continuous as functions of x and y. Prove the following “differentiation under the integral sign” theorem. Theorem 2.3. The function is of class C 1 and g(x) = � f (x, y) dy ∂g (x) = ∂xi � ∂f (x, y) dy . ∂xi Hints: For y fixed and h ∈ Rk , fi (x + h, y) − fi (x, y) = Dx fi (c)h for some point, c, on the line segment joining x to x + c. Using the fact that Dx f is continuous as a function of x and y and compactly supported, conclude: Lemma 2.4. Given ǫ > 0 there exists a δ > 0 such that for |h| ≤ δ |f (x + h, y) − f (x, y) − Dx f (x, g)h| ≤ ǫ|h| . Now let Q ⊆ Rℓ be a rectangle with supp f ⊆ Rk × Q and show that �� � |g(x + h) − g(x) − Dx f (x, y) dy h| ≤ ǫ vol (Q)|h| . Conclude that g is differentiable at x and that its derivative is � Dx f (x, y) dy . (3) Let f : Rk × Rℓ → R be a compactly supported continuous function. Prove Theorem 2.5. If all the partial derivatives of f (x, y) with respect to x of order ≤ r exist and are continuous as functions of x and y the function � g(x) = f (x, y) dy is of class C r . 10 3 The Poincare lemma for open subsets of Rn In this section we will generalize Theorem 2.1 to arbitrary connected open subsets of Rn . Theorem 3.1. Let U be a connected open subset of Rn and let ω be a compactly supported n-form of class C r with supp ω ⊂ U. The the following assertions are equivalent, � (a) ω = 0. (b) There exists a compactly supported (n − 1)-form µ of class C r with supp µ ⊆ U and ω = dµ. Proof. Proof that (b) ⇒ (a): Let Qi ⊆ U, i = 1, 2, 3, . . ., be a collection of rectangles with U = ∪ Int Qi , and let φi , i = 1, 2, 3, . . . be a partition �of unity with supp φi ⊆ Int φi µ, is finite; i.e., µ = �NQi . Since µ is compactly supported the sum, µ = φ µ for N large enough. Hence i=1 i dµ = N � d(φi µ) i=1 and since supp φi µ ⊆ Qi � dφiµ = 0 by Theorem 2.1. Proof that (a) ⇒ (b): Let ω1 and ω2 be compactly supported n-forms of class C r with support in U. We will write ω1 ∼ ω2 as shorthand notation for the statement: “There exists a compactly supported (n−1)form µ of class C r with support in U and with ω1 − ω2 = dµ.”, We will prove that (a) ⇒ (b) by proving an equivalent statement: Fix a rectangle Q0 ⊂ U and an n-form � r ω0 of class C with supp ω0 ⊆ Q0 and ω0 = 1. r Theorem � 3.2. If ω is a compactly supported n-form of class C with supp ω ⊆ U and c = ω then ω ∼ cω0 . Thus in particular if c = 0, Theorem 3.2 says that ω ∼ 0 proving that (a) ⇒ (b). To prove Theorem 3.2 let Qi ⊆ U be, as above, a collection of rectangles with U = ∪ Int Qi and let�φi be a partition of unity with supp φi ⊆ Int Qi . Replacing ω by the finite sum m i=1 φi ω, m large, it suffices to prove Theorem 3.2 for each of the summands φi ω. In other words we can assume that supp ω is contained in one of the open rectangles, Int Qi . Denote this rectangle by Q. Using the fact that U is connected we will also show that we can join Q0 to Q by a sequence of rectangles as in the figure below. 11 Q Q0 Lemma 3.3. There exists a sequence of rectangles, Ri , i = 0, . . . , N + 1 such that R0 = Q0 , RN +1 = Q and Int Ri ∩ Int Ri+1 is non-empty. Proof. Denote by A the set of points, x ∈ U, for which there exists a sequence of rectangles, Ri , i = 0, . . . , N + 1 with R0 = Q0 , with x ∈ Int RN +1 and with Int Ri ∩ Int Ri+1 non-empty. It is clear that this set is open and that its complement is open; so, by the connectivity of U, U = A. To prove Theorem 3.2 with supp ω ⊆ Q, select, for each i, a compactly supported � n-form νi of class C r with supp νi ⊆ Int Ri ∩ Int Ri+1 and with νi = 1. The difference, νi −νi+1 is supported in Int Ri+1 � , and its integral is zero; so by Theorem 2.1, νi ∼ νi+1 . Similarly, ω0 ∼ ν1 and, if c = ω, ω ∼ cνN . Thus, cω0 ∼ cν0 ∼ · · · ∼ cνN = ω, proving the theorem. 12 4 Proper mappings Let U and V be open subsets of Rn and Rk . A continuous mapping, f : U → V , is proper if, for every compact subset B of V , f −1 (B) is compact. Proper mappings have a number of nice properties which will be investigated in the exercises below. One obvious property is that if f is of class C r+1 and ω is a compactly supported k-form of class C r with support on V , f ∗ ω is a compactly supported k-form of class C r with support on U . Our goal in this section is to prove the following very general “change of variables theorem” for integrals of forms. Theorem 4.1. If U and V are connected open subsets of Rn and f : U → V is a proper C 2 mapping there exists a topological invariant of f called the degree of f , (denoted deg(f )) with the property that, for every compactly supported n-form, ω, of class C 1 with support in V � � f ∗ ω = deg(f ) U ω. (4.1) V Before we prove this identity let’s see what this “change of coordinates theorem” says in coordinates. If ω = ϕ(y) dy1 ∧ · · · ∧ dyn then at x ∈ U , f ∗ ω = (ϕ ◦ f )(x) det(Df (x)) dx1 ∧ · · · ∧ dxn ; so, in coordinates, (4.1) takes the form � � ϕ(y) dy = deg(f ) ϕ ◦ f (x) det(Df (x)) dx . V (4.2) U (See Munkres, §17.) Proof. Theorem 4.1 is an easy consequence of the Poincaré lemma that we proved in §2. Let ω0� be an n-form of class C 1 with support and with supp ω0 ⊂ V � compact ∗ and with ω0 = 1. If we set deg f = U f ω0 , then (4.1) clearly holds for ω0 . We 1 will prove that (4.1) holds � for every compactly supported n-form ω of class C with supp ω ⊆ V . Let c = V ω. Then by Theorem 3.1 ω − cω0 = dµ, where µ is an (n − 1)-form of class C 1 with supp µ ⊆ V . Hence f ∗ ω − cf ∗ ω0 = f ∗ dµ = df ∗ µ , and by part (a) of Theorem 3.1, � � � ∗ ∗ f ω = c f ω0 = deg(f ) ω . U V 13 We will next discuss a multiplicative property of the degree which is useful for computational purposes. Let U, V and W be connected open subsets of Rn and f : U → V and g : V → W proper C r mappings. We show below that deg(g ◦ f ) = deg(g) deg(f ) . (4.3) Proof. Let ω be a compactly supported n-form of class C 1 with support on W . Then (g ◦ f )∗ ω = g ∗ f ∗ ω ; so � U ∗ (g ◦ f ) ω = � ∗ ∗ g (f ω) = deg(g) U � = deg(g) deg(f ) ω. � f ∗ω V W From this multiplicative property it is easy to deduce the following result (which we will need in the next section). Theorem 4.2. Let A be a non-singular n × n matrix and fA : Rn → Rn the linear mapping associated with A. Then deg(fA ) = +1 if det A is positive and −1 if det A is negative. A proof of this result is outlined in exercises 5–8 below. Exercises for §4. (1) Let U be an open subset of Rn and φi, i = 1, 2, 3, . . ., a partition of unity on U. Show that the mapping, f : U → R defined by f= ∞ � kφk k=1 is a proper C ∞ mapping. (2) Let U and V be open subsets of Rn and Rk and let f : U → V be a proper continuous mapping. Prove: Theorem 4.3. If B is a compact subset of V and A = f −1 (B) then for every open subset, U0 , with A ⊆ U0 ⊆ U, there exists an open subset, V0 , with B ⊆ V0 ⊆ V and f −1 (V0 ) ⊆ U0 . 14 Hint: Let C be a compact subset of V with B ⊆ Int C. Then the set W = f −1 (C) − U0 is compact; so its image f (W ) is compact. Show that f (W ) and B are disjoint and let V0 = Int C − f (W ) . (3) Show that if f : U → V is a proper continuous mapping and X is a closed subset of U, then f (X) is closed. Hint: For p ∈ V − f (X) let C be a compact subset of V with p ∈ Int C. Show that if D = f −1 (C) then f (X ∩ D) is a compact set not containing p. Conclude that Int C − f (X) is an open neighborhood of p in V . (4) Let f : Rn → Rn be the translation, f (x) = x + a. Show that deg(f ) = 1. Hint: Let ψ : R → R be a compactly supported function of class C 1 . The identity � � ψ(t) dt = ψ(t − a) dt (4.4) is easy to prove by elementary calculus, and this identity proves the assertion above in dimension one. Now let φ(x) = ψ(x1 ) . . . ψ(xn ) (4.5) and compute the right and left sides of (4.2) Fubini’s theorem. (5) Let σ be a permutation of the numbers, 1, . . . , n and let fσ : Rn → Rn be the diffeomorphism, fσ (x1 , . . . , xn ) = (xσ(1) , . . . , xσ(n) ). Prove that deg fσ = sgn (σ). Hint: Let φ be the function (4.5). Show that if ω = φ(x) dx1 ∧ · · · ∧ dxn , then f ∗ ω = (sgn σ)ω. (6) Let f : Rn → Rn be the mapping f (x1 , . . . , xn ) = (x1 + λx2 , x2 , . . . , xn ). Prove that deg(f ) = 1. Hint: Let ω = φ(x1 , . . . , xn ) dx1 ∧ . . . ∧ dxn where φ : Rn → R is compactly supported and of class C 1 . Show that � � ∗ f ω = ϕ(x1 + x2 , x2 , . . . , xn ) dx1 . . . dxn and evaluate the integral on the right by Fubini’s theorem; i.e., by first integrating with respect to the x1 variable and then with respect to the remaining variables. Note that by (4.4) � � f (x1 + λx2 , x2 , . . . , xn ) dx1 = f (x1 , x2 , . . . , xn ) dx1 . 15 (7) Let f : Rn → Rn be the mapping f (x1 , . . . , xn ) = (λx1 , x2 , . . . , xn ) with λ �= 0. Show that deg f = +1 if λ is positive and −1 if λ is negative. Hint: In dimension 1 this is easy to prove by elementary calculus techniques. Prove it in d-dimensions by the same trick as in the previous exercise. (8) Let A be a non-singular n × n matrix and fA : Rn → Rn the linear mapping associated with A. Prove that deg(fA ) = +1 if det A is positive and −1 if det A is negative. Hint: Professor Munkres proves in §2 of chapter one that fA can be written as a composition of linear mappings, fE1 ◦ · · ·◦ fEk where the fEk ’s are mappings of the type described in the previous three exercises. Now use the identity (4.3). 16 5 The change of variables formula Let U and V be connected open subsets of Rn . If f : U → V is a C 2 -diffeomorphism, the determinant of Df (x) at x ∈ U is non-zero, and hence, since it is a continuous function of x, its sign is the same at every point. We will say that f is orientation preserving if this sign is positive and orientation reversing if it is negative. We will prove below: Theorem 5.1. The degree of f is +1 if f is orientation preserving and −1 if f is orientation reversing. We will then use this result to prove the following change of variables formula for diffeomorphisms. Theorem 5.2. Let φ : V → R be a compactly supported continuous function. Then � � φ ◦ f (x)| det(Df )(x)| = φ(y) dy . (5.1) U V Proof of Theorem 5.1. Given a point, a1 ∈ U, let a2 = −f (a1 ) and for i = 1, 2, let gi : Rn → Rn be the translation, gi (x) = x + ai . By (4.1) and exercise 4 of § 4 the composite diffeomorphism g2 ◦ f ◦ g1 (5.2) has the same degree as f , so it suffices to prove the theorem for this mapping. Notice however that this mapping maps the origin onto the origin. Hence, replacing f by this mapping, we can, without loss of generality, assume that 0 is in the domain of f and that f (0) = 0. Next notice that if A : Rn → Rn is a bijective linear mapping the theorem is true for A (by exercise 8 of § 4), and hence if we can prove the theorem for A−1 ◦ f , (4.1) will tell us that the theorem is true for f . In particular, letting A = Df (0), we have D(A−1 ◦ f )(0) = A−1 Df (0) = I where I is the identity mapping. Therefore, replacing f by A−1 f , we can assume that the mapping, f , for which we are attempting to prove Theorem 5.1 has the properties: f (0) = 0 and Df (0) = I. Let g(x) = x − f (x). Then these properties imply that g(0) = 0 and Df (0) = 0. Lemma 5.3. There exists a δ > 0 such that |g(x)| ≤ 12 |x| for |x| ≤ δ. Proof. Let g(x) = (g1 (x), . . . , gn (x)). Then ∂gi (0) = 0 ; ∂xj 17 so there exists a δ > 0 such that � � � � ∂gi 1 � � � ∂xj (x)� ≤ 2η for |x| ≤ δ. However, by the mean value theorem, gi (x) = � ∂gi (c)xj ∂xj for c = t0 x, 0 < t0 < 1. Thus |gi (x)| ≤ 1 1 sup |xi | = |x| , 2 2 so |g(x)| = sup |gi(x)| ≤ 1 |x| . 2 Let ρ be a compactly supported C ∞ function with 0 ≤ ρ ≤ 1 and with ρ(x) = 0 for |x| ≥ δ and ρ(x) = 1 for |x| ≤ 2δ , and let f˜ : Rn → Rn be the mapping It’s clear that f˜(x) = x + ρ(x)g(x) . (5.3) f˜(x) = x for |x| ≥ δ (5.4) δ f˜(x) = f (x) for |x| ≤ . 2 (5.5) 1 |f˜(x)| ≥ |x| . 2 (5.6) and, since f (x) = x + g(x), In addition, for all x ∈ Rn : Indeed, by (5.4), |f˜(x)| ≥ |x| for |x| ≥ δ, and for |x| ≤ δ |f˜(x)| ≥ |x| − ρ(x)|g(x)| ≥ |x| − |g(x)| ≥ |x| − 1 1 |x| = |x| 2 2 by Lemma 5.3. Now let Qr be the cube, {x ∈ Rn , |x| ≤ r}, and let Qcr = Rn − Qr . From (5.6) we easily deduce that f˜−1 (Qr ) ⊆ Q2r 18 (5.7) for all r, and hence that f˜ is proper. Also notice that for x ∈ Qδ , 3 |f˜(x)| ≤ |x| + |g(x)| ≤ |x| 2 by Lemma 5.3 and hence f˜−1 (Qc3 δ ) ⊆ Qcδ . (5.8) 2 Proof. We will now prove Theorem 5.1. Since f is a diffeomorphism mapping 0 to 0, it maps a neighborhood U0 of 0 in U diffeomorphically onto a neighborhood V0 of 0 in V , and by shrinking U0 if necessary we can assume that U0 is contained in Qδ/2 and V0 contained in Qδ/4 . Let ω be an n-form of class C 1 with support in V0 whose integral over Rn is equal to one. Then f ∗ ω is supported in U0 and hence in Qδ/2 . Also by (5.7) f˜∗ ω is supported in Qδ/2 . Thus both of these forms are zero outside Qδ/2 . However, on Qδ/2 , f˜ = f by (5.5), so these forms are equal everywhere, and hence � � ∗ deg(f ) = f ω = f˜∗ ω = deg(f˜) . Next let ω be a compactly supported n-form of class C 1 with support in Qc3δ/2 and with integral equal to one. Then f˜∗ ω is supported in Qδc by (5.8), and hence since f (x) = x on Qcδ f˜∗ ω = ω. Thus � � ∗ ˜ deg(f ) = f ω = ω = 1 . Putting these two identities together we conclude that deg(f ) = 1. If the function φ in Theorem 5.2 is a C 1 function, the identity (5.1) is an immediate consequence of the result above and the identity (4.2). If φ is not C 1 , but is just continuous, we will deduce Theorem 5.2 from the following result. Theorem 5.4. Let V be an open subset of Rn . If φ : Rn → R is a continuous function of compact support with supp φ ⊆ V ; then for every ǫ > 0 there exists a C ∞ function of compact support, ψ : Rn → R with supp ψ ⊆ V and sup |ψ(x) − φ(x)| < ǫ . Proof. Let A be the support of φ and let d be the distance in the sup norm from A to the complement of V . Since φ is continuous and compactly supported it is uniformly continuous; so for every ǫ > 0 there exists a δ > 0 with δ < d2 such that |φ(x)−φ(y)| < ǫ when |x − y| ≤ δ. Now let Q be the cube: |x| < δ and let ρ : Rn → R be a non-negative C ∞ function with supp ρ ⊆ Q and � ρ(y) dy = 1 . (5.9) 19 Set ψ(x) = � ρ(y − x)φ(y) dy . By Theorem 2.5, ψ is a C ∞ function. Moreover, if Aδ is the set of points in Rd whose / Aδ and y ∈ A , |x − y | > δ and distance in the sup norm from A is ≤ δ then for x ∈ hence ρ(y − x) = 0. Thus for x ∈ / Aδ � � ρ(y − x)φ(y) dy = ρ(y − x)φ(y) dy = 0 , A so ψ is supported on the compact set Aδ . Moreover, since δ < d2 , supp ψ is contained in V . Finally note that by (5.9) and exercise 4 of §4: � � ρ(y − x) dy = ρ(y) dy = 1 (5.10) and hence � φ(x)ρ(y − x) dy φ(x) − ψ(x) = � (φ(x) − φ(y))ρ(y − x) dy |φ(x) − ψ(x)| ≤ � |φ(x) − φ(y)| ρ(y − x) dy . φ(x) = so and But ρ(y − x) = 0 for |x − y | ≥ δ; and |φ(x) − φ(y)| < ǫ for |x − y | ≤ δ, so the integrand on the right is less than � ǫ ρ(y − x) dy , and hence by (5.10) |φ(x) − ψ(x)| ≤ ǫ . To prove the identity (5.1), let γ : Rn → R be a C ∞ cut-off function which is one on a neighborhood V1 of the support of φ, is non-negative, and is compactly supported with supp γ ⊆ V , and let � c = γ(y) dy . By Theorem 5.4 there exists, for every ǫ > 0, a C ∞ function ψ, with support on V1 satisfying (5.11) |φ − ψ | ≤ 2ǫc . 20 Thus � � �� � � |φ − ψ |(y) dy � (φ − ψ)(y) dy � ≤ V V � ≤ γ |φ − ψ |(y) dy V� γ(y) dy ≤ 2ǫ ≤ 2cǫ so � � �� � � φ(y) dy − ψ(y) dy � ≤ 2ǫ . � V Similarly, the expression (5.12) V � �� � � � (φ − ψ) ◦ f (x)| det Df (x)| dx� U is less than or equal to the integral � γ ◦ f (x)|(φ − ψ) ◦ f (x)| | det Df (x)| dx U and by (5.11), |(φ − ψ) ◦ f (x)| ≤ � ǫ 2c ǫ , 2c so this integral is less than or equal to γ ◦ f (x)| det Df (x)| dx and hence by (5.1) is less than or equal to 2ǫ . Thus � � �� � � � ψ ◦ f (x)| det Df (x)| dx� ≤ 2ǫ . � φ ◦ f (x) | det Df (x)|dx − � U U Combining (5.12), (5.13) and the identity � � ψ(y) dy = ψ ◦ f (x)| det Df (x)| dx V we get, for all ǫ > 0, � �� � � � φ(y) dy − φ ◦ f (x) | det Df (x) | dx �≤ǫ � and hence V U � � φ(y) dy = φ ◦ f (x)| det Df (x)| dx . 21 (5.13) Exercises for §5 (1) Let h : V → R be a non-negative continuous function. Show that if the improper integral � h(y) dy V is well-defined, then the improper integral � h ◦ f (x)| det Df (x)| dx U is well-defined and these two integrals are equal. Hint: If φi , i = 1, 2, 3, . . . is a partition of unity on V then ψi = φi ◦ f is a partition of unity on U and � � φi h dy = ψi (h ◦ f (x))| det Df (x)| dx . Now deduce the result above from Theorem 16.5 in Munkres’ book. (2) Show that the result above is true without the assumption that h is nonnegative. Hint: h = h+ − h− , where h+ = max(h, 0) and h− = max(−h, 0). (3) Show that, in the formula (4.2), one can allow the function, φ, to be a continuous compactly supported function rather than a C 1 compactly supported function. 22 6 The degree of a differentiable mapping Let U and V be open subsets of Rn and f : U → V a proper C 2 mapping. In this section we will show how to compute the degree of f and, in particular, show that it is always an integer. From this fact we will be able to conclude that the degree of f is a topological invariant of f : if we deform f smoothly, its degree doesn’t change. Definition 6.1. Let Cf be the set of critical points of f . A point, q ∈ V , is a regular value of f if it is not in the image, f (Cf ), of Cf . By Sard’s theorem “almost all” points, q in V are regular values of f ; i.e., the set of points which are not regular values of f is a set of measure zero. Notice, by the way, that a point, q, can qualify as a regular value of f by not being in the image of f . For instance, for the constant map, f : Rn → Rn , f (p) = c, the points, q ∈ Rn − {c} are all regular values of f . Picking a regular value, q, of f we will prove: Theorem 6.2. The set f −1 (q) is a finite set. Moreover, if f −1 (q) = {p1 , . . . , pn }, then there exist connected open neighborhoods Ui of pi in Y and an open neighborhood W of q in V such that: (i) for i �= j Ui and Uj are disjoint;  (ii) f −1 (W ) = Ui , (iii) f maps Ui diffeomorphically onto W . Proof. If p ∈ f −1 (q) then, since q is a regular value, p ∈ / Cf ; so Df (p) : Rn → Rn is bijective. Hence, by the inverse function theorem, f maps a neighborhood Up of p diffeomorphically onto a neighborhood of q. The open sets {Up , p ∈ f −1 (q)} are a covering of f −1 (q); and, since f is proper, f −1 (q) is compact; so we can extract a finite subcovering {Upi , i = 1, . . . , N}, and since pi is the only point in Upi which maps onto q, f −1 (q) = {p1 , . . . , pN }. Without loss of generality we can assume that the Upi ’s are disjoint from each other; for, if not, we can replace them by smaller neighborhoods of the pi ’s which have this property. By Theorem 4.3 there exists a connected open neighborhood W of q in V for which f −1 (W ) ⊂ Upi . To conclude the proof let Ui = f −1 (W ) ∩ Upi . 23 The main result of this section is a recipe for computing the degree of f by counting the number of pi ’s above, keeping track of orientation. Theorem 6.3. For each pi ∈ f −1 (q) let σpi = +1 if f : Ui → W is orientation preserving and −1 if f : Ui → W is orientation reversing. Then deg(f ) = N � σpi . (6.1) i=1 Proof. Let ω be a compactly supported n-form on W of class C 1 whose integral is one. Then � N � � ∗ deg(f ) = f ω= f ∗ω . U i=1 Ui Since f : Ui → W is a diffeomorphism � � ∗ f ω=± ω = +1 or − 1 Ui W depending on whether f : Ui → W is orientation preserving or not. Thus deg(f ) is equal to the sum (6.1). As we pointed out above, a point, q ∈ V can qualify as a regular value of f “by default”, i.e., by not being in the image of f . In this case the recipe (6.1) for computing the degree gives “by default” the answer zero. Let’s corroborate this directly. Theorem 6.4. If f : U → V isn’t onto, deg(f ) = 0. Proof. By exercise 3 of §4, V − f (U) is open; so if it is non-empty, there exists a compactly supported n-form, ω, of class C ∞ with support in V − f (U) and with integral equal to one. Since ω = 0 on the image of f , f ∗ ω = 0; so � � ∗ 0= f ω = deg(f ) ω = deg(f ) . U V Remark: In applications the contrapositive of this theorem is much more useful than the theorem itself. Theorem 6.5. If deg(f ) = � 0, then f maps U onto V . In other words if deg(f ) = � 0 the equation f (x) = y (6.2) has a solution, x ∈ U for every y ∈ V . We will now show that the degree of f is a topological invariant of f : if we deform f by a “homotopy” we don’t change its degree. To prove this we will need a slight generalization of the notion of “proper mapping”. 24 Definition 6.6. Let X be a subset of Rm and Y a subset of Rm . A continuous map f : X → Y is proper if for every compact subset, A, of Y , f −1 (A) is compact. In particular, let U and V be open subsets of Rn and let a be a positive real number. Suppose that g : [0, a] × U → V is a proper C 1 mapping. For t ∈ [0, a] let ft : U → V (6.3) be the mapping, ft (p) = g(p, t). Then, if A is a compact subset of V , ft−1 (A) is the intersection of the compact set {(s, p) ∈ [0, a] × U , g(s, p) ∈ A} with the set: s = t, and hence is compact. Therefore, for every t ∈ [0, a], ft : U → V is a proper C 1 mapping. If f0 = f , we will call the family of mappings ft , 0≤t≤a a “deformation” or “homotopy” of f . Theorem 6.7. For all t ∈ [0, a], the degree of ft is equal to the degree of f . Proof. Let ω = φ(y)dy1 ∧ · · · ∧ dyn be a compactly supported n-form of class C 1 on U with integral equal to one. Then the degree of ft is equal to the integral of ft∗ ω over U: � φ(g1 (x, t), . . . , gn (x, t)) det Dx g(x, t) dx . (6.4) U The integrand in (6.4) is continuous and is supported on a compact subset of [0, a]×U; hence (6.4) is continuous as a function of t. However, deg(ft ) is integer valued, so (6.4) is an (integer-valued) constant, not depending on t. There are many other nice applications of Theorem 6.3. We’ll content ourselves with two relatively simple and prosaic ones: Application 1. The Brouwer fixed point theorem Let B n be the closed unit ball in Rn : {x ∈ Rn , �x� ≤ 1} . 25 Theorem 6.8. If f : B n → B n is a C 2 mapping then f has a fixed point, i.e., f maps some point, x0 ∈ B n onto itself. The idea of the proof will be to assume that there isn’t a fixed point and show that this leads to a contradiction. Suppose that for every point, x ∈ B n f (x) = � x. Consider the ray through f (x) in the direction of x: f (x) + s(x − f (x)) , 0 ≤ s < ∞. This intersects the boundary, S n−1 , of B n in a unique point, γ(x), (see figure 1 below); and one of the exercises at the end of this section will be to show that the mapping γ : B n → S n−1, x → γ(x), is a C 2 mapping. Also it is clear from figure 1 that γ(x) = x if x ∈ S n−1 . f(x) x Ȗ(x) Figure 6.1. Let B n (r) be the ball, {x ∈ Rn , �x� ≤ r}. Since γ is a C 2 mapping of B n into Rn , there exists an open set, U, containing B n and a C 2 mapping of U into Rn whose restriction to B n is γ. (For the sake of economy of notation we’ll continue to call this map γ.) Since U is open and contains B n (1) it contains a slightly larger ball, B n (1 + δ0 ), δ0 > 0. We claim: Lemma 6.9. Given ǫ > 0 there exists 0 < δ < δ0 such that for 1 ≤ �x� ≤ 1 + δ, �γ(x) − x� is less than ǫ. Proof. Since γ is uniformly continuous on B(1 + δ0 ) there exists 0 < δ < δ0 such that for x, y ∈ B(1 + δ0 ) and �x − y� < δ, �γ(x) − γ(y)� is less than ǫ/2. Moreover we can assume δ < 2ǫ . Let 1 ≤ �x� ≤ 1 + δ and let y = x/�x�. Then x − y = �x�y − y = (�x� − 1)y 26 so �x − y� = �x� − 1 ≤ δ . Hence �γ(x) − γ(y)� is less than ǫ/2; therefore, since γ(y) = y, �γ(x) − y� is less than ǫ/2. Thus ǫ �γ(x) − x� ≤ �γ(x) − y� + �y − x� ≤ + δ ≤ ǫ . 2 Now let ϕ : Rn → R be a C ∞ function, with 0 ≤ ϕ ≤ 1, which is one on the set, �x� ≤ 1 + δ/2, and zero on the set, �x� ≥ 1 + δ, and let g(x) = ϕ(x)γ(x) + (1 − ϕ(x))x . (6.5) g(x) = γ(x) for �x� ≤ 1 + δ/2 (6.6) g(x) = (x) for �x� ≥ 1 + δ . (6.7) �g(x)� ≥ 1 − ǫ . (6.8) The mapping defined by (6.5) is a C 2 mapping of Rn into Rn with the properties and We claim that on the set 1 ≤ �x� ≤ 1 + δ Indeed, since g(x) = ϕ(x)(γ(x) − x) + x �g(x)� ≥ �x� − �γ(x) − x� and by Lemma 6.9, �γ(x) − x� < ǫ if 1 ≤ �x� ≤ 1 + δ. On the other hand, for �x� ≤ 1, g(x) = γ(x) and �γ(x)� = 1 so �g(x)� = 1 and for �x� ≥ 1 + δ g(x) = x, so �g(x)� ≥ 1 + δ. Hence �g(x)� ≥ 1 − ǫ (6.9) for all x. Moreover, by (6.6), g(x) is proper. Let’s compute its degree. (6.8) tells us that if ǫ < 1, the origin is not in the image of g, so by Theorem 6.4 the degree of g is zero. On the other hand g is equal to the identity map on the set �x� ≥ 1 + δ, and the degree of the identity map is one, hence so is the degree of g. This gives us he contradiction we were looking for a proves by contradiction that f has to have a fixed point. Application 2. The fundamental theorem of algebra Let p(z) = z n + an−1 z n−1 + · · · + a1 z + a0 be a polynomial of degree n with complex coefficients. If we identify the complex plane C = {z = x + iy ; x, y ∈ R} with R2 via the map, (x, y) ∈ R2 → z = x + iy, we can think of p as defining a mapping p : R2 → R2 , z → p(z) . We will prove 27 Theorem 6.10. The mapping, p, is proper and deg(p) = n. Proof. For 0 ≤ t ≤ 1 let pt (z) = (1 − t)z n + tp(z) n−1 � n = z +t ai z i . i=0 We will show that the mapping g : [0, 1] × R2 → R2 , z → pt (z) is a proper mapping. Let C = sup{|ai | , i = 0, . . . , n − 1} . Then for |z | ≥ 1 |a0 + · · · + an−1 z n−1 | ≤ |a0 | + |a1 ||z | + · · · + |an−1 | |z |n−1 ≤ C|z|n−1 , and hence, for |z | > 2C, |pt (z)| ≥ |z |n − C |z |n−1 ≥ C|z|n−1 . If A is a compact subset of C then for some R > 0, A is contained in the disk, |w | ≤ R and hence g −1(A) is contained in the set {(t, z) ; 0 ≤ t ≤ 1 , |pt (z)| ≤ R} ; and hence in the compact set {(t, z) ; 0 ≤ t ≤ 1 , C |z |n−1 ≤ R} ; and this shows that g is proper. Thus each of the mappings, pt : C → C , is proper and deg pt = deg p1 = deg p = deg p0 . However, p0 : C → C is just the mapping, z → z n and an elementary computation (see exercises 5 and 6 below) shows that the degree of this mapping is n. In particular for n > 0 the degree of p is non-zero; so by Theorem 6.4 we conclude that p : C → C is surjective and hence has zero in its image. Theorem 6.11. (fundamental theorem of algebra) Every polynomial, p(z) = z n + an−1 z n−1 + · · · + a0 , with complex coefficients has a complex root. 28 Exercises for §6 (1) Let W be a subset of Rn and let a(x), b(x) and c(x) be real-valued functions on W of class C r . Suppose that for every x ∈ W the quadratic polynomial a(x)s2 + b(x)s + c(x) (*) has two distinct real roots, s+ (x) and s− (x), with s+ (x) > s− (x). Prove that s+ and s− are functions of class C r . Hint: What are the roots of the quadratic polynomial: as2 + bs + c? (2) Show that the function, γ(x), defined in figure 1 is a C 1 mapping of B n onto S 2n−1 . Hint: γ(x) lies on the ray, f (x) + s(x − f (x)) , 0 ≤ s < ∞, and satisfies �γ(x)� = 1; so γ(x) is equal to f (x) + s0 (x − f (x)), where s0 is a non-negative root of the quadratic polynomial �f (x) + s(x − f (x))�2 − 1 . Argue from figure 1 that this polynomial has to have two distinct real roots. (3) Show that the Brouwer fixed point theorem isn’t true if one replaces the closed unit ball by the open unit ball. Hint: Let U be the open unit ball (i.e., the interior of B n ). Show that the map h : U → Rn , h(x) = x 1 − �x�2 is a diffeomorphism of U onto Rn , and show that there are lots of mappings of Rn onto Rn which don’t have fixed points. (4) Show that the fixed point in the Brouwer theorem doesn’t have to be an interior point of B n , i.e., show that it can lie on the boundary. (5) If we identify C with R2 via the mapping: (x, y) → z = x + iy, we can think of a C-linear mapping of C into itself, i.e., a mapping of the form z → cz , c∈C as being an R-linear mapping of R2 into itself. Show that the determinant of this mapping is |c|2 . 29 (6) (a) Let f : C → C be the mapping f (z) = z n . Show that Df (z) = nz n−1 . Hint: Argue from first principles. Show that for h ∈ C = R2 (z + h)n − z n − nz n−1 h |h| tends to zero as |h| → 0. (b) Conclude from the previous exercise that det Df (z) = n2 |z |2n−2 . (c) Show that at every point z ∈ C − 0, f is orientation preserving. (d) Show that every point, w ∈ C − 0 is a regular value of f and that f −1 (w) = {z1 , . . . , zn }, with σzi = +1. (e) Conclude that the degree of f is n. 30