Introduction to Complex Analysis Michael Taylor
2 Contents 0. Complex numbers, power series, and exponentials. Holomorphic functions, derivatives, and path integrals 2. Holomorphic functions defined by power series 3. Exponential and trigonometric functions: Euler s formula 4. Square roots, logs, and other inverse functions 5. The Cauchy integral theorem and the Cauchy integral formula 6. The maximum principle, Liouville s theorem, and the fundamental theorem of algebra 7. Harmonic functions on planar regions 8. Morera s theorem and the Schwarz reflection principle 9. Goursat s theorem 0. Uniqueness and analytic continuation. Singularities 2. Laurent series 3. Fourier series and the Poisson integral 4. Fourier transforms 5. Laplace transforms 6. Residue calculus 7. The argument principle 8. The Gamma function 9. The Riemann zeta function 20. Covering maps and inverse functions 2. Normal families 22. Conformal maps 23. The Riemann mapping theorem 24. Boundary behavior of conformal maps 25. The disk covers C \ {0, } 26. The Riemann sphere and other Riemann surfaces 27. Montel s theorem 28. Picard s theorems 29. Harmonic functions again: Harnack estimates and more Liouville theorems 30. Periodic and doubly periodic functions - infinite series representations 3. The Weierstrass in elliptic function theory 32. Theta functions and 33. Elliptic integrals 34. The Riemann surface of q(ζ) A. Metric spaces, convergence, and compactness B. Derivatives and diffeomorphisms C. Surfaces and metric tensors D. Green s theorem
E. Poincaré metrics F. The fundamental theorem of algebra (elementary proof) G. The Weierstrass approximation theorem H. Inner product spaces I. π 2 is irrational J. Euler s constant K. Rapid evaluation of the Weierstrass -function 3
4 Introduction This text covers material presented in complex analysis courses I have taught numerous times at UNC. The core idea of complex analysis is that all the basic functions that arise in calculus, first derived as functions of a real variable, such as powers and fractional powers, exponentials and logs, trigonometric functions and their inverses, and also a host of more sophisticated functions, are actually naturally defined for complex arguments, and are complex-differentiable (a.k.a. holomorphic). Furthermore, the study of these functions on the complex plane reveals their structure more truly and deeply than one could imagine by only thinking of them as defined for real arguments. An introductory 0 defines the algebraic operations on complex numbers, say z = x+iy and w = u + iv, discusses the magnitude z of z, defines convergence of infinite sequences and series, and derives some basic facts about power series (i.) f(z) = a k z k, k=0 such as the fact that if this converges for z = z 0, then it converges absolutely for z < R = z 0, to a continuous function. It is also shown that, for z = t real, f (t) = k ka kt k, for R < t < R. Here we allow a k C. We define the exponential function (i.2) e z = k=0 k! zk, and use these observations to deduce that, whenever a C, (i.3) d dt eat = ae at. We use this differential equation to derive further properties of the exponential function. While 0 develops calculus for complex valued functions of a real variable, introduces calculus for complex valued functions of a complex variable. We define the notion of complex differentiability. Given an open set Ω C, we say a function f : Ω C is holomorphic on Ω provided it is complex differentiable, with derivative f (z), and f is continuous on Ω. Writing f(z) = u(z) + iv(z), we discuss the Cauchy-Riemann equations for u and v. We also introduce the path integral and provide some versions of the fundamental theorem of calculus in the complex setting. (More definitive results will be given in 5.) In 2 we return to convergent power series and show they produce holomorphic functions. We extend results of 0 from functions of a real variable to functions of a complex variable. Section 3 returns to the exponential function e z, defined above. We extend (i.3) to (i.4) d dz eaz = ae az.
We show that t e t maps R one-to-one and onto (0, ), and define the logarithm on (0, ), as its inverse: (i.5) x = e t t = log x. We also examine the behavior of γ(t) = e it, for t R, showing that this is a unit-speed curve tracing out the unit circle. From this we deduce Euler s formula, (i.6) e it = cos t + i sin t. This leads to a direct, self-contained treatment of the trigonometric functions. In 4 we discuss inverses to holomorphic functions. In particular, we extend the logarithm from (0, ) to C \ (, 0], as a holomorphic function. We define fractional powers (i.7) z a = e a log z, a C, z C \ (, 0], and investigate their basic properties. We also discuss inverse trigonometric functions in the complex plane. In 5 we introduce a major theoretical tool of complex analysis, the Cauchy integral theorem. We provide a couple of proofs, one using Green s theorem and one based simply on the chain rule and the fundamental theorem of calculus. Cauchy s integral theorem leads to Cauchy s integral formula, and then to the general development of holomorphic functions on a domain Ω C in power series about any p Ω, convergent in any disk centered at p and contained in Ω. Results of 5 are applied in 6 to prove a maximum principle for holomorphic functions, and also a result called Liouville s theorem, stating that a holomorphic function on C that is bounded must be constant. We show that each of these results imply the fundamental theorem of algebra, that every non-constant polynomial p(z) must vanish somewhere in C. In 7 we discuss harmonic functions on planar regions, and their relationship to holomorphic functions. In 8 we establish Morera s theorem, a sort of converse to Cauchy s integral theorem. We use this in 9 to prove Goursat s theorem, to the effect that the C hypothesis can be dropped in the characterization of holomorphic functions. After a study of the zeros and isolated singularities of holomorphic functions in 0, we look at other infinite series developments of functions: Laurent series in 2 and Fourier series in 3. The method we use to prove a Fourier inversion formula also produces a Poisson integral formula for the unique harmonic function in a disk with given boundary values. Variants of Fourier series include the Fourier transform and the Laplace transform, discussed in 4 5. These transforms provide some interesting examples of integrals whose evaluations cannot be done with the techniques of elementary calculus. Residue calculus, studied in 6, provides a powerful tool for the evaluation of many definite integrals. A related tool with many important applications is the argument principle, studied in 7. In a sense these two sections lie at the heart of the course. In 8 and 9 we make use of many of the techniques developed up to this point to study two special functions, the Gamma function and the Riemann zeta function. In both 5
6 cases these functions are initially defined in a half-plane and then analytically continued as meromorphic functions on C. Sections 20 28 have a much more geometrical flavor than the preceding sections. We look upon holomorphic diffeomorphisms as conformal maps. We are interested in holomorphic covering maps and in implications of their existence. We also study the Riemann sphere, as a conformal compactification of the complex plane. We include two gems of nineteenth century analysis, the Riemann mapping theorem and Picard s theorem. An important tool is the theory of normal families, studied in 2. In 29 we return to a more function-theoretic point of view. We establish Harnack estimates for harmonic functions and use them to obtain Liouville theorems of a more general nature than obtained in 7. These tools help us produce some concrete illustrations of Picard s theorem, such as the fact that e z z takes on each complex value infinitely often. In 30 33 we provide an introduction to doubly periodic meromorphic functions, also known as elliptic functions, and their connection to theta functions and elliptic integrals, and in 34 we show how constructions of compact Riemann surfaces provide tools in elliptic function theory. This text concludes with several appendices. In Appendix A we collect material on metric spaces and compactness, including particularly the Arzela-Ascoli theorem, which is an important ingredient in the theory of normal families. We also prove the contraction mapping theorem, of use in Appendix B. In Appendix B we discuss the derivative of a function of several real variables and prove the Inverse Function Theorem, in the real context, which is used in 4 to get the Inverse Function Theorem for holomorphic functions on domains in C. Appendix C discusses metric tensors on surfaces. It is of use in 26, on the Riemann sphere and other Riemann surfaces. This material is also useful for Appendix E, which introduces a special metric tensor, called the Poincaré metric, on the unit disk and other domains in C, and discusses some connections with complex function theory, including another proof of Picard s big theorem. In between, Appendix D proves Green s theorem for planar domains, of use in one proof of the Cauchy integral theorem in 5. In Appendix F we give a proof of the Fundamental Theorem of Algebra that is somewhat different from that given in 7. It is elementary, in the sense that it does not rely on results from integral calculus. In Appendix G we show that a construction arising in 4 to prove the Fourier inversion formula also helps establish a classical result of Weierstrass on approximating continuous functions by polynomials. In fact, this proof is basically that of Weierstrass, and I prefer it to other proofs that manage to avoid complex function theory. Appendix H deals with inner product spaces, and presents some results of use in our treatments of Fourier series and the Fourier transform, in 3 4. In Appendix I, we present a proof that π 2 is irrational. Appendix J has material on Euler s constant. Appendix K complements 32 with a description of how to evaluate (z) rapidly via the use of theta functions. Acknowledgment
Thanks to Shrawan Kumar for testing this text in his Complex Analysis course, for pointing out corrections, and for other valuable advice. 7
8 0. Complex numbers, power series, and exponentials A complex number has the form (0.) z = x + iy, where x and y are real numbers. These numbers make up the complex plane, which is just the xy-plane with the real line forming the horizontal axis and the real multiples of i forming the vertical axis. See Figure 0.. We write (0.2) x = Re z, y = Im z. We write x, y R and z C. We identify x R with x + i0 C. If also w = u + iv with u, v R, we have addition and multiplication, given by (0.3) z + w = (x + u) + i(y + v), zw = (xu yv) + i(xv + yu), the latter rule containing the identity (0.4) i 2 =. One readily verifies the commutative laws (0.5) z + w = w + z, zw = wz, the associative laws (with also c C) (0.6) z + (w + c) = (z + w) + c, z(wc) = (zw)c, and the distributive law (0.7) c(z + w) = cz + cw, as following from their counterparts for real numbers. If c 0, we can perform division by c, z = w z = wc. c For z = x+iy, we define z to be the distance of z from the origin 0, via the Pythagorean theorem: (0.8) z = x 2 + y 2.
9 Note that (0.9) z 2 = z z, where we set (0.0) z = x iy, called the complex conjugate of z. One readily checks that (0.) z + z = 2 Re z, z z = 2i Im z, and (0.2) z + w = z + w, zw = z w. Hence zw 2 = zwz w = z 2 w 2, so (0.3) zw = z w. We also have, for c 0, z c = c 2 zc. The following result is known as the triangle inequality, as Figure 0.2 suggests. Proposition 0.. Given z, w C, (0.4) z + w z + w. Proof. We compare the squares of the two sides: (0.5) z + w 2 = (z + w)(z + w) = zz + ww + zw + wz = z 2 + w 2 + 2 Re(zw), while (0.6) ( z + w ) 2 = z 2 + w 2 + 2 z w = z 2 + w 2 + 2 zw. Thus (0.4) follows from the inequality Re(zw) zw, which in turn is immediate from the definition (0.8). (For any ζ C, Re ζ ζ.) We can define convergence of a sequence (z n ) in C as follows. We say (0.7) z n z if and only if z n z 0,
0 the latter notion involving convergence of a sequence of real numbers. Clearly if z n = x n + iy n and z = x + iy, with x n, y n, x, y R, then (0.8) z n z if and only if x n x and y n y. One readily verifies that (0.9) z n z, w n w = z n + w n z + w and z n w n zw, as a consequence of their counterparts for sequences of real numbers. We can define the notion of convergence of an infinite series (0.20) as follows. For each n Z +, set (0.2) s n = k=0 z k n z k. k=0 Then (0.20) converges if and only if the sequence (s n ) converges: (0.22) s n w = z k = w. Note that (0.23) so if (0.24) s n+m s n = k=0 n+m k=n+ n+m k=n+ z k <, k=0 then (0.20) converges. If (0.24) holds, we say the series (0.20) is absolutely convergent. An important class of infinite series is the class of power series (0.25) a k z k, k=0 with a k C. Note that if z 0 and (0.25) converges for z = z, then there exists C < such that (0.25A) a k z k C, k. Hence, if z r z, r <, we have (0.26) a k z k C k=0 r k = k=0 z k z k, C r <, the last identity being the classical geometric series computation. This yields the following.
Proposition 0.2. If (0.25) converges for some z 0, then either this series is absolutely convergent for all z C, or there is some R (0, ) such that the series is absolutely convergent for z < R and divergent for z > R. We call R the radius of convergence of (0.25). In case of convergence for all z, we say the radius of convergence is infinite. If R > 0 and (0.25) converges for z < R, it defines a function (0.27) f(z) = a k z k, z D R, k=0 on the disk of radius R centered at the origin, (0.28) D R = {z C : z < R}. Proposition 0.3. If the series (0.27) converges in D R, then f is continuous on D R, i.e., given z n, z D R, (0.29) z n z = f(z n ) f(z). Proof. For each z D R, there exists S < R such that z D S, so it suffices to show that f is continuous on D S whenever 0 < S < R. Pick T such that S < T < R. We know that there exists C < such that a k T k C for all k. Hence ( S ) k. (0.30) z D S = a k z k C T For each N, write (0.3) f(z) = S N (z) + R N (z), N S N (z) = a k z k, R N (z) = k=0 k=n+ a k z k. Each S N (z) is a polynomial in z, and it follows readily from (0.9) that S N is continuous. Meanwhile, ( S ) k (0.32) z D S R N (z) a k z k C = CεN, T k=n+ k=n+ and ε N 0 as N, independently of z D S. Continuity of f on D S follows. Remark. The estimate (0.32) says the series (0.27) converges uniformly on D S, for each S < R. A major consequence of material developed in 5 will be that a function on D R is given by a convergent power series (0.27) if and only if f has the property of being holomorphic on D R (a property that is defined in ). We will be doing differential and integral calculus on such functions. In this preliminary section, we restrict z to be real, and do some calculus, starting with the following.
2 Proposition 0.4. Assume a k C and (0.33) f(t) = a k t k k=0 converges for real t satisfying t < R. Then f is differentiable on the interval R < t < R, and (0.34) f (t) = ka k t k, k= the latter series being absolutely convergent for t < R. We first check absolute convergence of the series (0.34). Let S < T < R. Convergence of (0.33) implies there exists C < such that (0.35) a k T k C, k. Hence, if t S, (0.36) ka k t k C S k ( S T ) k, which readily yields absolute convergence. (See Exercise 3 below.) Hence (0.37) g(t) = ka k t k k= is continuous on ( R, R). To show that f (t) = g(t), by the fundamental theorem of calculus, it is equivalent to show (0.38) t 0 g(s) ds = f(t) f(0). The following result implies this. Proposition 0.5. Assume b k C and (0.39) g(t) = b k t k k=0 converges for real t, satisfying t < R. Then, for t < R, (0.40) t 0 g(s) ds = k=0 b k k + tk+,
3 the series being absolutely convergent for t < R. Proof. Since, for t < R, (0.4) b k k + tk+ R bk t k, convergence of the series in (0.40) is clear. Next, parallel to (0.3), write (0.42) g(t) = S N (t) + R N (t), N S N (t) = b k t k, R N (t) = b k t k. k=0 k=n+ Parallel to (0.32), if we pick S < R, we have (0.43) t S R N (t) Cε N 0 as N, so (0.44) and (0.45) t 0 t g(s) ds = 0 N k=0 R N (s) ds b t k k + tk+ + R N (s) ds, 0 t 0 R N (s) ds CRε N. This gives (0.40). We use Proposition 0.4 to solve some basic differential equations, starting with (0.46) f (t) = f(t), f(0) =. We look for a solution as a power series, of the form (0.33). If there is a solution of this form, (0.34) requires (0.47) a 0 =, a k+ = a k k +, i.e., a k = /k!, where k! = k(k ) 2. We deduce that (0.46) is solved by (0.48) f(t) = e t = k=0 k! tk, t R.
4 This defines the exponential function e t. Convergence for all t follows from the ratio test. (Cf. Exercise 4 below.) More generally, we define (0.49) e z = k=0 k! zk, z C. Again the ratio test shows that this series is absolutely convergent for all z C. Another application of Proposition 0.4 shows that (0.50) e at = solves (0.5) whenever a C. We claim that e at is the only solution to k=0 a k k! tk d dt eat = ae at, (0.52) f (t) = af(t), f(0) =. To see this, compute the derivative of e at f(t): (0.53) d ( e at f(t) ) = ae at f(t) + e at af(t) = 0, dt where we use the product rule, (0.5) (with a replaced by a), and (0.52). Thus e at f(t) is independent of t. Evaluating at t = 0 gives (0.54) e at f(t) =, t R, whenever f(t) solves (0.52). Since e at solves (0.52), we have e at e at =, hence (0.55) e at =, t R, a C. eat Thus multiplying both sides of (0.54) by e at gives the asserted uniqueness: (0.56) f(t) = e at, t R. We can draw further useful conclusions by applying d/dt to products of exponentials. Let a, b C. Then (0.57) d ( e at e bt e (a+b)t) dt = ae at e bt e (a+b)t be at e bt e (a+b)t + (a + b)e at e bt e (a+b)t = 0,
so again we are differentiating a function that is independent of t. Evaluation at t = 0 gives (0.58) e at e bt e (a+b)t =, t R. Using (0.55), we get (0.59) e (a+b)t = e at e bt, t R, a, b C, or, setting t =, (0.60) e a+b = e a e b, a, b C. We will resume study of the exponential function in 3, and derive further important properties. 5 Exercises. Supplement (0.9) with the following result. Assume there exists A > 0 such that z n A for all n. Then (0.6) z n z = z n z. 2. Letting s n = n k=0 rk, write the series for rs n and show that (0.62) ( r)s n = r n+, hence s n = rn+. r Deduce that (0.63) 0 < r < = s n, as n, r as stated in (0.26). 3. The absolute convergence said to follow from (0.36) can be stated as follows: (0.64) 0 < r < = kr k k= is absolutely convergent. Prove this. Hint. Writing r = s 2, 0 < s <, deduce (0.62) from the assertion (0.65) 0 < s < = ks k is bounded, for k N.
6 Note that this is equivalent to (0.66) a > 0 = k ( + a) k is bounded, for k N. Show that (0.67) ( + a) k = ( + a) ( + a) + ka, a > 0, k N. Use this to prove (0.66), hence (0.65), hence (0.64). 4. This exercise discusses the ratio test, mentioned in connection with the infinite series (0.49). Consider the infinite series (0.69) a k, a k C. k=0 Assume there exists r < and N < such that (0.70) k N = a k+ r. a k Show that (0.7) a k <. k=0 Hint. Show that (0.72) a k a N k=n l=0 r l = a N r. 5. In case (0.73) a k = zk k!, show that for each z C, there exists N < such that (0.70) holds, with r = /2. 6. This exercise discusses the integral test for absolute convergence of an infinite series, which goes as follows. Let f be a positive, monotonically decreasing, continuous function on [0, ), and suppose a k = f(k). Then a k < k=0 0 f(t) dt <.
7 Prove this. Hint. Use N a k k= N 0 f(t) dt N k=0 a k. 7. Use the integral test to show that, if a > 0, n= < a >. na 8. This exercise deals with alternating series. Assume b k 0. Show that ( ) k b k k=0 is convergent, be showing that, for m, n 0, n+m ( ) k b k bn. k=n 9. Show that k= ( )k /k is convergent, but not absolutely convergent. 0. Show that if f, g : (a, b) C are differentiable, then (0.74) d ( ) f(t)g(t) = f (t)g(t) + f(t)g (t). dt Note the use of this identity in (0.53) and (0.57).
8. Holomorphic functions, derivatives, and path integrals Let Ω C be open, i.e., if z 0 Ω, there exists ε > 0 such that D ε (z 0 ) = {z C : z z 0 < ε} is contained in Ω. Let f : Ω C. If z Ω, we say f is complex-differentiable at z, with derivative f (z) = a, if and only if (.) lim h 0 [f(z + h) f(z)] = a. h Here, h = h + ih 2, with h, h 2 R, and h 0 means h 0 and h 2 0. Note that (.2) lim h 0 and (.3) lim h 2 0 provided these limits exist. As a first set of examples, we have [f(z + h ) f(z)] = f h x (z), ih 2 [f(z + ih 2 ) f(z)] = i f y (z), (.4) f(z) = z = [f(z + h) f(z)] =, h f(z) = z = h [f(z + h) f(z)] = h h. In the first case, the limit exists and we have f (z) = for all z. In the second case, the limit does not exist. The function f(z) = z is not complex-differentiable. Definition. A function f : Ω C is holomorphic if and only if it is complex-differentiable and f is continuous on Ω. Adding the hypothesis that f is continuous makes for a convenient presentation of the basic results. In 9 it will be shown that every complex differentiable function has this additional property. So far, we have seen that f (z) = z is holomorphic. We produce more examples of holomorphic functions. For starters, we claim that f k (z) = z k is holomorphic on C for each k Z +, and (.5) d dz zk = kz k. One way to see this is inductively, via the following result.
9 Proposition.. If f and g are holomorphic on Ω, so is fg, and (.6) d dz (fg)(z) = f (z)g(z) + f(z)g (z). Proof. Just as in beginning calculus, we have (.7) [f(z + h)g(z + h) f(z)g(z)] h = [f(z + h)g(z + h) f(z)g(z + h) + f(z)g(z + h) f(z)g(z)] h = [f(z + h) f(z)]g(z + h) + f(z) [g(z + h) g(z)]. h h The first term in the last line tends to f (z)g(z), and the second term tends to f(z)g (z), as h 0. This gives (.6). If f and g are continuous, the right side of (.6) is also continuous, so f g is holomorphic. It is even easier to see that the sum of two holomorphic functions is holomorphic, and (.8) d dz (f(z) + g(z)) = f (z) + g (z), Hence every polynomial p(z) = a n z n + + a z + a 0 is holomorphic on C. We next show that f (z) = /z is holomorphic on C \ 0, with (.9) In fact, (.0) [ h z + h ] z d dz = h z = z 2. h z(z + h) = z(z + h), which tends to /z 2 as h 0, if z 0, and this gives (.9). Continuity on C \ 0 is readily established. From here, we can apply Proposition. inductively and see that z k is holomorphic on C \ 0 for k = 2, 3,..., and (.5) holds on C \ 0 for such k. Next, recall the exponential function (.) e z = k=0 k! zk, Introduced in 0. We claim that e z is holomorphic on C and (.2) d dz ez = e z.
20 To see this, we use the identity (0.60), which implies (.3) e z+h = e z e h. Hence (.4) h [ez+h e z ] = e z eh. h Now (.) implies (.5) and hence e h h = k= k! hk = k=0 (k + )! hk, (.6) lim h 0 e h h This gives (.2). Another proof of (.2) will follow from the results of 2. We next establish a chain rule for holomorphic functions. In preparation for this, we note that the definition (.) of complex differentiability is equivalent to the condition that, for h sufficiently small, =. (.7) f(z + h) = f(z) + ah + r(z, h), with (.8) lim h 0 r(z, h) h i.e., r(z, h) 0 faster than h. We write = 0, (.8) r(z, h) = o( h ). Here is the chain rule. Proposition.2. Let Ω, O C be open. If f : Ω C and g : O Ω are holomorphic, then f g : O C, given by (.9) f g(z) = f(g(z)), is holomorphic, and (.20) d dz f(g(z)) = f (g(z))g (z).
2 Proof. Since g is holomorphic, (.2) g(z + h) = g(z) + g (z)h + r(z, h), with r(z, h) = o( h ). Hence (.22) f(g(z + h)) = f(g(z) + g (z)h + r(z, h)) = f(g(z)) + f (g(z))(g (z)h + r(z, h)) + r 2 (z, h) = f(g(z)) + f (g(z))g (z)h + r 3 (z, h), with r 2 (z, h) = o( h ), because f is holomorphic, and then (.23) r 3 (z, h) = f (g(z))r(z, h) + r 2 (z, h) = o( h ). This implies f g is complex-differentiable and gives (.20). Since the right side of (.20) is continuous, f g is seen to be holomorphic. Combining Proposition.2 with (.9), we have the following. Proposition.3. If f : Ω C is holomorphic, then /f is holomorphic on Ω \ S, where (.24) S = {z Ω : f(z) = 0}, and, on Ω \ S, (.25) d dz f(z) = f (z) f(z) 2. We can also combine Proposition.2 with (.2) and get (.26) d dz ef(z) = f (z)e f(z). We next examine implications of (.2) (.3). The following is immediate. Proposition.4. If f : Ω C is holomorphic, then (.27) f x and f y exist, and are continuous on Ω, and (.28) f x = i f y on Ω, each side of (.28) being equal to f on Ω.
22 When (.27) holds, one says f is of class C and writes f C (Ω). Appendix B, if f C (Ω), it is R-differentiable, i.e., As shown in (.29) f((x + h ) + i(y + h 2 )) = f(x + iy) + Ah + Bh 2 + r(z, h), with z = x + iy, h = h + ih 2, r(z, h) = o( h ), and (.30) A = f x (z), B = f y (z). This has the form (.7), with a C, if and only if (.3) a(h + ih 2 ) = Ah + Bh 2, for all h, h 2 R, which holds if and only if (.32) A = i B = a, leading back to (.28). This gives the following converse to Proposition.4. Proposition.5. If f : Ω C is C and (.28) holds, then f is holomorphic. The equation (.28) is called the Cauchy-Riemann equation. Here is an alternative presentation. Write (.33) f(z) = u(z) + iv(z), u = Re f, v = Im f. Then (.28) is equivalent to the system of equations (.34) u x = v y, v x = u y. To pursue this a little further, we change perspective, and regard f as a map from an open subset Ω of R 2 into R 2. We represent an element of R 2 as a column vector. Objects on C and on R 2 correspond as follows. (.35) On C On R 2 ( ) x z = x + iy z = y ( ) u f = u + iv f = v ( ) h h = h + ih 2 h = h 2
23 As discussed in Appendix B, a map f : Ω R 2 is differentiable at z Ω if and only if there exists a 2 2 matrix L such that (.36) f(z + h) = f(z) + Lh + R(z, h), R(z, h) = o( h ). If such L exists, then L = Df(z), with ( u/ x u/ y (.37) Df(z) = v/ x v/ y ). The Cauchy-Riemann equations specify that ( ) α β (.38) Df =, α = u β α x, β = v x. Now the map z iz is a linear transformation on C R 2, whose 2 2 matrix representation is given by ( ) 0 (.39) J =. 0 ( α γ Note that, if L = β δ ), then ( β δ (.40) JL = α γ ), LJ = ( γ α δ β so JL = LJ if and only if α = δ and β = γ. (When JL = LJ, we say J and L commute.) When L = Df(z), this gives (.38), proving the following. Proposition.6. If f C (Ω), then f is holomorphic if and only if, for each z Ω, (.4) Df(z) and J commute. ), In the calculus of functions of a real variable, the interaction of derivatives and integrals, via the fundamental theorem of calculus, plays a central role. We recall the statement. Theorem.7. If f C ([a, b]), then (.42) b a f (t) dt = f(b) f(a). Furthermore, if g C([a, b]), then, for a < t < b, (.43) d t g(s) ds = g(t). dt a
24 In the study of holomorphic functions on an open set Ω C, the partner of d/dz is the integral over a curve, which we now discuss. A C curve (or path) in Ω is a C map γ : [a, b] Ω, where [a, b] = {t R : a t b}. If f : Ω C is continuous, we define (.44) γ f(z) dz = b a f(γ(t))γ (t) dt, the right side being the standard integral of a continuous function, as studied in beginning calculus (except that here the integrand is complex valued). More generally, if f, g : Ω C are continuous and γ = γ + iγ 2, with γ j real valued, we set (.44A) γ f(z) dx + g(z) dy = b a [ f(γ(t))γ (t) + g(γ(t))γ 2(t) ] dt. Then (.44) is the special case g = if (with dz = dx + i dy). The following result is a counterpart to (.42). Proposition.8. If f is holomorphic on Ω, and γ : [a, b] C is a C path, then (.45) γ f (z) dz = f(γ(b)) f(γ(a)). The proof will use the following chain rule. Proposition.9. If f : Ω C is holomorphic and γ : [a, b] Ω is C, then, for a < t < b, (.46) d dt f(γ(t)) = f (γ(t))γ (t). The proof of Proposition.9 is essentially the same as that of Proposition.2. address Proposition.8, we have b f (z) dz = f (γ(t))γ (t) dt a γ (.47) b d = f(γ(t)) dt a dt = f(γ(b)) f(γ(a)), To
25 the second identity by (.46) and the third by (.42). This gives Proposition.8. The second half of Theorem.7 involves producing an antiderivative of a given function g. In the complex context, we have the following. Definition. A holomorphic function g : Ω C is said to have an antiderivative f on Ω provided f : Ω C is holomorphic and f = g. Calculations done above show that g has an antiderivative f in the following cases: (.48) g(z) = z k, f(z) = k + zk+, k, g(z) = e z, f(z) = e z. A function g holomorphic on an open set Ω might not have an antiderivative f on all of Ω. In cases where it does, Proposition.8 implies (.49) γ g(z) dz = 0 for any closed path γ in Ω, i.e., any C path γ : [a, b] Ω such that γ(a) = γ(b). In 3, we will see that if γ is the unit circle centered at the origin, (.50) γ dz = 2πi, z so /z, which is holomorphic on C \ 0, does not have an antiderivative on C \ 0. In 4, we will construct log z as an antiderivative of /z on the smaller domain C \ (, 0]. We next show that each holomorphic function g : Ω C has an antiderivative for a significant class of open sets Ω C, namely sets with the following property. (.5) Whenever a + ib and x + iy Ω, (a, b, x, y R), the vertical line from a + ib to a + iy and the horizontal line from a + iy to x + iy belong to Ω. See Fig... Proposition.0. If Ω C is an open set satisfying (.5) and g : Ω C is holomorphic, then there exists a holomorphic f : Ω C such that f = g. Proof. Pick a + ib Ω, and set, for z = x + iy Ω, y x (.52) f(z) = i g(a + is) ds + g(t + iy) dt. b a
26 Theorem.7 readily gives (.53) f (z) = g(z). x We also have (.54) f x (z) = ig(a + iy) + y a g (t + iy) dt, y and applying the Cauchy-Riemann equation g/ y = i g/ x gives (.55) i f y x g = g(a + iy) + (t + iy) dt a t = g(a + iy) + [g(x + iy) g(a + iy)] = g(z). Comparing (.54) and (.55), we have the Cauchy-Riemann equations for f, and Proposition.0 follows. Examples of open sets satisfying (.5) include disks and rectangles, while C \ 0 does not satisfy (.5), as one can see by taking a + ib =, x + iy =. Exercises. Let f, g C (Ω), not necessarily holomorphic. Show that (.56) x (f(z)g(z)) = f x(z)g(z) + f(z)g x (z), y (f(z)g(z)) = f y(z)g(z) + f(z)g y (z), on Ω, where f x = f/ x, etc. 2. In the setting of Exercise, show that, on {z Ω : g(z) 0}, (.57) x g(z) = g x(z) g(z) 2, y g(z) = g y(z) g(z) 2. Derive formulas for x f(z) g(z) and y f(z) g(z).
27 3. In (a) (d), compute f/ x and f/ y. Determine whether f is holomorphic (and on what domain). If it is holomorphic, specify f (z). (a) (b) (c) (d) f(z) = z + z 2 +, f(z) = z + z 2 +, f(z) = e /z, f(z) = e z 2. 4. Find the antiderivative of each of the following functions. (a) (b) (c) f(z) = (z + 3) 2, f(z) = ze z2, f(z) = z 2 + e z, 5. Let γ : [, ] C be given by Compute f(z) dz in the following cases. γ γ(t) = t + it 2. (a) (b) (c) (d) f(z) = z, f(z) = z, f(z) = (z + 5) 2, f(z) = e z. 6. Do Exercise 5 with γ(t) = t 4 + it 2. 7. Recall the definition (.44) for f(z) dz when f C(Ω) and γ : [a, b] Ω is a C γ curve. Suppose s : [a, b] [α, β] is C, with C inverse, such that s(a) = α, s(b) = β. Set σ(s(t)) = γ(t). Show that f(z) dz = f(ζ) dζ, γ σ
28 so path integrals are invariant under change of parametrization. In the following exercises, let h f(z) = (f(z + h) f(z)). h 8. Show that h z z 2 uniformly on {z C : z > ε}, for each ε > 0. Hint. Use (.0). 9. Let Ω C be open and assume K Ω is compact. Assume f, g C(Ω) and h f(z) f (z), h g(z) g (z), uniformly on K. Show that Hint. Write h (f(z)g(z)) f (z)g(z) + f(z)g (z), uniformly on K. h (fg)(z) = h f(z) g(z + h) + f(z) h g(z). 0. Show that, for each ε > 0, A <, uniformly on {z C : ε < z A}. Hint. Use induction. h z n nz (n+)
29 2. Holomorphic functions defined by power series A power series has the form (2.) f(z) = a n (z z 0 ) n. n=0 Recall from 0 that to such a series there is associated a radius of convergence R [0, ], with the property that the series converges absolutely whenever z z 0 < R (if R > 0), and diverges whenever z z 0 > R (if R < ). We begin this section by identifying R as follows: (2.2) R = lim sup n a n /n. This is established in the following result, which reviews and complements Propositions 0.2 0.3. Proposition 2.. The series (2.) converges whenever z z 0 < R and diverges whenever z z 0 > R, where R is given by (2.2). If R > 0, the series converges uniformly on {z : z z 0 R }, for each R < R. Thus, when R > 0, the series (2.) defines a continuous function (2.3) f : D R (z 0 ) C, where (2.4) D R (z 0 ) = {z C : z z 0 < R}. Proof. If R < R, then there exists N Z + such that n N = a n /n < R = a n (R ) n <. Thus (2.5) z z 0 < R < R = a n (z z 0 ) n z z 0 n, R for n N, so (2.) is dominated by a convergent geometrical series in D R (z 0 ). For the converse, we argue as follows. Suppose R > R, so infinitely many a n /n /R, hence infinitely many a n (R ) n. Then z z 0 R > R = infinitely many a n (z z 0 ) n z z 0 n, R forcing divergence for z z 0 > R. The assertions about uniform convergence and continuity follow as in Proposition 0.3. The following result, which extends Proposition 0.4 from the real to the complex domain, is central to the study of holomorphic functions. A converse will be established in 5, as a consequence of the Cauchy integral formula.
30 Proposition 2.2. If R > 0, the function defined by (2.) is holomorphic on D R (z 0 ), with derivative given by (2.6) f (z) = na n (z z 0 ) n. n= Proof. Absolute convergence of (2.6) on D R (z 0 ) follows as in the proof of Proposition 0.4. Alternatively (cf. Exercise 3 below), we have (2.7) lim n n/n = = lim sup na n /n = lim sup a n /n, n n so the power series on the right side of (2.6) converges locally uniformly on D R (z 0 ), defining a continuous function g : D R (z 0 ) C. It remains to show that f (z) = g(z). To see this, consider (2.8) f k (z) = k a n (z z 0 ) n, g k (z) = n=0 k na n (z z 0 ) n. We have f k f and g k g locally uniformly on D R (z 0 ). By (.2) we have f k (z) = g k(z). Hence it follows from Proposition.8 that, for z D R (z 0 ), (2.9) f k (z) = a 0 + g k (ζ) dζ, for any path σ z : [a, b] D R (z 0 ) such that σ z (a) = z 0 and σ z (b) = z. Making use of the locally uniform convergence, we can pass to the limit in (2.9), to get (2.0) f(z) = a 0 + g(ζ) dζ. Taking σ z to approach z horizontally, we have (with z = x + iy, z 0 = x 0 + iy 0 ) and hence (2.) f(z) = a 0 + y σ z σ z y 0 g(x 0 + it) i dt + f (z) = g(z), x while taking σ z to approach z vertically yields f(z) = a 0 + x x 0 g(t + iy 0 ) dt + n= x y x 0 g(t + iy) dt, y 0 g(x + it) i dt,
and hence (2.2) f (z) = ig(z). y Thus f C (D R (z 0 )) and it satisfies the Cauchy-Riemann equation, so f is holomorphic and f (z) = g(z), as asserted. Remark. For a proof of Proposition 2.2 making a direct analysis of the difference quotient [f(z) f(w)]/(z w), see [Ahl]. It is useful to note that we can multiply power series with radius of convergence R > 0. In fact, there is the following more general result on products of absolutely convergent series. Proposition 2.3. Given absolutely convergent series (2.3) A = α n, B = β n, n=0 we have the absolutely convergent series (2.4) AB = γ n, γ n = n=0 Proof. Take A k = k n=0 α n, B k = k n=0 β n. Then (2.5) A k B k = with (2.6) R k = Hence (2.7) where (m,n) σ(k) α m β n, R k m k/2 k/2 n k A (2.8) A = n k/2 n=0 n α j β n j. j=0 k γ n + R k n=0 σ(k) = {(m, n) Z + Z + : m, n k, m + n > k}. α m β n + β n + B m k/2 α m, α n <, B = n=0 k/2 m k n k β n <. n=0 α m β n It follows that R k 0 as k. Thus the left side of (2.5) converges to AB and the right side to n=0 γ n. The absolute convergence of (2.4) follows by applying the same argument with α n replaced by α n and β n replaced by β n. 3
32 Corollary 2.4. Suppose the following power series converge for z < R: (2.9) f(z) = Then, for z < R, (2.20) f(z)g(z) = a n z n, g(z) = n=0 c n z n, c n = n=0 b n z n. n=0 n a j b n j. The following result, which is related to Proposition 2.3, has a similar proof. Proposition 2.5. If a jk C and j,k a jk <, then j a jk is absolutely convergent for each k, k a jk is absolutely convergent for each j, and ( ) (2.2) a jk = ( ) a jk = a jk. j k k j j,k Using this, we demonstrate the following. Proposition 2.6. If (2.) has a radius of convergence R > 0, and z D R (z 0 ), then f(z) has a convergent power series about z : (2.22) f(z) = j=0 b k (z z ) k, for z z < R z z 0. k=0 The proof of Proposition 2.6 will not use Proposition 2.2, and we can use this result to obtain a second proof of Proposition 2.2. Shrawan Kumar showed the author this argument. Proof of Proposition 2.6. There is no loss in generality in taking z 0 = 0, which we will do here, for notational simplicity. Setting f z (ζ) = f(z + ζ), we have from (2.) (2.23) f z (ζ) = = a n (ζ + z ) n n=0 n=0 k=0 n ( n a n k ) ζ k z n k, the second identity by the binomial formula (cf. (2.34) below). Now, (2.24) n=0 k=0 n ( n a n k ) ζ k z n k = a n ( ζ + z ) n <, n=0
33 provided ζ + z < R, which is the hypothesis in (2.22) (with z 0 = 0). Hence Proposition 2.5 gives (2.25) f z (ζ) = Hence (2.22) holds, with (2.26) b k = ( ( ) n a n k k=0 n=k n=k z n k ( ) n a n z n k. k This proves Proposition 2.6. Note in particular that (2.27) b = n= na n z n. ) ζ k. Second proof of Proposition 2.2. The result (2.22) implies f is complex differentiable at each z D R (z 0 ), and the computation (2.27) translates to (2.6), with z = z. Remark. The result of Proposition 2.6 is a special case of a general result on representing a holomorphic function by a convergent power series, which will be established in 5. Exercises. Determine the radius of convergence R for each of the following series. If 0 < R <, examine when convergence holds at points on z = R. (a) n=0 z n (b) z n n n= (c) n= z n n 2 (d) n=0 z n n!
34 (e) n=0 z n 2 n (f) n=0 z 2n 2 n 2. Show that if the power series (2.) has radius of convergence R > 0, then f, f,... are holomorphic on D R (z 0 ) and (2.28) f (n) (z 0 ) = n! a n. Here we set f (n) (z) = f (z) for n =, and inductively f (n+) (z) = (d/dz)f (n) (z). 3. Given a > 0, show that for n (2.29) ( + a) n + na. (Cf. Exercise 3 of 0.) Use (2.29) to show that (2.30) lim sup n and hence n /n, (2.3) lim n n/n =, a result used in (2.7). Hint. To get (2.30), deduce from (2.29) that n /n ( + a)/a /n. Then show that, for each a > 0, (2.32) lim n a/n =. For another proof of (2.3), see Exercise 4 of 4. 4. The following is a version of the binomial formula. If a C, n N, (2.33) ( + a) n = Another version is n k=0 (2.34) (z + w) n = ( ) n a k, k n k=0 ( ) n = k ( ) n z k w n k. k n! k!(n k)!.
35 Verify this identity and show that (2.33) implies (2.29) when a > 0. Hint. To verify (2.34), expand (2.35) (z + w) n = (z + w) (z + w) as a sum of monomials and count the number of terms equal to z k w n k. Use the fact that (2.36) ( ) n = number of combinations of n objects, taken k at a time. k 5. As a special case of Exercise 2, note that, given a polynomial (2.37) p(z) = a n z n + + a z + a 0, we have (2.38) p (k) (0) = k! a k, 0 k n. Apply this to (2.39) p n (z) = ( + z) n. Compute p (k) n (z), using (.5), then compute p (k) (0), and use this to give another proof of (2.33), i.e., (2.40) p n (z) = n k=0 ( ) n z k, k ( ) n = k n! k!(n k)!.
36 3. Exponential and trigonometric functions: Euler s formula Recall from 0 and that we define the exponential function by its power series: (3.) e z = + z + z2 2! + + zj j! + = By the ratio test this converges for all z, to a continuous function on C. Furthermore, the exponential function is holomorphic on C. This function satisfies j=0 z j j!. (3.2) d dz ez = e z, e 0 =. One derivation of this was given in. Alternatively, (3.2) can be established by differentiating term by term the series (3.) to get (by Proposition 2.2) (3.3) d dz ez = = j= k=0 (j )! zj = k! zk = e z. k=0 k! zk The property (3.2) uniquely characterizes e z. It implies (3.4) d j dz j ez = e z, j =, 2, 3,.... By (2.2), any function f(z) that is the sum of a convergent power series about z = 0 has the form (3.5) f(z) = j=0 f (j) (0) z j, j! which for a function satisfying (3.2) and (3.4) leads to (3.). A simple extension of (3.2) is (3.6) d dz eaz = a e az. Note how this also extends (0.5). As shown in (0.60), the exponential function satisfies the fundamental identity (3.7) e z e w = e z+w, z, w C.
37 For an alternative proof, we can expand the left side of (3.7) into a double series: (3.8) e z e w = j=0 z j j! k=0 w k k! = j,k=0 z j w k j!k!. We compare this with (3.9) e z+w (z + w) n =, n! n=0 using the binomial formula (cf. (2.34)) (3.0) (z + w) n = n j=0 ( ) n z j w n j, j ( ) n = j n! j!(n j)!. Setting k = n j, we have (3.) e z+w = n=0 j+k=n;j,k 0 n! n! j!k! zj w k = j,k=0 z j w k j!k!. Comparing (3.8) and (3.) again gives the identity (3.7). We next record some properties of exp(t) = e t for real t. The power series (3.) clearly gives e t > 0 for t 0. Since e t = /e t, we see that e t > 0 for all t R. Since de t /dt = e t > 0, the function is monotone increasing in t, and since d 2 e t /dt 2 = e t > 0, this function is convex. Note that (3.2) e t > + t, for t > 0. Hence (3.3) lim t + et = +. Since e t = /e t, (3.4) lim t et = 0. As a consequence, (3.5) exp : R (0, ) is smooth and one-to-one and onto, with positive derivative, so the inverse function theorem of one-variable calculus applies. There is a smooth inverse (3.6) L : (0, ) R.
38 We call this inverse the natural logarithm: (3.7) log x = L(x). See Figures 3. and 3.2 for graphs of x = e t and t = log x. Applying d/dt to (3.8) L(e t ) = t gives (3.9) L (e t )e t =, hence L (e t ) = e t, i.e., (3.20) Since log = 0, we get d dx log x = x. (3.2) log x = x dy y. An immediate consequence of (3.7) (for z, w R) is the identity (3.22) log xy = log x + log y, x, y (0, ). In 4, we will extend the logarithm into the complex domain, defining log z for z C \ (, 0]. We move next to a study of e z for purely imaginary z, i.e., of (3.23) γ(t) = e it, t R. This traces out a curve in the complex plane, and we want to understand which curve it is. Let us set (3.24) e it = c(t) + is(t), with c(t) and s(t) real valued. First we calculate e it 2 = c(t) 2 + s(t) 2. For x, y R, (3.25) z = x + iy = z = x iy = zz = x 2 + y 2 = z 2. It is elementary that (3.26) z, w C = zw = z w = z n = z n, and z + w = z + w.
39 Hence (3.27) e z = In particular, k=0 z k k! = ez. (3.28) t R = e it 2 = e it e it =. Hence t γ(t) = e it has image in the unit circle centered at the origin in C. Also (3.29) γ (t) = ie it = γ (t), so γ(t) moves at unit speed on the unit circle. We have (3.30) γ(0) =, γ (0) = i. Thus, for t between 0 and the circumference of the unit circle, the arc from γ(0) to γ(t) is an arc on the unit circle, pictured in Figure 3.3, of length (3.3) l(t) = t 0 γ (s) ds = t. Standard definitions from trigonometry say that the line segments from 0 to and from 0 to γ(t) meet at angle whose measurement in radians is equal to the length of the arc of the unit circle from to γ(t), i.e., to l(t). The cosine of this angle is defined to be the x-coordinate of γ(t) and the sine of the angle is defined to be the y-coordinate of γ(t). Hence the computation (3.3) gives (3.32) c(t) = cos t, s(t) = sin t. Thus (3.24) becomes (3.33) e it = cos t + i sin t, an identity known as Euler s formula. The identity (3.34) d dt eit = ie it, applied to (3.33), yields (3.35) d d cos t = sin t, dt sin t = cos t. dt
40 We can use (.3.7) to derive formulas for sin and cos of the sum of two angles. Indeed, comparing (3.36) e i(s+t) = cos(s + t) + i sin(s + t) with (3.37) e is e it = (cos s + i sin s)(cos t + i sin t) gives (3.38) cos(s + t) = (cos s)(cos t) (sin s)(sin t), sin(s + t) = (sin s)(cos t) + (cos s)(sin t). Derivations of the formulas (3.35) for the derivative of cos t and sin t given in first semester calculus courses typically make use of (3.38) and further limiting arguments, which we do not need with the approach used here. The standard definition of the number π is half the length of the unit circle. Hence π is the smallest positive number such that γ(2π) =. We also have ( π ) (3.39) γ(π) =, γ = i. 2 Furthermore, consideration of Fig. 3.4 shows that ( π ) (3.40) γ = 3 ( π ) 3 2 + 2 i, γ = 6 3 2 + 2 i. We now show how to compute an accurate approximation to π. This formula will arise by comparing two ways to compute the length of an arc of a circle. So consider the length of γ(t) over 0 t ϕ. By (3.29) we know it is equal to ϕ. Suppose 0 < ϕ < π/2 and parametrize this segment of the circle by (3.4) σ(s) = ( s 2, s), 0 s τ = sin ϕ. Then we know the length is also given by (3.42) l = τ 0 σ (s) ds = Comparing these two length calculations, we have τ 0 ds s 2. (3.43) τ 0 ds s 2 = ϕ, sin ϕ = τ,
when 0 < ϕ < π/2. As another way to see this, note that the substitution s = sin θ gives, by (3.35), ds = cos θ dθ, while s 2 = cos θ by (3.28), which implies (3.44) cos 2 t + sin 2 t =. Thus 4 (3.45) τ 0 ds ϕ = s 2 0 dθ = ϕ, again verifying (3.43). In particular, using sin(π/6) = /2, from (3.40), we deduce that (3.46) π /2 6 = 0 dx x 2. One can produce a power series for ( y) /2 and substitute y = x 2. (For more on this, see the exercises at the end of 5.) Integrating the resulting series term by term, one obtains (3.47) π 6 = a n ) 2n+, 2n + ( 2 n=0 where the numbers a n are defined inductively by (3.48) a 0 =, a n+ = 2n + 2n + 2 a n. Using a calculator, one can sum this series over 0 n 20 and show that (3.49) π = 3.4592653589. We leave the verification of (3.47) (3.49) as an exercise. Exercises Here s another way to demonstrate the formula (3.35) for the derivatives of sin t and cos t.. Suppose you define cos t and sin t so that γ(t) = (cos t, sin t) is a unit-speed parametrization of the unit circle centered at the origin, satisfying γ(0) = (, 0), γ (0) = (0, ), (as we did in (3.32)). Show directly (without using (3.35)) that γ (t) = ( sin t, cos t),
42 and hence deduce (3.35). (Hint. Differentiate γ(t) γ(t) = to deduce that, for each t, γ (t) γ(t). Meanwhile, γ(t) = γ (t) =.) 2. It follows from (3.33) and its companion e it = cos t i sin t that (3.50) cos z = eiz + e iz, sin z = eiz e iz 2 2i for z = t R. We define cos z and sin z as holomorphic functions on C by these identities. Show that they yield the series expansions (3.5) cos z = k=0 ( ) k (2k)! z2k, sin z = k=0 ( ) k (2k + )! z2k+. 3. Extend the identities (3.35) and (3.38) to complex arguments. In particular, for z C, we have (3.52) 4. We define cos(z + π 2 ) = sin z, cos(z + π) = cos z, cos(z + 2π) = cos z, sin(z + π 2 ) = cos z, sin(z + π) = sin z, sin(z + 2π) = sin z. (3.53) cosh y = ey + e y, sinh y = ey e y. 2 2 Show that cos iy = cosh y, sin iy = i sinh y, and hence (3.54) cos(x + iy) = cos x cosh y i sin x sinh y, sin(x + iy) = sin x cosh y + i cos x sinh y. 5. Define tan z for z (k + /2)π and cot z for z kπ, k Z, by (3.55) tan z = sin z cos z, Show that cos z cot z = sin z. (3.56) tan(z + π 2 ) = cot z, tan(z + π) = tan z, and (3.57) d dz tan z = cos 2 z = + tan2 z. 6. For each of the following functions g(z), find a holomorphic function f(z) such that f (z) = g(z). a) g(z) = z k e z, k Z +. b) g(z) = e az cos bz.
43 7. Concerning the identities in (3.40), verify algebraically that ( 2 + 3 2 i ) 3 =. Then use e πi/6 = e πi/2 e πi/3 to deduce the stated identity for γ(π/6) = e πi/6. 8. Let γ be the unit circle centered at the origin in C, going counterclockwise. Show that as stated in (.50). 9. One sets γ 2π z dz = 0 sec z = cos z, ie it dt = 2πi, eit so (3.57) yields (d/dz) tan z = sec 2 z. Where is sec z holomorphic? Show that d sec z = sec z tan z. dz 0. Show that + tan 2 z = sec 2 z.. Show that and d (sec z + tan z) = sec z (sec z + tan z), dz d dz (sec z tan z) = sec z tan2 z + sec 3 z = 2 sec 3 z sec z. (Hint. Use Exercise 0 for the last identity.) 2. The identities (3.53) serve to define cosh y and sinh y for y C, not merely for y R. Show that d d cosh z = sinh z, sinh z = cosh z, dz dz and cosh 2 z sinh 2 z =.
44 4. Square roots, logs, and other inverse functions We recall the Inverse Function Theorem for functions of real variables. Theorem 4.. Let Ω R n be open and let f : Ω R n be a C map. Take p Ω and assume Df(p) End(R n ) is invertible. Then there exists a neighborhood O of p and a neighborhood U of q = f(p) such that f : O U is one-to-one and onto, the inverse g = f : U O is C, and, for x O, y = f(x), (4.) Dg(y) = Df(x). A proof of this is given in Appendix B. This result has the following consequence, which is the Inverse Function Theorem for holomorphic functions. Theorem 4.2. Let Ω C be open and let f : Ω C be holomorphic. Take p Ω and assume f (p) 0. Then there exists a neighborhood O of p and a neighborhood U of q = f(p) such that f : O U is one-to-one and onto, the inverse g = f : U O is holomorphic, and, for z O, w = f(z), (4.2) g (w) = f (z). Proof. If we check that g is holomorphic, then (4.2) follows from the chain rule, Proposition.2, applied to g(f(z)) = z. We know g is C. By Proposition.6, g is holomorphic on U if and only if, for each w U, Dg(w) commutes with J, given by (.39). Also Proposition.6 implies Df(z) commutes with J. To finish, we need merely remark that if A is an invertible 2 2 matrix, (4.3) AJ = JA A J = JA. As a first example, consider the function Sq(z) = z 2. Note that we can use polar coordinates, (x, y) = (r cos θ, r sin θ), or equivalently z = re iθ, obtaining z 2 = r 2 e 2iθ. This shows that Sq maps the right half-plane (4.4) H = {z C : Re z > 0} bijectively onto C \ R (where R = (, 0]). Since Sq (z) = 2z vanishes only at z = 0, we see that we have a holomorphic inverse (4.5) Sqrt : C \ R H,
given by (4.6) Sqrt(re iθ ) = r /2 e iθ/2, r > 0, π < θ < π. We also write (4.7) z /2 = Sqrt(z). We can define other non-integral powers of z on C \ R. Before doing so, we take a look at log, the inverse function to the exponential function, exp(z) = e z. Consider the strip (4.8) Σ = {x + iy : x R, π < y < π}. Since e x+iy = e x e iy, we see that we have a bijective map (4.9) exp : Σ C \ R. Note that de z /dz = e z is nowhere vanishing, so (4.9) has a holomorphic inverse we denote log: (4.0) log : C \ R Σ. Note that (4.) log = 0. Applying (4.2) we have d (4.2) dz ez = e z = d dz log z = z. Thus, applying Proposition.8, we have (4.3) log z = z ζ dζ, where the integral is taken along any path from to z in C \ R. Now, given any a C, we can define (4.4) z a = Pow a (z), Pow a : C \ R C by (4.5) z a = e a log z. The identity e u+v = e u e v then gives (4.6) z a+b = z a z b, a, b C, z C \ R. In particular, for n Z, n 0, (4.7) (z /n ) n = z. Making use of (4.2) and the chain rule (.20), we see that d (4.8) dz za = a z a. While Theorem 4. and Corollary 4.2 are local in nature, the following result can provide global inverses, in some important cases. 45