SOLUTIONS TO HW6 Stein and Shakarchi, Chapter 5 Exercise 1. Let f (z) be a holomorphic function on the open unit disc D, and let {z1 , . . . , zN } be the zeros of f (z) inside D (counted with multiplicity). For α ∈ D, we let ψα (z) denote the function α−z ψα (z) = . 1 − αz This is similar to the function T (z) from Chapter 3, Problem 2. Consider the function f (z) g(z) = . ψz1 (z) · ψz2 (z) · · · ψzN (z) By construction, g(z) is nonvanishing on D. We proceed exactly as in Step 3 of the proof of Jensen’s formula to conclude that Z 2π 1 log(|g(eiθ )|) dθ. log(|g(0)|) = 2π 0 We are left to consider the Blaschke factors. Since the factors in Jensen’s formula are additive, we need only prove the formula for ψα (z) with α ∈ D. Since the only zero of ψα (z) occurs at z = α, this boils down to showing Z 2π Z 2π α − eiθ 1 1 iθ dθ. 0= log(|ψα (e )|) dθ = log 2π 0 2π 0 1 − αeiθ Indeed, we have 1 2π 2π Z 0 Z 2π iθ α − eiθ 1 1 |α − e | dθ = log log dθ 1 − αeiθ 2π 0 |eiθ | |e−iθ − α| ! Z 2π 1 |α − eiθ | = log dθ 2π 0 |α − eiθ | Z 2π 1 = log (1) dθ 2π 0 = 0. Exercise 3. Let τ be a fixed complex number with =(τ ) > 0, and set Θ(z|τ ) = ∞ X πin2 τ 2πinz e e n=−∞ = ∞ X 2 τ +2nz) eπi(n . n=−∞ Write τ = σ + it, with t > 0. We then have 2 τ +2nz) |eπi(n | = e<(πin 2 σ−πn2 t+2πinx−2πny) = eπ(−n 1 2 t−2ny) 2 t+2n|z|) ≤ eπ(−n 2 t/2 ≤ e−πn 2 SOLUTIONS TO HW6 for n ≥ 4|z|/t. Note that this type of inequality also holds if n ≤ −4|z|/t. Therefore, X X πi(n2 τ +2nz) πi(n2 τ +2nz) |Θ(z|τ )| = e + e |n|≤4|z|/t |n|≥4|z|/t X πi(n2 τ +2nz) X πi(n2 τ +2nz) + e e ≤ |n|≥4|z|/t |n|≤4|z|/t X X 2 2 ≤ eπ(−n t−2ny) + e−πn t/2 |n|≤4|z|/t ≤ 2 |n|≥4|z|/t −πn2 t 2πn|z| X e e + ≤ 2 e2πn|z| + Z ? ≤ ?? ≤ 2 t/2 e−πx dx + 1 r 2 +1 t 0≤n≤4|z|/t r 8|z| 2 2 + 2 e8π|z| /t + +1 t t ! r 8 2 2 2 |z|e8π|z| /t + + 3 e8π|z| /t t t ! r 2 8 8π|z|2 /t+|z| 2 e + + 3 e8π|z| /t+|z| t t ! r 8 2 2 + + 3 e1 e(8π/t+1)|z| , t t ≤ 2 ≤ ∞ −∞ 0≤n≤4|z|/t ≤ 2 t/2 e−πn n=−∞ 0≤n≤4|z|/t X ∞ X 8π|z|2 /t X e + where in the inequality (?) we have used the fact that |z| ≤ e|z| , and for the inequality 2 (??) we have used that |z| ≤ p|z| + 1. Therefore, we see that Θ(z|τ ) is of order at most 2 (with constants A = (8/t + 2/t + 3)e and B = (8π/t + 1)). We should remark that these computations can be done much less explicitly, but here we have instead opted for clarity. To show that the order of Θ(z|τ ) is exactly 2, we exploit the symmetry of the definition of Θ. Let α ∈ Z be an integer, and notice that, after reindexing the sum, we have ∞ X 2 Θ(ατ |τ ) = eπiτ (n +2nα) = n=−∞ ∞ X eπiτ ((n+α) 2 −α2 ) n=−∞ = e −πiτ α2 = e −πiτ α2 ∞ X 0 2 eπiτ (n ) n0 =−∞ Θ(0|τ ). SOLUTIONS TO HW6 3 Now assume that Θ(z|τ ) was of order ρ < 2. We would then have (for some positive constants A0 , B 0 ) 0 ρ A0 eB |z| ≥ |Θ(z|τ )|, and in particular 0 A0 eB |τ | ρ |α|ρ 2 2 ≥ |Θ(ατ |τ )| = |e−πiτ α Θ(0|τ )| = |Θ(0|τ )|eπtα . Taking the integer α sufficiently large gives us a contradiction. Exercise 9. Let |z| < 1, and set n ∞ Y Y k 2k fn (z) = (1 + z ), f (z) = (1 + z 2 ) = (1 + z)(1 + z 2 )(1 + z 4 )(1 + z 8 ) · · · . k=0 k=0 P 2k We first note that the infinite product for f (z) converges, since the sum ∞ converges. k=0 |z| Moreover, the sequence fn (z) converges to f (z) uniformly, by Proposition 3.2. We have n fn (z) = (1 + z)(1 + z 2 ) · · · (1 + z 2 ) n+1 −1 = 1 + a1 z + a2 z 2 + . . . + a2n+1 −1 z 2 , and by computing the expansion of the product, we see that aj is the number of ways to write the integer j as a sum j = α0 + α1 21 + α2 22 + . . . + αn 2n , with αk ∈ {0, 1}; that is, aj is the number of binary expansions of j. Since binary expansions are unique, we see that aj = 1 for every j. Hence n+1 2 fn (z) = 1 + z + z + . . . + z 2n+1 −1 1 − z2 . = 1−z Note that we may also prove this by induction; evidently f0 (z) = 1 + z = assume fn−1 (z) = n 1−z 2 1−z = 1 + z + ... + z 2n −1 1−z 2 , 1−z and if we , we have n n+1 1 − z2 1 − z2 n fn (z) = fn−1 (z)(1 + z ) = (1 + z 2 ) = . 1−z 1−z Using these results, we get 1 f (z) = lim fn (z) = . n→∞ 1−z 2n Exercise 10. (a) Let f (z) = ez − 1. One can easily check that the order of f (z) is 1. The zeros of f (z) are exactly at the complex numbers of the form zk = 2πik, k ∈ Z. Since f 0 (2πik) = e2πik = 1, we see that all of these zeros are simple. Therefore, Hadamard’s theorem tells us that f (z) must be of the form ∞ ∞ Y Y z z z2 z αz+β αz+β e 2πik = e z e − 1 = f (z) = e z 1− 1+ 2 2 2πik 4π k k=−∞,k6=0 k=1 for some α, β ∈ C. It remains to determine α and β. We have ∞ Y ez − 1 z2 αz+β =e 1+ 2 2 , z 4π k k=1 4 SOLUTIONS TO HW6 which implies ∞ Y z2 ez − 1 αz+β 1 + 2 2 = eβ . 1 = lim = lim e z→0 z→0 z 4π k k=1 Therefore, we may take β = 0 (any other choice of β differs by a multiple of 2πi, but this won’t affect the product). Expanding the first few terms of the product shows that we must also have α = 21 and hence ∞ Y z2 z z/2 e −1=e z 1+ 2 2 . 4π k k=1 (b) Now let f (z) = cos(πz). As in Example 1 in Section 3, f (z) has order equal to 1. 2k+1 The zeros of f (z) are exactly at complex numbers of the form zk = 2 , with k ∈ Z. Since f 0 2k+1 = −π sin (2k+1)π = ±π, all of the zeros are simple. Hadamard’s theorem again 2 2 gives ∞ ∞ 2 Y Y 2z 2z 4z αz+β αz+β cos(πz) = f (z) = e e 2k+1 = e 1− 1− 2k + 1 (2k + 1)2 k=−∞ k=0 for some α, β ∈ C. We again compute α and β: ∞ Y αz+β 1 = lim cos(πz) = lim e 1− z→0 z→0 k=0 4z 2 (2k + 1)2 = eβ which implies β = 0. To find α, we simply note that cos(πz) is even; therefore ∞ ∞ Y Y 4z 2 4z 2 αz −αz e = cos(πz) = cos(−πz) = e , 1− 1− (2k + 1)2 (2k + 1)2 k=0 k=0 which implies α = 0. Therefore ∞ Y cos(πz) = 1− k=0 4z 2 (2k + 1)2 . Exercise 13. Let f (z) = ez − z. One can easily check that f (z) has order 1. Assume that ez − z = 0 has finitely many solutions, meaning f (z) has finitely many roots a1 , . . . aN . Each of these roots is simple, and none are are equal to 0. Hadamard’s theorem gives us N Y z z z z γz+β z αz+β ak e =e 1− ··· 1 − e − z = f (z) = e 1− ak a1 aN k=1 P −1 where α, β ∈ C, and where we denote γ = α + N for brevity. This implies that k=1 ak (1−γ)z −γz th e − ze is a polynomial, so that if ` > N , its ` coefficient is 0. If γ = 0, we immediately obtain a contradiction, so we may assume that γ 6= 0. This gives ` (1 − γ)` (−γ)` 1 ` ` − = 0 ⇐⇒ (γ − 1) − `γ = 0 ⇐⇒ 1 − =` `! (` − 1)! γ for all ` > N . Since exponential functions grow (or decay) much faster than polynomials, we obtain a contradiction. SOLUTIONS TO HW6 5 Exercise 15. Let f (z) be a nonconstant meromorphic function on C, and let a1 , . . . , aN , . . . be the (possibly infinite) set of poles of f (z), counted with multiplicity. We note that these poles must necessarily be isolated by definition of meromorphicity, and we have limn→∞ an = ∞ (after potentially renumbering the an ), for otherwise our function f (z) would be constant. By Theorem 4.1, there exists an entire function g(z) that vanishes exactly at the points {an }n≥1 and nowhere else. Therefore, the function h(z) = f (z)g(z) has removable singularities at all the points an , and therefore is entire. Hence f (z) = h(z) g(z) is a quotient of two entire functions. Now assume a = {an }n≥1 and b = {bn }n≥1 be two disjoint sequences having no finite limit points (this implies in particular that we may rearrange the sequences {an }n≥1 and {bn }n≥1 so that limn→∞ an = limn→∞ bn = ∞). Then, by Theorem 4.1, there exist entire functions fa (z) and fb (z) which vanish exactly at {an }n≥1 and {bn }n≥1 , respectively. Therefore, the function fa (z) fb (z) has zeros precisely at {an }n≥1 and poles precisely at {bn }n≥1 . Problem 1. Let f (z) be holomorphic on D, bounded, and not identically 0. Let z1 , z2 , . . . be its zeros. We may assume that f (0) 6= 0. Fix R < 1 such that R 6= |zn | for every n. Then f (z) must have finitely many zeros on DR (0) (else f (z) would be identically 0). After renumbering the zeros so that |z1 | ≤ |z2 | ≤ . . ., we may assume these zeros contained in DR (0) are z1 , z2 , . . . , zNR . Note that NR approaches ∞ as R approaches 1 from the left. We’ll give two proofs of the statement using different techniques (but of the same flavor). First solution. Let |f (z)| ≤ M on D, and let 0 < r ≤ R < 1. By Jensen’s fomula, we have NR Nr X X R R log log ≤ |z |zk | k| k=1 k=1 Z 2π 1 = log(|f (Reiθ )|) dθ − log(|f (0)|) 2π 0 ≤ log(M ) − log(|f (0)|) M = log . |f (0)| Letting R approach 1 from the left, and using the inequality 1 − x ≤ − log(x) for 0 < x < 1, we obtain Nr Nr X X M 1 1 − |zk | ≤ log ≤ log . |z |f (0)| k| k=1 k=1 Finally, letting r approach 1 from the left gives ∞ X M 1 − |zk | ≤ log < ∞. |f (0)| k=1 6 SOLUTIONS TO HW6 Second solution. Assume the statement fails; that is, assume n X lim 1 − |zk | = ∞. n→∞ k=1 Since 1 − x ≤ − log(x) for 0 < x < 1, we obtain ∞ = lim n→∞ n X k=1 1 − |zk | ≤ lim n→∞ n X − log(|zk |) = lim − log n→∞ k=1 n Y ! |zk | , k=1 which implies lim n→∞ n Y |zk | = 0. k=1 Now let |f (z)| ≤ M on D, let 0 < r ≤ R < 1, and let k = 1, 2, . . . , Nr . We let z be of modulus R, and write z = Reiθ . We then have |R(zk − z)| = |Rzk − R2 eiθ | = |eiθ ||Re−iθ zk − R2 | = |R2 − zk z| = |R2 − zk z|. Now set Nr Y R2 − zk z R(zk − z) k=1 fR,r (z) = f (z) · ! ; the computation above implies |fR,r (z)| = |f (z)| for |z| = R. By the Maximum modulus principle, we have ! Nr Y RNr |f (0)| = |zk | |fR,r (0)| k=1 ≤ Nr Y ! |zk | max(|fR,r (z)|) |z|=R k=1 = Nr Y ! |zk | max(|f (z)|) |z|=R k=1 ≤ Nr Y ! |zk | M. k=1 Letting R approach 1 from the left gives |f (0)| ≤ Nr Y ! |zk | M. k=1 Letting r approach 1 from the left shows that |f (0)| ≤ lim− r→1 which contradicts the assumption f (0) 6= 0. Nr Y k=1 ! |zk | M = 0,