6 Series We call a normed space (X, k·k) a Banach space provided that every Cauchy sequence (xn ) in X converges. For example, R with the norm k·k = |·| is an example of Banach space. Now let (xn ) be a sequence in X. Define a new sequence (sn ) in X by n X xk , n ∈ N. sn = k=1 P P The sequence (sn ) is called a series in X and is written as xk or k xk . The n-th term sn of the sequence (sn) isPcalled the nth partial sum and xk is called the kth summand of the series k xk . P Definition 6.1. The series k xk converges or is convergent if the sequence (sn ) of partial limit limn→∞ sn is P sums converges. TheP Pcalled the value of ∞ the series k xk and is written as k=1 xk Finally, k xk diverges or is divergent if the sequence (sn ) diverges in X. Proposition 6.2 (Necessary condition for convergence). If the series P xk converges, then xn → 0. P Proof. Denoting by sn the nth partial sum of the series xk , we have sn → x for some x. Since xn = sn+1 − sn → x − x = 0, the result follows. Example 6.3. The series e. P P (1/k!) converges in R. Its value is ∞ k=0 1/k! = P P Example 6.4. The series 1/k2 converges in R. Indeed, if sn = nk=1 1/k2 , then n n n X X X 1 1 1 1 2 =1+ − = 1 + 1 − < 2. 1/k ≤ 1 + sn = k(k − 1) k−1 k n k=1 k=2 k=2 So, the sequence (sn ) of partial sums is bounded in R and since it is also increasing, it converges. P P Example 6.5. The harmonic series 1/k diverges. Indeed, if sn = nk=1 1/k, then 1 n 1 1 + ... + ≥ = |s2n − sn | = n+1 2n 2n 2 P showing that (sn ) is not a Cauchy sequence and this implies that k 1/k is dirvergent. 30 P k Example 6.6. The geometric series a , where a ∈ R satisfies |a| < 1, converges. Indeed, n X ak+1 − 1 ak = sn = a−1 k=0 and since |a| < 1, we have ak+1 → 0 showing that lim sn = n→∞ P ∞ X ak = k=0 1 . 1−a ak diverges. P P Proposition 6.7. Let ak and bk be convergent series in a Banach space X and let α ∈ R. Then P P∞ P∞ • The (ak + bk ) converges and k=1 (ak + bk ) = k=1 ak + P∞ series b . k=1 k P P∞ P∞ • The series ∞ k=1 (αak ) converges and k=1 (αak ) = α k=1 ak . P Theorem 6.8 (Cauchy criterion). For a series xk in a Banach space (X, k·k) the following conditions are equaivalent. P (a) The series xk converges. If |a| ≥ 1, then k (b) For every ε > 0 there is N ∈ N such that m X xk < ε k=n+1 for all m > n ≥ N . P P Proof. Let sn = nk=1 xk be the nth partial sum. Then sm −sn = m k=n+1 xk if m > n.Thus (sn ) is Cauchy in X if and only if the condition (b) holds true. By assumption, (X, k·k) is a Banach space and the result follows. P P Theorem 6.9. If ak is a series in R and ak ≥ 0 for all k ∈ N, then ak converges if and only if the sequence (sn ) of partial sums is bounded. In this case, the series has the value equal to supn∈N sn . Proof. Since ak ≥ 0 for all k ∈ N, the sequence (sn ) of partial sums is increasing. Hence (sn ) converges if and only if (sn ) is bounded. 31 6.1 Alternating series P Definition 6.10. A series ak in R is called alternatingP if ak and ak+1 have opposite signs. Any such series can be written as ± (−1)k ak with ak ≥ 0 for all k ∈ N. Theorem 6.11 (Leibniz criterion). Let (ak ) be a decreasing null Psequence such that ak ≥ 0 for all k ∈ N. Then the alternating series (−1)k ak converges in R. Proof. Since s2n+2 − s2n = −a2n+1 + a2n+2 ≤ 0, the sequence (s2n ) is decreasing. Similarly, s2n+3 − s2n+1 = a2n+2 − a2n+3 ≥ 0, implying that (s2n+1 ) is increasing. Moreover, s2n+1 ≤ s2n so that s2n+1 ≤ s2 and s2n ≥ 0 for all n ≥ 1. Hence, there are s and t such that s2n → s and s2n+1 → t. Also t − s = lim (s2n+1 − s2n ) = lim a2n+1 = 0 n→∞ n→∞ implying that t = s. Take ε > 0. Then there are N1 , N2 ∈ N such that |s2n − s| < ε, 2n ≥ N1 and |s2n+1 − s| < ε, 2n + 1 ≥ N2 . Thus, |sn − s| < ε for all n > 2 max{N1 , N2 }. Example 6.12. The assumption that (an ) is decreasing is necessary. For example, set 1 1 and a2n−1 = . a2n = 2n (2n − 1)2 1 1 1 Then a2n−1 = (2n−1) 2 ≤ 2n−1 so that 0 ≤ an ≤ n showing that an → 0. Since a2n−1 P< a2nn for n ≥ 2, the sequence (an ) is not decreasing. However, the series (−1) an diverges. Arguing by contradiction we assume that the sequence (sn ) of partial sums converges. We have s2n = 2n X (−1)k ak = k=1 n X =− k=1 n n X X (−1)a2k−1 + a2k k=1 1 + (2k − 1)2 = −An + Bn , 32 k=1 n X k=1 1 2k P Pn 1 1 1 1 Pn where An = − nk=1 (2k−1) 2 and Bn = k=1 k . Since the k=1 2k = 2 P 1 series k≥1 (2k−1) 2 converges, the sequence (An ) converges. Consequently, the sequence B = sn + An converges. But this contradicts the fact that the n P 1 series k≥1 k diverges. 6.2 Absolute convergence P Definition 6.13. The series xk in a BanachPspace (X, k·k) converges absolutely or Pis absolutely convergent if the seriesP kxk k is convergent P in R. The series xk is conditionally convergent if xk converges but kxk k diverges. P Example 6.14. The alternating harmonic series (−1)k /k converges in P view of Leibniz criterion,Phowever, the series of absolute values 1/k diverges. Hence the series (−1)k /k converges conditionally. Proposition 6.15. Every absolutely convergent series converges. Proof. The proposition is a consequence of Cauchy criterion and the triangle inequality. Indeed, m m X X xk ≤ kxk k . k=n+1 k=n+1 6.3 Comparison, Root, and Ratio Tests From now on we only consider real sequences, however, you are invited to figure out which of the results can be stated in the full generality of Banach spaces. P P Proposition 6.16 (Comparison test). Let ak and bn be two sequences of nonnegative terms and assume that an ≤ bn for all n. Then the following holds true, P P (a) If bk converges, then so is ak . P P (b) If ak diverges, then so is bk . 33 P m Example 6.17. Let m ≥ 2. Then the series P 1 1/k converges absolutely. Recall from Example 6.4 that the series converges. Since for m ≥ 2, k2 we have 1 1 ≤ 2 for all k, m k k P it follows from Proposition 6.16 that the series 1/km is indeed convergent. P P Proposition 6.18 (Limit comparison test). Let ak and bn be two sequences of positive terms and assume that an →L bn (Clearly, L ∈ [0, ∞) or L = ∞). Then the following holds true, P (a) P If L ∈ (0, ∞), then the convergence of bk implies the convergence of P P ak and the divergence of ak implies the divergence of bk . P P (b) Assume that L = 0. Then ak converges if bk converges. P P (c) Assume that L = ∞. Then ak diverges, if bk diverges. Proof. We only prove (a). If L ∈ (0, ∞), then there is N so that L/2 = L−L/2 < an /bn < L+L/2 = 3L/2 for n ≥ N . So, (L/2)bn < an < (3L/2)bn for n ≥ N and the result follows from Proposition 6.16. P Theorem 6.19 (Root test). Let ak be a series of nonnegative real numbers and let p α = lim sup n |an |. Then • • P P ak converges absolutely if α < 1. ak diverges if α > 1. P If α = 1, then both convergence and divergence of ak are possible. P Theorem 6.20 (Ratio test). Let ak be a series of real numbers such that ak 6= 0 for all k ≥ N . Then • If then P ak+1 < 1, lim k→∞ ak ak converges absolutely. 34 • If then P ak+1 > 1, lim k→∞ ak ak diverges. P a If limk→∞ k+1 ak are ak = 1, then both convergence and divergence of possible. Proposition 6.21 (Cauchy’s condensationPtest). Let (ak ) be a sequence of nonnegative ak converges if and only if P k and decreasing terms. Then the series 2 a2k converges. P P Proof. Denote by sn = nk=1 and tn = nk=1 2k a2k . Then, if n < 2m , we have sn ≤ a1 + (a2 + a3 ) + . . . + (a2m + . . . + a2m+1 −1 ) ≤ a1 + 2a2 + . . . + 2m a2m = tm . P If 2k a2k converges, then the sequence P (tm ) is bounded and above estimate shows that (sn ) is bounded. Hence ak converges. For the converse, note that if n > 2m , then sn ≥ a1 + a2 + (a3 + a4 ) + . . . + (a2m−1 +1 + . . . + a2m ) 1 1 ≥ a1 + a2 + 2a4 + . . . + 2m−1 a2m = tm . 2 2 P This P shows that if ak converges, then the sequence (tm ) is bounded so k that 2 a2k converges. P 1 Example 6.22. Consider series the np with p > 0.PThen n P condenP the 2n 1 1 converges if and only if = sation test implies that p np p−1 n 2 P 2 1 n converges converges. We already know that the geometric series 2p−1 1 < 1, i.e., p > 1. if and only if 2p−1 The next theorem is a generalization of the condensation test. Theorem 6.23 (Schlömlich theorem). Let (gk ) be a strictly increasing sequence of positive integers such that gk+1 − gk ≤ C(gk − gk−1 ) for some C > 0 and all k ∈ N. If (ak ) is a decreasing sequence of nonnegative numbers, then X X ak converges if and only if (gk+1 − gk )agk converges. 35 Proof. The proof is similar to the proof of the condensation test. Let (Sn ) P be the sequence of the partial sums of the series a and (T ) the sequence n k P of the partial sums of the series (gk+1 − gk )agk . Now, if n < gk , then Sn ≤ Sg k ≤ (a1 + . . . + ag1 −1 ) + (ag1 + . . . + ag2 −1 ) + . . . + (agk + . . . + agk+1 −1 ) ≤ (a1 + . . . + ag1 −1 ) + (g2 − g1 )ag1 + . . . + (gk+1 − gk )agk ≤ (a1 + . . . + ag1 −1 ) + Tk . Consequently, if the sequence (Tk ) converges and hence is bounded, then P the sequence (S ) is bounded so that a converges. Conversely, assume n k P that ak converges. If n > gk , then CSn ≥ CSgk ≥ C(ag1 +1 + . . . + ag2 ) + . . . + C(agk−1+1 + . . . + agk ≥ C(g2 − g1 )ag2 + . . . + C(gk − gk−1 )agk ≥ (g3 − g2 )ag2 + . . . + (gk+1 − gk )agk = Tk − (g2 − g1 )ag1 . P Consequently, the sequence (Tk ) is bounded and the series (gk+1 − gk )agk converges. Example 6.24. Take gk = 3k . Then the sequence (gk ) satisfies the assumptions of the above theorem and gk+1 − gk = 3k+1 − 3k = 2 · 3k . So, the following holds true. Assume that (ak ) is a decreasing sequence of nonnegative numbers, then X X ak converges if and only if 3k a3k converges. P 1 Now consider the series . Since the sequence ( 3ln1 n ) is decreasing P 3n P 1 3ln n P 3n and converges to 0, = ln n converges if and only if ln 3n = 3 3n ln 3 3 n P 1 ln 3−1 converges. Since ln 3 > 1, it follows that 3 > 1 and 3ln 3−1 n P P 1 1 1 < 1 so that the series conconverges. Hence 3ln 3−1 3ln 3−1 3ln n verges. Consider gk = k2 , then the sequence (gk ) satisfies the assumptions of the theorem. Then the following is true. If (ak ) is a decreasing sequence of nonnegative numbers, then X X ak converges if and only if kak2 converges. P Indeed, since (k + 1)2 −P k2 ) = 2k + 1, it follows from Theorem ?? that ak converges if and only if (2k + 1)ak2 converges. Since kak2 ≤ (2k + 1)ak2 ≤ 36 P if and only if 3kak2 , the comparison test shows that (2k + 1)ak2 converges P P 1 √ kak2 converges. showing our claim. Now consider the series . Since 2 nP √ 1 n √ the sequence 1/2 ) is decreasing P and converges to 0, the series 2 n P n √n = converges if and only if the series 2n converges. Using the 2 n2 P n P 1 √ ratio test, we see that converges. 2n converges and so 2 n 6.4 The Dirichlet and Abel Test We start with the following lemma. Lemma 6.25 (Abels’ lemma). Let (xn ) and (yn ) be two sequences of real numbers. Let (sn ) be a sequence of partial sums of (yn ) and s0 = 0. Then m X k=n+1 xk yk = (xm sm − xn+1 sn ) + m−1 X (xk − xk+1 )sk . k=n+1 Proof. Note that yk = sk − sk−1 . So, m X k=n+1 xk yk = m X k=n+1 xk (sk − sk−1 ) = xn+1 (sn+1 − sn ) + xn+2 (sn+2 − sn+1 ) + . . . + xm (sm − sm−1 ) = xm (xm − xn+1 sn ) + sn+1 (xn+1 − xn+2 ) + . . . + sm−1 (xm−1 − xm ) = xm (xm − xn+1 sn ) + m−1 X k=n+1 xk (sk − sk−1 ) We use this lemma to obtain a test for convergence of the series P xk yk . Theorem 6.26 (Dirichlet’s test). Suppose that (xn ) is a decreasing sequence such that xn → 0 and Pthe sequence (sn ) of the partial sums of (yn ) is bounded. Then the series xk yk converges. Proof. Since (sn ) is bounded, |sn | ≤ C for all n ≥ 1. Since (xk ) is decreasing, xk − xk+1 ≥ 0. Thus, by the above lemma, m m−1 X X ≤ |x s | + |x s | + x y (xk − xk+1 ) |sk | m m n+1 n k k k=n+1 k=n+1 ≤ ((xn+1 + xm ) + (xn+1 − xm ))C = 2xn+1 C. 37 P Example 6.27. Consider (−1)n xn where xn is a decreasing sequence of nonnegative numbers converging to 0. Then the series converges in view of alternating series test. However,Pthis also follows from Dirichlet’s test. Pn n n Indeed, let yn = (−1) . Then sn = k=1 bk = k=1 (−1)k is equal to −1 if n is odd and is equal to 0 if n is even. Hence (sn ) is bounded, P the sequence n and Dirichlet’s test implies that the series (−1) an converges. Example 6.28. One can show, using for example the principle of mathematical induction, that sin ϑ (n + 1)ϑ nϑ · (cos ϑ + cos 2ϑ + . . . + cos nϑ) = cos · sin . 2 2 2 Hence if ϑ 6= 2πk for all k ∈ Z, then (n+1)ϑ nϑ · sin cos 2 2 1 . ≤ |cos ϑ + cos 2ϑ + . . . + cos nϑ| = ϑ sin sin ϑ 2 2 Similarly, sin ϑ (n + 1)ϑ nϑ · (sin ϑ + sin 2ϑ + . . . + sin nϑ) = sin · sin 2 2 2 which implies that 1 |sin ϑ + sin 2ϑ + . . . + sin nϑ| ≤ sin ϑ 2 if ϑ 6= 2πk for all k ∈ Z. Hence, in view of Dirichlet’s test, one has the following result. Corollary 6.29. If (xn ) is a decreasing sequence converging to 0 and ϑ 6= P P 2πk for all k ∈ Z, then the series xn cos nϑ and xn sin nϑ converge. Theorem 6.30P(Abel’s test). If (xP n ) is a monotone convergent sequence and the series yk converges, then xk yk converges. Proof. Without loss of generality, we may assume that (xn ) is decreasing. If x = lim xn , set un = xn − x is decreasing and converging to 0. P Thus, xn = x + un and x y = xy + u y . Now note that the series xyk n n n P n n P converges since yk converges and the un yn converges in view of P series the Dirichlet’s test. Hence, the series xn yn converges. 38 6.5 Cauchy Product Definition 6.31. The Cauchy product of two series P is the series n≥0 cn where cn = n X ak bn−k . k=0 P n≥0 an and P n≥0 bn n ≥ 0. P Example 6.32. Let n≥0 cn be the Cauchy product of the alternating P (−1)n √ series with itself. That is, n+1 n n X X 1 (−1)n−k (−1)k n √ p = (−1) . ·p cn = k+1 (n − k) + 1 (k + 1)(n − k) k=0 k=0 P (−1)n √ Note that , in view of the alternating series test, the series converges n+1 conditionally. However, the Cauchy product diverges. To see this we show that the sequence (cn ) does not converge to 0. Indeed, we have (n − k + 1)(k + 1) = 2 n 2 +1 − −k 2 2 n and hence q p p n (n − k + 1)(k + 1) = (n/2 + 1)2 − (n/2 − k)2 ≤ n/2 + 1)2 = + 1. 2 Using this we find that |cn | = n X k=0 p 1 (k + 1)(n − k) ≥ n X k=0 2 1 = (n + 1) ≥1 n/2 + 1 n+2 showing that the sequence (cn ) does not converge to 0. P Theorem 6.33 (Mertens theorem). Assume that n≥0 converges absoP∞ P∞ P lutely, Cauchy product of n = A, n = B, and n=0 aP n=0 bP n≥0 is the P P∞ a and b . Then c converges and n n n n≥0 n≥0 n≥0 n=0 cn = AB. Proof. Define An = n X k=0 an , Bn = n X bn , Cn = k=0 n X k=0 39 cn , βn = Bn − B. Then Cn = c0 + c1 + . . . + cn = a0 b0 + (a0 b1 + a1 b0 ) + . . . + (a0 bn + a1 bn−1 + . . . + an b0 ) = a0 (b0 + b1 + . . . + bn ) + a1 (b0 + b1 + . . . + bn−1 ) + . . . + an b0 = a0 Bn + a1 Bn−1 + . . . + an B0 = a0 (βn + B) + a1 (βn−1 + B) + . . . + an (β0 + B) = (a0 + . . . + an )B + (a0 βn + a1 βn−1 + . . . + an βn ) = An B + γn where we have abbreviated γn = a0 βn + a1 βn−1 + . . . + an βn . Since An B → AB, it suffices to show that γn → 0. In order to show that γn → 0, pick ε > 0. Then, since βn → 0, there is N such that Let (The series Then P b= A n≥0 |an | |βn | < ε for all n ≥ N . ∞ X and n=0 |an | |βn | ≤ M. converges and (βn ) is bounded since it converges.) |γn | = |a0 βn + a1 βn−1 + . . . + an β0 | ≤ |a0 βn + a1 βn−1 + . . . + an−N βN | + |an−N +1 βN −1 + an−N +2 βN −2 + . . . + an β0 | b + M (|an−N +1 | + |an−N +2 | + . . . + |an |) ≤ε·A Keeping N fixed and letting n → ∞, we get b lim sup |γn | ≤ εA since an → 0. Since ε > 0 was arbitrary, we conclude that lim sup |γn | = 0 which implies that lim γn = 0 as claimed P P∞ Theorem 6.34 (Abel’s theorem). If ∞ n=0 an = A, n=0 bn = B, and P P n ∞ a b for all n ≥ 0, then C = AB. c = C, where c = n n k n−k k=0 n=0 40 6.6 Double Sums Let (X, k·k) be a normed P space, let (aij )(i,j)∈N×N be a double sequence in X, and the double series aij . We have a function a : N×N → N, (j, k) 7→ ajk . Since N × N is countable, there is a bijection σ : N → N × N. If σ is P Psuch bijection, the series n aσ(n) is called an ordering of the double series ajk . P If we fix j ∈ N (or k ∈ N ), then the series ajk is called the jth row series (or the kth column series). If every row seriesP(orP column series) converges, ∞ then we may consider the series of row sums ( j k=0 ajk P P∞ P) (or the series of column sums k ( j=0 ajk )). Finally, the double series ajk is summable if n X sup |ajk | < ∞. n Theorem 6.35. Let j,k P ajk be a summable double series. Then P P (a) Every ordering ajk is converges absolutely to a value n aσ(n) of s ∈ X which is independent of σ. P P (b) P The P series of row sums j ( ∞ k=0 ajk ) and the series of column sums ∞ ( a ) converge absolutely and k j=0 jk ∞ ∞ X ∞ X XX ( ajk ) = ajk ). ( j=0 k=0 k j=0 P To prove (a) we will need the following result. Let an be a series P and let σ : N → N be a bijection. Define b = a . Then the series bn is n σ(n) P called a rearrangement of of an . P Proposition 6.36. Assume that an converges absolutely. Then any reP arrangement of an converges absolutely. (They all converge to the same value). P Proof. Assume an converges absolutely. Take a bijection σ : N → N Pthat and consider bn P with bn =P aσ(n) . Denote by (An ) and P(Bn ) the sequences of partial sums of an and bn , respectively. Since an converges absolutely, given ε > 0, there is N such that m X k=n |ak | < ε 41 (1) for all m > n ≥ N . Let K be such that {1, . . . , N } ⊂ {σ(1), . . . , σ(K)}. Take n > K and consider the difference |An − Bn | Note that the terms a1 , . . . , aN appear in both partial sums An and Bn . So, by (1) |An − Bn | < ε. P Hence lim An = lim Bn , proving that bn converges to the same value P P as anP . The same argument in which P An and Bn are partial sums of |an | and |bn |, respectively, shows that bn converges absolutely. P P Proof of Theorem 6.35. (a) Abbreviate M = n nj,k=0 |ajk |. Let σ : N → N × N be a bijection and let N ∈ N. Then there is some K ∈ N such that {σ(0), . . . , σ(N )} ⊂ {(0, 0), (1, 0), . . . , (K, 0), . . . , (K, K)}. (2) P Together with the summability of ajk this imples N K X X aσ(n) ≤ |ajk | ≤ M. n=0 j,k=0 P This implies that n aσ(n) converges absolutely. Let δ : N → N × N be another bijection. Then α = σ −1 ◦ δ : N → N is a bijection. Set ym = xσ(m) . Then bα(n) = aσ(α(n)) = aδ(n) , n ∈ N. P P P a . Since of Thus n aδ(n) is a rearrangement σ(n) n aσ(n) converges n P absolutely, the series P n aδ(n) also converges absolutely. P∞ ∞ (b) The row series j=0 k=0 ajk , j ∈ N, and the column series P ajk , k ∈ N, converge absolutely. PThis follows from the summability of ajk . P∞ a ) and the series of column series ( So, the series of row series jk k=0 j P P∞ k ( j=0 ajk ). To see that these series converge absolutely, consider the following inequalities, m m m l X l X X X X |ajk | ≤ M, l ≤ m. |ajk | ≤ ajk ≤ j=0 k=0 j=0 k=0 j,k Pl P∞ Taking the limit m → ∞, we get l ∈PN. This j=0 | k=0 ajk | ≤ M ,P proves the absolute convergence of the series of row series j ( ∞ k=0 ajk ). A similar proves the absolute convergence of the series of column P argument P series k ( ∞ a ). j=0 jk 42 P∞ Now let σ : N → N × N be a bijection s = n aσ(n) . Then for P∞and let every ε > 0, there is N ∈ N such that n=N +1 aσ(n) < ∞. Also there is some K ∈ N such that (2) is satisfied. Then X m ∞ N X X l X aσ(n) < ε/2, l, m ≥ K. a ≤ a − jk σ(n) j=0 k=0 n=N +1 n=0 Taking the limit m → ∞ and l → ∞, we get X ∞ ∞ N ∞ X X X aσ(n) ≤ ε/2. ≤ a a − jk σ(n) n=N +1 j=0 k=0 n=0 Applying the triangle inequality to ∞ N X X aσ(n) < ε/2, aσ(n) ≤ s − n=0 n=N +1 we get ∞ ∞ X X ajk − s ≤ ε. j=0 k=0 P P A similar argument shows that the value of k ( ∞ j=0 ajk ) is also s. 43