Raabe's Test

Where the ratio test surrenders, this scene digs one level deeper. It is the tool for ratios that stick to 1.

Measure how fast the ratio approaches 1, and the verdict revives. The quantity doing the measuring is the first formula below. If it settles above 1, the series converges. If it settles below 1, the series diverges.

For example, a coarse πŸ“ ruler sometimes cannot separate two lengths. Switch to a ruler with finer marks, and the lengths finally split apart. Raabe's test is that finer ruler.

L=lim⁑nβ†’βˆžn(anan+1βˆ’1)L=\lim_{n\to\infty}n\left(\frac{a_n}{a_{n+1}}-1\right)
L>1β‡’βˆ‘anΒ converges,L<1β‡’Β divergesL>1\Rightarrow\sum a_n\ \text{converges},\quad L<1\Rightarrow\ \text{diverges}

Part of these learning paths

Examples & Exercises

State Raabe's test and prove it by comparison with the pp-series.

L=lim⁑kβ†’βˆžk(akak+1βˆ’1),L>1β‡’βˆ‘akΒ converges,L<1β‡’βˆ‘akΒ divergesL=\lim_{k\to\infty}k\left(\frac{a_k}{a_{k+1}}-1\right),\qquad L>1\Rightarrow\sum a_k\ \text{converges},\quad L<1\Rightarrow\sum a_k\ \text{diverges}
n=100n = 100: R100β‰ˆ1.4975R_{100} \approx 1.4975 β€” settling above the boundary 11, climbing toward 32\frac{3}{2}.
Show proof
  1. Raabe's test: if ak>0a_k > 0 and L=lim⁑kβ†’βˆžk(akak+1βˆ’1)L = \lim_{k\to\infty} k\left(\frac{a_k}{a_{k+1}} - 1\right) exists, then βˆ‘ak\sum a_k converges when L>1L > 1 and diverges when L<1L < 1. (Read every inequality below as holding for all large kk.)

  2. The comparison benchmark is the pp-series bk=1krb_k = \frac{1}{k^r}. Since bkbk+1=(k+1k)r=(1+1k)r\frac{b_k}{b_{k+1}} = \left(\frac{k+1}{k}\right)^r = \left(1+\frac{1}{k}\right)^r and (1+1k)r=1+rk+o(1k)\left(1+\frac{1}{k}\right)^r = 1+\frac{r}{k}+o\left(\frac{1}{k}\right), we get k(bkbk+1βˆ’1)β†’rk\left(\frac{b_k}{b_{k+1}}-1\right) \to r β€” so bkb_k is a series whose Raabe value is exactly rr.

  3. If L>1L > 1: pick rr with 1<r<L1 < r < L.

    Since k(akak+1βˆ’1)β†’Lk\left(\frac{a_k}{a_{k+1}}-1\right) \to L and k(bkbk+1βˆ’1)β†’rk\left(\frac{b_k}{b_{k+1}}-1\right) \to r, eventually akak+1βˆ’1>bkbk+1βˆ’1\frac{a_k}{a_{k+1}}-1 > \frac{b_k}{b_{k+1}}-1, i.e. ak+1ak<bk+1bk\frac{a_{k+1}}{a_k} < \frac{b_{k+1}}{b_k}.

    Then ak+1bk<akbk+1a_{k+1}b_k < a_k b_{k+1}, i.e. ak+1bk+1<akbk\frac{a_{k+1}}{b_{k+1}} < \frac{a_k}{b_k}, so akbk\frac{a_k}{b_k} decreases and ak≀Cbka_k \le Cb_k; comparing with the convergent βˆ‘bk\sum b_k (r>1r > 1), βˆ‘ak\sum a_k converges.

  4. If L<1L < 1: picking rr with L<r<1L < r < 1 reverses the inequalities to ak+1ak>bk+1bk\frac{a_{k+1}}{a_k} > \frac{b_{k+1}}{b_k}; then akbk\frac{a_k}{b_k} increases, so akβ‰₯cbka_k \ge cb_k (c>0c > 0), and comparing with the divergent βˆ‘bk\sum b_k (r<1r < 1) gives divergence of βˆ‘ak\sum a_k.

With uk=(2kk)4ku_k = \frac{\binom{2k}{k}}{4^k} and ak=uk2k+1a_k = \frac{u_k}{2k+1}, prove βˆ‘k=1∞ak\sum_{k=1}^{\infty} a_k converges via Raabe's test, and show why the ratio test stays silent.

akak+1=2(k+1)(2k+3)(2k+1)2,Rk=k(akak+1βˆ’1)=k(6k+5)4k2+4k+1β†’32>1\frac{a_k}{a_{k+1}}=\frac{2(k+1)(2k+3)}{(2k+1)^2},\qquad R_k=k\left(\frac{a_k}{a_{k+1}}-1\right)=\frac{k(6k+5)}{4k^2+4k+1}\to\frac{3}{2}>1
n=30n = 30: R30β‰ˆ1.4915R_{30} \approx 1.4915 (above the boundary 11), r30β‰ˆ0.9526r_{30} \approx 0.9526 β€” rkr_k clings to 1 while RkR_k settles above 1.
Show proof
  1. First the ratio: uk+1uk=(2k+2)(2k+1)(k+1)2β‹…14=2k+12k+2\frac{u_{k+1}}{u_k} = \frac{(2k+2)(2k+1)}{(k+1)^2}\cdot\frac{1}{4} = \frac{2k+1}{2k+2}, so akak+1=ukuk+1β‹…2k+32k+1=2k+22k+1β‹…2k+32k+1=2(k+1)(2k+3)(2k+1)2\frac{a_k}{a_{k+1}} = \frac{u_k}{u_{k+1}}\cdot\frac{2k+3}{2k+1} = \frac{2k+2}{2k+1}\cdot\frac{2k+3}{2k+1} = \frac{2(k+1)(2k+3)}{(2k+1)^2}.

  2. Expanding, the numerator is 2(k+1)(2k+3)=4k2+10k+62(k+1)(2k+3) = 4k^2+10k+6 and the denominator is (2k+1)2=4k2+4k+1(2k+1)^2 = 4k^2+4k+1, so akak+1βˆ’1=6k+54k2+4k+1\frac{a_k}{a_{k+1}}-1 = \frac{6k+5}{4k^2+4k+1}. Hence Rk=k6k+54k2+4k+1β†’64=32>1R_k = k\frac{6k+5}{4k^2+4k+1} \to \frac{6}{4} = \frac{3}{2} > 1, and βˆ‘ak\sum a_k converges by Raabe's test.

  3. Meanwhile ak+1ak=4k2+4k+14k2+10k+6β†’1\frac{a_{k+1}}{a_k} = \frac{4k^2+4k+1}{4k^2+10k+6} \to 1, so the ratio test is inconclusive β€” the limit is 11: starting from r1=920,r2=2542,r3=4972r_1 = \frac{9}{20}, r_2 = \frac{25}{42}, r_3 = \frac{49}{72} it crawls upward toward 11. Raabe's RkR_k looks one order deeper than the ratio.

For ak=1k(ln⁑k)pa_k = \frac{1}{k(\ln k)^p} (kβ‰₯2k \ge 2), show the Raabe limit is 11 for every pp, and prove the boundary L=1L = 1 is genuinely inconclusive β€” βˆ‘kβ‰₯21k(ln⁑k)2\sum_{k\ge 2}\frac{1}{k(\ln k)^2} converges while βˆ‘kβ‰₯21kln⁑k\sum_{k\ge 2}\frac{1}{k\ln k} diverges β€” via the integral test.

k(akak+1βˆ’1)=1+pln⁑k+o(1)β†’1(forΒ everyΒ p)k\left(\frac{a_k}{a_{k+1}}-1\right)=1+\frac{p}{\ln k}+o(1)\to 1\qquad(\text{for every }p)
n=200n = 200: R200β‰ˆ1.3786R_{200} \approx 1.3786 for p=2p = 2, R200β‰ˆ1.1892R_{200} \approx 1.1892 for p=1p = 1 β€” both crawling down to 11.
Show proof
  1. akak+1=k+1k(ln⁑(k+1)ln⁑k)p=(1+1k)(1+ln⁑(1+1/k)ln⁑k)p\frac{a_k}{a_{k+1}} = \frac{k+1}{k}\left(\frac{\ln(k+1)}{\ln k}\right)^p = \left(1+\frac{1}{k}\right)\left(1+\frac{\ln(1+1/k)}{\ln k}\right)^p. With ln⁑(1+1k)=1k+O(1k2)\ln\left(1+\frac{1}{k}\right) = \frac{1}{k}+O\left(\frac{1}{k^2}\right) and (1+x)p=1+px+o(x)(1+x)^p = 1+px+o(x), the parenthesis is 1+pkln⁑k+o(1k)1+\frac{p}{k\ln k}+o\left(\frac{1}{k}\right).

  2. Multiplying out, akak+1=1+1k+pkln⁑k+o(1k)\frac{a_k}{a_{k+1}} = 1+\frac{1}{k}+\frac{p}{k\ln k}+o\left(\frac{1}{k}\right), so k(akak+1βˆ’1)=1+pln⁑k+o(1)β†’1k\left(\frac{a_k}{a_{k+1}}-1\right) = 1+\frac{p}{\ln k}+o(1) \to 1 β€” the Raabe limit is 11 for every pp.

  3. Yet the two series have opposite fates.

    First verify the integral test's hypotheses: for xβ‰₯2x \ge 2, both xx and (ln⁑x)p(\ln x)^p (p=1,2p = 1, 2) are positive and increasing, so their product x(ln⁑x)px(\ln x)^p is increasing, hence the integrand 1x(ln⁑x)p\frac{1}{x(\ln x)^p} is positive and decreasing.

    Substituting t=ln⁑xt = \ln x, ∫2∞dxx(ln⁑x)p=∫ln⁑2∞dttp\int_2^{\infty}\frac{dx}{x(\ln x)^p} = \int_{\ln 2}^{\infty}\frac{dt}{t^p}; by the integral test βˆ‘kβ‰₯21k(ln⁑k)2\sum_{k\ge 2}\frac{1}{k(\ln k)^2} converges (∫ln⁑2∞tβˆ’2 dt=1ln⁑2\int_{\ln 2}^{\infty}t^{-2}\,dt = \frac{1}{\ln 2}) while βˆ‘kβ‰₯21kln⁑k\sum_{k\ge 2}\frac{1}{k\ln k} diverges (∫ln⁑2∞tβˆ’1 dt=∞\int_{\ln 2}^{\infty}t^{-1}\,dt = \infty).

    At L=1L = 1 Raabe's test is genuinely silent.

7. Series