Review : Taylor Series — Question 7

PDF ↗

Question 7

The direct Maclaurin expansion of ln⁡(1+x)\ln(1+x) converges slowly at x=1x=1. Seek a faster Taylor-series evaluation of ln⁡2\ln 2 using L(z)=ln⁡(1+z1−z),|z|<1.L(z)=\ln\!\left(\frac{1+z}{1-z}\right),\qquad |z|<1. Count terms by the number of nonzero summands, not by polynomial degree.

Tasks

  1. Derive the Maclaurin series for LL by differentiating and integrating a geometric series, with a justification on |z|<1|z|<1.

  2. Select a positive zz for which L(z)=ln⁡2L(z)=\ln 2. For the first NN terms, prove the tail bound 2z2N+1(2N+1)(1−z2)\frac{2z^{2N+1}}{(2N+1)(1-z^2)} and determine the error sign.

  3. Find the smallest NN for which this bound is strictly below 10−610^{-6} at the selected zz. Give the resulting approximation and a rigorous enclosure.

  4. For the direct alternating harmonic sum with NN terms, how many terms does the next-term bound require to certify error strictly below 10−610^{-6}? Compare the two certificates and explain the limits of this comparison.

Original worksheet page 1: question and worked solution for 6-2-007
Show solutionHide solution

Question 7 – Solution

Strategy. Change the evaluation point to shrink the geometric factor controlling the Taylor tail.

Step 1: Construct the odd-power series. Since L′(z)=2/(1−z2)=2∑k≥0z2kL'(z)=2/(1-z^2)=2\sum_{k\ge 0}z^{2k} and L(0)=0L(0)=0, L(z)=2∑k=0∞z2k+12k+1,|z|<1.\boxed{L(z)=2\sum_{k=0}^{\infty}\frac{z^{2k+1}}{2k+1},\quad |z|<1.} The geometric series converges uniformly on any closed segment inside (−1,1)(-1,1), so termwise integration is justified.

Step 2: Bound a positive tail. The equation (1+z)/(1−z)=2(1+z)/(1-z)=2 gives z=1/3z=1/3. With AN=2∑k=0N−1z2k+1/(2k+1)A_N=2\sum_{k=0}^{N-1}z^{2k+1}/(2k+1) and 0<z<10<z<1, 0<L(z)−AN=2∑k=N∞z2k+12k+1<2z2N+1(2N+1)(1−z2)=:BN.\begin{aligned} 0<L(z)-A_N&=2\sum_{k=N}^{\infty}\frac{z^{2k+1}}{2k+1}\\ &<\frac{2z^{2N+1}}{(2N+1)(1-z^2)}=:B_N. \end{aligned} Strictness follows because later denominators exceed 2N+12N+1.

Step 3: Count the accelerated terms. At z=1/3z=1/3, BN=9/[4(2N+1)32N+1]B_N=9/[4(2N+1)3^{2N+1}], strictly decreasing in NN. Here B5=9/7794468>10−6B_5=9/7794468>10^{-6} and B6=9/82904796<1.086×10−7B_6=9/82904796<1.086\times 10^{-7}. Thus six terms are the smallest number certified by this bound, and A6<ln⁡2<A6+982904796,A6≈0.6931470738.\boxed{A_6<\ln 2<A_6+\frac 9{82904796}},\qquad A_6\approx 0.6931470738.

Step 4: Make a fair certificate comparison. For the direct alternating sum with NN terms, the stated bound is 1/(N+1)1/(N+1). To make that bound strictly less than 10−610^{-6} requires N≥1,000,000N\ge 1{,}000{,}000. The comparison is six versus one million terms under these particular certificates. It does not assert that a million terms are necessary for the true direct-series error, nor that a loose bound is an optimal algorithm. The transformed series profits from the factor z2=1/9z^2=1/9.

Original worksheet page 2: question and worked solution for 6-2-007

Original worksheet layout. Use Enlarge or open the PDF for a closer view.