Topic 17

Taylor Polynomials & Newton's Method

A Taylor polynomial copies ff's value and first nn derivatives at x=ax = a: Tn(x)=∑k=0nf(k)(a)k!(x−a)kT_n(x) = \sum_{k=0}^{n} \dfrac{f^{(k)}(a)}{k!}(x - a)^k. Newton's method uses just the first-degree one, the tangent line, over and over to walk toward a root.

5 kinds of question. Open the one you are stuck on.

1

Find the Taylor polynomial of degree n about x = a

e.g. “Find the Taylor polynomial of degree nn for ff about x=ax = a (centered at aa)”
  1. 1
    Differentiate nn times: f,f′,…,f(n)f, f', \dots, f^{(n)}

    Look for a pattern; it saves work and catches slips.

  2. 2
    Evaluate each one at x=ax = a

    A small table: kk, f(k)(a)f^{(k)}(a).

  3. 3
    Divide the kkth value by k!k!

    0!=10! = 1, 1!=11! = 1, 2!=22! = 2, 3!=63! = 6, 4!=244! = 24.

  4. 4
    Multiply each by (x−a)k(x - a)^k and add

    Keep (x−a)(x - a) unexpanded unless asked.

Remember: Derivatives, plug in aa, divide by k!k!, powers of (x−a)(x - a).

See it in a problem3 worked examples

Find the third-degree Taylor polynomial of f(x)=log⁡xf(x) = \log x about x=1x = 1.

  1. 1
    Differentiate nn times: f,f′,…,f(n)f, f', \dots, f^{(n)}

    f=log⁡xf = \log x, f′=1xf' = \dfrac{1}{x}, f′′=−1x2f'' = -\dfrac{1}{x^2}, f′′′=2x3f''' = \dfrac{2}{x^3}.

  2. 2
    Evaluate each one at x=ax = a

    f(1)=0f(1) = 0, f′(1)=1f'(1) = 1, f′′(1)=−1f''(1) = -1, f′′′(1)=2f'''(1) = 2.

  3. 3
    Divide the kkth value by k!k!

    Coefficients: 01=0\dfrac{0}{1} = 0, 11=1\dfrac{1}{1} = 1, −12\dfrac{-1}{2}, 26=13\dfrac{2}{6} = \dfrac{1}{3}.

  4. 4
    Multiply each by (x−a)k(x - a)^k and add

    T3(x)=(x−1)−12(x−1)2+13(x−1)3T_3(x) = (x - 1) - \dfrac{1}{2}(x - 1)^2 + \dfrac{1}{3}(x - 1)^3.

Answer:

T3(x)=(x−1)−(x−1)22+(x−1)33T_3(x) = (x - 1) - \dfrac{(x - 1)^2}{2} + \dfrac{(x - 1)^3}{3}

Common mistakes3
  • Forgetting the k!k!. The x2x^2 coefficient is f′′(a)2\dfrac{f''(a)}{2}, not f′′(a)f''(a).
  • Writing powers of xx instead of (x−a)(x - a) when a≠0a \neq 0.
  • Plugging aa into ff before differentiating. Differentiate first, then plug in.
2

Approximate a number with a Taylor polynomial

e.g. “Use a quadratic approximation to estimate 26\sqrt{26}”
  1. 1
    Choose ff and a nice center aa near the target

    f(a)f(a) must be easy: a perfect square for x\sqrt{x} (a perfect cube for x3\sqrt[3]{x}), a=0a = 0 for exe^x, a=1a = 1 for log⁡x\log x, 00 or a standard angle for trig.

  2. 2
    Compute f(a),f′(a),…,f(n)(a)f(a), f'(a), \dots, f^{(n)}(a)
  3. 3
    Write Tn(x)T_n(x)

    Divide by k!k! and use powers of (x−a)(x - a).

  4. 4
    Plug in the target xx

    Only x−ax - a matters, and it is small: here is where (0.1)2=0.01(0.1)^2 = 0.01 makes life easy.

Remember: Easy center, derivatives, polynomial, plug in.

See it in a problem3 worked examples

Use a quadratic Taylor polynomial to estimate 26\sqrt{26}.

  1. 1
    Choose ff and a nice center aa near the target

    f(x)=xf(x) = \sqrt{x}, a=25a = 25.

  2. 2
    Compute f(a),f′(a),…,f(n)(a)f(a), f'(a), \dots, f^{(n)}(a)

    f(25)=5f(25) = 5; f′(x)=12xf'(x) = \dfrac{1}{2\sqrt{x}}, f′(25)=110f'(25) = \dfrac{1}{10}; f′′(x)=−14x3/2f''(x) = -\dfrac{1}{4x^{3/2}}, f′′(25)=−1500f''(25) = -\dfrac{1}{500}.

  3. 3
    Write Tn(x)T_n(x)

    T2(x)=5+110(x−25)−11000(x−25)2T_2(x) = 5 + \dfrac{1}{10}(x - 25) - \dfrac{1}{1000}(x - 25)^2.

  4. 4
    Plug in the target xx

    T2(26)=5+0.1−0.001=5.099T_2(26) = 5 + 0.1 - 0.001 = 5.099.

Answer:

26≈5.099\sqrt{26} \approx 5.099

Common mistakes3
  • Choosing the center at the target itself. f(26)=26f(26) = \sqrt{26} is the thing you don't know; center at 2525.
  • Trig in degrees. sin⁡(0.1)\sin(0.1) means 0.10.1 radians; change degrees to radians first.
  • Stopping at the linear term when the question says quadratic.
3

Bound the error of a Taylor approximation

e.g. “Give an upper bound for the error in your approximation”
  1. 1
    Identify nn, the center aa and the target xx
  2. 2
    Find the next derivative, f(n+1)f^{(n+1)}
  3. 3
    Find M≥∣f(n+1)(c)∣M \ge |f^{(n+1)}(c)| for every cc between aa and xx

    Use the worst point: if ∣f(n+1)∣|f^{(n+1)}| decreases, take the left end; if it increases, the right end. Any honest overestimate is fine.

  4. 4
    Plug in: ∣Rn(x)∣≤M(n+1)!∣x−a∣n+1|R_n(x)| \le \dfrac{M}{(n+1)!}|x - a|^{n+1}

Remember: One derivative past the polynomial, at its worst.

See it in a problem3 worked examples

Bound the error in the estimate 26≈5.099\sqrt{26} \approx 5.099 from the quadratic Taylor polynomial at a=25a = 25.

  1. 1
    Identify nn, the center aa and the target xx

    n=2n = 2, a=25a = 25, x=26x = 26.

  2. 2
    Find the next derivative, f(n+1)f^{(n+1)}

    f′′′(x)=38x−5/2f'''(x) = \dfrac{3}{8}x^{-5/2}.

  3. 3
    Find M≥∣f(n+1)(c)∣M \ge |f^{(n+1)}(c)| for every cc between aa and xx

    f′′′f''' is decreasing, so on [25,26][25, 26] it is largest at 2525: M=38⋅255/2=38⋅3125=325000M = \dfrac{3}{8 \cdot 25^{5/2}} = \dfrac{3}{8 \cdot 3125} = \dfrac{3}{25000}.

  4. 4
    Plug in: ∣Rn(x)∣≤M(n+1)!∣x−a∣n+1|R_n(x)| \le \dfrac{M}{(n+1)!}|x - a|^{n+1}

    ∣R2(26)∣≤325000⋅133!=150000=0.00002|R_2(26)| \le \dfrac{3}{25000} \cdot \dfrac{1^3}{3!} = \dfrac{1}{50000} = 0.00002.

Answer:

Error at most 0.000020.00002

Common mistakes3
  • Using f(n)f^{(n)} instead of f(n+1)f^{(n+1)}, or n!n! instead of (n+1)!(n+1)!.
  • Bounding ∣f(n+1)∣|f^{(n+1)}| at aa only. It must hold for every cc between aa and xx.
  • Using a calculator value of e0.2e^{0.2} to bound ece^c. Use something you know: ec<e<3e^c < e < 3.
4

Get a Taylor polynomial or a limit by substitution

e.g. “Find the Maclaurin polynomial of e−x2e^{-x^2}, xcos⁡(3x)x\cos(3x), 11−2x\dfrac{1}{\sqrt{1 - 2x}} …”
  1. 1
    Spot the standard function and what replaces xx

    e−x2e^{-x^2} is eue^u with u=−x2u = -x^2.

  2. 2
    Write its known Maclaurin polynomial in uu

    eue^u, sin⁡u\sin u, cos⁡u\cos u, log⁡(1+u)\log(1 + u), (1+u)k(1 + u)^k.

  3. 3
    Substitute and simplify, keeping terms up to the degree needed
  4. 4
    Use it: multiply as asked, or plug into the limit and cancel

    In a limit, the lowest power that survives decides the answer.

Remember: Known series, substitute, keep what you need.

See it in a problem4 worked examples

Find the Maclaurin polynomial of degree 66 for e−x2e^{-x^2}.

  1. 1
    Spot the standard function and what replaces xx

    eue^u with u=−x2u = -x^2.

  2. 2
    Write its known Maclaurin polynomial in uu

    eu≈1+u+u22+u36e^u \approx 1 + u + \dfrac{u^2}{2} + \dfrac{u^3}{6} (degree 33 in uu is degree 66 in xx).

  3. 3
    Substitute and simplify, keeping terms up to the degree needed

    1−x2+x42−x661 - x^2 + \dfrac{x^4}{2} - \dfrac{x^6}{6}.

  4. 4
    Use it: multiply as asked, or plug into the limit and cancel

    Nothing more to do: that is the polynomial.

Answer:

1−x2+x42−x661 - x^2 + \dfrac{x^4}{2} - \dfrac{x^6}{6}

Common mistakes3
  • Substituting into xx but not into the powers: (3x)3=27x3(3x)^3 = 27x^3, not 3x33x^3.
  • Stopping too early in a limit: if everything cancels, you needed one more term.
  • Degree confusion: e−x2e^{-x^2} to degree 66 needs uu only up to u3u^3.
5

Approximate a root with Newton's method

e.g. “Use Newton's method with x0=…x_0 = \dots to approximate …”
  1. 1
    Write the problem as f(x)=0f(x) = 0

    5\sqrt{5}: solve x2−5=0x^2 - 5 = 0. "g(x)=h(x)g(x) = h(x)": f=g−hf = g - h.

  2. 2
    Find f′(x)f'(x); write xn+1=xn−f(xn)f′(xn)x_{n+1} = x_n - \dfrac{f(x_n)}{f'(x_n)}
  3. 3
    Choose x0x_0: given, or a nearby easy guess
  4. 4
    Iterate as many times as asked, showing f(xn)f(x_n) and f′(xn)f'(x_n)

    Asked for a number of decimal places? Stop when two iterates agree to that many.

Remember: Follow the tangent line down to the axis, repeat.

See it in a problem3 worked examples

Use two iterations of Newton's method, starting at x0=2x_0 = 2, to approximate 5\sqrt{5}.

  1. 1
    Write the problem as f(x)=0f(x) = 0

    5\sqrt{5} is the positive root of f(x)=x2−5f(x) = x^2 - 5.

  2. 2
    Find f′(x)f'(x); write xn+1=xn−f(xn)f′(xn)x_{n+1} = x_n - \dfrac{f(x_n)}{f'(x_n)}

    f′(x)=2xf'(x) = 2x, so xn+1=xn−xn2−52xnx_{n+1} = x_n - \dfrac{x_n^2 - 5}{2x_n}.

  3. 3
    Choose x0x_0: given, or a nearby easy guess

    x0=2x_0 = 2 (given; 22=42^2 = 4 is close to 55).

  4. 4
    Iterate as many times as asked, showing f(xn)f(x_n) and f′(xn)f'(x_n)

    x1=2−−14=94x_1 = 2 - \dfrac{-1}{4} = \dfrac{9}{4}. x2=94−81/16−59/2=94−1/169/2=94−172=16172≈2.23611x_2 = \dfrac{9}{4} - \dfrac{81/16 - 5}{9/2} = \dfrac{9}{4} - \dfrac{1/16}{9/2} = \dfrac{9}{4} - \dfrac{1}{72} = \dfrac{161}{72} \approx 2.23611.

Answer:

5≈16172≈2.23611\sqrt{5} \approx \dfrac{161}{72} \approx 2.23611 (true value 2.236072.23607)

Common mistakes4
  • Sign slips: it is xnx_n minus f(xn)f′(xn)\dfrac{f(x_n)}{f'(x_n)}.
  • Writing the formula with f′(xn)f'(x_n) in the numerator. Value over slope.
  • Starting where f′(x0)=0f'(x_0) = 0: the tangent is flat and never meets the axis.
  • Rounding too early. Keep fractions or 6 digits until the end.

Need the formulas? Must memorize · Flashcards

Found this helpful? Leave us a review on Google ↗Spotted a mistake?

Which topics are you reviewing?

The topic list, Flashcards, Must memorize and Exam words will only show these. Change it any time with the Topics button above the topic list.

Limits & continuity
Derivatives
Applications
Theorems & approximation
10 of 17 topics selected