Navigation

Approximation · Topic 20 of 23

Higher-Degree Approximations and Taylor Polynomials

The degree-nn Taylor polynomial of ff centred at aa is Tn(x)=k=0nf(k)(a)k!(xa)kT_n(x)=\sum_{k=0}^{n}\frac{f^{(k)}(a)}{k!}(x-a)^{k} — the one polynomial of degree at most nn whose derivatives at aa all match ff's. Taylor's Inequality turns it into a guarantee: Rn(x)M(n+1)!xan+1|R_n(x)|\le\frac{M}{(n+1)!}|x-a|^{n+1}, where MM bounds f(n+1)|f^{(n+1)}| on the whole interval.

5 min readFrequent on exams8 formulas6 quick checks
01

Key ideas

5 things to remember
  1. 1

    Coefficients are derivatives divided by factorials

    Tn(x)=k=0nf(k)(a)k!(xa)kT_n(x)=\sum_{k=0}^{n}\frac{f^{(k)}(a)}{k!}(x-a)^{k}, so ck=f(k)(a)/k!c_k=f^{(k)}(a)/k!. The powers are always (xa)k(x-a)^{k}, never xkx^{k}, unless the centre is a=0a=0.

  2. 2

    The unique polynomial that matches derivatives

    TnT_n is the only polynomial of degree at most nn with P(k)(a)=f(k)(a)P^{(k)}(a)=f^{(k)}(a) for k=0,1,,nk=0,1,\dots,n. So any legitimate shortcut that lands on such a polynomial has produced TnT_n.

  3. 3

    Read derivatives off the coefficients

    Run the formula backwards: f(k)(a)=k!ckf^{(k)}(a)=k!\,c_k. From T3(x)=25(x1)+4(x1)2(x1)3T_3(x)=2-5(x-1)+4(x-1)^{2}-(x-1)^{3} you get f(1)=2!4=8f''(1)=2!\cdot 4=8, not 44.

  4. 4

    Each degree buys one more power

    With h=xah=x-a, the error of TnT_n is about Chn+1Ch^{n+1}. Halving hh divides that error by 2n+12^{n+1}, not by 22 — the whole reason to go past linear.

  5. 5

    Lagrange remainder turns closeness into proof

    Rn(x)=f(n+1)(c)(n+1)!(xa)n+1R_n(x)=\frac{f^{(n+1)}(c)}{(n+1)!}(x-a)^{n+1} for some cc strictly between aa and xx. You never find cc; you bound f(n+1)|f^{(n+1)}| and get a guaranteed error.

−π−π/2π/2π-3-2-11xycentre a = 0cos xT₂T₄
T2(x)=1x22T_2(x)=1-\frac{x^{2}}{2} and T4(x)=1x22+x424T_4(x)=1-\frac{x^{2}}{2}+\frac{x^{4}}{24} against cosx\cos x. All three agree exactly at the centre; T2T_2 has visibly left the curve by x1.5x\approx1.5, while T4T_4 tracks it out to about x2x\approx2 before it too climbs away. More degree buys reach, but only near the centre.
-0.20-0.15-0.10-0.050.050.100.150.2051015xyM = 6 read at the centre — falseM = 14.65 worksworst case at t = −0.2|f⁽⁴⁾(t)|
Bounding f(4)(t)=6(1+t)4|f^{(4)}(t)|=\frac{6}{(1+t)^{4}} for f(x)=ln(1+x)f(x)=\ln(1+x) on [0.2,0.2][-0.2,0.2]. Reading MM at the centre gives 66, but the fourth derivative climbs to 14.6514.65 at the left endpoint — the bound built from M=6M=6 is not merely loose, it is false. MM must dominate across the whole interval.
02

Formulas

What to have memorised
  • Taylor polynomial at x=ax=a

    Tn(x)=k=0nf(k)(a)k!(xa)kT_n(x)=\sum_{k=0}^{n}\frac{f^{(k)}(a)}{k!}(x-a)^{k}

    Centre a=0a=0 gives the Maclaurin polynomial; T1T_1 is the linearisation.

  • Coefficients and derivatives

    ck=f(k)(a)k!f(k)(a)=k!ckc_k=\frac{f^{(k)}(a)}{k!}\quad\Longleftrightarrow\quad f^{(k)}(a)=k!\,c_k

    Read a derivative off a given TnT_n by multiplying its coefficient by k!k!.

  • Lagrange remainder

    Rn(x)=f(x)Tn(x)=f(n+1)(c)(n+1)!(xa)n+1R_n(x)=f(x)-T_n(x)=\frac{f^{(n+1)}(c)}{(n+1)!}(x-a)^{n+1}

    Some cc strictly between aa and xx; the case n=0n=0 is the MVT.

  • Taylor's Inequality

    Rn(x)M(n+1)!xan+1|R_n(x)|\le\frac{M}{(n+1)!}\,|x-a|^{n+1}

    Needs f(n+1)(t)M|f^{(n+1)}(t)|\le M at every tt in the interval, not just at aa.

  • Maclaurin: exe^{x}

    ex1+x+x22!+x33!++xnn!e^{x}\approx 1+x+\frac{x^{2}}{2!}+\frac{x^{3}}{3!}+\cdots+\frac{x^{n}}{n!}

    Every derivative of exe^{x} equals 11 at 00, so every coefficient is 1/k!1/k!.

  • Maclaurin: sinx\sin x and cosx\cos x

    sinxxx33!+x55!,cosx1x22!+x44!\sin x\approx x-\frac{x^{3}}{3!}+\frac{x^{5}}{5!}-\cdots,\qquad \cos x\approx 1-\frac{x^{2}}{2!}+\frac{x^{4}}{4!}-\cdots

    Odd and even powers only, so T3=T4T_3=T_4 for sin\sin and T4=T5T_4=T_5 for cos\cos.

  • Maclaurin: ln(1+x)\ln(1+x) and 11x\frac{1}{1-x}

    ln(1+x)xx22+x33+(1)n1xnn,11x1+x+x2++xn\ln(1+x)\approx x-\frac{x^{2}}{2}+\frac{x^{3}}{3}-\cdots+\frac{(-1)^{n-1}x^{n}}{n},\qquad \frac{1}{1-x}\approx 1+x+x^{2}+\cdots+x^{n}

    lnx\ln x has no Maclaurin polynomial — it is undefined at 00. Shift or recentre.

  • Binomial polynomial

    (1+x)k1+kx+k(k1)2!x2+k(k1)(k2)3!x3+(1+x)^{k}\approx 1+kx+\frac{k(k-1)}{2!}x^{2}+\frac{k(k-1)(k-2)}{3!}x^{3}+\cdots

    Roots come from here: k=12k=\frac12 gives 1+x1+x2x28\sqrt{1+x}\approx 1+\frac{x}{2}-\frac{x^{2}}{8}.

03

Build a Taylor polynomial and bound its error

The steps, in order
  1. 1

    Pick the centre aa where ff and its derivatives are easy — 00, 11, a perfect square or cube — and note the degree nn asked for.

  2. 2

    Differentiate n+1n+1 times. Evaluate f,f,,f(n)f,f',\dots,f^{(n)} at aa, and keep f(n+1)f^{(n+1)} as a function for the bound.

  3. 3

    Assemble Tn(x)=k=0nf(k)(a)k!(xa)kT_n(x)=\sum_{k=0}^{n}\frac{f^{(k)}(a)}{k!}(x-a)^{k}: divide each derivative by k!k! and keep every power as (xa)k(x-a)^{k}.

  4. 4

    Substitute the point being approximated and evaluate TnT_n, carrying more decimals than the error bound you expect.

  5. 5

    Find M=maxf(n+1)(t)M=\max|f^{(n+1)}(t)| over the whole closed interval between aa and xx — check the endpoints, not the centre.

  6. 6

    State Rn(x)Mxan+1(n+1)!|R_n(x)|\le\frac{M|x-a|^{n+1}}{(n+1)!}. If Tn=Tn+1T_n=T_{n+1} — an odd or even ff — redo it with n+1n+1 for a sharper bound.

04

Watch out

The mistakes that cost marks
  • ✗ Wrong

    T3(x)=f(a)+f(a)(xa)+f(a)(xa)2+f(a)(xa)3T_3(x)=f(a)+f'(a)(x-a)+f''(a)(x-a)^{2}+f'''(a)(x-a)^{3}

    ✓ Right

    Divide by k!k!: the coefficients are f(a)2\frac{f''(a)}{2} and f(a)6\frac{f'''(a)}{6}.

    Why: Differentiating ck(xa)kc_k(x-a)^{k} kk times gives k!ckk!\,c_k, so ckc_k must carry the 1/k!1/k!.

  • ✗ Wrong

    Given T3(x)=25(x1)+4(x1)2(x1)3T_3(x)=2-5(x-1)+4(x-1)^{2}-(x-1)^{3}, writing f(1)=4f''(1)=4.

    ✓ Right

    f(1)=2!4=8f''(1)=2!\cdot 4=8 and f(1)=3!(1)=6f'''(1)=3!\cdot(-1)=-6: multiply the coefficient by k!k!.

  • ✗ Wrong

    T2(x)=ex+exx+ex2x2T_2(x)=e^{x}+e^{x}x+\frac{e^{x}}{2}x^{2}, or centring at a=2a=2 but writing powers of xx

    ✓ Right

    Derivatives are evaluated at the centre: T2(x)=1+x+x22T_2(x)=1+x+\frac{x^{2}}{2} for a=0a=0, and every power is (xa)k(x-a)^{k}.

    Why: A Taylor polynomial is a polynomial — numbers for coefficients, no ff left inside.

  • ✗ Wrong

    Taking M=f(n+1)(a)M=|f^{(n+1)}(a)|, the value at the centre.

    ✓ Right

    Maximise f(n+1)|f^{(n+1)}| over the entire interval between aa and xx; check both endpoints.

    Why: For ln(1+x)\ln(1+x) on [0.2,0.2][-0.2,0.2], M=6M=6 gives a bound the true error at x=0.2x=-0.2 breaks.

  • ✗ Wrong

    "The error is the next term, f(n+1)(a)(n+1)!(xa)n+1\frac{f^{(n+1)}(a)}{(n+1)!}(x-a)^{n+1}."

    ✓ Right

    Lagrange gives f(n+1)(c)f^{(n+1)}(c) with cc unknown between aa and xx; only bounding that is a proof.

    Why: The next term is a size heuristic, not an equality.

  • ✗ Wrong

    Expecting T4T_4 of sinx\sin x to have an x4x^{4} term.

    ✓ Right

    sin\sin is odd, so T3=T4=xx36T_3=T_4=x-\frac{x^{3}}{6} — then use n=4n=4 for a much sharper remainder bound.

05

Quick check

Commit to an answer before you reveal one
  1. Q1easy

    Working from the definition rather than a table, find the degree-44 Maclaurin polynomial of f(x)=cosxf(x)=\cos x and use it to estimate cos(0.2)\cos(0.2).

    Hint

    List f,f,f,f,f(4)f,f',f'',f''',f^{(4)} and evaluate each at the centre a=0a=0 before assembling anything.

    Show answer

    Answer

    T4(x)=1x22+x424T_4(x)=1-\frac{x^{2}}{2}+\frac{x^{4}}{24}; cos(0.2)0.98006667\cos(0.2)\approx 0.98006667.

    Steps

    At a=0a=0 the derivatives cos,sin,cos,sin,cos\cos,-\sin,-\cos,\sin,\cos give 1,0,1,0,11,0,-1,0,1, so

    T4(x)=1+0xx22!+0x3+x44!=1x22+x424T_4(x)=1+0\cdot x-\frac{x^{2}}{2!}+0\cdot x^{3}+\frac{x^{4}}{4!}=1-\frac{x^{2}}{2}+\frac{x^{4}}{24}

    T4(0.2)=10.02+0.001624=0.98006667T_4(0.2)=1-0.02+\frac{0.0016}{24}=0.98006667

    Since cos\cos is even the x5x^{5} coefficient vanishes, so T4=T5T_4=T_5 and R5(0.2)1(0.2)66!=8.9×108|R_5(0.2)|\le\frac{1\cdot(0.2)^{6}}{6!}=8.9\times10^{-8} — good to seven decimals.

  2. Q2easy

    The degree-33 Taylor polynomial of ff centred at a=1a=1 is T3(x)=25(x1)+4(x1)2(x1)3T_3(x)=2-5(x-1)+4(x-1)^{2}-(x-1)^{3}. Find f(1)f(1), f(1)f'(1), f(1)f''(1) and f(1)f'''(1), and estimate f(1.2)f(1.2).

    Hint

    Match coefficient by coefficient against c0+c1(x1)+c2(x1)2+c3(x1)3c_0+c_1(x-1)+c_2(x-1)^{2}+c_3(x-1)^{3} and recall ck=f(k)(1)/k!c_k=f^{(k)}(1)/k!.

    Show answer

    Answer

    f(1)=2f(1)=2, f(1)=5f'(1)=-5, f(1)=8f''(1)=8, f(1)=6f'''(1)=-6; f(1.2)1.152f(1.2)\approx 1.152.

    Steps

    Matching gives c0=2c_0=2, c1=5c_1=-5, c2=4c_2=4, c3=1c_3=-1, and f(k)(1)=k!ckf^{(k)}(1)=k!\,c_k:

    f(1)=2,f(1)=5,f(1)=2!4=8,f(1)=3!(1)=6f(1)=2,\quad f'(1)=-5,\quad f''(1)=2!\cdot 4=8,\quad f'''(1)=3!\cdot(-1)=-6

    At x=1.2x=1.2 the increment is x1=0.2x-1=0.2:

    T3(1.2)=25(0.2)+4(0.2)2(0.2)3=21+0.160.008=1.152T_3(1.2)=2-5(0.2)+4(0.2)^{2}-(0.2)^{3}=2-1+0.16-0.008=1.152

    The trap is reporting f(1)=4f''(1)=4; the coefficient is f(1)/2!f''(1)/2!, not f(1)f''(1).

  3. Q3medium

    Find the degree-66 Maclaurin polynomial of f(x)=ex2f(x)=e^{-x^{2}} without differentiating six times, and hence find f(4)(0)f^{(4)}(0) and f(6)(0)f^{(6)}(0).

    Hint

    Substitute into the Maclaurin polynomial of eue^{u} — first check the substitution sends 00 to 00, then count how many terms you need.

    Show answer

    Answer

    T6(x)=1x2+x42x66T_6(x)=1-x^{2}+\frac{x^{4}}{2}-\frac{x^{6}}{6}; f(4)(0)=12f^{(4)}(0)=12 and f(6)(0)=120f^{(6)}(0)=-120.

    Steps

    Start from eu1+u+u22+u36e^{u}\approx 1+u+\frac{u^{2}}{2}+\frac{u^{3}}{6} and put u=x2u=-x^{2}, legal for a Maclaurin polynomial because u=0u=0 when x=0x=0. Each uju^{j} becomes degree 2j2j, so three terms past the constant reach degree 66:

    T6(x)=1x2+x42x66T_6(x)=1-x^{2}+\frac{x^{4}}{2}-\frac{x^{6}}{6}

    This has degree at most 66 and differs from ff by less than any multiple of x6x^{6}, so by uniqueness it is T6T_6. Then f(k)(0)=k!ckf^{(k)}(0)=k!\,c_k:

    f(4)(0)=4!12=12,f(6)(0)=6!(16)=120f^{(4)}(0)=4!\cdot\tfrac12=12,\qquad f^{(6)}(0)=6!\cdot\left(-\tfrac16\right)=-120

    Also c3=c5=0c_3=c_5=0, so f(0)=f(5)(0)=0f'''(0)=f^{(5)}(0)=0 — as an even function requires.

  4. Q4medium

    Find the degree-33 Maclaurin polynomial of f(x)=(1+x)1/2f(x)=(1+x)^{1/2}, use it to approximate 1.1\sqrt{1.1}, and bound the error with Taylor's Inequality.

    Hint

    Binomial pattern with exponent 12\frac12; for the bound you need f(4)f^{(4)} and where it is largest on [0,0.1][0,0.1].

    Show answer

    Answer

    T3(x)=1+x2x28+x316T_3(x)=1+\frac{x}{2}-\frac{x^{2}}{8}+\frac{x^{3}}{16}; 1.11.0488125\sqrt{1.1}\approx 1.0488125 with R33.91×106|R_3|\le 3.91\times10^{-6}.

    Steps

    f=12(1+x)1/2f'=\frac12(1+x)^{-1/2}, f=14(1+x)3/2f''=-\frac14(1+x)^{-3/2}, f=38(1+x)5/2f'''=\frac38(1+x)^{-5/2}, so at 00 the values are 1,12,14,381,\frac12,-\frac14,\frac38 and

    T3(x)=1+x2x28+x316,T3(0.1)=1.0488125T_3(x)=1+\frac{x}{2}-\frac{x^{2}}{8}+\frac{x^{3}}{16},\qquad T_3(0.1)=1.0488125

    Now f(4)(t)=1516(1+t)7/2f^{(4)}(t)=-\frac{15}{16}(1+t)^{-7/2}, whose size decreases on [0,0.1][0,0.1], so M=1516M=\frac{15}{16} at t=0t=0:

    R3(0.1)1516(0.1)44!=3.91×106|R_3(0.1)|\le\frac{\frac{15}{16}(0.1)^{4}}{4!}=3.91\times10^{-6}

    So 1.0488085<1.1<1.04881651.0488085<\sqrt{1.1}<1.0488165.

  5. Q5medium

    Use the degree-22 Taylor polynomial of f(x)=x1/3f(x)=x^{1/3} centred at a=8a=8 to approximate 8.53\sqrt[3]{8.5}, and bound the error with Taylor's Inequality.

    Hint

    Centre at 88 because it is a perfect cube: 82/3=48^{2/3}=4, 85/3=328^{5/3}=32, 88/3=2568^{8/3}=256.

    Show answer

    Answer

    T2(x)=2+x812(x8)2288T_2(x)=2+\frac{x-8}{12}-\frac{(x-8)^{2}}{288}; 8.532.0407986\sqrt[3]{8.5}\approx 2.0407986 with R23.01×105|R_2|\le 3.01\times10^{-5}.

    Steps

    f(8)=2f(8)=2; f=13x2/3f'=\frac13x^{-2/3} gives f(8)=112f'(8)=\frac{1}{12}; f=29x5/3f''=-\frac29x^{-5/3} gives f(8)=1144f''(8)=-\frac{1}{144}, so c2=1288c_2=-\frac{1}{288}:

    T2(x)=2+x812(x8)2288,T2(8.5)=2+12411152=2.04079861T_2(x)=2+\frac{x-8}{12}-\frac{(x-8)^{2}}{288},\qquad T_2(8.5)=2+\frac{1}{24}-\frac{1}{1152}=2.04079861

    f(t)=1027t8/3f'''(t)=\frac{10}{27}t^{-8/3} is positive and decreasing, so on [8,8.5][8,8.5] its maximum is at t=8t=8: M=1027256=1.447×103M=\frac{10}{27\cdot 256}=1.447\times10^{-3}. With n=2n=2 and xa=0.5|x-a|=0.5,

    R2(8.5)M(0.5)33!=3.01×105|R_2(8.5)|\le\frac{M(0.5)^{3}}{3!}=3.01\times10^{-5}

  6. Q6hard

    How large must nn be so that the Maclaurin polynomial TnT_n of f(x)=exf(x)=e^{x} approximates exe^{x} to within 10510^{-5} for every xx in [1,1][-1,1]?

    Hint

    Write Taylor's Inequality with the worst-case f(n+1)|f^{(n+1)}| and the worst-case x|x| already inserted, then hunt through the factorials.

    Show answer

    Answer

    n=8n=8, with R8(x)8.27×106|R_8(x)|\le 8.27\times10^{-6} on [1,1][-1,1].

    Steps

    Every derivative of exe^{x} is exe^{x}, and on [1,1][-1,1] it is increasing, so f(n+1)(t)e<3=M|f^{(n+1)}(t)|\le e<3=M; also x01|x-0|\le 1. Hence

    Rn(x)31n+1(n+1)!=3(n+1)!|R_n(x)|\le\frac{3\cdot 1^{n+1}}{(n+1)!}=\frac{3}{(n+1)!}

    Force this under 10510^{-5}: (n+1)!>3×105(n+1)!>3\times10^{5}. Since 8!=403208!=40\,320 and 9!=3628809!=362\,880, the smallest choice is n+1=9n+1=9, so n=8n=8 and the bound is 3/9!=8.27×1063/9!=8.27\times10^{-6}.

    Check n=7n=7 genuinely fails: at x=1x=1, T7(1)=2.7182540T_7(1)=2.7182540 against e=2.7182818e=2.7182818, an error of 2.8×1052.8\times10^{-5}.

06

On the exam

How this topic is marked
  • Most marks are in the setup: show the derivatives evaluated at the centre, the division by k!k!, and powers of (xa)(x-a). A bare decimal answer rarely gets full credit.

  • An error-bound part is marked on MM: name the interval you maximised f(n+1)|f^{(n+1)}| over and say why (increasing, decreasing, cos1|\cos|\le 1), then substitute.

  • Look for the free upgrade — if the next coefficient is zero, as for sin\sin, cos\cos or any odd or even ff, use n+1n+1 in Taylor's Inequality for a far sharper bound.

Keep going