Taylor expansions and power-series endpoints
引入| English | 释义 | 拼音 |
|---|---|---|
| Taylor polynomial/ˈteɪlə ˌpɒlɪˈnəʊmɪəl/ | 泰勒多项式 | tài lēi duō xiàng shì |
| remainder/rɪˈmeɪndə/ | 余项 | yú xiàng |
| radius/ˈreɪdɪəs/ | 收敛半径 | shōu liǎn bàn jìng |
An approximation needs a decision rule
- A calculation uses $e^{0.5}\approx1+0.5+0.5^2/2=1.625$. Can it safely conclude that $e^{0.5}<1.63$? A nearby polynomial value is not a certified inequality.
- We will settle this question using Taylor polynomials 泰勒多项式 and remainders 余项, then distinguish the radius 收敛半径 of a power series from its endpoint behaviour and from equality with a given function.
Coefficients depend on the centre
- Write $c_k=f^{(k)}(a)/k!$. The degree-$m$ Taylor polynomial about $a$ is $T_m(x)=\sum_{k=0}^m c_k(x-a)^k$. A Maclaurin polynomial uses $a=0$. Each coefficient includes a factorial; it is not the derivative alone.
- For $e^{2x}$ at zero, the coefficient of $x^3$ is $2^3/3!=4/3$. At centre $a=1$, the same function is $e^2e^{2(x-1)}$, so the coefficient of $(x-1)^3$ is $4e^2/3$. Keep the centre and the power together.
What is the coefficient of $x^3$ in the Maclaurin series of $e^{2x}$? Give it to three decimal places.
难度 2/5 · 今日首次完成可获 140 金币及少量额外奖励
e^{2x} = Σ (2x)ⁿ/n!, so the x³ coefficient is 2³/3! = 8/6 = 4/3 ≈ 1.333.
The series to know by heart
| Function | Maclaurin series | Valid for |
|---|---|---|
| $\dfrac{1}{1-x}$ | $1+x+x^2+\cdots$ | $\lvert x\rvert<1$ |
| $e^x$ | $1+x+\dfrac{x^2}{2!}+\dfrac{x^3}{3!}+\cdots$ | all $x$ |
| $\sin x$ | $x-\dfrac{x^3}{3!}+\dfrac{x^5}{5!}-\cdots$ | all $x$ |
| $\cos x$ | $1-\dfrac{x^2}{2!}+\dfrac{x^4}{4!}-\cdots$ | all $x$ |
| $\ln(1+x)$ | $x-\dfrac{x^2}{2}+\dfrac{x^3}{3}-\cdots$ | $-1<x\le1$ |
| $\arctan x$ | $x-\dfrac{x^3}{3}+\dfrac{x^5}{5}-\cdots$ | $\lvert x\rvert\le1$ |
A high derivative from a series. Find $f^{(6)}(0)$ for $f(x)=e^{x^2}$.
- Substitute $x^2$ into the $e^x$ series: $e^{x^2}=1+x^2+\dfrac{x^4}{2}+\dfrac{x^6}{6}+\cdots$.
- The coefficient of $x^6$ is also $\dfrac{f^{(6)}(0)}{6!}$, so $f^{(6)}(0)=\dfrac{6!}{6}=120$. No repeated differentiation needed.
Let $f(x)=\sin(x^2)$. Find $f^{(6)}(0)$.
难度 2/5 · 今日首次完成可获 140 金币及少量额外奖励
sin(x²) = x² − x⁶/6 + …, so f⁽⁶⁾(0)/6! = −1/6 and f⁽⁶⁾(0) = −720/6 = −120.
What is $\displaystyle\sum_{n=0}^\infty\frac{(-1)^n}{2n+1}$?
难度 1/5 · 今日首次完成可获 100 金币及少量额外奖励
It is the arctan series at x = 1 (where it still converges): arctan 1 = π/4.
A finite Taylor formula includes an error
- Suppose $f$ has continuous derivatives through order $m+1$ on an interval containing the segment from $a$ to $x$. Put $h=x-a$. Taylor's theorem gives $f(x)=T_m(x)+R_m(x)$, where $R_m(x)=\frac{f^{(m+1)}(\xi)}{(m+1)!}h^{m+1}$ for some $\xi$ between the two points when $x\ne a$.
- If $|f^{(m+1)}(t)|\le M$ on that segment, then $|R_m(x)|\le\frac{M|h|^{m+1}}{(m+1)!}$. Here $M$ may depend on the degree and the segment. A bound on just $f^{(m+1)}(a)$ is insufficient; at $x=a$ the error is zero.
Settle the opening inequality from both sides
- At $x=1/2$, the cubic remainder for $e^x$ is positive. Hence $e^{1/2}>T_3(1/2)=79/48>163/100$. The proposed inequality $e^{0.5}<1.63$ is false. An upper bound on the error alone would not establish this lower bound.
- On $[0,1/2]$, all exponential derivatives are less than $2$: $e^{1/2}=\sum_{k\ge0}(1/2)^k/k!<\sum_{k\ge0}(1/2)^k=2$. With $T_4(1/2)=633/384$ and a positive error below $1/1920$, we obtain $1.6484375<e^{1/2}<1.648959$.
Which statement certifies that $e^{1/2}>1.63$?
难度 1/5 · 今日首次完成可获 100 金币及少量额外奖励
The cubic polynomial is 79/48 > 1.63 and its Lagrange remainder is positive. An upper error bound alone cannot supply a lower bound.
Choose the degree from the error budget
- To guarantee absolute error below $10^{-4}$ at $x=1/2$, use the same derivative bound $M=2$. Degree four gives the bound $1/1920$, which is too large to certify this tolerance. This does not prove its actual error is too large.
- Degree five gives $|R_5(1/2)|<1/23040<10^{-4}$. Thus $T_5(1/2)=6331/3840$ meets the target. An error bound is a guarantee; its failure to meet a target is not evidence that the approximation fails.
An error upper bound larger than the tolerance proves that the actual approximation error exceeds the tolerance.
难度 1/5 · 今日首次完成可获 100 金币及少量额外奖励
A loose upper bound cannot prove a lower bound on the actual error. A sharper estimate may still meet the tolerance.
Use $M=2$ on $[0,1/2]$ to certify $e^{1/2}>1.63$ and choose a Taylor degree guaranteeing error below $10^{-4}$. Explain what a failed upper-bound test does and does not show.
Compare your argument with the model and checklist.
The root formula handles missing coefficients
- For $\sum_{n\ge0}c_n(x-a)^n$, put $L=\limsup_{n\to\infty}|c_n|^{1/n}$. The radius is $R=1/L$, with $R=\infty$ if $L=0$ and $R=0$ if $L=\infty$. The series converges absolutely for $|x-a|<R$ and diverges for $|x-a|>R$; it always converges at its centre.
- The limit superior records the largest coefficient growth that keeps recurring. For $\sum_{n\ge0}x^{2n}$, the odd-power coefficients vanish and the even-power coefficients are $1$. Their successive ratios are unsuitable, but $L=1$, so $R=1$. Finitely many initial terms do not change the radius.
What is the radius of convergence of $\sum_{n\ge0}x^{2n}$, viewed as a power series in $x$?
难度 2/5 · 今日首次完成可获 140 金币及少量额外奖励
It is geometric in x², so |x²| < 1 means |x| < 1. Zero odd coefficients make the successive-coefficient ratio unsuitable.
Use the ratio shortcut only when it exists
- If the coefficients are eventually nonzero and $q=\lim |c_{n+1}/c_n|$ exists, then $R=1/q$, with the same zero/infinity conventions. For $c_n=3^{-n}$, $q=1/3$ and $R=3$; for $c_n=1/n!$, $q=0$ and $R=\infty$.
- For $c_n=n!$, the ratio tends to infinity and $R=0$: $\sum n!x^n$ converges only at zero. A ratio with no limit does not make the radius undefined; return to the root formula or test the nonzero terms directly.
What is the radius of convergence of $\sum x^n/3^n$?
难度 2/5 · 今日首次完成可获 140 金币及少量额外奖励
The ratio test gives |x|/3 < 1.
Shift and scale before testing endpoints
- Consider $\sum_{n\ge1}(x-2)^n/(n3^n)$. Set $u=(x-2)/3$. The series becomes $\sum u^n/n$, so its centre is $2$ and its radius is $3$. Its candidate endpoints are $-1$ and $5$, not $-3$ and $3$.
- At $x=-1$, $u=-1$ gives an alternating harmonic series, converging conditionally. At $x=5$, $u=1$ gives the divergent harmonic series. The interval is $[-1,5)$; absolute convergence holds only on $(-1,5)$.
What is the left endpoint of the convergence interval of $\sum_{n\ge1}(x-2)^n/(n3^n)$?
难度 2/5 · 今日首次完成可获 140 金币及少量额外奖励
The centre is 2 and radius 3, so the left endpoint is 2 - 3 = -1. It is included conditionally.
The same radius allows four endpoint patterns
- All sums here start at $n=1$ and have radius $1$. The geometric series $\sum x^n$ fails at both endpoints because its terms do not tend to zero. The series $\sum x^n/n$ includes only $-1$; replacing $x$ by $-x$ gives $\sum(-x)^n/n$, which includes only $1$.
- The series $\sum x^n/n^2$ includes both endpoints absolutely, by comparison with $\sum1/n^2$. Multiplying geometric terms by $n$ still gives neither endpoint. For $\sum x^{2n}$, both endpoints give terms equal to $1$, so neither is included.
哪些端点被包含?
求出半径后测试端点。
The interval of convergence of $\sum_{n\ge1} x^n/n^2$ is(是)
难度 1/5 · 今日首次完成可获 100 金币及少量额外奖励
At ±1 the series is dominated by Σ 1/n².
Find the centre, radius and full interval of convergence of $\sum_{n\ge1}(-1)^n(x+1)^n/(n2^n)$. State where convergence is absolute and where it is conditional.
Compare your argument with the model and checklist.
Termwise operations keep the radius
- A power series with positive radius can be differentiated and integrated term by term inside its open convergence interval. The resulting power series have the same radius, while endpoint convergence must be checked again. Integration also requires a constant, fixed by a value at the centre.
- On every closed subinterval strictly inside the radius, the series converges uniformly; its differentiated series also does so there. These local controls justify the operations. They do not claim uniform convergence on the whole open interval or permission to differentiate an endpoint sum.
Differentiating a power series term by term keeps its radius of convergence.
难度 1/5 · 今日首次完成可获 100 金币及少量额外奖励
True. The radius stays the same; only convergence at the endpoints can change.
Build logarithmic and derivative series
- For $|x|<1$, integrating the geometric series from $0$ to $x$ gives $-\ln(1-x)=\sum_{n\ge1}x^n/n$. Both sides vanish at zero, so the constant is fixed. Differentiating returns $1/(1-x)=\sum_{n\ge1}x^{n-1}$; neither endpoint survives for the derivative series.
- Differentiating the geometric series also gives $1/(1-x)^2=\sum_{n\ge1}n x^{n-1}$ on $(-1,1)$. A finite value of the rational function outside this interval does not mean its series converges there. The function's domain and a series representation's interval are different objects.
Substitution changes the allowed inputs
- Substituting $2x$ into the exponential series gives $e^{2x}=\sum_{n\ge0}2^nx^n/n!$ for all real $x$. Substituting $-4x^2$ into the geometric series gives $1/(1+4x^2)=\sum_{n\ge0}(-4)^nx^{2n}$ only for $|x|<1/2$.
- At $x=\pm1/2$, these geometric terms are $(-1)^n$, so both endpoints fail the term test. Similarly, $1/(1+x^2)=\sum_{n\ge0}(-1)^nx^{2n}$ has radius $1$, although the rational function is defined for every real $x$.
Where does $\sum_{n\ge0}(-4)^nx^{2n}$ converge?
难度 1/5 · 今日首次完成可获 100 金币及少量额外奖励
The geometric ratio is -4x². Its absolute value must be below 1, giving |x| < 1/2. At either endpoint the terms are (-1)^n and do not tend to zero.
Start with $1/(1-x)=\sum_{n\ge0}x^n$. Derive series for $1/(1-x)^2$ and $-\ln(1-x)$, state the open interval where the operations are justified, and explain why their endpoints need fresh checks.
Compare your argument with the model and checklist.
Control what remains after cancellation
- Taylor's theorem gives $e^x=1+x+x^2/2+O(x^3)$ near zero. After subtracting $1+x$ and dividing by $x^2$, the error is $O(x)$, so the limit is $1/2$. Here $O(x^k)$ means an absolute error at most $C|x|^k$ near zero for a fixed constant $C$.
- Likewise, $\cos x=1-x^2/2+O(x^4)$ gives $(\cos x-1)/x^2\to-1/2$. For a higher-order cancellation, $\sin x=x-x^3/6+O(x^5)$ gives $(\sin x-x)/x^3\to-1/6$. Retain the first nonzero term that survives division, with a smaller remaining error.
Evaluate $\lim_{x \to 0} \dfrac{\cos x - 1}{x^2}$.
难度 2/5 · 今日首次完成可获 140 金币及少量额外奖励
$\cos x - 1 = -x^2/2 + \cdots$.
Evaluate $\lim_{x\to0}(\sin x-x)/x^3$. Give a decimal to three places.
难度 2/5 · 今日首次完成可获 140 金币及少量额外奖励
Since sin x = x - x³/6 + O(x⁵), division gives -1/6 + O(x²), tending to -1/6 ≈ -0.167.
Smoothness does not guarantee Taylor equality
- Define $h(0)=0$ and $h(x)=e^{-1/x^2}$ for $x\ne0$. It is positive away from zero, yet every derivative at zero is zero. Thus every Maclaurin polynomial is the zero polynomial, and its infinite Taylor series is zero everywhere.
- To see why, every derivative away from zero is $e^{-1/x^2}$ times a polynomial in $1/x$. For every fixed $p\ge0$, $|x|^{-p}e^{-1/x^2}\to0$: put $u=1/x^2$ and use exponential growth faster than any power. The same bound on difference quotients proves inductively that all derivatives extend to zero with value zero.
Near zero, the positive curve is too small to distinguish from the zero line at this scale.
If a smooth function has a Taylor series with infinite radius of convergence, that series must equal the function everywhere.
难度 1/5 · 今日首次完成可获 100 金币及少量额外奖励
The function h(0)=0, h(x)=exp(-1/x²) for x≠0 has all Taylor coefficients zero. Its zero series converges everywhere but differs from h away from zero.
Check equality by sending the remainder to zero
- For an infinitely differentiable $f$, its Taylor series equals $f(x)$ exactly when $R_m(x)=f(x)-T_m(x)$ tends to zero as $m\to\infty$. Being analytic at the centre means this equality holds on some neighbourhood of the centre. Smoothness alone gives no such guarantee.
- For $e^x$ about zero and fixed real $x$, use $M=e^{|x|}$. Then $|R_m(x)|\le\frac{M|x|^{m+1}}{(m+1)!}\to0$, since the ratio of successive bounds tends to zero. For $h$ above and fixed $x\ne0$, the remainder is always $h(x)>0$, despite its Taylor series having infinite radius.
哪些保证是有效的?
分别列出半径、等式和误差保证。
Let $h(0)=0$ and $h(x)=e^{-1/x^2}$ for $x\ne0$, and define $g(x)=h(x)+e^x$. Find the Maclaurin coefficients of $g$, their radius of convergence, and where the resulting series equals $g$.
Compare your argument with the model and checklist.
Recap
- Taylor polynomial: A polynomial formed from the derivatives of a function at a chosen centre, divided by the corresponding factorials.
- remainder: The difference between a function and its Taylor polynomial at the point being approximated.
- radius: The distance from the centre within which a power series converges absolutely.
- Bound derivatives on the whole segment, test endpoints separately, and distinguish convergence of a Taylor series from equality with its source function.