Incremental Approximation and Error Estimation Methods | 增量近似与误差估计方法

📚 Incremental Approximation and Error Estimation Methods | 增量近似与误差估计方法

One of the most powerful applications of calculus is using the derivative to predict how a quantity changes when its input changes by a small amount. Instead of computing the exact new value through a lengthy formula, we can use a tangent line to approximate the result instantly. This idea, known as incremental approximation, also leads directly to methods for estimating and propagating errors in measurement.

微积分最强大的应用之一，就是利用导数来预测当输入发生微小变化时，一个量会如何变化。与其通过冗长的公式精确计算新值，我们不如用一条切线来即时逼近结果。这个思想被称为增量近似，它同时也直接引出了估计与传播测量误差的方法。

1. The Tangent Line as an Approximator | 切线的近似作用

Consider a function f(x) that is differentiable at a point a. Near the point (a, f(a)), the graph of f is very close to its tangent line. The equation of this tangent line is L(x) = f(a) + f'(a)(x – a). For values of x close to a, L(x) gives a very good approximation of f(x).

考虑一个在点 a 处可导的函数 f(x)。在点 (a, f(a)) 附近，f 的图像与它的切线非常接近。这条切线的方程为 L(x) = f(a) + f'(a)(x – a)。当 x 的值靠近 a 时，L(x) 能非常精确地逼近 f(x)。

Linear Approximation: f(x) ≈ f(a) + f'(a)(x – a)

线性近似：f(x) ≈ f(a) + f'(a)(x – a)

For example, let f(x) = √x and choose a = 25. Since f(25) = 5 and f'(25) = 1/(2√25) = 0.1, we can approximate √26 using L(26) = 5 + 0.1(1) = 5.1. The true value is 5.099, so the error is only about 0.001.

例如，设 f(x) = √x，取 a = 25。因为 f(25) = 5，f'(25) = 1/(2√25) = 0.1，我们可以用 L(26) = 5 + 0.1(1) = 5.1 来近似 √26。真实值约为 5.099，误差仅约 0.001。

2. From Δy to dy: The Differential Method | 从 Δy 到 dy：微分方法

Let x increase from a to a + Δx. The exact change in f is Δy = f(a + Δx) – f(a). When Δx is small, we can replace this exact change with the differential dy = f'(a)Δx. The notation dx is used instead of Δx to emphasize an infinitesimal increment, but in approximation problems we treat them the same way.

设 x 从 a 增加到 a + Δx。f 的精确变化量为 Δy = f(a + Δx) – f(a)。当 Δx 很小时，我们可以用微分 dy = f'(a)Δx 来替代这个精确变化量。记号 dx 用来强调无穷小增量，而在近似问题中我们将其与 Δx 等同处理。

δy ≈ dy = f'(x)δx

The symbol δ (delta) is often used to denote a small but measurable change, such as δx = 0.01 or δx = 0.001. The key idea is that the derivative f'(x) acts as a magnification factor: it tells us how many units y changes for each unit of change in x.

符号 δ（delta）通常用于表示一个微小但可测量的变化，例如 δx = 0.01 或 δx = 0.001。核心思想是：导数 f'(x) 扮演了一个放大系数的角色——它告诉我们 x 每变化一个单位，y 会相应变化多少个单位。

3. How Accurate Is the Approximation? The Taylor Bound | 近似有多准确？泰勒界

Since we are replacing a curve with a straight line, our estimate always contains some error. For a twice-differentiable function, the error of linear approximation is bounded by a quantity involving the second derivative. If |f”(t)| ≤ M for all t between a and a + Δx, then the absolute error satisfies:

既然我们用直线替代了曲线，估计值就总会包含一定误差。对于二阶可导的函数，线性近似的误差可以由一个涉及二阶导数的量来界定。如果对所有介于 a 与 a + Δx 之间的 t，都有 |f”(t)| ≤ M，那么绝对误差满足：

|Actual Error| = |f(a + Δx) – L(a + Δx)| ≤ (M/2)·(Δx)²

This quadratic behaviour explains why halving the step Δx reduces the error by a factor of four. It also tells us that linear approximation is only reliable when Δx is genuinely small.

这种二次行为解释了为什么将步长 Δx 减半会使误差缩小为原来的四分之一。它也告诉我们：只有当 Δx 真正很小时，线性近似才是可靠的。

For example, for f(x) = √x near a = 25, we have f”(x) = -1/(4x^(3/2)). The magnitude reaches its maximum at x = 25, giving M = 1/(4·125) = 0.002. For Δx = 1, the error bound is 0.001, matching our earlier observation.

例如，对于 a = 25 附近的 f(x) = √x，有 f”(x) = -1/(4x^(3/2))。其绝对值在 x = 25 处达到最大，得到 M = 1/(4·125) = 0.002。对于 Δx = 1，误差界为 0.001，与我们之前的观察一致。

4. Measuring Error: Absolute, Relative and Percentage | 误差度量：绝对误差、相对误差与百分误差

When a physical quantity is measured, there is always uncertainty. A measurement x has an absolute error δx, which is the maximum possible difference between the measured and true value. For instance, if a length is measured as 10.0 cm with a ruler accurate to 0.1 cm, then δx = 0.05 cm.

当我们测量一个物理量时，总会有不确定性。一个测量值 x 具有绝对误差 δx，即测量值与真实值之间可能的最大差值。例如，若用精确到 0.1 cm 的刻度尺测得长度为 10.0 cm，则 δx = 0.05 cm。

Absolute error: δx = difference between measured and true value (measured value uncertainty).

绝对误差：δx = 测量值与真实值之差（测量值的不确定度）。
Relative error: δx / |x|, a dimensionless ratio.

相对误差：δx / |x|，一个无量纲的比值。
Percentage error: (δx / |x|) × 100%.

百分误差：(δx / |x|) × 100%。

Relative and percentage errors are often more meaningful than absolute errors because they indicate the severity of the uncertainty relative to the size of the measurement.

相对误差和百分误差通常比绝对误差更有意义，因为它们反映了不确定性相对于测量值大小的严重程度。

5. Error Propagation: Addition and Subtraction | 误差传播：加法与减法

Suppose we calculate a derived quantity z = x + y, where x and y are measured with errors δx and δy. The worst-case absolute error in z is the sum of the individual errors: δz = δx + δy. This is because the errors could work in the same direction.

假设我们计算一个导出量 z = x + y，其中 x 和 y 的测量误差分别为 δx 和 δy。z 在最坏情况下的绝对误差等于各误差之和：δz = δx + δy。这是因为误差可能沿同一方向作用。

For z = x + y or z = x – y: δz = δx + δy

Note that the same rule applies to subtraction. Although a negative sign may cancel errors on average, the maximum possible error is still the sum of the magnitudes.

注意：减法遵循同样的规则。尽管负号在平均意义上可能抵消误差，但最大可能的误差仍然是各绝对误差之和。

For example, if we measure two sides of a rectangle as 3.0 ± 0.1 m and 4.0 ± 0.1 m, the perimeter P = 2(3.0 + 4.0) = 14.0 m. The error in the sum is 0.2 m, so δP = 2 × 0.2 = 0.4 m. The perimeter is 14.0 ± 0.4 m.

例如，测得矩形的两条边分别为 3.0 ± 0.1 m 和 4.0 ± 0.1 m，则周长 P = 2(3.0 + 4.0) = 14.0 m。和的误差为 0.2 m，因此 δP = 2 × 0.2 = 0.4 m，即周长为 14.0 ± 0.4 m。

6. Error Propagation: Multiplication and Division | 误差传播：乘法与除法

For products and quotients, relative errors combine by addition. If z = x·y, then we can take the natural logarithm of both sides and differentiate, or use the product rule with differentials:

对于乘法与除法，相对误差通过加法来合成。若 z = x·y，我们可以对两边取自然对数再微分，或使用带微分的乘法法则：

For z = x·y or z = x/y: δz/|z| = δx/|x| + δy/|y|

This rule is remarkably simple: the relative error of a product or quotient is the sum of the relative errors of its factors. The derivation comes directly from z = x·y giving dz/z = dx/x + dy/y, and similarly for division.

这条规则非常简洁：乘积或商量的相对误差等于各因子的相对误差之和。推导直接来自对 z = x·y 求微分得到 dz/z = dx/x + dy/y，商的情况同理。

For instance, to find the area of a rectangle, measure width = 5.0 ± 0.1 m and height = 2.0 ± 0.1 m. The relative errors are 0.1/5.0 = 0.02 and 0.1/2.0 = 0.05. The total relative error is 0.07, so the area is 10.0 m² with absolute error 0.7 m², giving A = 10.0 ± 0.7 m².

例如，为求矩形面积，测得宽 = 5.0 ± 0.1 m，高 = 2.0 ± 0.1 m。相对误差分别为 0.1/5.0 = 0.02 和 0.1/2.0 = 0.05。总相对误差为 0.07，故面积为 10.0 m²，绝对误差为 0.7 m²，即 A = 10.0 ± 0.7 m²。

7. Power Functions and Compound Errors | 幂函数与复合误差

When a quantity appears as a power, z = xⁿ, the differentiation rule dz = n·x^(n-1)dx can be converted directly into a relative-error statement:

当一个量以幂的形式出现时，如 z = xⁿ，微分规则 dz = n·x^(n-1)dx 可以直接转化为相对误差的表述：

For z = xⁿ: δz/|z| = |n|·(δx/|x|)

This formula is especially important in geometry. Since the volume of a sphere V = (4/3)πr³ involves r³, a relative error of 1% in the radius leads to a 3% relative error in the volume. Similarly for area A = πr², the relative error is doubled.

这个公式在几何中尤为重要。由于球体积 V = (4/3)πr³ 涉及 r³，半径的 1% 相对误差会导致体积的 3% 相对误差。类似地，对于面积 A = πr²，相对误差会加倍。

For more complex functions such as z = sin x, we cannot use the relative-error shortcut. Instead, we apply the general rule δz ≈ |f'(x)|·δx directly. For f(x) = sin x, this gives δz ≈ |cos x|·δx, which depends strongly on the value of x.

对于更复杂的函数，如 z = sin x，我们不能使用相对误差的快捷法则，而应直接应用一般规则 δz ≈ |f'(x)|·δx。对于 f(x) = sin x，有 δz ≈ |cos x|·δx，其值强烈依赖于 x 的大小。

8. The General Strategy for Estimating Errors | 误差估计的一般策略

Any differentiable relationship y = f(x) allows us to estimate output error from input error. The chain rule extends this to multiple variables. If y depends on two measured quantities u and v, the total differential provides a systematic estimate:

任何可导的关系 y = f(x) 都允许我们从输入误差估计输出误差。链式法则将其推广到多个变量。如果 y 依赖于两个被测量 u 和 v，全微分提供了一种系统性的估计：

dy = (∂y/∂u)du + (∂y/∂v)dv

Since partial derivatives may have either sign, and errors add in the worst case, we take absolute values of each term when computing maximum possible error:

由于偏导数可能为正或为负，而在最坏情况下误差相加，在计算最大可能误差时我们对每一项取绝对值：

δy = |∂y/∂u|·δu + |∂y/∂v|·δv

This method, called the linear error-propagation formula, is the backbone of experimental sciences and appears regularly in IB Paper 2 questions.

这种方法称为线性误差传播公式，是实验科学的核心工具，也经常出现在 IB 试卷二的题目中。

9. Worked IB Example: Surface Area of a Sphere | IB 例题：球的表面积

The radius of a sphere is measured as 10.0 cm with a maximum error of 0.05 cm. Estimate the maximum error in the surface area A = 4πr².

测得一个球的半径为 10.0 cm，最大测量误差为 0.05 cm。估计球的表面积 A = 4πr² 的最大误差。

Step 1: Differentiate A with respect to r.

第一步：对 r 求导。

dA/dr = 8πr, so δA ≈ 8πr·δr

Step 2: Substitute r = 10.0 and δr = 0.05.

第二步：代入 r = 10.0 与 δr = 0.05。

δA ≈ 8π(10.0)(0.05) = 4π ≈ 12.6 cm²

Step 3: Express as a percentage error. The measured area is A = 4π(10.0)² = 400π cm². The relative error is δA/A = (4π)/(400π) = 0.01, so the percentage error is 1%. This matches the power rule since A ∝ r².

第三步：用百分误差表示。测得面积为 A = 4π(10.0)² = 400π cm²。相对误差为 δA/A = (4π)/(400π) = 0.01，因此百分误差为 1%。这与幂函数法则一致，因为 A ∝ r²。

We conclude the surface area is 400π ± 4π cm², or approximately 1257 ± 12.6 cm².

我们得出结论：该球表面积为 400π ± 4π cm²，即约 1257 ± 12.6 cm²。

10. Common Pitfalls and Exam Tips | 常见错误与考试技巧

Students often confuse which formula to use. If the function is simple, such as z = x² or z = xy, use the power rule or relative-error rule directly. If the function is complicated, such as z = ln x or z = eˣ, return to the fundamental expression δz ≈ |f'(x)|·δx.

学生常常混淆应该使用哪个公式。如果函数简单，如 z = x² 或 z = xy，直接使用幂规则或相对误差法则。如果函数复杂，如 z = ln x 或 z = eˣ，则回到基本表达式 δz ≈ |f'(x)|·δx。

Always square the derivative for power functions: a 2% error in x does not produce a 2% error in x², it produces a 4% error.

对于幂函数，务必乘以指数：x 的 2% 误差不会导致 x² 的 2% 误差，而会产生 4% 的误差。
Do not forget absolute values. Relative error is defined as positive, so δx/|x| is always non-negative.

不要忘记绝对值。相对误差定义为正值，因此 δx/|x| 始终非负。
When adding or subtracting, absolute errors add; when multiplying or dividing, relative errors add.

加或减时，绝对误差相加；乘或除时，相对误差相加。
Express final answers in the form value ± error, and always state the unit.

最终结果以 “值 ± 误差” 的形式表示，并始终注明单位。

In exams, show the differential step explicitly rather than jumping to the final number. Markers award full marks for the method, and the approximation itself is only one mark.

在考试中，要清晰写出微分步骤，而不是直接跳到最终数字。阅卷者会为方法给出满分，而近似结果本身只占一分。

Summary | 总结

Incremental approximation replaces exact calculations with derivative-based estimates. The fundamental relation Δy ≈ f'(x)Δx gives us a fast and powerful way to predict changes and quantify errors. Errors propagate additively for sums and differences, and relative errors propagate additively for products, quotients and powers.

增量近似用基于导数的估计取代精确计算。基本关系 Δy ≈ f'(x)Δx 为我们提供了一种快速而强大的方式来预测变化并量化误差。对和与差，绝对误差相加；对积、商和幂，相对误差相加。

Mastering these two ideas together is essential for IB Mathematics Analysis and Approaches and Applications and Interpretation, since they connect calculus, algebra and measurement science in one coherent toolkit.

同时掌握这两种思想对 IB 数学分析与方法（AA）以及应用与解释（AI）都至关重要，因为它们将微积分、代数和测量科学连接成一个连贯的工具箱。

Published by TutorHao | Mathematics Revision Series | aleveler.com

更多咨询请联系16621398022（同微信）

Incremental Approximation and Error Estimation Methods | 增量近似与误差估计方法