📚 Logarithms and Non-Linear Data | 对数与非线性数据
In A-Level Mathematics, many real-world relationships are not linear. Logarithms provide a powerful tool to transform non-linear data into linear form, allowing us to estimate parameters and test models. This article explores exponential and power-law models, the process of linearisation using logarithms, and how to interpret log-linear and log-log graphs.
在A-Level数学中，许多现实世界的关系并非线性。对数提供了一种强大的工具，可以将非线性数据转化为线性形式，从而估计参数并检验模型。本文探讨指数模型和幂律模型，利用对数进行线性化的过程，以及如何解读半对数图和双对数图。
1. Recognising Non-Linear Patterns | 识别非线性模式
When plotting bivariate data, a scatter plot may reveal a curve rather than a straight line. Common non-linear patterns include rapid growth (exponential), decay, or a polynomial curve that bends systematically away from a line. Recognising these patterns is the first step before applying logarithms.
绘制双变量数据的散点图时，可能呈现出曲线而非直线。常见的非线性模式包括快速增长（指数型）、衰减，或者系统偏离直线的多项式曲线。识别这些模式是应用对数之前的第一步。
2. The Role of Logarithms in Linearisation | 对数在线性化中的作用
turn multiplicative relationships into additive ones. For an exponential function y = a bx ing logs gives log y = log a + x log b, which is linear in x. For a power function y = a xn , linear in log x. This allows us to use linear regression techniques on transformed data.
对数能够压缩宽广的范围，并将乘法关系转化为加法关系。对于指数函数 y = a bx，取对数得到 log y = log a + x log b，这是关于 x 的线性形式。对于幂函数 y = a xn，对两个轴取对数得到 log y = log a + n log x，关于 log x 呈线性。这使我们能够对转换后的数据使用线性回归方法。
3. Exponential Model 1: y = abx | 指数模型一：y = abx
Many growth and decay processes are modelled by y = a bx, where a is the initial value, b is the growth factor (b > 1 growth, 0 < b < 1 decay). To estimate a and b from data, we cannot use simple linear regression directly. Instead, we linearise: take base-10 logs (or any base) to obtain the logarithmic equation.
许多增长和衰退过程都可用 y = a bx 建模，其中 a 为初始值，b 为增长因子（b > 1 增长，0 < b < 1 衰退）。为从数据中估计 a 和 b，我们不能直接使用简单的线性回归。而是通过线性化：取以 10 为底的对数（或任意底数）得到对数方程。
log₁₀ y = log₁₀ a + x log₁₀ b
traight line with gradient log₁₀ b and vertical intercept log₁₀ a. We can then find a = 10log₁₀ a and b = 10log₁₀ b.
绘制 log₁₀ y 对 x 的图形应得到一条直线，其斜率为 log₁₀ b，截距为 log₁₀ a。然后我们可以求得 a = 10log₁₀ a，b = 10log₁₀ b。
4. Transforming y = abx to a Straight Line | 将 y = abx 转化为直线
. Perform linear regression on (x, Y) to find the best-fit line Y = m x + c. Then calculate a = 10c (if common log) and b = 10m. For natural logs, a = ec, b = em. Always use the same base consistently.
给定数据集 (x, y)，创建新变量 Y = log y（或 ln y）。然后绘制 (x, Y)。如果模型成立，数据点应大致落在一条直线上。对 (x, Y) 进行线性回归，求得最佳拟合直线 Y = m x + c。然后计算 a = 10c（若使用常用对数）和 b = 10m。使用自然对数时，a = ec, b = em。务必始终使用同一种底数。
5. Exponential Model 2: y = aekx | 指数模型二：y = aekx
The continuous exponential model y = a ek x is common in physics and biology. Taking natural logarithms gives a very simple linear form.
连续指数模型 y = a ek x 在物理和生物学中十分常见。取自然对数会得到一个非常简单的线性形式。
ln y = ln a + k x
ntinuous rate k. For growth, k > 0; for decay, k < 0. After finding ln a, calculate a = eln a.
绘制 ln y 对 x 的图形得到一条直线，其斜率为 k，截距为 ln a。这样做更简单，因为斜率直接给出连续速率 k。当 k > 0 时为增长；k < 0 时为衰减。求得 ln a 后，计算 a = eln a。
6. Power Law Model: y = axn | 幂律模型：y = axn
Power functions appear in allometry, economics, and physics. If y = a xn where a and n are constants, then taking logs (any base) of both sides gives a relationship that is linear in log x and log y.
幂函数广泛出现在异速生长、经济学和物理学中。若 y = a xn，其中 a 与 n 为常数，两边取对数（任意底数）可得到一个关于 log x 和 log y 的线性关系。
log y = log a + n log x
A log-log plot (log y vs log x) should produce a straight line with slope n and intercept log a. This transformation requires both x and y to be positive.
双对数图（log y 对 log x）应该得到一条直线，斜率为 n，截距为 log a。此变换要求 x 和 y 均为正数。
7. Transforming Power Law to Linear Form | 将幂律转化为线性形式
raight line. Use linear regression to estimate gradient n and intercept log a. Then a = 10intercept (or eintercept l scatter plot and residuals after back-transformation to ensure the model is appropriate.
要拟合幂律，计算每个数据点的 X = log x 和 Y = log y。绘制 Y 对 X 的图形。若关系拟合良好，点将位于一条直线上。利用线性回归估计斜率 n 和截距 log a。然后 a = 10截距（或 e截距），n = 斜率。通过回代逆变换后，检查原始散点图和残差以确保模型合适。
8. Worked Example: Exponential Decay Data | 实例分析：指数衰减数据
Data: (t=0, m=80), (t=1, m=56), (t=2, m=39), (t=3, m=27), (t=4, m=19). We suspect m = a bt. Take common log of m: log m values are approximately 1.903, 1.748, 1.591, 1.431, 1.279. Plot log m against t. The points are nearly linear.
假设测量放射性物质的质量 m 随时间 t 变化的数值。数据：(t=0, m=80), (t=1, m=56), (t=2, m=39), (t=3, m=27), (t=4, m=19)。我们猜想 m = a bt。取 m 的常用对数：log m 值约为 1.903, 1.748, 1.591, 1.431, 1.279。绘制 log m 对 t 的图形，点几乎呈直线。
Using linear regression, we find the slope ≈ -0.156, intercept ≈ 1.903. Then a = 101.903 ≈ 80.0, b = 10-0.156 ≈ 0.698. So the model is m = 80 × 0.698t, which describes exponential decay.
利用线性回归，求得斜率 ≈ -0.156，截距 ≈ 1.903。于是 a = 101.903 ≈ 80.0，b = 10-0.156 ≈ 0.698。因此模型为 m = 80 × 0.698t，描述了指数衰减。
9. Worked Example: Power Law Data | 实例分析：幂律数据
r similar shapes: (A=2, V=1.2), (A=5, V=4.5), (A=8, V=9.1), (A=12, V=16.8). Assume V = k An = log A, Y = log V. Log A: 0.301, 0.699, 0.903, 1.079; log V: 0.079, 0.653, 0.959, 1.225.
考虑相似形状的表面积 A 和体积 V 的测量值：(A=2, V=1.2), (A=5, V=4.5), (A=8, V=9.1), (A=12, V=16.8)。假设 V = k An。取对数：X = log A, Y = log V。log A: 0.301, 0.699, 0.903, 1.079；log V: 0.079, 0.653, 0.959, 1.225。
ight line. Linear regression gives gradient n ≈ 1.5, intercept log k ≈ -0.373. Thus k = 10-0.373 ≈ 0.424. The fitted model is V ≈ 0.424 A1.5. This matches the expected geometric scaling (volume ∝ surface area3/2).
绘制 Y 对 X 的图形得到一条直线。线性回归得出斜率 n ≈ 1.5，截距 log k ≈ -0.373。因此 k = 10-0.373 ≈ 0.424。拟合模型为 V ≈ 0.424 A1.5。这与预期的几何标度关系（体积 ∝ 表面积3/2）相符。
10. Choosing Between Semi-Log and Log-Log Plots | 在半对数图与双对数图之间选择
If the data suggest an exponential relationship, use a semi-log plot: plot log y (or ln y) against x. If the data follow a power law, use a log-log plot: plot log y against log x. Visual inspection of the original scatter plot can give clues; exponential curves usually show a constant multiplicative increase, while power laws appear as curves that do not
Published by TutorHao | A-Level Mathematics Revision Series | aleveler.com
更多咨询请联系16621398022（同微信）
或截图后用「扫一扫 → 相册」选择本图