📚 Applying the Central Limit Theorem to Other Distributions | 将中心极限定理应用于其他分布
The Central Limit Theorem (CLT) is a cornerstone of probability theory. It states that when independent random variables are sampled from any distribution with a finite mean and variance, the distribution of the sample mean approaches a normal distribution as the sample size increases. While originally studied in statistics, the CLT has profound implications in computer science — from validating simulations and randomised algorithms to performance modelling and error analysis. This article explores how the CLT can be applied to various distributions commonly encountered in computing contexts, and why this matters for A-Level Computer Science students.
中心极限定理是概率论的基石。该定理指出，从任何具有有限均值和方差的分布中抽取独立随机变量时，随着样本容量的增加，样本均值的分布会趋近于正态分布。虽然中心极限定理最初是在统计学中研究的，但它在计算机科学中有着深远的影响——从验证模拟和随机算法到性能建模和误差分析。本文探讨了中心极限定理如何应用于计算环境中常见的各种分布，以及为什么这对A-Level计算机科学学生很重要。
1. What Is the Central Limit Theorem? | 什么是中心极限定理？
The CLT can be formally stated as follows: let X1, X2, ..., Xn be independent and identically distributed random variables with mean μ and finite variance σ2. Define the sample mean X̄ = (1/n) ∑ Xi. Then, as n → ∞, the standardised statistic (X̄ − μ) / (σ/√n) converges in distribution to a standard normal distribution N(0,1). Equivalently, X̄ approximately follows N(μ, σ2/n) for large n, regardless of the shape of the original population distribution.
中心极限定理的正式表述如下：设 X1, X2, ..., Xn 是独立同分布的随机变量，均值为 μ，方差为有限值 σ2。定义样本均值 X̄ = (1/n) ∑ Xi。那么当 n → ∞ 时，标准化统计量 (X̄ − μ) / (σ/√n) 的分布收敛到标准正态分布 N(0,1)。等价地，对于大样本 n，X̄ 近似服从 N(μ, σ2/n)，而不论原始总体分布的形状如何。
ften considered sufficient for the CLT to provide a reasonable normal approximation, but this rule of thumb depends on how skewed or heavy-tailed the original distribution is. In computing experiments, we frequently verify this using random number generation.
在实践中，通常认为样本量 n ≥ 30 足以使中心极限定理提供合理的正态近似，但这条经验法则取决于原始分布的偏斜程度或厚尾程度。在计算实验中，我们经常通过随机数生成来验证这一点。
2. Why Does CLT Matter in Computer Science? | 中心极限定理在计算机科学中的重要性
Many problems in computer science involve randomness, whether in Monte Carlo simulations, machine learning model initialisation, or network traffic modelling. The CLT allows us to make probabilistic statements about the average behaviour of such systems without knowing the exact underlying distribution. For example, the average response time of a web server handling requests drawn from a heavy-tailed distribution will nevertheless be approximately normally distributed if we consider average response times over many requests.
计算机科学中的许多问题都涉及随机性，无论是在蒙特卡洛模拟、机器学习模型初始化还是网络流量建模中。中心极限定理使我们能够对这类系统的平均行为做出概率性陈述，而无需知道确切的底分布。例如，处理来自厚尾分布的请求的Web服务器的平均响应时间，如果我们考虑大量请求的平均响应时间，仍将近似服从正态分布。
Moreover, algorithm designers rely on the CLT when deriving average-case complexity bounds for randomised algorithms. The sum of costs across many independent runs tends to a normal distribution, which aids in setting confidence intervals on performance. Simulators and benchmarking tools routinely use normality assumptions to report average throughput, latency, and error margins.
此外，算法设计者在推导随机算法的平均情况复杂度界限时依赖中心极限定理。大量独立运行的成本之和趋于正态分布，这有助于设定性能的置信区间。模拟器和基准测试工具经常使用正态性假设来报告平均吞吐量、延迟和误差范围。
3. Applying CLT to Uniform Distribution | 将中心极限定理应用于均匀分布
, if we take the mean of n independent uniform(0,1) values, the distribution of this sample mean becomes more concentrated and bell-shaped as n grows. Even for n as small as 5, the density starts to show a peak near 0.5, and by n=30 it is virtually normal.
[0,1] 上的均匀分布是平坦的，而非钟形。如果我们只抽取一个值，区间内的每个结果出现的可能性相同。但如果取 n 个独立 uniform(0,1) 值的平均值，随着 n 的增加，这个样本均值的分布会变得更集中并呈钟形。即使 n 小至 5，密度曲线在 0.5 附近就开始出现峰值，到 n=30 时几乎就是正态分布。
This property is used in simple random number generators to create approximate normal deviates. For instance, by summing 12 uniform random numbers and subtracting 6, we get a value with mean 0 and variance 1, which closely follows a standard normal distribution — a handy trick in embedded systems without a dedicated normal RNG.
这个特性用于简单的随机数生成器中，以产生近似正态的偏离值。例如，通过将 12 个均匀随机数相加并减去 6，就得到一个均值为 0、方差为 1 的值，该值近似服从标准正态分布——这是在无专用正态随机数生成器的嵌入式系统中一个方便的诀窍。
4. Binomial Approximation via CLT | 通过中心极限定理近似二项分布
th mean np and variance np(1−p). Thus, we can estimate probabilities like P(X ≤ k) using the standard normal CDF after applying a continuity correction. This is fundamental in analysing error rates in data transmission where bit errors are modelled as Bernoulli trials.
二项分布 Bin(n, p) 是 n 次独立伯努利试验之和。根据中心极限定理，对于大 n，成功次数 X 近似服从正态分布，均值为 np，方差为 np(1−p)。因此，我们可以应用连续性校正后，使用标准正态累积分布函数估计像 P(X ≤ k) 这类概率。这在分析数据传输中的差错率时是基础性的，其中比特错误被建模为伯努利试验。
For example, in a network sending 10,000 bits with an error probability of 0.001, the probability of more than 15 errors can be approximated by normal methods, saving the computational cost of exact binomial calculations. This example illustrates why CLT-based approximations are embedded in network simulation software.
例如，在一个发送 10,000 比特、错误概率为 0.001 的网络中，出现超过 15 个错误的概率可以用正态方法近似，从而节省了精确二项式计算的计算开销。这个例子说明了为何基于中心极限定理的近似方法被嵌入到网络模拟软件中。
5. Poisson Distribution and the CLT | 泊松分布与中心极限定理
A Poisson distribution with parameter λ counts the number of events in a fixed interval. For large λ, a Poisson random variable can be approximated by a normal distribution with mean λ and variance λ, because the Poisson can be viewed as a sum of many independent rare-event indicators. In computer networks, packet arrival counts per second often follow a Poisson process; when λ is high, the number of arrivals is approximately normal, simplifying queueing analysis.
参数为 λ 的泊松分布计数固定间隔内的事件数。对于大 λ，泊松随机变量可以用均值为 λ、方差为 λ 的正态分布来近似，因为泊松可视为许多独立稀有事件指示变量之和。在计算机网络中，每秒的数据包到达数通常遵循泊松过程；当 λ 很大时，到达数量近似于正态分布，从而简化了排队分析。
This normal approximation also aids in server capacity planning. System administrators can estimate the probability that the number of concurrent users exceeds a threshold using the normal CDF, instead of summing many Poisson probabilities, leading to faster real-time monitoring algorithms.
这种正态近似也有助于服务器容量规划。系统管理员可以使用正态累积分布函数估计并发用户数超过某一阈值的概率，而无需对大量泊松概率求和，从而产生更快的实时监控算法。
6. Exponential Distribution: Sample Means Become Normal | 指数分布：样本均值趋于正态
The exponential distribution with rate λ is heavily right-skewed. Individual service times or hardware failure times often follow this distribution. Despite the skewness, if we collect the average of n independent exponential variables, the central limit theorem guarantees that the average becomes approximately normal with mean 1/λ and variance 1/(nλ2). Consequently, the reliability analysis of a cluster of servers can use normal-based confidence intervals for the mean time to failure.
参数为 λ 的指数分布是严重右偏的。单个服务时间或硬件故障时间往往服从这个分布。尽管存在偏斜，但如果我们收集 n 个独立指数变量的平均值，中心极限定理保证该平均值近似服从正态分布，均值为 1/λ，方差为 1/(nλ2)。因此，对服务器集群进行可靠性分析时，可以使用基于正态分布的置信区间来估计平均无故障时间。
In load-testing exercises, engineers often take repeat samples of transaction processing times and use the sample mean's normality to set performance targets. This approach is valid even if the raw timing data are far from normal; the averaging process invoked by CLT justifies the use of parametric statistical tests.
在负载测试中，工程师经常对事务处理时间进行重复采样，并利用样本均值的正态性来设定性能目标。即使原始时间数据远非正态，这种方法也是有效的；中心极限定理所调用的平均化过程证明了使用参数统计检验的合理性。
7. Simulating CLT with Code: A Uniform Example | 用代码模拟中心极限定理：以均匀分布为例
of sample means, each computed from, say, 30 uniform(0,1) random numbers. Plot a histogram of these 10,000 means. You will observe a bell-shaped curve centred near 0.5 with a standard deviation close to √1/12 / 30 ≈ 0.053. This simple experiment demonstrates convergence first-hand.
为了理解中心极限定理是如何运作的，你可以用 Python 编写一个简短的程序。生成 N 个（例如 N = 10,000）样本均值的样本，每个均值由 30 个 uniform(0,1) 随机数计算得到。画出这 10,000 个均值的直方图。你将看到一条以 0.5 为中心的钟形曲线，标准差接近 √1/12 / 30 ≈ 0.053。这个简单的实验第一手展示了收敛性。
Code outline: use random.random() to generate the raw values, compute sum(sample)/n, repeat, and feed the list into a plotting library. Even with small modifications — replacing the uniform distribution with a skewed beta or discrete empirical distribution — the mean's distribution still approaches normality, echoing the elegance of the theorem.
代码要点：使用 random.random() 生成原始值，计算 sum(sample)/n，重复该过程，然后将列表输入绘图库。即使稍作修改——用偏斜的 beta 分布或离散经验分布替换均匀分布——均值的分布依然会趋向正态，呼应了该定理的优雅性。
8. CLT in Randomised Algorithms: Quicksort Pivot Analysis | 随机算法中的中心极限定理：快速排序基准分析
Quicksort's average-case running time is O(n log n), but the actual number of comparisons varies randomly depending on pivot choices. Over many independent runs on random arrays of the same size, the total number of comparisons can be treated as a sum of many weakly dependent variables. Invoking CLT, the distribution of comparisons is approximately normal. This allows us to answer questions like 'what is the probability that quicksort requires more than 3n ln n comparisons?' using normal quantiles.
快速排序的平均情况运行时间是 O(n log n)，但实际的比较次数会因基准点的选择而随机波动。在对等长随机数组进行大量独立运行时，总的比较次数可视为许多弱相关变量之和。借助中心极限定理，比较次
Published by TutorHao | A-Level Computer Science 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