Blog

📚 Lightweight Neural Network Models Explained | 轻量级神经网络模型解析

In the era of edge computing and real-time AI applications, deploying large deep neural networks on resource-constrained devices such as smartphones, embedded systems, and IoT sensors has become a critical challenge. Lightweight neural network models are specifically designed to reduce computational cost, memory footprint, and energy consumption while preserving acceptable accuracy. This article provides a comprehensive, exam-focused analysis of the core architectures, mathematical principles, compression techniques, and practical applications of lightweight neural networks for computer science students.

在边缘计算和实时人工智能应用的时代，将大型深度神经网络部署到智能手机、嵌入式系统和物联网传感器等资源受限设备上，已成为一项关键挑战。轻量级神经网络模型专门设计用于降低计算成本、内存占用和能耗，同时保持可接受的精度。本文为计算机科学学生提供关于轻量级神经网络核心架构、数学原理、压缩技术和实际应用的全面且紧扣考点的分析。

---

1. The Motivation: Why Lightweight Models Matter | 动因：为什么轻量级模型至关重要

Traditional convolutional neural networks (CNNs) such as VGG-16 contain over 138 million parameters and require approximately 15.5 GFLOPs (giga floating-point operations) to classify a single 224×224 image. Running such models on a mobile device would exhaust battery life within minutes and exceed available memory. The fundamental motivation behind lightweight models is to enable on-device inference with low latency, enhanced privacy (no data upload required), and offline functionality.

传统的卷积神经网络（CNN），如VGG-16，包含超过1.38亿个参数，对单张224×224图像进行分类需要约15.5 GFLOPs（十亿次浮点运算）。在移动设备上运行此类模型会在几分钟内耗尽电池电量，并超出可用内存。轻量级模型背后的根本动机是实现在设备端的低延迟推理、增强隐私（无需上传数据）以及离线功能。

Key constraints on edge devices include:

边缘设备的关键限制包括：

• Processor clock speed often below 2 GHz, limiting arithmetic throughput.
• RAM typically between 1-8 GB, shared with the operating system and other apps.
• Battery capacity, making energy-efficient computation a priority.
• Thermal dissipation limits preventing sustained high-performance operation.

• 处理器时钟速度通常低于2 GHz，限制了算术吞吐量。
• 内存通常在1-8 GB之间，与操作系统和其他应用程序共享。
• 电池容量，使得能效计算成为优先事项。
• 散热限制，防止持续高性能运行。

---

2. Depthwise Separable Convolutions | 深度可分离卷积

Depthwise separable convolution is the cornerstone of many lightweight architectures, most notably MobileNet. It factorizes a standard convolution into two distinct operations: depthwise convolution and pointwise convolution. In depthwise convolution, a single filter is applied to each input channel independently. Then pointwise convolution, a 1×1 convolution, combines the output channels linearly.

深度可分离卷积是许多轻量级架构的基石，最著名的是MobileNet。它将标准卷积分解为两个独立操作：深度卷积和逐点卷积。在深度卷积中，单个滤波器独立应用于每个输入通道。随后，逐点卷积（即1×1卷积）线性组合输出通道。

Mathematically, a standard convolution with kernel size K × K, C input channels, and C output channels has:

从数学上讲，核大小为K × K、C个输入通道和C个输出通道的标准卷积具有：

Parameters (standard) = C × C × K × K

A depthwise separable convolution instead has:

深度可分离卷积则具有：

Parameters (depthwise) = C × K × K + C × C × 1 × 1

The reduction ratio is therefore:

因此压缩比为：

Ratio = (C × K × K + C²) / (C² × K²) = 1/K² + 1/C

For a typical case with K = 3 and C = 256, the ratio is approximately 1/9 + 1/256 ≈ 0.115, representing nearly a 9× reduction in parameters. The computational cost in floating-point operations (FLOPs) follows a similar reduction, making this factorization exceptionally efficient.

对于K = 3且C = 256的典型情况，该比值约为1/9 + 1/256 ≈ 0.115，表示参数减少近9倍。浮点运算（FLOPs）中的计算成本也有类似减少，使得这种分解极为高效。

---

3. MobileNet Architecture | MobileNet架构

MobileNet, introduced by Google in 2017, applies depthwise separable convolutions throughout the entire network architecture. The base MobileNetV1 structure begins with a standard 3×3 convolution followed by a stack of depthwise separable convolution blocks. Each block consists of a 3×3 depthwise convolution, batch normalization, ReLU activation, then a 1×1 pointwise convolution, followed again by batch normalization and ReLU.

MobileNet由Google在2017年提出，在整个网络架构中应用深度可分离卷积。MobileNetV1的基础结构始于一个标准3×3卷积，随后是一系列深度可分离卷积块。每个块包含一个3×3深度卷积、批归一化、ReLU激活，然后是一个1×1逐点卷积，再次进行批归一化和ReLU。

Two global hyper-parameters control the model size and latency:

两个全局超参数控制模型大小和延迟：

• Width multiplier α: Thins the network uniformly by reducing the number of channels at each layer. With α = 1.0, the standard model is used; α = 0.5 halves the channels, reducing computational cost by roughly 4×.
• Resolution multiplier ρ: Reduces the input image resolution, directly affecting the spatial dimension of feature maps and proportionally reducing FLOPs.

• 宽度乘数α：通过减少每层通道数来均匀缩减网络。α = 1.0时使用标准模型；α = 0.5时将通道减半，计算成本约减少4倍。
• 分辨率乘数ρ：降低输入图像分辨率，直接影响特征图的空间维度，并按比例减少FLOPs。

The total computational cost after applying these multipliers becomes:

应用这些乘数后的总计算成本变为：

FLOPs = α² × ρ² × (base FLOPs)

---

4. MobileNetV2: Inverted Residuals | MobileNetV2：倒残差结构

MobileNetV2 improves upon V1 by introducing inverted residual blocks with linear bottlenecks. The key insight is that the ReLU activation function causes information loss in low-dimensional feature spaces. Therefore, the bottleneck layers use linear activations before the residual connection. An inverted residual block expands the feature map dimension by a factor of t (typically 6), applies depthwise convolution, then projects back to a lower dimension.

MobileNetV2通过引入带线性瓶颈的倒残差块来改进V1。关键洞察是ReLU激活函数在低维特征空间中会导致信息丢失。因此，瓶颈层在残差连接之前使用线性激活。倒残差块将特征图维度扩展t倍（通常为6），应用深度卷积，然后投影回较低维度。

The structure of each block follows this sequence:

每个块的结构顺序如下：

Bottleneck Block: 1×1 expand → 3×3 depthwise → 1×1 project

This design separates the expressive capacity of the network from the computational cost. The expansion phase increases the channel count to allow rich feature extraction, while the projection phase compresses the representation to reduce memory and computation. The residual connection is applied only when the stride is 1 and the number of input channels matches the output channels.

这种设计将网络的表达能力与计算成本分离。扩展阶段增加通道数以允许丰富的特征提取，而投影阶段压缩表示以减少内存和计算。仅当步幅为1且输入通道数与输出通道数匹配时，才应用残差连接。

---

5. SqueezeNet: Fire Modules | SqueezeNet：Fire模块

SqueezeNet achieves a 50× reduction in parameters compared to AlexNet while maintaining comparable accuracy through a clever architectural design. The core building block is the Fire module, which consists of a squeeze layer (1×1 convolutions) followed by an expand layer containing both 1×1 and 3×3 convolutions.

SqueezeNet通过巧妙的设计实现了与AlexNet相当的精度，同时参数减少50倍。其核心构建块是Fire模块，由squeeze层（1×1卷积）和包含1×1与3×3卷积的expand层组成。

The squeeze layer reduces the number of input channels to a smaller set, using a ratio (squeeze ratio) typically around 0.125. For example, with 128 input channels, the squeeze layer outputs only 16 channels. The expand layer then mixes these compressed features through 64 filters of 1×1 and 64 filters of 3×3 convolution. This strategy decreases computation because the expensive 3×3 convolutions operate on far fewer channels.

Squeeze层将输入通道数减少为较小的集合，使用通常约为0.125的压缩比。例如，输入128个通道时，squeeze层仅输出16个通道。expend层随后通过64个1×1滤波器和64个3×3卷积滤波器混合这些压缩特征。该策略降低了计算量，因为昂贵的3×3卷积操作在更少的通道上进行。

Additionally, SqueezeNet employs delayed downsampling—the network intentionally avoids early pooling layers to maintain larger feature maps, which preserves spatial information and improves accuracy without increasing parameters.

此外，SqueezeNet采用延迟下采样——网络有意避免早期池化层以保持更大的特征图，从而保留空间信息并提高精度，而不会增加参数。

---

6. ShuffleNet: Channel Shuffle | ShuffleNet：通道混洗

ShuffleNet introduces two significant innovations: pointwise group convolution and channel shuffle. In group convolution, input channels are divided into multiple groups, and convolutions operate independently within each group, significantly reducing computation. However, if multiple group convolution layers are stacked, information cannot flow between groups—a problem known as the “group bottleneck.”

ShuffleNet引入两项重要创新：逐点分组卷积和通道混洗。在分组卷积中，输入通道被分为多个组，卷积在各组内独立操作，显著减少计算量。然而，如果堆叠多个分组卷积层，信息无法在组间流动——这称为”分组瓶颈”问题。

The channel shuffle operation solves this by randomly permuting the channel order after each group convolution. This allows cross-group information exchange at negligible computational cost (only tensor reshaping and transposition are required). The mathematical operation is described as follows: given input channels divided into g groups with each group having n channels, reshape the output to (g, n), transpose to (n, g), and flatten back to n × g channels.

通道混洗操作通过在每次分组卷积后随机排列通道顺序来解决此问题。这允许以可忽略的计算成本（仅需张量重塑和转置）进行跨组信息交换。该数学操作描述如下：假设输入通道分为g组，每组有n个通道，将输出重塑为(g, n)，转置为(n, g)，然后平铺回n × g个通道。

ShuffleNet units also integrate depthwise convolutions and residual connections, enabling the construction of very efficient networks. The computational savings are most pronounced when using a large number of groups (e.g., g = 8), reducing FLOPs by up to 13× compared to standard convolutions.

ShuffleNet单元还集成了深度卷积和残差连接，使得构建非常高效的网络成为可能。使用大量分组（例如g = 8）时，计算节省最为显著，与标准卷积相比，FLOPs减少高达13倍。

---

7. EfficientNet: Compound Scaling | EfficientNet：复合缩放

EfficientNet approaches model efficiency from a different angle—neural architecture search (NAS) and systematic scaling. The baseline network was discovered through NAS, optimizing both accuracy and FLOPs. More importantly, EfficientNet introduces a compound scaling method that uniformly scales all three dimensions: depth (number of layers), width (number of channels), and resolution (input image size).

EfficientNet从不同角度处理模型效率——神经架构搜索（NAS）和系统缩放。基线网络通过NAS发现，同时优化精度和FLOPs。更重要的是，EfficientNet引入了复合缩放方法，统一缩放所有三个维度：深度（层数）、宽度（通道数）和分辨率（输入图像大小）。

If we wish to target a computational budget of 2^φ times the baseline, then the scaling coefficients are determined by:

如果我们要以2^φ倍的基线计算预算为目标，则缩放系数由下式确定：

Depth: α^φ, Width: β^φ, Resolution: γ^φ, where α·β²·γ² ≈ 2

Here, β² and γ² account for the fact that FLOPs scale quadratically with channel width and spatial resolution, respectively, while α scales linearly with depth. The predefined constants α = 1.2, β = 1.1, γ = 1.15 were determined empirically. This systematic scaling allows EfficientNet to achieve superior accuracy per FLOP compared to both manually designed and NAS-searched networks.

其中β²和γ²反映FLOPs分别随通道宽度和空间分辨率二次方缩放，而α随深度线性缩放。预定义的常数α = 1.2、β = 1.1、γ = 1.15通过经验确定。这种系统缩放使EfficientNet在每FLOP精度上优于手工设计和NAS搜索的网络。

---

8. Model Compression Techniques | 模型压缩技术

Beyond architectural innovations, several post-training compression techniques can further reduce model size:

除了架构创新外，几种训练后压缩技术可以进一步减小模型大小：

8.1 Pruning | 剪枝

Pruning removes redundant weights or neurons that contribute minimally to the final prediction. Unstructured pruning zeroes out individual weights, creating sparse matrices that require specialized hardware for speedup. Structured pruning, conversely, removes entire channels or filters, enabling straightforward deployment on standard hardware. A common criterion for importance is the L1 or L2 norm of the weight tensor for each filter.

剪枝移除对最终预测贡献最小的冗余权重或神经元。非结构化剪枝将个权重置零，产生需要专用硬件才能加速的稀疏矩阵。相比之下，结构化剪枝移除整个通道或滤波器，能够在标准硬件上直接部署。常用的重要性标准是每个滤波器权重张量的L1或L2范数。

8.2 Quantization | 量化

Quantization reduces the numerical precision of weights and activations from 32-bit floating-point (FP32) to lower bit-widths such as INT8 or even binary values. Post-training quantization requires no retraining and can reduce model size by 4× while introducing minimal accuracy loss. Quantization-aware training simulates quantization during training, allowing the model to adapt, often achieving near-FP32 accuracy. The memory savings from INT8 quantization are: Memory = Parameters × 1 byte.

量化将权重和激活的数值精度从32位浮点（FP32）降低到更低的位宽，如INT8甚至二值。训练后量化无需重新训练，可将模型大小减少4倍，同时引入最小精度损失。量化感知训练在训练期间模拟量化，使模型自适应，通常可获得接近FP32的精度。INT8量化的内存节省为：内存 = 参数数量 × 1字节。

8.3 Knowledge Distillation | 知识蒸馏

Knowledge distillation transfers knowledge from a large, accurate teacher model to a small student model. The student is trained not only on the ground-truth labels but also on the soft output probabilities (logits) of the teacher. The distillation loss combines cross-entropy with the true labels and Kullback-Leibler (KL) divergence with the teacher’s softened probabilities:

知识蒸馏将知识从大型、精准的教师模型转移到小型学生模型。学生模型不仅基于真实标签训练，还基于教师的软输出概率（logits）训练。蒸馏损失结合了与真实标签的交叉熵以及与教师软化概率的Kullback-Leibler（KL）散度：

L = α·CE(y, σ(z_s)) + β·KL(σ(z_t/T), σ(z_s/T))

Here T is a temperature parameter that softens the probability distribution, revealing inter-class relationships that hard labels cannot encode. Typical values are T = 4-8 and α = 0.7-0.9.

其中T是温度参数，用于软化概率分布，揭示硬标签无法编码的类别间关系。典型值为T = 4-8，α = 0.7-0.9。

---

9. Comparative Analysis | 对比分析

The following table compares the key characteristics of major lightweight architectures discussed above:

以下表格比较了上述主要轻量级架构的关键特征：

Model | 模型	Core Technique | 核心技术	Parameters (approx.) | 参数量（约）	Accuracy (ImageNet) | 精度 (ImageNet)
MobileNetV1	Depthwise separable conv | 深度可分离卷积	4.2M	70.6% Top-1
MobileNetV2	Inverted residuals + linear bottlenecks | 倒残差+线性瓶颈	3.4M	72.0% Top-1
SqueezeNet	Fire modules | Fire模块	1.2M	57.5% Top-1
ShuffleNetV2	Channel shuffle + group conv | 通道混洗+分组卷积	2.3M	69.4% Top-1
EfficientNet-B0	Compound scaling + NAS | 复合缩放+NAS	5.3M	77.1% Top-1

A critical observation is that higher parameter counts do not always correlate with higher accuracy—EfficientNet demonstrates that efficient architecture design can outperform larger models. For exam purposes, understanding the architectural trade-offs is more important than memorizing specific numbers.

一个关键观察是，更高的参数数量并不总是与更高的精度相关——EfficientNet证明高效的架构设计可以超越更大的模型。就考试而言，理解架构的权衡比记忆具体数字更为重要。

---

10. Hardware-Aware Considerations | 硬件感知考量

Real-world deployment requires considering not only theoretical FLOPs but also memory access patterns and hardware characteristics. ShuffleNetV2, for instance, identified four practical guidelines for efficient network design through careful hardware benchmarking:

实际部署不仅需要考虑理论FLOPs，还需要考虑内存访问模式和硬件特性。例如，ShuffleNetV2通过仔细的硬件基准测试确定了高效网络设计的四个实用准则：

• Equal channel width minimizes memory access cost (MAC).
• Excessive group convolution increases MAC and should be avoided when complexity is high.
• Network fragmentation (parallel branches) reduces parallelism and is inefficient on GPUs.
• Element-wise operations (ReLU, addition, etc.) consume memory bandwidth and should be minimized.

• 相等的通道宽度使内存访问成本（MAC）最小化。
• 过多的分组卷积增加了MAC，在计算复杂度高时应避免。
• 网络碎片化（并行分支）降低了并行度，在GPU上效率低下。
• 逐元素操作（ReLU、加法等）消耗内存带宽，应尽量减少。

This highlights an important principle: actual inference time depends on the target hardware. A model optimized for a GPU may not be optimal for an ARM-based mobile processor. Thus, students should understand that engineering context is integral to model design.

这强调了一个重要原则：实际推理时间取决于目标硬件。为GPU优化的模型可能对基于ARM的移动处理器并非最佳。因此，学生应理解工程环境是模型设计不可分割的一部分。

---

11. Applications and Exam Relevance | 应用与考试相关性

Lightweight neural networks have found widespread adoption across numerous real-world applications:

轻量级神经网络已在众多现实应用中广泛采用：

• On-device image classification and object detection in smartphones and smart cameras.
• Real-time pose estimation for fitness tracking and augmented reality applications.
• Keyword spotting and wake-word detection for voice assistants.
• Autonomous drones and robots requiring real-time decision-making.
• Medical imaging analysis on portable diagnostic devices in remote areas.

• 智能手机和智能相机中的设备端图像分类与目标检测。
• 用于健身追踪和增强现实应用的实时姿态估计。
• 语音助手的关键词识别和唤醒词检测。
• 需要实时决策的自主无人机和机器人。
• 偏远地区便携式诊断设备上的医学影像分析。

For computer science examinations, students should be prepared to compare the FLOPs of standard versus depthwise separable convolutions, explain how bottleneck structures reduce computation, describe quantization and pruning algorithms, and analyze the accuracy-efficiency trade-off in different architectures. Numerical problems involving parameter counting and FLOP calculations are extremely common in A-Level examination papers.

对于计算机科学考试，学生应准备好比较标准卷积与深度可分离卷积的FLOPs，解释瓶颈结构如何减少计算，描述量化和剪枝算法，并分析不同架构中的精度-效率权衡。涉及参数计数和FLOP计算的数值问题在A-Level试卷中极为常见。

---

Published by TutorHao | Computer Science Revision Series | aleveler.com

更多咨询请联系16621398022（同微信）

📚 JavaScript Programming Techniques and Cutting-Edge Applications | JavaScript编程技巧与前沿应用

JavaScript has evolved from a simple scripting language for web pages to a versatile, high-level programming language powering everything from interactive frontends to scalable backends and even machine learning. This article explores key programming techniques and the newest frontiers of JavaScript, aligned with the Cambridge A-Level and IGCSE Computer Science curricula.

JavaScript 已从一种用于网页的简单脚本语言，演变为一种多功能、高级别的编程语言，驱动着从交互式前段到可扩展后端乃至机器学习的方方面面。本文围绕剑桥 A-Level 与 IGCSE 计算机科学考纲，探讨 JavaScript 的核心编程技巧与最前沿的应用方向。

---

1. Variable Declaration and Scope | 变量声明与作用域

Modern JavaScript offers three ways to declare variables: ‘var’, ‘let’, and ‘const’. Understanding their differences is essential for writing predictable and bug-free code. ‘var’ is function-scoped, meaning it is accessible throughout the entire function regardless of block boundaries. ‘let’ and ‘const’ are block-scoped, which keeps variables contained within the nearest pair of curly braces.

现代 JavaScript 提供三种声明变量的方式：var、let 和 const。理解它们的差异对于编写可预测且无 bug 的代码至关重要。var 具有函数作用域，即在整个函数内均可访问，而不受块级边界限制；let 与 const 具有块级作用域，使变量只能被限制在最近的一对花括号之内。

Variables declared with ‘const’ cannot be reassigned, which is a strong safeguard against accidental value changes. However, ‘const’ does not make the value immutable; for objects and arrays, their contents can still be mutated. The following table summarises the key differences:

使用 const 声明的变量不能被重新赋值，这是防止意外值变更的有效保障。然而，const 并不会使值不可变；对于对象和数组，其内容仍然可以被修改。下表总结了它们的主要区别：

Keyword	Scope	Reassignable	Hoisting
var	Function	Yes	Initialised as undefined
let	Block	Yes	Temporal dead zone
const	Block	No	Temporal dead zone

For A-Level candidates, it is crucial to recognise that variables declared with ‘let’ and ‘const’ exist in the temporal dead zone from the start of the block until the declaration is encountered, meaning they cannot be accessed before declaration.

对于 A-Level 考生来说，需特别留意：使用 let 与 const 声明的变量，在块起始到声明语句之间处于“暂时性死区”，在声明前无法被访问。

---

2. Arrow Functions and Lexical ‘this’ | 箭头函数与词法 this

Arrow functions provide a concise syntax for writing functions. Compared to traditional function expressions, they are shorter and more readable. For example:

箭头函数提供了一种简洁的函数书写语法。相较于传统的函数表达式，箭头函数更短小、更具可读性。例如：

const square = x → x × x;

Beyond syntax, the most important distinction is how ‘this’ is bound. In a regular function, ‘this’ is determined by how the function is called, which can lead to unexpected results in event handlers or callbacks. Arrow functions do not have their own ‘this’; instead, they inherit ‘this’ from the surrounding lexical scope.

除了语法差异外，最重要的区别在于 this 的绑定方式。在普通函数中，this 取决于函数的调用方式，在事件处理或回调中容易导致意外结果；而箭头函数没有自己的 this，它会从外层词法作用域继承 this。

In exam contexts, students often need to trace the value of ‘this’. A typical question might involve an object method that uses ‘setTimeout’. With a regular function, ‘this’ would refer to the global object, whereas with an arrow function, it correctly refers to the containing object.

在考试中，学生常需要追踪 this 的取值。典型问题可能涉及对象方法中使用 setTimeout。若使用普通函数，this 将指向全局对象；若使用箭头函数，this 将正确地指向所在的对象。

---

3. Destructuring and Template Literals | 解构赋值与模板字符串

Destructuring allows extracting values from arrays or properties from objects directly into individual variables. This technique reduces repetitive code and enhances readability. For arrays, position determines assignment; for objects, property names are used.

解构赋值允许从数组中提取值，或从对象中提取属性，并直接赋值给独立的变量。该技巧减少了重复代码并提高了可读性。对数组而言，按位置赋值；对对象而言，按属性名赋值。

const [a, b] = [10, 20];
const {name, age} = person;

Template literals, introduced in ES6, use backticks instead of quotes and allow embedded expressions via the ‘${…}’ syntax. This is especially useful when constructing complex strings that include variables, results of function calls, or even multi-line text.

ES6 引入的模板字符串使用反引号代替引号，允许通过 ${…} 语法嵌入表达式。这在构建包含变量、函数调用结果甚至多行文本的复杂字符串时特别有用。

In addition to basic interpolation, tagged templates allow a function to receive the string literals and interpolated values separately, enabling powerful custom string processing. This is an advanced but increasingly common interview topic for computer science students.

除基础插值外，带标签的模板允许函数分别接收字符串片段与插值，从而实现强大的自定义字符串处理。这是进阶但在计算机科学学生面试中愈发常见的主题。

---

4. Asynchronous JavaScript and ‘async/await’ | 异步 JavaScript 与 async/await

JavaScript is single-threaded, so it handles time-consuming tasks such as network requests or file reading through asynchronous callbacks. The original callback style can lead to deeply nested code, often called “callback hell.” The introduction of Promises and later ‘async/await’ has made asynchronous code far more manageable.

JavaScript 是单线程的，因此它通过异步回调来处理网络请求或文件读取等耗时任务。传统回调风格会导致深层嵌套的代码，常称为“回调地狱”。Promise 以及后来 async/await 的引入使异步代码变得易于管理。

A Promise is an object representing the eventual completion or failure of an asynchronous operation. It has three states: pending, fulfilled, and rejected. The ‘then()’ and ‘catch()’ methods allow chaining success and error handlers respectively.

Promise 是表示异步操作最终完成或失败的对象。它有三种状态：待定、已履行、已拒绝。then() 与 catch() 方法分别用于链式处理成功与错误。

fetch(‘/api/data’)
  .then(response → response.json())
  .then(data → console.log(data))
  .catch(error → console.error(error));

With ‘async/await’, code reads like synchronous code, making it easier to debug and maintain. An ‘async’ function always returns a Promise, and the ‘await’ keyword pauses execution until the Promise is settled. Proper error handling requires ‘try…catch’ blocks.

使用 async/await 后，代码读起来就像同步代码，更易于调试与维护。async 函数始终返回 Promise，await 关键字会暂停执行，直到 Promise 被解决。正确的错误处理需要 try…catch 块。

---

5. DOM Manipulation and Event Handling | DOM 操作与事件处理

The Document Object Model (DOM) represents the page structure as a tree of nodes. JavaScript can traverse, modify, and style these nodes dynamically. The modern ‘querySelector’ and ‘querySelectorAll’ methods provide flexibility and consistency in selecting elements.

文档对象模型（DOM）将页面结构表示为节点树。JavaScript 可以动态地遍历、修改这些节点并设置样式。现代 querySelector 与 querySelectorAll 方法在选取元素上提供了灵活性与一致性。

Event handling is central to interactive web pages. The ‘addEventListener’ method attaches a function to an event, allowing multiple listeners for the same event and the ability to remove them. Event delegation is an important technique: instead of binding listeners to each child, a single listener on a parent manages events for all children using the ‘target’ property.

事件处理是交互式网页的核心。addEventListener 方法将函数绑定到事件上，允许同一事件拥有多个监听器并可将它们移除。事件委托是一项重要技术：不在每个子元素上绑定监听器，而是在父元素上绑定一个监听器，通过 target 属性来管理所有子元素的事件。

For examination purposes, students should understand the event propagation phases: capturing, targeting, and bubbling. The bubbling phase is the most commonly used, where an event travels from the target element up to the root of the DOM tree.

备考时，学生应理解事件传播的三个阶段：捕获阶段、目标阶段与冒泡阶段。冒泡阶段最为常用，事件会从目标元素一路向上传播至 DOM 树的根节点。

---

6. Module Systems: ‘require’ vs ‘import’ | 模块系统：require 与 import 对比

As projects grow, splitting code into separate modules becomes essential. Two major module systems are CommonJS, primarily used in Node.js, and ES Modules, the official standard in modern browsers. CommonJS uses ‘require’ and ‘module.exports’, while ES Modules use ‘import’ and ‘export’.

随着项目发展，将代码拆分为独立模块变得至关重要。两种主要的模块系统为：主要用于 Node.js 的 CommonJS，以及现代浏览器的官方标准 ES Modules。CommonJS 使用 require 与 module.exports，而 ES Modules 使用 import 与 export。

Aspect	CommonJS (require)	ES Modules (import)
Loading	Synchronous	Asynchronous
Syntax	require(‘./m’)	import { x } from ‘./m’
Exports	module.exports = {}	export default / export
Static analysis	No	Yes

ES Modules support static analysis, meaning the dependency structure is known at parse time, enabling tree-shaking to remove unused code. This makes them preferable for front-end projects where bundle size matters.

ES Modules 支持静态分析，意味着依赖结构可在解析阶段得知，从而支持 tree-shaking 以移除未使用代码。这使得它们更适用于对打包体积敏感的前端项目。

---

7. Front-End Frameworks and Virtual DOM | 前端框架与虚拟 DOM

Frameworks like React, Vue, and Angular have transformed how user interfaces are built. They encourage a component-based architecture, where the UI is broken into isolated, reusable pieces. This directly improves code maintainability and testing.

React、Vue 和 Angular 等框架已经改变了用户界面的构建方式。它们提倡基于组件的架构，将 UI 切分为独立、可复用的部分。这直接提升了代码的可维护性与可测试性。

React’s virtual DOM is a lightweight in-memory representation of the actual DOM. When state changes, the framework computes the difference (diffing) between the previous and new virtual DOM and applies only the necessary updates to the real DOM. This minimises expensive browser reflows and repaints.

React 的虚拟 DOM 是真实 DOM 的一种轻量级内存表示。当状态改变时，框架计算新旧虚拟 DOM 之间的差异（diffing），并仅将必要的更新应用到真实 DOM 上。这最大限度地减少了浏览器昂贵的回流与重绘。

Hooks, such as ‘useState’ and ‘useEffect’, allow function components to manage state and side effects. For A-Level computer science, understanding the concept of one-way data flow and immutability in React is a frequent assessment point.

useState 与 useEffect 等 Hooks 允许函数组件管理状态与副作用。对于 A-Level 计算机科学，理解单向数据流与 React 中的不可变性是常见的考查点。

---

8. Node.js and Back-End Development | Node.js 与后端开发

Node.js brings JavaScript to the server, leveraging the V8 engine. Its non-blocking, event-driven I/O model makes it highly efficient for I/O-intensive applications. Unlike traditional multi-threaded servers, Node.js handles many concurrent connections using a small number of threads.

Node.js 借助 V8 引擎将 JavaScript 带到服务端。其非阻塞、事件驱动的 I/O 模型使其对 I/O 密集型应用极为高效。不同于传统的多线程服务器，Node.js 使用少量线程处理大量并发连接。

The core of Node.js includes the ‘events’ module and the ‘streams’ API. Streams allow data to be processed piece by piece, which reduces memory usage when handling large files. This is an important concept for understanding how Node.js scales.

Node.js 的核心包括 events 模块和 streams API。流允许数据逐块处理，从而在操作大文件时降低内存占用。这是理解 Node.js 如何进行扩展的重要概念。

The npm ecosystem is the largest package repository in the world. ‘express’ is a minimalist web framework; ‘socket.io’ enables real-time bidirectional communication. These libraries illustrate the power of open-source collaboration in modern software development.

npm 生态是全球最大的软件包仓库。express 是极简的 Web 框架；socket.io 支持实时的双向通信。这些库体现了现代软件开发中开源协作的力量。

---

9. JavaScript in Machine Learning | JavaScript 在机器学习中的应用

One of the most exciting frontiers is running machine learning models directly in the browser or in Node.js. TensorFlow.js is a popular library that provides a flexible API for building and training models in JavaScript. It supports both CPU and WebGL acceleration.

最激动人心的前沿之一是在浏览器或 Node.js 中直接运行机器学习模型。TensorFlow.js 是一个流行的库，提供灵活的 API 用于在 JavaScript 中构建和训练模型，并支持 CPU 与 WebGL 加速。

Transfer learning is particularly practical: a pre-trained model, such as MobileNet for image classification, can be loaded and fine-tuned on a small amount of custom data. This demonstrates the intersection of web development and artificial intelligence.

迁移学习尤其实用：可以加载一个预训练模型（例如用于图像分类的 MobileNet），并在少量自定义数据上微调。这展示了 Web 开发与人工智能的交叉融合。

For students, experimenting with simple classifiers using ‘ml5.js’ or ‘brain.js’ is an accessible entry point. These libraries abstract away complex mathematics while still demonstrating core concepts such as feature extraction and prediction confidence.

对学生而言，使用 ml5.js 或 brain.js 进行简单分类器实验是入门的好方式。这些库抽象了复杂的数学知识，同时仍然演示特征提取与预测置信度等核心概念。

---

10. Performance Optimisation Techniques | 性能优化技巧

Writing fast JavaScript requires deliberate strategies. Debouncing and throttling are two techniques that limit how often a function executes in response to frequent events like scrolling or typing. Debouncing waits for a pause, while throttling ensures execution at most once per interval.

编写高效的 JavaScript 需要刻意使用策略。防抖（debouncing）与节流（throttling）是两种限制高频率事件（如滚动或输入）触发函数执行次数的方法。防抖等待操作暂停后执行，而节流保证在固定时间间隔内最多执行一次。

Code splitting, lazy loading, and memoisation are other powerful tools. Memoisation caches the results of expensive function calls, so repeated calls with the same arguments return instantly instead of recomputing.

代码分割、懒加载与记忆化（memoisation）也是强大的工具。记忆化会缓存昂贵函数调用的结果，因此对相同参数的重复调用可立即返回，而不必重新计算。

function memoisedFactorial(n, cache = {}) {
  if (n in cache) return cache[n];
  if (n ≤ 1) return 1;
  cache[n] = n × memoisedFactorial(n-1, cache);
  return cache[n];
}

Understanding the time complexity of algorithms and the runtime behaviour of the event loop is crucial for answering questions about performance in final exams.

理解算法的时间复杂度以及事件循环的运行时行为，对于回答期末考试中有关性能的问题至关重要。

---

11. Security Considerations in JavaScript | JavaScript 安全注意事项

Web applications face several security threats, and JavaScript developers must be aware of them. Cross-Site Scripting (XSS) occurs when malicious scripts are injected into web pages viewed by other users. This can be prevented by escaping user input and using Content Security Policy (CSP) headers.

Web 应用面临多种安全威胁，JavaScript 开发者必须了解它们。跨站脚本攻击（XSS）发生在恶意脚本被注入到其他用户浏览的网页中时。这可以通过对用户输入进行转义以及使用内容安全策略（CSP）头来防范。

Cross-Site Request Forgery (CSRF) tricks authenticated users into submitting unwanted requests. Countermeasures include anti-CSRF tokens and the ‘SameSite’ cookie attribute. The principle of least privilege is also essential: only access the data and capabilities that are strictly necessary.

跨站请求伪造（CSRF）诱骗已认证用户提交非自愿的请求。对策包括反 CSRF 令牌以及 SameSite Cookie 属性。最小权限原则同样至关重要：只访问绝对必要的资源与功能。

For secure communication, HTTPS should always be used in production, and all sensitive data should be handled on the server side rather than in client-side JavaScript. A-Level students should know how to identify and mitigate basic OWASP vulnerabilities in code snippets.

为保障安全通信，生产环境应始终使用 HTTPS，所有敏感数据都应在服务端处理，而非在客户端 JavaScript 中。A-Level 学生应能识别代码片段中的基本 OWASP 漏洞并采取措施修复。

---

12. Future Trends and Emerging Technologies | 未来趋势与新兴技术

WebAssembly (Wasm) allows code written in languages like C, C++, and Rust to run in the browser at near-native speed. JavaScript and Wasm complement each other: heavy computations are delegated to Wasm, while JavaScript handles the dynamic logic and DOM interaction.

WebAssembly（Wasm）允许用 C、C++ 和 Rust 等语言编写的代码在浏览器中以接近原生的速度运行。JavaScript 与 Wasm 互补：重型计算交给 Wasm，而 JavaScript 处理动态逻辑与 DOM 交互。

TypeScript, a superset of JavaScript with static types, has become a dominant force in enterprise development. It catches type-related errors at compile time, improving reliability. Many interfaces and libraries are now written directly in TypeScript.

TypeScript 是 JavaScript 的超集，增加了静态类型，已成为企业开发中的主导力量。它在编译阶段捕获与类型相关的错误，提升了可靠性。如今许多接口与库直接用 TypeScript 编写。

Serverless architectures and edge computing further expand JavaScript’s reach. Functions-as-a-Service, such as AWS Lambda with Node.js, allow developers to focus on logic rather than infrastructure. These trends indicate that JavaScript will remain a vital and evolving language.

无服务器架构与边缘计算进一步拓展了 JavaScript 的应用范围。以 Node.js 支持的函数即服务（如 AWS Lambda）让开发者能够专注于逻辑而非基础设施。这些趋势表明 JavaScript 将始终是一门充满活力且持续演进的语言。

---

Published by TutorHao | Computer Science Revision Series | aleveler.com

更多咨询请联系16621398022（同微信）

📚 TOEFL Junior: Complete Test Format Analysis & Preparation Strategies | 小托福考试题型全解析与备考策略

The TOEFL Junior Standard test is an English proficiency assessment designed by ETS for students ages 11-15. It measures academic and social English skills in three core sections: Listening Comprehension, Language Form and Meaning, and Reading Comprehension. Understanding the exact test format is the first step toward a high score.

小托福标准考试是由 ETS 为 11-15 岁学生设计的英语能力测评，考查学术与社交场景中的英语技能，包含听力理解、语言形式与含义、阅读理解三大核心部分。准确把握考试题型是取得高分的第一步。

1. Test Structure Overview | 考试结构概览

The TOEFL Junior Standard test lasts about 110 minutes and consists of 126 multiple-choice questions. Each section is scored from 200 to 300, making a total score range of 600 to 900. The test uses American English and reflects both everyday school life and academic scenarios.

小托福标准考试时长约 110 分钟，共 126 道选择题。每个部分计分为 200-300 分，总分范围为 600-900 分。考试采用美式英语，内容贴近校园日常与学术场景。

Section	Questions	Time	Score Range
Listening Comprehension	42	40 min	200-300
Language Form and Meaning	42	25 min	200-300
Reading Comprehension	42	50 min	200-300

The table above provides a quick overview. Note that the total score is the sum of the three section scores, so no section should be ignored during preparation.

上表为整体概览。注意总分为三个部分分数之和，因此备考时任何一部分都不能忽视。

---

2. Listening Comprehension: Question Types and Focus | 听力理解：题型分类与考查重点

The listening section tests how well students understand classroom instructions, peer conversations, and short academic talks. It requires attention to both explicit details and implied meaning. Audio is played only once, so note-taking during the conversation is essential.

听力部分考查学生对课堂指令、同学对话和简短学术演讲的理解，既需要听懂细节，也要把握隐含含义。音频通常只播放一遍，因此边听边记笔记非常重要。

There are three main question categories: classroom instruction comprehension, communications in social or school settings, and academic lectures or discussions. Questions often ask about the purpose of a statement, the relationship between speakers, or the main idea of a talk.

主要包含三类题型：一是课堂指令理解，二是校园或社交场景对话，三是学术讲座或讨论。题目常考查说话目的、人物关系或演讲主旨。

Common question stems include: What is the teacher asking the student to do? Why does the student visit the teacher? What can be inferred from the conversation? To answer well, students should follow the logic of the recording rather than just matching keywords.

常见题干有：老师要求学生做什么？学生为什么去找老师？从对话中可以推断出什么？要答好这些题，需要跟随听力材料的逻辑，而不是单纯匹配个别关键词。

Students must also listen for signal words such as however, because, and first of all. These words help listeners organize information and predict the direction of the conversation.

考生还应注意 however、because、first of all 等信号词。这些词能帮助听者组织信息、预判对话走向。

---

3. Language Form and Meaning: A Combined Challenge of Grammar and Vocabulary | 语言形式与含义：语法与词汇的混合挑战

This section presents short passages such as emails, notices, diaries, and stories. Each passage has several blanks, and students must select the best word or phrase to fill each blank. The questions test grammar accuracy, word choice, and awareness of language patterns.

该部分给出邮件、通知、日记、故事等短文，每个短文中有若干空格，考生需选出最佳词语填入。题目考查语法准确性、词汇选择和语言模式意识。

Grammar items often cover verb tenses, subject-verb agreement, pronouns, conjunctions, prepositions, and sentence structure. Vocabulary items focus on choosing the correct word meaning from context, understanding collocations, and distinguishing similar words.

语法考点常涉及时态、主谓一致、代词、连词、介词与句子结构。词汇考点侧重根据上下文选择恰当词义、理解固定搭配并区分近义词。

Because the time limit is only 25 minutes, students cannot rewrite every sentence. A strong strategy is to read the whole passage briefly to grasp its theme, then examine the grammatical environment around each blank before choosing the answer.

由于该部分只有 25 分钟，不可能反复重读句子。有效策略是先快速通读全文把握主题，再仔细分析空格周围的语法环境，最后选择答案。

It is also helpful to study common affixes and word families. Knowing that the prefix re- means again or that the suffix -tion creates nouns can help students decode unfamiliar words quickly.

系统学习常见前后缀和词族也很有帮助。知道前缀 re- 表示“再次”、后缀 -tion 构成名词，可以帮助考生快速推断生词含义。

---

4. Reading Comprehension: Multiple Text Types and Question Skills | 阅读理解：多样文本与题型

The reading section includes both non-academic and academic texts. Non-academic passages may include school notices, letters, blogs, and short stories. Academic passages cover science, history, biography, and other subjects, with charts or tables sometimes included.

阅读部分包含非学术与学术两类文本。非学术文章包括校园通知、书信、博客和短篇故事；学术文章涉及科学、历史、人物传记等主题，有时还配有图表。

Question types include main idea, detail, inference, vocabulary-in-context, author purpose, and text organization. Students should practice moving quickly between reading the passage and locating evidence in the text.

题型包括主旨题、细节题、推断题、词义猜测题、作者目的题和篇章结构题。考生应练习在阅读文章与定位原文证据之间快速切换。

Understanding the structure of a passage is just as important as understanding its content. Transition words such as however, therefore, and for example often signal the relationship between ideas and help predict what comes next.

理解文章结构与理解内容同样重要。像 however、therefore、for example 这样的连接词往往提示观点之间的关系，有助于预判下文内容。

To improve speed, students should read the first sentence of each paragraph first to build a mental map of the passage, then answer details by scanning for specific information.

为提升速度，考生可以先读每段首句，建立文章结构图，再通过扫读查找细节信息来答题。

---

5. Score Reporting and Benchmarks | 评分机制与分数解读

Each section of the TOEFL Junior is scored on a scale of 200 to 300. The total score ranges from 600 to 900. Scores are based on the number of correct answers, and there is no penalty for wrong answers, so guessing is always better than leaving a blank.

小托福每个部分按 200-300 分计分，总分范围 600-900 分。成绩基于答对题目数量，答错不扣分，所以猜测作答优于留空不答。

The score report includes a total score, three section scores, and level descriptors. It also connects results to the Common European Framework of Reference (CEFR), which helps families understand the student’s English level in an international context.

成绩单提供总分、三个单项分和水平描述，并将分数对应到欧洲共同语言参考框架（CEFR），帮助家庭从国际视角了解学生的英语水平。

Reading section scores may also be linked to a Lexile measure. This useful tool suggests books at an appropriate difficulty level, allowing students to develop reading skills through graded materials.

阅读成绩还可关联蓝思阅读指数。这一工具能够推荐

Published by TutorHao | English Revision Series | aleveler.com

更多咨询请联系16621398022（同微信）

📚 Plant Structure: Cells, Tissues, and Organs | 植物结构考点：细胞、组织与器官层次

Understanding plant structure is fundamental to advanced biology study. From a single cell to a complex organ, plants exhibit a remarkable hierarchy of organisation. This article systematically reviews the key concepts of plant cells, tissues, and organs, ensuring you are fully prepared for exam questions on this topic.

理解植物结构是进阶生物学学习的基础。从单个细胞到复杂器官，植物展现出卓越的组织层次。本文系统复习植物细胞、组织和器官的关键概念，确保你为考试中相关问题做好充分准备。

---

1. The Plant Cell: Unique Organelles | 植物细胞：特有的细胞器

Plant cells are eukaryotic cells that share many features with animal cells, such as a nucleus, mitochondria, and endoplasmic reticulum. However, they possess three characteristic organelles that distinguish them: the cell wall, chloroplasts, and a large central vacuole.

植物细胞是真核细胞，与动物细胞共享许多特征，如细胞核、线粒体和内质网。然而，它们拥有三个区别性的细胞器：细胞壁、叶绿体和大型中央液泡。

The cell wall is primarily composed of cellulose microfibrils embedded in a matrix of hemicellulose and pectin. It provides structural support, maintains cell shape, and prevents excessive water uptake. Unlike animal cells, plant cells do not lyse in hypotonic solutions because the rigid cell wall resists expansion.

细胞壁主要由嵌入在半纤维素和果胶基质中的纤维素微纤维组成。它提供结构支撑，维持细胞形状，并防止过度吸水。与动物细胞不同，植物细胞在低渗溶液中不会裂解，因为刚性细胞壁抵抗膨胀。

Chloroplasts are the site of photosynthesis. They contain thylakoid membranes arranged in grana, where light-dependent reactions occur, and a stroma, where the Calvin cycle takes place. Each chloroplast is surrounded by a double membrane, and its internal membrane system creates a large surface area for photosynthetic pigments.

叶绿体是光合作用的场所。它们含有排列成基粒的类囊体膜，光依赖反应在此发生；基质中则进行卡尔文循环。每个叶绿体由双层膜包围，其内部膜系统为光合色素提供了巨大表面积。

The central vacuole can occupy up to 90% of a mature plant cell’s volume. It stores water, ions, pigments, and waste products. The vacuole also maintains turgor pressure, which is essential for plant support in non-woody tissues.

中央液泡可占据成熟植物细胞体积的90%。它储存水分、离子、色素和废物。液泡还维持膨压，对非木质组织的植物支撑至关重要。

---

2. Types of Plant Cells | 植物细胞的类型

Although all plant cells share basic features, they differentiate into specialised types based on their function. In exams, you are commonly asked to compare and contrast these cell types using annotated diagrams.

虽然所有植物细胞具有基本共同特征，但它们根据功能分化为特化类型。在考试中，常要求你用标注图表比较和对比这些细胞类型。

Parenchyma cells are the most common type. They have thin primary cell walls and are living at maturity. They function in photosynthesis (as chlorenchyma), storage, and tissue repair. Their simple structure allows them to differentiate into other cell types when required.

薄壁细胞是最常见的类型。它们具有薄的初生细胞壁，成熟时仍是活细胞。其功能包括光合作用（作为绿色薄壁组织）、储存和组织修复。其简单结构使其在需要时可分化成其他细胞类型。

Collenchyma cells have unevenly thickened primary cell walls, rich in pectin. They are living cells that provide flexible support to growing stems and petioles. Because their walls are not lignified, they can stretch as the plant grows.

厚角细胞具有不均匀增厚的初生细胞壁，富含果胶。它们是活细胞，为生长中的茎和叶柄提供柔性支撑。由于细胞壁未木质化，它们能随植物生长而伸展。

Sclerenchyma cells possess thick, lignified secondary cell walls. They are dead at maturity, losing their protoplasts. There are two types: fibres (long, slender cells) and sclereids (shorter, variable-shaped cells). Their primary function is rigid structural support. Lignin makes their walls waterproof and extremely strong.

厚壁细胞具有厚而木质化的次生细胞壁。它们在成熟时死亡，失去原生质体。有两种类型：纤维（长而细长的细胞）和石细胞（较短、形状多样）。其主要功能是刚性结构支撑。木质素使其细胞壁防水且极其坚固。

---

3. Simple Tissues: Coordination of Cells | 简单组织：细胞的协调组合

A tissue is a group of cells with a similar structure and function, working together. Simple tissues consist of one cell type only. In plants, the major simple tissues include the epidermis, parenchyma, collenchyma, and sclerenchyma — which we have already discussed in terms of individual cells.

组织是一组具有相似结构和功能的细胞协同工作。简单组织仅由一种细胞类型组成。在植物中，主要简单组织包括表皮、薄壁组织、厚角组织和厚壁组织——我们已从单个细胞角度进行了讨论。

The epidermis is the outermost layer of the plant body covering leaves, stems, and roots. It is typically one cell thick and covered by a waxy cuticle on aerial parts to reduce water loss. Specialised epidermal structures include stomata (pores flanked by guard cells), trichomes (hair-like projections), and root hairs (extensions of epidermal cells that increase surface area for absorption).

表皮是覆盖叶、茎和根的最外层结构。它通常只有一层细胞厚，在气生部分覆盖有蜡质角质层以减少水分流失。特化的表皮结构包括气孔（由保卫细胞围绕的气孔）、毛状体（毛状突起）和根毛（表皮细胞的延伸，增加吸收表面积）。

In roots, the epidermis is often called the rhizodermis or piliferous layer because it lacks cuticle and bears root hairs. The lack of cuticle allows efficient water and mineral absorption; root hairs increase the root surface area by several orders of magnitude.

在根中，表皮常被称为根表皮或毛根层，因为它缺乏角质层并生有根毛。缺乏角质层有利于高效吸收水分和矿物质；根毛使根的表面积增加数个数量级。

---

4. Complex Tissues: Xylem

Published by TutorHao | Biology Revision Series | aleveler.com

Find Biology Textbooks on eBay UK

New, used and second-hand copies of textbooks and revision guides are often much cheaper than retail — check current listings and prices before you buy.

Browse on eBay UK →

更多咨询请联系16621398022（同微信）

Share this:

• Email
• More

• LinkedIn
• X
• Facebook
• Tumblr
• Reddit
• Pinterest
• Telegram
• WhatsApp

Like this:

Like Loading…

📚 Euler’s Theorem and Its Applications in Problem Solving | 欧拉定理及其在题型中的应用

Euler’s Theorem is a cornerstone of number theory, offering a powerful tool for simplifying modular arithmetic. Named after the Swiss mathematician Leonhard Euler, this theorem generalizes Fermat’s Little Theorem and appears frequently in olympiad-style problems, A-Level further mathematics papers, and competitive examinations. Mastering Euler’s Theorem not only deepens your understanding of modular arithmetic but also unlocks elegant solutions to otherwise tedious calculations.

欧拉定理是数论的基石之一，为简化模运算提供了强大工具。该定理以瑞士数学家莱昂哈德·欧拉的名字命名，是费马小定理的推广，在奥林匹克风格题目、A-Level进阶数学试卷及各类竞赛中频繁出现。掌握欧拉定理不仅能加深你对模运算的理解，更能为原本繁冗的计算提供优雅的解答。

---

1. Statement of Euler’s Theorem | 欧拉定理的陈述

Euler’s Theorem states that if ( n ) is a positive integer and ( a ) is an integer coprime to ( n ) (i.e., gcd(a, n) = 1), then:

欧拉定理指出：若 ( n ) 为正整数，且 ( a ) 与 ( n ) 互质（即 gcd(a, n) = 1），则：

a^φ(n) ≡ 1 (mod n)

Here, φ(n) is Euler’s totient function, which counts the number of positive integers up to n that are coprime to n.

其中 φ(n) 为欧拉函数，表示不超过 n 且与 n 互质的正整数的个数。

Example: Let n = 10. The numbers coprime to 10 are 1, 3, 7, 9, so φ(10) = 4. For a = 3, we have 3⁴ = 81 ≡ 1 (mod 10). Indeed, 81 ≡ 1 (mod 10).

示例：取 n = 10，与 10 互质的数为 1、3、7、9，故 φ(10) = 4。取 a = 3，则 3⁴ = 81 ≡ 1 (mod 10)，验证成立。

---

2. Computing Euler’s Totient Function φ(n) | 欧拉函数 φ(n) 的计算

To apply Euler’s Theorem, you must first compute φ(n). The formula depends on the prime factorization of n.

要应用欧拉定理，首先需要计算 φ(n)。其计算公式依赖于 n 的质因数分解。

Formula: If n = p₁ᵏ¹ × p₂ᵏ² × … × pᵣᵏʳ, then

公式：若 n = p₁ᵏ¹ × p₂ᵏ² × … × pᵣᵏʳ，则

φ(n) = n × (1 − 1/p₁) × (1 − 1/p₂) × … × (1 − 1/pᵣ)

Example 1: Compute φ(12). Since 12 = 2² × 3,

例 1：计算 φ(12)。由于 12 = 2² × 3，

φ(12) = 12 × (1 − 1/2) × (1 − 1/3) = 12 × 1/2 × 2/3 = 4

Indeed, the numbers coprime to 12 are 1, 5, 7, 11.

确实，与 12 互质的数为 1、5、7、11。

Example 2: Compute φ(100). Since 100 = 2² × 5²,

例 2：计算 φ(100)。由于 100 = 2² × 5²，

φ(100) = 100 × (1 − 1/2) × (1 − 1/5) = 100 × 1/2 × 4/5 = 40

For a prime p, φ(p) = p − 1. This special case recovers Fermat’s Little Theorem.

对于素数 p，φ(p) = p − 1。这种特殊情况正是费马小定理。

---

3. Proof Strategy of Euler’s Theorem | 欧拉定理的证明思路

The classic proof uses the concept of reduced residue systems. Let r₁, r₂, …, r_φ(n) be the positive integers less than n and coprime to n. Multiply each by a (where gcd(a, n) = 1). The resulting set ar₁, ar₂, …, ar_φ(n) is again a reduced residue system modulo n.

经典证明利用既约剩余系的概念。设 r₁, r₂, …, r_φ(n) 是小于 n 且与 n 互质的全部正整数。将每个数乘以 a（其中 gcd(a, n) = 1），得到的集合 ar₁, ar₂, …, ar_φ(n) 仍然是模 n 的既约剩余系。

Thus, the product of the original set is congruent to the product of the multiplied set:

因此，原集合的乘积与乘以 a 后的集合的乘积同余：

r₁ × r₂ × … × r_φ(n) ≡ a^φ(n) × r₁ × r₂ × … × r_φ(n) (mod n)

Since each rᵢ is invertible modulo n, we cancel the common product to obtain a^φ(n) ≡ 1 (mod n). This proof is elegant and illustrates why the coprimality condition is essential.

由于每个 rᵢ 在模 n 下均可逆，约去公共乘积后即得 a^φ(n) ≡ 1 (mod n)。此证明简洁优美，也说明了互质条件为何必要。

Key insight: The theorem only applies when gcd(a, n) = 1. If they share a common factor, a^φ(n) ≡ 1 will not hold.

关键要点：该定理仅在 gcd(a, n) = 1 时成立。若 a 与 n 有公因子，则 a^φ(n) ≡ 1 不成立。

---

4. Euler’s Theorem vs. Fermat’s Little Theorem | 欧拉定理与费马小定理的对比

Fermat’s Little Theorem states that for a prime p and any integer a not divisible by p, a^p−1 ≡ 1 (mod p). This is a direct corollary of Euler’s Theorem because φ(p) = p − 1.

费马小定理指出：对于素数 p 及不被 p 整除的整数 a，有 a^p−1 ≡ 1 (mod p)。这是欧拉定理的直接推论，因为 φ(p) = p − 1。

Theorem	Condition	Result
Fermat’s Little Theorem	n = p (prime), p ∤ a	ap−1 ≡ 1 (mod p)
Euler’s Theorem	gcd(a, n) = 1	aφ(n) ≡ 1 (mod n)

In practice, when the modulus is composite, Euler’s Theorem is the appropriate tool; when the modulus is prime, both theorems work, but Fermat’s is often quicker.

在实际问题中，当模数为合数时，应使用欧拉定理；当模数为素数时，两个定理均适用，但费马小定理通常更快捷。

---

5. Application 1: Simplifying Large Exponents | 应用一：化简大指数幂

One of the most common exam problems asks you to compute a^b mod n for very large b. Euler’s Theorem reduces the exponent modulo φ(n).

最常见的考题之一是计算 a^b mod n，其中 b 非常大。欧拉定理可将指数缩小为模 φ(n) 的余数。

Example: Compute 7²²² mod 10.

例题：计算 7²²² mod 10。

Step 1: φ(10) = 4.

第一步：φ(10) = 4。

Step 2: Since gcd(7, 10) = 1, we have 7⁴ ≡ 1 (mod 10). Write 222 = 4 × 55 + 2.

第二步：因为 gcd(7, 10) = 1，所以 7⁴ ≡ 1 (mod 10)。将 222 写成 222 = 4 × 55 + 2。

Step 3: 7²²² = (7⁴)⁵⁵ × 7² ≡ 1⁵⁵ × 49 ≡ 9 (mod 10).

第三步：7²²² = (7⁴)⁵⁵ × 7² ≡ 1⁵⁵ × 49 ≡ 9 (mod 10)。

Thus 7²²² ends in 9.

因此 7²²² 的个位数字是 9。

Another example: Compute 3¹⁰⁰⁰ mod 7. Since 7 is prime, φ(7) = 6. Then 3⁶ ≡ 1 (mod 7). Since 1000 = 6 × 166 + 4, we have 3¹⁰⁰⁰ ≡ 3⁴ = 81 ≡ 4 (mod 7).

另一例：计算 3¹⁰⁰⁰ mod 7。由于 7 是素数，φ(7) = 6。于是 3⁶ ≡ 1 (mod 7)。由于 1000 = 6 × 166 + 4，故 3¹⁰⁰⁰ ≡ 3⁴ = 81 ≡ 4 (mod 7)。

---

6. Application 2: Finding Units Digits and Last Digits | 应用二：求个位数字与末位数字

Euler’s Theorem is particularly useful for determining the last digit (or last few digits) of large numbers without performing full multiplication.

欧拉定理在确定大数的末位数字（或末几位数字）时尤为实用，无需进行完整的大数乘法。

Problem: Find the last two digits of 19²⁰²⁵.

问题：求 19^{2025 的末两位数字。}

Solution: We need 19²⁰²⁵ mod 100. First, φ(100) = 40. Since gcd(19, 100) = 1, we have 19⁴⁰ ≡ 1 (mod 100). Write 2025 = 40 × 50 + 25. Thus 19²⁰²⁵ ≡ 19²⁵ (mod 100).

解：需要求 19²⁰²⁵ mod 100。首先 φ(100) = 40。由于 gcd(19, 100) = 1，所以 19⁴⁰ ≡ 1 (mod 100)。将 2025 = 40 × 50 + 25，所以 19²⁰²⁵ ≡ 19²⁵ (mod 100)。

Now compute 19² = 361 ≡ 61 (mod 100), 19⁴ ≡ 61² = 3721 ≡ 21 (mod 100), 19⁸ ≡ 21² = 441 ≡ 41 (mod 100), 19¹⁶ ≡ 41² = 1681 ≡ 81 (mod 100).

接着计算 19² = 361 ≡ 61 (mod 100)，19⁴ ≡ 61² = 3721 ≡ 21 (mod 100)，19⁸ ≡ 21² = 441 ≡ 41 (mod 100)，19¹⁶ ≡ 41² = 1681 ≡ 81 (mod 100)。

Then 19²⁵ = 19¹⁶ × 19⁸ × 19¹ ≡ 81 × 41 × 19 (mod 100). Since 81 × 41 = 3321 ≡ 21 (mod 100), and 21 × 19 = 399 ≡ 99 (mod 100). The last two digits are 99.

于是 19²⁵ = 19¹⁶ × 19⁸ × 19¹ ≡ 81 × 41 × 19 (mod 100)。由于 81 × 41 = 3321 ≡ 21 (mod 100)，而 21 × 19 = 399 ≡ 99 (mod 100)。末两位数字为 99。

---

7. Application 3: Solving Linear Congruences | 应用三：求解线性同余方程

Euler’s Theorem provides an explicit way to find modular inverses, which is essential for solving linear congruences of the form ax ≡ b (mod n).

欧拉定理为求模逆元提供了显式方法，而模逆元是求解形如 ax ≡ b (mod n) 的线性同余方程的关键。

If gcd(a, n) = 1, then a^φ(n) ≡ 1 (mod n), so a × a^φ(n)−1 ≡ 1 (mod n). Hence the modular inverse of a is:

若 gcd(a, n) = 1，则 a^φ(n) ≡ 1 (mod n)，即 a × a^φ(n)−1 ≡ 1 (mod n)。因此 a 的模逆元为：

a⁻¹ ≡ a^φ(n)−1 (mod n)

Example: Solve 5x ≡ 3 (mod 18).

例题：求解 5x ≡ 3 (mod 18)。

Step 1: Compute φ(18). Since 18 = 2 × 3², φ(18) = 18 × (1 − 1/2) × (1 − 1/3) = 6.

第一步：计算 φ(18)。因为 18 = 2 × 3²，φ(18) = 18 × (1 − 1/2) × (1 − 1/3) = 6。

Step 2: The inverse of 5 modulo 18 is 5⁶⁻¹ = 5⁵ = 3125. Since 3125 ≡ 11 (mod 18), the inverse is 11 (check: 5 × 11 = 55 ≡ 1 (mod 18)).

第二步：5 模 18 的逆元为 5⁶⁻¹ = 5⁵ = 3125。由于 3125 ≡ 11 (mod 18)，故逆元为 11（验证：5 × 11 = 55 ≡ 1 (mod 18)）。

Step 3: Multiply both sides by 11: x ≡ 3 × 11 = 33 ≡ 15 (mod 18).

第三步：两边乘以 11：x ≡ 3 × 11 = 33 ≡ 15 (mod 18)。

Thus x ≡ 15 (mod 18) is the solution.

因此 x ≡ 15 (mod 18) 即为方程的解。

---

8. Application 4: Euler’s Theorem in Cryptography (RSA) | 应用四：欧拉定理在密码学中的应用（RSA）

Euler’s Theorem underpins the RSA public-key cryptosystem. In RSA, two large primes p and q are chosen, and n = p × q. The encryption exponent e is chosen such that gcd(e, φ(n)) = 1, where φ(n) = (p − 1)(q − 1).

欧拉定理由 RSA 公钥密码体制奠定了理论基础。在 RSA 中，选择两个大素数 p 和 q，令 n = p × q。选择加密指数 e，满足 gcd(e, φ(n)) = 1，其中 φ(n) = (p − 1)(q − 1)。

To decrypt, one finds d such that e × d ≡ 1 (mod φ(n)). By Euler’s Theorem, for any message m coprime to n:

解密时，需要找到 d 使得 e × d ≡ 1 (mod φ(n))。根据欧拉定理，对于任何与 n 互质的消息 m：

m^e×d ≡ m^k×φ(n)+1 ≡ (m^φ(n))^k × m ≡ m (mod n)

This ensures that encrypting with e and decrypting with d restores the original message. The security of RSA relies on the difficulty of factoring n to compute φ(n).

这保证了使用 e 加密、d 解密后能恢复原始消息。RSA 的安全性依赖于对大数 n 进行因式分解以计算 φ(n) 的困难性。

Although the full RSA algorithm is beyond A-Level scope, understanding Euler’s Theorem gives you valuable insight into how modular arithmetic applies to real-world technology.

尽管完整的 RSA 算法超出 A-Level 范围，但理解欧拉定理能让你深刻体会模运算在现实技术中的应用。

---

9. Common Pitfalls and Exam Tips | 常见错误与应试技巧

Students often make the following mistakes when applying Euler’s Theorem:

学生在应用欧拉定理时常出现以下错误：

• Forgetting the coprimality condition: Euler’s Theorem fails if gcd(a, n) > 1. Always verify gcd(a, n) = 1 first.
• 忘记互质条件：当 gcd(a, n) > 1 时，欧拉定理不成立。务必先验证 gcd(a, n) = 1。
• Incorrect φ(n) computation: Remember φ(pᵏ) = pᵏ − pᵏ⁻¹, not pᵏ − 1 (unless k = 1). For n = 8 = 2³, φ(8) = 8 − 4 = 4, not 7.
• φ(n) 计算错误：注意 φ(pᵏ) = pᵏ − pᵏ⁻¹，而非 pᵏ − 1（除非 k = 1）。例如 n = 8 = 2³ 时，φ(8) = 8 − 4 = 4，而不是 7。
• Exponent reduction errors: When reducing a^b mod n, write b = q × φ(n) + r, then a^b ≡ a^r (mod n). Do not reduce a itself unless needed.
• 指数化简错误：化简 a^b mod n 时，将 b 写成 b = q × φ(n) + r，则 a^b ≡ a^r (mod n)。除非必要，不要对底数 a 进行化简。

Exam tip: If the modulus is small, listing the coprime numbers may be faster than using the prime factorization formula. For n = 10, φ(10) = 4 is obvious. Use the formula when n is large or composite with many factors.

应试技巧：若模数较小，列出互质数可能比使用质因数分解公式更快。例如 n = 10 时，φ(10) = 4 一目了然。当 n 较大或因子较多时，再使用公式计算。

---

10. Worked Exam-Style Problem | 典型考题详解

Problem: (A-Level Further Mathematics style) Find the remainder when 2²⁰²⁴ is divided by 15.

题目：（A-Level 进阶数学风格）求 2²⁰²⁴ 除以 15 的余数。

Solution:

解答：

Step 1: Compute φ(15). Since 15 = 3 × 5, φ(15) = 15 × (1 − 1/3) × (1 − 1/5) = 15 × 2/3 × 4/5 = 8.

第一步：计算 φ(15)。由于 15 = 3 × 5，φ(15) = 15 × (1 − 1/3) × (1 − 1/5) = 15 × 2/3 × 4/5 = 8。

Step 2: Check gcd(2, 15) = 1, so 2⁸ ≡ 1 (mod 15).

第二步：验证 gcd(2, 15) = 1，故 2⁸ ≡ 1 (mod 15)。

Step 3: Divide 2024 by 8: 2024 = 8 × 253, exactly. Thus 2²⁰²⁴ = (2⁸)²⁵³ ≡ 1²⁵³ ≡ 1 (mod 15).

第三步：用 8 除以 2024：2024 = 8 × 253，恰好整除。因此 2²⁰²⁴ = (2⁸)²⁵³ ≡ 1²⁵³ ≡ 1 (mod 15)。

Therefore the remainder is 1.

因此余数为 1。

Extension: What if the exponent were 2023? Then 2023 = 8 × 252 + 7, so 2²⁰²³ ≡ 2⁷ = 128 ≡ 8 (mod 15).

拓展：若指数为 2023 呢？则 2023 = 8 × 252 + 7，所以 2²⁰²³ ≡ 2⁷ = 128 ≡ 8 (mod 15)。

---

11. Euler’s Theorem with Non-Coprime Bases | 非互质底数的处理技巧

Sometimes the base a is not coprime to n. In such cases, Euler’s Theorem cannot be applied directly. However, we can often decompose the modulus or use the Chinese Remainder Theorem.

有时底数 a 与 n 不互质。此时不能直接应用欧拉定理。不过，我们通常可以分解模数或使用中国剩余定理。

Example: Compute 6¹⁰⁰ mod 12.

例题：计算 6¹⁰⁰ mod 12。

Observe that 6² = 36 = 3 × 12, so 6² ≡ 0 (mod 12). Therefore for any exponent k ≥ 2, 6^k ≡ 0 (mod 12). Hence 6¹⁰⁰ ≡ 0 (mod 12).

注意到 6² = 36 = 3 × 12，所以 6² ≡ 0 (mod 12)。因此对于任何指数 k ≥ 2，6^k ≡ 0 (mod 12)。故 6¹⁰⁰ ≡ 0 (mod 12)。

For a more systematic approach, factor the base and the modulus. If a shares all prime factors of n, the power eventually becomes divisible by n. In other cases, separate the shared and coprime parts and apply Euler’s Theorem only to the coprime part.

更系统的方法是将底数和模数分解质因数。若 a 含有 n 的全部质因子，则幂最终会被 n 整除。在其他情况下，将互质部分与非互质部分分离，只对互质部分应用欧拉定理。

General strategy: Write a = d × a’, where d contains all common prime factors with n, and gcd(a’, n) = 1. Then use Euler’s Theorem on a’ and handle d separately.

一般策略：将 a 写成 a = d × a’，其中 d 包含与 n 的全部公共质因子，且 gcd(a’, n) = 1。然后对 a’ 使用欧拉定理，并单独处理 d。

---

12. Summary and Further Practice | 总结与拓展练习

Euler’s Theorem is a versatile tool in number theory with applications ranging from simplifying powers to cryptographic protocols. Its essence lies in the beautiful relationship between multiplication modulo n and the totient function φ(n).

欧拉定理是数论中用途广泛的重要工具，其应用涵盖幂运算化简至密码协议等领域。它的精髓在于模 n 乘法与欧拉函数 φ(n) 之间的优美联系。

Key takeaways:

核心要点：

• Euler’s Theorem: a^φ(n) ≡ 1 (mod n) when gcd(a, n) = 1.
• 欧拉定理：当 gcd(a, n) = 1 时，a^φ(n) ≡ 1 (mod n)。
• φ(n) = n × ∏(1 − 1/p) over distinct primes p dividing n.
• φ(n) = n × ∏(1 − 1/p)，其中 p 取遍 n 的互异质因子。
• To reduce large exponents, reduce the exponent modulo φ(n).
• 要化简大指数幂，只需将指数对 φ(n) 取模。
• Always check coprimality before applying the theorem.
• 应用定理前务必检查互质条件。

Practice problem: Find the remainder when 11²⁰²⁶ is divided by 35. (Hint: φ(35) = 24; 2026 = 24 × 84 + 10; compute 11¹⁰ mod 35.)

练习题目：求 11²⁰²⁶ 除以 35 的余数。（提示：φ(35) = 24；2026 = 24 × 84 + 10；计算 11¹⁰ mod 35。）

By mastering Euler’s Theorem, you gain not only a scoring technique for exams but also a deeper appreciation for the elegance of number theory — a branch of mathematics that continues to shape modern technology and cryptography.

掌握欧拉定理，你不仅获得了一项考试得分技巧，更将深入领略数论的优雅之美——这门数学分支至今仍在深刻塑造现代科技与密码学的发展。

Published by TutorHao | Mathematics Revision Series | aleveler.com

更多咨询请联系16621398022（同微信）

📚 Understanding the Accounting Cycle | 会计周期的各阶段解析

The accounting cycle is a systematic sequence of steps used by businesses to record, classify, and summarize financial transactions during an accounting period. It transforms raw transaction data into reliable financial statements that stakeholders can use for decision-making. Mastering the accounting cycle is essential for success in business and accounting examinations.

会计周期是企业在某一会计期间内记录、分类和汇总财务交易的系统化步骤序列。它将原始交易数据转化为可靠的财务报表，供利益相关者用于决策。掌握会计周期是商科和会计考试获得高分的关键。

---

1. Transaction Occurrence and Source Documents | 交易发生与原始凭证

The accounting cycle begins when an economic event takes place. A transaction must involve a measurable exchange of value between two parties, such as a sale, purchase, payment, or receipt. Only genuine business transactions are recorded in the accounting system.

会计周期始于经济事件的发生。一项交易必须涉及两方之间可衡量的价值交换，例如销售、采购、付款或收款。只有真实发生的商业交易才会被记录到会计系统中。

Every transaction must be supported by a source document. Common source documents include:

每项交易必须有原始凭证作为支持。常见的原始凭证包括：

• Sales invoice 销售发票
• Purchase invoice 购货发票
• Receipt 收款收据
• Credit note 贷项通知单
• Debit note 借项通知单
• Bank statement 银行对账单

Source documents provide evidence of the transaction details, such as date, amount, parties involved, and the nature of the exchange. Without these documents, an accountant cannot reliably enter data into the system.

原始凭证提供交易详情证据，包括日期、金额、涉及双方及交换性质。没有这些凭证，会计人员无法可靠地将数据录入系统。

---

2. Journal Entries | 日记账分录

After identifying and measuring transactions, the next step is to record them in the journal. The journal is the book of original entry where all transactions are recorded in chronological order. Each entry follows the double-entry system, meaning every debit must have an equal credit.

识别并计量交易后，下一步是将交易记入日记账。日记账是原始账簿，所有交易按时间顺序记录。每笔分录遵循复式记账法，即每笔借方必须有等额的贷方。

The double-entry rule can be summarized as follows:

复式记账规则可总结如下：

Assets = Liabilities + Equity

For every transaction, the total value of debits must equal the total value of credits. A typical journal entry includes the date, account titles, a brief explanation, and the amounts in the debit and credit columns.

每笔交易中，借方总额必须等于贷方总额。典型的日记账分录包括日期、会计科目、简要说明以及借贷方金额。

Date 日期	Account Titles 科目	Debit 借方	Credit 贷方
Jan 5	Cash 现金	500	–
Jan 5	Sales Revenue 销售收入	–	500

The above example records a cash sale of $500. The debit increases cash, while the credit increases sales revenue. This dual effect maintains the accounting equation’s balance.

上述示例记录了一笔500美元的现金销售。借方增加现金，贷方增加销售收入。这种双重影响维持了会计等式的平衡。

---

3. Posting to the Ledger | 过账至分类账

Posting is the process of transferring journal entries to the general ledger. The ledger is a collection of all accounts, where each account shows its individual history and current balance. Ledger accounts are often drawn in the shape of a “T”, hence the name T-account.

过账是将日记账分录转移到总分类账的过程。分类账是所有账户的集合，每个账户显示其单独的历史记录和当前余额。分类账账户通常以字母’T’的形状绘制，因此称为T型账户。

Each T-account has a debit side (left) and a credit side (right). The difference between the two sides is the account balance. For asset and expense accounts, increases are recorded on the debit side; for liability, equity, and revenue accounts, increases are recorded on the credit side.

每个T型账户有借方（左侧）和贷方（右侧）。两方之间的差额就是账户余额。资产和费用账户的增加记在借方；负债、权益和收入账户的增加记在贷方。

Account Type 账户类型	Normal Debit 正常借方	Normal Credit 正常贷方
Asset 资产	Increase 增加	Decrease 减少
Expense 费用	Increase 增加	Decrease 减少
Liability 负债	Decrease 减少	Increase 增加
Equity 权益	Decrease 减少	Increase 增加
Revenue 收入	Decrease 减少	Increase 增加

After posting, the balances of all accounts are calculated. These balances will be used in the next stage to prepare a trial balance.

过账后，计算所有账户的余额。这些余额将在下一阶段用于编制试算平衡表。

---

4. Trial Balance | 试算平衡表

At the end of the accounting period, a trial balance is prepared to verify that total debits equal total credits across all ledger accounts. This is a crucial internal check for mathematical accuracy in the accounting records.

在会计期末，编制试算平衡表以验证所有分类账账户的借方总额等于贷方总额。这是对会计记录数学准确性的一项重要内部检查。

The trial balance is not a financial statement; it is simply a working paper. It helps detect errors such as posting a debit as a credit, omitting a transaction, or recording incorrect amounts. However, it cannot identify every error—for example, a transaction completely omitted from the books will not affect the balance.

试算平衡表不是财务报表，而是一份工作底稿。它帮助发现错误，例如将借方误记为贷方、遗漏交易或记录金额错误。然而，它无法识别所有错误——例如完全遗漏账本的交易不会影响平衡。

The format of a trial balance includes account names and their debit or credit balances:

试算平衡表的格式包括账户名称及其借方或贷方余额：

Account 账户	Debit 借方	Credit 贷方
Cash 现金	8,000	–
Accounts Receivable 应收账款	4,000	–
Accounts Payable 应付账款	–	3,000
Capital 资本	–	9,000
Total 合计	12,000	12,000

---

5. Adjusting Entries | 调整分录

Under the accrual basis of accounting, revenues and expenses must be recognized in the period they are earned or incurred, regardless of cash movement. Adjusting entries update accounts at the end of the period to reflect this principle. These entries are essential for matching revenue with expenses in the correct period.

在权责发生制基础上，收入和费用必须在其赚取或发生的期间确认，而无论现金是否收付。调整分录在期末更新账户以体现这一原则。这些分录对于在正确期间匹配收入与费用至关重要。

The four main types of adjusting entries are:

四类主要调整分录包括：

• Prepaid expenses 预付费用 – recognise the portion that has been consumed as an expense. 预付费用——确认已耗用的部分为费用。
• Unearned revenue 预收收入 – recognise the portion that has been earned as revenue. 预收收入——确认已赚取的部分为收入。
• Accrued expenses 应计费用 – record expenses that have been incurred but not yet paid. 应计费用——记录已发生但尚未支付的费用。
• Accrued revenues 应计收入 – record revenues that have been earned but not yet received. 应计收入——记录已赚取但尚未收到的收入。

For example, at the end of the period, a company discovers that it owes $300 for electricity used during the month but the bill has not yet arrived. The adjusting entry would be:

例如，期末时公司发现本月已使用300美元电费但账单尚未到达。调整分录为：

Dr Electricity Expense 电费费用 300 → Cr Accrued Expenses 应计费用 300

Adjusting entries also cover non-cash items such as depreciation of non-current assets and allowance for doubtful debts.

调整分录还包括非现金项目，例如非流动资产折旧和坏账准备。

---

6. Adjusted Trial Balance | 调整后试算平衡表

After all adjusting entries are made, a new trial balance is prepared. This adjusted trial balance includes the effects of all adjustments and confirms that debits still equal credits. It provides the final balances that appear directly in the financial statements.

所有调整分录完成后，编制新的试算平衡表。该调整后试算平衡表包含所有调整的影响，并确认借方仍等于贷方。它提供直接出现在财务报表中的最终余额。

This step is sometimes overlooked by students, but it is a critical control point. If the adjusted trial balance does not balance, the financial statements cannot be prepared accurately.

这一步有时被学生忽视，但它是一个关键控制点。如果调整后试算平衡表不平衡，就无法准确编制财务报表。

---

7. Preparing Financial Statements | 编制财务报表

The ultimate goal of the accounting cycle is to prepare financial statements. The key statements for a sole trader include:

会计周期的最终目标是编制财务报表。个体工商户的关键报表包括：

1. Income Statement (Profit and Loss Account) 损益表

This statement calculates net profit by deducting all expenses from total revenue. It shows the business’s operating performance over the period.

该表通过从总收入中扣除所有费用来计算净利润，显示企业在期间内的经营业绩。

Net Profit = Total Revenue − Total Expenses

2. Statement of Financial Position (Balance Sheet) 财务状况表（资产负债表）

This statement lists assets, liabilities, and equity at a specific date. It demonstrates that the accounting equation remains balanced after all transactions and adjustments.

该表列示特定日期的资产、负债和权益，证明在全部交易和调整后会计等式仍然平衡。

Assets = Liabilities + Equity

The income statement is prepared first because net profit is needed to update the capital balance in the balance sheet.

先编制损益表，因为需要净利润来更新资产负债表中的资本余额。

---

8. Closing Entries | 结账分录

Closing entries are made at the end of the accounting period to transfer the balances of temporary accounts—revenues, expenses, and drawings—to the permanent equity account (capital). Temporary accounts are then reduced to zero, ready for the next period.

结账分录在会计期末编制，将临时账户——收入、费用和提款——的余额转入永久性权益账户（资本）。随后临时账户归零，为下一期间做好准备。

The closing process involves three main steps:

结账过程包含三个主要步骤：

• Close revenue accounts to the Income Summary 将收入账户结转至损益汇总
• Close expense accounts to the Income Summary 将费用账户结转至损益汇总
• Close Income Summary to Capital (or owner’s equity) 将损益汇总结转至资本（或所有者权益）
• Close drawings to Capital 将提款结转至资本

For example, if total revenue is $10,000 and total expenses are $7,000, the closing entries are:

例如，若总收入为10,000美元，总费用为7,000美元，则结账分录为：

Entry 分录	Debit 借方	Credit 贷方
Close revenue 结转收入	Revenue 10,000	Income Summary 10,000
Close expenses 结转费用	Income Summary 7,000	Expenses 7,000
Close net profit 结转净利润	Income Summary 3,000	Capital 3,000

After closing, all temporary accounts have zero balances, and the capital account reflects both the beginning capital and the current period’s profit.

结账后，所有临时账户余额为零，资本账户反映期初资本和本期利润。

---

9. Post-Closing Trial Balance | 结账后试算平衡表

The final check in the accounting cycle is the post-closing trial balance. This is prepared after closing entries are posted to the ledger. It lists only permanent accounts—assets, liabilities, and capital—to confirm that the books remain in balance for the next accounting period.

会计周期的最终检查是结账后试算平衡表。该表在结账分录过账后编制，仅列示永久性账户——资产、负债和资本——以确认账簿在下一会计期间开始前保持平衡。

Balances in the post-closing trial balance are carried forward as the opening balances of the next period. The absence of temporary accounts confirms that the cycle has been completed accurately.

结账后试算平衡表中的余额将结转作为下一期间的期初余额。临时账户的消失确认了会计周期已准确完成。

---

10. Optional Reversing Entries | 可选转回分录

Some businesses choose to record reversing entries at the beginning of the new period. These entries reverse certain adjusting entries from the previous period, particularly accruals, to simplify routine recording when cash is later paid or received.

部分企业选择在新期间期初编制转回分录。这些分录冲销前期某些调整分录，特别是应计项目，以便在后续实际收付现金时简化日常记录。

For example, the accrued electricity expense of $300 recorded at the end of period 1 can be reversed at the start of period 2. When the actual bill is paid later, the entry will directly affect the expense account without needing to recall the prior accrual.

例如，第一期期末记录的应计电费300美元可以在第二期期初转回。当实际账单随后支付时，分录将直接影响费用账户，无需回溯前期的应计项目。

Reversing entries are optional; they do not affect the final financial statements. They are an internal convenience for accountants.

转回分录是可选的；它们不影响最终财务报表，只是会计人员的一种内部便利手段。

---

Published by TutorHao | Business Revision Series | aleveler.com

更多咨询请联系16621398022（同微信）