Chapter 1:Preliminaries

本文探讨了实数系统的性质及其在微积分中的核心作用。介绍了实数、有理数与无理数的概念,讨论了代数、顺序与完备性属性,并分析了这些属性如何支撑微积分的基本理论。

摘要生成于 C知道 ,由 DeepSeek-R1 满血版支持, 前往体验 >

1.1Real numbers and the real line

Real Numbers

Much of calculus is based on properties of the real number system. Real numbers are numbers that can be expressed as decimals, such as

{\color{Red} {\color{Red} }-\frac{3}{4}=-0.75000...}

{\color{Red} \frac{1}{3}=0.3333...}

{\color{Red} {\color{Red} }\sqrt{2}=1.4142...}

The dots in each case indicate that the sequence of decimal digits goes on forever. Every conceivable decimal expansion represents a real number, although some numbers have two representations. For instance, the infinite decimals and represent the same real number 1. A similar statement holds for any number with an infinite tail of 9’s.

The real numbers can be represented geometrically as points on a number line called the real line.

The symbol R denotes either the real number system or, equivalently, the real line

The properties of the real number system fall into three categories: algebraic properties, order properties, and completeness. The algebraic properties say that the real numbers can be added, subtracted, multiplied, and divided (except by 0) to produce more real numbers under the usual rules of arithmetic. You can never divide by 0. 

The order properties of real numbers are given in Appendix 4. The following useful rules can be derived from them, where the \Rightarrow symbol means “implies.”

Notice the rules for multiplying an inequality by a number. Multiplying by a positive number preserves the inequality; multiplying by a negative number reverses the inequality. Also, reciprocation reverses the inequality for numbers of the same sign. 

The completeness property of the real number system is deeper and harder to define precisely. However, the property is essential to the idea of a limit (Chapter 2). Roughly speaking, it says that there are enough real numbers to “complete” the real number line, in the sense that there are no “holes” or “gaps” in it. Many theorems of calculus would fail if the real number system were not complete. The topic is best saved for a more advanced course, but Appendix 4 hints about what is involved and how the real numbers are constructed. 

We distinguish three special subsets of real numbers. 

1. The natural numbers, namely 1, 2, 3, 4,...

2. The integers, namely 0,\pm 1,\pm 2,\pm3,...

3. The rational numbers,namely the numbers that can be expressed in the form of a fraction m/n, where m and n are integers and n\neq0. Examples are \frac{1}{3},  \frac{200}{13}.

The rational numbers are precisely the real numbers with decimal expansions that are either

(a) terminating (ending in an infinite string of zeros), for example,

\frac{3}{4}=0.75000...=0.75

or 

(b) eventually repeating (ending with a block of digits that repeats over and over), for example 

\frac{23}{11}=2.090909...=2.\bar{09}

(The bar indicates the block of repeating digits.)

A terminating decimal expansion is a special type of repeating decimal since the ending zeros repeat. 

The set of rational numbers has all the algebraic and order properties of the real numbers but lacks the completeness property. For example, there is no rational number whose square is 2; there is a “hole” in the rational line where \sqrt{2} should be. 

Real numbers that are not rational are called irrational numbers. They are characterized by having nonterminating and nonrepeating decimal expansions.  Examples are \pi, \sqrt{2},\sqrt[3]{7} and \lg{3}. Since every decimal expansion represents a real number, it should be clear that there are infinitely many irrational numbers. Both rational and irrational numbers are found arbitrarily close to any point on the real line. 

### 图对比学习的符号表示与预备知识 #### 符号定义 为了便于理解图对比学习(Graph Contrastive Learning, GCL),以下是常用的一些符号及其含义: - \( \mathcal{G}=(\mathcal{V}, \mathcal{E}) \) 表示一个图,其中 \( \mathcal{V}=\left\{v_{1}, v_{2}, \ldots, v_{|\mathcal{V}|}\right\} \) 是节点集合,\( \mathcal{E} \subseteq \mathcal{V} \times \mathcal{V} \) 是边集合[^1]。 - \( A \in \mathbb{R}^{N \times N} \) 或者 \( E \in \mathbb{R}^{N \times M} \),分别代表邻接矩阵和边特征矩阵。这里 \( N=|\mathcal{V}| \), \( M \) 则取决于具体的任务设定[^3]。 - \( X \in \mathbb{R}^{N \times F} \) 为节点特征矩阵,每一行对应于一个节点的特征向量,列数等于每个节点拥有的特征数量 \( F \)。 - 对于任意给定的图增强操作 \( T(\cdot) \),可以得到两个不同的视图 (views): \( \hat{\mathcal{G}}_i=T_i (\mathcal{G}), i={1,2} \),这两个视图为原始图的不同版本,通过特定的数据增强技术获得。 #### 预备知识 ##### 数据增强机制 在图对比学习框架下,数据增强是指通过对原图施加某种变换从而生成新的图实例的过程。常见的增强手段包括但不限于节点删除(Node Dropping)、边缘移除(Edge Removing)以及子图采样(Subgraph Sampling)等。这些方法旨在保持图形的整体结构特性的同时引入一定的变化,以便让模型能够捕捉到更鲁棒性的表征信息[^2]。 ##### 对比损失函数 对比损失通常用来度量同一张图经过不同转换后的相似程度。一种广泛使用的公式是InfoNCE Loss: \[ l(i,j)=−log\frac{{exp(sim(z_i,\tilde z_j)/τ)}}{{∑_{k∈Ω_N(j)} exp(sim(z_k ,\tilde z_j )/τ)}} \] 这里的 \( sim(a,b) \) 可以选择余弦相似度或其他合适的距离测度;而温度参数 \( τ>0 \) 控制着分布的尖锐程度;\( Ω_N(j) \) 定义了一个负样本池,它包含了除了正样本外的所有可能的选择。 ```python import torch.nn.functional as F def info_nce_loss(anchor_embeddings, positive_embeddings, negative_embeddings, temperature=0.5): pos_scores = F.cosine_similarity(anchor_embeddings, positive_embeddings) neg_scores = [] for neg_emb in negative_embeddings: score = F.cosine_similarity(anchor_embeddings, neg_emb.unsqueeze(dim=0)) neg_scores.append(score) logits = torch.cat([pos_scores.view(-1, 1)] + [torch.stack(neg_scores).T], dim=-1) / temperature labels = torch.zeros(logits.shape[0]).long().to(logits.device) loss = F.cross_entropy(logits, labels) return loss ```
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包

打赏作者

「已注销」

你的鼓励将是我创作的最大动力

¥1 ¥2 ¥4 ¥6 ¥10 ¥20
扫码支付:¥1
获取中
扫码支付

您的余额不足,请更换扫码支付或充值

打赏作者

实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值