Forgot password?
 Register account
View 1244|Reply 5

数学地图

[Copy link]

3159

Threads

7941

Posts

610K

Credits

Credits
63770
QQ

Show all posts

hbghlyj Posted 2025-5-7 07:25 |Read mode
chalkdustmagazine.com/features/an-invitation-to-category-theory/
范畴论是一个关于关系的理论,描述并研究关系的所有可能性质。Martin Kuppe曾绘制过一幅精美的数学地图,其中范畴论高高悬挂在天空,提供了整个地图的缩略图。它让我们能够看到在地面看不到的各个领域之间的关系,证明看似不相关的数学领域并不是完全不同。当你想解决某个领域(比如说拓扑)中的问题,但没有合适的工具可以使用时,这就变得非常有用。通过将问题转移到不同领域(比如群论),就能让你换个角度看问题,说不定还能发现新的工具,让问题变得更容易解决。事实上,范畴论就是这样产生的。它诞生于20世纪40年代,背景是人们试图用更简单的代数方法来解决一个困难的拓扑问题。

回到数学地图,你可以注意到各领域都包含一些对象:集合论有集合,群论有群,拓扑学有拓扑空间…… 这些对象彼此关联:集合通过映射关联,群通过同态关联,拓扑空间通过连续映射关联……

这条共同的线索贯穿了整个地图,将各领域统一到一起。范畴论将这种统一形式化了。更具体地说,范畴是一组对象及其关系(称为态射,morphisms)的集合,态射在组合(composition)和结合性(associativity)方面表现良好。这样就为数学提供了一个模板,将不同内容输入模板,就能重建一个数学领域:集合范畴由集合和它们之间的关系(映射)组成;群范畴由群和它们之间的关系(群同态)组成;拓扑空间范畴由拓扑空间和它们之间的关系(连续映射)组成;等等。

3159

Threads

7941

Posts

610K

Credits

Credits
63770
QQ

Show all posts

 Author| hbghlyj Posted 2025-5-7 07:31

一个数学对象完全由它与所有其他对象的关系决定(isomorphism 同构)

当且仅当两个对象以同样方式与范畴中的每个对象相关时,两个对象本质上是不可区分的。这其中的主旨 [这是著名的米田引理(Yoneda lemma)的一个推论] 与我们的日常经验并没有太大区别。你可以通过观察人们的关系来了解他们。

3159

Threads

7941

Posts

610K

Credits

Credits
63770
QQ

Show all posts

 Author| hbghlyj Posted 2025-5-7 07:34

High-order category

Morphism between a Category $C_1$ and another Category $C_2$ is called Functor $F$ (函子).
If $C_1 = C_2$, $F$ is Endo-Functor.

范畴之间的关系称为函子(functor),函子的关系称为自然变换(natural transformation),我们可以继续:“关系之间的关系之间的关系……?”这样做将使我们进入更高维的范畴论,这正是Eugenia Cheng的主要研究领域。

3159

Threads

7941

Posts

610K

Credits

Credits
63770
QQ

Show all posts

 Author| hbghlyj Posted 2025-5-7 07:43
nLab 着重范畴论有关的数学
这确实是一个很好的资源。不过要注意的是,很多文章都假设他们的受众已经知道这些基础知识。因此,在学习基础知识时,它们可能没什么用写这本书的人其实都是专家,但他们的读者大多是彼此。
nLab 在各种抽象和深奥的概念上花费的时间相对较多,而在某些基础知识上花费的时间相对较少。

3159

Threads

7941

Posts

610K

Credits

Credits
63770
QQ

Show all posts

 Author| hbghlyj Posted 2025-5-7 08:32
初学者可以阅读 Emily Riehl 的著作《Category Theory in Context 情境中的范畴论》。以下是摘录
In linear algebra, elementary row operations on a matrix with n rows ultimately arise from doing those row operations on the nxn identity matrix. This is because a special functor called the Yoneda Embedding has very nice properties. (It is "full" and "faithful")
$$
\left[\begin{array}{lll}
1 & 2 & 3 \\
4 & 5 & 6 \\
7 & 8 & 9
\end{array}\right] \xrightarrow[1\leftrightarrow2]{\text { switch rows }}\left[\begin{array}{lll}
4 & 5 & 6 \\
1 & 2 & 3 \\
7 & 8 & 9
\end{array}\right]=\left[\begin{array}{lll}
0 & 1 & 0 \\
1 & 0 & 0 \\
0 & 0 & 1
\end{array}\right]\left[\begin{array}{lll}
1 & 2 & 3 \\
4 & 5 & 6 \\
7 & 8 & 9
\end{array}\right]
$$
same as multiplying on the left by identity matrix usth rows #1 & #2 switched.

every group is a category with one object (the morphisms correspond to group elements)
$$\bullet\rightarrow C:\text{any category}$$
a functor from that group defines a group action on its image
and a natural transformation between two such functors G-equivariant map (where G = your group)

the Axiom of Choice is the statement that a certain construction called a limit (more specifically, a product) exists in the category of sets.

$\{Χ_α\}_{α\in A}$ = nonempty collection of sets

A functor called $π_1$, associates a group to every topological space $X$ the fundamental group of $X$:

Top$\to$Group
$X\mapsto π_1(Χ)$

You can use this functor to prove

①the fundamental theorem of algebra
every nonconstant polynomial in $\mathbb C[z]$ has a root in $\mathbb C$.

②Brouwer's Fixed Point Theorem
every homeomorphism of the disc has a fixed point

③ the Perron-Frobenius theorem
every 3x3 matrix with positive real entries has a positive eigenvalue

to name a few!

The universal property of a free group $F(S)$ on a set S arises as a special relationship between the categories Group and Set. (i.e. there is a pair of "adjoint functors" between them).

A free group on a set $S$ comes with a set map

F(S)

S

so that for any group and any set map...
$$\xymatrix{
  F(S) \ar@{-->}[dr] & \\
  S \ar[u] \ar[r] & G
}$$
...there is a unique homomorphism here causing this triangle to commute.

Viewing both $\mathbb Z$ and $\mathbb R$ as posets (and hence categories), the definitions of the floor $\lfloor\ \rfloor:\mathbb R→\mathbb Z$ and ceiling functions $\lceil\ \rceil:\mathbb R→\mathbb Z$ also arise from a special relationship (a pair of adjoint functors) between $\mathbb R$ and $\mathbb Z$.


The fundamental theorem of Galois Theory says that a certain functor is an isomorphism of categories":
$$\theta^\text{op}_G→\text{Field}_F^K$$
$\theta^\text{op}_G$ = orbit category of G
objects = quotients G/H by Subgroups H of G morphisms = G-equivariant maps

$\text{Field}_F^K$
objects = intermediate fields $F\subset E\subset K$ morphisms = field homomorphisms that fix $F$

K/F = finite Galois extension

G = corresponding Galois group

The Riesz Representation Theorem is the statement that a natural transformation between certain functors is quite special. (It's a "natural isomorphism")
\[
\xymatrix@C+2pc{
\mathsf{cHaus} \rtwocell& \mathsf{Ban}
}
\]
cHaus
objects = compact Hausdorff spaces
morphisms = continuous functions

Ban
objects = real Banach spaces
morphisms = continuous linear maps maps

3159

Threads

7941

Posts

610K

Credits

Credits
63770
QQ

Show all posts

 Author| hbghlyj Posted 2025-5-7 08:37

DeepLearning: KAN vs MLP

“KANs employ learnable activation functions on the edges themselves“

MLPs (Multi Layer Perceptrons 多层感知器) is improved by KANs with Weightages on Edges instead of at nodes in MLPs.

Current DeepLearning math (MLP, Cauchy Gradient Descent) which 丘成桐 comments as “over simplistic” of humain brain.There must be a higher Math for “Artificial General Intelligence” (AGI) approaching human intelligence.

If you know Category Theory (CT) applied in Functional Programming (FP) , you would recognize this KAN is in fact “functions” (called Functors 函子 in CT) composition and Monad pattern (Endo-functor in the Category of Monoid).

Read “Kolmogorov-Arnold Networks: a Critique“ by Rubens Zimbres on Medium: medium.com/@rubenszimbres/kolmogorov-arnold-n … ritique-2b37fea2112e

Mobile version|Discuz Math Forum

2025-5-31 11:07 GMT+8

Powered by Discuz!

× Quick Reply To Top Edit