Trance-0/NoteNextra-origin
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

updates

Browse Source
This commit is contained in:
Trance-0
2025-06-06 23:02:06 -05:00
parent 135575f0bf
commit a67c60f03e
6 changed files with 22 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
pages/Math401/Math401_EX1.md Normal file
View File

@@ -0,0 +1,20 @@
# Beginning of the Summer Program
## Schedule
Presentation starts next week
Start with examples, do exploratory work, it's just a summer.
Final work: Find certain topic you are interested in, and do a expository paper.
Find the motivation, background, definition, theorem, example, application for the theory you are interested in.
At least 3 presentations is required.
Collect the papers you interested in as you go the research, it is not linear.
Symposium on November.
Lightning talk (3 minutes) on end of July.

2
pages/Math401/Math401_EX2.md Normal file
View File

@@ -0,0 +1,2 @@
# Schedule Presentation

64
pages/Math401/Math401_N1.md Normal file
View File

@@ -0,0 +1,64 @@
# Node 1
_all the materials are recovered after the end of the course. I cannot split my mind away from those materials._
## Recap on familiar ideas
### Group
A group is a set $G$ with a binary operation $\cdot$ that satisfies the following properties:
1. **Closure**: For all $a, b \in G$, the result of the operation $a \cdot b$ is also in $G$.
2. **Associativity**: For all $a, b, c \in G$, $(a \cdot b) \cdot c = a \cdot (b \cdot c)$.
3. **Identity**: There exists an element $e \in G$ such that for all $a \in G$, $e \cdot a = a \cdot e = a$.
4. **Inverses**: For each $a \in G$, there exists an element $b \in G$ such that $a \cdot b = b \cdot a = e$.
### Ring
A ring is a set $R$ with two binary operations, addition and multiplication, that satisfies the following properties:
1. **Additive Closure**: For all $a, b \in R$, the result of the addition $a + b$ is also in $R$.
2. **Additive Associativity**: For all $a, b, c \in R$, $(a + b) + c = a + (b + c)$.
3. **Additive Identity**: There exists an element $0 \in R$ such that for all $a \in R$, $0 + a = a + 0 = a$.
4. **Additive Inverses**: For each $a \in R$, there exists an element $-a \in R$ such that $a + (-a) = (-a) + a = 0$.
5. **Multiplicative Closure**: For all $a, b \in R$, the result of the multiplication $a \cdot b$ is also in $R$.
6. **Multiplicative Associativity**: For all $a, b, c \in R$, $(a \cdot b) \cdot c = a \cdot (b \cdot c)$.
Others not shown since you don't need too much.
## Symmetric Group
### Definition
The symmetric group $S_n$ is the group of all permutations of $n$ elements. Or in other words, the group of all bijections from a set of $n$ elements to itself.
Example:
$$
e=1,2,3\\
(12)=2,1,3\\
(13)=3,2,1\\
(23)=1,3,2\\
(123)=3,1,2\\
(132)=2,3,1
$$
$(12)$ means that $1\to 2, 2\to 1, 3\to 3$ we follows the cyclic order for the elements in the set.
$S_3 = \{e, (12), (13), (23), (123), (132)\}$
The multiplication table of $S_3$ is:
|Element|e|(12)|(13)|(23)|(123)|(132)|
|---|---|---|---|---|---|---|
|e|e|(12)|(13)|(23)|(123)|(132)|
|(12)|(12)|e|(123)|(13)|(23)|(132)|
|(13)|(13)|(132)|e|(12)|(23)|(123)|
|(23)|(23)|(123)|(132)|e|(12)|(13)|
|(123)|(123)|(13)|(23)|(132)|e|(12)|
|(132)|(132)|(23)|(12)|(123)|(13)|e|
## Functions defined on $S_n$
### Symmetric Generating Set

9
pages/Math401/Math401_N2.md Normal file
View File

@@ -0,0 +1,9 @@
# Node 2
## Random matrix theory
### Wigner's semicircle law
## h-Inversion Polynomials for a Special Heisenberg Family
###

399
pages/Math401/Math401_N3.md Normal file
View File

@@ -0,0 +1,399 @@
# Coding and Information Theory Crash Course
## Encoding
Let $A,B$ be two finite sets with size $a,b$ respectively.
Let $S(A)=\bigcup_{r=1}^{\infty}A^r$ be the word semigroup generated by $A$.
A one-to-one mapping $f:A\to S(B)$ is called a code with message alphabet $A$ and encoded alphabet $B$.
Example:
- $A=$ RGB color space
- $B=\{0\sim 255\}$
- $f:A\to B^n$ is a code
For example, $f(white)=(255,255,255)$, $f(green)=(0,255,0)$
### Uniquely decipherable codes
A code $f:A\to S(B)$ is called uniquely decipherable if the extension code
$$
\tilde{f}:S(A)\to S(B)=f(a_1)f(a_2)\cdots f(a_n)
$$
is one-to-one.
Example:
- $A=\{a,b,c,d\}$
- $B=\{0,1\}$
- $f(a)=00$, $f(b)=01$, $f(c)=10$, $f(d)=11$
is uniquely decipherable.
- $f(a)=0$, $f(b)=1$, $f(c)=10$, $f(d)=11$
is not uniquely decipherable.
Since $\tilde{f}(ba)=10=\tilde{f}(c)$
#### Irreducible codes
A code $f:A\to S(B)$ is called irreducible if for any $x,y\in A$, $f(y)\neq f(x)w$ for some $w\in S(B)$.
This condition ensures that every message used in the encoding process is uniquely decipherable.
Means that there is no ambiguity in the decoding process.
#### Theorem 1.1.1 Sardinas and Patterson Theorem
Let $A,B$ be alphabet sets of size $a,b$ respectively.
Let $f:A\to S(B)$ be a code that is uniquely decipherable.
Then
$$
\sum_{x\in A}b^{-l(f(x))}\leq 1
$$
where $l(f(x))$ is the length of the codeword $f(x)$. (Number of elements used in the codeword for $x$)
Proof:
Let $L$ denote the max length of the codeword for any $x\in A$.
$L=\max\{l(f(x))|x\in A\}$
Let $c_r$ be the number of codewords of length $r$.
Then
$$
\begin{aligned}
\sum_{x\in A}b^{-l(f(x))}&=\sum_{r=1}^{L}\sum_{l(f(x))=r}b^{-l(f(x))}\\
&=\sum_{r=1}^{L}c_rb^{-r}
\end{aligned}
$$
Note that $r$ is the length of the codeword.
The max number of elements that can be represented by $r$ elements in $B$ is $b^r$, so $c_r\leq b^r$.
This gives that the power series
$$
\sum_{r=1}^{\infty}b^{-r}c_r
$$
converges to $R=\frac{1}{\limsup_{r\to\infty}\sqrt[r]{c_r}}\leq 1$.
So the sum of the probabilities of all the codewords must be less than or equal to 1.
#### Sardinas and Patterson Algorithm
Let $A=\{a_1,a_2,\cdots,a_n\}$ be the message alphabet and $B=\{b_1,b_2,\cdots,b_m\}$ be the encoded alphabet.
We test whether the code $f:A\to S(B)$ is uniquely decipherable.
```python
def is_uniquely_decipherable(f):
contain_common_prefix =
message_space=set()
for x in A:
if f(x) in message_space:
return False
message_space.add(f(x))
for i in range(1,len(f(x))):
if f(x)[:i] in message_space:
contain_common_prefix = True
while contain_common_prefix:
contain_common_prefix = False
for x in message_space:
for y in message_space:
code_length = min(len(x),len(y))
if x[:code_length] == y[:code_length]:
contain_common_prefix = True
if len(x) < len(y):
message_space.add(y[code_length:])
else:
message_space.add(x[code_length:])
break
return True
```
### Shannon's source coding theorem
#### Definition 1.1.4
An elementary information source is a pair $(A,\mu)$ where $A$ is an alphabet and $\mu$ is a probability distribution on $A$. $\mu$ is a function $\mu:A\to[0,1]$ such that $\sum_{a\in A}\mu(a)=1$.
The **mean code word length** of an information source $(A,\mu)$ given a code $f:A\to S(B)$ is defined as
$$
\overline{l}(\mu,f)=\sum_{a\in A}\mu(a)l(f(a))
$$
The $L(\mu)$ of the mean code word length is defined as
$$
L(\mu)=\min\{\overline{l}(\mu,f)|f:A\to S(B)\text{ is uniquely decipherable}\}
$$
#### Lemma: Jenson's inequality
Let $f$ be a convex function on the interval $(a,b)$. Then for any $x_1,x_2,\cdots,x_n\in (a,b)$ and $\lambda_1,\lambda_2,\cdots,\lambda_n\in [0,1]$ such that $\sum_{i=1}^{n}\lambda_i=1$, we have
$$f(\sum_{i=1}^{n}\lambda_ix_i)\leq \sum_{i=1}^{n}\lambda_if(x_i)$$
Proof:
If $f$ is a convex function, there are three properties that useful for the proof:
1. $f''(x)\geq 0$ for all $x\in (a,b)$
2. For any $x,y\in (a,b)$, $f(x)\geq f(y)+(x-y)f'(y)$ (Take tangent line at $y$)
3. For any $x,y\in (a,b)$ and $0<\lambda<1$, we have $f(\lambda x+(1-\lambda)y)\leq \lambda f(x)+(1-\lambda)f(y)$ (Take line connecting $f(x)$ and $f(y)$)
We use $f(x)\geq f(y)+(x-y)f'(y)$, we replace $y=\sum_{i=1}^{n}\lambda_ix_i$ and $x=x_j$, we have
$$f(x_j)\geq f(\sum_{i=1}^{n}\lambda_ix_i)+(x_j-\sum_{i=1}^{n}\lambda_ix_i)f'(\sum_{i=1}^{n}\lambda_ix_i)$$
We sum all the inequalities, we have
$$
\begin{aligned}
\sum_{j=1}^{n}\lambda_j f(x_j)&\geq \sum_{j=1}^{n}\lambda_jf(\sum_{i=1}^{n}\lambda_ix_i)+\sum_{j=1}^{n}\lambda_j(x_j-\sum_{i=1}^{n}\lambda_ix_i)f'(\sum_{i=1}^{n}\lambda_ix_i)\\
&\geq \sum_{j=1}^{n}\lambda_jf(\sum_{i=1}^{n}\lambda_ix_i)+0\\
&=f(\sum_{j=1}^{n}\lambda_ix_j)
\end{aligned}
$$
#### Theorem 1.1.5
Shannon's source coding theorem
Let $(A,\mu)$ be an elementary information source and let $b$ be the size of the encoded alphabet $B$.
Then
$$
\frac{-\sum_{x\in A}\mu(x)\log\mu(x)}{\log b}\leq L(\mu)<\frac{-\sum_{x\in A}\mu(x)\log\mu(x)}{\log b}+1
$$
where $L(\mu)$ is the minimum mean code word length of all uniquely decipherable codes for $(A,\mu)$.
Proof:
First, we show that
$$
\sum_{a\in A}m(a)\mu(a)\geq -(\log b)^{-1}\sum_{a\in A}\mu(a)\log \mu(a)
$$
Let $m:A\to \mathbb{Z}_+$ be a map satisfying
$$
\sum_{a\in A}b^{-m(a)}\leq 1
$$
We defined the probability distribution $v$ on $A$ as
$$
v(x)=\frac{b^{-m(x)}}{\sum_{a\in A}b^{-m(a)}}=\frac{b^{-m(x)}}{T}
$$
Write $T=\sum_{a\in A}b^{-m(a)}$ so that $T\leq 1$.
Since $b^{-m(a)}=T\cdot v(a)$,
$-m(a)\log b=\log T+\log v(a)$, $m(a)=-\frac{\log T}{\log b}-\frac{\log v(a)}{\log b}$
We have
$$
\begin{aligned}
\sum_{a\in A}m(a)\mu(a)&=\sum_{a\in A}\left(-\frac{\log T}{\log b}-\frac{\log v(a)}{\log b}\right)\mu(a)\\
&=-\frac{\log T}{\log b}\sum_{a\in A}\mu(a)-\frac{1}{\log b}\sum_{a\in A}\mu(a)\log v(a)\\
&=(\log b)^{-1}\left\{-\log T - \sum_{a\in A}\mu(a)\log v(a)\right\}
\end{aligned}
$$
Without loss of generality, we assume that $\mu(a)>0$ for all $a\in A$.
$$
\begin{aligned}
-\sum_{a\in A}\mu(a)\log v(a)&=-\sum_{a\in A}\mu(a)(\log \mu(a)+\log v(a)-\log \mu(a))\\
&=-\sum_{a\in A}\mu(a)\log \mu(a)-\sum_{a\in A}\mu(a)\log \frac{v(a)}{\mu(a)}\\
&=-\sum_{a\in A}\mu(a)\log \mu(a)-\log \prod_{a\in A}\left(\frac{v(a)}{\mu(a)}\right)^{\mu(a)}
\end{aligned}
$$
$\log \prod_{a\in A}\left(\frac{v(a)}{\mu(a)}\right)^{\mu(a)}=\sum_{a\in A}\mu(a)\log \frac{v(a)}{\mu(a)}$ is also called Kullback–Leibler divergence or relative entropy.
Since $\log$ is a concave function, by Jensen's inequality $f(\sum_{i=1}^{n}\lambda_ix_i)\leq \sum_{i=1}^{n}\lambda_if(x_i)$, we have
$$
\begin{aligned}
\sum_{a\in A}\mu(a)\log \frac{v(a)}{\mu(a)}&\leq\log\left(\sum_{a\in A}\mu(a) \frac{v(a)}{\mu(a)}\right)\\
\log\left(\prod_{a\in A}\left(\frac{v(a)}{\mu(a)}\right)^{\mu(a)}\right)&\leq \log\left(\sum_{a\in A}\mu(a) \frac{v(a)}{\mu(a)}\right)\\
\prod_{a\in A}\left(\frac{v(a)}{\mu(a)}\right)^{\mu(a)}&\leq \sum_{a\in A}\mu(a)\frac{v(a)}{\mu(a)}=1
\end{aligned}
$$
So
$$
-\sum_{a\in A}\mu(a)\log v(a)\geq -\sum_{a\in A}\mu(a)\log \mu(a)
$$
(This is also known as Gibbs' inequality: Put in words, the information entropy of a distribution $P$ is less than or equal to its cross entropy with any other distribution $Q$.)
Since $T\leq 1$, $-\log T\geq 0$, we have
$$
\sum_{a\in A}m(a)\mu(a)\geq -(\log b)^{-1}\sum_{a\in A}\mu(a)\log \mu(a)
$$
Second, we show that
$$
\sum_{a\in A}m(a)\mu(a)\leq -(\log b)^{-1}\sum_{a\in A}\mu(a)\log \mu(a)+1
$$
By Theorem 1.1.1, there exists an irreducible code $f:A\to S(B)$ such that $l(f(a))=m(a)$ for all $a\in A$.
So
$$
\begin{aligned}
\overline{l}(\mu,f)&=\sum_{a\in A}\mu(a)m(a)\\
&<\sum_{a\in A}\mu(a)\left(1-\frac{\log\mu(a)}{\log b}\right)\\
&=-\sum_{a\in A}\mu(a)\log\mu(a)\cdot\frac{1}{\log b}+1\\
&=-(\log b)^{-1}\sum_{a\in A}\mu(a)\log \mu(a)+1
\end{aligned}
$$
QED
### Entropy
Let $A$ be an alphabet and let $\mathcal{P}(A)$ be the set of all probability distributions on $A$. Any element $\mu\in \mathcal{P}(A)$ is thus a map from $A$ to $[0,1]$ such that $\sum_{a\in A}\mu(a)=1$.
Thus $\mathcal{P}(A)$ is a compact conves subset of real linear space $\mathbb{R}^A$.
We define the map $H:\mathcal{P}(A)\to\mathbb{R}$ as
$$
H(\mu)=-\sum_{a\in A}\mu(a)\log_2\mu(a)
$$
#### Basic properties of $H$
1. $H(\mu)\geq 0$
2. $H(\mu)=0$ if and only if $\mu$ is a point mass.
3. $H(\mu)=\log_2|A|$ if and only if $\mu$ is uniform.
### Huffman's coding algorithm
Huffman's coding algorithm is a greedy algorithm that constructs an optimal prefix code for a given probability distribution.
The algorithm is as follows:
1. Sort the symbols by their probabilities in descending order.
2. Merge the two symbols with the smallest probabilities and assign a new codeword to the merged symbol.
3. Repeat step 2 until there is only one symbol left.
```python
import heapq
def huffman(frequencies):
class Node:
def __init__(self, symbol=None, freq=0, left=None, right=None):
self.symbol = symbol
self.freq = freq
self.left = left
self.right = right
def __lt__(self, other): # For priority queue
return self.freq < other.freq
def is_leaf(self):
return self.symbol is not None
# Build the Huffman tree
heap = [Node(sym, freq) for sym, freq in frequencies.items()]
heapq.heapify(heap)
while len(heap) > 1:
left = heapq.heappop(heap)
right = heapq.heappop(heap)
merged = Node(freq=left.freq + right.freq, left=left, right=right)
heapq.heappush(heap, merged)
root = heap[0]
# Assign codes by traversing the tree
codebook = {}
def assign_codes(node, code=""):
if node.is_leaf():
codebook[node.symbol] = code
else:
assign_codes(node.left, code + "0")
assign_codes(node.right, code + "1")
assign_codes(root)
# Helper to pretty print the binary tree
def tree_repr(node, prefix=""):
if node.is_leaf():
return f"{prefix}→ {repr(node.symbol)} ({node.freq})\n"
else:
s = f"{prefix}• ({node.freq})\n"
s += tree_repr(node.left, prefix + " 0 ")
s += tree_repr(node.right, prefix + " 1 ")
return s
print("Huffman Codebook:")
for sym in sorted(codebook, key=lambda s: frequencies[s], reverse=True):
print(f"{repr(sym)}: {codebook[sym]}")
print("\nCoding Tree:")
print(tree_repr(root))
return codebook
```
#### Definition 1.2.1
Let $A=\{a_1,a_2,\cdots,a_n\}$ be an alphabet and let $\mu=(\mu(a_1),\mu(a_2),\cdots,\mu(a_n))$ be a probability distribution on $A$.
Proof:
We use mathematical induction on the number of symbols $n$.
Base Case: $n=2$
Only two symbols — assign one "0", the other "1". This is clearly optimal (only possible choice).
Inductive Step:
Assume that Huffman's algorithm produces an optimal code for $n-1$ symbols.
Now, given $n$ symbols, Huffman merges the two least probable symbols $a,b$ into $c$, and constructs an optimal code recursively for $n-1$ symbols.
n−1 symbols.
By the inductive hypothesis, the code on $A'$ is optimal.
is optimal.
By Step 2 above, assigning the two merged symbols $a$ and $b$ codewords $w_0$ and $w_1$ (based on
1
$w_1$ (based on $c$'s codeword $w$) results in the optimal solution for $A$.
Therefore, by induction, Huffman’s algorithm gives an optimal prefix code for any $n$.
QED

5
pages/Math401/index.md Normal file
View File

@@ -0,0 +1,5 @@
# Math 401
This is a course about symmetrical group and bunch of applications in other fields of math.
Prof. Renado Fere is teaching this course.