Blog 1

Random Talk on Random Thoughts

Two Diagrams Illustrating the Isomorphism Extension Theorem

| Comments |

Two weeks ago, I had proved that any two algebraic closures of a field are isomorphic to each other in a homework problem. To finish this problem, I opened my note book to view the diagram for the Isomorphism Extension Theorem (IET) drawn before I had understood the proof of the existence of algebraic closure.1

Your browser does not support SVG

Valid SVG 1.1

Drag the figure to translate it, and scroll to enlarge/reduce it.2
Source code: $\rm \LaTeX$, SVG

After I had read E. Artin’s construction of an algebraic closure of a field, I had also read the proof of IET.3 After that, I thought I understood this theorem. However, I couldn’t figure out how to make use of the above diagram to do this question.

\begin{equation*}\begin{CD} @. \overline{F'}\\ @. @AAA\\ E @>\tau>\cong> \tau[E]\\ @AAA @AAA\\ F @>\sigma>\cong> F' \end{CD}\end{equation*}

I can’t add dashed arrow for $\tau$.

I then opened John B. Fraleigh’s A First Course in Abstract Algebra and saw two diagrams which illustrated the IET. In those two figures, there’re only vertical and horizontal lines, no oblique lines were found.

Using these diagrams, I successfully answered this question by drawing a tower of five levels of algebraic extensions.


  1. Read a Proof of Existence of Algebraic Closure on Blog 1. 

  2. When it comes to drawing commutative diagrams, MathJax only supports AMScd, which doesn’t support diagonal arrows. Therefore, I used tikz-cd according to the last sentence of Guide to Commutative Diagram Packages by J.S. Milne to produce a standalone diagram in PDF format first. Then I converted it to an SVG file using the procedures described in the last paragraph in my earlier post about pdf2svg on Blog 1. Finally, I added the dragging and scrolling features to the SVG files after re-reading Zooming SVG in Web Browsers on Blog 1. 

  3. Same as footnote 1. 

A Non-surjective Embedding Mapping a Field to Itself

| Comments |

A week ago, I came up with an injective, but not surjective homomorphism which mapped a field to the same field: $\phi: \Q(e) \to \Q(e)$ defined by $\phi(e) = e^2$ and $\left.\phi\right|_\Q = \id_{\Q}$. It isn’t surjective because $\phi[\Q(e)] = \Q(e^2) \subsetneq \Q(e)$

Obviously, this kind of mapping wasn’t defined on a finite field.

After that, I found another non-surjective embedding which sends field $\Z_p [y]$, where $y$ is an indeterminate, to itself on Mathematics Stack Exchange.1

From this, I’ve understood that why the Isomorphism Extension Theomrem doesn’t apply to transcendental field extensions.


Powers of Roots of Irreducible Polynomials in a Field With Characteristic $p$

| Comments |

Suppose we have a field $F$ of characteristic $p$ and a degree $n$ irreducible polynomial $f \in F[x]$.

\[ f(x) = \sum\limits_{i = 0}^n a_i x^i \text{, where } a_i \in F \,\forall i = 0,1,\dots,n. \]

One can find a root $\alpha \notin F$ of $f$ in a field extension $E$ of $F$ by Kronecker’s Theorem. Then $f(\alpha) = 0$. How about the other $n - 1$ roots of $f$?

Half a month ago, I couldn’t find out the answer directly — I used the fact that all finite fields of order $p^n$ were isomorphic to $\F_{p^n}$, which was the collection of roots of $x^{p^n} - x$ in $\Z_p$ in $\overline{\Z_p}$. However, this makes use of too many abstract facts.

Yesterday night, by computing the $p$-th power of an element $\beta$ in $E$, I finally know the direct way of finding the other $n - 1$ roots of $f$ in $E$. Let $b_0,\dots,b_n \in F$ such that

\[ \beta = \sum\limits_{i = 0}^{n - 1} b_i \alpha^i. \]

Compute the $p$-th power of $\beta$.

\[ \beta^p = \left(\sum\limits_{i = 0}^{n - 1} b_i \alpha^i \right)^p = \sum\limits_{k_0,\dots,k_{n - 1}} \frac{p!}{\prod\limits_{i = 0}^{n - 1} k_i!} \prod\limits_{i = - 0}^{n - 1} (b_i \alpha^i)^{k_i}, \]
\begin{equation} \text{ where } 0 \le k_i \le p \,\forall i = 0,\dots,n \text{ and } \sum\limits_{i = 0}^{n - 1} k_i = p. \label{eq:cond} \end{equation}

Case 1: $\exists k_j = p$, then $k_i = 0 \,\forall i \ne j$.

\[ \frac{p!}{\prod\limits_{i = 0}^{n - 1} k_i!} \prod\limits_{i = 0}^{n - 1} (b_i \alpha^i)^{k_i} = 1 \cdot (b_j \alpha^j)^{k_j} = b_j \alpha^{pj} \]

Note that one can prove that $\forall b \in F, b^p = b$ by induction.

Case 2: $k_i \ne p \,\forall i = 0,\dots,n$. Since $p$ is a prime,

\[ \frac{p!}{\prod\limits_{i = 0}^{n - 1} k_i!} = 0 \text{ in } \Z_p \]

Since one only has $n$ choices of $k_0,\dots,k_{n - 1}$ which satisfy \eqref{eq:cond}, we conclude that

\begin{equation} \beta^p = \left(\sum\limits_{i = 0}^{n - 1} b_i \alpha^i \right)^p = \sum\limits_{i = 0}^{n - 1} b_i^p \alpha^{pi} \label{eq:powp} \end{equation}

Since $\alpha$ is a root of $f$ in $E$, $(f(\alpha))^p = 0$. Replacing “$n - 1$” by “$n$” in the derivation of \eqref{eq:powp}, one gets

\[ (f(\alpha))^p = \left(\sum\limits_{i = 0}^n a_i \alpha^i \right)^p = \sum\limits_{i = 0}^n a_i \alpha^{pi} = f(\alpha^p) = 0 \]

Therefore, without learning induction, one can sense that $f(\alpha^{p^m}) = 0 \,\forall m \in \N$. That’s not the end. Since the degree of $f$ is $n$, we expect to that the number of roots of $f$ is finite. Since we expect $n$ roots of $f$, we hope that $\alpha^{p^m}$ will repeat itself for sufficiently large $m$. This hope comes true due to Lagrange’s Theorem — $|F^\times| = p^n - 1$, so $\alpha^{p^n - 1} = 1$.

Unluckily, I’ve just found out that $f$ has some root $\alpha’$ which doesn’t hit either one of $\alpha,\alpha^p,\alpha^{p^2},\dots,\alpha^{p^{n - 1}}$. For example, if one sets $f(x) := x^{p^2} - x \in \Z_p[x]$ and let $\alpha$ be a root of $f$ in $\overline{\Z_p}$, then it’s trivial that $\alpha^{p^2} = \alpha$, thus the collection of the $p^m$-th power of $\alpha$ is just $\left\{ \alpha,\alpha^p \right\}$. (i.e. $\left\{ \alpha^{p^m} \mid m \in \N \right\} = \left\{ \alpha,\alpha^p \right\}$) Nevertheless, $f$ should have $p^2$ roots in $\overline{\Z_p}$.

Hence, I didn’t succeed in answering the above bolded question, but I still learn something about the roots of an irreducible polynomial in an algebraic extension.


(Edited on MAR 28, 2015)

With the results from perfect fields, we immediately know that $f$ has $n$ different roots.

Installed Jpegtran From Source With MinGW32

| Comments |

Background

To reduce the bandwidth usage while viewing pictures in this blog, I’ve already run OptiPNG on PNG pictures in my new posts.1 For JPEG files optimization, Google developers suggest jpegoptim or jpegtran.2 Since the first tool isn’t available for M$ Win* 7, I installed the second one. I couldn’t find a compiled version of the software, so I had to compile it from source code.

Problem

Since there’re too many words following the instructions for a *nix machine below the table of contents in install.txt, I tried installing it with Make like *nix, so that the installation could be finished in four simple commands: ./configure, make, make test and make install. The output of the first command was fine.3

Unluckily, I got trouble when I moved on to the next command.

$ make
c:/Temp/gnuwin32/bin/make  all-am
c:/Temp/gnuwin32/bin/make: Interrupt/Exception caught (code = 0xc00000fd, addr =
 0x4227d3)
make: *** [all] Error 255

Another Jekyll Error

| Comments |

Problem

Markdown encountered an error when I ran rake generate.

$ rake generate && rake preview
## Generating Site with Jekyll
    write source/stylesheets/screen.css
Configuration file: /home/owner/octopress/_config.yml
            Source: source
       Destination: public
      Generating... 
  Conversion error: Jekyll::Converters::Markdown encountered an error while conv
erting '_posts/2014-01-31-comparison-of-latexs-quote-and-quotation-environments.
markdown/#excerpt':
                    "\xC2" from ASCII-8BIT to UTF-8
jekyll 2.5.3 | Error:  "\xC2" from ASCII-8BIT to UTF-8
Starting to watch source with Jekyll and Compass. Starting Rack on port 4000
Configuration file: /home/owner/octopress/_config.yml
[2015-03-08 15:49:51] INFO  WEBrick 1.3.1
[2015-03-08 15:49:51] INFO  ruby 2.1.2 (2014-05-08) [i686-linux]
[2015-03-08 15:49:51] INFO  WEBrick::HTTPServer#start: pid=5560 port=4000
>>> Compass is watching for changes. Press Ctrl-C to Stop.
directory public/stylesheets
    write public/stylesheets/print.css
            Source: source
       Destination: public
      Generating... 
    write public/stylesheets/screen.css
  Conversion error: Jekyll::Converters::Markdown encountered an error while conv
erting '_posts/2014-04-05-latex-template-for-chinese-essays.markdown/#excerpt':
                    "\xE6" from ASCII-8BIT to UTF-8
jekyll 2.5.3 | Error:  "\xE6" from ASCII-8BIT to UTF-8

Learnt Cauchy–Bunyakovsky–Schwarz Inequality for Definite Integrals

| Comments |

When I was a secondary school student, I was quite satisfied with this proof of Cauchy–Schwarz Inequality.

\[ \begin{aligned} & \left(\sum_{i = 1}^n a_i b_i\right)^2 \le \left(\sum_{i = 1}^n a_i^2\right) \left(\sum_{j = 1}^n b_j^2\right) \\ \iff& \left(\sum_{i = 1}^n a_i b_i\right) \left(\sum_{j = 1}^n a_j b_j\right) \le \left(\sum_{i = 1}^n a_i^2\right) \left(\sum_{j = 1}^n b_j^2\right) \\ \iff& \sum_{i = 1}^n \sum_{j = 1}^n a_i a_j b_i b_j \le \sum_{i = 1}^n \sum_{j = 1}^n a_i^2 b_j^2 \\ \iff& \sum_{i = 1}^n \sum_{j = 1}^n a_i a_j b_i b_j \le \sum_{i = 1}^n \sum_{j = 1}^n \frac{1}{2} \left(a_i^2 b_j^2 + a_j^2 b_i^2 \right)\\ \iff& \sum_{i = 1}^n \sum_{j = 1}^n \frac{1}{2} (a_i^2 b_j^2 - 2 a_i b_j \cdot a_j b_i + a_j^2 b_i^2) \ge 0 \\ \iff& \sum_{i = 1}^n \sum_{j = 1}^n \frac{1}{2} (a_i b_j - a_j b_i)^2 \ge 0 \end{aligned} \]

Equality holds if and only if $\forall i,j = 1,\dots,n, a_i b_j - a_j b_i = 0$. I was so happy that I didn’t think of further generalisations. I didn’t realise that the inequality can be more concisely written as $\langle \vect{a}, \vect{b} \rangle^2 \le \norm{\vect{a}}^2 \norm{\vect{b}}^2$, where $\vect{a} = (a_1,\cdots,a_n), \vect{b} = (b_1,\cdots,b_n) \in \R^n$

This Friday evening, I did a question about the integral version of the inequality. After spending hours to come up with an idea, I realised why I needed to learn inner product spaces. The very first version of the inequality that I learnt has too many summation signs, and it can’t be easily generalised to other spaces. The second proof of the same inequality that I learnt makes use of the determinant of a quadratic polynomial $p(t) = \norm{\vect{u} - t \vect{v}}^2$. That proof is much more elegant, and it helped me a lot while I was doing that question. When equality holds, it’s very hard to imagine what happens by looking at the original equality. However, if we convert it into the determinant of $p(t)$, then one quickly knows that this is equivalent to $p(t) = 0$, and can easily conclude that equality holds if and only if the two functions $f$ and $g$ satisfy $f \equiv k g$ for some $k \in \R$ for almost all points.1


  1. To be more precise, if one accepts that the space $C([a,b])$ of continuous functions defined on a closed and bounded interval $[a,b]$ in an inner product space, and $f,g \in C([a,b])$, then the condition “for almost all points” can be dropped. If we don’t to want be to so strict on the functions $f$ and $g$ defined on $[a,b]$, and we just say that $f$ and $g$ are integrable functions defined on $[a,b]$, then we have to accept the fact that $\int_a^b f^2 = 0$ doesn’t imply that $f \equiv 0$. 

Evaluated a Convolution Integral

| Comments |

Yesterday afternoon, I did a convolution integral to show that a finite sum of i.i.d. $X_i \sim \Exp(\lambda)$ is Gamma distributed.

\[ \begin{aligned} f_{X_{i_1}} * f_{X_{i_2}}(x) =& \int_{-\infty}^{\infty} f_{X_{i_1}}(z) f_{X_{i_2}}(x - z) \ud z \\ =& \int_{0}^{x} (\lambda e^{-\lambda z}) (\lambda e^{-\lambda (x - z)}) \ud z \\ =& \int_{0}^{x} \lambda^2 e^{-\lambda x} \ud z \\ =& \lambda^2 x e^{-\lambda x} \end{aligned} \]

Using the same technique, one can prove the claim by induction. What’s written above is the base case.

\[ \begin{aligned} f_{\sum\limits_{j = 1}^{k + 1} X_{i_j}}(x) =& f_{\sum\limits_{j = 1}^{k} X_{i_j} + X_{i_{k + 1}}}(x) \\ =& f_{\sum\limits_{j = 1}^{k} X_{i_j}} * f_{X_{i_{k + 1}}}(x) \\ =& \int_{-\infty}^{\infty} f_{\sum\limits_{j = 1}^{k} X_{i_j}}(z) f_{X_{i_{k + 1}}}(x - z) \ud z \\ =& \int_{0}^{x} \left (\frac{\lambda^k z^{k - 1} e^{-\lambda z}}{(k - 1)!} \right ) (\lambda e^{-\lambda (x - z)}) \ud z \\ =& \int_{0}^{x} \frac{\lambda^{k + 1} e^{-\lambda x}}{(k - 1)!} z^{k - 1} \ud z \\ =& \frac{\lambda^{k + 1} e^{-\lambda x}}{(k - 1)!} \frac{x^k}{k} \\ =& \frac{\lambda^{k + 1} x^k e^{-\lambda x}}{k!} \end{aligned} \]

Another BoD on Win* 7

| Comments |

Yesterday afternoon, I switched on a public computer. After the desktop had been loaded, a dialog box popped up in the middle of the screen.

</source> got a BoD

I compared this dialog box with the one that I saw on my laptap a month ago.

</source> got a BoD

The difference between these two dialog boxes is the button “Check for solution” on the left of the “Close” button. I don’t know whether this is a difference between M$ Win* 7 Home Edition and Enterprise.

Saw a New Type of Ring

| Comments |

This Thursday evening, I saw a local ring in a commutative ring with unity for the first time. At the first glance, I didn’t know how to make use of it’s definition — a ring that has only one maximal ideal — to answer a question. If I was given a local ring $R$, then should I first assume that there’s a maximal ideal $M$, and then construct another proper ideal $I$ of $R$ so that no proper ideals of $R$ contained $I$ except for $I$ itself? I got stuck at this point for almost an hour, and couldn’t write down something more for that question.

I finally read another half of the question, and understood the statement of the whole question. It would be impossible for me to do the question without knowing the proper ideal that consisted merely of elements of R which don’t have a multiplicative inverse.

After I had proved the statement in that question, I was still not sure whether such a ring exists. I couldn’t give an example of a local ring. This evening, I found it in Wikipedia, and realised that even though I had already convinced myself that a finite field $F$ of prime characteristic $p$ has an algebraic closure $\overline{F} = \bigcup\limits_{i = 1}^\infty \F_{p^i}$, I still didn’t know many properties of the structure of fields.1 The simplest example of a local ring is a field, which has only the trivial ideal $\{0\}$ as its proper ideal.