Monday, May 21, 2018

[物理]申請物理博士班心得applying for Physics Ph,D. 2018



看很多同學今年都申請上啦


不免俗的我也來寫一下上的心得 applying for Physics Ph,D. 2018 雖說本人不是很強,但也希望可以給大家一些參考:

Admission(ta+fellowship)
UT Austin physics Ph.D

U Maryland Physics Ph.D

Rejection:

太多常春藤不好說

Princeton

Harvard

Stanford

Cornell

Yale

MIT

Johns

UPENN

UCLA

UCSB

STONY BROOK

UCB

.....太殘酷

Decision:

U Maryland Physics Ph.D

Education:

2012.9-2016.01 (大學三年半 早畢業半年)
National Chiao Tung University – B.S. in Electrophysics
GPA (4.30 point system): Overall - 4.18
2016.02~2017.06 (碩士一年半)
National Taiwan University – M.S. in Physics
GPA (4.30 point system): Overall - 4.17


Paper and honor:

申請時候,有兩邊published,一篇大學,一篇是碩班發的,期刊是 Jhep and PhyslettB。
申請結束前,有一篇put on arxiv,但沒有寫在cv。這裡都要感謝老師


Work Experience:
沒有,我最愛的工作是替代役: alternative military service. 說實在,比幾年下來奮鬥讀書輕鬆多了。







Test Scores:


GRE General Test V 153 /170 (61 %), Q 170 /170 (97%), AWA 3.0 /6.0 (18%)

GRE Subject Test Physics -990/990(94%)

TOEFL (iBT) R 29/30, L 29/30, S 22/30, W 20/30, Total 100/120
TOEFL 是在入伍之前匆忙考試 gre 是在專訓完以後兩個禮拜,整體是非常聰忙得,只有subject準備比較充分,整體還算可以,這些不多說了, 畢竟分數也不高,大家上網可以查到非常多 如何解決這些考試。

我個人建議toefl比gre重要 如果真的只能花時間準備一個,那就是托福盡量100,因為有些 學校會有托福門檻,如果門檻沒過,會連申請都無法申請:
還有博士班,美國人較在意speaking session

準備申請的物件:


1 Sop:最難搞的東西,因為一般人申請博士班,,除了前面大學gpa成績表現,修課紀錄等等,加上英文成績這都是死的,paper 也是,所以在申請前多半是 已經被決定,無改變空間了,所以 唯一能炫能秀能嘴砲的就是sop!!!
  1. 建議是要列出想跟的教授,不用太多,列幾個當意思意思,很有可能學校就是找這幾個人來看你的東西 
  2. 要言簡意賅,高中以前就隨緣往事不必提,生活瑣事不要提,直接快速點出你為何想讀物理,不要說太多老梗,什麼物理可以了解世界,物裡之美還啥的,徒然浪費空間 
  3. 偶爾提自己做過的研究經驗也很好 
  4. 有機會可以找其他人或是指導教授幫你看看,或是直接拿學長姊的來修,或是給伊些純正老美看看,因為畢竟我們華人角度,很難知道美國人是否真的覺得這sop很吸引人.


2 Curriculum Vitae:



簡稱CV,就是個履歷,我個人是寫,讀書背景哪個學校,Gpa,得過書卷獎等等,然後list research experience,paper publication,如果有興趣,可以加上課外經驗,社團或是領導活動,如果自己有個人website是更棒,或是你有paper researchgate的account,簡明扼要,兩頁之內,這個基本上也是死的資料。如果大家熱愛latex, 其實可以看overleaf 有很多很好的template可以參考。


3 推薦信: 

博士班三封,碩士班兩封到三封,碩士基本上會找指導教授,除非你鬧翻,如果不足三個,應該找修課成績或是對你有印象的老師,盡量是跟申請相同領域,然後國外都是把連結給老師,系統會問你是否想看老師推薦信,你記得寫waved,不必看老師如何描述您,這樣推薦信才有公信力。
兩個想法:如果有機會,某些學校推薦信可以到六封,這時要多給也很好,找一些課外教授,都有幫助,多也不會有損失,盡量找好找滿。

推薦信不用太care內容或擔心找德教授不夠大咖,除非你教授超級強,像是丘成桐之類,否則不太可能只靠推薦信就讓你去強校,只要教授對你滿意就可以,那種如果真的能靠推薦信就把你打敗的,就算你教授寫你是神人,我覺得也沒啥差別,所以這方面,要準備,但放寬心

,記得上了要感謝教授。


4是否聯絡美國教授?


不重要,我沒這樣做...理論上也是可以,但是ptt上,可以看到很多人都沒有connect,也沒啥關係,因為你到美國那裏,搶教授是很正常,不太可能會有人現在就指名要收你。


5選校?


據傳,博士班很多是會考量學校的funding,如果學校錢很多,可能會聘很多人,如果學校沒錢,今年位子很少,你可能就沒辦法上,不是你不夠優秀,有時候你想跟的領域今年教授收不了你,那就會比較可惜,及時你很優秀,還是可能會沒位子,所以申請學校跟太多因素有關,我們理解的資訊量太少,很難想像大陸新加坡日本南韓俄羅斯美國當地人的申請人強度是如何?

所以多半還是隨緣,盡力申請,別留遺憾,很多事情也不是偶們想怎樣就怎樣,一邊當替代役一邊申請是很痛苦的,六日也無法休息,有時候自己也茫了。


最後只上兩間,馬里蘭物理較好,環境較好,我也喜歡quantum computing,有時候人生或許當下不是最好的選擇,以後也很難講,我曾經想是否要再申請一年,賭看看更強的常春藤名校

,畢竟被拒絕十幾間,但是回首想來,自己已經累了,身心俱疲,而且要再拖一年,人生何必如此,當年我上交大,也是這樣,或許也有不同風景,沒必要去想遠在地球一端是否有自己的路,太遠了看不到,就一切隨緣。















Sunday, May 20, 2018

[Mathematics] Note for Riemann Hypothesis

intro for Riemann Hypothesis

Abstract

Since I am interested in Riemann Hypothesis. I want to write a note regarding this topics from today. The good reference is Lectures on The Riemann Zeta–Function 2018/5/21.

There are infinite Primes

Definition for \( Re s>1 \)

$$\zeta(s)=\sum_{n=1}^{\infty}\frac{1}{n^s}=\prod_{primes}\frac{1}{1-\frac{1}{p^{s}}}$$ We get $$log(\zeta(s))=-\sum_{primes}log(1-\frac{1}{p^{s}})$$ Use taylor expansion: \( log(1+x)=\sum_{i=1}^{n}\frac{x^n}{n}\) We get $$log(\zeta(s))=-\sum_{primes}\sum_{m=1}^{\infty}\frac{1}{m*p^{ms}}$$. We then back to the definition, first we noticed that $$ \zeta(1)=\sum_{n=1}^{\infty}\frac{1}{n}=\infty $$ so $$ \sum_{primes}\sum_{m=1}^{\infty}\frac{1}{m*p^{m}}=\infty$$ We write: $$ \sum_{primes}\sum_{m=1}^{\infty}\frac{1}{mp^{m}}= \sum_{primes}\sum_{m=2}^{\infty}\frac{1}{mp^{m}}+\sum_{primes}^{\infty}\frac{1}{p}=\infty$$ Now I claim $$\sum_{primes}\sum_{m=2}^{\infty}\frac{1}{mp^{m}} \text{ converges }$$ This is not hard since $$\sum_{primes}\sum_{m=2}^{\infty}\frac{1}{mp^{m}}=\sum_{primes}\frac{1}{2p^{2}}+\frac{1}{3p^{3}}... $$ Since $$ \frac{1}{2p^{2}}+\frac{1}{3p^{3}}...\leq \frac{1}{2p^{2}}+\frac{1}{2p^{3}}...=\sum_{primes} \frac{1}{2p^2-2p}, $$ By the p-series test, we know it converges. So I know: $$\sum_{primes}\frac{1}{p}=\infty $$ which leads to there are infinite primes.

Prime number Theorem

From the above discussion, we know that zeta function has deep connection to the prime number. But can we say more about prime numbers via prime number theorem? The answer is yes. $$-\frac{\zeta'(s)}{\zeta(s)}=\sum_{p} \frac{p^{-s}log(p)}{1-p^{-s}}=\sum_{p}\sum_{m=1}^{\infty}\frac{log(p)}{p^{ms}}$$ Let $$\Lambda(n)=log(p) \text {if } n=p^k \text{ for some positive k, otherwise} =0$$ $$-\frac{\zeta'(s)}{\zeta(s)}=\sum_{n=1}^{\infty}\frac{\Lambda(n)}{n^s}$$

functional equation formula and analytic continuation

step 1

In order to study this function more carefully, one need to analytic continuation on Riemann Zeta function: What? So we need to extend the definition of this function: first we try to extend it from $$ Re s>1 \to Re s>0 $$. Now we find $$ \zeta (s)=\sum_{n=1}^{\infty}\frac{1}{n^s}=\frac{1}{1}+\frac{1}{2^s}+\frac{1}{3^s}...$$ $$ \frac{1}{2^s}\zeta (s)=\frac{1}{2^s}+\frac{1}{4^s}+\frac{1}{6^s}...$$ $$ \zeta (s)-\frac{2}{2^s}\zeta (s)=\frac{1}{1}-\frac{1}{2^s}+\frac{1}{3^s}-\frac{1}{4^s}...$$. We find if we define for \( Res>0 \) $$\zeta (s)=\frac{1}{1-2^{1-s}}\sum _{n=1}^{\infty } \frac{(-1)^{n-1}}{n^s}$$. This series is convergent for \( Res>0 \) since it is alternating series. But this is not enough.

step 2: extend to all complex plane

functional equation formula and analytic continuation

We can define the functional equation $$\pi ^{-s/2} \zeta (s) \Gamma \left(\frac{s}{2}\right)=\pi ^{(1-s)/2} \zeta (1-s) \Gamma \left(\frac{1-s}{2}\right)$$ Why? the reason is complicated and we need techniques in complex analysis. Before we start, we first talk about gamma function: Usually, we define: $$\Gamma(s)=\int _0^{\infty }t^{s-1}e^{-t}dt$$ We know it converges when \( Re s \)>0. since we can write $$\Gamma(s)=\int _1^{\infty }t^{s-1}e^{-t}dt + \int _0^{1 }t^{s-1}e^{-t}dt $$.

Analytic continuation of \( \Gamma(z) \)

Consider the following contour integral: $$I(z)=\int _Ce^{-t}(-t)^{z-1}dt$$ This is so called Hankel contour This is very common contour dealing with the complex analysis. So now we tried to evaluate this integral along this coutour.

Lemma: Perron's formula

See Perron's formula

zero of Zeta function

we write $$ s=\sigma + it$$

first case :\( \sigma \geq 1 \)

if \( s \) is a real number and \( s \geq 1 \) definite \( \zeta(s)>1+... \) definitely nonzero. But what if \( s \) is complex number?

second case :\( \sigma=1 \)

This is not so easy, actually there are no zeros on \( \sigma=1 \) due to the following proof.

proof

we know $$ log(\zeta(s))=\sum_{p,m} \frac{p^{-ms}}{m}$$ we can write it as: $$ |log(\zeta(\sigma+it))|=|\sum_{p,m} \frac{p^{-ms}}{m}| $$ so $$ |(\zeta(\sigma+it))|=exp(\sum_{p,m} \frac{cos(mtlogp)}{mp^{m\sigma}})$$ here we use: $$ p^{ms}=exp(mslog(p))=exp((m\sigma+imt)log(p))$$. Now we consider the following $$3+4cos(\phi)+cos(2\phi)=2(1+cos(\phi))^2 \geq 0$$ we get $$|\zeta^{3}(\sigma)||\zeta^{4}(\sigma+it)||\zeta(\sigma+2it)|=exp(\sum_{p,m} \frac{(3+4cos(mtlogp)+cos(2mtlogp))}{mp^{m\sigma}}) \geq 1 $$ Since $$(3+4cos(mtlogp)+cos(2mtlogp) \geq 0 \text {for all p, m }$$ for all \( \sigma \geq 1 \). Suppose one that \( \zeta(1+it)=0 \), then approach this point from \( \sigma>1 \). This will contradict the above identity.

third case :\( \sigma \leq 0 \)

We know from the functional Equation. We know we can write $$\zeta(s)=\frac{\pi ^{(1-s)/2} \zeta (1-s) \Gamma \left(\frac{1-s}{2}\right)}{\pi ^{-s/2}\Gamma \left(\frac{s}{2}\right) } $$ This can be computed except for poles in gamma function. The numerator is nonzero obviously since gamma function only has poles but no zeros and zeta function are all greater than zero. so $$ \Gamma \left(\frac{s}{2}\right) $$ has pole when $$s=-2n \text{, n is integer} \geq 0$$. These zeros are called trivial zeros.

Conjecture (Riemann Hypothesis)

If we write $$s=\sigma+it \text{ both } \sigma \text{ and } t \text{ are reals } $$ The non-trivial zeros of \( \zeta(s) \) have real part only when $$\sigma= \frac{1}{2} $$
. Here blue line are real part of Riemann zeta along(1/2). Red line are imaginary part of Riemann zeta along(1/2). We can see the first zero at imaginary between 14 and 15.

ok, who cares?

Von Mangoldt’s Explicit Formula

we define: This is a function related to the distribution of prime numbers. The identity is:( \( \rho \) is a zero in the critical strip of zeta function), for \( x >1 \): $$\Psi(x) =x-\sum_{\rho}\frac{x^{\rho}}{\rho}-\frac{1}{2} log(1-\frac{1}{x^2})-log(2\pi)$$. This is too sophisticated to this blog. But, we can still make argument. The key identity is ( \( \rho \) is a zero in the critical strip): $$-\frac{\zeta'(s)}{\zeta(s)}=log(2\pi)+\frac{s}{1-s}+\sum_{n=1}^{\infty}\frac{-s}{2n(s+2n)}+\sum_{\rho}\frac{s}{\rho(s-\rho)}$$

key: Mellin tranformation:

definition \( Re s > 0 \): $$ M\{f(s)\}=\int_1^{\infty }f(s)x^{-1-s}dx $$. Now we are able to proof this identity in the following step: we already know that : $$n^{-s}=s\int _n^{\infty }x^{-s-1}\text{dx}$$ $$\frac{\zeta'(s)}{\zeta(s)}=-\sum_{n=1}^{\infty}\frac{\Lambda(n)}{n^s}=-\sum_{n=1}^{\infty}s \Lambda (n)\int _n^{\infty }x^{-s-1}\text{dx}=-s\int _1^{\infty }\sum _{n\leqslant x} \Lambda (n)x^{-s-1}dx$$ This is a little bit tricky, since \(x \text{is greater than} n \). we can replace $$ \sum _{n\leqslant x} \Lambda (n)=\Psi(x) $$ Finally, we have $$ \frac{\zeta'(s)}{\zeta(s)}=-s\int _1^{\infty }\Psi (x)x^{-s-1}dx$$ Than we can see $$ \frac{\zeta'(s)}{\zeta(s)}=-s M\{\Psi (s)\}$$. Now we can easily transform the rhs of $$-\frac{\zeta'(s)}{\zeta(s)}=log(2\pi)+\frac{s}{1-s}+\sum_{n=1}^{\infty}\frac{-s}{2n(s+2n)}+\sum_{\rho}\frac{s}{\rho(s-\rho)}$$
Combine four of them. we finally get our results.

Reference

Riemann_hypothesis in wiki
THE ZETA FUNCTION AND ITS RELATION TO THE PRIME NUMBER THEOREM
Riemann’s Zeta Function and the Prime Number Theorem
A Primer of Analytic Number Theory: From Pythagoras to Riemann

Saturday, May 19, 2018

物理(physics) Gre 990 滿分心得

gre Physics 成績單
Subject physics 990 tips
大概是前年暑假 to applied my physics Ph.D. I have to attend the Gre test physics in Taiwan.
那時剛到台大物理讀碩士沒多久,大概七月上GRE網站報名10月考試,準備期間大概兩三個月,基本上PHYSICS GRE 考得大多都是大一普通物理,大二電磁學大三量子物理,這些東西,偶爾會考一些他認為是常識的東西,像是核物理,或是粒子物理,問你QUARKS, LEPTONS' property之類,這些基本是要背的0.0
ets的一份試題


考試簡介: 除了幾百塊美金要用信用卡繳費以外(沒事情就會漲價哦)考試形式為70到200道單項選擇題,於170分鐘內完成,依據不同科目而定,有物理,化學,生物,英語文學,數學,心理學等等,
物理有100題,你一題沒有兩分鐘的時間,所以要秒殺,看到不會要跳過,很多東西要背。



  1. 線上資源 :敝人是讀高能理論物理,所以在準備方面上,除了複習普物之外,非常建議大家去坊間買那種物理公式的書籍,或是網路上我推薦一些美國大學的lecture note for preparing 這些是非常好的書籍,當然頁數比較多,你多看會發現,他會考你一些公式運用,都是那種很簡單但是你如果忘了公式,就會不知道怎麼做,不是那種大二大三電磁學考試方法,因為時間不夠
  2. 考古題,網路上有五份gre物理考古題 考古題:考古題下載
  3. ,這我不用多說了,然後你需要花時間找解答,一定要把五份全部做完,基本上會比你當天考試還要困難不少,至少我個人覺得。可看: Gre 物理 解答 (大家一起討論的論壇)
  4. 單位跟因次轉換一定會考
  5. 必考光學跟麥克生實驗,還有氫原子模型,還有轉動慣量,這些是我個人經驗
  6. 最後考試記得帶護照,如果一次不能990,也不用氣餒,計算分數方法是依照pr值,所以不用擔心運氣不好題目太難,題目太難也頂多是大家一起死得很難看,現在聽說沒有倒扣了,所以大家可以亂猜惹
  7. 以我申請美國學校經驗來講,亞洲人(特別是歹玩)考試自然一流,日韓中國台灣990人不計其數,我學長至少就有10人以上聽過可過990,我能一次考倒也算是運氣,考990大概要錯10-12題以內,但是如果去 gradcafe查一下,就會知道美國很多物理博士生都只會考8xx 就有很多好學校念,反觀華人,考990,可能人家還是不願意收你,當然盡量考高是最好,免得自己以後留下遺憾
  8. 祝願意考physics gre 的人,都能夠以好學校念且都990。



考試部分:

經典力學:(運動學、牛頓定律、功和能、定軸轉動、粒子系統動力學、質心運動、天體力學、三維粒子動力學、拉格朗日和哈密爾頓公式、非慣性參考系、流體力學初步)

電磁學:(靜電學、電流和直流電、自由空間磁場、洛倫茨力、電感、麥克斯韋方程組和應用、電磁波、交流電、物質裡面的電磁場)

光學和波現象:(波的特征、疊加、幹涉、衍射、幾何光學、偏振、多普勒效應)

熱統:(熱力學定律、熱力學過程、狀態方程、理想氣體、動能理論、系綜、統計學概念和熱力學量的計算、熱膨脹和熱傳導)

量子力學:(量子力學的基本概念、薛定鍔方程(包括方勢阱、諧振、氫原子)自旋、角動量、波函數對稱、基本擾動理論)

原子物理:(電子的特征、玻爾模型、能量量子化、原子結構、原子波普、選擇規則、黑體輻射、x射線、電磁場裡面的電子)

狹義相對論:(初步概念、時間膨脹、長度收縮、同時性、能量和動量、四維矢量、洛倫茨變換、矢量疊加)

實驗方法:(數據和錯誤分析、電器、設備、檢測輻射、數量統計學、帶電粒子和物質的作用、激光和光學幹涉儀、量綱分析、概率和統計的初步應用)

其他:(核物理和粒子物理(核的特征、衰變、裂變和聚變、反應、基本粒子的特征)、固體物理(晶體結構、x射線衍射、熱特性、金屬的電子理論、半導體、超導體)、其他(天體物理、數學方法、計算機應用))
如果你覺得 這文章對你有幫助的話: 可以點一下網站上的廣告(在右邊或是上面),我非常需要您的支持喔。

任何連結 含有廣告 都沒有惡意 畢竟 鬼島大家生活難過
版主也要奶茶跟把妹 


Online youtuber solution session:
有時候找不到解答
可以看一些人解給你看

網路上解答的 list:

Gre physics: Gre subject 0177 solution list: https://www.youtube.com/playlist?list=PLg9w7tItBlZs2I_yHrSdCITxZhJXYHGFM Gre subject 9677 solution list: https://www.youtube.com/playlist?list=PLg9w7tItBlZvW82UU2LQLynXdzPwF_ACm Gre subject 1777 solution list: https://www.youtube.com/playlist?list=PLg9w7tItBlZuYk3e_gmtGbTzHepusozYO Gre subject 0877 solution list: https://www.youtube.com/playlist?list=PLg9w7tItBlZsoor9zcHjEOnB5JYQjwUvi Gre subject 8677 solution list: https://www.youtube.com/playlist?list=PLg9w7tItBlZt1QFJRVC3XIk84e9pABQTY Gre subject 9277 solution list: https://www.youtube.com/playlist?list=PLg9w7tItBlZuG6e914nTMZpKN9yQnYgEQ Gre math: Gre subject 9768 solution list: https://www.youtube.com/playlist?list=PLg9w7tItBlZsvmw4WTAxvlbpzu1F_npex Gre subject 8767 solution list: https://www.youtube.com/playlist?list=PLg9w7tItBlZuCRSqoF5av7XFHuJ20sWoc Gre subject 0568 solution list: https://www.youtube.com/playlist?list=PLg9w7tItBlZsbq99hvsvh8ikrUjxZ9ePu Gre subject 1769 solution list: https://www.youtube.com/playlist?list=PLg9w7tItBlZsOcyFzf69sRycW40hYC3hu Gre subject 9676 solution list: https://www.youtube.com/playlist?list=PLg9w7tItBlZtB4NcyoHIyVrDEryHWpaBz





附上一個我考了 subject math gre 的成績
數學比較難 我這個業餘數學愛好者 也混到了95%

物理 94% 數學 95% 也算對得起理學院 
人生 儘管他們發不了光 沒有人愛 
心裡的眼淚 模糊了視線 妳已快看不見













當初考完試發在臉書的心得 2016/10/29

今天去考 subject gre physics 物理科人最多 14x人去考 我看到 研究室就派出五個高手 +上台大物理面熟的同學+助教們 三間教室擠滿滿的 好多好多人 我看著走廊最後的數學科 一間都都不滿

我突然不解了 比方物理台大申請 收3x個 來15x-16x 數學收15個 來24個 台灣太多人要搞物理了 14x來考這個人 都想出國 都要奮鬥 考試170分鐘超級累人的 100題 拚速度和正確率 寫到頭痛頭暈 考試100分鐘後 我眼已經花了 我好想 像以前周六 去拍展場 我想起我國中在拚pr99 對付一堆人 高中建中神人一大堆 還有交大 台大碩搬出去的美國top10的學長們 good guy們
還要對付多少人 多少競爭 才可以求生 而且目標是滿分 沒到標準 我想靠這個吃飯有這麼難嗎 可現實就是這麼難

你們要贏 去吧 我在這裡輸了 也夠了 從國中鬥到現在 真的累了
我們沒錢離開鬼島 就只能這樣 那乾脆輸了 小確幸 看直播 收集人面怪鳥 我常常不懂 我不懂 我到底是想輸還要贏 或許輸了 是更好的 我早有打算 輸了 當完兵 我就把大學存了錢拿去買東西 然後 我就把很多東西丟掉 我不想知道了 然後 忘了教育忘了我

就跟國小一樣 夢想太沉重 我不想要了 交給大安 升學主義大隊長們好了 我還有一些獎狀可以放在家裡自我安慰






****


鳥巢燈光設計師宗南 上節目說 他自己兒子今年考上中央!!!!!!!!! 好學校不錯 可是他 還是把他 小孩送去mit讀書 因為高薪 覺得"歹玩"平台不夠好
還說 留下來的只有幾種人 一種本事不夠 沒辦法出國 一種我家窮
沒辦法讓我出國 一種我有父母要養 一個是弱勢族群 .....

讚聲大國民

我今天去考物理subject gre 三間教室 來了14x多人 可見這群人大概以後都是要申請國外碩博的 我前段時間覺得很痛苦 因為人太多 他媽的太多人要讀物理要出國 要競爭 可我最後在考170分鐘的最後 迴光返照
我發現 我只是延平小咖
第一不夠優秀 第二我家不是宗南 友柏 勝文 大偉 季剛
第三 我確實有父母要養沒兄弟姊妹
第四 還好我不是弱勢族群 還可以靠自己讀點書
你們這些菁英去吧 把我留在這裡餓死 我還撐得下去的.....只要一直 迴光返照就好




收到成績的心得:
WOOOOO 今天,最高興的事情是 之前考的物理SUBJECT GRE成績出來啦滿分!!! 990 OMG 這考試是 ETS辦的想去美國要讀碩士博士就要考科目有物理化學數學等等要讀理學院一般要考專業科目
~~~ 台灣一年只有一場歐是要憑護照進場據說有不少人考不好就去日本南韓再考 考一次要150美金OMG
算分方式就是全世界都考完以後 一樣的考卷然後計算原始成績算PR值來 比方PR94以上 就是990滿分 (意思是100人你考試成績是前六個人啦) 其他資訊一概不知道...不公布
基本上考920以上就不錯囉 ~~ 大家可去P版查準備方法0. 偶是隨緣考的
100題 170分鐘 會滿有壓力一題五個選項考大一到大三基本的物理偶爾加點數學+各式各樣奇怪領域但也不算難可臨場有時真的靠真本事天賦我估計大概錯15題以內就有機會 雖說出國的前高能學長們五六個都說990簡單是基本但也不是真的那麼容易啦能否出國還要看其他資訊和緣分負面想法是這成績可以保存五年但可印出來存檔以後沒出國成功至少曾經盡過力享受過一些東西當作回憶吧忘了聯考忘了我。





[旅遊]澎湖沙灘路跑紀實

澎湖沙灘路跑

緣起

在一次偶然的機會之下, 因我老爸邀請, 去了澎湖的路跑在2018/ 4/13-4/15 沙灘路跑, 部分應該算是完玩票性質, 因為沙灘路跑主要跑沙灘, 腳會陷下去, 所以就會非常辛苦, 路跑時間是4/14號, 所以前面跟後面的時間, 就打算來騎車玩澎湖本島,一開始先來承租摩托車, 兩個人一台就很夠了唷

第一天

因為第一天中午才到本島, 所以就先去距離馬公市區稍微遠但不是很遠的地方玩, 去玩之間先填飽肚子XD 先來吃小管麵線: 這家就在馬公市區附近:

Thursday, May 17, 2018

[Mathematics]Cohomology and homology

Cohomology 和 homology 學習心得

緣起

之前偶不是數學系的,但對homology and cohomology 相關理論覺得很有興趣,且大學也修過分析代數相關課程,加上自己之前要讀一些string theory 相關的topics, 去過台大數學系修過一些Algebraic topology, 覺得有不少心得, 如果推薦open course, 可以看齊震宇教授的Youtube 影片
個人覺得非常硬, 需要很多代數知識, 如果完全沒有接觸過的話,可以看一個國外老先生教授的介紹, 是稍微簡單的, 而且喜歡用比較直覺的方式和畫圖來理解,

課本部分, 經典款式看 Allen Hatcher 的 algebraic topology 這本書也偏難, 但是中間部分可以連結到 sheaf之類(太難XD) 本書也是上網可以下載的唷(free download) 我在這文章指些微討論 cohomology 相關的東西, 也算替自己保留一些note, 有錯誤歡迎討教

definition

要搞cohomology/homology 在代數拓撲裡面, 是研究simplicial cohomology, general cohomology 牽涉到 exact sequence, 先介紹一下 exact sequence的概念. $$ 0 \overset{d}{\rightarrow }C_{0} \overset{d}{\rightarrow } C_{1}.....\overset{d}{\rightarrow } C_{n}\overset{d}{\rightarrow }0 $$ 滿足$$ d^2=0 \text{ for all abelian groups or vector spaces} C_{n}$$ 所以常識告訴你: $$Im(C_{n-1} \overset{d}{\rightarrow } C_{n}) \subset ker(C_{n} \overset{d}{\rightarrow } C_{n+1})$$ 這式子就告訴你, image 從左邊過來的, 是右邊那個 group or vector space的 subgroups or subspaces, 所以自然而然 想到定義: $$ H^n(C^{\cdot })=\frac{ker(C_{n} \overset{d}{\rightarrow } C_{n+1})}{Im(C_{n-1} \overset{d}{\rightarrow } C_{n})} $$

Wednesday, May 16, 2018

Intro: Integer factoring using order finding algorithm

= Integer factoring using order finding algorithm

Article By: En-Jui Kuo


What is a prime number?

In elementary school, we all know that the definition of prime numbers: A positive integer greater than one and it can't not be factorized into two integers both greater than one.
A natural number greater than 1 that is not prime is called a composite number. Prime number :2, 3, 5, 7, 11, 13,.....

Old story

Ancient Greek mathematics Euclid already knew that there are infinite prime numbers and any composite number can be factorized into a product of primes uniquely(up to their order). But how can we factorize a composite number in practice?

a simplest way

A simplest way is that we first prepare a list of primes: 2,3,5,..... . When you give me a composite number, then I test which primes can divide that composite number. For example, we want to factor 91, then we find 91 is divisible by 7. Then we use same algorithm on quotient until quotient is 1. So we find
$$ 91=13*7 $$

Theorem 1

A number N is composite number if it is divisible by a prime number which is smaller or equal to $$\sqrt{N}$$ Proof:
Suppose N can be factor into two numbers (NOT 1*N OR N*1): We can write $$N=a*b$$ Then either one of them is divisible by prime number which is smaller or equal to $$\sqrt{N}$$. Otherwise if both can not be factorized then:
$$a>\sqrt{N}, b>\sqrt{N}$$ then a*b> N.
We lead to a contradiction.
So if we want to factorized a composite number we need to prepare a list of primes smaller or equal to $$\sqrt{N}$$. Let $$\pi(x)$$ (called prime-counting function)denoted a function that gives the number of primes less than or equal to x. For example: $$\pi(10)=4 , \pi(11)=5$$. Since 2,3,5,7 are primes lower than 10 etc.
We have a bad news below:

Theorem 2: Prime Number Theorem

Roughly speaking(ln(x) means natural log): $$\pi(x) \approx \frac{x}{ln(x)} $$ So if you want to factorize a number which has 100 digits, there are approximately 10^97 primes since $$\pi(10^{100}) \approx \frac{10^{100}}{ln(10^{100})} \approx 4.34294\times 10^{97}$$. There is no hope to factorize this kind of large numbers using this stupid way. By the way, Prime Number Theorem is not easy to proof. There are famous conjectures related to prime-counting function which are not solved until now. You can see Prime-counting_function and Second Hardy–Littlewood conjecture The error term is related to a problem belonging to seven millennium prize problems called Riemann Hypothesis. I am very interested in this topics.

Order finding Algorithm

Theorem 3

Given an natural number a greater than 1 and coprime to N. Then there is an natural number x (called order) such that: $$a^{x}=1(\mathrm{mod} N)$$ proof: Simple group theory: $$ \text{consider } a^{1}(\mathrm{mod} N), a^{2}(\mathrm{mod} N),...a^{N}(\mathrm{mod} N), $$ We only have N-1 possibilities in the reminder (0 is impossible since a is coprime to N): But we have N values. So two of them are equal said $$\alpha ,\beta$$ such that $$ a^{\alpha}(\mathrm{mod} N) = a^{\beta}(\mathrm{mod} N) \text { lead to } a^{\alpha-\beta}=1 (\mathrm{mod} N))$$. We finish the proof.
Besides, we can find smallest natural number x which satisfy the equation. Since natural number has a lower bound 1.
By this theorem, we have an algorithm. pick any a which is coprime to N. We can find this x. If x is an even number then $$a^{x}= 1(\mathrm{mod} N)$$ so $$(a^{x/2}-1)(a^{x/2}+1)=0(\mathrm{mod} N)$$ There's a chance $$gcd(a^{x/2}-1, N) \text{, (gcd means the great common divisor)}$$ is not 1 or N.

Example

There is an example below using python: Suppose we want to factorize 606858167: pick a=2, first we can use brute force to find order: we find $$2^{50567308}=1 (\mathrm{mod} 606858167)$$ then we find: $$gcd(2^{\frac{50567308}{2}}-1, 606858167)=19759$$ Finally we get $$606858167=19759*30713$$ Both of them are large primes.

More fancy way factoting large integer?

Yes, we have quantum algorithm called Shor's algorithm. I will spend a small paragraph trying to explain this idea.

quantum computing

the difference of quantum computing and classical computing is that classical bit can be changed to vector $$ |0>, |1> $$ one can see this lies on complex planes: and a bit can exhibit on the superposition state $$ a|0> +b|1> $$. which a, b can be complex number. So you can think that quantum computer is like some algorithm and devices which can manipulate the quantum states and finally measure the quantum state which could give us the answer. but giving a successful quantum algorithm is not so easy since measuring states will cause the state to the eigenstate of that measurement. Suppose we have a state in our hand called $$\phi= a|0>+b|1>$$ and we don't know the coefficient. If we measure it, it will have some chance becoming state 0 or state 1 then we lost our coefficient a and b.

Tensor product

It is hard to explain the math detail here. But we can view tensor means different place in the biniary place. i,e we view: $$01 \to |0 > ;\mathop{\otimes}|1> ; $$ a computing means that we can construct some black box (function f)can manipulate state: $$ |x>|y> \to_{f} |x>|y+f(x)> $$. Now, we can start to illustrate this idea:

shor algorithm

first, we pick up a number called "a" before. then now we have quantum computing:
  1. create the superposition state $$ \frac{1}{\sqrt{2^n}}\sum_{0}^{2^n-1}|x> |0> $$
  2. suppose we have an operator satisfy $$ f|x> |y> \to |x> |a^x+ y(mod N)>$$
    , so now, we have $$ f \frac{1}{\sqrt{2^n}}\sum_{0}^{2^n-1}|x> |0>= \frac{1}{\sqrt{2^n}}\sum_{0}^{2^n-1}|x>|a^x(mod N)>$$ so actually if n is greater enough, the order must include in some $$ |a^x(mod N)>$$ then we measure the second qubit, suppose r=order, then we have $$|x0>+|x0+r>+|x0+2r>+|x0+3r>.... ...$$ I drop the normalization factor now. Finally, it seems like we are done since we can read the coefficient in the state to get our order and apply the order finding algorithm to solve this problem. However this is not true since we can not know the answer until we measure it. Finally, we need a devices called quantum fourier transform which can be seen at Quantum Fourier transform to extract the number.

[物理]CFT 簡介

little Introduction to conformal field theory conformal field theory (中文又稱作保角場論)是一種近代在高能物理和數學各方面都有重要應用的理論, 本文主要是想對這樣的理論做一個粗淺的介紹(盡量用中文), 更多lecture note 關於 conformal field theory 可以在 arxiv 找到。 若有任何錯誤或想討論可以告知。

What's CFT

A conformal field theory (CFT) is a quantum field theory that is invariant under conformal transformations. In two dimensions, there is an infinite-dimensional algebra of local conformal transformations, and conformal field theories can sometimes be exactly solved or classified. Conformal field theory has important applications[1] to condensed matter physics,statistical mechanics, and string theory. Statistical and condensed matter systems are indeed often conformally invariant at their thermodynamic or quantum critical points. (quote from wiki)
由此可見,CFT一個具有保角對稱的量子場論,這個不同於一般的Loretz場論和標準粒子模型(like QED), 這有什麼好處呢?首先就是因為對稱性加強了,所以有更多的計算是可以算的,cft裡的state可以被分類成兩種, 一種叫做primary operator,一種叫做descendent operator,primary operator決定了descendent operator, 所以在cft裡面,所有的field content就是它具有那些primary operator。 以下文章只focus在三維以上的保角場論(因為二為時候的conformal group 是 infinite generator 的 virasoro algebra)。

Correlation Function

首先要知道場論的目地就是要算correlation function,在一般場論裡面correlation function牽涉到scattering amplitude,這當然很正常,因為一個物理理論,最重要的就是把你能觀察的東西跟理論能夠計算的東西作連結,然後做實驗觀察它,所以一般的場論裡面,首先寫下有哪些field(比方電子,光子,一些夸克等等)再寫下一些interaction,然後開始計算它的correlation function利用費曼圖,需要把所有可能的費曼圖加起來,中間會遇到一些問題(renormalization:重整化),但原理上就是如此,所以計算上是非常繁瑣的,需要用到大量特殊函數和高維空間的積分等等等。
但保角場論就不一樣啦,由上述的statement,我們知道,只需要計算primary operator的correlation function 即可,而且不用利用費曼圖的技巧,也沒有重整化的問題,correlation function of two point 和 three point 都可以完全被保角對稱fixed住,這當然就非常強,和一般場論明顯不同 我們可用保角變換來確定兩點純量的函數,計算來自 Lectures on Conformal Field Theory: 一般的說 保角變換定義成:
$$ g_{\mu\nu}'(x')=c(x)\delta_{\mu\nu}(x), b(x)=\sqrt{c(x)} $$
一個scalar primary field定義成:
$$x \rightarrow x' , \mathcal{O}'(x')=b(x)^{-\Delta}\mathcal{O}(x) $$
We can use property to fixed the two point function (up to constant factor):
$$<\mathcal{O}_1\left(x_1\right)\mathcal{O}_2\left(x_2\right)>=\frac{\delta_{ij}}{\left|x_1-x_2\right|^{\Delta _1+\Delta _2}}. $$
同理三點純量函數也可以fix住:
$$<\mathcal{O}_1\left(x_1\right)\mathcal{O}_2\left(x_2\right)\mathcal{O}_3\left(x_3\right)> =\frac{\lambda _{123}}{\left|x_1-x_2\right|{}^a\left|x_2-x_3\right|{}^b\left|x_1-x_3\right|{}^c}. $$
此刻:
$$a=\frac{\left(\Delta _1+\Delta _2-\Delta _3\right)}{2},b=\frac{\left(\Delta_3+\Delta_2-\Delta_1\right)}{2},c=\frac{\left(\Delta_1+\Delta_3-\Delta_2\right)}{2}.$$
重點在於,當我們打算使用在四點的correlation function時候,會發現它沒辦法固定住:
$$<\mathcal{O}_1\left(x_1\right)\mathcal{O}_2\left(x_2\right)\mathcal{O}_3\left(x_3\right)\mathcal{O}_4\left(x_4\right)> =\left(\frac{x_{24}}{x_{14}}\right)^{\Delta_{12}}\left(\frac{x_{14}}{x_{13}}\right)^{\Delta_{34}}\frac{g(u,v)}{x_{12}^{\Delta 1+\Delta 2}x_{34}^{\Delta_3+\Delta_4}}.$$
此時 $$u=\left(\frac{x_{12}x_{34}}{x_{13}x_{24}}\right)^2,v=\left(\frac{x_{12}x_{34}}{x_{23}x_{14}}\right)^2.$$,$$(u, v)$$ 叫做 conformal cross ratio, 注意到: \(u, v\)(是conformal invariant. 所以g(u, v) can be any 函數. 對於具有spin的情況我們需要更有效率的做法,叫做embedding formalism 在這裡不多說惹.

OPE

似乎到了上一步,我們就做不下去了,這時需要一個強大的物件,就是"OPE",OPE是一個神奇的物件,如果我們假定OPE是正確的,那麼我們就有了很妙的東西,簡單的說ope 就是當你把兩個field成在一起時候可以把它展開成其他field: $$ \mathcal{O}_1\left(x_1\right)\mathcal{O}_2\left(x_2\right)=\sum_{\mathcal{O}} \lambda_{12\mathcal{O}}\mathcal{O}\left(x_{12},\partial y\right)\mathcal{O}(y)|{y=0} $$ 這有啥麼用? 簡單說: 就是 這樣我們就可以把四點的函數,related to 三點的函數

Scalar And Spinning Conformal Blocks

Conformal Blocks 是cft的最後一個重點物件之一,後面的conformal bootstrap都要依賴他,計算conformal block 主要是運用以下方法 Conformal Partial Waves: Further Mathematical Results using the Casimir differential equation: 使用 OPE expasion, g(u, v) 可以被分解為conformal blocks:
$$ \left(\sum_{i=4} M_{iAB}\right)<\mathcal{O}_1\left(P_1\right)\mathcal{O}_2\left(P_2\right)\mathcal{O}_3\left(P_3\right)\mathcal{O}_4\left(P_4\right)>=0 $$

$$ \frac{1}{2}M_{AB}M^{AB}\mathcal{O}_i\left(P_i\right)=c_{\Delta ,l}\mathcal{O}_i\left(P_i\right). $$

$$ c_{\Delta ,l}=\Delta (\Delta -d)+l(l+d-2). $$
We get
$$ \left(\frac{1}{2}\left(M_{1AB}+M_{2AB}\right){}^2-c_{\Delta ,l}\right)W_{\mathcal{O}_{\Delta, J}}\left(P_i\right)=0. $$
Dolan and Osborn 計算了 closed form expressions for the conformal blocks of an arbitrary spin-l primary in d = 2, 4. 奇數維度情況則沒有closed form, 只有積分表達式.
$$ W_{\mathcal{O}_{\Delta, l}}=\left(\frac{x_{24}}{x_{14}}\right)^{\Delta_{12}}\left(\frac{x_{14}}{x_{13}}\right)^{\Delta_{34}}\frac{g_{\Delta, l}(u,v)}{x_{12}^{\Delta _1+\Delta 2}x_{34}^{\Delta _3+\Delta _4}}. $$
此刻 $$(\Delta, l)$$ 是中間可能出現的primary field.

Conformal Bootstrap

Conformal Bootstrap 可以視為conformal field theory的心臟,我不知道bootstrap要怎麼翻譯,不過也不重要,重要的是Conformal Bootstrap是用來解保角場論,一開始先假定你的CFT的spectrum, 有哪些 primary fields,決定它的scaling dimension和representation, 然後開始Conformal Bootstrap來檢測是否滿足crossing symmetry,所以我們可以講, 一個well defined的CFT 就是自洽的OPE和a set of primary fields. 一定承認了這個定義,那這個場論原則上就被全部解出,因為更高點Correlation Function都可以被OPE簡化成四點的, 這樣就完整的解出這個場論(以下演示一下方法)。 We have the conformal block expansion (四個一樣的scalar): $$<\phi_1\left(x_1\right)\phi_2\left(x_2\right)\phi_3\left(x_3\right)\phi_4\left(x_4\right)>=\frac{g(u,v)}{x_{12}^{2\Delta_{\phi}}x_{34}^{2\Delta_{\phi}}}$$ $$g(u,v)=\Sigma\lambda_{\phi\phi\mathcal{O}}^2 g_{\Delta,l}(u, v)$$ 如果我們交換展開的順序1換成3,2換成4 則會有 $$g(u,v)=(\frac{u}{v})^{\Delta_{\phi}} g(v, u)$$ 這個複雜的條件就叫做crossing symmetry. 也就是bootstrap condition.一般會使用semidefinite programming來解決這個問題
explanation

一些目前研究的狀況和前景

CFT復甦大概是在2008年開始:計算 3d ising model 的臨界點所導出來的,然後大家開始做了4維的Conformal Bootstrap,不只限於4個scalar,也有人在研究三維的fermion和增加了超對稱的cft. 另一個方向是來自1995 Juan Maldacena 的 conjecture called AdS/CFT correspondence 可以藉由計算Witten Diagram(類似於費曼圖在AdS空間中)來得到對應的CFT的部分資訊。這些類似文章在arxiv都有很多介紹惹。