The h Bar

Kyle Kanos

3:46 PM

I wonder if this user Jerrold Franklin is this faculty Jerrold Franklin...

DanielSank

@KyleKanos Thrice.

Kyle Kanos

QMechanic beat you to it

DanielSank

@KyleKanos Indeed.

Physics Meta

0

Q: Did you know a intersting mail-list about physics?

I would like to join some mail-list that periodically tell you about online conferences, courses, opportunity,summer schools, jobs opportunity in topics related to physics, from universities and/or institution related... (it's that section appropriate for this kind of questions?)

Gaurav

4:02 PM

@DanielSank Repeat it too much and it might become common parlance. Nobody would be tired of it then.

0celo7

Just got a copy of Nature in the Senator's mail.

Kyle Kanos

Which Senator?

0celo7

I wonder if I'm allowed to share that with strangers.

Online while at work, that is.

Kyle Kanos

There's only 100 choices...

0celo7

Manchin

The senators get showered with magazines.

Kyle Kanos

4:07 PM

Had to Google him (probably would have had to do that for about 95 Senators anyways)

DanielSank

4:22 PM

Anyone understand why noise processes in Nature are often assumed to be Gaussian processes?

I think this goes beyond the central limit theorem because a Gaussian process not only has $\phi(t)$ Gaussian distributed, but actually $\phi(t_1)$ and $\phi(t_2)$ are jointly Gaussian for any $t_1$ and $t_2$.

My intuition says that this arises because uncorrelated forces in a physical differential equation must result in a Gaussian process, but I haven't found anything confirming that.

Kyle Kanos

I'm pretty sure it's just the CLT

DanielSank

@KyleKanos CLT only tells me that summing many random variables results in a normally distributed result.

(I think)

Kyle Kanos

If you add up enough i.i.d. random variables, you'll end up with a Normal (Gaussian) distribution

Late by a few seconds...

Though it does also assume a finite mean & finite variance

DanielSank

@KyleKanos Yes but as I said, a Gaussian process is not just Gaussian distributed at each time. It's also jointly Gaussian, which is important.

innisfree

@DanielSank See Jaynes, probability theory: the logic of science

Kyle Kanos

4:32 PM

The Wiki page on the multivariate normal distribution says it's the multidimensional CLT

DanielSank

@innisfree Is that a book or article?

Kyle Kanos

Articles

DanielSank

@KyleKanos typo.

innisfree

an absolute classic book

See the section, Why the ubiquitous use of Gaussian distributions?

Kyle Kanos

@DanielSank I see now

innisfree

4:34 PM

For example, today most accurate
experiments in physics take data electronically, and a physicist usually knows the mean
square error of those measurements because it is related to the temperature by the well-
known Nyquist thermal fluctuation law. 9 But he seldom knows any other property of the
noise. If he assigns the first two moments of a noise probability distribution to agree with such
information, but has no further information and therefore imposes no further constraints,
then a Gaussian distribution fit to those moments will, according to the principle of maximum

on the CLT:

n other fields, such as analysis of economic data, knowledge of the noise may be more
crude, consisting of its approximate general magnitude and nothing else. But for reasons
noted below (the central limit theorem), we still have good reasons to expect a Gaussian
functional form; so a Gaussian distribution fit to that magnitude is still a good representation
of one’s state of knowledge.

DanielSank

@innisfree This doesn't explain the assumption that these are Gaussian processes.

innisfree

Jaynes argues that the Gaussian most accurately reflects our state of knowledge about the noise

DanielSank

@innisfree This is missing the point of my question, I think.

I understand why the voltage on a wire at time $t$ is Gaussian distributed.

What I don't understand is why these are assumed to be Gaussian processes.

But maybe I just need to read some Wiki articles.

Interestingly the Ornstein-Uhlenbeck process is Gaussian, and it satisfies the Fokker Planck equation...

Kyle Kanos

Doesn't surprise me that a Gaussian satisfies the FP equation

innisfree

@Daniel Ah, I see

DanielSank

4:45 PM

@KyleKanos It's not the Gaussian, dude, it's the process.

@KyleKanos There are many processes whose unconditioned distribution is Gaussian, but which are not Gaussian processes!

@innisfree Yeah, I think this might be the key. My guess is that things like Johnson noise are Ornstein-Uhlenbeck processes.

Screw it, I'm going to post this as a question.

innisfree

whether or not something is a gaussian process must depend on how it evolves in time, so it's not a matter of a parameterizing our ignorance, as in the guassian distribution case

DanielSank

5:04 PM

@innisfree Yes.

Gaurav

@DanielSank In case you felt I was taking a jibe at you in the previous comment, I wasn't. :)

DanielSank

@Gaurav I didn't think you were.

@innisfree @KyleKanos ok, here's the question. We'll see what happens.

@innisfree any idea how I can get that Jaynes book for <$100?

I found a partial version online but I'm guessing it's not legal.

Gaurav

Go for it. If the FBI were to arrest everyone downloading illegal PDFs, they'd probably run out of agents.

DanielSank

@Gaurav Again, it's only a partial copy.

@Gaurav I'm also less worried about the FBI than I am about authors and associated institutions (publishers, etc.) receiving proper compensation for their work.

Gaurav

If you're going by propriety, you shouldn't. But I think god will probably pardon you for that.

BTW, publishers don't deserve such treatment, they sell digital copies for more or less the same cost of printed ones.

Kyle Kanos

5:22 PM

@Gaurav More likely to run out of jailspace (too late!) than agents

Gaurav

:D

Kyle Kanos

@Gaurav Which is reasonable given that the actual printing is actually a small portion of the net costs (advertising, editing, author royalties, etc)

innisfree

@DanielSank the partial version might be legal - the book took decades to write and Jaynes released a few unfinished editions in his lifetime. See bayes.wustl.edu

Gaurav

@KyleKanos Add binding, middlemen, etc.

Kyle Kanos

Binding is part of printing & middlemen are part of advertising & editing (probably also distributors as well)

But yes, one should never expect that a printed book be \$30 and the digital copy \$5 because of 'no printing costs.' That's extremely naive

(Though perhaps \$5 off that \$30 price is reasonable, I've not looked into digital books b/c I like thumbing the pages)

Gaurav

5:31 PM

Okay, cut the middlemen if you're buying online, but if you're buying from your local bookstore, the price is greatly hiked due to that.

I'm not asking for such a great difference in price, but maybe ten dollars less would do.

Okay, maybe at least five dollars.

alarge

@DanielSank So if you start with the SDE and add a random (Gaussian) noise term, that is an O-U process as it will have the reverting force. You can then transform this to Fokker-Planck or Feynman-Kac if you like. You could argue, with probably something along the lines of Mori-Zwanzig that this is how you should go about it.

Gaurav

For Indians, that's a big deal.

Kyle Kanos

But note also that, in order to get it into epub format, for instance, you have to have someone transcribe it

And how many people are going to the book publisher for the book? Aren't they still going to e.g., Amazon or Barnes & Noble?

Those guys are still going to charge for it

alarge

In the SDE you could drop in a more general distribution andbif I remember correctly fluctuation-dissipation should still work given some constrains on the moments

Kyle Kanos

There is a lot more to e-publishing than you seem to be giving credit to, though I cannot imagine that there's a smaller profit margin than with printing (that is, I imagine that companies make more with ebooks than paper ones)

DanielSank

5:39 PM

@alarge I understood some of those words.

Gaurav

They can publish the same file used for printing in the digital version, no ? What is the need to transcribe it again?

DanielSank

I don't suppose you know of a digestible reference... a book or something which you would recommend?

Gaurav

Assuming I'm reading the same definition of 'transcribe' as you do.

Kyle Kanos

@Gaurav You are accounting for only PDFs, how many e-readers are read that format?

DanielSank

By the way, guys, I would like to take a moment to thank you all for being a useful and generally supportive community. I've been trying to use the Signal Processing SE a bit lately and the tone is somewhat discouraging. Drive-by down votes etc. I'm glad that Physics has its act together and seems to be improving all the time.

@KyleKanos People who haven't actually worked in business frequently underestimate how difficult it is to actually make money on anything.

Kyle Kanos

5:41 PM

@DanielSank Exactly!

Gaurav

@KyleKanos What great labor is involved in converting a pdf to epub ? Or vice versa.

DanielSank

@Gaurav You know what would be a great exercise and really educational: try it yourself instead of speculating.

Really. Get a nice long pdf with figures etc. and write a computer program to convert it to epub with zero issues that you have to fix manually.

Kyle Kanos

Again, you are being naive. How many writers write in PDF format?

DanielSank

Make sure the table of contents links properly, make sure all the figures show up right, etc.

@KyleKanos Well, these days one hopes to have everything in some flavor of TeX or related.

Kyle Kanos

Scientists probably are writing with TeX, but what is the ratio of science books to every other f&^#ing book out there (to be blunt)

Gaurav

5:45 PM

@KyleKanos They have to transcribe it to pdf or epub for the print version anyway.

Kyle Kanos

@Gaurav Are you sure?

Why can't it be doc or docx or RTF or txt or one of a million other formats?

DanielSank

@KyleKanos Well I assume novels can be trivially converted to raw text and then shoved into whatever other format is needed.

Kyle Kanos

There are auto-conversion tools, but to get a quality job done (making sure to eliminate bad hyphenation, line-length formatting, etc) you have to have a human to do it

DanielSank

I mentioned TeX because math etc. is nontrivial to type set.

@KyleKanos Bingo. That human needs a salary.

Kyle Kanos

And humans cost money

Gaurav

5:47 PM

@KyleKanos It could be, but what stops them from being converted to pdf ?

DanielSank

Jinx!

Kyle Kanos

@Gaurav quality

DanielSank

@Gaurav Do some research. We're not publishers and we're all just speculating and making up random nonsense.

@Gaurav If you actually want to support an opinion that ebooks are over priced go find out how much each step actually costs. Science!

@Gaurav If it takes someone a year to write a book, the author alone is going to want to see many tens of thousands of dollars at least from that work.

At least, given my earning potential, if I were to write a technical book that would be my expectation.

I would only do it for less if I were retired and just having fun / trying to be helpful.

Gaurav

@DanielSank If I really wanted to meet a publisher to speak about this, I wouldn't be discussing here at all.

DanielSank

@Gaurav What I mean is to stop harping on the pdf thing and instead consider the actual costs.

After the author is paid, we still have an editor. Then we have design folks who deal with the layout, fonts, cover art (if there is any), etc.

Kyle Kanos

5:51 PM

And if you're Mary Higgins Clark, you're getting \$12 million per book

DanielSank

All of those people work in a company which employs secretaries, lawyers, the guy who goes and gets lunch...

@Gaurav Then we get to the conversion to digial media and distribution to various outlets such as e.g. Amazon, Barns and Nobel, whatever.

Those websites employ a bunch of programmers who want to earn $100,000 per year else they'd get another job.

There's the cost of the data center serving the book data, the contract with the online sales company which manages the flow of money and handles all of the encrypted online purchasing stuff.

There's the credit card company itself which takes a slice of all transactions.

Advertisements.

I'm not convinced that printing + trucking a book to a book store + running the book store costs that much more than the salary of the programmers etc. involved in online distribution.

Of course, I have no real idea what I'm talking about.

As I said, we're making this all up.

Kyle Kanos

But those are at least the correct lines of thinking, instead of 'no paper'

DanielSank

@KyleKanos Thank you, sir.

Kyle Kanos

There are way more costs involved in making a single book than paper

alarge

@DanielSank A recent book, Statistical Mechanics: Theory and Molecular Simulation covers all the stuff at a rather simple level. I think the auyhor has many of the relevant bits on his website. If you want to go in more detail with stochastic processes and the like, van Kampen's is a very readable book that people regularly use as a reference too.

Gaurav

6:02 PM

Yours is the right line of thinking. No disputing that. You are assuming that my apprehensions are the general high costs of books. No, they are not. I'm just saying compared to printed versions, the digital versions ought to be cheaper.

alarge

I can discuss this in more detail once I'm off my mobiLe. especially if you have more specific quetions

DanielSank

@alarge I'd be happy to post a series of questions on the main site. Please look out for them :D

Kyle Kanos

@Gaurav And we're saying that you shouldn't have that expectation; you are trading a cost from printing for a cost in epublishing

DanielSank

@Gaurav I just explained about paying all the tech people. They're definitely going to want more $$ than the truck driver who carries the books to the book store, or the cashier at the book store, or the guy who runs the printing press.

@alarge Heres one

Gaurav

Hmm.. okay I actually read through all those comments, and you're probably right.

Anyways, I have far less experience in anything 'business', so I'll have to take what you guys say as true.

Kyle Kanos

6:18 PM

Hmph...some people are bad at assumptions.

> Given a triangle, you know the lengths of two sides (6 and 8), what is the probability that the third side is between 3 and 10? (three sides are all integer)

Some people immediately jumped at P=1 because of the 3-4-5 triangle

Totally wrong though

Gaurav

8/10

Kyle Kanos

Nope, 8/9

You have to use law of cosines, $c^2=a^2+b^2-2ab\cos(\gamma)$. Since $\cos(\gamma)\in(0,1)$, then the minimum is 2 and the maximum is 10

The fact that the question gives you a range should be a strong indicator that you shouldn't assume Pythagorean holds

Gaurav

:D

Kyle Kanos

The question (found while studying for a quant job interview I have tomorrow with another bank) is just another one of those problems that is asked because it's going to trip up the interviewees

user54412

Ok, I just learned the 20-step procedure to vpn/ssh/vnc into the campus vis computer. Now to put my bash skills to the test and write a script to do it.

Kyle Kanos

6:28 PM

20 steps to vpn?

user54412

@KyleKanos Why can't I have a 6-8-13 triangle? $\cos\gamma \in (-1,1)$, no?

Kyle Kanos

-1 hmm

Forgot that one

8/13 then

Either way, 1 is wrong

And R is pretty sweet, now that I'm (kinda) trying to learn it

user54412

@KyleKanos tunnels within tunnels, displays within sessions, protocols on top of procedures across networks...

user54412

(it's been a very trying experience)

Kyle Kanos

Then the question simply becomes why do you need to vpn?

user54412

6:39 PM

The machines are very hard to access unless on the ethernet network. Even campus wifi doesn't have privileges to get to the cluster.

Kyle Kanos

That's silly

Gaurav

The cosine rule is one way to solve it, otherwise you can easily reason that a 6-8-(0 to 2) triangle is impossible since 6+n < 8, n $\epsilon$ (0,2)

Kyle Kanos

6:56 PM

I'm sure there's a few ways to solve it. You simply need to build a set of all possible combinations and then you easily get your ratio

Gaurav

Best of luck for the interview !

Kyle Kanos

Thanks

alarge

Is the interview over the phone or face to face?

innisfree

@KyleKanos doesn't it depend on a prior pdf for the angle?

i don't think the question is well-posed

Kyle Kanos

@alarge Phone

@innisfree AFAIK, all you need to know/determine is the maximal and minimal values of $\cos(\gamma)$ to get the minimum and maximum values of c.

I think it's intentionally ill-posed to see the thought process of the interviewee

innisfree

7:04 PM

yes but then how do you calculate a probability?

Kyle Kanos

Oh, that's just the ratio of anticipated values (3,4,...10) to total possible values (2,3,...,14)

Well, the size of the sets really

There are 8 acceptable numbers and 13 total options

innisfree

oh only integers allowed. but in any case, it depends on your prior weights for the possible values (2,3,...14)

Kyle Kanos

I think it's safe to assume equal probabilities of values

innisfree

why?

Kyle Kanos

Why would you assume otherwise?

alarge

7:07 PM

@innisfree Of course it depends. The whole point of an interview is to be a dialogue, so if you notice something like this, you ask. The interviewer will probably just pose the question back to as to what you might think is sensible.

Kyle Kanos

Principle of indifference

The principle of indifference (also called principle of insufficient reason) is a rule for assigning epistemic probabilities. Suppose that there are n > 1 mutually exclusive and collectively exhaustive possibilities. The principle of indifference states that if the n possibilities are indistinguishable except for their names, then each possibility should be assigned a probability equal to 1/n. In Bayesian probability, this is the simplest non-informative prior. The principle of indifference is meaningless under the frequency interpretation of probability, in which probabilities are relative...

innisfree

if only it were so simple, though. I might be indifferent to the angle in the triangle, and give those possibily lengths weights consistent with a flat pdf for the angle

bit strange though as we consider discrete lengths...

Kyle Kanos

I'd argue that it is that simple

Once you know the minimum and maximum values of cosine, it doesn't matter anything about the values of $\gamma$

innisfree

i doubt equal weights to the lenghths is non-informative with regards to the angle

Kyle Kanos

You've got your range, you're indifferent to the actual values $\gamma$ takes. All you know is that $c$ is an integer

user54412

7:15 PM

Ah the principle of indifference -- the line of thought that convinces so many people cosmological inflation is somehow not crackpottery.

innisfree

No, you aren't. If you equally weight the lengths, you end up favouring bigger values of \gamma.

Kyle Kanos

But $\gamma$ is irrelevant

innisfree

it's not so important in discrete cases, because the fact that the length is discrete already gives a preferred parametrization, in terms of the length. But in the case of a continuous length, you could apply principle of indifference to the angle or the length and get different results

also in discrete cases there's no jacobian for a change of variables between \gamma and length

Kyle Kanos

You are explicitly told to pick discrete values, so all you need is a maximal and a minimal range. The actual mathematics of the angles don't factor in, only the minimal & maximal value of cosine.

alarge

@KyleKanos You don't need cosine rule though as far as I can see: The maximum length is just the two sides added to one another and the minimum them subtracted from one another.

Kyle Kanos

7:27 PM

If my math serves me right, there are only 9 possible options: {14, 13, 11, 10, 8, 7, 5, 4, 2}. Of which 5 are in the range specified, so 5/9 ought to be the answer

innisfree

@alarge indeed

Kyle Kanos

@alarge True, going beyond that would break the length requirement. I suppose cosine rule is a bit more formal

(also, by "math" above, I mean scripting in R and taking steps of 1e8)

(but that's kinda cheating because you're not given the computer when answering the question)

alarge

I don't quite see why 3 wouldn't be a permissible length. Can't you just adjust the angle to fit all lengths?

Kyle Kanos

Kinda curious that all factors of 3 are missing: 3, 6, 9, 12

I would have suspected that all values would work, given tweaking the angle correctly

innisfree

I agree - you can make all the integers between the max and min lengths. but the possible angles here are between 0 and pi. equal weights for the lengths results in p(angle >= pi/2) << p(angle < pi/2). which, in terms of the angle, is informative. that's why the principle of indifference isn't so easy to apply.

Kyle Kanos

7:34 PM

Anything about the angles themselves is introducing a new variable that you are given no information on. Supposing anything about it is a waste of time

You know the minimum & maximum values of c, thus the range, plus the acceptable values

There's no need to introduce more variables into the problem

innisfree

It's not more variables. equally weighting all the lengths (looks uninformative, principle of indifference blah blah blah) <=> favouring angles of less than pi/2 (looks informative, what happened to the principle of indifference and the blah blah blah)

these are well-known issues in probability

Kyle Kanos

But you're not interested in the distribution of angles or lengths, just the possible values of lengths

alarge

@innisfree Again, it's an interview question and you are more than welcome to bring these issues up in the interview to create dialogue. The problem is entirely artificial and as such nobody is going to care which probability distribution you choose. Just that you'll be able to formulate and solve a version of the problem.

innisfree

:) all right @kyle good luck in your job hunting

Kyle Kanos

Thank you

paul23

8:18 PM

hey guys

quick question

How would you guys call a not-exactly-2d-world?

Kyle Kanos

1.5D?

paul23

But -like you often do- basically a 3d world where you only consider a plane in it?

So velocity vector has only 2 components, but you do use mass/parameters as they were 3d.

Kyle Kanos

That sounds more like 2.5D

paul23

Well a more formal definition would be nice lol

Kyle Kanos

That is formal

paul23

8:20 PM

Something that can be used in papers.

Kyle Kanos

There are research papers that use 2.5D

In the title, in the text, etc

paul23

Really?

Kyle Kanos

arxiv.org/abs/1412.6489

paul23

hmm

Kyle Kanos

IIRC, in that paper, the magnetic field is fully 3D while the simulation takes place in a plane

paul23

8:22 PM

@KyleKanos Well I do a even less 3d :P. It's only that "solar mass" is inherently based on 3d assumptions.

Kyle Kanos

That's generally how I see 2.5D used at least

My research used 3D assumptions of supernova remnant explosions but in 2D

paul23

I've been called back for calling the thing "2d simplification of orbits" once already - being told that "G & mass then should change accordingly".

Kyle Kanos

I thought that G would be the same, but the potential would change (cf this post)

alarge

9:58 PM

@DanielSank Well, I gave it a shot. I'm not sure that is what you were after though. I wanted to write about more stuff, too, but I think I was already veering off-topic... maybe.

DanielSank

10:32 PM

@alarge $&(# yeah! That's the kind of answer I really really like.

@alarge we need more stochastic processes / noise Q&A around here.

Kyle Kanos

@DanielSank: In RE your comment to alarge's post:

amazon.com/…

I've read some of that one and found it pretty good

Not sure if it's at the level you'd like

alarge

@DanielSank I'm happy that you liked it. There's not that many questions about stochastic stuff here, but I've answered a couple of questions. I think they have more stochastic processes over at crossvalidated.SE.

DanielSank

10:50 PM

@alarge Is crossvalidated good for these sorts of questions?

alarge

@DanielSank I don't follow it, so don't know for sure. But if DSP died out, maybe CV will take over some of the load.

DanielSank

@alarge There's a signal processing exchange. Not sure about DSP specifically. The signal processing one seems kind of low volume these days so I don't expect it to survive.

I actually had a perfectly good question which got a perfectly awesome answer down voted to -1 on SignalProcessing.SE. Does not bode well.

@alarge ordered a copy of Van Kampen.

Thanks.

@alarge I understand how to use the Caldeira-Leggett model to make an infinite set of oscillators look like a resistor, but I'm not sure how to make the leap you mention in your answer.

alarge

@DanielSank Which leap? To the generalized Langevin equation?

DanielSank

Yes.
I'm guessing the resolution is that a perfect resistor has, by fluctuation-dissipation, a white noise spectral density...

alarge

nyu.edu/classes/tuckerman/stat.mech/lectures/lecture_24/…

physnolimits

10:57 PM

Did you know a interesting mail-list about physics?

I would like to join some mail-list that periodically tell you about online conferences, courses, opportunity,summer schools, jobs opportunity in topics related to physics, from universities and/or institution related...

alarge

That's the website I kind of mentioned earlier. Lecture notes on the book on Statistical Mechanics.

DanielSank

@alarge Thanks. I stuck that in a comment under your answer.

alarge

@physnolimits Many of those are field-specific. You can subscribe to the RSS feed of Science jobs or Nature jobs or whatever, for example, but I think you'll see plenty of stuff you won't be interested in.

physnolimits

@alarge could you give me some links?

alarge

@physnolimits For the Science/Nature stuff? On their websites, sciencecareers.sciencemag.org nature.com/naturejobs/science. But again, I don't think these would be of much interest to you. The same is true for conference postings etc.

Transcript for