« first day (2521 days earlier)      last day (2796 days later) » 

16:00
Semiclassical let's make a markov matrix
@zed111 Here's how I understand them, to reiterate.
after youf inish with zed
Suppose we have t<0 and we close that integral in the upper half-plane.
Then $\omega$ will take on values with arbitrarily large positive imaginary part.
i.e. $\omega = x+i y$ with $y\to +\infty$.
Then $e^{-i \omega t}=e^{-i x+y t}$.
If $t>0$, then as $y\to\infty$ that exponent will be large and positive, so it'd diverge. But if $t<0$, then $+yt$ is actually negative.
Hence for $t<0$ we can close in the upper-half plane and the resulting integral over omega with positive imaginary part will be exponentially small.
Hence we can ignore it.
Great -- thanks @Semiclassical
16:06
we never actually did this in class so i hope its not on the final
but it doesn tlook that hard
(I imagine they have some further claim re: $\chi(\omega)$ being analytic in the upper half-plane so as to ensure that $\chi(t<0)=0$ i.e. causality)
a certain town never has two sunny days in a row. each day is classified as bveing either: sunny, cloudy (but dry), and rainy
if it is sunny one day, it is equally likely to be either cloudy or rainy the next day
if it is rainy or cloudy one day, then there is one chance in two that it will be the same the next day; if it changes, it is equally like to change to either of the two remaining possibilities
okay, so a 3-state system.
ya
our convention is:
rows index the state now and columns index the state later
so $p_{11} = 0$, that's immediate
first row is $[0,1,1]$
wait no
1=sunny, 2=cloudy, 3=rainy?
16:09
$[0,1/2,1/2]$
sure, thats the order its presented in
so row 1 is easy
$p_{12} = p_{13} = 1/2$
ooh, ooh, we're halfway there
$p_{j,k}=\Pr(j\to k)$?
i think
one thing to notice here is that, in all three cases, if you don't return to your initial state then the other two possibilities are equally likely.
So that simplifies things a bit.
16:13
not sure how that helps
Also, you're told what the probabilities of the states returning to themselves are.
So that corresponds to the main diagonal of your matrix
$p_{11} = 0$, and, uh
$p_{22} = p_{33} = 1/2$?
and the rest you can do by making the rows sum to 1
right, and by using the symmetry I just mentioned.
16:15
cool, alright
What would he actually want you to do with said matrix?
then the question asks about long run probabilities which we didnt do in class so Im sure it wont be on the exam, im not even sure he meant to do it as homework
ah.
The usual way to do this is to work out the eigenvalues + left eigenvectors of the matrix.
right, which we didnt do in class, so how can he have us do it on the test?
probably shouldn't.
16:16
he said that if we have to do any matrix multiplication on the test it will only be on triangular matrices
(not sure it's left vs right, i get mixed up on that. think it's left though)
and he for sure will ask us to partition a markov chain into states that intercommunicate
the easiest way I find to do that is to make a weighted directed graph
he said that we can use matrices and graphs interchangeably for any answer
states intercommunicate = cycle on the directed graph?
state $i$ intercommunicates with state $j$ if there is a nonzero probability that, starting with $i$, the state will eventually reach state $j$
16:19
theorem: "intercommunicates" is an equalivalence relation on the states of a markov chain
so i intercommunicate with j = i reaches j.
i guess that's a good way to think of it
anyways.
$i$ reaches $j$ with positive probability, given enough iterations
it doesnt have to reach $j$ directly
it's not a symmetric relation, though.
since it's directed.
16:20
well I mean
for a graph, I'd connect two nodes with two lines in that case, each one directed to the other
oh I see what your esaying
sorry, that's part of the definitiomn
if its not symmeric they dont intercommunicate
hi @Semiclassic, FauxPas
Ted, long time no see
16:23
Ohi
Hi @TedShifrin
heya @Astyx!
a state goes to itself in 0 steps by definition, so the only part of that theorem that isnt immediate is transitivity
hi @Justwin, whoever you be.
16:25
@Semiclassic: A journal sent me an article to referee on using differential geometry to figure out the normal derivative of the electrostatic potential. They didn't buy my excuse that I'm retired ... or that it looked like stuff already in a book.
@GFauxPas ahhh
@TedShifrin That looks like stuff that'd be in Jackson, yeah.
oh, not electrostatic potential ... magnitude of the field.
I mean, I remember one of the problems in the first chapter literally being to relate the principal curvatures at a point on a conductor to the field at the surface.
Right, but he's doing it in terms of principal curvatures of surfaces. It's still like stuff that's standard for understanding why soap bubbles are surfaces of constant mean curvature. But I guess I'll read it slightly carefully.
He references Jackson and says he does it "by an application of Gauss's theorem."
Semi, another matrix?
"3 out of every 4 trucks on the road are followed by a car. one out of every 5 cards is followed by a truck"
@Astyx: Alors, tu prends les examens oraux maintenant?
oh, thats easy
this is the part he didnt do cover in class, and that he has on the homework due tomorrow:
"What fraction of vehicles on the road are trucks? (set this up as a markov chain and computer the stationary distribution, whose components are the long run proprtion of time spent in each state)"
is it hard?
16:28
So, what's the transition matrix?
Hi @Ted
I always did a bit of Markov chains at the end of my linear algebra course.
Shouldn't be in this case.
Heya @Balarka
Home finally? :)
16:29
Yes, Balarka.
$\mathbf P = \begin{bmatrix} 1/4 \ \ 3/4 \\ 4/5 \ \ 1/5 \end{bmatrix}$
So now you can resume learning differential geometry, Balarka :D
Not quite. What's the probability of a car after another car?
@FauxPas: For me the columns should add up to 1, so you're transposed.
16:30
@TedShifrin Oui j'ai eu les oraux des Mines-Ponts lundi et mardi, et j'ai Polytechnique la semaine prochaine, les ENS celle d'après et Centrale en dernier si j'y vais
@ted Here it's $p_{ij}=\Pr(i\to j)$, so rows are normalized instead.
À la fin, tu vas beaucoup savoir, @Astyx :P
Ted: professor said both notations are acceptable and that we're using this notation in class just because its what our textbook uses
Ugh @Semiclassic
So we need eigenvectors of $A^\top$, not of $A$. Me no like.
I tend to agree with @ted, though. Column vectors are states and matrices act from left to right.
16:31
"one out of every five cars is followed by a truck"
Mes exercices aux mines étaient nullissimes, j'espère qu'ils seront mieux à X-ENS
Right. So Pr(C->C) = ?
@TedShifrin That's my plan, yup
oh, 3/4
16:32
just kidding?
Peut-être que tu sais déjà trop, @Astyx :)
its not 4/5 and 1/5?
3/4 is the probability of getting a car after a truck.
Ne sait-on jamais trop ?
what's the probability of getting a car after a car?
16:33
@Balarka: I assume lots of learning took place here in the month of my (mostly) absence.
you know that one out of every 5 cars is followed by a truck. so how many of those 5 cars are followed by cars?
1?
not seeing it :(
if 1 in 5 cars is followed by a truck, what are the other 4 cars followed by?
cars
right. so Pr(C->C)=4/5
16:35
thats what I wrote!
key point: That's one of the diagonal transitions
oooh
durrrrr i smart mathemagician
@Astyx: ooh, alors, moi je ne sais jamais assez ...
This is why I, in writing out my matrix, put T C above the columns and to the left of the rows
just to remind myself of which probability will be which.
16:36
$20^{-1} \begin{bmatrix} 5 \ \ 15 \\ 4 \ \ 16\end{bmatrix}$
@Semiclassic: Do physicists say "electric field intensity" for the field? This seems wrong to me.
first row isn't normalized to one anymore :P
@TedShifrin I learnt a lot of fun math from the good crew (Astyx, Akiva, Daminark, Fargle etc) here. In theory, a bit about foliations but right now I have, as a project of the week, learning homotopy theory using the Moore method.
was a "joke"
@TedShifrin I think I may have seen that? But I'd have expected 'electric field strength'
sure, let's go with that.
16:37
Intensity and strength sound like scalars, not fields.
I would say $\|\vec E\|$ is the electric field intensity/strength.
Well it's the force applied to a unit positive charge.
I'm not enjoying this paper so far.
I really just think of $\vec{E}$ as the electric field.
16:37
Ah ok, yeah, it's not just the magnitude
Ted I think that's the terms we used in physics ii
But I know that using 'intensity' comes up when distinguishing $\vec{E}$ versus $\vec{D}$
$\mathbf E$ is a function that maps a space to vectors, not a number
So it may be a consequence of that old issue.
@Balarka: How are you using the Moore method? I mean ... don't you usually try to figure things out?
@Semiclassic: OK, you got me. What's $\vec D$?
16:39
and $\Vert \mathbf E \Vert$ is the strength of $\mathbf E$
that's how we did it
the name I usually know for it is the electric displacement field.
but I think there's another terminology for it that shows up.
FauxPas, I would tend to agree. The words intensity and strength suggests magnitudes of vectors, not vectors.
or $E$ if we want ease of writing at the sacrifice of some clarity
@TedShifrin Eg my recent project is to not to have a clue about how Postnikov tower works but I am proving it.
Oh, I learned about those my first year of grad school and never thought about them ever again, @Balarka.
16:40
okay so I have a matrix, lets see what the book says about stationary distribution. unless you want to teach it to me?
Have you seen the book by Mosher and Tangora?
or should I use wikipedia? lets see what the book says
In the absence of polarizable materials, the relation between E and D is $\vec{D}=\epsilon_0 \vec{E}$.
I had fun proving stuff about Markov processes in my linear algebra book, but I never got to teach that stuff — no time.
OK, @Semiclassic. I'll move on with my unhappiness.
In a polarizable medium, though, it becomes $\vec{D}=\epsilon \vec{E}=\epsilon_0 \vec{E}+\vec{P}$ where $\vec{P}$ is the polarization field.
(I may be misremembering the definition slightly)
16:41
okay heres what my book says
Yeah I figured I should finally learn them. I was recently a little a bit stressed (it's less than depression, really :P) so I figured it's a good side project for distraction.
I'm not sure I ever was taught that in Purcell, Semiclassic
@TedShifrin Hm, no, I haven't.
Google tells me it's a book about cohomology operations in homotopy theory. Interesting!
let $\mathbf \pi$ be a column vector of probabilities that add up to 1
OK, @Balarka: I'm glad you're not too badly off. It's a book where I learned that stuff. It's on cohomology operations.
16:42
Setting aside units, $\vec{D}$ is the electric flux density due to free charges, and $\vec{P}$ is the material's response to that electric flux. Then $\vec{E}$ is the combination of these two effects.
FauxPas: If you're using column vectors, then your matrix should be transposed!
whats the mathjax to bold symbols, I forgot
mathbf
\mathbf
But I'm not saying it quite right. Wrong minus signs, I think.
Oh, speaking of, I accidentally derived the Gysin sequence from the Serre spectral sequence.
16:43
no, for symbols
not for leters
oh, you need AMSfonts and boldsymbol for that.
I don't understand the geometry of the map that goes backwards, though, unlike you probably do with forms :P
It comes algebraically from a splicing fact for me.
A flag on a bus is fluttering in north direction and wind is blowing in east direction. Then how do I determine the direction of the bus (using vectors)?
Um, integration over the fiber, most likely, @Balarka.
16:44
@TedShifrin I'll note it down. Thanks.
$\boldsymbol \pi$
there we go
\boldsymbol
Yeah, that.
@ted typically this comes up more with magnetic field stuff, where it's the difference between H and B.
It's the same thing that makes the Poincaré lemma work, @Balarka.
which again is another material-centric concept.
16:45
Sounds way too vague, @Abcd.
oooh I see the problem Semiclassical
$p_{ij} = p_{j \to i}$
But I have seen at least one way in which this does matter mathematically, @ted
@Abcd Check what side the door is on.
@TedShifrin Oh it's a question in my exercise on relative motion. the answer is "bus may be moving in any direction between south and east"
$p_{ij}$ means the probability of getting to $j$ from $i$
16:46
Well, it's beyond the scope of what I need, @Semiclassic. I'm having enough trouble with this guy's sloppiness. I feel rejection coming on.
namely, I've seen at least one presentation associate the electric field to the one-form $\mathcal{E}=\mathbf{E}\cdot d\mathbf{l}$
@SteamyRoot Not possible. It's a question in an exercise.
so $\mathbf P = [ p_{ij} ] = [ p_{j \to i} ]$
I know, it's a joke :P
and now Ted is happy
16:47
And the electric displacement field to the 2-form $\mathcal{D}=\mathbf{D}\cdot d\mathbf{S}$
hopefully
@Abcd: You want to add a vector pointing east and the velocity vector of the bus and get a vector pointing north. Draw a picture and see how that could happen.
And there are some reasons for that.
FauxPas: It takes more than that to make a Ted happy.
16:47
I'm fine with the standard way to turn vector field into 2-form, @Semiclassic. That part is no issue.
But in that case it makes a bit more sense to think of $\mathbf{D}$ as a flux density and $\mathbf{E}$ as a field intensity.
let $\pmb \pi$ (\pmb \pi)$ be a column vector of probabilities summing to one
@TedShifrin Oh. So north is resultant? That's where I was making a mistake.
Right, it's the resultant.
But honestly this is terminology which one avoids nowadays.
It's historical and not very appealing.
16:48
$\pmb \pi$ is a stationary distribution of a markov chain if $\pmb \pi = \pmb \pi \mathbf P$
Now you're back to row vectors, not column. Stop that!
Yeah, pick one and stick with it.
I said column vectors, not row vectors, I never wrote row vectors
you are lying
Oh hey @Ted
16:49
i fixed it :P
$\pi =\pi P$ only makes sense if $\pi$ is a row vector.
thanks Semiclassic for fighting his battle
hi @Krijn
what the heck
that's true
Let's clarify: How do you want to normalize P?
By rows, or by columns?
The chat has become 10x more active after Ted comes in
16:51
Don't blame me, @Balarka.
I'm trying to read this physics/baby diff geo paper.
my book just says "vector" withjout saying its a row vector or column, but i was taught that without qualification "vector" means column
let me see my notes
I know that people who do computer graphics tend to use row vectors. I know not why.
Also my english starts breaking somehow
i have in my notes
LOL ... you were never any good at English, @Balarka :D
16:52
@TedShifrin There's a really irritating instance of that, let me find it.
I just hit page 30 in my thesis
Time for a break, maybe
$p_{ij} = \Pr(X_n = j | X_{n-1} = i)$
Congrats, @Krijn. For a masters thesis, that should be almost enough :P
@Krijn When's the deadlne?
@TedShifrin why is north resultant?
16:52
@TedShifrin Wait what, 30 pages for a master's thesis?
als m englsh strts brkng smhw @BalarkaSen
and for matrices we have, in my notes
smacks Astyx
@Abcd: Because the flag "feels" both the motion of the bus and the wind.
$\mathbf P = [p_{ij}]$
English breaks starts Astyx is I do 'nt understand Sinner guns wake somehow
16:53
so each entry is the probability of going from $j$ to $i$
I see how it goes
@TedShifrin It should be, but unfortunately, I have now only hit the part which was unoriginal, and need to type like 10 pages or so on new stuff
@TedShifrin Ok. TY.
so which way is it normalized according to my notes' convention
@SteamyRoot I don' t really have one, although I hope to finish somewhere in september, october-ish
16:54
Right, IDL has some issues with matrix multiplication.
I edited to add a bad pun to make it more Joycean.
entry $ij$ is the probability of going from $j$ to $i$
And it's precisely because of its origins in image processing.
@BalarkaSen Mildly disturbing
Time to add that sentence to the English project for school that I am preparing.
16:56
You should do other writers as well with just that sentence
my book has the rows sum to 1
basically, the image processing pov takes $p_{ij}$ to have $i$ as the column and $j$ as the row
because I don't know
@Krijn Planning on a PhD afterwards?
@SteamyRoot Ehhh, let' s see how this part of my thesis goes first
anyways.
16:57
No "s", @Semiclassic.
wikipedia:
roll
@GFauxPas The test should be this: If I start with a configuration like (1,0), is the next state still normalized?
A stationary probability vector $\pmb \pi$ is defined as a distribution, written as a row vector, that does not change under application of the transition matrix
fml, my book says "vector" to mean row vector
every book ive had so far, and Khan Academy, all assume "column" when saying vector
@Krijn Well my last year's English project was basically a sampling of a bunch of lines/paragraphs from whatever works of literature or poetry that came to mind, combined with a bunch of nonsensical and horrid stream of consciousness.
If I have the matrix \begin{bmatrix} 1/4 & 3/4 \\ 1/5 & 4/5\end{bmatrix}
16:59
I often use row vectors in advanced differential geometry, but I blame my adviser for that, @GFauxPas. Most of the world tries to stick to columns.
well, if I multiply on the right by the column vector [1,0]^T then that comes out as [1/4,1/5]^T.

« first day (2521 days earlier)      last day (2796 days later) »