« first day (2745 days earlier)      last day (2196 days later) » 

12:02 AM
(Hmm, with Cows and co. resolved their problems, I think h bar should become a bit more stable)
(In fact, it might be possible the world is starting to fix itself as my PhD now keep having wired calculations that make no sense recently):P
(The Dorian Gray Effect means it's either me or the world, no 3rd option on the way The Tradeoff works, hmm...) :D
 
12:20 AM
lol
 
12:38 AM
Nastiest guitar rift ever, you can't even hum it.
Lol electromagnitivity
or drive thru prolapse
I was just sitting at a drive through when someone came down the hill at 90 and murdered me. in the drive through. I'm just a girl.. ahhhhhHHHHH... :(
AAhhhHHHHHHH
Sounds like a dying bird..
AhhhhhHHH
AhahahhhhhHHHH
11 Hurricanes destroy major islands last year... trump can't afford it
now volcano maybe VEI 2 or 3 opening up with fissures
116 acres covered in lava
are you still mascarading your prayer of peace
jesuss
 
Going to nap for a bit.
I have not done any work yet, but should do some coding tonight.
 
1:10 AM
@vzn "theory precedes experiment" Sometimes. Maybe even often, but not in general. Theory is data driven just as experiments are designed to probe theories. Ask Ray Davis. Or Henri Becquerel. Or think about the motivation for the motivation for the postulating neutrino.
29 messages moved to Trashcan
@Semiclassical I feel like I say this a lot, but if you watched the OPERA announcement or read the paper you would have seen how very careful there were to not claim to have seen FTL neutrinos.
The claimed exactly that their data were consistent with FTL neutrinos and that they had not at that point found a problem with the machine. And they listed a fairly extensive set of tests they had done in search of faults.
Then they invited other members of the community to look at the data and the machine.
The end result was, I'm sure a bit of an embarrassment because you can be sure that they ran connection checks on their electronic signal cables, but somehow the GPS timing fiber-optic line got overlooked.
 
vzn
@dmckee ok. theory drives experiments, experiments drive theories. its a synergy. both things happen. yin + yang.
 
@GettingNifty Enough, please.
These are not suitable topics for this space.
8 messages moved to Trashcan
 
vzn
1:27 AM
actually, looking up that idea, realize now am paraphrasing...
 
1 message moved to Trashcan
 
vzn
> Whether you can observe a thing or not depends on the theory which you use. It is the theory which decides what can be observed. ---Einstein en.wikiquote.org/wiki/Albert_Einstein
 
2 messages moved to Trashcan
 
vzn
1:55 AM
@Semiclassical in other words, even pro scientists can espouse an ultimate form of confirmation bias, possibly even more rigid than laymen. maybe there are no experimental violations of QM because physicists are ("only") searching under the lamppost so to speak...
 
2:17 AM
@vzn nothing stopping you or vixra from going beyond the lamppost, fields wide open, less competition by this logic - surely you wont get lost in the dark ;)
 
vzn
2:31 AM
@bolbteppa lol and how does Physics compare to vixra? already -1 for my effort and the coterie of world class phds are either off/ silent physics.stackexchange.com/questions/402680/… ... (rats, could almost )( have at least gotten a tumbleweed badge) :P
2
 
 
3 hours later…
5:12 AM
Dawn in Chester!
Colourful
 
Very beautiful
@JohnRennie Your yard is huge!
 
@Cows no, it's the wide angle camera on the phone making the garden look bigger than it is :-)
 
:D
 
5:33 AM
Big bell test: How do they ruled out the possibility of many people picking the same setting?
Or is that not matter because the researchers don't know what they will pick in advance?
 
6:32 AM
I waked from very pleasant dreams both yesterday and today--so pleasant that I wish to stay in my dreamland.
yesterday I dreamt my sister and I went taking adventure, finding a channel which can direct to Janpan easier than a usual route to Japan.
today I dreamt I was in a research lab when a group of research students, probably in the field of condensed matter physics, entered.
 
7:08 AM
what was the research about?
 
8:01 AM
@Secret I can't recall such details in my dream. I sometimes dream I am calculating something but I seldom remember what I exactly calculate when I wake.
 
 
1 hour later…
9:06 AM
BEE has arrived
hurray
Oh man
it's an old old version
With a very bad typesetting
No wonder it was cheap
 
9:40 AM
@Slereah I just read "In many languages, such as French, the verb in any given tense takes a different suffix for any of the various combinations of person and number of the subject." How complicated your language is!
 
9:56 AM
That's why we're so good at math
 
10:09 AM
@Slereah then by analogous reason, I know why we are so good at art.
We had calligraphy courses in primary and middle schools.
 
Everyone did
you ain't special
I spent like my entire first year of school just learning how to draw calligraphy upper case
A skill I have never used, except to write conformal infinities
$\mathscr{ABCDEFGHIJKLMNOPQRSTUVWXYZ}$
Oh and I guess the set of bounded operators on a Hilbert space
and Lagrangian densities
 
10:26 AM
these art words can now be produced easily by computer.
in calligraphy courses we were taught to use hairy pen dipped with ink to write art words.
Indeed, I have never been good at writing art words. I guess being good at calligraphy is an innate ability.
 
I was a nerd even in primary school
We had to do a presentation once
I tried doing one on nuclear power
It did not go well with the crowd
 
10:41 AM
sounds like an "over achiever" :P
 
0
Q: Can someone explain the concept of 'Negative Probabilities' in an intuitive manner?

AScientistCan someone explain the concept of Negative Probabilities in an intuitive manner? I can't seem to understand this concept. I hope someone can explain this concept in an intuitive manner.

ST
 
0
Q: What shape is the universe?

Samuel carsonMe and my friend have been discussing what the shape of the universe is and we cant come up an idea of it. can someone please help.

 
It does not exist
 
 
2 hours later…
12:27 PM
93
Q: Student caught cheating when leaving class after handing me the exam

JosephUsually I am quite clear that cheating and plagiarism is unacceptable. Although every semester I have to deal with several cases of plagiarism, I had not expected students to brazenly cheat in exams. A few seconds after one student handed me the exam, as she was leaving, I noticed that she had wr...

I think that's why they started "open book" exams.
 
If it's possible to cheat on an exam, maybe they're testing the wrong skill
 
ulimately, they're just cheating themselves
 
well, only if their job is related to their degree
which is unlikely
Apparently the EM bundle is trivial for electric charges, but it becomes twisted if there are magnetic charges
neat
Magnetic charge of $n$ gets the bundle with a structure group of $\mathbb Z_n$
 
12:52 PM
I don't know enough about bundles to understand what happens when an EM bundle becomes twisted, my guess it might be making things multivalued or something
 
@Slereah i meant in the ultimate ultimate sense; when they have forgotten everything they learned in school :-)
 
I wonder how much math I use in daily life that I don't use for like
Physics
I think mostly basic algebra and trigonometry and geometry
those are always useful
Make children code a game, I'm sure they'll learn bloody geometry then
You need a fair bit of geometry to make graphics
 
1:29 PM
0
Q: How do I make these questions better?

HemaI have been banned from Physics Stack Exchange, and I think it is mainly because of these two questions which were not received well. The thing is that for both these questions I have received hints from the comments section and I managed to work it out. What should I do about them? Would someone...

 
 
2 hours later…
3:18 PM
Why would you care to distinguish out spinors of $SL(2,C)$ which transform under the complex conjugate and put dots on them? Have a hand-wavey reason so far
 
Because you're using the complexified spinors, I suppose?
And you want to get real results
So you need to take some complex norm for that
I'm guessing you don't care about such things when you use Majorana spinors
 
3:35 PM
I think, because probability is now the time-component of a current vector, you don't need to preserve it, so unlike unitary transforms the complex conjugate doesn't need to be in any way related to the original transform, but you can show that a contravariant spinor transforming under the complex conjugate behaves, under spatial rotations, the same way as a direct transform of a covariant spinor :\
 
one thing I always wonder is, what does the Dirac equation look like if you're using the real Clifford algebra
 
Looks the same basically
 
different gamma matrices, I suppose
And I guess $\psi^* = \psi$
The Dirac equation, as the relativistic equation that describes spin 1/2 particles in quantum mechanics, can be written in terms of the Algebra of physical space (APS), which is a case of a Clifford algebra or geometric algebra that is based on the use of paravectors. The Dirac equation in APS, including the electromagnetic interaction, reads i ∂ ¯ Ψ e 3 + e ...
Oh there we go
Oh the spinors are still complex?
 
If you write it in standard form as a Schrodinger equation, there are complex $i$'s attached to the $\alpha_y$ and $i \beta$ matrices only, and you can just transform the basis so that you end up moving the $\beta$ to the $\alpha_y$ position and the $\alpha_y$ to the $ \beta$ position, multiplied by $i$, so everything is now real
The transform is something like $U \sim \alpha_y + \beta$
Then you have $U \alpha_x U^{-1} = - \alpha_x$, same with $\alpha_z$ but $U \alpha_y U^{-1}= \beta$ and $U \beta U^{-1} = \alpha_y$ or something
 
What's the group associated with the spacetime algebra, is it still SL(2,C)?
 
3:43 PM
What do you mean
 
reminds me. one of the things I want to do this summer as a side project is figure out how to numerically simulate scattering in various wave equations
e.g. schrodinger equation, dirac equation, NLS
with an eye towards getting code that I can use to visualize some of the dBB stuff b/c ofc that's what I'd do
 
From the Clifford algebra $\{ \gamma_a, \gamma_b \} = 2 \eta_{ab}$ you can multiply on the left by any invertible $U$ and right by $U^{-1}$ and then $U\{ \gamma_a, \gamma_b \} U^{-1} = \{ \gamma_a', \gamma_b' \} = U2 \eta_{ab}U^{-1} = 2 \eta_{ab}$ means the chosen basis is arbitrary
Related to that, I stupidly thought all first order qed processes were zero
 
The example I really want to implement is something like scattering of spin-up electrons off a potential of the form $V=S_z \tanh(x/a)$ along with an interaction term $S_x$
 
But then it turns out all the photoelectric effect stuff is all first order scattering
 
in which case the electron is most likely to either scatter off as a spin-up particle or transmit as a spin-up particle
mostly I'm interested in that case b/c it's similar to the Landau-Zener problem
 
3:48 PM
Apparently the Pauli matrices are just Clebsch-Gordan coefficients of the expansion of a vector in terms of spinors, want to see if you can get them by accident that way
 
I thought they were the solder form of the spinor bundle
 
Only from the perspective of solder mechanics...
 
solder mechanics?
 
Oh this is a real thing
In mathematics, more precisely in differential geometry, a soldering (or sometimes solder form) of a fiber bundle to a smooth manifold is a manner of attaching the fibres to the manifold in such a way that they can be regarded as tangent. Intuitively, soldering expresses in abstract terms the idea that a manifold may have a point of contact with a certain model Klein geometry at each point. In extrinsic differential geometry, the soldering is simply expressed by the tangency of the model space to the manifold. In intrinsic geometry, other techniques are needed to express it. Soldering was introduced...
haha
 
3:53 PM
omg another rarity
'A vielbein or solder form on a manifold X X is a linear identification of a tangent bundle with...' ncatlab.org/nlab/show/vielbein
An ncatlab which actually explains what an advanced thing is
 
those do exist occasionally
though usually only if you look at the 'introduction' section of the page
 
This might be the third I've seen so far
 
the page on hyperfunctions is decent: ncatlab.org/nlab/show/hyperfunction
the page on resurgence theory is nice insofar as it summarizes a lot of slogans: ncatlab.org/nlab/show/resurgence+theory
 
You must beware of the nlab loop
when you try to follow a definition and it ends up being a loop
 
Hi to all. @Slereah Could I take the courage and bother you for a bit with a question? If you have any idea you tell me.
 
4:04 PM
just ask and see
 
Okay; In Chiral Perturbation theory one gets the following first order Lagrangian: $tr[\partial_{\mu}U\partial^{\mu}U^{\dagger}]+ \bar{N}(i\partial_0 - \frac{g_a}{2f}\tau \cdot (\sigma \cdot \nabla)\pi - \frac{1}{4f}\tau \cdot(\pi \times \partial_0 \pi) )N -\frac{1}{2}C_S \bar{N}N\bar{N}N - \frac{1}{2}C_T(\bar{N}\sigma N)(\bar{N}\sigma N) $.
I can see that there are two contact terms giving a potential of the form $ V= C_S +C_T \sigma_1 \cdot \sigma_2 $. But, it is argued I should also have a one pion exchange term, equal to $ \simeg \tau_1 \cdot \tau_2 \frac{ \sigma_1 \cdot q \sigma_2 \c
 
Errr from what I remember the "real" lagrangian is of some exponential form
and you get theories of order $n$ by taking the taylor series
 
Does it seem sensible to you to take the square of the above term(as if I have second order perturbation theory)? Just your opinion.
 
I dunno
The Lagrangian of some $SU(N)$ effective model is something like $$\mathrm{Tr}(\partial_\mu U^\dagger \partial^\mu U)$$
For $U = \exp(ig\phi^a \lambda_a)$
With $\lambda_a$ the generators
That's usually how chiral theories are constructed
 
Ok. Thank you. Just for completeness, Weinberg shows this in these papers were he discusses nucleon interactions.
S. Weinberg, Phys. Lett. B251 (1990) 288
S. Weinberg, Nucl. Phys. B363 (1991) 3
So, I have to understand how he gets to that result...
 
4:17 PM
Well one of my thesis was on that topic, so lemme see if I can find the ref I liked
S. Scherer, Chiral Perturbation Theory : Introduction and Recent Results in the OneNucleon Sector, arXiv :0908.3425v1
I had this reference
 
how many theses u got
 
I did two master thesis
 
y tho
 
Couldn't enter a PhD due to funding, so I did a second year to try again
 
Anonymous
@Slereah I never asked. What was your undergraduate degree in?
 
4:25 PM
Thanks; I' ll have a look.
 
Errr what is that in european terms
First few years of uni?
It was just "physics"
generic physics degree
 
Anonymous
@Slereah I see. And masters in theoretical physics? (Any particular specialization?)
 
My master was in subatomic physics
 
Anonymous
I thought you had a computer science degree too
 
I do, yes
Gotta get paid
 
Anonymous
4:26 PM
Lol. So many degrees!
 
and so little jobs
 
Anonymous
Was it a masters in CS? Or a dual major?
 
It's a sad world
Well, not computer science
Software engineering
 
Anonymous
Well, that's close :)
 
I'd rather not soil science by calling it computer science :p
It wasn't the algorithm and abstract machine kind of degree
 
Anonymous
4:29 PM
Hehe. :P But well, I do think a CS related degree is a good investment for at least being employable at any point of time (in today's times mainly)
 
Nah, it's all a meme
if you don't have previous experience it's a nightmare finding anything
 
I got no CS degree
yet
I'm working as a data scientist
o.o it's doable without a degree
 
yeah, I did that too
But it's surprisingly rare
 
hmmm
 
Everyone says they want to get data science people, and that there aren't enough people
 
4:31 PM
actually several of my physics friends are data scientists lol
 
but they aren't that fond of hiring me!
mb I smell
 
D:
well you are in europe right
maybe it's different in europe
 
could be
France isn't really on the forefront of technology usually
 
Anonymous
@enumaris CS degrees at present don't really deal much with data analytics. There are new data science masters degrees coming up, but that's a very recent phenomenon. For now, they (data science) just look out for technically oriented people who can infer conclusions from data and can think logically. So physicists and mathematicians are good fits :)
 
@0celo7 pde question for you
 
4:32 PM
Ok
 
Let's use the usual xkcd
 
yeah, but I got hired to do NLP...not even that stats related LOL
 
consider the pde $\partial_t h = \partial_x^2 h +\lambda (\partial_x h)^2$
 
All the algorithmic part of machine learning and data science is pretty simple, really
 
I can't view xkcd at work :(
 
4:34 PM
if $\lambda=0$, that's the heat equation
 
It's choosing the right thing that's hard
 
simple conceptually
 
is there an obvious name for the nonlinear term? I want to say it's an elastic or gradient term or something like that
 
but doing backprop by hand is annoying as hell
 
Gradient term
 
Anonymous
4:34 PM
@Slereah *the mainstream algorithms :P There are lot of people (for example in Microsoft) who do fundamental level research work on algorithms too
 
Even the fancy algorithms are never that complicated
 
I remember seeing some neural network using the GRAHAM SCHMIDT THEOREM
Be still my beating heart
 
contrastive divergence using gibbs sampling is kinda complicated
but RBMs are not popular anymore...
 
really the hard part is like
Choosing the right data to analyze
and making it efficient
 
4:37 PM
also the api's make training an algorithm super simple
like if u use sci-kit learn
just load the data
and call .fit
done
 
I coded it all in C because dang it
 
I'm not familiar with ML packages in C
 
probably wasn't the best choice
 
is there good support?
 
Who said anything about ML packages
I coded the algorithm
 
4:38 PM
Why would you ever voluntarily choose C? :P
 
y tho
 
::codes in COBOL::
3
 
that's like reproducing work
 
It's the language I was taught!
 
D:
 
4:39 PM
@enumaris more flexibility?
 
@ACuriousMind we caught you
 
but unless you are coming up with your own algorithms...
it doesn't seem necessary
like I coded my own neural network before to "learn how neural networks work"...and it was just a giant pain in the ass
alternatively I could just use Keras and be done in 2 minutes...call .fit and wham
 
Oh well
it was a good learning opportunity
 
XD
 
@0celo7 I spent the last few days using reflection in ABAP, my soul is probably irredeemably lost now :P
 
4:42 PM
also a lesson I learned : if I ever try to recode some neural network, I'll use a graphics card next time
pretty slow if you don't have a lot of parallelization
 
well, depends on how big your network is
if your network is huge, then GPU speed up will be massive
if you have a tiny network, it's really no big deal
 
That and I had a shitload of data
 
I see
 
like 4GB
 
the dataset I work on in my free time is 300gb :P
 
4:44 PM
a lot of the work was preprocessing the data
@enumaris all your nude photos?
 
I spent 10 days of processing power to preprocess that data
and then I realized I had a much faster and better way to do it
LOL
 
At first I tried preprocessing in bash
 
wasted 10 days time
 
Anonymous
@enumaris I was trying to write a simple program related to counting clusters in a large matrices (100,000x100,000) a few months back. The first version took days to run using the "standard" algorithms took days to run. Took me 2 months of back and forth interaction with the SO guys to reduce that algorithm's run time to 2-3 hours. So, I feel that while standard algorithms are always available, they're not always beneficial due to their time and space complexity.
 
Anonymous
Being able to drastically reduce that complexity does require in-depth knowledge of algorithms.
 
4:44 PM
Then it was so slow that I coded it in C and it was like
ridiculously faster
 
yeah...there's a lot of details and work to be done in making algorithms efficient
there's stuff on the algorithm side, and then stuff on the computing side as well
but mostly I just let others do as much of the work as possible
 
I fell for the trap of thinking a faster activation function would help
I found one that was much faster to compute
but then turns out it converged much more slowly, too!
 
hehe
it does happen
 
Next time I think I'll try to get some piecewise linear one
Because the exp function is so slow
 
setting a large learning rate can have the same effect. It drops the loss quickly at first, but then plateaus and slows to a crawl.
 
4:47 PM
50% of CPU time was just computing exponentials
 
hmmm
 
would be nice if the CPU had some microcode exponential
 
I don't get into how CPUs actually execute code lol
I just let numpy take care of it
 
can be pretty important!
the exponential function is done by like...
Some Taylor series-like process
I forget the name
 
dang
I've read the word "gold" too many times
now the word feels weird to me
golld...gold...goollld....
 
4:51 PM
Ah, the wonder of semantic satiation.
 
I assume that np.exp() is as efficient as the exponential can be made to be
 
Chebyshev Polynomials
apparently
 
historically, trig functions were computed in the computer age with CORDIC
 
@enumaris well yes, but point is, still not very efficient
In a neural network, you don't need precision from the exponential function
the exp function gives you an exact result up to the LSB
but that is really not useful
 
so you want to write a custom exp function that is less precise?
feels like a pain
 
4:53 PM
CORDIC (for COordinate Rotation DIgital Computer), also known as Volder's algorithm, is a simple and efficient algorithm to calculate hyperbolic and trigonometric functions, typically converging with one digit (or bit) per iteration. It is therefore also a prominent example of digit-by-digit algorithms. CORDIC and closely related methods known as pseudo-multiplication and pseudo-division or factor combining are commonly used when no hardware multiplier is available (e.g. in simple microcontrollers and FPGAs), as the only operations it requires are addition, subtraction, bitshift and table lookup...
 
optimal is not nearly as useful as satisfactory in a lot of cases
 
^another method
 
I'll just go with the get a stronger computer method
 
Well it's not too hard, really
A piecewise linear function is just gonna be a bunch of if's
 
Sup guys. Weekend is finally here. Yay.
 
4:55 PM
but then you have to worry about vectorization
and then you have to go under the hood of the program to make sure if you have a huge matrix that you want to compute the element wise exponential of that you parallelize the if statements to utilize vectorization.
and parallelization is super annoying
 
that it is
but very useful for NNs
 
which is why I let the software worry about it
tensorflow can do all the parallelization for me, I won't worry about how to do it myself
 
I even looked into better matrix multiplication algorithms
Did you know there's an algorithm for matrix multiplication that works in $\approx O(n^{2.6})$ instead of $O(n^3)$
But it has so much overhead that it's not actually useful unless it's a matrix of a few million by million entries
 
@Blue Sup Blue. How was your day? Are the exams finally over :D?
 
few million by few million matrix is gonna be too big to keep in memory
unless it's a sparse matrix maybe
but that's like terrabytes of data lol
 
4:58 PM
It's the kind of stuff you're gonna get in climate simulations, I think
 
Anonymous
@Slereah The Strassen one?
 
@Blue there's a few
 
I have a 70k*70k matrix which is sparse. If I don't save it as a sparse matrix, it's like 5-10gb or something
 

« first day (2745 days earlier)      last day (2196 days later) »