The h Bar

0celo7

12:35 AM

@BernardMeurer maybe we should ask him more algebra questions

I think he's annoyed by all of the geometry

Bernard Meurer

@ACuriousMind will you teach me something?

0celo7

@BernardMeurer are you still learning magic

Bernard Meurer

@0celo7 It will be a while

Try cross-compiling the linux Kernel and you'll see

ACuriousMind

@BernardMeurer Depends on what it is

Bernard Meurer

@ACuriousMind I'll let you pick, I just want you to teach me something science related

ACuriousMind

12:38 AM

Uhhhhhhhhh

@FenderLesPaul, why don't you teach @BernardMeurer something sciency?

Bernard Meurer

Whoever teaches me something first get's a 3d printed goodie

ACuriousMind

*whoever

There, taught you grammar ;)

dmckee

@BernardMeurer I've got a lab monke^W^Wresearch student doing that for me already.

Bernard Meurer

@ACuriousMind SCIENCY

@dmckee Yeah just shred my heart to pieces

dmckee

It gets worse. This guy is on the verge of graduation and already has a development engineer's job nailed down.

ACuriousMind

12:45 AM

@BernardMeurer I'm having trouble thinking of something. No idea what to pick, and what qualifies as "teaching" (Linguistics, and hence the study of grammar, is a science, though ;P )

0celo7

@dmckee what's a "monke^W^Wresearch"

FenderLesPaul

@BernardMeurer what is Nature's light show?

ACuriousMind

@dmckee Shouldn't that be "^H^H"?

Bernard Meurer

@FenderLesPaul Northern Lights or Aurora Borealis

0celo7

@FenderLesPaul christmas tree

FenderLesPaul

12:58 AM

both wrong it's death

death is all around us

ACuriousMind

::screams::

Bernard Meurer

Seventh Seal

ACuriousMind

@FenderLesPaul Why would it be a light show, though?

FenderLesPaul

red light

0celo7

@ACuriousMind maybe the body produces luminescent gasses or something when decomposing

methane can be lit

3507

1:14 AM

is dirac notation a bad thing? should it be avoided as much as possible outside the context of qm?

ACuriousMind

@3507 What makes you ask that?

3507

because why don't people write linear algebra books with dirac notation?

well it's uncommon.

ACuriousMind

@3507 Mostly because the people who write those are mathematicians, not physicists.

FenderLesPaul

@ACuriousMind if I have a vector field $X$ orthogonal to another vector field $Y$ does $\mathcal{L}_Y X$ map to another vector field that is still orthogonal to $Y$?

3507

@ACuriousMind right though if it is a superior notation then shouldn't it be adopted everywhere?

FenderLesPaul

1:17 AM

where $\mathcal{L}_Y$ is the Lie derivative along $Y$

ACuriousMind

@FenderLesPaul I don't see how, the Lie derivative knows nothing of the metric structure

@3507 What would make it superior?

3507

@ACuriousMind that's what I'm asking.

FenderLesPaul

@ACuriousMind That's what I figured, thanks broski

ACuriousMind

@3507 Well, I don't see anything. And there's also not really something it does worse, if you are consistent with the directions of the actions of operators

user54412

1:42 AM

@tpg2114 Have you ever used a Blue Gene, or even any other IBM cluster? If so, do you have advice on optimization flags? The compiler seems to be strongly hinting that it takes no responsibility for screwing me over at -O3 or above.

dmckee

1:55 AM

@ACuriousMind Some terminals accept control-w as "erase word", so no. That's what I meant. It would have taken too many ^Hs.

user54412

^ TIL

dmckee

@ACuriousMind I know at least one professional mathematician who prefers dirac notation for linear algebra. But he's a bit of an iconoclast.

Bernard Meurer

2:17 AM

Sigh

No one taught me some magic

tpg2114

2:30 AM

@ChrisWhite Yeah, I have used BG/Q and BG/P systems

The compiler is actually being more helpful than others -- O3 could be screwing you everywhere, it's just xlf is nice enough to tell you about it

In our code, with both Intel and GNU on regular systems, O3 gives us straight up wrong answers and is no faster than O2

And O2 gives us wrong answers if some options are turned on... so we've been using O1 :/

And that's been true on SGI, Cray, Intel...

So -- I would recommend you run some timing and accuracy studies. Start at O2 and make sure it's cool and get a baseline

Then run O3 and see if you get the same answer and if it speeds up

If it does, run with it. If it doesn't, don't use it

dmckee

@BernardMeurer The conservation of angular momentum is close enough to magic for all practical purposes.

My department has this cast iron platform about 18 inches across mounted on very good bearings so that it spins freely in the horizontal plane.

At public demonstrations we stand people on that, hand them a spinning bicycle wheel and ask them to turn it to and fro. This generates wonder and grins from all and sundry.

People passing by simply stop and stare.

0celo7

@BernardMeurer ok, I'll teach you some magic

tpg2114

@dmckee I also like the bike wheel with one side of the axle attached to a string/wire and then you hold the string so the wheel is off the ground. When the wheel is stationary, it flops over. If you hold the wheel upright and spin it, then let it go, it stays vertical

MIT Physics Demo -- Bicycle Wheel Gyroscope

►

dmckee

2:47 AM

We do that one, too, and people ooh and aah. But they stay longer with the platform.

tpg2114

Oh, @ChrisWhite one other thing our email threads from the BG/P indicate is that using the mesh interconnect or the torus interconnect greatly changed our performance depending on our domain decomposition and distribution to the processors... I don't see how we changed that option, but it's something else to look into

The platform would be cool. I haven't seen that one before

dmckee

And of course the van de Graff generator is always a hit.

hwlau

@tpg2114 What is the reason of the wrong results? Or in what situation? I haven't notice there are problem of -O3

tpg2114

@hwlau We don't know. Our code is 700,000+ lines long over several hundred files and somewhere O3 makes the compiler do something that is unsafe and has side effects

dmckee

@hwlau At least with gcc the optimizer breaks various guarantees that IEEE standard says you're suppose to have about how floating point number behave (at -03).

tpg2114

2:50 AM

@dmckee Fortran is free to break it at lower values also. We've had to add flags to specifically require enforcement no matter what

dmckee

Most of these effects are subtle and don't come up often, but that just makes it worse because there is no easy way to test them.

tpg2114

We also had a bug where O3 would change the order of two lines inside a loop that shouldn't have been changed. Like A = B; A = A + B kind of thing

And changing the order changes the result because they had side effects

hwlau

How can I know if there are problem with the optimization flag in my program

dmckee

@hwlau Compare results with and without the flag.

Also, if you use a PRNG, force the seed to be the same between the two runs.

tpg2114

Yeah, run with O0 and compare the result to the other flags

hwlau

2:52 AM

Is -O0 the same as no flag?

tpg2114

Depends on the compiler. Some will do that, others may default to another option

So be explicit -- O0, O1, O2 and O3

And O0 is the "gold standard"

dmckee

^ That.

tpg2114

What language are you using?

hwlau

mainly C++

how likely will the wrong compilation arise when I just test it for a small problem size or a subset of problem

tpg2114

It may also be worth adding -fp-model source with Intel compilers and... need to look for the equivalent in gnu

dmckee

2:55 AM

A side effect of not trusting the compiler's optimizer is that you get to re-introduce many of the ugly hand optimizations that everyone tells you not to learn or bother with any more.

tpg2114

For us? It depends how bad it is. For example, we have some simulations that will flat out crash and blow up on O2/O3 but not O1

We have others where the reverse is true

dmckee

Though I admit that the only one I tell students about is loop re-ordering.

tpg2114

We have some where we can only see the difference in the 3rd decimal of things like time step/etc

dmckee

Because it can have such a dramatic effect so easily.

tpg2114

And then we have some where everything runs perfectly but never converges or converges to the wrong answer

But for our code, it is noticeable on any simulation if we compare the actual values in each cell. There's a difference every time, and not like the 16th decimal place either

It's usually in the 2nd or 3rd decimal

hwlau

2:57 AM

@tpg2114 so there are problem related to accuracy

tpg2114

Yes, but that can cause codes to crash and answers to be wrong

Just by changing the optimization flag

dmckee

@hwlau But remember that numeric algorithms may be making assumptions about the inaccuracy of individual steps. Break those assumptions and things that were guaranteed to converge on a correct result can lose that guarantee...

tpg2114

^ plus it means our end result can be flat out wrong. Like we compare to analytical or experimental data and aren't even in the ballpark

hwlau

If it is the case, it is hard to check

tpg2114

Like we get entropy violating conditions and stuff

hwlau

3:01 AM

unless I always run both version for all problem

tpg2114

Well... you can do what we did and see it on the very first problem you tried

And then you can say if it can't do one problem, then it's not reliable for any problem

hwlau

will this kind of problem occurs for those scripting language like python, and maybe matlab and mathematica

I did encounter problem in C++ before, but I don't really know the cause for some of them. So how likely will the -O3 cause these kind of problem

tpg2114

It's quite likely that optimization is to blame for incorrect answers that you cannot otherwise identify the source of

We run with O1 now for everything, and still have to specify that IEEE standards are strictly enforced, to make sure we get the correct answers.

user54412

I get non-associativity and all that. What scares me is references to "transformations that replace floating-point library functions with other library functions or with constants"

user54412

what single number could you possibly be replacing my function call with?

tpg2114

3:08 AM

@ChrisWhite That means that it would look for SQRT(2) function calls and replace it with 1.407

hwlau

which flag is related?

tpg2114

And hopefully include more digits than 1.407 but who knows

user54412

@tpg2114 the wording makes it sound like it will look up sqrt(x) and replace it with 1.5

0celo7

@ChrisWhite as long as $x^2\sim 1.5$ that makes sense

hwlau

so it related to all float, double, long double, or maybe __float128

tpg2114

3:09 AM

@ChrisWhite Well.... you told it you wanted the code to go fast and return 1.5 is much faster than sqrt :)

user54412

Robustness, accuracy, speed. Choose one.

tpg2114

@ChrisWhite When I publish the paper I'm working on, you'll have a method that gives you the first two together :)

The title will actually be something like "Improving the robustness and accuracy of the compressible MacCormack scheme at low-Mach numbers"

@hwlau All I can really advise you to do is develop a suite of canonical problems your code is supposed to be solving and that you know what the answer is supposed to be. Test the crap out of it on every combination of options for every compiler on every system you have access to

user54412

^ yes

tpg2114

We have automated test scripts that run on 15 different machines with 2 different compiler suites that runs every time a commit is made to the code

Some tests are unit tests -- we feed in MyFunc(X) and check that we get the number we want back out

These actually revealed most of the optimization flag issues

user54412

except I'm still at a loss as to how to test stuff on this fancy new large cluster we have access to

tpg2114

3:16 AM

Most are integration tests. So we run Sod's shock tube, we run vortex through a pipe, we run uniform flows

We have... uhhh, 200 test cases in our testing suite? And we only get like 40% code coverage because we have so many options in the code

And it's taken us years to get to this point

@ChrisWhite I wasn't impressed with the BG systems... that whole 512MB of memory thing per core really made things hard for us to run

user54412

they don't let us do anything smaller than 512-core jobs

0celo7

Coding nerds, can we please do something interesting like epsilon delta proofs

tpg2114

For a given delta, there is an epsilon in the neighborhood of optimization flags such that you get the right answer

2

user54412

All your mathematician friends are gone. This room is PhD-level software engineering now.

hwlau

usually, I switch the problem quite often that I do not have the time to find some canonical problem and develop some kind of unit test. Are there an easy to do so?

tpg2114

3:18 AM

Fluid dynamics?

I mean... what does you code do. Sorry, I don't remember

hwlau

mostly solving some complicated differential equation

tpg2114

With what approach? Runge-Kutta? Something implicit?

user54412

^^ umm, that describes all of physics, right?

0celo7

@ChrisWhite ...

tpg2114

Heh... yeah

hwlau

3:21 AM

quantum master equation, light, BEC

@tpg2114 Usually explicit, RK, and a bit implicit

tpg2114

I don't know anything about those... Is it an ODE or a PDE?

hwlau

PDE

tpg2114

Initial value, boundary value, mixed? Is it hyperbolic? Parabolic? Elliptic?

How many dimensions?

Smooth or discontinuous? What spatial discretization do you use?

user54412

On the topic of software -- someone in my department recently started using a code that deals with data dumps not in binary or ascii, but in a unique base-244 encoding or something (all the printable extended ascii characters). I felt so bad for him.

0celo7

Anyone here know any solid state physics?

tpg2114

3:24 AM

Wow, that's disgusting... was the original software implemented in a core-rope system?

user54412

haha for all I know it was

hwlau

@tpg2114 mostly diffusion type equation. I have coded for 1,2,and 3D. Just have a paper in 3D finished. It should be smooth, but I did look at the choatic regime. And all spatial discretization is simple uniform

tpg2114

@hwlau So if it's a diffusion type problem, can you run a heat diffusion problem on it? Simple constant coefficient

Those have analytical solutions

So you can check how your code converges, how the accuracy compares, etc

user54412

diffusion sounds nice, but it sounds like you mean Schrodinger-like

hwlau

@ChrisWhite Yes. Nonlinear one

3507

3:28 AM

@0celo7 depends... i only read like 100 pages but I can try to answer your question.

0celo7

@3507 what's the current state of research on La/Ac resistant amorphous materials

3507

@0celo7 idk about current states of research sorry.

user54412

I have a friend in applied math. In one grad-level numerics course, the prof asked everyone at the beginning "what equation would you like to solve?" Everyone said the standard "heat," "wave," "Navier-Stokes," etc., and the prof said each would be trivially solved in the course. Then my friend said "Schrodinger" and the prof replied "impossible."

hwlau

is it a joke?

the navier-stoke is not easy too

tpg2114

Okay... so let me show my ignorance here... What is it that makes Schrodinger that much harder than something like Navier-Stokes?

user54412

3:31 AM

The complex numbers, somehow. I don't really understand myself.

0celo7

If the Hamiltonian is time dependent everything goes to shit

hwlau

I don't really know. But the probability have to be conserved and it is straight Hamiltonian system

as long as I add some coupling with external system, the problem come

0celo7

@ChrisWhite only a fucking engineer would say heat

and not a very ambitious one at that

tpg2114

@0celo7 Analytically? From a numerical standpoint, we suffer from A) wide range of eigenvalues, B) wide range of scales (like fast/slow processes, or small/large wavenumbers, etc)... Does Schrodinger do that?

And I suppose we also have problems when multiple source terms exist, they are large, and they are opposite signs (sources and sinks)

0celo7

Off the top of my head, dunno.

user54412

3:34 AM

@hwlau Navier-Stokes is actually quite simple in principle. The discretization is straightforward, and if you have any viscosity it only stabilizes the numerics. Nonlinearities might make converting between conserved and primitive variables a pain, but any explicit method will evolve the PDE without caring about such details.

0celo7

But it's $\mathrm{i}\partial_t\psi=H\psi$

here $H$ is linear and self-adjoint but might be time-dependent

tpg2114

@ChrisWhite True -- the difficulty in Navier-Stokes numerically is being cost effective. Especially for high Reynolds numbers

user54412

Exactly. With fluids, your underlying scheme can't really be wrong. But I feel like with Schrodinger you can do something very wrong fundamentally.

tpg2114

I wouldn't go that far...

user54412

well, okay, you can always be wrong I guess

tpg2114

3:36 AM

So long as your scheme is consistent with the equations, it is okay. But I would think that's true for Schrodinger too

0celo7

@ChrisWhite where was the nerd who said "Einstein"

hwlau

@tpg2114 not related Schrodinger. It does mix the fast and slow time scale. We usually make some kind of elimination to remove the fast time scale. But it means the resulting equation can have many simliar time scale.

user54412

"high Reynolds numbers" -- I sometimes forget there are Reynolds numbers less than $\infty$.

tpg2114

@hwlau My thesis is on preconditioning to remove the fast time scales for low-Mach number flows so a compressible scheme can be used there

@ChrisWhite Do you know how angry it makes me that LaTeX uses $\infty$ and not $\infinity$? I cannot tell you how many times I mispell their ridiculous little shorthand. I end up taking more time fixing it than it would take to type the extra letters correctly

hwlau

@ChrisWhite Is it the same as superfuild of BEC?

user54412

3:39 AM

Uh, I don't know anything about superfluids. I meant normal, everyday things like air and the magnetized plasma accreting onto a black hole.

tpg2114

Superfluids are awesome. And quantum turbulence is really intriguing

hwlau

Well, a normal schrodinger equation can be separated into two coupled PDE, if it help your understanding

0celo7

this chat is ridiculous

tpg2114

But I don't have time to work on it

0celo7

this is a PHYSICS chat

not programming

tpg2114

3:40 AM

@0celo7 Have you seen a single line of code? I believe everything we've been talking about has been physics and how to solve physical problems.

Talking about making sure an optimization flag doesn't screw you is like making sure you check that your GPS clocks are synced before you tell the world about superliminal communication.

O3 -> loose wire

user54412

what's 05 then? running the experiment while drunk?

tpg2114

Anything cold fusion?

user54412

@hwlau Are you referring to that imaginary time method? Or something else?

hwlau

@ChrisWhite No, just writing Psi(x,t) = A(x,t) + i*B(x,t)

you can find it out

tpg2114

Imaginary numbers really bother me...

0celo7

3:44 AM

@tpg2114 why

tpg2114

I have a hard time discerning their meaning in physical systems. I only have some vague connection to phase lag/delaying in time

Which I can deal with for like a pure sine wave, I get that. But once it gets into more complicated things, it's confusing

0celo7

I've never gotten a good explanation of why there's an i in the Schroedinger equation

tpg2114

Like... if the eigenvalues of the velocity gradient tensor in a flow have a complex eigenvalue pair, it means the flow is swirling there. I know the rule, but I don't have a good feeling for why

0celo7

now ACM will come in and tell me something about a prequantum distribution and Lie theory

tpg2114

I made the mistake of taking a nap at 4pm and slept until 9... now it's going to be a long night

user54412

3:53 AM

@tpg2114 As long as you're here...

user54412

Tell me your thoughts on MPI 3

0celo7

Tell me how to do this analysis proof PLEASE

tpg2114

@ChrisWhite I haven't seen a need for most of the things in there

user54412

non-blocking collectives are sorta nice

tpg2114

The most appealing thing is nonblocking collectives

Huh, great minds...

user54412

3:55 AM

one person in the group wants to use them, but it seems the standard is a bit too new to be common

tpg2114

Was MPI-3 the one that introduced interprocess communication?

Or was that in MPI-2 also

user54412

this is why I always laugh when people on SO or Programmers or whatever say "use C++14" -- as though that will work anywhere

tpg2114

The appeal to the collectives for me would be in our time step calculation routine. About 3% of our runtime is spent finding the minimum value of time step in the domain

If we can overlap that, it would be great

user54412

MPI-2 has intra and intercommunicators or whatever they call them, but the latter seemed somewhat limited when I glanced at them

tpg2114

I use intercommuncation, it's really cool

user54412

3:58 AM

@tpg2114 You know what makes relativity awesome? The time step is set by the speed of light and never changes throughout my simulation :p

tpg2114

I have two completely independent codes and was able to couple their executables together with the interprocess communication

@ChrisWhite Yeah... but I'd rather spend 3% runtime than deal with relativistic effects that come with your work :)

hwlau

@tpg2114 @Chris I just think a bit about why the schroedinger equation are hard to solve. Consider a simple quantum system described by du/dt = i * u which has a solution on the unit circle u(t) = e^it. If you use either Euler or RK4, it is unstable and the solution either go to 0 or infinity. However, the law of physics constraint it on the unit circle and say it must have amplitude = 1 which cannot be fulfill by many numerical algorithm.

tpg2114

@hwlau But an implicit method should solve it just fine right? Like backward Euler

hwlau

In the system with very tiny dissipation and driving, that is larger than the numerical errror introduced, then everything is fine

tpg2114

Or IRK4

hwlau

4:08 AM

@tpg2114 No for implicit either. The algorithm must conserved phase space volume like the Crank-Nicolson

tpg2114

Well, Crank-Nicholson is implicit... but I guess what you mean is there are more constraints on it than just implicit

hwlau

ya

tpg2114

For what it's worth, Crank-Nicholson can also be written as an implicit RK scheme

So they are equivalent, at least for 2nd order

hwlau

CN is basically the mix of half forward and half backward second order

tpg2114

Right. Maybe that means other IMEX schemes would work too

Which is something we do in chemically reacting navier-stokes. Dealing with strong source terms requires different schemes than spatial derivatives

Secret

4:14 AM

WYCON AYUIM
The important thing is, it is shot down, which is good because it helps me to achieve a deeper understanding of quantum mechanics.
Having a correct thinking in quantum mechanics is more impotent for me to understand its formalism and gain intuition when doing quantum problems

hwlau

I expect there may have similar problem when you try the simulation fluid dynamic system without dissipation and source term. In such case, the energy have to be conserved. I guess there will also similar problem

tpg2114

Quite likely so

user54412

Well, there's no (explicit) dissipation in my stuff. Doing finite-volume differencing on flux-conservative form conserves energy pretty well.

tpg2114

What kinds of source terms do you have?

user54412

With quantum stuff it's like there's extra degrees of freedom, and correspondingly more constraints. Evolving constrained quantities via evolution equations that don't explicitly know about those constraints is where trouble happens, I think. Cf. evolving magnetic fields without introducing monopoles.

user54412

4:23 AM

@tpg2114 Often none, though heating and/or cooling are somewhat common. Also energy and momentum can be added by geometric source terms due to the whole curved spacetime thing.

tpg2114

I've actually been getting numerical instability problems for low speed flows with buoyancy terms and temperature changes of like 1000K

user54412

Actually, maintaining the divergence-free nature of the magnetic field, given that we're evolving it with the other Maxwell equation, is the hardest thing in what I do.

tpg2114

That's been strong enough to greatly upset the stability of our schemes

@ChrisWhite The guy who was working on MHD had a huge problem with \nabla \cdot B = 0

Spent years writing a solver that did it properly

Secret

(cont.) Since I am bad at intuition, any guess I have that pops up in the learning process will be treated by default as wrong and then via doing exercise and discussions, will try to find where it is wrong and shoot it down.

Only when after many attempts it is still standing will I asked whether such guess is actually a valid way to think about a problem

You guys have intuition that is better than I do. While I can do the maths, I don't know how to derive these intuitions

user54412

Nov 5 '15 at 2:32, by Chris White

There is a constraint in maxwell's equations (call it $\nabla \cdot B = 0$, or $\nabla_\mu {}^*F^{0\mu} = 0$ or whatever). And naively evolving a system according to the evolution equations tends to lead to violations in this constraint that accumulate over time. This is separate from the numerical stability of the integrator per se, and there are many volumes of literature written on how to properly evolve the electromagnetic field.

user54412

4:29 AM

Nov 5 '15 at 2:33, by Chris White

Basically, treating the 4-divergence of an antisymmetric tensor as the flux-conservative evolution of 3 (ideal MHD, perfect conductor case) or 6 independent scalars is... morally wrong. We do it anyway, but we have to be extremely careful.

user54412

Nov 5 '15 at 2:34, by Chris White

I actually wonder if some mathematician has figured out the right way to evolve tensors numerically, yet no one in physics or engineering has read the poor person's neglected thesis.

user54412

@tpg2114 I believe it.

Secret

@Secret or more importantly don't have the knowledge to tell whether I have already made a mistake while to you guys you can see the mistake

tpg2114

@ChrisWhite It was something about running on a staggered grid (since we are normally cell-centered) and integrating B on the control surface around each cell

And then making sure that everything matched on matching cell faces and that the integral was zero inside the domain

Not easy, and not inexpensive

user54412

Aha! We use a staggered grid too. I have nightmares about the bookkeeping.

tpg2114

4:34 AM

Only B is staggered if I remember correctly. All of our other things are cell-centered

user54412

Same. Well, B is on cell faces, and E is on edges (so the \dot B ~ curl E equation discretizes naturally), but E is sort of a derived quantity anyway when the fluid is a perfect conductor.

tpg2114

Yeah, that's probably how we do it too. I am (thankfully) not involved there :)

That's also how most incompressible flow solvers work too. Stagger P and velocity on the grid so they can enforce \nabla \cdot u = 0

user54412

What's ridiculous is that the method just works in curved spacetime. Actually one of the early papers pioneering it was done for GR.

tpg2114

Do you guys know where to find info on requests for proposals for grants?

Like... I know where to find SBIR/STTRs

And NIH/NSF have their things going on

But other things... like a NASA grant, or DoD or... maybe it's all field specific?

user54412

No idea actually.

tpg2114

4:46 AM

I may have the possibility to start writing grant applications

Mahnax

Hey folks, I have a really basic question about 3D particle in a box if anyone's willing to point me in the right direction.

tpg2114

And I'd like to find things I would like to apply for... but I have no idea where to look

Mahnax

tpg2114

Maybe that's a good question for @dmckee -- any advice for where to start looking for grants? Particularly 6.1 grants? Are they department/agency specific, or is there a general clearinghouse for them?

Hrm, grants.gov is a thing

DoE runs all their stuff through there, maybe that's the clearinghouse...

Secret

@Mahnax First wrote the hamiltonian of the system. Then apply this hamitonian to the given wavefunction. If it is an eigenfunction, your result should end up with a constant times the original wavefunction. That constant is your eigenvalue

Mahnax

4:56 AM

@Secret I have something like $\frac{-\hbar^2}{2m}\nabla^2\psi(x,y,z)=E\psi(x,y,z)$

But if I plug in $\frac{1}{\sqrt{2}}(\psi_{1,1,5}+\psi_{3,3,3})$ I don't really get anything useful.

I just get a big mess of second partials.

Oh, maybe I need to use this? $\psi(x,y,z)=\sqrt{\frac{8}{abc}}sin{\frac{n_x \pi{}x}{a}}sin{\frac{n_y \pi{}y}{b}}sin{\frac{n_z \pi{}z}{c}}$

That would… make sense.

Secret

You know that the E in the 3D particle in a box (where all sides are L) is $\frac{\hbar^2n_x^2\pi^2}{2mL^2}+\frac{\hbar^2n_y^2\pi^2}{2mL^2}+\frac{\hbar^2n_z^2\pi^2}{2mL^2}$

Now since the partial derivative is linear wrt addition, you can distribute it to eahc of the $\psi_{x,y,z}$ in your wavefuction. Thne using the schrodinger equation that you wrote out, each partials will evalued into some energy with the formula shown

If your $\psi$ is an eigenfunction, you should be able to pull out this common constant, thus you then left with constant x your wavefunction

$$\frac{-\hbar^2}{2m}\nabla^2\psi=\frac{1}{\sqrt{2}}(\frac{-\hbar^2}{2m}\nabla^2‌\psi_{1,1,5}+\frac{-\hbar^2}{2m}\nabla^2\psi_{3,3,3})$$ then use the schrodinger equation with the E mentioned above to evaluate the partials

Mahnax

What do you mean "distribute" to each $\psi_{x,y,z}$?

Can I substitute my formula for $\psi_{x,y,z}$ into the question, plugging in the values of $n$, and take the partial derivatives of all that?

hwlau

5:16 AM

why are you making the problem so complicated? It is just a simple linear algebra.

You know

Mahnax

Well, I am making it complicated because I don't understand it and am just kind of flailing my arms around and hoping something will work.

If we're being honest.

hwlau

H psi115 = E115 * psi115
H psi333 = E333 * psi333
So will H (psi115 + psi333) = E * (psi115 + psi333) ?

No, except E = E115 = E333. Done

Mahnax

But how do we know that? I'm sorry, I really don't understand.

Secret

ok let me write this out in full

hwlau

H is ANY linear operator/matrix, and Schrodinger equation is linear. So the problem can be reduce to this simple linear algebra

Ya, it takes time to understand

Mahnax

5:27 AM

I appreciate you folks helping me with this.

dmckee

@tpg2114 I'm not a great person to ask that. All the channels I know much about are DoE and NSF programs for either established and on-going groups or for professors in a position to try to start one.

That said, NSF does have programs for people just starting out in "small" money. By which they seem to mean $15k for a project that might be one to a few years in length.

Secret

@Mahnax You start with $$\psi=\frac{1}{\sqrt{2}}(\psi_{1,1,5}+\psi_{3,3,3}) \tag{1}$$

Now you know that the (time independent) Schrodinger equation is

$$\frac{-\hbar^2}{2m}\nabla^2\psi(x,y,z)=E\psi(x,y,z)\tag{2}$$

You also know that for a cubic particle in a box, the E is given as

$$E=\frac{\hbar^2n_x^2\pi^2}{2mL^2}+\frac{\hbar^2n_y^2\pi^2}{2mL^2}+\frac{\hbar^2n_z^2\pi^2}{2mL^2}=\frac{\hbar^2\left(n_x^2+n_y^2+n_z^2\right)\pi^2}{2mL^2}$$

Now plug (1) into (2) to get

$$\frac{-\hbar^2}{2m}\nabla^2\psi=\frac{-\hbar^2}{2m}\nabla^2\frac{1}{\sqrt{2}}(\psi_{1,1,5}+\psi_{3,3,3})$$

dmckee

DoE basically only funds significant projects.

And there is very little federal money for teaching laboratories (which is the kind I'm in a position to start for the time being), but if you buy it for science you can use it for teaching after you've completed the terms of the initial grant (they even like that).

I have a vague memory that NASA does have some money for small projects, but most of it is tied to getting OKed for space and mass on a launch.

Mahnax

@Secret Wow, this is actually very clear. Thank you.

This is all so new to me, I have no idea how to really think about it. That kind of clarifies some things.

dmckee

And similar comments about astronomy: there is some money for little stuff, but it is tied to getting approved for time on "real" scopes.

I think there are mechanism for getting grants of grid time or access to supercomputers, and there might be some travel money with that kind of thing, but I don't know where to start.

Secret

5:35 AM

Well, I only made aware of this method on my 3rd year. In the past, I do the exact same thign you did, by using the explicit form of the wavefunction and then crunch through the algebra

I owe my gratitude to my semiclassical QM professor

dmckee

If your school has a grant office they probably get a lot of literature from funders, and that might be the place to start.

Mahnax

@Secret I appreciate you spending time to help me all the same. Thank you!

I didn't realize you could manipulate things like that.

Secret

@Mahnax The substitution technique we learnt back in high school is very useful. It helps to avoid a signficant amount of algebra where careless mistakes will arise

@dmckee any news about the 750 Gev particle?

yuggib

9:37 AM

@0celo7 because you want a probability preserving solution for the dynamics

Danu

11:40 AM

God damnit Djokovic :(

user116211

12:31 PM

India whitewashes Aussies to make history by breaking the long 140 years of barren record of no team whitewashing Aus in their den. Check out: cricbuzz.com/live-cricket-scores/14884/…

David Z

1:07 PM

@Secret other than the thousand or so papers about it on arXiv? :-P

user116211

1:24 PM

@David Z: What would you do for this review? This is the link of the answer; the answer got +13 but it is link-only answer : /

David Z

@user36790 Delete it (or mark for deletion), definitely.

The score isn't (or shouldn't be) relevant when deciding whether an answer should be deleted for this type of reason

user116211

@David Z: Now, if it is destined for being deleted, then why those upvotes? Definitely there maybe reason for it; don't you think?

user116211

@David Z: Marked it for deletion.

user116211

Good point, BTW.

David Z

@user36790 Sure, it means that (at least) 13 people thought the answer was interesting or useful. But it does not mean that 13 people thought it should not be deleted. I doubt that any of the upvoters even stopped to consider whether the answer should actually be there or not.

user116211

1:30 PM

@Danu: I knew it Murray couldn't make it :P

Danu

1:49 PM

@DavidZ That was my flag ;D

(I must admit hesitating for a while because it was Eric Zaslow)

David Z

Understandable, but nobody gets around the rules - at least, not just for being famous or having a stellar IRL reputation. It could be Ed Witten or Steven Hawking posting, we'd still delete their answer if it's nothing more than a link.

Danu

@DavidZ I'd like it if the answer was at least preserved as a comment, though.

David Z

Comments are transient though. They're not meant for preserving anything.

Danu

But honestly, there should be a place for links (as in link-only "comments") on this site.

If it's not as an answer and not as a comment, then where?

And let's face it: Comments are semi-permanent almost all cases.

user116211

@Danu: That's my point; you can't simply kill that innocent link ; /

David Z

2:02 PM

@Danu in principle, yeah, it would be nice to have a "related resources" section, where the contributions would be permanent

But in this case, I don't think that's the solution. The real way to deal with this problem is to edit the answer so that it's not a link-only post. That sort of information belongs in an answer; it's just that the link itself is not enough for an answer.

Danu

@DavidZ Yeah, but in this case the OP hasn't been on the site for a few years.

David Z

So?

Danu

So that's not likely to happen.

David Z

What's not likely to happen?

Danu

Unless someone happens to know about this journal and additionally feels bored enough to put in the effort to edit it.

Editing the answer.

David Z

2:04 PM

Yeah, and that's the system working as intended. If nobody cares enough to put in the effort to make an answer valid, it doesn't get to be an answer at all.

Danu

Then we should work on instating some kind of official method of recommending resources by only providing a link, if you insist the comment section is not the way to go (which I disagree with!).

We clashed on something similar before, in the context of some resource recommendation answer about the QM book by Griffiths, IIRC.

By the way, shouldn't we change all resource-recommendation question into CW's? @DavidZ

I'm willing to go through the list and flag all that aren't CW yet, if you like.

David Z

@Danu I think that is the current policy...?

Danu

Is there a mod tool to convert all question featuring [some tag] to CW?

David Z

Nope, not that I know of

Danu

So I'll get flagging, then.

David Z

2:10 PM

I'd suggest double checking that that actually is the current policy, first

Danu

I'm pretty sure it is.

David Z

I know we were talking about it, but I don't remember if we actually put it into effect.

I guess it doesn't hurt to flag, anyway

(personally, I still don't like making recommendations CW, but I was outvoted)

Danu

The (by far) most highly upvoted answer to the most recent meta was CW, I think.

Danu

2:26 PM

@DavidZ I think I've never flagged this many posts in one day ;)

Btw, does anyone have the link with "Community" deletion criteria handy?

229

A: Enable automatic deletion of old, unanswered zero-score questions after a year?

Just to formally document the exact policies we have in place to remove old abandoned / dead questions, the Community user will delete questions in the following circumstances: If the question is more than 30 days old, and ... has −1 or lower score has no answers is not locked ...or... it ...

0celo7

3:01 PM

Does anyone else here use Chrome to access the chat on an iPhone?

Recently, it started going to the top of the page when I refresh instead of staying at the bottom. It's quite annoying. Anyone else experiencing this?

user116211

@Danu: Same to me; hands paining after repeatedly recommend deletions....

0celo7

3:27 PM

@BernardMeurer can you believe that book is $300

Transcript for