The h Bar

0celo7

00:01

Another courser?

@Slereah

Wonder if ACM has the game yet

vzn

(spking of "nuking"...) wow, thought had seen it all... a physics urban legend...!

doomsday scam aka "red mercury" / NYT

Slereah

00:26

Welp

Courser wasn't so tough

0celo7

00:43

@Slereah about to enter the Institute. What should I expect?

Slereah

Dunno

Building the teleporter myself

0celo7

@Slereah who are you building it with?

I fucked over the Railroad and chose Brohood

HDE 226868

I came across the equation$$\partial f(x)=\text{co}\left\{ \lim_{i\to+\infty}\mathrm{d}f(x_i)|x_i\mapsto x,x_i\notin S\cup\Omega_f \right\}$$Does anyone have an idea what the $\text{co}$ means?

0celo7

GIMME THE MONEY

I HAVE ALL THE SPOILERS

Slereah

Railroad

Should have picked the Brotherhood

I have to build it out of tincans

0celo7

00:57

Lel

The brotherhood gives you the parts

BTW this plot makes less sense than Bioshock

spoilers lol

uh

I just killed the son

@Slereah help

Slereah

get a new wife, make a new son

0celo7

I think the Brohood wants me to kill everyone here

Are you in yet?

I don't know what to do

Dunno who I want to betray

I have no clue what's happening

@Slereah where are you?

yay I can now fast travel to the Institute

Bethesda, why are your games so good...

Slereah

01:20

I'm in the institute

They'd better not fuck it up because I have a power armor, a nuclear launcher and 30 nukes

0celo7

Woah what level are you

They're not gonna do anything.

In interested to see what happens

Do you always run around in power armor?

Slereah

No

I put it on for the institute

About level 50

0celo7

01:36

@Slereah you killing everyone inside?

Slereah

Nah

Exploring a bit right now

If I don't like what I see though

There's gonna be heck to pay!

dmckee

I don't no whether to be impressed or dismayed that anyone bothered to watch the video

0celo7

01:51

@Slereah let me know what you end up doing

did the Railroad give you some kind of mission to do while in the Institute?

I got two from the brohood

dunno if I should do them!

this is nuts

obe

what game are you talking about?

0celo7

@obe do you live under a rock

obe

idk, do i?

0celo7

02:07

maybe...

ACuriousMind

02:17

@dmckee I'm more dismayed that there are people who actually vote those "free energy" posts up.

dmckee

@ACuriousMind It's possible that someone is running some socks. Time will tell, but I won't.

DanielSank

@dmckee It's good to help people understand how these free energy demos fail.

It can be really hard for a beginner/untrained person to understand.

dmckee

@DanielSank Oh, no argument from me.

DanielSank

@dmckee In fact, I asked this one

84

A: Where is the flaw in this machine that decreases the entropy of a closed system?

Thus, the air molecules contribute a small portion of their kinetic energy to the paddle, which is then expended as heat on the other side of the border, making the air molecules on the left colder, while air molecules on the right heat up. Doesn't this mean a decrease in entropy? Yes it doe...

to a bunch of physicists and nobody got it.

dmckee

But I don't understand the way the younger generation looks to video for information and just watch every thing that comes by.

DanielSank

02:24

@dmckee Oh well yeah. I find videos really inefficient for learning anything that isn't hands-on, e.g. fixing a car.

dmckee

Yeah. Low density, imprecise and hard to search.

DanielSank

Videos for demonstrating programming etc. are completely ridiculous.

"Here's a tutorial video on how to install our software."

::closes window immediately and doesn't use software::

3

David Z

02:50

@DanielSank good to know I'm not the only one

tpg2114

03:06

I could use some numerical analysis help if anybody is willing to take a stab at something... I'd ask on SciComp but unless it is about Petsc it won't get answered.

user54412

@tpg2114 me me me! pick me!

tpg2114

@ChrisWhite Brilliant! I need to do some von Neumann (Fourier) stability analysis for a scheme and I'm having an impossible time figuring it out

user54412

no guarantees though ;)

tpg2114

I need to generate contour plots of the amplification factor, let me find you an example...

user54412

while you're doing that, @tpg2114 did you ever see this message?

user54412

03:09

Nov 3 at 5:34, by Chris White

For anyone other than @tpg2114 and me who's interested in colormaps: matplotlib is now including perceptually uniform sequential colormaps, and in fact one of them will become the default!

user54412

I'm curious to hear your opinion on the new suggestions

tpg2114

@ChrisWhite Nope, I missed that one... Weird. I guess I hadn't been in chat recently enough?

Do you still have journal access to things?

user54412

yep

tpg2114

Or do I need to find a non-paywalled source

arc.aiaa.org/doi/abs/10.2514/6.1997-2101

I need to generate plots like Figure 1 and 2

For a dual-time scheme like Eq (2) in same

2D Euler equations, nonconservative form is okay (so \Gamma \partial Q /\partial \tau + T \partial Q / \partial t + A \partial Q /\partial x + B \partial Q /\partial y = 0) where \Gamma is a matrix, T is a matrix, and A and B are the flux jacobians

Where I am horribly confused is that I have 2 "time steps" -- a physical time t and a pseudo-time \tau, and so I end up with 2 amplification factors

And no idea how to deal with that.

user54412

I'm trying to wrap my head around this

user54412

03:13

\tau somehow accounts for substeps?

tpg2114

Yeah... okay, so a dual time framework adds a new, pseudo-time derivative to the Euler equations

And the physical time term becomes a source term as you iterate in pseudo time

As pseudo time -> infinity, \partial Q /\partial \tau -> 0

And you are left with the original equation

Jameson came up with it originally in like... 1981 or something. I'm using it to turn an explicit scheme into an implicit scheme. The physical time derivative is discretized with an implicit formulation, like a backward Euler, and you can choose a physical time step to be as large as you want (since it is implicit). And then you march in pseudo-time with an explicit scheme like RK4 and when your pseudo time derivative becomes close to zero, you are at your chosen physical time

Eq (7) is the discrete scheme you end up with -- the LHS is the update to the pseudo-time level and the RHS is the physical time derivative (using a 2nd order backward difference here) which acts as a source term, and the spatial fluxes R

user54412

As this is my first exposure to this particular scheme, I can't decide if it's madness or genius

tpg2114

Heh, I think it's brilliant

If you have an explicit code -- well validated, efficient, correct -- you can turn it into an arbitrary order implicit scheme by adding a source term

And then if it's low Mach number, you can use preconditioning as well to eliminate the stiffness, but you don't need to do that if it isn't stiff

user54412

::wonders if this would be so trivial in GR::

tpg2114

Although I may be biased in my view of it since it is the bulk of my thesis

user54412

03:22

"Extension of the method to 3-D is straightforward."

tpg2114

It actually is, don't worry :)

This paper: arc.aiaa.org/doi/abs/10.2514/3.12946 has everything you would need to do it

It has the 3D form written out

user54412

okay, so a lot of this paper is about preconditioning, but I'm thinking I can ignore that?

tpg2114

Yup

Just say \Gamma is a matrix and you don't really care what it looks like

Basically I need to figure out the amplification factor for Eq (7)

Assuming \partial A /\partial x = 0 and \partial B /\partial y = 0 because we're pretending the equations are linearized

And for fun, R can be central differences in X and Y (explicit, so based on the values of Q^k)

user54412

so there's some warning about the approximations needed for viscous terms to work

user54412

but you're not asking about that, right?

user54412

03:26

just assume everything is well diagonalized and hyperbolic and happy

tpg2114

Correct, Euler only

The problem I have is that I get an amplification factor for pseudo time from Q^{k+1} = G^k Q^k

And a second one from Q^{n} = G^n Q^{n-1}

which is the physical time amplification

user54412

so what to do with them?

tpg2114

And so I have no idea how to plot contours of |G^k| without knowing anything about G^n

Unless I just assume values for G^n, like -1 <= G^n <= 1 (because I am assuming my physical time scheme is stable)

But then they only show one set of contours as a function of Mach, CFL, and x- and y-wavenumbers

Not as a function of G^n

So maybe there's some assumption about it or relation or something that could be made that I'm missing

user54412

btw you keep using G

user54412

in the paper, G is just the z-fluxes?

tpg2114

03:35

G being the amplification factor matrix (not in the paper)

Sorry for the confusing notation

G is the amplification matrix arising from the von Neumann stability analysis -- Q^{k+1} = G^k Q^k

And the scheme is stable in pseudo time if | G^k | <= 1

Here, this one is easier to follow cause it uses the same notation I do:

arc.aiaa.org/doi/abs/10.2514/6.1995-78

Section D

The lead author on this paper is funding my thesis and when I asked him how to do it, he laughed and said "It's really easy, go look at (the paper I just sent you) and you can figure it out from there"

And... I can't figure it out

user54412

I've had people say something is "easy" and then they send me the actual code that does it and I still can't make heads or tails of what they're doing

tpg2114

Yeah, I'm afraid to ask again cause he's a really big name, and working to get funding for me for next year, and on my thesis committee...

But I've asked all 4 post docs in our lab and all the students in the lab and none of them even remember how to do von Neumann analysis so it was kind of a non-starter

(Side note -- arc.aiaa.org/doi/abs/10.2514/6.2001-2591 uses a dual-time scheme for the MHD equations; I bet you could do it for GR and with a proper preconditioning matrix, make everything hyperbolic and well conditioned)

user54412

GR is actually always hyperbolic and, as far as I've encountered, always well conditioned. And people think GR is hard ;)

tpg2114

Really? No low Mach number flows in GR?

I guess not... speed of sound is probably nowhere near speed of light in anything ever

MHD is not well conditioned, just because light is so much faster than sound

user54412

I think the important thing is that all speeds are bounded by c

user54412

03:46

I think of it as not that nothing is slow, but rather that nothing is too fast

tpg2114

But what are the eigenvalues of the characteristic system? You don't end up with the traditional Euler {u, u, u, u+c, u-c} (c being speed of sound there)?

user54412

adsabs.harvard.edu/abs/1996MNRAS.278..586F Appendix A covers SR hydro

tpg2114

Yeah -- so it's basically the same as the Euler system (plus some funny terms for the sound waves). So if you have a low "Mach" number (whatever that means exactly) then it will be ill-conditioned

user54412

adsabs.harvard.edu/abs/2010ApJS..188....1A Section 3 covers SR MHD

user54412

If anyone has written the equations in full GR, I don't want to see them

user54412

03:54

@tpg2114 yeah, but in practice this rarely happens

user54412

or maybe it's just ignored :p

tpg2114

That's what I figured... Although if you are doing radiation or MHD, it will show up because the speed of the Alfven waves is much bigger than the speed of sound or the particle velocity

user54412

well, I guess often our fluids are often radiation-dominated -- the "sound" waves have radiation pressure as a significant part of their restoring force

user54412

in the limit p_rad/p_gas -> \infty, iirc the hydro sound speed goes to c/\sqrt{3} or something

tpg2114

Yeah... I'm totally happy to be running problems at 5 m/s composed entirely of air

user54412

04:00

what's really awful is when the plasma is so hot you get pair production -- no more rest mass conservation

tpg2114

I'm running a jet in crossflow right now to compare the preconditioned scheme with the explicit scheme -- the solutions are night and day different

I'm actually pretty shocked at how different they are

Oh... more random questions for you @ChrisWhite... have you done any cosmological simulations? Or know anybody who has? I'm interested in doing some topological analysis using something like Reeb Graphs or Morse-Smale complexes, but I can't really figure it out

I found this library: www2.iap.fr/users/sousbie/web/html/indexd41d.html? and I'll play with it at some point, but I like to have a better understanding of some nuances to the algorithms before I just use them blindly. I've written a little code that can compute the complexes for really simple functions but then it doesn't work on real problems

user54412

I know of a lot of people doing cosmological simulations, but I thankfully haven't

user54412

actually they're quite straightforward except for the fact that they're essentially Newtonian with self-gravity

user54412

and of course Newtonian gravity is Poisson's equation, so it's all about elliptic solves and all-to-all MPI communication

tpg2114

Yeah, that's why I want to use preconditioning instead of incompressible... those damn global communications are brutal

user54412

04:11

(there's also lots of "subgrid modeling" that needs to be done better, like accounting for how a newly-formed galaxy will have a burst of star formation and supernovae that put out energy and momentum, but all of this on much smaller scales than a single cell, but this doesn't affect the post-processing topology stuff)

user54412

@tpg2114 the vast, vast majority of the matter in the universe is a highly compressible plasma :p

tpg2114

I also dislike subgrid modeling, but since I work in large-eddy simulation, I didn't just say that

user54412

There's an emeritus at my department (and has been emeritus for a long time if you catch my drift) -- Rich Gott -- who's done a lot with cosmology topology

user54412

that said, I don't think there's any sort of final word on it

tpg2114

I just want a non-peer-reviewed, non-jargon explanation of how to do it... hah

user54412

04:17

btw I'm no less confused than you about these amplification factors

user54412

all these plots show the explicit scheme and the explicit+pseudo time scheme, right?

user54412

or did I miss something already

user54412

actually, I need terminology help -- what does your field mean when you say CFL = 10?

tpg2114

CFL = \lambda dt /dx

user54412

ok, so CFL=1 is marginal stability for explicit hyperbolic equations

tpg2114

04:21

And no, you didn't miss anything -- the plots are the non-dual-time scheme and the dual-time scheme. I think both of those papers use an implicit scheme for pseudo time and an implicit scheme for physical time as well

Correct. So in my code, my pseudo-time is explicit and physical time is implicit. So I run my pseudo-time at CFL_pseudo = 0.5 and CFL_physical = whatever I want (usually 100-1000)

If my pseudo time were implicit (like these papers), I could run at whatever is stable for it (so CFL_pseudo = 1000 or whatever)

user54412

ok, so CFL_\tau is just set to be 3 usually in the paper

user54412

and the timesteps are chosen either to make the particle/contact/entropy wave have a certain CFL number (CFL_u) or to do the same for the sound wave (CFL_{u+c})

user54412

so if CFL_{u+c}=1, CFL_u will be really small and you'll wast cycles in some sense

user54412

and if CFL_u=1, CFL_{u+c} will be large and you'll have large amplification?

tpg2114

Basically, yes

Although with preconditioning you get good amplification at CFL_u = 1 (that's the benefit)

user54412

04:30

right

user54412

so Eqn 6 in that last paper is the amplification for the whole scheme?

user54412

it's certainly aware of both timesteps

tpg2114

It's for a 1D scheme, but basically

What drives me crazy about that is he doesn't explain how he gets it -- if you take Eq (3)

And make it 1D

You get \Gamma_p \partial Q_p/\partial \tau + \Gamma_e \partial Q_p /\partial t + A \partial Q_p /\partial x

Where \Gamma_e = \partial Q/\partial Q_p converts conservative to primitive

And then, if you discretize the pseudo time derivative with a backward euler, the physical time with another backward euler, and the spatial with a central scheme, you get:

\Gamma_p (Q^{k+1} - Q^k) + \Gamma_e \Delta \tau / \Delta t (Q^{n+1} - Q^n) + A \Delta \tau/(2\Delta x) {Q^{k+1}_{i+1} - Q^{k+1}_{i-1}) = 0

And you assume that you are marching to Q^{n+1} and Q^{k+1} is an approximation to it, so you replace Q^{n+1} by Q^{k+1}

To give you:

\Gamma_p (Q^{k+1} - Q^k) + \Gamma_e \Delta \tau / \Delta t (Q^{k+1} - Q^n) + A \Delta \tau/(2\Delta x) {Q^{k+1}_{i+1} - Q^{k+1}_{i-1}) = 0

And then you do the von Neumann analysis on that and hopefully end up with Eq 6

Which I can do, but only if I make a weird assumption to deal with Q^n that I don't understand...

So my total VN analysis is:

\Gamma_p (G Q^k - Q^k) + \Gamma_e \Delta \tau / \Delta t (G Q^k - Q^n) + A \Delta \tau/(2\Delta x) {G Q^k exp{i w\Delta x} - G Q^k exp{-iw \Delta x}) = 0

And if these were scalars, you would divide by Q^k, since they are vectors/matricies I guess you right-multiply by inv(Q^k)

But either way, I get:

[\Gamma_p + \Gamma_e \Delta \tau / \Delta t + A \Delta \tau/\Delta x i sin(w \Delta x)] G - \Gamma_p - \Gamma_e \Delta \tau/\Delta t Q^n inv(Q^k)= 0

Which gives Eq 6 if somehow \Gamma_e \Delta \tau/\Delta t Q^n inv(Q^k) = 0, but no idea how the heck that happens

I know to start things up, at the very first pseudo iteration, Q^k = Q^n, but then that term would be = \Gamma_e \Delta \tau / \Delta t and not 0 so I don't know how that works

user54412

04:45

2 hours ago, by tpg2114

As pseudo time -> infinity, \partial Q /\partial \tau -> 0

tpg2114

Yeah, but G is the amplification factor in pseudo time

So I don't know how that related to Q^n

user54412

I was going to say this might help but on second thought I don't know

tpg2114

So my actual problem has both a Q^n and Q^{n-1}

And since I can't even figure this one out, I'm super lost on my real problem

user54412

So \Gamma_e is some order-unity thing

tpg2114

\Gamma_e converts conservative to primitive variables

user54412

04:48

yeah

tpg2114

It doesn't really have an order to it

user54412

Can I not just say \Delta\tau << \Delta t

tpg2114

You could, but then why retain it in G?

If you do that, it would go away everywhere

Not just in that Q^n term

user54412

good point

tpg2114

And it's not required that \Delta \tau << \Delta t

Although it often is in practice (in fact, you can march to the time invariant problems by taking \Delta t >> 1. I'm running simulations to steady state with \Delta t = 1e20 right now)

But for time accurate simulations, those terms stick around

user54412

04:53

have you tried petsc? :p

tpg2114

Haha, good one

I'm supposed to have a meeting with the sponsor on Tuesday. I may just have to walk through the derivation to this point and ask WTF he did to that term

It's a bummer nobody else seems to know how to do it either, but it's a little bit reassuring that I'm not like... stupid or something

4 post docs and a number of PhD students couldn't answer it

How's your real work coming?

user54412

I feel there's something fundamental I'm not understanding with the two times floating around -- something about simultaneously discretizing them seems... off?

0celo7

If I have something like this i.gyazo.com/d832feb08f8dc5262d96d6fff3237f32.png, is the only torque about G due to T? It's not slipping.

user54412

Anyway, my work is going okay. I finally got the method paper out for our code, so now I can focus on doing science.

0celo7

05:08

Part of me feel like friction has to come in...

user54412

And my thesis committee has made it clear that I had better start doing science, ~~now~~ yesterday.

0celo7

But I have no clue how to get friction

user54412

rolling w/o slipping -> static friction

tpg2114

@ChrisWhite This analysis is holding up my methodology paper...

0celo7

@ChrisWhite exactly

so I have way too many unknowns...

user54412

05:09

static friction is whatever it has to be -- you solve for it

0celo7

I don't know the acceleration

user54412

@tpg2114 your field is clearly more numerically advanced than mine

user54412

there's a delay of 20-30 years between when methods are invented and when they are brought into GR MHD

tpg2114

You mean "Everybody else but you in your field..." since I can't figure this out :)

0celo7

@ChrisWhite Do I consider $T$ and $f$ for Neton's law on the CM?

that might work

tpg2114

05:12

Well, this dual time thing is 30 years old, preconditioning is 25 years old and I'm the first one to apply a scheme from the 60's in the dual-time framework, so I can't say I'm super modern

user54412

@0celo7 maybe -- I hate these sorts of problems

0celo7

this might work

I get $T-f=ma$ and $rT-Rf=I\alpha$

o.o how are $a$ and $\alpha$ related when there are two radii

probably use the biggger one

yolo

haha it's rolling backwards

ha

crap

tpg2114

@0celo7 Assume a sign is wrong and flip it at the end ;)

0celo7

ha!

the denominator is negative too

but wow 23 rads per sec^2 seems like a lot...

user54412

numbers...

tpg2114

05:20

I know right? Symbols or floating point values on a grid.

Anything else, not worth it

@ChrisWhite You use Visit for your visualization, right? Or am I mis-remembering...

0celo7

ok

assuming all of that is correct

now I have to find the min coeff of static friction

that's when $f=\mu N$, right?

user54412

@tpg2114 yes, especially for on-the-fly stuff, 3D, and complex grids

user54412

I never quite liked the final output, though, so pretty pictures I try to do in matplotlib

tpg2114

Do you have licenses for FieldView? Apparently they partnered with Visit now and can do in-situ visualization

We abandoned FV years ago, I use Paraview now and I have it hooked up to our code for in-situ analysis

0celo7

....

user54412

05:22

FieldView? have not used it

0celo7

negative coefficient of friction

that's awkward

user54412

yay perpetual motion!

0celo7

ah, it would help if I used linear acceleration in Newton's law

yay .24

that's not crazy

I think I'm correct

tpg2114

@ChrisWhite Big presentation file: mcs.anl.gov/~hereld/doecgf2014/slides/… and a lecture on using it in-situ: nas.nasa.gov/publications/ams/2015/01-16-15.html

user54412

you know, I'm chatting on my windows machine, but I'm turning around to my mac in order to load ppt

tpg2114

05:25

Really?

0celo7

good man.

tpg2114

I have a 181MB PPT about Paraview and in-situ processing

user54412

wow this presentation reminds me how much industry cares about CFD

tpg2114

Holy crap that's big

user54412

(it doesn't care about astro, or fusion plasmas, and that's all I see these days)

tpg2114

05:28

Yeah -- FieldView is having a webinar next week I think about visualization for the Formula1 teams they work with

Good visualization is important everywhere though, not just industry

It just isn't given proper consideration in my field

It's all 2D slices colored in rainbow

user54412

^ all too true

tpg2114

People should lose their funding if they write a report or paper that uses rainbow or spectrum

It should be as bad as falsifying data.

Do you guys do anything with tensor invariants? Like for turbulence there are P, Q and R which are the 3 invariants of the velocity gradient tensor

P is the dilation, Q is the relationship between strain rate and rotation rate, and R... nobody has a good explanation of what R is really

user54412

some people might

tpg2114

I was trying to think of what the invariants of the vorticity gradient tensor would be... but I can't wrap my head around it

And nobody has published anything about it

user54412

but mostly I see vorticity, helicity, curl, divergence -- all much nicer to visualize :)

tpg2114

05:34

We use Q to visualize coherent structures in a flow

user54412

lemme guess: butterfly?

tpg2114

The top is Q -- 1/2 (|\Omega|^2 - |S|^2)

The lower left is vorticity, lower right is species mass fraction of the jet

Butterfly?

user54412

looked like a Rorschach test :p

tpg2114

Ah, haha... Jet into a crossflow

From the side (top), from behind the jet (lower left) and isometric of the front of the jet

The top and bottom left are two different ways to visualize strong vortices. It's really cool, it creates those upright tornado vortices

They drop from the jet and make contact with the bottom wall

user54412

Oh random thing I remembered to tell you: If you want cosmological simulation output, some big names are Millenium (though I don't know how much of the data is public) and Illustris (which has all the data available for download).

tpg2114

05:44

Sweet, more data to play with

user54412

the latter one uses an... interesting grid

user54412

moving mesh, with voronoi tesselation redone every timestep or something

tpg2114

We have some demographic data based on zip code for a non-profit I am part of and so I was learning fun data analysis stuff in Python. I have a script that downloads all the zip codes in a state, all the highways in the country, and then plots them on a map and I'm working on contour/heatmaps for the zip codes and stuff

I could get it to download counties, congressional districts, area codes... all kinds of stuff

user54412

data visualization can be kind of fun

tpg2114

Python and matplotlib is magic

Yeah, I really like data visualization and trying to come up with good ways to analyze high-dimensional datasets

The in-situ stuff I'm doing with Paraview is to try and identify structures of interest during the simulation and them trigger file and image output when things start to appear or disappear

user54412

05:48

That's something I've never played around with, but I can see it becoming more necessary.

tpg2114

That's why I've been digging into the invariants

user54412

You don't want to be post-processing yottabytes of data

tpg2114

Exactly -- that 3 view of the jet was done during the simulation. It was outputting those frames every 1000 time steps

It can also be setup so I can connect to the simulation while it's running and look at the solution interactively, that's pretty cool too

user54412

ooo fancy

tpg2114

It's actually really easy to set it up in a code if you are interested :) I'd be happy to help if you go that route

If you know how to use VTK, it's not bad. If your code uses VTK data structures already, it's super simple

user54412

05:51

I don't suppose it's on Mira? We just got a chunk of time on that.

user54412

VTK was our default output for many years (though now we favor HDF5)

tpg2114

Doesn't look like it's installed on Mira, but you can build paraview from source (although it can be annoying sometimes)

It is on Tukey though

Ohh, and the paraview mailing list claims it is on Mira: public.kitware.com/pipermail/paraview/2014-February/030600.html

There is a presentation that says:

> /soft/visualization/paraview/v4.1.0/catalyst/<edition>
Where <edition> is:
base[+essentials][+extras][+python]

Catalyst is the co-processing library

on Mira

I don't have access to that machine so I can't poke around

user54412

I won't have access for a couple weeks, but I'll keep it in mind

tpg2114

If you do want to try the in-situ thing, let me know and I can walk you through the steps of getting your code ready.

Is it in C/C++ or Fortran?

user54412

C++

tpg2114

05:59

Psh, that will be simple then

user54412

^ words not often said of C++

tpg2114

Ours is Fortran and never used VTK... so I had to take the C++ VTK stuff, extern it to C, wrap the extern C routines into C routines that Fortran could call, then write the Fortran module to call them from our code

And then I had to dig into the guts of VTK to figure out how to not duplicate the data in the VTK data structures from our arrays and to instead store a pointer to our memory. So I have to get a pointer to Fortran memory to go from Fortran -> C -> C-wrapped C++ -> C++

And the pipeline used during the simulation is a python script, so it's really a big jumble of languages working together

user54412

Hmm, I have a meeting in too few hours. I should head off.

user54412

Good seeing you again @tpg2114 And I'll think more about this pseudo-time thing.

tpg2114

@ChrisWhite Thanks for you help, if you have an epiphany, ping me :)

Physics Meta

07:10

1

Q: An incomprehensible warning

I was about to ask a question, when this warning appeared: Wait! Some of your past questions have not been well-received, and you're in danger of being blocked from asking any more. For help formulating a clear, useful question, see: How do I ask a good question? Also, edit your...

Balarka Sen

09:42

What is the physical "meaning" of the Lagrangian? I would think $T + V$ instead of $T - V$ would be more "geometrically meaningful" in the sense that $T + V$ of a particle moving through a specific trajectory represents the total energy of the particle on that path. The intuition I have for $T - V$ is that it's the total energy the particle "uses" during moving through that path. Is that the correct interpretation? If not, what is the correct "meaning"?

That said, the action $S = \int L dt$ should represent the total energy use during traversing the path, right?

yuggib

I am not sure that the Lagrangian has a strict physical meaning. Physically, it is rather "intuitive" (but that could be disputable) to use position and velocities as the data of motion, since the equations of motion are of second order in time. And the lagrangian is the "generator" of dynamics (via E-L equations) with position and velocities as variables.

Balarka Sen

I mean, the whole point of Lagrangian formalism is to minimize $S$, right? Why try to do that in the first place if we don't know what $S$ means? I am completely befuddled.

yuggib

minimize $S$ ;-)

Balarka Sen

Whoops, typo.

yuggib

historically, for sure it went the other way around

first there were Newton equations

Balarka Sen

09:52

In my interpretation, it means particles traverse in path in which the total energy use is minimized. That I can believe.

yuggib

I may say that work is much more closely related to the concept of "energy used"

at least for my personal intuition

Balarka Sen

Yeah, that's precisely what I think it means.

yuggib

but the work and the lagrangian are not the same thing :-P

Balarka Sen

oh, I misread.

what do you mean by work?

yuggib

Work (physics)

In physics, a force is said to do work if, when acting on a body, there is a displacement of the point of application in the direction of the force. For example, when a ball is held above the ground and then dropped, the work done on the ball as it falls is equal to the weight of the ball (a force) multiplied by the distance to the ground (a displacement). The term work was introduced in 1826 by the French mathematician Gaspard-Gustave Coriolis as "weight lifted through a height", which is based on the use of early steam engines to lift buckets of water out of flooded ore mines. The SI unit of...

of course only the displacement in the direction of the force counts (so for example the magnetic field does no work)

Transcript for