The h Bar

2:31 AM

@KyleKanos Is there a standard way or program to give a triangulation of the dicretization of space? How can you specify the region with more points? I don't really understand this so I never try to use FVM

Kyle Kanos

Usually FVM uses rectangular grids

hwlau

What is the point of using FVM then?

Kyle Kanos

Ease of programming + Adaptive Mesh Refinement

hwlau

Do you mean that you can change the grid size during simulation?

Kyle Kanos

Yes

hwlau

2:34 AM

I have to use small grid size sometimes during runtime because the method is not accurate, but I have not use FVM

So, what is the difference?

I use two difference grid size and compare and determine whether I want to use a small one.

Kyle Kanos

AMR kicks in when the difference between adjacent cells is above a certain value

The grids around that jump are cut in half (in a new struct)

The new solutions with the smaller grids are then later merged with the larger grid

hwlau

Usually, I resize the lattice size by a factor of 2 for all lattice for easy implementation. So, you mean FVM can resize only a local region?

BTW, what is AMR?

Kyle Kanos

No, FVM doesn't do anything like that. It's the AMR that does it

AMR = Adaptive Mesh Refinement

It's a separate thing from the FVM

hwlau

So, can AMR adapt only a local region?

Kyle Kanos

Yes

The idea is that you don't want to run a 5000x5000 simulation because it will just waste computing time in the regions where nothing is really happening

hwlau

2:41 AM

I know the core idea of FVM, but I don't know how the AMR really work out

DO you have a reference or link to a real example that I can follow?

Kyle Kanos

So if you instead run a 50x50 base with say 7 levels of refinement, then you get 50^2 * 2^7 = 6,400 x 6400 resolution

hwlau

@KyleKanos Ya, change dx also require dt to change, so the total simulation time is even worse

Kyle Kanos

@hwlau Nope, it's better

The resolution is only high in regions of interest and very low everywhere else

Suppose your domain is from 0 to 10

hwlau

@KyleKanos Are you talking about the FVM?

Kyle Kanos

@hwlau No, the AMR makes the run time faster than without it at high resolution

hwlau

2:46 AM

@KyleKanos Is it still stable?

Kyle Kanos

Of course!

But any way...if your simulation is really strong in the range 0 to 2

Then your resolution would be very high in that range and at the coarsest level everywhere else

hwlau

@KyleKanos > DO you have a reference or link to a real example that I can follow?

Kyle Kanos

I'm trying to find a really good one

Here: homes.esat.kuleuven.be/~keppens/amrstructure.html

And some pictorials: homes.esat.kuleuven.be/~keppens/Examples.html

hwlau

Is the AMR criteria universal for different nonlinear differential equation?

And how can I choose the error estimator?

hwlau

3:13 AM

The 2d and 3d implementation is complicated

it is easy to messy thing up

BTW, there is no good answer on it on scicomp.SE

Awal Garg

7:52 AM

chat.stackexchange.com/rooms/13871/the-alchemy-lab

Waffle's Crazy Peanut

@KyleKanos: I guess you know about "fluids" very well...

(as @tpg isn't here) :D

Waffle's Crazy Peanut

8:04 AM

On the other day, I saw tpg's "nozzle" answer, where he mentioned something about supersonic flows

Got it...

physics.stackexchange.com/questions/107191/…

> When we look at subsonic flow, the gas speeds up as the area decreases and slows down as the area increases. This is the traditional Venturi effect. However, when the flow is supersonic, the opposite happens. The flow accelerates as the area increases and slows as it decreases.

^^^ This thing gets me...

I'm unable to comprehend that :/

Qmechanic

8:24 AM

To @Brandon Enright (and other editors): I would suggest to use the motion tag as little as possible, as explained in its tag wiki.

user54412

10:02 AM

@Waffle'sCrazyPeanut Maybe this will help: nakka-rocketry.net/th_nozz.html

user54412

well, it doesn't offer any good intuition, but it derives the appropriate equations (they're not too complicated)

Kyle Kanos

10:59 AM

@hwlau AFAIK, AMR should be useful for any partial differential equation. Usually one of the fluid variables (density or pressure really) are chosen as the error check with discrepancies larger than say 0.02 between the two cells flagging it for refinement.

I've never actually programmed an AMR engine in any dimension, just taken what other people have done. I have done a SMR (Statis Mesh Refinement) which is where only a certain region is refined and the solution is smoothed onto the coarse solution.

I've been thinking of throwing that one up on github, but I've been lazy

@hwlau Yes, no answers on SciComp because most of the answerers use the FEM method

jinawee

4:18 PM

Do you guys think that the question: "Can we measure the rotation of the Earth around the Sun from inside a laboratory?" would be on-topic?

user54412

5:16 PM

@jinawee why not?

Glen The Udderboat

5:48 PM

The Riemannium atom? Is it possibly real or does it not abide to certain conditions for existence?

user54412

6:21 PM

@dmckee If I see a user profile that says "Please delete me" should I inform the mods? (maybe the user forgot to notify someone through the proper channels)

dmckee

@jinawee Sure, but the answer is going to be massively complicated by competition from the rotation of the planet. Maybe a really big Foucault pendulum at the rotational pole, but protecting it against noise and backgrounds for months is not going to be easy.

@ChrisWhite According to physics.stackexchange.com/help/deleting-account they are suppose to also use the contact link to bring it to the attention of the team. We can't kill most accounts.

On the pendulum...would the equator be better. One involves subtracting of the daily rotation.

user54412

@dmckee ok, so it's a team thing, not a mod thing

user54412

it's weird, I found an account like this, but the user posted a question a few minutes ago

user54412

@GlenTheUdderboat you're going to have to define Riemannium - I've never heard of it

Kishor Bharti

Hey all

Anyone Interested in QCQI

Physics Meta

8:24 PM

0

Q: Why can we not downvote comments, when we can upvote them?

Simple as that. I think that since we can only upvote comments, it leads to skew, where comments, even if they are unhelpful or inaccurate cannot be downvoted. Given that the whole mechanics of this website is, more-or-less "people vote for things and the good things rise to the top, while the ...

David Z

8:42 PM

@ChrisWhite yeah, I think we know about one case. Anyway, it never hurts to bring it to moderators' attention, but it's not really something we can deal with.

Physics Meta

9:01 PM

1

Q: Should pure math questions be on topic?

We've previously had some discussion about the suitability of pure math questions for this site. Currently, pure math questions - by which I mean those that have no physics content, even if they arise in a physical context - are off topic. We migrate those to Mathematics. Here are some recent exa...

discussion featured scope

hwlau

9:15 PM

@KyleKanos I know now that why people usually use finite different because it can be easily written from scratch. In particular when you get new differential equations. The AMR is hard to implement and it costs months to make it correct.

Kyle Kanos

@hwlau FDM is trivial because it's simple differences of adjacent cells

hwlau

@KyleKanos Are you using some kind of public library or in-house one?

Kyle Kanos

FVM requires smoothing of the data (interpolation) for more accuracy

hwlau

@KyleKanos yup, but the PDE can't be very non-trivial. You will certainty want to gurantee it is correct before you do anything else

Kyle Kanos

I am using a modification of a public library (full disclosure: I didn't do a drop of work on the AMR engine, that was all done by other people)

The code bases the AMR on Chombo

hwlau

9:17 PM

@KyleKanos Is the smoothing gurantee some kind of conservation of mass/energy/probabilty?

Kyle Kanos

@hwlau No, it improves the accuracy of the computation.

hwlau

What is the point if the mass disappear after some time of simulation?

I just find that I confused with the FVM and FEM somehow

Kyle Kanos

Unless you have free boundaries, mass won't just disappear

hwlau

@KyleKanos You know the FDM of diffusion equation (or similar) has stability criteria, how can it be solved in AMR

In particular, can I use implicit method in AMR?

Kyle Kanos

I don't think it would change any

The AMR engine still detects if there is a discrepancy and adds new structs as necessary

Your code would call the solver regardless

hwlau

9:29 PM

The criteria is something like dx/dt^2, how can it adapt a different time step to satisfy the stability criteria

Kyle Kanos

The criterion is 0.5 <= k*dt/dx/dx

hwlau

sor, yes

but if you use small dx, would you change dt?

Kyle Kanos

It depends

Some AMR engines would change it and others would not

hwlau

So, do you mean that I only need to give the relation between different nearest neighbour gird point, and the AMR engines will calculate other things for me? Is it the reason you think that the method can be used for all partial differential equation?

Sklivvz

11:32 PM

@dimension10 I noticed you have a "fake" Sklivvz over at PO. Can you please delete the user and only leave the attribution on posts/comments (unless you want to remove them as well, but that's up to you).

@dilaton ^

Transcript for