Is fluid dynamics a true physics subject?

MagnetoBLI

By providing such an initial statement (albeit incorrect), a discussion was stimulated that actually uncovered a lot of interesting information. The emergent phenomenon is spot on there, startlingly interesting. The link is below for any curious eyes. Thanks for all your replies and I hope this helps others with a similar thought pattern.

http://www.nobelprize.org/nobel_priz...in-lecture.pdf

twofish-quant

Yes, but it involves a 18th century view of physics that's wrong.

In modern particle physics, the electron exists as a field and in some sense all electron behavior is (or could be) collective. One thing that happens in CFD is that instead of dealing with atoms, you start seeing collective "objects" like eddies which interact with each other. That's something like the way that electrons in a field interact with each other except that the electron field is mostly linear hence much easier to deal with than CFD.

You start see the connections a bit more with things like superfluids.

Also most engineering specifically avoids "weird stuff" for good reason. If you are building a bridge you want to do it with time tested principles and you want to avoid as much "research" as possible. If you are using an 18th century physics model instead of a 21-st century one, and it matters, then you really have to ask whether you want to build the bridge that way.

I don't see that sort of division. One thing that will also give people headaches are magnetic fields and GR, and you run into the same sorts of issues that you do with solving the NS equation.

One thing I love about the field is that you run into very practical problems. You have X CPU/GPU cycles and you have to make some decisions on what is the important physics that you want to simulate. You also get *very* *very* familiar with the computer architecture to squeeze out some more performance. For example, if you write a "for-loop" if you do it the wrong way, you are hosed because you want the CPU to pack those numbers in one bunch and execute them in one instruction. So you get familiar with things like the AVX instruction set, and what versions of what compilers will output that instruction

MagnetoBLI

Yes to be honest I have to agree, it does all seem great. The technical side complements the theory very nicely from the sound of it. Is it normal to focus on one particular area (application or theoretical issue) for a whole career or is it common to apply knowledge in many areas? It seems that fluid dynamics theory is very adaptable.

How soon will (likely to be) quantum computing play a significant role in fluid dynamics modelling? Is it likely to be a prerequisite to solving turbulence in your opinions?

twofish-quant

At the very, very best case, ten years. Also, if you want my opinion right now, I can't possibly see how quantum computing would *ever* be useful to CFD. If I'm wrong about this, ten years is the minimum time it would take for someone show that I'm wrong.

I think "solving" turbulence is the wrong way of describing what people are looking for. I wouldn't be surprised if it turns out that the NS-equations are "unsolvable" in the traditional sense.

Again going back to the limits of the clock work universe. Suppose I'm modelling a car with wind going over it. A "solution" of the problem would have me specify the exact initial conditions, run through the simulation, and then come up the exact trajectory of every atom. I strongly suspect that this sort of "solution" doesn't exist, because if I change the initial condition slightly, then the trajectories of the particles are going to be extremely different.

But I really don't want a "solution." What I want is to calculate drag coefficients, so I don't *need* an exact solution. I just need a model that replicates the physics that I'm interested in, that can be done with the computing power that I have. So I'm not interested in an "exact solution" but rather a "good enough approximation."

There's also an "artistic" element to this. When an artist paints with oil paints or water color, they are trying to get to the "artistic truth" of a situation. A painting of the sun isn't a real sun, but if you are a good artist, it's a good enough simulation. CFD works a lot of the same way. You use a palette of techniques to get to the "essential physical truth" of a situation.

Studiot

I take it my observations were not of interest|?

MagnetoBLI

Are there any areas in physics that are looking for exact solutions? It seems that for every application there is another fluid theory that applies to it, which intuitively seems like a job for an engineering-physicist rather than a pure physicist. What do you think?

I am unaware of the details of solar prominence, although they appear to be represent-able by a specific solution rather than a general underlying property.

Studiot

I have no idea how this relates to my comments.

I offered two areas of fundamental theory that have arisen recently from fluid dynamics.

We have barely scratched the surface of these phenomena to date. Much of our knowledge here is in the pure theory stage ('Faraday's what use is a newborn baby'), known applications are very limited.

I also offered two areas of applied theory that have not been solved to date, not to be sneered at but as examples of how little we know.

MagnetoBLI

But surely this means all solutions that are true for initial condition changes are over simplified and provide an approximate solution in reality? Is it that fluids are so complex that this over-simplification is more pronounced when initial conditions are altered?

twofish-quant

What happens with turbulence and chaotic systems in general is that if your initial conditions change a little bit, then your final result becomes random. One way of thinking about this is imagine a particle near a turbulent field. If you change the initial conditions slightly, that particle could end up *anywhere*. If you care where all of the particles end up, you are stuffed, because where an individual particle ends up is random.

What this says is that the "reductionist" approach wouldn't work, you have to try something else.

You usually don't care about tracking every particle. What you care about are "generalized quantities" (i.e. if I put this shape in this gas, what's the drag coefficient). There are a number of techniques for figuring that out.

It's not that fluids are particular complex, it's just undergraduate physics classes are deliberately oversimplified. The types of systems that you learn about in undergraduate physics are *deliberately* selected not to show complex behavior, and this is in part because engineers *deliberately* build systems that minimize weird behavior. So what happens is that physics undergraduates learn a "cookbook" of tools that work for certain rather simple systems. The mistake is thinking that you can use the cookbook for everything.

MagnetoBLI

Is anything truly random or is it just effectively random to today's computing power? But yes I understand what your saying about real systems being unique.

twofish-quant

The thing about similarity variables is that they let you show that one physical situation is basically the same as another, so you can run an experiment in one situation and scale it up to another.

For example, turbulence is controlled by the Reynolds number. So I can run water through a tube and get some numbers, and then apply them to some other physical situation. As long as the Reynolds number is the same, the behavior will be similar.

The problem is that there are some situations where you can't run an experiment. I can put a wing in a wind tunnel, but I can't drop gas down a black hole.

I don't think that the bulk properties are well understood. Rather they are well defined.

One thing about this is that most of this is experimental/observationally driven. You see something, and you try to model it. Our understanding of the equations aren't good enough so that you can;t take the equations and *predict* some complex behavior.

You talk water, you run it through a tube. At some critical number, it turns turbulent. Now suppose you are a space alien that doesn't know about water and tubes. Someone gives you the NS equations. Looking at those equations, with the math techniques that we have, that space alien wouldn't be able to figure out that AHHH, at X situation, I see turbulence.

This matters for things like black holes and neutron stars. With GR, magnetic fields, and all sorts of quark phenonmenon, I'm sure that "something weird" happens.

What usually happens in astrophysics is that people create particular approximations. For example, gravity works with GR, but GR is too hard to build bridges. However, by making some assumptions end up with Newtonian gravity which you can work with. Same thing happens in astrophysics. You find that with certain situations, magentic field lines get trapped with fluid, and you end up with the MHD approximation.

MagnetoBLI

Cheers for the replies, very useful. How does one go about choosing an area of fluid dynamics for a career? There seems to be select groups within uni departments that focus on specific areas (although maybe some overlap), presumably due to the time required to understand each area properly, and from reading over the next year (all I have until application time) I will barely scratch the surface.

reasonableman

This has been an interesting thread to read but I have to echo MagnetoBLI's view of the field. When you look at the literature it seems to, largely, be phenomenology (when you do this, this happens).

If a problem needs to be solved either a simple, soluble case is used, or it's modelled using CFD (yes, this can be an interesting challenge). Yes, it's led to new theories but there seems to be few fundamental ideas (NS equations describe everything, but are too complicated) that are used.

For example, take a simple problem of mixing milk in my tea. If I stir how long will it take to mix, if I stir twice how will that change the time? As far as I'm aware the answer to this question would be highly specific with no 'general principles'. Which is why it doesn't 'feel' like physics. This isn't necessarily a criticism, though.

twofish-quant

Usually someone specializes in a particular department that uses CFD in a certain way. There are some university centers for CFD and numerical programming.

One reason that it's subject focuses is that different departments look at different types of problems and those call for different techniques. For example aeronautical engineering typically looks at problems in which the fluid does do not weird things (i.e you have to worry about antimatter appearing near your airplane) but you have very complex shapes.

In astrophysics you *do* have to worry about antimatter and tau neutrinos popup out in your fluid, but you have simple boundary conditions and shapes (i.e. stars are more or less spheres). What happens is that his why AE using finite element methods whereas astrophysicists use finite element models.

Then there are the CFD simulations of the entire universe.

http://www.mpa-garching.mpg.de/gadget/

where your grids are 100s kiloparsecs and your timesteps are tens of thousands of years.

chiro

I don't want to hi-jack this thread, but I was wondering if anyone can give some resources for CFD, in particular in relation to your own specific sub-area considering your own issues if you have them handy or in memory.

Getting a small insight into these intracacies has been very enlightening.

MagnetoBLI

From the below two quotes it seems that fundamental ideas in the current age of physics are based upon processes rather than intrinsic properties and that new found particles are systems of the same substance such as emergence phenomena. Is this the same with other ares of physics?

Robert Laughlin's Nobel Prize lecture on solid state physics mentions that his students feel betrayed when he tells them that fits to experiment/deductive techniques are perfectly valid , as the students are trained to think in reductionist terms and think non-amenable things are unimportant. He explains that 'emergent phenomena' can be described as a particle that has it's own characteristics and properties - like an eddy, which I thought was interesting.

And

AlephZero

You can't minimize weird behaviour in a rational way until you can predict something about it. Otherwise you may be dancing blindfold on the edge of a cilff, and of course there are plenty of examples in engineering (some from thousands of years ago) where engineering cliff-edges were discovered by falling off them.

I agrre that predicting the "details" (whatever that means) of a turbulent flow is probably not useful, but predicting the onset of turbulence, or the boundaries of turbulent regions in a flow pattern, would be a useful start. In mechanical engineering people are using nonlinear dynamics models to track how solutions depend on system parameters, to predict unstable or chaotic behavour, locate multiple solutions to the response of a system for the "same" conditions, etc. It hasn't yet been reduced to a black box method that anybody can use wthout understanding it, but it works in real-world situations, and on models with 200,000 DOF not two. But AFAIK, CFD hasn't got to that point yet.

As a simple CFD example, think about flow through a uniform pipe. The laminar flow solution of the NS equations is well known, and mathematicallly that solution exists for all Reynolds numbers. But experiment says there is some critical value Re* such that if Re > Re* this flow pattern doesn't physically happen. It would be very useful to have a way to compute the value of Re* from a simulation of the flow. And it might even be even more useful in the long term if somebody comes up with insight into why doing that calculation based on a continuum model of the fluid (i.e. NS) is impossible!

Probably doing "only" that wouln't earn you a Millennium Prize, but you have to start somewhere...