Talk-Polywell.org

Diogenes wrote:Or the incident in which the Patriot batteries didn't fire to take out a scud missile in Iraq because someone had used a float when they should have used an int. It showed up as a failure after days of operation. The Scud hit a barracks and killed a lot of our people.


But I think Plasma physics simulation written for a super computer is an entirely different level of programing than that of a weapons system controller or that of a robot.

It's not (different). The fundamentals are the same. The usage of algorithms and understanding of various equations may be different. Most likely, the coders are your average contract coders. There is no special emphasis on getting top talent in these cases and even if you wanted top talent, how does a person who isn't a coder determine the level of talent. The market is flooded with adequate, yet not particularly great coders. Your banking software is written by guys that are just good enough to get the general work done, but have no real passion for the work or attention to detail. The same goes for your security software, drivers, etc. To clarify my point, there would be no difference between me coding enterprise business software and me writing simulation software. A coder is a coder, a good coder is a good coder, a great coder is a great coder. All can produce anything you ask, but the quality may vary.

Tom Ligon wrote:About this matter of physicists not being good at code and software geeks not being good at physics, why would anyone believe you can't have both on a team and have them learn something about each other's craft?

This is why Dr. Bussard hired Kitty King and later Loren Jameson to work with him prior to me coming aboard.
...

Truncated your post for readability. Tom, you're perfectly correct that a physicist could be a great coder or a great coder could learn the prerequisite physics. I don't think anyone was arguing that point. What we were arguing is that without enough information about who is doing the coding to make a truly informed comment or refute what is being said, we cannot be sure. Sorry, but this forum....or well Rossi has taught us not to believe claims with out substantiation. I think it is probably that they do have good coders, but there is a confusion about what constitutes a good coder or how they're hired. I'm willing to bet that most coders used in these scenarios are contract coders. In my experience, having interviewed several, they are generally adequate to get the job done, but not particularly efficient or great at what they do. This is because they approach the work as a job that they have no passion. I say this as a full time software architect focusing on enterprise business solutions for a major university. In my experience, making a claim along the lines of "we got the best coders" is often complete bs. I also posted above to clarify that it doesn't take a special coder to write simulator software, however; finding a talented one is most beneficial, albeit very difficult.

Most likely, the coders are your average contract coders.

In this case they are not "your average coders". And, they have a great understanding of the physics involved.
In my experience, having programmers who a.) care about the project and b.) understand the project's objectives and c.) grasp the project's concepts and theories makes for way better outcomes.
I am sorry I can not specify who is doing and what they have done at this point. However, rest assured that it will come out, and maybe sooner than you guys expect. Personally, I am looking for a nice peer-reviewed article in a respectable journal, and another patent. And who knows, maybe the code will be used by other projects working in a similar lane...

If all the optimistic comments are true then scalability is a non-issue. So then why was it said that scalability is "next"?

Maybe what was really meant is that next is that simulation parameters will be varied to test -machine- scalability as opposed to -algorithm- scalability?!? Two entirely different things.

The first you consider computational complexity time ala big-O notation. The second you just run a bunch of aims with different inputs.

But if you're big-O answer is O(NxN) or worse, it's back to the drawing board. If not, well then the algorithm is scalable.

The intent is to test machine scaling. And that is why it runs on thousands of cores on a supercomputer.

ladajo wrote: I am sorry I can not specify who is doing and what they have done at this point. However, rest assured that it will come out, and maybe sooner than you guys expect. Personally, I am looking for a nice peer-reviewed article in a respectable journal, and another patent. And who knows, maybe the code will be used by other projects working in a similar lane...

Will it be worthwhile then to get A. Carlson's attention again?

I think he would be interested. I am not sure he actually responded the last time we reached out.
I believe that some of his questions and concerns have been addressed, especially regarding diamagnetic behaviors at the boundaries and within the cusps.

If anyone remembers exactly what it was he deferred his involvement for, and if enough of it is addressed by then, I think it would be worthwhile having him at least take a look. I remember him arguing the way someone does when interested but too skeptical to stay involved.

ladajo wrote:I think he would be interested. I am not sure he actually responded the last time we reached out.
I believe that some of his questions and concerns have been addressed, especially regarding diamagnetic behaviors at the boundaries and within the cusps.

I contacted him about the Q&A with Park, but he never replied

I think we should try again. Be very interesting to hear his take. I suspect, however, that he will only comment on published work.
He may not have enough interest until the next publication.

I thought Carlson changed his focus entirely.

Back to the question of the potential well in the current simulation, how good is it?

As I know, he is still at Max Planck, but possibly supporting Biochemistry work.

The well depth is a function of confinement and electron sourcing as well as other loss factors. As in all science, we strive to identify and control significant factors. In the case of Polywell, confinement is the leading factor in this regard, with injection efficiemcy probably in second, and other losses in third place.
Be patient.

ladajo wrote:The intent is to test machine scaling. And that is why it runs on thousands of cores on a supercomputer.

Wait, so it _was_ run on a thousand cores, or it soon will be?

Sounds like maybe scalability has already been tested or at least started to?

Speculating here... Sounds like the problem has been divided up into finite volume units. Then the static fields are applied well... statically, and then the dynamic parts are applied in addition to that, as a set of functions that describe a statistical spatial distribution. So this is like particle in cell without the particles. Presumably with added constraints to maintain mass (and inertia?) conservation to high accuracy, thus limiting drift from rounding errors.

If the physics of the cells are experimentally confirmed - even in edge cases - this could enable truly massive scalability, bridging the extreme gap between atomic scale and -- well -- not atomic scale.

As I understand this sort of simulation, you may use a finite number of cells to simulate the volume, and probably in a limited part of the volume. Perhaps one quadrant is modeled and the rest are presumed symmetrical. Then you put discrete test charges in motion. So you model some number of discrete particles operating in the finite element volume.

Though maybe they have enough computing power now to really push for a particle by particle approach? There wouldn't be enough computing power on the planet to simulate every ion and electron, but maybe you can model enough of them to understand the bulk behavior.

But if I really knew I'd be working on it rather than guessing, and then I couldn't talk about it.

It has run on thousands of cores, and will continue to.
It is a full 3d model, not 2d or a quadrant.