?

Log in

No account? Create an account

Previous Entry | Next Entry

miscommunications, doom

Sometimes I feel like I'm a character in the movie "Office Space":


I copied a whole stack of those forms... after borrowing one from you
earlier.  I always attach one to the copy I give to e-res, but when I
originally asked you whether I should attach another one to the copy
for the physics department I thought you had said no.  I guess this was a
miscommunication.  (If I remember correctly, I was confused the first
time you said no so I asked again to make sure... but maybe you thought
I was asking something different.)  I will attach one to both copies
from now on.

Jeff

On Fri, 20 May 2005, Julie Reiner wrote:

> Jeff:
> Could I ask you to kindly write the name of the instructor, the # HW
> solutions (e.g., HW solutions #7),
> the name of the course (e.g., Physics 6C) and the quarter on all HW
> solutions? There are template
> pages which we give out to TAs that are already made up for this
> purpose--you can write the
> solutions on one of those pages, and there is a space at the top
> available for writing in this info.
> It makes it a lot easier for the Science Library scanners and us in
> the office who file the material,
> to identify course material when it comes in, and helps our student
> workers as well, who may not
> know otherwise how to identify it. Next time you're by, please ask me
> for one of these template
> pages for solutions. If I'm not here, anyone in the office can help you.
> Thanks,
> Julie
>




The clock is still ticking down to my doom at the end of the quarter... no way in hell I can do three more assignments in the next 2 and a half weeks (considering it took me the first 7 weeks to do the first 2 assignments). I think I might take an "incomplete" for this class.

I did however figure out that the QCD calculations we have to do aren't quite as bad as I thought... apparently I was making things a bit too complicated.

Comments

( 13 comments — Leave a comment )
lars_larsen
May. 21st, 2005 03:19 am (UTC)
Wait, you spend weeks doing calculations? Don't we have computers for that sort of thing?

We really should cut you out of the loop. You're slow.
spoonless
May. 21st, 2005 03:40 am (UTC)

Wait, you spend weeks doing calculations? Don't we have computers for that sort of thing?

These are not calculations like "add, subtract, multiply, divide" or even solving a system of differential equations, which a computer can do just fine. It's stuff like integrating over infinite-dimensional spaces where the basis vectors are functions themselves on other spaces... or integrating over fractional or even complex dimensions (for instance, "pi+i" dimensions). In QCD, there are at least 3 indicies (subscripts and superscripts) on every variable which need to be implicitly summed over... and most of the work is in keeping track of the various indicies and only writing down the minimal amount (since writing everything out in full would take a half a page or so for each "step"). There are a lot of tricks and identities that make all this a lot easier... once I get some practice with it it'll go a lot quicker. And yes, someone should be able to make a computer program do stuff like this eventually. But handling all the notation would be somewhat of a bitch. Another part of the problem is... going from one step to the next is rather trivial, it's just writing it all out that takes the time. So typing it all into the computer might take just as much time. I'd love to work on automating stuff like this... I think it would be a great idea. But so far I don't know that anyone knows of a nice clean way to have the computer do the work without requiring almost as much typing as you'd need to write it out yourself.
onhava
May. 21st, 2005 03:48 am (UTC)
The color indices usually straightforwardly just give you some Casimir, so they don't trouble you too much. With a little practice, it's pretty straightforward to reduce loop diagrams to a few simple factors multiplying some integral, for which you then have to introduce Feynman parameters and whatnot....

Of course, the hard part (at least for me) is always getting factors of i and -1 and 2 in the right places. That can be a mess.

People seem to have pretty much automated all sorts of tree-level computations, at least for low numbers of particles (for collider physics, there are various things like CompHEP and Madgraph that do this sort of thing). One loop computations are a different story. From some talks that I've heard, I get the impression that people who do NLO and NNLO calculations for a living do use some automated tools, but there's no general-purpose way to automate things. I think combinatorics tend to overwhelm you, so you have to be really clever to do any but the simplest calculations.

The twistor string people will tell you they're going to change all this, but I remain skeptical.
spoonless
May. 22nd, 2005 02:37 am (UTC)

People seem to have pretty much automated all sorts of tree-level computations, at least for low numbers of particles (for collider physics, there are various things like CompHEP and Madgraph that do this sort of thing). One loop computations are a different story. From some talks that I've heard, I get the impression that people who do NLO and NNLO calculations for a living do use some automated tools, but there's no general-purpose way to automate things. I think combinatorics tend to overwhelm you, so you have to be really clever to do any but the simplest calculations.

Interesting. I'll have to check out CompHEP and Madgraph. I'm doing everything to 1-loop, which I imagine wouldn't be too difficult to automate if they've already got tree-level working. But I guess somebody just has to sit down and do it... and anyone who's qualified to write such a pogram is too busy working on other stuff. I would be interested in getting involved in that type of thing since I have a strong background in computers, and somewhat enjoy programming. But then again, if I spent my time as a graduate student working on something like that, I can't imagine where I'd be able to get a HET postdoc. So I guess that leaves me back at square one :)

The twistor string people will tell you they're going to change all this, but I remain skeptical.

Hmmm, that makes me very curious about twistor string theory. Even though from what I gather you don't like it. I still don't have any clue what it is really.
lars_larsen
May. 21st, 2005 04:52 am (UTC)
Thats what I was wondering. Thanks for explaining it. Sounds hard!

Back in highschool I spent hours doing endless lists of algebra problems. Which was really quite pointless considering computers can easily do that for me. I'm a programmer, and programmers are lazy. There must be a better way!

Clearly this has to do with the size of the input vs the size of the output and the relationship between the size of the input and the time required to calculate the output. Which are classic computer science problems.

The inputs along the way are just found through deterministic rules you're following to get those numbers right? There is no creative input by the user along the way? You say that you write things down in each step. But the output of one step can be automagically typed in in the next step if a computer does it.

The real reason it hasn't been automated yet might be that so few people on earth actually do this kind of thing, there isn't a huge demand. Just like there aren't a large number of computer programs that design, for example, the hull of a sailboat... Since very few people design sailboats. There are however lots and lots of programs that do word processing, since almost everyone does that.

I still don't understand why its a difficult problem for a computer since I dont understand the math involved. Here is a site that talks about the complexity of algorithms:

http://mathworld.wolfram.com/topics/ComplexityofAlgorithms.html

You should be pretty damn pround if you are really doing calculations that are too difficult for a computer! Thats quite an accomplishment!

I also found this, which may or may not be relevant to your multidimensional problems:


http://www.idsia.ch/~marcus/official/projects.htm

"Optimization of functions on very large/high-dimensional spaces is crucial in a plethora of applications, including operations research and artificial intelligence, to name only two. The functions often involve all known difficulties at once: multimodality, deceptiveness, discontinuities, ... Usually one has to be satisfied with approximate solutions. For concrete, often NP-complete problems, polynomial time approximation algorithms (PTAS) can often be found..."


spoonless
May. 22nd, 2005 02:43 am (UTC)

The real reason it hasn't been automated yet might be that so few people on earth actually do this kind of thing, there isn't a huge demand.

Yes, I think that's the main reason. There are only a small number of people in the world who need it, and they're all too busy on the verge of proving really cool new things about physics, so none of them wants to take time out to write a general purpose software package for it. I'm sure it'll get done eventually, but that doesn't help me very much right now.

I don't think either of your links are very relevant. It's not really an issue of complexity, it's just an issue of trying to organize how things are done... we use so many "creative" tricks along the way that we've picked up over the years that it's hard to build them all into a deterministic program. Hard, but not impossible.
lars_larsen
May. 22nd, 2005 05:34 am (UTC)
Well, I think its an issue of how much you have to manipulate the process along the way. If there is "one true way" to calculate it, and all you're doing is going through the motions, it should be a lot faster and easier for software to do. And programming it shouldn't take much longer than doing it once. I understand what you're saying about the interface. The calculations were designed to be done with pen and paper, so it would be difficult to design a software interface.

If only I knew a 1 percent of the math you do, I might tinker with it.

Don't some physicists do simulations? Thats basically what we're talking about here right?

The people teaching you probably do have an automagic quantum solution machine. They just dont want you to know it exists or you will never learn how this all works. Just like my sadistic algebra teacher wouldn't tell me about mathematica.
spoonless
May. 22nd, 2005 06:48 am (UTC)

If there is "one true way" to calculate it, and all you're doing is going through the motions, it should be a lot faster and easier for software to do.

There's a lot of different ways to do most physics problems. The harder they get, the more ways there are to solve them. That's part of the problem... you can give ten different physicists the same problem, and if they know what they're doing they'll all get the same answer, but they'll do it 10 different ways. Often it's not obvious which way is going to be the fastest until the end. Sometimes you just have to try different things until something works. It helps to know a lot of tricks, but for every easy way there's a ton of hard ways that lead to confusion and frustration. This is what makes physics challenging!

Don't some physicists do simulations? Thats basically what we're talking about here right?

Yup... you're talking to one. That's what my research last summer was about. But that's NOT what we're talking about here. There are people who do QCD simulations, but the types of problems which can be solved by doing simulations are a different class of problems than those done by hand. The ones the computer can solve by simulations are called "non-perturbative" whereas almost all of the ones done by hand are "perturbative". Simulating QCD takes a mind-bogglingly huge amount of computing power and long amounts of time to run, and it only can be done for very small system sizes. That said, they do give us answers to stuff nobody could solve by hand. But the stuff we do solve by hand sometimes gives us answers that could not be realistically simulated in any reasonable amount of time with current supercomputing limits. You could teach a computer to do the calculations we're doing by hand (with enough effort), but that's different from simulating... it's not exact, and it works in a very different way than the physical process works... and it's usually aimed at only finding the answer to one specific question about something (such as the lifetime of a particle)... it doesn't tell you exactly what went on in the rest of the process.

If we ever get quantum computers working on a large scale, then all this will change. Processes which couldn't be simulated on a classical computer in any time less than the age of the universe might become tractable. But this is a long ways off. Decades, at the very least.
lars_larsen
May. 22nd, 2005 07:04 am (UTC)
Sometimes you just have to try different things until something works.

That translates to very slow computer programs. Now I get it.

A tool that you could use would require a complex interface allowing you to switch between different mathematical tools and visualize the results. Sort of a word processor for math. That sounds like a pain in the ass. No wonder nobody has ever done it.
spoonless
May. 22nd, 2005 08:14 am (UTC)

A tool that you could use would require a complex interface allowing you to switch between different mathematical tools and visualize the results. Sort of a word processor for math. That sounds like a pain in the ass. No wonder nobody has ever done it.

Exactly. There are several mathematical packages that do this for more general math & physics applications (Mathematica, Maple, Matlab, etc.) but when you get to a high enough level math, such as the type used in particle physics, it just doesn't have nearly enough of the features available.

Yet another way to say it is... computers are really good at crunching finite numbers, but people are better at working with infinite numbers. A lot of quantum field theory is about manipulating infinities and knowing which infinities can be ignored since they'll be cancelled or overshadowed by other larger infinities. Not a lot of software is out there to deal with such things since it doesn't directly fall under the category of "number crunching" (although it is systematic enough to be translated into a manipulation of finite quantities.)
detacheduser
May. 27th, 2005 05:53 am (UTC)
What type of math do you use at such a high level? Tensor analysis, exterior calculus, or something even more obscure?
spoonless
May. 27th, 2005 07:14 am (UTC)
Everything in Quantum Field Theory has multiple tensor indicies, so tensors are a part of it. No exterior calculus though, that's used in general relativity where spacetime is curved (although there are some people who do "QFT in curved spacetime" where it would be useful.)

The other math in QFT is Clifford Algebra, Lie Groups and Lie Algebras and their representations, Grassman Algebra, calculus of variations, functional integration... and of course other stuff like Hilbert spaces that carries over from quantum mechanics. Various other weird stuff like dimensional regularization (integrating in fractional or complex dimensions) which I'm not sure any mathematicians use. The hard part isn't doing any one of these, it's mixing them all together... you have to constantly be aware of which indicies refer to the Lie groups and which indicies refer to the Lorentz tensors and which refer to the Clifford Algebra. Gets confusing as hell sometimes!
spoonless
May. 27th, 2005 07:46 am (UTC)
Actually, I meant to say "Grassman numbers" rather than Grassman Algebra as that is apparently another name for an exterior algebra. I don't know if the two are related, but the things we do with the grassman numbers seem very different to me than what I know about exterior algebra (aside from them both being anticommuting). With the Grassman numbers, we define an anticommuting field that can be mapped to from ordinary spacetime. We integrate over them using certain rules, even though it isn't something very visualizable. Then we use functional integration to integrate over the infinite dimensional space of functions mapping 4D spacetime into the Grassman numbers which is another step more abstract.
( 13 comments — Leave a comment )

Profile

blueshirt
spoonless
domino plural

Latest Month

May 2017
S M T W T F S
 123456
78910111213
14151617181920
21222324252627
28293031   

Tags

Page Summary

Powered by LiveJournal.com
Designed by Lizzy Enger