sci.op-research Google Group

Re: numerical optimization and data-fitting and over fitting

2008-10-13T09:56:44Z

In article <336d0453-5d71-4db1-b63e-ba7a8 0455...@a3g2000prm.googlegroup s.com>,
old stuff snipped
as Sebastian already told you: it is a pitty, but you can check
this _not_ beforehand.
the reason:
if your problem is convex _and feasible_ (!) , then the Karush-Kuhn-
Tucker conditions are a necessary and sufficient characterization of the

Penalty method

2008-10-13T08:16:18Z

Hi Everybody,
I have a question about penalty function method. In the convergence
theorem of penalty function, it says, any limit point of a sequence
generated by the penalty method is also an optimal solution of
original nonlinear programming problem.
Here, I want to know that is this theorem true vice versa for

Re: Out-of-print math books: An Update

2008-10-13T08:06:56Z

Glad to know and thank you for your feedback.
Yes it was. In the meantime I reworked the site:
[link]
If you find out any other glitches, please let me know.
-- Klaus Schmid

Reference for results in polyhedral combinatorics

2008-10-12T12:45:09Z

Dear experts,
I am currently working on a class of combinatorial problems related to
connected subgraphs where I had some success in proving polyhedral
results and obtaining an efficient algorithm. However, despite
extensive literature search I have found only a very few scattered
results related to the problem. Therefore I have the feeling that my

Re: numerical optimization and data-fitting and over fitting

2008-10-11T16:47:28Z

Hi,
To the problem, you can easily obtain one optimal solution using
numerical solvers.
To check whether this solution is unique, you can check the second
order sufficiency conditions for strict optimality based on the
Lagrangian at the solution. A good explanation with examples can be
found in Bertsekas "Nonlinear Programming", section 3.2. It also

Re: numerical optimization and data-fitting and over fitting

2008-10-10T22:44:18Z

Exactly, in our original problem,
we have a LS problem
min || Ax-b||^2
s.t. Cx-d=0
and x>=0.
where A is n x n^2, and C is 2n x n^2, x is n^2.
Of course, without the constraint x>=0, the problem is under-
determined.
With the addition of constraint "x>=0" into this problem, how exactly
shall we check if the problem is still under-determined and therefore

Re: MPL, AMPL, or OPL?

2008-10-10T20:26:48Z

Actually all examples are listed at:
[link]
JDB
There is a side by side comparison of each modeling language (AMPL,
GAMS, MPL, OPL, Xpress-Mosel) modeling an mrp formulation model in:

Re: Creating a cone

2008-10-10T12:30:05Z

On Thu, 9 Oct 2008 19:04:03 -0700 (PDT), Stephen
Coal mining?...
A.L.

Operations Research Openings at SAS Institute

2008-10-10T02:35:47Z

Dear Colleagues,
Apologies for the mass email and any cross postings!
SAS Institute announces several openings for senior and junior
operations research professionals, many of which are in or related to
the new Center for Excellence in Operations Research Applications.
Specific positions are in research and development, consulting,

Re: Creating a cone

2008-10-10T02:04:03Z

Yes, paper and glue... Of course I'm looking for a method that defines
the smooth non-regular ellipsoid created by taking a horizonal cross
section of the cone. It's for a real world mining application and
isn't homework.

Re: Out-of-print math books: An Update

2008-10-09T14:58:55Z

On Wed, 8 Oct 2008 23:51:07 -0700 (PDT), outofprintmath
Much better, thanks! I can now keep the page open in my browser
without any ill effects.
I notice that when I click on "Show results", no matter what the book,
the text "posted by Tim Chow 7/01/2008 06:58:00 PM" invariably appears
(along with the results, of course!): is this a bug?

Re: Creating a cone

2008-10-09T12:37:19Z

On Wed, 8 Oct 2008 19:46:05 -0700 (PDT), Stephen
Use paper and glue. This should be not too hard
A.L.

Re: numerical optimization and data-fitting and over fitting

2008-10-09T10:02:32Z

In article <f1429a5e-876a-46fa-9b0e-2d28f 40f6...@z6g2000pre.googlegroup s.com>,
in this case this is simple: after regularization your objective function
is strictly convex, hence has a unique minimizer on every convex feasible set,
of which x>=0 is the simplest one
it could even be the case that your problem becomes uniquely solvable _without

Re: numerical optimization and data-fitting and over fitting

2008-10-09T07:21:49Z

Hello,
Adding constraints (a positive orthant constraint) simply makes the
feasible set smaller. You still minimize a strongly convex function
over a non-empty convex set, hence there exist a minimum and it is
unique.
Not just convex but strictly convex.
Sebastian

Re: Out-of-print math books: An Update

2008-10-09T06:51:07Z

You challenged me to find out a better solution and I think the new
version is remarkably faster and requires much less resources.
[link]
Does it work for you?
-- Klaus Schmid