I have written a fairly simple algorithm for solving a linear system Ax=b. I have a parameter rkind that defines kind of my real variables.
Setting rkind=8 creates no problems, but rkind=kind(0.e0) outputs NaNs in my solution. I changed the initialisation with rkind so I don't understand where the problem is.
I'm quite new to Fortran90, any suggestions?
Related
I have a relatively simple question regarding the linear solver built into Armadillo. I am a relative newcomer to C++ but have experience coding in other languages. I am solving a fluid flow problem by successive linearization, using the armadillo function Solve(A,b) to get the solution at each iteration.
The issue that I am running into is that my matrix is very ill-conditioned. The determinant is on the order of 10^-20 and the condition number is 75000. I know these are terrible conditions but it's what I've got. Does anyone know if it is possible to specify the precision in my A matrix and in the solve function to something beyond double (long double perhaps)? I know that there are double matrix classes in Armadillo but I haven't found any documentation for higher levels of precision.
To approach this from another angle, I wrote some code in Mathematica and the LinearSolve worked very well and the program converged to the correct answer. My reasoning is that Mathematica variables have higher precision which can handle the higher levels of rounding error.
If anyone has any insight on this, please let me know. I know there are other ways to approach a poorly conditioned matrix (like preconditioning and pivoting), but my work is more in the physics than in the actual numerical solution so I'm trying to steer clear of that.
EDIT: I just limited the precision in the Mathematica version to 15 decimal places and the program still converges. This leads me to believe it is NOT a variable precision question but rather an issue with the method.
As you said "your work is more in the physics": rather than trying to increase the accuracy, I would use the Moore-Penrose Pseudo-Inverse, which in Armadillo can be obtained by the function pinv. You should then experience a bit with the parameter tolerance to set it to a reasonable level.
The geometrical interpretation is as follows: bad condition numbers are due to the fact that the row/column-vectors are linearly dependent. In physics, such linearly dependencies usually have an origin which at least needs to be interpreted. The pseudoinverse first projects the matrix onto a lower dimensional space in which the vectors are "less linearly dependent" by dropping all singular vectors with singular values smaller than the parameter tolerance. The reulting matrix has a better condition number such that the standard inverse can be constructed with less problems.
My professor gave me a binary linear programming problem, but this problem is slightly different from optimization problems I used to solve(i.e. this is probably not maximizing or minimizing the object function.)
The problem is as follows,
Given a matrix M, for entries m_ij != 0, there are corresponding x_ijk variables.
Entries m_ij = 0 can be ignored.
x_ijk is either 0 or 1, and I want to try 5 x_ijk variables for each m_ij (that is, x_ij1, x_ij2, x_ij3, x_ij4, and x_ij5. One of them is 1 and the others are 0) are enough to satisfy some conditions(a set of inequalities).
More simply, this is to check if the set of constraints involving 5 x_ijk variables for each m_ij is a valid(or feasible) constraints.
I have solved some optimization problems, but I have never solved a problem without an objective function.
What should I set as my objective function here?
0? nothing?
I might be using lp_solve or CPLEX.
Thank you in advance for your advice!
That is correct, you can set an arbitrary constant value as an objective function.
Most of the solvers I have tried allow an empty objective function. Simply leave it out from your model.
Depending on the solver and the API you are using, it can happen that you have to set the coefficients of all variables in the objective to zero.
Don't worry, it has to work.
In response to your comment: Yes, constraint programming tools can provide better performance on feasibility problems than LP solvers (such as CPLEX). I have played with the IBM ILOG CPLEX CP Optimizer a few months ago, it is free for Academic users. Both the LP solver and the CP solver failed on my problems. Don't expect a miracle from constraint programming.
Keep in mind the that time needed to solve a constraint program grows exponentially with the size of the problem in the worse case. Sooner or later, your problems will most likely become unsolvable with either tool.
Just for your information: in the end, the constraint programming solver will call the LP solver (for example CPLEX).
My advice is: try the tool you already have / use the problem formulation that is more natural to you. Check whether the tool can solve your problem. Switch tool only if the tool fails and you cannot improve your model.
How fast is simplex method compared with brute-force or any other algorithm to solve a ts problem?
You can't model a TS problem with a "pure" LP problem (with continuous variables). You can use an integer-programming formulation, wich will use the simplex method at each node of a research tree (branch and bound or branch and cut method). It will work for small problems, but it is slow because the problem is hard: with one binary variable for each edge for instance, you need a lot of constraints to model the fact that the path is a cycle.
Brute-force is intractable (the problem is exponential), do not even try it unless you have a very small problem. Use the MIP formulation, even for small problems.
For big problems, you should use some kind of heuristic (I think simulated annealing give good results on this one), or a "smart" modelization of you problem (column generation for instance) if you want an exact solution.
I created a program using dev-cpp and wxwidgets which solves a puzzle.
The user must fill the operations blocks and the results blocks, and the program will solve it. I'm solving it using brute force, I generate all non-repeated 9 length number combinations using a recursive algorithm. It does it pretty fast.
Up to here all is great!
But the problem is when my program operates depending the character on the blocks. Its extremely slow (it never gets the answer), because of the chars comparation against +, -, *, etc. I'm doing a CASE.
Is there some way or some programming language which allows dynamic creation of operators? So I can define the operator ROW1COL2 to be a +, and the same way to all other operations.
I leave a screenshot of the app, so its easier to understand how the puzzle works.
http://www.imageshare.web.id/images/9gg5cev8vyokp8rhlot9.png
PD: The algorithm works, I tried it with a trivial puzzle, and solved it in a second.
Not sure that this is really what you're looking for but..
Any Object Oriented language such as C++ or C# will allow you to create an "Operator" base class and then to derive from this base class a "PlusOperator" or "MinusOperator" etc'. this is the standard way to avoid such case statements.
However I am not sure this will solve your performance problem.
Using plain brute force for such a problem will result you in an exponential solution. this will seem to work fast for small input - say completing all the numbers. But if you want to complete the operations its a much larger problem with alot more possibilities.
So its likely that even without the CASE your program is not going to be able to solve it.
The right way to try to solve this kind of problems is using some advanced search methods which use some Heuristic function. See the A* (A-star) algorithm for example.
Good luck!
You can represent the numbers and operators as objects, so the parsing is done only once in the beginning of the solving.
I need to solve a few mathematical equations in my application. Here's a typical example of such an equation:
a + b * c - d / e = a
Additional rules:
b % 10 = 0
b >= 0
b <= 100
Each number must be integer
...
I would like to get the possible solution sets for a, b, c, d and e.
Are there any libraries out there, either open source or commercial, which I can use to solve such an equation? If yes, what kind of result do they provide?
Solving linear systems can generally be solved using linear programming. I'd recommend taking a look at Boost uBLAS for starters - it has a simple triangular solver. Then you might checkout libraries targeting more domain specific approaches, perhaps QSopt.
You're venturing into the world of numerical analysis, and here be dragons. Seemingly small differences in specification can make a huge difference in what is the right approach.
I hesitate to make specific suggestions without a fairly precise description of the problem domain. It sounds superficiall like you are solving constrained linear problems that are simple enough that there are a lot of ways to do it but "..." could be a problem.
A good resource for general solvers etc. would be GAMS. Much of the software there may be a bit heavy weight for what you are asking.
You want a computer algebra system.
See https://stackoverflow.com/questions/160911/symbolic-math-lib, the answers to which are mostly as relevant to c++ as to c.
I know it is not your real question, but you can simplify the given equation to:
d = b * c * e with e != 0
Pretty sure Numerical Recipes will have something
You're looking for a computer algebra system, and that's not a trivial thing.
Lot's of them are available, though, try this list at Wikipedia:
http://en.wikipedia.org/wiki/Comparison_of_computer_algebra_systems
-Adam
This looks like linear programming. Does this list help?
In addition to the other posts. Your constraint sets make this reminiscent of an integer programming problem, so you might want to check that kind of thing out as well. Perhaps your problem can be (re-)stated as one.
You must know, however that the integer programming problems tends to be one of the harder computational problems so you might end up using many clock cycles to crack it.
Looking only at the "additional rules" part it does look like linear programming, in which case LINDO or a similar program implementing the simplex algorithm should be fine.
However, if the first equation is really typical it shows yours is NOT a linear algebra problem - no 2 variables multiplying or dividing each other should appear on a linear equation!
So I'd say you definitely need either a computer algebra system or solve the problem using a genetic algorithm.
Since you have restrictions similar to those found in linear programming though you're not quite there, if you just want a solution to your specific problem I'd say pick up any of the libraries mentioned at the end of Wikipedia's article on genetic algorithms and develop an app to give you the result. If you want a more generalist approach, then you've got to simulate algebraic manipulations on your computer, no other way around.
The TI-89 Calculator has a 'solver' application.
It was built to solve problems like the one in your example.
I know its not a library. But there are several TI-89 emulators out there.