Re: Boundary value problem.

Peter Spellucci wrote:
In article <f2nk4e$f3a$1@xxxxxxxxxxxxxxxxxxx>,
Amit Bhatia <abhatia@xxxxxxxxxxxxxxxxx> writes:
>Peter Spellucci wrote:
>> In article <f2hui6$gio$1@xxxxxxxxxxxxxxxxxx>,
>> Amit Bhatia <abhatia@xxxxxxxxxxxxxxxxx> writes:
>> >
>> >Hi,
>> > I am trying to identify the parameters for the following two point >> >boundary value problem:
>> >
>> >y' = f(x,y)=[cos(y(3));sin(y(3));u(y)]
>> >where -1<=u(y)<=1,
>> >u(y)= c1*y(2)-c2*y(1)+c3 OR sign(c1*y-c2*x+c3) , c1,c2,c3 are constants >> >to be determined.
>> >
>> >y(a), y(b) are given to me.
>> >
> > >It is a given that there exists smooth differentiable solution to the >> >problem, but there can be more than one such solutions.
>> >
>> >
>> >If I solve this problem using bvp4c in matlab, I get the error
>> >"Unable to solve the collocation equations -- a singular Jacobian >> >encountered"
>> >
>> >So apparently my initial guess is not good or there is some other >> >problem. I was wondering if I should be doing something else or using >> >something different then this Matlab routine?
>> > The problem is more of parameter estimation probably.
>> >
>> >thanks,
>> >--a.
>> >> what does this "OR" mean : two cases , a switch in the definition of y(3)' >> or what? the sign-case might be harmful if you have a bad initial guess >> and a sign change occurs, such that a jump occurs in the course of integration
>> this will led to failure. also, if there exists several (isolated) solutions
>> then somewhere the jacobian of the system must become singular, and again with >> a bad choice you will get trouble. did you try "shooting" to get an initial guess?
>> maybe if you begin with a low precision requirement first and then increase this
>> step by step, using interpolation of the old solution to get the new guess?
>> hth
>> peter
>
>OR means that when the value of u(y) gets saturated we use the >maximum/minimum possible, i.e.,
>
>u(y)=c1*y(2)-c2*y(1)+c3, if -1 < c1*y(2)-c2*y(1)+c3 < 1
> =1 if c1*y(2)-c2*y(1)+c3 >1
> =-1 if c1*y(2)-c2*y(1)+c3 < -1
>
>I am not sure what you meant by initial guess: I have y(0), y(T) already >specified. So I need the constants c1, c2, c3 only. I do start with an >initial guess for them, but that gives me that Jacobian is singular.
> I do not think there are any jumps, however there is non linearity in >form of saturation. Also, we can have sines and cosines in solutions, so >I am not sure but it may cause problems?
>In any case, it is a given that the solution y() is continuous and >differentiable.
>
>--a.

so your problem has a nonsmooth right hand side, this means that the collocation based solvers will fail. you must identify the switch points and
solve a multipoint boundary value problem (both tasks must even be mixed here)
and due to the nonlinearity you need indeed an initial grid, not containing the
two boundary points only, and the guess of the solution there. otherwise again the solver may (and will) run in the desert.
you should take a closer look on a textbook on solving nonlinear two point
boundary value problems , for example Ascher, Mattheij and Russell


hth
peter

So that means that even if I had smooth non linear right hand side, I will still need good initial guesses at some intermediate points rather than just the end point conditions, i.e. for non linear problems grid of just two end points is not going to work either?

thanks,
--a.
.

Relevant Pages

  • Re: Error propagation through a iterative root-finding function
    ... (Peter Spellucci) ... P have standard deviations associated with them. ... of the original distribution, depeding on x. otherwise things can be become ... function can be assumed linear in the vicinity of x. ...
    (sci.math.num-analysis)
  • Re: time complexity for iterative methods
    ... (Peter Spellucci) ... a good enough precision is enough ... =A0and the required precision, and hopefully with some idea about the con= ... it will be linear .... ...
    (sci.math.num-analysis)
  • Re: Damn you, FEDEX! or Nikon D40 lost in Springfield, MO blackhole.
    ... the 2 mp Mavica he had been using with a Nikon D40. ... After shopping around, he got me to order one for him. ... The shipper had it insured, but from what I have read it could take weeks to sort this crap out. ... You may get your insurance from FedEx and a couple weeks later they find it and deliver it. ...
    (alt.photography)
  • Re: Linear Equations Similar to 1D Poisson
    ... (Peter Spellucci) ... how to deal with a linear system with matrix ...  you write the coefficients of the linear equation on top of the matrix.. ... constraint when ATA is invertible, ...
    (sci.math.num-analysis)
  • Re: ? estimate newton step
    ... (Peter Spellucci) ... convergent result. ... f is s linear function. ... The very first question is: can we always get correct result using ...
    (sci.math.num-analysis)