Varun Chopra wrote:
> Fricas-1.2.2 downloaded.
> Can you send the pdf on my mail. The link is directing but pdf is not
> opening.
> and how should I proceed now? Can there be any use of solving differential
> equation model in simulink using MATLAB in this project?
> What should I read and learn ?
>
I have sent you pdf in the mail, I hope you can read it. The project
is about _exact_ (symbolic) computations. We start from differential
operator L which with coefficient beeing rational functions of x.
The coefficient are known exactly (in simplest case rational functions
have rational coefficients which can be represented exactly in
the computer). Fundamental thing is that equaltion Lf = 0
has n-dimensional solution space if we allow sufficiently many
f as solutions. One could try to find solutions among rational
functions, but many equations have no rational solutions.
Larger class is elementary functions (build from rational
functions, exp and log by composition) or Liouvillian
functions (beside rational functions, exp and log one can
use integration to express them). There are differential
equation which have no Liouvillian solutions, so we need larger
class of solutions. A lot of equations can be solved by
power series. However, even power series are not enough.
For example f' - 2*x*f = 0 has x^(1/2) as solutions,
which is not a normal power series at 0 (it needs fractional
exponents). f' + (1/x^2)*f = 0 has exp(1/x) as a solutions
and exp(1/x) has no power series expansion at 0. Sometimes
solutions have logarithmic singularity. When
fractional (possibly irrational) power are allowed then
there are n linearly independent solutions of form
f = e*\sum log(x)^i*p_i
where e = exp(r) where r is rational function of 1/k for some
integer k, and p_i are power series in fractional powers
(Puiseux series). This looks similar to traditional theorem
about existence of solutions to differential equations, but
here coeffients of L may have pole at 0, so most traditonal
results do not apply (they need smooth coeffiecients), and
we say nothing about convergence of p_i. Our power series
are formal power series, so we do only operations on coefficients.
Generalized exponents capture e and leading term of power
series. They tell you how solutions behave close to x = 0.
And they contain important information about possible
factorizations of L. Generalized exponents may be found
by factoring L over ring of Puiseux series (that is we allow
as factors operators having series as coefficients). And
factors with series coefficients help find normal factors.
BTW. All of the above is in van Hoeij's thesis.
To implement van Hoeij's algorithms one needs to operate
with series. FriCAS has apropriate domains to do this.
Series are represented in lazy way, namely there is some
number of already computed coefficients and a rule to
compute next term. So on demand we can compute as many
coefficients as needed. Basic operations like addition,
inverse, multiplication are already done. van Hoeij's
algorithms will need some special operations on series.
OK, I think it is enough for today.
--
Waldek Hebisch
heb...@math.uni.wroc.pl