Re: TWO questions to experts, (academics & pratical)

[vuca@envi.env.civil.saitama-u.ac.jp (Vu Thanh Ca)]

Dear Sonnet,

I feels that your questions are too general to be answered. That's why I did not
answer. I think you know very well that the most important thing for a 
programmer is the numerical algorithm. Programming is very easy if you 
understand the angorithm and the physical meaning of your problem.

It is EXTREMELY difficult to say that you have the best algorithm. There are
manythings that are still UNDER STUDYING. Yet, we have a lot of 
well-tested, stable and accurate algorithms in text books of numerical
computation and journals. It is not possible that one can use the latest
knowledge. However, if we use some well-known algorithms, people
can understand how we treat the problem.

We are writing programs for sell in Japan and sell them to Japanese
companies, research institutes. What we do is to make a mannual 
about all equations, the way we treat the boundary conditions, and 
options to run the program. You can change the boundary conditions 
by just make subroutines with different algorithms and specify the 
options for your boundary conditions.

Even in fluid mechanics, there is something called "Computational 
Fluid Mechanics" which was established and is being developed by many
great scientists in the world. Even I am one of people who are working
in the field, I do not understand all things in this field. However, you 
can take almost all necessary knowledge about a stable, accurate 
scheme for the Navier-Stokes equations in the newest text books.


Cheers. Ca.


At  6:14 AM 97.3.27 -0600, Sonnet Nguyen wrote:
>Hi,
>On Wed, 26 Mar 1997, Aiviet Nguyen wrote:
>
>> Sure, there are quite a lot in Fortran and in C. You can also write one 
>> by yourself if no additional requirements are needed.
>
>
>Your answer is as usually "too general".  Each person perfectly knows
>about that.  
>Of course, I can write (and I've written a lot), but my products are 
>useful and convenient only for my uses.  (For solving only special types
>of equations (for eg: fixed functional coefficients, etc.), and
>convergence of algorithm is not verified for variety of boundary
>conditions)   
>
>Let consider for example the Navier-Stokes equation.  Problems of
>hidrodynamics and fluid dynamics (and many problems of NASA, BOEING) are
>reduced to solving this equation with different (arbitrary) boundary
>conditions. 
>But could you imagine yourself a patological boundary consisting of 
>100.000 different elements? (For interesting applications, boundary
>consists of more than 1 million diff. elements).  Do you know how many
>institutes of fluid dynamics are in the world? 
>For what, if each student can write a program solving Navier-Stokes
>equation.   The problem is that program must be convenient (for
>engeeners) and algorithm of solving is tested and optimalized.  
>Not everything is "simple" and "beauty" as in Pure Math or in theoretical
>physics.
>
>regards,
>SN