Newton-Levenberg-Marquardt method in Ada?

Newton-Levenberg-Marquardt method in Ada?

[van.snyder@jpl.nasa.gov]

A colleague is developing an instrument to measure CO2 and oxygen concentration in the International Space Station.

In order to do that, he needs to invert a nonlinear function.

The usual way to do that is a Newton-Levenberg-Marquardt iteration.

Many numerical libraries have software for it in Fortran or C.

Does anybody have a highly-reliable Newton-Levenberg-Marquardt method in Ada?

Re: Newton-Levenberg-Marquardt method in Ada?

[Lucretia]

On Thursday, 9 August 2018 00:35:56 UTC+1, van.s...@jpl.nasa.gov  wrote:
> A colleague is developing an instrument to measure CO2 and oxygen concentration in the International Space Station.
> 
> In order to do that, he needs to invert a nonlinear function.
> 
> The usual way to do that is a Newton-Levenberg-Marquardt iteration.
> 
> Many numerical libraries have software for it in Fortran or C.
> 
> Does anybody have a highly-reliable Newton-Levenberg-Marquardt method in Ada?

You can import C and Fortran until you write your own in Ada.

Re: Newton-Levenberg-Marquardt method in Ada?

[gautier_niouzes]

Le jeudi 9 août 2018 01:35:56 UTC+2, van.s...@jpl.nasa.gov a écrit :
> A colleague is developing an instrument to measure CO2 and oxygen concentration in the International Space Station.
> 
> In order to do that, he needs to invert a nonlinear function.
> The usual way to do that is a Newton-Levenberg-Marquardt iteration.
> Many numerical libraries have software for it in Fortran or C.
> Does anybody have a highly-reliable Newton-Levenberg-Marquardt method in Ada?

Hi,
As suggested, you can use the pragma Import for your C or Fortran code.
Often I prefer to translate these algorithms (preferably from Fortran or Pascal using an automatic translator, then reviewing manually):
  - often they contain magic numbers, typically epsilon's for floating point types of some VAX machine that won't match the floating point type(s) you are actually using
  - you can spot border-line cases where variants of algorithms are switched, sometimes not in a smooth way...
  - you sometimes find that cannot be detected by Fortran or C, like indexes going outside their intended range
And don't forget to test extensively these numerical algorithms (in whichever language). Even code that seems to be "safe" because it is in a famous decades-old library can contain awful bugs - used by everybody, tested by nobody!