topic Hi Yuri, in Intel® oneAPI Math Kernel Library

Elements of functional analysis

Yuri_D_ — Wed, 08 Oct 2014 12:29:43 GMT

Good day!

I'm sorry if my question being improper but I really want to know if there is any features for functional manipulating in Intel MKL? More precisely, I've got now a task to solve a system of non-linear equations. To do that I have to obtain partial derivatives of my functions. Can I do this using a standart method of MKL and what is the way of such a realization?

Hi Yuri D,

Ying_H_Intel — Thu, 09 Oct 2014 01:47:01 GMT

Hi Yuri D,

As far as i know, no, mkl don't provide feature for functional manipulation. you may refer to the mkl manual at https://software.intel.com/en-us/articles/intel-math-kernel-library-documentation. The no-linear solvers supported by MKL are mainly based on interation, jacobian etc, not direct partial derivatives of one functions

Intel MKL provides tools for solving nonlinear least squares problems using the
Trust-Region (TR) algorithms. The solver routines are grouped according to their purpose as follows:
• Nonlinear Least Squares Problem without Constraints
• Nonlinear Least Squares Problem with Linear (Boundary) Constraints
• Jacobian Matrix Calculation Routines
For more information on the key concepts required to understand the use of the Intel MKL nonlinear least
squares problem solver routines, see [Conn00].

Best Regards,

Ying

Thank you, Ying!

Yuri_D_ — Thu, 09 Oct 2014 20:17:50 GMT

Thank you, Ying!

I found a way of realization of my task; I had completely overlooked the fact that MKL can manage solution of nonlinear equations' systems.

But I caught myself not being capable to realize more simple things. I hope you will forgive me some newbie questions such as:

1) how can I find [Conn00]?

2) how can I handle this problem:

In function 'func':

.../func.f: 75: underfined reference to 'dtrnlsp_init_'

I don't even imagine the reason for this problem to appear, because I wrote the include operator (include 'mkl_rci.fi') in my code and, also, gave the compiler the full address of the file, that has been included.

I'm very sorry if my questions are too annoying - this is my second day of Intel MKL usage and I'm just trying to sort the things out.

Hi Yuri

Ying_H_Intel — Sat, 11 Oct 2014 03:45:11 GMT

Hi Yuri

You can download the mkl manual and mkl userguide from the website. at https://software.intel.com/en-us/articles/intel-math-kernel-library-documentation.

Then you will see the reference doc is from

1) [Conn00] A. R. Conn, N. I.M. Gould, P. L. Toint.Trust-region Methods.SIAM Society for
Industrial & Applied Mathematics, Englewood Cliffs, New Jersey, MPS-SIAM
Series on Optimization edition, 2000.

2) Are you building the lexample code provided by MKL? What is your build environment and compiler command line?

It is right, Underfined reference error hints that the problem in the include path or the function name. DTRNLSP_INIT ( dtrnlsp_init_, the underline seems redundant, it can it be left out? , could you please attach the code or please try the code under mkl example : [MKL Dir]\example\solverf\source\ex_nlsqp_f.f ( attached).

Best Regards,

Ying

Thank you, Ying! You've

Yuri_D_ — Fri, 17 Oct 2014 06:18:38 GMT

Thank you, Ying! You've helped me so much!

I've remastered this program and tried to use it, but then I found out, that I obtained complex values on some steps of iterations. And this fact brought me to the necessity of changing types of the solution vector's coordinates - as well as the function value vector's ones.

So, would I be right if I change these types to COMPLEX*16?

Hi Yuri,

Alexander_K_Intel2 — Mon, 20 Oct 2014 08:01:21 GMT

Hi Yuri,

Current version of MKL support nonlinear solver only for real variables. For sure one can calculate complex equation but after conversion it to real one (value.complex -> (value.re, value.im)). So could you explain what does following sentence mean "I've remastered this program and tried to use it, but then I found out, that I obtained complex values on some steps of iterations"?

Thanks,

Alex