IMSL routine DGVCRG error

[prashantsondkar]

I am using IMSL routine "DGVCRG" before passing on variables to this routine. I perform some calculations. One simple calculation involves division of two double precision number. When the new number is passed on to this routine it fails giving error

"error 1 from DGVCRG. The iteration for an eigenvalue failed to converge. "

But instead of performing that division, if I insert a number there, which is suppose to come out of that division, it runs fine and I get results

But why it does it mess it up when I have that division.

It goes like this

Run which fails --------------------------------------------
variable1 = variable2 / variable3
variable4 = f (variable1, variable..,........ ) Function of variables


DGVCRG(variable4, ...........)

Run which goes through --------------------------------------------
variable1 = 1.235689 !(which is same as variable2 / variable3)
variable4 = f (variable1, variable..,........ ) Function of variables


DGVCRG(variable4, ...........)


Thanks
Prashant Sondkar

[Steven_L_Intel1]

We don't have general knowledge of the IMSL library internals here. Yours is a good question to ask in the IMSL forum at http://forums.vni.com/forumdisplay.php?f=8

[mecej4]

>I am using IMSL routine "DGVCRG" before passing on variables to this routine. I perform some calculations.

The misplaced punctuation completely changed the meaning of what you probably intended to say.

>it runs fine and I get results

Did you check those results?

The IMSL routine DGVCRG does not take any scalar real arguments. All of its arguments are either integer, or double precision real/complex arrays.

No help can be given when the problem description is incomplete and/or misleading.

We need to see the declarations of the arguments passed to DGVCRG. We need to know which version of IMSL you used, and particulars of the compiler, compiler switches, OS version,...

It would be best for you to make up a small example that shows the problematic behavior that you mentioned and to post the complete source code for the example using the Syntax Highlighting tool (the slanted yellow pen/pencil).

[prashantsondkar]

[fxfortran]
	REAL*8 ratio1, test, rbs, rbp


	! rbs and rbp are user defined for this example their values are
	! rbs = 0.07326, rbp = 0.06678

     ratio1 = rbs/(rbp)
     test = 1.09703504043127

      OmegaS = (OmegaSN * 2 * PI)/60             ! takes values from user OmegaSN is defined by user
      OmegaP=(ratio1)*OmegaS


      Ggr(4,5)= OmegaS * EMSD				! EMSD is user defined

! Here Ggr forms the part of array EK2 and EM2 , rest of the variables N2P, ALPHA, BETAV, EVEC2 are defined


     	CALL DGVCRG (N2P,EK2,N2P, EM2,N2P, ALPHA, BETAV, EVEC2,N2P)[/fxfortran]
 
Comments
If I use ratio1 program fails
but if I use test instead of ratio1 it goes through



Intel Compiler 10.1
IMSL version 6.0

[mecej4]

You have not provided enough information. There is nothing in the code fragment to indicate that any of the arguments to DGVCRG are related to the variables ratio1, test and OmegaP. You may be well-intentioned in not showing us too much detail, but lack of relevant detail remains a roadblock.

[prashantsondkar]

soon I will provide you with simplified code , because if I post entire code it will fill up pages
Thanks
Prashant

[mecej4]

For long source code files, it is better to use these instructions and attach the files, rather than placing them in line:

Attaching or including files in a post

The procedure has two steps: first, you upload your file(s) as a .zip or .tgz file to the Intel server using the instructions, and then place a link to the uploaded file in a post in this forum.

[prashantsondkar]

uploaded_intel_forum.zip

Here are all files of my code. There is INTEL project file also, so you can open it directly.

Now in the file named : DH_FRORCED_VIB_Model_1

if you search for variables ratio and ratio1 you will find following entries
Here are these lines

ratio1 = rbs/(rbp)
ratio = 1.09703504043127

Now 3 lines below you will find
OmegaP=(ratio)*OmegaS

Here if I use ratio, run will go through, but if I use ratio1 it will fail
Both ratio and ratio1 have same values, only thing is rbs and rbp are user input

Hope this will clear my problem

Thanks
Prashant

[mecej4]

I cannot try out your combination of compiler and IMSL versions.

You have the statement

ratio = 1.09703504043127

in which the constant is of type REAL*4, rather than the intended REAL*8. Replace it by

ratio = 1.09703504043127D0

which matches the statement

ratio = rbs/rbp

better, since both rbs and rbp are REAL*8.

The questions

i) why are the eigenvalue calculations are so sensitive to this ratio?

ii) how to fix the problem?

iii) given the sensitiveness, do the input data have sufficient precision?

need to be addressed by you or someone versed in the application area.

[prashantsondkar]

YOu are suggesting changing the constant type, but run goes through with
ratio = 1.09703504043127

and fails with

ratio = rbs/rbp

lets say run goes through by entering the value of rbs/rbp but fails when calculations are perfomed by fortran


ratio = rbs/rbp (fails)
ratio = 1.09703504043127 (goes through)

is it possible while doing this calculation it changes data type.

[mecej4]

You are drawing unfounded conclusions. Whether the calculation "goes through" or not is irrelevant at this point.

The rules of Fortran are quite explicit as to assignment statements. The right hand side expression is evaluated using mixed mode rules, the result is converted if necessary to the type of the variable on the left hand side.

In this case, the expression on the right has type REAL*4, regardless of how many digits you typed in for the constant. The variable on the left is REAL*8. To convince yourself, print the value of the variable to, say, 15 digits:

ratio = 1.09703504043127
write(*,'(1x,1p,D20.14)') ratio

You will see the printed value as

1.09703505039215D+00

You have another statement in the same file which is waiting to give you trouble:

PISI(i)=(i-1)*(360/Nplanets)

If Nplanets is, for example, 7, and i = 2, the value assigned will be 50.0D0

[prashantsondkar]

No NPlanets always assume value of 5, even though user has control over it, it is limited to 5, somehow told to user through separate intructions



So what shall I do, just to make it work with using ratio1 =rbs/rbp

Thanks
Prashant

[mecej4]

Given the difficulty with others being able to match your combination of Fortran compiler and IMSL versions, here is a suggestion.

If you use IMSL only for the call to DGVCRG, you can set IMSL aside and use MKL instead (if your Intel Fortran came with it or you obtained it separately).

1. Remove all references to IMSL modules, include files and libraries.

2. In dh_frorced_vib_model_1.f, add the following before any declarations:

use mkl95_lapack

3. Instead of allocating the complex array ALPHA, allocate REAL*8 arrays ALPHAR and ALPHAI.

4. Instead of calling DGVCRG, do

call ggev(EK2, EM2, alphar, alphai, betav)
EVAL = cmplx(ALPHAR,alphaI) / BETAV

5. If your compiler version allows a /Qmkl switch, do

ifort /Qmkl *.f mkl_lapack95.lib

If not, read your MKL user's guide to see how to compile and link to MKL.

I have tried this and it works fine with Intel Composer XE 12.0.4 and the accompanying MKL 10.3.

[prashantsondkar]

I am getting this error
Error 1 Error: Error in opening the compiled module file. Check INCLUDE paths. [MKL_LAPACK]

Header lines from code are,

[fxfortran]      PROGRAM DHFORVIB
      
C      USE imsl_libraries    
C
C	INCLUDE 'link_fnl_static.h'
C!	INCLUDE 'link_fnl_shared.h'
C	
C	!DEC$ OBJCOMMENT LIB:'libiomp5md.lib' 
      USE MKL_LAPACK
      
C      INCLUDE 'link_fnl_shared_hpc.h'[/fxfortran]

[mecej4]

use mkl95_lapack

Check whether you have Lapack95. If not, you may need to build it from sources (provided with MKL). Alternatively, you can make a call to the F77 routine DGGEV -- see the MKL documentation.

[psantos]

Hello prashantsondkar,

In order the use statement to work you must go to your Project Properties>Linker>Input>Additional Dependencies and add "mkl_lapack95.lib" if you are compiling in 32bit or "mkl_lapack95_lp64.lib" if you are compiling in 64bit. You should also verify that you have activated MKL. Just go to your project properties>Fortran>Libraries>Use Intel Math Kernel and select either Parallel or Sequential.

Pedro