Matlab coder - wrong answer

Hi guys,

I'm tring to export matlab code to C code, in my matlab code i have Eigen function in order to calculate the Eigen vector right matrix.

Although when i compile my code in C compiler (atollic true studio) i received a wrong answers.

Can anyone please help me to solve this problem ?

input matrix and function :[cN,~]=eig([6506.1709323862497 + 0 * i, 352.72061684167767 + -3710.3651327887619 * i, 2011.5655185722712 + 384.26985324363284 * i, 1725.5363173165599 + -5930.123440499453 * i; 352.72061684167767 + 3710.3651327887619 * i, 6199.5160239281295 + 0 * i, 356.19284194608099 + -3663.5190286699631 * i, 1768.5916083618101 + 331.40000184453999 * i; 2011.5655185722712 + -384.26985324363284 * i, 356.19284194608099 + 3663.5190286699631 * i, 6447.9241828420545 + 0 * i, 382.01986570001651 + -4005.4837002418008 * i; 1725.5363173165599 + 5930.123440499453 * i, 1768.5916083618101 + -331.40000184453999 * i, 382.01986570001651 + 4005.4837002418008 * i, 6611.8755981240902 + 0 * i])

The first picture is the values that i received after the eigen function, and the second picture the the correct values that i expect to receive

[Wilson A N]

Hi Tamir,

Can you share the exact values you had used with the eigen function. Additionally, providing the commands used for code generation would help.

- Wilson

[Tamir Mizrahi]

Thanks Wilson , i updated my post

Answer 1

[Areej Varamban Kallan]

Hi Tamir,

Thanks for sharing the values. I am able to reproduce the results.

The computed results in the generated code are correct. For a given matrix, the eigenvector corresponding to a particular eigevalue is not unique. If v is an eigenvector, then alpha*v, which is a scaled version of v is also an eigenvector for the same eigenvalue.

For the inputs considered here, each eigenvector returned by the C code is a scaled version of the corresponding eigenvector returned by MATLAB.

To better understand the results, let us compute eigenvalues as well.

function [v,d] = myeig(a)
%#codegen
[v,d] = eig(a);

>> cfg = coder.config('lib');
>> cfg.VerificationMode = 'SIL';
>> codegen myeig -args {coder.typeof(1i,[4 4])} -config cfg

>> [vmatlab,dmatlab] = myeig(A)

vmatlab =

  Columns 1 through 2

 -0.376796540203342 - 0.355408671476361i -0.398541780958145 + 0.374702617926804i
  0.001275398603249 + 0.561417931091213i -0.143284708674997 + 0.311017096268371i
  0.480804533726921 - 0.184887065618415i  0.248485110194001 + 0.324562032274043i
 -0.388794414636469 + 0.000000000000000i  0.645302171278108 + 0.000000000000000i

  Columns 3 through 4

 -0.045415333593944 + 0.168282082335769i  0.176847576029146 - 0.608945780485925i
 -0.699329885477756 - 0.125756891058523i  0.246152272181504 + 0.045586320593164i
  0.056552956858085 - 0.672545130245288i  0.032517800265024 - 0.333145334730278i
  0.096053707488093 + 0.000000000000000i  0.650536467977805 + 0.000000000000000i


dmatlab =

   1.0e+04 *

   0.000172468967993                   0                   0                   0
                   0   0.000229944404698                   0                   0
                   0                   0   1.036674139012830                   0
                   0                   0                   0   1.539472121342531

      
[vccode,dccode] = myeig_sil(A)

vccode =

  Columns 1 through 2

 -0.634105692072699 + 0.000000000000000i -0.174302644158693 - 0.000000000000000i
 -0.024872565084105 - 0.249099195565696i -0.060800388869246 - 0.707940955930921i
 -0.328995848224188 + 0.061684303010346i  0.664049974356741 - 0.120635072441042i
 -0.181430002787685 - 0.624724619538480i  0.025027223135727 + 0.092735930600402i

  Columns 3 through 4

  0.547026144759982 + 0.000000000002153i -0.517968103718098 + 0.000000000008760i
  0.317432475274703 - 0.128447520612772i  0.386149968271705 + 0.407529044359775i
  0.041282386650599 - 0.406670784952457i  0.222899861257117 - 0.464404864894278i
 -0.470141837021641 - 0.442019847222879i -0.282828979683157 + 0.266774933423510i


dccode =

   1.0e+04 *

   1.539472121342530                   0                   0                   0
                   0   1.036674139012831                   0                   0
                   0                   0   0.000229944404698                   0
                   0                   0                   0   0.000172468967994
                                                                                                

The eigenvalues returned by MATLAB and C code are the same but they appear in a different order.

Now we must check if the computed results are correct by verifying if they satisfy the eigenvalue equation, ie for

[V,D] = eig(A), A*V - V*D = 0 must be satisfied within some tolerance.

   >> norm(A*vmat- vmat*dmat)

ans =

     3.294189142814234e-12

>> norm(A*vccode- vccode*dccode)

ans =

     1.220174523947488e-11
                

We see that the computed values are solutions to the given eigenvalue problem

[Tamir Mizrahi]

Thanks alot ! , but my C algorithm expect to receive the not scaled eigenvector ,and this is the only eigenvector that works well in the next C algorithm code.

so how can i extract the first eigenvector matrix in this case?

Answer 2

[Areej Varamban Kallan]

Hi Tamir,

Could you please share the inputs that were passed to eig. Which version of MATLAB are you using?

Please note that the generated code might return eigenvalues in a different sorted order compared to MATLAB. Moreover, the matrix of eigen vectors could be returned in a different basis.

Please check this link for details.

[Tamir Mizrahi]

Thanks Areej, i update my post.

my version is : '9.6.0.1114505 (R2019a) Update 2'