Error in <TDecompChol::Decompose()>: matrix not positive definite

Roofit and RooStats
1

Hi Experts,

I was trying to invert TMatrixD matricies with small elements. Initially it was giving the following error[*]

[*]
Error in TDecompLU::InvertLU: matrix is singular, 6 diag elements < tolerance of 2.2204e-16
Error in TDecompChol::Decompose(): matrix not positive definite

then I put a .IsValid() to check whether the inversion was successful and also set matrix tolerance value to 2.2e-10. There was no more the “InverLU” message but still it shows the “matrix not positive definite” message[**]. Could you please tell me how to take care of it ??

[**]
Error in TDecompChol::Decompose(): matrix not positive definite

cheers - Niladri

2

Hi Niladri,

Could you show some code. I am puzzled about usage and TMatrixD and
the apperance of TDecompChol .

Best Regards, Eddy

3

Hi Eddy,

Here is the snipped code attached.

TMatrixDSym cov(dim);
for (i = 0; i < dim; i++) {
for (j = 0; j < dim; j++) {
double errPar = readParam(iBin,TString::Format(“EffErr%d”,i+1), j);
//cout << cov.IsValid() << endl;
if (! cov.IsValid()) continue;
//cov(i,j) =
errPar;

  if (errPar <= 0){
    cov(i,j) = 2.2e-10 ;
  } else {
    cov(i,j) = errPar;
  }

}

}

cheers - Niladri

4

Hi Niladri,

Are you sure that you are filling your matrix symmetrically through this ReadParam
routine ?

I do not see a call to a matrix inversion routine !

Best Eddy

5

Hi Eddy,

I am saving the matrix elements in the EffPar file and then I’m reading the elements in this script …

cheers – Niladri

6

Hi Eddy,

Still I’m getting the same message “Error in TDecompChol::Decompose(): matrix not positive definite”. When I changed the #parameters to 15, it worked absolutely fine. But when I am giving the #parameters to be 20, then this message shows off. Please tell me how to get rid of this message.

best regards – Niladri

7

Hi Niladri,

You will have to post something with which this error can be reproduced,
preferably short. Up to now you did not show any code where the inversion
is being done.

-Eddy