mdaakir
Registered Member

Hello,
I have one problem while computing some matrix inversion with Eigen library. I implement a least square algorithm, and I do it iteratively. I did the same thing under matlab and it's working well, so the problem is that I can't succeed to do the inversion under Eigen. This how it look like : for(int i=0; i<50; i++) //iterations { //all the other matrices are fine ..... //eqnormal equation Eigen::MatrixXd S(24,12); //declare the matrice S=(Qll*G.transpose())*(G*Qll*G.transpose()).inverse(); //where I do the inversion } the first time, I got the good result for S. After the second iteration, I have only "NaN" inside. Can you please help me ? Thank you. 
ggael
Moderator

You matrix is likely not invertible. You can use inverseWithCheck to verify that. However the best is to avoid the explicit computation of the inverse by factorizing it and call solve method to apply A^1. E.g.:
Also, you should store Qll*G.transpose() into a temporary matrix to avoid computing it twice. 
mdaakir
Registered Member

thank you for your answer.
The problem is, I want to do something like, b*A^1 I took you advice into account and I have : Eigen::MatrixXd s1(24,12); s1.setZero(); //temporary matrix s1=Qll*G.transpose(); Eigen::MatrixXd s2(12,12); s2.setZero();//matrice temporaire s2=G*s1; //Eigen::MatrixXd S(24,12); S.setZero(); //the matrix with inversion //S=(s1)*(s2.inverse()); //not working How should I use : HouseholderQR<MAtrixXd> ? HouseholderQR<MAtrixXd> qr(s2); S=qr.solve(s1); // here the result is then S=s2^1 * s1 and what I want is : S=s1*s2^1 thanks again for your help 
mdaakir
Registered Member

sorry that was stupid question, i can use identity to compute : s3=s2^1 * id
Now I do : Eigen::MatrixXd s1(24,12); s1.setZero();//matrice temporaire s1=Qll*G.transpose(); Eigen::MatrixXd s2(12,12); s2.setZero();//matrice temporaire s2=G*s1; Eigen::MatrixXd id(12,12); id.setIdentity(); Eigen::MatrixXd s3(12,12); s3.setZero(); Eigen::HouseholderQR<Eigen::MatrixXd> qr(s2); s3=qr.solve(id); //s3=s2^1 * id Eigen::MatrixXd S(24,12); S.setZero(); S=s1*s3; I have one message error : variable 'Eigen::HouseholderQR<Eigen::Matrix<double, 0x00000000000000001, 0x00000000000000001> > qr' has initializer but incomplete type can you help me with this please ? Thanks again 
ggael
Moderator

You need to include Eigen/QR and transpose everything:
s1=G*Qll.transpose(); s2=s1*G1.transpose(); Eigen::HouseholderQR<Eigen::MatrixXd> qr(s2); s.transpose() = qr.solve(s1); 
Registered users: Andreas Mair, Baidu [Spider], Bing [Bot], capslock, cloose, edmael, Exabot [Bot], Google [Bot], GreatEmerald, HmpfCBR, johngates, koriun, Majestic12 [Bot], piedro, psbot [Picsearch], sledz, Sogou [Bot], TheraHedwig, tparrott, Yahoo [Bot]