I'm using the QR decomposition to solve a non-least squares but similar looking problem. (Finding the minimum norm element of a plane.) To do this I'm computing the QR decomposition of A.transpose(), and for simplicity (since I use it several times) I had a typedef. So the following code

Code: Select all


typedef Eigen::MatrixXd Matrix;
typedef Eigen::ColPivHouseholderQR<Matrix> cph_qr;

const Matrix A;

//fill A in

const cph_qr qr_fact =  A.transpose().colPivHouseholderQr();


is failing with the error:

error: conversion from ‘const Eigen::ColPivHouseholderQR<Eigen::Matrix<double, -0x00000000000000001, -0x00000000000000001, 1, -0x00000000000000001, -0x00000000000000001> >’ to non-scalar type ‘cph_qr {aka const Eigen::ColPivHouseholderQR<Eigen::Matrix<double, -0x00000000000000001, -0x00000000000000001> >}’ requested

A similar problem happens when I attempt to assign qr_fact.matrixQR().triangularView<Eigen::Upper>() to a triangular view of type Eigen::TriangularView<const Matrix, Eigen::Upper>. (Which presumably is just a consequence of being constructed from a QR factorization of A.transpose().)

It looks like the Option, MaxRows, and MaxCols template parameters got set when I took the transpose. I'm wondering (1) Is there a reason for this, (2) Is there an easier way to solve my problem, and (3) failing that, what are my best options? I could use c++11 syntax and just declare the variable types with auto, or I could replace the typedef for cph_qr with

Code: Select all

typedef Eigen::ColPivHouseholderQR<Eigen::Matrix<double, Eigen::Dynamic, Eigen::Dynamic, Eigen::RowMajor, Eigen::Dynamic, Eigen::Dynamic> > cph_qr;


I unsure which would be preferred. Thanks in advance for any responses.

This is because A is a column major matrix, while A.transpose() is rowmajor. You should rather do:

Code: Select all

const cph_qr qr_fact(A.transpose());
// ...
qr_fact.compute(A);
//...
qr_fact.compute(A.transpose());
//...


Thanks! That fixed it. But I don't understand why I'd make the two calls to compute later. (Though it's good to know that function existed. It isn't in the online docs.)