Eigen Basics

xWe can import it like this:

#include <iostream>
using namespace std;
 
#include <ctime>
 
// Eigen core
#include <Eigen/Core>
 
// Algebraic operations of dense matrices (inverse, eigenvalues, etc.)
#include <Eigen/Dense>
 
using namespace Eigen;

All vectors and matrices in Eigen are Eigen::Matrix, which is a template class. Its first three parameters are: data type, row, column. Declaring a 2x3 float matrix:

Matrix<float, 2, 3> matrix_23;

At the same time, Eigen provides many built-in types via typdef, but the bottom layer is still Eigen::Matrix. For example, Vector3d is essentially Eigen::Matrix<double, 3, 1>.

// These are the same
Vector3d v_3d;
Matrix<double, 3, 1> vd_3d;

• Matrix3d is also essentially the same as Eigen::Matrix<double, 3,3>.

If we’re not sure about the size of the matrix, we can use one with dynamic size:

Matrix<double, Dynamic, Dynamic> matrix_dynamic;
MatrixXd matrix_x; // simpler version of the above

Input and output operations:

// input data (initialization)
matrix_23 << 1, 2, 3, 4, 5, 6;
 
// output
cout << "matrix 2x3 from 1 to 6: \n" << matrix_23 << endl;

Using loops to access elements in the matrix:

cout << "print matrix 2x3: " << endl;
for (int i = 0; i < 2; i++) {
	for (int j = 0; j < 3; j++)
	cout << matrix_23(i, j) << "\t";
	cout << endl;
}

We can easily multiply a matrix with a vector (but actually still matrices and matrices). In Eigen you can’t mix two different types of matrices, like this is wrong:

Matrix<double, 2, 1> result_wrong_type = matrix_23 * v_3d;

We need to explicitly convert by casting:

v_3d << 3, 2, 1;
vd_3d << 4, 5, 6;
 
Matrix<double, 2, 1> result = matrix_23.cast<double>() * v_3d;
cout << "[1,2,3;4,5,6]*[3,2,1]=" << result.transpose() << endl;
 
Matrix<float, 2, 1> result2 = matrix_23 * vd_3d.cast<float>();
cout << "[1,2,3;4,5,6]*[4,5,6]: " << result2.transpose() << endl;

Also you can’t misjudge the dimensions of the matrix like this:

Eigen::Matrix<double, 2, 3> result_wrong_dimension = matrix_23.cast<double>() * v_3d;

Matrix operations:

matrix_33 = Matrix3d::Random(); // Random Number Matrix
cout << "random matrix: \n" << matrix_33 << endl;
cout << "transpose: \n" << matrix_33.transpose() << endl;
cout << "sum: " << matrix_33.sum() << endl;
cout << "trace: " << matrix_33.trace() << endl;
cout << "times 10: \n" << 10 * matrix_33 << endl;
cout << "inverse: \n" << matrix_33.inverse() << endl;
cout << "det: " << matrix_33.determinant() << endl;

Eigenvalues;

// Real symmetric matrix can guarantee successful diagonalization
SelfAdjointEigenSolver<Matrix3d> eigen_solver(matrix_33.transpose() * matrix_33);
cout << "Eigen values = \n" << eigen_solver.eigenvalues() << endl;
cout << "Eigen vectors = \n" << eigen_solver.eigenvectors() << endl;

Solving equations also works! We can solve the equation of matrix_NN * x = v_Nd. The size of N is defined in the previous macro, which is generated by a random number. Direct inversion is the most direct but the number of operations is large.

Matrix<double, MATRIX_SIZE, MATRIX_SIZE> matrix_NN =
	MatrixXd::Random(MATRIX_SIZE, MATRIX_SIZE);
matrix_NN =
	matrix_NN * matrix_NN.transpose(); // Guarantee semi-positive definite
Matrix<double, MATRIX_SIZE, 1> v_Nd = MatrixXd::Random(MATRIX_SIZE, 1);

Benchmarking different inversion methods:

clock_t time_stt = clock(); // timing
// Direct inversion
Matrix<double, MATRIX_SIZE, 1> x = matrix_NN.inverse() * v_Nd;
cout << "time of normal inverse is "
	<< 1000 * (clock() - time_stt) / (double)CLOCKS_PER_SEC << "ms" << endl;
cout << "x = " << x.transpose() << endl;
 
// Usually solved by matrix decomposition, such as QR decomposition, the speed
// will be much faster
 
time_stt = clock();
x = matrix_NN.colPivHouseholderQr().solve(v_Nd);
cout << "time of Qr decomposition is "
	<< 1000 * (clock() - time_stt) / (double)CLOCKS_PER_SEC << "ms" << endl;
cout << "x = " << x.transpose() << endl;
 
// For positive definite matrices, you can also use cholesky decomposition to
// solve equations.
 
time_stt = clock();
x = matrix_NN.ldlt().solve(v_Nd);
cout << "time of ldlt decomposition is "
	<< 1000 * (clock() - time_stt) / (double)CLOCKS_PER_SEC << "ms" << endl;
cout << "x = " << x.transpose() << endl;