RigidElasticSim/src/problems/barrier_problem.cpp

// Solve the optimization general_problem using Newton's Method with barriers
// for the constraints.
#include "barrier_problem.hpp"

#include <finitediff.hpp>

namespace ipc::rigid {

// Compute the objective function f(x) = E(x) + κ ∑_{k ∈ C} b(d(x_k))
double BarrierProblem::compute_objective(
    const Eigen::VectorXd& x,
    Eigen::VectorXd& grad,
    Eigen::SparseMatrix<double>& hess,
    bool compute_grad,
    bool compute_hess)
{
    // Compute each term
    double Ex = compute_energy_term(x, grad, hess, compute_grad, compute_hess);

    // The following is used to disable constraints if desired
    // (useful for testing).
    if (!m_use_barriers) {
        return Ex;
    }

    Eigen::VectorXd grad_Bx;
    Eigen::SparseMatrix<double> hess_Bx;
    int num_constraints;
    double Bx = compute_barrier_term(
        x, grad_Bx, hess_Bx, num_constraints, compute_grad, compute_hess);

    double kappa = barrier_stiffness();

    if (compute_grad) {
        grad += kappa * grad_Bx;
    }
    if (compute_hess) {
        hess += kappa * hess_Bx;
    }
    return Ex + kappa * Bx;
}

Eigen::VectorXd
eval_grad_energy_approx(BarrierProblem& problem, const Eigen::VectorXd& x)
{
    auto f = [&](const Eigen::VectorXd& xk) -> double {
        return problem.compute_energy_term(xk);
    };
    Eigen::VectorXd grad(x.size());
    fd::finite_gradient(x, f, grad);
    return grad;
}

Eigen::MatrixXd
eval_hess_energy_approx(BarrierProblem& problem, const Eigen::VectorXd& x)
{
    Eigen::MatrixXd hess;
    auto func_grad_f = [&](const Eigen::VectorXd& xk) -> Eigen::VectorXd {
        Eigen::VectorXd grad;
        problem.compute_energy_term(xk, grad);
        return grad;
    };

    fd::finite_jacobian(x, func_grad_f, hess, fd::AccuracyOrder::SECOND);
    return hess;
}

} // namespace ipc::rigid