GenericProblem.cpp

Go to the documentation of this file.

#include "GenericProblem.hpp"
 
#include <polyfem/utils/JSONUtils.hpp>
#include <polyfem/utils/Logger.hpp>
#include <polyfem/utils/StringUtils.hpp>
#include <polyfem/io/MatrixIO.hpp>
 
namespace polyfem
{
    using namespace utils;
 
    namespace assembler
    {
        namespace
        {
            bool matches_fe_space(const int entry_fe_space_id, const int fe_space_id)
            {
                return entry_fe_space_id < 0 || fe_space_id < 0 || entry_fe_space_id == fe_space_id;
            }
 
            int fe_space_id(const json &entry)
            {
                return entry.value("fe_space", -1);
            }
 
            bool is_body_value_entry(const json &entry)
            {
                if (!entry.is_object() || !entry.contains("id") || !entry.contains("value"))
                    return false;
 
                const json &id = entry["id"];
                if (id.is_array())
                {
                    for (const json &value : id)
                        if (value.is_number_integer() && value.get<int>() != -1)
                            return true;
                    return false;
                }
 
                return id.is_number_integer() && id.get<int>() != -1;
            }
 
            bool has_body_value_entries(const json &value)
            {
                return is_body_value_entry(value) || (value.is_array() && !value.empty() && is_body_value_entry(value.front()));
            }
 
            template <typename Map>
            const typename Map::mapped_type *find_for_fe_space(const Map &values, const int fe_space_id)
            {
                if (const auto it = values.find(fe_space_id); it != values.end())
                    return &it->second;
                if (const auto it = values.find(-1); it != values.end())
                    return &it->second;
                if (fe_space_id < 0 && values.size() == 1)
                    return &values.begin()->second;
                return nullptr;
            }
 
            std::vector<json> flatten_ids(const json &p_j_boundary_tmp)
            {
                const std::vector<json> j_boundary_tmp = utils::json_as_array(p_j_boundary_tmp);
 
                std::vector<json> j_boundary;
 
                for (size_t i = 0; i < j_boundary_tmp.size(); ++i)
                {
                    const auto &tmp = j_boundary_tmp[i];
 
                    if (tmp.is_string())
                    {
                        j_boundary.push_back(tmp);
                        continue;
                    }
                    if (!tmp.contains("id"))
                        continue;
 
                    if (tmp["id"].is_array())
                    {
                        for (size_t j = 0; j < tmp["id"].size(); ++j)
                        {
                            json newj = tmp;
                            newj["id"] = tmp["id"][j].get<int>();
                            j_boundary.push_back(newj);
                        }
                    }
                    else
                        j_boundary.push_back(tmp);
                }
 
                return j_boundary;
            }
        } // namespace
 
        double TensorBCValue::eval(const RowVectorNd &pts, const int dim, const double t, const int el_id) const
        {
            double x = pts(0);
            double y = pts(1);
            double z = pts.size() == 2 ? 0 : pts(2);
 
            double val = value[dim](x, y, z, t, el_id);
 
            if (interpolation.empty())
            {
            }
            else if (interpolation.size() == 1)
                val *= interpolation[0]->eval(t);
            else
            {
                assert(dim < interpolation.size());
                val *= interpolation[dim]->eval(t);
            }
 
            return val;
        }
 
        double ScalarBCValue::eval(const RowVectorNd &pts, const double t) const
        {
            assert(pts.size() == 2 || pts.size() == 3);
            double x = pts(0), y = pts(1), z = pts.size() == 3 ? pts(2) : 0.0;
            return value(x, y, z, t) * interpolation->eval(t);
        }
 
        GenericTensorProblem::GenericTensorProblem(const std::string &name)
            : Problem(name), is_all_(false)
        {
        }
 
        void GenericTensorProblem::set_units(const assembler::Assembler &assembler, const Units &units)
        {
            if (assembler.is_fluid())
            {
                for (int i = 0; i < 3; ++i)
                {
                    for (auto &[fe_space_id, rhs] : rhs_)
                        rhs[i].set_unit_type(units.force());
                    exact_[i].set_unit_type(units.velocity());
                }
                for (int i = 0; i < 3; ++i)
                    exact_grad_[i].set_unit_type("");
 
                for (auto &v : displacements_)
                    v.set_unit_type(units.velocity());
 
                for (auto &v : forces_)
                    v.set_unit_type(units.force());
 
                for (auto &v : normal_aligned_forces_)
                    v.set_unit_type(units.pressure());
 
                for (auto &v : pressures_)
                    v.set_unit_type(units.pressure());
 
                for (auto &v : cavity_pressures_)
                    v.second.set_unit_type(units.pressure());
 
                for (auto &v : initial_position_)
                    for (int i = 0; i < 3; ++i)
                        v.value[i].set_unit_type(units.velocity());
 
                for (auto &v : initial_velocity_)
                    for (int i = 0; i < 3; ++i)
                        v.value[i].set_unit_type(units.velocity());
 
                for (auto &v : initial_acceleration_)
                    for (int i = 0; i < 3; ++i)
                        v.value[i].set_unit_type(units.acceleration());
 
                for (auto &v : nodal_dirichlet_)
                    v.second.set_unit_type(units.velocity());
 
                for (auto &v : nodal_neumann_)
                    v.second.set_unit_type(units.force());
            }
            else
            {
                for (int i = 0; i < 3; ++i)
                {
                    for (auto &[fe_space_id, rhs] : rhs_)
                        rhs[i].set_unit_type(units.acceleration());
                    exact_[i].set_unit_type(units.length());
                }
                for (int i = 0; i < 3; ++i)
                    exact_grad_[i].set_unit_type("");
 
                for (auto &v : displacements_)
                    v.set_unit_type(units.length());
 
                for (auto &v : forces_)
                    v.set_unit_type(units.force());
 
                for (auto &v : normal_aligned_forces_)
                    v.set_unit_type(units.pressure());
 
                for (auto &v : pressures_)
                    v.set_unit_type(units.pressure());
 
                for (auto &v : cavity_pressures_)
                    v.second.set_unit_type(units.pressure());
 
                for (auto &v : initial_position_)
                    for (int i = 0; i < 3; ++i)
                        v.value[i].set_unit_type(units.length());
 
                for (auto &v : initial_velocity_)
                    for (int i = 0; i < 3; ++i)
                        v.value[i].set_unit_type(units.velocity());
 
                for (auto &v : initial_acceleration_)
                    for (int i = 0; i < 3; ++i)
                        v.value[i].set_unit_type(units.acceleration());
 
                for (auto &v : nodal_dirichlet_)
                    v.second.set_unit_type(units.length());
 
                for (auto &v : nodal_neumann_)
                    v.second.set_unit_type(units.force());
            }
        }
 
        void GenericTensorProblem::rhs(const assembler::Assembler &assembler, const Eigen::MatrixXd &pts, const double t, Eigen::MatrixXd &val, const int fe_space_id) const
        {
            const auto *rhs = find_for_fe_space(rhs_, fe_space_id);
            const auto *rhs_size = find_for_fe_space(rhs_size_, fe_space_id);
            const int size = rhs_size == nullptr ? pts.cols() : *rhs_size;
            val.resize(pts.rows(), size);
 
            if (rhs == nullptr || is_rhs_zero(fe_space_id))
            {
                val.setZero();
                return;
            }
 
            const bool planar = pts.cols() == 2;
            for (int i = 0; i < pts.rows(); ++i)
            {
                for (int j = 0; j < size; ++j)
                {
                    double x = pts(i, 0), y = pts(i, 1), z = planar ? 0 : pts(i, 2);
                    val(i, j) = (*rhs)[j](x, y, z, t);
                }
            }
        }
 
        void GenericTensorProblem::rhs(const assembler::Assembler &assembler, const mesh::Mesh &mesh, const int element_id, const Eigen::MatrixXd &pts, const double t, Eigen::MatrixXd &val, const int fe_space_id) const
        {
            if (body_rhs_.empty())
            {
                rhs(assembler, pts, t, val, fe_space_id);
                return;
            }
 
            int value_size = pts.cols();
            for (const TensorInitialValue &entry : body_rhs_)
            {
                if (matches_fe_space(entry.fe_space_id, fe_space_id))
                {
                    value_size = entry.size;
                    break;
                }
            }
 
            val.resize(pts.rows(), value_size);
            val.setZero();
 
            const int body_id = mesh.get_body_id(element_id);
            const TensorInitialValue *body_rhs = nullptr;
            for (const TensorInitialValue &entry : body_rhs_)
            {
                if (entry.body_id == body_id && matches_fe_space(entry.fe_space_id, fe_space_id))
                {
                    body_rhs = &entry;
                    break;
                }
            }
 
            if (body_rhs == nullptr)
                return;
 
            const bool planar = pts.cols() == 2;
            for (int i = 0; i < pts.rows(); ++i)
            {
                for (int j = 0; j < val.cols(); ++j)
                {
                    const double x = pts(i, 0), y = pts(i, 1), z = planar ? 0 : pts(i, 2);
                    val(i, j) = body_rhs->value[j](x, y, z, t);
                }
            }
        }
 
        bool GenericTensorProblem::is_rhs_zero(const int fe_space_id) const
        {
            const auto *rhs = find_for_fe_space(rhs_, fe_space_id);
            const auto *rhs_size = find_for_fe_space(rhs_size_, fe_space_id);
            if (rhs != nullptr && rhs_size != nullptr)
            {
                for (int i = 0; i < *rhs_size; ++i)
                    if (!(*rhs)[i].is_zero())
                        return false;
            }
 
            for (const TensorInitialValue &entry : body_rhs_)
            {
                if (!matches_fe_space(entry.fe_space_id, fe_space_id))
                    continue;
                for (int i = 0; i < entry.size; ++i)
                    if (!entry.value[i].is_zero())
                        return false;
            }
 
            return true;
        }
 
        bool GenericTensorProblem::has_boundary(const BoundaryKind kind, const int tag, const int fe_space_id)
        {
            if (tag <= 0)
                return false;
 
            if (kind == BoundaryKind::Dirichlet)
            {
                for (size_t i = 0; i < boundary_ids_.size(); ++i)
                    if ((boundary_ids_[i] < 0 || boundary_ids_[i] == tag) && matches_fe_space(displacements_[i].fe_space_id, fe_space_id))
                        return true;
                return false;
            }
 
            for (size_t i = 0; i < neumann_boundary_ids_.size(); ++i)
                if (neumann_boundary_ids_[i] == tag && matches_fe_space(forces_[i].fe_space_id, fe_space_id))
                    return true;
            for (size_t i = 0; i < normal_aligned_neumann_boundary_ids_.size(); ++i)
                if (normal_aligned_neumann_boundary_ids_[i] == tag && matches_fe_space(normal_aligned_forces_[i].fe_space_id, fe_space_id))
                    return true;
            return false;
        }
 
        bool GenericTensorProblem::is_dimension_dirichet(const int tag, const int dim, const int fe_space_id) const
        {
            if (all_dimensions_dirichlet(fe_space_id))
                return true;
 
            for (size_t b = 0; b < boundary_ids_.size(); ++b)
            {
                if ((boundary_ids_[b] < 0 || tag == boundary_ids_[b]) && matches_fe_space(displacements_[b].fe_space_id, fe_space_id))
                {
                    auto &tmp = displacements_[b].dirichlet_dimension;
                    return tmp[dim];
                }
            }
 
            assert(false);
            return true;
        }
 
        bool GenericTensorProblem::all_dimensions_dirichlet(const int fe_space_id) const
        {
            for (const TensorBCValue &displacement : displacements_)
            {
                if (!matches_fe_space(displacement.fe_space_id, fe_space_id))
                    continue;
 
                for (int d = 0; d < displacement.dirichlet_dimension.size(); ++d)
                {
                    if (!displacement.dirichlet_dimension(d))
                        return false;
                }
            }
 
            for (const auto &n_dirichlet : nodal_dirichlet_mat_)
            {
                for (int i = 0; i < n_dirichlet.rows(); ++i)
                {
                    for (int d = 1; d < n_dirichlet.cols(); ++d)
                    {
                        if (std::isnan(n_dirichlet(i, d)))
                            return false;
                    }
                }
            }
 
            return true;
        }
 
        void GenericTensorProblem::dirichlet_bc(const mesh::Mesh &mesh, const Eigen::MatrixXi &global_ids, const Eigen::MatrixXd &uv, const Eigen::MatrixXd &pts, const double t, Eigen::MatrixXd &val, const int fe_space_id) const
        {
            int value_size = mesh.dimension();
            for (const TensorBCValue &displacement : displacements_)
            {
                if (displacement.size > 0 && matches_fe_space(displacement.fe_space_id, fe_space_id))
                {
                    value_size = displacement.size;
                    break;
                }
            }
            val = Eigen::MatrixXd::Zero(pts.rows(), value_size);
 
            for (long i = 0; i < pts.rows(); ++i)
            {
                const int id = mesh.get_boundary_id(global_ids(i));
                for (size_t b = 0; b < boundary_ids_.size(); ++b)
                {
                    if ((boundary_ids_[b] < 0 || id == boundary_ids_[b]) && matches_fe_space(displacements_[b].fe_space_id, fe_space_id))
                    {
                        for (int d = 0; d < std::min<int>(val.cols(), displacements_[b].size); ++d)
                        {
                            val(i, d) = displacements_[b].eval(pts.row(i), d, t);
                        }
                        break;
                    }
                }
            }
        }
 
        void GenericTensorProblem::neumann_bc(const mesh::Mesh &mesh, const Eigen::MatrixXi &global_ids, const Eigen::MatrixXd &uv, const Eigen::MatrixXd &pts, const Eigen::MatrixXd &normals, const double t, Eigen::MatrixXd &val, const int fe_space_id) const
        {
            val = Eigen::MatrixXd::Zero(pts.rows(), mesh.dimension());
 
            for (long i = 0; i < pts.rows(); ++i)
            {
                const int id = mesh.get_boundary_id(global_ids(i));
 
                for (size_t b = 0; b < neumann_boundary_ids_.size(); ++b)
                {
                    if (id == neumann_boundary_ids_[b] && matches_fe_space(forces_[b].fe_space_id, fe_space_id))
                    {
                        for (int d = 0; d < std::min<int>(val.cols(), forces_[b].size); ++d)
                        {
                            val(i, d) = forces_[b].eval(pts.row(i), d, t);
                        }
 
                        break;
                    }
                }
 
                for (size_t b = 0; b < normal_aligned_neumann_boundary_ids_.size(); ++b)
                {
                    if (id == normal_aligned_neumann_boundary_ids_[b] && matches_fe_space(normal_aligned_forces_[b].fe_space_id, fe_space_id))
                    {
                        for (int d = 0; d < val.cols(); ++d)
                        {
                            val(i, d) = normal_aligned_forces_[b].eval(pts.row(i), t) * normals(i, d);
                        }
                        break;
                    }
                }
            }
        }
 
        void GenericTensorProblem::pressure_bc(const mesh::Mesh &mesh, const Eigen::MatrixXi &global_ids, const Eigen::MatrixXd &uv, const Eigen::MatrixXd &pts, const Eigen::MatrixXd &normals, const double t, Eigen::MatrixXd &val) const
        {
            val = Eigen::MatrixXd::Zero(pts.rows(), 1);
 
            for (long i = 0; i < pts.rows(); ++i)
            {
                const int id = mesh.get_boundary_id(global_ids(i));
 
                for (size_t b = 0; b < pressure_boundary_ids_.size(); ++b)
                {
                    if (id == pressure_boundary_ids_[b])
                    {
                        val(i) = pressures_[b].eval(pts.row(i), t);
                        break;
                    }
                }
            }
        }
 
        double GenericTensorProblem::pressure_cavity_bc(const int boundary_id, const double t) const
        {
            Eigen::VectorXd pt;
            pt.setZero(3);
            return cavity_pressures_.at(boundary_id).eval(pt, t);
        }
 
        void GenericTensorProblem::exact(const Eigen::MatrixXd &pts, const double t, Eigen::MatrixXd &val) const
        {
            assert(has_exact_sol());
            const bool planar = pts.cols() == 2;
            val.resize(pts.rows(), pts.cols());
 
            for (int i = 0; i < pts.rows(); ++i)
            {
                for (int j = 0; j < pts.cols(); ++j)
                {
                    double x = pts(i, 0), y = pts(i, 1), z = planar ? 0 : pts(i, 2);
                    val(i, j) = exact_[j](x, y, z, t);
                }
            }
        }
 
        void GenericTensorProblem::exact_grad(const Eigen::MatrixXd &pts, const double t, Eigen::MatrixXd &val) const
        {
            const int size = pts.cols();
            val.resize(pts.rows(), pts.cols() * size);
            if (!has_exact_grad_)
                return;
 
            const bool planar = size == 2;
            for (int i = 0; i < pts.rows(); ++i)
            {
                for (int j = 0; j < pts.cols() * size; ++j)
                {
                    double x = pts(i, 0), y = pts(i, 1), z = planar ? 0 : pts(i, 2);
                    val(i, j) = exact_grad_[j](x, y, z, t);
                }
            }
        }
 
        void GenericTensorProblem::dirichlet_nodal_value(const mesh::Mesh &mesh, const int node_id, const RowVectorNd &pt, const double t, Eigen::MatrixXd &val, const int fe_space_id) const
        {
            val = Eigen::MatrixXd::Zero(1, mesh.dimension());
            const int tag = mesh.get_node_id(node_id);
 
            for (size_t i = 0; i < boundary_ids_.size(); ++i)
            {
                if ((boundary_ids_[i] < 0 || boundary_ids_[i] == tag) && matches_fe_space(displacements_[i].fe_space_id, fe_space_id))
                {
                    val = Eigen::MatrixXd::Zero(1, displacements_[i].size);
                    for (int d = 0; d < val.cols(); ++d)
                        val(d) = displacements_[i].eval(pt, d, t);
                    return;
                }
            }
 
            for (const auto &n_dirichlet : nodal_dirichlet_mat_)
            {
                for (int i = 0; i < n_dirichlet.rows(); ++i)
                {
                    if (n_dirichlet(i, 0) == node_id)
                    {
                        for (int d = 0; d < val.cols(); ++d)
                        {
                            val(d) = n_dirichlet(i, d + 1);
                        }
 
                        return;
                    }
                }
            }
 
            assert(false);
        }
 
        void GenericTensorProblem::neumann_nodal_value(const mesh::Mesh &mesh, const int node_id, const RowVectorNd &pt, const Eigen::MatrixXd &normal, const double t, Eigen::MatrixXd &val, const int fe_space_id) const
        {
            val = Eigen::MatrixXd::Zero(1, mesh.dimension());
            const int tag = mesh.get_node_id(node_id);
 
            if (const auto it = nodal_neumann_.find(tag); it != nodal_neumann_.end() && matches_fe_space(it->second.fe_space_id, fe_space_id))
            {
                val = Eigen::MatrixXd::Zero(1, it->second.size);
                for (int d = 0; d < val.cols(); ++d)
                    val(d) = it->second.eval(pt, d, t);
                return;
            }
 
            for (const auto &n_neumann : nodal_neumann_mat_)
            {
                for (int i = 0; i < n_neumann.rows(); ++i)
                {
                    if (n_neumann(i, 0) == node_id)
                    {
                        val.resize(1, n_neumann.cols() - 1);
                        for (int d = 0; d < val.cols(); ++d)
                            val(d) = n_neumann(i, d + 1);
                        return;
                    }
                }
            }
 
            assert(false);
        }
 
        bool GenericTensorProblem::is_nodal_dirichlet_boundary(const int n_id, const int tag, const int fe_space_id)
        {
            for (size_t i = 0; i < boundary_ids_.size(); ++i)
                if ((boundary_ids_[i] < 0 || boundary_ids_[i] == tag) && matches_fe_space(displacements_[i].fe_space_id, fe_space_id))
                    return true;
 
            for (const auto &n_dirichlet : nodal_dirichlet_mat_)
            {
                for (int i = 0; i < n_dirichlet.rows(); ++i)
                {
                    if (n_dirichlet(i, 0) == n_id)
                        return true;
                }
            }
 
            return false;
        }
 
        bool GenericTensorProblem::is_nodal_neumann_boundary(const int n_id, const int tag, const int fe_space_id)
        {
            if (const auto it = nodal_neumann_.find(tag); it != nodal_neumann_.end())
                return matches_fe_space(it->second.fe_space_id, fe_space_id);
 
            for (const auto &n_neumann : nodal_neumann_mat_)
            {
                for (int i = 0; i < n_neumann.rows(); ++i)
                {
                    if (n_neumann(i, 0) == n_id)
                        return true;
                }
            }
 
            return false;
        }
 
        bool GenericTensorProblem::has_nodal_dirichlet(const int fe_space_id)
        {
            return !nodal_dirichlet_mat_.empty();
        }
 
        bool GenericTensorProblem::has_nodal_neumann(const int fe_space_id)
        {
            if (!nodal_neumann_mat_.empty())
                return true;
 
            for (const auto &[id, force] : nodal_neumann_)
            {
                (void)id;
                if (matches_fe_space(force.fe_space_id, fe_space_id))
                    return true;
            }
            return false;
        }
 
        bool GenericTensorProblem::is_nodal_dimension_dirichlet(const int n_id, const int tag, const int dim, const int fe_space_id) const
        {
            for (size_t i = 0; i < boundary_ids_.size(); ++i)
                if ((boundary_ids_[i] < 0 || boundary_ids_[i] == tag) && matches_fe_space(displacements_[i].fe_space_id, fe_space_id))
                    return displacements_[i].dirichlet_dimension(dim);
 
            for (const auto &n_dirichlet : nodal_dirichlet_mat_)
            {
                for (int i = 0; i < n_dirichlet.rows(); ++i)
                {
                    if (n_dirichlet(i, 0) == n_id)
                    {
                        return !std::isnan(n_dirichlet(i, dim + 1));
                    }
                }
            }
 
            assert(false);
            return true;
        }
 
        void GenericTensorProblem::update_nodes(const Eigen::VectorXi &in_node_to_node)
        {
            if (updated_dirichlet_node_ordering_)
            {
                if (!nodal_dirichlet_mat_.empty() || !nodal_neumann_mat_.empty())
                    logger().debug("Skipping updating in nodes to nodes in problem, already done once...");
                return;
            }
            for (auto &n_dirichlet : nodal_dirichlet_mat_)
            {
                for (int n = 0; n < n_dirichlet.rows(); ++n)
                {
                    const int node_id = in_node_to_node[n_dirichlet(n, 0)];
                    n_dirichlet(n, 0) = node_id;
                }
            }
            for (auto &n_neumann : nodal_neumann_mat_)
            {
                for (int n = 0; n < n_neumann.rows(); ++n)
                {
                    const int node_id = in_node_to_node[n_neumann(n, 0)];
                    n_neumann(n, 0) = node_id;
                }
            }
            updated_dirichlet_node_ordering_ = true;
        }
 
        void GenericTensorProblem::set_parameters(const json &params, const std::string &root_path)
        {
            if (is_param_valid(params, "is_time_dependent"))
            {
                is_time_dept_ = params["is_time_dependent"];
            }
 
            if (is_param_valid(params, "rhs"))
            {
                const json &rr = params["rhs"];
                const bool has_fe_spaces = rr.is_array() && !rr.empty() && rr.front().is_object() && rr.front().contains("fe_space") && rr.front().contains("value");
                const bool has_body_ids = has_body_value_entries(rr);
                if (has_body_ids)
                {
                    const std::vector<json> entries = flatten_ids(rr);
                    body_rhs_.resize(entries.size());
                    for (size_t k = 0; k < entries.size(); ++k)
                    {
                        body_rhs_[k].body_id = entries[k]["id"];
                        body_rhs_[k].fe_space_id = fe_space_id(entries[k]);
                        const auto &value = entries[k]["value"];
                        body_rhs_[k].size = value.is_array() ? int(value.size()) : 1;
                        if (body_rhs_[k].size > 3)
                            log_and_throw_error("RHS for body {} and FE space {} has {} components; at most 3 are supported.", body_rhs_[k].body_id, body_rhs_[k].fe_space_id, body_rhs_[k].size);
                        for (int d = 0; d < body_rhs_[k].size; ++d)
                            body_rhs_[k].value[d].init(value.is_array() ? value[d] : value, root_path);
                    }
                }
                else if (has_fe_spaces)
                {
                    for (const json &entry : rr)
                    {
                        const int id = fe_space_id(entry);
                        const json &value = entry["value"];
                        const int size = value.is_array() ? int(value.size()) : 1;
                        if (size > 3)
                            log_and_throw_error("RHS for FE space {} has {} components; at most 3 are supported.", id, size);
                        rhs_size_[id] = size;
                        for (int k = 0; k < size; ++k)
                            rhs_[id][k].init(value.is_array() ? value[k] : value, root_path);
                    }
                }
                else if (rr.is_array() && !rr.empty())
                {
                    if (rr.size() > 3)
                        log_and_throw_error("RHS has {} components; at most 3 are supported.", rr.size());
                    rhs_size_[-1] = int(rr.size());
                    for (size_t k = 0; k < rr.size(); ++k)
                        rhs_[-1][k].init(rr[k], root_path);
                }
                else if (!rr.is_array())
                    log_and_throw_error("Invalid tensor problem RHS: expected an array.");
            }
 
            if (is_param_valid(params, "reference") && is_param_valid(params["reference"], "solution"))
            {
                auto ex = params["reference"]["solution"];
                has_exact_ = ex.size() > 0;
                if (ex.is_array())
                {
                    for (size_t k = 0; k < ex.size(); ++k)
                        exact_[k].init(ex[k], root_path);
                }
                else
                {
                    assert(false);
                }
            }
 
            if (is_param_valid(params, "reference") && is_param_valid(params["reference"], "gradient"))
            {
                auto ex = params["reference"]["gradient"];
                has_exact_grad_ = ex.size() > 0;
                if (ex.is_array())
                {
                    for (size_t k = 0; k < ex.size(); ++k)
                        exact_grad_[k].init(ex[k], root_path);
                }
                else
                {
                    assert(false);
                }
            }
 
            if (is_param_valid(params, "dirichlet_boundary"))
            {
                // boundary_ids_.clear();
                int offset = boundary_ids_.size();
                std::vector<json> j_boundary = flatten_ids(params["dirichlet_boundary"]);
 
                boundary_ids_.resize(offset + j_boundary.size());
                displacements_.resize(offset + j_boundary.size());
 
                for (size_t i = offset; i < boundary_ids_.size(); ++i)
                {
                    if (j_boundary[i - offset].is_string())
                    {
                        const std::string path = resolve_path(j_boundary[i - offset], params["root_path"]);
                        if (!std::filesystem::is_regular_file(path))
                            log_and_throw_error("unable to open {} file", path);
 
                        Eigen::MatrixXd tmp;
                        io::read_matrix(path, tmp);
                        nodal_dirichlet_mat_.emplace_back(tmp);
 
                        continue;
                    }
 
                    int current_id = -1;
                    displacements_[i].fe_space_id = fe_space_id(j_boundary[i - offset]);
 
                    if (j_boundary[i - offset]["id"] == "all")
                    {
                        boundary_ids_[i] = -1;
                        is_all_ = true;
                        nodal_dirichlet_[current_id] = TensorBCValue();
                    }
                    else
                    {
                        boundary_ids_[i] = j_boundary[i - offset]["id"];
                        current_id = boundary_ids_[i];
                        nodal_dirichlet_[current_id] = TensorBCValue();
                    }
                    nodal_dirichlet_[current_id].fe_space_id = displacements_[i].fe_space_id;
 
                    auto ff = j_boundary[i - offset]["value"];
                    if (ff.is_array())
                    {
                        if (ff.size() > 3)
                            log_and_throw_error("Dirichlet condition for FE space {} has {} components; at most 3 are supported.", displacements_[i].fe_space_id, ff.size());
                        displacements_[i].size = int(ff.size());
                        for (size_t k = 0; k < ff.size(); ++k)
                        {
                            displacements_[i].value[k].init(ff[k], root_path);
                            if (j_boundary[i - offset].contains("time_reference") && j_boundary[i - offset]["time_reference"].size() > 0)
                                displacements_[i].value[k].set_t(j_boundary[i - offset]["time_reference"]);
                            nodal_dirichlet_[current_id].value[k].init(ff[k], root_path);
                        }
                    }
                    else
                    {
                        displacements_[i].size = 1;
                        displacements_[i].value[0].init(ff, root_path);
                        nodal_dirichlet_[current_id].value[0].init(ff, root_path);
                    }
                    nodal_dirichlet_[current_id].size = displacements_[i].size;
 
                    displacements_[i].dirichlet_dimension.setConstant(true);
                    nodal_dirichlet_[current_id].dirichlet_dimension.setConstant(true);
                    if (j_boundary[i - offset].contains("dimension"))
                    {
                        auto &tmp = j_boundary[i - offset]["dimension"];
                        assert(tmp.is_array());
                        for (size_t k = 0; k < tmp.size(); ++k)
                        {
                            displacements_[i].dirichlet_dimension[k] = tmp[k];
                            nodal_dirichlet_[current_id].dirichlet_dimension[k] = tmp[k];
                        }
                    }
 
                    if (j_boundary[i - offset]["interpolation"].is_array())
                    {
                        for (int ii = 0; ii < j_boundary[i - offset]["interpolation"].size(); ++ii)
                            displacements_[i].interpolation.push_back(Interpolation::build(j_boundary[i - offset]["interpolation"][ii]));
                    }
                    else
                        displacements_[i].interpolation.push_back(Interpolation::build(j_boundary[i - offset]["interpolation"]));
 
                    nodal_dirichlet_[current_id].interpolation = displacements_[i].interpolation;
                }
            }
 
            if (is_param_valid(params, "neumann_boundary"))
            {
                // neumann_boundary_ids_.clear();
                const int offset = neumann_boundary_ids_.size();
 
                auto j_boundary_tmp = params["neumann_boundary"];
                std::vector<json> j_boundary = flatten_ids(j_boundary_tmp);
 
                neumann_boundary_ids_.resize(offset + j_boundary.size());
                forces_.resize(offset + j_boundary.size());
 
                for (size_t i = offset; i < neumann_boundary_ids_.size(); ++i)
                {
                    neumann_boundary_ids_[i] = j_boundary[i - offset]["id"];
                    forces_[i].fe_space_id = fe_space_id(j_boundary[i - offset]);
 
                    auto ff = j_boundary[i - offset]["value"];
                    forces_[i].size = ff.is_array() ? int(ff.size()) : 1;
                    if (forces_[i].size > 3)
                        log_and_throw_error("Neumann condition for FE space {} has {} components; at most 3 are supported.", forces_[i].fe_space_id, forces_[i].size);
                    for (int k = 0; k < forces_[i].size; ++k)
                        forces_[i].value[k].init(ff.is_array() ? ff[k] : ff, root_path);
 
                    if (j_boundary[i - offset]["interpolation"].is_array())
                    {
                        for (int ii = 0; ii < j_boundary[i - offset]["interpolation"].size(); ++ii)
                            forces_[i].interpolation.push_back(Interpolation::build(j_boundary[i - offset]["interpolation"][ii]));
                    }
                    else
                    {
                        forces_[i].interpolation.push_back(Interpolation::build(j_boundary[i - offset]["interpolation"]));
                    }
                }
            }
 
            if (is_param_valid(params, "nodal_neumann_boundary"))
            {
                std::vector<json> j_boundary = flatten_ids(params["nodal_neumann_boundary"]);
 
                for (size_t i = 0; i < j_boundary.size(); ++i)
                {
                    if (j_boundary[i].is_string())
                    {
                        const std::string path = resolve_path(j_boundary[i], params["root_path"]);
                        if (!std::filesystem::is_regular_file(path))
                            log_and_throw_error("unable to open {} file", path);
 
                        Eigen::MatrixXd tmp;
                        io::read_matrix(path, tmp);
                        nodal_neumann_mat_.emplace_back(tmp);
 
                        continue;
                    }
 
                    const int id = j_boundary[i]["id"];
                    TensorBCValue nodal_neumann;
                    nodal_neumann.fe_space_id = fe_space_id(j_boundary[i]);
 
                    auto ff = j_boundary[i]["value"];
                    nodal_neumann.size = ff.is_array() ? int(ff.size()) : 1;
                    if (nodal_neumann.size > 3)
                        log_and_throw_error("Nodal Neumann condition for FE space {} has {} components; at most 3 are supported.", nodal_neumann.fe_space_id, nodal_neumann.size);
                    for (int k = 0; k < nodal_neumann.size; ++k)
                        nodal_neumann.value[k].init(ff.is_array() ? ff[k] : ff, root_path);
 
                    if (j_boundary[i]["interpolation"].is_array())
                    {
                        for (int ii = 0; ii < j_boundary[i]["interpolation"].size(); ++ii)
                            nodal_neumann.interpolation.push_back(Interpolation::build(j_boundary[i]["interpolation"][ii]));
                    }
                    else
                    {
                        nodal_neumann.interpolation.push_back(Interpolation::build(j_boundary[i]["interpolation"]));
                    }
 
                    nodal_neumann_[id] = nodal_neumann;
                }
            }
 
            if (is_param_valid(params, "normal_aligned_neumann_boundary"))
            {
                const int offset = normal_aligned_neumann_boundary_ids_.size();
 
                auto j_boundary_tmp = params["normal_aligned_neumann_boundary"];
                std::vector<json> j_boundary = flatten_ids(j_boundary_tmp);
 
                normal_aligned_neumann_boundary_ids_.resize(offset + j_boundary.size());
                normal_aligned_forces_.resize(offset + j_boundary.size());
 
                for (size_t i = offset; i < normal_aligned_neumann_boundary_ids_.size(); ++i)
                {
                    normal_aligned_neumann_boundary_ids_[i] = j_boundary[i - offset]["id"];
                    normal_aligned_forces_[i].fe_space_id = fe_space_id(j_boundary[i - offset]);
 
                    auto ff = j_boundary[i - offset]["value"];
                    normal_aligned_forces_[i].value.init(ff, root_path);
 
                    if (j_boundary[i - offset].contains("interpolation"))
                        normal_aligned_forces_[i].interpolation = Interpolation::build(j_boundary[i - offset]["interpolation"]);
                    else
                        normal_aligned_forces_[i].interpolation = std::make_shared<NoInterpolation>();
                }
            }
 
            if (is_param_valid(params, "pressure_boundary"))
            {
                // pressure_boundary_ids_.clear();
                const int offset = pressure_boundary_ids_.size();
 
                auto j_boundary_tmp = params["pressure_boundary"];
                std::vector<json> j_boundary = flatten_ids(j_boundary_tmp);
 
                pressure_boundary_ids_.resize(offset + j_boundary.size());
                pressures_.resize(offset + j_boundary.size());
 
                for (size_t i = offset; i < pressure_boundary_ids_.size(); ++i)
                {
                    pressure_boundary_ids_[i] = j_boundary[i - offset]["id"];
 
                    auto ff = j_boundary[i - offset]["value"];
                    pressures_[i].value.init(ff, root_path);
                    if (j_boundary[i - offset].contains("time_reference") && j_boundary[i - offset]["time_reference"].size() > 0)
                        pressures_[i].value.set_t(j_boundary[i - offset]["time_reference"]);
 
                    pressures_[i].interpolation = std::make_shared<NoInterpolation>();
                }
            }
 
            if (is_param_valid(params, "pressure_cavity"))
            {
                const int offset = pressure_cavity_ids_.size();
 
                auto j_boundary_tmp = params["pressure_cavity"];
                std::vector<json> j_boundary = flatten_ids(j_boundary_tmp);
 
                pressure_cavity_ids_.resize(offset + j_boundary.size());
 
                for (size_t i = offset; i < pressure_cavity_ids_.size(); ++i)
                {
                    int boundary_id = j_boundary[i - offset]["id"];
                    pressure_cavity_ids_[i] = boundary_id;
 
                    if (cavity_pressures_.find(boundary_id) == cavity_pressures_.end())
                    {
                        cavity_pressures_[boundary_id] = ScalarBCValue();
 
                        auto ff = j_boundary[i - offset]["value"];
                        cavity_pressures_[boundary_id].value.init(ff, root_path);
 
                        cavity_pressures_[boundary_id].interpolation = std::make_shared<NoInterpolation>();
                    }
                }
            }
 
            if (is_param_valid(params, "solution"))
            {
                auto rr = params["solution"];
                initial_position_.resize(rr.size());
                assert(rr.is_array());
 
                for (size_t k = 0; k < rr.size(); ++k)
                {
                    initial_position_[k].body_id = rr[k]["id"];
                    initial_position_[k].fe_space_id = fe_space_id(rr[k]);
                    const auto v = rr[k]["value"];
                    initial_position_[k].size = v.is_array() ? int(v.size()) : 1;
                    if (initial_position_[k].size > 3)
                        log_and_throw_error("Initial solution for FE space {} has {} components; at most 3 are supported.", initial_position_[k].fe_space_id, initial_position_[k].size);
                    for (int d = 0; d < initial_position_[k].size; ++d)
                        initial_position_[k].value[d].init(v.is_array() ? v[d] : v, root_path);
                }
            }
 
            if (is_param_valid(params, "velocity"))
            {
                auto rr = params["velocity"];
                initial_velocity_.resize(rr.size());
                assert(rr.is_array());
 
                for (size_t k = 0; k < rr.size(); ++k)
                {
                    initial_velocity_[k].body_id = rr[k]["id"];
                    initial_velocity_[k].fe_space_id = fe_space_id(rr[k]);
                    const auto v = rr[k]["value"];
                    initial_velocity_[k].size = v.is_array() ? int(v.size()) : 1;
                    if (initial_velocity_[k].size > 3)
                        log_and_throw_error("Initial velocity for FE space {} has {} components; at most 3 are supported.", initial_velocity_[k].fe_space_id, initial_velocity_[k].size);
                    for (int d = 0; d < initial_velocity_[k].size; ++d)
                        initial_velocity_[k].value[d].init(v.is_array() ? v[d] : v, root_path);
                }
            }
 
            if (is_param_valid(params, "acceleration"))
            {
                auto rr = params["acceleration"];
                initial_acceleration_.resize(rr.size());
                assert(rr.is_array());
 
                for (size_t k = 0; k < rr.size(); ++k)
                {
                    initial_acceleration_[k].body_id = rr[k]["id"];
                    initial_acceleration_[k].fe_space_id = fe_space_id(rr[k]);
                    const auto v = rr[k]["value"];
                    initial_acceleration_[k].size = v.is_array() ? int(v.size()) : 1;
                    if (initial_acceleration_[k].size > 3)
                        log_and_throw_error("Initial acceleration for FE space {} has {} components; at most 3 are supported.", initial_acceleration_[k].fe_space_id, initial_acceleration_[k].size);
                    for (int d = 0; d < initial_acceleration_[k].size; ++d)
                        initial_acceleration_[k].value[d].init(v.is_array() ? v[d] : v, root_path);
                }
            }
        }
 
        void GenericTensorProblem::initial_solution(const mesh::Mesh &mesh, const Eigen::MatrixXi &global_ids, const Eigen::MatrixXd &pts, Eigen::MatrixXd &val, const int fe_space_id) const
        {
            int value_size = pts.cols();
            for (const TensorInitialValue &entry : initial_position_)
                if (matches_fe_space(entry.fe_space_id, fe_space_id))
                {
                    value_size = entry.size;
                    break;
                }
            val.resize(pts.rows(), value_size);
            if (initial_position_.empty())
            {
                val.setZero();
                return;
            }
 
            const bool planar = pts.cols() == 2;
            for (int i = 0; i < pts.rows(); ++i)
            {
                const int id = mesh.get_body_id(global_ids(i));
                int index = -1;
                for (int j = 0; j < initial_position_.size(); ++j)
                {
                    if (initial_position_[j].body_id == id && matches_fe_space(initial_position_[j].fe_space_id, fe_space_id))
                    {
                        index = j;
                        break;
                    }
                }
                if (index < 0)
                {
                    val.row(i).setZero();
                    continue;
                }
 
                for (int j = 0; j < val.cols(); ++j)
                    val(i, j) = planar ? initial_position_[index].value[j](pts(i, 0), pts(i, 1)) : initial_position_[index].value[j](pts(i, 0), pts(i, 1), pts(i, 2));
            }
        }
 
        void GenericTensorProblem::initial_velocity(const mesh::Mesh &mesh, const Eigen::MatrixXi &global_ids, const Eigen::MatrixXd &pts, Eigen::MatrixXd &val, const int fe_space_id) const
        {
            int value_size = pts.cols();
            for (const TensorInitialValue &entry : initial_velocity_)
                if (matches_fe_space(entry.fe_space_id, fe_space_id))
                {
                    value_size = entry.size;
                    break;
                }
            val.resize(pts.rows(), value_size);
            if (initial_velocity_.empty())
            {
                val.setZero();
                return;
            }
 
            const bool planar = pts.cols() == 2;
            for (int i = 0; i < pts.rows(); ++i)
            {
                const int id = mesh.get_body_id(global_ids(i));
                int index = -1;
                for (int j = 0; j < initial_velocity_.size(); ++j)
                {
                    if (initial_velocity_[j].body_id == id && matches_fe_space(initial_velocity_[j].fe_space_id, fe_space_id))
                    {
                        index = j;
                        break;
                    }
                }
                if (index < 0)
                {
                    val.row(i).setZero();
                    continue;
                }
 
                for (int j = 0; j < val.cols(); ++j)
                    val(i, j) = planar ? initial_velocity_[index].value[j](pts(i, 0), pts(i, 1)) : initial_velocity_[index].value[j](pts(i, 0), pts(i, 1), pts(i, 2));
            }
        }
 
        void GenericTensorProblem::initial_acceleration(const mesh::Mesh &mesh, const Eigen::MatrixXi &global_ids, const Eigen::MatrixXd &pts, Eigen::MatrixXd &val, const int fe_space_id) const
        {
            int value_size = pts.cols();
            for (const TensorInitialValue &entry : initial_acceleration_)
                if (matches_fe_space(entry.fe_space_id, fe_space_id))
                {
                    value_size = entry.size;
                    break;
                }
            val.resize(pts.rows(), value_size);
            if (initial_acceleration_.empty())
            {
                val.setZero();
                return;
            }
 
            const bool planar = pts.cols() == 2;
            for (int i = 0; i < pts.rows(); ++i)
            {
                const int id = mesh.get_body_id(global_ids(i));
                int index = -1;
                for (int j = 0; j < initial_acceleration_.size(); ++j)
                {
                    if (initial_acceleration_[j].body_id == id && matches_fe_space(initial_acceleration_[j].fe_space_id, fe_space_id))
                    {
                        index = j;
                        break;
                    }
                }
                if (index < 0)
                {
                    val.row(i).setZero();
                    continue;
                }
 
                for (int j = 0; j < val.cols(); ++j)
                    val(i, j) = planar ? initial_acceleration_[index].value[j](pts(i, 0), pts(i, 1)) : initial_acceleration_[index].value[j](pts(i, 0), pts(i, 1), pts(i, 2));
            }
        }
 
        void GenericTensorProblem::clear()
        {
            has_exact_ = false;
            has_exact_grad_ = false;
            is_time_dept_ = false;
 
            forces_.clear();
            displacements_.clear();
            normal_aligned_forces_.clear();
            pressures_.clear();
            cavity_pressures_.clear();
 
            nodal_dirichlet_.clear();
            nodal_neumann_.clear();
            nodal_dirichlet_mat_.clear();
            nodal_neumann_mat_.clear();
 
            initial_position_.clear();
            initial_velocity_.clear();
            initial_acceleration_.clear();
 
            rhs_.clear();
            rhs_size_.clear();
            body_rhs_.clear();
            for (int i = 0; i < exact_.size(); ++i)
                exact_[i].clear();
            for (int i = 0; i < exact_grad_.size(); ++i)
                exact_grad_[i].clear();
            is_all_ = false;
            updated_dirichlet_node_ordering_ = false;
        }
 
        GenericScalarProblem::GenericScalarProblem(const std::string &name)
            : Problem(name), is_all_(false)
        {
        }
 
        void GenericScalarProblem::set_units(const assembler::Assembler &assembler, const Units &units)
        {
            // TODO?
 
            for (auto &v : neumann_)
                v.set_unit_type("");
 
            for (auto &v : dirichlet_)
                v.set_unit_type("");
 
            for (auto &v : nodal_dirichlet_)
                v.second.set_unit_type("");
 
            for (auto &v : nodal_neumann_)
                v.second.set_unit_type("");
 
            for (auto &v : initial_solution_)
                v.value.set_unit_type("");
 
            for (auto &[fe_space_id, rhs] : rhs_)
                rhs.set_unit_type("");
            exact_.set_unit_type("");
 
            for (int i = 0; i < 3; ++i)
                exact_grad_[i].set_unit_type("");
        }
 
        void GenericScalarProblem::rhs(const assembler::Assembler &assembler, const Eigen::MatrixXd &pts, const double t, Eigen::MatrixXd &val, const int fe_space_id) const
        {
            val.resize(pts.rows(), 1);
            const auto *rhs = find_for_fe_space(rhs_, fe_space_id);
            if (rhs == nullptr || rhs->is_zero())
            {
                val.setZero();
                return;
            }
            const bool planar = pts.cols() == 2;
            for (int i = 0; i < pts.rows(); ++i)
            {
                double x = pts(i, 0), y = pts(i, 1), z = planar ? 0 : pts(i, 2);
                val(i) = (*rhs)(x, y, z, t);
            }
        }
 
        void GenericScalarProblem::rhs(const assembler::Assembler &assembler, const mesh::Mesh &mesh, const int element_id, const Eigen::MatrixXd &pts, const double t, Eigen::MatrixXd &val, const int fe_space_id) const
        {
            if (body_rhs_.empty())
            {
                rhs(assembler, pts, t, val, fe_space_id);
                return;
            }
 
            val.resize(pts.rows(), 1);
            val.setZero();
 
            const int body_id = mesh.get_body_id(element_id);
            const ScalarInitialValue *body_rhs = nullptr;
            for (const ScalarInitialValue &entry : body_rhs_)
            {
                if (entry.body_id == body_id && matches_fe_space(entry.fe_space_id, fe_space_id))
                {
                    body_rhs = &entry;
                    break;
                }
            }
 
            if (body_rhs == nullptr)
                return;
 
            const bool planar = pts.cols() == 2;
            for (int i = 0; i < pts.rows(); ++i)
            {
                const double x = pts(i, 0), y = pts(i, 1), z = planar ? 0 : pts(i, 2);
                val(i) = body_rhs->value(x, y, z, t);
            }
        }
 
        bool GenericScalarProblem::is_rhs_zero(const int fe_space_id) const
        {
            const auto *rhs = find_for_fe_space(rhs_, fe_space_id);
            if (rhs != nullptr && !rhs->is_zero())
                return false;
 
            for (const ScalarInitialValue &entry : body_rhs_)
                if (matches_fe_space(entry.fe_space_id, fe_space_id) && !entry.value.is_zero())
                    return false;
 
            return true;
        }
 
        bool GenericScalarProblem::has_boundary(const BoundaryKind kind, const int tag, const int fe_space_id)
        {
            if (tag <= 0)
                return false;
            if (kind == BoundaryKind::Dirichlet)
            {
                for (size_t i = 0; i < boundary_ids_.size(); ++i)
                    if ((boundary_ids_[i] < 0 || boundary_ids_[i] == tag) && matches_fe_space(dirichlet_[i].fe_space_id, fe_space_id))
                        return true;
                return false;
            }
            for (size_t i = 0; i < neumann_boundary_ids_.size(); ++i)
                if (neumann_boundary_ids_[i] == tag && matches_fe_space(neumann_[i].fe_space_id, fe_space_id))
                    return true;
            return false;
        }
 
        void GenericScalarProblem::dirichlet_bc(const mesh::Mesh &mesh, const Eigen::MatrixXi &global_ids, const Eigen::MatrixXd &uv, const Eigen::MatrixXd &pts, const double t, Eigen::MatrixXd &val, const int fe_space_id) const
        {
            val = Eigen::MatrixXd::Zero(pts.rows(), 1);
 
            for (long i = 0; i < pts.rows(); ++i)
            {
                const int id = mesh.get_boundary_id(global_ids(i));
                for (size_t b = 0; b < boundary_ids_.size(); ++b)
                {
                    if ((boundary_ids_[b] < 0 || id == boundary_ids_[b]) && matches_fe_space(dirichlet_[b].fe_space_id, fe_space_id))
                    {
                        val(i) = dirichlet_[b].eval(pts.row(i), t);
                        break;
                    }
                }
            }
        }
 
        void GenericScalarProblem::neumann_bc(const mesh::Mesh &mesh, const Eigen::MatrixXi &global_ids, const Eigen::MatrixXd &uv, const Eigen::MatrixXd &pts, const Eigen::MatrixXd &normals, const double t, Eigen::MatrixXd &val, const int fe_space_id) const
        {
            val = Eigen::MatrixXd::Zero(pts.rows(), 1);
 
            for (long i = 0; i < pts.rows(); ++i)
            {
                const int id = mesh.get_boundary_id(global_ids(i));
 
                for (size_t b = 0; b < neumann_boundary_ids_.size(); ++b)
                {
                    if (id == neumann_boundary_ids_[b] && matches_fe_space(neumann_[b].fe_space_id, fe_space_id))
                    {
                        double x = pts(i, 0), y = pts(i, 1), z = pts.cols() == 2 ? 0 : pts(i, 2);
                        val(i) = neumann_[b].eval(pts.row(i), t);
                        break;
                    }
                }
            }
        }
 
        void GenericScalarProblem::initial_solution(const mesh::Mesh &mesh, const Eigen::MatrixXi &global_ids, const Eigen::MatrixXd &pts, Eigen::MatrixXd &val, const int fe_space_id) const
        {
            val.resize(pts.rows(), 1);
            if (initial_solution_.empty())
            {
                val.setZero();
                return;
            }
 
            const bool planar = pts.cols() == 2;
            for (int i = 0; i < pts.rows(); ++i)
            {
                const int id = mesh.get_body_id(global_ids(i));
                int index = -1;
                for (int j = 0; j < initial_solution_.size(); ++j)
                {
                    if (initial_solution_[j].body_id == id && matches_fe_space(initial_solution_[j].fe_space_id, fe_space_id))
                    {
                        index = j;
                        break;
                    }
                }
                if (index < 0)
                {
                    val(i) = 0;
                    continue;
                }
 
                val(i) = planar ? initial_solution_[index].value(pts(i, 0), pts(i, 1)) : initial_solution_[index].value(pts(i, 0), pts(i, 1), pts(i, 2));
            }
        }
 
        void GenericScalarProblem::exact(const Eigen::MatrixXd &pts, const double t, Eigen::MatrixXd &val) const
        {
            assert(has_exact_sol());
            const bool planar = pts.cols() == 2;
            val.resize(pts.rows(), 1);
 
            for (int i = 0; i < pts.rows(); ++i)
            {
                double x = pts(i, 0), y = pts(i, 1), z = pts.cols() == 2 ? 0 : pts(i, 2);
                val(i) = exact_(x, y, z, t);
            }
        }
 
        void GenericScalarProblem::exact_grad(const Eigen::MatrixXd &pts, const double t, Eigen::MatrixXd &val) const
        {
            val.resize(pts.rows(), pts.cols());
            if (!has_exact_grad_)
                return;
 
            const bool planar = pts.cols() == 2;
            for (int i = 0; i < pts.rows(); ++i)
            {
                for (int j = 0; j < pts.cols(); ++j)
                {
                    double x = pts(i, 0), y = pts(i, 1), z = pts.cols() == 2 ? 0 : pts(i, 2);
                    val(i, j) = exact_grad_[j](x, y, z, t);
                }
            }
        }
 
        void GenericTensorProblem::update_pressure_boundary(const int id, const int time_step, const double val)
        {
            int index = -1;
            for (int i = 0; i < pressure_boundary_ids_.size(); ++i)
            {
                if (pressure_boundary_ids_[i] == id)
                {
                    index = i;
                    break;
                }
            }
            if (index == -1)
            {
                throw "Invalid boundary id for pressure update";
            }
 
            if (pressures_[index].value.is_mat())
            {
                Eigen::MatrixXd curr_val = pressures_[index].value.get_mat();
                assert(time_step <= curr_val.size());
                assert(curr_val.cols() == 1);
                curr_val(time_step) = val;
                pressures_[index].value.set_mat(curr_val);
            }
            else
            {
                pressures_[index].value.init(val);
            }
        }
 
        void GenericTensorProblem::update_dirichlet_boundary(const int id, const int time_step, const Eigen::VectorXd &val)
        {
            int index = -1;
            for (int i = 0; i < boundary_ids_.size(); ++i)
            {
                if (boundary_ids_[i] == id)
                {
                    index = i;
                    break;
                }
            }
            if (index == -1)
            {
                throw "Invalid boundary id for dirichlet update";
            }
 
            for (int i = 0; i < val.size(); ++i)
            {
                if (displacements_[index].value[i].is_mat())
                {
                    Eigen::MatrixXd curr_val = displacements_[index].value[i].get_mat();
                    assert(time_step <= curr_val.size());
                    assert(curr_val.cols() == 1);
                    curr_val(time_step) = val(i);
                    displacements_[index].value[i].set_mat(curr_val);
                }
                else
                {
                    displacements_[index].value[i].init(val(i));
                }
            }
        }
 
        void GenericTensorProblem::update_dirichlet_nodes(const Eigen::VectorXi &in_node_to_node, const Eigen::VectorXi &node_ids, const Eigen::MatrixXd &nodal_dirichlet)
        {
            assert(node_ids.size() == nodal_dirichlet.rows());
            // NOTE!!! Update nodes called in `State::build_basis()` so the first row of this mat
            // must be set to input node ordering if `build_basis` will be called, like in optimization
            for (auto &n_dirichlet : nodal_dirichlet_mat_)
                for (int i = 0; i < node_ids.size(); ++i)
                {
                    int mapped_node_id = in_node_to_node(node_ids(i));
                    for (int j = 0; j < n_dirichlet.rows(); ++j)
                        if (mapped_node_id == n_dirichlet(j, 0))
                            for (int k = 0; k < n_dirichlet.cols() - 1; ++k)
                                n_dirichlet(j, k + 1) = nodal_dirichlet(i, k);
                }
        }
 
        void GenericScalarProblem::dirichlet_nodal_value(const mesh::Mesh &mesh, const int node_id, const RowVectorNd &pt, const double t, Eigen::MatrixXd &val, const int fe_space_id) const
        {
            val = Eigen::MatrixXd::Zero(1, 1);
            const int tag = mesh.get_node_id(node_id);
 
            for (size_t i = 0; i < boundary_ids_.size(); ++i)
            {
                if ((boundary_ids_[i] < 0 || boundary_ids_[i] == tag) && matches_fe_space(dirichlet_[i].fe_space_id, fe_space_id))
                {
                    val(0) = dirichlet_[i].eval(pt, t);
                    return;
                }
            }
 
            for (const auto &n_dirichlet : nodal_dirichlet_mat_)
            {
                for (int i = 0; i < n_dirichlet.rows(); ++i)
                {
                    if (n_dirichlet(i, 0) == node_id)
                    {
                        val(0) = n_dirichlet(i, 1);
                        return;
                    }
                }
            }
 
            assert(false);
        }
 
        void GenericScalarProblem::neumann_nodal_value(const mesh::Mesh &mesh, const int node_id, const RowVectorNd &pt, const Eigen::MatrixXd &normal, const double t, Eigen::MatrixXd &val, const int fe_space_id) const
        {
            val = Eigen::MatrixXd::Zero(1, 1);
            const int tag = mesh.get_node_id(node_id);
 
            if (const auto it = nodal_neumann_.find(tag); it != nodal_neumann_.end() && matches_fe_space(it->second.fe_space_id, fe_space_id))
            {
                val(0) = it->second.eval(pt, t);
                return;
            }
 
            for (const auto &n_neumann : nodal_neumann_mat_)
            {
                for (int i = 0; i < n_neumann.rows(); ++i)
                {
                    if (n_neumann(i, 0) == node_id)
                    {
                        val(0) = n_neumann(i, 1);
                        return;
                    }
                }
            }
 
            assert(false);
        }
 
        bool GenericScalarProblem::is_nodal_dirichlet_boundary(const int n_id, const int tag, const int fe_space_id)
        {
            for (size_t i = 0; i < boundary_ids_.size(); ++i)
                if ((boundary_ids_[i] < 0 || boundary_ids_[i] == tag) && matches_fe_space(dirichlet_[i].fe_space_id, fe_space_id))
                    return true;
 
            for (const auto &n_dirichlet : nodal_dirichlet_mat_)
            {
                for (int i = 0; i < n_dirichlet.rows(); ++i)
                {
                    if (n_dirichlet(i, 0) == n_id)
                        return true;
                }
            }
 
            return false;
        }
 
        bool GenericScalarProblem::is_nodal_neumann_boundary(const int n_id, const int tag, const int fe_space_id)
        {
            if (const auto it = nodal_neumann_.find(tag); it != nodal_neumann_.end())
                return matches_fe_space(it->second.fe_space_id, fe_space_id);
 
            for (const auto &n_neumann : nodal_neumann_mat_)
            {
                for (int i = 0; i < n_neumann.rows(); ++i)
                {
                    if (n_neumann(i, 0) == n_id)
                        return true;
                }
            }
 
            return false;
        }
 
        bool GenericScalarProblem::has_nodal_dirichlet(const int fe_space_id)
        {
            return nodal_dirichlet_mat_.size() > 0;
        }
 
        bool GenericScalarProblem::has_nodal_neumann(const int fe_space_id)
        {
            if (!nodal_neumann_mat_.empty())
                return true;
 
            for (const auto &[id, neumann] : nodal_neumann_)
            {
                (void)id;
                if (matches_fe_space(neumann.fe_space_id, fe_space_id))
                    return true;
            }
            return false;
        }
 
        void GenericScalarProblem::update_nodes(const Eigen::VectorXi &in_node_to_node)
        {
            if (updated_dirichlet_node_ordering_)
                return;
            for (auto &n_dirichlet : nodal_dirichlet_mat_)
            {
                for (int n = 0; n < n_dirichlet.rows(); ++n)
                {
                    const int node_id = in_node_to_node[n_dirichlet(n, 0)];
                    n_dirichlet(n, 0) = node_id;
                }
            }
            for (auto &n_neumann : nodal_neumann_mat_)
            {
                for (int n = 0; n < n_neumann.rows(); ++n)
                {
                    const int node_id = in_node_to_node[n_neumann(n, 0)];
                    n_neumann(n, 0) = node_id;
                }
            }
            updated_dirichlet_node_ordering_ = true;
        }
 
        void GenericScalarProblem::set_parameters(const json &params, const std::string &root_path)
        {
            if (is_param_valid(params, "is_time_dependent"))
            {
                is_time_dept_ = params["is_time_dependent"];
            }
 
            if (is_param_valid(params, "rhs"))
            {
                const json &rr = params["rhs"];
                const bool has_fe_spaces = rr.is_array() && !rr.empty() && rr.front().is_object() && rr.front().contains("fe_space") && rr.front().contains("value");
                const bool has_body_ids = has_body_value_entries(rr);
                if (has_body_ids)
                {
                    const std::vector<json> entries = flatten_ids(rr);
                    body_rhs_.resize(entries.size());
                    for (size_t k = 0; k < entries.size(); ++k)
                    {
                        body_rhs_[k].body_id = entries[k]["id"];
                        body_rhs_[k].fe_space_id = fe_space_id(entries[k]);
                        body_rhs_[k].value.init(entries[k]["value"], root_path);
                    }
                }
                else if (has_fe_spaces)
                {
                    for (const json &entry : rr)
                        rhs_[fe_space_id(entry)].init(entry["value"], root_path);
                }
                else if (!rr.is_array() || !rr.empty())
                    rhs_[-1].init(rr, root_path);
            }
 
            if (is_param_valid(params, "reference") && is_param_valid(params["reference"], "solution"))
            {
                has_exact_ = !params["reference"]["solution"].empty();
                exact_.init(params["reference"]["solution"], root_path);
            }
 
            if (is_param_valid(params, "reference") && is_param_valid(params["reference"], "gradient"))
            {
                auto ex = params["reference"]["gradient"];
                has_exact_grad_ = ex.size() > 0;
                if (ex.is_array())
                {
                    for (size_t k = 0; k < ex.size(); ++k)
                        exact_grad_[k].init(ex[k], root_path);
                }
                else
                {
                    assert(false);
                }
            }
 
            if (is_param_valid(params, "dirichlet_boundary"))
            {
                // boundary_ids_.clear();
                const int offset = boundary_ids_.size();
                std::vector<json> j_boundary = flatten_ids(params["dirichlet_boundary"]);
 
                boundary_ids_.resize(offset + j_boundary.size());
                dirichlet_.resize(offset + j_boundary.size());
 
                for (size_t i = offset; i < boundary_ids_.size(); ++i)
                {
                    if (j_boundary[i - offset].is_string())
                    {
                        const std::string path = resolve_path(j_boundary[i - offset], params["root_path"]);
                        if (!std::filesystem::is_regular_file(path))
                            log_and_throw_error(fmt::format("unable to open {} file", path));
 
                        Eigen::MatrixXd tmp;
                        io::read_matrix(path, tmp);
                        nodal_dirichlet_mat_.emplace_back(tmp);
 
                        continue;
                    }
 
                    int current_id = -1;
                    dirichlet_[i].fe_space_id = fe_space_id(j_boundary[i - offset]);
 
                    if (j_boundary[i - offset]["id"] == "all")
                    {
                        boundary_ids_[i] = -1;
                        is_all_ = true;
                        nodal_dirichlet_[current_id] = ScalarBCValue();
                    }
                    else
                    {
                        boundary_ids_[i] = j_boundary[i - offset]["id"];
                        current_id = boundary_ids_[i];
                        nodal_dirichlet_[current_id] = ScalarBCValue();
                    }
                    nodal_dirichlet_[current_id].fe_space_id = dirichlet_[i].fe_space_id;
 
                    auto ff = j_boundary[i - offset]["value"];
                    dirichlet_[i].value.init(ff, root_path);
                    nodal_dirichlet_[current_id].value.init(ff, root_path);
 
                    if (j_boundary[i - offset]["interpolation"].is_array())
                    {
                        if (j_boundary[i - offset]["interpolation"].size() == 0)
                            dirichlet_[i].interpolation = std::make_shared<NoInterpolation>();
                        else if (j_boundary[i - offset]["interpolation"].size() == 1)
                            dirichlet_[i].interpolation = Interpolation::build(j_boundary[i - offset]["interpolation"][0]);
                        else
                            log_and_throw_error("Only one Dirichlet interpolation supported");
                    }
                    else
                        dirichlet_[i].interpolation = Interpolation::build(j_boundary[i - offset]["interpolation"]);
 
                    nodal_dirichlet_[current_id].interpolation = dirichlet_[i].interpolation;
                }
            }
 
            if (is_param_valid(params, "neumann_boundary"))
            {
                // neumann_boundary_ids_.clear();
                const int offset = neumann_boundary_ids_.size();
                auto j_boundary_tmp = params["neumann_boundary"];
                std::vector<json> j_boundary = flatten_ids(j_boundary_tmp);
 
                neumann_boundary_ids_.resize(offset + j_boundary.size());
                neumann_.resize(offset + j_boundary.size());
 
                for (size_t i = offset; i < neumann_boundary_ids_.size(); ++i)
                {
                    neumann_boundary_ids_[i] = j_boundary[i - offset]["id"];
                    neumann_[i].fe_space_id = fe_space_id(j_boundary[i - offset]);
 
                    auto ff = j_boundary[i - offset]["value"];
                    neumann_[i].value.init(ff, root_path);
 
                    if (j_boundary[i - offset]["interpolation"].is_array())
                    {
                        if (j_boundary[i - offset]["interpolation"].size() == 0)
                            neumann_[i].interpolation = std::make_shared<NoInterpolation>();
                        else if (j_boundary[i - offset]["interpolation"].size() == 1)
                            neumann_[i].interpolation = Interpolation::build(j_boundary[i - offset]["interpolation"][0]);
                        else
                            log_and_throw_error("Only one Neumann interpolation supported");
                    }
                    else
                        neumann_[i].interpolation = Interpolation::build(j_boundary[i - offset]["interpolation"]);
                }
            }
 
            if (is_param_valid(params, "nodal_neumann_boundary"))
            {
                std::vector<json> j_boundary = flatten_ids(params["nodal_neumann_boundary"]);
 
                for (size_t i = 0; i < j_boundary.size(); ++i)
                {
                    if (j_boundary[i].is_string())
                    {
                        const std::string path = resolve_path(j_boundary[i], params["root_path"]);
                        if (!std::filesystem::is_regular_file(path))
                            log_and_throw_error(fmt::format("unable to open {} file", path));
 
                        Eigen::MatrixXd tmp;
                        io::read_matrix(path, tmp);
                        nodal_neumann_mat_.emplace_back(tmp);
 
                        continue;
                    }
 
                    const int id = j_boundary[i]["id"];
                    ScalarBCValue nodal_neumann;
                    nodal_neumann.fe_space_id = fe_space_id(j_boundary[i]);
                    nodal_neumann.value.init(j_boundary[i]["value"], root_path);
 
                    if (j_boundary[i]["interpolation"].is_array())
                    {
                        if (j_boundary[i]["interpolation"].size() == 0)
                            nodal_neumann.interpolation = std::make_shared<NoInterpolation>();
                        else if (j_boundary[i]["interpolation"].size() == 1)
                            nodal_neumann.interpolation = Interpolation::build(j_boundary[i]["interpolation"][0]);
                        else
                            log_and_throw_error("Only one nodal Neumann interpolation supported");
                    }
                    else
                        nodal_neumann.interpolation = Interpolation::build(j_boundary[i]["interpolation"]);
 
                    nodal_neumann_[id] = nodal_neumann;
                }
            }
 
            if (is_param_valid(params, "solution"))
            {
                auto rr = params["solution"];
                initial_solution_.resize(rr.size());
                assert(rr.is_array());
 
                for (size_t k = 0; k < rr.size(); ++k)
                {
                    initial_solution_[k].body_id = rr[k]["id"];
                    initial_solution_[k].fe_space_id = fe_space_id(rr[k]);
                    initial_solution_[k].value.init(rr[k]["value"], root_path);
                }
            }
        }
 
        void GenericScalarProblem::clear()
        {
            neumann_.clear();
            dirichlet_.clear();
 
            nodal_dirichlet_.clear();
            nodal_neumann_.clear();
            nodal_dirichlet_mat_.clear();
            nodal_neumann_mat_.clear();
 
            rhs_.clear();
            body_rhs_.clear();
            exact_.clear();
            for (int i = 0; i < exact_grad_.size(); ++i)
                exact_grad_[i].clear();
            is_all_ = false;
            has_exact_ = false;
            has_exact_grad_ = false;
            is_time_dept_ = false;
            updated_dirichlet_node_ordering_ = false;
        }
    } // namespace assembler
} // namespace polyfem