problem.hpp

// Copyright (c) Sleipnir contributors
 
#pragma once
 
#include <algorithm>
#include <array>
#include <cmath>
#include <concepts>
#include <cstddef>
#include <functional>
#include <iterator>
#include <memory>
#include <optional>
#include <ranges>
#include <utility>
 
#include <Eigen/Core>
#include <Eigen/SparseCore>
#include <gch/small_vector.hpp>
 
#include "sleipnir/autodiff/expression_type.hpp"
#include "sleipnir/autodiff/gradient.hpp"
#include "sleipnir/autodiff/hessian.hpp"
#include "sleipnir/autodiff/jacobian.hpp"
#include "sleipnir/autodiff/variable.hpp"
#include "sleipnir/autodiff/variable_matrix.hpp"
#include "sleipnir/optimization/solver/exit_status.hpp"
#include "sleipnir/optimization/solver/ipm.hpp"
#include "sleipnir/optimization/solver/ipm_matrix_callbacks.hpp"
#include "sleipnir/optimization/solver/iteration_info.hpp"
#include "sleipnir/optimization/solver/newton.hpp"
#include "sleipnir/optimization/solver/newton_matrix_callbacks.hpp"
#include "sleipnir/optimization/solver/options.hpp"
#include "sleipnir/optimization/solver/sqp.hpp"
#include "sleipnir/optimization/solver/sqp_matrix_callbacks.hpp"
#include "sleipnir/optimization/solver/util/bounds.hpp"
#include "sleipnir/optimization/solver/util/problem_scaling.hpp"
#include "sleipnir/optimization/solver/util/time_trace.hpp"
#include "sleipnir/util/empty.hpp"
#include "sleipnir/util/print.hpp"
#include "sleipnir/util/print_diagnostics.hpp"
#include "sleipnir/util/spy.hpp"
#include "sleipnir/util/symbol_exports.hpp"
#include "sleipnir/util/time_tracer.hpp"
 
namespace slp {
 
template <typename Scalar>
class Problem {
 public:
  Problem() noexcept = default;
 
  [[nodiscard]]
  Variable<Scalar> decision_variable() {
    m_decision_variables.emplace_back();
    return m_decision_variables.back();
  }
 
  [[nodiscard]]
  VariableMatrix<Scalar> decision_variable(int rows, int cols = 1) {
    m_decision_variables.reserve(m_decision_variables.size() + rows * cols);
 
    VariableMatrix<Scalar> vars{detail::empty, rows, cols};
 
    for (int row = 0; row < rows; ++row) {
      for (int col = 0; col < cols; ++col) {
        m_decision_variables.emplace_back();
        vars[row, col] = m_decision_variables.back();
      }
    }
 
    return vars;
  }
 
  [[nodiscard]]
  VariableMatrix<Scalar> symmetric_decision_variable(int rows) {
    // We only need to store the lower triangle of an n x n symmetric matrix;
    // the other elements are duplicates. The lower triangle has (n² + n)/2
    // elements.
    //
    //   n
    //   Σ k = (n² + n)/2
    //  k=1
    m_decision_variables.reserve(m_decision_variables.size() +
                                 (rows * rows + rows) / 2);
 
    VariableMatrix<Scalar> vars{detail::empty, rows, rows};
 
    for (int row = 0; row < rows; ++row) {
      for (int col = 0; col <= row; ++col) {
        m_decision_variables.emplace_back();
        vars[row, col] = m_decision_variables.back();
        vars[col, row] = m_decision_variables.back();
      }
    }
 
    return vars;
  }
 
  void minimize(const Variable<Scalar>& cost) { m_f = cost; }
 
  void minimize(Variable<Scalar>&& cost) { m_f = std::move(cost); }
 
  void maximize(const Variable<Scalar>& objective) {
    // Maximizing a cost function is the same as minimizing its negative
    m_f = -objective;
  }
 
  void maximize(Variable<Scalar>&& objective) {
    // Maximizing a cost function is the same as minimizing its negative
    m_f = -std::move(objective);
  }
 
  void subject_to(const EqualityConstraints<Scalar>& constraint) {
    m_equality_constraints.reserve(m_equality_constraints.size() +
                                   constraint.constraints.size());
    std::ranges::copy(constraint.constraints,
                      std::back_inserter(m_equality_constraints));
  }
 
  void subject_to(EqualityConstraints<Scalar>&& constraint) {
    m_equality_constraints.reserve(m_equality_constraints.size() +
                                   constraint.constraints.size());
    std::ranges::copy(constraint.constraints,
                      std::back_inserter(m_equality_constraints));
  }
 
  void subject_to(const InequalityConstraints<Scalar>& constraint) {
    m_inequality_constraints.reserve(m_inequality_constraints.size() +
                                     constraint.constraints.size());
    std::ranges::copy(constraint.constraints,
                      std::back_inserter(m_inequality_constraints));
  }
 
  void subject_to(InequalityConstraints<Scalar>&& constraint) {
    m_inequality_constraints.reserve(m_inequality_constraints.size() +
                                     constraint.constraints.size());
    std::ranges::copy(constraint.constraints,
                      std::back_inserter(m_inequality_constraints));
  }
 
  ExpressionType cost_function_type() const {
    if (m_f) {
      return m_f.value().type();
    } else {
      return ExpressionType::NONE;
    }
  }
 
  ExpressionType equality_constraint_type() const {
    if (!m_equality_constraints.empty()) {
      return std::ranges::max(m_equality_constraints, {},
                              &Variable<Scalar>::type)
          .type();
    } else {
      return ExpressionType::NONE;
    }
  }
 
  ExpressionType inequality_constraint_type() const {
    if (!m_inequality_constraints.empty()) {
      return std::ranges::max(m_inequality_constraints, {},
                              &Variable<Scalar>::type)
          .type();
    } else {
      return ExpressionType::NONE;
    }
  }
 
  ExitStatus solve(const Options<Scalar>& options = Options<Scalar>{},
                   bool spy = false) {
    if (options.diagnostics) {
      print_exit_conditions(options);
      print_problem_analysis();
    }
 
    // If the problem is empty or constant, there's nothing to do
    if (cost_function_type() <= ExpressionType::CONSTANT &&
        equality_constraint_type() <= ExpressionType::CONSTANT &&
        inequality_constraint_type() <= ExpressionType::CONSTANT) {
      if (options.diagnostics) {
        slp::println("\nInvoking no-op solver\n");
      }
      return ExitStatus::SUCCESS;
    }
 
    // Create the initial value column vector
    DenseVector x{m_decision_variables.size()};
    for (size_t i = 0; i < m_decision_variables.size(); ++i) {
      x[i] = m_decision_variables[i].value();
    }
 
    // Solve the optimization problem
    ExitStatus status;
    if (m_equality_constraints.empty() && m_inequality_constraints.empty()) {
      if (options.diagnostics) {
        slp::println("\nInvoking Newton solver\n");
      }
 
      status = solve_newton(options, spy, x);
    } else if (m_inequality_constraints.empty()) {
      if (options.diagnostics) {
        slp::println("\nInvoking SQP solver\n");
      }
 
      status = solve_sqp(options, spy, x);
    } else {
      if (options.diagnostics) {
        slp::println("\nInvoking IPM solver\n");
      }
 
      status = solve_ipm(options, spy, x);
    }
 
    if (options.diagnostics) {
      slp::println("\nExit: {}", status);
    }
 
    // Assign the solution to the original Variable instances
    VariableMatrix<Scalar>{m_decision_variables}.set_value(x);
 
    return status;
  }
 
  template <typename F>
    requires requires(F callback, const IterationInfo<Scalar>& info) {
      { callback(info) } -> std::same_as<void>;
    }
  void add_callback(F&& callback) {
    m_iteration_callbacks.emplace_back(
        [=, callback =
                std::forward<F>(callback)](const IterationInfo<Scalar>& info) {
          callback(info);
          return false;
        });
  }
 
  template <typename F>
    requires requires(F callback, const IterationInfo<Scalar>& info) {
      { callback(info) } -> std::same_as<bool>;
    }
  void add_callback(F&& callback) {
    m_iteration_callbacks.emplace_back(std::forward<F>(callback));
  }
 
  void clear_callbacks() { m_iteration_callbacks.clear(); }
 
  template <typename F>
    requires requires(F callback, const IterationInfo<Scalar>& info) {
      { callback(info) } -> std::same_as<bool>;
    }
  void add_persistent_callback(F&& callback) {
    m_persistent_iteration_callbacks.emplace_back(std::forward<F>(callback));
  }
 
 private:
  using DenseVector = Eigen::VectorX<Scalar>;
 
  // The list of decision variables, which are the root of the problem's
  // expression tree
  gch::small_vector<Variable<Scalar>> m_decision_variables;
 
  // The cost function: f(x)
  std::optional<Variable<Scalar>> m_f;
 
  // The list of equality constraints: cₑ(x) = 0
  gch::small_vector<Variable<Scalar>> m_equality_constraints;
 
  // The list of inequality constraints: cᵢ(x) ≥ 0
  gch::small_vector<Variable<Scalar>> m_inequality_constraints;
 
  // The iteration callbacks
  gch::small_vector<std::function<bool(const IterationInfo<Scalar>& info)>>
      m_iteration_callbacks;
  gch::small_vector<std::function<bool(const IterationInfo<Scalar>& info)>>
      m_persistent_iteration_callbacks;
 
  ExitStatus solve_newton(const Options<Scalar>& options,
                          [[maybe_unused]] bool spy, DenseVector& x) {
    using SparseMatrix = Eigen::SparseMatrix<Scalar>;
    using SparseVector = Eigen::SparseVector<Scalar>;
    using enum SetupTimeTracerID;
 
    VariableMatrix<Scalar> x_ad{m_decision_variables};
 
    // Set up cost function
    Variable f = m_f.value_or(Scalar(0));
 
    int num_decision_variables = m_decision_variables.size();
 
    gch::small_vector<std::function<bool(const IterationInfo<Scalar>& info)>>
        iteration_callbacks;
    for (const auto& callback : m_iteration_callbacks) {
      iteration_callbacks.emplace_back(callback);
    }
    for (const auto& callback : m_persistent_iteration_callbacks) {
      iteration_callbacks.emplace_back(callback);
    }
 
    TimeTracerTree<SetupTimeTracer, newton_setup_time_trace> ad_setup_tracers{
        options.diagnostics};
    ScopedTimeTracer setup_tracer{ad_setup_tracers[SETUP]};
 
    // Set up gradient autodiff
    PhaseTimeTracer phase_tracer{ad_setup_tracers[GRADIENT]};
    Gradient g{f, x_ad};
 
    // Set up Lagrangian Hessian autodiff
    phase_tracer.next(ad_setup_tracers[HESSIAN]);
    Hessian<Scalar, Eigen::Lower> H{f, x_ad};
 
    phase_tracer.stop();
    setup_tracer.stop();
 
    if (options.diagnostics) {
      print_setup_diagnostics(ad_setup_tracers);
    }
 
#ifndef SLEIPNIR_DISABLE_DIAGNOSTICS
    // Sparsity pattern files written when spy flag is set
    std::unique_ptr<Spy<Scalar>> H_spy;
 
    if (spy) {
      H_spy = std::make_unique<Spy<Scalar>>(
          "H.spy", "Hessian", "Decision variables", "Decision variables",
          num_decision_variables, num_decision_variables);
      iteration_callbacks.push_back(
          [&](const IterationInfo<Scalar>& info) -> bool {
            H_spy->add(info.H);
            return false;
          });
    }
#endif
 
    // Automatically scale the cost. The problem scaling procedure is
    // described in more detail in docs/algorithms.md#problem-scaling.
    x_ad.set_value(x);
    const ProblemScaling<Scalar> scaling{g.value()};
 
    NewtonMatrixCallbacks<Scalar> matrix_callbacks{
        num_decision_variables,
        [&](const DenseVector& x) -> Scalar {
          x_ad.set_value(x);
          return scaling.f * f.value();
        },
        [&](const DenseVector& x) -> SparseVector {
          x_ad.set_value(x);
          return scaling.f * g.value();
        },
        [&](const DenseVector& x) -> SparseMatrix {
          x_ad.set_value(x);
          return scaling.f * H.value();
        },
        scaling};
 
    // Invoke Newton solver
    return newton<Scalar>(matrix_callbacks, iteration_callbacks, options, x);
  }
 
  ExitStatus solve_sqp(const Options<Scalar>& options,
                       [[maybe_unused]] bool spy, DenseVector& x) {
    using SparseMatrix = Eigen::SparseMatrix<Scalar>;
    using SparseVector = Eigen::SparseVector<Scalar>;
    using enum SetupTimeTracerID;
 
    VariableMatrix<Scalar> x_ad{m_decision_variables};
 
    // Set up cost function
    Variable f = m_f.value_or(Scalar(0));
 
    int num_decision_variables = m_decision_variables.size();
    int num_equality_constraints = m_equality_constraints.size();
 
    gch::small_vector<std::function<bool(const IterationInfo<Scalar>& info)>>
        iteration_callbacks;
    for (const auto& callback : m_iteration_callbacks) {
      iteration_callbacks.emplace_back(callback);
    }
    for (const auto& callback : m_persistent_iteration_callbacks) {
      iteration_callbacks.emplace_back(callback);
    }
 
    VariableMatrix<Scalar> c_e_ad{m_equality_constraints};
    VariableMatrix<Scalar> y_ad(num_equality_constraints);
 
    TimeTracerTree<SetupTimeTracer, sqp_setup_time_trace> ad_setup_tracers{
        options.diagnostics};
    ScopedTimeTracer setup_tracer{ad_setup_tracers[SETUP]};
 
    // Set up gradient autodiff
    PhaseTimeTracer phase_tracer{ad_setup_tracers[GRADIENT]};
    Gradient g{f, x_ad};
 
    phase_tracer.next(ad_setup_tracers[HESSIAN]);
 
    // Set up cost part of Lagrangian Hessian autodiff
    PhaseTimeTracer hessian_phase_tracer{ad_setup_tracers[HESSIAN_COST]};
    Hessian<Scalar, Eigen::Lower> H_f{f, x_ad};
 
    // Set up constraint part of Lagrangian Hessian autodiff
    hessian_phase_tracer.next(ad_setup_tracers[HESSIAN_CONSTRAINTS]);
    Hessian<Scalar, Eigen::Lower> H_c{-y_ad.T() * c_e_ad, x_ad};
 
    hessian_phase_tracer.stop();
 
    // Set up equality constraint Jacobian autodiff
    phase_tracer.next(ad_setup_tracers[EQUALITY_JACOBIAN]);
    Jacobian A_e{c_e_ad, x_ad};
 
    phase_tracer.stop();
    setup_tracer.stop();
 
    if (options.diagnostics) {
      print_setup_diagnostics(ad_setup_tracers);
    }
 
#ifndef SLEIPNIR_DISABLE_DIAGNOSTICS
    // Sparsity pattern files written when spy flag is set
    std::unique_ptr<Spy<Scalar>> H_spy;
    std::unique_ptr<Spy<Scalar>> A_e_spy;
 
    if (spy) {
      H_spy = std::make_unique<Spy<Scalar>>(
          "H.spy", "Hessian", "Decision variables", "Decision variables",
          num_decision_variables, num_decision_variables);
      A_e_spy = std::make_unique<Spy<Scalar>>(
          "A_e.spy", "Equality constraint Jacobian", "Constraints",
          "Decision variables", num_equality_constraints,
          num_decision_variables);
      iteration_callbacks.push_back(
          [&](const IterationInfo<Scalar>& info) -> bool {
            H_spy->add(info.H);
            A_e_spy->add(info.A_e);
            return false;
          });
    }
#endif
 
    // Automatically scale the cost and constraints. The problem scaling
    // procedure is described in more detail in
    // docs/algorithms.md#problem-scaling.
    x_ad.set_value(x);
    const ProblemScaling<Scalar> scaling{g.value(), A_e.value()};
 
    SQPMatrixCallbacks<Scalar> matrix_callbacks{
        num_decision_variables,
        num_equality_constraints,
        [&](const DenseVector& x) -> Scalar {
          x_ad.set_value(x);
          return scaling.f * f.value();
        },
        [&](const DenseVector& x) -> SparseVector {
          x_ad.set_value(x);
          return scaling.f * g.value();
        },
        [&](const DenseVector& x, const DenseVector& y) -> SparseMatrix {
          x_ad.set_value(x);
          y_ad.set_value(scaling.c_e.cwiseProduct(y));
          return scaling.f * H_f.value() + H_c.value();
        },
        [&](const DenseVector& x, const DenseVector& y) -> SparseMatrix {
          x_ad.set_value(x);
          y_ad.set_value(scaling.c_e.cwiseProduct(y));
          return H_c.value();
        },
        [&](const DenseVector& x) -> DenseVector {
          x_ad.set_value(x);
          return scaling.c_e.cwiseProduct(c_e_ad.value());
        },
        [&](const DenseVector& x) -> SparseMatrix {
          x_ad.set_value(x);
          return scaling.c_e.asDiagonal() * A_e.value();
        },
        scaling};
 
    // Invoke SQP solver
    return sqp<Scalar>(matrix_callbacks, iteration_callbacks, options, x);
  }
 
  ExitStatus solve_ipm(const Options<Scalar>& options,
                       [[maybe_unused]] bool spy, DenseVector& x) {
    using SparseMatrix = Eigen::SparseMatrix<Scalar>;
    using SparseVector = Eigen::SparseVector<Scalar>;
    using enum SetupTimeTracerID;
 
    VariableMatrix<Scalar> x_ad{m_decision_variables};
 
    // Set up cost function
    Variable f = m_f.value_or(Scalar(0));
 
    int num_decision_variables = m_decision_variables.size();
    int num_equality_constraints = m_equality_constraints.size();
    int num_inequality_constraints = m_inequality_constraints.size();
 
    gch::small_vector<std::function<bool(const IterationInfo<Scalar>& info)>>
        iteration_callbacks;
    for (const auto& callback : m_iteration_callbacks) {
      iteration_callbacks.emplace_back(callback);
    }
    for (const auto& callback : m_persistent_iteration_callbacks) {
      iteration_callbacks.emplace_back(callback);
    }
 
    VariableMatrix<Scalar> c_e_ad{m_equality_constraints};
    VariableMatrix<Scalar> c_i_ad{m_inequality_constraints};
    VariableMatrix<Scalar> y_ad(num_equality_constraints);
    VariableMatrix<Scalar> z_ad(num_inequality_constraints);
 
    TimeTracerTree<SetupTimeTracer, ipm_setup_time_trace> ad_setup_tracers{
        options.diagnostics};
    ScopedTimeTracer setup_tracer{ad_setup_tracers[SETUP]};
 
    // Set up gradient autodiff
    PhaseTimeTracer phase_tracer{ad_setup_tracers[GRADIENT]};
    Gradient g{f, x_ad};
 
    phase_tracer.next(ad_setup_tracers[HESSIAN]);
 
    // Set up cost part of Lagrangian Hessian autodiff
    PhaseTimeTracer hessian_phase_tracer{ad_setup_tracers[HESSIAN_COST]};
    Hessian<Scalar, Eigen::Lower> H_f{f, x_ad};
 
    // Set up constraint part of Lagrangian Hessian autodiff
    hessian_phase_tracer.next(ad_setup_tracers[HESSIAN_CONSTRAINTS]);
    Hessian<Scalar, Eigen::Lower> H_c{-y_ad.T() * c_e_ad - z_ad.T() * c_i_ad,
                                      x_ad};
 
    hessian_phase_tracer.stop();
 
    // Set up equality constraint Jacobian autodiff
    phase_tracer.next(ad_setup_tracers[EQUALITY_JACOBIAN]);
    Jacobian A_e{c_e_ad, x_ad};
 
    // Set up inequality constraint Jacobian autodiff
    phase_tracer.next(ad_setup_tracers[INEQUALITY_JACOBIAN]);
    Jacobian A_i{c_i_ad, x_ad};
 
    phase_tracer.stop();
    setup_tracer.stop();
 
    if (options.diagnostics) {
      print_setup_diagnostics(ad_setup_tracers);
    }
 
#ifndef SLEIPNIR_DISABLE_DIAGNOSTICS
    // Sparsity pattern files written when spy flag is set
    std::unique_ptr<Spy<Scalar>> H_spy;
    std::unique_ptr<Spy<Scalar>> A_e_spy;
    std::unique_ptr<Spy<Scalar>> A_i_spy;
 
    if (spy) {
      H_spy = std::make_unique<Spy<Scalar>>(
          "H.spy", "Hessian", "Decision variables", "Decision variables",
          num_decision_variables, num_decision_variables);
      A_e_spy = std::make_unique<Spy<Scalar>>(
          "A_e.spy", "Equality constraint Jacobian", "Constraints",
          "Decision variables", num_equality_constraints,
          num_decision_variables);
      A_i_spy = std::make_unique<Spy<Scalar>>(
          "A_i.spy", "Inequality constraint Jacobian", "Constraints",
          "Decision variables", num_inequality_constraints,
          num_decision_variables);
      iteration_callbacks.push_back(
          [&](const IterationInfo<Scalar>& info) -> bool {
            H_spy->add(info.H);
            A_e_spy->add(info.A_e);
            A_i_spy->add(info.A_i);
            return false;
          });
    }
#endif
 
    const auto [bound_constraint_mask, bounds, conflicting_bound_indices] =
        get_bounds<Scalar>(m_decision_variables, m_inequality_constraints,
                           A_i.value());
    if (!conflicting_bound_indices.empty()) {
      if (options.diagnostics) {
        print_bound_constraint_global_infeasibility_error(
            conflicting_bound_indices);
      }
      return ExitStatus::GLOBALLY_INFEASIBLE;
    }
 
#ifdef SLEIPNIR_ENABLE_BOUND_PROJECTION
    project_onto_bounds(x, bounds);
#endif
 
    // Automatically scale the cost and constraints. The problem scaling
    // procedure is described in more detail in
    // docs/algorithms.md#problem-scaling.
    x_ad.set_value(x);
    const ProblemScaling<Scalar> scaling{g.value(), A_e.value(), A_i.value()};
 
    IPMMatrixCallbacks<Scalar> matrix_callbacks{
        num_decision_variables,
        num_equality_constraints,
        num_inequality_constraints,
        [&](const DenseVector& x) -> Scalar {
          x_ad.set_value(x);
          return scaling.f * f.value();
        },
        [&](const DenseVector& x) -> SparseVector {
          x_ad.set_value(x);
          return scaling.f * g.value();
        },
        [&](const DenseVector& x, const DenseVector& y,
            const DenseVector& z) -> SparseMatrix {
          x_ad.set_value(x);
          y_ad.set_value(scaling.c_e.cwiseProduct(y));
          z_ad.set_value(scaling.c_i.cwiseProduct(z));
          return scaling.f * H_f.value() + H_c.value();
        },
        [&](const DenseVector& x, const DenseVector& y,
            const DenseVector& z) -> SparseMatrix {
          x_ad.set_value(x);
          y_ad.set_value(scaling.c_e.cwiseProduct(y));
          z_ad.set_value(scaling.c_i.cwiseProduct(z));
          return H_c.value();
        },
        [&](const DenseVector& x) -> DenseVector {
          x_ad.set_value(x);
          return scaling.c_e.cwiseProduct(c_e_ad.value());
        },
        [&](const DenseVector& x) -> SparseMatrix {
          x_ad.set_value(x);
          return scaling.c_e.asDiagonal() * A_e.value();
        },
        [&](const DenseVector& x) -> DenseVector {
          x_ad.set_value(x);
          return scaling.c_i.cwiseProduct(c_i_ad.value());
        },
        [&](const DenseVector& x) -> SparseMatrix {
          x_ad.set_value(x);
          return scaling.c_i.asDiagonal() * A_i.value();
        },
        scaling};
 
    // Invoke interior-point method solver
    return ipm<Scalar>(matrix_callbacks, iteration_callbacks, options,
#ifdef SLEIPNIR_ENABLE_BOUND_PROJECTION
                       bound_constraint_mask,
#endif
                       x);
  }
 
  void print_exit_conditions(const Options<Scalar>& options) {
    // Print possible exit conditions
    slp::println("User-configured exit conditions:");
    slp::println("  ↳ error below {}", options.tolerance);
    if (!m_iteration_callbacks.empty() ||
        !m_persistent_iteration_callbacks.empty()) {
      slp::println("  ↳ iteration callback requested stop");
    }
    if (std::isfinite(options.max_iterations)) {
      slp::println("  ↳ executed {} iterations", options.max_iterations);
    }
    if (std::isfinite(options.timeout.count())) {
      slp::println("  ↳ {} elapsed", options.timeout);
    }
  }
 
  void print_problem_analysis() {
    constexpr std::array types{"no", "constant", "linear", "quadratic",
                               "nonlinear"};
 
    // Print problem structure
    slp::println("\nProblem structure:");
    slp::println("  ↳ {} cost function",
                 types[std::to_underlying(cost_function_type())]);
    slp::println("  ↳ {} equality constraints",
                 types[std::to_underlying(equality_constraint_type())]);
    slp::println("  ↳ {} inequality constraints",
                 types[std::to_underlying(inequality_constraint_type())]);
 
    if (m_decision_variables.size() == 1) {
      slp::print("\n1 decision variable\n");
    } else {
      slp::print("\n{} decision variables\n", m_decision_variables.size());
    }
 
    auto print_constraint_types =
        [](const gch::small_vector<Variable<Scalar>>& constraints) {
          std::array<size_t, 5> counts{};
          for (const auto& constraint : constraints) {
            ++counts[std::to_underlying(constraint.type())];
          }
          for (const auto& [count, name] :
               std::views::zip(counts, std::array{"empty", "constant", "linear",
                                                  "quadratic", "nonlinear"})) {
            if (count > 0) {
              slp::println("  ↳ {} {}", count, name);
            }
          }
        };
 
    // Print constraint types
    if (m_equality_constraints.size() == 1) {
      slp::println("1 equality constraint");
    } else {
      slp::println("{} equality constraints", m_equality_constraints.size());
    }
    print_constraint_types(m_equality_constraints);
    if (m_inequality_constraints.size() == 1) {
      slp::println("1 inequality constraint");
    } else {
      slp::println("{} inequality constraints",
                   m_inequality_constraints.size());
    }
    print_constraint_types(m_inequality_constraints);
  }
};
 
extern template class EXPORT_TEMPLATE_DECLARE(SLEIPNIR_DLLEXPORT)
Problem<double>;
 
}  // namespace slp