sqp.hpp

// Copyright (c) Sleipnir contributors
 
#pragma once
 
#include <chrono>
#include <cmath>
#include <functional>
#include <limits>
#include <span>
 
#include <Eigen/Core>
#include <Eigen/SparseCore>
#include <gch/small_vector.hpp>
 
#include "sleipnir/optimization/solver/exit_status.hpp"
#include "sleipnir/optimization/solver/iteration_info.hpp"
#include "sleipnir/optimization/solver/options.hpp"
#include "sleipnir/optimization/solver/sqp_matrix_callbacks.hpp"
#include "sleipnir/optimization/solver/util/all_finite.hpp"
#include "sleipnir/optimization/solver/util/append_as_triplets.hpp"
#include "sleipnir/optimization/solver/util/filter.hpp"
#include "sleipnir/optimization/solver/util/is_locally_infeasible.hpp"
#include "sleipnir/optimization/solver/util/kkt_error.hpp"
#include "sleipnir/optimization/solver/util/regularized_ldlt.hpp"
#include "sleipnir/util/assert.hpp"
#include "sleipnir/util/print_diagnostics.hpp"
#include "sleipnir/util/profiler.hpp"
#include "sleipnir/util/scope_exit.hpp"
#include "sleipnir/util/symbol_exports.hpp"
 
// See docs/algorithms.md#Works_cited for citation definitions.
 
namespace slp {
 
template <typename Scalar>
ExitStatus sqp(const SQPMatrixCallbacks<Scalar>& matrix_callbacks,
               std::span<std::function<bool(const IterationInfo<Scalar>& info)>>
                   iteration_callbacks,
               const Options& options,
               Eigen::Vector<Scalar, Eigen::Dynamic>& x) {
  using DenseVector = Eigen::Vector<Scalar, Eigen::Dynamic>;
 
  DenseVector y = DenseVector::Zero(matrix_callbacks.num_equality_constraints);
 
  return sqp(matrix_callbacks, iteration_callbacks, options, x, y);
}
 
template <typename Scalar>
ExitStatus sqp(const SQPMatrixCallbacks<Scalar>& matrix_callbacks,
               std::span<std::function<bool(const IterationInfo<Scalar>& info)>>
                   iteration_callbacks,
               const Options& options, Eigen::Vector<Scalar, Eigen::Dynamic>& x,
               Eigen::Vector<Scalar, Eigen::Dynamic>& y) {
  using DenseVector = Eigen::Vector<Scalar, Eigen::Dynamic>;
  using SparseMatrix = Eigen::SparseMatrix<Scalar>;
  using SparseVector = Eigen::SparseVector<Scalar>;
 
  struct Step {
    DenseVector p_x;
    DenseVector p_y;
  };
 
  using std::isfinite;
 
  const auto solve_start_time = std::chrono::steady_clock::now();
 
  gch::small_vector<SolveProfiler> solve_profilers;
  solve_profilers.emplace_back("solver");
  solve_profilers.emplace_back("  ↳ setup");
  solve_profilers.emplace_back("  ↳ iteration");
  solve_profilers.emplace_back("    ↳ feasibility ✓");
  solve_profilers.emplace_back("    ↳ iter callbacks");
  solve_profilers.emplace_back("    ↳ KKT matrix build");
  solve_profilers.emplace_back("    ↳ KKT matrix decomp");
  solve_profilers.emplace_back("    ↳ KKT system solve");
  solve_profilers.emplace_back("    ↳ line search");
  solve_profilers.emplace_back("      ↳ SOC");
  solve_profilers.emplace_back("    ↳ next iter prep");
  solve_profilers.emplace_back("    ↳ f(x)");
  solve_profilers.emplace_back("    ↳ ∇f(x)");
  solve_profilers.emplace_back("    ↳ ∇²ₓₓL");
  solve_profilers.emplace_back("    ↳ cₑ(x)");
  solve_profilers.emplace_back("    ↳ ∂cₑ/∂x");
 
  auto& solver_prof = solve_profilers[0];
  auto& setup_prof = solve_profilers[1];
  auto& inner_iter_prof = solve_profilers[2];
  auto& feasibility_check_prof = solve_profilers[3];
  auto& iter_callbacks_prof = solve_profilers[4];
  auto& kkt_matrix_build_prof = solve_profilers[5];
  auto& kkt_matrix_decomp_prof = solve_profilers[6];
  auto& kkt_system_solve_prof = solve_profilers[7];
  auto& line_search_prof = solve_profilers[8];
  auto& soc_prof = solve_profilers[9];
  auto& next_iter_prep_prof = solve_profilers[10];
 
  // Set up profiled matrix callbacks
#ifndef SLEIPNIR_DISABLE_DIAGNOSTICS
  auto& f_prof = solve_profilers[11];
  auto& g_prof = solve_profilers[12];
  auto& H_prof = solve_profilers[13];
  auto& c_e_prof = solve_profilers[14];
  auto& A_e_prof = solve_profilers[15];
 
  SQPMatrixCallbacks<Scalar> matrices{
      matrix_callbacks.num_decision_variables,
      matrix_callbacks.num_equality_constraints,
      [&](const DenseVector& x) -> Scalar {
        ScopedProfiler prof{f_prof};
        return matrix_callbacks.f(x);
      },
      [&](const DenseVector& x) -> SparseVector {
        ScopedProfiler prof{g_prof};
        return matrix_callbacks.g(x);
      },
      [&](const DenseVector& x, const DenseVector& y) -> SparseMatrix {
        ScopedProfiler prof{H_prof};
        return matrix_callbacks.H(x, y);
      },
      [&](const DenseVector& x) -> DenseVector {
        ScopedProfiler prof{c_e_prof};
        return matrix_callbacks.c_e(x);
      },
      [&](const DenseVector& x) -> SparseMatrix {
        ScopedProfiler prof{A_e_prof};
        return matrix_callbacks.A_e(x);
      }};
#else
  const auto& matrices = matrix_callbacks;
#endif
 
  solver_prof.start();
  setup_prof.start();
 
  Scalar f = matrices.f(x);
  SparseVector g = matrices.g(x);
  SparseMatrix H = matrices.H(x, y);
  DenseVector c_e = matrices.c_e(x);
  SparseMatrix A_e = matrices.A_e(x);
 
  // Ensure matrix callback dimensions are consistent
  slp_assert(g.rows() == matrices.num_decision_variables);
  slp_assert(H.rows() == matrices.num_decision_variables);
  slp_assert(H.cols() == matrices.num_decision_variables);
  slp_assert(c_e.rows() == matrices.num_equality_constraints);
  slp_assert(A_e.rows() == matrices.num_equality_constraints);
  slp_assert(A_e.cols() == matrices.num_decision_variables);
 
  // Check for overconstrained problem
  if (matrices.num_equality_constraints > matrices.num_decision_variables) {
    if (options.diagnostics) {
      print_too_few_dofs_error(c_e);
    }
 
    return ExitStatus::TOO_FEW_DOFS;
  }
 
  // Check whether initial guess has finite cost, constraints, and derivatives
  if (!isfinite(f) || !all_finite(g) || !all_finite(H) || !c_e.allFinite() ||
      !all_finite(A_e)) {
    return ExitStatus::NONFINITE_INITIAL_GUESS;
  }
 
  int iterations = 0;
 
  Filter<Scalar> filter;
 
  // Kept outside the loop so its storage can be reused
  gch::small_vector<Eigen::Triplet<Scalar>> triplets;
 
  RegularizedLDLT<Scalar> solver{matrices.num_decision_variables,
                                 matrices.num_equality_constraints};
 
  // Variables for determining when a step is acceptable
  constexpr Scalar α_reduction_factor(0.5);
  constexpr Scalar α_min(1e-7);
 
  int full_step_rejected_counter = 0;
 
  // Error
  Scalar E_0 = std::numeric_limits<Scalar>::infinity();
 
  setup_prof.stop();
 
  // Prints final solver diagnostics when the solver exits
  scope_exit exit{[&] {
    if (options.diagnostics) {
      solver_prof.stop();
      if (iterations > 0) {
        print_bottom_iteration_diagnostics();
      }
      print_solver_diagnostics(solve_profilers);
    }
  }};
 
  while (E_0 > Scalar(options.tolerance)) {
    ScopedProfiler inner_iter_profiler{inner_iter_prof};
    ScopedProfiler feasibility_check_profiler{feasibility_check_prof};
 
    // Check for local equality constraint infeasibility
    if (is_equality_locally_infeasible(A_e, c_e)) {
      if (options.diagnostics) {
        print_c_e_local_infeasibility_error(c_e);
      }
 
      return ExitStatus::LOCALLY_INFEASIBLE;
    }
 
    // Check for diverging iterates
    if (x.template lpNorm<Eigen::Infinity>() > Scalar(1e10) || !x.allFinite()) {
      return ExitStatus::DIVERGING_ITERATES;
    }
 
    feasibility_check_profiler.stop();
    ScopedProfiler iter_callbacks_profiler{iter_callbacks_prof};
 
    // Call iteration callbacks
    for (const auto& callback : iteration_callbacks) {
      if (callback({iterations, x, {}, y, {}, g, H, A_e, {}})) {
        return ExitStatus::CALLBACK_REQUESTED_STOP;
      }
    }
 
    iter_callbacks_profiler.stop();
    ScopedProfiler kkt_matrix_build_profiler{kkt_matrix_build_prof};
 
    // lhs = [H   Aₑᵀ]
    //       [Aₑ   0 ]
    //
    // Don't assign upper triangle because solver only uses lower triangle.
    triplets.clear();
    triplets.reserve(H.nonZeros() + A_e.nonZeros());
    append_as_triplets(triplets, 0, 0, {H, A_e});
    SparseMatrix lhs(
        matrices.num_decision_variables + matrices.num_equality_constraints,
        matrices.num_decision_variables + matrices.num_equality_constraints);
    lhs.setFromSortedTriplets(triplets.begin(), triplets.end());
 
    // rhs = −[∇f − Aₑᵀy]
    //        [   cₑ    ]
    DenseVector rhs{x.rows() + y.rows()};
    rhs.segment(0, x.rows()) = -g + A_e.transpose() * y;
    rhs.segment(x.rows(), y.rows()) = -c_e;
 
    kkt_matrix_build_profiler.stop();
    ScopedProfiler kkt_matrix_decomp_profiler{kkt_matrix_decomp_prof};
 
    Step step;
    constexpr Scalar α_max(1);
    Scalar α(1);
 
    // Solve the Newton-KKT system
    //
    // [H   Aₑᵀ][ pˣ] = −[∇f − Aₑᵀy]
    // [Aₑ   0 ][−pʸ]    [   cₑ    ]
    if (solver.compute(lhs).info() != Eigen::Success) [[unlikely]] {
      return ExitStatus::FACTORIZATION_FAILED;
    }
 
    kkt_matrix_decomp_profiler.stop();
    ScopedProfiler kkt_system_solve_profiler{kkt_system_solve_prof};
 
    auto compute_step = [&](Step& step) {
      // p = [ pˣ]
      //     [−pʸ]
      DenseVector p = solver.solve(rhs);
      step.p_x = p.segment(0, x.rows());
      step.p_y = -p.segment(x.rows(), y.rows());
    };
    compute_step(step);
 
    kkt_system_solve_profiler.stop();
    ScopedProfiler line_search_profiler{line_search_prof};
 
    α = α_max;
 
    // Loop until a step is accepted
    while (1) {
      DenseVector trial_x = x + α * step.p_x;
      DenseVector trial_y = y + α * step.p_y;
 
      Scalar trial_f = matrices.f(trial_x);
      DenseVector trial_c_e = matrices.c_e(trial_x);
 
      // If f(xₖ + αpₖˣ) or cₑ(xₖ + αpₖˣ) aren't finite, reduce step size
      // immediately
      if (!isfinite(trial_f) || !trial_c_e.allFinite()) {
        // Reduce step size
        α *= α_reduction_factor;
 
        if (α < α_min) {
          return ExitStatus::LINE_SEARCH_FAILED;
        }
        continue;
      }
 
      // Check whether filter accepts trial iterate
      if (filter.try_add(FilterEntry{trial_f, trial_c_e}, α)) {
        // Accept step
        break;
      }
 
      Scalar prev_constraint_violation = c_e.template lpNorm<1>();
      Scalar next_constraint_violation = trial_c_e.template lpNorm<1>();
 
      // Second-order corrections
      //
      // If first trial point was rejected and constraint violation stayed the
      // same or went up, apply second-order corrections
      if (α == α_max &&
          next_constraint_violation >= prev_constraint_violation) {
        // Apply second-order corrections. See section 2.4 of [2].
        auto soc_step = step;
 
        Scalar α_soc = α;
        DenseVector c_e_soc = c_e;
 
        bool step_acceptable = false;
        for (int soc_iteration = 0; soc_iteration < 5 && !step_acceptable;
             ++soc_iteration) {
          ScopedProfiler soc_profiler{soc_prof};
 
          scope_exit soc_exit{[&] {
            soc_profiler.stop();
 
            if (options.diagnostics) {
              print_iteration_diagnostics(
                  iterations,
                  step_acceptable ? IterationType::ACCEPTED_SOC
                                  : IterationType::REJECTED_SOC,
                  soc_profiler.current_duration(),
                  kkt_error<Scalar, KKTErrorType::INF_NORM_SCALED>(
                      g, A_e, trial_c_e, trial_y),
                  trial_f, trial_c_e.template lpNorm<1>(), Scalar(0), Scalar(0),
                  solver.hessian_regularization(), α_soc, Scalar(1),
                  α_reduction_factor, Scalar(1));
            }
          }};
 
          // Rebuild Newton-KKT rhs with updated constraint values.
          //
          // rhs = −[∇f − Aₑᵀy]
          //        [  cₑˢᵒᶜ  ]
          //
          // where cₑˢᵒᶜ = αc(xₖ) + c(xₖ + αpₖˣ)
          c_e_soc = α_soc * c_e_soc + trial_c_e;
          rhs.bottomRows(y.rows()) = -c_e_soc;
 
          // Solve the Newton-KKT system
          compute_step(soc_step);
 
          trial_x = x + α_soc * soc_step.p_x;
          trial_y = y + α_soc * soc_step.p_y;
 
          trial_f = matrices.f(trial_x);
          trial_c_e = matrices.c_e(trial_x);
 
          // Constraint violation scale factor for second-order corrections
          constexpr Scalar κ_soc(0.99);
 
          // If constraint violation hasn't been sufficiently reduced, stop
          // making second-order corrections
          next_constraint_violation = trial_c_e.template lpNorm<1>();
          if (next_constraint_violation > κ_soc * prev_constraint_violation) {
            break;
          }
 
          // Check whether filter accepts trial iterate
          if (filter.try_add(FilterEntry{trial_f, trial_c_e}, α)) {
            step = soc_step;
            α = α_soc;
            step_acceptable = true;
          }
        }
 
        if (step_acceptable) {
          // Accept step
          break;
        }
      }
 
      // If we got here and α is the full step, the full step was rejected.
      // Increment the full-step rejected counter to keep track of how many full
      // steps have been rejected in a row.
      if (α == α_max) {
        ++full_step_rejected_counter;
      }
 
      // If the full step was rejected enough times in a row, reset the filter
      // because it may be impeding progress.
      //
      // See section 3.2 case I of [2].
      if (full_step_rejected_counter >= 4 &&
          filter.max_constraint_violation >
              filter.back().constraint_violation / Scalar(10)) {
        filter.max_constraint_violation *= Scalar(0.1);
        filter.reset();
        continue;
      }
 
      // Reduce step size
      α *= α_reduction_factor;
 
      // If step size hit a minimum, check if the KKT error was reduced. If it
      // wasn't, report line search failure.
      if (α < α_min) {
        Scalar current_kkt_error =
            kkt_error<Scalar, KKTErrorType::ONE_NORM>(g, A_e, c_e, y);
 
        trial_x = x + α_max * step.p_x;
        trial_y = y + α_max * step.p_y;
 
        trial_c_e = matrices.c_e(trial_x);
 
        Scalar next_kkt_error = kkt_error<Scalar, KKTErrorType::ONE_NORM>(
            matrices.g(trial_x), matrices.A_e(trial_x), trial_c_e, trial_y);
 
        // If the step using αᵐᵃˣ reduced the KKT error, accept it anyway
        if (next_kkt_error <= Scalar(0.999) * current_kkt_error) {
          α = α_max;
 
          // Accept step
          break;
        }
 
        return ExitStatus::LINE_SEARCH_FAILED;
      }
    }
 
    line_search_profiler.stop();
 
    // If full step was accepted, reset full-step rejected counter
    if (α == α_max) {
      full_step_rejected_counter = 0;
    }
 
    // xₖ₊₁ = xₖ + αₖpₖˣ
    // yₖ₊₁ = yₖ + αₖpₖʸ
    x += α * step.p_x;
    y += α * step.p_y;
 
    // Update autodiff for Jacobians and Hessian
    f = matrices.f(x);
    A_e = matrices.A_e(x);
    g = matrices.g(x);
    H = matrices.H(x, y);
 
    ScopedProfiler next_iter_prep_profiler{next_iter_prep_prof};
 
    c_e = matrices.c_e(x);
 
    // Update the error
    E_0 = kkt_error<Scalar, KKTErrorType::INF_NORM_SCALED>(g, A_e, c_e, y);
 
    next_iter_prep_profiler.stop();
    inner_iter_profiler.stop();
 
    if (options.diagnostics) {
      print_iteration_diagnostics(iterations, IterationType::NORMAL,
                                  inner_iter_profiler.current_duration(), E_0,
                                  f, c_e.template lpNorm<1>(), Scalar(0),
                                  Scalar(0), solver.hessian_regularization(), α,
                                  α_max, α_reduction_factor, α);
    }
 
    ++iterations;
 
    // Check for max iterations
    if (iterations >= options.max_iterations) {
      return ExitStatus::MAX_ITERATIONS_EXCEEDED;
    }
 
    // Check for max wall clock time
    if (std::chrono::steady_clock::now() - solve_start_time > options.timeout) {
      return ExitStatus::TIMEOUT;
    }
  }
 
  return ExitStatus::SUCCESS;
}
 
extern template SLEIPNIR_DLLEXPORT ExitStatus
sqp(const SQPMatrixCallbacks<double>& matrix_callbacks,
    std::span<std::function<bool(const IterationInfo<double>& info)>>
        iteration_callbacks,
    const Options& options, Eigen::Vector<double, Eigen::Dynamic>& x);
 
}  // namespace slp