gw.h

Go to the documentation of this file.

/*
 *            Copyright 2009-2020 The VOTCA Development Team
 *                       (http://www.votca.org)
 *
 *      Licensed under the Apache License, Version 2.0 (the "License")
 *
 * You may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *              http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 *
 */
 
#pragma once
#ifndef VOTCA_XTP_GW_H
#define VOTCA_XTP_GW_H
 
#include <unordered_set>
 
// Local VOTCA includes
#include "logger.h"
#include "orbitals.h"
#include "qp_solver_utils.h"
#include "rpa.h"
#include "sigma_base.h"
#include "threecenter.h"
 
namespace votca {
namespace xtp {
 
class GW {
 
  using EvalStage = qp_solver::EvalStage;
  using QPStats = qp_solver::Stats;
  using QPRootCandidate = qp_solver::RootCandidate;
  using QPWindowDiagnostics = qp_solver::WindowDiagnostics;
 
 public:
  GW(Logger& log, TCMatrix_gwbse& Mmn, const Eigen::MatrixXd& vxc,
     const Eigen::VectorXd& dft_energies)
      : log_(log),
        Mmn_(Mmn),
        vxc_(vxc),
        dft_energies_(dft_energies),
        rpa_(log, Mmn) {};
 
  struct options {
    Index homo;
    Index qpmin;
    Index qpmax;
    Index rpamin;
    Index rpamax;
    double eta;
    double g_sc_limit;
    Index g_sc_max_iterations;
    double gw_sc_limit;
    Index gw_sc_max_iterations;
    double shift = 0;
    double ScaHFX = 0.0;
    std::string sigma_integration;
    Index reset_3c;  // how often the 3c integrals in iterate should be
                     // rebuilt
    std::string qp_solver;
    double qp_solver_alpha = 0.75;
    // Legacy aliases from the original grid-search implementation.
    // They used to define both the total QP window and the scan resolution at
    // the same time. The new implementation keeps them only so that old XML
    // files and existing tests still work; configuration code maps them onto
    // the decoupled controls below.
    Index qp_grid_steps = 0;
    double qp_grid_spacing = 0.0;
 
    // Decoupled QP search controls:
    //  - qp_full_window_half_width defines the outer full search interval
    //  - qp_dense_spacing defines the dense scan used for robust sign changes
    //  - qp_adaptive_shell_width / count define the outward shell search
    // All values are normalized in configure(); negative values mean "unset"
    // and trigger either legacy mapping or robust defaults.
    double qp_full_window_half_width = -1.0;  // Ha; <=0 means "unset"
    double qp_dense_spacing = -1.0;           // Ha; <=0 means "unset"
    double qp_adaptive_shell_width = -1.0;    // Ha; <=0 means "unset"
    Index qp_adaptive_shell_count = 0;        // 0 means "use shell width"
 
    Index gw_mixing_order;          // mixing order
    double gw_mixing_alpha;         // mixing alpha, also linear mixing
    std::string quadrature_scheme;  // Kind of Gaussian-quadrature scheme to use
    Index order;   // only needed for complex integration sigma CDA
    double alpha;  // smooth tail in complex integration sigma CDA
    bool qp_restrict_search = true;
    double qp_zero_margin = 1e-6;
    double qp_virtual_min_energy = -0.1;
    std::string qp_root_finder = "bisection";
    std::string qp_grid_search_mode = "adaptive_with_dense_fallback";
  };
 
  void configure(const options& opt);
 
  Eigen::VectorXd getGWAResults() const;
  // Calculates the diagonal elements up to self consistency
  void CalculateGWPerturbation();
 
  // Calculated offdiagonal elements as well
  void CalculateHQP();
 
  Eigen::MatrixXd getHQP() const;
 
  // Diagonalize QP particle Hamiltonian
  Eigen::SelfAdjointEigenSolver<Eigen::MatrixXd> DiagonalizeQPHamiltonian()
      const;
 
  void PlotSigma(std::string filename, Index steps, double spacing,
                 std::string states) const;
 
  Eigen::VectorXd RPAInputEnergies() const {
    return rpa_.getRPAInputEnergies();
  }
 
 private:
  Index qptotal_;
 
  Eigen::MatrixXd Sigma_x_;
  Eigen::MatrixXd Sigma_c_;
 
  options opt_;
 
  std::unique_ptr<Sigma_base> sigma_ = nullptr;
  Logger& log_;
  TCMatrix_gwbse& Mmn_;
  const Eigen::MatrixXd& vxc_;
  const Eigen::VectorXd& dft_energies_;
 
  Index gw_sc_iteration_;
 
  RPA rpa_;
  // small class which calculates f(w) with and df/dw(w)
  // f=Sigma_c(w)+offset-w
  // offset= e_dft+Sigma_x-Vxc
  class QPFunc {
   public:
    QPFunc(Index gw_level, const Sigma_base& sigma, double offset)
        : gw_level_(gw_level), offset_(offset), sigma_c_func_(sigma) {}
 
    std::pair<double, double> operator()(double frequency) const {
      std::pair<double, double> result;
      result.first = value(frequency, EvalStage::Other);
      result.second = deriv(frequency);
      return result;
    }
 
    double sigma(double frequency, EvalStage stage = EvalStage::Other) const {
      const std::uint64_t key = FrequencyKey(frequency);
 
      auto insert_result = seen_frequencies_.insert(key);
      if (!insert_result.second) {
        ++stats_.sigma_repeat_calls;
      } else {
        ++stats_.sigma_unique_frequencies;
      }
 
      CountSigmaStage(stage);
      return sigma_c_func_.CalcCorrelationDiagElement(gw_level_, frequency);
    }
 
    double value(double frequency, EvalStage stage = EvalStage::Other) const {
      return sigma(frequency, stage) + offset_ - frequency;
    }
 
    double deriv(double frequency) const {
      ++stats_.deriv_calls;
      return sigma_c_func_.CalcCorrelationDiagElementDerivative(gw_level_,
                                                                frequency) -
             1.0;
    }
 
    const QPStats& GetStats() const { return stats_; }
 
   private:
    static std::uint64_t FrequencyKey(double x) {
      std::uint64_t key = 0;
      static_assert(sizeof(double) == sizeof(std::uint64_t),
                    "Unexpected double size");
      std::memcpy(&key, &x, sizeof(double));
      return key;
    }
 
    void CountSigmaStage(EvalStage stage) const {
      switch (stage) {
        case EvalStage::Scan:
          ++stats_.sigma_scan_calls;
          break;
        case EvalStage::Refine:
          ++stats_.sigma_refine_calls;
          break;
        case EvalStage::Derivative:
          ++stats_.sigma_derivative_calls;
          break;
        case EvalStage::Other:
        default:
          ++stats_.sigma_other_calls;
          break;
      }
    }
 
    Index gw_level_;
    double offset_;
    const Sigma_base& sigma_c_func_;
 
    mutable std::unordered_set<std::uint64_t> seen_frequencies_;
    mutable QPStats stats_;
  };
 
  double CalcHomoLumoShift(Eigen::VectorXd frequencies) const;
  Eigen::VectorXd ScissorShift_DFTlevel(
      const Eigen::VectorXd& dft_energies) const;
  void PrintQP_Energies(const Eigen::VectorXd& qp_diag_energies) const;
  void PrintGWA_Energies() const;
 
  Eigen::VectorXd SolveQP(const Eigen::VectorXd& frequencies) const;
  boost::optional<double> SolveQP_Grid(double intercept0, double frequency0,
                                       Index gw_level,
                                       QPStats* stats = nullptr) const;
 
  boost::optional<double> SolveQP_Grid_Windowed(
      double intercept0, double frequency0, Index gw_level, double left_limit,
      double right_limit, bool allow_rejected_return = true,
      QPStats* stats = nullptr) const;
 
  boost::optional<double> SolveQP_Grid_Windowed_Adaptive(
      double intercept0, double frequency0, Index gw_level, double left_limit,
      double right_limit, bool allow_rejected_return = true,
      QPStats* stats = nullptr) const;
 
  boost::optional<double> SolveQP_Grid_Windowed_Dense(
      double intercept0, double frequency0, Index gw_level, double left_limit,
      double right_limit, bool allow_rejected_return = true,
      QPStats* stats = nullptr) const;
 
  boost::optional<double> SolveQP_FixedPoint(double intercept0,
                                             double frequency0, Index gw_level,
                                             QPStats* stats = nullptr) const;
  boost::optional<double> SolveQP_Linearisation(double intercept0,
                                                double frequency0,
                                                Index gw_level,
                                                QPStats* stats = nullptr) const;
  bool Converged(const Eigen::VectorXd& e1, const Eigen::VectorXd& e2,
                 double epsilon) const;
 
  boost::optional<QPRootCandidate> RefineQPInterval(
      double lowerbound, double f_lowerbound, double upperbound,
      double f_upperbound, const QPFunc& f, double reference) const;
};
}  // namespace xtp
}  // namespace votca
 
#endif  // VOTCA_XTP_GW_H