hsolver_lrtd.hpp

Go to the documentation of this file.

#pragma once
#include "source_io/module_parameter/parameter.h"
#include "source_hsolver/diago_david.h"
#include "source_hsolver/diago_dav_subspace.h"
#include "source_hsolver/diago_cg.h"
#include "source_hsolver/diago_iter_assist.h"
#include "source_hsolver/diago_cg.h"
#include "source_lcao/module_lr/utils/lr_util.h"
#include "source_lcao/module_lr/utils/lr_util_print.h"
#include "source_base/module_container/ATen/core/tensor_map.h"
 
namespace LR
{
    template<typename T> using Real = typename GetTypeReal<T>::type;
 
    namespace HSolver
    {
        template<typename T>
        inline void print_eigs(const std::vector<T>& eigs, const std::string& label = "", const double factor = 1.0)
        {
            std::cout << label << std::endl;
            for (auto& e : eigs) { std::cout << e * factor << " "; }
            std::cout << std::endl;
        }
 
        template<typename T, typename THamilt>
        void solve(const THamilt& hm,
            T* psi,
            const int& dim, 
            const int& nband,   
            double* eig,
            const std::string method,
            const Real<T>& diag_ethr, 
            const std::vector<Real<T>>& precondition,
            const bool hermitian = true)
        {
            ModuleBase::TITLE("HSolverLR", "solve");
            const std::vector<std::string> spin_types = { "singlet", "triplet" };
            // note: if not TDA, the eigenvalues will be complex
            // then we will need a new constructor of DiagoDavid
 
            // 1. allocate eigenvalue
            std::vector<Real<T>> eigenvalue(nband);   //nstates
            // 2. select the method
#ifdef __MPI
            const hsolver::diag_comm_info comm_info = { POOL_WORLD, GlobalV::RANK_IN_POOL, GlobalV::NPROC_IN_POOL };
#else
            const hsolver::diag_comm_info comm_info = { GlobalV::RANK_IN_POOL, GlobalV::NPROC_IN_POOL };
#endif
 
            if (method == "lapack")
            {
                std::vector<T> Amat_full = hm.matrix();
                const int gdim = std::sqrt(Amat_full.size());
                eigenvalue.resize(gdim);
                if (hermitian) { LR_Util::diag_lapack(gdim, Amat_full.data(), eigenvalue.data()); }
                else
                {
                    std::vector<std::complex<double>> eig_complex(gdim);
                    LR_Util::diag_lapack_nh(gdim, Amat_full.data(), eig_complex.data());
                    print_eigs(eig_complex, "Right eigenvalues: of the non-Hermitian matrix: (Ry)");
                    for (int i = 0; i < gdim; i++) { eigenvalue[i] = eig_complex[i].real(); }
                }
                // copy eigenvectors
                hm.global2local(psi, Amat_full.data(), nband);
            }
            else
            {
                // 3. set maxiter and funcs
                const int maxiter = hsolver::DiagoIterAssist<T>::PW_DIAG_NMAX;
 
                auto hpsi_func = [&hm](T* psi_in, T* hpsi, const int ld_psi, const int nvec) {hm.hPsi(psi_in, hpsi, ld_psi, nvec);};
                auto spsi_func = [&hm](const T* psi_in, T* spsi, const int ld_psi, const int nvec)
                    { std::memcpy(spsi, psi_in, sizeof(T) * ld_psi * nvec); };
 
                if (method == "dav")
                {
                    // Allow 5 tries at most. If ntry > ntry_max = 5, exit diag loop.
                    const int ntry_max = 5;
                    // In non-self consistent calculation, do until totally converged. Else allow 5 eigenvecs to be NOT
                    // converged.
                    const int notconv_max = ("nscf" == PARAM.inp.calculation) ? 0 : 5;
                    // do diag and add davidson iteration counts up to avg_iter
                    hsolver::DiagoDavid<T> david(precondition.data(),
                                                 nband,
                                                 dim,
                                                 PARAM.inp.pw_diag_ndim,
                                                 false,
                                                 comm_info);
                    std::vector<double> ethr_band(nband, diag_ethr);
                    hsolver::DiagoIterAssist<T>::avg_iter += static_cast<double>(david.diag(hpsi_func, spsi_func,
                        dim, psi, eigenvalue.data(), ethr_band, maxiter, ntry_max, 0));
                }
                else if (method == "dav_subspace") //need refactor
                {
                    hsolver::Diago_DavSubspace<T> dav_subspace(precondition,
                        nband,
                        dim,
                        PARAM.inp.pw_diag_ndim,
                        diag_ethr,
                        maxiter,
                        false, //always do the subspace diag (check the implementation)
                        comm_info,
                        PARAM.inp.diag_subspace,
                        PARAM.inp.nb2d);
                    std::vector<double> ethr_band(nband, diag_ethr);
                    hsolver::DiagoIterAssist<T>::avg_iter += static_cast<double>(
                        dav_subspace.diag(hpsi_func, spsi_func, psi, dim, eigenvalue.data(), ethr_band, false /*scf*/));
                }
                else if (method == "cg")
                {
 
                    // auto subspace_func = [&hm](const ct::Tensor& psi_in, ct::Tensor& psi_out) {
                    //     const auto ndim = psi_in.shape().ndim();
                    //     REQUIRES_OK(ndim == 2, "dims of psi_in should be less than or equal to 2");
                    //     // Convert a Tensor object to a psi::Psi object
                    //     auto psi_in_wrapper = psi::Psi<T>(psi_in.data<T>(),
                    //         1,
                    //         psi_in.shape().dim_size(0),
                    //         psi_in.shape().dim_size(1));
                    //     auto psi_out_wrapper = psi::Psi<T>(psi_out.data<T>(),
                    //         1,
                    //         psi_out.shape().dim_size(0),
                    //         psi_out.shape().dim_size(1));
                    //     auto eigen = ct::Tensor(ct::DataTypeToEnum<Real<T>>::value,
                    //         ct::DeviceType::CpuDevice,
                    //         ct::TensorShape({ psi_in.shape().dim_size(0) }));
                    //     hsolver::DiagoIterAssist<T>::diagH_subspace(hm, psi_in_wrapper, psi_out_wrapper, eigen.data<Real<T>>());
                    //     };
 
                    // hsolver::DiagoCG<T> cg("lcao", "nscf", true, subspace_func, diag_ethr, maxiter, GlobalV::NPROC_IN_POOL);
 
                    auto subspace_func = [](const ct::Tensor& psi_in, ct::Tensor& psi_out) {};
                    hsolver::DiagoCG<T> cg("lcao", "nscf", false, subspace_func, diag_ethr, maxiter, GlobalV::NPROC_IN_POOL);
 
                    auto psi_tensor = ct::TensorMap(psi, ct::DataTypeToEnum<T>::value, ct::DeviceType::CpuDevice, ct::TensorShape({ nband, dim }));
                    auto eigen_tensor = ct::TensorMap(eigenvalue.data(), ct::DataTypeToEnum<Real<T>>::value, ct::DeviceType::CpuDevice, ct::TensorShape({ nband }));
                    std::vector<Real<T>> precondition_(precondition);   //since TensorMap does not support const pointer
                    auto precon_tensor = ct::TensorMap(precondition_.data(), ct::DataTypeToEnum<Real<T>>::value, ct::DeviceType::CpuDevice, ct::TensorShape({ dim }));
                    auto hpsi_func = [&hm](const ct::Tensor& psi_in, ct::Tensor& hpsi) {hm.hPsi(psi_in.data<T>(), hpsi.data<T>(), psi_in.shape().dim_size(0) /*nbasis_local*/, 1/*band-by-band*/);};
                    auto spsi_func = [&hm](const ct::Tensor& psi_in, ct::Tensor& spsi) {
                        std::memcpy(spsi.data<T>(), psi_in.data<T>(), sizeof(T) * psi_in.NumElements());
                    };
 
                    std::vector<double> ethr_band(nband, diag_ethr);
                    cg.diag(hpsi_func, spsi_func, psi_tensor, eigen_tensor, ethr_band, precon_tensor);
                }
                else { throw std::runtime_error("HSolverLR::solve: method not implemented"); }
            }
 
            // 5. copy eigenvalues
            for (int ist = 0;ist < nband;++ist) { eig[ist] = eigenvalue[ist]; }
 
            // 6. output eigenvalues and eigenvectors
            print_eigs(eigenvalue, "eigenvalues: (Ry)");
            print_eigs(eigenvalue, "eigenvalues: (eV)", ModuleBase::Ry_to_eV);
 
            // normalization is already satisfied
            // std::cout << "check normalization of eigenvectors:" << std::endl;
            // for (int ist = 0;ist < nband;++ist)
            // {
            //     double norm2 = 0;
            //     for (int ik = 0;ik < psi.get_nk();++ik)
            //     {
            //         for (int ib = 0;ib < psi.get_nbasis();++ib)
            //         {
            //             norm2 += std::norm(psi(ist, ik, ib));
            //             // std::cout << "norm2_now=" << norm2 << std::endl;
            //         }
            //     }
            //     std::cout << "state " << ist << ", norm2=" << norm2 << std::endl;
            // }
 
            // output iters
            std::cout << " Average iterative diagonalization steps: " << hsolver::DiagoIterAssist<T>::avg_iter
                << "; current threshold: " << hsolver::DiagoIterAssist<T>::PW_DIAG_THR << std::endl;
        }
    }
}