! Copyright © 2025-2026 Dynare Team ! ! This file is part of Dynare. ! ! Dynare is free software: you can redistribute it and/or modify it under the terms of the ! GNU General Public License as published by the Free Software Foundation, either version 3 of ! the License, or (at your option) any later version. ! ! Dynare is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without ! even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the ! GNU General Public License for more details. ! ! You should have received a copy of the GNU General Public License along with Dynare. If not, ! see . ! ! Original author: Normann Rion ! ! Distribution derivative computation (curly D matrices) module curlyDs use iso_fortran_env, only: real64, int32 use gray_code use blas implicit none (type, external) contains !--------------------------------------------------------------------------- ! Compute distribution derivative matrix (curly D) for one-state case ! ! Computes D(i,j',s) = Σⱼ (∂Bⁱ)ⱼ',ⱼ · vᵢ,ⱼ,ₖ,ₛ · ωᵢ,ⱼ ! then D(i',j',s) = Σᵢ μᵢ',ᵢ · D(i,j',s) ! ! Arguments: ! D [out] : Distribution derivative [N_om × S] ! ind [in] : Bracket indices [N_om × 1] ! inv_h [in] : Inverse grid spacing [N_om × 1] ! a_hat [in] : Policy function derivatives [N_om × S × 1] ! om [in] : Distribution [N_om] ! Mu [in] : Transition matrix [N_e × N_e] !--------------------------------------------------------------------------- subroutine compute_curlyDs_1d(D, ind, inv_h, a_hat, om, Mu) ! Inputs/Outputs real(real64), contiguous, target, intent(out) :: D(:,:) real(real64), contiguous, intent(in) :: inv_h(:,:), a_hat(:,:,:), Mu(:,:), om(:) integer(int32), contiguous, intent(in) :: ind(:,:) ! Local variables real(real64) :: inc real(real64), allocatable, target :: acc(:,:) real(real64), pointer, contiguous :: D_mat(:,:), D_rhs(:,:) integer(int32) :: N_om, N_e, N_a, S, i, j, k, m, a, a_next ! Get dimensions N_e = size(Mu, 1, int32) N_om = size(D, 1, int32) S = size(D, 2, int32) N_a = N_om / N_e allocate(acc(N_om,S)) acc = 0.0_real64 do i=1,N_e do concurrent (k=1:S, j=1:N_a) m = (j-1)*N_e + i a = (ind(i+(j-1)*N_e,1)-1)*N_e + i a_next = a+N_e inc = inv_h(m,1)*om(m)*a_hat(m,k,1) acc(a,k) = acc(a,k) - inc acc(a_next,k) = acc(a_next,k) + inc end do end do D_rhs(1:N_e,1:(N_a*S)) => acc D_mat(1:N_e,1:(N_a*S)) => D call matmul_add("T", "N", 1.0_real64, Mu, D_rhs, 0.0_real64, D_mat) end subroutine compute_curlyDs_1d !--------------------------------------------------------------------------- ! Compute distribution derivative matrix (curly D) for n-state case ! ! Uses Gray code traversal for efficient hypercube corner iteration. ! Handles hard boundaries (weight = 0 or 1) with special cases. ! ! Arguments: ! D [out] : Distribution derivative [N_om × S] ! ind [in] : Multi-dimensional bracket indices [N_om × n] ! w [in] : Interpolation weights [N_om × n] ! inv_h [in] : Inverse grid spacing [N_om × n] ! a_hat [in] : Policy function derivatives [N_om × S × n] ! om [in] : Distribution [N_om] ! mu [in] : Transition matrix [N_e × N_e] ! dims [in] : Grid dimensions for each state [n] !--------------------------------------------------------------------------- subroutine compute_curlyDs_nd(D, ind, w, inv_h, a_hat, om, mu, dims) ! Inputs/Outpus real(real64), contiguous, target, intent(out) :: D(:,:) real(real64), contiguous, intent(in) :: a_hat(:,:,:), w(:,:), inv_h(:,:), mu(:,:), om(:) integer(int32), contiguous, intent(in) :: ind(:,:), dims(:) ! Local variables real(real64), allocatable, target :: acc(:,:) real(real64), allocatable :: beta_k(:), beta_0(:), inc(:), gamma(:), rhs(:,:), r_up(:,:), r_down(:,:) real(real64), pointer, contiguous :: D_mat(:,:) integer(int32) :: N_om, N_e, N_a, S, n, Kcorn, t, i, j, k, l, m, kf, z, lin integer(int32), allocatable :: flip_idx(:), a(:) integer(int32), allocatable, target :: stride(:) logical, allocatable :: sk(:), is_hard_one(:,:), is_hard_zero(:,:) N_om = size(D, 1, int32) S = size(D, 2, int32) n = size(dims, 1, int32) N_e = size(mu, 1, int32) N_a = N_om / N_e ! Generate Gray code flip indices call generate_flip_indices(n, flip_idx, Kcorn) ! ---- Strides for state-only linear indices (1..N_a) ---- call compute_strides(dims, stride) ! Low-corner linear indices in 1, ..., N_a allocate(a(N_om)) call compute_linear_indices(ind, stride, a) ! Fetch it into 1, ..., N_om do concurrent (j=1:N_a, i=1:N_e) a((j-1_int32)*N_e+i) = (a((j-1_int32)*N_e+i)-1)*N_e+i end do ! Coefficient updates and hard dimensions allocate(r_up(N_om, n), r_down(N_om, n), is_hard_one(N_om, n), is_hard_zero(N_om, n)) call compute_coefficient_updates(w, r_up, r_down, is_hard_one, is_hard_zero) ! Low corner initial weight allocate(beta_0(N_om)) beta_0 = product(w, dim=2, mask=(.not. is_hard_zero)) ! ---- Expectations ---- allocate(sk(n), beta_k(N_om), acc(S,N_om), inc(S), gamma(N_om)) acc = 0.0_real64 stride = stride*N_e do k=1, n beta_k = inv_h(:,k)*beta_0*om where (.not. is_hard_zero(:,k)) beta_k = beta_k / w(:,k) do i=1, N_e do j=1,N_a sk = .false. lin = (j-1)*N_e + i z = count(is_hard_zero(lin,:)) m = a(lin) inc = -beta_k(lin)*a_hat(lin,:,k) if (z == 0_int32) acc(:,m) = acc(:,m) + inc do t=1, Kcorn-1 kf = flip_idx(t) if (kf==k) then inc = -inc if (.not. sk(kf)) then ! lower -> upper on dim kf m = m+stride(kf) sk(kf) = .true. else ! upper -> lower on dim kf m = m-stride(kf) sk(kf) = .false. end if else if (.not. sk(kf)) then ! lower -> upper on dim kf m = m + stride(kf) sk(kf) = .true. if (is_hard_one(lin,kf)) then ! No contribution. The number of ! hard-one dims flipped to upper increases z = z+1_int32 else if (is_hard_zero(lin,kf)) then ! No contribution. The number of ! hard-zero dims flipped to lower decreases z = z-1_int32 else ! Both sides are higher than 0, but ! there is a contribution if and only ! if no hard dims is flipped to upper inc = inc*r_up(lin,kf) end if else ! upper -> lower on dim kf m = m - stride(kf) sk(kf) = .false. if (is_hard_one(lin,kf)) then ! The number of hard-one dims flipped to upper decreases z = z-1_int32 else if (is_hard_zero(lin,kf)) then ! The number of hard-zero dims flipped to lower increases z = z+1_int32 else ! Both sides are higher than 0, but ! there is a contribution if and only ! if no hard dims is flipped to upper inc = inc*r_down(lin,kf) end if end if end if if (z==0_int32) acc(:,m) = acc(:,m)+inc end do end do end do end do ! Weighting with the Markov kernel allocate(rhs(S*N_a,N_e)) do concurrent(l=1:S,j=1:N_a,i=1:N_e) rhs((l-1)*N_a+j,i) = acc(l,(j-1)*N_e+i) end do D_mat(1:N_e,1:(N_a*S))=>D call matmul_add("T", "T", 1.0_real64, Mu, rhs, 0.0_real64, D_mat) end subroutine compute_curlyDs_nd end module curlyDs