! 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 ! ! Parameter calibration using Broyden quasi-Newton solver module calibration use iso_fortran_env, only: real64, int32 use iso_c_binding use matlab_mex ! For mexPrintf use calibration_types use aggregate_residuals_tensor, only: agg_resid_tensor ! Broyden quasi-Newton solver for parameter calibration use broyden_solver, only: broyden_solve, & BROYDEN_SUCCESS, BROYDEN_MAX_ITER, BROYDEN_SINGULAR_JAC, & BROYDEN_MAX_BACKTRACK, BROYDEN_FCN_ERROR implicit none (type, external) private public :: calibrate contains !--------------------------------------------------------------------------- ! Calibrate parameters to match targets using Broyden quasi-Newton method ! ! This implements the pattern from Python SSJ hank_1a.py lines 223-224: ! (beta, vphi), _ = utils.broyden_solver(res, initial_guess) ! ! where res() calls household_trans.ss() which is like our solve_steady_state() ! ! Returns: status (0=success, >0=error code) !--------------------------------------------------------------------------- subroutine calibrate(input, output) type(calibration_input), intent(inout) :: input type(calibration_output), intent(inout) :: output integer(int32) :: status ! Local variables for Broyden solver real(real64), allocatable :: x(:), fvec(:) integer(int32) :: info, nfev integer :: i character(len=256) :: msg ! Handle no-calibration case: just compute residuals if (input%dims%n_free_parameters == 0) then status = agg_resid_tensor(input, output) output%converged = (status == 0) output%iterations = 1 ! residual_norm computed from ALL aggregate residuals output%residual_norm = maxval(abs(output%agg_output%residuals)) if (input%cal_verbosity == 2) & call mexPrintf('No calibration parameters - computing residuals only'//NEW_LINE('A')) return end if if (input%cal_verbosity == 2) & call mexPrintf('=== Parameter Calibration (Broyden) ==='//NEW_LINE('A')) ! Allocate local solution and residual vectors allocate(x(input%dims%n_free_parameters), fvec(input%dims%n_free_parameters)) ! Set initial guess (work directly in θ-space, no transformation) do i = 1, input%dims%n_free_parameters x(i) = input%free_parameters_init(i) end do ! Print initial guesses and bounds if (input%cal_verbosity == 2) then call mexPrintf(' Initial parameter guesses:'//NEW_LINE('A')) do i = 1, input%dims%n_free_parameters write(msg, '(A,A,A,ES12.5,A,ES12.5,A,ES12.5,A)') & ' ', trim(input%free_parameters_names(i)), ' = ', input%free_parameters_init(i), & ' (lb=', input%free_parameters_bounds(1, i), & ', ub=', input%free_parameters_bounds(2, i), ')' call mexPrintf(trim(msg)//NEW_LINE('A')) end do end if ! Solve using Broyden method with bounds (clamping) info = broyden_solve(input%broyden_ws, x, fvec, residual_wrapper, & input%solver_opts%tol, input%solver_opts%max_iter, nfev, & verbosity=input%cal_verbosity, & lower_bounds=input%free_parameters_bounds(1,:), & upper_bounds=input%free_parameters_bounds(2,:)) ! Check convergence if (info == BROYDEN_SUCCESS) then output%converged = .true. if (input%cal_verbosity == 2) & call mexPrintf('Calibration converged successfully!'//NEW_LINE('A')) else output%converged = .false. if (input%cal_verbosity == 2) then call mexPrintf('Calibration failed to converge'//NEW_LINE('A')) call print_broyden_error(info) end if end if ! Extract calibrated parameters (already in θ-space) if (.not. allocated(output%params)) allocate(output%params(input%dims%n_free_parameters)) do i = 1, input%dims%n_free_parameters output%params(i) = x(i) end do ! Print calibrated parameters if (input%cal_verbosity == 2 .and. output%converged) then call mexPrintf('Calibrated parameters:'//NEW_LINE('A')) do i = 1, input%dims%n_free_parameters write(msg, '(A,A,A,ES12.5)') ' ', trim(input%free_parameters_names(i)), ' = ', output%params(i) call mexPrintf(trim(msg)//NEW_LINE('A')) end do write(msg, '(A,I4,A)') 'Converged in ', nfev, ' function evaluations' call mexPrintf(trim(msg)//NEW_LINE('A')) end if ! Extract calibration target equations residuals if (.not. allocated(output%residuals)) allocate(output%residuals(input%dims%n_free_parameters)) output%residuals = fvec output%residual_norm = maxval(abs(output%residuals)) ! Store iteration count output%iterations = nfev if (input%cal_verbosity == 2) call mexPrintf('Calibration complete'//NEW_LINE('A')) contains ! Residual wrapper (nested procedure to access input/output) subroutine residual_wrapper(nvar, x_params, resid, iflag) integer, intent(in) :: nvar real(real64), intent(in) :: x_params(nvar) real(real64), intent(out) :: resid(nvar) integer, intent(inout) :: iflag integer(int32) :: ss_status, k, n_target_eqs ! Update parameters with current guess (already in θ-space) do k = 1, nvar input%params(input%free_parameters_ind(k)) = x_params(k) end do ! Solve full steady state with these parameters (time iteration + distribution + aggregation) ss_status = agg_resid_tensor(input, output) if (ss_status /= 0) then iflag = -1 ! Signal error to Broyden solver return end if ! Populate resid with calibration target equations residuals n_target_eqs = size(input%target_equations, 1) do k = 1, n_target_eqs resid(k) = output%agg_output%residuals(input%target_equations(k)) end do ! Print iteration info if verbosity == 2 (use parameter values from params) if (input%cal_verbosity == 2) & call print_calibration_iteration(nvar, input%params, input%free_parameters_ind, & input%free_parameters_names) end subroutine residual_wrapper subroutine print_broyden_error(code) integer(int32), intent(in) :: code select case (code) case (BROYDEN_MAX_ITER) call mexPrintf(' Broyden: Maximum iterations exceeded'//NEW_LINE('A')) case (BROYDEN_SINGULAR_JAC) call mexPrintf(' Broyden: Singular Jacobian encountered'//NEW_LINE('A')) case (BROYDEN_MAX_BACKTRACK) call mexPrintf(' Broyden: Too many backtracks'//NEW_LINE('A')) case (BROYDEN_FCN_ERROR) call mexPrintf(' Broyden: Function evaluation error'//NEW_LINE('A')) case default call mexPrintf(' Broyden: Unknown error code'//NEW_LINE('A')) end select end subroutine print_broyden_error subroutine print_calibration_iteration(n, params, free_parameters_ind, param_names) integer, intent(in) :: n real(real64), intent(in) :: params(:) ! Full params array integer(int32), intent(in) :: free_parameters_ind(:) ! Indices of free parameters in params character(len=*), intent(in) :: param_names(:) character(len=256) :: msg integer :: i do i = 1, n write(msg, '(A,A,A,ES12.5)') ' ', trim(param_names(i)), ' = ', & params(free_parameters_ind(i)) call mexPrintf(trim(msg)//NEW_LINE('A')) end do end subroutine print_calibration_iteration end subroutine calibrate end module calibration