/* nag_opt_estimate_deriv (e04xac) Example Program.
 *
 * Copyright 2023 Numerical Algorithms Group.
 *
 * Mark 29.3, 2023.
 *
 */
#include <nag.h>
#include <stdio.h>
#include <string.h>

#ifdef __cplusplus
extern "C" {
#endif
static void NAG_CALL objfun(Integer n, const double x[], double *objf,
                            double g[], Nag_Comm *comm);
#ifdef __cplusplus
}
#endif

#define H(I, J) h[(I)*tdh + J]

int main(void) {
  Integer exit_status = 0, i, j, n, tdh;
  double *g = 0, *h = 0, *h_central = 0, *h_forward = 0, *hess_diag = 0, objf,
         *x = 0;
  Nag_Comm comm;
  Nag_DerivInfo *deriv_info = 0;
  Nag_E04_Opt options;
  NagError fail;

  INIT_FAIL(fail);

  printf("nag_opt_estimate_deriv (e04xac) Example Program Results\n");
  n = 4;

  if (!(x = NAG_ALLOC(n, double)) || !(h_central = NAG_ALLOC(n, double)) ||
      !(h_forward = NAG_ALLOC(n, double)) || !(g = NAG_ALLOC(n, double)) ||
      !(h = NAG_ALLOC(n * n, double)) || !(hess_diag = NAG_ALLOC(n, double)) ||
      !(deriv_info = NAG_ALLOC(n, Nag_DerivInfo))) {
    printf("Allocation failure\n");
    exit_status = -1;
    goto END;
  }
  tdh = n;

  x[0] = 3.0;
  x[1] = -1.0;
  x[2] = 0.0;
  x[3] = 1.0;

  /* nag_opt_init (e04xxc).
   * Initialization function for option setting
   */
  nag_opt_init(&options);
  options.list = Nag_FALSE;
  options.print_deriv = Nag_D_NoPrint;
  options.deriv_want = Nag_Grad_HessDiag;

  printf("\nEstimate gradient and Hessian diagonals given function only\n");

  /* Note: it is acceptable to pass an array of length n (hess_diag)
   * as the Hessian parameter in this case.
   */
  /* nag_opt_estimate_deriv (e04xac), see above. */
  nag_opt_estimate_deriv(n, x, objfun, &objf, g, h_forward, h_central,
                         hess_diag, tdh, deriv_info, &options, &comm, &fail);
  if (fail.code != NE_NOERROR) {
    printf("Error from nag_opt_estimate_deriv (e04xac).\n%s\n", fail.message);
    exit_status = 1;
    goto END;
  }

  printf("\nFunction value: %13.4e\n", objf);
  printf("Estimated gradient vector\n");
  for (i = 0; i < n; ++i)
    printf("%13.4e  ", g[i]);
  printf("\nEstimated Hessian matrix diagonal\n");
  for (i = 0; i < n; ++i)
    printf("%13.4e  ", hess_diag[i]);
  printf("\n");

  options.deriv_want = Nag_HessFull;

  printf("\nEstimate full Hessian given function and gradients\n");
  /* nag_opt_estimate_deriv (e04xac), see above. */
  nag_opt_estimate_deriv(n, x, objfun, &objf, g, h_forward, h_central, h, tdh,
                         deriv_info, &options, &comm, &fail);
  if (fail.code != NE_NOERROR) {
    printf("Error from nag_opt_estimate_deriv (e04xac).\n%s\n", fail.message);
    exit_status = 1;
    goto END;
  }

  printf("\nFunction value: %13.4e\n", objf);
  printf("Computed gradient vector\n");
  for (i = 0; i < n; ++i)
    printf("%13.4e  ", g[i]);
  printf("\nEstimated Hessian matrix\n");
  for (i = 0; i < n; ++i) {
    for (j = 0; j < n; ++j)
      printf("%13.4e  ", H(i, j));
    printf("\n");
  }

END:
  NAG_FREE(x);
  NAG_FREE(h_central);
  NAG_FREE(h_forward);
  NAG_FREE(g);
  NAG_FREE(h);
  NAG_FREE(hess_diag);
  NAG_FREE(deriv_info);

  return exit_status;
}

static void NAG_CALL objfun(Integer n, const double x[], double *objf,
                            double g[], Nag_Comm *comm) {
  double a, asq, b, bsq, c, csq, d, dsq;

  a = x[0] + 10.0 * x[1];
  b = x[2] - x[3];
  c = x[1] - 2.0 * x[2];
  d = x[0] - x[3];
  asq = a * a;
  bsq = b * b;
  csq = c * c;
  dsq = d * d;
  *objf = asq + 5.0 * bsq + csq * csq + 10.0 * dsq * dsq;
  if (comm->flag == 2) {
    g[0] = 2.0 * a + 40.0 * d * dsq;
    g[1] = 20.0 * a + 4.0 * c * csq;
    g[2] = 10.0 * b - 8.0 * c * csq;
    g[3] = -10.0 * b - 40.0 * d * dsq;
  }
}

/* objfun */