How to make `acceptable_obj_change_tol` work? #631

type-null · 2023-01-20T21:52:39Z

type-null
Jan 20, 2023

I found my objective be flat between iterations so wanted to terminate it early based on the percentage change of the objective between iterations. I found acceptable_obj_change_tol option from https://coin-or.github.io/Ipopt/OPTIONS.html.

It says its default value is 10^20. But I tried setting acceptable_obj_change_tol to 10^500 and 10^-500, nothing seemed happen. It still do not stop even when I set the acceptable_iter to 5.

Here is the first 30 iterations with acceptable_obj_change_tol = 1e-500.

This is Ipopt version 3.14.11, running with linear solver ma97.

Number of nonzeros in equality constraint Jacobian...:    77700
Number of nonzeros in inequality constraint Jacobian.:    27380
Number of nonzeros in Lagrangian Hessian.............:        0

Total number of variables............................:      370
                     variables with only lower bounds:       37
                variables with lower and upper bounds:      167
                     variables with only upper bounds:        0
Total number of equality constraints.................:      210
Total number of inequality constraints...............:       74
        inequality constraints with only lower bounds:       74
   inequality constraints with lower and upper bounds:        0
        inequality constraints with only upper bounds:        0

iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
   0 -4.6174315e+04 1.39e+02 1.86e+00   0.0 0.00e+00    -  0.00e+00 0.00e+00   0
   1 -4.6231622e+04 1.39e+02 1.46e+01  -4.2 1.15e+03    -  1.56e-02 9.84e-04f  1
   2 -4.6233102e+04 1.39e+02 1.55e+01  -0.3 1.47e+03    -  7.51e-04 2.40e-04f  1
   3 -4.6234870e+04 1.39e+02 1.28e+02  -0.1 1.39e+03    -  7.75e-03 3.78e-04f  1
   4 -4.6268767e+04 1.38e+02 1.17e+03  -0.1 1.39e+03    -  4.75e-02 1.08e-02f  1
   5 -4.6252092e+04 1.34e+02 6.39e+03   0.3 1.37e+03    -  7.59e-02 2.79e-02f  1
   6 -4.6179486e+04 1.27e+02 3.82e+04   0.7 1.30e+03    -  1.72e-01 4.79e-02f  1
   7 -4.6155402e+04 1.19e+02 5.39e+04   0.5 1.17e+03    -  1.72e-01 6.87e-02f  1
   8 -4.5742218e+04 1.18e+02 1.21e+07   2.3 1.23e+03    -  9.08e-01 8.64e-03f  1
   9 -4.5746556e+04 1.16e+02 9.87e+06   0.9 9.95e+02    -  2.44e-01 1.14e-02h  1
iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
  10 -2.7267327e+04 1.92e+01 3.79e+06   2.0 1.05e+03    -  5.58e-01 8.35e-01f  1
  11 -2.7104740e+04 1.45e+01 1.61e+06   2.0 3.10e+02    -  5.57e-01 2.41e-01h  1
  12 -2.6599621e+04 4.61e-01 3.96e+05   2.0 2.41e+02    -  7.32e-01 9.68e-01h  1
  13 -2.5934027e+04 4.45e-01 4.53e+07   2.2 6.54e+01    -  1.00e+00 3.58e-02f  1
  14 -2.9832864e+04 8.85e-02 2.50e+06   1.9 3.11e+01    -  1.00e+00 9.62e-01h  1
  15 -3.7297683e+04 2.58e-02 9.61e+06   1.5 3.99e+01    -  8.81e-01 7.00e-01f  1
  16 -4.5499460e+04 2.67e-03 3.50e+05   0.5 3.09e+01    -  1.00e+00 9.95e-01f  1
  17 -4.5864138e+04 7.21e-04 9.11e+05  -0.0 4.08e+01    -  1.00e+00 9.63e-01f  1
  18 -4.6142299e+04 5.43e-03 2.01e+01  -0.4 2.68e+01    -  1.00e+00 1.00e+00f  1
  19 -4.6234709e+04 5.52e-03 1.40e+04  -0.6 1.05e+01    -  9.81e-01 1.00e+00f  1
iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
  20 -4.6296392e+04 8.35e-03 4.25e+01  -0.9 4.76e+00    -  1.00e+00 1.00e+00f  1
  21 -4.6250560e+04 4.87e-02 1.09e+05  -0.6 1.66e+01    -  1.00e+00 8.49e-01f  1
  22 -4.6284598e+04 1.92e-02 1.31e+05  -0.7 7.05e+00    -  5.64e-01 1.00e+00f  1
  23 -4.6291279e+04 3.22e-02 4.07e+04  -0.7 8.82e+00    -  6.90e-01 5.73e-01f  1
  24 -4.6287744e+04 3.01e-02 3.39e+04  -0.7 3.16e+00    -  1.67e-01 5.34e-01h  1
  25 -4.6286423e+04 2.01e-02 2.47e+04  -0.7 8.45e-01    -  2.70e-01 8.77e-01h  1
  26 -4.6279553e+04 3.87e-02 2.18e+04  -0.7 6.12e+00    -  1.18e-01 1.00e+00h  1
  27 -4.6279344e+04 2.93e-02 1.32e+03  -0.7 3.97e+00    -  1.00e+00 8.40e-01f  1
  28 -4.6267139e+04 2.44e-02 5.96e+03  -0.7 3.29e+00    -  1.09e-01 5.94e-01H  1
  29 -4.6268290e+04 2.74e-02 4.50e+03  -0.7 7.46e-01    -  4.39e-01 1.00e+00f  1

Here is the first 30 iterations with acceptable_obj_change_tol = 1e500.

This is Ipopt version 3.14.11, running with linear solver ma97.

Number of nonzeros in equality constraint Jacobian...:    77700
Number of nonzeros in inequality constraint Jacobian.:    27380
Number of nonzeros in Lagrangian Hessian.............:        0

Total number of variables............................:      370
                     variables with only lower bounds:       37
                variables with lower and upper bounds:      167
                     variables with only upper bounds:        0
Total number of equality constraints.................:      210
Total number of inequality constraints...............:       74
        inequality constraints with only lower bounds:       74
   inequality constraints with lower and upper bounds:        0
        inequality constraints with only upper bounds:        0

iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
   0 -4.6174315e+04 1.39e+02 1.86e+00   0.0 0.00e+00    -  0.00e+00 0.00e+00   0
   1 -4.6231622e+04 1.39e+02 1.46e+01  -4.2 1.15e+03    -  1.56e-02 9.84e-04f  1
   2 -4.6233102e+04 1.39e+02 1.55e+01  -0.3 1.47e+03    -  7.51e-04 2.40e-04f  1
   3 -4.6234870e+04 1.39e+02 1.28e+02  -0.1 1.39e+03    -  7.75e-03 3.78e-04f  1
   4 -4.6268767e+04 1.38e+02 1.17e+03  -0.1 1.39e+03    -  4.75e-02 1.08e-02f  1
   5 -4.6252092e+04 1.34e+02 6.39e+03   0.3 1.37e+03    -  7.59e-02 2.79e-02f  1
   6 -4.6179486e+04 1.27e+02 3.82e+04   0.7 1.30e+03    -  1.72e-01 4.79e-02f  1
   7 -4.6155402e+04 1.19e+02 5.39e+04   0.5 1.17e+03    -  1.72e-01 6.87e-02f  1
   8 -4.5742218e+04 1.18e+02 1.21e+07   2.3 1.23e+03    -  9.08e-01 8.64e-03f  1
   9 -4.5746556e+04 1.16e+02 9.87e+06   0.9 9.95e+02    -  2.44e-01 1.14e-02h  1
iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
  10 -2.7267327e+04 1.92e+01 3.79e+06   2.0 1.05e+03    -  5.58e-01 8.35e-01f  1
  11 -2.7104740e+04 1.45e+01 1.61e+06   2.0 3.10e+02    -  5.57e-01 2.41e-01h  1
  12 -2.6599621e+04 4.61e-01 3.96e+05   2.0 2.41e+02    -  7.32e-01 9.68e-01h  1
  13 -2.5934027e+04 4.45e-01 4.53e+07   2.2 6.54e+01    -  1.00e+00 3.58e-02f  1
  14 -2.9832864e+04 8.85e-02 2.50e+06   1.9 3.11e+01    -  1.00e+00 9.62e-01h  1
  15 -3.7297683e+04 2.58e-02 9.61e+06   1.5 3.99e+01    -  8.81e-01 7.00e-01f  1
  16 -4.5499460e+04 2.67e-03 3.50e+05   0.5 3.09e+01    -  1.00e+00 9.95e-01f  1
  17 -4.5864138e+04 7.21e-04 9.11e+05  -0.0 4.08e+01    -  1.00e+00 9.63e-01f  1
  18 -4.6142299e+04 5.43e-03 2.01e+01  -0.4 2.68e+01    -  1.00e+00 1.00e+00f  1
  19 -4.6234709e+04 5.52e-03 1.40e+04  -0.6 1.05e+01    -  9.81e-01 1.00e+00f  1
iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
  20 -4.6296392e+04 8.35e-03 4.25e+01  -0.9 4.76e+00    -  1.00e+00 1.00e+00f  1
  21 -4.6250560e+04 4.87e-02 1.09e+05  -0.6 1.66e+01    -  1.00e+00 8.49e-01f  1
  22 -4.6284598e+04 1.92e-02 1.31e+05  -0.7 7.05e+00    -  5.64e-01 1.00e+00f  1
  23 -4.6291279e+04 3.22e-02 4.07e+04  -0.7 8.82e+00    -  6.90e-01 5.73e-01f  1
  24 -4.6287744e+04 3.01e-02 3.39e+04  -0.7 3.16e+00    -  1.67e-01 5.34e-01h  1
  25 -4.6286423e+04 2.01e-02 2.47e+04  -0.7 8.45e-01    -  2.70e-01 8.77e-01h  1
  26 -4.6279553e+04 3.87e-02 2.18e+04  -0.7 6.12e+00    -  1.18e-01 1.00e+00h  1
  27 -4.6279344e+04 2.93e-02 1.32e+03  -0.7 3.97e+00    -  1.00e+00 8.40e-01f  1
  28 -4.6267139e+04 2.44e-02 5.96e+03  -0.7 3.29e+00    -  1.09e-01 5.94e-01H  1
  29 -4.6268290e+04 2.74e-02 4.50e+03  -0.7 7.46e-01    -  4.39e-01 1.00e+00f  1

As you can see, the objective did not move much in either cases. I wonder what is the correct setting to make it stop at a acceptable (feasible) point in fewer iterations so I can speed up the program. Thanks!

Answered by svigerske

Jan 25, 2023

The acceptable_obj_change_tol is only one criterion that needs to be satisfied to make Ipopt stop with an "acceptable" point. In addition, various infeasibility measures need to be satisfied and this needs to happen for at least acceptable_iter many subsequent iterations.
Check https://coin-or.github.io/Ipopt/OPTIONS.html#OPT_Termination for all other options you may need to set. In your log, the primal and dual infeasibility are still quite high, preventing Ipopt from stopping.

Looking at the code for the acceptable check may also help:

Ipopt/src/Algorithm/IpOptErrorConvCheck.cpp

Lines 286 to 347 in e622421

     bool OptimalityErrorConvergenceCheck::CurrentIsAcceptable()  
   {  
   

View full answer

svigerske · 2023-01-25T04:43:31Z

svigerske
Jan 25, 2023
Maintainer

The acceptable_obj_change_tol is only one criterion that needs to be satisfied to make Ipopt stop with an "acceptable" point. In addition, various infeasibility measures need to be satisfied and this needs to happen for at least acceptable_iter many subsequent iterations.
Check https://coin-or.github.io/Ipopt/OPTIONS.html#OPT_Termination for all other options you may need to set. In your log, the primal and dual infeasibility are still quite high, preventing Ipopt from stopping.

Looking at the code for the acceptable check may also help:

Ipopt/src/Algorithm/IpOptErrorConvCheck.cpp

Lines 286 to 347 in e622421

    
           bool OptimalityErrorConvergenceCheck::CurrentIsAcceptable() 
        
           { 
        
              DBG_START_METH("OptimalityErrorConvergenceCheck::CurrentIsAcceptable", 
        
                             dbg_verbosity); 
        
              Number overall_error = IpCq().curr_nlp_error(); 
        
              Number dual_inf = IpCq().unscaled_curr_dual_infeasibility(NORM_MAX); 
        
              Number constr_viol = IpCq().unscaled_curr_nlp_constraint_violation(NORM_MAX); 
        
              Number compl_inf = IpCq().unscaled_curr_complementarity(mu_target_, NORM_MAX); 
        
              if( IpData().iter_count() != last_obj_val_iter_ ) 
        
              { 
        
                 //Jnlst().Printf(J_MOREDETAILED, J_MAIN, 
        
                 //               "obj val update iter = %" IPOPT_INDEX_FORMAT "\n", IpData().iter_count()); 
        
                 last_obj_val_ = curr_obj_val_; 
        
                 curr_obj_val_ = IpCq().curr_f(); 
        
                 last_obj_val_iter_ = IpData().iter_count(); 
        
              } 
        
              DBG_PRINT((1, "overall_error = %e\n", overall_error)); 
        
              DBG_PRINT((1, "dual_inf = %e\n", dual_inf)); 
        
              DBG_PRINT((1, "constr_viol = %e\n", constr_viol)); 
        
              DBG_PRINT((1, "compl_inf = %e\n", compl_inf)); 
        
              DBG_PRINT((1, "acceptable_tol_ = %e\n", acceptable_tol_)); 
        
              DBG_PRINT((1, "acceptable_dual_inf_tol_ = %e\n", acceptable_dual_inf_tol_)); 
        
              DBG_PRINT((1, "acceptable_constr_viol_tol_ = %e\n", acceptable_constr_viol_tol_)); 
        
              DBG_PRINT((1, "acceptable_compl_inf_tol_ = %e\n", acceptable_compl_inf_tol_)); 
        
              if( IpData().curr()->x()->Dim() == IpData().curr()->y_c()->Dim() ) 
        
              { 
        
                 // the problem is square, there is no point in looking at dual 
        
                 // infeasibility and complementarity as termination criterion 
        
                 acceptable_dual_inf_tol_ = 1e300; 
        
                 acceptable_compl_inf_tol_ = 1e300; 
        
              } 
        
              if( Jnlst().ProduceOutput(J_MOREDETAILED, J_MAIN) ) 
        
              { 
        
                 Jnlst().Printf(J_MOREDETAILED, J_MAIN, 
        
                                "Acceptable Check:\n"); 
        
                 Jnlst().Printf(J_MOREDETAILED, J_MAIN, 
        
                                "  overall_error = %23.16e   acceptable_tol_             = %23.16e\n", overall_error, acceptable_tol_); 
        
                 Jnlst().Printf(J_MOREDETAILED, J_MAIN, 
        
                                "  dual_inf      = %23.16e   acceptable_dual_inf_tol_    = %23.16e\n", dual_inf, acceptable_dual_inf_tol_); 
        
                 Jnlst().Printf(J_MOREDETAILED, J_MAIN, 
        
                                "  constr_viol   = %23.16e   acceptable_constr_viol_tol_ = %23.16e\n", constr_viol, acceptable_constr_viol_tol_); 
        
                 Jnlst().Printf(J_MOREDETAILED, J_MAIN, 
        
                                "  compl_inf     = %23.16e   acceptable_compl_inf_tol_   = %23.16e\n", compl_inf, acceptable_compl_inf_tol_); 
        
                 Jnlst().Printf(J_MOREDETAILED, J_MAIN, 
        
                                "  curr_obj_val_ = %23.16e   last_obj_val                = %23.16e\n", curr_obj_val_, last_obj_val_); 
        
                 Jnlst().Printf(J_MOREDETAILED, J_MAIN, 
        
                                "  std::abs(curr_obj_val_-last_obj_val_)/Max(1., std::abs(curr_obj_val_)) = %23.16e acceptable_obj_change_tol_ = %23.16e\n", 
        
                                std::abs(curr_obj_val_ - last_obj_val_) / Max(Number(1.), std::abs(curr_obj_val_)), acceptable_obj_change_tol_); 
        
                 //Jnlst().Printf(J_MOREDETAILED, J_MAIN, 
        
                 //               "test iter = %" IPOPT_INDEX_FORMAT "\n", IpData().iter_count()); 
        
              } 
        
              return (overall_error <= acceptable_tol_ && dual_inf <= acceptable_dual_inf_tol_ 
        
                      && constr_viol <= acceptable_constr_viol_tol_ && compl_inf <= acceptable_compl_inf_tol_ 
        
                      && std::abs(curr_obj_val_ - last_obj_val_) / Max(Number(1.), std::abs(curr_obj_val_)) <= acceptable_obj_change_tol_); 
        
           }

You see that if you increase the print_level to 7, you will get additional info on whether the check was successful or not to the log.

0 replies

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

How to make `acceptable_obj_change_tol` work? #631

{{title}}

{{editor}}'s edit

{{editor}}'s edit

Replies: 1 comment

{{title}}

Select a reply

	bool OptimalityErrorConvergenceCheck::CurrentIsAcceptable()
	{

How to make acceptable_obj_change_tol work? #631

type-null Jan 20, 2023

Replies: 1 comment

svigerske Jan 25, 2023 Maintainer

How to make `acceptable_obj_change_tol` work? #631

type-null
Jan 20, 2023

svigerske
Jan 25, 2023
Maintainer