nls_subproblem_control_type structure#

    struct nls_subproblem_control_type{T}
      error::Int32
      out::Int32
      print_level::Int32
      start_print::Int32
      stop_print::Int32
      print_gap::Int32
      maxit::Int32
      alive_unit::Int32
      alive_file::NTuple{31,Cchar}
      jacobian_available::Int32
      hessian_available::Int32
      model::Int32
      norm::Int32
      non_monotone::Int32
      weight_update_strategy::Int32
      stop_c_absolute::T
      stop_c_relative::T
      stop_g_absolute::T
      stop_g_relative::T
      stop_s::T
      power::T
      initial_weight::T
      minimum_weight::T
      initial_inner_weight::T
      eta_successful::T
      eta_very_successful::T
      eta_too_successful::T
      weight_decrease_min::T
      weight_decrease::T
      weight_increase::T
      weight_increase_max::T
      reduce_gap::T
      tiny_gap::T
      large_root::T
      switch_to_newton::T
      cpu_time_limit::T
      clock_time_limit::T
      subproblem_direct::Bool
      renormalize_weight::Bool
      magic_step::Bool
      print_obj::Bool
      space_critical::Bool
      deallocate_error_fatal::Bool
      prefix::NTuple{31,Cchar}
      rqs_control::rqs_control_type{T}
      glrt_control::glrt_control_type{T}
      psls_control::psls_control_type{T}
      bsc_control::bsc_control_type
      roots_control::roots_control_type{T}

detailed documentation#

subproblem_control derived type as a Julia structure

components#

Int32 error

error and warning diagnostics occur on stream error

Int32 out

general output occurs on stream out

Int32 print_level

the level of output required.

\(\leq\) 0 gives no output,
= 1 gives a one-line summary for every iteration,
= 2 gives a summary of the inner iteration for each iteration,
\(\geq\) 3 gives increasingly verbose (debugging) output

Int32 start_print

any printing will start on this iteration

Int32 stop_print

any printing will stop on this iteration

Int32 print_gap

the number of iterations between printing

Int32 maxit

the maximum number of iterations performed

Int32 alive_unit

removal of the file alive_file from unit alive_unit terminates execution

char alive_file[31]

see alive_unit

Int32 jacobian_available

is the Jacobian matrix of first derivatives available (\(\geq\) 2), is access only via matrix-vector products (=1) or is it not available (\(\leq\) 0) ?

Int32 hessian_available

is the Hessian matrix of second derivatives available (\(\geq\) 2), is access only via matrix-vector products (=1) or is it not available (\(\leq\) 0) ?

Int32 model

the model used.

Possible values are

0 dynamic (not yet implemented)
1 first-order (no Hessian)
2 barely second-order (identity Hessian)
3 Gauss-Newton (\(J^T J\) Hessian)
4 second-order (exact Hessian)
5 Gauss-Newton to Newton transition
6 tensor Gauss-Newton treated as a least-squares model
7 tensor Gauss-Newton treated as a general model
8 tensor Gauss-Newton transition from a least-squares to a general mode

Int32 norm

the regularization norm used.

The norm is defined via \(\|v\|^2 = v^T S v\), and will define the preconditioner used for iterative methods. Possible values for \(S\) are

-3 user’s own regularization norm
-2 \(S\) = limited-memory BFGS matrix (with .PSLS_control.lbfgs_vectors history) (not yet implemented)
-1 identity (= Euclidan two-norm)
0 automatic (not yet implemented)
1 diagonal, \(S\) = diag( max(\(J^TJ\) Hessian, .PSLS_control.min_diagonal ) )
2 diagonal, \(S\) = diag( max( Hessian, .PSLS_control.min_diagonal ) )
3 banded, \(S\) = band( Hessian ) with semi-bandwidth .PSLS_control.semi_bandwidth
4 re-ordered band, P=band(order(A)) with semi-bandwidth .PSLS_control.semi_bandwidth
5 full factorization, \(S\) = Hessian, Schnabel-Eskow modification
6 full factorization, \(S\) = Hessian, GMPS modification (not yet implemented)
7 incomplete factorization of Hessian, Lin-More’
8 incomplete factorization of Hessian, HSL_MI28
9 incomplete factorization of Hessian, Munskgaard (not yet implemented)
10 expanding band of Hessian (not yet implemented)

Int32 non_monotone

non-monotone \(\leq\) 0 monotone strategy used, anything else non-monotone strategy with this history length used

Int32 weight_update_strategy

define the weight-update strategy: 1 (basic), 2 (reset to zero when very successful), 3 (imitate TR), 4 (increase lower bound), 5 (GPT)

T stop_c_absolute

overall convergence tolerances. The iteration will terminate when \(||c(x)||_2 \leq\) MAX( .stop_c_absolute, .stop_c_relative \(* \|c(x_{\mbox{initial}})\|_2\), or when the norm of the gradient, \(g = J^T(x) c(x) / \|c(x)\|_2\), of \|\|c\|\|_2, satisfies \(\|g\|_2 \leq\) MAX( .stop_g_absolute, .stop_g_relative \(* \|g_{\mbox{initial}}\|_2\), or if the step is less than .stop_s

T stop_c_relative

see stop_c_absolute

T stop_g_absolute

see stop_c_absolute

T stop_g_relative

see stop_c_absolute

T stop_s

see stop_c_absolute

T power

the regularization power (<2 => chosen according to the model)

T initial_weight

initial value for the regularization weight (-ve => \(1/\|g_0\|)\))

T minimum_weight

minimum permitted regularization weight

T initial_inner_weight

initial value for the inner regularization weight for tensor GN (-ve => 0)

T eta_successful

a potential iterate will only be accepted if the actual decrease f - f(x_new) is larger than .eta_successful times that predicted by a quadratic model of the decrease. The regularization weight will be decreaed if this relative decrease is greater than .eta_very_successful but smaller than .eta_too_successful

T eta_very_successful

see eta_successful

T eta_too_successful

see eta_successful

T weight_decrease_min

on very successful iterations, the regularization weight will be reduced by the factor .weight_decrease but no more than .weight_decrease_min while if the iteration is unsucceful, the weight will be increased by a factor .weight_increase but no more than .weight_increase_max (these are delta_1, delta_2, delta3 and delta_max in Gould, Porcelli and Toint, 2011)

T weight_decrease

see weight_decrease_min

T weight_increase

see weight_decrease_min

T weight_increase_max

see weight_decrease_min

T reduce_gap

expert parameters as suggested in Gould, Porcelli and Toint, “Updating t regularization parameter in the adaptive cubic regularization algorithm” RAL-TR-2011-007, Rutherford Appleton Laboratory, England (2011), http://epubs.stfc.ac.uk/bitstream/6181/RAL-TR-2011-007.pdf (these are denoted beta, epsilon_chi and alpha_max in the paper)

T tiny_gap

see reduce_gap

T large_root

see reduce_gap

T switch_to_newton

if the Gauss-Newto to Newton model is specified, switch to Newton as soon as the norm of the gradient g is smaller than switch_to_newton

T cpu_time_limit

the maximum CPU time allowed (-ve means infinite)

T clock_time_limit

the maximum elapsed clock time allowed (-ve means infinite)

Bool subproblem_direct

use a direct (factorization) or (preconditioned) iterative method to find the search direction

Bool renormalize_weight

should the weight be renormalized to account for a change in scaling?

Bool magic_step

allow the user to perform a “magic” step to improve the objective

Bool print_obj

print values of the objective/gradient rather than ||c|| and its gradien

Bool space_critical

if .space_critical true, every effort will be made to use as little space as possible. This may result in longer computation time

Bool deallocate_error_fatal

if .deallocate_error_fatal is true, any array/pointer deallocation error will terminate execution. Otherwise, computation will continue

NTuple{31,Cchar} prefix

all output lines will be prefixed by .prefix(2:LEN(TRIM(.prefix))-1) where .prefix contains the required string enclosed in quotes, e.g. “string” or ‘string’

struct rqs_control_type rqs_control

control parameters for RQS

struct glrt_control_type glrt_control

control parameters for GLRT

struct psls_control_type psls_control

control parameters for PSLS

struct bsc_control_type bsc_control

control parameters for BSC

struct roots_control_type roots_control

control parameters for ROOTS