Variational Inequality (VI)

Variational Inequality (VI)#

Problem statement#

Given

  • an integer \(n\) , the dimension of the ambient space,

  • a mapping \(F\colon \mathrm{I\!R}^n \rightarrow \mathrm{I\!R}^n\)

  • a set \({X} \in {{\mathrm{I\!R}}}^n\)

the variational inequality problem consists in finding a vector \(z\in{{\mathrm{I\!R}}}^n\) , such that

\[\begin{equation*} F(z)^T(y-z) \geq 0,\quad \text{ for all } y \in X \end{equation*}\]

or equivalently,

\[\begin{equation*} - F(z) \in \mathcal{N}_X(z) \end{equation*}\]

where \(\mathcal{N}_X\) is the normal cone to \(X\) at \(z\) .

References : [6], [1].

Implementation in numerics#

Structure to define the problem: VariationalInequality.

The generic driver for all VI is variationalInequality_driver().

solvers list VI_SOLVER

VI Available solvers#

Extra gradient (SICONOS_VI_EG)#

Extra Gradient solver forvariational inequality problem based on the De Saxce Formulation

driver variationalInequality_ExtraGradient()

parameters:

  • iparam[SICONOS_IPARAM_MAX_ITER] = 20000

  • iparam[SICONOS_VI_IPARAM_ERROR_EVALUATION] = SICONOS_VI_ERROR_EVALUATION_LIGHT_WITH_FULL_FINAL

  • iparam[SICONOS_VI_IPARAM_ERROR_EVALUATION_FREQUENCY] (set but not used)

  • iparam[SICONOS_VI_IPARAM_LINESEARCH_METHOD] = SICONOS_VI_LS_ARMIJO

    allowed values :

    • SICONOS_VI_LS_ARMIJO : Armijo rule with Khotbotov ratio (default)

    • SICONOS_VI_LS_SOLODOV : Armijo rule with Solodov.Tseng ratio

    • SICONOS_VI_LS_HANSUN : Armijo rule with Han.Sun ratio

  • iparam[SICONOS_VI_IPARAM_ACTIVATE_UPDATE] = 0;

  • iparam[SICONOS_VI_IPARAM_DECREASE_RHO] = 0;

  • dparam[SICONOS_DPARAM_TOL] = 1e-3, in-out parameter

  • dparam[SICONOS_VI_DPARAM_RHO] = -1., in-out parameter

  • dparam[SICONOS_VI_DPARAM_LS_TAU] = 2/3

  • dparam[SICONOS_VI_DPARAM_LS_TAUINV] = 3/2

  • dparam[SICONOS_VI_DPARAM_LS_L] = 0.9

  • dparam[SICONOS_VI_DPARAM_LS_LMIN] = 0.3

Fixed-point projection (SICONOS_VI_FPP)#

Fixed Point Projection solver for variational inequality problem based on the De Saxce Formulation.

driver: variationalInequality_FixedPointProjection()

parameters: same as SICONOS_VI_EG.

Hyperplane projection (SICONOS_VI_HP)#

driver: variationalInequality_HyperplaneProjection()

parameters:

  • iparam[SICONOS_IPARAM_MAX_ITER] = 20000

  • iparam[SICONOS_VI_IPARAM_LS_MAX_ITER] = 100

  • dparam[SICONOS_DPARAM_TOL] = 1e-3

  • dparam[SICONOS_VI_DPARAM_LS_TAU] = 1.0, tau

  • dparam[SICONOS_VI_DPARAM_SIGMA] = 0.8, sigma

out :

  • iparam[SICONOS_IPARAM_ITER_DONE] : number of iterations

SICONOS_VI_BOX_QI (SICONOS_VI_BOX_QI)#

Solver using the merit function proposed by Qi for box-constrained Newton QI LSA

id:

driver : variationalInequality_box_newton_QiLSA()

parameters:

  • iparam[SICONOS_IPARAM_MAX_ITER] = 1000

  • iparam[SICONOS_IPARAM_PREALLOC] = 0

  • iparam[SICONOS_IPARAM_STOPPING_CRITERION] = SICONOS_STOPPING_CRITERION_USER_ROUTINE;

  • iparam[SICONOS_IPARAM_LSA_NONMONOTONE_LS] = 0

  • iparam[SICONOS_IPARAM_LSA_NONMONOTONE_LS_M] = 0 (set but not used)

  • iparam[SICONOS_IPARAM_LSA_FORCE_ARCSEARCH] = 1

  • dparam[SICONOS_DPARAM_LSA_ALPHA_MIN] = 1e-16

  • dparam[SICONOS_DPARAM_TOL] = 1e-10

SICONOS_VI_BOX_AVI_LSA (SICONOS_VI_BOX_AVI_LSA)#

driver : vi_box_AVI_LSA()

parameters:

  • iparam[SICONOS_IPARAM_MAX_ITER] = 100

  • iparam[SICONOS_IPARAM_LSA_FORCE_ARCSEARCH] = 1

  • iparam[SICONOS_IPARAM_LSA_NONMONOTONE_LS] = 0

  • iparam[SICONOS_IPARAM_LSA_NONMONOTONE_LS_M] = 0 (set but not used)

  • iparam[SICONOS_IPARAM_STOPPING_CRITERION] = SICONOS_STOPPING_CRITERION_USER_ROUTINE;

  • dparam[SICONOS_DPARAM_TOL] = 1e-12

  • dparam[SICONOS_DPARAM_LSA_ALPHA_MIN] = 1e-16

internal solver : SICONOS_RELAY_AVI_CAOFERRIS

SICONOS_VI_BOX_PATH (SICONOS_VI_BOX_PATH)#

driver : vi_box_path()

parameters:

  • iparam[SICONOS_IPARAM_MAX_ITER] = 10000

  • dparam[SICONOS_DPARAM_TOL] = 1e-12

Contents