xref: /petsc/src/tao/complementarity/impls/asls/asfls.c (revision 7d5fd1e4d9337468ad3f05b65b7facdcd2dfd2a4)
1 #include <../src/tao/complementarity/impls/ssls/ssls.h>
2 /*
3    Context for ASXLS
4      -- active-set      - reduced matrices formed
5                           - inherit properties of original system
6      -- semismooth (S)  - function not differentiable
7                         - merit function continuously differentiable
8                         - Fischer-Burmeister reformulation of complementarity
9                           - Billups composition for two finite bounds
10      -- infeasible (I)  - iterates not guaranteed to remain within bounds
11      -- feasible (F)    - iterates guaranteed to remain within bounds
12      -- linesearch (LS) - Armijo rule on direction
13 
14    Many other reformulations are possible and combinations of
15    feasible/infeasible and linesearch/trust region are possible.
16 
17    Basic theory
18      Fischer-Burmeister reformulation is semismooth with a continuously
19      differentiable merit function and strongly semismooth if the F has
20      lipschitz continuous derivatives.
21 
22      Every accumulation point generated by the algorithm is a stationary
23      point for the merit function.  Stationary points of the merit function
24      are solutions of the complementarity problem if
25        a.  the stationary point has a BD-regular subdifferential, or
26        b.  the Schur complement F'/F'_ff is a P_0-matrix where ff is the
27            index set corresponding to the free variables.
28 
29      If one of the accumulation points has a BD-regular subdifferential then
30        a.  the entire sequence converges to this accumulation point at
31            a local q-superlinear rate
32        b.  if in addition the reformulation is strongly semismooth near
33            this accumulation point, then the algorithm converges at a
34            local q-quadratic rate.
35 
36    The theory for the feasible version follows from the feasible descent
37    algorithm framework.
38 
39    References:
40      Billups, "Algorithms for Complementarity Problems and Generalized
41        Equations," Ph.D thesis, University of Wisconsin  Madison, 1995.
42      De Luca, Facchinei, Kanzow, "A Semismooth Equation Approach to the
43        Solution of Nonlinear Complementarity Problems," Mathematical
44        Programming, 75, pages 407439, 1996.
45      Ferris, Kanzow, Munson, "Feasible Descent Algorithms for Mixed
46        Complementarity Problems," Mathematical Programming, 86,
47        pages 475497, 1999.
48      Fischer, "A Special Newton type Optimization Method," Optimization,
49        24, 1992
50      Munson, Facchinei, Ferris, Fischer, Kanzow, "The Semismooth Algorithm
51        for Large Scale Complementarity Problems," Technical Report,
52        University of Wisconsin  Madison, 1999.
53 */
54 
55 static PetscErrorCode TaoSetUp_ASFLS(Tao tao)
56 {
57   TAO_SSLS       *asls = (TAO_SSLS *)tao->data;
58   PetscErrorCode ierr;
59 
60   PetscFunctionBegin;
61   ierr = VecDuplicate(tao->solution,&tao->gradient);CHKERRQ(ierr);
62   ierr = VecDuplicate(tao->solution,&tao->stepdirection);CHKERRQ(ierr);
63   ierr = VecDuplicate(tao->solution,&asls->ff);CHKERRQ(ierr);
64   ierr = VecDuplicate(tao->solution,&asls->dpsi);CHKERRQ(ierr);
65   ierr = VecDuplicate(tao->solution,&asls->da);CHKERRQ(ierr);
66   ierr = VecDuplicate(tao->solution,&asls->db);CHKERRQ(ierr);
67   ierr = VecDuplicate(tao->solution,&asls->t1);CHKERRQ(ierr);
68   ierr = VecDuplicate(tao->solution,&asls->t2);CHKERRQ(ierr);
69   ierr = VecDuplicate(tao->solution, &asls->w);CHKERRQ(ierr);
70   asls->fixed = NULL;
71   asls->free = NULL;
72   asls->J_sub = NULL;
73   asls->Jpre_sub = NULL;
74   asls->r1 = NULL;
75   asls->r2 = NULL;
76   asls->r3 = NULL;
77   asls->dxfree = NULL;
78   PetscFunctionReturn(0);
79 }
80 
81 static PetscErrorCode Tao_ASLS_FunctionGradient(TaoLineSearch ls, Vec X, PetscReal *fcn,  Vec G, void *ptr)
82 {
83   Tao            tao = (Tao)ptr;
84   TAO_SSLS       *asls = (TAO_SSLS *)tao->data;
85   PetscErrorCode ierr;
86 
87   PetscFunctionBegin;
88   ierr = TaoComputeConstraints(tao, X, tao->constraints);CHKERRQ(ierr);
89   ierr = VecFischer(X,tao->constraints,tao->XL,tao->XU,asls->ff);CHKERRQ(ierr);
90   ierr = VecNorm(asls->ff,NORM_2,&asls->merit);CHKERRQ(ierr);
91   *fcn = 0.5*asls->merit*asls->merit;
92   ierr = TaoComputeJacobian(tao,tao->solution,tao->jacobian,tao->jacobian_pre);CHKERRQ(ierr);
93 
94   ierr = MatDFischer(tao->jacobian, tao->solution, tao->constraints,tao->XL, tao->XU, asls->t1, asls->t2,asls->da, asls->db);CHKERRQ(ierr);
95   ierr = VecPointwiseMult(asls->t1, asls->ff, asls->db);CHKERRQ(ierr);
96   ierr = MatMultTranspose(tao->jacobian,asls->t1,G);CHKERRQ(ierr);
97   ierr = VecPointwiseMult(asls->t1, asls->ff, asls->da);CHKERRQ(ierr);
98   ierr = VecAXPY(G,1.0,asls->t1);CHKERRQ(ierr);
99   PetscFunctionReturn(0);
100 }
101 
102 static PetscErrorCode TaoDestroy_ASFLS(Tao tao)
103 {
104   TAO_SSLS       *ssls = (TAO_SSLS *)tao->data;
105   PetscErrorCode ierr;
106 
107   PetscFunctionBegin;
108   ierr = VecDestroy(&ssls->ff);CHKERRQ(ierr);
109   ierr = VecDestroy(&ssls->dpsi);CHKERRQ(ierr);
110   ierr = VecDestroy(&ssls->da);CHKERRQ(ierr);
111   ierr = VecDestroy(&ssls->db);CHKERRQ(ierr);
112   ierr = VecDestroy(&ssls->w);CHKERRQ(ierr);
113   ierr = VecDestroy(&ssls->t1);CHKERRQ(ierr);
114   ierr = VecDestroy(&ssls->t2);CHKERRQ(ierr);
115   ierr = VecDestroy(&ssls->r1);CHKERRQ(ierr);
116   ierr = VecDestroy(&ssls->r2);CHKERRQ(ierr);
117   ierr = VecDestroy(&ssls->r3);CHKERRQ(ierr);
118   ierr = VecDestroy(&ssls->dxfree);CHKERRQ(ierr);
119   ierr = MatDestroy(&ssls->J_sub);CHKERRQ(ierr);
120   ierr = MatDestroy(&ssls->Jpre_sub);CHKERRQ(ierr);
121   ierr = ISDestroy(&ssls->fixed);CHKERRQ(ierr);
122   ierr = ISDestroy(&ssls->free);CHKERRQ(ierr);
123   ierr = PetscFree(tao->data);CHKERRQ(ierr);
124   tao->data = NULL;
125   PetscFunctionReturn(0);
126 }
127 
128 static PetscErrorCode TaoSolve_ASFLS(Tao tao)
129 {
130   TAO_SSLS                     *asls = (TAO_SSLS *)tao->data;
131   PetscReal                    psi,ndpsi, normd, innerd, t=0;
132   PetscInt                     nf;
133   PetscErrorCode               ierr;
134   TaoLineSearchConvergedReason ls_reason;
135 
136   PetscFunctionBegin;
137   /* Assume that Setup has been called!
138      Set the structure for the Jacobian and create a linear solver. */
139 
140   ierr = TaoComputeVariableBounds(tao);CHKERRQ(ierr);
141   ierr = TaoLineSearchSetObjectiveAndGradientRoutine(tao->linesearch,Tao_ASLS_FunctionGradient,tao);CHKERRQ(ierr);
142   ierr = TaoLineSearchSetObjectiveRoutine(tao->linesearch,Tao_SSLS_Function,tao);CHKERRQ(ierr);
143   ierr = TaoLineSearchSetVariableBounds(tao->linesearch,tao->XL,tao->XU);CHKERRQ(ierr);
144 
145   ierr = VecMedian(tao->XL, tao->solution, tao->XU, tao->solution);CHKERRQ(ierr);
146 
147   /* Calculate the function value and fischer function value at the
148      current iterate */
149   ierr = TaoLineSearchComputeObjectiveAndGradient(tao->linesearch,tao->solution,&psi,asls->dpsi);CHKERRQ(ierr);
150   ierr = VecNorm(asls->dpsi,NORM_2,&ndpsi);CHKERRQ(ierr);
151 
152   tao->reason = TAO_CONTINUE_ITERATING;
153   while (1) {
154     /* Check the converged criteria */
155     ierr = PetscInfo3(tao,"iter %D, merit: %g, ||dpsi||: %g\n",tao->niter,(double)asls->merit,(double)ndpsi);CHKERRQ(ierr);
156     ierr = TaoLogConvergenceHistory(tao,asls->merit,ndpsi,0.0,tao->ksp_its);CHKERRQ(ierr);
157     ierr = TaoMonitor(tao,tao->niter,asls->merit,ndpsi,0.0,t);CHKERRQ(ierr);
158     ierr = (*tao->ops->convergencetest)(tao,tao->cnvP);CHKERRQ(ierr);
159     if (TAO_CONTINUE_ITERATING != tao->reason) break;
160 
161     /* Call general purpose update function */
162     if (tao->ops->update) {
163       ierr = (*tao->ops->update)(tao, tao->niter, tao->user_update);CHKERRQ(ierr);
164     }
165     tao->niter++;
166 
167     /* We are going to solve a linear system of equations.  We need to
168        set the tolerances for the solve so that we maintain an asymptotic
169        rate of convergence that is superlinear.
170        Note: these tolerances are for the reduced system.  We really need
171        to make sure that the full system satisfies the full-space conditions.
172 
173        This rule gives superlinear asymptotic convergence
174        asls->atol = min(0.5, asls->merit*sqrt(asls->merit));
175        asls->rtol = 0.0;
176 
177        This rule gives quadratic asymptotic convergence
178        asls->atol = min(0.5, asls->merit*asls->merit);
179        asls->rtol = 0.0;
180 
181        Calculate a free and fixed set of variables.  The fixed set of
182        variables are those for the d_b is approximately equal to zero.
183        The definition of approximately changes as we approach the solution
184        to the problem.
185 
186        No one rule is guaranteed to work in all cases.  The following
187        definition is based on the norm of the Jacobian matrix.  If the
188        norm is large, the tolerance becomes smaller. */
189     ierr = MatNorm(tao->jacobian,NORM_1,&asls->identifier);CHKERRQ(ierr);
190     asls->identifier = PetscMin(asls->merit, 1e-2) / (1 + asls->identifier);
191 
192     ierr = VecSet(asls->t1,-asls->identifier);CHKERRQ(ierr);
193     ierr = VecSet(asls->t2, asls->identifier);CHKERRQ(ierr);
194 
195     ierr = ISDestroy(&asls->fixed);CHKERRQ(ierr);
196     ierr = ISDestroy(&asls->free);CHKERRQ(ierr);
197     ierr = VecWhichBetweenOrEqual(asls->t1, asls->db, asls->t2, &asls->fixed);CHKERRQ(ierr);
198     ierr = ISComplementVec(asls->fixed,asls->t1, &asls->free);CHKERRQ(ierr);
199 
200     ierr = ISGetSize(asls->fixed,&nf);CHKERRQ(ierr);
201     ierr = PetscInfo1(tao,"Number of fixed variables: %D\n", nf);CHKERRQ(ierr);
202 
203     /* We now have our partition.  Now calculate the direction in the
204        fixed variable space. */
205     ierr = TaoVecGetSubVec(asls->ff, asls->fixed, tao->subset_type, 0.0, &asls->r1);CHKERRQ(ierr);
206     ierr = TaoVecGetSubVec(asls->da, asls->fixed, tao->subset_type, 1.0, &asls->r2);CHKERRQ(ierr);
207     ierr = VecPointwiseDivide(asls->r1,asls->r1,asls->r2);CHKERRQ(ierr);
208     ierr = VecSet(tao->stepdirection,0.0);CHKERRQ(ierr);
209     ierr = VecISAXPY(tao->stepdirection, asls->fixed, 1.0,asls->r1);CHKERRQ(ierr);
210 
211     /* Our direction in the Fixed Variable Set is fixed.  Calculate the
212        information needed for the step in the Free Variable Set.  To
213        do this, we need to know the diagonal perturbation and the
214        right hand side. */
215 
216     ierr = TaoVecGetSubVec(asls->da, asls->free, tao->subset_type, 0.0, &asls->r1);CHKERRQ(ierr);
217     ierr = TaoVecGetSubVec(asls->ff, asls->free, tao->subset_type, 0.0, &asls->r2);CHKERRQ(ierr);
218     ierr = TaoVecGetSubVec(asls->db, asls->free, tao->subset_type, 1.0, &asls->r3);CHKERRQ(ierr);
219     ierr = VecPointwiseDivide(asls->r1,asls->r1, asls->r3);CHKERRQ(ierr);
220     ierr = VecPointwiseDivide(asls->r2,asls->r2, asls->r3);CHKERRQ(ierr);
221 
222     /* r1 is the diagonal perturbation
223        r2 is the right hand side
224        r3 is no longer needed
225 
226        Now need to modify r2 for our direction choice in the fixed
227        variable set:  calculate t1 = J*d, take the reduced vector
228        of t1 and modify r2. */
229 
230     ierr = MatMult(tao->jacobian, tao->stepdirection, asls->t1);CHKERRQ(ierr);
231     ierr = TaoVecGetSubVec(asls->t1,asls->free,tao->subset_type,0.0,&asls->r3);CHKERRQ(ierr);
232     ierr = VecAXPY(asls->r2, -1.0, asls->r3);CHKERRQ(ierr);
233 
234     /* Calculate the reduced problem matrix and the direction */
235     ierr = TaoMatGetSubMat(tao->jacobian, asls->free, asls->w, tao->subset_type,&asls->J_sub);CHKERRQ(ierr);
236     if (tao->jacobian != tao->jacobian_pre) {
237       ierr = TaoMatGetSubMat(tao->jacobian_pre, asls->free, asls->w, tao->subset_type, &asls->Jpre_sub);CHKERRQ(ierr);
238     } else {
239       ierr = MatDestroy(&asls->Jpre_sub);CHKERRQ(ierr);
240       asls->Jpre_sub = asls->J_sub;
241       ierr = PetscObjectReference((PetscObject)(asls->Jpre_sub));CHKERRQ(ierr);
242     }
243     ierr = MatDiagonalSet(asls->J_sub, asls->r1,ADD_VALUES);CHKERRQ(ierr);
244     ierr = TaoVecGetSubVec(tao->stepdirection, asls->free, tao->subset_type, 0.0, &asls->dxfree);CHKERRQ(ierr);
245     ierr = VecSet(asls->dxfree, 0.0);CHKERRQ(ierr);
246 
247     /* Calculate the reduced direction.  (Really negative of Newton
248        direction.  Therefore, rest of the code uses -d.) */
249     ierr = KSPReset(tao->ksp);CHKERRQ(ierr);
250     ierr = KSPSetOperators(tao->ksp, asls->J_sub, asls->Jpre_sub);CHKERRQ(ierr);
251     ierr = KSPSolve(tao->ksp, asls->r2, asls->dxfree);CHKERRQ(ierr);
252     ierr = KSPGetIterationNumber(tao->ksp,&tao->ksp_its);CHKERRQ(ierr);
253     tao->ksp_tot_its+=tao->ksp_its;
254 
255     /* Add the direction in the free variables back into the real direction. */
256     ierr = VecISAXPY(tao->stepdirection, asls->free, 1.0,asls->dxfree);CHKERRQ(ierr);
257 
258     /* Check the projected real direction for descent and if not, use the negative
259        gradient direction. */
260     ierr = VecCopy(tao->stepdirection, asls->w);CHKERRQ(ierr);
261     ierr = VecScale(asls->w, -1.0);CHKERRQ(ierr);
262     ierr = VecBoundGradientProjection(asls->w, tao->solution, tao->XL, tao->XU, asls->w);CHKERRQ(ierr);
263     ierr = VecNorm(asls->w, NORM_2, &normd);CHKERRQ(ierr);
264     ierr = VecDot(asls->w, asls->dpsi, &innerd);CHKERRQ(ierr);
265 
266     if (innerd >= -asls->delta*PetscPowReal(normd, asls->rho)) {
267       ierr = PetscInfo1(tao,"Gradient direction: %5.4e.\n", (double)innerd);CHKERRQ(ierr);
268       ierr = PetscInfo1(tao, "Iteration %D: newton direction not descent\n", tao->niter);CHKERRQ(ierr);
269       ierr = VecCopy(asls->dpsi, tao->stepdirection);CHKERRQ(ierr);
270       ierr = VecDot(asls->dpsi, tao->stepdirection, &innerd);CHKERRQ(ierr);
271     }
272 
273     ierr = VecScale(tao->stepdirection, -1.0);CHKERRQ(ierr);
274     innerd = -innerd;
275 
276     /* We now have a correct descent direction.  Apply a linesearch to
277        find the new iterate. */
278     ierr = TaoLineSearchSetInitialStepLength(tao->linesearch, 1.0);CHKERRQ(ierr);
279     ierr = TaoLineSearchApply(tao->linesearch, tao->solution, &psi,asls->dpsi, tao->stepdirection, &t, &ls_reason);CHKERRQ(ierr);
280     ierr = VecNorm(asls->dpsi, NORM_2, &ndpsi);CHKERRQ(ierr);
281   }
282   PetscFunctionReturn(0);
283 }
284 
285 /* ---------------------------------------------------------- */
286 /*MC
287    TAOASFLS - Active-set feasible linesearch algorithm for solving
288        complementarity constraints
289 
290    Options Database Keys:
291 + -tao_ssls_delta - descent test fraction
292 - -tao_ssls_rho - descent test power
293 
294    Level: beginner
295 M*/
296 PETSC_EXTERN PetscErrorCode TaoCreate_ASFLS(Tao tao)
297 {
298   TAO_SSLS       *asls;
299   PetscErrorCode ierr;
300   const char     *armijo_type = TAOLINESEARCHARMIJO;
301 
302   PetscFunctionBegin;
303   ierr = PetscNewLog(tao,&asls);CHKERRQ(ierr);
304   tao->data = (void*)asls;
305   tao->ops->solve = TaoSolve_ASFLS;
306   tao->ops->setup = TaoSetUp_ASFLS;
307   tao->ops->view = TaoView_SSLS;
308   tao->ops->setfromoptions = TaoSetFromOptions_SSLS;
309   tao->ops->destroy = TaoDestroy_ASFLS;
310   tao->subset_type = TAO_SUBSET_SUBVEC;
311   asls->delta = 1e-10;
312   asls->rho = 2.1;
313   asls->fixed = NULL;
314   asls->free = NULL;
315   asls->J_sub = NULL;
316   asls->Jpre_sub = NULL;
317   asls->w = NULL;
318   asls->r1 = NULL;
319   asls->r2 = NULL;
320   asls->r3 = NULL;
321   asls->t1 = NULL;
322   asls->t2 = NULL;
323   asls->dxfree = NULL;
324   asls->identifier = 1e-5;
325 
326   ierr = TaoLineSearchCreate(((PetscObject)tao)->comm, &tao->linesearch);CHKERRQ(ierr);
327   ierr = PetscObjectIncrementTabLevel((PetscObject)tao->linesearch, (PetscObject)tao, 1);CHKERRQ(ierr);
328   ierr = TaoLineSearchSetType(tao->linesearch, armijo_type);CHKERRQ(ierr);
329   ierr = TaoLineSearchSetOptionsPrefix(tao->linesearch,tao->hdr.prefix);CHKERRQ(ierr);
330   ierr = TaoLineSearchSetFromOptions(tao->linesearch);CHKERRQ(ierr);
331 
332   ierr = KSPCreate(((PetscObject)tao)->comm, &tao->ksp);CHKERRQ(ierr);
333   ierr = PetscObjectIncrementTabLevel((PetscObject)tao->ksp, (PetscObject)tao, 1);CHKERRQ(ierr);
334   ierr = KSPSetOptionsPrefix(tao->ksp,tao->hdr.prefix);CHKERRQ(ierr);
335   ierr = KSPSetFromOptions(tao->ksp);CHKERRQ(ierr);
336 
337   /* Override default settings (unless already changed) */
338   if (!tao->max_it_changed) tao->max_it = 2000;
339   if (!tao->max_funcs_changed) tao->max_funcs = 4000;
340   if (!tao->gttol_changed) tao->gttol = 0;
341   if (!tao->grtol_changed) tao->grtol = 0;
342 #if defined(PETSC_USE_REAL_SINGLE)
343   if (!tao->gatol_changed) tao->gatol = 1.0e-6;
344   if (!tao->fmin_changed)  tao->fmin = 1.0e-4;
345 #else
346   if (!tao->gatol_changed) tao->gatol = 1.0e-16;
347   if (!tao->fmin_changed)  tao->fmin = 1.0e-8;
348 #endif
349   PetscFunctionReturn(0);
350 }
351