xref: /petsc/src/tao/complementarity/impls/asls/asfls.c (revision 609bdbee21ea3be08735c64dbe00a9ab27759925)
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 
56 static PetscErrorCode TaoSetUp_ASFLS(Tao tao)
57 {
58   TAO_SSLS       *asls = (TAO_SSLS *)tao->data;
59   PetscErrorCode ierr;
60 
61   PetscFunctionBegin;
62   ierr = VecDuplicate(tao->solution,&tao->gradient);CHKERRQ(ierr);
63   ierr = VecDuplicate(tao->solution,&tao->stepdirection);CHKERRQ(ierr);
64   ierr = VecDuplicate(tao->solution,&asls->ff);CHKERRQ(ierr);
65   ierr = VecDuplicate(tao->solution,&asls->dpsi);CHKERRQ(ierr);
66   ierr = VecDuplicate(tao->solution,&asls->da);CHKERRQ(ierr);
67   ierr = VecDuplicate(tao->solution,&asls->db);CHKERRQ(ierr);
68   ierr = VecDuplicate(tao->solution,&asls->t1);CHKERRQ(ierr);
69   ierr = VecDuplicate(tao->solution,&asls->t2);CHKERRQ(ierr);
70   ierr = VecDuplicate(tao->solution, &asls->w);CHKERRQ(ierr);
71   asls->fixed = NULL;
72   asls->free = NULL;
73   asls->J_sub = NULL;
74   asls->Jpre_sub = NULL;
75   asls->r1 = NULL;
76   asls->r2 = NULL;
77   asls->r3 = NULL;
78   asls->dxfree = NULL;
79   PetscFunctionReturn(0);
80 }
81 
82 static PetscErrorCode Tao_ASLS_FunctionGradient(TaoLineSearch ls, Vec X, PetscReal *fcn,  Vec G, void *ptr)
83 {
84   Tao            tao = (Tao)ptr;
85   TAO_SSLS       *asls = (TAO_SSLS *)tao->data;
86   PetscErrorCode ierr;
87 
88   PetscFunctionBegin;
89   ierr = TaoComputeConstraints(tao, X, tao->constraints);CHKERRQ(ierr);
90   ierr = VecFischer(X,tao->constraints,tao->XL,tao->XU,asls->ff);CHKERRQ(ierr);
91   ierr = VecNorm(asls->ff,NORM_2,&asls->merit);CHKERRQ(ierr);
92   *fcn = 0.5*asls->merit*asls->merit;
93   ierr = TaoComputeJacobian(tao,tao->solution,tao->jacobian,tao->jacobian_pre);CHKERRQ(ierr);
94 
95   ierr = MatDFischer(tao->jacobian, tao->solution, tao->constraints,tao->XL, tao->XU, asls->t1, asls->t2,asls->da, asls->db);CHKERRQ(ierr);
96   ierr = VecPointwiseMult(asls->t1, asls->ff, asls->db);CHKERRQ(ierr);
97   ierr = MatMultTranspose(tao->jacobian,asls->t1,G);CHKERRQ(ierr);
98   ierr = VecPointwiseMult(asls->t1, asls->ff, asls->da);CHKERRQ(ierr);
99   ierr = VecAXPY(G,1.0,asls->t1);CHKERRQ(ierr);
100   PetscFunctionReturn(0);
101 }
102 
103 static PetscErrorCode TaoDestroy_ASFLS(Tao tao)
104 {
105   TAO_SSLS       *ssls = (TAO_SSLS *)tao->data;
106   PetscErrorCode ierr;
107 
108   PetscFunctionBegin;
109   ierr = VecDestroy(&ssls->ff);CHKERRQ(ierr);
110   ierr = VecDestroy(&ssls->dpsi);CHKERRQ(ierr);
111   ierr = VecDestroy(&ssls->da);CHKERRQ(ierr);
112   ierr = VecDestroy(&ssls->db);CHKERRQ(ierr);
113   ierr = VecDestroy(&ssls->w);CHKERRQ(ierr);
114   ierr = VecDestroy(&ssls->t1);CHKERRQ(ierr);
115   ierr = VecDestroy(&ssls->t2);CHKERRQ(ierr);
116   ierr = VecDestroy(&ssls->r1);CHKERRQ(ierr);
117   ierr = VecDestroy(&ssls->r2);CHKERRQ(ierr);
118   ierr = VecDestroy(&ssls->r3);CHKERRQ(ierr);
119   ierr = VecDestroy(&ssls->dxfree);CHKERRQ(ierr);
120   ierr = MatDestroy(&ssls->J_sub);CHKERRQ(ierr);
121   ierr = MatDestroy(&ssls->Jpre_sub);CHKERRQ(ierr);
122   ierr = ISDestroy(&ssls->fixed);CHKERRQ(ierr);
123   ierr = ISDestroy(&ssls->free);CHKERRQ(ierr);
124   ierr = PetscFree(tao->data);CHKERRQ(ierr);
125   tao->data = NULL;
126   PetscFunctionReturn(0);
127 }
128 
129 static PetscErrorCode TaoSolve_ASFLS(Tao tao)
130 {
131   TAO_SSLS                     *asls = (TAO_SSLS *)tao->data;
132   PetscReal                    psi,ndpsi, normd, innerd, t=0;
133   PetscInt                     nf;
134   PetscErrorCode               ierr;
135   TaoConvergedReason           reason;
136   TaoLineSearchConvergedReason ls_reason;
137 
138   PetscFunctionBegin;
139   /* Assume that Setup has been called!
140      Set the structure for the Jacobian and create a linear solver. */
141 
142   ierr = TaoComputeVariableBounds(tao);CHKERRQ(ierr);
143   ierr = TaoLineSearchSetObjectiveAndGradientRoutine(tao->linesearch,Tao_ASLS_FunctionGradient,tao);CHKERRQ(ierr);
144   ierr = TaoLineSearchSetObjectiveRoutine(tao->linesearch,Tao_SSLS_Function,tao);CHKERRQ(ierr);
145   ierr = TaoLineSearchSetVariableBounds(tao->linesearch,tao->XL,tao->XU);CHKERRQ(ierr);
146 
147   ierr = VecMedian(tao->XL, tao->solution, tao->XU, tao->solution);CHKERRQ(ierr);
148 
149   /* Calculate the function value and fischer function value at the
150      current iterate */
151   ierr = TaoLineSearchComputeObjectiveAndGradient(tao->linesearch,tao->solution,&psi,asls->dpsi);CHKERRQ(ierr);
152   ierr = VecNorm(asls->dpsi,NORM_2,&ndpsi);CHKERRQ(ierr);
153 
154   while (1) {
155     /* Check the converged criteria */
156     ierr = PetscInfo3(tao,"iter %D, merit: %g, ||dpsi||: %g\n",tao->niter,(double)asls->merit,(double)ndpsi);CHKERRQ(ierr);
157     ierr = TaoMonitor(tao,tao->niter,asls->merit,ndpsi,0.0,t,&reason);CHKERRQ(ierr);
158     if (TAO_CONTINUE_ITERATING != reason) break;
159     tao->niter++;
160 
161     /* We are going to solve a linear system of equations.  We need to
162        set the tolerances for the solve so that we maintain an asymptotic
163        rate of convergence that is superlinear.
164        Note: these tolerances are for the reduced system.  We really need
165        to make sure that the full system satisfies the full-space conditions.
166 
167        This rule gives superlinear asymptotic convergence
168        asls->atol = min(0.5, asls->merit*sqrt(asls->merit));
169        asls->rtol = 0.0;
170 
171        This rule gives quadratic asymptotic convergence
172        asls->atol = min(0.5, asls->merit*asls->merit);
173        asls->rtol = 0.0;
174 
175        Calculate a free and fixed set of variables.  The fixed set of
176        variables are those for the d_b is approximately equal to zero.
177        The definition of approximately changes as we approach the solution
178        to the problem.
179 
180        No one rule is guaranteed to work in all cases.  The following
181        definition is based on the norm of the Jacobian matrix.  If the
182        norm is large, the tolerance becomes smaller. */
183     ierr = MatNorm(tao->jacobian,NORM_1,&asls->identifier);CHKERRQ(ierr);
184     asls->identifier = PetscMin(asls->merit, 1e-2) / (1 + asls->identifier);
185 
186     ierr = VecSet(asls->t1,-asls->identifier);CHKERRQ(ierr);
187     ierr = VecSet(asls->t2, asls->identifier);CHKERRQ(ierr);
188 
189     ierr = ISDestroy(&asls->fixed);CHKERRQ(ierr);
190     ierr = ISDestroy(&asls->free);CHKERRQ(ierr);
191     ierr = VecWhichBetweenOrEqual(asls->t1, asls->db, asls->t2, &asls->fixed);CHKERRQ(ierr);
192     ierr = ISComplementVec(asls->fixed,asls->t1, &asls->free);CHKERRQ(ierr);
193 
194     ierr = ISGetSize(asls->fixed,&nf);CHKERRQ(ierr);
195     ierr = PetscInfo1(tao,"Number of fixed variables: %D\n", nf);CHKERRQ(ierr);
196 
197     /* We now have our partition.  Now calculate the direction in the
198        fixed variable space. */
199     ierr = TaoVecGetSubVec(asls->ff, asls->fixed, tao->subset_type, 0.0, &asls->r1);CHKERRQ(ierr);
200     ierr = TaoVecGetSubVec(asls->da, asls->fixed, tao->subset_type, 1.0, &asls->r2);CHKERRQ(ierr);
201     ierr = VecPointwiseDivide(asls->r1,asls->r1,asls->r2);CHKERRQ(ierr);
202     ierr = VecSet(tao->stepdirection,0.0);CHKERRQ(ierr);
203     ierr = VecISAXPY(tao->stepdirection, asls->fixed, 1.0,asls->r1);CHKERRQ(ierr);
204 
205     /* Our direction in the Fixed Variable Set is fixed.  Calculate the
206        information needed for the step in the Free Variable Set.  To
207        do this, we need to know the diagonal perturbation and the
208        right hand side. */
209 
210     ierr = TaoVecGetSubVec(asls->da, asls->free, tao->subset_type, 0.0, &asls->r1);CHKERRQ(ierr);
211     ierr = TaoVecGetSubVec(asls->ff, asls->free, tao->subset_type, 0.0, &asls->r2);CHKERRQ(ierr);
212     ierr = TaoVecGetSubVec(asls->db, asls->free, tao->subset_type, 1.0, &asls->r3);CHKERRQ(ierr);
213     ierr = VecPointwiseDivide(asls->r1,asls->r1, asls->r3);CHKERRQ(ierr);
214     ierr = VecPointwiseDivide(asls->r2,asls->r2, asls->r3);CHKERRQ(ierr);
215 
216     /* r1 is the diagonal perturbation
217        r2 is the right hand side
218        r3 is no longer needed
219 
220        Now need to modify r2 for our direction choice in the fixed
221        variable set:  calculate t1 = J*d, take the reduced vector
222        of t1 and modify r2. */
223 
224     ierr = MatMult(tao->jacobian, tao->stepdirection, asls->t1);CHKERRQ(ierr);
225     ierr = TaoVecGetSubVec(asls->t1,asls->free,tao->subset_type,0.0,&asls->r3);CHKERRQ(ierr);
226     ierr = VecAXPY(asls->r2, -1.0, asls->r3);CHKERRQ(ierr);
227 
228     /* Calculate the reduced problem matrix and the direction */
229     ierr = TaoMatGetSubMat(tao->jacobian, asls->free, asls->w, tao->subset_type,&asls->J_sub);CHKERRQ(ierr);
230     if (tao->jacobian != tao->jacobian_pre) {
231       ierr = TaoMatGetSubMat(tao->jacobian_pre, asls->free, asls->w, tao->subset_type, &asls->Jpre_sub);CHKERRQ(ierr);
232     } else {
233       ierr = MatDestroy(&asls->Jpre_sub);CHKERRQ(ierr);
234       asls->Jpre_sub = asls->J_sub;
235       ierr = PetscObjectReference((PetscObject)(asls->Jpre_sub));CHKERRQ(ierr);
236     }
237     ierr = MatDiagonalSet(asls->J_sub, asls->r1,ADD_VALUES);CHKERRQ(ierr);
238     ierr = TaoVecGetSubVec(tao->stepdirection, asls->free, tao->subset_type, 0.0, &asls->dxfree);CHKERRQ(ierr);
239     ierr = VecSet(asls->dxfree, 0.0);CHKERRQ(ierr);
240 
241     /* Calculate the reduced direction.  (Really negative of Newton
242        direction.  Therefore, rest of the code uses -d.) */
243     ierr = KSPReset(tao->ksp);CHKERRQ(ierr);
244     ierr = KSPSetOperators(tao->ksp, asls->J_sub, asls->Jpre_sub);CHKERRQ(ierr);
245     ierr = KSPSolve(tao->ksp, asls->r2, asls->dxfree);CHKERRQ(ierr);
246     ierr = KSPGetIterationNumber(tao->ksp,&tao->ksp_its);CHKERRQ(ierr);
247     tao->ksp_tot_its+=tao->ksp_its;
248 
249     /* Add the direction in the free variables back into the real direction. */
250     ierr = VecISAXPY(tao->stepdirection, asls->free, 1.0,asls->dxfree);CHKERRQ(ierr);
251 
252 
253     /* Check the projected real direction for descent and if not, use the negative
254        gradient direction. */
255     ierr = VecCopy(tao->stepdirection, asls->w);CHKERRQ(ierr);
256     ierr = VecScale(asls->w, -1.0);CHKERRQ(ierr);
257     ierr = VecBoundGradientProjection(asls->w, tao->solution, tao->XL, tao->XU, asls->w);CHKERRQ(ierr);
258     ierr = VecNorm(asls->w, NORM_2, &normd);CHKERRQ(ierr);
259     ierr = VecDot(asls->w, asls->dpsi, &innerd);CHKERRQ(ierr);
260 
261     if (innerd >= -asls->delta*PetscPowReal(normd, asls->rho)) {
262       ierr = PetscInfo1(tao,"Gradient direction: %5.4e.\n", (double)innerd);CHKERRQ(ierr);
263       ierr = PetscInfo1(tao, "Iteration %D: newton direction not descent\n", tao->niter);CHKERRQ(ierr);
264       ierr = VecCopy(asls->dpsi, tao->stepdirection);CHKERRQ(ierr);
265       ierr = VecDot(asls->dpsi, tao->stepdirection, &innerd);CHKERRQ(ierr);
266     }
267 
268     ierr = VecScale(tao->stepdirection, -1.0);CHKERRQ(ierr);
269     innerd = -innerd;
270 
271     /* We now have a correct descent direction.  Apply a linesearch to
272        find the new iterate. */
273     ierr = TaoLineSearchSetInitialStepLength(tao->linesearch, 1.0);CHKERRQ(ierr);
274     ierr = TaoLineSearchApply(tao->linesearch, tao->solution, &psi,asls->dpsi, tao->stepdirection, &t, &ls_reason);CHKERRQ(ierr);
275     ierr = VecNorm(asls->dpsi, NORM_2, &ndpsi);CHKERRQ(ierr);
276   }
277   PetscFunctionReturn(0);
278 }
279 
280 /* ---------------------------------------------------------- */
281 /*MC
282    TAOASFLS - Active-set feasible linesearch algorithm for solving
283        complementarity constraints
284 
285    Options Database Keys:
286 + -tao_ssls_delta - descent test fraction
287 - -tao_ssls_rho - descent test power
288 
289    Level: beginner
290 M*/
291 PETSC_EXTERN PetscErrorCode TaoCreate_ASFLS(Tao tao)
292 {
293   TAO_SSLS       *asls;
294   PetscErrorCode ierr;
295   const char     *armijo_type = TAOLINESEARCHARMIJO;
296 
297   PetscFunctionBegin;
298   ierr = PetscNewLog(tao,&asls);CHKERRQ(ierr);
299   tao->data = (void*)asls;
300   tao->ops->solve = TaoSolve_ASFLS;
301   tao->ops->setup = TaoSetUp_ASFLS;
302   tao->ops->view = TaoView_SSLS;
303   tao->ops->setfromoptions = TaoSetFromOptions_SSLS;
304   tao->ops->destroy = TaoDestroy_ASFLS;
305   tao->subset_type = TAO_SUBSET_SUBVEC;
306   asls->delta = 1e-10;
307   asls->rho = 2.1;
308   asls->fixed = NULL;
309   asls->free = NULL;
310   asls->J_sub = NULL;
311   asls->Jpre_sub = NULL;
312   asls->w = NULL;
313   asls->r1 = NULL;
314   asls->r2 = NULL;
315   asls->r3 = NULL;
316   asls->t1 = NULL;
317   asls->t2 = NULL;
318   asls->dxfree = NULL;
319   asls->identifier = 1e-5;
320 
321   ierr = TaoLineSearchCreate(((PetscObject)tao)->comm, &tao->linesearch);CHKERRQ(ierr);
322   ierr = TaoLineSearchSetType(tao->linesearch, armijo_type);CHKERRQ(ierr);
323   ierr = TaoLineSearchSetOptionsPrefix(tao->linesearch,tao->hdr.prefix);CHKERRQ(ierr);
324   ierr = TaoLineSearchSetFromOptions(tao->linesearch);CHKERRQ(ierr);
325 
326   ierr = KSPCreate(((PetscObject)tao)->comm, &tao->ksp);CHKERRQ(ierr);
327   ierr = KSPSetOptionsPrefix(tao->ksp,tao->hdr.prefix);CHKERRQ(ierr);
328   ierr = KSPSetFromOptions(tao->ksp);CHKERRQ(ierr);
329 
330   /* Override default settings (unless already changed) */
331   if (!tao->max_it_changed) tao->max_it = 2000;
332   if (!tao->max_funcs_changed) tao->max_funcs = 4000;
333   if (!tao->gttol_changed) tao->gttol = 0;
334   if (!tao->grtol_changed) tao->grtol = 0;
335 #if defined(PETSC_USE_REAL_SINGLE)
336   if (!tao->gatol_changed) tao->gatol = 1.0e-6;
337   if (!tao->fmin_changed)  tao->fmin = 1.0e-4;
338 #else
339   if (!tao->gatol_changed) tao->gatol = 1.0e-16;
340   if (!tao->fmin_changed)  tao->fmin = 1.0e-8;
341 #endif
342   PetscFunctionReturn(0);
343 }
344 
345