ROL
ROL_NewtonKrylovStep.hpp
Go to the documentation of this file.
1// @HEADER
2// ************************************************************************
3//
4// Rapid Optimization Library (ROL) Package
5// Copyright (2014) Sandia Corporation
6//
7// Under terms of Contract DE-AC04-94AL85000, there is a non-exclusive
8// license for use of this work by or on behalf of the U.S. Government.
9//
10// Redistribution and use in source and binary forms, with or without
11// modification, are permitted provided that the following conditions are
12// met:
13//
14// 1. Redistributions of source code must retain the above copyright
15// notice, this list of conditions and the following disclaimer.
16//
17// 2. Redistributions in binary form must reproduce the above copyright
18// notice, this list of conditions and the following disclaimer in the
19// documentation and/or other materials provided with the distribution.
20//
21// 3. Neither the name of the Corporation nor the names of the
22// contributors may be used to endorse or promote products derived from
23// this software without specific prior written permission.
24//
25// THIS SOFTWARE IS PROVIDED BY SANDIA CORPORATION "AS IS" AND ANY
26// EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
28// PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL SANDIA CORPORATION OR THE
29// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
30// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
31// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
32// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
33// LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
34// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
35// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
36//
37// Questions? Contact lead developers:
38// Drew Kouri (dpkouri@sandia.gov) and
39// Denis Ridzal (dridzal@sandia.gov)
40//
41// ************************************************************************
42// @HEADER
43
44#ifndef ROL_NEWTONKRYLOVSTEP_H
45#define ROL_NEWTONKRYLOVSTEP_H
46
47#include "ROL_Types.hpp"
48#include "ROL_Step.hpp"
49
50#include "ROL_Secant.hpp"
51#include "ROL_KrylovFactory.hpp"
53
54#include <sstream>
55#include <iomanip>
56
63namespace ROL {
64
65template <class Real>
66class NewtonKrylovStep : public Step<Real> {
67private:
68
69 ROL::Ptr<Secant<Real> > secant_;
70 ROL::Ptr<Krylov<Real> > krylov_;
71
74
75 ROL::Ptr<Vector<Real> > gp_;
76
80 const bool computeObj_;
81
83
84 std::string krylovName_;
85 std::string secantName_;
86
87
88 class HessianNK : public LinearOperator<Real> {
89 private:
90 const ROL::Ptr<Objective<Real> > obj_;
91 const ROL::Ptr<Vector<Real> > x_;
92 public:
93 HessianNK(const ROL::Ptr<Objective<Real> > &obj,
94 const ROL::Ptr<Vector<Real> > &x) : obj_(obj), x_(x) {}
95 void apply(Vector<Real> &Hv, const Vector<Real> &v, Real &tol) const {
96 obj_->hessVec(Hv,v,*x_,tol);
97 }
98 };
99
100 class PrecondNK : public LinearOperator<Real> {
101 private:
102 const ROL::Ptr<Objective<Real> > obj_;
103 const ROL::Ptr<Vector<Real> > x_;
104 public:
105 PrecondNK(const ROL::Ptr<Objective<Real> > &obj,
106 const ROL::Ptr<Vector<Real> > &x) : obj_(obj), x_(x) {}
107 void apply(Vector<Real> &Hv, const Vector<Real> &v, Real &tol) const {
108 Hv.set(v.dual());
109 }
110 void applyInverse(Vector<Real> &Hv, const Vector<Real> &v, Real &tol) const {
111 obj_->precond(Hv,v,*x_,tol);
112 }
113 };
114
115public:
116
117 using Step<Real>::initialize;
118 using Step<Real>::compute;
119 using Step<Real>::update;
120
128 NewtonKrylovStep( ROL::ParameterList &parlist, const bool computeObj = true )
129 : Step<Real>(), secant_(ROL::nullPtr), krylov_(ROL::nullPtr),
130 gp_(ROL::nullPtr), iterKrylov_(0), flagKrylov_(0),
131 verbosity_(0), computeObj_(computeObj), useSecantPrecond_(false) {
132 // Parse ParameterList
133 ROL::ParameterList& Glist = parlist.sublist("General");
134 useSecantPrecond_ = Glist.sublist("Secant").get("Use as Preconditioner", false);
135 verbosity_ = Glist.get("Print Verbosity",0);
136 // Initialize Krylov object
137 krylovName_ = Glist.sublist("Krylov").get("Type","Conjugate Gradients");
139 krylov_ = KrylovFactory<Real>(parlist);
140 // Initialize secant object
141 secantName_ = Glist.sublist("Secant").get("Type","Limited-Memory BFGS");
143 if ( useSecantPrecond_ ) {
144 secant_ = SecantFactory<Real>(parlist);
145 }
146 }
147
158 NewtonKrylovStep(ROL::ParameterList &parlist,
159 const ROL::Ptr<Krylov<Real> > &krylov,
160 const ROL::Ptr<Secant<Real> > &secant,
161 const bool computeObj = true)
162 : Step<Real>(), secant_(secant), krylov_(krylov),
164 gp_(ROL::nullPtr), iterKrylov_(0), flagKrylov_(0),
165 verbosity_(0), computeObj_(computeObj), useSecantPrecond_(false) {
166 // Parse ParameterList
167 ROL::ParameterList& Glist = parlist.sublist("General");
168 useSecantPrecond_ = Glist.sublist("Secant").get("Use as Preconditioner", false);
169 verbosity_ = Glist.get("Print Verbosity",0);
170 // Initialize secant object
171 if ( useSecantPrecond_ ) {
172 if(secant_ == ROL::nullPtr ) {
173 secantName_ = Glist.sublist("Secant").get("Type","Limited-Memory BFGS");
175 secant_ = SecantFactory<Real>(parlist);
176 }
177 else {
178 secantName_ = Glist.sublist("Secant").get("User Defined Secant Name",
179 "Unspecified User Defined Secant Method");
180 }
181 }
182 // Initialize Krylov object
183 if ( krylov_ == ROL::nullPtr ) {
184 krylovName_ = Glist.sublist("Krylov").get("Type","Conjugate Gradients");
186 krylov_ = KrylovFactory<Real>(parlist);
187 }
188 else {
189 krylovName_ = Glist.sublist("Krylov").get("User Defined Krylov Name",
190 "Unspecified User Defined Krylov Method");
191 }
192 }
193
194 void initialize( Vector<Real> &x, const Vector<Real> &s, const Vector<Real> &g,
196 AlgorithmState<Real> &algo_state ) {
197 Step<Real>::initialize(x,s,g,obj,bnd,algo_state);
198 if ( useSecantPrecond_ ) {
199 gp_ = g.clone();
200 }
201 }
202
205 AlgorithmState<Real> &algo_state ) {
206 Real one(1);
207 ROL::Ptr<StepState<Real> > step_state = Step<Real>::getState();
208
209 // Build Hessian and Preconditioner object
210 ROL::Ptr<Objective<Real> > obj_ptr = ROL::makePtrFromRef(obj);
211 ROL::Ptr<LinearOperator<Real> > hessian
212 = ROL::makePtr<HessianNK>(obj_ptr,algo_state.iterateVec);
213 ROL::Ptr<LinearOperator<Real> > precond;
214 if ( useSecantPrecond_ ) {
215 precond = secant_;
216 }
217 else {
218 precond = ROL::makePtr<PrecondNK>(obj_ptr,algo_state.iterateVec);
219 }
220
221 // Run Krylov method
222 flagKrylov_ = 0;
223 krylov_->run(s,*hessian,*(step_state->gradientVec),*precond,iterKrylov_,flagKrylov_);
224
225 // Check Krylov flags
226 if ( flagKrylov_ == 2 && iterKrylov_ <= 1 ) {
227 s.set((step_state->gradientVec)->dual());
228 }
229 s.scale(-one);
230 }
231
232 void update( Vector<Real> &x, const Vector<Real> &s,
234 AlgorithmState<Real> &algo_state ) {
235 Real tol = std::sqrt(ROL_EPSILON<Real>());
236 ROL::Ptr<StepState<Real> > step_state = Step<Real>::getState();
237 step_state->SPiter = iterKrylov_;
238 step_state->SPflag = flagKrylov_;
239
240 // Update iterate
241 algo_state.iter++;
242 x.plus(s);
243 (step_state->descentVec)->set(s);
244 algo_state.snorm = s.norm();
245
246 // Compute new gradient
247 if ( useSecantPrecond_ ) {
248 gp_->set(*(step_state->gradientVec));
249 }
250 obj.update(x,true,algo_state.iter);
251 if ( computeObj_ ) {
252 algo_state.value = obj.value(x,tol);
253 algo_state.nfval++;
254 }
255 obj.gradient(*(step_state->gradientVec),x,tol);
256 algo_state.ngrad++;
257
258 // Update Secant Information
259 if ( useSecantPrecond_ ) {
260 secant_->updateStorage(x,*(step_state->gradientVec),*gp_,s,algo_state.snorm,algo_state.iter+1);
261 }
262
263 // Update algorithm state
264 (algo_state.iterateVec)->set(x);
265 algo_state.gnorm = step_state->gradientVec->norm();
266 }
267
268 std::string printHeader( void ) const {
269 std::stringstream hist;
270
271 if( verbosity_>0 ) {
272 hist << std::string(109,'-') << "\n";
274 hist << " status output definitions\n\n";
275 hist << " iter - Number of iterates (steps taken) \n";
276 hist << " value - Objective function value \n";
277 hist << " gnorm - Norm of the gradient\n";
278 hist << " snorm - Norm of the step (update to optimization vector)\n";
279 hist << " #fval - Cumulative number of times the objective function was evaluated\n";
280 hist << " #grad - Number of times the gradient was computed\n";
281 hist << " iterCG - Number of Krylov iterations used to compute search direction\n";
282 hist << " flagCG - Krylov solver flag" << "\n";
283 hist << std::string(109,'-') << "\n";
284 }
285
286 hist << " ";
287 hist << std::setw(6) << std::left << "iter";
288 hist << std::setw(15) << std::left << "value";
289 hist << std::setw(15) << std::left << "gnorm";
290 hist << std::setw(15) << std::left << "snorm";
291 hist << std::setw(10) << std::left << "#fval";
292 hist << std::setw(10) << std::left << "#grad";
293 hist << std::setw(10) << std::left << "iterCG";
294 hist << std::setw(10) << std::left << "flagCG";
295 hist << "\n";
296 return hist.str();
297 }
298 std::string printName( void ) const {
299 std::stringstream hist;
300 hist << "\n" << EDescentToString(DESCENT_NEWTONKRYLOV);
301 hist << " using " << krylovName_;
302 if ( useSecantPrecond_ ) {
303 hist << " with " << ESecantToString(esec_) << " preconditioning";
304 }
305 hist << "\n";
306 return hist.str();
307 }
308 std::string print( AlgorithmState<Real> &algo_state, bool print_header = false ) const {
309 std::stringstream hist;
310 hist << std::scientific << std::setprecision(6);
311 if ( algo_state.iter == 0 ) {
312 hist << printName();
313 }
314 if ( print_header ) {
315 hist << printHeader();
316 }
317 if ( algo_state.iter == 0 ) {
318 hist << " ";
319 hist << std::setw(6) << std::left << algo_state.iter;
320 hist << std::setw(15) << std::left << algo_state.value;
321 hist << std::setw(15) << std::left << algo_state.gnorm;
322 hist << "\n";
323 }
324 else {
325 hist << " ";
326 hist << std::setw(6) << std::left << algo_state.iter;
327 hist << std::setw(15) << std::left << algo_state.value;
328 hist << std::setw(15) << std::left << algo_state.gnorm;
329 hist << std::setw(15) << std::left << algo_state.snorm;
330 hist << std::setw(10) << std::left << algo_state.nfval;
331 hist << std::setw(10) << std::left << algo_state.ngrad;
332 hist << std::setw(10) << std::left << iterKrylov_;
333 hist << std::setw(10) << std::left << flagKrylov_;
334 hist << "\n";
335 }
336 return hist.str();
337 }
338}; // class NewtonKrylovStep
339
340} // namespace ROL
341
342#endif
Contains definitions of custom data types in ROL.
Provides the interface to apply upper and lower bound constraints.
Provides definitions for Krylov solvers.
Provides the interface to apply a linear operator.
const ROL::Ptr< Objective< Real > > obj_
HessianNK(const ROL::Ptr< Objective< Real > > &obj, const ROL::Ptr< Vector< Real > > &x)
const ROL::Ptr< Vector< Real > > x_
void apply(Vector< Real > &Hv, const Vector< Real > &v, Real &tol) const
Apply linear operator.
void applyInverse(Vector< Real > &Hv, const Vector< Real > &v, Real &tol) const
Apply inverse of linear operator.
const ROL::Ptr< Vector< Real > > x_
void apply(Vector< Real > &Hv, const Vector< Real > &v, Real &tol) const
Apply linear operator.
const ROL::Ptr< Objective< Real > > obj_
PrecondNK(const ROL::Ptr< Objective< Real > > &obj, const ROL::Ptr< Vector< Real > > &x)
Provides the interface to compute optimization steps with projected inexact Newton's method using lin...
ROL::Ptr< Vector< Real > > gp_
std::string print(AlgorithmState< Real > &algo_state, bool print_header=false) const
Print iterate status.
ROL::Ptr< Secant< Real > > secant_
Secant object (used for quasi-Newton)
int flagKrylov_
Termination flag for Krylov method (used for inexact Newton)
int verbosity_
Verbosity level.
void compute(Vector< Real > &s, const Vector< Real > &x, Objective< Real > &obj, BoundConstraint< Real > &bnd, AlgorithmState< Real > &algo_state)
Compute step.
bool useSecantPrecond_
Whether or not a secant approximation is used for preconditioning inexact Newton.
void initialize(Vector< Real > &x, const Vector< Real > &s, const Vector< Real > &g, Objective< Real > &obj, BoundConstraint< Real > &bnd, AlgorithmState< Real > &algo_state)
Initialize step with bound constraint.
int iterKrylov_
Number of Krylov iterations (used for inexact Newton)
std::string printName(void) const
Print step name.
NewtonKrylovStep(ROL::ParameterList &parlist, const bool computeObj=true)
Constructor.
void update(Vector< Real > &x, const Vector< Real > &s, Objective< Real > &obj, BoundConstraint< Real > &bnd, AlgorithmState< Real > &algo_state)
Update step, if successful.
std::string printHeader(void) const
Print iterate header.
NewtonKrylovStep(ROL::ParameterList &parlist, const ROL::Ptr< Krylov< Real > > &krylov, const ROL::Ptr< Secant< Real > > &secant, const bool computeObj=true)
Constructor.
ROL::Ptr< Krylov< Real > > krylov_
Krylov solver object (used for inexact Newton)
Provides the interface to evaluate objective functions.
virtual void gradient(Vector< Real > &g, const Vector< Real > &x, Real &tol)
Compute gradient.
virtual Real value(const Vector< Real > &x, Real &tol)=0
Compute value.
virtual void update(const Vector< Real > &x, UpdateType type, int iter=-1)
Update objective function.
Provides interface for and implements limited-memory secant operators.
Provides the interface to compute optimization steps.
Definition ROL_Step.hpp:68
virtual void initialize(Vector< Real > &x, const Vector< Real > &g, Objective< Real > &obj, BoundConstraint< Real > &con, AlgorithmState< Real > &algo_state)
Initialize step with bound constraint.
Definition ROL_Step.hpp:88
ROL::Ptr< StepState< Real > > getState(void)
Definition ROL_Step.hpp:73
Defines the linear algebra or vector space interface.
virtual Real norm() const =0
Returns where .
virtual void set(const Vector &x)
Set where .
virtual void scale(const Real alpha)=0
Compute where .
virtual const Vector & dual() const
Return dual representation of , for example, the result of applying a Riesz map, or change of basis,...
virtual void plus(const Vector &x)=0
Compute , where .
virtual ROL::Ptr< Vector > clone() const =0
Clone to make a new (uninitialized) vector.
EKrylov StringToEKrylov(std::string s)
@ DESCENT_NEWTONKRYLOV
ESecant StringToESecant(std::string s)
@ SECANT_USERDEFINED
std::string EDescentToString(EDescent tr)
std::string ESecantToString(ESecant tr)
State for algorithm class. Will be used for restarts.
ROL::Ptr< Vector< Real > > iterateVec