doc/glpk05.tex
author Alpar Juttner <alpar@cs.elte.hu>
Mon, 06 Dec 2010 13:09:21 +0100
changeset 1 c445c931472f
permissions -rw-r--r--
Import glpk-4.45

- Generated files and doc/notes are removed
alpar@1
     1
%* glpk05.tex *%
alpar@1
     2
alpar@1
     3
\chapter{Branch-and-Cut API Routines}
alpar@1
     4
alpar@1
     5
\section{Introduction}
alpar@1
     6
alpar@1
     7
\subsection{Using the callback routine}
alpar@1
     8
alpar@1
     9
The GLPK MIP solver based on the branch-and-cut method allows the
alpar@1
    10
application program to control the solution process. This is attained
alpar@1
    11
by means of the user-defined callback routine, which is called by the
alpar@1
    12
solver at various points of the branch-and-cut algorithm.
alpar@1
    13
alpar@1
    14
The callback routine passed to the MIP solver should be written by the
alpar@1
    15
user and has the following specification:\footnote{The name
alpar@1
    16
{\tt foo\_bar} used here is a placeholder for the callback routine
alpar@1
    17
name.}
alpar@1
    18
alpar@1
    19
\begin{verbatim}
alpar@1
    20
   void foo_bar(glp_tree *tree, void *info);
alpar@1
    21
\end{verbatim}
alpar@1
    22
alpar@1
    23
\noindent
alpar@1
    24
where \verb|tree| is a pointer to the data structure \verb|glp_tree|,
alpar@1
    25
which should be used on subsequent calls to branch-and-cut interface
alpar@1
    26
routines, and \verb|info| is a transit pointer passed to the routine
alpar@1
    27
\verb|glp_intopt|, which may be used by the application program to pass
alpar@1
    28
some external data to the callback routine.
alpar@1
    29
alpar@1
    30
The callback routine is passed to the MIP solver through the control
alpar@1
    31
parameter structure \verb|glp_iocp| (see Chapter ``Basic API Routines'',
alpar@1
    32
Section ``Mixed integer programming routines'', Subsection ``Solve MIP
alpar@1
    33
problem with the branch-and-cut method'') as follows:
alpar@1
    34
alpar@1
    35
\newpage
alpar@1
    36
alpar@1
    37
\begin{verbatim}
alpar@1
    38
   glp_prob *mip;
alpar@1
    39
   glp_iocp parm;
alpar@1
    40
   . . .
alpar@1
    41
   glp_init_iocp(&parm);
alpar@1
    42
   . . .
alpar@1
    43
   parm.cb_func = foo_bar;
alpar@1
    44
   parm.cb_info = ... ;
alpar@1
    45
   ret = glp_intopt(mip, &parm);
alpar@1
    46
   . . .
alpar@1
    47
\end{verbatim}
alpar@1
    48
alpar@1
    49
To determine why it is being called by the MIP solver the callback
alpar@1
    50
routine should use the routine \verb|glp_ios_reason| (described in this
alpar@1
    51
section below), which returns a code indicating the reason for calling.
alpar@1
    52
Depending on the reason the callback routine may perform necessary
alpar@1
    53
actions to control the solution process.
alpar@1
    54
alpar@1
    55
The reason codes, which correspond to various point of the
alpar@1
    56
branch-and-cut algorithm implemented in the MIP solver, are described
alpar@1
    57
in Subsection ``Reasons for calling the callback routine'' below.
alpar@1
    58
alpar@1
    59
To ignore calls for reasons, which are not processed by the callback
alpar@1
    60
routine, it should just return to the MIP solver doing nothing. For
alpar@1
    61
example:
alpar@1
    62
alpar@1
    63
\begin{verbatim}
alpar@1
    64
void foo_bar(glp_tree *tree, void *info)
alpar@1
    65
{     . . .
alpar@1
    66
      switch (glp_ios_reason(tree))
alpar@1
    67
      {  case GLP_IBRANCH:
alpar@1
    68
            . . .
alpar@1
    69
            break;
alpar@1
    70
         case GLP_ISELECT:
alpar@1
    71
            . . .
alpar@1
    72
            break;
alpar@1
    73
         default:
alpar@1
    74
            /* ignore call for other reasons */
alpar@1
    75
            break;
alpar@1
    76
      }
alpar@1
    77
      return;
alpar@1
    78
}
alpar@1
    79
\end{verbatim}
alpar@1
    80
alpar@1
    81
To control the solution process as well as to obtain necessary
alpar@1
    82
information the callback routine may use the branch-and-cut API
alpar@1
    83
routines described in this chapter. Names of all these routines begin
alpar@1
    84
with `\verb|glp_ios_|'.
alpar@1
    85
alpar@1
    86
\subsection{Branch-and-cut algorithm}
alpar@1
    87
alpar@1
    88
This section gives a schematic description of the branch-and-cut
alpar@1
    89
algorithm as it is implemented in the GLPK MIP solver.
alpar@1
    90
alpar@1
    91
\medskip
alpar@1
    92
alpar@1
    93
{\it 1. Initialization}
alpar@1
    94
alpar@1
    95
Set $L:=\{P_0\}$, where $L$ is the {\it active list} (i.e. the list of
alpar@1
    96
active subproblems), $P_0$ is the original MIP problem to be solved.
alpar@1
    97
alpar@1
    98
Set $z^{\it best}:=+\infty$ (in case of minimization) or
alpar@1
    99
$z^{\it best}:=-\infty$ (in case of maximization), where $z^{\it best}$
alpar@1
   100
is {\it incumbent value}, i.e. an upper (minimization) or lower
alpar@1
   101
(maximization) global bound for $z^{\it opt}$, the optimal objective
alpar@1
   102
value for $P^0$.
alpar@1
   103
alpar@1
   104
\medskip
alpar@1
   105
alpar@1
   106
{\it 2. Subproblem selection}
alpar@1
   107
alpar@1
   108
If $L=\varnothing$ then GO TO 9.
alpar@1
   109
alpar@1
   110
Select $P\in L$, i.e. make active subproblem $P$ current.
alpar@1
   111
alpar@1
   112
\medskip
alpar@1
   113
alpar@1
   114
{\it 3. Solving LP relaxation}
alpar@1
   115
alpar@1
   116
Solve $P^{\it LP}$, which is LP relaxation of $P$.
alpar@1
   117
alpar@1
   118
If $P^{\it LP}$ has no primal feasible solution then GO TO 8.
alpar@1
   119
alpar@1
   120
Let $z^{\it LP}$ be the optimal objective value for $P^{\it LP}$.
alpar@1
   121
alpar@1
   122
If $z^{\it LP}\geq z^{\it best}$ (in case of minimization) or
alpar@1
   123
$z^{\it LP}\leq z^{\rm best}$ (in case of maximization) then GO TO 8.
alpar@1
   124
alpar@1
   125
\medskip
alpar@1
   126
alpar@1
   127
{\it 4. Adding ``lazy'' constraints}
alpar@1
   128
alpar@1
   129
Let $x^{\it LP}$ be the optimal solution to $P^{\it LP}$.
alpar@1
   130
alpar@1
   131
If there are ``lazy'' constraints (i.e. essential constraints not
alpar@1
   132
included in the original MIP problem $P_0$), which are violated at the
alpar@1
   133
optimal point $x^{\it LP}$, add them to $P$, and GO TO 3.
alpar@1
   134
alpar@1
   135
\medskip
alpar@1
   136
alpar@1
   137
{\it 5. Check for integrality}
alpar@1
   138
alpar@1
   139
Let $x_j$ be a variable, which is required to be integer, and let
alpar@1
   140
$x^{\it LP}_j\in x^{\it LP}$ be its value in the optimal solution to
alpar@1
   141
$P^{\it LP}$.
alpar@1
   142
alpar@1
   143
If $x^{\it LP}_j$ are integral for all integer variables, then a better
alpar@1
   144
integer feasible solution is found. Store its components, set
alpar@1
   145
$z^{\it best}:=z^{\it LP}$, and GO TO 8.
alpar@1
   146
alpar@1
   147
\medskip
alpar@1
   148
alpar@1
   149
{\it 6. Adding cutting planes}
alpar@1
   150
alpar@1
   151
If there are cutting planes (i.e. valid constraints for $P$),
alpar@1
   152
which are violated at the optimal point $x^{\it LP}$, add them to $P$,
alpar@1
   153
and GO TO 3.
alpar@1
   154
alpar@1
   155
\medskip
alpar@1
   156
alpar@1
   157
{\it 7. Branching}
alpar@1
   158
alpar@1
   159
Select {\it branching variable} $x_j$, i.e. a variable, which is
alpar@1
   160
required to be integer, and whose value $x^{\it LP}_j\in x^{\it LP}$ is
alpar@1
   161
fractional in the optimal solution to $P^{\it LP}$.
alpar@1
   162
alpar@1
   163
Create new subproblem $P^D$ (so called {\it down branch}), which is
alpar@1
   164
identical to the current subproblem $P$ with exception that the upper
alpar@1
   165
bound of $x_j$ is replaced by $\lfloor x^{\it LP}_j\rfloor$. (For
alpar@1
   166
example, if $x^{\it LP}_j=3.14$, the new upper bound of $x_j$ in the
alpar@1
   167
down branch will be $\lfloor 3.14\rfloor=3$.)
alpar@1
   168
alpar@1
   169
Create new subproblem $P^U$ (so called {\it up branch}), which is
alpar@1
   170
identical to the current subproblem $P$ with exception that the lower
alpar@1
   171
bound of $x_j$ is replaced by $\lceil x^{\it LP}_j\rceil$. (For example,
alpar@1
   172
if $x^{\it LP}_j=3.14$, the new lower bound of $x_j$ in the up branch
alpar@1
   173
will be $\lceil 3.14\rceil=4$.)
alpar@1
   174
alpar@1
   175
Set $L:=(L\backslash\{P\})\cup\{P^D,P^U\}$, i.e. remove the current
alpar@1
   176
subproblem $P$ from the active list $L$ and add two new subproblems
alpar@1
   177
$P^D$ and $P^U$ to it. Then GO TO 2.
alpar@1
   178
alpar@1
   179
\medskip
alpar@1
   180
alpar@1
   181
{\it 8. Pruning}
alpar@1
   182
alpar@1
   183
Remove from the active list $L$ all subproblems (including the current
alpar@1
   184
one), whose local bound $\widetilde{z}$ is not better than the global
alpar@1
   185
bound $z^{\it best}$, i.e. set $L:=L\backslash\{P\}$ for all $P$, where
alpar@1
   186
$\widetilde{z}\geq z^{\it best}$ (in case of minimization) or
alpar@1
   187
$\widetilde{z}\leq z^{\it best}$ (in case of maximization), and then
alpar@1
   188
GO TO 2.
alpar@1
   189
alpar@1
   190
The local bound $\widetilde{z}$ for subproblem $P$ is an lower
alpar@1
   191
(minimization) or upper (maximization) bound for integer optimal
alpar@1
   192
solution to {\it this} subproblem (not to the original problem). This
alpar@1
   193
bound is local in the sense that only subproblems in the subtree rooted
alpar@1
   194
at node $P$ cannot have better integer feasible solutions. Note that
alpar@1
   195
the local bound is not necessarily the optimal objective value to LP
alpar@1
   196
relaxation $P^{\it LP}$.
alpar@1
   197
alpar@1
   198
\medskip
alpar@1
   199
alpar@1
   200
{\it 9. Termination}
alpar@1
   201
alpar@1
   202
If $z^{\it best}=+\infty$ (in case of minimization) or
alpar@1
   203
$z^{\it best}=-\infty$ (in case of maximization), the original problem
alpar@1
   204
$P_0$ has no integer feasible solution. Otherwise, the last integer
alpar@1
   205
feasible solution stored on step 5 is the integer optimal solution to
alpar@1
   206
the original problem $P_0$ with $z^{\it opt}=z^{\it best}$. STOP.
alpar@1
   207
alpar@1
   208
\subsection{The search tree}
alpar@1
   209
alpar@1
   210
On the branching step of the branch-and-cut algorithm the current
alpar@1
   211
subproblem is divided into two\footnote{In more general cases the
alpar@1
   212
current subproblem may be divided into more than two subproblems.
alpar@1
   213
However, currently such feature is not used in GLPK.} new subproblems,
alpar@1
   214
so the set of all subproblems can be represented in the form of a rooted
alpar@1
   215
tree, which is called the {\it search} or {\it branch-and-bound} tree.
alpar@1
   216
An example of the search tree is shown on Fig.~1. Each node of the
alpar@1
   217
search tree corresponds to a subproblem, so the terms `node' and
alpar@1
   218
`subproblem' may be used synonymously.
alpar@1
   219
alpar@1
   220
\newpage
alpar@1
   221
alpar@1
   222
\begin{figure}[t]
alpar@1
   223
\noindent\hfil
alpar@1
   224
\xymatrix @R=20pt @C=10pt
alpar@1
   225
{&&&&&&*+<14pt>[o][F=]{A}\ar@{-}[dllll]\ar@{-}[dr]\ar@{-}[drrrr]&&&&\\
alpar@1
   226
&&*+<14pt>[o][F=]{B}\ar@{-}[dl]\ar@{-}[dr]&&&&&*+<14pt>[o][F=]{C}
alpar@1
   227
\ar@{-}[dll]\ar@{-}[dr]\ar@{-}[drrr]&&&*+<14pt>[o][F-]{\times}\\
alpar@1
   228
&*+<14pt>[o][F-]{\times}\ar@{-}[dl]\ar@{-}[d]\ar@{-}[dr]&&
alpar@1
   229
*+<14pt>[o][F-]{D}&&*+<14pt>[o][F=]{E}\ar@{-}[dl]\ar@{-}[dr]&&&
alpar@1
   230
*+<14pt>[o][F=]{F}\ar@{-}[dl]\ar@{-}[dr]&&*+<14pt>[o][F-]{G}\\
alpar@1
   231
*+<14pt>[o][F-]{\times}&*+<14pt>[o][F-]{\times}&*+<14pt>[o][F-]{\times}
alpar@1
   232
&&*+<14pt>[][F-]{H}&&*+<14pt>[o][F-]{I}&*+<14pt>[o][F-]{\times}&&
alpar@1
   233
*+<14pt>[o][F-]{J}&\\}
alpar@1
   234
alpar@1
   235
\bigskip
alpar@1
   236
alpar@1
   237
\noindent\hspace{.8in}
alpar@1
   238
\xymatrix @R=11pt
alpar@1
   239
{*+<20pt>[][F-]{}&*\txt{\makebox[1in][l]{Current}}&&
alpar@1
   240
*+<20pt>[o][F-]{}&*\txt{\makebox[1in][l]{Active}}\\
alpar@1
   241
*+<20pt>[o][F=]{}&*\txt{\makebox[1in][l]{Non-active}}&&
alpar@1
   242
*+<14pt>[o][F-]{\times}&*\txt{\makebox[1in][l]{Fathomed}}\\
alpar@1
   243
}
alpar@1
   244
alpar@1
   245
\begin{center}
alpar@1
   246
Fig. 1. An example of the search tree.
alpar@1
   247
\end{center}
alpar@1
   248
\end{figure}
alpar@1
   249
alpar@1
   250
In GLPK each node may have one of the following four statuses:
alpar@1
   251
alpar@1
   252
$\bullet$ {\it current node} is the active node currently being
alpar@1
   253
processed;
alpar@1
   254
alpar@1
   255
$\bullet$ {\it active node} is a leaf node, which still has to be
alpar@1
   256
processed;
alpar@1
   257
alpar@1
   258
$\bullet$ {\it non-active node} is a node, which has been processed,
alpar@1
   259
but not fathomed;
alpar@1
   260
alpar@1
   261
$\bullet$ {\it fathomed node} is a node, which has been processed and
alpar@1
   262
fathomed.
alpar@1
   263
alpar@1
   264
In the data structure representing the search tree GLPK keeps only
alpar@1
   265
current, active, and non-active nodes. Once a node has been fathomed,
alpar@1
   266
it is removed from the tree data structure.
alpar@1
   267
alpar@1
   268
Being created each node of the search tree is assigned a distinct
alpar@1
   269
positive integer called the {\it subproblem reference number}, which
alpar@1
   270
may be used by the application program to specify a particular node of
alpar@1
   271
the tree. The root node corresponding to the original problem to be
alpar@1
   272
solved is always assigned the reference number 1.
alpar@1
   273
alpar@1
   274
\subsection{Current subproblem}
alpar@1
   275
alpar@1
   276
The current subproblem is a MIP problem corresponding to the current
alpar@1
   277
node of the search tree. It is represented as the GLPK problem object
alpar@1
   278
(\verb|glp_prob|) that allows the application program using API routines
alpar@1
   279
to access its content in the standard way. If the MIP presolver is not
alpar@1
   280
used, it is the original problem object passed to the routine
alpar@1
   281
\verb|glp_intopt|; otherwise, it is an internal problem object built by
alpar@1
   282
the MIP presolver.
alpar@1
   283
alpar@1
   284
Note that the problem object is used by the MIP solver itself during
alpar@1
   285
the solution process for various purposes (to solve LP relaxations, to
alpar@1
   286
perfom branching, etc.), and even if the MIP presolver is not used, the
alpar@1
   287
current content of the problem object may differ from its original
alpar@1
   288
content. For example, it may have additional rows, bounds of some rows
alpar@1
   289
and columns may be changed, etc. In particular, LP segment of the
alpar@1
   290
problem object corresponds to LP relaxation of the current subproblem.
alpar@1
   291
However, on exit from the MIP solver the content of the problem object
alpar@1
   292
is restored to its original state.
alpar@1
   293
alpar@1
   294
To obtain information from the problem object the application program
alpar@1
   295
may use any API routines, which do not change the object. Using API
alpar@1
   296
routines, which change the problem object, is restricted to stipulated
alpar@1
   297
cases.
alpar@1
   298
alpar@1
   299
\subsection{The cut pool}
alpar@1
   300
alpar@1
   301
The {\it cut pool} is a set of cutting plane constraints maintained by
alpar@1
   302
the MIP solver. It is used by the GLPK cut generation routines and may
alpar@1
   303
be used by the application program in the same way, i.e. rather than
alpar@1
   304
to add cutting plane constraints directly to the problem object the
alpar@1
   305
application program may store them to the cut pool. In the latter case
alpar@1
   306
the solver looks through the cut pool, selects efficient constraints,
alpar@1
   307
and adds them to the problem object.
alpar@1
   308
alpar@1
   309
\subsection{Reasons for calling the callback routine}
alpar@1
   310
alpar@1
   311
The callback routine may be called by the MIP solver for the following
alpar@1
   312
reasons.
alpar@1
   313
alpar@1
   314
\subsubsection*{Request for subproblem selection}
alpar@1
   315
alpar@1
   316
The callback routine is called with the reason code \verb|GLP_ISELECT|
alpar@1
   317
if the current subproblem has been fathomed and therefore there is no
alpar@1
   318
current subproblem.
alpar@1
   319
alpar@1
   320
In response the callback routine may select some subproblem from the
alpar@1
   321
active list and pass its reference number to the solver using the
alpar@1
   322
routine \verb|glp_ios_select_node|, in which case the solver continues
alpar@1
   323
the search from the specified active subproblem. If no selection is made
alpar@1
   324
by the callback routine, the solver uses a backtracking technique
alpar@1
   325
specified by the control parameter \verb|bt_tech|.
alpar@1
   326
alpar@1
   327
To explore the active list (i.e. active nodes of the branch-and-bound
alpar@1
   328
tree) the callback routine may use the routines \verb|glp_ios_next_node|
alpar@1
   329
and \verb|glp_ios_prev_node|.
alpar@1
   330
alpar@1
   331
\subsubsection*{Request for preprocessing}
alpar@1
   332
alpar@1
   333
The callback routine is called with the reason code \verb|GLP_IPREPRO|
alpar@1
   334
if the current subproblem has just been selected from the active list
alpar@1
   335
and its LP relaxation is not solved yet.
alpar@1
   336
alpar@1
   337
In response the callback routine may perform some preprocessing of the
alpar@1
   338
current subproblem like tightening bounds of some variables or removing
alpar@1
   339
bounds of some redundant constraints.
alpar@1
   340
alpar@1
   341
\subsubsection*{Request for row generation}
alpar@1
   342
alpar@1
   343
The callback routine is called with the reason code \verb|GLP_IROWGEN|
alpar@1
   344
if LP relaxation of the current subproblem has just been solved to
alpar@1
   345
optimality and its objective value is better than the best known integer
alpar@1
   346
feasible solution.
alpar@1
   347
alpar@1
   348
In response the callback routine may add one or more ``lazy''
alpar@1
   349
constraints (rows), which are violated by the current optimal solution
alpar@1
   350
of LP relaxation, using API routines \verb|glp_add_rows|,
alpar@1
   351
\verb|glp_set_row_name|, \verb|glp_set_row_bnds|, and
alpar@1
   352
\verb|glp_set_mat_row|, in which case the solver will perform
alpar@1
   353
re-optimization of LP relaxation. If there are no violated constraints,
alpar@1
   354
the callback routine should just return.
alpar@1
   355
alpar@1
   356
Optimal solution components for LP relaxation can be obtained with API
alpar@1
   357
routines \verb|glp_get_obj_val|, \verb|glp_get_row_prim|,
alpar@1
   358
\verb|glp_get_row_dual|, \verb|glp_get_col_prim|, and
alpar@1
   359
\verb|glp_get_col_dual|.
alpar@1
   360
alpar@1
   361
\subsubsection*{Request for heuristic solution}
alpar@1
   362
alpar@1
   363
The callback routine is called with the reason code \verb|GLP_IHEUR|
alpar@1
   364
if LP relaxation of the current subproblem being solved to optimality
alpar@1
   365
is integer infeasible (i.e. values of some structural variables of
alpar@1
   366
integer kind are fractional), though its objective value is better than
alpar@1
   367
the best known integer feasible solution.
alpar@1
   368
alpar@1
   369
In response the callback routine may try applying a primal heuristic
alpar@1
   370
to find an integer feasible solution,\footnote{Integer feasible to the
alpar@1
   371
original MIP problem, not to the current subproblem.} which is better
alpar@1
   372
than the best known one. In case of success the callback routine may
alpar@1
   373
store such better solution in the problem object using the routine
alpar@1
   374
\verb|glp_ios_heur_sol|.
alpar@1
   375
alpar@1
   376
\subsubsection*{Request for cut generation}
alpar@1
   377
alpar@1
   378
The callback routine is called with the reason code \verb|GLP_ICUTGEN|
alpar@1
   379
if LP relaxation of the current subproblem being solved to optimality
alpar@1
   380
is integer infeasible (i.e. values of some structural variables of
alpar@1
   381
integer kind are fractional), though its objective value is better than
alpar@1
   382
the best known integer feasible solution.
alpar@1
   383
alpar@1
   384
In response the callback routine may reformulate the {\it current}
alpar@1
   385
subproblem (before it will be splitted up due to branching) by adding to
alpar@1
   386
the problem object one or more {\it cutting plane constraints}, which
alpar@1
   387
cut off the fractional optimal point from the MIP
alpar@1
   388
polytope.\footnote{Since these constraints are added to the current
alpar@1
   389
subproblem, they may be globally as well as locally valid.}
alpar@1
   390
alpar@1
   391
Adding cutting plane constraints may be performed in two ways.
alpar@1
   392
One way is the same as for the reason code \verb|GLP_IROWGEN| (see
alpar@1
   393
above), in which case the callback routine adds new rows corresponding
alpar@1
   394
to cutting plane constraints directly to the current subproblem.
alpar@1
   395
alpar@1
   396
The other way is to add cutting plane constraints to the {\it cut pool},
alpar@1
   397
a set of cutting plane constraints maintained by the solver, rather than
alpar@1
   398
directly to the current subproblem. In this case after return from the
alpar@1
   399
callback routine the solver looks through the cut pool, selects
alpar@1
   400
efficient cutting plane constraints, adds them to the current
alpar@1
   401
subproblem, drops other constraints, and then performs re-optimization.
alpar@1
   402
alpar@1
   403
\subsubsection*{Request for branching}
alpar@1
   404
alpar@1
   405
The callback routine is called with the reason code \verb|GLP_IBRANCH|
alpar@1
   406
if LP relaxation of the current subproblem being solved to optimality
alpar@1
   407
is integer infeasible (i.e. values of some structural variables of
alpar@1
   408
integer kind are fractional), though its objective value is better than
alpar@1
   409
the best known integer feasible solution.
alpar@1
   410
alpar@1
   411
In response the callback routine may choose some variable suitable for
alpar@1
   412
branching (i.e. integer variable, whose value in optimal solution to
alpar@1
   413
LP relaxation of the current subproblem is fractional) and pass its
alpar@1
   414
ordinal number to the solver using the routine
alpar@1
   415
\verb|glp_ios_branch_upon|, in which case the solver splits the current
alpar@1
   416
subproblem in two new subproblems and continues the search. If no choice
alpar@1
   417
is made by the callback routine, the solver uses a branching technique
alpar@1
   418
specified by the control parameter \verb|br_tech|.
alpar@1
   419
alpar@1
   420
\subsubsection*{Better integer solution found}
alpar@1
   421
alpar@1
   422
The callback routine is called with the reason code \verb|GLP_IBINGO|
alpar@1
   423
if LP relaxation of the current subproblem being solved to optimality
alpar@1
   424
is integer feasible (i.e. values of all structural variables of integer
alpar@1
   425
kind are integral within the working precision) and its objective value
alpar@1
   426
is better than the best known integer feasible solution.
alpar@1
   427
alpar@1
   428
Optimal solution components for LP relaxation can be obtained in the
alpar@1
   429
same way as for the reason code \verb|GLP_IROWGEN| (see above).
alpar@1
   430
alpar@1
   431
Components of the new MIP solution can be obtained with API routines
alpar@1
   432
\verb|glp_mip_obj_val|, \verb|glp_mip_row_val|, and
alpar@1
   433
\verb|glp_mip_col_val|. Note, however, that due to row/cut generation
alpar@1
   434
there may be additional rows in the problem object.
alpar@1
   435
alpar@1
   436
The difference between optimal solution to LP relaxation and
alpar@1
   437
corresponding MIP solution is that in the former case some structural
alpar@1
   438
variables of integer kind (namely, basic variables) may have values,
alpar@1
   439
which are close to nearest integers within the working precision, while
alpar@1
   440
in the latter case all such variables have exact integral values.
alpar@1
   441
alpar@1
   442
The reason \verb|GLP_IBINGO| is intended only for informational
alpar@1
   443
purposes, so the callback routine should not modify the problem object
alpar@1
   444
in this case.
alpar@1
   445
alpar@1
   446
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
alpar@1
   447
alpar@1
   448
\newpage
alpar@1
   449
alpar@1
   450
\section{Basic routines}
alpar@1
   451
alpar@1
   452
\subsection{glp\_ios\_reason---determine reason for calling the
alpar@1
   453
callback routine}
alpar@1
   454
alpar@1
   455
\subsubsection*{Synopsis}
alpar@1
   456
alpar@1
   457
\begin{verbatim}
alpar@1
   458
int glp_ios_reason(glp_tree *tree);
alpar@1
   459
\end{verbatim}
alpar@1
   460
alpar@1
   461
\subsubsection*{Returns}
alpar@1
   462
alpar@1
   463
The routine \verb|glp_ios_reason| returns a code, which indicates why
alpar@1
   464
the user-defined callback routine is being called:
alpar@1
   465
alpar@1
   466
\verb|GLP_ISELECT|---request for subproblem selection;
alpar@1
   467
alpar@1
   468
\verb|GLP_IPREPRO|---request for preprocessing;
alpar@1
   469
alpar@1
   470
\verb|GLP_IROWGEN|---request for row generation;
alpar@1
   471
alpar@1
   472
\verb|GLP_IHEUR  |---request for heuristic solution;
alpar@1
   473
alpar@1
   474
\verb|GLP_ICUTGEN|---request for cut generation;
alpar@1
   475
alpar@1
   476
\verb|GLP_IBRANCH|---request for branching;
alpar@1
   477
alpar@1
   478
\verb|GLP_IBINGO |---better integer solution found.
alpar@1
   479
alpar@1
   480
\subsection{glp\_ios\_get\_prob---access the problem object}
alpar@1
   481
alpar@1
   482
\subsubsection*{Synopsis}
alpar@1
   483
alpar@1
   484
\begin{verbatim}
alpar@1
   485
glp_prob *glp_ios_get_prob(glp_tree *tree);
alpar@1
   486
\end{verbatim}
alpar@1
   487
alpar@1
   488
\subsubsection*{Description}
alpar@1
   489
alpar@1
   490
The routine \verb|glp_ios_get_prob| can be called from the user-defined
alpar@1
   491
callback routine to access the problem object, which is used by the MIP
alpar@1
   492
solver. It is the original problem object passed to the routine
alpar@1
   493
\verb|glp_intopt| if the MIP presolver is not used; otherwise it is an
alpar@1
   494
internal problem object built by the presolver.
alpar@1
   495
alpar@1
   496
\subsubsection*{Returns}
alpar@1
   497
alpar@1
   498
The routine \verb|glp_ios_get_prob| returns a pointer to the problem
alpar@1
   499
object used by the MIP solver.
alpar@1
   500
alpar@1
   501
\subsubsection*{Comments}
alpar@1
   502
alpar@1
   503
To obtain various information about the problem instance the callback
alpar@1
   504
routine can access the problem object (i.e. the object of type
alpar@1
   505
\verb|glp_prob|) using the routine \verb|glp_ios_get_prob|. It is the
alpar@1
   506
original problem object passed to the routine \verb|glp_intopt| if the
alpar@1
   507
MIP presolver is not used; otherwise it is an internal problem object
alpar@1
   508
built by the presolver.
alpar@1
   509
alpar@1
   510
\subsection{glp\_ios\_row\_attr---determine additional row attributes}
alpar@1
   511
alpar@1
   512
\subsubsection*{Synopsis}
alpar@1
   513
alpar@1
   514
\begin{verbatim}
alpar@1
   515
void glp_ios_row_attr(glp_tree *tree, int i, glp_attr *attr);
alpar@1
   516
\end{verbatim}
alpar@1
   517
alpar@1
   518
\subsubsection*{Description}
alpar@1
   519
alpar@1
   520
The routine \verb|glp_ios_row_attr| retrieves additional attributes of
alpar@1
   521
$i$-th row of the current subproblem and stores them in the structure
alpar@1
   522
\verb|glp_attr|, which the parameter \verb|attr| points to.
alpar@1
   523
alpar@1
   524
The structure \verb|glp_attr| has the following fields:
alpar@1
   525
alpar@1
   526
\medskip
alpar@1
   527
alpar@1
   528
\noindent\begin{tabular}{@{}p{17pt}@{}p{120.5mm}@{}}
alpar@1
   529
\multicolumn{2}{@{}l}{{\tt int level}}\\
alpar@1
   530
&Subproblem level at which the row was created. (If \verb|level| = 0,
alpar@1
   531
the row was added either to the original problem object passed to the
alpar@1
   532
routine \verb|glp_intopt| or to the root subproblem on generating
alpar@1
   533
``lazy'' or/and cutting plane constraints.)\\
alpar@1
   534
\end{tabular}
alpar@1
   535
alpar@1
   536
\medskip
alpar@1
   537
alpar@1
   538
\noindent\begin{tabular}{@{}p{17pt}@{}p{120.5mm}@{}}
alpar@1
   539
\multicolumn{2}{@{}l}{{\tt int origin}}\\
alpar@1
   540
&The row origin flag:\\
alpar@1
   541
&\verb|GLP_RF_REG |---regular constraint;\\
alpar@1
   542
&\verb|GLP_RF_LAZY|---``lazy'' constraint;\\
alpar@1
   543
&\verb|GLP_RF_CUT |---cutting plane constraint.\\
alpar@1
   544
\end{tabular}
alpar@1
   545
alpar@1
   546
\medskip
alpar@1
   547
alpar@1
   548
\noindent\begin{tabular}{@{}p{17pt}@{}p{120.5mm}@{}}
alpar@1
   549
\multicolumn{2}{@{}l}{{\tt int klass}}\\
alpar@1
   550
&The row class descriptor, which is a number passed to the routine
alpar@1
   551
\verb|glp_ios_add_row| as its third parameter. If the row is a cutting
alpar@1
   552
plane constraint generated by the solver, its class may be the
alpar@1
   553
following:\\
alpar@1
   554
&\verb|GLP_RF_GMI |---Gomory's mixed integer cut;\\
alpar@1
   555
&\verb|GLP_RF_MIR |---mixed integer rounding cut;\\
alpar@1
   556
&\verb|GLP_RF_COV |---mixed cover cut;\\
alpar@1
   557
&\verb|GLP_RF_CLQ |---clique cut.\\
alpar@1
   558
\end{tabular}
alpar@1
   559
alpar@1
   560
\newpage
alpar@1
   561
alpar@1
   562
\subsection{glp\_ios\_mip\_gap---compute relative MIP gap}
alpar@1
   563
alpar@1
   564
\subsubsection*{Synopsis}
alpar@1
   565
alpar@1
   566
\begin{verbatim}
alpar@1
   567
double glp_ios_mip_gap(glp_tree *tree);
alpar@1
   568
\end{verbatim}
alpar@1
   569
alpar@1
   570
\subsubsection*{Description}
alpar@1
   571
alpar@1
   572
The routine \verb|glp_ios_mip_gap| computes the relative MIP gap (also
alpar@1
   573
called {\it duality gap}) with the following formula:
alpar@1
   574
$${\tt gap} = \frac{|{\tt best\_mip} - {\tt best\_bnd}|}
alpar@1
   575
{|{\tt best\_mip}| + {\tt DBL\_EPSILON}}$$
alpar@1
   576
where \verb|best_mip| is the best integer feasible solution found so
alpar@1
   577
far, \verb|best_bnd| is the best (global) bound. If no integer feasible
alpar@1
   578
solution has been found yet, \verb|gap| is set to \verb|DBL_MAX|.
alpar@1
   579
alpar@1
   580
\subsubsection*{Returns}
alpar@1
   581
alpar@1
   582
The routine \verb|glp_ios_mip_gap| returns the relative MIP gap.
alpar@1
   583
alpar@1
   584
\subsubsection*{Comments}
alpar@1
   585
alpar@1
   586
The relative MIP gap is used to measure the quality of the best integer
alpar@1
   587
feasible solution found so far, because the optimal solution value
alpar@1
   588
$z^*$ for the original MIP problem always lies in the range
alpar@1
   589
$${\tt best\_bnd}\leq z^*\leq{\tt best\_mip}$$
alpar@1
   590
in case of minimization, or in the range
alpar@1
   591
$${\tt best\_mip}\leq z^*\leq{\tt best\_bnd}$$
alpar@1
   592
in case of maximization.
alpar@1
   593
alpar@1
   594
To express the relative MIP gap in percents the value returned by the
alpar@1
   595
routine \verb|glp_ios_mip_gap| should be multiplied by 100\%.
alpar@1
   596
alpar@1
   597
\newpage
alpar@1
   598
alpar@1
   599
\subsection{glp\_ios\_node\_data---access application-specific data}
alpar@1
   600
alpar@1
   601
\subsubsection*{Synopsis}
alpar@1
   602
alpar@1
   603
\begin{verbatim}
alpar@1
   604
void *glp_ios_node_data(glp_tree *tree, int p);
alpar@1
   605
\end{verbatim}
alpar@1
   606
alpar@1
   607
\subsubsection*{Description}
alpar@1
   608
alpar@1
   609
The routine \verb|glp_ios_node_data| allows the application accessing a
alpar@1
   610
memory block allocated for the subproblem (which may be active or
alpar@1
   611
inactive), whose reference number is $p$.
alpar@1
   612
alpar@1
   613
The size of the block is defined by the control parameter \verb|cb_size|
alpar@1
   614
passed to the routine \verb|glp_intopt|. The block is initialized by
alpar@1
   615
binary zeros on creating corresponding subproblem, and its contents is
alpar@1
   616
kept until the subproblem will be removed from the tree.
alpar@1
   617
alpar@1
   618
The application may use these memory blocks to store specific data for
alpar@1
   619
each subproblem.
alpar@1
   620
alpar@1
   621
\subsubsection*{Returns}
alpar@1
   622
alpar@1
   623
The routine \verb|glp_ios_node_data| returns a pointer to the memory
alpar@1
   624
block for the specified subproblem. Note that if \verb|cb_size| = 0, the
alpar@1
   625
routine returns a null pointer.
alpar@1
   626
alpar@1
   627
\subsection{glp\_ios\_select\_node---select subproblem to continue the
alpar@1
   628
search}
alpar@1
   629
alpar@1
   630
\subsubsection*{Synopsis}
alpar@1
   631
alpar@1
   632
\begin{verbatim}
alpar@1
   633
void glp_ios_select_node(glp_tree *tree, int p);
alpar@1
   634
\end{verbatim}
alpar@1
   635
alpar@1
   636
\subsubsection*{Description}
alpar@1
   637
alpar@1
   638
The routine \verb|glp_ios_select_node| can be called from the
alpar@1
   639
user-defined callback routine in response to the reason
alpar@1
   640
\verb|GLP_ISELECT| to select an active subproblem, whose reference
alpar@1
   641
number is $p$. The search will be continued from the subproblem
alpar@1
   642
selected.
alpar@1
   643
alpar@1
   644
\newpage
alpar@1
   645
alpar@1
   646
\subsection{glp\_ios\_heur\_sol---provide solution found by heuristic}
alpar@1
   647
alpar@1
   648
\subsubsection*{Synopsis}
alpar@1
   649
alpar@1
   650
\begin{verbatim}
alpar@1
   651
int glp_ios_heur_sol(glp_tree *tree, const double x[]);
alpar@1
   652
\end{verbatim}
alpar@1
   653
alpar@1
   654
\subsubsection*{Description}
alpar@1
   655
alpar@1
   656
The routine \verb|glp_ios_heur_sol| can be called from the user-defined
alpar@1
   657
callback routine in response to the reason \verb|GLP_IHEUR| to provide
alpar@1
   658
an integer feasible solution found by a primal heuristic.
alpar@1
   659
alpar@1
   660
Primal values of {\it all} variables (columns) found by the heuristic
alpar@1
   661
should be placed in locations $x[1]$, \dots, $x[n]$, where $n$ is the
alpar@1
   662
number of columns in the original problem object. Note that the routine
alpar@1
   663
\verb|glp_ios_heur_sol| does {\it not} check primal feasibility of the
alpar@1
   664
solution provided.
alpar@1
   665
alpar@1
   666
Using the solution passed in the array $x$ the routine computes value
alpar@1
   667
of the objective function. If the objective value is better than the
alpar@1
   668
best known integer feasible solution, the routine computes values of
alpar@1
   669
auxiliary variables (rows) and stores all solution components in the
alpar@1
   670
problem object.
alpar@1
   671
alpar@1
   672
\subsubsection*{Returns}
alpar@1
   673
alpar@1
   674
If the provided solution is accepted, the routine
alpar@1
   675
\verb|glp_ios_heur_sol| returns zero. Otherwise, if the provided
alpar@1
   676
solution is rejected, the routine returns non-zero.
alpar@1
   677
alpar@1
   678
\subsection{glp\_ios\_can\_branch---check if can branch upon specified
alpar@1
   679
variable}
alpar@1
   680
alpar@1
   681
\subsubsection*{Synopsis}
alpar@1
   682
alpar@1
   683
\begin{verbatim}
alpar@1
   684
int glp_ios_can_branch(glp_tree *tree, int j);
alpar@1
   685
\end{verbatim}
alpar@1
   686
alpar@1
   687
\subsubsection*{Returns}
alpar@1
   688
alpar@1
   689
If $j$-th variable (column) can be used to branch upon, the routine
alpar@1
   690
returns non-zero, otherwise zero.
alpar@1
   691
alpar@1
   692
\newpage
alpar@1
   693
alpar@1
   694
\subsection{glp\_ios\_branch\_upon---choose variable to branch upon}
alpar@1
   695
alpar@1
   696
\subsubsection*{Synopsis}
alpar@1
   697
alpar@1
   698
\begin{verbatim}
alpar@1
   699
void glp_ios_branch_upon(glp_tree *tree, int j, int sel);
alpar@1
   700
\end{verbatim}
alpar@1
   701
alpar@1
   702
\subsubsection*{Description}
alpar@1
   703
alpar@1
   704
The routine \verb|glp_ios_branch_upon| can be called from the
alpar@1
   705
user-defined callback routine in response to the reason
alpar@1
   706
\verb|GLP_IBRANCH| to choose a branching variable, whose ordinal number
alpar@1
   707
is $j$. Should note that only variables, for which the routine
alpar@1
   708
\verb|glp_ios_can_branch| returns non-zero, can be used to branch upon.
alpar@1
   709
alpar@1
   710
The parameter \verb|sel| is a flag that indicates which branch
alpar@1
   711
(subproblem) should be selected next to continue the search:
alpar@1
   712
alpar@1
   713
\verb|GLP_DN_BRNCH|---select down-branch;
alpar@1
   714
alpar@1
   715
\verb|GLP_UP_BRNCH|---select up-branch;
alpar@1
   716
alpar@1
   717
\verb|GLP_NO_BRNCH|---use general selection technique.
alpar@1
   718
alpar@1
   719
\subsubsection*{Comments}
alpar@1
   720
alpar@1
   721
On branching the solver removes the current active subproblem from the
alpar@1
   722
active list and creates two new subproblems ({\it down-} and {\it
alpar@1
   723
up-branches}), which are added to the end of the active list. Note that
alpar@1
   724
the down-branch is created before the up-branch, so the last active
alpar@1
   725
subproblem will be the up-branch.
alpar@1
   726
alpar@1
   727
The down- and up-branches are identical to the current subproblem with
alpar@1
   728
exception that in the down-branch the upper bound of $x_j$, the variable
alpar@1
   729
chosen to branch upon, is replaced by $\lfloor x_j^*\rfloor$, while in
alpar@1
   730
the up-branch the lower bound of $x_j$ is replaced by
alpar@1
   731
$\lceil x_j^*\rceil$, where $x_j^*$ is the value of $x_j$ in optimal
alpar@1
   732
solution to LP relaxation of the current subproblem. For example, if
alpar@1
   733
$x_j^*=3.14$, the new upper bound of $x_j$ in the down-branch is
alpar@1
   734
$\lfloor 3.14\rfloor=3$, and the new lower bound in the up-branch is
alpar@1
   735
$\lceil 3.14\rceil=4$.)
alpar@1
   736
alpar@1
   737
Additionally the callback routine may select either down- or up-branch,
alpar@1
   738
from which the solver will continue the search. If none of the branches
alpar@1
   739
is selected, a general selection technique will be used.
alpar@1
   740
alpar@1
   741
\newpage
alpar@1
   742
alpar@1
   743
\subsection{glp\_ios\_terminate---terminate the solution process}
alpar@1
   744
alpar@1
   745
\subsubsection*{Synopsis}
alpar@1
   746
alpar@1
   747
\begin{verbatim}
alpar@1
   748
void glp_ios_terminate(glp_tree *tree);
alpar@1
   749
\end{verbatim}
alpar@1
   750
alpar@1
   751
\subsubsection*{Description}
alpar@1
   752
alpar@1
   753
The routine \verb|glp_ios_terminate| sets a flag indicating that the
alpar@1
   754
MIP solver should prematurely terminate the search.
alpar@1
   755
alpar@1
   756
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
alpar@1
   757
alpar@1
   758
\newpage
alpar@1
   759
alpar@1
   760
\section{The search tree exploring routines}
alpar@1
   761
alpar@1
   762
\subsection{glp\_ios\_tree\_size---determine size of the search tree}
alpar@1
   763
alpar@1
   764
\subsubsection*{Synopsis}
alpar@1
   765
alpar@1
   766
\begin{verbatim}
alpar@1
   767
void glp_ios_tree_size(glp_tree *tree, int *a_cnt, int *n_cnt,
alpar@1
   768
      int *t_cnt);
alpar@1
   769
\end{verbatim}
alpar@1
   770
alpar@1
   771
\subsubsection*{Description}
alpar@1
   772
alpar@1
   773
The routine \verb|glp_ios_tree_size| stores the following three counts
alpar@1
   774
which characterize the current size of the search tree:
alpar@1
   775
alpar@1
   776
\verb|a_cnt| is the current number of active nodes, i.e. the current
alpar@1
   777
size of the active list;
alpar@1
   778
alpar@1
   779
\verb|n_cnt| is the current number of all (active and inactive) nodes;
alpar@1
   780
alpar@1
   781
\verb|t_cnt| is the total number of nodes including those which have
alpar@1
   782
been already removed from the tree. This count is increased whenever
alpar@1
   783
a new node appears in the tree and never decreased.
alpar@1
   784
alpar@1
   785
If some of the parameters \verb|a_cnt|, \verb|n_cnt|, \verb|t_cnt| is
alpar@1
   786
a null pointer, the corresponding count is not stored.
alpar@1
   787
alpar@1
   788
\subsection{glp\_ios\_curr\_node---determine current active subprob-\\
alpar@1
   789
lem}
alpar@1
   790
alpar@1
   791
\subsubsection*{Synopsis}
alpar@1
   792
alpar@1
   793
\begin{verbatim}
alpar@1
   794
int glp_ios_curr_node(glp_tree *tree);
alpar@1
   795
\end{verbatim}
alpar@1
   796
alpar@1
   797
\subsubsection*{Returns}
alpar@1
   798
alpar@1
   799
The routine \verb|glp_ios_curr_node| returns the reference number of the
alpar@1
   800
current active subproblem. However, if the current subproblem does not
alpar@1
   801
exist, the routine returns zero.
alpar@1
   802
alpar@1
   803
\newpage
alpar@1
   804
alpar@1
   805
\subsection{glp\_ios\_next\_node---determine next active subproblem}
alpar@1
   806
alpar@1
   807
\subsubsection*{Synopsis}
alpar@1
   808
alpar@1
   809
\begin{verbatim}
alpar@1
   810
int glp_ios_next_node(glp_tree *tree, int p);
alpar@1
   811
\end{verbatim}
alpar@1
   812
alpar@1
   813
\subsubsection*{Returns}
alpar@1
   814
alpar@1
   815
If the parameter $p$ is zero, the routine \verb|glp_ios_next_node|
alpar@1
   816
returns the reference number of the first active subproblem. However,
alpar@1
   817
if the tree is empty, zero is returned.
alpar@1
   818
alpar@1
   819
If the parameter $p$ is not zero, it must specify the reference number
alpar@1
   820
of some active subproblem, in which case the routine returns the
alpar@1
   821
reference number of the next active subproblem. However, if there is
alpar@1
   822
no next active subproblem in the list, zero is returned.
alpar@1
   823
alpar@1
   824
All subproblems in the active list are ordered chronologically, i.e.
alpar@1
   825
subproblem $A$ precedes subproblem $B$ if $A$ was created before $B$.
alpar@1
   826
alpar@1
   827
\subsection{glp\_ios\_prev\_node---determine previous active subproblem}
alpar@1
   828
alpar@1
   829
\subsubsection*{Synopsis}
alpar@1
   830
alpar@1
   831
\begin{verbatim}
alpar@1
   832
int glp_ios_prev_node(glp_tree *tree, int p);
alpar@1
   833
\end{verbatim}
alpar@1
   834
alpar@1
   835
\subsubsection*{Returns}
alpar@1
   836
alpar@1
   837
If the parameter $p$ is zero, the routine \verb|glp_ios_prev_node|
alpar@1
   838
returns the reference number of the last active subproblem. However, if
alpar@1
   839
the tree is empty, zero is returned.
alpar@1
   840
alpar@1
   841
If the parameter $p$ is not zero, it must specify the reference number
alpar@1
   842
of some active subproblem, in which case the routine returns the
alpar@1
   843
reference number of the previous active subproblem. However, if there
alpar@1
   844
is no previous active subproblem in the list, zero is returned.
alpar@1
   845
alpar@1
   846
All subproblems in the active list are ordered chronologically, i.e.
alpar@1
   847
subproblem $A$ precedes subproblem $B$ if $A$ was created before $B$.
alpar@1
   848
alpar@1
   849
\newpage
alpar@1
   850
alpar@1
   851
\subsection{glp\_ios\_up\_node---determine parent subproblem}
alpar@1
   852
alpar@1
   853
\subsubsection*{Synopsis}
alpar@1
   854
alpar@1
   855
\begin{verbatim}
alpar@1
   856
int glp_ios_up_node(glp_tree *tree, int p);
alpar@1
   857
\end{verbatim}
alpar@1
   858
alpar@1
   859
\subsubsection*{Returns}
alpar@1
   860
alpar@1
   861
The parameter $p$ must specify the reference number of some (active or
alpar@1
   862
inactive) subproblem, in which case the routine \verb|iet_get_up_node|
alpar@1
   863
returns the reference number of its parent subproblem. However, if the
alpar@1
   864
specified subproblem is the root of the tree and, therefore, has
alpar@1
   865
no parent, the routine returns zero.
alpar@1
   866
alpar@1
   867
\subsection{glp\_ios\_node\_level---determine subproblem level}
alpar@1
   868
alpar@1
   869
\subsubsection*{Synopsis}
alpar@1
   870
alpar@1
   871
\begin{verbatim}
alpar@1
   872
int glp_ios_node_level(glp_tree *tree, int p);
alpar@1
   873
\end{verbatim}
alpar@1
   874
alpar@1
   875
\subsubsection*{Returns}
alpar@1
   876
alpar@1
   877
The routine \verb|glp_ios_node_level| returns the level of the
alpar@1
   878
subproblem,\linebreak whose reference number is $p$, in the
alpar@1
   879
branch-and-bound tree. (The root subproblem has level 0, and the level
alpar@1
   880
of any other subproblem is the level of its parent plus one.)
alpar@1
   881
alpar@1
   882
\subsection{glp\_ios\_node\_bound---determine subproblem local\\bound}
alpar@1
   883
alpar@1
   884
\subsubsection*{Synopsis}
alpar@1
   885
alpar@1
   886
\begin{verbatim}
alpar@1
   887
double glp_ios_node_bound(glp_tree *tree, int p);
alpar@1
   888
\end{verbatim}
alpar@1
   889
alpar@1
   890
\subsubsection*{Returns}
alpar@1
   891
alpar@1
   892
The routine \verb|glp_ios_node_bound| returns the local bound for
alpar@1
   893
(active or inactive) subproblem, whose reference number is $p$.
alpar@1
   894
alpar@1
   895
\subsubsection*{Comments}
alpar@1
   896
alpar@1
   897
The local bound for subproblem $p$ is an lower (minimization) or upper
alpar@1
   898
(maximization) bound for integer optimal solution to {\it this}
alpar@1
   899
subproblem (not to the original problem). This bound is local in the
alpar@1
   900
sense that only subproblems in the subtree rooted at node $p$ cannot
alpar@1
   901
have better integer feasible solutions.
alpar@1
   902
alpar@1
   903
On creating a subproblem (due to the branching step) its local bound is
alpar@1
   904
inherited from its parent and then may get only stronger (never weaker).
alpar@1
   905
For the root subproblem its local bound is initially set to
alpar@1
   906
\verb|-DBL_MAX| (minimization) or \verb|+DBL_MAX| (maximization) and
alpar@1
   907
then improved as the root LP relaxation has been solved.
alpar@1
   908
alpar@1
   909
Note that the local bound is not necessarily the optimal objective value
alpar@1
   910
to corresponding LP relaxation.
alpar@1
   911
alpar@1
   912
\subsection{glp\_ios\_best\_node---find active subproblem with best
alpar@1
   913
local bound}
alpar@1
   914
alpar@1
   915
\subsubsection*{Synopsis}
alpar@1
   916
alpar@1
   917
\begin{verbatim}
alpar@1
   918
int glp_ios_best_node(glp_tree *tree);
alpar@1
   919
\end{verbatim}
alpar@1
   920
alpar@1
   921
\subsubsection*{Returns}
alpar@1
   922
alpar@1
   923
The routine \verb|glp_ios_best_node| returns the reference number of
alpar@1
   924
the active subproblem, whose local bound is best (i.e. smallest in case
alpar@1
   925
of minimization or largest in case of maximization). However, if the
alpar@1
   926
tree is empty, the routine returns zero.
alpar@1
   927
alpar@1
   928
\subsubsection*{Comments}
alpar@1
   929
alpar@1
   930
The best local bound is an lower (minimization) or upper (maximization)
alpar@1
   931
bound for integer optimal solution to the original MIP problem.
alpar@1
   932
alpar@1
   933
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
alpar@1
   934
alpar@1
   935
\newpage
alpar@1
   936
alpar@1
   937
\section{The cut pool routines}
alpar@1
   938
alpar@1
   939
\subsection{glp\_ios\_pool\_size---determine current size of the cut\\
alpar@1
   940
pool}
alpar@1
   941
alpar@1
   942
\subsubsection*{Synopsis}
alpar@1
   943
alpar@1
   944
\begin{verbatim}
alpar@1
   945
int glp_ios_pool_size(glp_tree *tree);
alpar@1
   946
\end{verbatim}
alpar@1
   947
alpar@1
   948
\subsubsection*{Returns}
alpar@1
   949
alpar@1
   950
The routine \verb|glp_ios_pool_size| returns the current size of the
alpar@1
   951
cut pool, that is, the number of cutting plane constraints currently
alpar@1
   952
added to it.
alpar@1
   953
alpar@1
   954
\subsection{glp\_ios\_add\_row---add constraint to the cut pool}
alpar@1
   955
alpar@1
   956
\subsubsection*{Synopsis}
alpar@1
   957
alpar@1
   958
\begin{verbatim}
alpar@1
   959
int glp_ios_add_row(glp_tree *tree, const char *name,
alpar@1
   960
      int klass, int flags, int len, const int ind[],
alpar@1
   961
      const double val[], int type, double rhs);
alpar@1
   962
\end{verbatim}
alpar@1
   963
alpar@1
   964
\subsubsection*{Description}
alpar@1
   965
alpar@1
   966
The routine \verb|glp_ios_add_row| adds specified row (cutting plane
alpar@1
   967
constraint) to the cut pool.
alpar@1
   968
alpar@1
   969
The cutting plane constraint should have the following format:
alpar@1
   970
$$\sum_{j\in J}a_jx_j\left\{\begin{array}{@{}c@{}}\geq\\\leq\\
alpar@1
   971
\end{array}\right\}b,$$
alpar@1
   972
where $J$ is a set of indices (ordinal numbers) of structural variables,
alpar@1
   973
$a_j$ are constraint coefficients, $x_j$ are structural variables, $b$
alpar@1
   974
is the right-hand side.
alpar@1
   975
alpar@1
   976
The parameter \verb|name| specifies a symbolic name assigned to the
alpar@1
   977
constraint (1 up to 255 characters). If it is \verb|NULL| or an empty
alpar@1
   978
string, no name is assigned.
alpar@1
   979
alpar@1
   980
The parameter \verb|klass| specifies the constraint class, which must
alpar@1
   981
be either zero or a number in the range from 101 to 200. The application
alpar@1
   982
may use this attribute to distinguish between cutting plane constraints
alpar@1
   983
of different classes.\footnote{Constraint classes numbered from 1 to 100
alpar@1
   984
are reserved for GLPK cutting plane generators.}
alpar@1
   985
alpar@1
   986
The parameter \verb|flags| currently is not used and must be zero.
alpar@1
   987
alpar@1
   988
Ordinal numbers of structural variables (i.e. column indices) $j\in J$
alpar@1
   989
and numerical values of corresponding constraint coefficients $a_j$ must
alpar@1
   990
be placed in locations \verb|ind[1]|, \dots, \verb|ind[len]| and
alpar@1
   991
\verb|val[1]|, \dots, \verb|val[len]|, respectively, where
alpar@1
   992
${\tt len}=|J|$ is the number of constraint coefficients,
alpar@1
   993
$0\leq{\tt len}\leq n$, and $n$ is the number of columns in the problem
alpar@1
   994
object. Coefficients with identical column indices are not allowed.
alpar@1
   995
Zero coefficients are allowed, however, they are ignored.
alpar@1
   996
alpar@1
   997
The parameter \verb|type| specifies the constraint type as follows:
alpar@1
   998
alpar@1
   999
\verb|GLP_LO| means inequality constraint $\Sigma a_jx_j\geq b$;
alpar@1
  1000
alpar@1
  1001
\verb|GLP_UP| means inequality constraint $\Sigma a_jx_j\leq b$;
alpar@1
  1002
alpar@1
  1003
The parameter \verb|rhs| specifies the right-hand side $b$.
alpar@1
  1004
alpar@1
  1005
All cutting plane constraints in the cut pool are identified by their
alpar@1
  1006
ordinal numbers 1, 2, \dots, $size$, where $size$ is the current size
alpar@1
  1007
of the cut pool. New constraints are always added to the end of the cut
alpar@1
  1008
pool, thus, ordinal numbers of previously added constraints are not
alpar@1
  1009
changed.
alpar@1
  1010
alpar@1
  1011
\subsubsection*{Returns}
alpar@1
  1012
alpar@1
  1013
The routine \verb|glp_ios_add_row| returns the ordinal number of the
alpar@1
  1014
cutting plane constraint added, which is the new size of the cut pool.
alpar@1
  1015
alpar@1
  1016
\subsubsection*{Example}
alpar@1
  1017
alpar@1
  1018
\begin{verbatim}
alpar@1
  1019
/* generate triangle cutting plane:
alpar@1
  1020
   x[i] + x[j] + x[k] <= 1 */
alpar@1
  1021
. . .
alpar@1
  1022
/* add the constraint to the cut pool */
alpar@1
  1023
ind[1] = i, val[1] = 1.0;
alpar@1
  1024
ind[2] = j, val[2] = 1.0;
alpar@1
  1025
ind[3] = k, val[3] = 1.0;
alpar@1
  1026
glp_ios_add_row(tree, NULL, TRIANGLE_CUT, 0, 3, ind, val,
alpar@1
  1027
                GLP_UP, 1.0);
alpar@1
  1028
\end{verbatim}
alpar@1
  1029
alpar@1
  1030
\subsubsection*{Comments}
alpar@1
  1031
alpar@1
  1032
Cutting plane constraints added to the cut pool are intended to be then
alpar@1
  1033
added only to the {\it current} subproblem, so these constraints can be
alpar@1
  1034
globally as well as locally valid. However, adding a constraint to the
alpar@1
  1035
cut pool does not mean that it will be added to the current
alpar@1
  1036
subproblem---it depends on the solver's decision: if the constraint
alpar@1
  1037
seems to be efficient, it is moved from the pool to the current
alpar@1
  1038
subproblem, otherwise it is simply dropped.\footnote{Globally valid
alpar@1
  1039
constraints could be saved and then re-used for other subproblems, but
alpar@1
  1040
currently such feature is not implemented.}
alpar@1
  1041
alpar@1
  1042
Normally, every time the callback routine is called for cut generation,
alpar@1
  1043
the cut pool is empty. On the other hand, the solver itself can generate
alpar@1
  1044
cutting plane constraints (like Gomory's or mixed integer rounding
alpar@1
  1045
cuts), in which case the cut pool may be non-empty.
alpar@1
  1046
alpar@1
  1047
\subsection{glp\_ios\_del\_row---remove constraint from the cut pool}
alpar@1
  1048
alpar@1
  1049
\subsubsection*{Synopsis}
alpar@1
  1050
alpar@1
  1051
\begin{verbatim}
alpar@1
  1052
void glp_ios_del_row(glp_tree *tree, int i);
alpar@1
  1053
\end{verbatim}
alpar@1
  1054
alpar@1
  1055
\subsubsection*{Description}
alpar@1
  1056
alpar@1
  1057
The routine \verb|glp_ios_del_row| deletes $i$-th row (cutting plane
alpar@1
  1058
constraint) from the cut pool, where $1\leq i\leq size$ is the ordinal
alpar@1
  1059
number of the constraint in the pool, $size$ is the current size of the
alpar@1
  1060
cut pool.
alpar@1
  1061
alpar@1
  1062
Note that deleting a constraint from the cut pool leads to changing
alpar@1
  1063
ordinal numbers of other constraints remaining in the pool. New ordinal
alpar@1
  1064
numbers of the remaining constraints are assigned under assumption that
alpar@1
  1065
the original order of constraints is not changed. Let, for example,
alpar@1
  1066
there be four constraints $a$, $b$, $c$ and $d$ in the cut pool, which
alpar@1
  1067
have ordinal numbers 1, 2, 3 and 4, respectively, and let constraint
alpar@1
  1068
$b$ have been deleted. Then after deletion the remaining constraint $a$,
alpar@1
  1069
$c$ and $d$ are assigned new ordinal numbers 1, 2 and 3, respectively.
alpar@1
  1070
alpar@1
  1071
To find the constraint to be deleted the routine \verb|glp_ios_del_row|
alpar@1
  1072
uses ``smart'' linear search, so it is recommended to remove constraints
alpar@1
  1073
in a natural or reverse order and avoid removing them in a random order.
alpar@1
  1074
alpar@1
  1075
\subsubsection*{Example}
alpar@1
  1076
alpar@1
  1077
\begin{verbatim}
alpar@1
  1078
/* keep first 10 constraints in the cut pool and remove other
alpar@1
  1079
   constraints */
alpar@1
  1080
while (glp_ios_pool_size(tree) > 10)
alpar@1
  1081
   glp_ios_del_row(tree, glp_ios_pool_size(tree));
alpar@1
  1082
\end{verbatim}
alpar@1
  1083
alpar@1
  1084
\newpage
alpar@1
  1085
alpar@1
  1086
\subsection{glp\_ios\_clear\_pool---remove all constraints from the cut
alpar@1
  1087
pool}
alpar@1
  1088
alpar@1
  1089
\subsubsection*{Synopsis}
alpar@1
  1090
alpar@1
  1091
\begin{verbatim}
alpar@1
  1092
void glp_ios_clear_pool(glp_tree *tree);
alpar@1
  1093
\end{verbatim}
alpar@1
  1094
alpar@1
  1095
\subsubsection*{Description}
alpar@1
  1096
alpar@1
  1097
The routine \verb|glp_ios_clear_pool| makes the cut pool empty deleting
alpar@1
  1098
all existing rows (cutting plane constraints) from it.
alpar@1
  1099
alpar@1
  1100
%* eof *%