File cplex is a student version of CPLEX 10.1.0 (for use with AMPL).
See file LICENSE, which pertains to cplex; for commercial versions
of CPLEX, see http://www.ilog.com/products/cplex/ for contact details.
Details on using "cplex" appear in /netlib/ampl/solvers/cplex/README.cplex,
which for convenience is appended below. ILOG has also kindly made older
documentation on using AMPL and cplex available as
http://netlib.bell-labs.com/netlib/ampl/solvers/cplex/ampl70.pdf
Please note that ampl70.pdf refers to "cplexamp" rather than "cplex".
Unrestricted versions of cplex.exe sold by ILOG/CPLEX are called
"cplexamp" when delivered. You are free to name things as you wish
(with suitable changes to "option solver" lines in AMPL scripts if
necessary). One approach that avoids changes to AMPL scripts is to
call the interactive CPLEX front end provided by ILOG/CPLEX "icplex"
rather than "cplex", and to use "cplex" for the solver that works with
AMPL.
--------------------------------------------------------
/netlib/ampl/solvers/cplex/README.cplex (8 October 2006)
--------------------------------------------------------
The solver "cplex" uses CPLEX (a trademark of ILOG) to solve integer,
mixed-integer, linear programming, and quadratic problems, including
(with version 9) problems with quadratic constraints possibly
involving integer variables. Normally cplex is invoked by AMPL's
solve command, which gives the invocation
cplex stub -AMPL
in which stub.nl is an AMPL generic output file (possibly written
by "ampl -obstub" or "ampl -ogstub"). After solving the problem,
cplex writes a stub.sol file for use by ampl's solve and solution
commands. When you run ampl, this all happens automatically if you
give the AMPL commands
option solver cplex;
solve;
You can control cplex by setting the environment variable cplex_options
appropriately (either by using ampl's option command, or by using the
shell's set and export commands before you invoke ampl). You can put
one or more (white-space separated) phrases in $cplex_options. A few
of the phrases are single words:
Phrase Meaning
autoopt Use CPLEX's automatic choice of optimizer (dualopt).
baropt Use the barrier algorithm (unless there are discrete
variables).
bestbound Return in suffix .bestbound the best known bound on
the objective value. For MIP problems with a finite
bestnode value (see below), bestbound = bestnode.
bestnode For MIP problems, return the best node value in
suffix .bestnode of the current objective and problem.
For non-MIP problems, set .bestnode to Infinity for
minimization, -Infinity for maximization. If CPLEX
has not yet found a best node value for a MIP problem,
or has completely solved the problem, it similarly
supplies 1e75 or -1e75 as the best node value.
concurrentopt
When hardware and licensing permit, try several methods in
parallel (with CPLEX versions >= 8).
dual Solve the dual problem.
dualopt Use a dual simplex algorithm.
maximize Maximize the objective, regardless of model specification.
minimize Minimize the objective, regardless of model specification.
siftopt Solve ever larger sequences of subproblems until the
whole LP is solved (versions of CPLEX >= 8).
primal Solve the primal problem (default, via "dualratio"
described below). Note that "primal", "dual", and
"dualratio" determine what problem is presented to
CPLEX, whereas "primalopt" or "dualopt" instructs CPLEX to
use a primal or dual simplex algorithm, rather than making
an automatic choice (which currently is always "dualopt").
Thus "primalopt" and "dualopt" are orthogonal to "primal",
"dual", and "dualratio". The best settings depend on
the problem.
primalopt Use a primal simplex algorithm.
relax Ignore integrality: treat integer variables as continuous.
version Show the current version.
Others are name-value pairs, possibly separated by '=', as in
iterations 600
or
iterations=600
or
iterations = 600
any of which limits cplex to 600 iterations. The following names may
occur in name-value pairs. For convenience, the list also repeats the
above single-word phrases, denoting them with "Single-word phrase".
Name Comment
absmipgap Absolute mixed-integer optimality gap tolerance
(for difference between current best integer solution
and optimal value of LP relaxation). Default 0.
advance Whether to use advance basis information (initial
primal and dual variable values and basis indicators).
Default 1 (yes).
aggcutlim Bound on the number of constraints aggregated to
generate flow-cover and mixed-integer-rounding cuts;
default = 3.
aggfill Synonym for "agglim".
aggregate 0 or 1 (default 1): Whether to make substitutions
to reduce the number of rows.
agglim Variables that appear in more than agglim rows
(default 10) will not be substituted away by the
"aggregate" algorithm.
aggtol Pivot tolerance for aggregating. It seldom needs
fiddling. Default = .05; must be in [1e-10, .99].
autoopt Single-word phrase: use CPLEX's automatic choice of
optimizer (currently dualopt for LPs).
autopt Synonym for "autoopt".
backtrack Tolerance (> 0, default 0.85) for when to backtrack
during branch & bound. Low values tend to pure
best-bound search. High values (> 1) tend to pure
depth-first search. Values less than the default
are often good when subproblems are expensive.
baralg How to start the barrier algorithm:
0 (default) = 1 for MIP subproblems, else 3
1 = infeasibility-estimate start
2 = infeasibility-constant start
3 = standard start
barcorr Limit on centering corrections in each iteration
of the barrier algorithm:
-1 = decide automatically (default)
nonnegative = at most that many
bardisplay Specifies how much the barrier algorithm chatters:
0 = no output (default)
1 = one line per iteration
2 = more output
bargrowth Tolerance for detecting unbounded faces in the
barrier algorithm: higher values make the test
for unbounded faces harder to satisfy.
Default = 1e6.
bariterlim Maximum barrier iterations allowed (default 200).
barobjrange Limit on the absolute objective value before the
barrier algorithm considers the problem unbounded.
Default = 1e15.
baropt Single-word phrase: use the barrier algorithm
(unless there are discrete variables).
barstart Barrier starting-point algorithm:
1 = dual is 0 (default)
2 = estimate dual
3 = average of primal estimate, 0 dual
4 = average of primal and dual estimates
barstartalg Synonym for "barstart".
baroutofcore New in CPLEX 7.1: whether the barrier solver should
use disk (out-of-core) storage for Cholesky factors:
0 = no (default)
1 = yes.
Withdrawn in CPLEX 10.0.
barthreads Maximum threads for the barrier algorithm
(1 unless you have a CPLEX license for multiple
threads).
barvarup Upper bound imposed by barrier algorithm on
variables with infinite upper bound; used to
prevent trouble with problems having unbounded
optimal faces. Default = 1e20.
Withdrawn from CPLEX 9.0.
basis_cond Whether to show the condition number of the simplex
basis in the solve_message and to return its value
in the problem.basis_cond and objective.basis_cond
suffixes. (Default = 0 = no; 1 = yes).
basisinterval Number of interations between savings of current
simplex basis to a file. Default = 50000.
Deprecated in 10.0.
bbinterval For nodeselect = 2, select the best-bound node,
rather than the best-estimate node, every
bbinterval iterations (default 7); 0 means
always use the best-estimate node.
bestbound Single-word phrase requesting return of .bestbound
suffix. See the description above.
bestnode Single-word phrase requesting return of .bestnode
suffix. See the description above.
boundstr Whether to use bound strengthening in solving MIPs:
-1 (default) = automatic choice
0 = never
1 = always
branch Branching direction for integer variables:
-1 = down, 0 = algorithm decides, 1 = up;
default = 0.
branchdir Synonym for "branch".
cliquecuts Synonym for "cliques".
cliques Whether to use clique cuts in solving MIPs:
-1 = never
0 = automatic choice (default)
1 = at root node only
2 = at all nodes
clocktype Kind of times CPLEX reports during the solution
process:
1 = CPU time (default)
2 = wall clock time (total elapsed time)
coeffreduce Whether to use coefficient reduction when
preprocessing MIPS:
0 = no
1 = yes (default)
comptol Convergence tolerance for barrier algorithm:
the algorithm stops when the relative
complementarity is < bartol (default 1e-8).
concurrentopt Single-word phrase: with CPLEX versions >= 8
and when hardware and licensing permit, try
several methods in parallel.
conflictdisplay New in CPLEX 10.0. What to report when the
conflict finder is working:
0 = nothing
1 = summary (default)
2 = detailed display
covercuts Synonym for "covers".
covers Whether to use cover cuts in solving MIPs:
-1 = never
0 = automatic choice (default)
1 = at root node only
2 = at all nodes
crash Crash strategy (used to obtain starting basis);
possible values = -1, 0, 1; default = 1.
The best setting is problem-dependent and
can only be found by experimentation.
0 completely ignores the objective.
crossover Causes the barrier algorithm to be run (in the
absence of discrete variables) and specifies
whether to "crossover" to an optimal simplex
basis afterwards:
0 = no crossover
1 = crossover with primal simplex
(default for baropt)
2 = crossover with dual simplex
cutpass Number of passes permitted when generating MIP
cutting plane:
-1 = none
0 = automatic choice (default)
positive = at most that many passes
cutsfactor Limit MIP cuts added to (cutsfactor-1)*m, where m
is the original number of rows (after presolve).
Default = 4.
dgradient Pricing algorithm for dual simplex (default 0):
0 = choose automatically
1 = standard dual pricing
2 = steepest-edge pricing
3 = steepest-edge pricing in slack space
4 = steepest-edge with unit initial norms.
dense Synonym for "densecol".
densecol If positive, minimum nonzeros in a column for
the barrier algorithm to consider the column dense.
If 0 (default), this tolerance is selected
automatically.
dependency Whether to use CPLEX's presolve dependency
checker:
0 = no (default)
1 = yes
disjcuts Whether to generate MIP disjunctive cuts:
-1 = no
0 = automatic choice (default)
1, 2, 3 = ever more aggressive generation
display Frequency of displaying LP progress information:
0 (default) = never; 1 = each factorization;
2 = each iteration.
doperturb 1 means initially perturb the problem (by an
amount governed by "perturbation", which is
described below). 0 (default) means let the
algorithm decide. Setting doperturb to 1
is occasionally helpful for highly degenerate
problems.
dparam Used with syntax "dparam=n=d" (no spaces), where n
is a decimal integer, the number of a CPLEX "double"
(i.e., floating-point valued) parameter. If d is a
decimal floating-point value, assign d to "double"
parameter n. If d is ?, report the current value of
"double" paramter n. This facility provides a way
to modify "double" parameters that have not (yet)
been assigned a keyword.
dual Single-word phrase: solve the dual problem.
dualopt Single-word phrase: use a dual simplex algorithm.
dualratio If neither "primal" nor "dual" was specified and
"dual" is possible (e.g., no integer variables and
no node and arc declarations), choose between
"primal" and "dual" as follows.
Let m = number of rows, n = number of columns;
if m - n > dualthresh > 0 or m > dualratio*n,
solve the dual; otherwise solve the primal.
Defaults: dualthresh = 0, dualratio = 3.
dualthresh See dualratio.
endbasis "endbasis foo" writes the final basis to
file "foo" (in BAS format).
endtree File for writing final branch & bound search tree.
Withdrawn from CPLEX 10.
endvector File for writing solution from barrier algorithm
without crossover: only meaningful with
"baropt crossover=0".
feasibility Amount by which basic variables can violate
their bounds. Default = 1e-6; possible
values are between 1e-9 and 1e-4.
feasopt New in CPLEX 10.0. For infeasible problems,
whether to find a feasible point for a relaxed
problem (see feasoptobj):
0 = no (default)
1 = find a feasible point
2 = find a "best" feasible point
feasoptobj New in CPLEX 10.0. Objective for "feasopt":
1 = minimize sum of constraint and variable
bound relaxations
2 = minimize number of constraint and variable
bounds relaxed (a MIP problem,
generally harder than feasoptobj = 1)
3 = minimize sum of squares of constraint and
variable bound relaxations
file Synonym for "writeprob".
finalfactor Whether to factor the basis after "uncrushing"
the problem at the end.
0 = no
1 = yes (default)
Withdrawn in CPLEX 10.0.
flowcuts Whether to use flow cuts in solving MIPs:
-1 = never
0 = automatic choice (default)
1 = at root node only
2 = at all nodes
flowpathcuts Whether to generate MIP flow-path cuts:
-1 = no
0 = automatic choice (default)
1 = moderate generation
2 = aggressive generation
fraccand Limit on number of candidate variables when
generating Gomory cuts for MIP problems:
default = 200
fraccuts Whether to generate MIP fractional Gomory
cuts:
-1 = no
0 = decide automatically (default)
1 = generate moderately
2 = generate aggressively
fracpass Limit on number of passes to generate MIP
fractional Gomory cuts:
0 = automatic choice (default)
positive = at most that many passes
fractionalcuts Synonym for "fracpass".
growth Synonym for "bargrowth".
gubcuts Whether to use GUB cuts in solving MIPs:
-1 = never
0 = automatic choice (default)
1 = at root node only
2 = at all nodes
heurfreq Withdrawn from CPLEX 9.0. How often to apply
"node heuristics" for MIPS:
-1 = never
0 = automatic choice (default)
n > 0 = every n nodes
heuristic Deprecated: withdrawn from CPLEX 9.0.
What heuristic to use on MIPS for finding an
initial integer solution:
-1 = none
0 = decide automatically (default)
1 = use a rounding heuristic at node 0
heuristicfreq Synonym for "heurfreq".
iisfind Whether to find and return an IIS (irreducible
infeasible set of variables and constraints) if
the problem is infeasible:
0 = no (default)
1 = find an IIS quickly
2 = try to find a small IIS.
IIS details are returned in suffix .iis, which
assumes one of the values "non" for variables
and constraints not in the IIS; "low" for
variables or inequality constraint bodies whose lower
bounds are in the IIS; "upp" for variables and
inequality constraint bodies whose upper bounds are
in the IIS; and "fix" for equality constraints that
are in the IIS.
impliedcuts Whether to use implied cuts in solving MIPs:
-1 = never
0 = automatic choice (default)
1 = at root node only
2 = at all nodes
integrality Amount by which an integer variable can differ
from the nearest integer and still be considered
feasible. Default = 1e-5; must be >= 1e-9.
intwarntol Do not warn about perturbations to "integer"
variables to make them integers when the maximum
perturburation is at most intwarntol (default 1e-9);
see "round".
iparam Used with syntax "iparam=n=i" (no spaces), where n
is a decimal integer, the number of a CPLEX integer
parameter. If i is a decimal integer, assign i to
integer parameter n. If i is ?, report the current
value of integer paramter n. This facility provides
a way to modify integer parameters that have not (yet)
been assigned a keyword.
iterations Limit on total LP iterations; default 5000000.
iterlim Synonym for "iterations".
limitperturb Synonym for "perturblimit".
lazy Whether to recognize suffix .lazy on constraints
(new for CPLEX 10): sum of
1 ==> treat .lazy = 1 as lazy constraint
2 ==> treat .lazy = 2 as user cut
Default lazy = 3 ==> treat both. (Suffix .lazy on
constraints is ignored if not 0, 1, or 2 modulo 3.)
lbheur Whether to use a local branching heuristic in an
attempt to improve new incumbents found during a
MIP search. (Default = 0 = no; 1 = yes.)
logfile Name of file to receive all CPLEX messages.
lowercutoff For maximization problems involving integer
variables, skip any branch whose LP relaxation's
optimal value is less than lowercutoff. Warning:
if lowercutoff is too large, the problem will
appear infeasible. Default = -Infinity.
lowerobj Stop minimizing when the objective value
goes below lowerobj. Default = -1e75.
lowerobjlim Synonym for "lowerobj".
lpdisplay Synonym for "display".
lpiterlim Synonym for "iterations".
lptimelim Synonym for "time".
markowitz Pivot tolerance; default = 0.01; must be between
0.0001 and 0.99999. Bigger values may improve
numerical properties of the solution (and may
take more time).
maximize Single-word phrase: maximize the objective,
regardless of model specifications.
memoryemphasis New in CPLEX 10.0. Whether to compress data to reduce
the memory used, which may make some information (e.g.,
basis condition) unavailable:
0 = no (default)
1 = yes
minimize Single-word phrase: minimize the objective,
regardless of model specifications.
mipalg Algorithm used on mixed-integer subproblems
(default 2):
1 = primal simplex
2 = dual simplex (plus primal, if dual fails)
3 = network alg., then dual simplex
4 = barrier with crossover
The next two are not available in CPLEX versions >= 8;
specify mipcrossover instead.
5 = dual to iteration limit, then barrier
6 = barrier without crossover
mipalgorithm Synonym for "mipalg".
mipbasis Whether to compute a basis and dual variables for MIP
problems when endbasis is not specified:
0 = no
1 = yes (default)
This keyword is new with driver version 20040716.
When endbasis is specified, mipbasis=1 is assumed.
Otherwise, when mipbasis=0 is specified for a MIP
problem, no solver-status values for variables are
returned to AMPL.
mipcrossover Crossover method used when using the barrier
method for MIP subproblems:
-1 = no crossover
0 (default) = automatic choice
1 = primal
2 = dual
mipcuts Sets all nine of cliques, covers, disjcuts,
flowcuts, flowpathcuts, fraccuts, gubcuts,
impliedcuts and mircuts to the specified value.
mipdisplay Frequency of displaying branch-and-bound
information (for optimizing integer variables):
0 (default) = never
1 = each integer feasible solution
2 = every "mipinterval" nodes
3 = every "mipinterval" nodes plus
information on LP relaxations
(as controlled by "display")
4 = same as 2, plus LP relaxation info.
5 = same as 2, plus LP subproblem info.
mipemphasis Whether to emphasize seeking optimality (0 =
default) or finding feasible solutions (1).
For CPLEX versions >= 8, two other values are
possible: emphasizing optimality over
feasibility (2) and emphasizing best bound (3).
mipgap Relative tolerance for optimizing integer
variables: stop if
abs((best bound) - (best integer))
< mipgap * (1 + abs(best bound)).
Default = 1e-4; must be between 1e-9 and 1.
mipinterval Frequency of node logging for mipdisplay 2 or 3.
Default = 1.
mipordertype Synonym for "ordertype".
mipsolutions Stop branch-and-bound for integer variables
after finding "mipsolutions" feasible solutions.
Default = 2100000000.
mipstart Synonym for "mipstartvalue".
mipstartalg Which LP algorithm to use in solving the initial
MIP subproblem:
1 = primal simplex
2 = dual simplex (default)
3 = netopt
4 = barrier with crossover
The next two are not available in CPLEX versions >= 8;
specify mipcrossover instead.
5 = dual simplex to iteration limit,
then barrier
6 = barrier without crossover
mipstartstatus Whether to use incoming variable and constraint
statuses if the problem has integer variables:
0 = no
1 = yes (default).
mipstartvalue Whether to use initial guesses in problems with
integer variables:
0 = no
1 = yes (default).
mipsubalg Synonym for "mipalg".
mipthreads Maximum threads for the MIP algorithm
(1 unless you have a CPLEX license for multiple
threads).
mircuts Whether to generate MIP rounding cuts:
-1 = no
0 = automatic choice (default)
1 = moderate generation
2 = aggressive generation
nameround Whether to mangle variable and constraint names
by turning [ and ] into ( and ), respectively:
0 = no (default)
1 = yes.
This only matters if you specify endbasis=...
or startbasis=... or perhaps writeprob=something.lp
and have instructed AMPL to write .row and .col files.
(It is usually better to let AMPL's status facilities
convey basis information.) An alternative under Unix
is to use the "tr" command to make the above changes
if they are needed.
netdisplay Which objective value to show when using the
network simplex algorithm with display > 0:
0 = none
1 = true objective
2 = penalized objective (default)
netfeasibility Feasibility tolerance for the network simplex
algorithm. Default = 1e-6; possible values are
between 1e-11 and 1e-1.
netfind Algorithm for finding embedded networks:
1 = extract only the natural network (default)
2 = use reflection scaling
3 = use general scaling
netfinder Synonym for "netfind".
netiterations Limit on network simplex iterations.
Default = large (e.g., 2100000000).
netopt 0 means never invoke the network optimizer.
1 (default) means invoke the network optimizer
only if the model had node and arc declarations.
2 means always invoke the network optimizer
(unless there are integer variables); the network
optimizer may be able to find and exploit an
embedded network.
netoptimality Tolerance for optimality of reduced costs in the
network simplex algorithm. Default 1e-6; must be
between 1e-11 and 1e-1.
netpricing How to price in the network simplex algorithm:
0 = automatic choice (default)
1 = partial pricing
2 = multiple partial pricing
3 = multiple partial pricing with sorting
node Synonym for "nodes".
nodefile Whether to save node information in a temporary file:
0 = no
1 (default) = compressed node file in memory
2 = node file on disk
3 = compressed node file on disk.
nodefiledir Synonym for workfiledir. Prior to CPLEX 7.1,
this directory is just for node information files.
nodefilelim Maximum size for the node file. Default 1e75.
Removed from CPLEX 7.1. See workfilelim.
nodefilesize Synonym for "nodefilelim".
nodelim Synonym for "nodes".
nodes Stop branch-and-bound for integer variables
after "nodes" LP relaxations. Default = 2100000000.
nodesel Strategy for choosing next node while optimizing
integer variables:
0 = depth-first search;
1 = breadth-first search (default);
2 = best-estimate search;
3 = alternate best-estimate search.
nodeselect Synonym for "nodesel".
numericalemphasis New in CPLEX 10.0. Whether to try to improve
numerical accuracy (at a possible cost of time or
memory):
0 = no (default)
1 = yes
objdifference Amount added to (for maximizing) or subtracted
from (for minimizing) the best (so far) feasible
objective value while optimizing integer variables.
Subsequent nodes will be ignored if their LP
relaxations have optimal values worse than this
sum. Default = 0. Positive values may speed
the search -- and may cause the optimal solution
to be missed.
objno 0 (default) = first objective, 1 = second, etc.
This only matters if you declare several
objectives and do not use AMPL's drop or objective
commands to select a single objective for the solver
(CPLEX) to see.
optimality Tolerance for optimality of reduced costs.
Default 1e-6; must be between 1e-9 and 1e-1.
optimize Synonym for "primal".
ordering Ordering algorithm used by the barrier algorithm
0 = automatic choice (default)
1 = approximate minimum degree
2 = approximate minimum fill
3 = nested disection.
ordertype How to generate default priorities for integer
variables when no .priority suffix is specified:
0 = do not generate priorities (default)
1 = use decreasing costs
2 = use increasing bound range
3 = use coefficient count.
outlev Synonym for "display".
pdswitch Whether to switch algorithms (from primal to dual
or vice versa) when undoing perturbations or shifts
made during the primal or dual simplex algorithm:
-1 = no
0 = automatic choice (default)
1 = yes
pdswitch=1 may help if there are many cycles
of undoing perturbations. New in CPLEX 7.0,
hidden in CPLEX 7.1.
perturb Synonym for "doperturb".
perturbation Amount by which to perturb variable bounds
when perturbing problems (see "doperturb").
Default 1e-6; must be positive.
perturbconst Synonym for "perturbation".
perturblim Number of stalled simplex iterations before the
problem is perturbed. Default = 0 = automatic.
perturblimit Synonym for "perturblim".
pgradient Pricing algorithm for primal simplex (default 0):
-1 = reduced-cost pricing
0 = hybrid reduced-cost and Devex pricing
1 = Devex pricing
2 = steepest-edge pricing
3 = steepest-edge with slack initial norms
4 = full pricing.
plconpri Priority (default 1) for SOS2 constraints for nonconvex
piecewise-linear terms in constraints.
plobjpri Priority (default 2) for SOS2 constraints for nonconvex
piecewise-linear terms in objectives.
polishtime New in CPLEX 10.0: seconds to spend "polishing" integer
solutions. Default 0.
precompress New in CPLEX 7.1: whether to compress the original
problem after CPLEX's presolve:
-1 = no
0 = automatic choice (default)
1 = yes
Withdrawn in CPLEX 10.0.
predual Whether CPLEX's presolve phase should present the
CPLEX solution algorithm with the primal (-1) or
dual (1) problem or (default = 0) should decide
which automatically. Specifying "predual=1" often
gives better performance than specifying just "dual",
but sometimes "dual predual=1" is still better.
prelinear Whether CPLEX's presolve should do full reductions
or only linear ones. Default = 1 = full.
prereduce Kinds of reductions permitted during CPLEX presolve:
0 = none
1 = only primal
2 = only dual
3 = both primal and dual (default)
prerelax Whether to use CPLEX's presolve on the initial LP
relaxation of a MIP:
0 = no (default)
1 = yes.
prepass Limit on number of CPLEX presolve passes.
Default = -1 = decide limit automatically.
presolve 0 or 1 (default 1): Whether to run CPLEX's presolve
algorithm.
presolvedual Synonym for "predual".
presolvenode -1, 0, or 1 (default 0): Whether to run CPLEX's
presolve at each node of the MIP branch-and-bound
tree: -1 = no; 1 = yes; 0 = automatic choice.
prestats 0 or 1 (default 0): Whether to print results of
and time for CPLEX's "aggregate" and "presolve"
algorithms.
pricing Size of pricing candidate list (for partial pricing).
0 (default) means the algorithm decides.
primal Single-word phrase: solve the primal problem.
primalopt Use the primal simplex algorithm.
priorities Whether to consider priorities for MIP branching:
0 = no
1 = yes (default)
probe Whether to do variable probing when solving MIPs
(which sometimes dramatically affects performance,
for better or worse):
-1 = no
0 = automatic choice (default)
1, 2, or 3 = ever more probing.
probetime New in CPLEX 10.0. Limit in seconds on time spent
probing. Default = 1e75.
qcpconvergetol Convergence tolerance on relative complementarity for
problems with quadratic constraints. Default = 1e-6.
rays Whether to return suffix .unbdd when the objective is
unbounded or suffix .dunbdd when the constraints are
infeasible:
0 = neither
1 = just .unbdd
2 = just .dunbdd
3 = both (default)
readbasis BAS file containing starting basis.
readvector VEC file containing starting point for barrier alg.
refactor LP iterations between refactorizing the basis.
0 (default) means the algorithm decides.
relax Single-word phrase: ignore integrality.
relaxpresolve Synonym for "prerelax".
relobjdif Synonym for "relobjdiff".
relobjdiff If the objdifference parameter is 0,
relobjdiff times the absolute value of the objective
value is added to (for maximizing) or subtracted
from (for minimizing) the best (so far) feasible
objective value while optimizing integer variables.
Subsequent nodes will be ignored if their LP
relaxations have optimal values worse than this
sum. Default = 0. Positive values may speed
the search -- and may cause the optimal solution
to be missed.
repairtries New in CPLEX 10.0. How many times to try to
repair in infeasible MIP starting guess:
-1 = none
0 = automatic choice (default)
> 0 = that many times
repeatpresolve New in CPLEX 10.0. Whether to repeat CPLEX's
presolve at MIP nodes:
-1 = automatic choice (default)
0 = no
1 = presolve again ignoring cuts
2 = presolve again considering cuts
3 = presolve again considering cuts and
allowing new root cuts
return_mipgap Whether to return mipgap suffixes or include
mipgap values in the solve_message: sum of
1 = return relmapgap suffix
2 = return absmipgap suffix
4 = suppress mipgap values in solve_message
The suffixes are on the objective and problem;
returned suffix values are +Infinity if no integer-
feasible solution has been found, in which case no
mipgap values are reported in the solve_message.
Default = 0. See also bestbound and bestnode above.
rinsheur Relaxation INduced neighborhood Search HEURistic
for MIP problems:
-1 = none
0 = automatic choice of interval (default)
n (for n > 0) = every n nodes.
relpresolve Synonym for "prerelax".
rootheuristic Withdrawn from CPLEX 9.0. Synonym for "heuristic".
round Whether to round integer variables to integral
values before returning the solution, and whether
to report that CPLEX returned noninteger values
for integer values (default 1): sum of
1 ==> round nonintegral integer variables
2 ==> do not modify solve_result
4 ==> do not modify solve_message
8 ==> modify solve_result and solve_message
even if maxerr < intwarntol (default 1e-9).
Modifications take place only if CPLEX assigned
nonintegral values to one or more integer variables.
scale How to scale the problem:
-1 = no scaling
0 (default) = equilibration
1 = a more aggressive scheme that sometimes helps.
sensitivity Whether to return sensitivy information for the
objective (in suffixes .up for the largest value
of a variable's cost coefficient or constraint's
right-hand side before the optimal basis changes,
.down for the smallest such value, and .current for
the current cost coefficient or right-hand side):
0 = no (default)
1 = yes.
siftingopt Synonym for "siftopt".
siftopt Single-word phrase: on LPs with CPLEX versions >= 8,
solve ever larger sequences of subproblems until the
whole LP is solved.
simthreads Maximum threads for the dual simplex algorithm
(currently always 1). Withdrawn from CPLEX 9.0.
singular Maximum number of times CPLEX should try to
repair the basis when it encounters singularities.
Default = 10.
singularlim Synonym for "singular".
solutionlim Synonym for "mipsolutions".
sos 0 or 1 (default 1): Whether to honor declared
suffixes .sosno and .ref describing SOS sets.
Each distinct nonzero .sosno value designates an SOS
set, of type 1 for positive .sosno values and of type
2 for negative values. The .ref suffix contains
corresponding reference values.
sos2 0 or 1 (default 1): Whether to tell CPLEX about SOS2
constraints for nonconvex piecewise-linear terms.
startalg Synonym for "mipstartalg".
startalgorithm Synonym for "mipstartalg".
startbasis "startbasis foo" reads the initial basis
(in BAS format) from file "foo".
starttree File for reading initial branch-and-bound search tree.
Withdrawn from CPLEX 10.
strongcand Length of the candidate list for "strong branching"
when solving MIPs: default 10.
strongit Number of simplex iterations on each variable in
the candidate list during strong branching.
Default = 0 = automatic choice.
strongthreads Maximum threads during strong branching.
Default = 1.
subalg Synonym for "mipalg".
subalgorithm Synonym for "mipalg".
submipnodelim Limit on nodes searched by relaxation induced
neighborhood search (RINS) heuristic for MIP
problems. Default = 500.
symmetry New in CPLEX 9.1? Whether to break symmetry during
preprocessing of MIP problems:
-1 = automatic choice (default)
0 = no
1 = moderate effort
2 = more effort
3 = still more effort
threads Default maximum number of threads for any of
the parallel CPLEX optimizers (limited also
by licensing). Default = 1. May be overridden
by more specific limits, such as barthreads or
mipthreads.
time Time limit in seconds; default = 1e75.
timelimit Synonym for "time".
timing 1 means write time to stdout, 2 to stderr,
3 to both stdout and stderr, 0 to neither.
Default = 0.
tranopt Synonym for "dualopt".
treelimit Synonym for "treememory".
treememlim Synonym for "treememory".
treememory Max. megabytes of memory (default 128) to use for
branch-and-bound tree.
uppercutoff For minimization problems involving integer
variables, skip any branch whose LP relaxation's
optimal value is more than uppercutoff. Warning:
if uppercutoff is too small, the problem will
appear infeasible. Default = Infinity.
upperobj Stop maximizing when the objective value
goes above upperobj. Default = 1e75.
upperobjlim Synonym for "upperobj".
varsel Strategy for selecting the next branching
variable during integer branch-and-bound:
-1 = branch on variable with smallest
integer infeasibility;
0 = algorithm decides (default);
1 = branch on variable with largest
integer infeasibility;
2 = branch based on pseudo costs (new in CPLEX 6.5);
3 = strong branching;
4 = branch based on pseudo reduced costs
(with CPLEX 6.5; varsel 2 had this meaning
with earlier versions of CPLEX).
varselect Synonym for "varsel".
version Single-word phrase: show the current version.
wantsol In a stand-alone invocation (no -AMPL on the
command line), what solution information to write.
Sum of
1 = write .sol file
2 = primal variables to stdout
4 = dual variables to stdout
8 = suppress solution message
workfiledir Directory where CPLEX creates a temporary
subdirectory for temporary files, e.g., for
node information and Cholesky factors.
workfilelim Maximum size in megabytes for in-core work "files".
Default 128.
writebas Synonym for "endbasis".
writeprob Name of file to which the problem is written
in a format determined by the name's suffix:
.sav = binary SAV file;
.mps = MPS file, original names;
.lp = LP file, original names;
.rmp = MPS file, generic names;
.rew = MPS file, generic names;
.rlp = LP file, generic names.
SAV and LP formats are peculiar to CPLEX.
writevector Synonym for "endvector".
xxxstart Whether to read .xxx files to resume an
interrupted run of CPLEX (deprecated in 10.0):
0 = no (default)
1 = yes.
When CPLEX finds the problem unbounded and a basis is available,
a ray is returned in suffix unbdd.
For problems with several integer (or binary) variable declarations,
it sometimes helps to specify branching priorities for the integer
variables. When CPLEX has a choice of which integer variable to
bound (or fix) during its branch-and-bound algorithm, it chooses
a variable with the highest priority. You can declare suffix priority
and assign individual priorities (between 0 and 2147483647) to each
integer variable. Assigning .priority values is the preferred
approach, but you can also use the older and less flexible scheme of
specifying priorities in the environment variable $mip_priorities,
which should contain zero or more white-space separated pairs of the
form
variable-name priority
Each priority should be an integer between 1 and 2147483647.
All components of an indexed variable have the same priority;
the variable-name should not have a subscript.
If suffix priority is present, option mip_priorities is ignored.
You can also declare suffix direction and use it to convey
individual branching directions (-1, 0, or 1, as with the branch
keyword described above) for each integer variable. If suffix
direction is present, the branch keyword applies only to variables
whose .direction is 0.
When using $mip_priorities, you must arrange for AMPL to write a .col
file. See the discussion of auxiliary files on p. 333 of the AMPL book:
unless $cplex_auxfiles already contains 'c', issue the AMPL command
option cplex_auxfiles c;
before saying "solve;". Similarly, if you use AMPL's write command,
first make sure $auxfiles contains 'c'; if not, issue the AMPL command
option auxfiles c;
The keywords '!quit' and '!echo' may appear at the beginning of
$mip_priorities. If cplex finds an error in $mip_priorities, it bails
out unless '!quit' appears in $mip_priorities before the error.
Normally cplex reports the priorities it acquires from $mip_priorities,
but '!echo' suppresses this reporting. For example, if Buy is an
integer variable, you might try the AMPL commands
option mip_priorities 'Buy 3', cplex_auxfiles ''; solve;
option mip_priorities '!quit Buy 3'; solve;
option cplex_auxfiles c; solve;
option mip_priorities '!echo Buy 3'; solve;
option mip_priorities '!quit !echo Buy 3'; solve;
-----------------------
solve_result_num values
=======================
Here is a table of solve_result_num values that "cplex" can return
to an AMPL session, along with the text that appears in the associated
solve_message.
Value Message
0 optimal solution
1 primal has unbounded optimal face
2 optimal integer solution
3 optimal integer solution within mipgap or absmipgap
4 primal optimal (no dual solution available)
100 best solution found, primal-dual feasible
102 optimal (non-)integer solution
103 optimal (non-)integer solution within mipgap or absmipgap
110 optimal with unscaled infeasibilities
111 integer optimal with unscaled infeasibilities
121 feasible relaxed sum in feasopt
122 optimal relaxed sum in feasopt
123 feasible relaxed infeasibility count in feasopt
124 optimal relaxed infeasibility count in feasopt
125 feasible relaxed quadratic penalty in feasopt
126 optimal relaxed quadratic penalty in feasopt
200 infeasible problem
201 infeasible with phase II singularities
202 infeasible with phase I singularities
204 converged, dual feasible, primal infeasible
205 converged, primal and dual infeasible
206 best solution found, primal infeasible
207 best solution found, primal-dual infeasible
208 infeasible or unbounded in presolve
209 integer infeasible or unbounded in presolve
210 infeasible problem found by dualopt; .dunbdd returned
220 integer infeasible
300 unbounded problem
301 converged, primal feasible, dual infeasible
302 best solution found, dual infeasible
310 unbounded problem found by primalopt; .unbdd returned
320 integer unbounded ray
400 objective limit exceeded in phase II
401 iteration limit in phase II
402 iteration limit in phase I
403 time limit in phase II
404 time limit in phase I
405 primal objective limit reached
406 dual objective limit reached
410 node limit with no integer solution
411 time limit with no integer solution
412 treememory limit with no integer solution
413 node file limit with no integer solution
420 mixed-integer solutions limit
421 node limit with integer solution
422 time limit with integer solution
423 treememory limit with integer solution
424 node file limit with integer solution
500 unrecoverable failure
501 aborted in phase II
502 aborted in phase I
503 aborted in barrier, dual infeasible
504 aborted in barrier, primal infeasible
505 aborted in barrier, primal and dual infeasible
506 aborted in barrier, primal and dual feasible
507 aborted in crossover
508 solution found, numerical difficulties
509 solution found, inconsistent equations
510 unrecoverable failure with no integer solution
511 aborted, no integer solution
512 out of memory, no tree; no integer solution
520 unrecoverable failure with integer solution
521 aborted, integer solution exists
523 out of memory, no tree; solution may exist
530 bug? problem has no variables
531 bug? Error return from named CPLEX routine
540 Diagonal QP Hessian has elements of the wrong sign
541 QP Hessian has diag. elements of the wrong sign
542 QP Hessian is not positive definite
550 problem has (nonquadratic) nonlinear constraints
551 problem has a nonlinear objective
552 problem has nonlinear integer variables
553 problem has integer variables and a quadratic objective
554 problem has unlinearized piecewise-linear terms
555 problem has a quadratic objective involving division by 0
556 nonlinear objective without CPLEX Barrier option (for QPs)
557 CPLEX MIP option needed to handle piecewise-linear terms
558 quadratic constraint involves division by zero
559 bug: no quadratic terms in "nonlinear" constraint
560 error in $cplex_options
561 surprise return from a CPLEX routine (perhaps a driver bug)
562 constraint is not convex quadratic
563 logical constraint is not an indicator constraint
570 CPLEX licensing problem
571 not licensed to solve MIP problems
572 not licensed to use the barrier algorithm
573 not licensed to solve MIQP or QCP problems
The CPLEX driver is supposed to distinguish infeasibility from unboundedness,
so solve_result_num should never be 208 or 209. If you find an example where
one of these values does occur, please send it to us.
-------------------
If you have a CPLEX 4.0 manual, you may find the following table
helpful. It relates the above names (in name-value pairs) to
SET commands described in chapters 8 and 9 of the CPLEX 4.0 manual.
Name CPLEX SET command
agglim PREPROCESSING colnonzeros
bardisplay SET BARRIER display
bargrowth SET BARRIER limits growth
bariterlim SET BARRIER limits iterations
barobjrange SET BARRIER limits objrange
barthreads SET BARRIER LIMIT THREADS
barvarup SET BARRIER LIMITS varupper
branch SET MIP STRATEGY branch
cliques MIP STRATEGY cliques
coeffreduce PREPROCESSING coeffreduce
comptol SET BARRIER convergetol
covers MIP STRATEGY covers
crash SET SIMPLEX crash
dense SET BARRIER colnonzeros
dependency PREPROCESSING dependency
dgradient SET SIMPLEX dgradient
display SET SIMPLEX display
doperturb SET SIMPLEX perturbation
feasibility SET SIMPLEX TOLERANCE feasibility
heuristic MIP STRATEGY heuristic
integrality SET MIP TOLERANCE integrality
iterations SET SIMPLEX LIMIT iterations
iterlim SET SIMPLEX LIMIT iterations
lowercutoff SET MIP STRATEGY lowercutoff
lowerobj SET SIMPLEX LIMIT lowerobj
markowitz SET SIMPLEX TOLERANCE markowitz
mipdisplay SET MIP display
mipgap SET MIP TOLERANCE mipgap
mipinterval SET MIP interval
mipsolutions SET MIP LIMIT solutions
mipstartalg MIP STRATEGY startalgorithm
mipsubalg MIP STRATEGY subalgorithm
mipthreads SET MIP LIMIT THREADS
netfind SIMPLEX netfind
nodefile SET MIP STRATEGY FILE
nodes SET MIP LIMIT nodes
nodesel SET MIP STRATEGY nodeselect
objdifference SET MIP TOLERANCE objdifference
optimality SET SIMPLEX TOLERANCE optimality
ordering SET BARRIER ordering
outlev SET SIMPLEX display
perturbation SET SIMPLEX perturbation
pgradient SET SIMPLEX pgradient
pricing SET SIMPLEX pricing
priorities MIP STRATEGY order
refactor SET SIMPLEX refactor
relobjdiff MIP TOLERANCES relobjdifference
scale SET read scale
simthreads SET SIMPLEX LIMIT THREADS
singular SET SIMPLEX LIMIT singular
sosmin MIP LIMITS minsossize
time SET timelimit
uppercutoff SET MIP TOLERANCE uppercutoff
upperobj SET SIMPLEX LIMIT upperobj
varsel SET MIP STRATEGY variableselect
-------------------
Keywords added for CPLEX 6.5:
barstart
bbinterval
boundstr
cutsfactor
flowcuts
gubcuts
heuristicfreq
impliedcuts
mipcrossover
mipcuts
nodefilelim
ordertype
perturblimit
prerelax
probe
strongcand
strongit
strongthreads
Keywords no longer available in CPLEX 6.5:
reducecostfix Whether to use reduced cost fixing in MIP problems:
0 = no
1 = yes (default)
sosmin Minimum number of members in a set for the
set to be considered of SOS type 3.
sosscan 0 or 1 (default 0): Whether to scan for SOS3 sets.
Keywords added for CPLEX 7.0:
aggcutlim
barstartalg
cutpass
disjcuts
flowpathcuts
fraccand
fracpass
heurfreq
limitperturb
mipemphasis
mipordertype
mipstart
mircuts
nodefiledir
nodefilesize
pdswitch
perturblim
prereduce
presolvedual
presolvenode
relaxpresolve
relpresolve
Keywords removed for CPLEX 7.1:
nodefilelim
nodefilesize
Keywords added for CPLEX 8.0 (excluding undocumented keywords,
which are discussed below):
autoopt
autopt
baroutofcore
concurrent
concurrentopt
precompress
rays
siftingopt
siftopt
threads
workfilelim
Keywords added for CPLEX 9.0:
finalfactor
qcpconvergetol
rinsheur
submipnodelim
Keywords removed for CPLEX 9.0:
barvarup
heuristic
rootheuristic
simthreads
Keywords added for CPLEX 9.1:
basis_cond
lbheur
prelinear
prepass
-------------------
If you invoke "cplex stub -AMPL" or "cplex stub", you can also supply
additional command-line arguments of the form name=value. Such
arguments override specifications in $cplex_options. Example:
ampl -obfoo foo.model foo.data
nohup cplex foo timing=2 2>>times&
to solve a problem whose solution will take a while; after cplex
finishes,
ampl foo.model foo.data -
solution foo.sol;
display ... /* things involving the computed solution */;
(Here, - denotes standard input, and ampl reads the "solution..."
and "display..." lines.)
When invoked with no arguments, cplex either operates as described
in the CPLEX manual (when compiled with -DStand_alone on systems
where the CPLEX callable library provides cpxmain()), or complains
about not having a stub.
-------------------
In addition to the keyword phrases described above, cplex recognizes
some directives (name-value pairs, separated by white space and/or an
= sign, appearing either in $cplex_options or on the command line) for
settings not described in the official CPLEX documentation. These
"undocumented" directives all begin with underscore. You are unlikely
ever to need to use them, as the default settings suffice in most
cases. They are provided for the rare cases where CPLEX technical
support may recommend their use. The directive names appear at the
beginning of the output you will get by invoking
cplex "-="
(or 'cplexamp "-="' if you follow the ILOG naming practice).
*************************
If you have questions about or find bugs with this stuff,
please contact:
David M. Gay
dmg@ampl.com