AMPL-Gurobi Parameter Reference

aggfill amount of fill allowed during aggregation during Gurobi's presolve (default: -1)
aggregate whether to use aggregation during Gurobi presolve:
0 = no (sometimes reduces numerical errors)
1 = yes (default)
ams_eps relative tolerance for reporting alternate MIP solutions (default = no limit)
ams_epsabs absolute tolerance for reporting alternate MIP solutions (default = no limit)
ams_limit limit on number of alternate MIP solutions written (default = 10)
ams_mode search mode for MIP solutions when several are desired:
0 = just focus on finding an optimal solution (default)
1 = make some effort at finding additional solutions
2 = try to find the "ams_limit" best solutions.
ams_stub stub for alternate MIP solutions. The number of alternative MIP solution files written is determined by three keywords:
ams_limit gives the maximum number of files written;
ams_eps gives a relative tolerance on the objective values of alternative solutions;
ams_epsabs gives an absolute tolerance on how much worse the objectives can be.
barconvtol tolerance on the relative difference between the primal and dual objectives for stopping the barrier algorithm (default: 1e-8)
barcorrectors Limit on the number of central corrections done in each barrier iteration (default -1 = automatic choice)
barhomogeneous Whether to use the homogeneous barrier algorithm (e.g., when lpmethod=2 or nodemethod=2 is specified):
-1 = only when solving a MIP node relaxation (default)
0 = never
1 = always.
The homogeneous barrier algorithm can detect infeasibility or unboundedness directly, without crossover, but is a bit slower than the nonhomogeneous barrier algorithm.
bariterlim Limit on the number of barrier iterations (default: none)
barorder Ordering used to reduce fill in sparse-matrix factorizations during the barrier algorithm:
-1 = automatic choice
0 = approximate minimum degree
1 = nested dissection
barqcptol convergence tolerance on the relative difference between primal and dual objective values for barrier algorithms when solving problems with quadratic constraints (default: 1e-6)
basis whether to use or return a basis:
0 = no
1 = use incoming basis (if provided)
2 = return final basis
3 = both (1 + 2 = default)
For problems with integer variables and quadratic constraints, basis = 0 is assumed quietly.
basisdebug whether to honor basis and solnsens when an optimal solution was not found:
0 = only if a feasible solution was found (default)
1 = yes
2 = no
bestbndstop stop once the best bound on the objective value is at least as good as this value.
bestbound whether to return suffix .bestbound for the best known bound on the objective value:
0 = no (default)
1 = yes
bestobjstop stop after a feasible solution with objective value at least as good as this value has been found.
branchdir which child node to explore first when branching:
-1 = explore "down" branch first
0 = explore "most promising" branch first (default)
1 = explore "up" branch first
cliquecuts clique cuts: overrides "cuts"; choices as for "cuts"
cloudid use Gurobi Instant Cloud with this "accessID".
cloudkey use Gurobi Instant Cloud with this "secretKey". Both cloudid and cloudkey are required.
cloudpool optional "machine pool" to use with Gurobi Instant Cloud.
concurrentmip how many independent MIP solves to allow at once when multiple threads are available. The available threads are divided as evenly as possible among the concurrent solves. (default: 1)
covercuts cover cuts: overrides "cuts"; choices as for "cuts"
crossover how to transform a barrier solution to a basic one:
-1 = automatic choice (default)
0 = none: return an interior solution
1 = push dual vars first, finish with primal simplex
2 = push dual vars first, finish with dual simplex
3 = push primal vars first, finish with primal simplex
4 = push primal vars first, finish with dual simplex
crossoverbasis strategy for initial basis construction during crossover:
0 = favor speed (default)
1 = favor numerical stability
cutagg maximum number of constraint aggregation passes during cut generation (-1 = default = no limit); overrides "cuts"
cutoff If the optimal objective value is no better than cutoff, report "objective cutoff" and do not return a solution; default: -∞ for minimizing, +∞ for maximizing
cutpasses maximum number of cutting-plane passes to do during root-cut generation (default: -1 = automatic choice)
cuts global cut generation control, valid unless overridden by individual cut-type controls:
-1 = automatic choice (default)
0 = no cuts
1 = conservative cut generation
2 = aggressive cut generation
3 = very aggressive cut generation
degenmoves limit on the number of degenerate simplex moves -- for use when too much time is taken after solving the initial root relaxation of a MIP problem and before cut generation or root heuristics have started.
disconnected Whether to exploit independent MIP sub-models:
-1 = automatic choice (default)
0 = no
1 = use moderate effort
2 = use aggressive effort
dualreductions whether Gurobi's presolve should use dual reductions, which may be useful on a well-posed problem but can prevent distinguishing whether a problem is infeasible or unbounded:
0 = no
1 = yes (default)
feasrelaxbigm Value of "big-M" sometimes used with constraints when doing a feasibility relaxation (default: 1e6)
feasrelax Whether to modify the problem into a feasibility relaxation problem:
0 = no (default)
1 = yes, minimizing the weighted sum of violations
2 = yes, minimizing the weighted count of violations
3 = yes, minimizing the sum of squared violations
4-6 = same objective as 1-3, but also optimize the original objective, subject to the violation objective being minimized
Weights are given by suffixes .lbpen and .ubpen on variables and .rhspen on constraints (when positive), else by keywords lbpen, ubpen, and rhspen, respectively (default values = 1). Weights ≤ 0 are treated as ∞, allowing no violation.
feastol primal feasibility tolerance (default: 1e-6)
flowcover flowcover cuts: overrides "cuts"; choices as for "cuts"
flowpath flowpath cuts: overrides "cuts"; choices as for "cuts"
gomory maximum number of Gomory cut passes during cut generation
(-1 = default = no limit); overrides "cuts"
gubcover GUB cover cuts: overrides "cuts"; choices as for "cuts"
heurfrac fraction of time to spend in MIP heuristics (default: 0.05)
iisfind whether to return an IIS (via suffix .iis) when the problem is infeasible:
0 = no (default)
1 = yes
iismethod which method to use when finding an IIS (irreducible infeasible set of constraints, including variable bounds):
-1 = automatic choice (default)
0 = often faster than method 1
1 = can find a smaller IIS than method 0
implied implied cuts: overrides "cuts"; choices as for "cuts"
improvegap optimality gap below which the MIP solver switches from trying to improve the best bound to trying to find better feasible solutions (default: 0)
improvetime execution seconds after which the MIP solver switches from trying to improve the best bound to trying to find better feasible solutions (default: ∞)
impstartnodes number of MIP nodes after which the solution strategy will change from improving the best bound to finding better feasible solutions (default: 0)
infproofcuts whether to generate infeasibility proof cuts:
-1 = automatic choice (default)
0 = no
1 = moderate cut generation
2 = aggressive cut generation
intfeastol feasibility tolerance for integer variables (default: 1e-05)
intstart when there are integer variables, whether to use an initial guess (if available):
0 = no
1 = yes (default)
iterlim iteration limit (default: no limit)
lazy whether to honor suffix .lazy on linear constraints in problems with binary or integer variables:
0 = no (ignore .lazy)
1 = yes (default)

Lazy constraints are indicated with .lazy values of 1, 2, or 3 and are ignored until a solution feasible to the remaining constraints is found. What happens next depends on the values of .lazy:
1 ==> the constraint may still be ignored if another lazy constraint cuts off the current solution;
2 ==> the constraint will henceforth be enforced if it is violated by the current solution;
3 ==> the constraint will henceforth be enforced.:
lbpen See feasrelax
logfile name of file to receive log lines (default: none)
logfreq number of seconds between log lines (default: 5)
lpmethod which algorithm to use for continuous models (LP, QP) and for the root node of integer models (MIP, MIQP):
0 = primal simplex
1 = dual simplex (default)
2 = barrier
3 = concurrent
4 = deterministic concurrent
maxmipsub maximum number of nodes for RINS heuristic to explore on MIP problems (default: 500)
minrelnodes number of nodes for the Minimum Relaxation heuristic to explore at the MIP root node when a feasible solution has not been found by any other heuristic (default: 0)
mipfocus MIP solution strategy:
0 = balance finding good feasible solutions and proving optimality (default)
1 = favor finding feasible solutions
2 = favor proving optimality
3 = focus on improving the best objective bound
mipgap maximum relative MIP optimality gap (default: 1e-4)
mipgapabs absolute MIP optimality gap (default: 1e-10)
mipsep MIPsep cuts: overrides "cuts"; choices as for "cuts"
mipstart whether to use initial guesses in problems with integer variables:
0 = no
1 = yes (default)
mircuts MIR cuts: overrides "cuts"; choices as for "cuts"
modkcuts mod-k cuts: overrides "cuts"; choices as for "cuts"
multiobj whether to do multi-objective optimization:
0 = no (default)
1 = yes
When multiobj = 1 and several objectives are present, suffixes .objpriority, .objweight, .objreltol, and .objabstol on the objectives are relevant. Objectives with greater .objpriority values (integer values) have higher priority. Objectives with the same .objpriority are weighted by .objweight. Objectives with positive .objabstol or .objreltol are allowed to be degraded by lower priority objectives by amounts not exceeding the .objabstol (absolute) and .objreltol (relative) limits. The objective indicated by objno can be general; all others must be linear. Objective-specific convergence tolerances and method values may be assigned via keywords of the form obj_n_name, such as obj_1_method for the first objective.
multiobjmethod choice of optimization algorithm for lower-priority objectives:
-1 = automatic choice (default)
0 = primal simplex
1 = dual simplex
2 = ignore warm-start information; use the algorithm specified by the method keyword.
The method keyword determines the algorithm to use for the highest priority objective.
multiobjpre how to apply Gurobi's presolve when doing multi-objective optimization:
-1 = automatic choice (default)
0 = do not use Gurobi's presolve
1 = conservative presolve
2 = aggressive presolve, which may degrade lower- priority objectives.
multprice_norm choice of norm used in multiple pricing:
-1 = automatic choice (default)
0,1,2,3 = alternate norm pricing
networkcuts Network cuts: overrides "cuts"; choices as for "cuts"
nodefiledir directory where MIP tree nodes are written after memory for them exceeds nodefilestart (default: .)
nodefilestart gigabytes of memory to use for MIP tree nodes
(default = ∞ - no limit, i.e., no node files written)
nodelim maximum MIP nodes to explore (default: no limit)
nodemethod algorithm used to solve relaxed MIP node problems:
0 = primal simplex
1 = dual simplex (default)
2 = barrier
normadjust synonym for multprice_norm
numericfocus how much to try detecting and managing numerical issues:
0 = automatic choice (default)
1-3 = increasing focus on more stable computations
objno objective to optimize:
0 = none
1 = first (default, if available),
2 = second (if available), etc.
objscale how to scale the objective:
objscale=0: automatic choice (default)
-1 ≤ objscale < 0: divide by max abs. coefficient raised to this power
objscale > 0: divide by this value
opttol optimality tolerance on reduced costs (default: 1e-6)
outlev whether to write Gurobi log lines (chatter) to stdout:
0 = no (default)
1 = yes (see logfreq)
param general way to specify values of both documented and undocumented Gurobi parameters; value should be a quoted string (delimited by ' or ") containing a parameter name, a space, and the value to be assigned to the parameter. Can appear more than once. Cannot be used to query current parameter values.
paramfile name of file (surrounded by 'single' or "double" quotes if the name contains blanks) of parameter names and values for them. Lines that start with # are ignored. Otherwise, each nonempty line should contain a name and a value, separated by a space.
perturb magnitude of simplex perturbation (when needed; default: 2e-4)
pivtol Markowitz pivot tolerance (default: 7.8125e-3)
pl_bigm When some variables applear in piecewise-linear terms in the objective and AMPL's "option pl_lineraize 0" is specified, lower bounds of -pl_bigm are assumed for such variables that are not bounded below and upper bounds of +pl_bigm are assumed for such variables that are not bounded above. (Default = 1e6)
pool_mip number of independent MIP jobs (default 0) to generate and solve using the server pool (if specified by pool_servers); Gurobi automatically chooses different algorithm parameter values for each job
pool_password password for the server pool (if needed)
pool_servers comma-separated list of server names or IP addresses of machines in the server pool (default "" = none)
pool_tunejobs number of parallel tuning jobs (default 0) to run on the server (if specified by pool_servers); tuning results are not normalized by server performance, so tuning is most effective when all the servers in the server pool have similar performance characteristics
poolsearchmode synonym for ams_mode
poolsolutions synonym for ams_limit
predeprow whether Gurobi's presolve should remove linearly dependent constraint-matrix rows:
-1 = only for continuous models
0 = never
1 = for all models
predual whether gurobi's presolve should form the dual of a continuous model:
-1 = automatic choice (default)
0 = no
1 = yes
2 = form both primal and dual and use two threads to choose heuristically between them
prepasses limit on the number of Gurobi presolve passes:
-1 = automatic choice (default)
n ≥ 0: at most n passes
preqlinearize How Gurobi's presolve should treat quadratic problems:
-1 = automatic choice (default)
0 = do not modify the quadratic part(s)
1 = try to linearize quadratic parts
presolve whether to use Gurobi's presolve:
-1 = automatic choice (default)
0 = no
1 = conservative presolve
2 = aggressive presolve
presos1bigm Big-M for converting SOS1 constraints to binary form:
-1 = automatic choice (default)
0 = no conversion
Large values (e.g., 1e8) may cause numeric trouble
presos2bigm Big-M for converting SOS2 constraints to binary form:
-1 = automatic choice
0 = no conversion (default)
Large values (e.g., 1e8) may cause numeric trouble
presparsify whether Gurobi's presolve should use its sparsify reduction, which sometimes gives significant problem-size reductions:
0 = no (default)
1 = yes
pricing pricing strategy:
-1 = automatic choice (default)
0 = partial pricing
1 = steepest edge
2 = Devex
3 = quick-start steepest edge
priorities whether to use the variable.priority suffix with MIP problems. When several branching candidates are available, a variable with the highest .priority is chosen for the next branch. Priorities are nonnegative integers, with a default priority of 0:
0 = ignore .priority; assume priority 0 for all vars
1 = use .priority if present (default)
psdtol maximum diagonal perturbation to correct indefiniteness in quadratic objectives (default: 1e-6)
pumppasses number of feasibility-pump passes to do after the MIP root when no other root heuristoc found a feasible solution (default: 0)
qcpdual Whether to compute dual variables when the problem
has quadratic constraints (which can be expensive):
0 = no (default)
1 = yes
quad whether simplex should use quad-precision:
-1 = automatic choice (default)
0 = no
1 = yes
rays Whether to return suffix .unbdd if the objective is unbounded or suffix .dunbdd if the constraints are infeasible:
0 = neither
1 = just .unbdd
2 = just .dunbdd
3 = both (default)
relax whether to enforce integrality:
0 = yes (default)
1 = no: treat integer and binary variables as continuous
resultfile name of a file of extra information written after completion of optimization. The name's suffix determines what is written:
.sol - solution vector
.bas - simplex basis
.mst - integer variable solution vector
rhspen See feasrelax
rins how often to apply the RINS heuristic for MIP problems:
-1 = automatic choice (default)
0 = never
n > 0: every n-th node
scale whether to scale the problem:
0 = no
1 = yes (default)
seed random number seed, affecting perturbations that may influence the solution path (default: 0)
server Comma-separated list of Gurobi compute servers, specified either by name or by IP address. (default: unspecified - solve according to license key)
server_password Password (if needed) for specified Gurobi Compute Server(s)
server_port Network port to use for Gurobi Compute Server(s) (default: -1 = use default port)
server_priority Priority for Gurobi Compute Server(s) (default: 1; maximum 100)
server_timeout Report job as rejected by Gurobi Compute Server if the job is not started within server_timeout seconds (default: -1 = no limit)
sifting whether to use sifting within the dual simplex algorithm, which can be useful when there are many more variables than constraints:
-1 = automatic choice (default)
0 = no
1 = yes, moderate sifting
2 = yes, aggressive sifting
siftmethod algorithm to use for sifting with the dual simplex method:
-1 = automatic choice (default)
0 = primal simplex
1 = dual simplex
2 = barrier
simplex synonym for lpmethod
solnlimit maximum MIP solutions to find (default: 2e9)
solnsens whether to return suffixes for solution sensitivities, i.e., ranges of values for which the optimal basis remains optimal:
0 = no (default)
1 = yes: suffixes return on variables are
.sensobjlo = smallest objective coefficient
.sensobjhi = greatest objective coefficient
.senslblo = smallest variable lower bound
.senslbhi = greatest variable lower bound
.sensublo = smallest variable upper bound
.sensubhi = greatest variable upper bound
suffixes for constraints are
.sensrhslo = smallest right-hand side value
.sensrhshi = greatest right-hand side value
For problems with integer variables and quadratic constraints, solnsens = 0 is assumed quietly.
sos whether to honor declared suffixes .sosno and .ref describing SOS sets:
0 = no
1 = yes (default): 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 whether to tell Gurobi about SOS2 constraints for nonconvex piecewise-linear terms
1 = no
2 = yes (default), using suffixes .sos and .sosref provided by AMPL
startnodelimit limit on how many branch-and-bound nodes to explore when doing a partial MIP start:
-2 = suppress MIP start processing
-1 = use submipnodes (default)
>= 0 ==> specific node limit
submipcuts sub-MIP cuts: overrides "cuts"; choices as for "cuts"
submipnodes limit on nodes explored by MIP-based heuristics, e.g., RINS. Default = 500.
symmetry MIP symmetry detection:
-1 = automatic choice (default)
0 = none
1 = conservative
2 = agressive
threads maximum threads to use on MIP problems (default: 0 - maximum possible)
timelim limit on solve time (in seconds; default: no limit)
timing whether to report timing:
0 (default) = no
1 = report times on stdout
2 = report times on stderr
tunebase base name for results of running Gurobi's search for better parameter settings. The search is run only when tuneparbase is specified. Results are written to files with names derived from tunebase by appending ".prm" if ".prm" does not occur in tuneparbase and inserting 1, 2, ... (for the first, second, ... set of parameter settings) before the right-most ".prm". The file with "1" inserted is the best set and the solve results returned are for this set. In a subsequent "solve;", you can use paramfile=... to apply the settings in results file ... .
tuneoutput amount of tuning output when tunebase is specified:
0 = none
1 = summarize each new best parameter set
2 = summarize each set tried (default)
3 = summary plus detailed solver output for each trial
tuneresults limit on the number of tuning result files to write when tunerbase is specified. The default (-1) is to write results for all parameter sets on the efficient frontier.
tunetimelimit time limit (in seconds) on tuning when tunebase is specified. (default: -1 = automatic choice of time limit)
tunetrials number of trials for each parameter set when tunebase is specified, each with a different random seed value. (default = 2)
ubpen See feasrelax
varbranch MIP branch variable selection strategy:
-1 = automatic choice (default)
0 = pseudo reduced-cost branching
1 = pseudo shadow-price branching
2 = maximum infeasibility branching
3 = strong branching
wantsol solution report without -AMPL: sum of
1 => write .sol file
2 => print primal variable values
4 => print dual variable values
8 => do not print solution message
warmstart Whether to use incoming primal and dual variable values (if both are available) in a simplex warm start:
0 = no
1 = yes if there is no incoming basis (default)
2 = yes, ignoring the incoming basis (if any)
3 = no, but on MIP problems, use the incoming primal values as hints, ignoring the .hintpri suffix
4 = similar to 3, but use the .hintpri suffix on variables: larger (integer) values give greater priority to the initial value of the associated variable.
Note that specifying basis=0 or basis=2 causes there to be no incoming basis. This is relevant to warmstart values 1, 3, and 4. For continuous problems, warmstart values >= 2 are treated as 1.
writeprob name of problem file to be written (for debugging); must end in one of ".bas", ".lp", ".mps", ".prm", or ".sol"; can appear more than once (with different filenames).
zerohalfcuts zero-half cuts: overrides "cuts"; choices as for "cuts"
zeroobjnodes number of nodes for the zero objective heuristic to explore at the MIP root node when a feasible solution has not been found by any other heuristic (default: 0)

Try Gurobi for Free

Choose the evaluation license that fits you best, and start working with our Expert Team for technical guidance and support.

Evaluation License
Get a free, full-featured license of the Gurobi Optimizer to experience the performance, support, benchmarking and tuning services we provide as part of our product offering.
Academic License
Gurobi supports the teaching and use of optimization within academic institutions. We offer free, full-featured copies of Gurobi for use in class, and for research.
Cloud Trial

Request free trial hours, so you can see how quickly and easily a model can be solved on the cloud.

Search