Ipopt

Initial setup

from pyoptinterface import ipopt

model = ipopt.Model()

You need to follow the instructions in Getting Started to set up the optimizer correctly.

Attention

Due to the internal API design of Ipopt, the ipopt.Model lacks some features compared to other solvers. Some of these restrictions may be lifted in the future.

• It does not support delete_variable and delete_constraint methods.

• It does not support incremental modification of linear constraint like set_normalized_rhs and set_normalized_coefficient.

• It only supports to minimize the objective function. If you want to maximize the objective function, you need to multiply the objective function by -1 manually.

The capability of ipopt.Model

Supported constraints

Constraint	Supported
model.add_linear_constraint	✅
model.add_quadratic_constraint	✅
model.add_second_order_cone_constraint	❌
model.add_sos_constraint	❌
model.add_nl_constraint	✅

Supported model attribute

Attribute	Get	Set
Name	❌	❌
ObjectiveSense	❌	❌
DualStatus	✅	❌
PrimalStatus	✅	❌
RawStatusString	✅	❌
TerminationStatus	✅	❌
BarrierIterations	❌	❌
DualObjectiveValue	❌	❌
NodeCount	❌	❌
NumberOfThreads	❌	❌
ObjectiveBound	❌	❌
ObjectiveValue	✅	❌
RelativeGap	❌	❌
Silent	❌	✅
SimplexIterations	❌	❌
SolverName	✅	❌
SolverVersion	❌	❌
SolveTimeSec	❌	❌
TimeLimitSec	❌	✅

Supported variable attribute

Attribute	Get	Set
Value	✅	❌
LowerBound	✅	✅
UpperBound	✅	✅
Domain	❌	❌
PrimalStart	✅	✅
Name	✅	✅

Supported constraint attribute

Attribute	Get	Set
Name	❌	❌
Primal	✅	❌
Dual	✅	❌

Solver-specific operations

Parameter

For solver-specific parameters, we provide set_raw_parameter methods to get and set the parameters.

model = ipopt.Model()

# set the value of the parameter
model.set_raw_parameter("tol", 1e-5)
model.set_raw_parameter("max_iter", 200)

# For HSL library
model.set_raw_parameter("hsllib", "/path/to/libhsl.so")
model.set_raw_parameter("linear_solver", "ma27")

JIT compiler used by Ipopt interface

The interface of Ipopt uses the JIT compiler to compile the nonlinear objective function, constraints and their derivatives. We have two implementations of JIT based on llvmlite and tccbox(Tiny C Compiler). The default JIT compiler is llvmlite and we advise you to use it for better performance brought by optimization capability of LLVM. If you want to use tccbox, you can specify jit="C" when creating the ipopt.Model object.

model = ipopt.Model()
# equivalent to
model = ipopt.Model(jit="LLVM")

# If you want to use tccbox
model = ipopt.Model(jit="C")