Model Building Functions

Model Building

m = GEKKO([server], [name]):

Creates a GEKKO model m.

c = m.Const(value, [name]):
p = m.Param([value], [name])
v = m.Var([value], [lb], [ub], [integer], [name])
m = m.MV([value], [lb], [ub], [integer], [name])
f = m.FV([value], [lb], [ub], [integer], [name])
s = m.SV([value] [lb], [ub], [integer], [name])
c = m.CV([value] [lb], [ub], [integer], [name])
i = m.Intermediate(equation, [name])
eq = m.Equation(equation)

Add a constraint equation built from GEKKO Parameters, Variables and Intermediates, and python scalars. Valid operators include python math and comparisons (+,-,*,/,**,==,<,>). Available functions are listed below in Equation Functions.

[eqs] = m.Equations(equations)

Accepts a list or array of equations.

classmethod m.Obj(obj)

The problem objective to minimize. If multiple objective are provided, they are summed.


Sets the time array indicating the discrete elements of time discretization for dynamic modes (IMODE > 3). Accepts a python list of a numpy array.

classmethod dt()

Ordinary differential equations are specified by differentiation a variable with the dt() method. For example, velocity v is the derivative of position x:

m.Equation( v == x.dt() )

Discretization is determined by the model time attribute. For example, m.time = [0,1,2,3] will discretize all equations and variable at the 4 points specified. Only ordinary differential equations discretized by time are available internally. Other discretization must be performed manually.

a = m.Array(type,dimension,**args)
Create an n-dimensional array (as defined in tuple input dimension ) of GEKKO variables of type type .

The optional keyword arguments (**args) are applied to each element of the array.

classmethod m.solve(remote=True, disp=True, debug=False)

Solve the optimization problem.

This function has these substeps:

  • Validates the model and write .apm file
  • Validate and write .csv file
  • Write options to overrides.dbs
  • Solve the problem using the apm.exe commandline interface.
  • Load results into python variables.

If remote is True, the problem is sent to self.server to be solved. If False, GEKKO looks for local binaries of APMonitor.

If disp is True, APM and solve output are printed.

If debug is True, variable names are checked for problems, tuning parameters are checked for common errors, and user-defined input options are compared against options used by APM. This is useful in debugging strange results.

classmethod m.Connection(var1, var2, pos1=None, pos2=None, node1='end', node2='end')

var1 must be a GEKKO variable, but var2 can be a static value. If pos1 or pos2 is not None, the associated var must be a GEKKO variable and the position is the (0-indexed) time-discretized index of the variable.

Connections are processed after the parameters and variables are parsed, but before the initialization of the values. Connections are the merging of two variables or connecting specific nodes of a discretized variable. Once the variable is connected to another, the variable is only listed as an alias. Any other references to the connected value are referred to the principal variable (var1). The alias variable (var2) can be referenced in other parts of the model, but will not appear in the solution files.

classmethod m.fix(var, pos, val)

This function facilitates the Connection function when var2 is a static value (val).


A list of strings to pass to the solver; one string for each option name and value. For example:

m = GEKKO()
m.solver_options = ['max_iter 100','max_cpu_time 100']

Equation Functions

Special function besides algebraic operators are available through GEKKO functions. These must be used (not numpy or other equivalent functions):

classmethod m.sin(other)
classmethod m.cos(other)
classmethod m.tan(other)
classmethod m.asin(other)
classmethod m.acos(other)
classmethod m.atan(other)
classmethod m.sinh(other)
classmethod m.mcosh(other)
classmethod m.tanh(other)
classmethod m.exp(other)
classmethod m.log(other)
classmethod m.log10(other)
classmethod m.sqrt(other)

Pre-built Objects

x,y,u = state_space(A,B,C,D=None,discrete=False,dense=False)

For State Space models, input SS matricies A,B,C, and optionally D. Returns a GEKKO array of states (SV) x, array of outputs (CV) y and array of inputs (MV) u. A,B,C and D must be 2-dimensional matricies of the appropriate size.

The discrete Boolean parameter indicates a discrete-time model, which requires constant time steps and 2 NODES. The dense Boolean parameter indicates if A,B,C,D should be written as dense or sparse matrices. Sparse matricies will be faster unless it is known that the matricies are very dense.

classmethod m.cspline(x, y, x_data, y_data, bound_x=False)
Generate a 1d cubic spline with continuous first and seconds derivatives

from arrays of x and y data which link to GEKKO variables x and y with a constraint that y=f(x).

This function is currently only available through remote solves to the default server.


x: GEKKO variable

y: GEKKO variable

x_data: array of x data

y_data: array of y data that matches x_data

bound_x: boolean to state that x should be bounded at the upper and lower bounds of x_data to avoid

extrapolation error of the cspline.

classmethod periodic(v)

Makes the variable argument periodic by adding an equation to constrains v[end] = v[0]. This does not affect the default behavior of fixing initial conditions (v[0]).

Internal Methods

These are the major methods used internal by GEKKO. They are not intended for external use, but may prove useful for highly customized applications.

static build_model(self)

Write the .apm model file for the executable to read. The .apm file contains all constants, parameters, variables, intermediates, equations and objectives. Single values and/or initializations, along with variable bounds, are passed throught the .apm model file.

static write_csv()

Any array values are passed through the csv, including variable initializations. If imode > 3 then time must be discretized in the csv.

static generate_overrides_dbs_file()

All global and local variable options are listed in the overrides database file.

static load_json()

Reads back global and variable options from the options.json file. Stores output and input/output options to the associated variable.

static load_results()

The executable returns variable value results in a json. This function reads the json and loads the results back into local python variables.

static verify_input_options()

Called when optional solve argument verify_input=True. Compares input options of the model and variables from GEKKO to those reported by APM to find discrepencies.

Internal Attributes

These are GEKKO model attributes used internally. They are not intended for external use, but may prove useful in advanced applications.


String representation of the server url where the model is solved. The default is ‘’. This is set by the optional argument server when intializing a model.


A python list of pointers to GEKKO Constants attributed to the model.


A python list of pointers to GEKKO Parameters, FVs and MVs attributed to the model.


A python list of pointers to GEKKO Variables, SVs and CVs attributed to the model.


A python list of pointers to GEKKO Intermediate variables attributed to the model.


A python list of the explicit intermediate equations. The order of this list must match the order of intermediates in the intermediates attribute.


A python list of equations


A python list of objective.


A python list of connections


Set to ‘generated’ if any time, parameter or variable data is communicated through the csv file. Otherwise set to ‘none’.


The name of the model as a string. Used in local temporary file name and application name for remote solves. This is set by the optional argument name when intializing a model. Default names include the model id attribute to maintain unique names.


The absolute path of the temporary file used to store all input/output files for the APMonitor executable.