Jackson
- 3 state
- 1 control
- Mayer cost
- fixed final time
- 3 state box
- 1 control box
- 3 boundary
- OptimalControl valid
- OptimalControl, scalarized valid
- JuMP valid
- ModelingToolkit valid
The Jackson problem is a classical benchmark in optimal control. It consists of controlling a three-dimensional system in which the first two states interact linearly under the effect of a single control input, while the third state accumulates based on the complementary control. The objective is to minimise the third state at the final time, while satisfying bounds on states and control, as well as initial and terminal conditions. The problem exhibits singular arcs, making it a useful benchmark for testing direct transcription and nonlinear programming methods.
Mathematical formulation
The problem can be stated as
where
Qualitative behaviour
The first two states interact linearly under the effect of the control, while the third state accumulates according to
.The problem exhibits singular arcs, where the optimal control is neither at its lower nor upper bound but satisfies Hamiltonian conditions.
Optimal trajectories typically include bang–bang segments interleaved with singular arcs.
State and control constraints are respected at all times, and the final cost depends only on
.
Characteristics
Linear–nonlinear dynamics.
Terminal cost depends on a single state.
Serves as a benchmark for testing solvers handling singular arcs and constrained optimal control.
References
Jackson, E. A. (1968). The existence of singular extremals. Journal of Optimization Theory and Applications. Discusses the theoretical existence of singular extremals, forming the basis of the Jackson benchmark problem.
Biegler, L. T. (2010). Nonlinear Programming: Concepts, Algorithms, and Applications to Chemical Processes. SIAM. Provides background on nonlinear programming methods applicable to problems like Jackson.
BOCOP Repository: Jackson Example. https://github.com/control-toolbox/bocop/tree/main/bocop Contains the Jackson problem implementation for testing direct transcription and NLP solvers.
Parameters
| Parameter | Default value | Unit | Meaning |
|---|---|---|---|
t0 | 0 | – | initial time |
tf | 4 | – | final time |
k1 | 1 | – | rate of the reaction from a to b |
k2 | 10 | – | rate of the reverse reaction from b to a |
k3 | 1 | – | rate of the reaction from b to x₃ |
a_l | 0 | – | lower bound of the concentration a |
a_u | 1.1 | – | upper bound of the concentration a |
b_l | 0 | – | lower bound of the concentration b |
b_u | 1.1 | – | upper bound of the concentration b |
x₃_l | 0 | – | lower bound of x₃ |
x₃_u | 1.1 | – | upper bound of x₃ |
u_l | 0 | – | lower bound of the control u, the fraction of the first catalyst |
u_u | 1 | – | upper bound of the control u |
a_t0 | 1 | – | initial concentration a |
b_t0 | 0 | – | initial concentration b |
x₃_t0 | 0 | – | initial value of x₃ |
Every modeling takes other values with the keyword parameters, merged with these: jackson(backend; parameters=(t0=…,)).
Reference objective
The best objective among the solves that converged, in each modeling, on the default grid of the modeling: the modelings are independent, each has its own reference. See how the references are computed.
| Modeling | Scheme | Grid | Reference objective | Solves | Distinct solutions | Configurations agree |
|---|---|---|---|---|---|---|
| JuMP | :trapeze | 500 | -0.191821767 | 8 | 1 | yes, within 1.5e-06 |
| OptimalControl | :trapeze | 500 | -0.1918219474 | 18 | 1 | yes, within 1.5e-06 |
| OptimalControl, scalarized | :trapeze | 500 | -0.1918216613 | 3 | 1 | yes, within 1.1e-13 |