Skip to content

Beam ​

  • 2 state
  • 1 control
  • Lagrange cost
  • fixed final time
  • 1 state box
  • 4 boundary
  • OptimalControl valid
  • OptimalControl, scalarized valid
  • JuMP valid
  • ModelingToolkit valid

The clamped beam problem is a classical benchmark in constrained optimal control. It models the deflection of a flexible beam that is fixed (clamped) at both ends and subject to an external force. Originating from the Euler–Bernoulli beam model Bryson et al. 1963, it is widely used in benchmark suites. Both the state trajectory   and the control   are decision variables, and the aim is to minimise the control effort under boundary conditions and a state inequality constraint on the beam deflection.

Mathematical formulation ​

The problem can be stated as

where denotes the maximal admissible deflection (e.g.   in the current implementation).

Qualitative behaviour ​

This problem involves a second-order state constraint. Differentiating the inequality twice reveals that the control is directly related to the curvature of the state:

Along a boundary arc, where remains constant, both derivatives vanish. Consequently, the control along such an arc is zero:

In optimal control terminology, a touch point refers to the case where the state hits the bound at a single instant, while a boundary arc designates an interval where the state remains on the bound.

The qualitative behaviour of the optimal trajectory depends on the parameter :

  • Case 1:  . The inequality constraint is inactive. The optimal solution is simply the unconstrained trajectory:    .

  • Case 2:   . The trajectory touches the boundary exactly once (a touch point) at  . There is no interval of constraint activation, but the single contact reflects the influence of the state bound.

  • Case 3:  . A boundary arc appears: the state constraint is active over a finite interval. Along this interval, the control vanishes:  .

These cases illustrate how second-order state inequality constraints can yield unconstrained solutions, isolated touch points, or extended boundary arcs, thereby shaping both the trajectory and the control.

Characteristics ​

  • Linear–quadratic dynamics and cost.

  • State inequality constraint induces either unconstrained motion, touch points, or boundary arcs depending on the parameter .

  • Serves as a benchmark for optimal control methods handling state constraints.

References ​

  • Bryson, A.E., Denham, W.F., & Dreyfus, S.E. (1963). Optimal programming problems with inequality constraints I: necessary conditions for extremal solutions. AIAA Journal. doi.org/10.2514/3.2107 This seminal work develops the theoretical foundation for optimal control problems with inequality constraints, including the derivation of necessary conditions for extremal solutions. It provides the basis for analysing state-constrained control problems like the clamped beam.

  • BOCOP examples: Clamped Beam problem. Examples-BOCOP.pdf This example illustrates the practical implementation of the clamped beam problem in BOCOP, providing a benchmark for testing optimal control methods under state inequality constraints.

Parameters ​

ParameterDefault valueUnitMeaning
t00–initial time
tf1–final time
x₁_l0–lower bound of the position x₁
x₁_u0.1–upper bound of the position x₁ (the state constraint)
x₁_t00–initial position x₁
x₂_t01–initial velocity x₂
x₁_tf0–final position x₁
x₂_tf-1–final velocity x₂

Every modeling takes other values with the keyword parameters, merged with these: beam(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.

ModelingSchemeGridReference objectiveSolvesDistinct solutionsConfigurations agree
JuMP:trapeze5008.88928223281yes, within 4.7e-08
OptimalControl:trapeze5008.889281809181yes, within 6.9e-08
OptimalControl, scalarized:trapeze5008.88928179331yes, within 6.4e-09

Analytic solution. The continuous problem has a known solution, of objective 8.888888889 (Bryson, Denham & Dreyfus (1963), https://doi.org/10.2514/3.2107). The reference objective above is the one of the discretized problem: the gap between the two is the discretization error.