11.12 Bilevel Sensitivities and Inner-Loop Convergence

Core idea¶

Bilevel Sensitivities and Inner-Loop Convergence must be treated as a system-level decision rather than an isolated technique. For nested and simultaneous suspension formulations using identical meshes, derivatives, tolerances, and starts, state what is fixed, what is optimized, what information is available, and what equations define feasibility.

The relevant quantities are outer $p$ , trajectory $w=[X,U]$ , and sparse residuals $R(p,w)$ . The chapter-level formulation is

\frac{dJ^*}{dp}=\frac{\partial J}{\partial p}+\frac{\partial J}{\partial w}\frac{dw^*}{dp}.

For this section, trace how the choice changes inner control solve, the active constraints, and the implementable engineering design. A method is useful only when its assumptions are explicit and its result answers the same system question as the baseline.

Engineering interpretation¶

Ask three questions:

Which physical, informational, computational, or economic resource changed?
Which objective component or active constraint made the change valuable?
Does the conclusion survive model, disturbance, initialization, uncertainty, and implementation checks?

A practical action is to verify derivatives. Record units and assumptions before optimization, report component objectives and margins afterward, and verify the result using an independent calculation or higher-fidelity model.

Activity 11.12: quantify bilevel sensitivities and inner-loop convergence¶

Activity 11.12: quantify bilevel sensitivities and inner-loop convergence

For a nonlinear CCD solver benchmark, compare two complete alternatives for Bilevel Sensitivities and Inner-Loop Convergence.

Alternative	Objective	Minimum margin	CPU	Validation error
Baseline	1.120	12%	54 s	2.0%
Candidate	0.924	5%	108 s	3.2%

Compute objective improvement and CPU increase.
Apply a minimum 5% feasibility margin.
Inflate validation error by 50% and reassess.
Report improvement that survives validation.
Identify the decision variable responsible for the trade.

Chapter summary¶

The chapter connected outer plant loop, inner control solve, sparse simultaneous NLP, derivatives and scaling, verified comparison through one system formulation. Engineering conclusions require aligned models, information, numerical accuracy, and validation.

Common mistakes¶

changing assumptions while comparing alternatives;
reporting objective improvement without verified feasibility;
hiding information, architecture, or uncertainty;
treating solver convergence as validation; and
reporting runtime without accuracy, derivatives, and tolerances.

Exercises¶

Recreate the workflow for nested and simultaneous suspension formulations using identical meshes, derivatives, tolerances, and starts.
State every variable, unit, dependency, and constraint.
Construct a common sequential or nominal baseline.
Identify active constraints and the physical bottleneck.
Design a test that could falsify the claimed benefit.

Principal sources¶

Herber and Allison; Sundarrajan and Herber; Allison, Guo, and Han.

Open research question¶

How can inner convergence and bilevel derivative accuracy be controlled adaptively?