Sensitivity Analysis in Code Workflows
Sensitivity analysis ranks which inputs most influence an output, focusing effort on the parameters that actually control a prediction.
Which inputs matter
A workflow may have dozens of uncertain inputs, but usually only a few drive the answer. Sensitivity analysis quantifies how much each input affects a chosen output, ranking them so that measurement effort, model refinement, and uncertainty reduction target the parameters that count instead of being spread thinly.
It complements uncertainty propagation: propagation gives the size of the output uncertainty, sensitivity attributes that uncertainty to its sources.
Local vs global
Local sensitivity computes partial derivatives of the output with respect to each input at a nominal point, cheap but valid only nearby. Global sensitivity explores the whole input space and apportions the output variance among inputs and their interactions, using variance-based measures such as Sobol indices. Global methods reveal interactions that local derivatives miss.
Sobol indices
A first-order Sobol index gives the fraction of output variance explained by one input alone; a total-order index adds that input's contribution through interactions. If total exceeds first-order substantially, interactions matter and inputs cannot be treated independently. Estimating these indices typically uses structured Monte Carlo sampling.
Design relevance
For the Hyperion breeder, sensitivity analysis identifies which uncertain inputs, for example the turbulent transport model or the pedestal boundary condition, most control the predicted fusion power, so that verification and future measurement focus there. This sharpens the honest, simulation-based case ahead of construction.
- Ranks inputs by influence on an output
- Attributes output variance to its sources
- Local derivatives vs global variance-based measures
- Sobol indices capture interactions