What problem does it solve? Physics models often depend on many parameters, but results are typically dominated by only a few. Without sensitivity analysis, error bars are numbers without actionable meaning, and effort may be wasted improving parameters that barely affect the output. This Skill identifies which input parameters most strongly affect a target quantity and where precision-improvement effort should be directed. ## Core Features & Use Cases - Sensitivity Coefficient Computation: Calculates dimensionless sensitivity coefficients S_i = (df/dp_i)(p_i/f) via analytical (SymPy symbolic differentiation) or numerical (central finite differences) methods. - Parameter Ranking & Uncertainty Budget: Ranks parameters by contribution to output uncertainty, identifies stiff and null directions in parameter space, and flags divergent sensitivities indicating critical points or ill-conditioned formulations. - Approximation Sensitivity: Estimates systematic errors from active approximations (perturbation theory, continuum limits, truncations) as controlling parameters approach validity boundaries. - Use Case: A researcher computing a predicted energy value of 3.7 +/- 0.2 eV uses this Skill to discover that 90% of the uncertainty comes from the coupling constant, directing measurement effort where it has the greatest impact. ## Quick Start Ask the AI to run a sensitivity analysis on a target quantity, for example: analyze which parameters most affect the turbulence energy output and rank them by uncertainty contribution.
