Conduction Intensity and Structural Fragility of Heston Volatility Surfaces

We develop an operator-theoretic measure of structural fragility for implied volatility surfaces under the Heston stochastic volatility model. The \emph{conduction intensity} $\tau$, defined as the operator norm of the Fr\'{e}chet derivative of the implied volatility mapping, captures worst-case parameter sensitivity of the entire surface. We prove Fr\'{e}chet differentiability via a three-link chain (characteristic function analyticity, Fourier integral regularity, Black--Scholes inversion) with explicit bounds on the Riccati function derivatives, establish a first-order duality with local model uncertainty, and provide five main results. First, $\tau$ is functionally independent from the industry-standard triple (ATM vol, skew, butterfly): linear regression yields $R^2=0.129$, meaning 87\% of $\tau$-information is invisible to standard monitors. Second, the Heston conduction matrix has effective rank two: the $(v_0,\bar{v})$ subspace explains $>$99.7\% of total conduction variance, a consequence of ATM dominance quantified by the wings-to-ATM Vega ratio bound $\eta_w/\eta_{\mathrm{ATM}}<0.03$. Third, a $\kappa$ direction-crossing theorem: the most dangerous perturbation direction rotates continuously from $e_{v_0}$ to $e_{\bar{v}}$ as $\kappa$ increases, with critical $\kappa^*$ satisfying $\int_\cS(2w(\kappa^*,T)-1)\,dT=0$, numerically $\kappa^*\approx 3.25$. Fourth, a high-$\kappa$ fragility paradox: strong mean-reversion renders the model extremely sensitive to $\bar{v}$, with $\tau^{(\bar{v})}\sim\kappa^{0.52}$. Fifth, the Feller boundary $2\kappa\bar{v}=\sigma_v^2$ produces Lipschitz-continuous analytic robustness, contrasting with the hard transitions found in CEV and CDO models. Stylized analyses of the 2018 Volmageddon and 2020 March crisis demonstrate that raw $\tau$ misleadingly decreases during crises (Vega-inflation bias). The $v_0$-normalized $\tilde\tau=\tau\sqrt{v_0}$ corrects this, while a directional anomaly indicator amplifies 3.69-fold (2018) and 4.2-fold (2020) before the crash.