 Derivative of a constrained value function

Forsuppose the optimizer and multiplier vary smoothly and use $L=f-\lambda (g-b)$. Differentiating the optimum value and using stationarity gives