Calculating Value Function Gradients Across Capital Trajectories
Value function gradients across capital trajectories compute the marginal sensitivity of terminal discounted returns to resource shifts at each state coordinate.
generalized subgradient
Meaning
Mathematical generalizations of the classical derivative to non-smooth, continuous functions represent the set of all directional slopes at a non-differentiable point. A generalized subgradient allows optimization algorithms to navigate sharp bends, steps, and corners where standard gradients are undefined. In non-convex and non-smooth optimization, this construct provides the directional information needed to find local minima.
Nondifferentiable Slope
Convex optimization often encounters objective functions with discontinuous derivatives, such as those arising in dual decomposition or robust cost modeling. The generalized subgradient replaces the single gradient vector with a closed, convex set of vectors known as the subdifferential. Each element in this set acts as a linear underestimator of the function at that specific point.
Set-valued derivatives allow the optimizer to handle piecewise linear functions without smoothing.
Descent Direction
Selecting a search path when the standard gradient does not exist requires evaluating multiple potential directions of descent. The generalized subgradient provides these vectors, and choosing any vector from the subdifferential ensures a step that does not increase the objective value. Because non-smooth paths can cause standard algorithms to stall, using these generalized vectors guarantees convergence in subgradient methods.
Asymmetric Boundary
Calculating these derivatives becomes more complex when the function is non-convex. The generalized subgradient in such cases is defined using Clarke subdifferentials, which extend the concept to locally Lipschitz functions. Such a broader definition ensures that even in highly volatile cost landscapes, optimization remains mathematically tractable.