Library / Advanced Mathematics
The gradient of a scalar-valued function is the vector of its first partial derivatives. It is the object that packages local rate-of-change information into a direction-sensitive form.
If f(x_1, ..., x_n) is a scalar-valued function, its gradient is the vector
grad f whose components are the partial derivatives of f with respect to
each input variable.
This turns local derivative information into a geometric object that can be compared with directions, constraints, and local approximations.
The gradient points in the direction where the function increases fastest locally. Its magnitude reflects how steep that increase is. That is why gradients appear everywhere in optimization, machine learning, and constrained analysis.
Geometrically, the gradient is normal to level curves or level surfaces of the function. This makes it a natural object for reasoning about constrained optimization and local geometry.
Symbolic computation can derive gradient expressions exactly, simplify them, and expose repeated substructure before the result is evaluated numerically.
The gradient is closely related to the Jacobian and Hessian. For a scalar-valued function, the Jacobian can be viewed as the row-vector version of first derivatives, while the Hessian organizes second derivatives and local curvature.
This is one reason the gradient belongs naturally beside the existing Jacobian and Hessian pages. It completes the basic local-geometry chain for multivariable analysis.
Gradient descent, local sensitivity, constrained optimization, and many symbolic derivation tasks all depend on having a clean first-order description of how a function changes. The gradient is the standard object for that job.