|
Size: 1172
Comment: Relink
|
Size: 1783
Comment: Add comparison
|
| Deletions are marked like this. | Additions are marked like this. |
| Line 23: | Line 23: |
| Also note that, technically, this is the [[LinearAlgebra/Transposition|transpose]] of the gradient; vectors are by default columns. === Relation to Jacobian Matrices === A [[Calculus/JacobianMatricesAndDeterminants|Jacobian matrix]] is a square matrix of partial derivatives. Consider a vector-valued function, such as a [[Calculus/ParametricEquation|parameterized] function, but for simplicity let ''F(X) = <f,,1,,(X), f,,2,,(X)>'' where ''X'' is a vector as ''X = <x,,1,,``, x,,2,,``, ... x,,n,,>''. A gradient can be calculated for each of the component function. {{attachment:jacobian.svg}} |
Gradient
A gradient is a vector of partial derivatives. It describes the direction of steepest ascent for a differentiable function.
Notation
The gradient of function f is notated as ∇f. In terms of partial derivatives, the gradient of f(x1, x2, ... xn) is:
At a given point p, as long as the function f is differentiable at p, the gradient vector is:
Note the assumption; it is not negligible. For example, (xy)/(x2 + y2) is partially derivable but is itself not totally derivable at point p = [0 0]. Furthermore, it is not derivable if rotated; the basis must be orthonormal.
Also note that, technically, this is the transpose of the gradient; vectors are by default columns.
Relation to Jacobian Matrices
A Jacobian matrix is a square matrix of partial derivatives. Consider a vector-valued function, such as a [[Calculus/ParametricEquation|parameterized] function, but for simplicity let F(X) = <f1(X), f2(X)> where X is a vector as X = <x1, x2, ... xn>. A gradient can be calculated for each of the component function.
Usage
By setting a gradient to 0, critical points (local minima, local maxima, and inflections) can be calculated.
More generally, gradient descent can be used to estimate minima.