Descent direction

From HandWiki

In optimization, a descent direction is a vector 𝐩n that points towards a local minimum 𝐱* of an objective function f:n. Computing 𝐱* by an iterative method, such as line search defines a descent direction 𝐩kn at the kth iterate to be any 𝐩k such that 𝐩k,f(𝐱k)<0, where , denotes the inner product. The motivation for such an approach is that small steps along 𝐩k guarantee that f is reduced, by Taylor's theorem.

Using this definition, the negative of a non-zero gradient is always a descent direction, as f(𝐱k),f(𝐱k)=f(𝐱k),f(𝐱k)<0.

Numerous methods exist to compute descent directions, all with differing merits, such as gradient descent or the conjugate gradient method.

More generally, if P is a positive definite matrix, then pk=Pf(xk) is a descent direction at xk.[1] This generality is used in preconditioned gradient descent methods.

See also

References

  1. J. M. Ortega and W. C. Rheinbold (1970). Iterative Solution of Nonlinear Equations in Several Variables. pp. 243. doi:10.1137/1.9780898719468.