multivariate production systems optimization - Stanford University
multivariate production systems optimization - Stanford University
multivariate production systems optimization - Stanford University
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
F(x + hj ej - hi ei)<br />
F(x - hi ei)<br />
F(x - hj ej - hi ei)<br />
gi = F x+hiei - F x-hiei<br />
2hi<br />
The diagonal elements of the Hessian are given by<br />
Hii = F x+hiei - 2F x + Fx-hiei<br />
h i 2<br />
and the off-diagonal elements of the Hessian are given by<br />
Hij = F x+hiei+hjej + F x-hiei-hjej - F x-hiei+hjej - F x+hiei-hjej<br />
4hihj<br />
hj<br />
F(x + hj ej) F(x + hj ej + hi ei)<br />
F(x)<br />
F(x - hj ej) F(x - hj ej + hi ei)<br />
hi<br />
67<br />
F(x + hi ei)<br />
Figure 6.2: Discretization Scheme Used for Finite Difference Approximations<br />
(6.20)<br />
(6.21)<br />
(6.22)<br />
All of the finite difference approximations are second-order accurate which preserves the<br />
second-order accuracy of the quadratic model. If first-order accuracy is acceptable in the<br />
mixed partial derivatives, significant savings can be realized by evaluating the off-diagonal<br />
elements of the Hessian as<br />
Hij = F x+hiei+hjej + F x - Fx+hjej - F x+hiei<br />
hihj<br />
Notice that after determining the elements of the gradient with Equation 6.20 and the<br />
diagonal elements of the Hessian with Equation 6.21, each off-diagonal element can be<br />
obtained with Equation 6.23 by a single additional function evaluation as opposed to the<br />
four additional function evaluations that would be required by Equation 6.22.<br />
(6.23)