descent direction
To show that dk is a descent direction for f (x) at xk , it suffices to show that f (xk )dk < 0:
2x1
f (x) = x31 + x1 x2 +
x2
" #
2
3x21 + x2 + x2
∇f (x)0 =
x1 − 2x
x22
1
1
xk =
1
f (xk ) = 4
6
∇f (xk )0 =
−1
0
dk =
−1
k k
0
∇f (x )d = 6 −1 ∗ =1>0 (1)
−1
As can be seen from equation (1), dk is not a descending direction for f (x) at
xk. To find one, we
0
can take any dk such that the inequality ∇f (xk )dk < 0 holds. If we take dk = for instance,
1
0
∇f (xk )dk = 6 −1 ∗
= −1 < 0
1
Quadratic interpolation
Let’s find the coefficients a, b and c that define the quadratic function q(α) = aα2 + bα + c such
that:
q(0) = f (xk )
q0(0) = ∇f (xk )dk
q(αmax ) = f (xk + αmax dk )
α=0:
q(0) = c = f (xk ) = 4 ; c = 4
q(α)0 = 2aα + b ; q(0)0 = b = ∇f (xk )dk = −1 ; b = −1
1
α=1:
q(1) = a + b + c = f (xk + dk ); a − 1 + 4 = 4; a = 1
Acceptability test
Given the parameters c1 = 0.01 and c2 = 0.5 the acceptability test over αq : is performed by veri-
fying the A-G conditions as follows:
1
The trial steplength αq satisfies the acceptability condition and therefore αk = αq = 2