The two tables below correspond to the performance of the method
described in
[1] "A Newton-like method with mixed factorizations and cubic
regularization for unconstrained minimization" by E. G. Birgin and
J. M. Martínez (submitted).
when applied to the problems considered in
[2] "Adaptive cubic regularisation methods for unconstrained
optimization. Part I: motivation, convergence and numerical results"
by C. Cartis, N. I. M. Gould, and Ph. L. Toint (Math. Program., Ser. A
(2011) 127:245–295).
Only problem STREG is missing, since we were unable to find it.
In the first table, the stopping criteria are the one described in
[1]; while in the second table the stopping criteria described in [2]
were considered, i.e. |g(xk)|f(xk) with i=argmax{|g(xk)_i|} and y = xk - sign(1,g(xk)_i) macheps^1/3 max(1,|xk_i|) e_i.
STOP = 8: xk and xk+1 are such that f(xk+1)=f(xk) during 10 consecutive iterations.
STOP = 9: xk is such that |x|>xlarge, with xlarge=1.0d+10.
STOP = 10: xk+s is such that xk+s did not give sufficient decrease and |s|