1: \begin{abstract}

2: In many practical decision-making problems it happens that

3: functions involved in optimization process  are black-box with

4: unknown analytical representations and hard to evaluate. In this

5: paper, a global optimization problem is considered where both the

6: goal function~$f(x)$ and its gradient $f'(x)$ are black-box

7: functions. It is supposed that $f'(x)$ satisfies the Lipschitz

8: condition over the search hyperinterval with an unknown Lipschitz

9: constant~$K$. A new deterministic `Divide-the-Best' algorithm

10: based on efficient diagonal partitions and smooth auxiliary

11: functions is proposed in its basic version, its convergence

12: conditions are studied and numerical experiments executed on eight

13: hundred test functions are presented.

14: \end{abstract}