1: \begin{abstract}

2: We study the complexity of 

3: high-dimensional approximation 

4: in the $L_2$-norm when different classes of information are available;

5: we compare the power of function evaluations with the power of arbtirary continuous linear measurements.

6: Here, we discuss the situation when the number of 

7: linear measurements

8: required to achieve an error $\varepsilon \in (0,1)$ in dimension $d\in\IN$

9: depends only 

10: poly-logarithmically on $\eps^{-1}$. 

11: This corresponds to an exponential order of convergence of the approximation error, 

12: which often happens

13: in applications. 

14: However, it does not mean that the high-dimensional approximation problem is easy,

15: the main difficulty usually

16: lies within

17: the dependence on the dimension $d$.

18: We determine to which extent the required

19: amount of information changes if 

20: we allow only function evaluation instead of arbitrary linear information.

21: It turns out that in this case we only lose very little, and we can even 

22: restrict to linear algorithms. 

23: In particular, several notions of tractability hold simultaneously for both types 

24: of available information. 

25: 

26: \smallskip

27: \noindent \textbf{Keywords.} Approximation, Multivariate problems, Tractability, Complexity

28: \end{abstract}

29: