Approximation by sums of piecewise linear polynomials (Q2252933)

Let \(f\) be an \(L_p\) function defined on \(U\), the \(d\)-dimensional unit cube (\(d>1\)). Let \(P\) be a partition of \(U\) into \(N\) convex sub-regions. A function is called piecewise linear with respect to \(P\) if its restriction to each sub-region, \(v\), in \(P\) is linear. Let \(q\) be a best piecewise linear approximation to \(f\) with respect to the \(L_p\) norm. There is general interest in determining the order of the resulting error, \(\|f-q\|_p\), as \(N\) increases. The authors' setting is more intricate. Let \(\{P_1, P_2, \dots , P_n\}\) be a collection of such convex partitions of \(U\). Assume each \(P_k\) comprises \(N_k\) sub-regions. Let \(N\) be the sum of the \(N_k\). Let \(S = \{g: g = \sum_k g_k,\) where \( g_k\) is piecewise linear on \(P_k \}\). This paper focuses on approximating \(f\) by members of \(S\). This work describes two algorithms for constructing partitions \(\{P_k\}\) so that the resulting error function, \(\|f-q\|_p\), approaches zero rapidly as \(N\) increases. That is, the error is of the order \([ N^{{6}\over {2d+1} }] ^{-1} \). The authors remark that this ``makes a marked improvement over the \([ N^{{2}\over {d }}] ^{-1} \) order achievable on a single convex partition''. Putting \(P_2 = P_3 = \dots = P_n = \emptyset\), the setting of this paper includes approximation by the piecewise linear functions with respect to a single convex partition of \(U\). So it is not surprising that the order of approximation in the setting of this work is better than the order related to a single partition. The achievement of this work calculates a specific order and algorithms. The construction of the first algorithm depends on properties of \(f\), the function being approximated; the second is independent of \(f\) but employs more partitions. Not all of the results are limited exclusively to piecewise linear functions. Some preliminary results are proven for piecewise polynomials and one conclusion provides estimates for piecewise constant functions. Reviewers Comment: Let \(P^*\) be the convex partition of \(U\) that is the common refinement of the partitions \(\{P_1, P_2, \dots , P_n\}\). Then the approximating space, \(S\), studied here is a subspace of the piecewise linear functions on \(P^*\). The dimension of \(S\) is the same as that of the piecewise linear functions on a single partition comprising \(N\) sub-regions (i.e., the dimension is \(dN + N \)). Where the number of sub-regions of \(P^*\) can be the product of the \(N_k\).