Best approximation by continuous n-convex functions (Q583531)

Let C([0,1]) be the space, equipped with the uniform norm, of continuous real functions on [0,1]. Let \({\mathcal C}_ n\) (n\(\geq 2)\) be the set of continuous n-convex functions on [0,1] (a function f is said to be n- convex if its nth divided differences are non-negative). A characterization and a partial uniqueness theorem (given below) for a best approximation to a function \(\phi\in C([0,1])\setminus {\mathcal C}_ n\) from \({\mathcal C}_ n\) is obtained in this paper: Main Theorem: A function \(f_ 0\in {\mathcal C}_ n\) is a best approximation to \(\phi\in C([0,1])| {\mathcal C}_ n\) from \({\mathcal C}_ n\) if and only if there exist \(n\geq n+1\) and \(0\leq \xi_ 1<...<\xi_ m\leq 1\) such that (i) m-n-1 is even, (ii) \((f_ 0-\phi)(\xi_ i)=(-1)^{j-1-n}\| f_ 0-\phi \|\) for \(j=1,...,m,\) (iii) \(f_ 0| (\xi_ 1,\xi_ m)\) is a spline of degree n-1 with simple knots and satisfies the conditions (1) \(k(f_ 0,(\xi_ 1,\xi_ m))\leq (m-1-n),\) (2) \(k(f_ 0,(\xi_ 1,\xi_{\ell}))\leq (\ell_ 2)\) for \(\ell =2,...,m-1,\) (3) \(k(f_ 0,[\xi_ k,\xi_ m))\leq (m-k-1)\) for \(k=2,...,m-1,\) (4) \(k(f_ 0,[\xi_{k_ k},\xi_{\ell}])\leq (1-k+n-2)\) if \(k+n\) is even and \(2\leq k<\ell \leq m-1\) and \(<(\ell -k+n-2)\) if \(k+n\) is odd and \(2\leq k<\ell \leq m-1.\) If these conditions are satisfied then each best approximation to \(\phi\) from \({\mathcal C}_ n\) coincides with \(f_ 0\) on the interval \([\xi_ 1,\xi_ m]\).