Convex hulls of curves of genus one (Q645211): Difference between revisions

Let \(C\) be a real irreducible nonsingular affine algebraic curve of geometric genus one, embedded in affine \(n\)-space. Assume the curve's set of real points \(C(\mathbb{R})\) is compact and let \(\mathbb{R}[C] =\mathbb{R}[x_1,\dots,x_n]/\mathfrak{I}(C),\) be its coordinate ring. The author showed in [Trans. Am. Math. Soc. 352, No. 3, 1039--1069 (2000; Zbl 0941.14024)] and [Math. Z. 245, No. 4, 725--760 (2003; Zbl 1056.14078)] that if \(f\) is nonnegative (=positive semidefinite(psd))(on \(C(\mathbb{R})\)), then \(f\) is a sum of squares: \(f=\sum_i f_i^2\) for some \(f_i\in \mathbb{R}[C].\) In this paper the author shows that the preordering \(\Sigma \mathbb{R}[C]^2\) is stable (see below) and the degrees of the \(f_i\) can all be bounded in terms of \(\deg(f).\) As a consequence \textit{J. B. Lasserre}'s relaxation method [Math. Program. 120, No. 2 (A), 457--477 (2009; Zbl 1198.14055)] applies to find an explicit lifted linear matrix inequality (LMI) representation for the convex hull of \(C(\mathbb{R});\) i.o.w. \(\text{conv}\, C(\mathbb{R})\) is a projected spectrahedron. This is the first time such a representation is given for a nonrational variety. In section 1, the context from general real algebra is provided. Let \(A\) be a finitely generated real algebra \(A=\mathbb{R}[{\mathtt x}]/I,\) \({\mathtt x}=(x_1,\dots,x_n),\) say. Via \(\text{Hom}(A,\mathbb{R})\) one defines the cone \(A_+\) of positive semidefinite elements; it has the expected property that the sums of squares in \(A\) lies in \(A_+,\) i.e. \(\Sigma A^2\subseteq A_+.\) Preordering \(\Sigma A^2\) is called saturated if \(A_+=\Sigma A^2.\) If for some finite dimensional subspace \(L\) of \(A\) we have \(L\cap A_+ =L\cap \Sigma A^2,\) we have partial saturation; if there exists a finite dimensional \(W\) (to \(L\)) so that \(L\cap\Sigma A^2 \subset M_W=\{\sum a_i^2: a_i\in W\}\) then the preordering \(\Sigma A^2\) has a partial stability property -- by definition it is fully stable if this holds for all such \(L.\) Partial saturation and stability are prerequisites for applying Lasserre's relaxation method. In particular, if the real variety \(V_{\mathbb{R}}(I) \) is compact, and \(L=A_1=\{f+I:\deg f\leq 1\}\) (for example) a space satisfying with appropriate \(W\)s above conditions, then the relaxation method guarantees a representation of \(\text{conv}\,V_{\mathbb{R}}(I)\) as the intersection of a shrinking family of projected spectrahedra \(K_W\) obtained by making \(W\) larger and larger. That is there exist symmetric real matrices \(A_i\) \((0\leq i\leq n)\) and \(B_j\) \((1\leq j \leq m)\) so that \(\text{conv}\,V_{\mathbb{R}}(I)=\{x\in \mathbb{R}^n: \exists y\in \mathbb{R}^m \text{ so that } A_0+\sum_i x_iA_i+\sum_j y_jB_j\succeq 0\}.\) The theory is also connected to results on theta bodies by \textit{J. Gouveia, P. A. Parillo} and \textit{R. R. Thomas} [Siam J. Optim. 20, No. 4, 2097--2118 (2010; Zbl 1213.90190)] Sections 2 to 4 establish information that allows to pass from this general result to explicit representations in the case of curves as above (bounds for \(\dim W\) for example). In view of Scheiderer's earlier results [loc. cit.] on saturation, section 2 concentrates on showing stability of preordering \(\Sigma\mathbb{R}[C]^2\) in case that \(C\) has at least one pair of conjugate nonreal points: by the Riemann Roch theorem it is sufficient to show stability for plane affine curves of equation \(y^2+q(x)=0\) with \(q\in \mathbb{R}[x],\) monic of degree 4, indefinite and without multiple roots. \(q\) depends on \(C.\) Defining in the function field \(\mathbb{R}(C),\) \(v_p:\mathbb{R}(C)^*\rightarrow \mathbb{Z}\) as the discrete valuation associated to \(p\in C(\mathbb{C}),\) let \(\delta(f):=-v_{\infty}(f)\) and if \(f\in \mathbb{R}[C],\) let \(\theta(f)\) be the least integer \(d\) for which there exists a sum of square representation \(f=f_1^2+\cdots+f_r^2\) with all \(\delta(f_i)\leq d.\) Theorem 2.13 shows that there is a nonnegative integer \(N\) (explicitly obtainable from \(q\) and hence from \(C\)) so that \(\theta(f)\leq N+ \lceil \frac{\delta(f)}{2} \rceil \) for every psd regular \(f\in \mathbb{R}[C].\) This is easily seen to imply stability of \(\Sigma R[C]^2.\) An explicit sos representation of a linear polynomial that is nonnegative on the set of real points of a plane affine curve is provided. By this theory, in Section 3 the author is able to indicate \(\text{conv} C(\mathbb{R})\) the form of a projected spectrahedron. For explicit examples he borrows results from section 4 where a detailed study of bounds for the beforementioned \(N=N_C\) is made. There it is also shown that for a special class of curves depending on parameters, \(C=C(a,b),\) \(N_C\) can tend to infinity if the curve degenerates to a singular curve. As usual, Scheiderer treats his subject very thoroughly. After a search it seems to the reviewer that \textit{I. R. Shafarevich}'s [Basic algebraic geometry. Berlin-Heidelberg-New York: Springer-Verlag (1974; Zbl 0284.14001)] and \textit{K. Kendig}'s [Elementary algebraic geometry. New York - Heidelberg - Berlin: Springer-Verlag (1977; Zbl 0364.14001), Chapter 5], cover much of some (partially deep) background knowledge on algebraic curves that is necessary for detailed understanding, including normalization and proof of Riemann-Roch.