Sums of squares of linear forms (Q5898338)

Let \(A\) be a commutative ring with 1. For a quadratic form \(f=\sum_{i,j} a_{ij} x_ix_j \in A[x_1,\dots,x_n]\) with \(a_{ij}\in A\) define its length \[ \ell(f):=\min\{r: \text{\(f\) is sos (sum of squares) of \(r\) linear forms}\} \] and define \[ g_n(A)=\sup\{\ell(f): \text{\(f\) is quadratic form in \(A[x_1,\ldots,x_n]\)}\}. \] So \(p(A)=g_1(A)\) is the Pythagoras number of \(A.\) For a non-real field \(E\) its level (or `Stufe') \(s(E)\) is defined as the smallest \(n\) such that \(-1\) is a sum of \(n\) squares. If \(k\) is a real field, let \(\tau(k)=\sup\{s(E): E/k\) is a finite extension with nonreal \(E\}\). Recall that a polynomial in \(A[x_1,\dots, x_n]\) is called psd if for every homomorphism \(\varphi: A\rightarrow R\) (\(R\) real closed) the induced polynomial \(f^\varphi\) takes nonnegative values on \(R^n.\) Theorem: \(g_n(k[t]) \leq 2n \tau(k).\) Furthermore every positive semidefinite quadratic form in \(k[t][x_1,\dots, x_n]\) is a sum of squares of linear forms. Since \(\tau(\mathbb R)=1,\) one recovers for the case \(k=\mathbb R\) a result from \textit{M.-D. Choi, T.-Y. Lam} and \textit{B. Reznick}, Math. Z. 171, 1--26 (1980; Zbl 0415.10018)] (see their theorem 5.6). The constructive proof works by induction over \(n\) and seeking for a symmetric matrix \(S\) over \(k[t]\) that encodes the quadratic form a decomposition \(S=T^t T\) with \(T\) and \(m\times n\) matrix. It makes use of basic facts of quadratic form theory (Witt decomposition, Pfister forms). The authors say for their proof they reconsidered the one by Choi et.al. [loc. cit.], but the reviewer sees little similarity; the present paper uses more abstract arguments. As a consequence, for any one-dimensional \(k\)-algebra \(A\) of finite type, the Pythagoras number \(p(A)<\infty.\) Also if \(A\) is an excellent Henselian local ring of dimension two whose residue field \(k\) satisfies \(\tau(k)<\infty,\) then \(p(A)<\infty.\) The work also provides positive partial answers to a question found in the influential paper by [\textit{M.-D. Choi, Z. D. Dai, T.-Y. Lam} and \textit{B. Reznick}, J. Reine Angew. Math. 336, 45--82 (1982; Zbl 0499.12018)]; namely, whether for a finite \(A\)-Algebra \(B\), we have \(p(A)<\infty \Rightarrow p(B)<\infty\)?