Types of linkage of quadratic Pfister forms (Q1729764)

Let \(F\) be a field of characteristic \(p>0\). In this paper, the authors continue their study of linkage properties of symbols in the Kato-Milne cohomology groups \(H^n_p(F)\) and \(\nu_F(n)\). The logarithmic differential forms \(\frac{da_1}{a_1}\wedge\cdots\wedge \frac{da_{n}}{a_{n}}\), \(a_i\in F^*\), generate \(\nu_F(n)\) additively and are called symbols in \(\nu_F(n)\). A symbol in \(H^n_p(F)\) is the class of a logarithmic differential form \(a_1\frac{da_2}{a_2}\wedge\cdots\wedge \frac{da_{n}}{a_{n}}\), \(a_i\in F^*\). We call two symbols \(\omega_1\), \(\omega_2\in H^n_p(F)\) separably (resp. inseparably) \(m\)-linked (\(1\leq m< {n}\)) if there exist symbols \(\theta\in H^{m}_p(F)\) and \(\eta_1,\eta_2\in \nu_F(n-m)\) (resp. symbols \(\theta_1,\theta_2\in H^{n-m}_p(F)\) and \(\eta\in \nu_F(m)\)) such that for \(i=1,2\) one has \(\omega_i=\theta\wedge\eta_i\in H^n_P(F)\) (resp. \(\omega_i=\theta_i\wedge\eta\)) for \(i=1,2\). Furthermore, we call \(\omega_1\) and \(\omega_2\) totally separably (resp. totally inseparably) \(m\)-linked if for any symbol \(\theta\in H^{m}_p(F)\) (resp. \(\eta\in \nu_F(m)\)), there exists \(\eta_1\in \nu_F(n-m)\) (resp. \(\theta_1\in H^{n-m}_p(F)\)) with \(\omega_1=\theta\wedge \eta_1\) (resp. \(\omega_1=\theta_1\wedge\eta\)) if and only if there exists \(\eta_2\in \nu_F(n-m)\) (resp. \(\theta_2\in H^{n-m}_p(F)\)) with \(\omega_2=\theta\wedge \eta_2\) (resp. \(\omega_2=\theta_2\wedge\eta\)). The authors show the following. If two separably \((n-1)\)-linked symbols \(\omega\wedge \frac{da_{i}}{a_{i}}\in H^n_p(F)\) (\(i=1,2\), \(a_i\in F^*\), \(\omega\in H^{n-1}_p(F)\)) are totally separably \(1\)-linked, then \(\omega\wedge \frac{da_{1}}{a_{1}} \wedge \frac{da_{2}}{a_{2}}=0\in H^{n+1}_p(F)\). For \(p=2\), by a famous theorem due to Kato, symbols in \(H^n_2(F)\) (resp. \(\nu_F(n)\)) correspond bijectively to isometry classes of \(n\)-fold quadratic (resp. bilinear) Pfister forms, and the above notions of linkage carry over naturally to Pfister forms. In order to study linkage properties one can then employ more manageable techniques from the algebraic theory of quadratic forms. Using the above results on differential symbols, the authors show that any two totally separably \((n-1)\)-linked \(n\)-fold quadratic Pfister forms are inseparably \((n-1)\)-linked. This is significant because there are examples that show that separable \((n-1)\)-linkage alone generally does not imply inseparable \((n-1)\)-linkage. They also construct examples of nonisometric anisotropic \(n\)-fold quadratic Pfister forms that are totally separably (resp. totally inseparably) \((n-1)\)-linked and state a few open problems concerning linkage.