Levels of quaternion algebras (Q5900385)

The main theorem of the author reads: Let \(m\) be a nonnegative integer. Then there exists a quaternion division algebra \(D\) with \[ m+1 \leq s(D)\leq m+1+[m/3]. \] In particular, at least two ``new values'' for the level \(s(D)\) exist in the interval \([2^k+2,2^{k+1}-1]\) for any \(k\geq 4\). The main part of this fine paper is elementary and constructive using a clever idea, as is typical for most of Hoffmann's papers. Starting with a quaternion division algebra \(Q=(a,b)\) over some field \(F\) one considers the quadratic forms \[ \psi_n=\langle 1\rangle\perp n\times \langle 1,a,b,-ab\rangle\quad\text{of dimension }4n+1 \] \[ \varphi_n= \langle 1\rangle\perp n\times\langle a,b,-ab\rangle\quad\text{of dimension }3n+1 \] for \(n\in\mathbb N\). Then the following criterion holds: Let \(n>m\geq 0\) be integers, let \(K\) be an extension field of \(F\). If \(\varphi_n\) is isotropic and \(\psi_m\) is anisotropic over \(K\), then \[ s(Q_K)\in[m+1,n]. \] In order to fulfill this criterion in a generic situation one takes \(K_n=F(\varphi_n)\), the function field of the projective quadric \(\varphi_n=0\). The difficult part is to show that \(\psi_m\) remains anisotropic over \(K_n\) for suitable \(m<n\). At this point the author has to use a deep theorem of Karpenko-Merkurjev on the ``essential dimension'' of quadrics. Final remark: Up to now it is not possible to specify explicit numbers \(s\) different from \(2^k\), \(2^k+1\) which are values \(s=s(D)\) of quaternion division algebras \(D\).