Common subfields of \(p\)-algebras of prime degree (Q908229)

Fix a prime number \(p\). Let \(F\) be a field and \(A\) a central simple algebra with center \(F\), of degree \(p^e\). We say \(A\) is a cyclic algebra if it contains a commutative \(F\)-subalgebra \(K\subset A\), Galois over \(F\) such that \(\mathrm{Gal}(K/F)=\langle \sigma \rangle \cong C_{p^e}\). The study of central simple algebras usually divides into two cases according to the characteristic of the base field being \(p\) or prime to \(p\). In both cases (although usually using different methods), cyclic algebras play an important role, expressed in the following theorems: {\parindent=6mm \begin{itemize}\item[(1)] Teichmüller: Let \(A\) be a f.d. \(F\)-c.s.a of exponent \(p^e\) and assume the characteristic of \(F\) is \(p\), such algebras are called \(p\)-algebras. Then \(A\) is Brauer equivalent to the tensor product of cyclic algebras of degree \(p^e\). \item[(2)] Merkurjev and Suslin: Let \(A\) be a f.d. \(F\)-c.s.a of exponent \(p^e\) and assume \(F\) contains a primitive \(p^e\)-root of unity, in particular the characteristic of \(F\) is prime to \(p\). Then \(A\) is Brauer equivalent to the tensor product of cyclic algebras of degree \(p^e\). \end{itemize}} Assume that \(\mathrm{char}(F)\neq p\) (the so called non-modular case) and that \(F\) contains a primitive \(p\)-th root of one, \(\rho\). Then any cyclic algebra of degree \(p\) over \(F\) has a presentation of the form \[ (\alpha, \beta)_{p,F}=F[x,y \mid x^p=\alpha; \;y^p=\beta; \;yx=\rho xy]. \] In this situation, one can see that the slots are symmetric in the sense that both \(F[x], F[y]\) are cyclic Galois extensions and the study of presentations of \(A\) mounts to understanding pairs of Kummer extensions of \(F\), splitting the algebra \(A\). In the modular case, that is when \(\mathrm{char}(F)=p\), the situation is very different, as inseparable extensions of dimension \(p\) arise. In particular, the presentation of a cyclic algebra, \(A\), involves both separable and inseparable extensions and \(A\) takes the following form, called the Artin-Schreier symbol, \[ [\alpha,\beta)_{p,F}=F[x,y \mid x^p-x=\alpha; \;y^p=\beta; \;yx-xy=y]. \] Note that the left slot corresponds to a cyclic Galois extension and the right one corresponds to a purely inseparable extension. Thus, understanding the presentations of \(A\) is a bit more delicate as it requires the understanding of both separable and inseparable extensions and the interplay between them. Call two algebras as above \textit{left linked} if they have a presentation sharing the left (separable) slot and \textit{right linked} if they have a presentation sharing the right (purely inseparable) slot, notice that in the non-modular case both concepts are the same. The above leads to the following question: Suppose that \(A\) and \(B\) are left linked, are they also right linked? and vice versa. In [Nachr. Akad. Wiss. Gött., II. Math.-Phys. Kl. 1975, 251--259 (1975; Zbl 0316.16018)] the case of \(p=2\) was studied by \textit{P. Draxl} who proved that if \(A\) and \(B\) are right linked then they are also left linked. Later, in [Bull. Belg. Math. Soc. - Simon Stevin 9, No. 3, 415--418 (2002; Zbl 1044.16012)], \textit{T. Y. Lam} simplified the proof of Draxl and showed that the converse is not true, that is that for \(p=2\), left linked algebras are not necessarily right linked. In this work, the author generalizes the above results to odd primes, namely that right linkage implies left linkage and that the converse is not true in general. For the first part of the statement the author starts with a pair of right linked algebras and explicitly builds presentations which are left linked by standard manipulations of symbol presentations. For the second part, the author constructs two algebras \[ [1,\alpha)_{p,F_0}, \;[1,\beta)_{p,F_0}, \] where \(F_0=\mathbb{F}_p(\alpha,\beta)\), and \(\mathbb{F}_p\) is the field with \(p\) elements. Clearly these algebras are left linked. To see that they are not right linked the author uses the theory of Henselian valuations by extending scalars to \(F=\mathbb{F}_p((\alpha))((\beta))\), and computing the value groups for the separable extensions contained in each algebra and showing they are not the same. This example for the case \(p=2\) recovers the example given by Lam in [loc. cit.] but the proof is different and enables the generalization to odd primes.