1. Introduction
 Let 
 $p$
 be a prime number, let
$p$
 be a prime number, let 
 $F$
 be a field of characteristic different from
$F$
 be a field of characteristic different from 
 $p$
 and containing a primitive
$p$
 and containing a primitive 
 $p$
th root of unity
$p$
th root of unity 
 $\zeta$
, and let
$\zeta$
, and let 
 $\Gamma _F$
 be the absolute Galois group of
$\Gamma _F$
 be the absolute Galois group of 
 $F$
. The norm-residue isomorphism theorem of Voevodsky and Rost [Reference Haesemeyer and WeibelHW19] gives an explicit presentation by generators and relations of the cohomology ring
$F$
. The norm-residue isomorphism theorem of Voevodsky and Rost [Reference Haesemeyer and WeibelHW19] gives an explicit presentation by generators and relations of the cohomology ring 
 $H^{* }(F,\mathbb Z/p\mathbb Z)=H^{* }(\Gamma _F,\mathbb Z/p\mathbb Z)$
. In view of this complete description of the cup product, the research on
$H^{* }(F,\mathbb Z/p\mathbb Z)=H^{* }(\Gamma _F,\mathbb Z/p\mathbb Z)$
. In view of this complete description of the cup product, the research on 
 $H^{* }(F, \mathbb Z/p\mathbb Z)$
 shifted in recent years to external operations, defined in terms of the differential graded ring of continuous cochains
$H^{* }(F, \mathbb Z/p\mathbb Z)$
 shifted in recent years to external operations, defined in terms of the differential graded ring of continuous cochains 
 $C^{*}(\Gamma _F, \mathbb Z/p\mathbb Z)$
.
$C^{*}(\Gamma _F, \mathbb Z/p\mathbb Z)$
.
 Hopkins and Wickelgren [Reference Hopkins and WickelgrenHW15] asked whether 
 $C^{*}(\Gamma _F, \mathbb Z/p\mathbb Z)$
 is formal for every field
$C^{*}(\Gamma _F, \mathbb Z/p\mathbb Z)$
 is formal for every field 
 $F$
 and every prime
$F$
 and every prime 
 $p$
. Loosely speaking, this amounts to saying that no essential information is lost when passing from
$p$
. Loosely speaking, this amounts to saying that no essential information is lost when passing from 
 $C^{*}(\Gamma _F, \mathbb Z/p\mathbb Z)$
 to
$C^{*}(\Gamma _F, \mathbb Z/p\mathbb Z)$
 to 
 $H^{* }(F, \mathbb Z/p\mathbb Z)$
. The authors of [Reference Hopkins and WickelgrenHW15] were unaware of earlier work of Positselski, who had already shown in [Reference PositselskiPos11, Section 9.11] that
$H^{* }(F, \mathbb Z/p\mathbb Z)$
. The authors of [Reference Hopkins and WickelgrenHW15] were unaware of earlier work of Positselski, who had already shown in [Reference PositselskiPos11, Section 9.11] that 
 $C^{*}(\Gamma _F, \mathbb Z/p\mathbb Z)$
 is not formal for some finite extensions
$C^{*}(\Gamma _F, \mathbb Z/p\mathbb Z)$
 is not formal for some finite extensions 
 $F$
 of
$F$
 of 
 $\mathbb Q_{\ell }$
 and
$\mathbb Q_{\ell }$
 and 
 ${\mathbb F}_{\ell }((z))$
, where
${\mathbb F}_{\ell }((z))$
, where 
 $\ell \neq p$
. Positselski then wrote a detailed exposition of his counterexamples in [Reference PositselskiPos17].
$\ell \neq p$
. Positselski then wrote a detailed exposition of his counterexamples in [Reference PositselskiPos17].
 For Positselski’s method to work, it seemed important that 
 $F$
 did not contain all the roots of unity of
$F$
 did not contain all the roots of unity of 
 $p$
-power order. This motivated the following question; see [Reference PositselskiPos17, p. 226].
$p$
-power order. This motivated the following question; see [Reference PositselskiPos17, p. 226].
Question 1.1 (Positselski). Does there exist a field 
 $F$
 containing all roots of unity of
$F$
 containing all roots of unity of 
 $p$
-power order such that
$p$
-power order such that 
 $C^{*}(\Gamma _F,\mathbb Z/p\mathbb Z)$
 is not formal?
$C^{*}(\Gamma _F,\mathbb Z/p\mathbb Z)$
 is not formal?
 We showed in [Reference Merkurjev and ScaviaMS22, Theorem 1.6] that Question 1.1 has a positive answer when 
 $p=2$
. In the present work, we provide examples showing that the answer to Question 1.1 is affirmative for all primes
$p=2$
. In the present work, we provide examples showing that the answer to Question 1.1 is affirmative for all primes 
 $p$
.
$p$
.
Theorem 1.2. 
Let 
 $p$
 be a prime number and let
$p$
 be a prime number and let 
 $F$
 be a field of characteristic different from
$F$
 be a field of characteristic different from 
 $p$
. There exists a field
$p$
. There exists a field 
 $L$
 containing
$L$
 containing 
 $F$
 such that the differential graded ring
$F$
 such that the differential graded ring 
 $C^{*}(\Gamma _L,\mathbb Z/p\mathbb Z)$
 is not formal.
$C^{*}(\Gamma _L,\mathbb Z/p\mathbb Z)$
 is not formal.
 To detect non-formality of the cochain differential graded ring, we use Massey products. For any 
 $n\geq 2$
 and all
$n\geq 2$
 and all 
 $\chi _1,\ldots ,\chi _n\in H^1(F,\mathbb Z/p\mathbb Z)$
, the Massey product of
$\chi _1,\ldots ,\chi _n\in H^1(F,\mathbb Z/p\mathbb Z)$
, the Massey product of 
 $\chi _1,\ldots ,\chi _n$
 is a certain subset
$\chi _1,\ldots ,\chi _n$
 is a certain subset 
 $\langle {\chi _1,\ldots ,\chi _n}\rangle \subset H^2(F,\mathbb Z/p\mathbb Z)$
; see Section 2.2 for the definition. We say that
$\langle {\chi _1,\ldots ,\chi _n}\rangle \subset H^2(F,\mathbb Z/p\mathbb Z)$
; see Section 2.2 for the definition. We say that 
 $\langle {\chi _1,\ldots ,\chi _n}\rangle$
 is defined if it is not empty, and that it vanishes if it contains
$\langle {\chi _1,\ldots ,\chi _n}\rangle$
 is defined if it is not empty, and that it vanishes if it contains 
 $0$
. When
$0$
. When 
 $\operatorname {char}(F)\neq p$
 and
$\operatorname {char}(F)\neq p$
 and 
 $F$
 contains a primitive
$F$
 contains a primitive 
 $p$
th root of unity
$p$
th root of unity 
 $\zeta$
, Kummer theory gives an identification
$\zeta$
, Kummer theory gives an identification 
 $H^1(F,\mathbb Z/p\mathbb Z)=F^{\times }/F^{\times p}$
, and we may thus consider Massey products
$H^1(F,\mathbb Z/p\mathbb Z)=F^{\times }/F^{\times p}$
, and we may thus consider Massey products 
 $\langle {a_1,\ldots ,a_n}\rangle$
, where
$\langle {a_1,\ldots ,a_n}\rangle$
, where 
 $a_i\in F^\times$
 for
$a_i\in F^\times$
 for 
 $1\leq i\leq n$
.
$1\leq i\leq n$
.
 Let 
 $n\geq 3$
 be an integer, let
$n\geq 3$
 be an integer, let 
 $\chi _1,\ldots ,\chi _n\in H^1(F,\mathbb Z/p\mathbb Z)$
 and consider the following assertions.
$\chi _1,\ldots ,\chi _n\in H^1(F,\mathbb Z/p\mathbb Z)$
 and consider the following assertions.
 $$\begin{align}\textrm{The Massey product $\langle {\chi _1,\ldots ,\chi _n}\rangle $ vanishes.}\end{align}$$
$$\begin{align}\textrm{The Massey product $\langle {\chi _1,\ldots ,\chi _n}\rangle $ vanishes.}\end{align}$$
 $$\begin{align}\textrm {The Massey product $\langle {\chi _1,\ldots ,\chi _n}\rangle $ is defined.}\end{align}$$
$$\begin{align}\textrm {The Massey product $\langle {\chi _1,\ldots ,\chi _n}\rangle $ is defined.}\end{align}$$
 $$\begin{align}\textrm {We have $\chi _i\cup \chi _{i+1}=0$ for all $1\leq i\leq n-1$.}\end{align}$$
$$\begin{align}\textrm {We have $\chi _i\cup \chi _{i+1}=0$ for all $1\leq i\leq n-1$.}\end{align}$$
We have that (1.1) implies (1.2), and that (1.2) implies (1.3). The Massey vanishing conjecture, due to Mináč and Tân [Reference Mináč and TânMT17b] and inspired by the earlier work of Hopkins and Wickelgren [Reference Hopkins and WickelgrenHW15], predicts that (1.2) implies (1.1). This conjecture has sparked a lot of activity in recent years. When 
 $F$
 is an arbitrary field, the conjecture was shown when either
$F$
 is an arbitrary field, the conjecture was shown when either 
 $n=3$
 and
$n=3$
 and 
 $p$
 is arbitrary, by Efrat and Matzri and Mináč and Tân [Reference MatzriMat18, Reference Efrat and MatzriEM17, Reference Mináč and TânMT16], or when
$p$
 is arbitrary, by Efrat and Matzri and Mináč and Tân [Reference MatzriMat18, Reference Efrat and MatzriEM17, Reference Mináč and TânMT16], or when 
 $n=4$
 and
$n=4$
 and 
 $p=2$
, by [Reference Merkurjev and ScaviaMS23]. When
$p=2$
, by [Reference Merkurjev and ScaviaMS23]. When 
 $F$
 is a number field, the conjecture was proved for all
$F$
 is a number field, the conjecture was proved for all 
 $n\geq 3$
 and all primes
$n\geq 3$
 and all primes 
 $p$
 by Harpaz and Wittenberg [Reference Harpaz and WittenbergHW23].
$p$
 by Harpaz and Wittenberg [Reference Harpaz and WittenbergHW23].
 When 
 $n=3$
, it is a direct consequence of the definition of the Massey product that (1.3) implies (1.2). Thus, (1.1), (1.2) and (1.3) are equivalent when
$n=3$
, it is a direct consequence of the definition of the Massey product that (1.3) implies (1.2). Thus, (1.1), (1.2) and (1.3) are equivalent when 
 $n=3$
.
$n=3$
.
 In [Reference Mináč and TânMT17a, Question 4.2], Mináč and Tân asked whether (1.3) implies (1.1). This became known as the strong Massey vanishing conjecture (see, e.g., [Reference Pál and SzabóPS18]). If 
 $F$
 is a field,
$F$
 is a field, 
 $p$
 is a prime number and
$p$
 is a prime number and 
 $n\geq 3$
 is an integer, then, for all characters
$n\geq 3$
 is an integer, then, for all characters 
 $\chi _1,\ldots ,\chi _n\in H^1(F,\mathbb Z/p\mathbb Z)$
 such that
$\chi _1,\ldots ,\chi _n\in H^1(F,\mathbb Z/p\mathbb Z)$
 such that 
 $\chi _i\cup \chi _{i+1}=0$
 for all
$\chi _i\cup \chi _{i+1}=0$
 for all 
 $1\leq i\leq n-1$
, the Massey product
$1\leq i\leq n-1$
, the Massey product 
 $\langle {\chi _1,\ldots ,\chi _n}\rangle$
 vanishes.
$\langle {\chi _1,\ldots ,\chi _n}\rangle$
 vanishes.
 The strong Massey vanishing conjecture implies the Massey vanishing conjecture. However, Harpaz and Wittenberg produced a counterexample to the strong Massey vanishing conjecture, for 
 $n=4$
,
$n=4$
, 
 $p=2$
 and
$p=2$
 and 
 $F=\mathbb Q$
; see [Reference Guillot, Mináč and TopazGMT18, Example A.15]. More precisely, if we let
$F=\mathbb Q$
; see [Reference Guillot, Mináč and TopazGMT18, Example A.15]. More precisely, if we let 
 $b=2$
,
$b=2$
, 
 $c=17$
 and
$c=17$
 and 
 $a=d=bc=34$
, then
$a=d=bc=34$
, then 
 $(a,b)=(b,c)=(c,d)=0$
 in
$(a,b)=(b,c)=(c,d)=0$
 in 
 $\operatorname {Br}(\mathbb Q)$
 but
$\operatorname {Br}(\mathbb Q)$
 but 
 $\langle {a,b,c,d}\rangle$
 is not defined over
$\langle {a,b,c,d}\rangle$
 is not defined over 
 $\mathbb Q$
. In this example, the classes of
$\mathbb Q$
. In this example, the classes of 
 $a,b,c,d$
 in
$a,b,c,d$
 in 
 $F^{\times }/F^{\times 2}$
 are not
$F^{\times }/F^{\times 2}$
 are not 
 ${\mathbb F}_2$
-linearly independent modulo squares. In fact, by a theorem of Guillot, Mináč, Topaz and Wittenberg [Reference Guillot, Mináč and TopazGMT18], if
${\mathbb F}_2$
-linearly independent modulo squares. In fact, by a theorem of Guillot, Mináč, Topaz and Wittenberg [Reference Guillot, Mináč and TopazGMT18], if 
 $F$
 is a number field and
$F$
 is a number field and 
 $a,b,c,d$
 are independent in
$a,b,c,d$
 are independent in 
 $F^\times /F^{\times 2}$
 and satisfy
$F^\times /F^{\times 2}$
 and satisfy 
 $(a,b)=(b,c)=(c,d)=0$
 in
$(a,b)=(b,c)=(c,d)=0$
 in 
 $\operatorname {Br}(F)$
, then
$\operatorname {Br}(F)$
, then 
 $\langle {a,b,c,d}\rangle$
 vanishes.
$\langle {a,b,c,d}\rangle$
 vanishes.
 If 
 $F$
 is a field for which the strong Massey vanishing conjecture fails, for some
$F$
 is a field for which the strong Massey vanishing conjecture fails, for some 
 $n\geq 3$
 and some prime
$n\geq 3$
 and some prime 
 $p$
, then
$p$
, then 
 $C^{*}(\Gamma _F,\mathbb Z/p\mathbb Z)$
 is not formal; see Lemma 2.3. Therefore, Theorem 1.2 follows from the next more precise result.
$C^{*}(\Gamma _F,\mathbb Z/p\mathbb Z)$
 is not formal; see Lemma 2.3. Therefore, Theorem 1.2 follows from the next more precise result.
Theorem 1.3. 
Let 
 $p$
 be a prime number and let
$p$
 be a prime number and let 
 $F$
 be a field of characteristic different from
$F$
 be a field of characteristic different from 
 $p$
. There exist a field
$p$
. There exist a field 
 $L$
 containing
$L$
 containing 
 $F$
 and
$F$
 and 
 $\chi _1,\chi _2,\chi _3,\chi _4\in H^1(L,\mathbb Z/p\mathbb Z)$
 such that
$\chi _1,\chi _2,\chi _3,\chi _4\in H^1(L,\mathbb Z/p\mathbb Z)$
 such that 
 $\chi _1\cup \chi _2=\chi _2\cup \chi _3=\chi _3\cup \chi _4=0$
 in
$\chi _1\cup \chi _2=\chi _2\cup \chi _3=\chi _3\cup \chi _4=0$
 in 
 $H^2(L,\mathbb Z/p\mathbb Z)$
 but
$H^2(L,\mathbb Z/p\mathbb Z)$
 but 
 $\langle {\chi _1,\chi _2,\chi _3,\chi _4}\rangle$
 is not defined. Thus, the strong Massey vanishing conjecture at
$\langle {\chi _1,\chi _2,\chi _3,\chi _4}\rangle$
 is not defined. Thus, the strong Massey vanishing conjecture at 
 $n=4$
 and the prime
$n=4$
 and the prime 
 $p$
 fails for
$p$
 fails for 
 $L$
, and
$L$
, and 
 $C^{*}(\Gamma _L,\mathbb Z/p\mathbb Z)$
 is not formal.
$C^{*}(\Gamma _L,\mathbb Z/p\mathbb Z)$
 is not formal.
 This gives the first counterexamples to the strong Massey vanishing conjecture for all odd primes 
 $p$
. We easily deduce that (1.3) does not imply (1.2) for all
$p$
. We easily deduce that (1.3) does not imply (1.2) for all 
 $n\geq 4$
 in general: indeed, if the fourfold Massey product
$n\geq 4$
 in general: indeed, if the fourfold Massey product 
 $\langle {\chi _1,\chi _2,\chi _3,\chi _4}\rangle$
 is not defined, neither is the
$\langle {\chi _1,\chi _2,\chi _3,\chi _4}\rangle$
 is not defined, neither is the 
 $n$
-fold Massey product
$n$
-fold Massey product 
 $\langle {\chi _1,\chi _2,\chi _3,\chi _4,0,\ldots ,0}\rangle$
. Theorem 1.3 was proved in [Reference Merkurjev and ScaviaMS22, Theorem 1.6] when
$\langle {\chi _1,\chi _2,\chi _3,\chi _4,0,\ldots ,0}\rangle$
. Theorem 1.3 was proved in [Reference Merkurjev and ScaviaMS22, Theorem 1.6] when 
 $p=2$
, and is new when
$p=2$
, and is new when 
 $p$
 is odd. Our proof of Theorem 1.3 is uniform in
$p$
 is odd. Our proof of Theorem 1.3 is uniform in 
 $p$
.
$p$
.
 We now describe the main ideas that go into the proof of Theorem 1.3. We may assume, without loss of generality, that 
 $F$
 contains a primitive
$F$
 contains a primitive 
 $p$
th root of unity. In § 2, we collect preliminaries on formality, Massey products and Galois algebras. In particular, we recall Dwyer’s theorem (see Theorem 2.4): a Massey product
$p$
th root of unity. In § 2, we collect preliminaries on formality, Massey products and Galois algebras. In particular, we recall Dwyer’s theorem (see Theorem 2.4): a Massey product 
 $\langle {\chi _1,\ldots ,\chi _n}\rangle \subset H^2(F,\mathbb Z/p\mathbb Z)$
 vanishes (respectively, is defined) if and only if the homomorphism
$\langle {\chi _1,\ldots ,\chi _n}\rangle \subset H^2(F,\mathbb Z/p\mathbb Z)$
 vanishes (respectively, is defined) if and only if the homomorphism 
 $(\chi _1,\ldots ,\chi _n)\colon \Gamma _F\to (\mathbb Z/p\mathbb Z)^n$
 lifts to the group
$(\chi _1,\ldots ,\chi _n)\colon \Gamma _F\to (\mathbb Z/p\mathbb Z)^n$
 lifts to the group 
 $U_{n+1}$
 of upper unitriangular matrices in
$U_{n+1}$
 of upper unitriangular matrices in 
 $\operatorname {GL}_{n+1}({\mathbb F}_p)$
 (respectively, to the group
$\operatorname {GL}_{n+1}({\mathbb F}_p)$
 (respectively, to the group 
 $\overline {U}_{n+1}$
 of upper unitriangular matrices in
$\overline {U}_{n+1}$
 of upper unitriangular matrices in 
 $\operatorname {GL}_{n+1}({\mathbb F}_p)$
 with top-right corner removed). As for [Reference Merkurjev and ScaviaMS22, Theorem 1.6], our approach is based on Corollary 2.5, which is a restatement of Theorem 2.4 in terms of Galois algebras.
$\operatorname {GL}_{n+1}({\mathbb F}_p)$
 with top-right corner removed). As for [Reference Merkurjev and ScaviaMS22, Theorem 1.6], our approach is based on Corollary 2.5, which is a restatement of Theorem 2.4 in terms of Galois algebras.
 In § 3, we show that a fourfold Massey product 
 $\langle {a,b,c,d}\rangle$
 is defined over
$\langle {a,b,c,d}\rangle$
 is defined over 
 $F$
 if and only if a certain system of equations admits a solution over
$F$
 if and only if a certain system of equations admits a solution over 
 $F$
. Moreover, the variety defined by these equations is a torsor under a torus; see Proposition 3.7. This equivalence is proved by using Dwyer’s Theorem 2.4 to rephrase the property that
$F$
. Moreover, the variety defined by these equations is a torsor under a torus; see Proposition 3.7. This equivalence is proved by using Dwyer’s Theorem 2.4 to rephrase the property that 
 $\langle {a,b,c,d}\rangle$
 is defined in terms of
$\langle {a,b,c,d}\rangle$
 is defined in terms of 
 $\overline {U}_5$
-Galois algebras, and then by a detailed study of Galois
$\overline {U}_5$
-Galois algebras, and then by a detailed study of Galois 
 $G$
-algebras, for
$G$
-algebras, for 
 $G=U_3,\overline {U}_4,U_4,\overline {U}_5$
. As a consequence, we also obtain an alternative proof of the Massey vanishing conjecture for
$G=U_3,\overline {U}_4,U_4,\overline {U}_5$
. As a consequence, we also obtain an alternative proof of the Massey vanishing conjecture for 
 $n=3$
 and any prime
$n=3$
 and any prime 
 $p$
; see Proposition 3.6.
$p$
; see Proposition 3.6.
 In § 4, we use the work of § 3.4 to construct a ‘generic variety’ for the property that 
 $\langle {a,b,c,d}\rangle$
 is defined. More precisely, under the assumption that
$\langle {a,b,c,d}\rangle$
 is defined. More precisely, under the assumption that 
 $(a,b)=(c,d)=0$
 in
$(a,b)=(c,d)=0$
 in 
 $\operatorname {Br}(F)$
 and letting
$\operatorname {Br}(F)$
 and letting 
 $X$
 be the Severi–Brauer variety of
$X$
 be the Severi–Brauer variety of 
 $(b,c)$
, we construct an
$(b,c)$
, we construct an 
 $F$
-torus
$F$
-torus 
 $T$
 and a
$T$
 and a 
 $T_{F(X)}$
-torsor
$T_{F(X)}$
-torsor 
 $E_w$
 such that, if
$E_w$
 such that, if 
 $E_w$
 is non-trivial, then
$E_w$
 is non-trivial, then 
 $\langle {a,b,c,d}\rangle$
 is not defined over
$\langle {a,b,c,d}\rangle$
 is not defined over 
 $F(X)$
; see Corollary 4.5. The definition of
$F(X)$
; see Corollary 4.5. The definition of 
 $E_w$
 depends on a rational function
$E_w$
 depends on a rational function 
 $w\in F(X)^\times$
, which we construct in Lemma 4.1(3).
$w\in F(X)^\times$
, which we construct in Lemma 4.1(3).
 Since 
 $(a,b)=(b,c)=(c,d)=0$
 in
$(a,b)=(b,c)=(c,d)=0$
 in 
 $\operatorname {Br}(F(X))$
, the proof of Theorem 1.3 will be complete once we give an example of
$\operatorname {Br}(F(X))$
, the proof of Theorem 1.3 will be complete once we give an example of 
 $a,b,c,d$
 for which the corresponding torsor
$a,b,c,d$
 for which the corresponding torsor 
 $E_w$
 is non-trivial. Here, we consider the generic quadruple
$E_w$
 is non-trivial. Here, we consider the generic quadruple 
 $a,b,c,d$
 such that
$a,b,c,d$
 such that 
 $(a,b)$
 and
$(a,b)$
 and 
 $(c,d)$
 are trivial. More precisely, we let
$(c,d)$
 are trivial. More precisely, we let 
 $x,y$
 be two independent variables over
$x,y$
 be two independent variables over 
 $F$
, and replace
$F$
, and replace 
 $F$
 by
$F$
 by 
 $E:= F(x,y)$
. We then set
$E:= F(x,y)$
. We then set 
 $a:= 1-x$
,
$a:= 1-x$
, 
 $b:= x$
,
$b:= x$
, 
 $c:= y$
 and
$c:= y$
 and 
 $d:= 1-y$
 over
$d:= 1-y$
 over 
 $E$
. We have
$E$
. We have 
 $(a,b)=(c,d)=0$
 in
$(a,b)=(c,d)=0$
 in 
 $\operatorname {Br}(E)$
. The class
$\operatorname {Br}(E)$
. The class 
 $(b,c)$
 is not zero in
$(b,c)$
 is not zero in 
 $\operatorname {Br}(E)$
, so the Severi–Brauer variety
$\operatorname {Br}(E)$
, so the Severi–Brauer variety 
 $X/E$
 of
$X/E$
 of 
 $(b,c)$
 is non-trivial, but
$(b,c)$
 is non-trivial, but 
 $(b,c)=0$
 over
$(b,c)=0$
 over 
 $L:= E(X)$
.
$L:= E(X)$
.
 It is natural to attempt to prove that 
 $E_w$
 is non-trivial over
$E_w$
 is non-trivial over 
 $L$
 by performing residue calculations to deduce that this torsor is ramified. However, the torsor
$L$
 by performing residue calculations to deduce that this torsor is ramified. However, the torsor 
 $E_w$
 is in fact unramified. We are thus led to consider a finer obstruction to the triviality of
$E_w$
 is in fact unramified. We are thus led to consider a finer obstruction to the triviality of 
 $E_w$
. This ‘secondary obstruction’ is only defined for unramified torsors. We describe the necessary homological algebra in Appendix A, and we define the obstruction and give a method to compute it in Appendix B. In § 5, an explicit calculation shows that the obstruction is non-zero on
$E_w$
. This ‘secondary obstruction’ is only defined for unramified torsors. We describe the necessary homological algebra in Appendix A, and we define the obstruction and give a method to compute it in Appendix B. In § 5, an explicit calculation shows that the obstruction is non-zero on 
 $E_w$
, and hence
$E_w$
, and hence 
 $E_w$
 is non-trivial, as desired.
$E_w$
 is non-trivial, as desired.
1.1 Notation
 Let 
 $F$
 be a field, let
$F$
 be a field, let 
 $F_s$
 be a separable closure of
$F_s$
 be a separable closure of 
 $F$
 and denote by
$F$
 and denote by 
 $\Gamma _F:= \operatorname {Gal}(F_s/F)$
 the absolute Galois group of
$\Gamma _F:= \operatorname {Gal}(F_s/F)$
 the absolute Galois group of 
 $F$
.
$F$
.
 If 
 $E$
 is an
$E$
 is an 
 $F$
-algebra, we let
$F$
-algebra, we let 
 $H^i(E,-)$
 be the étale cohomology of
$H^i(E,-)$
 be the étale cohomology of 
 $\operatorname {Spec}(E)$
 (possibly non-abelian if
$\operatorname {Spec}(E)$
 (possibly non-abelian if 
 $i\leq 1$
). If
$i\leq 1$
). If 
 $E$
 is a field,
$E$
 is a field, 
 $H^i(E,-)$
 may be identified with the continuous cohomology of
$H^i(E,-)$
 may be identified with the continuous cohomology of 
 $\Gamma _E$
.
$\Gamma _E$
.
 We fix a prime 
 $p$
, and we suppose that
$p$
, and we suppose that 
 $\operatorname {char}(F)\neq p$
. If
$\operatorname {char}(F)\neq p$
. If 
 $E$
 is an
$E$
 is an 
 $F$
-algebra and
$F$
-algebra and 
 $a_1,\ldots ,a_n\in E^{\times }$
, we define the étale
$a_1,\ldots ,a_n\in E^{\times }$
, we define the étale 
 $E$
-algebra
$E$
-algebra 
 $E_{a_1,\ldots ,a_n}$
 by
$E_{a_1,\ldots ,a_n}$
 by
 \begin{equation*}E_{a_1,\ldots ,a_n}:= E[x_1,\ldots ,x_n]/(x_1^p-a_1,\ldots ,x_n^p-a_n),\end{equation*}
\begin{equation*}E_{a_1,\ldots ,a_n}:= E[x_1,\ldots ,x_n]/(x_1^p-a_1,\ldots ,x_n^p-a_n),\end{equation*}
and we set 
 $(a_i)^{1/p}:= x_i$
. More generally, for all integers
$(a_i)^{1/p}:= x_i$
. More generally, for all integers 
 $d$
, we set
$d$
, we set 
 $(a_i)^{d/p}:= x_i^d$
. We denote by
$(a_i)^{d/p}:= x_i^d$
. We denote by 
 $R_{a_1,\ldots ,a_n}(-)$
 the functor of Weil restriction along
$R_{a_1,\ldots ,a_n}(-)$
 the functor of Weil restriction along 
 $F_{a_1,\ldots ,a_n}/F$
. In particular,
$F_{a_1,\ldots ,a_n}/F$
. In particular, 
 $R_{a_1,\ldots ,a_n}({\mathbb G}_{\operatorname {m}})$
 is the quasi-trivial torus associated to
$R_{a_1,\ldots ,a_n}({\mathbb G}_{\operatorname {m}})$
 is the quasi-trivial torus associated to 
 $F_{a_1,\ldots ,a_n}/F$
, and we denote by
$F_{a_1,\ldots ,a_n}/F$
, and we denote by 
 $R^{(1)}_{a_1,\ldots ,a_n}({\mathbb G}_{\operatorname {m}})$
 the norm-one subtorus of
$R^{(1)}_{a_1,\ldots ,a_n}({\mathbb G}_{\operatorname {m}})$
 the norm-one subtorus of 
 $R_{a_1,\ldots ,a_n}({\mathbb G}_{\operatorname {m}})$
. We denote by
$R_{a_1,\ldots ,a_n}({\mathbb G}_{\operatorname {m}})$
. We denote by 
 $N_{a_1,\ldots ,a_n}$
 the norm map from
$N_{a_1,\ldots ,a_n}$
 the norm map from 
 $F_{a_1,\ldots ,a_n}$
 to
$F_{a_1,\ldots ,a_n}$
 to 
 $F$
.
$F$
.
 We write 
 $\operatorname {Br}(F)$
 for the Brauer group of
$\operatorname {Br}(F)$
 for the Brauer group of 
 $F$
. If
$F$
. If 
 $\operatorname {char}(F)\neq p$
 and
$\operatorname {char}(F)\neq p$
 and 
 $F$
 contains a primitive
$F$
 contains a primitive 
 $p$
th root of unity, for all
$p$
th root of unity, for all 
 $a,b\in F^\times$
 we denote by
$a,b\in F^\times$
 we denote by 
 $(a,b)$
 the corresponding degree-
$(a,b)$
 the corresponding degree-
 $p$
 cyclic algebra and also its class in
$p$
 cyclic algebra and also its class in 
 $\operatorname {Br}(F)$
; see § 2.1. We denote by
$\operatorname {Br}(F)$
; see § 2.1. We denote by 
 $N_{a_1,\ldots ,a_n}\colon \operatorname {Br}(F_{a_1,\ldots ,a_n})\to \operatorname {Br}(F)$
 the corestriction map along
$N_{a_1,\ldots ,a_n}\colon \operatorname {Br}(F_{a_1,\ldots ,a_n})\to \operatorname {Br}(F)$
 the corestriction map along 
 $F_{a_1,\ldots ,a_n}/F$
.
$F_{a_1,\ldots ,a_n}/F$
.
 An 
 $F$
-variety is a separated integral
$F$
-variety is a separated integral 
 $F$
-scheme of finite type. If
$F$
-scheme of finite type. If 
 $X$
 is an
$X$
 is an 
 $F$
-variety, we let
$F$
-variety, we let 
 $F(X)$
 be the function field of
$F(X)$
 be the function field of 
 $X$
, and we write
$X$
, and we write 
 $X^{(1)}$
 for the collection of all points of codimension
$X^{(1)}$
 for the collection of all points of codimension 
 $1$
 in
$1$
 in 
 $X$
. We set
$X$
. We set 
 $X_s:= X\times _FF_s$
. If
$X_s:= X\times _FF_s$
. If 
 $K$
 is an étale algebra over
$K$
 is an étale algebra over 
 $F$
, we write
$F$
, we write 
 $X_K$
 for
$X_K$
 for 
 $X\times _FK$
. For all
$X\times _FK$
. For all 
 $a_1,\ldots ,a_n\in F^\times$
, we write
$a_1,\ldots ,a_n\in F^\times$
, we write 
 $X_{a_1,\ldots ,a_n}$
 for
$X_{a_1,\ldots ,a_n}$
 for 
 $X_{F_{a_1,\ldots ,a_n}}$
. When
$X_{F_{a_1,\ldots ,a_n}}$
. When 
 $X=\mathbb P^d_F$
 is a
$X=\mathbb P^d_F$
 is a 
 $d$
-dimensional projective space, we denote by
$d$
-dimensional projective space, we denote by 
 $\mathbb P^d_{a_1,\ldots ,a_n}$
 the base change of
$\mathbb P^d_{a_1,\ldots ,a_n}$
 the base change of 
 $\mathbb P^d_F$
 to
$\mathbb P^d_F$
 to 
 $F_{a_1,\ldots ,a_d}$
.
$F_{a_1,\ldots ,a_d}$
.
2. Preliminaries
2.1 Galois algebras and Kummer theory
 Let 
 $F$
 be a field and let
$F$
 be a field and let 
 $G$
 be a finite group. A
$G$
 be a finite group. A 
 $G$
-algebra is an étale
$G$
-algebra is an étale 
 $F$
-algebra
$F$
-algebra 
 $L$
 on which
$L$
 on which 
 $G$
 acts via
$G$
 acts via 
 $F$
-algebra automorphisms. The
$F$
-algebra automorphisms. The 
 $G$
-algebra
$G$
-algebra 
 $L$
 is Galois if
$L$
 is Galois if 
 $|G|={\dim }_F(L)$
 and
$|G|={\dim }_F(L)$
 and 
 $L^G=F$
; see [Reference Knus, Merkurjev, Rost, Tignol and TitsKMRT98, Definitions (18.15)]. A
$L^G=F$
; see [Reference Knus, Merkurjev, Rost, Tignol and TitsKMRT98, Definitions (18.15)]. A 
 $G$
-algebra
$G$
-algebra 
 $L/F$
 is Galois if and only if the morphism of schemes
$L/F$
 is Galois if and only if the morphism of schemes 
 $\operatorname {Spec}(L)\to \operatorname {Spec}(F)$
 is an étale
$\operatorname {Spec}(L)\to \operatorname {Spec}(F)$
 is an étale 
 $G$
-torsor. If
$G$
-torsor. If 
 $L/F$
 is a Galois
$L/F$
 is a Galois 
 $G$
-algebra, then the group algebra
$G$
-algebra, then the group algebra 
 $\mathbb Z[G]$
 acts on the multiplicative group
$\mathbb Z[G]$
 acts on the multiplicative group 
 $L^{\times }$
: an element
$L^{\times }$
: an element 
 ${{\sum }}_{i=1}^r m_ig_i\in \mathbb Z[G]$
, where
${{\sum }}_{i=1}^r m_ig_i\in \mathbb Z[G]$
, where 
 $m_i\in \mathbb Z$
 and
$m_i\in \mathbb Z$
 and 
 $g_i\in G$
, sends
$g_i\in G$
, sends 
 $x\in L^{\times }$
 to
$x\in L^{\times }$
 to 
 ${{\prod }}_{i=1}^r g_i(x)^{m_i}$
.
${{\prod }}_{i=1}^r g_i(x)^{m_i}$
.
By [Reference Knus, Merkurjev, Rost, Tignol and TitsKMRT98, Example (28.15)], we have a canonical bijection
 \begin{align} \operatorname {Hom}_{\operatorname {cont}}(\Gamma _F,G)/_{\sim }\xrightarrow {\sim }\left \{\text {Isomorphism classes of Galois}\,\,G\text{-algebras over}\,\,F\right \}, \end{align}
\begin{align} \operatorname {Hom}_{\operatorname {cont}}(\Gamma _F,G)/_{\sim }\xrightarrow {\sim }\left \{\text {Isomorphism classes of Galois}\,\,G\text{-algebras over}\,\,F\right \}, \end{align}
where, if 
 $f_1,f_2\colon \Gamma _F\to G$
 are continuous group homomorphisms, we say that
$f_1,f_2\colon \Gamma _F\to G$
 are continuous group homomorphisms, we say that 
 $f_1\sim f_2$
 if there exists
$f_1\sim f_2$
 if there exists 
 $g\in G$
 such that
$g\in G$
 such that 
 $gf_1(\sigma )g^{-1}=f_2(\sigma )$
 for all
$gf_1(\sigma )g^{-1}=f_2(\sigma )$
 for all 
 $\sigma \in \Gamma _F$
.
$\sigma \in \Gamma _F$
.
 Let 
 $H$
 be a normal subgroup of
$H$
 be a normal subgroup of 
 $G$
. Under the correspondence (2.1), the map
$G$
. Under the correspondence (2.1), the map 
 $\operatorname {Hom}_{\operatorname {cont}}(\Gamma _F,G)/_{\sim }\to \operatorname {Hom}_{\operatorname {cont}}(\Gamma _F,G/H)/_{\sim }$
 sends the class of a Galois
$\operatorname {Hom}_{\operatorname {cont}}(\Gamma _F,G)/_{\sim }\to \operatorname {Hom}_{\operatorname {cont}}(\Gamma _F,G/H)/_{\sim }$
 sends the class of a Galois 
 $G$
-algebra
$G$
-algebra 
 $L$
 to the class of the Galois
$L$
 to the class of the Galois 
 $G/H$
-algebra
$G/H$
-algebra 
 $L^H$
.
$L^H$
.
Lemma 2.1. 
Let 
 $G$
 be a finite group, and let
$G$
 be a finite group, and let 
 $H, H', S$
 be normal subgroups of
$H, H', S$
 be normal subgroups of 
 $G$
 such that
$G$
 such that 
 $H\subset S$
,
$H\subset S$
, 
 $H'\subset S$
, and the following square is cartesian.
$H'\subset S$
, and the following square is cartesian.

- 
(1) Let  $L$
 be a Galois $L$
 be a Galois $G$
-algebra. Then $G$
-algebra. Then $L^H \otimes _{L^S} L^{H'}$
 has a Galois $L^H \otimes _{L^S} L^{H'}$
 has a Galois $G$
-algebra structure given by $G$
-algebra structure given by $g(x\otimes x'):= g(x)\otimes g(x')$
 for all $g(x\otimes x'):= g(x)\otimes g(x')$
 for all $x\in L^H$
 and $x\in L^H$
 and $x'\in L^{H'}$
, and the inclusions $x'\in L^{H'}$
, and the inclusions $L^H\to L$
 and $L^H\to L$
 and $L^{H'}\to L$
 induce an isomorphism of Galois $L^{H'}\to L$
 induce an isomorphism of Galois $G$
-algebras $G$
-algebras $L^H \otimes _{L^S} L^{H'} \rightarrow L$
. $L^H \otimes _{L^S} L^{H'} \rightarrow L$
.
- 
(2) Conversely, let  $K$
 be a Galois $K$
 be a Galois $G/H$
-algebra, let $G/H$
-algebra, let $K'$
 be a Galois $K'$
 be a Galois $G/H'$
-algebra and let $G/H'$
-algebra and let $E$
 be a Galois $E$
 be a Galois $G/S$
-algebra. Suppose we are given $G/S$
-algebra. Suppose we are given $G$
-equivariant algebra homomorphisms $G$
-equivariant algebra homomorphisms $E \rightarrow K$
 and $E \rightarrow K$
 and $E \rightarrow K'$
. Endow the tensor product $E \rightarrow K'$
. Endow the tensor product $L := K \otimes _{E} K'$
 with the structure of a $L := K \otimes _{E} K'$
 with the structure of a $G$
-algebra given by $G$
-algebra given by $g(x\otimes x'):= g(x)\otimes g(x')$
 for all $g(x\otimes x'):= g(x)\otimes g(x')$
 for all $x\in K$
 and $x\in K$
 and $x'\in K'$
. Then $x'\in K'$
. Then $L$
 is a Galois $L$
 is a Galois $G$
-algebra such that $G$
-algebra such that $L^H\simeq K$
 as $L^H\simeq K$
 as $G/H$
-algebras and $G/H$
-algebras and $L^{H'}\simeq K'$
 as $L^{H'}\simeq K'$
 as $G/H'$
-algebras. $G/H'$
-algebras.
 The condition that (2.2) is cartesian is equivalent to 
 $H\cap H'= \left \{1\right \}$
 and
$H\cap H'= \left \{1\right \}$
 and 
 $S=HH'$
.
$S=HH'$
.
Proof.
 (1) It is clear that the formula 
 $g(x\otimes x'):= g(x)\otimes g(x')$
 makes
$g(x\otimes x'):= g(x)\otimes g(x')$
 makes 
 $L^H \otimes _{L^S} L^{H'}$
 into a
$L^H \otimes _{L^S} L^{H'}$
 into a 
 $G$
-algebra. Consider the following commutative square of
$G$
-algebra. Consider the following commutative square of 
 $F$
-schemes.
$F$
-schemes.

 After base change to a separable closure of 
 $F$
, this square becomes the cartesian square (2.2), and therefore it is cartesian. Passing to coordinate rings, we deduce that the homomorphism
$F$
, this square becomes the cartesian square (2.2), and therefore it is cartesian. Passing to coordinate rings, we deduce that the homomorphism 
 $L^H \otimes _{L^S} L^{H'} \rightarrow L$
 is an isomorphism of
$L^H \otimes _{L^S} L^{H'} \rightarrow L$
 is an isomorphism of 
 $G$
-algebras. In particular, since
$G$
-algebras. In particular, since 
 $L$
 is a Galois
$L$
 is a Galois 
 $G$
-algebra, so is
$G$
-algebra, so is 
 $L^H \otimes _{L^S} L^{H'}$
.
$L^H \otimes _{L^S} L^{H'}$
.
 (2) We have the following 
 $G$
-equivariant cartesian diagram.
$G$
-equivariant cartesian diagram.

 Every 
 $G$
-equivariant morphism between
$G$
-equivariant morphism between 
 $G/H$
 and
$G/H$
 and 
 $G/S$
 is isomorphic to the projection map
$G/S$
 is isomorphic to the projection map 
 $G/H\to G/S$
. Therefore, the base change of
$G/H\to G/S$
. Therefore, the base change of 
 $\operatorname {Spec}(K) \to \operatorname {Spec}(E)$
 to
$\operatorname {Spec}(K) \to \operatorname {Spec}(E)$
 to 
 $F_s$
 is
$F_s$
 is 
 $G$
-equivariantly isomorphic to the projection
$G$
-equivariantly isomorphic to the projection 
 $G/H \to G/S$
. Similarly for
$G/H \to G/S$
. Similarly for 
 $\operatorname {Spec}(K')\to \operatorname {Spec}(E)$
. Therefore, the base change of
$\operatorname {Spec}(K')\to \operatorname {Spec}(E)$
. Therefore, the base change of 
 $\operatorname {Spec}(L)\to \operatorname {Spec}(F)$
 over
$\operatorname {Spec}(L)\to \operatorname {Spec}(F)$
 over 
 $F_s$
 is
$F_s$
 is 
 $G$
-equivariantly isomorphic to
$G$
-equivariantly isomorphic to 
 $(G/H)\times _{G/S}(G/H')\simeq G$
, that is, the morphism
$(G/H)\times _{G/S}(G/H')\simeq G$
, that is, the morphism 
 $\operatorname {Spec}(L)\to \operatorname {Spec}(F)$
 is an étale
$\operatorname {Spec}(L)\to \operatorname {Spec}(F)$
 is an étale 
 $G$
-torsor.
$G$
-torsor.
 Suppose that 
 $\operatorname {char}(F)\neq p$
 and that
$\operatorname {char}(F)\neq p$
 and that 
 $F$
 contains a primitive
$F$
 contains a primitive 
 $p$
th root of unity. We fix a primitive
$p$
th root of unity. We fix a primitive 
 $p$
th root of unity
$p$
th root of unity 
 $\zeta \in F^\times$
. This determines an isomorphism of Galois modules
$\zeta \in F^\times$
. This determines an isomorphism of Galois modules 
 $\mathbb Z/p\mathbb Z \simeq \mu _p$
, given by
$\mathbb Z/p\mathbb Z \simeq \mu _p$
, given by 
 $1\mapsto \zeta$
, and so the Kummer sequence yields an isomorphism
$1\mapsto \zeta$
, and so the Kummer sequence yields an isomorphism
 \begin{align} \operatorname {Hom}_{\operatorname {cont}}(\Gamma _F,\mathbb Z/p\mathbb Z)=H^1(F,\mathbb Z/p\mathbb Z )\simeq H^1(F,\mu _p)\simeq F^{\times }/F^{\times p}. \end{align}
\begin{align} \operatorname {Hom}_{\operatorname {cont}}(\Gamma _F,\mathbb Z/p\mathbb Z)=H^1(F,\mathbb Z/p\mathbb Z )\simeq H^1(F,\mu _p)\simeq F^{\times }/F^{\times p}. \end{align}
 For every 
 $a\in F^{\times }$
, we let
$a\in F^{\times }$
, we let 
 $\chi _a\colon \Gamma _F\to \mathbb Z/p\mathbb Z$
 be the homomorphism corresponding to the coset
$\chi _a\colon \Gamma _F\to \mathbb Z/p\mathbb Z$
 be the homomorphism corresponding to the coset 
 $a F^{\times p}$
 under (2.3). Explicitly, letting
$a F^{\times p}$
 under (2.3). Explicitly, letting 
 $a'\in F_{{s}}^\times$
 be such that
$a'\in F_{{s}}^\times$
 be such that 
 $(a')^p=a$
, we have
$(a')^p=a$
, we have 
 $g(a')=\zeta ^{\chi _a(g)}a'$
 for all
$g(a')=\zeta ^{\chi _a(g)}a'$
 for all 
 $g\in \Gamma _F$
. This definition does not depend on the choice of
$g\in \Gamma _F$
. This definition does not depend on the choice of 
 $a'$
.
$a'$
.
 Now let 
 $n\geq 1$
 be an integer. For all
$n\geq 1$
 be an integer. For all 
 $i=1,\ldots ,n$
, let
$i=1,\ldots ,n$
, let 
 $\sigma _i$
 be the canonical generator of the
$\sigma _i$
 be the canonical generator of the 
 $i$
th factor
$i$
th factor 
 $\mathbb Z/p\mathbb Z$
 of
$\mathbb Z/p\mathbb Z$
 of 
 $(\mathbb Z/p\mathbb Z)^n$
. By (2.3), all Galois
$(\mathbb Z/p\mathbb Z)^n$
. By (2.3), all Galois 
 $(\mathbb Z/p\mathbb Z)^n$
-algebras over
$(\mathbb Z/p\mathbb Z)^n$
-algebras over 
 $F$
 are of the form
$F$
 are of the form 
 $F_{a_1,\ldots ,a_n}$
, where
$F_{a_1,\ldots ,a_n}$
, where 
 $a_1,\ldots ,a_n\in F^\times$
 and the Galois
$a_1,\ldots ,a_n\in F^\times$
 and the Galois 
 $(\mathbb Z/p\mathbb Z)^n$
-algebra structure is defined by
$(\mathbb Z/p\mathbb Z)^n$
-algebra structure is defined by 
 $(\sigma _i-1)a_i^{1/p}=\zeta$
 for all
$(\sigma _i-1)a_i^{1/p}=\zeta$
 for all 
 $i$
 and by
$i$
 and by 
 $(\sigma _i-1)a_j^{1/p}=1$
 for all
$(\sigma _i-1)a_j^{1/p}=1$
 for all 
 $j\neq i$
.
$j\neq i$
.
 We write 
 $(a,b)$
 for the cyclic degree-
$(a,b)$
 for the cyclic degree-
 $p$
 central simple algebra over
$p$
 central simple algebra over 
 $F$
 generated, as an
$F$
 generated, as an 
 $F$
-algebra, by
$F$
-algebra, by 
 $F_a$
 and an element
$F_a$
 and an element 
 $y$
 such that
$y$
 such that
 \begin{equation*}y^p=b,\!\quad ty=y\sigma _a(t)\quad \text {for all $t\in F_a$}.\end{equation*}
\begin{equation*}y^p=b,\!\quad ty=y\sigma _a(t)\quad \text {for all $t\in F_a$}.\end{equation*}
We also write 
 $(a,b)$
 for the class of
$(a,b)$
 for the class of 
 $(a,b)$
 in
$(a,b)$
 in 
 $\operatorname {Br}(F)$
. The Kummer sequence yields a group isomorphism
$\operatorname {Br}(F)$
. The Kummer sequence yields a group isomorphism
 \begin{equation*}\iota \colon H^2(F,\mathbb Z/p\mathbb Z)\xrightarrow {\sim }\operatorname {Br}(F)[p].\end{equation*}
\begin{equation*}\iota \colon H^2(F,\mathbb Z/p\mathbb Z)\xrightarrow {\sim }\operatorname {Br}(F)[p].\end{equation*}
For all 
 $a,b\in F^{\times }$
, we have
$a,b\in F^{\times }$
, we have 
 $\iota (\chi _a\cup \chi _b)=(a,b)$
 in
$\iota (\chi _a\cup \chi _b)=(a,b)$
 in 
 $\operatorname {Br}(F)$
; see [Reference Serre and GreenbergSer79, Chapter XIV, Proposition 5].
$\operatorname {Br}(F)$
; see [Reference Serre and GreenbergSer79, Chapter XIV, Proposition 5].
Lemma 2.2. 
Let 
 $p$
 be a prime, and let
$p$
 be a prime, and let 
 $F$
 be a field of characteristic different from
$F$
 be a field of characteristic different from 
 $p$
 and containing a primitive
$p$
 and containing a primitive 
 $p$
th root of unity
$p$
th root of unity 
 $\zeta$
. The following are equivalent.
$\zeta$
. The following are equivalent.
- 
(i) We have  $(a,b)=0$
 in $(a,b)=0$
 in $\operatorname {Br}(F)$
. $\operatorname {Br}(F)$
.
- 
(ii) There exists  $\alpha \in F_a^\times$
 such that $\alpha \in F_a^\times$
 such that $b=N_a(\alpha )$
. $b=N_a(\alpha )$
.
- 
(iii) There exists  $\beta \in F_b^\times$
 such that $\beta \in F_b^\times$
 such that $a=N_b(\beta )$
. $a=N_b(\beta )$
.
Proof. See [Reference Serre and GreenbergSer79, Chapter XIV, Proposition 4(iii)].
2.2 Formality and Massey products
 Let 
 $(A,\partial )$
 be a differential graded ring, that is,
$(A,\partial )$
 be a differential graded ring, that is, 
 $A=\oplus _{i\geq 0}A^i$
 is a non-negatively graded abelian group with an associative multiplication which respects the grading, and
$A=\oplus _{i\geq 0}A^i$
 is a non-negatively graded abelian group with an associative multiplication which respects the grading, and 
 $\partial \colon A\to A$
 is a group homomorphism of degree
$\partial \colon A\to A$
 is a group homomorphism of degree 
 $1$
 such that
$1$
 such that 
 $\partial \circ \partial =0$
 and
$\partial \circ \partial =0$
 and 
 $\partial (ab)=\partial (a)b+(-1)^ia\partial (b)$
 for all
$\partial (ab)=\partial (a)b+(-1)^ia\partial (b)$
 for all 
 $i\geq 0$
,
$i\geq 0$
, 
 $a\in A^i$
 and
$a\in A^i$
 and 
 $b\in A$
. We denote by
$b\in A$
. We denote by 
 $H^{* }(A):= \operatorname {Ker}(\partial )/\operatorname {Im}(\partial )$
 the cohomology of
$H^{* }(A):= \operatorname {Ker}(\partial )/\operatorname {Im}(\partial )$
 the cohomology of 
 $(A,\partial )$
, and we write
$(A,\partial )$
, and we write 
 $\cup$
 for the multiplication (cup product) on
$\cup$
 for the multiplication (cup product) on 
 $H^{* }(A)$
.
$H^{* }(A)$
.
 We say that 
 $A$
 is formal if it is quasi-isomorphic, as a differential graded ring, to
$A$
 is formal if it is quasi-isomorphic, as a differential graded ring, to 
 $H^{* }(A)$
 with the zero differential.
$H^{* }(A)$
 with the zero differential.
 Let 
 $n\geq 2$
 be an integer and let
$n\geq 2$
 be an integer and let 
 $a_1,\ldots ,a_n\in H^1(A)$
. A defining system for the
$a_1,\ldots ,a_n\in H^1(A)$
. A defining system for the 
 $n$
th order Massey product
$n$
th order Massey product 
 $\langle {a_1,\ldots ,a_n}\rangle$
 is a collection
$\langle {a_1,\ldots ,a_n}\rangle$
 is a collection 
 $M$
 of elements
$M$
 of elements 
 $a_{ij}\in A^1$
, where
$a_{ij}\in A^1$
, where 
 $1\leq i\lt j\leq n+1$
,
$1\leq i\lt j\leq n+1$
, 
 $(i,j)\neq (1,n+1)$
, such that:
$(i,j)\neq (1,n+1)$
, such that:
- 
(1)  $\partial (a_{i,i+1})=0$
 and $\partial (a_{i,i+1})=0$
 and $a_{i,i+1}$
 represents $a_{i,i+1}$
 represents $a_i$
 in $a_i$
 in $H^1(A)$
; and $H^1(A)$
; and
- 
(2)  $\partial (a_{ij})=-{{\sum }}_{l=i+1}^{j-1}a_{il}a_{lj}$
 for all $\partial (a_{ij})=-{{\sum }}_{l=i+1}^{j-1}a_{il}a_{lj}$
 for all $i\lt j-1$
. $i\lt j-1$
.
 It follows from (2) that 
 $-{{\sum }}_{l=2}^na_{1l}a_{l,n+1}$
 is a
$-{{\sum }}_{l=2}^na_{1l}a_{l,n+1}$
 is a 
 $2$
-cocycle: we write
$2$
-cocycle: we write 
 $\langle {a_1,\ldots ,a_n}\rangle _M$
 for its cohomology class in
$\langle {a_1,\ldots ,a_n}\rangle _M$
 for its cohomology class in 
 $H^2(A)$
, called the value of
$H^2(A)$
, called the value of 
 $\langle {a_1,\ldots ,a_n}\rangle$
 corresponding to
$\langle {a_1,\ldots ,a_n}\rangle$
 corresponding to 
 $M$
. By definition, the Massey product of
$M$
. By definition, the Massey product of 
 $a_1,\ldots ,a_n$
 is the subset
$a_1,\ldots ,a_n$
 is the subset 
 $\langle {a_1,\ldots ,a_n}\rangle$
 of
$\langle {a_1,\ldots ,a_n}\rangle$
 of 
 $H^2(A)$
 consisting of the values
$H^2(A)$
 consisting of the values 
 $\langle {a_1,\ldots ,a_n}\rangle _M$
 of all defining systems
$\langle {a_1,\ldots ,a_n}\rangle _M$
 of all defining systems 
 $M$
. We say that the Massey product
$M$
. We say that the Massey product 
 $\langle {a_1,\ldots ,a_n}\rangle$
 is defined if it is non-empty, and that it vanishes if
$\langle {a_1,\ldots ,a_n}\rangle$
 is defined if it is non-empty, and that it vanishes if 
 $0\in \langle {a_1,\ldots ,a_n}\rangle$
.
$0\in \langle {a_1,\ldots ,a_n}\rangle$
.
Lemma 2.3. 
Let 
 $(A,\partial )$
 be a differential graded ring, let
$(A,\partial )$
 be a differential graded ring, let 
 $n\geq 3$
 be an integer and let
$n\geq 3$
 be an integer and let 
 $\alpha _1,\ldots ,\alpha _n$
 be elements of
$\alpha _1,\ldots ,\alpha _n$
 be elements of 
 $H^1(A)$
 satisfying
$H^1(A)$
 satisfying 
 $\alpha _i\cup \alpha _{i+1}=0$
 for all
$\alpha _i\cup \alpha _{i+1}=0$
 for all 
 $1\leq i\leq n-1$
. If
$1\leq i\leq n-1$
. If 
 $A$
 is formal, then
$A$
 is formal, then 
 $\langle {\alpha _1,\ldots ,\alpha _n}\rangle$
 vanishes.
$\langle {\alpha _1,\ldots ,\alpha _n}\rangle$
 vanishes.
Proof. See [Reference Pál and QuickPQ22, Theorem 3.8].
2.3 Dwyer’s theorem
 Let 
 $p$
 be a prime, and let
$p$
 be a prime, and let 
 $U_{n+1}\subset \operatorname {GL}_{n+1}({\mathbb F}_p)$
 be the subgroup of
$U_{n+1}\subset \operatorname {GL}_{n+1}({\mathbb F}_p)$
 be the subgroup of 
 $(n+1)\times (n+1)$
 upper unitriangular matrices. For all
$(n+1)\times (n+1)$
 upper unitriangular matrices. For all 
 $1\leq i\lt j\leq n+1$
, we denote by
$1\leq i\lt j\leq n+1$
, we denote by 
 $e_{ij}\in U_{n+1}$
 the matrix whose non-diagonal entries are all zero except for the entry
$e_{ij}\in U_{n+1}$
 the matrix whose non-diagonal entries are all zero except for the entry 
 $(i,j)$
, which is equal to
$(i,j)$
, which is equal to 
 $1$
. We set
$1$
. We set 
 $\sigma _i:= e_{i,i+1}$
 for all
$\sigma _i:= e_{i,i+1}$
 for all 
 $1\leq i\leq n$
. By [Reference Biss and DasguptaBD01, Theorem 1], the group
$1\leq i\leq n$
. By [Reference Biss and DasguptaBD01, Theorem 1], the group 
 $U_{n+1}$
 admits a presentation with generators the
$U_{n+1}$
 admits a presentation with generators the 
 $\sigma _i$
 and the following relations:
$\sigma _i$
 and the following relations:
 \begin{align} \sigma _i^p=1\quad \text {for all $1\leq i\leq n$,} \\[-32pt] \nonumber \end{align}
\begin{align} \sigma _i^p=1\quad \text {for all $1\leq i\leq n$,} \\[-32pt] \nonumber \end{align}
 \begin{align} [\sigma _i,\sigma _j]=1\quad \text {for all $1\leq i\leq j-2\leq n-2$,} \\[-24pt] \nonumber \end{align}
\begin{align} [\sigma _i,\sigma _j]=1\quad \text {for all $1\leq i\leq j-2\leq n-2$,} \\[-24pt] \nonumber \end{align}
 \begin{align} [\sigma _i,[\sigma _i,\sigma _{i+1}]]=[\sigma _{i+1},[\sigma _i,\sigma _{i+1}]]=1\quad \text {for all $1\leq i\leq n-2$,} \\[-24pt] \nonumber \end{align}
\begin{align} [\sigma _i,[\sigma _i,\sigma _{i+1}]]=[\sigma _{i+1},[\sigma _i,\sigma _{i+1}]]=1\quad \text {for all $1\leq i\leq n-2$,} \\[-24pt] \nonumber \end{align}
 \begin{align} [[\sigma _i,\sigma _{i+1}],[\sigma _{i+1},\sigma _{i+2}]]=1\quad \text {for all $1\leq i\leq n-3$.} \\[6pt] \nonumber \end{align}
\begin{align} [[\sigma _i,\sigma _{i+1}],[\sigma _{i+1},\sigma _{i+2}]]=1\quad \text {for all $1\leq i\leq n-3$.} \\[6pt] \nonumber \end{align}
 The following relations holds in 
 $U_{n+1}$
:
$U_{n+1}$
:
 \begin{equation*} [e_{ij},e_{jk}]=e_{ik}\quad \text {for all $1\leq i\lt j\lt k\leq n+1$.} \end{equation*}
\begin{equation*} [e_{ij},e_{jk}]=e_{ik}\quad \text {for all $1\leq i\lt j\lt k\leq n+1$.} \end{equation*}
By induction, we deduce that
 \begin{equation*}e_{1,n+1}=[\sigma _1,[\sigma _2,\ldots ,[\sigma _{n-2},[\sigma _{n-1},\sigma _n]]\ldots ]].\end{equation*}
\begin{equation*}e_{1,n+1}=[\sigma _1,[\sigma _2,\ldots ,[\sigma _{n-2},[\sigma _{n-1},\sigma _n]]\ldots ]].\end{equation*}
 The center 
 $Z_{n+1}$
 of
$Z_{n+1}$
 of 
 $U_{n+1}$
 is the subgroup generated by
$U_{n+1}$
 is the subgroup generated by 
 $e_{1,n+1}$
. The factor group
$e_{1,n+1}$
. The factor group 
 $\overline {U}_{n+1}:= U_{n+1}/Z_{n+1}$
 may be identified with the group of all
$\overline {U}_{n+1}:= U_{n+1}/Z_{n+1}$
 may be identified with the group of all 
 $(n+1)\times (n+1)$
 upper unitriangular matrices with entry
$(n+1)\times (n+1)$
 upper unitriangular matrices with entry 
 $(1,n+1)$
 omitted. For all
$(1,n+1)$
 omitted. For all 
 $1\leq i\lt j\leq n+1$
, let
$1\leq i\lt j\leq n+1$
, let 
 $\overline {e}_{ij}$
 be the coset of
$\overline {e}_{ij}$
 be the coset of 
 $e_{ij}$
 in
$e_{ij}$
 in 
 $\overline {U}_{n+1}$
, and set
$\overline {U}_{n+1}$
, and set 
 $\overline {\sigma }_i:= \overline {e}_{i,i+1}$
 for all
$\overline {\sigma }_i:= \overline {e}_{i,i+1}$
 for all 
 $1\leq i\leq n$
. Then
$1\leq i\leq n$
. Then 
 $\overline {U}_{n+1}$
 is generated by all the
$\overline {U}_{n+1}$
 is generated by all the 
 $\overline{\sigma}_i$
 modulo the relations
$\overline{\sigma}_i$
 modulo the relations
 \begin{align} \overline {\sigma }_i^p=1\quad \text {for all $1\leq i\leq n$,} \\[-24pt] \nonumber \end{align}
\begin{align} \overline {\sigma }_i^p=1\quad \text {for all $1\leq i\leq n$,} \\[-24pt] \nonumber \end{align}
 \begin{align} [\overline {\sigma }_i,\overline {\sigma }_j]=1\quad \text {for all $1\leq i\leq j-2\leq n-2$,} \\[-24pt] \nonumber \end{align}
\begin{align} [\overline {\sigma }_i,\overline {\sigma }_j]=1\quad \text {for all $1\leq i\leq j-2\leq n-2$,} \\[-24pt] \nonumber \end{align}
 \begin{align} [\overline {\sigma }_i,[\overline {\sigma }_i,\overline {\sigma }_{i+1}]=[\overline {\sigma }_{i+1},[\overline {\sigma }_i,\overline {\sigma }_{i+1}]]=1\quad \text {for all $1\leq i\leq n-2$,} \\[-24pt] \nonumber \end{align}
\begin{align} [\overline {\sigma }_i,[\overline {\sigma }_i,\overline {\sigma }_{i+1}]=[\overline {\sigma }_{i+1},[\overline {\sigma }_i,\overline {\sigma }_{i+1}]]=1\quad \text {for all $1\leq i\leq n-2$,} \\[-24pt] \nonumber \end{align}
 \begin{align} [[\overline {\sigma }_i,\overline {\sigma }_{i+1}],[\overline {\sigma }_{i+1},\overline {\sigma }_{i+2}]]=1\quad \text {for all $1\leq i\leq n-3$,} \\[-24pt] \nonumber \end{align}
\begin{align} [[\overline {\sigma }_i,\overline {\sigma }_{i+1}],[\overline {\sigma }_{i+1},\overline {\sigma }_{i+2}]]=1\quad \text {for all $1\leq i\leq n-3$,} \\[-24pt] \nonumber \end{align}
 \begin{align} [\overline {\sigma }_1,[\overline {\sigma }_2,\ldots ,[\overline {\sigma }_{n-2},[\overline {\sigma }_{n-1},\overline {\sigma }_n]]\ldots ]]=1. \end{align}
\begin{align} [\overline {\sigma }_1,[\overline {\sigma }_2,\ldots ,[\overline {\sigma }_{n-2},[\overline {\sigma }_{n-1},\overline {\sigma }_n]]\ldots ]]=1. \end{align}
 We write 
 $u_{ij}\colon U_{n+1}\to \mathbb Z/p\mathbb Z$
 for the
$u_{ij}\colon U_{n+1}\to \mathbb Z/p\mathbb Z$
 for the 
 $(i,j)$
th coordinate function on
$(i,j)$
th coordinate function on 
 $U_{n+1}$
. Note that
$U_{n+1}$
. Note that 
 $u_{ij}$
 is not a group homomorphism unless
$u_{ij}$
 is not a group homomorphism unless 
 $j=i+1$
. We have a commutative diagram
$j=i+1$
. We have a commutative diagram

 where the row is a central exact sequence and the homomorphism 
 $U_{n+1}\to (\mathbb Z/p\mathbb Z)^n$
 is given by
$U_{n+1}\to (\mathbb Z/p\mathbb Z)^n$
 is given by 
 $(u_{12},u_{23},\ldots , u_{n,n+1})$
. We also let
$(u_{12},u_{23},\ldots , u_{n,n+1})$
. We also let
 \begin{equation*}Q_{n+1}:= \operatorname {Ker}[U_{n+1}\to (\mathbb Z/p\mathbb Z)^n],\quad \overline {Q}_{n+1}:= \operatorname {Ker}[\overline {U}_{n+1}\to (\mathbb Z/p\mathbb Z)^n]=Q_{n+1}/Z_{n+1}.\end{equation*}
\begin{equation*}Q_{n+1}:= \operatorname {Ker}[U_{n+1}\to (\mathbb Z/p\mathbb Z)^n],\quad \overline {Q}_{n+1}:= \operatorname {Ker}[\overline {U}_{n+1}\to (\mathbb Z/p\mathbb Z)^n]=Q_{n+1}/Z_{n+1}.\end{equation*}
Note that 
 $Z_{n+1}\subset Q_{n+1}$
, with equality when
$Z_{n+1}\subset Q_{n+1}$
, with equality when 
 $n=2$
.
$n=2$
.
 Let 
 $G$
 be a profinite group. The complex
$G$
 be a profinite group. The complex 
 $(C^{*}(G,\mathbb Z/p\mathbb Z),\partial )$
 of mod
$(C^{*}(G,\mathbb Z/p\mathbb Z),\partial )$
 of mod 
 $p$
 non-homogeneous continuous cochains of
$p$
 non-homogeneous continuous cochains of 
 $G$
 with the standard cup product is a differential graded ring. Therefore,
$G$
 with the standard cup product is a differential graded ring. Therefore, 
 $H^{* }(G,\mathbb Z/p\mathbb Z)=H^{* }(C^{*}(G,\mathbb Z/p\mathbb Z),\partial )$
 is endowed with Massey products. The following theorem is due to Dwyer [Reference DwyerDwy75].
$H^{* }(G,\mathbb Z/p\mathbb Z)=H^{* }(C^{*}(G,\mathbb Z/p\mathbb Z),\partial )$
 is endowed with Massey products. The following theorem is due to Dwyer [Reference DwyerDwy75].
Theorem 2.4 (Dwyer). Let 
 $p$
 be a prime number, let
$p$
 be a prime number, let 
 $G$
 be a profinite group, let
$G$
 be a profinite group, let 
 $\chi _1,\ldots ,\chi _n\in H^1(G,\mathbb Z/p\mathbb Z)$
 and write
$\chi _1,\ldots ,\chi _n\in H^1(G,\mathbb Z/p\mathbb Z)$
 and write 
 $\chi \colon G\to (\mathbb Z/p\mathbb Z)^n$
 for the continuous homomorphism with components
$\chi \colon G\to (\mathbb Z/p\mathbb Z)^n$
 for the continuous homomorphism with components 
 $(\chi _1,\ldots ,\chi _n)$
. Consider diagram (
2.13
).
$(\chi _1,\ldots ,\chi _n)$
. Consider diagram (
2.13
).
- 
(1) The Massey product  $\langle {\chi _1,\ldots ,\chi _n}\rangle$
 is defined if and only if $\langle {\chi _1,\ldots ,\chi _n}\rangle$
 is defined if and only if $\chi$
 lifts to a continuous homomorphism $\chi$
 lifts to a continuous homomorphism $G\to \overline {U}_{n+1}$
. $G\to \overline {U}_{n+1}$
.
- 
(2) The Massey product  $\langle {\chi _1,\ldots ,\chi _n}\rangle$
 vanishes if and only if $\langle {\chi _1,\ldots ,\chi _n}\rangle$
 vanishes if and only if $\chi$
 lifts to a continuous homomorphism $\chi$
 lifts to a continuous homomorphism $G\to U_{n+1}$
. $G\to U_{n+1}$
.
Proof. See [Reference DwyerDwy75] for Dwyer’s original proof in the setting of abstract groups, and see [Reference EfratEfr14] or [Reference Harpaz and WittenbergHW23, Proposition 2.2] for the statement in the case of profinite groups.
Theorem 2.4 may be rephrased as follows.
Corollary 2.5. 
Let 
 $p$
 be a prime, let
$p$
 be a prime, let 
 $F$
 be a field of characteristic different from
$F$
 be a field of characteristic different from 
 $p$
 and containing a primitive
$p$
 and containing a primitive 
 $p$
th root of unity
$p$
th root of unity 
 $\zeta$
, and let
$\zeta$
, and let 
 $a_1,\ldots ,a_n\in F^\times$
. The Massey product
$a_1,\ldots ,a_n\in F^\times$
. The Massey product 
 $\langle {a_1,\ldots ,a_n}\rangle \subset H^2(F,\mathbb Z/p\mathbb Z)$
 is defined (respectively, vanishes) if and only if there exists a Galois
$\langle {a_1,\ldots ,a_n}\rangle \subset H^2(F,\mathbb Z/p\mathbb Z)$
 is defined (respectively, vanishes) if and only if there exists a Galois 
 $\overline {U}_{n+1}$
-algebra
$\overline {U}_{n+1}$
-algebra 
 $K/F$
 (respectively, a Galois
$K/F$
 (respectively, a Galois 
 $U_{n+1}$
-algebra
$U_{n+1}$
-algebra 
 $L/F$
) such that
$L/F$
) such that 
 $K^{\overline {Q}_{n+1}}\simeq F_{a_1,\ldots ,a_n}$
 (respectively,
$K^{\overline {Q}_{n+1}}\simeq F_{a_1,\ldots ,a_n}$
 (respectively, 
 $L^{{Q}_{n+1}}\simeq F_{a_1,\ldots ,a_n}$
) as
$L^{{Q}_{n+1}}\simeq F_{a_1,\ldots ,a_n}$
) as 
 $(\mathbb Z/p\mathbb Z)^n$
-algebras.
$(\mathbb Z/p\mathbb Z)^n$
-algebras.
We will apply Lemma 2.1 to the cartesian square of groups

 where 
 ${\varphi }_{n+1}$
 (respectively,
${\varphi }_{n+1}$
 (respectively, 
 ${\varphi }'_{n+1}$
) is the restriction homomorphism from
${\varphi }'_{n+1}$
) is the restriction homomorphism from 
 $U_{n+1}$
 or from
$U_{n+1}$
 or from 
 $\overline {U}_{n+1}$
 to the top-left (respectively, bottom-right)
$\overline {U}_{n+1}$
 to the top-left (respectively, bottom-right) 
 $n\times n$
 subsquare
$n\times n$
 subsquare 
 $U_n$
 in
$U_n$
 in 
 $U_{n+1}$
.
$U_{n+1}$
.
 The fact that the square (2.14) is cartesian is proved in [Reference Merkurjev and ScaviaMS22, Proposition 2.7] when 
 $p=2$
. The proof extends to odd
$p=2$
. The proof extends to odd 
 $p$
 without change.
$p$
 without change.
Remark 2.6. The presentations of 
 $U_{n+1}$
 and
$U_{n+1}$
 and 
 $\overline {U}_{n+1}$
 of [Reference Biss and DasguptaBD01] given above will allow us to avoid lengthy calculations in § 3, but they are not essential for our arguments. One could instead use the following classical presentations of
$\overline {U}_{n+1}$
 of [Reference Biss and DasguptaBD01] given above will allow us to avoid lengthy calculations in § 3, but they are not essential for our arguments. One could instead use the following classical presentations of 
 $U_{n+1}$
 and
$U_{n+1}$
 and 
 $\overline {U}_{n+1}$
, which are reminiscent of the Steinberg relations for the Steinberg group of a ring in algebraic K-theory.
$\overline {U}_{n+1}$
, which are reminiscent of the Steinberg relations for the Steinberg group of a ring in algebraic K-theory.
 The group 
 $U_{n+1}$
 admits a presentation with generators
$U_{n+1}$
 admits a presentation with generators 
 $\{e_{ij}:1\leq i\lt j\leq n+1\}$
 and the following relations:
$\{e_{ij}:1\leq i\lt j\leq n+1\}$
 and the following relations:
 \begin{align}e_{ij}^p=1\quad \text {for all $1\leq i\lt j\leq n+1$},\end{align}
\begin{align}e_{ij}^p=1\quad \text {for all $1\leq i\lt j\leq n+1$},\end{align}
 \begin{align}[e_{ij},e_{jk}]=e_{ik}\quad \text {for all $1\leq i\lt j\lt k\leq n+1$},\end{align}
\begin{align}[e_{ij},e_{jk}]=e_{ik}\quad \text {for all $1\leq i\lt j\lt k\leq n+1$},\end{align}
 \begin{align}[e_{ij},e_{kl}]=1\quad \text {for all $1\leq i\lt j\leq n+1$, $1\leq k\lt l\leq n+1$, $i\neq l$, $j\neq k$}.\end{align}
\begin{align}[e_{ij},e_{kl}]=1\quad \text {for all $1\leq i\lt j\leq n+1$, $1\leq k\lt l\leq n+1$, $i\neq l$, $j\neq k$}.\end{align}
This is a particular case of [Reference Abramenko and BrownAB08, Proposition 7.108], where we choose 
 $w$
 to be the longest element of the Weyl group of
$w$
 to be the longest element of the Weyl group of 
 $\operatorname {GL}_{n+1}$
 over
$\operatorname {GL}_{n+1}$
 over 
 ${\mathbb F}_p$
. One obtains a presentation of
${\mathbb F}_p$
. One obtains a presentation of 
 $\overline {U}_{n+1}$
 with generating set
$\overline {U}_{n+1}$
 with generating set 
 $\{\overline {e}_{ij}:1\leq i\lt j\leq n+1\}$
, modulo the relations induced by the above relations for the
$\{\overline {e}_{ij}:1\leq i\lt j\leq n+1\}$
, modulo the relations induced by the above relations for the 
 $e_{ij}$
, together with the relation
$e_{ij}$
, together with the relation 
 $\overline {e}_{1,n+1}=1$
.
$\overline {e}_{1,n+1}=1$
.
3. Massey products and Galois algebras
 In this section, we let 
 $p$
 be a prime number and we let
$p$
 be a prime number and we let 
 $F$
 be a field. With the exception of Proposition 3.6, we assume that
$F$
 be a field. With the exception of Proposition 3.6, we assume that 
 $\operatorname {char}(F)\neq p$
 and that
$\operatorname {char}(F)\neq p$
 and that 
 $F$
 contains a primitive
$F$
 contains a primitive 
 $p$
th root of unity
$p$
th root of unity 
 $\zeta$
.
$\zeta$
.
3.1 Galois 
 $U_3$
-algebras
$U_3$
-algebras
 Let 
 $a,b\in F^\times$
, and suppose that
$a,b\in F^\times$
, and suppose that 
 $(a,b)=0$
 in
$(a,b)=0$
 in 
 $\operatorname {Br}(F)$
. By Lemma 2.2, we may fix
$\operatorname {Br}(F)$
. By Lemma 2.2, we may fix 
 $\alpha \in F_a^\times$
 and
$\alpha \in F_a^\times$
 and 
 $\beta \in F_b^\times$
 such that
$\beta \in F_b^\times$
 such that 
 $N_a(\alpha )=b$
 and
$N_a(\alpha )=b$
 and 
 $N_b(\beta )=a$
.
$N_b(\beta )=a$
.
 We write 
 $(\mathbb Z/p\mathbb Z)^2=\langle {\sigma _a,\sigma _b}\rangle$
, and we view
$(\mathbb Z/p\mathbb Z)^2=\langle {\sigma _a,\sigma _b}\rangle$
, and we view 
 $F_{a,b}$
 as a Galois
$F_{a,b}$
 as a Galois 
 $(\mathbb Z/p\mathbb Z)^2$
-algebra as in § 2.1. The projection
$(\mathbb Z/p\mathbb Z)^2$
-algebra as in § 2.1. The projection 
 $U_3\to \overline {U}_3=(\mathbb Z/p\mathbb Z)^2$
 sends
$U_3\to \overline {U}_3=(\mathbb Z/p\mathbb Z)^2$
 sends 
 $e_{12}\mapsto \sigma _a$
 and
$e_{12}\mapsto \sigma _a$
 and 
 $e_{23}\mapsto \sigma _b$
. We define the following elements of
$e_{23}\mapsto \sigma _b$
. We define the following elements of 
 $U_3$
:
$U_3$
:
 \begin{equation*} \sigma _a:= e_{12},\quad \sigma _b:= e_{23},\quad \tau := e_{13}=[\sigma _a,\sigma _b]. \end{equation*}
\begin{equation*} \sigma _a:= e_{12},\quad \sigma _b:= e_{23},\quad \tau := e_{13}=[\sigma _a,\sigma _b]. \end{equation*}
Suppose we are given 
 $x\in F_a^\times$
 such that
$x\in F_a^\times$
 such that
 \begin{align} (\sigma _a-1)x=\frac {b}{\alpha ^p}. \end{align}
\begin{align} (\sigma _a-1)x=\frac {b}{\alpha ^p}. \end{align}
The étale 
 $F$
-algebra
$F$
-algebra 
 $K:= (F_{a,b})_x$
 has the structure of a Galois
$K:= (F_{a,b})_x$
 has the structure of a Galois 
 $U_3$
-algebra, such that the Galois
$U_3$
-algebra, such that the Galois 
 $(\mathbb Z/p\mathbb Z)^2$
-algebra
$(\mathbb Z/p\mathbb Z)^2$
-algebra 
 $K^{Q_3}$
 is equal to
$K^{Q_3}$
 is equal to 
 $F_{a,b}$
 and
$F_{a,b}$
 and
 \begin{align} (\sigma _a-1)x^{1/p}=\frac {b^{1/p}}{\alpha },\quad (\sigma _b-1)x^{1/p}=1,\quad (\tau -1)x^{1/p}=\zeta ^{-1}. \end{align}
\begin{align} (\sigma _a-1)x^{1/p}=\frac {b^{1/p}}{\alpha },\quad (\sigma _b-1)x^{1/p}=1,\quad (\tau -1)x^{1/p}=\zeta ^{-1}. \end{align}
Similarly, suppose we are given 
 $y\in F_b^\times$
 such that
$y\in F_b^\times$
 such that
 \begin{align} (\sigma _b-1)y=\frac {a}{\beta ^p}. \end{align}
\begin{align} (\sigma _b-1)y=\frac {a}{\beta ^p}. \end{align}
The étale 
 $F$
-algebra
$F$
-algebra 
 $K:= (F_{a,b})_y$
 has the structure of a Galois
$K:= (F_{a,b})_y$
 has the structure of a Galois 
 $U_3$
-algebra, such that the Galois
$U_3$
-algebra, such that the Galois 
 $(\mathbb Z/p\mathbb Z)^2$
-algebra
$(\mathbb Z/p\mathbb Z)^2$
-algebra 
 $K^{Q_3}$
 is equal to
$K^{Q_3}$
 is equal to 
 $F_{a,b}$
 and
$F_{a,b}$
 and
 \begin{align} (\sigma _a-1)y^{1/p}=1,\quad (\sigma _b-1)y^{1/p}=\frac {a^{1/p}}{\beta },\quad (\tau -1)y^{1/p}=\zeta . \end{align}
\begin{align} (\sigma _a-1)y^{1/p}=1,\quad (\sigma _b-1)y^{1/p}=\frac {a^{1/p}}{\beta },\quad (\tau -1)y^{1/p}=\zeta . \end{align}
In (3.2) and (3.4), the relations involving 
 $\tau$
 follows from the first two.
$\tau$
 follows from the first two.
 If 
 $x\in F_a^\times$
 satisfies (3.1), then so does
$x\in F_a^\times$
 satisfies (3.1), then so does 
 $ax$
. We may thus apply (3.2) to
$ax$
. We may thus apply (3.2) to 
 $(F_{a,b})_{ax}$
. Therefore,
$(F_{a,b})_{ax}$
. Therefore, 
 $(F_{a,b})_{ax}$
 has the structure of a Galois
$(F_{a,b})_{ax}$
 has the structure of a Galois 
 $U_3$
-algebra, where
$U_3$
-algebra, where 
 $U_3$
 acts via
$U_3$
 acts via 
 $\overline {U}_3=\operatorname {Gal}(F_{a,b}/F)$
 on
$\overline {U}_3=\operatorname {Gal}(F_{a,b}/F)$
 on 
 $F_{a,b}$
 and
$F_{a,b}$
 and
 \begin{equation*}(\sigma _a-1)(ax)^{1/p}=\frac {b^{1/p}}{\alpha },\quad (\sigma _b-1)(ax)^{1/p}=1,\quad (\tau -1)(ax)^{1/p}=\zeta ^{-1}.\end{equation*}
\begin{equation*}(\sigma _a-1)(ax)^{1/p}=\frac {b^{1/p}}{\alpha },\quad (\sigma _b-1)(ax)^{1/p}=1,\quad (\tau -1)(ax)^{1/p}=\zeta ^{-1}.\end{equation*}
Similarly, if 
 $y\in F_b^\times$
 satisfies (3.3), we may apply (3.4) to
$y\in F_b^\times$
 satisfies (3.3), we may apply (3.4) to 
 $(F_{a,b})_{by}$
. Therefore,
$(F_{a,b})_{by}$
. Therefore, 
 $(F_{a,b})_{by}$
 admits a Galois
$(F_{a,b})_{by}$
 admits a Galois 
 $U_3$
-algebra structure, where
$U_3$
-algebra structure, where 
 $U_3$
 acts via
$U_3$
 acts via 
 $\overline {U}_3=\operatorname {Gal}(F_{a,b}/F)$
 on
$\overline {U}_3=\operatorname {Gal}(F_{a,b}/F)$
 on 
 $F_{a,b}$
 and
$F_{a,b}$
 and
 \begin{equation}(\sigma _a-1)(by)^{1/p}=1,\quad (\sigma _b-1)(by)^{1/p}=\frac {a^{1/p}}{\beta },\quad (\tau -1)(by)^{1/p}=\zeta.\end{equation}
\begin{equation}(\sigma _a-1)(by)^{1/p}=1,\quad (\sigma _b-1)(by)^{1/p}=\frac {a^{1/p}}{\beta },\quad (\tau -1)(by)^{1/p}=\zeta.\end{equation}
Lemma 3.1.
- 
(1) Let  $x\in F_a^\times$
 satisfy (
3.1
), and consider the Galois $x\in F_a^\times$
 satisfy (
3.1
), and consider the Galois $U_3$
-algebras $U_3$
-algebras $(F_{a,b})_x$
 and $(F_{a,b})_x$
 and $(F_{a,b})_{ax}$
 as in (
3.2
). Then $(F_{a,b})_{ax}$
 as in (
3.2
). Then $(F_{a,b})_x\simeq (F_{a,b})_{ax}$
 as Galois $(F_{a,b})_x\simeq (F_{a,b})_{ax}$
 as Galois $U_3$
-algebras. $U_3$
-algebras.
- 
(2) Let  $y\in F_b^\times$
 satisfy (
3.3
), and consider the Galois $y\in F_b^\times$
 satisfy (
3.3
), and consider the Galois $U_3$
-algebras $U_3$
-algebras $(F_{a,b})_y$
 and $(F_{a,b})_y$
 and $(F_{a,b})_{by}$
 as in (
3.4
). Then $(F_{a,b})_{by}$
 as in (
3.4
). Then $(F_{a,b})_y\simeq (F_{a,b})_{by}$
 as Galois $(F_{a,b})_y\simeq (F_{a,b})_{by}$
 as Galois $U_3$
-algebras. $U_3$
-algebras.
Proof.
 (1) The automorphism 
 $\sigma _b\colon F_{a,b}\to F_{a,b}$
 extends to an isomorphism of étale algebras
$\sigma _b\colon F_{a,b}\to F_{a,b}$
 extends to an isomorphism of étale algebras 
 $f\colon (F_{a,b})_{ax}\to (F_{a,b})_x$
 by sending
$f\colon (F_{a,b})_{ax}\to (F_{a,b})_x$
 by sending 
 $(ax)^{1/p}$
 to
$(ax)^{1/p}$
 to 
 $a^{1/p}x^{1/p}$
. The map
$a^{1/p}x^{1/p}$
. The map 
 $f$
 is well defined because
$f$
 is well defined because 
 $f((ax)^{1/p})^p=(a^{1/p}x^{1/p})^p=ax$
. We now show that
$f((ax)^{1/p})^p=(a^{1/p}x^{1/p})^p=ax$
. We now show that 
 $f$
 is
$f$
 is 
 $U_3$
-equivariant. The restriction of
$U_3$
-equivariant. The restriction of 
 $f$
 to
$f$
 to 
 $F_{a,b}$
 is
$F_{a,b}$
 is 
 $U_3$
-equivariant because
$U_3$
-equivariant because 
 $\sigma _a\sigma _b=\sigma _b\sigma _a$
 on
$\sigma _a\sigma _b=\sigma _b\sigma _a$
 on 
 $F_{a,b}$
. We have
$F_{a,b}$
. We have
 \begin{equation*} \sigma _a(f((ax)^{1/p}))=\sigma _a(a^{1/p})\cdot \sigma _a(x^{1/p})=\zeta \cdot a^{1/p}\cdot \frac {b^{1/p}}{\alpha }\cdot x^{1/p}=\frac {\zeta a^{1/p}b^{1/p}x^{1/p}}{\alpha } \end{equation*}
\begin{equation*} \sigma _a(f((ax)^{1/p}))=\sigma _a(a^{1/p})\cdot \sigma _a(x^{1/p})=\zeta \cdot a^{1/p}\cdot \frac {b^{1/p}}{\alpha }\cdot x^{1/p}=\frac {\zeta a^{1/p}b^{1/p}x^{1/p}}{\alpha } \end{equation*}
and
 \begin{equation*}f(\sigma _a((ax)^{1/p}))=f((b^{1/p}/\alpha )\cdot (ax)^{1/p})=\zeta \cdot \frac {b^{1/p}}{\alpha }\cdot a^{1/p}\cdot x^{1/p}=\frac {\zeta a^{1/p}b^{1/p}x^{1/p}}{\alpha }.\end{equation*}
\begin{equation*}f(\sigma _a((ax)^{1/p}))=f((b^{1/p}/\alpha )\cdot (ax)^{1/p})=\zeta \cdot \frac {b^{1/p}}{\alpha }\cdot a^{1/p}\cdot x^{1/p}=\frac {\zeta a^{1/p}b^{1/p}x^{1/p}}{\alpha }.\end{equation*}
Thus, 
 $f$
 is
$f$
 is 
 $\langle {\sigma _a}\rangle$
-equivariant. We also have
$\langle {\sigma _a}\rangle$
-equivariant. We also have
 \begin{equation*}\sigma _b(f((ax)^{1/p}))= \sigma _b(a^{1/p}) \cdot \sigma _b(x^{1/p}) = a^{1/p}\cdot x^{1/p} \end{equation*}
\begin{equation*}\sigma _b(f((ax)^{1/p}))= \sigma _b(a^{1/p}) \cdot \sigma _b(x^{1/p}) = a^{1/p}\cdot x^{1/p} \end{equation*}
and
 \begin{equation*}f(\sigma _b((ax)^{1/p}))=f((ax)^{1/p})=a^{1/p}\cdot x^{1/p}.\end{equation*}
\begin{equation*}f(\sigma _b((ax)^{1/p}))=f((ax)^{1/p})=a^{1/p}\cdot x^{1/p}.\end{equation*}
Thus, 
 $f$
 is
$f$
 is 
 $\langle {\sigma _b}\rangle$
-equivariant. Since
$\langle {\sigma _b}\rangle$
-equivariant. Since 
 $\sigma _a$
 and
$\sigma _a$
 and 
 $\sigma _b$
 generate
$\sigma _b$
 generate 
 $U_3$
, we conclude that
$U_3$
, we conclude that 
 $f$
 is
$f$
 is 
 $U_3$
-equivariant, as desired.
$U_3$
-equivariant, as desired.
(2) The proof is similar to that of (1).
Proposition 3.2. 
Let 
 $a,b\in F^\times$
 be such that
$a,b\in F^\times$
 be such that 
 $(a,b)=0$
 in
$(a,b)=0$
 in 
 $\operatorname {Br}(F)$
, and fix
$\operatorname {Br}(F)$
, and fix 
 $\alpha \in F_a^\times$
 and
$\alpha \in F_a^\times$
 and 
 $\beta \in F_b^\times$
 such that
$\beta \in F_b^\times$
 such that 
 $N_a(\alpha )=b$
 and
$N_a(\alpha )=b$
 and 
 $N_b(\beta )=a$
.
$N_b(\beta )=a$
.
- 
(1) Every Galois  $U_3$
-algebra $U_3$
-algebra $K$
 over $K$
 over $F$
 such that $F$
 such that $K^{Q_3}\simeq F_{a,b}$
 as $K^{Q_3}\simeq F_{a,b}$
 as $(\mathbb Z/p\mathbb Z)^2$
-algebras is of the form $(\mathbb Z/p\mathbb Z)^2$
-algebras is of the form $(F_{a,b})_x$
 for some $(F_{a,b})_x$
 for some $x\in F_a^\times$
 as in (
3.1
), with $x\in F_a^\times$
 as in (
3.1
), with $U_3$
-action given by (
3.2
). $U_3$
-action given by (
3.2
).
- 
(2) Every Galois  $U_3$
-algebra $U_3$
-algebra $K$
 over $K$
 over $F$
 such that $F$
 such that $K^{Q_3}\simeq F_{a,b}$
 as $K^{Q_3}\simeq F_{a,b}$
 as $(\mathbb Z/p\mathbb Z)^2$
-algebras is of the form $(\mathbb Z/p\mathbb Z)^2$
-algebras is of the form $(F_{a,b})_y$
 for some $(F_{a,b})_y$
 for some $y\in F_b^\times$
 as in (
3.3
), with $y\in F_b^\times$
 as in (
3.3
), with $U_3$
-action given by (
3.4
). $U_3$
-action given by (
3.4
).
- 
(3) Let  $(F_{a,b})_x$
 and $(F_{a,b})_x$
 and $(F_{a,b})_y$
 be Galois $(F_{a,b})_y$
 be Galois $U_3$
-algebras as in (
3.2
) and (
3.4
), respectively. The Galois $U_3$
-algebras as in (
3.2
) and (
3.4
), respectively. The Galois $U_3$
-algebras $U_3$
-algebras $(F_{a,b})_x$
 and $(F_{a,b})_x$
 and $(F_{a,b})_y$
 are isomorphic if and only if there exists $(F_{a,b})_y$
 are isomorphic if and only if there exists $w\in F_{a,b}^\times$
 such that $w\in F_{a,b}^\times$
 such that \begin{align*}w^p=xy,\quad (\sigma _a-1)(\sigma _b-1)w=\zeta . \\[-24pt] \end{align*} \begin{align*}w^p=xy,\quad (\sigma _a-1)(\sigma _b-1)w=\zeta . \\[-24pt] \end{align*}
Proof.
 (1) Since 
 $Q_3=\langle {\tau }\rangle \simeq \mathbb Z/p\mathbb Z$
 and
$Q_3=\langle {\tau }\rangle \simeq \mathbb Z/p\mathbb Z$
 and 
 $K^{Q_3}\simeq F_{a,b}$
 as
$K^{Q_3}\simeq F_{a,b}$
 as 
 $(\mathbb Z/p\mathbb Z)^2$
-algebras, we have an isomorphism of étale
$(\mathbb Z/p\mathbb Z)^2$
-algebras, we have an isomorphism of étale 
 $F_{a,b}$
-algebras
$F_{a,b}$
-algebras 
 $K\simeq (F_{a,b})_z$
 for some
$K\simeq (F_{a,b})_z$
 for some 
 $z\in F_{a,b}^\times$
 such that
$z\in F_{a,b}^\times$
 such that 
 $(\tau -1)z^{1/p}=\zeta ^{-1}$
. We may suppose that
$(\tau -1)z^{1/p}=\zeta ^{-1}$
. We may suppose that 
 $K=(F_{a,b})_z$
. As
$K=(F_{a,b})_z$
. As 
 $\tau$
 commutes with
$\tau$
 commutes with 
 $\sigma _b$
,
$\sigma _b$
,
 \begin{equation*} (\tau -1)(\sigma _b-1)z^{1/p}=(\sigma _b-1)(\tau -1)z^{1/p}=(\sigma _b-1)\zeta ^{-1}=1, \end{equation*}
\begin{equation*} (\tau -1)(\sigma _b-1)z^{1/p}=(\sigma _b-1)(\tau -1)z^{1/p}=(\sigma _b-1)\zeta ^{-1}=1, \end{equation*} 
and hence 
 $(\sigma _b-1)z^{1/p}\in F_{a,b}^\times$
. By Hilbert’s Theorem 90 for the extension
$(\sigma _b-1)z^{1/p}\in F_{a,b}^\times$
. By Hilbert’s Theorem 90 for the extension 
 $F_{a,b}/F_a$
, there is
$F_{a,b}/F_a$
, there is 
 $t\in F_{a,b}^\times$
 such that
$t\in F_{a,b}^\times$
 such that 
 $(\sigma _b-1)z^{1/p}=(\sigma _b-1)t$
. Replacing
$(\sigma _b-1)z^{1/p}=(\sigma _b-1)t$
. Replacing 
 $z$
 by
$z$
 by 
 $zt^{-p}$
, we may thus assume that
$zt^{-p}$
, we may thus assume that 
 $(\sigma _b-1)z^{1/p}=1$
. In particular,
$(\sigma _b-1)z^{1/p}=1$
. In particular, 
 $z\in F_a^\times$
. Since
$z\in F_a^\times$
. Since 
 $(\tau -1)z^{1/p}=\zeta ^{-1}$
, we have
$(\tau -1)z^{1/p}=\zeta ^{-1}$
, we have 
 $\sigma _b\sigma _a(z^{1/p})=\zeta \sigma _a\sigma _b(z^{1/p})$
. Thus,
$\sigma _b\sigma _a(z^{1/p})=\zeta \sigma _a\sigma _b(z^{1/p})$
. Thus,
 \begin{equation*}(\sigma _b-1) (\sigma _a-1)z^{1/p} =(\sigma _b\sigma _a-\sigma _a\sigma _b+ (\sigma _a-1)(\sigma _b-1))z^{1/p}=\zeta (\sigma _a-1) (\sigma _b-1)z^{1/p} = \zeta ,\end{equation*}
\begin{equation*}(\sigma _b-1) (\sigma _a-1)z^{1/p} =(\sigma _b\sigma _a-\sigma _a\sigma _b+ (\sigma _a-1)(\sigma _b-1))z^{1/p}=\zeta (\sigma _a-1) (\sigma _b-1)z^{1/p} = \zeta ,\end{equation*} 
and hence 
 $(\sigma _a-1)z^{1/p} = b^{1/p}/\alpha '$
 for some
$(\sigma _a-1)z^{1/p} = b^{1/p}/\alpha '$
 for some 
 $\alpha '\in F_a^\times$
. Moreover,
$\alpha '\in F_a^\times$
. Moreover, 
 $N_a(\alpha '/\alpha )=b/b=1$
, and so by Hilbert’s Theorem 90 there exists
$N_a(\alpha '/\alpha )=b/b=1$
, and so by Hilbert’s Theorem 90 there exists 
 $\theta \in F_a^{\times }$
 such that
$\theta \in F_a^{\times }$
 such that 
 $\alpha '/\alpha =(\sigma _a-1)\theta$
. We define
$\alpha '/\alpha =(\sigma _a-1)\theta$
. We define 
 $x:= z\theta ^p\in F_a^\times$
, and we set
$x:= z\theta ^p\in F_a^\times$
, and we set 
 $x^{1/p}:= z^{1/p}\theta \in (F_{a,b})_z^\times$
. Then
$x^{1/p}:= z^{1/p}\theta \in (F_{a,b})_z^\times$
. Then 
 $K=(F_{a,b})_x$
, where
$K=(F_{a,b})_x$
, where
 \begin{equation*}(\sigma _a-1)x^{1/p}=(\sigma _a-1)z^{1/p}\cdot (\sigma _a-1)\theta =\frac {b^{1/p}}{\alpha '}\cdot \frac {\alpha '}{\alpha }=\frac {b^{1/p}}{\alpha },\end{equation*}
\begin{equation*}(\sigma _a-1)x^{1/p}=(\sigma _a-1)z^{1/p}\cdot (\sigma _a-1)\theta =\frac {b^{1/p}}{\alpha '}\cdot \frac {\alpha '}{\alpha }=\frac {b^{1/p}}{\alpha },\end{equation*}
and 
 $(\sigma _b-1)x^{1/p}=1$
, as desired.
$(\sigma _b-1)x^{1/p}=1$
, as desired.
(2) The proof is analogous to that of (1).
 (3) Suppose we are given an isomorphism of Galois 
 $U_3$
-algebras between
$U_3$
-algebras between 
 $(F_{a,b})_x$
 and
$(F_{a,b})_x$
 and 
 $(F_{a,b})_y$
. Let
$(F_{a,b})_y$
. Let 
 $t\in (F_{a,b})_x$
 be the image of
$t\in (F_{a,b})_x$
 be the image of 
 $y^{1/p}$
 under the isomorphism and set
$y^{1/p}$
 under the isomorphism and set
 \begin{equation*}w':= x^{1/p} t \in (F_{a,b})_x.\end{equation*}
\begin{equation*}w':= x^{1/p} t \in (F_{a,b})_x.\end{equation*} 
Set 
 $y':= t^p$
. We have
$y':= t^p$
. We have 
 $(\tau -1)w'=\zeta ^{-1}\cdot \zeta =1$
, and hence
$(\tau -1)w'=\zeta ^{-1}\cdot \zeta =1$
, and hence 
 $w'\in F_{a,b}^\times$
. We have
$w'\in F_{a,b}^\times$
. We have 
 $(w')^p=xy'$
. Since
$(w')^p=xy'$
. Since 
 $F_b$
 coincides with the
$F_b$
 coincides with the 
 $\langle {\sigma _a,\tau }\rangle$
-invariant subalgebra of
$\langle {\sigma _a,\tau }\rangle$
-invariant subalgebra of 
 $(F_{a,b})_x$
 and
$(F_{a,b})_x$
 and 
 $(F_{a,b})_y$
, the isomorphism
$(F_{a,b})_y$
, the isomorphism 
 $(F_{a,b})_y\to (F_{a,b})_x$
 restricts to an isomorphism of Galois
$(F_{a,b})_y\to (F_{a,b})_x$
 restricts to an isomorphism of Galois 
 $\mathbb Z/p\mathbb Z$
-algebras
$\mathbb Z/p\mathbb Z$
-algebras 
 $F_b \to F_b$
. Since the automorphism group of
$F_b \to F_b$
. Since the automorphism group of 
 $F_b$
 as a Galois
$F_b$
 as a Galois 
 $(\mathbb Z/p\mathbb Z)$
-algebra is
$(\mathbb Z/p\mathbb Z)$
-algebra is 
 $\mathbb Z/p\mathbb Z$
, generated by
$\mathbb Z/p\mathbb Z$
, generated by 
 $\sigma _b$
, this isomorphism
$\sigma _b$
, this isomorphism 
 $F_b \to F_b$
 is equal to
$F_b \to F_b$
 is equal to 
 $\sigma _b^i$
 for some integer
$\sigma _b^i$
 for some integer 
 $i\geq 0$
. Thus,
$i\geq 0$
. Thus, 
 $y'=\sigma _b^i(y)$
. Define
$y'=\sigma _b^i(y)$
. Define
 \begin{equation*}w:= w' a^{-i/p}\mathop {{\prod }}\limits _{j=0}^i\sigma _b^{{\kern1.1pt}j}(\beta )\in F_{a,b}^\times .\end{equation*}
\begin{equation*}w:= w' a^{-i/p}\mathop {{\prod }}\limits _{j=0}^i\sigma _b^{{\kern1.1pt}j}(\beta )\in F_{a,b}^\times .\end{equation*} 
We have 
 $(1-\sigma _b)y=\beta ^p/a$
, and hence
$(1-\sigma _b)y=\beta ^p/a$
, and hence
 \begin{align*}(1-\sigma _b^i)y= \biggl(\mathop {{\sum }}\limits _{j=0}^{i-1}\sigma _b^{{\kern1.1pt}j}(1-\sigma _b) \biggr)y= \biggl(\mathop {{\prod }}\limits _{j=0}^i\sigma _b^{{\kern1.1pt}j}(\beta ^p) \biggr)/a^i=w^p/(w')^p.\end{align*}
\begin{align*}(1-\sigma _b^i)y= \biggl(\mathop {{\sum }}\limits _{j=0}^{i-1}\sigma _b^{{\kern1.1pt}j}(1-\sigma _b) \biggr)y= \biggl(\mathop {{\prod }}\limits _{j=0}^i\sigma _b^{{\kern1.1pt}j}(\beta ^p) \biggr)/a^i=w^p/(w')^p.\end{align*} 
Therefore,
 \begin{align} w^p=(w')^p(1-\sigma _b^i)y=x\sigma _b^i(y)(1-\sigma _b^i)y=xy. \end{align}
\begin{align} w^p=(w')^p(1-\sigma _b^i)y=x\sigma _b^i(y)(1-\sigma _b^i)y=xy. \end{align}
We have 
 $(\sigma _b-1)x^{1/p}=1$
 and
$(\sigma _b-1)x^{1/p}=1$
 and
 \begin{equation*}(\sigma _a-1)(\sigma _b-1)t=(\sigma _a-1)(\sigma _b-1)y^{1/p}=(\sigma _a-1)(a^{1/p}/\beta )=\zeta .\end{equation*}
\begin{equation*}(\sigma _a-1)(\sigma _b-1)t=(\sigma _a-1)(\sigma _b-1)y^{1/p}=(\sigma _a-1)(a^{1/p}/\beta )=\zeta .\end{equation*}
Therefore,
 \begin{equation*}(\sigma _a-1)(\sigma _b-1)w'=(\sigma _a-1)(\sigma _b-1)t=\zeta .\end{equation*}
\begin{equation*}(\sigma _a-1)(\sigma _b-1)w'=(\sigma _a-1)(\sigma _b-1)t=\zeta .\end{equation*} 
Since 
 $(\sigma _a-1)(\sigma _b-1)a^{1/p}=1$
 and
$(\sigma _a-1)(\sigma _b-1)a^{1/p}=1$
 and 
 $(\sigma _a-1)(\sigma _b-1)\beta =1$
, we conclude that
$(\sigma _a-1)(\sigma _b-1)\beta =1$
, we conclude that
 \begin{align} (\sigma _a-1)(\sigma _b-1)w=(\sigma _a-1)(\sigma _b-1)w'=\zeta . \end{align}
\begin{align} (\sigma _a-1)(\sigma _b-1)w=(\sigma _a-1)(\sigma _b-1)w'=\zeta . \end{align}
Putting (3.5) and (3.6) together, we see that 
 $w$
 satisfies the conditions of (3). Conversely, suppose we are given
$w$
 satisfies the conditions of (3). Conversely, suppose we are given 
 $w'\in F_{a,b}^\times$
 such that
$w'\in F_{a,b}^\times$
 such that
 \begin{equation*}xy=(w')^p,\quad (\sigma _a-1)(\sigma _b-1)w'=\zeta .\end{equation*}
\begin{equation*}xy=(w')^p,\quad (\sigma _a-1)(\sigma _b-1)w'=\zeta .\end{equation*} 
Claim 3.3. 
There exists 
 $w\in F_{a,b}^\times$
 such that
$w\in F_{a,b}^\times$
 such that
 \begin{equation*}xy=w^p,\quad (\sigma _a-1)w=\zeta ^{-i}\frac {b^{1/p}}{\alpha },\quad (\sigma _b-1)w=\zeta ^{-j}\frac {a^{1/p}}{\beta }, \end{equation*}
\begin{equation*}xy=w^p,\quad (\sigma _a-1)w=\zeta ^{-i}\frac {b^{1/p}}{\alpha },\quad (\sigma _b-1)w=\zeta ^{-j}\frac {a^{1/p}}{\beta }, \end{equation*}
for some integers 
 $i$
 and
$i$
 and 
 $j$
.
$j$
.
Proof of Claim 3.3
. First, we find 
 $\eta _a\in F_a^\times$
 such that
$\eta _a\in F_a^\times$
 such that
 \begin{align} \eta _a^p=1, \quad (\sigma _a-1)(w'/\eta _a)=\zeta ^{-i}\frac {b^{1/p}}{\alpha }. \end{align}
\begin{align} \eta _a^p=1, \quad (\sigma _a-1)(w'/\eta _a)=\zeta ^{-i}\frac {b^{1/p}}{\alpha }. \end{align}
We have
 \begin{equation*} (\sigma _a - 1) (w')^p=(\sigma _a - 1) x=\frac {b}{\alpha ^p}. \end{equation*}
\begin{equation*} (\sigma _a - 1) (w')^p=(\sigma _a - 1) x=\frac {b}{\alpha ^p}. \end{equation*}
Let
 \begin{equation*}\zeta _a:= (\sigma _a-1)w'\cdot \alpha \cdot b^{-1/p}\in F_{a,b}^\times .\end{equation*}
\begin{equation*}\zeta _a:= (\sigma _a-1)w'\cdot \alpha \cdot b^{-1/p}\in F_{a,b}^\times .\end{equation*}
We have 
 $\zeta _a^p=1$
. Moreover,
$\zeta _a^p=1$
. Moreover, 
 $(\sigma _b-1)\zeta _a=\zeta \cdot 1\cdot \zeta ^{-1}=1$
, that is,
$(\sigma _b-1)\zeta _a=\zeta \cdot 1\cdot \zeta ^{-1}=1$
, that is, 
 $\zeta _a$
 belongs to
$\zeta _a$
 belongs to 
 $F_a^\times$
. If
$F_a^\times$
. If 
 $F_a$
 is a field, this implies that
$F_a$
 is a field, this implies that 
 $\zeta _a=\zeta ^{-i}$
 for some integer
$\zeta _a=\zeta ^{-i}$
 for some integer 
 $i$
, and (3.7) holds for
$i$
, and (3.7) holds for 
 $\eta _a=1$
.
$\eta _a=1$
.
 Suppose that 
 $F_a$
 is not a field. Then
$F_a$
 is not a field. Then 
 $F_a\simeq F^p$
, where
$F_a\simeq F^p$
, where 
 $\sigma _a$
 acts on Fp
 by cyclically permuting the coordinates, that is,
$\sigma _a$
 acts on Fp
 by cyclically permuting the coordinates, that is,
 \begin{equation*}\sigma _a(x_1,x_2,\ldots ,x_p)=(x_2,\ldots ,x_p, x_1).\end{equation*}
\begin{equation*}\sigma _a(x_1,x_2,\ldots ,x_p)=(x_2,\ldots ,x_p, x_1).\end{equation*}
We have 
 $\zeta _a=(\zeta _1,\ldots ,\zeta _p)$
 in
$\zeta _a=(\zeta _1,\ldots ,\zeta _p)$
 in 
 $F_a=F^p$
, where
$F_a=F^p$
, where 
 $\zeta _i\in F^\times$
 is a
$\zeta _i\in F^\times$
 is a 
 $p$
th root of unity for all
$p$
th root of unity for all 
 $i$
. We have
$i$
. We have 
 $N_a(\zeta _a)=N_a(\alpha )/b=1$
, and so
$N_a(\zeta _a)=N_a(\alpha )/b=1$
, and so 
 $\zeta _1\cdots \zeta _p=1$
. Inductively, define
$\zeta _1\cdots \zeta _p=1$
. Inductively, define 
 $\eta _1:= 1$
 and
$\eta _1:= 1$
 and 
 $\eta _{i+1}:= \zeta _i\eta _i$
 for all
$\eta _{i+1}:= \zeta _i\eta _i$
 for all 
 $i=1,\ldots ,p-1$
. Then
$i=1,\ldots ,p-1$
. Then
 \begin{equation*}\eta _1/\eta _p=(\eta _1/\eta _2)\cdot (\eta _2/\eta _3)\cdots (\eta _{p-1}/\eta _p)=\zeta _1^{-1}\zeta _2^{-1}\cdots \zeta _{p-1}^{-1}=\zeta _p.\end{equation*}
\begin{equation*}\eta _1/\eta _p=(\eta _1/\eta _2)\cdot (\eta _2/\eta _3)\cdots (\eta _{p-1}/\eta _p)=\zeta _1^{-1}\zeta _2^{-1}\cdots \zeta _{p-1}^{-1}=\zeta _p.\end{equation*}
Therefore, the element 
 $\eta _a:= (\eta _1,\ldots ,\eta _p)\in F^p=F_a$
 satisfies
$\eta _a:= (\eta _1,\ldots ,\eta _p)\in F^p=F_a$
 satisfies 
 $\eta _a^p=1$
 and
$\eta _a^p=1$
 and
 \begin{equation*}(\sigma _a-1)\eta _a=(\eta _2/\eta _1,\ldots ,\eta _p/\eta _{p-1},\eta _1/\eta _p)=(\zeta _1,\ldots ,\zeta _{p-1},\zeta _p)=\zeta _a.\end{equation*}
\begin{equation*}(\sigma _a-1)\eta _a=(\eta _2/\eta _1,\ldots ,\eta _p/\eta _{p-1},\eta _1/\eta _p)=(\zeta _1,\ldots ,\zeta _{p-1},\zeta _p)=\zeta _a.\end{equation*}
Thus,
 \begin{equation*}\eta _a^p=1,\quad (\sigma _a-1)(w'/\eta _a)=(\sigma _a-1)w'\cdot \zeta _a^{-1}=\frac {b^{1/p}}{\alpha },\end{equation*}
\begin{equation*}\eta _a^p=1,\quad (\sigma _a-1)(w'/\eta _a)=(\sigma _a-1)w'\cdot \zeta _a^{-1}=\frac {b^{1/p}}{\alpha },\end{equation*}
All in all, independent of whether 
 $F_a$
 is a field or not, we have found
$F_a$
 is a field or not, we have found 
 $\eta _a$
 satisfying (3.7).
$\eta _a$
 satisfying (3.7).
 Similarly, we construct 
 $\eta _b\in F_b^\times$
 such that
$\eta _b\in F_b^\times$
 such that
 \begin{align} \eta _b^p=1,\quad (\sigma _b-1)(w'/\eta _b)=\zeta ^{-j}\frac {a^{1/p}}{\beta }, \end{align}
\begin{align} \eta _b^p=1,\quad (\sigma _b-1)(w'/\eta _b)=\zeta ^{-j}\frac {a^{1/p}}{\beta }, \end{align}
for some integer 
 $j$
. Set
$j$
. Set 
 $w:= w'/(\eta _a\eta _b)\in F_{a,b}^\times$
. Putting together (3.7) and (3.8), we deduce that
$w:= w'/(\eta _a\eta _b)\in F_{a,b}^\times$
. Putting together (3.7) and (3.8), we deduce that 
 $w$
 satisfies the conclusion of Claim 3.3.
$w$
 satisfies the conclusion of Claim 3.3.
 Let 
 $w\in F_{a,b}^\times$
 be as in Claim 3.3. By Lemma 3.1(1), applied
$w\in F_{a,b}^\times$
 be as in Claim 3.3. By Lemma 3.1(1), applied 
 $i$
 times, the Galois
$i$
 times, the Galois 
 $U_3$
-algebra
$U_3$
-algebra 
 $(F_{a,b})_x$
 is isomorphic to
$(F_{a,b})_x$
 is isomorphic to 
 $(F_{a,b})_{a^ix}$
, where
$(F_{a,b})_{a^ix}$
, where
 \begin{equation*}(\sigma _a-1)(a^ix)^{1/p}=\frac {b^{1/p}}{\alpha },\quad (\sigma _b-1)(a^ix)^{1/p}=1.\end{equation*}
\begin{equation*}(\sigma _a-1)(a^ix)^{1/p}=\frac {b^{1/p}}{\alpha },\quad (\sigma _b-1)(a^ix)^{1/p}=1.\end{equation*}
By Lemma 3.1(2), applied 
 $j$
 times, the Galois
$j$
 times, the Galois 
 $U_3$
-algebra
$U_3$
-algebra 
 $(F_{a,b})_y$
 is isomorphic to
$(F_{a,b})_y$
 is isomorphic to 
 $(F_{a,b})_{b^{{\kern1.1pt}j}y}$
, where
$(F_{a,b})_{b^{{\kern1.1pt}j}y}$
, where
 \begin{equation*}(\sigma _a-1)(b^{{\kern1.1pt}j}y)^{1/p}=1,\quad (\sigma _b-1)(b^{{\kern1.1pt}j}y)^{1/p}=\frac {a^{1/p}}{\beta }.\end{equation*}
\begin{equation*}(\sigma _a-1)(b^{{\kern1.1pt}j}y)^{1/p}=1,\quad (\sigma _b-1)(b^{{\kern1.1pt}j}y)^{1/p}=\frac {a^{1/p}}{\beta }.\end{equation*}
Thus, it suffices to construct an isomorphism of 
 $U_3$
-algebras
$U_3$
-algebras 
 $(F_{a,b})_{a^ix}\simeq (F_{a,b})_{b^{{\kern1.1pt}j}y}$
. Let
$(F_{a,b})_{a^ix}\simeq (F_{a,b})_{b^{{\kern1.1pt}j}y}$
. Let
 \begin{equation*}\tilde {w}:= wa^{i/p}b^{{\kern1.1pt}j/p}\in F_{a,b}^\times ,\end{equation*}
\begin{equation*}\tilde {w}:= wa^{i/p}b^{{\kern1.1pt}j/p}\in F_{a,b}^\times ,\end{equation*}
so that
 \begin{equation*}(\sigma _a-1)\tilde {w}=\frac {b^{1/p}}{\alpha },\quad (\sigma _b-1)\tilde {w}=\frac {a^{1/p}}{\beta }.\end{equation*}
\begin{equation*}(\sigma _a-1)\tilde {w}=\frac {b^{1/p}}{\alpha },\quad (\sigma _b-1)\tilde {w}=\frac {a^{1/p}}{\beta }.\end{equation*}
Let 
 $f\colon (F_{a,b})_{a^ix}\to (F_{a,b})_{b^{{\kern1.1pt}j}y}$
 be the isomorphism of étale algebras which is the identity on
$f\colon (F_{a,b})_{a^ix}\to (F_{a,b})_{b^{{\kern1.1pt}j}y}$
 be the isomorphism of étale algebras which is the identity on 
 $F_{a,b}$
 and sends
$F_{a,b}$
 and sends 
 $(a^ix)^{1/p}$
 to
$(a^ix)^{1/p}$
 to 
 $\tilde {w}/(b^{{\kern1.1pt}j}y)^{1/p}$
. Note that
$\tilde {w}/(b^{{\kern1.1pt}j}y)^{1/p}$
. Note that 
 $f$
 is well defined because
$f$
 is well defined because
 \begin{equation*}(\tilde {w})^p=wa^ib^{{\kern1.1pt}j}=(a^ix)(b^{{\kern1.1pt}j}y).\end{equation*}
\begin{equation*}(\tilde {w})^p=wa^ib^{{\kern1.1pt}j}=(a^ix)(b^{{\kern1.1pt}j}y).\end{equation*}
Moreover,
 \begin{align*}(\sigma _a-1) (\tilde {w}/(b^{{\kern1.1pt}j}y)^{1/p})=\frac {b^{1/p}}{\alpha }=(\sigma _a-1)(a^ix)^{1/p}, \\[-24pt] \end{align*}
\begin{align*}(\sigma _a-1) (\tilde {w}/(b^{{\kern1.1pt}j}y)^{1/p})=\frac {b^{1/p}}{\alpha }=(\sigma _a-1)(a^ix)^{1/p}, \\[-24pt] \end{align*}
 \begin{equation*}(\sigma _b-1) (\tilde {w}/(b^{{\kern1.1pt}j}y)^{1/p})=\frac {a^{1/p}}{\beta }\cdot \frac {\beta }{a^{1/p}}=1=(\sigma _b-1)(a^ix)^{1/p},\end{equation*}
\begin{equation*}(\sigma _b-1) (\tilde {w}/(b^{{\kern1.1pt}j}y)^{1/p})=\frac {a^{1/p}}{\beta }\cdot \frac {\beta }{a^{1/p}}=1=(\sigma _b-1)(a^ix)^{1/p},\end{equation*}
and hence 
 $f$
 is
$f$
 is 
 $U_3$
-equivariant.
$U_3$
-equivariant.
3.2 Galois 
 $\overline {U}_4$
-algebras
$\overline {U}_4$
-algebras
 Let 
 $a,b,c\in F^\times$
 be such that
$a,b,c\in F^\times$
 be such that 
 $(a,b)=(b,c)=0$
 in
$(a,b)=(b,c)=0$
 in 
 $\operatorname {Br}(F)$
. By Lemma 2.2, we may fix
$\operatorname {Br}(F)$
. By Lemma 2.2, we may fix 
 $\alpha \in F_a^\times$
 and
$\alpha \in F_a^\times$
 and 
 $\gamma \in F_c^\times$
 such that
$\gamma \in F_c^\times$
 such that 
 $N_a(\alpha )=N_c(\gamma )=b$
. We have
$N_a(\alpha )=N_c(\gamma )=b$
. We have 
 $\operatorname {Gal}(F_{a,b,c}/F)=(\mathbb Z/p\mathbb Z)^3=\langle {\sigma _a,\sigma _b,\sigma _c}\rangle$
. The projection map
$\operatorname {Gal}(F_{a,b,c}/F)=(\mathbb Z/p\mathbb Z)^3=\langle {\sigma _a,\sigma _b,\sigma _c}\rangle$
. The projection map 
 $\overline {U}_4\to (\mathbb Z/p\mathbb Z)^3$
 is given by
$\overline {U}_4\to (\mathbb Z/p\mathbb Z)^3$
 is given by 
 $\overline {e}_{12}\mapsto \sigma _a$
,
$\overline {e}_{12}\mapsto \sigma _a$
, 
 $\overline {e}_{23}\mapsto \sigma _b$
,
$\overline {e}_{23}\mapsto \sigma _b$
, 
 $\overline {e}_{34}\mapsto \sigma _c$
. Its kernel
$\overline {e}_{34}\mapsto \sigma _c$
. Its kernel 
 $\overline {Q}_4\subset \overline {U}_4$
 is isomorphic to
$\overline {Q}_4\subset \overline {U}_4$
 is isomorphic to 
 $(\mathbb Z/p\mathbb Z)^2$
, generated by
$(\mathbb Z/p\mathbb Z)^2$
, generated by 
 $\overline {e}_{13}$
 and
$\overline {e}_{13}$
 and 
 $\overline {e}_{24}$
. We define the following elements of
$\overline {e}_{24}$
. We define the following elements of 
 $\overline {U}_4$
:
$\overline {U}_4$
:
 \begin{equation*}\sigma _a:= \overline {e}_{12},\quad \sigma _b:= \overline {e}_{23},\quad \sigma _c:= \overline {e}_{34},\quad \tau _{ab}:= \overline {e}_{13},\quad \tau _{bc}:= \overline {e}_{24}.\end{equation*}
\begin{equation*}\sigma _a:= \overline {e}_{12},\quad \sigma _b:= \overline {e}_{23},\quad \sigma _c:= \overline {e}_{34},\quad \tau _{ab}:= \overline {e}_{13},\quad \tau _{bc}:= \overline {e}_{24}.\end{equation*}
Let 
 $x\in F_a^\times$
 and
$x\in F_a^\times$
 and 
 $x'\in F_c^\times$
 be such that
$x'\in F_c^\times$
 be such that
 \begin{align} (\sigma _a-1)x=\frac {b}{\alpha ^p},\quad (\sigma _c-1)x'=\frac {b}{\gamma ^p}, \end{align}
\begin{align} (\sigma _a-1)x=\frac {b}{\alpha ^p},\quad (\sigma _c-1)x'=\frac {b}{\gamma ^p}, \end{align}
and consider the Galois 
 $\overline {U}_4$
-algebra
$\overline {U}_4$
-algebra 
 $K:= (F_{a,b,c})_{x,x'}$
, where
$K:= (F_{a,b,c})_{x,x'}$
, where 
 $\overline {U}_4$
 acts on
$\overline {U}_4$
 acts on 
 $F_{a,b,c}$
 via the surjection onto
$F_{a,b,c}$
 via the surjection onto 
 $\operatorname {Gal}(F_{a,b,c}/F)$
, and
$\operatorname {Gal}(F_{a,b,c}/F)$
, and
 \begin{align} (\sigma _a-1)x^{1/p}=\frac {b^{1/p}}{\alpha },\quad (\sigma _b-1)x^{1/p}=1,\quad (\sigma _c-1)x^{1/p}=1, \\[-24pt] \nonumber \end{align}
\begin{align} (\sigma _a-1)x^{1/p}=\frac {b^{1/p}}{\alpha },\quad (\sigma _b-1)x^{1/p}=1,\quad (\sigma _c-1)x^{1/p}=1, \\[-24pt] \nonumber \end{align}
 \begin{align} (\tau _{ab}-1)x^{1/p}=\zeta ^{-1},\quad (\tau _{bc}-1)x^{1/p}=1, \\[-24pt] \nonumber \end{align}
\begin{align} (\tau _{ab}-1)x^{1/p}=\zeta ^{-1},\quad (\tau _{bc}-1)x^{1/p}=1, \\[-24pt] \nonumber \end{align}
 \begin{align} (\sigma _a-1)(x')^{1/p}=1,\quad (\sigma _b-1)(x')^{1/p}=1,\quad (\sigma _c-1)(x')^{1/p}=\frac {b^{1/p}}{\gamma }, \end{align}
\begin{align} (\sigma _a-1)(x')^{1/p}=1,\quad (\sigma _b-1)(x')^{1/p}=1,\quad (\sigma _c-1)(x')^{1/p}=\frac {b^{1/p}}{\gamma }, \end{align}
 \begin{align} (\tau _{ab}-1)(x')^{1/p}=1,\quad (\tau _{bc}-1)(x')^{1/p}=\zeta . \\[6pt] \nonumber \end{align}
\begin{align} (\tau _{ab}-1)(x')^{1/p}=1,\quad (\tau _{bc}-1)(x')^{1/p}=\zeta . \\[6pt] \nonumber \end{align}
Note that (3.11) follows from (3.10) and (3.13) follows from (3.12). We leave to the reader to check that the relations (2.8)–(2.12) are satisfied.
Proposition 3.4. 
Let 
 $a,b,c\in F^\times$
 be such that
$a,b,c\in F^\times$
 be such that 
 $(a,b)=(b,c)=0$
 in
$(a,b)=(b,c)=0$
 in 
 $\operatorname {Br}(F)$
. Fix
$\operatorname {Br}(F)$
. Fix 
 $\alpha \in F_a^\times$
 and
$\alpha \in F_a^\times$
 and 
 $\gamma \in F_c^\times$
 such that
$\gamma \in F_c^\times$
 such that 
 $N_a(\alpha )=N_c(\gamma )=b$
. Let
$N_a(\alpha )=N_c(\gamma )=b$
. Let 
 $K$
 be a Galois
$K$
 be a Galois 
 $\overline {U}_4$
-algebra such that
$\overline {U}_4$
-algebra such that 
 $K^{\overline {Q}_4}\simeq F_{a,b,c}$
 as
$K^{\overline {Q}_4}\simeq F_{a,b,c}$
 as 
 $(\mathbb Z/p\mathbb Z)^3$
-algebras. Then there exist
$(\mathbb Z/p\mathbb Z)^3$
-algebras. Then there exist 
 $x\in F_a^\times$
 and
$x\in F_a^\times$
 and 
 $x'\in F_c^\times$
 such that
$x'\in F_c^\times$
 such that 
 $K\simeq (F_{a,b,c})_{x,x'}$
 as Galois
$K\simeq (F_{a,b,c})_{x,x'}$
 as Galois 
 $\overline {U}_4$
-algebras, where
$\overline {U}_4$
-algebras, where 
 $\overline {U}_4$
 acts on
$\overline {U}_4$
 acts on 
 $(F_{a,b,c})_{x,x'}$
 by (
3.10
)–(
3.13
).
$(F_{a,b,c})_{x,x'}$
 by (
3.10
)–(
3.13
).
Proof.
 Let 
 $H$
 (respectively,
$H$
 (respectively, 
 $H'$
) be the subgroup of
$H'$
) be the subgroup of 
 $\overline {U}_4$
 generated by
$\overline {U}_4$
 generated by 
 $\sigma _c$
 and
$\sigma _c$
 and 
 $\tau _{bc}$
 (respectively,
$\tau _{bc}$
 (respectively, 
 $\sigma _a$
 and
$\sigma _a$
 and 
 $\tau _{ab}$
), and let
$\tau _{ab}$
), and let 
 $S$
 be the subgroup of
$S$
 be the subgroup of 
 $\overline {U}_4$
 generated by
$\overline {U}_4$
 generated by 
 $H$
 and
$H$
 and 
 $H'$
. Note that
$H'$
. Note that 
 $K^H$
 is a Galois
$K^H$
 is a Galois 
 $U_3$
-algebra over
$U_3$
-algebra over 
 $F$
 such that
$F$
 such that 
 $(K^H)^{Q_3}\simeq F_{a,b}$
 as
$(K^H)^{Q_3}\simeq F_{a,b}$
 as 
 $(\mathbb Z/p\mathbb Z)^2$
-algebras and
$(\mathbb Z/p\mathbb Z)^2$
-algebras and 
 $K^S\simeq F_b$
 as
$K^S\simeq F_b$
 as 
 $(\mathbb Z/p\mathbb Z)$
-algebras. Thus, by Proposition 3.2(1), there exists
$(\mathbb Z/p\mathbb Z)$
-algebras. Thus, by Proposition 3.2(1), there exists 
 $x\in F_a^\times$
 such that
$x\in F_a^\times$
 such that 
 $K^H\simeq (F_{a,b})_x$
 as Galois
$K^H\simeq (F_{a,b})_x$
 as Galois 
 $U_3$
-algebras. Similarly, by Proposition 3.2(2), there exists
$U_3$
-algebras. Similarly, by Proposition 3.2(2), there exists 
 $x'\in F_c^\times$
 such that
$x'\in F_c^\times$
 such that 
 $K^{H'}\simeq (F_{b,c})_{x'}$
 as Galois
$K^{H'}\simeq (F_{b,c})_{x'}$
 as Galois 
 $U_3$
-algebras. Therefore,
$U_3$
-algebras. Therefore, 
 $x$
 satisfies (3.10) and
$x$
 satisfies (3.10) and 
 $x'$
 satisfies (3.12). We apply Lemma 2.1(2) to (2.14). We obtain the isomorphisms of
$x'$
 satisfies (3.12). We apply Lemma 2.1(2) to (2.14). We obtain the isomorphisms of 
 $\overline {U}_4$
-algebras
$\overline {U}_4$
-algebras
 \begin{equation*}K\simeq K^H\otimes _{K^S}K^{H'}\simeq (F_{a,b,c})_{x,x'},\end{equation*}
\begin{equation*}K\simeq K^H\otimes _{K^S}K^{H'}\simeq (F_{a,b,c})_{x,x'},\end{equation*}
where 
 $(F_{a,b,c})_{x,x'}$
 is the
$(F_{a,b,c})_{x,x'}$
 is the 
 $\overline {U}_4$
-algebra determined by (3.10) and (3.12).
$\overline {U}_4$
-algebra determined by (3.10) and (3.12).
3.3 Galois 
 $U_4$
-algebras
$U_4$
-algebras
 Let 
 $a,b,c\in F^\times$
, and suppose that
$a,b,c\in F^\times$
, and suppose that 
 $(a,b)=(b,c)=0$
 in
$(a,b)=(b,c)=0$
 in 
 $\operatorname {Br}(F)$
. We write
$\operatorname {Br}(F)$
. We write 
 $(\mathbb Z/p\mathbb Z)^3=\langle {\sigma _a,\sigma _b,\sigma _c}\rangle$
 and view
$(\mathbb Z/p\mathbb Z)^3=\langle {\sigma _a,\sigma _b,\sigma _c}\rangle$
 and view 
 $F_{a,b,c}$
 as a Galois
$F_{a,b,c}$
 as a Galois 
 $(\mathbb Z/p\mathbb Z)^3$
-algebra over
$(\mathbb Z/p\mathbb Z)^3$
-algebra over 
 $F$
, as in § 2.1. The quotient map
$F$
, as in § 2.1. The quotient map 
 $U_4\to (\mathbb Z/p\mathbb Z)^3$
 is given by
$U_4\to (\mathbb Z/p\mathbb Z)^3$
 is given by 
 $e_{12}\mapsto \sigma _a$
,
$e_{12}\mapsto \sigma _a$
, 
 $e_{23}\mapsto \sigma _b$
 and
$e_{23}\mapsto \sigma _b$
 and 
 $e_{34}\mapsto \sigma _c$
. The kernel
$e_{34}\mapsto \sigma _c$
. The kernel 
 $Q_4$
 of this homomorphism is generated by
$Q_4$
 of this homomorphism is generated by 
 $e_{13}$
,
$e_{13}$
, 
 $e_{24}$
 and
$e_{24}$
 and 
 $e_{14}$
 and is isomorphic to
$e_{14}$
 and is isomorphic to 
 $(\mathbb Z/p\mathbb Z)^3$
. We define the following elements of
$(\mathbb Z/p\mathbb Z)^3$
. We define the following elements of 
 $U_4$
:
$U_4$
:
 \begin{align*} \sigma _a:= e_{12},\quad \sigma _b:= e_{23},\quad \sigma _c:= e_{34}, \\[-24pt] \end{align*}
\begin{align*} \sigma _a:= e_{12},\quad \sigma _b:= e_{23},\quad \sigma _c:= e_{34}, \\[-24pt] \end{align*}
 \begin{equation*} \tau _{ab}:= e_{13}=[\sigma _a,\sigma _b],\quad \tau _{bc}:= e_{24}=[\sigma _b,\sigma _c],\quad \rho := e_{14}=[\sigma _a,\tau _{bc}]=[\tau _{ab},\sigma _c]. \end{equation*}
\begin{equation*} \tau _{ab}:= e_{13}=[\sigma _a,\sigma _b],\quad \tau _{bc}:= e_{24}=[\sigma _b,\sigma _c],\quad \rho := e_{14}=[\sigma _a,\tau _{bc}]=[\tau _{ab},\sigma _c]. \end{equation*}
Proposition 3.5. 
Let 
 $a,b,c\in F^\times$
 be such that
$a,b,c\in F^\times$
 be such that 
 $(a,b)=(b,c)=0$
 in
$(a,b)=(b,c)=0$
 in 
 $\operatorname {Br}(F)$
. Let
$\operatorname {Br}(F)$
. Let 
 $\alpha \in F_a^\times$
 and
$\alpha \in F_a^\times$
 and 
 $\gamma \in F_c^\times$
 be such that
$\gamma \in F_c^\times$
 be such that 
 $N_a(\alpha )=b$
 and
$N_a(\alpha )=b$
 and 
 $N_c(\gamma )=b$
. Let
$N_c(\gamma )=b$
. Let 
 $K$
 be a Galois
$K$
 be a Galois 
 $\overline {U}_4$
-algebra such that
$\overline {U}_4$
-algebra such that 
 $K^{\overline {Q}_4}\simeq F_{a,b,c}$
 as
$K^{\overline {Q}_4}\simeq F_{a,b,c}$
 as 
 $(\mathbb Z/p\mathbb Z)^3$
-algebras.
$(\mathbb Z/p\mathbb Z)^3$
-algebras.
 
There exists a Galois 
 $U_4$
-algebra
$U_4$
-algebra 
 $L$
 over
$L$
 over 
 $F$
 such that
$F$
 such that 
 $L^{Z_4}\simeq K$
 as
$L^{Z_4}\simeq K$
 as 
 $\overline {U}_4$
-algebras if and only if there exist
$\overline {U}_4$
-algebras if and only if there exist 
 $u,u'\in F_{a,c}^\times$
 such that
$u,u'\in F_{a,c}^\times$
 such that
 \begin{equation*}\alpha \cdot (\sigma _a-1)u=\gamma \cdot (\sigma _c-1)u'\end{equation*}
\begin{equation*}\alpha \cdot (\sigma _a-1)u=\gamma \cdot (\sigma _c-1)u'\end{equation*}
and such that 
 $K$
 is isomorphic to the Galois
$K$
 is isomorphic to the Galois 
 $\overline {U}_4$
-algebra
$\overline {U}_4$
-algebra 
 $(F_{a,b,c})_{x,x'}$
 determined by (
3.10
)–(
3.13
), where
$(F_{a,b,c})_{x,x'}$
 determined by (
3.10
)–(
3.13
), where 
 $x=N_c(u)\in F_a^\times$
 and
$x=N_c(u)\in F_a^\times$
 and 
 $x'=N_a(u')\in F_c^\times$
.
$x'=N_a(u')\in F_c^\times$
.
Proof.
 Suppose that 
 $K=(F_{a,b,c})_{x,x'}$
, with
$K=(F_{a,b,c})_{x,x'}$
, with 
 $\overline {U}_4$
-action determined by (3.10)–(3.13). Let
$\overline {U}_4$
-action determined by (3.10)–(3.13). Let 
 $L$
 be a Galois
$L$
 be a Galois 
 $U_4$
-algebra over
$U_4$
-algebra over 
 $F$
 such that
$F$
 such that 
 $L^{Z_4}=K$
, and let
$L^{Z_4}=K$
, and let 
 $y\in K^\times$
 be such that
$y\in K^\times$
 be such that 
 $L=K_y$
.
$L=K_y$
.
 We have 
 $\operatorname {Gal}(L/F_{a,b,c})=Q_4=\langle {\tau _{ab}, \tau _{bc},\rho }\rangle \simeq (\mathbb Z/p\mathbb Z)^3$
, and hence one may choose
$\operatorname {Gal}(L/F_{a,b,c})=Q_4=\langle {\tau _{ab}, \tau _{bc},\rho }\rangle \simeq (\mathbb Z/p\mathbb Z)^3$
, and hence one may choose 
 $y$
 in
$y$
 in 
 $F_{a,b,c}^\times$
 and such that
$F_{a,b,c}^\times$
 and such that
 \begin{equation*} (\tau _{ab}-1)y^{1/p}=1,\quad (\tau _{bc}-1)y^{1/p}=1,\quad (\rho -1)y^{1/p}=\zeta ^{-1}. \end{equation*}
\begin{equation*} (\tau _{ab}-1)y^{1/p}=1,\quad (\tau _{bc}-1)y^{1/p}=1,\quad (\rho -1)y^{1/p}=\zeta ^{-1}. \end{equation*}
The element 
 $\sigma _b$
 commutes with
$\sigma _b$
 commutes with 
 $\tau _{ab}, \tau _{bc}$
 and
$\tau _{ab}, \tau _{bc}$
 and 
 $\rho$
. Hence,
$\rho$
. Hence,
 \begin{equation*} \tau _{ab}(\sigma _b-1)(y^{1/p})=(\sigma _b-1)\tau _{ab}(y^{1/p})=(\sigma _b-1)(y^{1/p}). \end{equation*}
\begin{equation*} \tau _{ab}(\sigma _b-1)(y^{1/p})=(\sigma _b-1)\tau _{ab}(y^{1/p})=(\sigma _b-1)(y^{1/p}). \end{equation*}
Similarly,
 \begin{equation*} \tau _{bc}(\sigma _b-1)(y^{1/p})=(\sigma _b-1)(y^{1/p}) \end{equation*}
\begin{equation*} \tau _{bc}(\sigma _b-1)(y^{1/p})=(\sigma _b-1)(y^{1/p}) \end{equation*}
and
 \begin{equation*}\rho (\sigma _b-1)(y^{1/p})=(\sigma _b-1)(\zeta \cdot y^{1/p})=(\sigma _b-1)(y^{1/p}).\end{equation*}
\begin{equation*}\rho (\sigma _b-1)(y^{1/p})=(\sigma _b-1)(\zeta \cdot y^{1/p})=(\sigma _b-1)(y^{1/p}).\end{equation*}
It follows that 
 $(\sigma _b-1)(y^{1/p})\in F_{a,b,c}^\times$
. By Hilbert’s Theorem 90, applied to
$(\sigma _b-1)(y^{1/p})\in F_{a,b,c}^\times$
. By Hilbert’s Theorem 90, applied to 
 $F_{a,b,c}/F_{a,c}$
, there is
$F_{a,b,c}/F_{a,c}$
, there is 
 $q\in F_{a,b,c}^\times$
 such that
$q\in F_{a,b,c}^\times$
 such that 
 $(\sigma _b-1)(y^{1/p})=(\sigma _b-1)q$
. Replacing
$(\sigma _b-1)(y^{1/p})=(\sigma _b-1)q$
. Replacing 
 $y$
 by
$y$
 by 
 $y/ q^p$
, we may assume that
$y/ q^p$
, we may assume that 
 $\sigma _b(y^{1/p})=y^{1/p}$
. In particular,
$\sigma _b(y^{1/p})=y^{1/p}$
. In particular, 
 $y\in F_{a,c}^\times$
. We have
$y\in F_{a,c}^\times$
. We have
 \begin{align*} \rho (\sigma _a-1) (y^{1/p})= &\ (\sigma _a -1)\rho (y^{1/p})=(\sigma _a -1)(\zeta ^{-1}\cdot y^{1/p})=(\sigma _a -1)(y^{1/p}), \\ \sigma _b(\sigma _a-1) (y^{1/p})= &\ (\sigma _a\sigma _b{\tau _{ab}}^{-1} -\sigma _b)(y^{1/p})=(\sigma _a-1) (y^{1/p}),\\ \tau _{ab}(\sigma _a-1) (y^{1/p})= &\ (\sigma _a -1)\tau _{ab}(y^{1/p})=(\sigma _a -1)(y^{1/p}),\\ \tau _{bc}(\sigma _a-1) (y^{1/p})= &\ (\rho ^{-1} \sigma _a -1)\tau _{bc}(y^{1/p})=( \sigma _a \rho ^{-1} -1)(y^{1/p})=\zeta \cdot (\sigma _a-1) (y^{1/p}). \end{align*}
\begin{align*} \rho (\sigma _a-1) (y^{1/p})= &\ (\sigma _a -1)\rho (y^{1/p})=(\sigma _a -1)(\zeta ^{-1}\cdot y^{1/p})=(\sigma _a -1)(y^{1/p}), \\ \sigma _b(\sigma _a-1) (y^{1/p})= &\ (\sigma _a\sigma _b{\tau _{ab}}^{-1} -\sigma _b)(y^{1/p})=(\sigma _a-1) (y^{1/p}),\\ \tau _{ab}(\sigma _a-1) (y^{1/p})= &\ (\sigma _a -1)\tau _{ab}(y^{1/p})=(\sigma _a -1)(y^{1/p}),\\ \tau _{bc}(\sigma _a-1) (y^{1/p})= &\ (\rho ^{-1} \sigma _a -1)\tau _{bc}(y^{1/p})=( \sigma _a \rho ^{-1} -1)(y^{1/p})=\zeta \cdot (\sigma _a-1) (y^{1/p}). \end{align*}
By (3.12)–(3.13), analogous identities are satisfied by 
 $(x')^{1/p}$
, that is,
$(x')^{1/p}$
, that is,
 \begin{equation*}(\rho -1)(x')^{1/p}=(\sigma _b-1)(x')^{1/p}=(\tau _{ab}-1)(x')^{1/p}=1,\quad (\tau _{bc}-1)(x')^{1/p}=\zeta .\end{equation*}
\begin{equation*}(\rho -1)(x')^{1/p}=(\sigma _b-1)(x')^{1/p}=(\tau _{ab}-1)(x')^{1/p}=1,\quad (\tau _{bc}-1)(x')^{1/p}=\zeta .\end{equation*}
Therefore,
 \begin{equation*} (\sigma _a-1) (y^{1/p})=\frac {(x')^{1/p}}{u'}, \end{equation*}
\begin{equation*} (\sigma _a-1) (y^{1/p})=\frac {(x')^{1/p}}{u'}, \end{equation*}
for some 
 $u'\in F_{a,c}^\times$
. In particular,
$u'\in F_{a,c}^\times$
. In particular, 
 $x'= N_a(u')$
. A similar computation shows that
$x'= N_a(u')$
. A similar computation shows that
 \begin{equation*} (\sigma _c-1) (y^{1/p})=\frac {x^{1/p}}{u}, \end{equation*}
\begin{equation*} (\sigma _c-1) (y^{1/p})=\frac {x^{1/p}}{u}, \end{equation*}
for some 
 $u\in F_{a,c}^\times$
. In particular,
$u\in F_{a,c}^\times$
. In particular, 
 $x= N_c(u)$
. In addition,
$x= N_c(u)$
. In addition,
 \begin{equation*} \frac {b^{1/p}}{\alpha }=(\sigma _a-1) (x^{1/p})=(\sigma _a-1) [u\cdot (\sigma _c-1)(y^{1/p})], \end{equation*}
\begin{equation*} \frac {b^{1/p}}{\alpha }=(\sigma _a-1) (x^{1/p})=(\sigma _a-1) [u\cdot (\sigma _c-1)(y^{1/p})], \end{equation*}
 \begin{equation*} \frac {b^{1/p}}{\gamma }=(\sigma _c-1) ((x')^{1/p})=(\sigma _c-1) [u'\cdot (\sigma _a-1)(y^{1/p})]. \end{equation*}
\begin{equation*} \frac {b^{1/p}}{\gamma }=(\sigma _c-1) ((x')^{1/p})=(\sigma _c-1) [u'\cdot (\sigma _a-1)(y^{1/p})]. \end{equation*}
Therefore,
 \begin{equation*}\alpha \cdot (\sigma _a-1)u=\gamma \cdot (\sigma _c-1)u'.\end{equation*}
\begin{equation*}\alpha \cdot (\sigma _a-1)u=\gamma \cdot (\sigma _c-1)u'.\end{equation*}
 Conversely, suppose we are given 
 $u,u'\in F_{a,c}^\times$
 such that
$u,u'\in F_{a,c}^\times$
 such that
 \begin{equation*}\alpha \cdot (\sigma _a-1)u=\gamma \cdot (\sigma _c-1)u',\quad x=N_c(u),\quad x'=N_a(u').\end{equation*}
\begin{equation*}\alpha \cdot (\sigma _a-1)u=\gamma \cdot (\sigma _c-1)u',\quad x=N_c(u),\quad x'=N_a(u').\end{equation*}
Then
 \begin{equation*} (\sigma _a-1)x=(\sigma _a-1)N_c(u)=N_c(\sigma _a-1)u=N_c\Big(\frac {\gamma }{\alpha }\Big)=\frac {b}{\alpha ^p}, \end{equation*}
\begin{equation*} (\sigma _a-1)x=(\sigma _a-1)N_c(u)=N_c(\sigma _a-1)u=N_c\Big(\frac {\gamma }{\alpha }\Big)=\frac {b}{\alpha ^p}, \end{equation*}
 \begin{equation*} (\sigma _c-1)x'=(\sigma _c-1)N_a(u')=N_a(\sigma _c-1)u'=N_a\Big(\frac {\alpha }{\gamma }\Big)=\frac {b}{\gamma ^p}. \end{equation*}
\begin{equation*} (\sigma _c-1)x'=(\sigma _c-1)N_a(u')=N_a(\sigma _c-1)u'=N_a\Big(\frac {\alpha }{\gamma }\Big)=\frac {b}{\gamma ^p}. \end{equation*}
We have
 \begin{equation*} N_c \Big(\frac {x}{u^p}\Big)=\frac {N_c(x)}{N_c(u^p)}=\frac {x^p}{x^p}=1, \end{equation*}
\begin{equation*} N_c \Big(\frac {x}{u^p}\Big)=\frac {N_c(x)}{N_c(u^p)}=\frac {x^p}{x^p}=1, \end{equation*}
 \begin{equation*} N_a \Big(\frac {x'}{(u')^p}\Big)=\frac {N_a(x')}{N_a((u')^p)}=\frac {(x')^p}{(x')^p}=1, \end{equation*}
\begin{equation*} N_a \Big(\frac {x'}{(u')^p}\Big)=\frac {N_a(x')}{N_a((u')^p)}=\frac {(x')^p}{(x')^p}=1, \end{equation*}
 \begin{equation*} (\sigma _a-1)\Big(\frac {x}{u^p}\Big)=\frac {b}{\alpha ^p\cdot (\sigma _a-1)u^p}= \frac {b}{\gamma ^p\cdot (\sigma _c-1)(u')^p}=(\sigma _c-1)\Big(\frac {x'}{(u')^p}\Big). \end{equation*}
\begin{equation*} (\sigma _a-1)\Big(\frac {x}{u^p}\Big)=\frac {b}{\alpha ^p\cdot (\sigma _a-1)u^p}= \frac {b}{\gamma ^p\cdot (\sigma _c-1)(u')^p}=(\sigma _c-1)\Big(\frac {x'}{(u')^p}\Big). \end{equation*}
By Hilbert’s Theorem 90 applied to 
 $F_{a,c}/F$
, there is
$F_{a,c}/F$
, there is 
 $y\in F_{a,c}^\times$
 such that
$y\in F_{a,c}^\times$
 such that
 \begin{equation*} (\sigma _a-1)y=\frac {x'}{(u')^p} \quad \text {and}\quad (\sigma _c-1)y=\frac {x}{u^p}. \end{equation*}
\begin{equation*} (\sigma _a-1)y=\frac {x'}{(u')^p} \quad \text {and}\quad (\sigma _c-1)y=\frac {x}{u^p}. \end{equation*}
We consider the étale 
 $F$
-algebra
$F$
-algebra 
 $L:= K_y$
 and make it into a Galois
$L:= K_y$
 and make it into a Galois 
 $U_4$
-algebra such that
$U_4$
-algebra such that 
 $L^{Z_4}=K$
. It suffices to describe the
$L^{Z_4}=K$
. It suffices to describe the 
 $U_4$
-action on
$U_4$
-action on 
 $y^{1/p}$
. We set
$y^{1/p}$
. We set
 \begin{equation*} (\sigma _a-1)(y^{1/p})=\frac {(x')^{1/p}}{u'},\quad (\sigma _b-1)(y^{1/p})=1,\quad (\sigma _c-1)(y^{1/p})=\frac {x^{1/p}}{u}. \end{equation*}
\begin{equation*} (\sigma _a-1)(y^{1/p})=\frac {(x')^{1/p}}{u'},\quad (\sigma _b-1)(y^{1/p})=1,\quad (\sigma _c-1)(y^{1/p})=\frac {x^{1/p}}{u}. \end{equation*}
One can check that this definition is compatible with relations (2.4)–(2.7), and hence that it makes 
 $L$
 into a Galois
$L$
 into a Galois 
 $U_4$
-algebra such that
$U_4$
-algebra such that 
 $L^{Z_4}=K$
.
$L^{Z_4}=K$
.
 We use Proposition 3.5 to give an alternative proof for the Massey vanishing conjecture for 
 $n=3$
 and arbitrary
$n=3$
 and arbitrary 
 $p$
.
$p$
.
Proposition 3.6. 
Let 
 $p$
 be a prime, let
$p$
 be a prime, let 
 $F$
 be a field and let
$F$
 be a field and let 
 $\chi _1,\chi _2,\chi _3\in H^1(F,\mathbb Z/p\mathbb Z)$
. The following are equivalent.
$\chi _1,\chi _2,\chi _3\in H^1(F,\mathbb Z/p\mathbb Z)$
. The following are equivalent.
- 
(1) We have  $\chi _1\cup \chi _2=\chi _2 \cup \chi _3=0$
 in $\chi _1\cup \chi _2=\chi _2 \cup \chi _3=0$
 in $H^2(F,\mathbb Z/p\mathbb Z)$
. $H^2(F,\mathbb Z/p\mathbb Z)$
.
- 
(2) The Massey product  $\langle {\chi _1,\chi _2,\chi _3}\rangle \subset H^2(F,\mathbb Z/p\mathbb Z)$
 is defined. $\langle {\chi _1,\chi _2,\chi _3}\rangle \subset H^2(F,\mathbb Z/p\mathbb Z)$
 is defined.
- 
(3) The Massey product  $\langle {\chi _1,\chi _2,\chi _3}\rangle \subset H^2(F,\mathbb Z/p\mathbb Z)$
 vanishes. $\langle {\chi _1,\chi _2,\chi _3}\rangle \subset H^2(F,\mathbb Z/p\mathbb Z)$
 vanishes.
Proof.
 It is clear that (3) implies (2) and that (2) implies (1). We now prove that (1) implies (3). The first part of the proof is a reduction to the case when 
 $\operatorname {char}(F)\neq p$
 and
$\operatorname {char}(F)\neq p$
 and 
 $F$
 contains a primitive
$F$
 contains a primitive 
 $p$
th root of unity, and it follows [Reference Mináč and TânMT16, Proposition 4.14].
$p$
th root of unity, and it follows [Reference Mináč and TânMT16, Proposition 4.14].
Consider the short exact sequence
 \begin{align} 1\to Q_4\to U_4\to (\mathbb Z/p\mathbb Z)^3\to 1, \end{align}
\begin{align} 1\to Q_4\to U_4\to (\mathbb Z/p\mathbb Z)^3\to 1, \end{align}
where the map 
 $U_4\to (\mathbb Z/p\mathbb Z)^3$
 comes from (2.13). Recall from the paragraph preceding Proposition 3.5 that the group
$U_4\to (\mathbb Z/p\mathbb Z)^3$
 comes from (2.13). Recall from the paragraph preceding Proposition 3.5 that the group 
 $Q_4$
 is abelian. Therefore, the group homomorphism
$Q_4$
 is abelian. Therefore, the group homomorphism 
 $\chi :=(\chi _1,\chi _2,\chi _3)\colon \Gamma _F \to (\mathbb Z/p\mathbb Z)^3$
 induces a pullback map
$\chi :=(\chi _1,\chi _2,\chi _3)\colon \Gamma _F \to (\mathbb Z/p\mathbb Z)^3$
 induces a pullback map
 \begin{equation*}H^2((\mathbb Z/p\mathbb Z)^3, Q_4)\to H^2(F, Q_4).\end{equation*}
\begin{equation*}H^2((\mathbb Z/p\mathbb Z)^3, Q_4)\to H^2(F, Q_4).\end{equation*}
We let 
 $A\in H^2(F, Q_4)$
 be the image of the class of (3.14) under this map. By Theorem 2.4, for every finite extension
$A\in H^2(F, Q_4)$
 be the image of the class of (3.14) under this map. By Theorem 2.4, for every finite extension 
 $F'/F$
 the Massey product
$F'/F$
 the Massey product 
 $\langle {\chi _1,\chi _2,\chi _3}\rangle$
 vanishes over
$\langle {\chi _1,\chi _2,\chi _3}\rangle$
 vanishes over 
 $F'$
 if and only if the restriction of
$F'$
 if and only if the restriction of 
 $\chi$
 to
$\chi$
 to 
 $\Gamma _{F'}$
 lifts to
$\Gamma _{F'}$
 lifts to 
 $U_4$
, and this happens if and only if
$U_4$
, and this happens if and only if 
 $A$
 restricts to zero in
$A$
 restricts to zero in 
 $H^2(F', Q_4)$
.
$H^2(F', Q_4)$
.
 When 
 $\operatorname {char}(F)=p$
, we have
$\operatorname {char}(F)=p$
, we have 
 $\operatorname {cd}_p(F)\leq 1$
 by [Reference Serre and IonSer97, § 2.2, Proposition 3]. Therefore,
$\operatorname {cd}_p(F)\leq 1$
 by [Reference Serre and IonSer97, § 2.2, Proposition 3]. Therefore, 
 $H^2(F,Q_4)=0$
 and hence
$H^2(F,Q_4)=0$
 and hence 
 $A=0$
. Thus, (1) implies (3) when
$A=0$
. Thus, (1) implies (3) when 
 $\operatorname {char}(F)=p$
.
$\operatorname {char}(F)=p$
.
 Suppose that 
 $\operatorname {char}(F)\neq p$
. There exists an extension
$\operatorname {char}(F)\neq p$
. There exists an extension 
 $F'/F$
 of prime-to-
$F'/F$
 of prime-to-
 $p$
 degree such that
$p$
 degree such that 
 $F'$
 contains a primitive
$F'$
 contains a primitive 
 $p$
th root of
$p$
th root of 
 $1$
. If (1) implies (3) for
$1$
. If (1) implies (3) for 
 $F'$
, then
$F'$
, then 
 $A$
 restricts to zero in
$A$
 restricts to zero in 
 $H^2(F', Q_4)$
. By a restriction-corestriction argument, we deduce that
$H^2(F', Q_4)$
. By a restriction-corestriction argument, we deduce that 
 $A$
 vanishes, that is, (1) implies (3) for
$A$
 vanishes, that is, (1) implies (3) for 
 $F$
. Thus, we may assume that
$F$
. Thus, we may assume that 
 $F$
 contains a primitive
$F$
 contains a primitive 
 $p$
th root of unity
$p$
th root of unity 
 $\zeta$
.
$\zeta$
.
 Let 
 $a,b,c\in F^\times$
 be such that
$a,b,c\in F^\times$
 be such that 
 $\chi _a=\chi _1$
,
$\chi _a=\chi _1$
, 
 $\chi _b=\chi _2$
 and
$\chi _b=\chi _2$
 and 
 $\chi _c=\chi _3$
 in
$\chi _c=\chi _3$
 in 
 $H^1(F,\mathbb Z/p\mathbb Z)$
. Since
$H^1(F,\mathbb Z/p\mathbb Z)$
. Since 
 $(a,b)=(b,c)=0$
 in
$(a,b)=(b,c)=0$
 in 
 $\operatorname {Br}(F)$
, there exists
$\operatorname {Br}(F)$
, there exists 
 $\alpha \in F_a^\times$
 and
$\alpha \in F_a^\times$
 and 
 $\gamma \in F_c^\times$
 such that
$\gamma \in F_c^\times$
 such that 
 $N_a(\alpha )=N_c(\gamma )=b$
. Since
$N_a(\alpha )=N_c(\gamma )=b$
. Since 
 $N_{ac}(\gamma /\alpha )=N_c(\gamma )/N_a(\alpha )=1$
 in
$N_{ac}(\gamma /\alpha )=N_c(\gamma )/N_a(\alpha )=1$
 in 
 $F_{ac}^\times$
, by Hilbert’s Theorem 90 there exists
$F_{ac}^\times$
, by Hilbert’s Theorem 90 there exists 
 $t\in F_{a,c}^\times$
 such that
$t\in F_{a,c}^\times$
 such that 
 $\gamma /\alpha = (\sigma _a\sigma _c - 1)t$
. Define
$\gamma /\alpha = (\sigma _a\sigma _c - 1)t$
. Define 
 $u,u'\in F_{a,c}^\times$
 by
$u,u'\in F_{a,c}^\times$
 by 
 $u:= \sigma _c(t)$
 and
$u:= \sigma _c(t)$
 and 
 $u':= t^{-1}$
. Then
$u':= t^{-1}$
. Then
 \begin{equation*}\alpha \cdot (\sigma _a-1)u=\alpha \cdot (\sigma _a\sigma _c-\sigma _c)t= \alpha \cdot (\sigma _a\sigma _c-1)t\cdot (\sigma _c-1)t^{-1}=\gamma \cdot (\sigma _c-1)u'.\end{equation*}
\begin{equation*}\alpha \cdot (\sigma _a-1)u=\alpha \cdot (\sigma _a\sigma _c-\sigma _c)t= \alpha \cdot (\sigma _a\sigma _c-1)t\cdot (\sigma _c-1)t^{-1}=\gamma \cdot (\sigma _c-1)u'.\end{equation*}
Let 
 $x:= N_c(u)\in F_a^\times$
 and
$x:= N_c(u)\in F_a^\times$
 and 
 $x':= N_a(u')\in F_c^\times$
. Since
$x':= N_a(u')\in F_c^\times$
. Since 
 $\sigma _a\sigma _c=\sigma _c\sigma _a$
 on
$\sigma _a\sigma _c=\sigma _c\sigma _a$
 on 
 $F_{a,c}^\times$
,
$F_{a,c}^\times$
, 
 \begin{equation*}(\sigma _a-1)x=N_c((\sigma _a-1)u)=N_c((\sigma _c-1)u'\cdot (\gamma /\alpha ))=N_c(\gamma )/N_c(\alpha )=b/\alpha ^p.\end{equation*}
\begin{equation*}(\sigma _a-1)x=N_c((\sigma _a-1)u)=N_c((\sigma _c-1)u'\cdot (\gamma /\alpha ))=N_c(\gamma )/N_c(\alpha )=b/\alpha ^p.\end{equation*}
Similarly, 
 $(\sigma _c-1)x'=b/\gamma ^p$
. Therefore,
$(\sigma _c-1)x'=b/\gamma ^p$
. Therefore, 
 $x,x'$
 satisfy (3.9). Let
$x,x'$
 satisfy (3.9). Let 
 $K:= (F_{a,b,c})_{x,x'}$
 be the Galois
$K:= (F_{a,b,c})_{x,x'}$
 be the Galois 
 $\overline {U}_4$
-algebra over
$\overline {U}_4$
-algebra over 
 $F$
, with the
$F$
, with the 
 $\overline {U}_4$
-action given by (3.10)–(3.13). By Proposition 3.5, there exists a Galois
$\overline {U}_4$
-action given by (3.10)–(3.13). By Proposition 3.5, there exists a Galois 
 $U_4$
-algebra
$U_4$
-algebra 
 $L$
 over
$L$
 over 
 $F$
 such that
$F$
 such that 
 $L^{Z_4}\simeq (F_{a,b,c})_{x,x'}$
 as
$L^{Z_4}\simeq (F_{a,b,c})_{x,x'}$
 as 
 $\overline {U}_4$
-algebras. In particular,
$\overline {U}_4$
-algebras. In particular, 
 $L^{Q_4}\simeq F_{a,b,c}$
 as
$L^{Q_4}\simeq F_{a,b,c}$
 as 
 $(\mathbb Z/p\mathbb Z)^3$
-algebras. By Corollary 2.5, we conclude that
$(\mathbb Z/p\mathbb Z)^3$
-algebras. By Corollary 2.5, we conclude that 
 $\langle {a,b,c}\rangle$
 vanishes.
$\langle {a,b,c}\rangle$
 vanishes.
3.4 Galois 
 $\overline {U}_5$
-algebras
$\overline {U}_5$
-algebras
 Let 
 $a,b,c,d\in F^\times$
. We write
$a,b,c,d\in F^\times$
. We write 
 $(\mathbb Z/p\mathbb Z)^4=\langle {\sigma _a,\sigma _b,\sigma _c,\sigma _d}\rangle$
 and regard
$(\mathbb Z/p\mathbb Z)^4=\langle {\sigma _a,\sigma _b,\sigma _c,\sigma _d}\rangle$
 and regard 
 $F_{a,b,c,d}$
 as a Galois
$F_{a,b,c,d}$
 as a Galois 
 $(\mathbb Z/p\mathbb Z)^4$
-algebra over
$(\mathbb Z/p\mathbb Z)^4$
-algebra over 
 $F$
 as in § 2.1.
$F$
 as in § 2.1.
Proposition 3.7. 
Let 
 $a,b,c,d \in F^\times$
 be such that
$a,b,c,d \in F^\times$
 be such that 
 $(a,b)=(b,c)=(c,d)=0$
 in
$(a,b)=(b,c)=(c,d)=0$
 in 
 $\operatorname {Br}(F)$
. The Massey product
$\operatorname {Br}(F)$
. The Massey product 
 $\langle {a,b,c,d}\rangle$
 is defined if and only if there exist
$\langle {a,b,c,d}\rangle$
 is defined if and only if there exist 
 $u\in F_{a,c}^\times$
,
$u\in F_{a,c}^\times$
, 
 $v\in F_{b,d}^\times$
 and
$v\in F_{b,d}^\times$
 and 
 $w\in F_{b,c}^\times$
 such that
$w\in F_{b,c}^\times$
 such that
 \begin{equation*}N_a(u)\cdot N_d(v) = w^p,\quad (\sigma _b - 1)(\sigma _c - 1)w = \zeta .\end{equation*}
\begin{equation*}N_a(u)\cdot N_d(v) = w^p,\quad (\sigma _b - 1)(\sigma _c - 1)w = \zeta .\end{equation*}
Proof.
 Denote by 
 $U_4^+$
 and
$U_4^+$
 and 
 $U_4^-$
 the top-left and bottom-right
$U_4^-$
 the top-left and bottom-right 
 $4\times 4$
 corners of
$4\times 4$
 corners of 
 $U_5$
, respectively, and let
$U_5$
, respectively, and let 
 $S:= U_4^+\cap U_4^-$
 be the middle subgroup
$S:= U_4^+\cap U_4^-$
 be the middle subgroup 
 $U_3$
. Let
$U_3$
. Let 
 $Q_4^+$
 and
$Q_4^+$
 and 
 $Q_4^-$
 be the kernels of the maps
$Q_4^-$
 be the kernels of the maps 
 $U_4^+\to (\mathbb Z/p\mathbb Z)^3$
 and
$U_4^+\to (\mathbb Z/p\mathbb Z)^3$
 and 
 $U_4^-\to (\mathbb Z/p\mathbb Z)^3$
, respectively, and let
$U_4^-\to (\mathbb Z/p\mathbb Z)^3$
, respectively, and let 
 $P_4^+$
 and
$P_4^+$
 and 
 $P_4^-$
 be the kernels of the maps
$P_4^-$
 be the kernels of the maps 
 $U_4^+\to U_3$
 and
$U_4^+\to U_3$
 and 
 $U_4^-\to U_3$
, respectively.
$U_4^-\to U_3$
, respectively.
 Suppose 
 $\langle a,b,c,d \rangle$
 is defined. By Corollary 2.5, there exists a
$\langle a,b,c,d \rangle$
 is defined. By Corollary 2.5, there exists a 
 $\overline {U}_5$
-algebra
$\overline {U}_5$
-algebra 
 $L$
 such that
$L$
 such that 
 $L^{\overline {Q}_5}\simeq F_{a,b,c,d}$
 as
$L^{\overline {Q}_5}\simeq F_{a,b,c,d}$
 as 
 $(\mathbb Z/p\mathbb Z)^4$
-algebras. Using Lemma 2.2, we fix
$(\mathbb Z/p\mathbb Z)^4$
-algebras. Using Lemma 2.2, we fix 
 $\alpha \in F_a^\times$
 and
$\alpha \in F_a^\times$
 and 
 $\gamma \in F_c^\times$
 such that
$\gamma \in F_c^\times$
 such that 
 $N_a(\alpha )=b$
 and
$N_a(\alpha )=b$
 and 
 $N_c(\gamma )=b$
. By Proposition 3.5, there exist
$N_c(\gamma )=b$
. By Proposition 3.5, there exist 
 $u,u'\in F_{a,c}^\times$
 such that, letting
$u,u'\in F_{a,c}^\times$
 such that, letting 
 $x':= N_c(u')$
 and
$x':= N_c(u')$
 and 
 $x:= N_a(u)$
, the
$x:= N_a(u)$
, the 
 $\overline {U}_4^+$
-algebra
$\overline {U}_4^+$
-algebra 
 $K_1$
 induced by
$K_1$
 induced by 
 $L$
 is isomorphic to the
$L$
 is isomorphic to the 
 $\overline {U}_4^+$
-algebra
$\overline {U}_4^+$
-algebra 
 $(F_{a,b,c})_{x',x}$
, where
$(F_{a,b,c})_{x',x}$
, where 
 $\overline {U}_4^+$
 acts via (3.10)–(3.13), and where the roles of
$\overline {U}_4^+$
 acts via (3.10)–(3.13), and where the roles of 
 $x$
 and
$x$
 and 
 $x'$
 have been switched.
$x'$
 have been switched.
 Similarly, there exist 
 $v,v'\in F_{b,d}^\times$
 such that, letting
$v,v'\in F_{b,d}^\times$
 such that, letting 
 $z:= N_d(v)$
 and
$z:= N_d(v)$
 and 
 $z':= N_b(v')$
, the
$z':= N_b(v')$
, the 
 $\overline {U}^-_4$
-algebra
$\overline {U}^-_4$
-algebra 
 $K_2$
 induced by
$K_2$
 induced by 
 $L$
 is isomorphic to
$L$
 is isomorphic to 
 $(F_{b,c,d})_{z,z'}$
. Since the
$(F_{b,c,d})_{z,z'}$
. Since the 
 $U_3$
-algebras
$U_3$
-algebras 
 $(K_1)^{P_4^+}$
 and
$(K_1)^{P_4^+}$
 and 
 $(K_2)^{P_4^-}$
 are equal, by Proposition 3.2(3) there exists
$(K_2)^{P_4^-}$
 are equal, by Proposition 3.2(3) there exists 
 $w\in F_{b,c}^\times$
 such that
$w\in F_{b,c}^\times$
 such that
 \begin{equation*} N_a(u)\cdot N_d(v) = xz = w^p,\quad (\sigma _b -1)(\sigma _c -1)w = \zeta . \end{equation*}
\begin{equation*} N_a(u)\cdot N_d(v) = xz = w^p,\quad (\sigma _b -1)(\sigma _c -1)w = \zeta . \end{equation*}
Conversely, let 
 $u\in F_{a,c}^\times$
,
$u\in F_{a,c}^\times$
, 
 $v\in F_{b,d}^\times$
, and
$v\in F_{b,d}^\times$
, and 
 $w\in F_{b,c}^\times$
 be such that
$w\in F_{b,c}^\times$
 be such that
 \begin{equation*} N_a(u)\cdot N_d(v) = w^p,\quad (\sigma _b -1)(\sigma _c -1)w = \zeta .\end{equation*}
\begin{equation*} N_a(u)\cdot N_d(v) = w^p,\quad (\sigma _b -1)(\sigma _c -1)w = \zeta .\end{equation*}
By Lemma 2.2, there exist 
 $\alpha \in F_a^\times$
 and
$\alpha \in F_a^\times$
 and 
 $\delta \in F_d^\times$
 such that
$\delta \in F_d^\times$
 such that 
 $N_a(\alpha )=b$
 and
$N_a(\alpha )=b$
 and 
 $N_d(\delta )=c$
. We may write
$N_d(\delta )=c$
. We may write
 \begin{equation*} (\sigma _b - 1)w = \frac {c^{1/p}}{\beta },\quad (\sigma _c - 1)w = \frac {b^{1/p}}{\gamma }, \end{equation*}
\begin{equation*} (\sigma _b - 1)w = \frac {c^{1/p}}{\beta },\quad (\sigma _c - 1)w = \frac {b^{1/p}}{\gamma }, \end{equation*}
for some 
 $\beta \in F_b^\times$
 and
$\beta \in F_b^\times$
 and 
 $\gamma \in F_c^\times$
. We have
$\gamma \in F_c^\times$
. We have
 \begin{equation*} N_a((\sigma _c - 1)u\cdot (\gamma /\alpha )) = (\sigma _c - 1)N_a(u)\cdot N_a(\gamma /\alpha )= (\sigma _c - 1)w^p \cdot (\gamma ^p/b) = 1. \end{equation*}
\begin{equation*} N_a((\sigma _c - 1)u\cdot (\gamma /\alpha )) = (\sigma _c - 1)N_a(u)\cdot N_a(\gamma /\alpha )= (\sigma _c - 1)w^p \cdot (\gamma ^p/b) = 1. \end{equation*}
By Hilbert’s Theorem 90, there is 
 $u'\in F_{a,c}^\times$
 such that
$u'\in F_{a,c}^\times$
 such that
 \begin{equation*} \alpha \cdot (\sigma _a - 1)u' = \gamma \cdot (\sigma _c - 1)u. \end{equation*}
\begin{equation*} \alpha \cdot (\sigma _a - 1)u' = \gamma \cdot (\sigma _c - 1)u. \end{equation*}
By Proposition 3.5, we obtain a Galois 
 $U_4^+$
-algebra
$U_4^+$
-algebra 
 $K_1$
 over
$K_1$
 over 
 $F$
 with the property that
$F$
 with the property that 
 $(K_1)^{Q_4^+}\simeq F_{a,b,c}$
 as
$(K_1)^{Q_4^+}\simeq F_{a,b,c}$
 as 
 $(\mathbb Z/p\mathbb Z)^3$
-algebras. Similarly, we get a Galois
$(\mathbb Z/p\mathbb Z)^3$
-algebras. Similarly, we get a Galois 
 $U_4^-$
-algebra over
$U_4^-$
-algebra over 
 $F$
 such that
$F$
 such that 
 $(K_2)^{Q_4^-}\simeq F_{b,c,d}$
 as
$(K_2)^{Q_4^-}\simeq F_{b,c,d}$
 as 
 $(\mathbb Z/p\mathbb Z)^3$
-algebras. Since
$(\mathbb Z/p\mathbb Z)^3$
-algebras. Since 
 $N_a(u)\cdot N_d(v) = w^p$
 and
$N_a(u)\cdot N_d(v) = w^p$
 and 
 $(\sigma_b-1)(\sigma_c-1)w=\zeta$
, by Proposition 3.2(3) the
$(\sigma_b-1)(\sigma_c-1)w=\zeta$
, by Proposition 3.2(3) the 
 $U_3$
-algebras
$U_3$
-algebras 
 $(K_1)^{P_4^+}$
 and
$(K_1)^{P_4^+}$
 and 
 $(K_2)^{P_4^-}$
 are isomorphic. Now Lemma 2.1 applied to the cartesian square (2.14) for
$(K_2)^{P_4^-}$
 are isomorphic. Now Lemma 2.1 applied to the cartesian square (2.14) for 
 $n=4$
 yields a
$n=4$
 yields a 
 $\overline {U}_5$
-Galois algebra
$\overline {U}_5$
-Galois algebra 
 $L$
 such that
$L$
 such that 
 $L^{Q_5}\simeq F_{a,b,c,d}$
 as
$L^{Q_5}\simeq F_{a,b,c,d}$
 as 
 $(\mathbb Z/p\mathbb Z)^4$
-algebras. By Corollary 2.5, this implies that
$(\mathbb Z/p\mathbb Z)^4$
-algebras. By Corollary 2.5, this implies that 
 $\langle {a,b,c,d}\rangle$
 is defined.
$\langle {a,b,c,d}\rangle$
 is defined.
Lemma 3.8. 
Let 
 $b,c\in F^\times$
 and
$b,c\in F^\times$
 and 
 $w\in F_{b,c}^{\times }$
. We have
$w\in F_{b,c}^{\times }$
. We have 
 $(\sigma _b-1)(\sigma _c-1)w=1$
 if and only if there exist
$(\sigma _b-1)(\sigma _c-1)w=1$
 if and only if there exist 
 $w_b\in F_b^\times$
 and
$w_b\in F_b^\times$
 and 
 $w_c\in F_c^\times$
 such that
$w_c\in F_c^\times$
 such that 
 $w=w_bw_c$
 in
$w=w_bw_c$
 in 
 $F_{b,c}^\times$
.
$F_{b,c}^\times$
.
Proof.
 We have 
 $(\sigma _b-1)(\sigma _c-1)(w_bw_c)=(\sigma _b-1)w_c=1$
 for all
$(\sigma _b-1)(\sigma _c-1)(w_bw_c)=(\sigma _b-1)w_c=1$
 for all 
 $w_b\in F_b^\times$
 and
$w_b\in F_b^\times$
 and 
 $w_c\in F_c^\times$
. Conversely, if
$w_c\in F_c^\times$
. Conversely, if 
 $w\in F_{b,c}^\times$
 satisfies
$w\in F_{b,c}^\times$
 satisfies 
 $(\sigma _b-1)(\sigma _c-1)w=1$
, then
$(\sigma _b-1)(\sigma _c-1)w=1$
, then 
 $(\sigma _c-1)w\in F_c^\times$
 and
$(\sigma _c-1)w\in F_c^\times$
 and 
 $N_c((\sigma _c-1)w)=1$
, and hence by Hilbert’s Theorem 90 there exists
$N_c((\sigma _c-1)w)=1$
, and hence by Hilbert’s Theorem 90 there exists 
 $w_c\in F_c^\times$
 such that
$w_c\in F_c^\times$
 such that 
 $(\sigma _c-1)w_c=(\sigma _c-1)w$
. Letting
$(\sigma _c-1)w_c=(\sigma _c-1)w$
. Letting 
 $w_b:= w/w_c\in F_{b,c}^\times$
, we have
$w_b:= w/w_c\in F_{b,c}^\times$
, we have
 \begin{equation*}(\sigma _c-1)w_b=(\sigma _c-1)(w/w_c)=1,\end{equation*}
\begin{equation*}(\sigma _c-1)w_b=(\sigma _c-1)(w/w_c)=1,\end{equation*}
that is, 
 $w_b\in F_b^{\times }$
.
$w_b\in F_b^{\times }$
.
From Proposition 3.7, we derive the following necessary condition for a fourfold Massey product to be defined.
Proposition 3.9. 
Let 
 $p$
 be a prime, let
$p$
 be a prime, let 
 $F$
 be a field of characteristic different from
$F$
 be a field of characteristic different from 
 $p$
 and containing a primitive
$p$
 and containing a primitive 
 $p$
th root of unity
$p$
th root of unity 
 $\zeta$
, let
$\zeta$
, let 
 $a,b,c,d\in F^\times$
, and suppose that
$a,b,c,d\in F^\times$
, and suppose that 
 $\langle {a,b,c,d}\rangle$
 is defined over
$\langle {a,b,c,d}\rangle$
 is defined over 
 $F$
. For every
$F$
. For every 
 $w\in F_{b,c}^\times$
 such that
$w\in F_{b,c}^\times$
 such that 
 $(\sigma _b-1)(\sigma _c-1)w=\zeta$
, there exist
$(\sigma _b-1)(\sigma _c-1)w=\zeta$
, there exist 
 $u\in F_{a,c}^\times$
 and
$u\in F_{a,c}^\times$
 and 
 $v\in F_{b,d}^\times$
 such that
$v\in F_{b,d}^\times$
 such that 
 $N_a(u)N_d(v)=w^p$
.
$N_a(u)N_d(v)=w^p$
.
Proof.
 By Proposition 3.7, there exist 
 $u_0\in F_{a,c}^\times$
,
$u_0\in F_{a,c}^\times$
, 
 $v_0\in F_{b,d}^\times$
 and
$v_0\in F_{b,d}^\times$
 and 
 $w_0\in F_{b,c}^\times$
 such that
$w_0\in F_{b,c}^\times$
 such that
 \begin{equation*}N_a(u_0) \cdot N_d(v_0) =w_0^p,\quad (\sigma _b-1)(\sigma _c-1)w_0=\zeta .\end{equation*}
\begin{equation*}N_a(u_0) \cdot N_d(v_0) =w_0^p,\quad (\sigma _b-1)(\sigma _c-1)w_0=\zeta .\end{equation*}
We have 
 $(\sigma _b-1)(\sigma _c-1)(w_0/w)=1$
. By Lemma 3.8, this implies that
$(\sigma _b-1)(\sigma _c-1)(w_0/w)=1$
. By Lemma 3.8, this implies that 
 $w_0=w w_b w_c$
, where
$w_0=w w_b w_c$
, where 
 $w_b\in F_b^\times$
 and
$w_b\in F_b^\times$
 and 
 $w_c\in F_c^\times$
. If we define
$w_c\in F_c^\times$
. If we define 
 $u=u_0 w_c$
 and
$u=u_0 w_c$
 and 
 $v=v_0 w_b$
, then
$v=v_0 w_b$
, then
 \begin{equation}N_a(u)N_d(v)=N_a(u_0)N_a(w_c)N_d(v_0)N_d(w_b)=w_0^pw_c^pw_b^p=w^p.\end{equation}
\begin{equation}N_a(u)N_d(v)=N_a(u_0)N_a(w_c)N_d(v_0)N_d(w_b)=w_0^pw_c^pw_b^p=w^p.\end{equation}
4. A generic variety
 In this section, we let 
 $p$
 be a prime number, and we let
$p$
 be a prime number, and we let 
 $F$
 be a field of characteristic different from
$F$
 be a field of characteristic different from 
 $p$
 and containing a primitive
$p$
 and containing a primitive 
 $p$
th root of unity
$p$
th root of unity 
 $\zeta$
.
$\zeta$
.
 Let 
 $b,c\in F^\times$
, and let
$b,c\in F^\times$
, and let 
 $X$
 be the Severi–Brauer variety associated to
$X$
 be the Severi–Brauer variety associated to 
 $(b,c)$
 over
$(b,c)$
 over 
 $F$
; see [Reference Gille and SzamuelyGS17, Chapter 5]. For every étale
$F$
; see [Reference Gille and SzamuelyGS17, Chapter 5]. For every étale 
 $F$
-algebra
$F$
-algebra 
 $K$
, we have
$K$
, we have 
 $(b,c)=0$
 in
$(b,c)=0$
 in 
 $\operatorname {Br}(K)$
 if and only if
$\operatorname {Br}(K)$
 if and only if 
 $X_K\simeq \mathbb P^{p-1}_K$
 over
$X_K\simeq \mathbb P^{p-1}_K$
 over 
 $K$
. In particular,
$K$
. In particular, 
 $X_b\simeq \mathbb{P}^{p-1}_b\ \text{over}\ {F_b}$
. (Recall that we write (
$X_b\simeq \mathbb{P}^{p-1}_b\ \text{over}\ {F_b}$
. (Recall that we write (
 $\mathbb{P}^{p-1}_b\ \textrm{for}\ \mathbb{P}^{p-1}_{F_b}$
.) By [Reference Gille and SzamuelyGS17, Theorem 5.4.1], the central simple algebra
$\mathbb{P}^{p-1}_b\ \textrm{for}\ \mathbb{P}^{p-1}_{F_b}$
.) By [Reference Gille and SzamuelyGS17, Theorem 5.4.1], the central simple algebra 
 $(b,c)$
 is split over
$(b,c)$
 is split over 
 $F(X)$
.
$F(X)$
.
 We define the degree map 
 $\deg \colon \operatorname {Pic}(X)\to \mathbb Z$
 as the composite of the pullback map
$\deg \colon \operatorname {Pic}(X)\to \mathbb Z$
 as the composite of the pullback map 
 $\operatorname {Pic}(X)\to \operatorname {Pic}(X_b)\simeq \operatorname {Pic}(\mathbb P^{p-1}_b)$
 and the degree isomorphism
$\operatorname {Pic}(X)\to \operatorname {Pic}(X_b)\simeq \operatorname {Pic}(\mathbb P^{p-1}_b)$
 and the degree isomorphism 
 $\operatorname {Pic}(\mathbb P^{p-1}_b)\to \mathbb Z$
. This does not depend on the choice of isomorphism
$\operatorname {Pic}(\mathbb P^{p-1}_b)\to \mathbb Z$
. This does not depend on the choice of isomorphism 
 $X_b\simeq \mathbb P^{p-1}_b$
.
$X_b\simeq \mathbb P^{p-1}_b$
.
Lemma 4.1. 
Let 
 $b,c\in F^\times$
, let
$b,c\in F^\times$
, let 
 $G_b:= \operatorname {Gal}(F_b/F)$
 and let
$G_b:= \operatorname {Gal}(F_b/F)$
 and let 
 $X$
 be the Severi–Brauer variety of
$X$
 be the Severi–Brauer variety of 
 $(b,c)$
 over
$(b,c)$
 over 
 $F$
. Let
$F$
. Let 
 $s_1,\ldots ,s_p$
 be homogeneous coordinates on
$s_1,\ldots ,s_p$
 be homogeneous coordinates on 
 $\mathbb P^{p-1}_F$
.
$\mathbb P^{p-1}_F$
.
- 
(1) There exists a  $G_b$
-equivariant isomorphism $G_b$
-equivariant isomorphism $X_b\xrightarrow {\sim } \mathbb P_b^{p-1}$
, where $X_b\xrightarrow {\sim } \mathbb P_b^{p-1}$
, where $G_b$
 acts on $G_b$
 acts on $X_b$
 via its action on $X_b$
 via its action on $F_b$
, and on $F_b$
, and on $\mathbb P^{p-1}_b$
 by $\mathbb P^{p-1}_b$
 by \begin{equation*}\sigma _b^*(s_1)=cs_p,\quad \sigma _b^*(s_i)=s_{i-1}\quad (i=2,\ldots ,p).\end{equation*} \begin{equation*}\sigma _b^*(s_1)=cs_p,\quad \sigma _b^*(s_i)=s_{i-1}\quad (i=2,\ldots ,p).\end{equation*}
- 
(2) If  $(b,c)\neq 0$
 in $(b,c)\neq 0$
 in $\operatorname {Br}(F)$
, the image of $\operatorname {Br}(F)$
, the image of $\deg \colon \operatorname {Pic}(X)\to \mathbb Z$
 is equal to $\deg \colon \operatorname {Pic}(X)\to \mathbb Z$
 is equal to $p\mathbb Z$
. $p\mathbb Z$
.
- 
(3) There exists a rational function  $w\in F_{b,c}(X)^\times$
 such that
and $w\in F_{b,c}(X)^\times$
 such that
and \begin{equation*}(\sigma _b-1)(\sigma _c-1)w=\zeta \end{equation*}
where \begin{equation*}(\sigma _b-1)(\sigma _c-1)w=\zeta \end{equation*}
where \begin{equation*}\operatorname {div}(w)=x-y\quad \text {in $\operatorname {Div}(X_{b,c})$},\end{equation*} \begin{equation*}\operatorname {div}(w)=x-y\quad \text {in $\operatorname {Div}(X_{b,c})$},\end{equation*} $x,y\in (X_{b,c})^{(1)}$
 satisfy $x,y\in (X_{b,c})^{(1)}$
 satisfy $\deg (x)=\deg (y)=1$
, $\deg (x)=\deg (y)=1$
, $\sigma _b(x)=x$
 and $\sigma _b(x)=x$
 and $\sigma _c(y)=y$
. $\sigma _c(y)=y$
.
Proof.
 (1) Consider the 
 $1$
-cocycle
$1$
-cocycle 
 $z\colon G_b\to \operatorname {PGL}_p(F_b)$
 given by
$z\colon G_b\to \operatorname {PGL}_p(F_b)$
 given by
 \begin{align*}\sigma _b\mapsto \left[\begin{array}{l@{\quad}l@{\quad}l@{\quad}l@{\quad}l} 0 & 0 & \ldots & 0 & c \\ 1 & 0 & \ldots & 0 & 0 \\ 0 & 1 & \ldots & 0 & 0 \\ \vdots & \vdots & \ddots & \vdots & \vdots \\ 0 & 0 & \ldots & 1 & 0 \\ \end{array} \right]. \end{align*}
\begin{align*}\sigma _b\mapsto \left[\begin{array}{l@{\quad}l@{\quad}l@{\quad}l@{\quad}l} 0 & 0 & \ldots & 0 & c \\ 1 & 0 & \ldots & 0 & 0 \\ 0 & 1 & \ldots & 0 & 0 \\ \vdots & \vdots & \ddots & \vdots & \vdots \\ 0 & 0 & \ldots & 1 & 0 \\ \end{array} \right]. \end{align*}
By [Reference Gille and SzamuelyGS17, Construction 2.5.1], the class 
 $[z]\in H^1(G_b,\operatorname {PGL}_p(F_b))$
 coincides with the class of the degree-
$[z]\in H^1(G_b,\operatorname {PGL}_p(F_b))$
 coincides with the class of the degree-
 $p$
 central simple algebra over
$p$
 central simple algebra over 
 $F$
 with Brauer class
$F$
 with Brauer class 
 $(b,c)$
, and hence with the class of the associated Severi–Brauer variety
$(b,c)$
, and hence with the class of the associated Severi–Brauer variety 
 $X$
. It follows that we have a
$X$
. It follows that we have a 
 $G_b$
-equivariant isomorphism
$G_b$
-equivariant isomorphism 
 $X_b\simeq \mathbb P^{p-1}_b$
, where
$X_b\simeq \mathbb P^{p-1}_b$
, where 
 $G_b$
 acts on
$G_b$
 acts on 
 $X_b$
 via its action on
$X_b$
 via its action on 
 $F_b$
, and on
$F_b$
, and on 
 $\mathbb P^{p-1}_b$
 via the cocycle
$\mathbb P^{p-1}_b$
 via the cocycle 
 $z$
. This proves (1).
$z$
. This proves (1).
(2) By a theorem of Lichtenbaum [Reference Gille and SzamuelyGS17, Theorem 5.4.10], we have an exact sequence
 \begin{equation*}\operatorname {Pic}(X)\xrightarrow {\deg } \mathbb Z\xrightarrow {\delta }\operatorname {Br}(F),\end{equation*}
\begin{equation*}\operatorname {Pic}(X)\xrightarrow {\deg } \mathbb Z\xrightarrow {\delta }\operatorname {Br}(F),\end{equation*}
where 
 $\delta (1)=(b,c)$
. Since
$\delta (1)=(b,c)$
. Since 
 $(b,c)$
 has exponent
$(b,c)$
 has exponent 
 $p$
, we conclude that the image of
$p$
, we conclude that the image of 
 $\deg$
 is equal to
$\deg$
 is equal to 
 $p\mathbb Z$
.
$p\mathbb Z$
.
 (3) Let 
 $G_{b,c}:= \operatorname {Gal}(F_{b,c}/F)=\langle {\sigma _b,\sigma _c}\rangle$
. By (1), there is a
$G_{b,c}:= \operatorname {Gal}(F_{b,c}/F)=\langle {\sigma _b,\sigma _c}\rangle$
. By (1), there is a 
 $G_{b,c}$
-equivariant isomorphism
$G_{b,c}$
-equivariant isomorphism 
 $f\colon \mathbb P^{p-1}_{b,c}\to X_{b,c}$
, where
$f\colon \mathbb P^{p-1}_{b,c}\to X_{b,c}$
, where 
 $G_{b,c}$
 acts on
$G_{b,c}$
 acts on 
 $X_{b,c}$
 via its action on
$X_{b,c}$
 via its action on 
 $F_{b,c}$
, the action of
$F_{b,c}$
, the action of 
 $\sigma _c$
 on
$\sigma _c$
 on 
 $\mathbb P^{p-1}_{b,c}$
 is trivial and the action of
$\mathbb P^{p-1}_{b,c}$
 is trivial and the action of 
 $\sigma _b$
 on
$\sigma _b$
 on 
 $\mathbb P^{p-1}_{b,c}$
 is determined by
$\mathbb P^{p-1}_{b,c}$
 is determined by
 \begin{equation*}\sigma _b^*(s_1)=cs_p,\quad \sigma _b^*(s_i)=s_{i-1}\quad (i=2,\ldots ,p).\end{equation*}
\begin{equation*}\sigma _b^*(s_1)=cs_p,\quad \sigma _b^*(s_i)=s_{i-1}\quad (i=2,\ldots ,p).\end{equation*}
Consider the linear form 
 $l:={{\sum }}_{i=1}^{p} c^{i/p}\cdot s_i$
 on
$l:={{\sum }}_{i=1}^{p} c^{i/p}\cdot s_i$
 on 
 $\mathbb P^{p-1}_{b,c}$
 and set
$\mathbb P^{p-1}_{b,c}$
 and set 
 $w':= l/s_p\in F_{b,c}(\mathbb P^{p-1})^\times$
. We have
$w':= l/s_p\in F_{b,c}(\mathbb P^{p-1})^\times$
. We have 
 $\sigma _b^*(l)=c^{1/p}\cdot l$
, and hence
$\sigma _b^*(l)=c^{1/p}\cdot l$
, and hence 
 $(\sigma _b-1)w'=c^{1/p}\cdot (s_p/s_{p-1})$
. It follows that
$(\sigma _b-1)w'=c^{1/p}\cdot (s_p/s_{p-1})$
. It follows that 
 $(\sigma _c-1)(\sigma _b-1)w'=\xi$
. Let
$(\sigma _c-1)(\sigma _b-1)w'=\xi$
. Let 
 $x',y'\in \operatorname {Div}(\mathbb P^{p-1}_{b,c})$
 be the classes of the linear subspaces of
$x',y'\in \operatorname {Div}(\mathbb P^{p-1}_{b,c})$
 be the classes of the linear subspaces of 
 $\mathbb P^{p-1}_{b,c}$
 given by
$\mathbb P^{p-1}_{b,c}$
 given by 
 $l=0$
 and
$l=0$
 and 
 $s_p=0$
, respectively. Then
$s_p=0$
, respectively. Then
 \begin{equation*}\operatorname {div}(w')=x'-y',\quad \sigma _b(x')=x',\quad \sigma _c(y')=y'.\end{equation*}
\begin{equation*}\operatorname {div}(w')=x'-y',\quad \sigma _b(x')=x',\quad \sigma _c(y')=y'.\end{equation*}
Define
 \begin{equation*}w:= w'\circ f^{-1}\in F_{b,c}(X)^\times ,\quad x':= f_*(x)\in (X_{b,c})^{(1)},\quad y':= f_*(y)\in (X_{b,c})^{(1)}.\end{equation*}
\begin{equation*}w:= w'\circ f^{-1}\in F_{b,c}(X)^\times ,\quad x':= f_*(x)\in (X_{b,c})^{(1)},\quad y':= f_*(y)\in (X_{b,c})^{(1)}.\end{equation*}
Then 
 $w,x,y$
 satisfy the conclusion of (3).
$w,x,y$
 satisfy the conclusion of (3).
Lemma 4.2. 
Let 
 $a,b,c,d\in F^\times$
. The complex of tori
$a,b,c,d\in F^\times$
. The complex of tori
 \begin{equation*}R_{a,c}({\mathbb G}_{\operatorname {m}})\times R_{b,d}({\mathbb G}_{\operatorname {m}})\xrightarrow {\varphi }R_{b,c}({\mathbb G}_{\operatorname {m}})\xrightarrow {\psi } R_{b,c}({\mathbb G}_{\operatorname {m}}),\end{equation*}
\begin{equation*}R_{a,c}({\mathbb G}_{\operatorname {m}})\times R_{b,d}({\mathbb G}_{\operatorname {m}})\xrightarrow {\varphi }R_{b,c}({\mathbb G}_{\operatorname {m}})\xrightarrow {\psi } R_{b,c}({\mathbb G}_{\operatorname {m}}),\end{equation*}
where 
 $\varphi (u,v):= N_a(u)N_d(v)$
 and
$\varphi (u,v):= N_a(u)N_d(v)$
 and 
 $\psi (z)=(\sigma _b-1)(\sigma _c-1)z$
, is exact. Furthermore, the torus
$\psi (z)=(\sigma _b-1)(\sigma _c-1)z$
, is exact. Furthermore, the torus 
 $\operatorname {Im}({\varphi })=\operatorname {Ker}(\psi )$
 has dimension
$\operatorname {Im}({\varphi })=\operatorname {Ker}(\psi )$
 has dimension 
 $2p-1$
.
$2p-1$
.
Proof. By Lemma 3.8, we have an exact sequence
 \begin{equation*}R_c({\mathbb G}_{\operatorname {m}})\times R_b({\mathbb G}_{\operatorname {m}})\xrightarrow {\varphi '} R_{b,c}({\mathbb G}_{\operatorname {m}})\xrightarrow {\psi } R_{b,c}({\mathbb G}_{\operatorname {m}}),\end{equation*}
\begin{equation*}R_c({\mathbb G}_{\operatorname {m}})\times R_b({\mathbb G}_{\operatorname {m}})\xrightarrow {\varphi '} R_{b,c}({\mathbb G}_{\operatorname {m}})\xrightarrow {\psi } R_{b,c}({\mathbb G}_{\operatorname {m}}),\end{equation*}
where 
 $\varphi '(x,y)=xy$
. The homomorphism
$\varphi '(x,y)=xy$
. The homomorphism 
 $\varphi$
 factors as
$\varphi$
 factors as
 \begin{equation*}R_{a,c}({\mathbb G}_{\operatorname {m}})\times R_{b,d}({\mathbb G}_{\operatorname {m}})\xrightarrow {N_a\times N_d}R_c({\mathbb G}_{\operatorname {m}})\times R_b({\mathbb G}_{\operatorname {m}})\xrightarrow {\varphi '} R_{b,c}({\mathbb G}_{\operatorname {m}}).\end{equation*}
\begin{equation*}R_{a,c}({\mathbb G}_{\operatorname {m}})\times R_{b,d}({\mathbb G}_{\operatorname {m}})\xrightarrow {N_a\times N_d}R_c({\mathbb G}_{\operatorname {m}})\times R_b({\mathbb G}_{\operatorname {m}})\xrightarrow {\varphi '} R_{b,c}({\mathbb G}_{\operatorname {m}}).\end{equation*}
Since the homomorphisms 
 $N_a$
 and
$N_a$
 and 
 $N_d$
 are surjective, so is
$N_d$
 are surjective, so is 
 $N_a\times N_d$
. We conclude that
$N_a\times N_d$
. We conclude that 
 $\operatorname {Im}(\varphi )=\operatorname {Im}(\varphi ')=\operatorname {Ker}(\psi )$
, as desired. Finally, it is immediate to see that
$\operatorname {Im}(\varphi )=\operatorname {Im}(\varphi ')=\operatorname {Ker}(\psi )$
, as desired. Finally, it is immediate to see that 
 $\operatorname {Ker}(\varphi ')={\mathbb G}_{\operatorname {m}}$
, embedded anti-diagonally in
$\operatorname {Ker}(\varphi ')={\mathbb G}_{\operatorname {m}}$
, embedded anti-diagonally in 
 $R_c({\mathbb G}_{\operatorname {m}})\times R_b({\mathbb G}_{\operatorname {m}})$
. Thus,
$R_c({\mathbb G}_{\operatorname {m}})\times R_b({\mathbb G}_{\operatorname {m}})$
. Thus,
 \begin{equation*}{\dim }(\operatorname {Im}(\varphi ))={\dim }(\operatorname {Im}(\varphi '))=2p-{\dim }(\operatorname {Ker}(\varphi '))=2p-1.\end{equation*}
\begin{equation*}{\dim }(\operatorname {Im}(\varphi ))={\dim }(\operatorname {Im}(\varphi '))=2p-{\dim }(\operatorname {Ker}(\varphi '))=2p-1.\end{equation*}
 Let 
 $a,b,c,d\in F^\times$
, and consider the complex of tori of Lemma 4.2. We define the following groups of multiplicative type over
$a,b,c,d\in F^\times$
, and consider the complex of tori of Lemma 4.2. We define the following groups of multiplicative type over 
 $F$
:
$F$
:
 \begin{equation*}P:= R_{a,c}({\mathbb G}_{\operatorname {m}})\times R_{b,d}({\mathbb G}_{\operatorname {m}}),\quad S:= \operatorname {Ker}(\psi )=\operatorname {Im}(\varphi ),\quad T:= \operatorname {Ker}(\varphi )\subset P.\end{equation*}
\begin{equation*}P:= R_{a,c}({\mathbb G}_{\operatorname {m}})\times R_{b,d}({\mathbb G}_{\operatorname {m}}),\quad S:= \operatorname {Ker}(\psi )=\operatorname {Im}(\varphi ),\quad T:= \operatorname {Ker}(\varphi )\subset P.\end{equation*}
By Lemma 4.2, we get a short exact sequence
 \begin{align} 1\to T\xrightarrow {\iota } P\xrightarrow {\pi } S\to 1, \end{align}
\begin{align} 1\to T\xrightarrow {\iota } P\xrightarrow {\pi } S\to 1, \end{align}
where 
 $\iota$
 is the inclusion map and
$\iota$
 is the inclusion map and 
 $\pi$
 is induced by
$\pi$
 is induced by 
 $\varphi$
.
$\varphi$
.
Lemma 4.3. 
The groups of multiplicative type 
 $T$
,
$T$
, 
 $P$
 and
$P$
 and 
 $S$
 are tori.
$S$
 are tori.
Proof.
 It is clear that 
 $P$
 and
$P$
 and 
 $S$
 are tori. We now prove that
$S$
 are tori. We now prove that 
 $T$
 is a torus. Consider the subgroup
$T$
 is a torus. Consider the subgroup 
 $Q\subset R_{a,c}({\mathbb G}_{\operatorname {m}})$
, which makes the following commutative square cartesian.
$Q\subset R_{a,c}({\mathbb G}_{\operatorname {m}})$
, which makes the following commutative square cartesian.

 Here the bottom horizontal map is the obvious inclusion. It follows that 
 $Q$
 is an
$Q$
 is an 
 $R_c(R^{(1)}_a({\mathbb G}_{\operatorname {m}}))$
-torsor over
$R_c(R^{(1)}_a({\mathbb G}_{\operatorname {m}}))$
-torsor over 
 ${\mathbb G}_{\operatorname {m}}$
, and hence it is smooth and connected. Therefore,
${\mathbb G}_{\operatorname {m}}$
, and hence it is smooth and connected. Therefore, 
 $Q$
 is a torus.
$Q$
 is a torus.
 The image of the projection 
 $T \stackrel {\iota }{\hookrightarrow } P\to R_{a,c}({\mathbb G}_{\operatorname {m}})$
 is contained in the torus
$T \stackrel {\iota }{\hookrightarrow } P\to R_{a,c}({\mathbb G}_{\operatorname {m}})$
 is contained in the torus 
 $Q$
. Moreover, the kernel
$Q$
. Moreover, the kernel 
 $U$
 of the projection is
$U$
 of the projection is 
 $R_b(R^{(1)}_{F_{b,d}/F_b}({\mathbb G}_{\operatorname {m}}))$
, and hence it is also a torus. We have an exact sequence
$R_b(R^{(1)}_{F_{b,d}/F_b}({\mathbb G}_{\operatorname {m}}))$
, and hence it is also a torus. We have an exact sequence
 \begin{equation*}1\to U\to T\to Q.\end{equation*}
\begin{equation*}1\to U\to T\to Q.\end{equation*}
We have 
 ${\dim }(U) = p(p-1)$
, and we see from the cartesian square (4.2) that
${\dim }(U) = p(p-1)$
, and we see from the cartesian square (4.2) that 
 ${\dim } (Q) = p^2 - p +1$
. By Lemma 4.2, we have
${\dim } (Q) = p^2 - p +1$
. By Lemma 4.2, we have 
 ${\dim }(S)=2p-1$
. From (4.1), we deduce that
${\dim }(S)=2p-1$
. From (4.1), we deduce that
 \begin{equation*}{\dim } (T) = {\dim }(P) - {\dim }(S) = 2p^2 -(2p-1)=2p^2-2p+1.\end{equation*}
\begin{equation*}{\dim } (T) = {\dim }(P) - {\dim }(S) = 2p^2 -(2p-1)=2p^2-2p+1.\end{equation*}
Therefore, 
 ${\dim } (T) = {\dim } (U) + {\dim } (Q)$
, and so the sequence
${\dim } (T) = {\dim } (U) + {\dim } (Q)$
, and so the sequence
 \begin{equation*}1 \to U \to T \to Q \to 1\end{equation*}
\begin{equation*}1 \to U \to T \to Q \to 1\end{equation*}
is exact. As 
 $U$
 and
$U$
 and 
 $Q$
 are tori, so is
$Q$
 are tori, so is 
 $T$
.
$T$
.
Proposition 4.4. 
Let 
 $p$
 be a prime, let
$p$
 be a prime, let 
 $F$
 be a field of characteristic different from
$F$
 be a field of characteristic different from 
 $p$
 and containing a primitive
$p$
 and containing a primitive 
 $p$
th root of unity
$p$
th root of unity 
 $\zeta$
, and let
$\zeta$
, and let 
 $a,b,c,d\in F^\times$
. Suppose that
$a,b,c,d\in F^\times$
. Suppose that 
 $(a,b)=(b,c)=(c,d)=0$
 in
$(a,b)=(b,c)=(c,d)=0$
 in 
 $\operatorname {Br}(F)$
, and let
$\operatorname {Br}(F)$
, and let 
 $w\in F_{b,c}^\times$
 be such that
$w\in F_{b,c}^\times$
 be such that 
 $(\sigma _b-1)(\sigma _c-1)w = \zeta$
. Let
$(\sigma _b-1)(\sigma _c-1)w = \zeta$
. Let 
 $T$
 and
$T$
 and 
 $P$
 be the tori appearing in (
4.1
), and let
$P$
 be the tori appearing in (
4.1
), and let 
 $E_w\subset P$
 be the
$E_w\subset P$
 be the 
 $T$
-torsor given by the equation
$T$
-torsor given by the equation 
 $N_a(u)N_d(v)=w^p$
. Then the mod
$N_a(u)N_d(v)=w^p$
. Then the mod 
 $p$
 Massey product
$p$
 Massey product 
 $\langle {a,b,c,d}\rangle$
 is defined over
$\langle {a,b,c,d}\rangle$
 is defined over 
 $F$
 if and only if
$F$
 if and only if 
 $E_w$
 is trivial.
$E_w$
 is trivial.
 The construction of 
 $E_w$
 is functorial in
$E_w$
 is functorial in 
 $F$
. Therefore, for every field extension
$F$
. Therefore, for every field extension 
 $K/F$
, the mod
$K/F$
, the mod 
 $p$
 Massey product
$p$
 Massey product 
 $\langle {a,b,c,d}\rangle$
 is defined over
$\langle {a,b,c,d}\rangle$
 is defined over 
 $K$
 if and only if
$K$
 if and only if 
 $E_w$
 is split by
$E_w$
 is split by 
 $K$
. We may thus call
$K$
. We may thus call 
 $E_w$
 a generic variety for the property ‘the Massey product
$E_w$
 a generic variety for the property ‘the Massey product 
 $\langle {a,b,c,d}\rangle$
 is defined’.
$\langle {a,b,c,d}\rangle$
 is defined’.
Proof.
 Suppose that the Massey product 
 $\langle {a,b,c,d}\rangle$
 is defined over
$\langle {a,b,c,d}\rangle$
 is defined over 
 $F$
. By Proposition 3.9, there exist
$F$
. By Proposition 3.9, there exist 
 $u\in F_{a,c}^\times$
,
$u\in F_{a,c}^\times$
, 
 $v\in F_{b,d}^\times$
 such that
$v\in F_{b,d}^\times$
 such that 
 $N_a(u)N_d(v) = w^p$
. This means precisely that
$N_a(u)N_d(v) = w^p$
. This means precisely that 
 $E_w\subset P$
 has the
$E_w\subset P$
 has the 
 $F$
-point
$F$
-point 
 $(u,v)$
. Thus, the
$(u,v)$
. Thus, the 
 $T$
-torsor
$T$
-torsor 
 $E_w$
 is trivial.
$E_w$
 is trivial.
 Conversely, suppose that the 
 $T$
-torsor
$T$
-torsor 
 $E_w$
 is trivial and let
$E_w$
 is trivial and let 
 $(u,v)$
 be an
$(u,v)$
 be an 
 $F$
-point of
$F$
-point of 
 $E_w$
. Then we have
$E_w$
. Then we have 
 $N_a(u)N_d(v) = w^p$
 and, by assumption, we also have
$N_a(u)N_d(v) = w^p$
 and, by assumption, we also have 
 $(\sigma _b-1)(\sigma _c-1)w = \zeta$
. Proposition 3.7 now implies that the Massey product
$(\sigma _b-1)(\sigma _c-1)w = \zeta$
. Proposition 3.7 now implies that the Massey product 
 $\langle {a,b,c,d}\rangle$
 is defined over
$\langle {a,b,c,d}\rangle$
 is defined over 
 $F$
.
$F$
.
Corollary 4.5. 
Let 
 $p$
 be a prime, let
$p$
 be a prime, let 
 $F$
 be a field of characteristic different from
$F$
 be a field of characteristic different from 
 $p$
 and containing a primitive
$p$
 and containing a primitive 
 $p$
th root of unity
$p$
th root of unity 
 $\zeta$
, and let
$\zeta$
, and let 
 $a,b,c,d\in F^\times$
 be such that
$a,b,c,d\in F^\times$
 be such that 
 $(a,b)=(c,d)=0$
 in
$(a,b)=(c,d)=0$
 in 
 $\operatorname{Br}(F)$
. Let
$\operatorname{Br}(F)$
. Let 
 $X$
 be the Severi–Brauer variety of
$X$
 be the Severi–Brauer variety of 
 $(b,c)$
 over
$(b,c)$
 over 
 $F$
, fix
$F$
, fix 
 $w\in F_{b,c}(X)^\times$
 as in Lemma 4.1(3) and let
$w\in F_{b,c}(X)^\times$
 as in Lemma 4.1(3) and let 
 $E_w\subset P_{F(X)}$
 be the
$E_w\subset P_{F(X)}$
 be the 
 $T_{F(X)}$
-torsor given by the equation
$T_{F(X)}$
-torsor given by the equation 
 $N_a(u)N_d(v)=w^p$
.
$N_a(u)N_d(v)=w^p$
.
 
The Massey product 
 $\langle {a,b,c,d}\rangle$
 is defined over
$\langle {a,b,c,d}\rangle$
 is defined over 
 $F(X)$
 if and only if
$F(X)$
 if and only if 
 $E_w$
 is trivial over
$E_w$
 is trivial over 
 $F(X)$
.
$F(X)$
.
Proof.
 This is a special case of Proposition 4.4, applied over the ground field 
 $F(X)$
.
$F(X)$
.
5. Proof of Theorem 1.3
 Let 
 $p$
 be a prime and let
$p$
 be a prime and let 
 $F$
 be a field of characteristic different from
$F$
 be a field of characteristic different from 
 $p$
 and containing a primitive
$p$
 and containing a primitive 
 $p$
th root of unity
$p$
th root of unity 
 $\zeta$
. Let
$\zeta$
. Let 
 $a,b,c,d\in F^\times$
 be such that their cosets in
$a,b,c,d\in F^\times$
 be such that their cosets in 
 $F^\times /F^{\times p}$
 are
$F^\times /F^{\times p}$
 are 
 ${\mathbb F}_p$
-linearly independent. Consider the field
${\mathbb F}_p$
-linearly independent. Consider the field 
 $K:= F_{a,b,c,d}$
, and write
$K:= F_{a,b,c,d}$
, and write 
 $G=\operatorname {Gal}(K/F)=\langle {\sigma _a,\sigma _b,\sigma _c,\sigma _d}\rangle$
 as in § 2.1. We set
$G=\operatorname {Gal}(K/F)=\langle {\sigma _a,\sigma _b,\sigma _c,\sigma _d}\rangle$
 as in § 2.1. We set 
 $N_a:= {{\sum }}_{j=0}^{p-1}\sigma _a^{{\kern1.1pt}j}\in \mathbb Z[G]$
. For every subgroup
$N_a:= {{\sum }}_{j=0}^{p-1}\sigma _a^{{\kern1.1pt}j}\in \mathbb Z[G]$
. For every subgroup 
 $H$
 of
$H$
 of 
 $G$
, we also write
$G$
, we also write 
 $N_a$
 for the image of
$N_a$
 for the image of 
 $N_a\in \mathbb Z[G]$
 under the canonical map
$N_a\in \mathbb Z[G]$
 under the canonical map 
 $\mathbb Z[G]\to \mathbb Z[G/H]$
. We define
$\mathbb Z[G]\to \mathbb Z[G/H]$
. We define 
 $N_b$
,
$N_b$
, 
 $N_c$
 and
$N_c$
 and 
 $N_d$
 in a similar way.
$N_d$
 in a similar way.
Let
 \begin{equation*}1\to T\xrightarrow {\iota } P\xrightarrow {\pi } S\to 1 \end{equation*}
\begin{equation*}1\to T\xrightarrow {\iota } P\xrightarrow {\pi } S\to 1 \end{equation*}
be the short exact sequence of 
 $F$
-tori (4.1). It induces a short exact sequence of cocharacter
$F$
-tori (4.1). It induces a short exact sequence of cocharacter 
 $G$
-lattices
$G$
-lattices
 \begin{equation*}0\to T_*\xrightarrow {\iota _*} P_*\xrightarrow {\pi _*} S_*\to 1.\end{equation*}
\begin{equation*}0\to T_*\xrightarrow {\iota _*} P_*\xrightarrow {\pi _*} S_*\to 1.\end{equation*}
By definition of 
 $P$
 and
$P$
 and 
 $S$
,
$S$
, 
 \begin{equation*}P_*=\mathbb Z[G/\langle \sigma _b,\sigma _d\rangle ]\oplus \mathbb Z[G/\langle \sigma _a,\sigma _c\rangle ],\quad S_*=\langle N_b,N_c\rangle \subset \mathbb Z[G/\langle \sigma _a,\sigma _d\rangle ].\end{equation*}
\begin{equation*}P_*=\mathbb Z[G/\langle \sigma _b,\sigma _d\rangle ]\oplus \mathbb Z[G/\langle \sigma _a,\sigma _c\rangle ],\quad S_*=\langle N_b,N_c\rangle \subset \mathbb Z[G/\langle \sigma _a,\sigma _d\rangle ].\end{equation*}
Let 
 $X$
 be the Severi–Brauer variety associated to
$X$
 be the Severi–Brauer variety associated to 
 $(b,c)\in \operatorname {Br}(F)$
. Since
$(b,c)\in \operatorname {Br}(F)$
. Since 
 $X_K\simeq \mathbb P^{p-1}_K$
, the degree map
$X_K\simeq \mathbb P^{p-1}_K$
, the degree map 
 $\operatorname {Pic}(X_K)\to \mathbb Z$
 is an isomorphism, and so the map
$\operatorname {Pic}(X_K)\to \mathbb Z$
 is an isomorphism, and so the map 
 $\operatorname {Div}(X_K)\to \operatorname {Pic}(X_K)$
 is identified with the degree map
$\operatorname {Div}(X_K)\to \operatorname {Pic}(X_K)$
 is identified with the degree map 
 $\deg \colon \operatorname {Div}(X_K)\to \mathbb Z$
. Thus, the sequence (B.2) for the torus
$\deg \colon \operatorname {Div}(X_K)\to \mathbb Z$
. Thus, the sequence (B.2) for the torus 
 $T$
 takes the form
$T$
 takes the form
 \begin{align} 1\to T(K) \to T(K(X))\xrightarrow {\operatorname {div}}\operatorname {Div}(X_K)\otimes T_*\xrightarrow {\deg } T_*\to 0, \end{align}
\begin{align} 1\to T(K) \to T(K(X))\xrightarrow {\operatorname {div}}\operatorname {Div}(X_K)\otimes T_*\xrightarrow {\deg } T_*\to 0, \end{align}
where 
 $T_*$
 denotes the cocharacter lattice of
$T_*$
 denotes the cocharacter lattice of 
 $T$
.
$T$
.
Lemma 5.1.
- 
(1) We have  $(T_*)^G=\mathbb Z\cdot \eta$
, where $(T_*)^G=\mathbb Z\cdot \eta$
, where $\iota _*(\eta )=(N_a N_c,-N_b N_d)$
 in $\iota _*(\eta )=(N_a N_c,-N_b N_d)$
 in $(P_*)^G$
. $(P_*)^G$
.
- 
(2) If  $(b,c)\neq 0$
 in $(b,c)\neq 0$
 in $\operatorname {Br}(F)$
, the image of $\operatorname {Br}(F)$
, the image of $\deg \colon (\operatorname {Div}(X_{b,c})\otimes T_*)^G\to (T_*)^G$
 is equal to $\deg \colon (\operatorname {Div}(X_{b,c})\otimes T_*)^G\to (T_*)^G$
 is equal to $p(T_*)^G$
. $p(T_*)^G$
.
Proof.
 (1) The free 
 $\mathbb Z$
-module
$\mathbb Z$
-module 
 $(P_*)^G$
 has a basis consisting of the elements
$(P_*)^G$
 has a basis consisting of the elements 
 $(N_aN_c,0)$
 and
$(N_aN_c,0)$
 and 
 $(0,N_bN_d)$
. The map
$(0,N_bN_d)$
. The map 
 $\pi _*\colon P_* \to S_*\subset \mathbb Z[G/\langle {\sigma _a,\sigma _d}\rangle ]$
 takes
$\pi _*\colon P_* \to S_*\subset \mathbb Z[G/\langle {\sigma _a,\sigma _d}\rangle ]$
 takes 
 $(1, 0)$
 to
$(1, 0)$
 to 
 $N_b$
 and
$N_b$
 and 
 $(0, 1)$
 to
$(0, 1)$
 to 
 $N_c$
. It follows that
$N_c$
. It follows that 
 $\operatorname {Ker}(\pi _*)^G$
 is generated by
$\operatorname {Ker}(\pi _*)^G$
 is generated by 
 $(N_aN_c, -N_bN_d)$
.
$(N_aN_c, -N_bN_d)$
.
(2) By Lemma 4.1(2), the image of the composition
 \begin{equation*}\operatorname {Div}(X)\otimes T_*^G=(\operatorname {Div}(X)\otimes T_*)^G\to (\operatorname {Div}(X_{b,c})\otimes T_*)^G\xrightarrow {\operatorname {\deg }} (T_*)^G \end{equation*}
\begin{equation*}\operatorname {Div}(X)\otimes T_*^G=(\operatorname {Div}(X)\otimes T_*)^G\to (\operatorname {Div}(X_{b,c})\otimes T_*)^G\xrightarrow {\operatorname {\deg }} (T_*)^G \end{equation*}
is equal to 
 $p(T_*)^G$
. Thus, the image of the degree map contains
$p(T_*)^G$
. Thus, the image of the degree map contains 
 $p(T_*)^G$
.
$p(T_*)^G$
.
 We now show that the image of the degree map is contained in 
 $p(T_*)^G$
.
$p(T_*)^G$
.
 For every 
 $x\in X^{(1)}$
, pick
$x\in X^{(1)}$
, pick 
 $x'\in (X_{b,c})^{(1)}$
 lying over
$x'\in (X_{b,c})^{(1)}$
 lying over 
 $x$
, and write
$x$
, and write 
 $H_x$
 for the
$H_x$
 for the 
 $G$
-stabilizer of
$G$
-stabilizer of 
 $x'$
. The injective homomorphisms of
$x'$
. The injective homomorphisms of 
 $G$
-modules
$G$
-modules
 \begin{equation*}j_x\colon \mathbb Z[G/H_x]\hookrightarrow \operatorname {Div}(X_{b,c}),\quad gH_x\mapsto g(x'), \end{equation*}
\begin{equation*}j_x\colon \mathbb Z[G/H_x]\hookrightarrow \operatorname {Div}(X_{b,c}),\quad gH_x\mapsto g(x'), \end{equation*}
yield an isomorphism of 
 $G$
-modules
$G$
-modules
 \begin{equation*}\oplus _{x\in X^{(1)}}j_x\colon \oplus _{x\in X^{(1)}}\mathbb Z[G/H_x]\xrightarrow {\sim } \operatorname {Div}(X_{b,c}).\end{equation*}
\begin{equation*}\oplus _{x\in X^{(1)}}j_x\colon \oplus _{x\in X^{(1)}}\mathbb Z[G/H_x]\xrightarrow {\sim } \operatorname {Div}(X_{b,c}).\end{equation*}
To conclude, it suffices to show that the image of
 \begin{align} (T_*)^{H_x}=(\mathbb Z[G/H_x]\otimes T_*)^G\to (\operatorname {Div}(X_{b,c})\otimes T_*)^G\xrightarrow {\deg } (T_*)^G \end{align}
\begin{align} (T_*)^{H_x}=(\mathbb Z[G/H_x]\otimes T_*)^G\to (\operatorname {Div}(X_{b,c})\otimes T_*)^G\xrightarrow {\deg } (T_*)^G \end{align}
is contained in 
 $p(T_*)^G$
 for all
$p(T_*)^G$
 for all 
 $x\in X^{(1)}$
. Set
$x\in X^{(1)}$
. Set 
 $H:=H_x$
.
$H:=H_x$
.
 The composition (5.2) takes a cocharacter 
 $q\in (T_*)^{H}$
 to
$q\in (T_*)^{H}$
 to
 \begin{equation*}\deg \biggl(\mathop {{\sum }}\limits _{gH\in G/H} gx'\otimes gq \biggr)=\deg (x')\cdot N_{G/H}(q).\end{equation*}
\begin{equation*}\deg \biggl(\mathop {{\sum }}\limits _{gH\in G/H} gx'\otimes gq \biggr)=\deg (x')\cdot N_{G/H}(q).\end{equation*}
Thus, (5.2) coincides with the norm map 
 $N_{G/H}$
 times the degree of
$N_{G/H}$
 times the degree of 
 $x'$
.
$x'$
.
 Suppose that 
 $G=H$
. Then
$G=H$
. Then 
 $\deg (x')=\deg (x)$
 and, since
$\deg (x')=\deg (x)$
 and, since 
 $(b,c)\neq 0$
, the degree of
$(b,c)\neq 0$
, the degree of 
 $x$
 is divisible by
$x$
 is divisible by 
 $p$
 by Lemma 4.1(2).
$p$
 by Lemma 4.1(2).
 Suppose that 
 $G\neq H$
. Then either
$G\neq H$
. Then either 
 $\langle \sigma _a,\sigma _c\rangle$
 or
$\langle \sigma _a,\sigma _c\rangle$
 or 
 $\langle \sigma _b,\sigma _d\rangle$
 is not contained in
$\langle \sigma _b,\sigma _d\rangle$
 is not contained in 
 $H$
. Suppose that
$H$
. Suppose that 
 $\langle \sigma _b,\sigma _d\rangle$
 is not contained in
$\langle \sigma _b,\sigma _d\rangle$
 is not contained in 
 $H$
 and let
$H$
 and let 
 $N$
 be the subgroup generated by
$N$
 be the subgroup generated by 
 $H,\sigma _b,\sigma _d$
. Note that
$H,\sigma _b,\sigma _d$
. Note that 
 $H$
 is a proper subgroup of
$H$
 is a proper subgroup of 
 $N$
.
$N$
.
 The norm map 
 $N_{G/H}:(T_*)^H\to (T_*)^G$
 is the composition of the two norm maps
$N_{G/H}:(T_*)^H\to (T_*)^G$
 is the composition of the two norm maps
 \begin{equation*} (T_*)^H\xrightarrow {N_{N/H}} (T_*)^N\xrightarrow {N_{G/N}} (T_*)^G. \end{equation*}
\begin{equation*} (T_*)^H\xrightarrow {N_{N/H}} (T_*)^N\xrightarrow {N_{G/N}} (T_*)^G. \end{equation*}
Since 
 $\mathbb Z[G/\langle {\sigma _b,\sigma _d}\rangle ]^H=\mathbb Z[G/\langle {\sigma _b,\sigma _d}\rangle ]^N$
, the norm map
$\mathbb Z[G/\langle {\sigma _b,\sigma _d}\rangle ]^H=\mathbb Z[G/\langle {\sigma _b,\sigma _d}\rangle ]^N$
, the norm map 
 $(T_*)^H\to (T_*)^N$
 is multiplication by
$(T_*)^H\to (T_*)^N$
 is multiplication by 
 $[N:H]\in p\mathbb Z$
 on the first component of
$[N:H]\in p\mathbb Z$
 on the first component of 
 $T_*$
 with respect to the inclusion
$T_*$
 with respect to the inclusion 
 $\iota _*$
 of
$\iota _*$
 of 
 $T_*$
 into
$T_*$
 into 
 $P_*=\mathbb Z[G/\langle {\sigma _b,\sigma _d}\rangle ]\oplus \mathbb Z[G/\langle {\sigma _a,\sigma _c}\rangle ]$
.
$P_*=\mathbb Z[G/\langle {\sigma _b,\sigma _d}\rangle ]\oplus \mathbb Z[G/\langle {\sigma _a,\sigma _c}\rangle ]$
.
 By Lemma 5.1(1), 
 $(T_*)^G=\mathbb Z\cdot \eta$
, where
$(T_*)^G=\mathbb Z\cdot \eta$
, where 
 $\iota _*(\eta )=(N_a N_c,-N_b N_d)$
 in
$\iota _*(\eta )=(N_a N_c,-N_b N_d)$
 in 
 $(P_*)^G$
. Since
$(P_*)^G$
. Since 
 $N_a N_c$
 is not divisible by
$N_a N_c$
 is not divisible by 
 $p$
 in
$p$
 in 
 $\mathbb Z[G/\langle {\sigma _b,\sigma _d}\rangle ]$
, the image of (5.2) is contained in
$\mathbb Z[G/\langle {\sigma _b,\sigma _d}\rangle ]$
, the image of (5.2) is contained in 
 $p\mathbb Z\cdot \eta =p(T_*)^G$
, as desired. The proof in the case when
$p\mathbb Z\cdot \eta =p(T_*)^G$
, as desired. The proof in the case when 
 $\langle{\sigma_a,\sigma_c}\rangle $
 is not contained in H is entirely analogous.
$\langle{\sigma_a,\sigma_c}\rangle $
 is not contained in H is entirely analogous.
We write
 \begin{equation*}\overline {\eta } \in \operatorname {Coker}[(\operatorname {Div}(X_{b,c})\otimes T_*)^G\xrightarrow {\deg } (T_*)^G] \end{equation*}
\begin{equation*}\overline {\eta } \in \operatorname {Coker}[(\operatorname {Div}(X_{b,c})\otimes T_*)^G\xrightarrow {\deg } (T_*)^G] \end{equation*}
for the coset of the generator 
 $\eta \in (T_*)^G$
 appearing in Lemma 5.1(1). If
$\eta \in (T_*)^G$
 appearing in Lemma 5.1(1). If 
 $(b,c)\neq 0$
, then we have
$(b,c)\neq 0$
, then we have 
 $\overline {\eta }\neq 0$
 by Lemma 5.1(2). We consider the subgroup of unramified torsors
$\overline {\eta }\neq 0$
 by Lemma 5.1(2). We consider the subgroup of unramified torsors
 \begin{equation*}H^1(G,T(K(X)))_{\operatorname {nr}}:= \operatorname {Ker}[H^1(G,T(K(X)))\xrightarrow {\operatorname {div}}H^1(G,\operatorname {Div}(X_K\otimes T_*))]\end{equation*}
\begin{equation*}H^1(G,T(K(X)))_{\operatorname {nr}}:= \operatorname {Ker}[H^1(G,T(K(X)))\xrightarrow {\operatorname {div}}H^1(G,\operatorname {Div}(X_K\otimes T_*))]\end{equation*}
and the homomorphism
 \begin{equation*}\theta \colon H^1(G,T(K(X)))_{\operatorname {nr}}\to \operatorname {Coker}[(\operatorname {Div}(X_K)\otimes T_*)^G\xrightarrow {\deg } (T_*)^G],\end{equation*}
\begin{equation*}\theta \colon H^1(G,T(K(X)))_{\operatorname {nr}}\to \operatorname {Coker}[(\operatorname {Div}(X_K)\otimes T_*)^G\xrightarrow {\deg } (T_*)^G],\end{equation*}
which are defined in (B.3).
Lemma 5.2. 
Let 
 $b,c\in F^\times$
 be such that
$b,c\in F^\times$
 be such that 
 $(b,c)\neq 0$
 in
$(b,c)\neq 0$
 in 
 $\operatorname {Br}(F)$
 and let
$\operatorname {Br}(F)$
 and let 
 $w\in F_{b,c}(X)^{\times }$
 be such that
$w\in F_{b,c}(X)^{\times }$
 be such that 
 $(\sigma _b-1)(\sigma _c-1)w=\zeta$
 and
$(\sigma _b-1)(\sigma _c-1)w=\zeta$
 and 
 $\operatorname {div}(w)=x-y$
, where
$\operatorname {div}(w)=x-y$
, where 
 $\deg (x)=\deg (y)=1$
 and
$\deg (x)=\deg (y)=1$
 and 
 $\sigma _b(x)=x$
 and
$\sigma _b(x)=x$
 and 
 $\sigma _c(y)=y$
. Let
$\sigma _c(y)=y$
. Let 
 $E_w\subset P_{F(X)}$
 be the
$E_w\subset P_{F(X)}$
 be the 
 $T_{F(X)}$
-torsor given by the equation
$T_{F(X)}$
-torsor given by the equation 
 $N_a(u)N_d(v)=w^p$
, and write
$N_a(u)N_d(v)=w^p$
, and write 
 $[E_w]$
 for the class of
$[E_w]$
 for the class of 
 $E_w$
 in
$E_w$
 in 
 $H^1(G,T(K(X)))$
.
$H^1(G,T(K(X)))$
.
- 
(1) We have  $[E_w]\in H^1(G,T(K(X)))_{\operatorname {nr}}$
. $[E_w]\in H^1(G,T(K(X)))_{\operatorname {nr}}$
.
- 
(2) Let  $\theta$
 be the homomorphism of (
B.3
). We have $\theta$
 be the homomorphism of (
B.3
). We have $\theta ([E_w])=-\overline {\eta }\neq 0$
. $\theta ([E_w])=-\overline {\eta }\neq 0$
.
Proof.
 The 
 $F$
-tori
$F$
-tori 
 $T$
,
$T$
, 
 $P$
 and
$P$
 and 
 $S$
 of (4.1) are split by
$S$
 of (4.1) are split by 
 $K=F_{a,b,c,d}$
. Therefore, we may consider diagram (B.6) for the short exact sequence (4.1), the splitting field
$K=F_{a,b,c,d}$
. Therefore, we may consider diagram (B.6) for the short exact sequence (4.1), the splitting field 
 $K/F$
 and the Severi–Brauer variety
$K/F$
 and the Severi–Brauer variety 
 $X$
 of
$X$
 of 
 $(b,c)$
 over
$(b,c)$
 over 
 $F$
.
$F$
.

 Since 
 $(\sigma _b-1)(\sigma _c-1)w^p=1$
, we have
$(\sigma _b-1)(\sigma _c-1)w^p=1$
, we have 
 $w^p\in S(F(X))$
. The image of
$w^p\in S(F(X))$
. The image of 
 $w^p$
 under
$w^p$
 under 
 $\partial$
 is equal to
$\partial$
 is equal to 
 $[E_w]\in H^1(G,T(K(X)))$
.
$[E_w]\in H^1(G,T(K(X)))$
.
 Let 
 $H\subset G$
 be the subgroup generated by
$H\subset G$
 be the subgroup generated by 
 $\sigma _a$
 and
$\sigma _a$
 and 
 $\sigma _d$
. The canonical isomorphism
$\sigma _d$
. The canonical isomorphism
 \begin{equation*} \operatorname {Div}(X_{b,c})=\operatorname {Div}(X_K)^H=(\operatorname {Div}(X_K)\otimes \mathbb Z[G/H])^G \end{equation*}
\begin{equation*} \operatorname {Div}(X_{b,c})=\operatorname {Div}(X_K)^H=(\operatorname {Div}(X_K)\otimes \mathbb Z[G/H])^G \end{equation*}
sends the divisor 
 $\operatorname {div}(w)=x-y$
 to
$\operatorname {div}(w)=x-y$
 to 
 $ \sum_{i,j=0}^{p-1} \sigma _b^i\sigma _c^{{\kern1.1pt}j} (x-y)\otimes \sigma _b^i\sigma _c^{{\kern1.1pt}j}$
. Therefore, the element
$ \sum_{i,j=0}^{p-1} \sigma _b^i\sigma _c^{{\kern1.1pt}j} (x-y)\otimes \sigma _b^i\sigma _c^{{\kern1.1pt}j}$
. Therefore, the element 
 $\operatorname {div}(w^p)$
 in
$\operatorname {div}(w^p)$
 in 
 $(\operatorname {Div}(X_K)\otimes S_*)^G\subset (\operatorname {Div}(X_K)\otimes \mathbb Z[G/H])^G$
 is equal to
$(\operatorname {Div}(X_K)\otimes S_*)^G\subset (\operatorname {Div}(X_K)\otimes \mathbb Z[G/H])^G$
 is equal to
 \begin{equation*} e:=p\mathop {{\sum }}\limits _{i,j=0}^{p-1}(\sigma _b^i\sigma _c^{{\kern1.1pt}j}(x-y)\otimes \sigma _b^i\sigma _c^{{\kern1.1pt}j})= p\mathop {{\sum }}\limits _{j=0}^{p-1}(\sigma _c^{{\kern1.1pt}j} x \otimes \sigma _c^{{\kern1.1pt}j} N_b)-p\mathop {{\sum }}\limits _{i=0}^{p-1}(\sigma _b^i y \otimes \sigma _b^i N_c). \end{equation*}
\begin{equation*} e:=p\mathop {{\sum }}\limits _{i,j=0}^{p-1}(\sigma _b^i\sigma _c^{{\kern1.1pt}j}(x-y)\otimes \sigma _b^i\sigma _c^{{\kern1.1pt}j})= p\mathop {{\sum }}\limits _{j=0}^{p-1}(\sigma _c^{{\kern1.1pt}j} x \otimes \sigma _c^{{\kern1.1pt}j} N_b)-p\mathop {{\sum }}\limits _{i=0}^{p-1}(\sigma _b^i y \otimes \sigma _b^i N_c). \end{equation*}
Since 
 $S_*$
 is the G-sublattice of
$S_*$
 is the G-sublattice of 
 $\mathbb Z[G/\langle \sigma _a,\sigma _d\rangle ]$
 generated by
$\mathbb Z[G/\langle \sigma _a,\sigma _d\rangle ]$
 generated by 
 $N_b$
 and
$N_b$
 and 
 $N_c$
, this implies that
$N_c$
, this implies that 
 $e$
 belongs to
$e$
 belongs to 
 $(\operatorname {Div}(X_K)\otimes S_*)^G$
. Then
$(\operatorname {Div}(X_K)\otimes S_*)^G$
. Then 
 $e=\pi _*(f)$
, where
$e=\pi _*(f)$
, where
 \begin{equation*} f:=\mathop {{\sum }}\limits _{j=0}^{p-1}(\sigma _c^{{\kern1.1pt}j} x \otimes \sigma _c^{{\kern1.1pt}j} N_a)-\mathop {{\sum }}\limits _{i=0}^{p-1}(\sigma _b^i y \otimes \sigma _b^i N_d)\in (\operatorname {Div}(X_K)\otimes P_*)^G. \end{equation*}
\begin{equation*} f:=\mathop {{\sum }}\limits _{j=0}^{p-1}(\sigma _c^{{\kern1.1pt}j} x \otimes \sigma _c^{{\kern1.1pt}j} N_a)-\mathop {{\sum }}\limits _{i=0}^{p-1}(\sigma _b^i y \otimes \sigma _b^i N_d)\in (\operatorname {Div}(X_K)\otimes P_*)^G. \end{equation*}
It follows that 
 $\operatorname {div}([E_w])=\partial (e)=\partial (\pi _*(f))=0$
, which proves (1).
$\operatorname {div}([E_w])=\partial (e)=\partial (\pi _*(f))=0$
, which proves (1).
 Moreover, since 
 $\deg (x)=\deg (y)=1$
, we have
$\deg (x)=\deg (y)=1$
, we have
 \begin{equation*} \deg (f)=(N_a N_c, - N_b N_d)=\iota _*(\eta )\quad \text {in $(P_*)^G$.} \end{equation*}
\begin{equation*} \deg (f)=(N_a N_c, - N_b N_d)=\iota _*(\eta )\quad \text {in $(P_*)^G$.} \end{equation*}
In view of (B.7), this implies that 
 $\theta ([E_w])=-\overline {\eta }$
. We know from Lemma 5.1(2) that
$\theta ([E_w])=-\overline {\eta }$
. We know from Lemma 5.1(2) that 
 $\overline {\eta }\neq 0$
. This completes the proof of (2).
$\overline {\eta }\neq 0$
. This completes the proof of (2).
Proof of Theorem
 
1.3
. Replacing 
 $F$
 by a finite extension, if necessary, we may suppose that
$F$
 by a finite extension, if necessary, we may suppose that 
 $F$
 contains a primitive
$F$
 contains a primitive 
 $p$
th root of unity
$p$
th root of unity 
 $\zeta$
. Let
$\zeta$
. Let 
 $E:= F(x,y)$
, where
$E:= F(x,y)$
, where 
 $x$
 and
$x$
 and 
 $y$
 are independent variables over
$y$
 are independent variables over 
 $F$
, let
$F$
, let 
 $X$
 be the Severi–Brauer variety of the degree-
$X$
 be the Severi–Brauer variety of the degree-
 $p$
 cyclic algebra
$p$
 cyclic algebra 
 $(x,y)$
 over
$(x,y)$
 over 
 $E$
 and let
$E$
 and let 
 $L:= E(X)$
. Consider the following elements of
$L:= E(X)$
. Consider the following elements of 
 $E^\times$
:
$E^\times$
:
 \begin{equation*}a:= 1-x,\quad b:= x,\quad c:= y,\quad d:= 1-y.\end{equation*}
\begin{equation*}a:= 1-x,\quad b:= x,\quad c:= y,\quad d:= 1-y.\end{equation*}
We have 
 $(a,b)=(c,d)=0$
 in
$(a,b)=(c,d)=0$
 in 
 $\operatorname {Br}(E)$
 by the Steinberg relations [Reference Serre and GreenbergSer79, Chapter XIV, Proposition 4(iv)], and hence
$\operatorname {Br}(E)$
 by the Steinberg relations [Reference Serre and GreenbergSer79, Chapter XIV, Proposition 4(iv)], and hence 
 $(a,b)=(b,c)=0$
 in
$(a,b)=(b,c)=0$
 in 
 $\operatorname {Br}(L)$
. Moreover,
$\operatorname {Br}(L)$
. Moreover, 
 $(b,c)\neq 0$
 in
$(b,c)\neq 0$
 in 
 $\operatorname {Br}(E)$
 because the residue of
$\operatorname {Br}(E)$
 because the residue of 
 $(b,c)$
 along
$(b,c)$
 along 
 $x=0$
 is non-zero, whereas
$x=0$
 is non-zero, whereas 
 $(b,c)=0$
 in
$(b,c)=0$
 in 
 $\operatorname {Br}(L)$
 by [Reference Gille and SzamuelyGS17, Theorem 5.4.1]. Thus,
$\operatorname {Br}(L)$
 by [Reference Gille and SzamuelyGS17, Theorem 5.4.1]. Thus, 
 $(a,b)=(b,c)=(c,d)=0$
 in
$(a,b)=(b,c)=(c,d)=0$
 in 
 $\operatorname {Br}(L)$
.
$\operatorname {Br}(L)$
.
 Consider the sequence of tori (4.1) over the ground field 
 $E$
 associated to the scalars
$E$
 associated to the scalars 
 $a,b,c,d\in E^\times$
 chosen above:
$a,b,c,d\in E^\times$
 chosen above:
 \begin{equation*}1\to T\to P\to S\to 1.\end{equation*}
\begin{equation*}1\to T\to P\to S\to 1.\end{equation*}
Let 
 $w\in L_{b,c}(X)$
 be as in Lemma 4.1(3), and let
$w\in L_{b,c}(X)$
 be as in Lemma 4.1(3), and let 
 $E_w\subset P_L$
 be the
$E_w\subset P_L$
 be the 
 $T_L$
-torsor given by the equation
$T_L$
-torsor given by the equation 
 $N_a(u)N_d(v)=w^p$
. By Lemma 5.2(2), the torsor
$N_a(u)N_d(v)=w^p$
. By Lemma 5.2(2), the torsor 
 $E_w$
 is non-trivial over
$E_w$
 is non-trivial over 
 $L$
. Now Corollary 4.5 implies that the Massey product
$L$
. Now Corollary 4.5 implies that the Massey product 
 $\langle a,b,c,d\rangle$
 is not defined over
$\langle a,b,c,d\rangle$
 is not defined over 
 $L$
. In particular, by Lemma 2.3, the differential graded ring
$L$
. In particular, by Lemma 2.3, the differential graded ring 
 $C^{*}(\Gamma _L,\mathbb Z/p\mathbb Z)$
 is not formal.
$C^{*}(\Gamma _L,\mathbb Z/p\mathbb Z)$
 is not formal.
Appendix A. Homological algebra
 Let 
 $G$
 be a profinite group, and let
$G$
 be a profinite group, and let
 \begin{align} 0\to A_0\xrightarrow {\alpha _0} A_1\xrightarrow {\alpha _1} A_2\xrightarrow {\alpha _2} A_3\to 0 \end{align}
\begin{align} 0\to A_0\xrightarrow {\alpha _0} A_1\xrightarrow {\alpha _1} A_2\xrightarrow {\alpha _2} A_3\to 0 \end{align}
be an exact sequence of discrete 
 $G$
-modules. We break (A.1) into two short exact sequences
$G$
-modules. We break (A.1) into two short exact sequences
 \begin{equation*}0\to A_0\xrightarrow {\alpha _0} A_1\to A\to 0,\end{equation*}
\begin{equation*}0\to A_0\xrightarrow {\alpha _0} A_1\to A\to 0,\end{equation*}
 \begin{equation*}0\to A\to A_2\xrightarrow {\alpha _2} A_3\to 0.\end{equation*}
\begin{equation*}0\to A\to A_2\xrightarrow {\alpha _2} A_3\to 0.\end{equation*}
We obtain a homomorphism
 \begin{align} \theta \colon \operatorname {Ker}[H^1(G,A_1)\xrightarrow {\alpha _1} H^1(G,A_2)]\to \operatorname {Coker}[A_2^G\xrightarrow {\alpha _2} A_3^G], \end{align}
\begin{align} \theta \colon \operatorname {Ker}[H^1(G,A_1)\xrightarrow {\alpha _1} H^1(G,A_2)]\to \operatorname {Coker}[A_2^G\xrightarrow {\alpha _2} A_3^G], \end{align}
which is defined as the composition of the map
 \begin{equation*}\operatorname {Ker}[H^1(G,A_1)\xrightarrow {\alpha _1} H^1(G,A_2)]\to \operatorname {Ker}[H^1(G,A)\to H^1(G,A_2)], \end{equation*}
\begin{equation*}\operatorname {Ker}[H^1(G,A_1)\xrightarrow {\alpha _1} H^1(G,A_2)]\to \operatorname {Ker}[H^1(G,A)\to H^1(G,A_2)], \end{equation*}
and the inverse of the isomorphism
 \begin{align} \operatorname {Coker}[A_2^G\xrightarrow {\alpha _2} A_3^G]\xrightarrow {\sim } \operatorname {Ker}[H^1(G,A)\to H^1(G,A_2)], \end{align}
\begin{align} \operatorname {Coker}[A_2^G\xrightarrow {\alpha _2} A_3^G]\xrightarrow {\sim } \operatorname {Ker}[H^1(G,A)\to H^1(G,A_2)], \end{align}
which is induced by the connecting homomorphism 
 $A_3^G\to H^1(G,A)$
.
$A_3^G\to H^1(G,A)$
.
Lemma A.1. We have an exact sequence
 \begin{align*} H^1(G,A_0)\xrightarrow {\alpha _0} \operatorname {Ker}[H^1(G,A_1)\xrightarrow {\alpha _1}H^1(G,A_2)]\xrightarrow {\theta } \operatorname {Coker}[A_2^G\to A_3^G]\to H^2(G,A_0), \end{align*}
\begin{align*} H^1(G,A_0)\xrightarrow {\alpha _0} \operatorname {Ker}[H^1(G,A_1)\xrightarrow {\alpha _1}H^1(G,A_2)]\xrightarrow {\theta } \operatorname {Coker}[A_2^G\to A_3^G]\to H^2(G,A_0), \end{align*}
where the last map is defined as the composition of (A.3) and the connecting homomorphism 
 $H^1(G,A)\to H^2(G,A_0)$
.
$H^1(G,A)\to H^2(G,A_0)$
.
Proof.
 The proof follows from the definition of 
 $\theta$
 and the exactness of (A.1).
$\theta$
 and the exactness of (A.1).
Consider a commutative diagram of discrete 
 $G$
-modules with exact rows and columns.
$G$
-modules with exact rows and columns.

It yields a commutative diagram of abelian groups where the columns are exact and the rows are complexes.

 Suppose that the connecting homomorphism 
 $\partial _1\colon C_1^G\to H^1(G,A_1)$
 is surjective. We define a function
$\partial _1\colon C_1^G\to H^1(G,A_1)$
 is surjective. We define a function
 \begin{equation*}\theta '\colon \operatorname {Ker}[H^1(G,A_1)\xrightarrow {\alpha _1} H^1(G,A_2)]\to \operatorname {Coker}(A_2^G\xrightarrow {\alpha _2} A_3^G) \end{equation*}
\begin{equation*}\theta '\colon \operatorname {Ker}[H^1(G,A_1)\xrightarrow {\alpha _1} H^1(G,A_2)]\to \operatorname {Coker}(A_2^G\xrightarrow {\alpha _2} A_3^G) \end{equation*}
as follows. Let 
 $z\in H^1(G,A_1)$
 such that
$z\in H^1(G,A_1)$
 such that 
 $\alpha _1(z)=0$
 in
$\alpha _1(z)=0$
 in 
 $H^1(G,A_2)$
. By assumption, there exists
$H^1(G,A_2)$
. By assumption, there exists 
 $c_1\in C_1^G$
 such that
$c_1\in C_1^G$
 such that 
 $\partial _1(c_1)=z$
. By the exactness of the second column, there exists
$\partial _1(c_1)=z$
. By the exactness of the second column, there exists 
 $b_2\in B_2^G$
 such that
$b_2\in B_2^G$
 such that 
 $\pi _2(b_2)=\gamma _1(c_1)$
. By the exactness of the third column and the injectivity of
$\pi _2(b_2)=\gamma _1(c_1)$
. By the exactness of the third column and the injectivity of 
 $\iota _3$
, there exists a unique element
$\iota _3$
, there exists a unique element 
 $a_3\in A_3^G$
 such that
$a_3\in A_3^G$
 such that 
 $\beta _2(b_2)=\iota _3(a_3)$
. We set
$\beta _2(b_2)=\iota _3(a_3)$
. We set
 \begin{equation*}\theta '(z):= a_3+\alpha _2(A_2^G).\end{equation*}
\begin{equation*}\theta '(z):= a_3+\alpha _2(A_2^G).\end{equation*}
A diagram chase shows that 
 $\theta '$
 is a well-defined homomorphism.
$\theta '$
 is a well-defined homomorphism.
Lemma A.2. 
Let 
 $G$
 be a profinite group, and suppose that we are given an exact sequence (
A.1
) and a commutative diagram (
A.4
) such that the connecting homomorphism
$G$
 be a profinite group, and suppose that we are given an exact sequence (
A.1
) and a commutative diagram (
A.4
) such that the connecting homomorphism 
 $\partial _1\colon C_1^G\to H^1(G,A_1)$
 is surjective. Then
$\partial _1\colon C_1^G\to H^1(G,A_1)$
 is surjective. Then 
 $\theta =-\theta '$
.
$\theta =-\theta '$
.
Proof.
 Let 
 $z\in H^1(G,A_1)$
 be such that
$z\in H^1(G,A_1)$
 be such that 
 $\alpha _1(z)=0$
 in
$\alpha _1(z)=0$
 in 
 $H^1(G,A_2)$
. Since the map
$H^1(G,A_2)$
. Since the map 
 $\partial _1\colon C_1^G\to H^1(G,A_1)$
 is surjective, there exists
$\partial _1\colon C_1^G\to H^1(G,A_1)$
 is surjective, there exists 
 $c_1\in C_1^G$
 such that
$c_1\in C_1^G$
 such that 
 $\partial _1(c_1)=z$
. Let
$\partial _1(c_1)=z$
. Let 
 $b_1\in B_1$
 be such that
$b_1\in B_1$
 be such that 
 $\pi _1(b_1)=c_1$
, and, for all
$\pi _1(b_1)=c_1$
, and, for all 
 $g\in G$
, let
$g\in G$
, let 
 $a_{1g}$
 be the unique element of
$a_{1g}$
 be the unique element of 
 $A_1$
 such that
$A_1$
 such that 
 $\iota _1(a_{1g})=gb_1-b_1$
. Then
$\iota _1(a_{1g})=gb_1-b_1$
. Then 
 $\partial _1(c_1)$
 is represented by the
$\partial _1(c_1)$
 is represented by the 
 $1$
-cocycle
$1$
-cocycle 
 $\left \{a_{1g}\right \}_{g\in G}$
.
$\left \{a_{1g}\right \}_{g\in G}$
.
 Define 
 $b_2:= \beta _1(b_1)$
 and
$b_2:= \beta _1(b_1)$
 and 
 $c_2:= \gamma _1(c_1)$
, so that
$c_2:= \gamma _1(c_1)$
, so that 
 $\pi _2(b_2)=c_2$
. Since
$\pi _2(b_2)=c_2$
. Since 
 $\alpha _1(z)=0$
 is represented by the cocycle
$\alpha _1(z)=0$
 is represented by the cocycle 
 $\left \{\alpha _1(a_{1g})\right \}_{g\in G}$
, we deduce that there exists
$\left \{\alpha _1(a_{1g})\right \}_{g\in G}$
, we deduce that there exists 
 $a_2\in A_2$
 such that
$a_2\in A_2$
 such that 
 $\alpha _1(a_{1g})=ga_2-a_2$
 for all
$\alpha _1(a_{1g})=ga_2-a_2$
 for all 
 $g\in G$
. It follows that
$g\in G$
. It follows that 
 $gb_2-b_2=\iota _2(ga_2-a_2)$
 for all
$gb_2-b_2=\iota _2(ga_2-a_2)$
 for all 
 $g\in G$
, that is,
$g\in G$
, that is, 
 $b_2-\iota _2(a_2)$
 belongs to
$b_2-\iota _2(a_2)$
 belongs to 
 $B_2^G$
. Moreover,
$B_2^G$
. Moreover,
 \begin{equation*}\pi _2(b_2-\iota _2(a_2))=\pi _2(b_2)=\gamma _1(c_1).\end{equation*}
\begin{equation*}\pi _2(b_2-\iota _2(a_2))=\pi _2(b_2)=\gamma _1(c_1).\end{equation*}
Finally,
 \begin{equation*}\beta _2(b_2-\iota _2(a_2))=\beta _2(\beta _1(b_1))-\iota _3(\alpha _2(a_2))=\iota _3(-\alpha _2(a_2)).\end{equation*}
\begin{equation*}\beta _2(b_2-\iota _2(a_2))=\beta _2(\beta _1(b_1))-\iota _3(\alpha _2(a_2))=\iota _3(-\alpha _2(a_2)).\end{equation*}
By definition, 
 $\theta '(z)=-\alpha _2(a_2)+\alpha _2(A_2^G)$
. Observe that
$\theta '(z)=-\alpha _2(a_2)+\alpha _2(A_2^G)$
. Observe that 
 $\alpha _2(a_2)$
 belongs to
$\alpha _2(a_2)$
 belongs to 
 $A_3^G$
 because, for every
$A_3^G$
 because, for every 
 $g\in G$
,
$g\in G$
,
 \begin{equation*}g\alpha _2(a_2)-\alpha _2(a_2)=\alpha _2(ga_2-a_2)=\alpha _2(\alpha _1(a_{1g}))=0.\end{equation*}
\begin{equation*}g\alpha _2(a_2)-\alpha _2(a_2)=\alpha _2(ga_2-a_2)=\alpha _2(\alpha _1(a_{1g}))=0.\end{equation*}
For all 
 $g\in G$
, let
$g\in G$
, let 
 $a_g\in A$
 be the image of
$a_g\in A$
 be the image of 
 $a_{1g}$
. The homomorphism
$a_{1g}$
. The homomorphism
 \begin{equation*}\operatorname {Ker}[H^1(G,A_1)\xrightarrow {\alpha _1} H^1(G,A_2)]\to \operatorname {Ker}[H^1(G,A)\to H^1(G,A_2)], \end{equation*}
\begin{equation*}\operatorname {Ker}[H^1(G,A_1)\xrightarrow {\alpha _1} H^1(G,A_2)]\to \operatorname {Ker}[H^1(G,A)\to H^1(G,A_2)], \end{equation*}
induced by the map 
 $A_1\to A$
, sends the class of
$A_1\to A$
, sends the class of 
 $\left \{a_{1g}\right \}_{g\in G}$
 to the class of
$\left \{a_{1g}\right \}_{g\in G}$
 to the class of 
 $\left \{a_g\right \}_{g\in G}$
.
$\left \{a_g\right \}_{g\in G}$
.
 The element 
 $a_2\in A_2$
 is a lift of
$a_2\in A_2$
 is a lift of 
 $\alpha _2(a_2)$
. As
$\alpha _2(a_2)$
. As 
 $ga_2-a_2=\alpha _1(a_{1g})$
 for all
$ga_2-a_2=\alpha _1(a_{1g})$
 for all 
 $g\in G$
, the injective map
$g\in G$
, the injective map 
 $A\to A_2$
 sends
$A\to A_2$
 sends 
 $a_g$
 to
$a_g$
 to 
 $ga_2-a_2$
 for all
$ga_2-a_2$
 for all 
 $g\in G$
. Therefore, the connecting map
$g\in G$
. Therefore, the connecting map 
 $A_3^G\to H^1(G,A)$
 sends
$A_3^G\to H^1(G,A)$
 sends 
 $\alpha _2(a_2)$
 to the class of
$\alpha _2(a_2)$
 to the class of 
 $\left \{a_g\right \}_{g\in G}$
. It follows that the isomorphism
$\left \{a_g\right \}_{g\in G}$
. It follows that the isomorphism
 \begin{equation*}\operatorname {Coker}[A_2^G\xrightarrow {\alpha _2} A_3^G]\xrightarrow {\sim } \operatorname {Ker}[H^1(G,A)\to H^1(G,A_2)], \end{equation*}
\begin{equation*}\operatorname {Coker}[A_2^G\xrightarrow {\alpha _2} A_3^G]\xrightarrow {\sim } \operatorname {Ker}[H^1(G,A)\to H^1(G,A_2)], \end{equation*}
induced by 
 $A_3^G\to H^1(G,A)$
, sends
$A_3^G\to H^1(G,A)$
, sends 
 $\alpha _2(a_2)+\alpha _2(A_2^G)$
 to the class of
$\alpha _2(a_2)+\alpha _2(A_2^G)$
 to the class of 
 $\left \{a_g\right \}_{g\in G}$
. By the definition of
$\left \{a_g\right \}_{g\in G}$
. By the definition of 
 $\theta$
, we conclude that
$\theta$
, we conclude that 
 $\theta (z)=\alpha _2(a_2)+\alpha _2(A_2^G)=-\theta '(z)$
.
$\theta (z)=\alpha _2(a_2)+\alpha _2(A_2^G)=-\theta '(z)$
.
Appendix B. Unramified torsors under tori
 Let 
 $F$
 be a field, let
$F$
 be a field, let 
 $X$
 be a smooth projective geometrically connected
$X$
 be a smooth projective geometrically connected 
 $F$
-variety, let
$F$
-variety, let 
 $K$
 be a Galois extension of
$K$
 be a Galois extension of 
 $F$
 (possibly of infinite degree over
$F$
 (possibly of infinite degree over 
 $F$
) and let
$F$
) and let 
 $G:= \operatorname {Gal}(K/F)$
. We have an exact sequence of discrete
$G:= \operatorname {Gal}(K/F)$
. We have an exact sequence of discrete 
 $G$
-modules
$G$
-modules
 \begin{align} 1\to K^\times \to K(X)^\times \xrightarrow {\operatorname {div}} \operatorname {Div}(X_K)\xrightarrow {\lambda } \operatorname {Pic}(X_K)\to 0, \end{align}
\begin{align} 1\to K^\times \to K(X)^\times \xrightarrow {\operatorname {div}} \operatorname {Div}(X_K)\xrightarrow {\lambda } \operatorname {Pic}(X_K)\to 0, \end{align}
where 
 $\operatorname {div}$
 takes a non-zero rational function
$\operatorname {div}$
 takes a non-zero rational function 
 $f\in K(X)^\times$
 to its divisor and
$f\in K(X)^\times$
 to its divisor and 
 $\lambda$
 takes a divisor on
$\lambda$
 takes a divisor on 
 $X_K$
 to its class in
$X_K$
 to its class in 
 $\operatorname {Pic}(X_K)$
.
$\operatorname {Pic}(X_K)$
.
 Let 
 $T$
 be an
$T$
 be an 
 $F$
-torus split by
$F$
-torus split by 
 $K$
. Write
$K$
. Write 
 $T_*$
 for the cocharacter lattice of
$T_*$
 for the cocharacter lattice of 
 $T$
: it is a finitely generated
$T$
: it is a finitely generated 
 $\mathbb Z$
-free
$\mathbb Z$
-free 
 $G$
-module. Tensoring (B.1) with
$G$
-module. Tensoring (B.1) with 
 $T_*$
, we obtain an exact sequence of
$T_*$
, we obtain an exact sequence of 
 $G$
-modules
$G$
-modules
 \begin{align} 1\to T(K) \to T(K(X))\xrightarrow {\operatorname {div}} \operatorname {Div}(X_K)\otimes T_*\xrightarrow {\lambda } \operatorname {Pic}(X_K)\otimes T_*\to 0, \end{align}
\begin{align} 1\to T(K) \to T(K(X))\xrightarrow {\operatorname {div}} \operatorname {Div}(X_K)\otimes T_*\xrightarrow {\lambda } \operatorname {Pic}(X_K)\otimes T_*\to 0, \end{align}
where we have used the fact that 
 $K^\times \otimes T_*=T(K)$
.
$K^\times \otimes T_*=T(K)$
.
We define the subgroup of unramified torsors
 \begin{equation*}H^1(G,T(K(X)))_{\operatorname {nr}}:= \operatorname {Ker}[H^1(G,T(K(X)))\xrightarrow {\operatorname {div}}H^1(G,\operatorname {Div}(X_K\otimes T_*))].\end{equation*}
\begin{equation*}H^1(G,T(K(X)))_{\operatorname {nr}}:= \operatorname {Ker}[H^1(G,T(K(X)))\xrightarrow {\operatorname {div}}H^1(G,\operatorname {Div}(X_K\otimes T_*))].\end{equation*}
The sequence (B.1) is a special case of (A.1). In this case, the map 
 $\theta$
 of (A.1) takes the form
$\theta$
 of (A.1) takes the form
 \begin{align} \theta \colon H^1(G,T(K(X)))_{\operatorname {nr}}\to \operatorname {Coker}[(\operatorname {Div}(X_K)\otimes T_*)^G\xrightarrow {\lambda }(\operatorname {Pic}(X_K)\otimes T_*)^G]. \end{align}
\begin{align} \theta \colon H^1(G,T(K(X)))_{\operatorname {nr}}\to \operatorname {Coker}[(\operatorname {Div}(X_K)\otimes T_*)^G\xrightarrow {\lambda }(\operatorname {Pic}(X_K)\otimes T_*)^G]. \end{align}
Proposition B.1. We have an exact sequence
 \begin{align*} H^1(G, T(K)) \to H^1(G, T(K(X)))_{\operatorname {nr}} & \xrightarrow {\theta } \operatorname {Coker} [(\operatorname {Div}(X_K)\otimes T_*)^G \xrightarrow {\lambda }(\operatorname {Pic}(X_K)\otimes T_*)^G ] \\ &\to H^2(G, T(K)), \end{align*}
\begin{align*} H^1(G, T(K)) \to H^1(G, T(K(X)))_{\operatorname {nr}} & \xrightarrow {\theta } \operatorname {Coker} [(\operatorname {Div}(X_K)\otimes T_*)^G \xrightarrow {\lambda }(\operatorname {Pic}(X_K)\otimes T_*)^G ] \\ &\to H^2(G, T(K)), \end{align*}
where the first map and the last map are induced by (B.2).
Proof. This is a special case of Lemma A.1.
 By Lemma A.2, the map 
 $\theta$
 may be computed as follows. Let
$\theta$
 may be computed as follows. Let
 \begin{align} 1\to T\xrightarrow {\iota } P\xrightarrow {\pi } S\to 1 \end{align}
\begin{align} 1\to T\xrightarrow {\iota } P\xrightarrow {\pi } S\to 1 \end{align}
be a short exact sequence of 
 $F$
-tori split by
$F$
-tori split by 
 $K$
 such that
$K$
 such that 
 $P$
 is a quasi-trivial torus. Passing to cocharacter lattices, we obtain a short exact sequence of
$P$
 is a quasi-trivial torus. Passing to cocharacter lattices, we obtain a short exact sequence of 
 $G$
-modules
$G$
-modules
 \begin{align} 0\to T_*\xrightarrow {\iota _*} P_*\xrightarrow {\pi _*} S_*\to 0. \end{align}
\begin{align} 0\to T_*\xrightarrow {\iota _*} P_*\xrightarrow {\pi _*} S_*\to 0. \end{align}
We tensor (B.1) with 
 $T_*$
,
$T_*$
, 
 $P_*$
 and
$P_*$
 and 
 $S_*$
, respectively, and pass to group cohomology to obtain the following commutative diagram, where the columns are exact and the rows are complexes.
$S_*$
, respectively, and pass to group cohomology to obtain the following commutative diagram, where the columns are exact and the rows are complexes.

Note that 
 $\operatorname {Gal}(K(X)/F(X))=G$
. Therefore,
$\operatorname {Gal}(K(X)/F(X))=G$
. Therefore, 
 $H^1(G,P(K(X)))$
 is trivial, and hence
$H^1(G,P(K(X)))$
 is trivial, and hence 
 $\partial \colon S(F(X))\to H^1(G,T(K(X)))$
 is surjective.
$\partial \colon S(F(X))\to H^1(G,T(K(X)))$
 is surjective.
 Let 
 $\tau \in H^1(G, T(K(X)))_{\operatorname {nr}}$
 and choose
$\tau \in H^1(G, T(K(X)))_{\operatorname {nr}}$
 and choose 
 $\sigma \in S(F(X))$
 such that
$\sigma \in S(F(X))$
 such that 
 $\partial (\sigma )=\tau$
. Then pick
$\partial (\sigma )=\tau$
. Then pick 
 $\rho \in (\operatorname {Div}(X_K)\otimes P_*)^G$
 such that
$\rho \in (\operatorname {Div}(X_K)\otimes P_*)^G$
 such that 
 $\pi _*(\rho )=\operatorname {div}(\sigma )$
, and let
$\pi _*(\rho )=\operatorname {div}(\sigma )$
, and let 
 $t$
 be the unique element in
$t$
 be the unique element in 
 $(\operatorname {Pic}(X_K)\otimes T_*)^G$
 such that
$(\operatorname {Pic}(X_K)\otimes T_*)^G$
 such that 
 $\lambda (\rho )=\iota _*(t)$
. Lemma A.2 implies that
$\lambda (\rho )=\iota _*(t)$
. Lemma A.2 implies that
 \begin{align} \theta (\tau )=-t. \end{align}
\begin{align} \theta (\tau )=-t. \end{align}
 Finally, suppose that 
 $K=F_s$
 is a separable closure of
$K=F_s$
 is a separable closure of 
 $F$
, so that
$F$
, so that 
 $G=\Gamma _F$
, and write
$G=\Gamma _F$
, and write 
 $X_s$
 for
$X_s$
 for 
 $X\times _FF_s$
. The exact sequence (B.2) for
$X\times _FF_s$
. The exact sequence (B.2) for 
 $K=F_s$
 takes the form
$K=F_s$
 takes the form
 \begin{align} 1\to T(F_s) \to T(F_s(X))\xrightarrow {\operatorname {div}} \operatorname {Div}(X_s)\otimes T_*\xrightarrow {\lambda } \operatorname {Pic}(X_s)\otimes T_*\to 0. \end{align}
\begin{align} 1\to T(F_s) \to T(F_s(X))\xrightarrow {\operatorname {div}} \operatorname {Div}(X_s)\otimes T_*\xrightarrow {\lambda } \operatorname {Pic}(X_s)\otimes T_*\to 0. \end{align}
We have the inflation–restriction sequence
 \begin{equation*}0\to H^1(F,T(F_s(X)))\xrightarrow {\operatorname {Inf}}H^1(F(X),T)\xrightarrow {\operatorname {Res}}H^1(F_s(X),T).\end{equation*}
\begin{equation*}0\to H^1(F,T(F_s(X)))\xrightarrow {\operatorname {Inf}}H^1(F(X),T)\xrightarrow {\operatorname {Res}}H^1(F_s(X),T).\end{equation*}
Since 
 $T$
 is defined over
$T$
 is defined over 
 $F$
, it is split by
$F$
, it is split by 
 $F_s$
, and hence by Hilbert’s Theorem 90 we have
$F_s$
, and hence by Hilbert’s Theorem 90 we have 
 $H^1(F_s(X),T)$
 = 0. Thus, the inflation map
$H^1(F_s(X),T)$
 = 0. Thus, the inflation map 
 $H^1(F,T(F_s(X)))\to H^1(F(X),T)$
 is an isomorphism. We identify
$H^1(F,T(F_s(X)))\to H^1(F(X),T)$
 is an isomorphism. We identify 
 $H^1(F,T(F_s(X)))$
 with
$H^1(F,T(F_s(X)))$
 with 
 $H^1(F(X),T)$
 via the inflation map. If we define
$H^1(F(X),T)$
 via the inflation map. If we define
 \begin{equation*}H^1(F(X),T)_{\operatorname {nr}}:= \operatorname {Ker}[H^1(F(X),T)\xrightarrow {\operatorname {div}}H^1(F,\operatorname {Div}(X_s)\otimes T_*)],\end{equation*}
\begin{equation*}H^1(F(X),T)_{\operatorname {nr}}:= \operatorname {Ker}[H^1(F(X),T)\xrightarrow {\operatorname {div}}H^1(F,\operatorname {Div}(X_s)\otimes T_*)],\end{equation*}
the map 
 $\theta$
 of (A.2) takes the form
$\theta$
 of (A.2) takes the form
 \begin{equation}\theta \colon H^1(F(X),T)_{\operatorname {nr}}\to \operatorname {Coker}[(\operatorname {Div}(X_s)\otimes T_*)^{\Gamma _F}\to (\operatorname {Pic}(X_s)\otimes T_*)^{\Gamma _F}].\end{equation}
\begin{equation}\theta \colon H^1(F(X),T)_{\operatorname {nr}}\to \operatorname {Coker}[(\operatorname {Div}(X_s)\otimes T_*)^{\Gamma _F}\to (\operatorname {Pic}(X_s)\otimes T_*)^{\Gamma _F}].\end{equation}
Corollary B.2. We have an exact sequence
 \begin{equation*} H^1(F, T) \to H^1(F(X), T)_{\operatorname {nr}} \xrightarrow {\theta } \operatorname {Coker} [(\operatorname {Div}(X_s)\otimes T_*)^{\Gamma _F} \xrightarrow {\lambda } (\operatorname {Pic}(X_s)\otimes T_*)^{\Gamma _F}] \to H^2(F, T), \end{equation*}
\begin{equation*} H^1(F, T) \to H^1(F(X), T)_{\operatorname {nr}} \xrightarrow {\theta } \operatorname {Coker} [(\operatorname {Div}(X_s)\otimes T_*)^{\Gamma _F} \xrightarrow {\lambda } (\operatorname {Pic}(X_s)\otimes T_*)^{\Gamma _F}] \to H^2(F, T), \end{equation*}
where the first and last map are induced by (B.8).
Proof. This is a special case of Proposition B.1.
Acknowledgements
We thank the anonymous referee for carefully reading our manuscript and for sending us comments which greatly improved the exposition.
Conflicts of interest
None.
Journal information
Compositio Mathematica is owned by the Foundation Compositio Mathematica and published by the London Mathematical Society in partnership with Cambridge University Press. All surplus income from the publication of Compositio Mathematica is returned to mathematics and higher education through the charitable activities of the Foundation, the London Mathematical Society and Cambridge University Press.
 
 














 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
