Similitudes over fields with 𝖨 4 = 0 \mathsf{I}^{4}=0 M. Archita, Karim Johannes Becher Abstract . This article studies the set of R R -equivalence classes of the group of proper projective similitudes of an algebra with involution of the first kind. The main results concern base fields of characteristic different from 2 2 over which every 9 9 -dimensional quadratic form has a nontrivial zero. This includes function fields of p p -adic curves and extensions of transcendence degree 3 3 of ℂ \mathbb{C} . Main results of [ 28 ] and [ 29 ] are extended by relaxing the condition on the base field as well as on the Clifford invariant for orthogonal involutions. Keywords: Classical adjoint algebraic group, R R -equivalence, stably rational, algebra with involution, orthogonal, symplectic, hyperbolic, quadratic form, Pfister form, similarity factor, 2 2 -extension, reduced norm, cohomological dimension, u u -invariant, field ordering, signature, weakly hyperbolic Classification (MSC 2020): 11E04, 11E57, 11E81, 14E08, 20G15 1. Introduction In this article we study adjoint semisimple classical linear algebraic groups of type C and D over a field K of characteristic different from 2 . In view of Weil’s classification results [ 31 ] , such a group is given as {\bf PSim}^{+}(A,\sigma) , the connected component of the identity in the group projective similitudes {\bf PSim}(A,\sigma) , for some central simple K -algebra A of even degree and a K -linear involution \sigma on A which is symplectic (type C ) or orthogonal (type D ). A finite field extension L/K is called a 2 -extension if [L:K]=2^{r} for some r\in\mathbb{N} and there exists a sequence of intermediate fields (K_{i})_{i=0}^{r} such that K_{0}=K , K_{r}=L and K_{i}/K_{i-1} is a quadratic extension for 1\leqslant i\leqslant r . At the center of our work is the following general question. Question 1.1 . Which conditions on (A,\sigma) and on K imply that the group of similarity factors \mathsf{G}(A,\sigma) is generated by {K}^{\times 2} and the elements of {K}^{\times} which are norms from finite 2 -extensions L/K such that \sigma_{L} is hyperbolic? By Merkurjev’s criterion from [ 23 , Theorem 1] , a positive answer to 1.1 implies that the group of K -rational R -equivalence classes {\bf PSim}^{+}(A,\sigma)(K)/R is trivial. See Section 3 for details. The focus on 2 -extensions in 1.1 is a central feature of this article. Merkurjev’s characterization does not impose this. It is motivated by the fact that we rely crucially on tools from classical quadratic form theory, where there is more control on 2 -extensions than on arbitrary finite extensions. In this context we show that various results known under a given bound on the cohomological 2 -dimension remain valid with a milder hypothesis in terms of Pfister forms. We denote by \mathsf{W}K the Witt ring of quadratic forms over K and by \mathsf{I}K its fundamental ideal. For n\in\mathbb{N} , we denote \mathsf{I}^{n}K=(\mathsf{I}K)^{n} , the n th power of \mathsf{I}K , generated by the Witt equivalence classes of n -fold Pfister forms over K . Our main results concern fields K where \mathsf{I}^{3}K(\sqrt{-1})=0 or \mathsf{I}^{4}K=0 . In either of these situations, we achieve refinements and extensions of existing results. We will treat central simple algebras with involutions of the first kind (i.e. trivial on the center). For orthogonal involutions, we restrict our attention to those of trivial discriminant, and we look at conditions on their Clifford algebra. The behavior of orthogonal involutions of trivial Clifford invariant resembles that of symplectic involutions in that they have trivial cohomological invariants in degree 1 and 2 . In Section 5, we revisit positive results [ 17 ] and [ 2 ] for orthogonal and symplectic involutions regarding 1.1 , assuming only that \mathsf{I}^{3}K(\sqrt{-1})=0 . In Section 6, we obtain some partial analogues under the assumption that \mathsf{I}^{n+1}K(\sqrt{-1})=0 , \operatorname{\mathsf{ind}}A\leqslant 2 and (A,\sigma) has trivial cohomological invariants in degree less than n , for some arbitrary n\in\mathbb{N} . For a K -quaternion algebra Q , we denote by K_{Q} the function field of the projective conic \mathcal{C}_{Q} associated to Q . If Q=(a,b)_{K} with a,b\in{K}^{\times} , then \mathcal{C}_{Q} is given by aX^{2}+bY^{2}=Z^{2} . When A is Brauer equivalent to Q and \sigma is orthogonal, there is much control on \sigma under scalar extension to K_{Q} . Using this to reduce to the split case, we show in 6.4 that, whenever \mathsf{I}^{n+1}K(\sqrt{-1})=0 , A splits over every real closure of K and \operatorname{\mathsf{ind}}A\leqslant 2 , then the answer to 1.1 is positive for K -linear involutions on A having trivial cohomological invariants up to degree n-1 . It remains yet to be investigated whether this positive answer can be extended by relaxing the conditions on A . In Section 7, we study the case where \mathsf{I}^{4}K=0 . Question 1.2 . Assume that \mathsf{I}^{4}K=0 . If \sigma is orthogonal, then assume that it has trivial discriminant. Does it follow that {\bf PSim}^{+}(A,\sigma)(K)/R=1 ? In 7.5 , we show that the answer is positive whenever K has u -invariant at most 8 and \sigma is symplectic or has trivial Clifford invariant. This was previously shown [ 28 ] and [ 29 ] in the special case where K is the function field of a curve over a p -adic number field \mathbb{Q}_{p} for an odd prime p . In 7.2 , only assuming that \mathsf{I}^{4}K=0 , we obtain a positive answer to 1.2 when \sigma is orthogonal and both components of the Clifford algebra \mathcal{C}(A,\sigma) have index at most 2 (which implies that \operatorname{\mathsf{ind}}A\leqslant 4 ). In our exposition, we strive to present cases of orthogonal and symplectic involutions together whenever the same conclusion can be taken. We further circumvent case distinctions according to the parity of the rank of an associated hermitian form, which occur in [ 28 ] and [ 29 ] . 2. Quadratic forms and cohomological invariants We follow standard terminology and notation from quadratic form theory, taking [ 11 ] and [ 18 ] as our main references. By a quadratic form we always mean a regular quadratic form. Let K be a field of characteristic different from 2 . The nonzero squares in K form the subgroup {K}^{\times 2} of the multiplicative group {K}^{\times} , giving rise to square class group {K}^{\times}/{K}^{\times 2} . Consider a quadratic form \varphi over K . It is considered as a homogeneous quadratic polynomial in a given number n\in\mathbb{N} of variables, where n is called the dimension of \varphi and denoted \mathsf{dim}(\varphi) . It gives rise to a quadratic map K^{n}\to K . For a field extension L/K , we denote by \varphi_{L} the same polynomial considered with its evaluation map L^{n}\to L . For k\in\mathbb{N} , we denote by k\times\varphi the k -fold multiple \varphi\perp\dots\perp\varphi . We denote by \varphi_{\mathsf{an}} the anisotropic part and by \mathsf{i}(\varphi) the Witt index of \varphi . They are determined by having that \varphi_{\mathsf{an}} is anisotropic and \varphi\simeq\varphi_{\mathsf{an}}\perp\mathsf{i}(\varphi)\times\operatorname{\mathbb{H}} , where \operatorname{\mathbb{H}} denotes the hyperbolic plane over K . We call \varphi isotropic if \mathsf{i}(\varphi)>0 , and anisotropic otherwise. We call \varphi hyperbolic if \varphi_{\mathsf{an}} is trivial, i.e. if \varphi\simeq i\times\operatorname{\mathbb{H}} for i=\mathsf{i}(\varphi) . We denote by \mathsf{disc}(\varphi) the discriminant of \varphi , considered as a class in {K}^{\times}/{K}^{\times 2} . We say that \mathsf{disc}(\varphi) is trivial if 1\in\mathsf{disc}(\varphi) , that is, if \mathsf{disc}(\varphi)=1{K}^{\times 2} . We denote by \mathsf{D}(\varphi) the set of elements of {K}^{\times} that are represented by \varphi , and we set \mathsf{G}(\varphi)=\{a\in{K}^{\times}\mid a\varphi\simeq\varphi\} , which is a subgroup of {K}^{\times} . If \mathsf{dim}(\varphi)=2 we also call \varphi a binary form . For an ideal J of the Witt ring \mathsf{W}K , we write \varphi\in J to indicate that the Witt equivalence class of \varphi belongs to J . For a central simple K -algebra A , we denote by \mathsf{deg}A its degree and by \operatorname{\mathsf{ind}}A its index , and we call A split if A\sim K , or equivalently, if \operatorname{\mathsf{ind}}A=1 . We denote by \operatorname{\mathsf{Br}}(K) the Brauer group of K and by \operatorname{\mathsf{Br}}_{2}(K) its 2 -torsion part. The elements of \operatorname{\mathsf{Br}}(K) are the classes [A] of central simple K -algebras A . The tensor product induces the group operation in \operatorname{\mathsf{Br}}(K) , which we write additively, and we denote with neutral element 0=[K] . Given two central simple K -algebras A and A^{\prime} , we write A\sim A^{\prime} to indicate that [A]=[A^{\prime}] . For the definition and the basic properties of the Clifford invariant of quadratic forms, we refer to [ 18 , Chap. V] . Let us recall what we shall need. The Clifford invariant is a map c:\mathsf{W}K\to\operatorname{\mathsf{Br}}_{2}(K). If \varphi is even-dimensional, then its Clifford algebra \mathcal{C}(\varphi) is a central simple K -algebra and c(\varphi)=[\mathcal{C}(\varphi)] . Theorem 2.1 (Merkurjev) . The restriction of c to \mathsf{I}^{2}K gives a surjective group homomorphism c:\mathsf{I}^{2}K\rightarrow\operatorname{\mathsf{Br}}_{2}(K) whose its kernel is equal to \mathsf{I}^{3}K . Proof. See [ 22 ] . ∎ For k\in\mathbb{N} , we denote by \mu_{k} the group of k th roots of unity in an algebraic closure of K . Let n\in\mathbb{N} . For an abelian group A endowed with an action of the absolute Galois group \mathcal{G}_{K} of K , we denote by H^{n}(K,A) the n th cohomology group H^{n}(\mathcal{G}_{K},A) . Here, we consider the natural action of \mathcal{G}_{K} on \mu_{k}^{\otimes n-1} . For n\in\mathbb{N} and a_{1},\dots,a_{n}\in{K}^{\times} , we denote by \langle\!\langle a_{1},\dots,a_{n}\rangle\!\rangle the n -fold Pfister form \langle 1,-a_{1}\rangle\otimes\dots\otimes\langle 1,-a_{n}\rangle over K . Theorem 2.2 (Orlov-Vishik-Voevodsky) . There is a natural surjective group homomorphism e_{n}:\mathsf{I}^{n}K\to H^{n}(K,\mu_{2}) whose kernel is equal to \mathsf{I}^{n+1}K and which maps the class of \langle\!\langle a_{1},\dots,a_{n}\rangle\!\rangle to the symbol (a_{1})\cup\cdots\cup(a_{n}) , for any a_{1},\ldots,a_{n}\in{K}^{\times} . Proof. See [ 30 ] in combination with [ 25 , Theorem 4.1] or [ 24 , Theorem 1.1] . ∎ Note that H^{1}(K,\mu_{2}) is naturally isomorphic to {K}^{\times}/{K}^{\times 2} , and this isomorphism makes e_{1} correspond to the discriminant. Furthermore, H^{2}(K,\mu_{2}) can be naturally identified with \operatorname{\mathsf{Br}}_{2}(K) (cf. [ 14 , Cor. 4.4.5] ), and this identifies e_{2} with the map c of 2.1 . We write \mathbb{N}^{+} for \mathbb{N}\smallsetminus\{0\} . We denote by \mathsf{X}(K) the set of all field orderings of K . For P\in\mathsf{X}(K) , we denote by K_{P} the corresponding real closure of K . For a quadratic form \varphi over K , we denote by \mathsf{sign}_{P}(\varphi) its signature at an ordering P\in\mathsf{X}(K) . We call the field K real if \mathsf{X}(K)\neq\emptyset , and nonreal otherwise. We denote by \mathsf{\Sigma}{K}^{2} the set of sums of squares in K . By the Artin-Schreier Theorem, [ 18 , Chap. VIII, Theorem 1.10] , K is real if and only if -1\notin\mathsf{\Sigma}{K}^{2} . Corollary 2.3 . Let n\in\mathbb{N}^{+} be such that \mathsf{I}^{n}K(\sqrt{-1})=0 . Then the natural homomorphism H^{n}(K,\mu_{2})\to\prod_{P\in\mathsf{X}(K)}H^{n}(K,\mu_{2}) is injective. Furthermore, for any r,d\in\mathbb{N} , we have H^{n}(K,\mu_{2^{r}}^{\otimes d})=H^{n}(K,\mu_{2})=(-1)\cup H^{n-1}(K,\mu_{2}) . Proof. Since \mathsf{I}^{n}K(\sqrt{-1})=0 , we have that \mathsf{I}^{n}K=2\times\mathsf{I}^{n-1}K , by [ 11 , Cor. 35.27] . Hence, using the surjectivity of e_{n}:\mathsf{I}^{n}K\to H^{n}(K,\mu_{2}) , we obtain that H^{n}(K,\mu_{2})=(-1)\cup H^{n-1}(K,\mu_{2}) . Let r,d\in\mathbb{N} . Consider \xi\in\mathsf{H}^{n}(K,\mu_{2^{r}}^{\otimes d}) and assume that \xi\neq 0 . Let i\in\mathbb{N} be minimal such that 2^{i}\xi=0 . Then i>0 . Let \xi^{\prime}=2^{i-1}\xi . Then \xi^{\prime}\neq 0 and 2\xi^{\prime}=0 . Therefore \xi^{\prime}\in H^{n}(K,\mu_{2})\smallsetminus\{0\} , and it follows by 2.2 that \xi^{\prime}=e_{n}(\varphi) for a quadratic form \varphi\in\mathsf{I}^{n}K\smallsetminus\mathsf{I}^{n+1}K . As \mathsf{I}^{n}K(\sqrt{-1})=0 and \varphi\notin\mathsf{I}^{n+1}K , it follows by [ 11 , Cor. 35.27] together with [ 16 , Theorem 1] that there exists P\in\mathsf{X}(K) such that \mathsf{sign}_{P}(\varphi)\notin 2^{n+1}K . Then \xi^{\prime}_{K_{P}}=e_{n}(\varphi_{K_{P}})\neq 0 , and consequently \xi_{K_{P}}\neq 0 . This proves that the natural homomorphism H^{n}(K,\mu_{2^{r}}^{\otimes d})\to\prod_{P\in\mathsf{X}(K)}H^{n}(K_{P},\mu_{2^{r}}^{\otimes d}) is injective. Since for any real closed field R we have H^{n}(R,\mu_{2^{r}}^{\otimes d})=H^{n}(R,\mu_{2})\simeq\mathbb{Z}/2\mathbb{Z} , the statement follows. ∎ As a consequence of 2.2 , the cohomological 2 -dimension of K , \mathsf{cd}_{2}(K) , can be characterised as follows (see e.g. [ 5 , 3.4. Remark] ): \mathsf{cd}_{2}(K)=\mathsf{sup}\{n\in\mathbb{N}\mid\mathsf{I}^{n}K^{\prime}\neq 0\mbox{ for some finite field extension }K^{\prime}/K\}\,. Proposition 2.4 . Let n\in\mathbb{N} be such that \mathsf{I}^{n}K=0 . Then \mathsf{I}^{n}L=0 for every finite 2 -extension L/K . Proof. See [ 11 , Cor. 35.8] for the case where [L:K]=2 . The general case follows immediatley by induction on r\in\mathbb{N} such that [L:K]=2^{r} . ∎ 3. Algebras with involution, similitudes and norms Our main reference for algebras with involution is [ 15 ] . For terminology and basic facts concerning central simple algebras, we refer to [ 14 ] and [ 27 ] . Let A be a central simple K -algebra and let \sigma be an involution on A . We call \sigma hyperbolic if there exists an element e\in A such that e^{2}=e and \sigma(e)=1-e . We restrict our attention to involutions of the first kind , that is, which are K -linear. By a K -algebra with involution we mean a pair (A,\sigma) where A is a central simple K -algebra and \sigma is a K -linear involution on A . Such involutions are distinguished into orthogonal and symplectic ones, see [ 15 , §2] . Let (A,\sigma) be a K -algebra with involution. We call (A,\sigma) hyperbolic if \sigma is hyperbolic. This can only occur when \mathsf{deg}A is a multiple of 2\operatorname{\mathsf{ind}}A , so in particular \mathsf{deg}A must be even. We call (A,\sigma) split hyperbolic if A is split and \sigma is hyperbolic. For a field extension L/K , we obtain an L -algebra with involution (A,\sigma)_{L}=(A_{L},\sigma_{L}) , where A_{L}=A\otimes_{K}L and \sigma_{L}=\sigma\otimes\operatorname{\mathsf{id}}_{L} . We denote {\bf\mathsf{Sim}}(A,\sigma)=\{x\in A\mid\sigma(x)x\in{K}^{\times}\} and obtain that this is a subgroup of {A}^{\times} . We denote by {\bf\mathsf{PSim}}(A,\sigma) the quotient group {\bf\mathsf{Sim}}(A,\sigma)/{K}^{\times} . The elements of {\bf\mathsf{Sim}}(A,\sigma) are called similitudes of (A,\sigma) , those of {\bf\mathsf{PSim}}(A,\sigma) are called projective similitudes of (A,\sigma) . Letting {\bf PSim}(A,\sigma)(K^{\prime})={\bf\mathsf{PSim}}(A_{K^{\prime}},\sigma_{K^{\prime}}) for every field extension K^{\prime}/K defines an algebraic group over K whose set of K -rational points is {\bf\mathsf{PSim}}(A,\sigma) . We denote by {\bf PSim^{+}}(A,\sigma) be the connected component of the identity in {\bf PSim}(A,\sigma) . This is a classical semisimple adjoint linear algebraic group. We mention that {\bf PSim}^{+}(A,\sigma)={\bf PSim}(A,\sigma) unless \sigma is orthogonal and \mathsf{deg}A is even, and in the latter case, {\bf PSim}^{+}(A,\sigma)(K^{\prime}) has index 2 in {\bf PSim}(A,\sigma)(K^{\prime}) for every field extension K^{\prime}/K . (See [ 15 , Prop. 12.23] .) We obtain a natural group homomorphism \mu:{\bf\mathsf{Sim}}(A,\sigma)\to{K}^{\times},x\mapsto\sigma(x)x\,. Its image is denoted by \mathsf{G}(A,\sigma) . In other terms, \mathsf{G}(A,\sigma)=\{\sigma(x)x\mid x\in{\bf\mathsf{Sim}}(A,\sigma)\}\,. The elements of this subgroup of {K}^{\times} are called similarity factors of \sigma . We further set \mathsf{G}^{+}(A,\sigma)=\{\mu(a)\mid a{K}^{\times}\in{\bf PSim}^{+}(A,\sigma)(K)\}\,. This is a subgroup of \mathsf{G}(A,\sigma) of index at most 2 . Consider a quadratic form over K , given on the K -vector space V_{\varphi} . The split central simple K -algebra \operatorname{\mathsf{End}}_{K}(V_{\varphi}) then is endowed with a K -linear involution \operatorname{\mathsf{ad}}_{\varphi} , called the adjoint involution on \varphi , which is characterized by the property that \mathsf{\mathfrak{b}}_{\varphi}(x,f(y))=\mathsf{\mathfrak{b}}_{\varphi}(\operatorname{\mathsf{ad}}_{\varphi}(f)(x),y) for every f\in\operatorname{\mathsf{End}}_{K}(V_{\varphi}) and every x,y\in V_{\varphi} , where \mathsf{\mathfrak{b}}_{\varphi}:V_{\varphi}\times V_{\varphi}\to K is the symmetric K -bilinear form associated to \varphi given by \mathsf{\mathfrak{b}}_{\varphi}(x,y)=\varphi(x+y)-\varphi(x)-\varphi(y) . We set \operatorname{\mathsf{Ad}}(\varphi)=(\operatorname{\mathsf{End}}_{K}(V_{\varphi}),\operatorname{\mathsf{ad}}_{\varphi})\,. This is a split K -algebra with orthogonal involution. By [ 15 , p.1, Theorem] , every split K -algebra with orthogonal involution is isomorphic to \operatorname{\mathsf{Ad}}(\varphi) for a quadratic form \varphi over K , and \operatorname{\mathsf{Ad}}(\varphi) is hyperbolic if and only if the quadratic form \varphi is hyperbolic in the usual sense, by [ 15 , Prop. 6.7] . Proposition 3.1 . If \sigma is orthogonal of nontrivial discriminant, \operatorname{\mathsf{ind}}A=2 and A_{K(\sqrt{d})} is split for d\in\mathsf{disc}(\sigma) , then \mathsf{G}^{+}(A,\sigma) has index 2 in \mathsf{G}(A,\sigma) . In every other case \mathsf{G}^{+}(A,\sigma)=\mathsf{G}(A,\sigma) . Proof. See [ 15 , Prop. 12.23] for the case where \sigma is symplectic, [ 15 , Cor. 13.43] for the case where A is not split and \sigma is orthogonal. It remains to consider the case where A is split and \sigma is orthogonal. Then (A,\sigma)\simeq\operatorname{\mathsf{Ad}}(\varphi) for a quadratic form \varphi over K . If \mathsf{deg}A is odd, then \mathsf{G}(A,\sigma)=\mathsf{G}^{+}(A,\sigma)={K}^{\times 2} . Assume that \mathsf{deg}A is even. We choose a hyperplane reflection a of \varphi . Then a represents a class in {\bf PSim}(A,\sigma)\smallsetminus{\bf PSim}^{+}(A,\sigma) . Hence \mathsf{G}(A,\sigma)=\mathsf{G}^{+}(A,\sigma)\cup\mu(a)\mathsf{G}^{+}(A,\sigma) , and since \mathsf{deg}A is even, we have \mu(a)=1 . ∎ For a finite field extension L/K , we denote by \mathsf{N}_{L/K}:L\to K the norm map, and we fix the notation \mathsf{N}_{L/K}^{\ast}=\mathsf{N}_{L/K}({L}^{\times})\,, and note that this is a subgroup of {K}^{\times} . For a subset S\subseteq{K}^{\times} , we denote by \langle S\rangle the subgroup of {K}^{\times} generated by S . Let \Phi be a quadratic form over K or a K -algebra with involution. Hence, for any field extension K^{\prime}/K , we have an object \Phi_{K^{\prime}} obtained by scalar extension, and we can consider whether it is hyperbolic. We denote by \mathsf{Hyp}^{\ast}(\Phi) (resp. by \mathsf{Hyp}_{2}^{\ast}(\Phi) ) the union of the subgroups \mathsf{N}_{L/K}^{\ast} where L/K is a finite field extension (resp. a finite 2 -extension) such that \Phi_{L} is hyperbolic. We further set \mathsf{Hyp}(\Phi)=\langle\mathsf{Hyp}^{\ast}(\Phi)\rangle and \mathsf{Hyp}_{2}(\Phi)=\langle\mathsf{Hyp}^{\ast}(\Phi)\rangle . Note that \mathsf{Hyp}_{2}(\Phi)\subseteq\mathsf{Hyp}(\Phi) . We do not know of an example where these two groups are different. Cases where \mathsf{Hyp}^{\ast}(\Phi)\subsetneq\mathsf{Hyp}(\Phi) are easy to obtain, e.g. from 4.2 (a) below. While the main attention lies on the groups \mathsf{Hyp}(A,\sigma) and \mathsf{Hyp}_{2}(A,\sigma) , the \ast -notation will allow us to formulate more precise statements without becoming too cumbersome. If \Phi is a quadratic form of odd dimension over K or (equivalently) a K -algebra with involution of odd degree, then \mathsf{Hyp}_{2}^{\ast}(\Phi)=\emptyset and \mathsf{Hyp}_{2}(\Phi)=\{1\} . For a quadratic form \varphi over K , the adjoint K -algebra with involution \operatorname{\mathsf{Ad}}(\varphi) is hyperbolic if and only if the quadratic form \varphi is hyperbolic, and thus we have \mathsf{Hyp}_{2}^{\ast}(\operatorname{\mathsf{Ad}}(\varphi))=\mathsf{Hyp}_{2}^{\ast}(\varphi) and \mathsf{Hyp}^{\ast}(\operatorname{\mathsf{Ad}}(\varphi))=\mathsf{Hyp}^{\ast}(\varphi) . Proposition 3.2 (Norm Principle) . Let \Phi be a quadratic form or a K -algebra with involution. Then {K}^{\times 2}\mathsf{Hyp}(\Phi)\subseteq\mathsf{G}^{+}(\Phi) . Proof. This is covered by [ 23 , Theorem 1] . See also [ 18 , Chap. VII, Cor. 4.4] for the case where \Phi is a quadratic form. ∎ Theorem 3.3 (Merkurjev) . Let \Phi be a quadratic form or a K -algebra with involution. For any field extension K^{\prime}/K , the homomorphism \mu induces an isomorphism on the group of K^{\prime} -rational R -equivalence classes {\bf PSim}^{+}(\Phi)(K^{\prime})/R\stackrel{{\scriptstyle\simeq}}{{\longrightarrow}}\mathsf{G}^{+}(\Phi_{K^{\prime}})/{K}^{\times 2}\mathsf{Hyp}(\Phi_{K^{\prime}})\,. Proof. See [ 23 , Theorem 1] . ∎ 4. Reduced norms For a K -algebra with orthogonal involution (A,\sigma) of trivial discriminant, we look into the general relations (without conditions on K ) between \mathsf{G}(A,\sigma) , \mathsf{Hyp}(A,\sigma) as well as the groups of reduced norms of A and of the two components of \mathcal{C}(A,\sigma) . Let A be a central simple K -algebra. We denote by \operatorname{\mathsf{Trd}}_{A}:A\to K the reduced trace and by \operatorname{\mathsf{Nrd}}_{A}:A\to K the reduced norm map of A , and we set \operatorname{\mathsf{Nrd}}_{A}^{\ast}=\operatorname{\mathsf{Nrd}}_{A}({A}^{\times}) , which is a subgroup of {K}^{\times} . If A is a K -quaternion algebra, then \operatorname{\mathsf{Nrd}}_{A} is a quadratic form over K , isometric to \langle\!\langle a,b\rangle\!\rangle for any a,b\in{K}^{\times} with A\simeq(a,b)_{K} . Proposition 4.1 . The following hold: (a) \operatorname{\mathsf{Nrd}}_{A}^{\ast}=\operatorname{\mathsf{Nrd}}_{B}^{\ast} for every central simple K -algebra B\sim A . (b) \operatorname{\mathsf{Nrd}}_{A}^{\ast} is the union of the subgroups \mathsf{N}_{L/K}^{\ast} for finite field extensions L/K such that A_{L} is split. (c) If B is a central simple K -algebra and every finite field extension splitting A also splits B , then \operatorname{\mathsf{Nrd}}_{A}^{\ast}\subseteq\operatorname{\mathsf{Nrd}}_{B}^{\ast} . Proof. See [ 14 , Lemma 2.6.4] for (a) . Part (b) follows from (a) , using that a finite field extension L/K splits A if and only if A\sim B for a central simple K -algebra B containing L as a maximal commutative subring (cf. [ 27 , Theorem 13.3] ), and in this case \operatorname{\mathsf{Nrd}}_{B}|_{L}=\mathsf{N}_{L/K} . Parts (c) is immediate from (b) . ∎ Assume that \mathsf{deg}A is even and consider an orthogonal involution \sigma on A . We denote by \mathsf{disc}(\sigma) the discriminant of \sigma , given as a coset in {K}^{\times}/{K}^{\times 2} . We may write d\in\mathsf{disc}(\sigma) to indicate that d\in{K}^{\times} is such that \mathsf{disc}(\sigma)=d{K}^{\times 2} . We denote by \mathcal{C}(A,\sigma) the Clifford algebra of (A,\sigma) . If \sigma has trivial discriminant (i.e. 1\in\mathsf{disc}(\sigma) ), then \mathcal{C}(A,\sigma)\simeq C_{1}\times C_{2} for two central simple K -algebras C_{1} and C_{2} , which we call the components of \mathcal{C}(A,\sigma) , and by [ 15 , Theorem 9.12] , we have in \operatorname{\mathsf{Br}}(K) that [C_{1}]+[C_{2}]=[A] if \mathsf{deg}A\equiv 0\,\mathsf{mod}\,4 and 2[C_{1}]=2[C_{2}]=[A] if \mathsf{deg}A\equiv 2\,\mathsf{mod}\,4 . Proposition 4.2 . Let (A,\sigma) be a K -algebra with orthogonal involution of trivial discriminant and let C_{1} and C_{2} be the two components of \mathcal{C}(A,\sigma) . Then: (a) \mathsf{Hyp}(A,\sigma)\subseteq\operatorname{\mathsf{Nrd}}_{C_{1}}^{\ast}\cdot\operatorname{\mathsf{Nrd}}_{C_{2}}^{\ast} . (b) If \operatorname{\mathsf{ind}}(C_{i})\leqslant 2 for i\in\{1,2\} , then \mathsf{G}(A,\sigma)\subseteq\operatorname{\mathsf{Nrd}}_{C_{1}}^{\ast}\cdot\operatorname{\mathsf{Nrd}}_{C_{2}}^{\ast} . (c) If \mathsf{deg}A=4 , then \mathsf{G}(A,\sigma)=\operatorname{\mathsf{Nrd}}_{C_{1}}^{\ast}\cdot\operatorname{\mathsf{Nrd}}_{C_{2}}^{\ast}=\mathsf{Hyp}_{2}(A,\sigma) . Proof. (a) Consider a\in\mathsf{Hyp}^{\ast}(A,\sigma) . Then there exists a finite field extension L/K such that \sigma_{L} is hyperbolic and a\in\mathsf{N}_{L/K}^{\ast} . It follows by [ 15 , Prop. 8.31] that either (C_{1})_{L} or (C_{2})_{L} is split. This yields that \mathsf{Hyp}^{\ast}(A,\sigma)\subseteq\operatorname{\mathsf{Nrd}}_{C_{1}}^{\ast}\cup\operatorname{\mathsf{Nrd}}_{C_{2}}^{\ast} , whereby \mathsf{Hyp}(A,\sigma)\subseteq\operatorname{\mathsf{Nrd}}_{C_{1}}^{\ast}\cdot\operatorname{\mathsf{Nrd}}_{C_{2}}^{\ast} . (b) By the hypothesis, for i\in\{1,2\} , we have C_{i}\sim Q_{i} for a K -algebra Q_{i} , and we denote by \pi_{i} its norm form, to obtain that \operatorname{\mathsf{Nrd}}_{C_{i}}^{\ast}=\operatorname{\mathsf{Nrd}}_{Q_{i}}^{\ast}=\mathsf{D}(\pi_{i}) . Let a\in\mathsf{G}(A,\sigma) . We consider the quadratic form \varphi=\pi_{1}\perp-a\pi_{2} over K . By [ 15 , Theorem 9.15] , we have \mathcal{C}(\varphi)\sim Q_{1}\otimes Q_{2}\sim A . If A is split, then Q_{1}\simeq Q_{2} and (A,\sigma)\simeq\operatorname{\mathsf{Ad}}(\rho) for a quadratic form \rho over K with \mathcal{C}(\rho)\sim Q_{1}\sim Q_{2} , and it follows that \mathsf{G}(A,\sigma)=\mathsf{G}(\rho)\subseteq\mathsf{G}(\pi_{1})=\mathsf{G}(\pi_{2}) , whereby \varphi is hyperbolic. In the general case, let K_{A}/K denote the splitting field of the Severi-Brauer variety associated to A . We conclude from the split case that \varphi_{K_{A}} is hyperbolic. By [ 19 , Théorème 4] , this implies that \varphi is isotropic over K . This yields that a\in\mathsf{D}(\pi_{1})\cdot\mathsf{D}(\pi_{2})=\operatorname{\mathsf{Nrd}}_{C_{1}}^{\ast}\cdot\operatorname{\mathsf{Nrd}}_{C_{2}}^{\ast} . (c) Assume that \mathsf{deg}A=4 . By [ 15 , Corollary 15.13] , then C_{1} and C_{2} are K -quaternion algebras and (A,\sigma)\simeq(C_{1},\operatorname{\mathsf{can}}_{C_{1}})\otimes(C_{2},\operatorname{\mathsf{can}}_{C_{2}}) , and by [ 15 , Corollary 15.13] , we have \mathsf{G}(A,\sigma)=\operatorname{\mathsf{Nrd}}_{C_{i}}^{\ast}\cdot\operatorname{\mathsf{Nrd}}_{C_{2}}^{\ast} . From (a) we obtain that \mathsf{Hyp}_{2}(A,\sigma)\subseteq\operatorname{\mathsf{Nrd}}_{C_{i}}^{\ast}\cdot\operatorname{\mathsf{Nrd}}_{C_{2}}^{\ast} . To show the converse inclusion, consider a=a_{1}\cdot a_{2} with a_{1}\in\operatorname{\mathsf{Nrd}}_{C_{1}}^{\ast} and a_{2}\in\operatorname{\mathsf{Nrd}}_{C_{2}}^{\ast} . For i\in\{1,2\} , there exists a field extension L_{i}/K with [L_{i}:K]\leqslant 2 such that a_{i}\in\mathsf{N}_{L_{i}/K}^{\ast} and (C_{i})_{L_{i}} is split, whereby (\operatorname{\mathsf{can}}_{C_{i}})_{L_{i}} and \sigma_{L_{i}} are hyperbolic, and consequently a_{i}\in\mathsf{N}_{L_{i}/K}^{\ast}\subseteq\mathsf{Hyp}_{2}^{\ast}(A,\sigma) . Hence a=a_{1}a_{2}\in\mathsf{Hyp}_{2}(A,\sigma) . ∎ In the remainder of this section, we look at the case where \mathsf{deg}A\equiv 2\,\mathsf{mod}\,4 . Proposition 4.3 . Let (A,\sigma) be a K -algebra with orthogonal involution with \mathsf{deg}A\equiv 2\,\mathsf{mod}\,4 . Then: (a) \mathsf{G}^{+}(A,\sigma)\subseteq\operatorname{\mathsf{Nrd}}_{A}^{\ast} . (b) If \mathsf{disc}(\sigma) is trivial discriminant and C is a component of \mathcal{C}(A,\sigma) , then \mathcal{C}(A,\sigma)\simeq C\times C^{\operatorname{\mathsf{op}}} and C^{\otimes 2}\sim A , and in particular A is split or \operatorname{\mathsf{ind}}C\geqslant 4 . Proof. (a) Write \mathsf{deg}A=2n with n\in\mathbb{N} and \lambda\in\mathsf{G}^{+}(A,\sigma) . Let a\in{\bf\mathsf{Sim}}^{+}(A,\sigma) be such that \lambda=a\sigma(a) . Then \operatorname{\mathsf{Nrd}}_{A}(\sigma(a))=\operatorname{\mathsf{Nrd}}_{A}(a)=+\lambda^{n} . This implies that \lambda^{2n}=\operatorname{\mathsf{Nrd}}_{A}(\lambda)=\operatorname{\mathsf{Nrd}}_{A}(a)\operatorname{\mathsf{Nrd}}_{A}(\sigma(a))=\operatorname{\mathsf{Nrd}}_{A}(a)^{2}\,. By the hypothesis, n is odd. We conclude that \lambda\in{K}^{\times 2}\cdot\operatorname{\mathsf{Nrd}}_{A}^{\ast} . Since A carries a K -linear involution, \operatorname{\mathsf{ind}}A is a 2 -power. Since \operatorname{\mathsf{ind}}A divides \mathsf{deg}A=2n , it follows that \operatorname{\mathsf{ind}}A\leqslant 2 . Hence A\sim Q for a K -quaternion algebra Q . Using 4.1 , we obtain that {K}^{\times 2}\subseteq\operatorname{\mathsf{Nrd}}_{Q}^{\ast}=\operatorname{\mathsf{Nrd}}_{A}^{\ast} . (b) Assume that \mathsf{disc}(\sigma) is trivial. Then \mathcal{C}(A,\sigma)\simeq C_{1}\times C_{2} for two central simple K -algebras C_{1} and C_{2} of the same degree. By [ 15 , Theorem 9.11] , we have [C_{1}]+[C_{2}]=0 and 2[C_{1}]=2[C_{2}]=[A] in \operatorname{\mathsf{Br}}(K) . Hence C_{2}\simeq C_{1}^{\operatorname{\mathsf{op}}} . Since 2[A]=0 , it follows that either [A]=0 or the class given by [C_{i}] has order 4 in \operatorname{\mathsf{Br}}(K) for i\in\{1,2\} , and in that case \operatorname{\mathsf{ind}}C_{i} is a multiple of 4 . ∎ Example 4.4 . Consider the case where (A,\sigma) is orthogonal of degree 6 with trivial discriminant. Then \mathcal{C}(A,\sigma)\simeq C\times C^{\operatorname{\mathsf{op}}} for a central simple K -algebra C of degree 4 . In this case it follows that \mathsf{G}(A,\sigma)={K}^{\times 2}\operatorname{\mathsf{Nrd}}_{C}^{\ast} ; see [ 23 , Prop. 6] and the references there, along with 3.1 . Recall that, for a K -quaternion algebra Q , we denote by K_{Q} the function field of the projective conic associated to Q . Theorem 4.5 . Let (A,\sigma) be a K -algebra with orthogonal involution of trivial discriminant and such that \mathsf{deg}A\equiv 2\,\mathsf{mod}\,4 . Let C be one component of \mathcal{C}(A,\sigma) and Q the K -quaternion algebra such that A\sim Q . Assume that \operatorname{\mathsf{ind}}C_{K_{Q}}\leqslant 4 . Then \mathsf{G}(A,\sigma)\subseteq{K}^{\times 2}\operatorname{\mathsf{Nrd}}_{C}^{\ast}\,. Proof. Note that \mathcal{C}(A,\sigma)\simeq C\times C^{\operatorname{\mathsf{op}}} . Since \operatorname{\mathsf{ind}}C_{K_{Q}}\leqslant 4 , we may choose a central simple K -algebra C^{\prime} with \mathsf{deg}C^{\prime}=4 such that either C^{\prime}\sim C or C^{\prime}\sim C\otimes Q . Then there exists a central simple K -algebra with orthogonal involution (A^{\prime},\sigma^{\prime}) such that \mathcal{C}(A^{\prime},\sigma^{\prime})\simeq C^{\prime}\times C^{\prime\operatorname{\mathsf{op}}} . Note that A^{\prime}\sim C^{\prime\otimes 2}\sim C^{\otimes 2}\sim A\sim Q . There exist skew-hermitian forms (V,h) and (V^{\prime},h^{\prime}) over (Q,\operatorname{\mathsf{can}}_{Q}) such that (A,\sigma)\simeq\operatorname{\mathsf{Ad}}(V,h) and (A^{\prime},\sigma^{\prime})\simeq\operatorname{\mathsf{Ad}}(V^{\prime},h^{\prime}) . We obtain a skew-hermitian form (V\oplus V^{\prime},h\perp h^{\prime}) over (Q,\operatorname{\mathsf{can}}_{Q}) and consider its adjoint K -algebra with orthogonal involution (A^{\prime\prime},\sigma^{\prime\prime})=\operatorname{\mathsf{Ad}}(V\oplus V^{\prime},h\perp h^{\prime}) . (This is an orthogonal sum of (A,\sigma) and (A^{\prime},\sigma^{\prime}) in the sense of Dejaiffe [ 9 ] .) Since \mathsf{disc}(\sigma) is trivial, so is \mathsf{disc}(\sigma^{\prime}) , by [ 9 , Prop. 2.3] . Furthermore \mathcal{C}(A^{\prime\prime},\sigma^{\prime\prime})\simeq(C\otimes C^{\prime},C^{\operatorname{\mathsf{op}}}\otimes C^{\prime\operatorname{\mathsf{op}}}) . If C^{\prime}\sim C , then C\otimes C^{\prime}\sim Q , and if C^{\prime}\sim C\otimes Q , then C\otimes C^{\prime} is split. Therefore \mathcal{C}(A^{\prime\prime}_{K_{Q}},\sigma^{\prime\prime}_{K_{Q}}) is trivial. Since A^{\prime}\sim A\sim Q , we have that A,A^{\prime} and A^{\prime\prime} split over K_{Q} . Hence there exist quadratic forms \varphi and \varphi^{\prime} over K_{Q} such that (A,\sigma)_{K_{Q}}\simeq\operatorname{\mathsf{Ad}}(\varphi) and (A^{\prime},\sigma^{\prime})_{K_{Q}}\simeq\operatorname{\mathsf{Ad}}(\varphi^{\prime}) . It follows that \operatorname{\mathsf{Ad}}(\varphi\perp\varphi^{\prime})\simeq(A^{\prime\prime},\sigma^{\prime\prime})_{K_{Q}} . Note that \varphi and \varphi^{\prime} are even-dimension of trivial discriminant, that is, \varphi,\varphi^{\prime}\in\mathsf{I}^{2}K_{Q} . Furthermore, c(\varphi)=[C_{K_{Q}}] and c(\varphi^{\prime})=[C^{\prime}_{K_{Q}}] . Since C_{K_{Q}}\sim C^{\prime}_{K_{Q}} , we conclude that c(\varphi\perp\varphi^{\prime})=0 . Hence \varphi\perp\varphi^{\prime}\in\mathsf{I}^{3}K_{Q} , by 2.1 . Consider now an element a\in\mathsf{G}(A,\sigma) . Then a\in\mathsf{G}(\varphi) , whereby \langle\!\langle a\rangle\!\rangle\otimes\varphi is hyperbolic. Since \langle\!\langle a\rangle\!\rangle\otimes(\varphi\perp\varphi^{\prime})\in\mathsf{I}^{4}K_{Q} , we conclude that \langle\!\langle a\rangle\!\rangle\otimes\varphi^{\prime}\in\mathsf{I}^{4}K_{Q} . Since \mathsf{dim}(\varphi^{\prime})=\mathsf{deg}A^{\prime}=6 , we have \mathsf{dim}(\langle\!\langle a\rangle\!\rangle\otimes\varphi^{\prime})=12 . We conclude by the Arason-Pfister Hauptsatz [ 11 , Theorem 23.7] that \langle\!\langle a\rangle\!\rangle\otimes\varphi^{\prime} is hyperbolic. Hence \left(\operatorname{\mathsf{Ad}}\langle\!\langle a\rangle\!\rangle\otimes(A^{\prime},\sigma^{\prime})\right)_{K_{Q}} is hyperbolic. Hence, Dejaiffe’s Theorem [ 10 ] yields that \operatorname{\mathsf{Ad}}\langle\!\langle a\rangle\!\rangle\otimes(A^{\prime},\sigma^{\prime}) is hyperbolic, whereby a\in\mathsf{G}(A^{\prime},\sigma^{\prime}) . This argument shows that \mathsf{G}(A,\sigma)\subseteq G(A^{\prime},\sigma^{\prime}) . Since \mathsf{deg}A^{\prime}=6 and \mathcal{C}(A^{\prime},\sigma^{\prime})=C^{\prime}\times C^{\prime\operatorname{\mathsf{op}}} , we have \mathsf{G}(A^{\prime},\sigma^{\prime})={K}^{\times 2}\operatorname{\mathsf{Nrd}}_{C^{\prime}}^{\ast} , by 4.4 . Note that C and C^{\prime} are split over the same field extensions. Hence \operatorname{\mathsf{Nrd}}_{C^{\prime}}^{\ast}=\operatorname{\mathsf{Nrd}}_{C}^{\ast} , by 4.1 , and consequently \mathsf{G}(A,\sigma)\subseteq\mathsf{G}(A^{\prime},\sigma^{\prime})={K}^{\times 2}\operatorname{\mathsf{Nrd}}_{C^{\prime}}^{\ast}={K}^{\times 2}\operatorname{\mathsf{Nrd}}_{C}^{\ast} . ∎ 5. Norms from splitting 2 -extensions We study the situation where \mathsf{I}^{3}K(\sqrt{-1})=0 . We revisit some results from [ 17 ] and [ 2 ] , replacing the hypothesis that \mathsf{cd}_{2}(K(\sqrt{-1}))\leqslant 2 there by the milder condition that \mathsf{I}^{3}K(\sqrt{-1})=0 . We denote by \mathsf{I}_{\mathsf{tors}}K the torsion ideal of \mathsf{W}K and set \mathsf{I}^{n}_{\mathsf{tors}}K=\mathsf{I}^{n}K\cap\mathsf{I}_{\mathsf{tors}}K for n\in\mathbb{N} . Proposition 5.1 . Let n\in\mathbb{N} be such that \mathsf{I}^{n+1}K(\sqrt{-1})=0 . Let \varphi\in\mathsf{I}^{n}_{\mathsf{tors}}K . Then \mathsf{Hyp}_{2}^{\ast}(\varphi)={K}^{\times} . Proof. This is [ 1 , Prop. 4.1] . ∎ A central theme of this and the next section is the quest for extensions of 5.1 to algebras with involution, replacing conditions such as being torsion accordingly, in particular for n=2 . Let A be a central simple K -algebra. We associate the quadratic form \operatorname{\mathsf{T}}_{A}:A\to K,x\mapsto\operatorname{\mathsf{Trd}}_{A}(x^{2}) over K , called the trace form of A . For an ordering P\in\mathsf{X}(K) , one has \mathsf{sign}_{P}(\operatorname{\mathsf{T}}_{A})=\pm\mathsf{deg}A , or more precisely, \mathsf{sign}_{P}(\operatorname{\mathsf{T}}_{A})=\begin{cases}+\mathsf{deg}(A)&\mbox{ if $A_{K_{P}}$ is split,}\\ -\mathsf{deg}(A)&\mbox{ if $A_{K_{P}}\sim(-1,-1)_{K_{P}}$\,.}\end{cases} We call the central simple K -algebra A totally positive if \mathsf{sign}_{P}(\operatorname{\mathsf{T}}_{A})>0 for all P\in\mathsf{X}(K) , or equivalently, if A splits over every real closure of K . Note that, if K is nonreal, then every central simple K -algebra is totally positive. (Without giving it a name, the condition of being totally positive has been considered in various preceding works, in particular [ 3 ] . In [ 4 ] , the term ‘nonreal’ was used instead of ‘totally positive’, but that leads to a conflict of terminology when applied to A=K .) The following is a refinement of a characterization of reduced norms given in [ 3 , Theorem 2.1] under the stronger assumption that \mathsf{cd}_{2}(K(\sqrt{-1}))\leqslant 2 . When K is nonreal, the statement corresponds to [ 17 , Prop. 2.1] . Proposition 5.2 . Assume that \mathsf{I}^{3}K(\sqrt{-1})=0 . Let A be a central simple K -algebra representing a 2 -primary torsion element of \operatorname{\mathsf{Br}}(K) . For a\in{K}^{\times} , the following are equivalent: (i) a\in\operatorname{\mathsf{Nrd}}_{A}^{\ast} . (ii) a\in P for all P\in\mathsf{X}(K) with \mathsf{sign}_{P}(\operatorname{\mathsf{T}}_{A})<0 . (iii) a\in\mathsf{N}_{L/K}^{\ast} for a finite 2 -extension L/K such that A_{L} is split. In particular, if A is totally positive, then \operatorname{\mathsf{Nrd}}_{A}^{\ast}={K}^{\times} . Proof. (iii\Rightarrow i) This follows from 4.1 (b) . (i\Rightarrow ii) Let P\in\mathsf{X}(K) be an ordering such that \mathsf{sign}_{P}(A)<0 . Then A_{K_{P}}\sim(-1,-1)_{K_{P}} , and hence \operatorname{\mathsf{Nrd}}_{A_{K_{P}}}^{\ast}=\operatorname{\mathsf{Nrd}}_{(-1,-1)_{K_{P}}}^{\ast}={(K_{P})}^{\times 2} , in view of 4.1 (a) . Therefore \operatorname{\mathsf{Nrd}}_{A}^{\ast}\subseteq{K}^{\times}\cap{(K_{P})}^{\times 2}\subseteq P . (ii\Rightarrow iii) Consider a\in{K}^{\times} such that a\in P for every P\in\mathsf{X}(K) with \mathsf{sign}_{P}(\operatorname{\mathsf{T}}_{A})<0 . Let K^{\prime}=K(\sqrt{-a}) . For P^{\prime}\in\mathsf{X}(K^{\prime}) , letting P=P^{\prime}\cap K , we have P\in\mathsf{X}(K) and a\notin P , whereby \mathsf{sign}_{P^{\prime}}(\operatorname{\mathsf{T}}_{A_{K^{\prime}}})=\mathsf{sign}_{P}(\operatorname{\mathsf{T}}_{A})>0 . Hence A_{K^{\prime}} is totally positive. We proceed by induction on the smallest r\in\mathbb{N} such that A^{\otimes 2^{r}} is split. If r=0 then A_{K^{\prime}} is split and we may take L=K^{\prime} . Assume now that r>0 . The induction hypothesis applied to A^{\prime}=A^{\otimes 2^{r-1}} yields that \sqrt{-a}=\mathsf{N}_{K^{\prime\prime}/K^{\prime}}(c) for a finite 2 -extension K^{\prime\prime}/K^{\prime} such that A^{\prime}_{K^{\prime\prime}} is split and some c\in{K^{\prime\prime}}^{\times} . Then K^{\prime\prime}/K is a finite 2 -extension, whereby \mathsf{I}^{3}K^{\prime\prime}(\sqrt{-1})=0 , in view of 2.4 . Since A^{\prime}_{K^{\prime\prime}} is split, we have [A_{K^{\prime\prime}}]\in\operatorname{\mathsf{Br}}_{2}(K^{\prime\prime}) . Moreover, A_{K^{\prime\prime}} is totally positive. It follows by [ 4 , Prop. 5.5] that [A_{K^{\prime\prime}}]=c(\psi) for a quadratic form \psi\in\mathsf{I}^{2}_{\mathsf{tors}}K^{\prime\prime} . By 5.1 , \mathsf{Hyp}_{2}^{\ast}(\psi)={K^{\prime\prime}}^{\times} , so there exists a finite 2 -extension L/K^{\prime\prime} such that c\in\mathsf{N}_{L/K^{\prime\prime}}^{\ast} and \psi_{L} is hyperbolic. Then L/K is a finite 2 -extension with [A_{L}]=c(\psi_{L})=0 . So a=\mathsf{N}_{K^{\prime}/K}(\sqrt{-a})=\mathsf{N}_{K^{\prime\prime}/K}(c)\in\mathsf{N}_{K^{\prime\prime}/K}(\mathsf{N}_{L/K}^{\ast})=\mathsf{N}_{L/K}^{\ast} and A_{L} is split. ∎ Consider a K -algebra with involution (A,\sigma) . If \sigma is orthogonal, then we say (A,\sigma) (or simply \sigma ) has trivial Clifford invariant if \mathsf{deg}A is even, \mathsf{disc}(\sigma) is trivial and at least one of the two components of the Clifford algebra \mathcal{C}(A,\sigma) is split. We write (A,\sigma)\in\mathsf{I}^{2} to indicate that \mathsf{deg}A is even and either \sigma is orthogonal with trivial discriminant or \sigma is symplectic. We write (A,\sigma)\in\mathsf{I}^{3} to indicate that either \sigma is orthogonal with trivial Clifford invariant or \sigma is symplectic with \mathsf{deg}A\equiv 0\,\mathsf{mod}\,4 . We refer to [ 15 ] and [ 6 ] for the definition of signatures of an involution of the first kind. Recall that n\times\langle 1\rangle is the n -dimensional quadratic form \langle 1,\dots,1\rangle , for n\in\mathbb{N} . Let \Phi be a K -algebra with involution. We denote the K -algebra with involution \operatorname{\mathsf{Ad}}(n\times\langle 1\rangle)\otimes\Phi by n\times\Phi . We call \Phi weakly hyperbolic if n\times\Phi is hyperbolic for some n\in\mathbb{N}^{+} . Note that, if K is nonreal, then every K -algebra with involution \Phi is weakly hyperbolic, because for r\in\mathbb{N} such that -1 is a sum of 2^{r} squares in K , we have that 2^{r+1}\times\langle 1\rangle is hyperbolic, whereby 2^{r+1}\times\Phi is hyperbolic. Weakly hyperbolic K -algebras with involution were characterized in [ 21 ] and [ 6 ] . The following statement extends [ 2 , Theorem 4.5] in the orthogonal case. Note that the split case corresponds to 5.1 for n=2 . Proposition 5.3 . Assume that \mathsf{I}^{3}K(\sqrt{-1})=0 . Let (A,\sigma) be a K -algebra with involution. Assume that A is totally positive, (A,\sigma) is weakly hyperbolic and (A,\sigma)\in\mathsf{I}^{2} . Then \mathsf{Hyp}_{2}^{\ast}(A,\sigma)=\mathsf{G}(A,\sigma)={K}^{\times}\,. Proof. In view of 3.2 , we need only to show that {K}^{\times}\subseteq\mathsf{Hyp}_{2}^{\ast}(A,\sigma) . Since A carries a K -linear involution, we have [A]\in\operatorname{\mathsf{Br}}_{2}(K) . Consider a\in{K}^{\times} arbitrary. By 5.2 , since A is totally positive, there exists a finite 2 -extension L/K such that a\in\mathsf{N}_{L/K}^{\ast} and A_{L} is split. If \sigma is symplectic, then \sigma_{L} is hyperbolic, whereby a\in\mathsf{N}_{L/K}^{\ast}\subseteq\mathsf{Hyp}_{2}^{\ast}(A,\sigma) . Assume that \sigma is orthogonal. Then (A_{L},\sigma_{L})\simeq\operatorname{\mathsf{Ad}}(\varphi) for a quadratic form \varphi over L . It follows from the hypothesis on (A,\sigma) that \operatorname{\mathsf{Ad}}(\varphi) is weakly hyperbolic and belongs to \mathsf{I}^{2} . This means that \varphi\in\mathsf{I}^{2}_{\mathsf{tors}}L . By 2.4 , as \mathsf{I}^{3}K(\sqrt{-1})=0 , we have \mathsf{I}^{3}L(\sqrt{-1})=0 . Hence, 5.1 yields that \mathsf{Hyp}_{2}^{\ast}(A_{L},\sigma_{L})=\mathsf{Hyp}_{2}^{\ast}(\varphi)={L}^{\times} . Therefore a\in\mathsf{N}_{L/K}^{\ast}\subseteq\mathsf{Hyp}_{2}^{\ast}(A,\sigma) . ∎ We set {(\mathsf{\Sigma}{K}^{2})}^{\times}=\mathsf{\Sigma}{K}^{2}\cap{K}^{\times} , which is a subgroup of {K}^{\times} . Corollary 5.4 . Assume that \mathsf{I}^{3}K(\sqrt{-1})=0 . Let (A,\sigma) be K -algebra with orthogonal involution with (A,\sigma)\in\mathsf{I}^{2} . Then {K}^{\times 2}\subseteq{(\mathsf{\Sigma}{K}^{2})}^{\times}\subseteq\mathsf{Hyp}_{2}^{\ast}(A,\sigma)\,. Furthermore, if A is totally positive, then \mathsf{G}(A,\sigma)=\mathsf{Hyp}_{2}^{\ast}(A,\sigma)\,. Proof. Consider first a\in{(\mathsf{\Sigma}{K}^{2})}^{\times} . Set L=K(\sqrt{-a}) . Then L is nonreal and \mathsf{I}^{3}L(\sqrt{-1})=0 , by 2.4 . Hence \mathsf{Hyp}_{2}^{\ast}(A_{L},\sigma_{L})={L}^{\times} , by 5.3 . In particular a=\mathsf{N}_{L/K}(\sqrt{-a})\subseteq\mathsf{Hyp}_{2}^{\ast}(A,\sigma) . Consider now an arbitrary element a\in\mathsf{G}(A,\sigma) . Let K^{\prime}=K(\sqrt{-a}) . Then -1\in\mathsf{G}(A_{L},\sigma_{L}) , or in different terms, 2\times(A_{L},\sigma_{L}) is hyperbolic. In particular, (A_{L},\sigma_{L}) is weakly hyperbolic. Moreover, since A is totally positive, so is A_{L} . By 2.4 , as \mathsf{I}^{3}K(\sqrt{-1})=0 , we have \mathsf{I}^{3}K^{\prime}(\sqrt{-1})=0 . We obtain by 5.3 that \mathsf{Hyp}_{2}^{\ast}(A_{K^{\prime}},\sigma_{K^{\prime}})={K^{\prime}}^{\times} . In particular, we have a=\mathsf{N}_{K^{\prime}/K}(\sqrt{-a})\in\mathsf{N}_{K^{\prime}/K}(\mathsf{Hyp}_{2}^{\ast}(A_{K^{\prime}},\sigma_{K^{\prime}}))\subseteq\mathsf{Hyp}_{2}^{\ast}(A,\sigma) . This shows that \mathsf{G}(A,\sigma)\subseteq\mathsf{Hyp}_{2}^{\ast}(A,\sigma) , and the opposite inclusion holds by 3.2 . ∎ When aiming to show that \mathsf{Hyp}_{2}(\Phi)={K}^{\times} for a K -algebra with involution \Phi of a certain type under the assumption that \mathsf{I}^{3}K(\sqrt{-1})=0 , a reduction can be made to the case where A (or another central simple K -algebra of exponent 2 ) splits over K(\sqrt{-1}) . This reduction method stems from [ 8 , Prop. 2.3] and [ 3 , Prop. 7.7] , and was further refined in [ 17 , Section 2] and ingeniously used in [ 17 , Prop. 4.1 & Lemma 6.5] . We simplify this method and adapt it to the hypothesis that \mathsf{I}^{3}K(\sqrt{-1})=0 . The following is a refinement of [ 8 , Prop. 2.3] . Lemma 5.5 . Let \psi be a 4 -dimensional quadratic form over K and c\in\mathsf{disc}(\psi) . Then {K}^{\times}\cap\mathsf{G}(\psi_{K(\sqrt{c})}) is the union of the subgroups \mathsf{N}_{K_{1}/K}^{\ast}\cdot\mathsf{N}_{K_{2}/K}^{\ast} where K_{i}/K are field extensions with [K_{i}:K]\leqslant 2 such that \psi_{K_{i}} is isotropic for i\in\{1,2\} . Proof. Note that \psi_{K(\sqrt{c})} is a scaled 2 -fold Pfister form. For any field extension K^{\prime}/K such that \psi_{K^{\prime}} is isotropic, we obtain that \psi_{K^{\prime}(c)} is hyperbolic, whereby \mathsf{N}_{K^{\prime}/K}^{\ast}\subseteq{K}^{\times}\cap\mathsf{N}_{K^{\prime}(\sqrt{c})/K(\sqrt{c})}^{\ast}\subseteq{K}^{\times}\cap\mathsf{G}(\psi_{K(\sqrt{c})}) , by 3.2 . Consider now a\in{K}^{\times}\cap\mathsf{G}(\psi_{K(\sqrt{c})}) . Since the claim is invariant under scaling of \psi , we may assume that 1\in\mathsf{D}(\psi) . Then \mathsf{G}(\psi_{K(\sqrt{c})})=\mathsf{D}(\psi_{K(\sqrt{c})}) . Hence, there exist x,y\in K^{4} such that a=\psi(x+\sqrt{c}y) . It follows that x and y are orthogonal with respect to \psi and a=\psi(x)+c\psi(y) . We construct field extensions K_{i}/K for i\in\{1,2\} such that \psi_{K_{i}} is isotropic, [K_{i}:K]\leqslant 2 , and such that a\in\mathsf{N}_{K_{1}/K}^{\ast}\cdot\mathsf{N}_{K_{2}/K}^{\ast} holds. Assume first that \psi(y)=0 . There exists a (regular) binary subform \beta of \psi which represents 1 and \psi(x) . We take d\in\mathsf{disc}(\beta) and set K_{1}=K_{2}=K(\sqrt{d}) to satisfy the claim. Assume now that \psi(x)=0 . We then choose a (regular) binary subform \beta of \psi representing 1 and \psi(y) and then set K_{1}=K(\sqrt{d}) for some d\in\mathsf{disc}(\beta) . Then -acd=(-d)c^{2}\psi(y)\in\mathsf{N}_{K_{1}/K}^{\ast} and \psi_{K_{1}} is isotropic. Furthermore \psi\simeq\beta\perp\beta^{\prime} for a binary subform \beta^{\prime} of \psi with \mathsf{disc}(\beta^{\prime})=cd{K}^{\times 2} . Letting K_{2}=K(\sqrt{cd}) , we obtain that \psi_{K_{2}} is isotropic and -cd\in\mathsf{N}_{K_{2}/K}^{\ast} , whereby a\in\mathsf{N}_{K_{1}/K}^{\ast}\cdot\mathsf{N}_{K_{2}/K}^{\ast} . Assume finally that \psi(x)\psi(y)\neq 0 . We set a_{1}=\psi(x) and a_{2}=\psi(y) , and obtain that \langle a_{1},a_{2}\rangle is a binary subform of \psi . It follows that \psi\simeq\langle a_{1},a_{2},b,a_{1}a_{2}bc\rangle for some b\in{K}^{\times} . There further exists a (regular) binary subform \beta of \psi with 1,a_{1}\in\mathsf{D}(\beta) . We set K_{1}=K(\sqrt{d}) for d\in\mathsf{disc}(\beta) , so that a_{1}\in\mathsf{N}_{K_{1}/K}^{\ast} . Let K_{2}=K(\sqrt{-a_{1}a_{2}c}) . Since a=a_{1}+ca_{2} , we have aa_{1}\in\mathsf{N}_{K_{2}/K}^{\ast} , so a\in\mathsf{N}_{K_{1}/K}^{\ast}\cdot\mathsf{N}_{K_{2}/K}^{\ast} , and \psi_{K_{i}} is isotropic for i\in\{1,2\} . ∎ Lemma 5.6 . Assume that \mathsf{I}^{3}K(\sqrt{-1})=0 . Let k\in\mathbb{N} . Let \varphi be a quadratic form over K with \mathsf{dim}(\varphi)=2k+2 and \mathsf{disc}(\varphi)=\pm 1\cdot{K}^{\times 2} . Then there exists a set \mathcal{L} of 2 -extensions of K such that [L:K]\leqslant 2^{k} and \varphi_{L(\sqrt{-1})} is hyperbolic for every L\in\mathcal{L} and such that {K}^{\times}=\langle\bigcup_{L\in\mathcal{L}}\mathsf{N}_{L/K}^{\ast}\rangle . Proof. The proof is by induction on k . If k=0 , we take \mathcal{L}=\{K\} . Assume that k>0 . We fix a_{1},a_{2},a_{3}\in{K}^{\times} such that \langle a_{1},a_{2},a_{3}\rangle is a subform of \varphi and set \psi=\langle a_{1},a_{2},a_{3},-a_{1}a_{2}a_{3}\rangle . Then \psi_{K(\sqrt{-1})}\in\mathsf{I}^{2}K(\sqrt{-1}) , and since \mathsf{I}^{3}K(\sqrt{-1})=0 , it follows that \mathsf{G}(\psi_{K(\sqrt{-1})})={K(\sqrt{-1})}^{\times} . Consider a\in{K}^{\times} arbitrary. 5.5 applied with c=-1 yields that there exist field extensions K_{i}/K with [K_{i}:K]\leqslant 2 such that \psi_{K_{i}} is isotropic and elements a_{i}\in{K_{i}}^{\times} for i\in\{1,2\} such that a=\mathsf{N}_{K_{1}/K}(a_{1})\cdot\mathsf{N}_{K_{2}/K}(a_{2}) . Fix i\in\{1,2\} . Since \varphi_{K_{i}(\sqrt{-1})} is isotropic. It follows that \varphi_{K_{i}}\simeq\varphi^{\prime}\perp\langle d,d\rangle for some d\in{K_{i}}^{\times} and a quadratic form \varphi^{\prime}\in\mathsf{I}^{2}K_{i} with \mathsf{dim}(\varphi^{\prime})=2k . Since \mathsf{I}^{3}K_{i}(\sqrt{-1})=0 , by 2.4 , the induction hypothesis for k-1 applies to \varphi^{\prime} over K_{i} . We obtain a family \mathcal{L}_{i}^{a} of 2 -extensions L/K_{i} with [L:K_{i}]\leqslant 2^{k-1} such that \varphi_{L(\sqrt{-1})} is hyperbolic and a_{i}\in\langle\bigcup_{L\in\mathcal{L}^{a}_{i}}\mathsf{N}_{L/K}^{\ast}\rangle . We set \mathcal{L}^{a}=\mathcal{L}_{1}^{a}\cup\mathcal{L}_{2}^{a} . Then a\in\langle\bigcup_{L\in\mathcal{L}^{a}}\mathsf{N}_{L/K}^{\ast}\rangle and \mathcal{L}^{a} consists of 2 -extensions L/K with [L:K]\leqslant 2^{k} and such that \varphi_{L(\sqrt{-1})} is hyperbolic. Constructing \mathcal{L}^{a} for each a\in{K}^{\times} in this way, the claim is satisfied for \mathcal{L}=\bigcup_{a\in{K}^{\times}}\mathcal{L}^{a} . ∎ For symplectic involutions on A , the following recovers [ 17 , Prop. 4.1] , where \mathsf{cd}_{2}(K(\sqrt{-1}))\leqslant 2 is assumed. We follow the ideas of [ 17 , Prop. 4.1 & Lemma 6.5] . Theorem 5.7 . Assume that \mathsf{I}^{3}K(\sqrt{-1})=0 . Let (A,\sigma) be a weakly hyperbolic K -algebra with involution. If \sigma is orthogonal, then assume that (A,\sigma)\in\mathsf{I}^{3} . Then \mathsf{Hyp}_{2}(A,\sigma)={K}^{\times}\,. Proof. Consider a central simple K -algebra D with [D]\in\operatorname{\mathsf{Br}}_{2}(K) . 2.1 yields that [D]=c(\varphi) for a quadratic form \varphi\in\mathsf{I}^{2}K . By 5.6 , we can find a family \mathcal{L}_{D} of finite 2 -extensions L/K for which \varphi_{L(\sqrt{-1})} is hyperbolic and such that {K}^{\times}=\langle\bigcup_{L\in\mathcal{L}_{D}}\mathsf{N}_{L/K}^{\ast}\rangle . Then, for L\in\mathcal{L}_{D} , D_{L(\sqrt{-1})} is split and \mathsf{I}^{3}L(\sqrt{-1})=0 , by 2.4 . If now \mathsf{N}_{L/K}^{\ast}\subseteq\mathsf{Hyp}_{2}(A,\sigma) holds for each L\in\mathcal{L}_{D} , then we conclude that \mathsf{Hyp}_{2}(A,\sigma)={K}^{\times} . If \sigma is symplectic, we apply this with D=A , associating a set of finite 2 -extensions \mathcal{L}_{A} accordingly. Let L\in\mathcal{L}_{A} . Then (A_{L},\sigma_{L})\simeq\operatorname{\mathsf{Ad}}(\varphi)\otimes(Q,\operatorname{\mathsf{can}}_{Q}) for an L -quaternion algebra Q and a quadratic form \varphi over L . Let \pi denote the norm form of Q . Since (A_{L},\sigma_{L}) is weakly hyperbolic, we have \varphi\otimes\pi\in\mathsf{I}_{\mathsf{tors}}K . If \mathsf{dim}(\varphi) is even, then \varphi\otimes\pi\in\mathsf{I}^{3}_{\mathsf{tors}}L , and as \mathsf{I}^{3}L(\sqrt{-1})=0 , we have \mathsf{I}^{3}_{\mathsf{tors}}L=0 , by [ 11 , Cor. 35.27] , whence \varphi\otimes\pi is hyperbolic, whereby (A,\sigma)_{L} is hyperbolic. If \mathsf{dim}(\varphi) is odd, then \pi\in\mathsf{I}_{\mathsf{tors}}^{2}K , and we conclude that \operatorname{\mathsf{Nrd}}_{Q}^{\ast}={L}^{\times} . In either case, we obtain that \mathsf{Hyp}_{2}(A_{L},\sigma_{L})={L}^{\times} , and hence \mathsf{N}_{L/K}^{\ast}\subseteq\mathsf{Hyp}_{2}(A,\sigma) . Assume now that \sigma is orthogonal. We consider the Rost invariant of (A,\sigma) . This is a class in H^{3}(K,\mu_{4}^{\otimes 2})/([A]\cup({K}^{\times})) . Let \xi\in H^{3}(K,\mu_{4}^{\otimes 2}) be a representative of this class. Since \mathsf{I}^{3}K(\sqrt{-1})=0 , it follows by 2.3 that \xi=(-1)\cup[D] for some central simple K -algebra D with [D]\in\operatorname{\mathsf{Br}}_{2}(K) . We now associate the set \mathcal{L}_{D} as above. Consider L\in\mathcal{L}_{D} . Then D\sim(-1,c)_{L} for some c\in{L}^{\times} , and consequently \xi=(-1)\cup(-1)\cup(c) . For every P\in\mathsf{X}(K) such that A_{K_{P}} is split, \sigma_{K_{P}} is hyperbolic, whereby \xi_{K_{P}}=0 and thus c\in P . This implies that \xi_{K_{P}}=[A_{K_{P}}]\cup(c) for every P\in\mathsf{X}(K) . We conclude by 2.3 that \xi_{L}=[A_{L}]\cup(c) . Hence the Rost invariant of (A_{L},\sigma_{L}) is trivial. Since \mathsf{I}^{3}K(\sqrt{-1})=0 , we have \mathsf{I}^{3}L(\sqrt{-1})=0 , by 2.4 . We conclude by [ 3 , Theorem 7.3] that \sigma_{L} is hyperbolic. Therefore \mathsf{N}_{L/K}^{\ast}\subseteq\mathsf{Hyp}_{2}(A,\sigma) . ∎ Corollary 5.8 . Assume that \mathsf{I}^{3}K(\sqrt{-1})=0 . Let (A,\sigma) be a K -algebra with orthogonal involution with (A,\sigma)\in\mathsf{I}^{3} . Then \mathsf{G}(A,\sigma)=\mathsf{Hyp}_{2}(A,\sigma). Proof. By 3.2 , we only need to show that \mathsf{G}(A,\sigma)\subseteq\mathsf{Hyp}_{2}(A,\sigma) . Consider a\in\mathsf{G}(A,\sigma) . Set K^{\prime}=K(\sqrt{-a}) . Then a=\mathsf{N}_{K^{\prime}/K}(\sqrt{-a}) . Moreover -1\in\mathsf{G}(A_{K^{\prime}},\sigma_{K^{\prime}}) , so 2\times(A_{K^{\prime}},\sigma_{K^{\prime}}) is hyperbolic. Hence (A_{K^{\prime}},\sigma_{K^{\prime}}) is weakly hyperbolic. Since \mathsf{I}^{3}K(\sqrt{-1})=0 , we have \mathsf{I}^{3}K^{\prime}(\sqrt{-1})=0 , by 2.4 . Therefore 5.7 yields that \mathsf{Hyp}_{2}(A_{K^{\prime}},\sigma_{K^{\prime}})={K^{\prime}}^{\times} , whereby a\in\mathsf{N}_{K^{\prime}/K}(\mathsf{Hyp}_{2}(A_{K^{\prime}},\sigma_{K^{\prime}}))\subseteq\mathsf{Hyp}_{2}(A,\sigma) . ∎ The following result extends [ 17 , Prop. 6.15] . Theorem 5.9 . Assume that \mathsf{I}^{3}K(\sqrt{-1})=0 . Let (A,\sigma) be a weakly hyperbolic K -algebra with orthogonal involution with (A,\sigma)\in\mathsf{I}^{2} . Let C_{1} and C_{2} be the two components of \mathcal{C}(A,\sigma) . Then \mathsf{G}(A,\sigma)=\mathsf{Hyp}_{2}(A,\sigma)=\operatorname{\mathsf{Nrd}}_{C_{1}}^{\ast}\cdot\operatorname{\mathsf{Nrd}}_{C_{2}}^{\ast}\,. Proof. We first show that \mathsf{G}(A,\sigma)\subseteq\operatorname{\mathsf{Nrd}}_{C_{1}}^{\ast}\cdot\operatorname{\mathsf{Nrd}}_{C_{2}}^{\ast} . Consider a\in\mathsf{G}(A,\sigma) . Then the K -algebra with involution \operatorname{\mathsf{Ad}}(\langle\!\langle a\rangle\!\rangle)\otimes(A,\sigma) is hyperbolic, so in particular its Rost invariant in H^{3}(K,\mu_{4}^{\otimes 2})/([A]\cup({K}^{\times})) is trivial. By [ 17 , Lemma 6.9] , this means that there exists x\in{K}^{\times} such that [C_{1}]\cup(a)=[A]\cup(x)\,. Recall that, by [ 15 , Theorem 9.12] , either 2[C_{1}]=[A] or [C_{1}]+[C_{2}]=[A] . Assume first that 2[C_{1}]=[A] . Then [C_{1}]\cup(ax^{-2})=0 . Hence, for every P\in\mathsf{X}(K) with \mathsf{sign}_{P}(\operatorname{\mathsf{T}}_{C_{1}})<0 , since (C_{1})_{K_{P}}\sim(-1,-1)_{K_{P}} , we must have a\in P . Therefore, in this case, 5.2 yields that a\in\operatorname{\mathsf{Nrd}}_{C_{1}}^{\ast} , and we are done. Assume now that [C_{1}]+[C_{2}]=[A] . Then [C_{1}]\cup(ax)=[C_{2}]\cup(x)\,. Consider an ordering P\in\mathsf{X}(K) . Since \sigma_{K_{P}} is hyperbolic, (C_{1})_{K_{P}} or (C_{2})_{K_{P}} is split, whereby [(C_{1})_{K_{P}}]\cup(ax)=[(C_{2})_{K_{P}}]\cup(x)=0\,. Therefore, on the one hand, either (C_{1})_{K_{P}} is split or ax\in P , and on the other hand, either (C_{2})_{K_{P}} is split or x\in P . Having this for all P\in\mathsf{X}(K) , we conclude by 5.2 that ax\in\operatorname{\mathsf{Nrd}}_{C_{1}}^{\ast} and x\in\operatorname{\mathsf{Nrd}}_{C_{2}}^{\ast} , whereby a=axx^{-1}\in\operatorname{\mathsf{Nrd}}_{C_{1}}^{\ast}\cdot\operatorname{\mathsf{Nrd}}_{C_{2}}^{\ast} . It remains to show that \operatorname{\mathsf{Nrd}}_{C_{1}}^{\ast}\cdot\operatorname{\mathsf{Nrd}}_{C_{2}}^{\ast}\subseteq\mathsf{Hyp}_{2}(A,\sigma) . Consider i\in\{1,2\} and a\in\operatorname{\mathsf{Nrd}}_{C_{i}}^{\ast} . By 5.2 , there exists a finite 2 -extension L/K such that a\in\mathsf{N}_{L/K}^{\ast} and (C_{i})_{L} is split. Then (A_{L},\sigma_{L}) has trivial Clifford invariant. Since \mathsf{I}^{3}K(\sqrt{-1})=0 , we have \mathsf{I}^{3}L(\sqrt{-1})=0 , by 2.4 . Since (A_{L},\sigma_{L}) is weakly hyperbolic, we obtain by 5.7 that \mathsf{Hyp}_{2}(A_{L},\sigma_{L})={L}^{\times} . Hence a\in\mathsf{N}_{L/K}^{\ast}\subseteq\mathsf{Hyp}_{2}(A,\sigma) . ∎ Following [ 26 ] , K is called an ED-field if every 2 -dimensional torsion form over K represents an element of {(\mathsf{\Sigma}{K}^{2})}^{\times} . This property is stable under 2 -extensions, by [ 26 , Cor. 3] , and it implies the Strong Approximation Property ( SAP ) introduced in [ 12 ] . Assuming that \mathsf{I}^{3}K(\sqrt{-1})=0 and K is an ED -field, it is shown in [ 17 , Theorem 7.11] that \mathsf{Hyp}(A,\sigma)=\mathsf{G}(A,\sigma)={K}^{\times} if (A,\sigma) is weakly hyperbolic. It is apparent from the discussion in [ 17 , Section 7] that the ED -property is crucial for this conclusion. The following general example confirms this. Example 5.10 . Assume that \mathsf{I}^{3}K(\sqrt{-1})=0 , but K is not a SAP -field (and in particular not an ED -field). (For example, let K be either \mathbb{R}(x,y) or \mathbb{R}(\!(x)\!)(\!(y)\!) where x and y are variables.) Hence, there exist x,y\in{K}^{\times} (e.g. the two variables in the examples) such that every multiple of the quadratic form \langle 1,x,y,-xy\rangle over K is anisotropic. Consider then the K -quaternion algebras Q_{1}=(-1,y)_{K} , Q_{2}=(-1,-y)_{K} and the K -algebra with orthogonal involution (A,\sigma)_{K}=(Q_{1},\operatorname{\mathsf{can}}_{Q_{1}})\otimes(Q_{2},\operatorname{\mathsf{can}}_{Q_{2}}) . Note that \mathsf{G}(A,\sigma)=\operatorname{\mathsf{Nrd}}_{Q_{1}}^{\ast}\cdot\operatorname{\mathsf{Nrd}}_{Q_{2}}^{\ast} and 2\times(A,\sigma) is hyperbolic, whereby (A,\sigma) is weakly hyperbolic. On the other hand, since the form \langle\!\langle-1,y\rangle\!\rangle\perp x\langle\!\langle-1,-y\rangle\!\rangle=2\times\langle 1,x,y,-xy\rangle over K is anisotropic, we have that -x\notin\operatorname{\mathsf{Nrd}}_{Q_{1}}^{\ast}\cdot\operatorname{\mathsf{Nrd}}_{Q_{2}}^{\ast}=\mathsf{G}(A,\sigma) (see also [ 17 , Cor. 6.5] ), whereby \mathsf{G}(A,\sigma)\neq{K}^{\times} . Hence, the condition in 5.3 that A is totally positive is not superfluous. 6. Involutions on index- 2 algebras Consider a K -quaternion algebra Q and a K -algebra with orthogonal involution (A,\sigma) such that A\sim Q . Then A_{K_{Q}} is split and (A,\sigma)_{K_{Q}}\simeq\operatorname{\mathsf{Ad}}(\varphi) for a quadratic form \varphi defined over K_{Q} , which is unique up to similarity. For m\in\mathbb{N} write (A,\sigma)\in\mathsf{I}^{m} to indicate that (A,\sigma)_{K_{Q}}\simeq\operatorname{\mathsf{Ad}}(\varphi) for a quadratic form \varphi over K_{Q} whose class lies in \mathsf{I}^{m}K_{Q} . If A is split, then so is Q , whereby K_{Q}/K is rational, and it follows that (A,\sigma)\in\mathsf{I}^{m} holds if and only if (A,\sigma)\simeq\operatorname{\mathsf{Ad}}(\psi) for a quadratic form \psi\in\mathsf{I}^{m}K . Starting from a K -algebra with orthogonal involution (A,\sigma) with \operatorname{\mathsf{ind}}A\leqslant 2 , there is up to isomorphism a unique K -quaternion algebra Q\sim A , and hence it unnecessary to mention Q for stating that (A,\sigma)\in\mathsf{I}^{m} . Note that these conventions for being in \mathsf{I}^{m} are compatible with those we took before for m\in\{2,3\} . Theorem 6.1 . Let n\in\mathbb{N} be such that \mathsf{I}^{n+1}K=0 . Let (A,\sigma) a K -algebra with orthogonal involution with \operatorname{\mathsf{ind}}A\leqslant 2 and (A,\sigma)\in\mathsf{I}^{n} . Then \sigma is hyperbolic if and only if (A,\sigma)\in\mathsf{I}^{n+1} . Proof. Let Q be the K -quaternion algebra with A\sim Q and \varphi a quadratic form over K_{Q} such that (A,\sigma)\simeq\operatorname{\mathsf{Ad}}(\varphi) . If \sigma is hyperbolic, then \sigma_{K_{Q}} is hyperbolic, whereby \varphi_{K_{Q}}=0\in\mathsf{I}^{n+1}K_{Q} , so (A,\sigma)\in\mathsf{I}^{n+1} . Assume that \sigma is not hyperbolic. By Dejaiffe’s Theorem [ 10 ] , then \sigma_{K_{Q}} is not hyperbolic, hence \varphi is not hyperbolic. By the Arason-Pfister Hauptsatz [ 11 , Theorem 6.18] , there exists m\in\mathbb{N} such that \varphi\in\mathsf{I}^{m}K_{Q}\smallsetminus\mathsf{I}^{m+1}K_{Q} . By 2.2 , then e_{m}(\varphi)\neq 0 in H^{m}(K,\mu_{2}) . It follows by [ 7 , Prop. 7 & Prop. 9] that e_{m}([\varphi_{K_{Q}}]) lies in the image of the restriction homomorphism H^{m}(K,\mu_{4}^{\otimes m-1})\to H^{m}(K_{Q},\mu_{2}) . Hence H^{m}(K,\mu_{4}^{\otimes m-1})\neq 0 . This implies that \mathsf{H}^{m}(K,\mu_{2})\neq 0 and hence, by 2.2 , that \mathsf{I}^{m}K\neq 0 . Therefore m\leqslant n and \varphi\notin\mathsf{I}^{n+1}K_{Q} , so that (A,\sigma)\notin\mathsf{I}^{n+1} . ∎ Proposition 6.2 . Let n\in\mathbb{N} be such that \mathsf{I}^{n+1}K(\sqrt{-1})=0 . Let k,d\in\mathbb{N} and \xi\in H^{n}(K,\mu_{2^{k}}^{\,\otimes d}) such that \xi_{K_{P}}=0 for every P\in\mathsf{X}(K) . Then, for any c\in{K}^{\times} , there exists a finite 2 -extension L/K with c\in\mathsf{N}_{L/K}^{\ast} and \xi_{L}=0 . Proof. The proof is by induction on k . If k=0 , then \xi=0 and the claim holds with L=K . Assume now that k>0 . As \xi\in H^{n}(K,\mu_{2^{k}}^{\,\otimes d}) , we have 2^{k-1}\cdot\xi\in H^{n}(K,\mu_{2}) . The short exact sequence 1\rightarrow\mu_{2}\rightarrow\mu_{2^{k}}^{\otimes d}\stackrel{{\scriptstyle\cdot^{2}}}{{\rightarrow}}\mu_{2^{k-1}}^{\otimes d}\rightarrow 1 induces an exact sequence of cohomology groups: \cdots\rightarrow H^{n}(K,\mu_{2})\rightarrow H^{n}(K,\mu_{2^{k}}^{\otimes d})\stackrel{{\scriptstyle\cdot 2^{k-1}}}{{\rightarrow}}H^{n}(K,\mu_{2^{k-1}}^{\otimes d})\rightarrow\cdots By the surjectivity of e_{n}:\mathsf{I}^{n}K\to H^{n}(K,\mu_{2}) , there exists \phi\in\mathsf{I}^{n}K such that e_{n}(\phi)=2^{k-1}\cdot\xi . For any ordering P\in\mathsf{X}(K) , we have e_{n}(\phi_{K_{P}})=2^{k-1}\cdot\xi_{K_{P}}=0 , whereby \mathsf{sign}_{P}(\phi)\in 2^{n+1}\mathbb{Z} . It follows by [ 11 , Cor. 35.27] together with [ 16 , Theorem 1] that \phi\equiv\psi\,\mathsf{mod}\,\mathsf{I}^{n+1}K for some quadratic form \psi\in\mathsf{I}^{n}_{\mathsf{tors}}K . Consider some c\in{K}^{\times} . By 5.1 , there exists a finite 2 -extension K^{\prime}/K such that \psi_{K^{\prime}} is hyperbolic and c\in\mathsf{N}_{K^{\prime}/K}^{\ast} . Then \phi_{K^{\prime}}\in\mathsf{I}^{n+1}K^{\prime} , whereby 2^{k-1}\cdot\xi_{K^{\prime}}=e_{n}(\phi_{K^{\prime}})=0 . Since K^{\prime}/K is a 2 -extension and \mathsf{I}^{n+1}K(\sqrt{-1})=0 , we have \mathsf{I}^{n+1}K^{\prime}(\sqrt{-1})=0 , by 2.4 . Fix c^{\prime}\in{K^{\prime}}^{\times} with c=\mathsf{N}_{K^{\prime}/K}(c^{\prime}) . By the induction hypothesis, there exists a finite 2 -extension L/K^{\prime} such that \xi^{\prime}_{L}=0 and c^{\prime}\in\mathsf{N}_{L/K^{\prime}}^{\ast} . Then L/K is a finite 2 -extension, \xi_{L}=0 and we have c=\mathsf{N}_{K^{\prime}/K}(c^{\prime})\in\mathsf{N}_{K^{\prime}/K}(\mathsf{N}_{L/K^{\prime}}^{\ast})=\mathsf{N}_{L/K}^{\ast} . ∎ In the next two statements, we want to include the symplectic case. Assume that (A,\sigma) is a K -algebra with symplectic involution with \operatorname{\mathsf{ind}}A\leqslant 2 . Let Q be the K -quaternion algebra with Q\sim A . We denote by \operatorname{\mathsf{can}}_{Q} the canonical involution on Q . Then (A,\sigma)\simeq\operatorname{\mathsf{Ad}}(\varphi)\otimes(Q,\operatorname{\mathsf{can}}_{Q}) for a quadratic form \varphi over K . Let \pi denote the norm form of Q . By Jacobson’s Theorem (see e.g. [ 6 , Theorem 4.6] ), the properties of the involution \sigma and of the quadratic form \pi\otimes\varphi concerning hyperbolicity and similarity factors are the same, and this also over all field extensions. In particular, we have \mathsf{G}(A,\sigma)=\mathsf{G}(\pi\otimes\varphi) , \mathsf{Hyp}_{2}^{\ast}(A,\sigma)=\mathsf{Hyp}_{2}^{\ast}(\pi\otimes\varphi) , etc. For m\in\mathbb{N} , we write (A,\sigma)\in\mathsf{I}^{m} if \pi\otimes\varphi\in\mathsf{I}^{m}K . Theorem 6.3 . Assume that \mathsf{I}^{n+1}K(\sqrt{-1})=0 . Let (A,\sigma) be a weakly hyperbolic K -algebra with involution with (A,\sigma)\in\mathsf{I}^{n} . Assume that A is totally positive and \operatorname{\mathsf{ind}}A\leqslant 2 . Then \mathsf{G}(A,\sigma)=\mathsf{Hyp}_{2}^{\ast}(A,\sigma)={K}^{\times}\,. Proof. Let Q denote the K -quaternion algebra Brauer equivalent to A . Assume first that \sigma is orthogonal. Then (A,\sigma)_{K_{Q}}\simeq\operatorname{\mathsf{Ad}}(\varphi) for some quadratic form \varphi\in\mathsf{I}^{n}K_{Q} . Consider e_{n}(\varphi)\in H^{n}(K_{Q},\mu_{2}) . By [ 7 , Prop. 7 and Prop. 9] , this class lies in the image of the restriction map H^{n}(K,\mu_{4}^{\otimes n-1})\to H^{n}(K_{Q},\mu_{4}^{\otimes n-1}) . Let \xi\in H^{n}(K,\mu_{4}^{\otimes n-1}) be a preimage of e_{n}(\varphi) . Consider an arbitrary ordering P\in\mathsf{X}(K) . Since A_{K_{P}} is split, the conic \mathcal{C}_{Q} has a rational point over K_{P} , and hence P extends to an ordering P^{\prime}\in\mathsf{X}(K_{Q}) . Set R=K_{P} and R^{\prime}=(K_{Q})_{P^{\prime}} . Since \sigma_{R} is hyperbolic, so is \sigma_{R^{\prime}} . Hence \varphi_{R^{\prime}} is hyperbolic, whereby \xi_{R^{\prime}}=0 . Since the restriction map \mathsf{H}^{n}(R,\mu_{2})\to\mathsf{H}^{n}(R^{\prime},\mu_{2}) is an isomorphism (between groups of order 2 ), we conclude that \xi_{R}=0 . This shows that \xi_{K_{P}}=0 for every P\in\mathsf{X}(K) . Consider c\in{K}^{\times} . By 6.2 , we have c\in\mathsf{N}_{K^{\prime}/K}^{\ast} for some finite 2 -extension K^{\prime}/K with \xi_{K^{\prime}}=0 . Since \mathsf{I}^{n+1}K=0 , we have \mathsf{I}^{n+1}K^{\prime}=0 , by 2.4 . Since \xi_{K^{\prime}}=0 , we have e_{n}(\varphi_{K^{\prime}_{Q}})=0 , hence by 2.2 , we obtain that \varphi_{K^{\prime}_{Q}}\in\mathsf{I}^{n+1}K^{\prime}_{Q} . It follows by 6.1 that \sigma_{K^{\prime}_{Q}} is hyperbolic. By Dejaiffe’s Theorem [ 10 ] , this implies that the involution \sigma_{K^{\prime}} is hyperbolic. Therefore c\in\mathsf{N}_{K^{\prime}/K}^{\ast}\subseteq\mathsf{Hyp}_{2}^{\ast}(A,\sigma) . Assume now that \sigma is symplectic. Let \pi denote the norm form of Q . Then (A,\sigma)\simeq\operatorname{\mathsf{Ad}}(\varphi)\otimes(Q,\operatorname{\mathsf{can}}_{Q}) for a quadratic form \varphi over K , and the properties of the involution \sigma correspond to those of the quadratic form \pi\otimes\varphi . In particular, \mathsf{Hyp}_{2}^{\ast}(A,\sigma)=\mathsf{Hyp}_{2}^{\ast}(\varphi\otimes\pi) . Since (A,\sigma)\in\mathsf{I}^{n} , we have \pi\otimes\varphi\in\mathsf{I}^{n}K , and since (A,\sigma) is weakly hyperbolic, we get that \pi\otimes\varphi\in\mathsf{I}^{n}_{\mathsf{tors}}K . Since \mathsf{I}^{n+1}K(\sqrt{-1})=0 , we obtain by 5.1 that \mathsf{Hyp}_{2}^{\ast}(A,\sigma)=\mathsf{Hyp}_{2}^{\ast}(\varphi\otimes\pi)={K}^{\times} . ∎ Corollary 6.4 . Assume that \mathsf{I}^{n+1}K(\sqrt{-1})=0 . Let (A,\sigma) be a K -algebra with involution with (A,\sigma)\in\mathsf{I}^{n} . Assume that A is totally positive and \operatorname{\mathsf{ind}}A\leqslant 2 . Then \mathsf{G}(A,\sigma)=\mathsf{Hyp}_{2}^{\ast}(A,\sigma)\,. In particular {\bf PSim}^{+}(A,\sigma)(K)/R=\{1\} . Proof. Consider a\in\mathsf{G}(A,\sigma) . Set K^{\prime}=K(\sqrt{-a}) . Then A_{K^{\prime}} is totally positive and (A,\sigma)_{K^{\prime}} is weakly hyperbolic. Hence \mathsf{Hyp}_{2}^{\ast}(A_{K^{\prime}},\sigma_{K^{\prime}})={K^{\prime}}^{\times} , by 6.3 , whereby a\in\mathsf{N}_{K^{\prime}/K}^{\ast}\subseteq\mathsf{Hyp}_{2}^{\ast}(A,\sigma) . In view 3.2 , this shows the claim. ∎ Remark 6.5 . For any n\geqslant 3 , the hypotheses of 6.3 imply that A is split or \mathsf{deg}A is a multiple of 4 . Indeed, if A is nonsplit and \mathsf{deg}A\equiv 2\,\mathsf{mod}\,4 , then (A,\sigma)_{K(Q)} cannot have trivial Clifford invariant, by 4.3 . 6.4 raises the question whether any of the two assumptions that A is totally positive and \operatorname{\mathsf{ind}}A\leqslant 2 can be relaxed or dropped. For n\geqslant 3 , this presupposes a choice for the definition of belonging to \mathsf{I}^{n} . Question 6.6 . Assume that \mathsf{I}^{n+1}K(\sqrt{-1})=0 . Let (A,\sigma) be a K -algebra with involution with (A,\sigma)\in\mathsf{I}^{n} . Is {\bf PSim}^{+}(A,\sigma)(K)/R=\{1\} ? 7. Orthogonal involutions of trivial discriminant when \mathsf{I}^{4}=0 In [ 29 , Example 6.1] , an example is given for an algebra with orthogonal involution (A,\sigma) over k=\mathbb{Q}_{p}(X) for which {\bf PSim}^{+}(A,\sigma)(k)/R is nontrivial. Hence the underlying variety of the group is not rational. In that example, \mathsf{disc}(\sigma) is non-trivial. In [ 28 , Theorem 7.2] , on the other hand, assuming that \mathsf{cd}_{2}(K)\leqslant 3 , it is shown that {\bf PSim}^{+}(A,\sigma)(K)/R is trivial for any algebra K -algebra with orthogonal involution (A,\sigma) of trivial discriminant and trivial Clifford invariant. A similar result is achieved for symplectic involutions. Here we revisit these results with milder hypotheses. In particluar, we study {\bf PSim}^{+}(A,\sigma)(K)/R when (A,\sigma) is orthogonal of even degree and trivial discriminant, but with milder assumptions on the Clifford algebra. By 3.3 and 3.1 , triviality of {\bf PSim}^{+}(A,\sigma)(K)/R amounts to the equality \mathsf{G}(A,\sigma)={K}^{\times 2}\mathsf{Hyp}(A,\sigma)\,. The results of this section establish even some refinements of this equality, under different hypotheses, assuming that \mathsf{I}^{4}K=0 . We first consider the split case. Proposition 7.1 . Assume that \mathsf{I}^{4}K=0 . Let \varphi be a quadratic form over K with \varphi\in\mathsf{I}^{2}K and \operatorname{\mathsf{ind}}(\mathcal{C}(\varphi))\leqslant 4 . Then \mathsf{G}(\varphi)={K}^{\times 2}\mathsf{Hyp}_{2}^{\ast}(\varphi) . Proof. By 3.2 , we have {K}^{\times 2}\mathsf{Hyp}_{2}^{\ast}(\varphi)\subseteq\mathsf{G}(\varphi) , so only the other inclusion needs to be shown. Since \operatorname{\mathsf{ind}}(\mathcal{C}(\varphi))\leqslant 4 and [\mathcal{C}(\varphi)]\in\operatorname{\mathsf{Br}}_{2}(K) , Albert’s Theorem [ 15 , Theorem 16.1] together with [ 15 , Prop. 16.3] yield that \mathcal{C}(\alpha)\sim\mathcal{C}(\varphi) for a 6 -dimensional quadratic form \alpha\in\mathsf{I}^{2}K . By 2.1 , then \varphi\perp-\alpha\in\mathsf{I}^{3}K , and since \mathsf{I}^{4}K=0 , we conclude that \mathsf{G}(\varphi\perp-\alpha)={K}^{\times} . Therefore \mathsf{G}(\varphi)=\mathsf{G}(\alpha) . Consider c\in\mathsf{G}(\varphi)=\mathsf{G}(\alpha) . By [ 13 , Cor. 2.3] , we can write \alpha\simeq\beta_{1}\perp\beta_{2}\perp\beta_{3} with binary forms \beta_{1},\beta_{2},\beta_{3} over K such that c\in\bigcap_{i=1}^{3}\mathsf{G}(\beta_{i}) . Let d_{i}\in\mathsf{disc}(\beta_{i}) for 1\leqslant i\leqslant 3 . Then d_{1}d_{2}d_{3}\in{K}^{\times 2} . We set L=K(\sqrt{d_{1}},\sqrt{d_{2}}) . We obtain that [L:K]\leqslant 4 and c\in\mathsf{N}_{K(\sqrt{d_{1}})/K}^{\ast}\cap\mathsf{N}_{K(\sqrt{d_{2}})/K}^{\ast}={K}^{\times 2}\mathsf{N}_{L/K}^{\ast} , by [ 1 , Lemma 3.3] . Furthermore \alpha_{L} is hyperbolic, whereby \varphi_{L}\in\mathsf{I}^{3}L . Since \mathsf{I}^{4}K=0 , we have \mathsf{I}^{4}L=0 , by 2.4 . It follows by 5.1 that \mathsf{Hyp}_{2}^{\ast}(\varphi_{L})={L}^{\times} . Therefore c\in{K}^{\times 2}\mathsf{N}_{L/K}(\mathsf{Hyp}_{2}^{\ast}(\varphi_{L}))\subseteq{K}^{\times 2}\mathsf{Hyp}_{2}^{\ast}(\varphi) . ∎ The following statement extends [ 28 , Theorem 7.2] by relaxing the condition on the Clifford invariant from trivial to index at most 2 . Hence it provides a partial positive answer to a question asked in [ 29 , p. 1028] . Theorem 7.2 . Assume that \mathsf{I}^{4}K=0 . Let (A,\sigma) be a K -algebra with orthogonal involution with (A,\sigma)\in\mathsf{I}^{2} . Let C_{1} and C_{2} be the components of \mathcal{C}(A,\sigma) . Assume that \operatorname{\mathsf{ind}}(C_{i})\leqslant 2 for i\in\{1,2\} . Then \mathsf{G}(A,\sigma)=\operatorname{\mathsf{Nrd}}_{C_{1}}^{\ast}\cdot\operatorname{\mathsf{Nrd}}_{C_{2}}^{\ast}=\mathsf{Hyp}_{2}(A,\sigma)\,. Proof. The inclusions \mathsf{Hyp}_{2}(A,\sigma)\subseteq\mathsf{G}(A,\sigma)\subseteq\operatorname{\mathsf{Nrd}}_{C_{1}}^{\ast}\cdot\operatorname{\mathsf{Nrd}}_{C_{2}}^{\ast} hold by 3.2 and 4.2 , respectively. It remains to show the inclusion \operatorname{\mathsf{Nrd}}_{C_{i}}^{\ast}\subseteq\mathsf{Hyp}_{2}(A,\sigma) for i\in\{1,2\} . Consider i\in\{1,2\} and c\in\operatorname{\mathsf{Nrd}}_{C_{i}}^{\ast} . Since \operatorname{\mathsf{ind}}C_{i}\leqslant 2 , there exists a field extension L/K with [L:K]\leqslant 2 such that c\in\mathsf{N}_{L/K}^{\ast} and (C_{i})_{L} is split. Then (A_{L},\sigma_{L}) has trivial Clifford invariant. Since \mathsf{I}^{4}K=0 and [L:K]\leqslant 2 , we have \mathsf{I}^{4}L=0 , by 2.4 . Therefore \mathsf{Hyp}_{2}(A_{L},\sigma_{L})={L}^{\times} , by 6.3 , and we conclude that c\in\mathsf{N}_{L/K}(\mathsf{Hyp}_{2}(A_{L},\sigma_{L}))\subseteq\mathsf{Hyp}_{2}(A,\sigma) . ∎ Theorem 7.3 . Assume that \mathsf{I}^{4}K^{\prime}=0 for every field extension K^{\prime}/K of degree at most 4 . Let (A,\sigma) be a K -algebra with orthogonal involution trivial discriminant and such that \mathsf{deg}A\equiv 2\,\mathsf{mod}\,4 . Let C be a component of \mathcal{C}(A,\sigma) . Assume that \operatorname{\mathsf{ind}}C_{K_{Q}}\leqslant 4 for the unique K -quaternion algebra Q\sim A . Then \mathsf{Hyp}^{\ast}(A,\sigma)=\operatorname{\mathsf{Nrd}}_{C}^{\ast}\quad\mbox{ and }\quad\mathsf{G}(A,\sigma)={K}^{\times 2}\mathsf{Hyp}^{\ast}(A,\sigma)={K}^{\times 2}\operatorname{\mathsf{Nrd}}_{C}^{\ast}\,. Proof. We have \mathsf{Hyp}(A,\sigma)\subseteq\mathsf{G}(A,\sigma)\subseteq{K}^{\times 2}\operatorname{\mathsf{Nrd}}_{C}^{\ast} , by 3.2 and 4.5 , respectively. Hence, to complete the proof, it suffices to show that \operatorname{\mathsf{Nrd}}_{C}^{*}\subseteq\mathsf{Hyp}^{*}(A,\sigma) . Consider a\in\operatorname{\mathsf{Nrd}}_{C}^{\ast} . Let D be the central K -division algebra with D\sim C . Then a\in\operatorname{\mathsf{Nrd}}_{D}^{\ast} , by 4.1 , so that a\in\mathsf{N}_{L/K}^{\ast} for some maximal subfield L of D . Then [L:K]=\mathsf{deg}D\leqslant\operatorname{\mathsf{ind}}C\leqslant 4 and C_{L} is split. Hence \mathsf{I}^{4}L=0 , by our hypothesis. Since \mathsf{deg}A\equiv 2\,\mathsf{mod}\,4 , we have C^{\otimes 2}\sim A , whereby A_{L} is split and (A_{L},\sigma_{L})\simeq\operatorname{\mathsf{Ad}}(\varphi) for a quadratic form \varphi over L . Since (A_{L},\sigma_{L}) has trivial Clifford invariant, it follows by 2.1 that \varphi\in\mathsf{I}^{3}L . Using 5.1 , we obtain that {L}^{\times}=\mathsf{Hyp}_{2}^{\ast}(\varphi)=\mathsf{Hyp}_{2}^{\ast}(A_{L},\sigma_{L}) . Therefore a\in\mathsf{N}_{L/K}(\mathsf{Hyp}^{\ast}(A,\sigma))\subseteq\mathsf{Hyp}^{\ast}(A,\sigma) . ∎ We recall the classical definition of the u -invariant of K : u(K)=\mathsf{sup}\{\mathsf{dim}(\varphi)\mid\varphi\mbox{ anisotropic quadratic form over }K\}\in\mathbb{N}\cup\{\infty\}\,. We refer to [ 18 , Chap. XI, §6] for an introduction to the study and the interest of this field invariant. If n\in\mathbb{N} is such that u(K)<2^{n} , then every n -fold Pfister form over K is hyperbolic, whereby \mathsf{I}^{n}K=0 . Lemma 7.4 . Assume that u(K)\leqslant 8 . Let \varphi be a quadratic form over K and a\in{K}^{\times} . Then there exist 2 -extensions L_{1}/K and L_{2}/K of degree at most 4 such that a\in\mathsf{N}_{L_{1}/K}^{\ast}\cdot\mathsf{N}_{L_{2}/K}^{\ast} and \mathsf{dim}((\varphi_{L_{i}})_{\mathsf{an}})\leqslant 4 for i\in\{1,2\} . Proof. Without loss of generality, we may assume that \varphi is anisotropic. If \mathsf{dim}(\varphi)\leqslant 4 then we may take L_{1}=L_{2}=K . Assume that \mathsf{dim}(\varphi)>4 . Then \mathsf{dim}(\langle\!\langle a\rangle\!\rangle\otimes\varphi)>8\geqslant u(K) , whereby \langle\!\langle a\rangle\!\rangle\otimes\varphi is isotropic. By [ 13 , Prop. 2.2] , it follows that \varphi\simeq c\langle 1,-d\rangle\perp\psi for a quadratic form \psi over K and c,d\in{K}^{\times} such that \langle\!\langle a,d\rangle\!\rangle is hyperbolic. As \varphi is anisotropic, we have \mathsf{dim}(\varphi)\leqslant u(K)\leqslant 8 . Hence \mathsf{dim}(\psi)\leqslant 6 . Set L=K(\sqrt{d}) . Then a\in\mathsf{N}_{L/K}^{\ast} and (\varphi_{L})_{\mathsf{an}}\simeq(\psi_{L})_{\mathsf{an}} . If \mathsf{dim}((\psi_{L})_{\mathsf{an}})\leqslant 4 , we may take L_{1}=L_{2}=L . Assume now that \mathsf{dim}((\psi_{L})_{\mathsf{an}})>4 . Then \psi is anisotropic and of dimension 5 or 6 . It follows that \mathsf{dim}(\langle\!\langle a\rangle\!\rangle\otimes\psi)>8\geqslant u(K) , whereby \langle\!\langle a\rangle\!\rangle\otimes\psi is isotropic. By [ 13 , Prop. 2.2] , we have \psi\simeq c_{1}\langle 1,-d_{1}\rangle\perp\tau_{1} for a quadratic form \tau_{1} over K and c_{1},d_{1}\in{K}^{\times} such that \langle\!\langle a,d_{1}\rangle\!\rangle is hyperbolic. Since a\in\mathsf{N}_{L/K}^{\ast} , we fix b\in{L}^{\times} with a=\mathsf{N}_{L/K}(b) . Since \mathsf{N}_{L/K}(b)=a\in\mathsf{D}(\langle\!\langle d_{1}\rangle\!\rangle_{K}) , it follows by [ 18 , Chap. VII, Theorem 5.10] that b\in{K}^{\times}\cdot\mathsf{D}(\langle\!\langle d_{1}\rangle\!\rangle_{L}) . We fix b_{1}\in\mathsf{D}(\langle\!\langle d_{1}\rangle\!\rangle_{L}) and b_{2}\in{K}^{\times} such that b=b_{1}\cdot b_{2} . We set L_{1}=L(\sqrt{d_{1}}) . Then b_{1}\in\mathsf{N}_{L_{1}/L}^{\ast} . Since \mathsf{dim}(\langle\!\langle b_{2}\rangle\!\rangle\otimes\psi)>8=u(K) , the quadratic form \langle\!\langle b_{2}\rangle\!\rangle\otimes\psi over K is isotropic. Again by [ 13 , Prop. 2.2] , there exists a quadratic form \tau_{2} over K and c_{2},d_{2}\in{K}^{\times} such that \psi\simeq c_{2}\langle 1,-d_{2}\rangle\perp\tau_{2} and \langle\!\langle b_{2},d_{2}\rangle\!\rangle is hyperbolic, whereby b_{2}\in\mathsf{D}(\langle\!\langle d_{2}\rangle\!\rangle_{L}) . We set L_{2}=L(\sqrt{d_{2}}) . Then b_{2}\in\mathsf{N}_{L_{2}/L}^{\ast} , and we obtain that a=\mathsf{N}_{L/K}(b)=\mathsf{N}_{L/K}(b_{1})\cdot\mathsf{N}_{L/K}(b_{2})\in\mathsf{N}_{L_{1}/K}^{\ast}\cdot\mathsf{N}_{L_{2}/K}^{\ast}\,. For i\in\{1,2\} , since \varphi\simeq c\langle 1,-d\rangle\perp c_{i}\langle 1,-d_{i}\rangle\perp\tau_{i} and \varphi is anisotropic, we have that \mathsf{dim}(\tau_{i})=\mathsf{dim}(\varphi)-4\leqslant u(K)-4\leqslant 4 , and as d,d_{i}\in{L}^{\times 2}_{i} , we obtain that \mathsf{dim}((\varphi_{L_{i}})_{\mathsf{an}})\leqslant\mathsf{dim}(\tau_{i})\leqslant 4 . ∎ The following statement generalises [ 28 , Theorem 6.1, Theorem 7.2] and [ 29 , Theorem 5.1] , where it is shown for function fields in one variable over \mathbb{Q}_{p} for an odd prime p . By [ 20 , Theorem 3.4] , u(K)=8 holds when K is a function field in one variable over \mathbb{Q}_{p} for any prime p . Hence this case is covered, but our result applies at the same time to arbitrary extensions of transcendence degree at most 3 of an algebraically closed field. Note that the hypothesis that u(K)\leqslant 8 implies that the index of elements in \operatorname{\mathsf{Br}}_{2}(K) is bounded by 8 , and no stronger bound on the index is required for the proof. Theorem 7.5 . Assume that u(K)\leqslant 8 . Let (A,\sigma) be a K -algebra with involution with (A,\sigma)\in\mathsf{I}^{3} . Then \mathsf{Hyp}_{2}(A,\sigma)=\mathsf{G}(A,\sigma)={K}^{\times}\,. Proof. Since A carries a K -linear involution, we have that [A]\in\operatorname{\mathsf{Br}}_{2}(K) . Hence 2.1 yields that A\sim\mathcal{C}(\rho) for some anisotropic quadratic form \rho\in\mathsf{I}^{2}K . By 7.4 , {K}^{\times} is generated by the subgroups \mathsf{N}_{K^{\prime}/K}^{\ast} where K^{\prime}/K is a finite 2 -extension with \mathsf{dim}((\rho_{K^{\prime}})_{\mathsf{an}})\leqslant 4 , and consequently \operatorname{\mathsf{ind}}(A_{K^{\prime}})\leqslant 2 . Let K^{\prime}/K be such a finite 2 -extension. By 2.4 , we have that \mathsf{I}^{4}K^{\prime}=0 . In view of the hypotheses, we obtain by 6.3 that \mathsf{Hyp}_{2}^{\ast}(A_{K^{\prime}},\sigma_{K^{\prime}})={K^{\prime}}^{\times} . Therefore \mathsf{N}_{K^{\prime}/K}^{\ast}\subseteq\mathsf{Hyp}_{2}^{\ast}(A,\sigma) . We conclude that {K}^{\times}=\mathsf{Hyp}_{2}(A,\sigma) . ∎ One may ask whether the equality \mathsf{Hyp}(A,\sigma)=\mathsf{G}(A,\sigma) still holds (whereby {\bf PSim}^{+}(A,\sigma)(K)/R would be trivial) for K -algebras with orthogonal involution (A,\sigma)\in\mathsf{I}^{2} (as opposed to \mathsf{I}^{3} ) when u(K)\leqslant 8 . Assuming that the u -invariant of all finite extensions of K is bounded by 8 (which is the case when K is the function field of a p -adic curve), we can reduce this problem to the relation between \mathsf{G}(A,\sigma) and the reduced norms of the components of \mathcal{C}(A,\sigma) . Corollary 7.6 . Assume that u(K^{\prime})\leqslant 8 for every finite field extension K^{\prime}/K . Let (A,\sigma) be a K -algebra with orthogonal involution with (A,\sigma)\in\mathsf{I}^{2} . Let C_{1} and C_{2} be the two components of \mathcal{C}(A,\sigma) . Then \mathsf{Hyp}(A,\sigma)=\operatorname{\mathsf{Nrd}}_{C_{1}}^{\ast}\cdot\operatorname{\mathsf{Nrd}}_{C_{2}}^{\ast}\,. Proof. By 4.2 , we have \mathsf{Hyp}(A,\sigma)\subseteq\operatorname{\mathsf{Nrd}}_{C_{1}}^{\ast}\cdot\operatorname{\mathsf{Nrd}}_{C_{2}}^{\ast} . For the opposite inclusion, consider i\in\{1,2\} and a\in\operatorname{\mathsf{Nrd}}_{C_{i}}^{\ast} . Hence, there exists a finite field extension L/K such that (C_{i})_{L} is split and a\in\mathsf{N}_{L/K}^{*} . Then u(L)\leqslant 8 and (A_{L},\sigma_{L}) is an L -algebra with orthogonal involution of even degree with trivial discriminant and trivial Clifford invariant, so \mathsf{Hyp}^{\ast}(A_{L},\sigma_{L})={L}^{\times} , by 7.5 , and a\in\mathsf{N}_{L/K}(\mathsf{Hyp}(A_{L},\sigma_{L}))\subseteq\mathsf{Hyp}(A,\sigma) . This shows that \operatorname{\mathsf{Nrd}}_{C_{i}}^{\ast}\subseteq\mathsf{Hyp}^{\ast}(A,\sigma) for i\in\{1,2\} . Hence \operatorname{\mathsf{Nrd}}_{C_{1}}^{*}\cdot\operatorname{\mathsf{Nrd}}_{C_{2}}^{*}\subseteq\mathsf{Hyp}(A,\sigma) . ∎