Single conjugacy classes of isometries in orthogonal groups over local fields Thanks: The first author is partially supported by the National Key R & \And D Program of China No. 2023YFA1009702 and National Natural Science Foundation of China No. 12231009 Fei Xu, Bo Zhang Abstract . All isometries σ \sigma in a quadratic space over a non-archimedean local field of characteristic not 2 satisfying that any isometry τ \tau which is conjugate to σ \sigma in the general linear group is conjugate to σ \sigma in the orthogonal group are determined. This extends [ 3 , Theorem 2.1] to arbitary cases. 1. Introduction Motivated by studying knot cobordism in [ 2 ] , Milnor developed classification theory of conjugacy classes of orthogonal groups in [ 3 ] and proved that two isometries of a quadratic space (V,\langle\ ,\ \rangle) over a local field k of char(k)\neq 2 with the same irreducible minimal polynomial are conjugate in {\rm O}(V) in [ 3 , Theorem 2.1] , which was conjectured by Levine. It is natural to have the following more general question. Question 1.1 . Determine all isometries \sigma\in{\rm O}(V) satisfying that any isometry in {\rm O}(V) which is conjugate to \sigma in {\rm GL}(V) is conjugate to \sigma in {\rm O}(V) . When \sigma is semi-simple and k is a non-archimedean local field, an answer to Question 1.1 is given in [ 1 ] by using the moduli spaces of pairs of a quadratic space with a semi-simple isometry. In this paper, we give a complete answer to Question 1.1 over a non-archimedean local field by applying Milnor’s theory for classification conjugacy classes of orthogonal groups in [ 3 ] . Let f(x) be the characteristic polynomial of \sigma\in{\rm O}(V) . Then (1) f(x)=(x+1)^{m_{+}}(x-1)^{m_{-}}\prod_{i=1}^{m_{1}}p_{i}(x)^{e_{i}}\prod_{j=1}^{m_{2}}(\tilde{p}_{j}(x)\tilde{p}_{j}^{*}(x))^{\tilde{e}_{j}} by [ 2 , 7. Lemma (a)] , where p_{i}(x) is a monic irreducible polynomial of degree bigger than 1 over k such that p_{i}(x)=x^{\deg(p_{i})}p_{i}(x^{-1})\ \ \ \text{ for $1\leq i\leq m_{1}$} and \tilde{p}_{j}(x) is a monic irreducible polynomial over k and \tilde{p}_{j}^{*}(x)=\tilde{p}_{j}(0)^{-1}x^{\deg(\tilde{p}_{j})}\tilde{p}_{j}(x^{-1})\ \ \ \text{with}\ \ \ \tilde{p}_{j}(x)\neq\tilde{p}_{j}^{*}(x) for 1\leq j\leq m_{2} . For a monic irreducible factor q(x) of f(x) , one defines V_{q(x),\sigma}=\{v\in V:\ q(\sigma)^{l}v=0\ \text{for sufficiently large integer $l$}\}. When q(x)=x\pm 1 or p_{i}(x) for 1\leq i\leq m_{1} , then there is an orthogonal decomposition V_{q(x),\sigma}=V_{q(x),\sigma}^{(1)}\perp\cdots\perp V_{q(x),\sigma}^{(l)}\perp\cdots where V_{q(x),\sigma}^{(l)} is a free k[x]/(q(x)^{l}) -module by [ 3 , Theorem 3.2] . For q(x)=x\pm 1 , the well-defined bilinear map (2) b_{\sigma}^{(l)}:(V_{q(x),\sigma}^{(l)}/q(x)V_{q(x),\sigma}^{(l)})\times(V_{q(x),\sigma}^{(l)}/q(x)V_{q(x),\sigma}^{(l)})\rightarrow k b_{\sigma}^{(l)}(\bar{u},\bar{v})=\langle(\sigma-\sigma^{-1})^{l-1}u,v\rangle is non-degenerate, symmetric when l is odd by [ 3 , Theorem 3.4] . The main result of this paper gives a complete answer to Question 1.1 over a non-archimedean local field, which extends [ 3 , Theorem 2.1] to arbitrary cases. Theorem 1.2 . Let V be a non-degenerated quadratic space over a non-archimedean local field k with char(k)\neq 2 . Suppose \sigma\in{\rm O}(V) and f(x) is the characteristic polynomial of \sigma with the factorization of ( 1 ). Write m_{0}=\sharp\{2l+1:V_{x+1,\sigma}^{(2l+1)}\neq 0\}+\sharp\{2l+1:V_{x-1,\sigma}^{(2l+1)}\neq 0\}. Then any isometry in {\rm O}(V) which is conjugate to \sigma in {\rm GL}(V) is conjugate to \sigma in {\rm O}(V) if and only if one of the following conditions holds (i) m_{0}\leq 1 and m_{1}=0 . (ii) m_{0}=0 and m_{1}=1 and V_{p_{1}(x),\sigma}=V_{p_{1}(x),\sigma}^{(2l_{1}+1)} for some integer l_{1}\geq 0 . (iii) m_{0}=m_{1}=1 and V_{p_{1}(x),\sigma}=V_{p_{1}(x),\sigma}^{(2l_{1}+1)} for some integer l_{1}\geq 0 and the unique non-zero quadratic space (V_{x\pm 1,\sigma}^{(2l_{0}+1)}/(x\pm 1)V_{x\pm 1,\sigma}^{(2l_{0}+1)},b_{\sigma}^{(2l_{0}+1)}) defined by ( 2 ) is either one dimensional or a hyperbolic plane. Terminology and notation are standard if not explained and are adopted from [ 4 ] and [ 3 ] . Let k be a field k of char(k)\neq 2 . A non-degenerate quadratic space (V,\langle\ ,\ \rangle) is a finite dimensional vector space V over k with a symmetric bilinear map \langle\ ,\ \rangle:\ V\times V\longrightarrow k\ \ \ \text{with}\ \ \ \det(\langle e_{i},e_{j}\rangle)\neq 0 for a basis \{e_{1},\cdots,e_{n}\} of V . Write {\rm O}(V)=\{\sigma\in{\rm GL}(V):\ \langle\sigma u,\sigma v\rangle=\langle u,v\rangle,\ \forall u,v\in V\}. A non-degenerate two dimensional quadratic space (V,\langle\ ,\ \rangle) is called a hyperbolic plane if there is v\in V with v\neq 0 such that \langle v,v\rangle=0 . If f(x) is the characteristic polynomial of \sigma\in{\rm O}(V) and q(x) is a monic irreducible factor of f(x) , we write V_{q(x),\sigma}=\{v\in V:\ q(\sigma)^{l}v=0\ \text{for sufficiently large integer $l$}\}. A separable k -algebra is defined to be a finite dimensional semi-simple k -algebra such that the center of each simple component is a separable field extension of k . For a separable k -algebra A , the reduced trace of A to k is denoted by {\rm Tr}_{A/k} (see [ 5 , §9] ). For a polynomial f(x) of degree n over k with f(0)\neq 0 , we define f^{*}(x)=f(0)^{-1}x^{n}f(x^{-1}). 2. Hermitian structures for non-reciprocal cases It is well-known that the classification of conjugacy classes of {\rm GL}(V) is equivalent to classify all k[x] -module structure of V via \sigma\in{\rm GL}(V) for a finite dimensional vector space V over k . When V is a quadratic space over k , the classification of conjugacy classes of {\rm O}(V) becomes classification of k[x] -module structure of V via \sigma\in{\rm O}(V) compatible with the quadratic structure. This leads to Milnor’s classification in [ 3 , §3] . A brief explanation for Case 3 in [ 3 , §3] is given in the last paragraph in [ 3 , p.94] . We will provide the similar hermitian structure for Case 3 in [ 3 , §3] which makes this theory in a uniform way. Definition 2.1 . Let p(x) be an irreducible monic polynomial over k with p(0)\neq 0 such that p(x)\neq p^{*}(x) and E=k[x]/(p(x)p^{*}(x)) . A free E -module M of finite rank is called a hermitian space over E if there is a map h:M\times M\rightarrow E;\ (u,v)\mapsto h(u,v) such that h(u,v) is linear in u and h(v,u)=\iota(h(u,v)) where \iota is an involution of E over k induced by x\mapsto x^{-1} . A hermitian space (M,h) over E is called non-degenerate if h(u,v)=0,\ \forall v\in M\ \ \ \Leftrightarrow\ \ \ u=0. The unitary group U(M,h) of (M,h) is defined by {\rm U}(M,h)=\{\sigma\in\Aut_{E}(M):\ h(\sigma u,\sigma v)=h(u,v),\ \forall u,v\in M\} The following result gives the equivalent condition that a hermitian space is non-degenerate. Proposition 2.2 . Let p(x) be an irreducible monic polynomial over k with p(0)\neq 0 such that p(x)\neq p^{*}(x) and E=k[x]/(p(x)p^{*}(x)) . Suppose (M,h) is a hermitian space over E . Then (M,h) is non-degenerate if and only if \det(h(e_{i},e_{j}))_{1\leq i,j\leq n} is invertible in E for a basis \{e_{1},\cdots,e_{n}\} of M over E . Proof. ( \Rightarrow ) Suppose that \det(h(e_{i},e_{j}))_{1\leq i,j\leq n} is a zero divisor in E for a basis \{e_{1},\cdots,e_{n}\} of M over E . Since E\cong k[x]/(p(x))\times k[x]/(p^{*}(x)), one can write h(e_{i},e_{j})=(a_{ij},b_{ij})\ \ \text{ with }\ \ a_{ij}\in k[x]/(p(x)),\ b_{ij}\in k[x]/(p^{*}(x)) for 1\leq i,j\leq n . Since \det(h(e_{i},e_{j}))=(\det(a_{ij}),\det(b_{ij})) is a zero divisor in E , one has \det(a_{ij})=0 or \det(b_{ij})=0 . Without loss of generality, we assume \det(a_{ij})=0 . Then there is (\xi_{1},\cdots,\xi_{n})\neq(0,\cdots,0)\ \ \ \text{such that}\ \ \ (\xi_{1},\cdots,\xi_{n})\cdot(a_{ij})=(0,\cdots,0) where \xi_{1},\cdots,\xi_{n} are all in k[x]/(p(x)) . Therefore u_{0}=(\xi_{1},0)e_{1}+\cdots+(\xi_{n},0)e_{n}\neq 0. For any v=\alpha_{1}e_{1}+\cdots+\alpha_{n}e_{n}\in M , one has \displaystyle h(u_{0},v)=((\xi_{1},0),\cdots,(\xi_{n},0))\cdot(h(e_{i},e_{j}))\cdot(\iota(\alpha_{1}),\cdots,\iota(\alpha_{n}))^{T} \displaystyle=((\xi_{1},\cdots,\xi_{n})(a_{ij}),(0,\cdots,0)(b_{ij}))\cdot(\iota(\alpha_{1}),\cdots,\iota(\alpha_{n}))^{T}=0. A contradiction is derived. ( \Leftarrow ) Let u=\beta_{1}e_{1}+\cdots+\beta_{n}e_{n} such that h(u,v)=0 for all v\in M . Then (\beta_{1},\cdots,\beta_{n})(h(e_{i},e_{j}))^{T}=(0,\cdots,0). Since \det(h(e_{i},e_{j}))_{1\leq i,j\leq n} is invertible in E , one obtains (\beta_{1},\cdots,\beta_{n})=(0,\cdots,0). Therefore u=0 as required. ∎ This kind of non-degenerate hermitian space of the same rank is unique up to isomorphism. Proposition 2.3 . Let p(x) be an irreducible monic polynomial over k with p(0)\neq 0 such that p(x)\neq p^{*}(x) and E=k[x]/(p(x)p^{*}(x)) . Assume that char(k)\neq 2 . If (M,h) is a non-degenerate hermitian space over E with respect to the involution induced by x\mapsto x^{-1} , then M has an orthogonal basis \{v_{1},\cdots,v_{n}\} over E such that h(v_{1},v_{1})=\cdots=h(v_{n},v_{n})=1. In particular, any two non-degenerate hermitian spaces of the same rank over E with respect to the involution induced by x\mapsto x^{-1} are isomorphic. Proof. Let \{e_{1},\cdots,e_{n}\} be a basis of M over E . If h(e_{1},e_{1}) is invertible in E , then M=Ee_{1}\perp N where N=E(e_{2}-\frac{h(e_{1},e_{2})}{h(e_{1},e_{1})}e_{1})+\cdots+E(e_{n}-\frac{h(e_{1},e_{n})}{h(e_{1},e_{1})}e_{1}). Since \det(M)=\det(Ee_{1})\cdot\det(N) , one obtains that N is also a non-degenerate hermitian space over E by Proposition 2.2 . The result follows from induction. Otherwise, one has that h(e_{1},e_{1}) is a zero divisor of E . Then there is 2\leq i_{0}\leq n such that h(e_{1},e_{i_{0}})\neq 0 by the assumption that \det(h(e_{i},e_{j})) is invertible in E . Without loss of generality, we simply assume that h(e_{1},e_{2})\neq 0 . Claim: There is \lambda\in E such that h(\lambda e_{1}+e_{2},\lambda e_{1}+e_{2}) is invertible in E . Indeed, since E\cong k[x]/(p(x))\times k[x]/(p^{*}(x))\cong K\times K as k -algebras where K=k(\alpha) and \alpha is a root of p(x) , the involution of E induced by x\mapsto x^{-1} gives the involution of K\times K by interchanging two coordinates. Under this identification, one can write h(e_{1},e_{1})=(a_{1},a_{2}),\ \ \ h(e_{1},e_{2})=(c_{1},c_{2})\ \ \ \text{and}\ \ \ h(e_{2},e_{2})=(b_{1},b_{2}) where a_{i},b_{i},c_{i}\in K for i=1,2 . Write \lambda=(\lambda_{1},\lambda_{2}) . Then h(\lambda e_{1}+e_{2},\lambda e_{1}+e_{2})=(a_{1}\lambda_{1}\lambda_{2}+c_{1}\lambda_{1}+c_{2}\lambda_{2}+b_{1},a_{2}\lambda_{1}\lambda_{2}+c_{1}\lambda_{1}+c_{2}\lambda_{2}+b_{2}). Since h(e_{1},e_{1}) is a zero divisor of E , one can simply assume that a_{1}=0 . By h(e_{1},e_{2})=(c_{1},c_{2})\neq(0,0) , the linear form c_{1}\lambda_{1}+c_{2}\lambda_{2} is not zero. When a_{2}\neq 0 , one considers the equation c_{1}\lambda_{1}+c_{2}\lambda_{2}+b_{1}=1. Then the equation a_{2}\lambda_{1}\lambda_{2}+c_{1}\lambda_{1}+c_{2}\lambda_{2}+b_{2}=0 has at most two solutions in K . Since char(k)\neq 2 , the field K contains more than two elements. Therefore there are \lambda_{1},\lambda_{2} in K such that a_{2}\lambda_{1}\lambda_{2}+c_{1}\lambda_{1}+c_{2}\lambda_{2}+b_{2}\neq 0. This implies that h(\lambda e_{1}+e_{2},\lambda e_{1}+e_{2}) is invertible in E . When a_{2}=0 , there are \lambda_{1},\lambda_{2}\in K such that c_{1}\lambda_{1}+c_{2}\lambda_{2}\in K\setminus\{-b_{1},-b_{2}\} by char(k)\neq 2 . Therefore the claim follows. By repeating the beginning arguments to the basis \{\lambda e_{1}+e_{2},e_{1},e_{3},\cdots,e_{n}\} of M , one obtains an orthogonal basis \{\eta_{1},\cdots,\eta_{n}\} . Write h(\eta_{1},\eta_{1})=(\alpha_{1},\alpha_{1}),\cdots,h(e_{n},e_{n})=(\alpha_{n},\alpha_{n}) with \alpha_{1},\cdots,\alpha_{n}\in K . Then v_{1}=(\alpha_{1},1)^{-1}\eta_{1},\cdots,v_{n}=(\alpha_{n},1)^{-1}\eta_{n} is as required. ∎ Corollary 2.4 . If (M,h) is the non-degenerate hermitian space over E in the above Proposition 2.3 , then {\rm U}(M,h)\cong{\rm GL}_{n}(K) where n={\rm rank}_{E}(M) and K=k(\alpha) for a root \alpha of p(x) . Proof. Suppose that \{v_{1},\cdots,v_{n}\} is an orthogonal basis over E such that h(v_{1},v_{1})=\cdots=h(v_{n},v_{n})=1 . Let \sigma\in\Aut_{E}(M) and write \sigma(v_{i})=\sum_{j=1}^{n}(a_{ij},b_{ij})\cdot v_{j} where a_{ij},b_{ij}\in K for 1\leq i\leq n . Then \sigma\in{\rm U}(M,h)\ \ \ \Leftrightarrow\ \ \ ((a_{ij},b_{ij}))\cdot((b_{ji},a_{ji}))=I_{n}. This is equivalent to (a_{ij})=(b_{ji})^{-1}\in{\rm GL}_{n}(K) . The map \sigma\mapsto(a_{ij}) gives the desired isomorphism. ∎ One also has an analogue of [ 3 , Theorem 3.2] for Case 3 in [ 3 , §3] . Theorem 2.5 . Let (V,\langle\ ,\ \rangle) be a non-degenerate quadratic space over a field of k with char(k)\neq 2 and f(x) be the characteristic polynomial of \sigma\in{\rm O}(V) . If \tilde{p}(x) is a monic irreducible factor of f(x) with \tilde{p}(x)\neq\tilde{p}^{*}(x) , then there is a direct sum decomposition V_{\tilde{p}(x),\sigma}=V_{\tilde{p}(x),\sigma}^{(1)}\oplus\cdots\oplus V_{\tilde{p}(x),\sigma}^{(i)}\cdots where V_{\tilde{p}(x),\sigma}^{(i)} is a free k[x]/(\tilde{p}(x)^{i}) -module and V_{\tilde{p}^{*}(x),\sigma}=V_{\tilde{p}^{*}(x),\sigma}^{(1)}\oplus\cdots\oplus V_{\tilde{p}^{*}(x),\sigma}^{(i)}\cdots where V_{\tilde{p}^{*}(x),\sigma}^{i} is a free k[x]/(\tilde{p}^{*}(x)^{i}) -module such that V_{\tilde{p}(x),\sigma}\oplus V_{\tilde{p}^{*}(x),\sigma}=(V_{\tilde{p}(x),\sigma}^{(1)}\oplus V_{\tilde{p}^{*}(x),\sigma}^{(1)})\perp\cdots\perp(V_{\tilde{p}(x),\sigma}^{(i)}\oplus V_{\tilde{p}^{*}(x),\sigma}^{(i)})\perp\cdots where V_{\tilde{p}(x),\sigma}^{(i)}\oplus V_{\tilde{p}^{*}(x),\sigma}^{(i)} is a free k[x]/(\tilde{p}(x)^{i}\tilde{p}^{*}(x)^{i}) -module. Proof. Let {\rm Ann}_{k[x]}(V_{\tilde{p}(x),\sigma})=\{a(x)\in k[x]:a(x)V_{\tilde{p}(x),\sigma}=0\}=(\tilde{p}(x)^{s}) for some positive integer s and {\rm Ann}_{k[x]}(V_{\tilde{p}^{*}(x),\sigma})=\{a(x)\in k[x]:a(x)V_{\tilde{p}^{*}(x),\sigma}=0\}=(\tilde{p}^{*}(x)^{t}) for some positive integer t . Then s=t . Indeed, suppose s>t . Since there is e\in V_{\tilde{p}(x),\sigma} such that \tilde{p}(x)^{s-1}e\neq 0 and V_{\tilde{p}(x),\sigma}\oplus V_{\tilde{p}^{*}(x),\sigma} is a non-degenerate subspace of V , there is f\in V_{\tilde{p}^{*}(x),\sigma} such that \langle\tilde{p}(x)^{s-1}e,f\rangle=1 by [ 3 , Lemma 3.1] . Then 1=\langle\tilde{p}(x)^{s-1}e,f\rangle=\langle e,\tilde{p}^{*}(x)^{(s-1)}\tilde{p}^{*}(0)^{s-1}\sigma^{(1-s)\deg(\tilde{p}(x))}(f)\rangle=\langle e,0\rangle=0 and a contradiction is derived. By the structure theorem of finitely generated modules over PID, there are a free k[x]/(\tilde{p}(x)^{s}) -module V_{\tilde{p}(x),\sigma}^{(s)} and a free k[x]/(\tilde{p}^{*}(x)^{s}) -module V_{\tilde{p}^{*}(x),\sigma}^{(s)} such that V_{\tilde{p}(x),\sigma}=V_{\tilde{p}(x),\sigma}^{\prime}\oplus V_{\tilde{p}(x),\sigma}^{(s)}\ \ \ \text{and}\ \ \ V_{\tilde{p}^{*}(x),\sigma}=V_{\tilde{p}^{*}(x),\sigma}^{\prime}\oplus V_{\tilde{p}^{*}(x),\sigma}^{(s)} where V_{\tilde{p}(x),\sigma}^{\prime} and V_{\tilde{p}^{*}(x),\sigma}^{\prime} are submodules of V_{\tilde{p}(x),\sigma} and V_{\tilde{p}^{*}(x),\sigma} respectively satisfying \tilde{p}(x)^{s-1}V_{\tilde{p}(x),\sigma}^{\prime}=\tilde{p}^{*}(x)^{s-1}V_{\tilde{p}^{*}(x),\sigma}^{\prime}=0. Then V_{\tilde{p}(x),\sigma}^{(s)}\oplus V_{\tilde{p}^{*}(x),\sigma}^{(s)} is a non-degenerate quadratic space over k . Indeed, suppose there is v_{0}\in V_{\tilde{p}(x),\sigma}^{(s)}\ \ \text{ with }\ \ v_{0}\neq 0\ \ \text{ such that }\ \ \langle v_{0},v\rangle=0,\ \forall v\in V_{\tilde{p}^{*}(x),\sigma}^{(s)}. Write (\tilde{p}(x)^{m})={\rm Ann}_{k[x]}(v_{0})=\{a(x)\in k[x]:a(x)v_{0}=0\}. Then there is w_{0}\in V_{\tilde{p}(x),\sigma}^{(s)} such that \tilde{p}(x)^{m-1}v_{0}=\tilde{p}(x)^{s-1}w_{0}\neq 0 . Since \tilde{p}(\sigma^{-1})^{m-1}v\in V_{\tilde{p}^{*}(x),\sigma}^{(s)}\ \ \ \text{for}\ \ \ v\in V_{\tilde{p}^{*}(x),\sigma}^{(s)}, one obtains \langle\tilde{p}(x)^{m-1}v_{0},v\rangle=\langle v_{0},\tilde{p}(\sigma^{-1})^{m-1}v\rangle=0 for all v\in V_{\tilde{p}^{*}(x),\sigma}^{(s)} . Since \tilde{p}^{*}(0)^{s-1}\sigma^{(1-s)\deg(\tilde{p}(x))}(w)\in V_{\tilde{p}^{*}(x),\sigma}^{\prime} for any w\in V_{\tilde{p}^{*}(x),\sigma}^{\prime} , one has \displaystyle\langle\tilde{p}(x)^{m-1}v_{0},w\rangle=\langle\tilde{p}(x)^{s-1}w_{0},w\rangle \displaystyle= \displaystyle\langle w_{0},\tilde{p}^{*}(x)^{s-1}(\tilde{p}^{*}(0)^{s-1}\sigma^{(1-s)\deg(\tilde{p}(x))}(w))\rangle=\langle w_{0},0\rangle=0 for any w\in V_{\tilde{p}^{*}(x),\sigma}^{\prime} . Since V is non-degenerate, one concludes that \tilde{p}(x)^{m-1}v_{0}=\tilde{p}(x)^{s-1}w_{0}=0 by [ 3 , Lemma 3.1] . A contradiction is derived. Therefore V_{\tilde{p}(x),\sigma}\oplus V_{\tilde{p}^{*}(x),\sigma}=(V_{\tilde{p}(x),\sigma}^{(s)}\oplus V_{\tilde{p}^{*}(x),\sigma}^{(s)})\perp(W_{\tilde{p}(x),\sigma}\oplus W_{\tilde{p}^{*}(x),\sigma}) where W_{\tilde{p}(x),\sigma}=\{v\in V_{\tilde{p}(x),\sigma}:\ \langle v,V_{\tilde{p}^{*}(x),\sigma}^{(s)}\rangle=0\} and W_{\tilde{p}^{*}(x),\sigma})=\{v\in V_{\tilde{p}^{*}(x),\sigma}:\ \langle v,V_{\tilde{p}(x),\sigma}^{(s)}\rangle=0\}. The result follows from induction on W_{\tilde{p}(x),\sigma}\oplus W_{\tilde{p}^{*}(x),\sigma} . ∎ The following lemma is an analogue of [ 3 , Lemma 1.1] for Case 3 in [ 3 , §3] which is also given in [ 1 , Remark 2.10] . For convenience, we provide a proof which is close to the original proof [ 3 , Lemma 1.1] . Lemma 2.6 . Let (V,\langle\ ,\ \rangle) be a non-degenerate quadratic space over a field of k with char(k)\neq 2 . Suppose \sigma\in{\rm O}(V) such that the minimal polynomial of \sigma is \tilde{p}(x)\tilde{p}^{*}(x) where \tilde{p}(x) is a monic irreducible separable polynomial with \tilde{p}(x)\neq\tilde{p}^{*}(x) . If E=k[x]/(\tilde{p}(x)\tilde{p}^{*}(x)) and V is E -module via \sigma , then V over E admits one and only one hermitian inner product h with respect to the involution \iota of E over k induced by x\mapsto x^{-1} such that \langle u,v\rangle={\rm Tr}_{E/k}(h(u,v))\ \ \ \forall\ u,v\in V. Proof. For any u,v\in V , one can define a linear map l_{u,v}:\ E\longrightarrow k;\ \lambda\mapsto\langle\lambda u,v\rangle over k . Since E/k is an etale algebra, the reduced trace map E\times E\longrightarrow k;\ \ (\alpha,\beta)\mapsto{\rm Tr}_{E/k}(\alpha\beta) is non-degenerate by [ 5 , (9.9) Theorem] . Then there is a unique h(u,v)\in E\ \ \text{such that}\ \ l_{u,v}(\lambda)=\langle\lambda u,v\rangle={\rm Tr}_{E/k}(h(u,v)\lambda) for all \lambda\in E . In particular, \langle u,v\rangle={\rm Tr}_{E/k}(h(u,v)) . We claim that h:\ V\times V\longrightarrow E;\ (u,v)\mapsto h(u,v) is a hermitian form over E with respect to the involution \iota induced by sending x\mapsto x^{-1} . First of all, V is a free E -module of finite rank by Theorem 2.5 . Since any \lambda\in E can be written in terms of x over k , one has \langle\lambda u,v\rangle=\langle u,\iota(\lambda)v\rangle=\langle\iota(\lambda)v,u\rangle for all u,v\in V . This implies that {\rm Tr}_{E/k}(h(u,v)\lambda)={\rm Tr}_{E/k}(h(v,u)\iota(\lambda))={\rm Tr}_{E/k}(\iota(h(v,u))\lambda) for all \lambda\in E . Therefore h(u,v)=\iota(h(v,u)) as desired. Since (V,\langle\ ,\ \rangle) is non-degenerate, one concludes that h is also non-degenerate. Suppose there is another hermitian form h^{\prime} on V over E such that \langle u,v\rangle={\rm Tr}_{E/k}(h(u,v))={\rm Tr}_{E/k}(h^{\prime}(u,v)) for all u,v\in V . Then {\rm Tr}_{E/k}(\lambda h(u,v))={\rm Tr}_{E/k}(\lambda h^{\prime}(u,v)) for any \lambda\in E . This implies that h=h^{\prime} and the uniqueness property for h holds. ∎ The following result is an analogue of [ 3 , Theorem 3.3] for Case 3 in [ 3 , §3] . Theorem 2.7 . Let (V,\langle\ ,\ \rangle) be a non-degenerate quadratic space over a field of k with char(k)\neq 2 and f(x) be the characteristic polynomial of \sigma\in{\rm O}(V) . Suppose that \tilde{p}(x) is a monic irreducible separable factor of f(x) with \tilde{p}(x)\neq\tilde{p}^{*}(x) . If V_{\tilde{p}(x),\sigma}\oplus V_{\tilde{p}^{*}(x),\sigma}=(V_{\tilde{p}(x),\sigma}^{(1)}\oplus V_{\tilde{p}^{*}(x),\sigma}^{(1)})\perp\cdots\perp(V_{\tilde{p}(x),\sigma}^{(i)}\oplus V_{\tilde{p}^{*}(x),\sigma}^{(i)})\perp\cdots is an orthogonal splitting in Theorem 2.5 , then the space H_{\sigma}^{(i)}=(V_{\tilde{p}(x),\sigma}^{(i)}/\tilde{p}(x)V_{\tilde{p}(x),\sigma}^{(i)})\oplus(V_{\tilde{p}^{*}(x),\sigma}^{(i)}/\tilde{p}^{*}(x)V_{\tilde{p}^{*}(x),\sigma}^{(i)}) over E=k[x]/(\tilde{p}(x)\tilde{p}^{*}(x)) admits a unique hermitian inner product h_{\sigma}^{(i)} with respect to the involution \iota induced by x\mapsto x^{-1} satisfying \langle\tilde{p}(x)^{i-1}u+u^{*},\tilde{p}(x)^{i-1}v+v^{*}\rangle={\rm Tr}_{E/k}(h_{\sigma}^{(i)}(\overline{u+u^{*}},\overline{v+v^{*}})) for u,v\in V_{\tilde{p}(x),\sigma}^{(i)} and u^{*},v^{*}\in V_{\tilde{p}^{*}(x),\sigma}^{(i)} with i=1,2,\cdots . Moreover, the sequence of isomorphism classes of the hermitian spaces (H_{\sigma}^{(i)},h_{\sigma}^{(i)}) forms a complete invariant of pair (V_{\tilde{p}(x),\sigma}\oplus V_{\tilde{p}^{*}(x),\sigma},\sigma) . Proof. Define a bilinear map b_{\sigma}^{(i)} on the vector space H_{\sigma}^{(i)} over k by b_{\sigma}^{(i)}(\bar{\xi}+\bar{\xi}^{*},\bar{\eta}+\bar{\eta}^{*})=\langle\tilde{p}(x)^{i-1}\xi+\xi^{*},\ \tilde{p}(x)^{i-1}\eta+\eta^{*}\rangle where \xi,\eta\in V_{\tilde{p}(x),\sigma}^{(i)} and \xi^{*},\eta^{*}\in V_{\tilde{p}^{*}(x),\sigma}^{(i)} . It can be verified that b_{\sigma}^{(i)} is well-defined and symmetric. Suppose \xi_{0}\in V_{\tilde{p}(x),\sigma}^{(i)} satisfying \bar{\xi}_{0}\neq 0 in H^{(i)} . Then \tilde{p}(x)^{i-1}\xi_{0}\neq 0 . Since V_{\tilde{p}(x),\sigma}^{(i)}\oplus V_{\tilde{p}^{*}(x),\sigma}^{(i)} is non-degenerate, there is \eta_{0}^{*}\in V_{\tilde{p}^{*}(x),\sigma}^{(i)} such that \langle\tilde{p}(x)^{i-1}\xi_{0},\eta_{0}^{*}\rangle\neq 0 . Therefore b_{\sigma}^{(i)}(\bar{\xi}_{0},\bar{\eta_{0}^{*}})=\langle\tilde{p}(x)^{i-1}\xi_{0},\eta_{0}^{*}\rangle\neq 0. This implies that b_{\sigma}^{(i)} is non-degenerate. Since \sigma induces an isometry of quadratic space (H_{\sigma}^{(i)},b_{\sigma}^{(i)}) with the minimal polynomial \tilde{p}(x)\tilde{p}^{*}(x) , one obtains the required hermitian form h_{\sigma}^{(i)} by applying Lemma 2.6 . Since (H_{\sigma}^{(i)},b_{\sigma}^{(i)}) is orthogonal sum of hyperbolic planes by [ 3 , Lemma 3.1] , the hermitian space (H_{\sigma}^{(i)},h_{\sigma}^{(i)}) does not depend on the choice of V_{\tilde{p}(x),\sigma}^{(i)}\oplus V_{\tilde{p}^{*}(x),\sigma}^{(i)} up to isomorphism by Lemma 2.6 . Since the isomorphism classes of hermitian spaces (H_{\sigma}^{(i)},h_{\sigma}^{(i)}) determine the ranks of free k[x]/(p(x)^{i}) -modules V_{\tilde{p}(x),\sigma}^{(i)} for i=1,2,\cdots , the sequence of isomorphism classes of hermitian spaces (H_{\sigma}^{(i)},h_{\sigma}^{(i)}) determines the isomorphic class V_{\tilde{p}(x),\sigma} as k[x] -module which is a complete invariant for the pair (V_{\tilde{p}(x),\sigma}\oplus V_{\tilde{p}^{*}(x),\sigma},\sigma) . ∎ 3. More complements about Milnor’s classification of conjugacy classes of orthogonal groups Since a complete invariant for isomorphic classes of pair of a quadratic space with an isometry is given by a sequence of hermitian spaces by [ 3 , Theorem 3.3] , it is natural to ask if any sequence of hermitian spaces can be realized as a complete invariant of some pair of a quadratic space with an isometry. This is proved for semi-simple isometries by [ 3 , Lemma 1.2] . In order to establish this result for general cases, we first need the following lemma. Lemma 3.1 . Let p(x) be a monic irreducible polynomial of degree 2d over a field k with p(0)\neq 0 . If l is a positive integer and s(x)=x^{-d}p(x) , then \{x^{i}s(x)^{j}:\ 0\leq i\leq 2d-1,\ 0\leq j\leq l-1\} is a basis of k[x]/(p(x)^{l}) as k -vector space. Proof. Since \dim_{k}(k[x]/(p(x)^{l}))=2dl as a k -vector space, one only needs to show that the given vectors are linearly independent over k . Suppose \sum_{i=0}^{2d-1}\sum_{j=0}^{l-1}a_{ij}\bar{x}^{i}s(\bar{x})^{j}=0\ \ \ \text{in}\ \ \ k[x]/(p(x)^{l}) where a_{ij}\in k for 0\leq i\leq 2d-1 and 0\leq j\leq l-1 . Since the following natural maps give a short exact sequence 0\longrightarrow(p(x))/(p(x)^{l})\longrightarrow k[x]/(p(x)^{l})\longrightarrow k[x]/(p(x))\longrightarrow 0 as k[x] -modules, one obtains a_{i0}=0 for 0\leq i\leq 2d-1 . This implies that s(x)\sum_{i=0}^{2d-1}\sum_{j=1}^{l-1}a_{ij}\bar{x}^{i}s(\bar{x})^{j-1}=0\ \ \ \text{in}\ \ \ k[x]/(p(x)^{l}). Since k[x]/(p(x)^{l-1})\xrightarrow{\cdot s(x)}(p(x))/(p(x)^{l});\ \alpha(x)\mapsto\alpha(x)s(x) is an isomorphism of k[x] -modules, one concludes \sum_{i=0}^{2d-1}\sum_{j=1}^{l-1}a_{ij}\bar{x}^{i}s(\bar{x})^{j-1}=0\ \ \ \text{in}\ \ \ k[x]/(p(x)^{l-1}). Then a_{ij}=0 for 0\leq i\leq 2d-1 and 1\leq j\leq l-1 by induction on l and the result follows. ∎ The following construction gives an answer to the realization of [ 3 , Theorem 3.3] , which can also be viewed as generalization of [ 3 , Lemma 1.2] . Proposition 3.2 . Let p(x) be a monic irreducible polynomial of degree bigger than 1 over a field k with char(k)\neq 2 and p^{*}(x)=p(x) . Suppose that M is a free k[x]/(p(x)^{l}) -module of rank n with a positive integer l and F=k[x]/(p(x)) . If (M/p(x)M,h) is a hermitian space over F with respect to the involution \iota induced by x\mapsto x^{-1} , there is a non-degenerate inner product (M,\langle\ ,\rangle) over k with an isometry \sigma such that the minimal polynomial of \sigma is p(x)^{l} and {\rm Tr}_{F/k}(h(\bar{u},\bar{v}))=\langle(\sigma^{-1}p(\sigma))^{l-1}u,v\rangle,\ \ \forall u,v\in M. Proof. Let \bar{v}_{1},\cdots,\bar{v}_{n} be an orthogonal basis of (M/p(x)M,h) . Since M is a free k[x]/(p(x)^{l}) -module of rank n , one has M=(k[x]/(p(x)^{l}))v_{1}\oplus\cdots\oplus(k[x]/(p(x)^{l}))v_{n}. By Lemma 3.1 , each k -vector subspace (k[x]/(p(x)^{l}))v_{s} has a basis \{x^{i}s(x)^{j}v_{s}:\ 0\leq i\leq 2d-1,\ 0\leq j\leq l-1\} over k for 1\leq s\leq n . Define an inner product on this subspace satisfying (3) \langle x^{i}s(x)^{j}v_{s},x^{i^{\prime}}s(x)^{j^{\prime}}v_{s}\rangle_{s}=\langle x^{|i-i^{\prime}|}s(x)^{j+j^{\prime}}v_{s},v_{s}\rangle_{s} and (4) \langle x^{i}s(x)^{j}v_{s},v_{s}\rangle_{s}=\begin{cases}{\rm Tr}_{F/k}(x^{i}h(\bar{v}_{s},\bar{v}_{s}))\ \ \ &j=l-1,\ 0\leq i0 , where q(x)=(x+1) or (x-1) . If (M/q(x)M,b) is a non-degenerate quadratic space over k with char(k)\neq 2 , there is a non-degenerate quadratic space (M,\langle\ ,\rangle) over k with an isometry \sigma such that the minimal polynomial of \sigma is q(x)^{l} and b(\bar{u},\bar{v})=\langle(\sigma-\sigma^{-1})^{l-1}u,v\rangle,\ \ \forall u,v\in M. Proof. Let \bar{v}_{1},\cdots,\bar{v}_{n} be an orthogonal basis of (M/q(x)M,b) . Since M is a free k[x]/(q(x)^{l}) -module of rank n , one has M=(k[x]/(q(x)^{l}))v_{1}\oplus\cdots\oplus(k[x]/(q(x)^{l}))v_{n}. Since each k -vector subspace (k[x]/(p(x)^{l}))v_{s} has a basis \{(x-x^{-1})^{i}v_{s}:\ 0\leq i\leq l-1\} over k for 1\leq s\leq n , one defines an inner product on this subspace by (5) \langle(x-x^{-1})^{i}v_{s},(x-x^{-1})^{j}v_{s}\rangle_{s}=(-1)^{j}\langle(x-x^{-1})^{i+j}v_{s},v_{s}\rangle_{s} and (6) \langle(x-x^{-1})^{i}v_{s},v_{s}\rangle_{s}=\begin{cases}b(\bar{v}_{s},\bar{v}_{s})\ \ \ &i=l-1\\ 0\ \ \ &\text{otherwise}\end{cases} for 1\leq s\leq n . Since the determinant of the inner product \langle\ ,\rangle_{s} with respect to this basis is equal to \pm b(\bar{v}_{s},\bar{v}_{s})^{l}\neq 0 , the quadratic space (M,\langle\ ,\rangle) is non-degenerate where \langle\ ,\rangle=\langle\ ,\rangle_{1}\perp\cdots\perp\langle\ ,\rangle_{n} . Write x=\sum_{i=0}^{l-1}\alpha_{i}(x-x^{-1})^{i}\ \ \ \text{in}\ \ \ k[x]/(p(x)^{l}) where \alpha_{i}\in k for 0\leq i\leq l-1 . Then x=\sum_{i=0}^{l-1}\alpha_{i}(x-x^{-1})^{i}+(x-x^{-1})^{r}x^{a}g(x) where r,a\in\mathbb{Z} with r\geq l and g(x)\in k[x] with (g(x),q(x))=1 . Replacing x with x^{-1} in the above identity, one obtains x^{-1}=\sum_{i=0}^{l-1}(-1)^{i}\alpha_{i}(x-x^{-1})^{i}+(x-x^{-1})^{r}x^{b}h(x) where b\in\mathbb{Z} and h(x)\in k[x] with (q(x),h(x))=1 . Then \langle(x-x^{-1})^{i}xv_{s},v_{s}\rangle_{s}=\alpha_{l-1-i}b(\bar{v}_{s},\bar{v}_{s})=\langle(x-x^{-1})^{i}v_{s},x^{-1}v_{s}\rangle_{s} by ( 6 ). This implies that the map \sigma:M\longrightarrow M;\ m\mapsto xm\ \ \ \ \forall m\in M is an isometry of (M,\langle\ ,\rangle) as desired. ∎ In order to give a proof of Theorem 1.2 , one needs the similar result as the remark after [ 3 , Theorem 3.3] for Case 2) in [ 3 ] . Proposition 3.4 . Let (V,\langle\ ,\ \rangle) be a non-degenerate quadratic space over a field of k with char(k)\neq 2 and f(x) be the characteristic polynomial of \sigma\in{\rm O}(V) . Suppose that q(x)=(x+1) or (x-1) is a factor of f(x) and V_{q(x),\sigma}=V_{q(x),\sigma}^{(1)}\perp\cdots\perp V_{q(x),\sigma}^{(l)}\perp\cdots where (V_{q(x),\sigma}^{(l)},\langle\ ,\rangle) is non-degenerate and free as k[x]/(q(x)^{l}) -module by [ 3 , Theorem 3.2] . When l is even, then V_{q(x),\sigma}^{(l)} is an orthogonal sum of hyperbolic planes. When l is odd, then V_{q(x),\sigma}^{(l)}=W\perp H where (W,\langle\ ,\rangle)\cong(V_{q(x),\sigma}^{(l)}/q(x)V_{q(x),\sigma}^{(l)},(-1)^{\frac{l-1}{2}}b_{\sigma}^{(l)}) defined by ( 2 ) and H is an orthogonal sum of hyperbolic planes. Proof. When l is even, then \dim_{k}(q(x)^{\frac{l}{2}}V_{q(x),\sigma}^{(l)})=\frac{1}{2}\dim_{k}(V_{q(x),\sigma}^{(l)}). Since q(x)^{\frac{l}{2}}V_{q(x),\sigma}^{(l)} is isotropic, one has that V_{q(x),\sigma}^{(l)} is an orthogonal sum of hyperbolic planes as desired. When l is odd, then q(x)^{\frac{l+1}{2}}V_{q(x),\sigma}^{(l)}\subset q(x)^{\frac{l-1}{2}}V_{q(x),\sigma}^{(l)}\subseteq(q(x)^{\frac{l+1}{2}}V_{q(x),\sigma}^{(l)})^{\perp}. Since \dim_{k}(q(x)^{\frac{l+1}{2}}V_{q(x),\sigma}^{(l)})+\dim_{k}(q(x)^{\frac{l+1}{2}}V_{q(x),\sigma}^{(l)})^{\perp}=\dim_{k}(V_{q(x),\sigma}^{(l)}) by [ 4 , 42:6] , one concludes that q(x)^{\frac{l-1}{2}}V_{q(x),\sigma}^{(l)}=(q(x)^{\frac{l+1}{2}}V_{q(x),\sigma}^{(l)})^{\perp}. Since q(x)^{\frac{l+1}{2}}V_{q(x),\sigma}^{(l)} is isotropic, there is an orthogonal sum H of hyperbolic planes such that q(x)^{\frac{l+1}{2}}V_{q(x),\sigma}^{(l)}\subset H\ \ \ \text{and}\ \ \ V_{q(x),\sigma}^{(l)}=H\perp H^{\perp}. Therefore q(x)^{\frac{l-1}{2}}V_{q(x),\sigma}^{(l)}=(q(x)^{\frac{l+1}{2}}V_{q(x),\sigma}^{(l)})^{\perp}=q(x)^{\frac{l+1}{2}}V_{q(x),\sigma}^{(l)}\perp H^{\perp}. This induces a canonical isomorphism of k -vector spaces V_{q(x),\sigma}^{(l)}/q(x)V_{q(x),\sigma}^{(l)}\xrightarrow{\Delta^{\frac{l-1}{2}}\cdot}q(x)^{\frac{l-1}{2}}V_{q(x),\sigma}^{(l)}/q(x)^{\frac{l+1}{2}}V_{q(x),\sigma}^{(l)}\cong H^{\perp} where \Delta=\sigma-\sigma^{-1} . Since \langle\Delta u,v\rangle=-\langle u,\Delta v\rangle and b_{\sigma}^{(l)}(\bar{u},\bar{v})=\langle\Delta^{l-1}u,v\rangle=(-1)^{\frac{l-1}{2}}\langle\Delta^{\frac{l-1}{2}}u,\Delta^{\frac{l-1}{2}}v\rangle,\ \ \forall u,v\in V_{q(x),\sigma}^{(l)} by ( 2 ), the canonical isomorphism is also an isomorphism of the corresponding quadratic spaces up to (-1)^{\frac{l-1}{2}} as desired. ∎ As an application, one has the following sufficient conditions for Theorem 1.2 over any field k with char(k)\neq 2 . Corollary 3.5 . Let (V,\langle\ ,\ \rangle) be a non-degenerate quadratic space over a field of k with char(k)\neq 2 and f(x) be the characteristic polynomial of \sigma\in{\rm O}(V) with the factorization ( 1 ). Write m_{0}=\sharp\{2l+1:V_{x+1,\sigma}^{(2l+1)}\neq 0\}+\sharp\{2l+1:V_{x-1,\sigma}^{(2l+1)}\neq 0\}. If m_{0}\leq 1 and m_{1}=0 , then any \tau\in{\rm O}(V) which is conjugate to \sigma in {\rm GL}(V) is conjugate to \sigma in {\rm O}(V) . Proof. By [ 3 , Lemma 3.1] , one has \displaystyle V= \displaystyle V_{x+1,\sigma}\perp V_{x-1,\sigma}\underset{1\leq j\leq m_{2}}{\perp}(V_{\tilde{p}_{j}(x),\sigma}\oplus V_{\tilde{p}_{j}^{*}(x),\sigma}) \displaystyle= \displaystyle V_{x+1,\tau}\perp V_{x-1,\tau}\underset{1\leq j\leq m_{2}}{\perp}(V_{\tilde{p}_{j}(x),\tau}\oplus V_{\tilde{p}_{j}^{*}(x),\tau})). Since \sigma and \tau are conjugates in {\rm GL}(V) , the orthogonal splittings V_{x\pm 1,\sigma}=V_{x\pm 1,\sigma}^{(1)}\perp\cdots\perp V_{x\pm 1,\sigma}^{(l)}\perp\cdots and V_{x\pm 1,\tau}=V_{x\pm 1,\tau}^{(1)}\perp\cdots\perp V_{x\pm 1,\tau}^{(l)}\perp\cdots satisfy that V_{x\pm 1,\sigma}^{(l)} and V_{x\pm 1,\tau}^{(l)} are free k[x]/(x\pm 1)^{l} -modules with the same rank by [ 3 , Theorem 3.2] . The structure theorem of finitely generated modules also implies that V_{\tilde{p}_{j}(x),\sigma}\cong V_{\tilde{p}_{j}(x),\tau} as k[x] -modules. Then there is an isomorphism of quadratic spaces \phi_{j}:(V_{\tilde{p}_{j}(x),\sigma}\oplus V_{\tilde{p}_{j}^{*}(x),\sigma},\langle,\rangle)\xrightarrow{\cong}(V_{\tilde{p}_{j}(x),\tau}\oplus V_{\tilde{p}_{j}^{*}(x),\tau},\langle,\rangle) such that \phi_{j}\circ\sigma=\tau\circ\phi_{j} for 1\leq j\leq m_{2} . Since m_{0}\leq 1 , one can simply assume that there is at most one l_{0} such that V_{x-1,\sigma}^{(2l_{0}+1)}\neq 0\ \ \ \text{and}\ \ \ V_{x-1,\tau}^{(2l_{0}+1)}\neq 0. Applying Proposition 3.4 , one obtains (V_{x-1,\sigma}^{(2l_{0}+1)}/(x-1)V_{x-1,\sigma}^{(2l_{0}+1)},b_{\sigma}^{(2l_{0}+1)})\cong(V_{x-1,\tau}^{(2l_{0}+1)}/(x-1)V_{x-1,\tau}^{(2l_{0}+1)},b_{\tau}^{(2l_{0}+1)}) by Witt’s theorem [ 4 , 42:16. Theorem] . Since the non-degenerate sympletic spaces given by b_{\sigma}^{(2l)} and b_{\tau}^{(2l)} are automotically isomorphic for all l , there is an isomorphism of quadratic spaces \rho_{\pm}:\ (V_{x\pm 1,\sigma},\langle\ ,\ \rangle)\xrightarrow{\cong}(V_{x\pm 1,\tau},\langle\ ,\ \rangle) such that \rho_{\pm}\circ\sigma=\tau\circ\rho_{\pm} by [ 3 , Theorem 3.4] . Then \phi=\rho_{+}\perp\rho_{-}\perp\phi_{1}\perp\cdots\perp\phi_{m_{2}}\in{\rm O}(V)\ \ \ \text{and}\ \ \ \tau=\phi\circ\sigma\circ\phi^{-1} as desired. ∎ 4. Proof of main theorem In this section, we assume that k is a non-archimedean local field. Theorem 4.1 . Let V be a non-degenerated quadratic space over a non-archimedean local field k with char(k)\neq 2 . Suppose \sigma\in{\rm O}(V) and f(x) is the characteristic polynomial of \sigma with the factorization ( 1 ). Write m_{0}=\sharp\{2l+1:V_{x+1,\sigma}^{(2l+1)}\neq 0\}+\sharp\{2l+1:V_{x-1,\sigma}^{(2l+1)}\neq 0\}. Then any isometry in {\rm O}(V) which is conjugate to \sigma in {\rm GL}(V) is conjugate to \sigma in {\rm O}(V) if and only if one of the following conditions holds (i) m_{0}\leq 1 and m_{1}=0 . (ii) m_{0}=0 and m_{1}=1 and V_{p_{1}(x),\sigma}=V_{p_{1}(x),\sigma}^{(2l_{1}+1)} for some integer l_{1}\geq 0 . (iii) m_{0}=m_{1}=1 and V_{p_{1}(x),\sigma}=V_{p_{1}(x),\sigma}^{(2l_{1}+1)} for some integer l_{1}\geq 0 and the unique non-zero quadratic space (V_{x\pm 1,\sigma}^{(2l_{0}+1)}/(x\pm 1)V_{x\pm 1,\sigma}^{(2l_{0}+1)},b_{\sigma}^{(2l_{0}+1)}) defined by ( 2 ) is either one dimensional or a hyperbolic plane. Proof. Let \tau\in{\rm O}(V) such that \tau is conjugate to \sigma in {\rm GL}(V) . Then the characteristic polynomial of \tau is also f(x) and \displaystyle V= \displaystyle V_{x+1,\sigma}\perp V_{x-1,\sigma}\perp(\underset{1\leq i\leq m_{1}}{\perp}V_{p_{i}(x),\sigma})\perp(\underset{1\leq j\leq m_{2}}{\perp}(V_{\tilde{p}_{j}(x),\sigma}\oplus V_{\tilde{p}_{j}^{*}(x),\sigma})) \displaystyle= \displaystyle V_{x+1,\tau}\perp V_{x-1,\tau}\perp(\underset{1\leq i\leq m_{1}}{\perp}V_{p_{i}(x),\tau})\perp(\underset{1\leq j\leq m_{2}}{\perp}(V_{\tilde{p}_{j}(x),\tau}\oplus V_{\tilde{p}_{j}^{*}(x),\tau})) by [ 3 , Lemma 3.1] . Moreover, there are orthogonal splittings V_{x\pm 1,\sigma}=V_{x\pm 1,\sigma}^{(1)}\perp\cdots\perp V_{x\pm 1,\sigma}^{(l)}\perp\cdots and V_{x\pm 1,\tau}=V_{x\pm 1,\tau}^{(1)}\perp\cdots\perp V_{x\pm 1,\tau}^{(l)}\perp\cdots where V_{x\pm 1,\sigma}^{(l)} and V_{x\pm 1,\tau}^{(l)} are free k[x]/(x\pm 1)^{l} -modules with the same rank by [ 3 , Theorem 3.2] . Similarly, V_{p_{i}(x),\sigma}=V_{p_{i}(x),\sigma}^{(1)}\perp\cdots\perp V_{p_{i}(x),\sigma}^{(l)}\perp\cdots and V_{p_{i}(x),\tau}=V_{p_{i}(x),\tau}^{(1)}\perp\cdots\perp V_{p_{i}(x),\tau}^{(l)}\perp\cdots where V_{p_{i}(x),\sigma}^{(l)} and V_{p_{i}(x),\tau}^{(l)} are free k[x]/(p_{i}(x)^{l}) -modules of the same rank by [ 3 , Theorem 3.2] for 1\leq i\leq m_{1} . By the same argument as those in Corollary 3.5 , there is an isomorphism of quadratic spaces \phi_{j}:\ (V_{\tilde{p}_{j}(x),\sigma}\oplus V_{\tilde{p}_{j}^{*}(x),\sigma},\langle\ ,\ \rangle)\xrightarrow{\cong}(V_{\tilde{p}_{j}(x),\tau}\oplus V_{\tilde{p}_{j}^{*}(x),\tau},\langle\ ,\ \rangle) such that \phi_{j}\circ\sigma=\tau\circ\phi_{j} for 1\leq j\leq m_{2} . Sufficiency . One only needs to treat (ii) and (iii) by Corollary 3.5 . If m_{0}=0 and m_{1}=1 and V_{p_{1}(x),\sigma}=V_{p_{1}(x),\sigma}^{(2l_{1}+1)} , then V_{p_{1}(x),\tau}=V_{p_{1}(x),\tau}^{(2l_{1}+1)} and (V_{p_{1}(x),\sigma}^{(2l_{1}+1)}/p_{1}(x)V_{p_{1}(x),\sigma}^{(2l_{1}+1)},\ b_{\sigma}^{(2l_{1}+1)})\cong(V_{p_{1}(x),\tau}^{(2l_{1}+1)}/p_{1}(x)V_{p_{1}(x),\tau}^{(2l_{1}+1)},\ b_{\tau}^{(2l_{1}+1)}) where (7) b_{\sigma}^{(2l_{1}+1)}(\bar{u},\bar{v})=\langle(\sigma^{-d}p_{1}(\sigma))^{2l_{1}}u,v\rangle,\ \ \ \forall u,v\in V_{p_{1}(x),\sigma}^{(2l_{1}+1)} and (8) b_{\tau}^{(2l_{1}+1)}(\bar{u},\bar{v})=\langle(\tau^{-d}p_{1}(\tau))^{2l_{1}}u,v\rangle,\ \ \ \forall u,v\in V_{p_{1}(x),\tau}^{(2l_{1}+1)} with \deg(p_{1}(x))=2d by the remark after [ 3 , Theorem 3.3] and Witt’s theorem [ 4 , 42:16. Theorem] . Let (V_{p_{1}(x),\sigma}^{(2l_{1}+1)}/p_{1}(x)V_{p_{1}(x),\sigma}^{(2l_{1}+1)},h_{\sigma}^{(2l_{1}+1)})\ \text{ and }\ (V_{p_{1}(x),\tau}^{(2l_{1}+1)}/p_{1}(x)V_{p_{1}(x),\tau}^{(2l_{1}+1)},h_{\tau}^{(2l_{1}+1)}) be the corresponding hermitian spaces such that b_{\sigma}^{(2l_{1}+1)}={\rm Tr}_{F/k}\circ h_{\sigma}^{(2l_{1}+1)}\ \ \ \text{and}\ \ \ b_{\tau}^{(2l_{1}+1)}={\rm Tr}_{F/k}\circ h_{\tau}^{(2l_{1}+1)} with F=k[x]/(p_{1}(x)) by [ 3 , Lemma 1.1] . One has an isomorphism of hermitian spaces (V_{p_{1}(x),\sigma}^{(2l_{1}+1)}/p_{1}(x)V_{p_{1}(x),\sigma}^{(2l_{1}+1)},h_{\sigma}^{(2l_{1}+1)})\cong(V_{p_{1}(x),\tau}^{(2l_{1}+1)}/p_{1}(x)V_{p_{1}(x),\tau}^{(2l_{1}+1)},h_{\tau}^{(2l_{1}+1)}) by [ 3 , Theorem 2.7] . Then there is an isomorphism of quadratic spaces \psi_{1}:(V_{p_{1}(x),\sigma},\langle\ ,\rangle)\xrightarrow{\cong}(V_{p_{1}(x),\tau},\langle\ ,\rangle) such that \psi_{1}\circ\sigma=\tau\circ\psi_{1} by [ 3 , Theorem 3.3] . Therefore \phi=(\rho_{+}\perp\rho_{-}\perp\psi_{1}\perp\phi_{1}\perp\cdots\perp\phi_{m_{2}})\in{\rm O}(V) and \tau=\phi\circ\sigma\circ\phi^{-1} as desired. If m_{0}=m_{1}=1 and V_{p_{1}(x),\sigma}=V_{p_{1}(x),\sigma}^{(2l_{1}+1)} for some integer l_{1}\geq 0 and one simply assume that (V_{x-1,\sigma}^{(2l_{0}+1)}/(x-1)V_{x-1,\sigma}^{(2l_{0}+1)},b_{\sigma}^{(2l_{0}+1)}) is either one dimensional or a hyperbolic plane, then V_{p_{1}(x),\tau}=V_{p_{1}(x),\tau}^{(2l_{1}+1)}\ \ \ \text{ and }\ \ \ 1\leq\dim_{k}(V_{x-1,\tau}^{(2l_{0}+1)}/(x-1)V_{x-1,\tau}^{(2l_{0}+1)})\leq 2 respectively. Since the determinants of quadratic spaces (V_{p_{1}(x),\sigma}^{(2l_{1}+1)}/p_{1}(x)V_{p_{1}(x),\sigma}^{(2l_{1}+1)},\ b_{\sigma}^{(2l_{1}+1)})\ \ \text{and}\ \ (V_{p_{1}(x),\tau}^{(2l_{1}+1)}/p_{1}(x)V_{p_{1}(x),\tau}^{(2l_{1}+1)},\ b_{\tau}^{(2l_{1}+1)}) defined by ( 7 ) and ( 8 ) are equal by [ 3 , Theorem 2.7] , one obtains the determinants of quadratic spaces (V_{x-1,\sigma}^{(2l_{0}+1)}/(x-1)V_{x-1,\sigma}^{(2l_{0}+1)},b_{\sigma}^{(2l_{0}+1)})\ \text{and}\ (V_{x-1,\tau}^{(2l_{0}+1)}/(x-1)V_{x-1,\tau}^{(2l_{0}+1)},b_{\tau}^{(2l_{0}+1)}) are equal by the remark after [ 3 , Theorem 3.3] and Proposition 3.4 . Then (V_{x-1,\sigma}^{(2l_{0}+1)}/(x-1)V_{x-1,\sigma}^{(2l_{0}+1)},b_{\sigma}^{(2l_{0}+1)})\cong(V_{x-1,\tau}^{(2l_{0}+1)}/(x-1)V_{x-1,\tau}^{(2l_{0}+1)},b_{\tau}^{(2l_{0}+1)}) as quadratic spaces by the assumption and [ 4 , 42:9] . One further obtains an isomorphism of quadratic spaces (V_{p_{1}(x),\sigma}^{(2l_{1}+1)}/p_{1}(x)V_{p_{1}(x),\sigma}^{(2l_{1}+1)},\ b_{\sigma}^{(2l_{1}+1)})\cong(V_{p_{1}(x),\tau}^{(2l_{1}+1)}/p_{1}(x)V_{p_{1}(x),\tau}^{(2l_{1}+1)},\ b_{\tau}^{(2l_{1}+1)}) by Proposition 3.4 and the remark after [ 3 , Theorem 3.3] and Witt’s theorem [ 4 , 42:16. Theorem] . The result follows from the same arguments as above and those in Corollary 3.5 . Necessity . Suppose m_{1}\geq 1 and there is a positive integer l_{1} such that V_{p_{1}(x),\sigma}^{(2l_{1})}\neq 0 . Let (V_{p_{1}(x),\sigma}^{(2l_{1})}/p_{1}(x)V_{p_{1}(x),\sigma}^{(2l_{1})},h^{(2l_{1})}) be the corresponding hermitian space given in [ 3 , Theorem 3.3] . Since k[x]/(p_{1}(x)) is a non-archimedean local field, there is another hermitian space (V_{p_{1}(x),\sigma}^{(2l_{1})}/p_{1}(x)V_{p_{1}(x),\sigma}^{(2l_{1})},h_{0}^{(2l_{1})}) with a different determinant by [ 3 , Example 3, §1] . By Proposition 3.2 , there is a non-degenerate quadratic space (V_{p_{1}(x),\sigma}^{(2l_{1})},\langle\ ,\rangle^{\prime}) with an isometry \tau_{0} such that (V_{p_{1}(x),\sigma}^{(2l_{1})}/p_{1}(x)V_{p_{1}(x),\sigma}^{(2l_{1})},\ h_{0}^{(2l_{1})}) is a complete invariant of this pair. By the remark after [ 3 , Theorem 3.3] , one has (V_{p_{1}(x),\sigma}^{(2l_{1})},\langle\ ,\rangle)\cong(V_{p_{1}(x),\sigma}^{(2l_{1})},\langle\ ,\rangle^{\prime}). Replacing \sigma|_{V_{p_{1}(x),\sigma}^{(2l_{1})}} with \tau_{0} at V_{p_{1}(x),\sigma}^{(2l_{1})} for \sigma , one obtains a new isometry \tau of (V,\langle\ ,\rangle) . Then \tau is conjugate to \sigma in {\rm GL}(V) but not in {\rm O}(V) . Suppose m_{1}\geq 1 and there are non-negative integers l_{1}\neq l_{2} such that V_{p_{1}(x),\sigma}^{(2l_{1}+1)}\neq 0 and V_{p_{1}(x),\sigma}^{(2l_{2}+1)}\neq 0 . Let (V_{p_{1}(x),\sigma}^{(2l_{1}+1)}/p_{1}(x)V_{p_{1}(x),\sigma}^{(2l_{1}+1)},h^{(2l_{1}+1)})\ \ \ \text{and}\ \ \ (V_{p_{1}(x),\sigma}^{(2l_{2}+1)}/p_{1}(x)V_{p_{1}(x),\sigma}^{(2l_{2}+1)},h^{(2l_{2}+1)}) be the corresponding hermitian spaces given in [ 3 , Theorem 3.3] respectively. Then there are hermitian spaces (V_{p_{1}(x),\sigma}^{(2l_{1}+1)}/p_{1}(x)V_{p_{1}(x),\sigma}^{(2l_{1}+1)},h_{0}^{(2l_{1}+1)})\ \ \ \text{and}\ \ \ (V_{p_{1}(x),\sigma}^{(2l_{2}+1)}/p_{1}(x)V_{p_{1}(x),\sigma}^{(2l_{2}+1)},h_{0}^{(2l_{2}+1)}) with different determinants by [ 3 , Example 3, §1] respectively. By Proposition 3.2 , there are two pairs of a non-degenerate quadratic space with an isometry ((V_{p_{1}(x),\sigma}^{(2l_{1}+1)},\langle\ ,\rangle^{\prime}),\ \tau_{1})\ \ \ \text{and}\ \ \ ((V_{p_{1}(x),\sigma}^{(2l_{2}+1)},\langle\ ,\rangle^{\prime}),\ \tau_{2}) such that h_{0}^{(2l_{1}+1)} and h_{0}^{(2l_{2}+1)} are the complete invariants respectively. Since (V_{p_{1}(x),\sigma}^{(2l_{1}+1)},\langle\ ,\rangle)\perp(V_{p_{1}(x),\sigma}^{(2l_{2}+1)},\langle\ ,\rangle)\cong(V_{p_{1}(x),\sigma}^{(2l_{1}+1)},\langle\ ,\rangle^{\prime})\perp(V_{p_{1}(x),\sigma}^{(2l_{2}+1)},\langle\ ,\rangle^{\prime}) by comparing the Hasse symbols with [ 4 , 58:3 Remark] , the remark after [ 3 , Theorem 3.3] and [ 3 , Theorem 2.7] , one obtains a new isometry \tau of (V,\langle\ ,\rangle) by replacing \sigma|_{V_{p_{1}(x),\sigma}^{(2l_{1}+1)}}\perp\sigma|_{V_{p_{1}(x),\sigma}^{(2l_{2}+1)}}\ \ \ \text{ with }\ \ \ \tau_{1}\perp\tau_{2} for \sigma . Then \tau is conjugate to \sigma in {\rm GL}(V) but not in {\rm O}(V) . Suppose m_{0}\geq 2 . Without loss of generality, we assume that there are non-negative integers l_{-} and l_{+} such that V_{x-1,\sigma}^{(2l_{-}+1)}\neq 0\ \ \ \text{ and }\ \ \ V_{x+1,\sigma}^{(2l_{+}+1)}\neq 0. Write W_{-}=V_{x-1,\sigma}^{(2l_{-}+1)}/(x-1)V_{x-1,\sigma}^{(2l_{-}+1)}\ \ \text{and}\ \ W_{+}=V_{x+1,\sigma}^{(2l_{+}+1)}/(x+1)V_{x+1,\sigma}^{(2l_{+}+1)}. In the following arguments, we’ll construct two quadratic spaces (W_{-},b_{-}) and (W_{+},b_{+}) satisfying (W_{-},b_{-})\not\cong(W_{-},b_{\sigma}^{(2l_{-}+1)})\ \ \ \text{ and }\ \ \ (W_{+},b_{+})\not\cong(W_{+},b_{\sigma}^{(2l_{+}+1)}) defined by (\ref{bil}) . By Proposition 3.3 , one obtains two pairs of a quadratic space with an isometry ((V_{x-1,\sigma}^{(2l_{-}+1)},\langle\ ,\rangle^{\prime}),\ \tau_{-})\ \ \ \text{and}\ \ \ ((V_{x+1,\sigma}^{(2l_{+}+1)},\langle\ ,\rangle^{\prime}),\ \tau_{+}) such that (W_{-},b_{-}) and (W_{+},b_{+}) are the complete invariants respectively. We further need (9) \displaystyle(V_{x-1,\sigma}^{(2l_{-}+1)},\langle\ ,\rangle)\perp(V_{x+1,\sigma}^{(2l_{+}+1)},\langle\ ,\rangle) \displaystyle\cong \displaystyle(V_{x-1,\sigma}^{(2l_{-}+1)},\langle\ ,\rangle^{\prime})\perp(V_{x+1,\sigma}^{(2l_{+}+1)},\langle\ ,\rangle^{\prime}) in the above construction. A new isometry \tau of (V,\langle\ ,\rangle) is obtained by replacing \sigma|_{V_{x-1,\sigma}^{(2l_{-}+1)}}\perp\sigma|_{V_{x+1,\sigma}^{(2l_{+}+1)}}\ \ \ \text{ with }\ \ \ \tau_{-}\perp\tau_{+} for \sigma . Then \tau is conjugate to \sigma in {\rm GL}(V) but not in {\rm O}(V) . If \min\{\dim_{k}(W_{-}),\dim_{k}(W_{+})\}\geq 2 and neither of (W_{-},b_{\sigma}^{(2l_{-}+1)}) nor (W_{+},b_{\sigma}^{(2l_{+}+1)}) is a hyperbolic plane, we choose (W_{-},\ b_{-}) and (W_{+},\ b_{+}) with the same determinants as b_{\sigma}^{(2l_{-}+1)} and b_{\sigma}^{(2l_{+}+1)} but the different Hasse symbols respectively by [ 4 , 63:22. Theorem] . The property ( 9 ) is satisfied by comparing the Hasse symbols with [ 4 , 58:3 Remark] and Proposition 3.4 . If \dim_{k}(W_{-})=\dim_{k}(W_{+})=2 and one of (W_{-},b_{\sigma}^{(2l_{-}+1)})\ \ \ \text{ and }\ \ \ (W_{+},b_{\sigma}^{(2l_{+}+1)}) is a hyperbolic plane with (W_{-},\ b_{\sigma}^{(2l_{-}+1)})\not\cong(W_{+},\ b_{\sigma}^{(2l_{+}+1)}) , one can assume that (W_{-},\ b_{\sigma}^{(2l_{-}+1)}) is a hyperbolic plane. Choose (W_{-},b_{-}) to be (W_{-},b_{-})\cong(W_{+},(-1)^{(l_{+}-l_{-})}b_{\sigma}^{(2l_{+}+1)}) and (W_{+},b_{+}) to be a hyperbolic plane. The property ( 9 ) follows from Proposition 3.4 . If both (W_{-},b_{\sigma}^{(2l_{-}+1)}) and (W_{+},b_{\sigma}^{(2l_{+}+1)}) are hyperbolic planes, then one can choose (W_{-},b_{-}) and (W_{+},b_{+}) such that (W_{-},(-1)^{l_{-}}b_{-})\cong(W_{+},(-1)^{l_{+}}b_{+}) which is not a hyperbolic plane but represents all units of k by [ 4 , 63:15. Example (i)] . Then the property ( 9 ) follows from Proposition 3.4 . If \dim_{k}(W_{-})=\dim_{k}(W_{+})=1 , one can choose \gamma\in N_{k(\sqrt{-\delta})/k}(k(\sqrt{-\delta})^{\times})\setminus(k^{\times})^{2} where \delta is the determinant of binary quadratic space (W_{-},(-1)^{l_{-}}b_{\sigma}^{(2l_{-}+1)})\perp(W_{+},(-1)^{l_{+}}b_{\sigma}^{(2l_{+}+1)}). Define (W_{-},b_{-}) and (W_{+},b_{+}) to be (W_{-},b_{-})\cong(W_{-},\gamma b_{\sigma}^{(2l_{-}+1)})\ \ \ \text{and}\ \ \ (W_{+},b_{+})\cong(W_{+},\gamma b_{\sigma}^{(2l_{+}+1)}). Then the property ( 9 ) follows from Proposition 3.4 and [ 4 , 63:15. Example (iii)] . The last possibility is one of W_{-} and W_{+} is one dimensional and the other one is two dimensional. Without loss of generality, we assume \dim_{k}(W_{-})=1 and \dim_{k}(W_{+})=2 . Choose \xi\in(-1)^{l_{-}+l_{+}}b_{\sigma}^{(2l_{+}+1)}(W_{+})\setminus b_{\sigma}^{(2l_{-}+1)}(W_{-}) by [ 4 , 63:15. Example (ii)] . Then (W_{+},b_{\sigma}^{(2l_{+}+1)})\cong<(-1)^{l_{-}+l_{+}}\xi>\perp<(-1)^{l_{-}+l_{+}}\eta> for some \eta\in k^{\times} . Let (W_{-},b_{-})\cong<\xi>\ \ \text{and}\ \ (W_{+},b_{+})\cong<(-1)^{l_{-}+l_{+}}\alpha>\perp<(-1)^{l_{-}+l_{+}}\eta> where \alpha is the determinant of (W_{-},b_{\sigma}^{(2l_{-}+1)}) . Then the property ( 9 ) follows from Proposition 3.4 . Suppose m_{0}=m_{1}=1 and V_{p_{1}(x),\sigma}=V_{p_{1}(x),\sigma}^{(2l_{1}+1)} for some integer l_{1}\geq 0 and the unique non-zero quadratic space (V_{x\pm 1,\sigma}^{(2l_{0}+1)}/(x\pm 1)V_{x\pm 1,\sigma}^{(2l_{0}+1)},b_{\sigma}^{(2l_{0}+1)}) defined by ( 2 ) is more than one dimensional and is not a hyperbolic plane. By [ 4 , 63:22. Theorem] , there is a quadratic space (V_{x\pm 1,\sigma}^{(2l_{0}+1)}/(x\pm 1)V_{x\pm 1,\sigma}^{(2l_{0}+1)},b_{\pm}) with the same determinants as b_{\sigma}^{(2l_{0}+1)} but the different Hasse symbol. Applying Proposition 3.3 , one obtains a pairs of a quadratic space with an isometry ((V_{x\pm 1,\sigma}^{(2l_{0}+1)},\langle\ ,\rangle^{\prime}),\ \tau_{\pm}) such that (V_{x\pm 1,\sigma}^{(2l_{0}+1)}/(x\pm 1)V_{x\pm 1,\sigma}^{(2l_{0}+1)},b_{\pm}) is the complete invariant. Let (V_{p_{1}(x),\sigma}^{(2l_{1}+1)}/p(x)V_{p_{1}(x),\sigma}^{(2l_{1}+1)},h_{\sigma}^{(2l_{1}+1)}) be the corresponding hermitian space given in [ 3 , Theorem 3.3] . Then there is another hermitian structure (V_{p_{1}(x),\sigma}^{(2l_{1}+1)}/p_{1}(x)V_{p_{1}(x),\sigma}^{(2l_{1}+1)},h_{0}^{(2l_{1}+1)}) with a different determinant by [ 3 , Example 3, §1] . Applying Proposition 3.2 , one obtains a pair ((V_{p_{1}(x),\sigma}^{(2l_{1}+1)},\langle\ ,\rangle^{\prime}),\tau_{1}) of a non-degenerate quadratic space with an isometry such that (V_{p_{1}(x),\sigma}^{(2l_{1}+1)}/p_{1}(x)V_{p_{1}(x),\sigma}^{(2l_{1}+1)},\ h_{0}^{(2l_{1}+1)}) is a complete invariant of this pair. Applying the remark after [ 3 , Theorem 3.3] and Proposition 3.4 , one has (V_{p_{1}(x),\sigma}^{(2l_{1}+1)},\langle\ ,\rangle)\perp(V_{x\pm 1,\sigma}^{(2l_{0}+1)},\langle\ ,\rangle)\cong(V_{p_{1}(x),\sigma}^{(2l_{1}+1)},\langle\ ,\rangle^{\prime})\perp(V_{x\pm 1,\sigma}^{(2l_{0}+1)},\langle\ ,\rangle^{\prime}) as quadratic spaces by comparing the determinants using [ 3 , Theorem 2.7] and the Hasse symbols using [ 4 , 58:3 Remark] . Replacing \sigma|_{V_{p_{1}(x),\sigma}^{(2l_{1}+1)}}\perp\sigma|_{V_{x\pm 1,\sigma}^{(2l_{0}+1)}}\ \ \ \text{ with }\ \ \ \tau_{1}\perp\tau_{\pm} for \sigma , one obtains \tau\in{\rm O}(V) which conjugate to \sigma in {\rm GL}(V) but not in {\rm O}(V) as desired. ∎