Structure and unique factorization in concordance groups of links Kouki Sato, Akira Yasuhara Abstract . Donald and Owens introduced two link concordance groups with a marked component and showed that they contain the knot concordance group as a direct summand with infinitely generated complements. While not explicitly posed by Donald and Owens, the problem of determining the structure of these complements arises naturally from their work. In this paper, we completely resolve this problem by proving that both complements are isomorphic to ℤ ∞ ⊕ ( ℤ / 2 ​ ℤ ) ∞ \mathbb{Z}^{\infty}\oplus(\mathbb{Z}/2\mathbb{Z})^{\infty} . Moreover, we introduce a notion of prime element and establish a unique prime decomposition theorem. This yields a canonical normal form, providing a complete description of the group structure. 1. Introduction A partly oriented link is a link in S^{3} with a marked oriented component and the remaining components unoriented. A marked oriented link is an oriented link in S^{3} with a marked component. If L and L^{\prime} are partly oriented links (resp. marked oriented link), then the connected sum L\#L^{\prime} with respect to the marked components is well-defined and commutative up to isotopy, and also regarded as a partly oriented link (resp. marked oriented link). In addition, the mirror L^{*} and the orientation reversal -L of L are also regarded as a partly oriented link (resp. marked oriented link). On such objects, Donald-Owens [ 1 ] defined the following equivalence relation. Definition 1.1 ( [ 1 , Definition 2] ) . Let L_{0} and L_{1} be partly oriented links (resp. marked oriented links). We say L_{0} and L_{1} are \chi -concordant if there is a smoothly, properly embedded compact surface F in the 4-ball B^{4} without closed components, and with Euler characteristic \chi(F)=1 such that • F consists of a single disk D and unoriented surfaces (resp. oriented surfaces) that are not disks, • \partial F=-L^{*}_{0}\#L_{1} , and \partial D is the marked component of -L^{*}_{0}\#L_{1} . In the definition above, since \chi(F\setminus D)=0 , we note that F\setminus D consists of annuli and/or Möbius bands in the partly oriented case, and of oriented annuli in the marked oriented case. We denote by \mathcal{L} (resp. \widetilde{\mathcal{L}} ) the set of \chi -concordance classes of partly oriented links (resp. marked oriented links), and call them the \chi -concordance groups . As proved in [ 1 ] , \mathcal{L} and \widetilde{\mathcal{L}} form abelian groups under connected sum respectively. Here, the trivial element 0 in both \mathcal{L} and \widetilde{\mathcal{L}} is represented by the trivial knot. Every trivial link represents 0 in \mathcal{L} . In contrast, a trivial link represents 0 in \widetilde{\mathcal{L}} if and only if the number of its components is odd, since any two marked oriented links in the same class of \widetilde{\mathcal{L}} have numbers of components congruent modulo 2 . Since a link L is \chi -concordant to K\#((-K)^{*}\#L) , the equivalent class [L] is equal to [K]+[(-K)^{*}\#L] in each of the \chi -concordance groups \mathcal{L} and \widetilde{\mathcal{L}} . It is shown in [ 1 ] that this observation induces direct sum decompositions of \mathcal{L} and \widetilde{\mathcal{L}} : \mathcal{L}=\mathcal{C}\oplus\mathcal{L}_{0}\quad\text{and}\quad\widetilde{\mathcal{L}}=\mathcal{C}\oplus\widetilde{\mathcal{L}}_{0}, where \mathcal{C} denotes the knot concordance group and \mathcal{L}_{0} and \widetilde{\mathcal{L}}_{0} consist of \chi -concordance classes of links whose marked component is a slice knot. This implies that the intrinsic properties of \mathcal{L} (resp. \widetilde{\mathcal{L}} ) are concentrated in \mathcal{L}_{0} (resp. \widetilde{\mathcal{L}}_{0} ). Since \mathcal{L}_{0} and \widetilde{\mathcal{L}}_{0} still contain classes of links with non-slice unmarked components, it appears very difficult to determine these groups, as determining \mathcal{C} is a widely open problem. Surprisingly, we obtain the following unexpected result, which determines their isomorphism classes. Theorem 1.2 (Structure Theorem) . Both \mathcal{L}_{0} and \widetilde{\mathcal{L}}_{0} are isomorphic to \mathbb{Z}^{\infty}\oplus(\mathbb{Z}/2\mathbb{Z})^{\infty} . In [ 1 ] , it is shown that both \mathcal{L}_{0} and \widetilde{\mathcal{L}}_{0} are infinitely generated and contain subgroups isomorphic to \mathbb{Z}^{\infty}\oplus(\mathbb{Z}/2\mathbb{Z})^{\infty} . However, it remained completely open to determine the isomorphism classes of these groups. For example, it was unknown which orders of elements occur in these abelian groups. Theorem 1.2 settles these problems completely. In the following, the notation \mathcal{L}_{0}^{*} denotes either \mathcal{L}_{0} or \widetilde{\mathcal{L}}_{0} , and marked links (or simply links ) refers either partly oriented links or marked oriented links. We assume that the marked component of any marked link is slice knot. For proving Theorem 1.2 , we introduce the primeness of elements in \mathcal{L}^{*}_{0} . An element x in \mathcal{L}^{*}_{0} is non-prime if x can be realized by the connected sum L_{1}\#L_{2} of two marked links L_{1} and L_{2} such that • each L_{i} has at least two components, including a marked slice component, and • the number of components of L_{1}\#L_{2} is minimal among all representatives of x . An element x\in\mathcal{L}^{*}_{0} is prime if x is neither zero nor non-prime. We stress that a non-trivial element x\in\mathcal{L}^{*}_{0} containing a 2-component link is prime. A remarkable consequence of defining primeness is the following. Theorem 1.3 (Prime Decomposition Theorem) . Any non-zero element x\in\mathcal{L}^{*}_{0} can be written in the form x=\sum_{i=1}^{n}k_{i}x_{i} where x_{i} is prime, k_{i}x_{i}\neq 0 and x_{s}\neq\pm x_{t} for any 1\leq i\leq n and 1\leq s0 , the marked oriented link W_{n} (resp. B_{n} ) is not equal to any trivial link and has order \infty (resp. order 2) in \mathcal{L}^{*}_{0} . Moreover, W_{2k-1}~(k\in\mathbb{N}) (resp. B_{2k-1}~(k\in\mathbb{N}) ) are mutually distinct up to \chi -concordance. Remark 1.6 . We note that W_{1} and B_{1} are the Whitehead link and the Borromean rings respectively. In particular, Proposition 1.5 answers to Donald-Owens’ question [ 1 , Section 4] asking what the orders of W_{1} and B_{1} are. Figure 1 . Marked oriented links W_{n} and B_{n} By ignoring the orientations on the unmarked component of a marked oriented link, we have the natural epimorphism \rho:\widetilde{\mathcal{L}}_{0}\longrightarrow\mathcal{L}_{0} . We also regard \rho as a map from the set of marked oriented links to the set of partly oriented links. The structure of \ker\rho is described as follows. Theorem 1.7 . The kernel of the projection \rho:\widetilde{\mathcal{L}}_{0}\to\mathcal{L}_{0} is isomorphic to \mathbb{Z}^{\infty}\oplus(\mathbb{Z}/2\mathbb{Z})^{\infty} . Furthermore, let \mathcal{B} denote the set of nontrivial elements in \widetilde{\mathcal{L}}_{0} represented by Brunnian links. Then \widetilde{\mathcal{L}}_{0} contains the subgroup generated by (\ker\rho)\cap\mathcal{B} as a direct summand, which is isomorphic to \mathbb{Z}^{\infty}\oplus(\mathbb{Z}/2\mathbb{Z})^{\infty} . 2. Prime decomposition in \mathcal{L}^{*}_{0} In this section, we prove Theorem 1.3 and Proposition 1.4 . We first introduce several notions and prove lemmas. For marked links L_{0} and L_{1} , a properly embedded surface F in B^{4} is called a \chi -concordance from L_{0} to L_{1} if F satisfies the conditions in Definition 1.1 . Throughout this section, for given marked link L with marked component K , we denote by \check{L} the sublink L\setminus K of L . Then, -L^{*}_{0}\#L_{1} can be regarded as the union of the marked component and (-\check{L_{0}^{*}})\cup\check{L_{1}} . A marked link L is a minimizer of [L]\in\mathcal{L}^{*}_{0} if the number of the components of L is minimal among all representatives of [L] . Let c(L) denote the number of the components of L . Lemma 2.1 . Let L and L^{\prime} be \chi -concordant links. If L is a minimizer of [L]\in\mathcal{L}_{0}^{*} , then there does not exist a \chi -concordance F from L^{\prime} to L such that F contains a connected component bounded by a sublink of \check{L} . Proof. If a \chi -concordance F from L^{\prime} to L contains a connected component F^{\prime} with \partial F^{\prime}\subset\check{L} , then F\setminus F^{\prime} is a \chi -concordance from L^{\prime} to L\setminus\partial F^{\prime} . This contradicts the assumption that L is a minimizer, since L and L^{\prime} are \chi -concordant. ∎ Lemma 2.2 . If \chi -concordant links L and L^{\prime} are minimizers of [L]=[L^{\prime}]\in\mathcal{L}^{*}_{0} , then a \chi -concordance F from L^{\prime} to L induces a marked-concordance from L^{\prime} to L . Proof. It immediately follows from Lemma 2.1 that F consists of one disk and several annuli, each of which connects a component of (-L^{\prime})^{*} to a component of L . Since F is bounded by (-L^{\prime})^{*}\#L , by the definition of the connected sum of marked links, F induces a marked-concordance from L^{\prime} to L in a canonical way. ∎ Lemma 2.3 . Let L_{1} and L_{2} be marked links representing prime elements in \mathcal{L}^{*}_{0} with [L_{1}\#L_{2}]\neq 0 . If both L_{1} and L_{2} are minimizers, then L_{1}\#L_{2} is a minimizer of [L_{1}\#L_{2}] . Proof. Let L be a minimizer of [L_{1}\#L_{2}]\in\mathcal{L}^{*}_{0} . Suppose that L_{1}\#L_{2} is not a minimizer. Then c(\check{L})0 , [\rho(C_{n,1}(W))] is prime and has infinite order in \mathcal{L}_{0} . Moreover, [\rho(C_{2k-1,1}(W))]~(k\in\mathbb{N}) are mutually distinct in \mathcal{L}_{0} . (3) For any n\geq 1 , [\rho(C^{n}(W))]=0 in \mathcal{L}_{0} . To prove Proposition 3.1 , we need the following three lemmas. Here, for given knot K , denote by \sigma(K) the knot signature of K and by \Arf(K) the Arf invariant of K . Lemma 3.2 . Let L be a 2-component marked oriented link and K a knot with \sigma(K)+4\Arf(K)\equiv 4\mod 8 . If there is a planer surface F in S^{3}\times[0,1] from L to K , then the partly oriented link \rho(L) is non-zero in \mathcal{L}_{0} . Proof. Suppose that L=K_{1}\cup K_{2} and K_{1} is the marked component. If [\rho(L)]=0 in \mathcal{L}_{0} , then \rho(L) bounds a disjoint union F^{\prime} of a disk and a Möbius band in B^{4} . By gluing (B^{4},F^{\prime}) and (S^{3}\times[0,1],F) along \rho(L) , we obtain a Möbius band F\cup F^{\prime} in B^{4} with boundary K . However, it is proved in [ 5 ] that any knot K with \sigma(K)+4\Arf(K)\equiv 4(\mod 8) cannot bound a Möbius band in B^{4} , a contradiction. Thus we have [L]\neq 0 in \mathcal{L}_{0} . ∎ Definition 3.3 . Let L=O\cup K_{1}\cup\cdots\cup K_{n} be an (n+1) -component marked oriented link such that the marked component O is an unknot, and \lk(O,K_{i}) is even for any i=1,...,n . Then the double branched cover \Sigma_{2}(O) of S^{3} branched over O is homeomorphic to S^{3} . Hence, the linking number of any pair of lifts K_{i} and K_{j} in \Sigma_{2}(O) takes values in \mathbb{Z} . Since all \lk(K,K_{i})~(i=1,...,n) are even, O bounds an unoriented surface F in S^{3} disjoint from K_{1}\cup\cdots\cup K_{n} . Then \Sigma_{2}(O) decomposes as the union of M_{1} and M_{2} , each of which is a copy of S^{3} cut along F . For each i,~k\in\{1,2\} , let K_{ik} denote the lift of K_{i} contained in M_{k} . In the following we denote \lk(K_{ik},K_{jl}) by \lambda^{F}_{L}(ik,jl) , or simply \lambda_{L}(ik,jl) . We note that \lambda_{L} is symmetric, i.e., \lambda_{L}(ik,jl)=\lambda_{L}(jl,ik) . We remark that if for some i\in\{1,...,n\} the sublink O\cup K_{i} is a split link, then \lambda(i1,i2)=0 . Remark 3.4 . In the case that \lk(K_{i},K_{j})=0 for 1\leq i\leq j\leq m ( \lk(K_{i},K_{j}) is defined to be 0 if i=j ), by [ 2 , Theorem 2.3] , we can calculate \lambda^{F}_{L}(ik,jl) by using a Goeritz matrix [ 3 , 4 ] of F as follows: Let G_{\alpha} be the Goeritz matrix with respect to a basis \alpha=(a_{1},\dots,a_{n}) of H_{1}(F) , i.e., the (i,j) -entry of G_{\alpha} is given by \mathrm{lk}(a_{i},\tau a_{j}) , where \tau a_{j} is the 1-cycle in S^{3}-F obtained by pushing off 2a_{j} in both normal directions. For each K_{i} , define the vector V_{\alpha}(K_{i})=(\mathrm{lk}(K_{i},a_{1}),\dots,\mathrm{lk}(K_{i},a_{n})). Then \lambda^{F}_{L}(ik,jl)=(-1)^{\delta_{kl}}V_{\alpha}(K_{i})\,G_{\alpha}^{-1}\,V_{\alpha}(K_{j})^{T}, where \delta_{kl} is the Kronecker delta. Lemma 3.5 . Let L and L^{\prime} be 2-component marked links such that their marked components are unknots and their linking numbers are even. If [L]=[L^{\prime}] in \widetilde{\mathcal{L}}_{0} , then \lambda_{L}({11},{12})=\lambda_{L^{\prime}}({11},{12}) . Proof. Since both L and L^{\prime} are 2 -component links, they are minimizers of [L]=[L^{\prime}]\in\widetilde{\mathcal{L}}_{0} . Hence Lemma 2.2 shows that there exists a marked-concordance A\cup A_{1} from L to L^{\prime} such that A_{1} connects the unmarked components of L and L^{\prime} . Moreover, there are two lifts {A}_{11} , {A}_{12} of A_{1} in the double branched cover \Sigma_{2}(A) of S^{3}\times[0,1] over A_{1} . Since \Sigma_{2}(A) of S^{3}\times[0,1] is a two punctured rational homology 4-sphere, |\lambda_{L}(11,12)-\lambda_{L^{\prime}}(11,12)|=|A_{11}\cdot A_{12}|=0 , where A_{1}\cdot A_{2} is the intersection number of A_{1} and A_{2} . ∎ Lemma 3.6 . Let L and L^{\prime} be 2-component marked links as in Lemma 3.5 . If [\rho(L)]=[\rho(L^{\prime})]\neq 0 in \mathcal{L}_{0} , then \lambda_{L}({11},{12})=\lambda_{L^{\prime}}({11},{12}) . Proof. Let L=O\cup K_{1} and L^{\prime}=O\cup K^{\prime}_{1} , where O is the marked component of each. Since [L] and [L^{\prime}] are non-zero elements in \mathcal{L}_{0} , L and L^{\prime} are minimizers. Thus, we can adopt the same argument as the proof of Lemma 3.5 , except for that the annulus A_{1} possibly connects K_{1} to -K^{\prime}_{1} . Even for that case, we reaches the same conclusion as the original case since \lambda_{O\cup K^{\prime}_{1}}({11},{12})=\lk(K_{11},K_{12})=\lk(-{K}_{11},-{K}_{12})=\lambda_{O\cup(-K^{\prime}_{1})}({11},{12}). ∎ Proof of Proposition 3.1 . We first prove the assertion (1). Since P_{n}(W) is a 2-component link, [P_{n}(W)]\in\widetilde{\mathcal{L}}_{0} is prime. Thus, by Corollary 2.5 , we only need to show [P_{n}(W)]\neq[-P_{n}(W)^{*}] and [P_{n}(W)]\neq[P_{m}(W)] in \widetilde{\mathcal{L}}_{0} . Let L=O\cup K_{1} denote P_{n}(W) . If [P_{n}(W)]=[-P_{n}(W)^{*}] , then Lemma 3.5 shows \lambda_{L}(11,12)=\lambda_{-L^{*}}(11,12). Since (S^{3},-L^{*})\cong(-S^{3},-L) , this homeomorphism induces \lambda_{-L^{*}}(11,12)=-\lk(-K_{11},-K_{12})~(=-\lambda_{L}(11,12)). Hence we have \lambda_{L}(11,12)=0 . On the other hand, for a surface F with \partial F=O , and for a basis \alpha=(a_{1},a_{2}) illustrated in Figure 3 , we have G_{\alpha}=\begin{pmatrix}0&-1\\ -1&2\end{pmatrix}. Hence, by Remark 3.4 , \lambda_{W}(11,12)=(1\;\;0)G_{\alpha}^{-1}\begin{pmatrix}1\\ 0\end{pmatrix}=(1\;\;0)\begin{pmatrix}-2&-1\\ -1&0\end{pmatrix}\begin{pmatrix}1\\ 0\end{pmatrix}=-2. Furthermore, \lambda_{P_{n}(W)}(11,12)=(n\;\;0)\begin{pmatrix}-2&-1\\ -1&0\end{pmatrix}\begin{pmatrix}n\\ 0\end{pmatrix}=-2n^{2}. This shows [P_{n}(W)]\neq[-P_{n}(W)^{*}] and also implies that if n\neq m then [P_{n}(W)]\neq[P_{m}(W)] . We next prove the assertion (2). Figure 3 shows that there is a planar surface in S^{3}\times[0,1] from C_{n,1}(W) to a twist knot K_{n} , where \sigma(K_{n})=0 and \Arf(K_{n})\equiv 1\mod 2 for any odd n>0 . Moreover, we have w(C_{n,1})=n . Now, it follows from Lemma 3.2 , Lemma 3.6 and arguments in the previous paragraph that the assertion (2) holds. Finally, one can easily check the assertion (3) by constructing a disjoint union of a disk and a Möbius band in B^{4} with boundary \rho(C^{n}(W)) . ∎ Figure 2 . Surfaces F bounded by marked components and a basis a_{1},a_{2} of H_{1}(F) Figure 3 . A planar surface in S^{3}\times[0,1] from C_{n,1}(W) to a twist knot K_{n} Next, let us consider the assertion for B_{n} in Proposition 1.5 . Let B=O\cup K_{1}\cup K_{2} be the Borromean rings with marked component O and P_{n}\subset S^{1}\times D^{2} a pattern with winding number n . Set P_{n}(B):=O\cup P_{n}(K_{1})\cup P_{n}^{*}(K_{2}) , where P_{n}^{*} is the mirror of P_{n} . Then we see C_{n,1}(B)=B_{n} , and the assertion for B_{n} in Proposition 1.5 is a consequence of the following proposition. Proposition 3.7 . The following assertions hold: (1) For any P_{n} , we have 2[P_{n}(B)]=0 in \widetilde{\mathcal{L}}_{0} . (2) For any odd n>0 , [C_{n,1}(B)] and [\rho(C_{n,1}(B))] are prime in \widetilde{\mathcal{L}}_{0} and \mathcal{L}_{0} , respectively. Moreover, [\rho(C_{2k-1,1}(B))]~(k\in\mathbb{N}) are mutually distinct in \mathcal{L}_{0} . (3) For any n\geq 1 , [\rho(C^{n}(B))]=0 in \mathcal{L}_{0} , and [C^{n}(B)] is prime in \widetilde{\mathcal{L}}_{0} . Moreover, [C^{n}(B)]~(n\in\mathbb{N}) are mutually distinct in \widetilde{\mathcal{L}}_{0} . To prove Proposition 3.7 , we first show lemmas for 3-component links. Definition 3.8 . Let L=O\cup K_{1}\cup\cdots\cup K_{n} be an (n+1) -component marked oriented link as in Definition 3.3 , i.e., the marked component O is an unknot and \lk(O,K_{i}) is even for any i=1,...,n . Furthermore, suppose that \lk(K_{i},K_{j})=0 for some i\neq j . Then by Remark 3.4 , we have \lambda(i1,j1)=\lambda(i2,j2)=-\lambda(i1,j2)=-\lambda(i2,j1) . Hence we define \bar{\lambda}(i,j) as \bar{\lambda}(i,j):=|\lambda(i1,j1)|. In the following lemmas, we consider Brunnian links. We stress that Brunnian links satisfy the condition in the definition above. Lemma 3.9 . If L is Brunnian and [L]=0 in \widetilde{\mathcal{L}}_{0} , then \bar{\lambda}_{L}(1,2)=0. Proof. Since L is Brunnian, for each i , O\cup K_{i} is a trivial link, and hence \lambda(i1,i2)=0 . If [L]=0 in \widetilde{\mathcal{L}}_{0} , then O and K_{1}\cup K_{2} bounds a disk D and annulus A in B^{4} respectively, which are mutually disjoint. Moreover, there are two lifts {A}_{1} , {A}_{2} of A in the double branched cover \Sigma_{2}(D) of B^{4} branched over D . Note that \Sigma_{2}(D) is a rational homology 4-ball, and hence \lk(\partial{A}_{1},\partial{A}_{2})={A}_{1}\cdot{A}_{2}=0 . Since (\partial A_{1},\partial A_{2})=(K_{11}\cup K_{21},K_{12}\cup K_{22}) or (K_{11}\cup K_{22},K_{12}\cup K_{21}) , we have \mathrm{lk}(\partial A_{1},\partial A_{2})=\lambda_{L}(11,21)+\lambda_{L}(12,22)\quad\text{or}\quad=\lambda_{L}(11,22)+\lambda_{L}(12,21). As we mentioned in Definition 3.8 , \left|\lambda_{L}(11,21)+\lambda_{L}(12,22)\right|=\left|\lambda_{L}(11,22)+\lambda_{L}(12,21)\right|=2\bar{\lambda}_{L}(1,2). This completes the proof. ∎ Lemma 3.10 . If L is Brunnian and [\rho(L)]=0 in \mathcal{L}_{0} , then \bar{\lambda}(1,2) is even. Proof. If [\rho(L)]=0 in \mathcal{L}_{0} , then at least one of the following holds: • L bounds a disjoint union of a disk D and an annulus A , or • L bounds a disjoint union of a disk D and two Möbius bands M_{1} , M_{2} . For the first case, we can adopt the same argument as the proof of Lemma 3.9 , except for that the boundary of the annulus A is possibly K_{1}\cup-K_{2} . Even for the case, we have 2\bar{\lambda}(1,2)=|\lambda_{L}(11,21)+\lambda_{L}(12,22)|=|\lambda_{L}(11,22)+\lambda_{L}(12,21)|=0. Next, for the second case, there are two lifts {M}_{i1} , {M}_{i2} of M_{i} in \Sigma_{2}(D) for each i\in\{1,2\} . Without loss of generality, we may assume that \partial{M}_{ik}=K_{ik} . Now, since \Sigma_{2}(D) is a rational homology 4-ball, we have \bar{\lambda}(1,2)=|\lk(K_{11},K_{21})|=|\lk(\partial{M}_{11},\partial{M}_{21})|\equiv|{M}_{11}\cdot{M}_{21}|=0\mod 2. ∎ The following two lemmas can be proved similarly to Lemmas 3.5 and 3.6 respectively. Lemma 3.11 . If L and L^{\prime} are Brunnian and [L]=[L^{\prime}]\neq 0 in \widetilde{\mathcal{L}}_{0} , then \bar{\lambda}_{L}(1,2)=\bar{\lambda}_{L^{\prime}}(1,2) . Lemma 3.12 . If L and L^{\prime} are Brunnian and [\rho(L)]=[\rho(L^{\prime})]\neq 0 in \mathcal{L}_{0} , then \bar{\lambda}_{L}(1,2)=\bar{\lambda}_{L^{\prime}}(1,2) . Proof of Proposition 3.7 . Figure 4 describes an isotopy from B to -B^{*} which sends the triple (O,K_{1},K_{2}) to (-O^{*},-K^{*}_{2},-K^{*}_{1}) . By taking satellite operations, we also have an isotopy from P_{n}(B) to -P_{n}(B)^{*} , which proves the assertion (1). To prove the assertions (2) and (3), let us compute \bar{\lambda}(1,2) for any P_{n}(B) . For a surface F with \partial F=O , and for a basis \alpha=(a_{1},a_{2}) illustrated in Figure 3 , we have G_{\alpha}=\begin{pmatrix}0&-1\\ -1&0\end{pmatrix}. Hence, by Remark 3.4 , \bar{\lambda}_{B}(1,2)=|\lambda_{B}(11,21)|=\left|(1\;\;0)G_{\alpha}^{-1}\begin{pmatrix}0\\ -1\end{pmatrix}\right|=\left|(1\;\;0)\begin{pmatrix}0&-1\\ -1&0\end{pmatrix}\begin{pmatrix}0\\ -1\end{pmatrix}\right|=1. Furthermore, \bar{\lambda}_{P_{n}(B)}(1,2)=|\lambda_{P_{n}(B)}(11,21)|=\left|(n\;\;0)\begin{pmatrix}0&-1\\ -1&0\end{pmatrix}\begin{pmatrix}0\\ -n\end{pmatrix}\right|=n^{2}. Now, since w(C_{n,1})=n and w(C^{n})=2^{n} , the assertions (2) and (3) directly follows from Lemmas 3.9 , 3.10 , 3.11 and 3.12 . (Note that the primeness follows from Proposition 1.4 and the fact that C_{n,1}(B) and C^{n}(B) are Brunnian for any n .) ∎ Figure 4 . An isotopy from B to -B^{*} Proof of Proposition 1.5 . Since W_{n}=C_{n,1}(W) , B_{n}=C_{n,1}(B) and w(C_{n,1})=n , Proposition 1.5 immediately follows from Propositions 3.1 and 3.7 . ∎ Proof of Theorem 1.2 . This immediately follows from Theorem 2.6 and Proposition 1.5 . ∎ Proof of Theorem 1.7 . By Theorem 2.6 , we have a direct decomposition \Ker\rho=(\Ker\rho\cap\widetilde{\mathcal{L}}_{\infty})\oplus(\Ker\rho\cap\widetilde{\mathcal{L}}_{2}) such that • \Ker\rho\cap\widetilde{\mathcal{L}}_{\infty} is an free abelian group containing \{C^{n}(W)~|~n\in\mathbb{N}\} , and • \Ker\rho\cap\widetilde{\mathcal{L}}_{2} is an \mathbb{Z}/2\mathbb{Z} -vector space containing \{C^{n}(B)~|~n\in\mathbb{N}\} . Since \{C^{n}(W)~|~n\in\mathbb{N}\} (resp. \{C^{n}(B)~|~n\in\mathbb{N}\} ) is a linearly independent set of \Ker\rho\cap\widetilde{\mathcal{L}}_{\infty} (resp. \Ker\rho\cap\widetilde{\mathcal{L}}_{2} ), the equality \Ker\rho\cong\mathbb{Z}^{\infty}\oplus(\mathbb{Z}/2\mathbb{Z})^{\infty} holds. Next, denote by \widetilde{\mathcal{L}}_{B} the subgroup of \widetilde{\mathcal{L}}_{0} generated by \Ker\rho\cap\mathcal{B} , and let us prove that \widetilde{\mathcal{L}}_{B} is a direct summand of \widetilde{\mathcal{L}}_{0} isomorphic to \mathbb{Z}^{\infty}\oplus(\mathbb{Z}/2\mathbb{Z})^{\infty} . It follows from Proposition 1.4 that \mathcal{B} is contained in the set of prime elements in \widetilde{\mathcal{L}} . Therefore, by Theorem 2.6 , we see that \widetilde{\mathcal{L}}_{B}\cap\widetilde{\mathcal{L}}_{\infty} (resp. \widetilde{\mathcal{L}}_{B}\cap\widetilde{\mathcal{L}}_{2} ) is a direct summand of \widetilde{\mathcal{L}}_{\infty} (resp. \widetilde{\mathcal{L}}_{2} ) with a basis \Ker\rho\cap\mathcal{B}\cap\widetilde{\mathcal{P}}_{\infty} (resp. \Ker\rho\cap\mathcal{B}\cap\widetilde{\mathcal{P}}_{2} ). Moreover, we note that • \Ker\rho\cap\mathcal{B}\cap\widetilde{\mathcal{P}}_{\infty} contains \{\varepsilon(n)C^{n}(W)~|~n\in\mathbb{N}\} for some \varepsilon\colon\mathbb{N}\to\{\pm 1\} , and • \Ker\rho\cap\mathcal{B}\cap\widetilde{\mathcal{P}}_{2} contains \{C^{n}(B)~|~n\in\mathbb{N}\} . As a conclusion, we can say that \widetilde{\mathcal{L}}_{B} is a direct summand of \widetilde{\mathcal{L}}_{0} and \widetilde{\mathcal{L}}_{B}=(\widetilde{\mathcal{L}}_{B}\cap\widetilde{\mathcal{L}}_{\infty})\oplus(\widetilde{\mathcal{L}}_{B}\cap\widetilde{\mathcal{L}}_{2})\cong\mathbb{Z}^{\infty}\oplus(\mathbb{Z}/2\mathbb{Z})^{\infty}. ∎