Independence of homogeneous GKM manifolds and symmetric spaces Shintarô Kuroki, Grigory Solomadin Abstract . Let G / H G/H be a simply connected homogeneous space of maximal rank. Then the maximal torus T T -action on G / H G/H is a GKM manifold. We call the T T -action j j -independent if any i ( ≤ j ) i(\leq j) pairwise distinct isotropy weights at a fixed point are linearly independent. Using weighted graphs, we show that the maximal independence of G / H G/H is 2 2 , 3 3 or n = dim T n=\dim T , and that the cases of 3 3 or n = dim T n=\dim T correspond to some symmetric spaces of rank > 2 >2 . As a corollary, using the results of Ayzenberg and Masuda, the lower-degree reduced homology groups (with appropriate coefficients) of the orbit space T \ G / H T\backslash G/H vanish. 2020 Mathematics Subject Classification Primary: 57S12, 17B22. Secondary: 55P91, 57R91 1. Introduction 1.1. Motivation and Main theorem Let M be a 2m -dimensional, smooth T -manifold with a nonempty set of isolated fixed points M^{T} , where T\cong(S^{1})^{n} and we assume that the T -action is almost effective , i.e., the kernel of the T -action is finite. There is the following irreducible decomposition (1.1) \displaystyle T_{p}M\cong\bigoplus_{i=1}^{m}V_{\alpha_{p,i}}, of the isotropy representation at any fixed point p\in M^{T} , where V_{\alpha_{p,i}} is the (complex) irreducible representation space of T , and \alpha_{p,i}\in\mathfrak{t}_{\mathbb{Z}}^{*}/\pm 1 for i=1,\ldots,m is determined up to a sign. Here, \mathfrak{t}_{\mathbb{Z}}^{*}\cong{\mathbb{Z}}^{n} denotes the character lattice of the dual of the Lie algebra \mathfrak{t}^{*} of T . Set S_{p}:=\{\alpha_{p,i}\ |\ i=1,\ldots,m\}\subset\mathbb{Z}^{n} and define the following numbers: (1.2) \displaystyle k(p):=\max\{q\ |\ \text{every $q$ elements in $S_{p}$ are linearly independent}\},\ k(M):=\min\limits_{p\in M^{T}}k(p). If 2\leq k(M) , then this manifold is called a GKM manifold [ GKM98 ] , [ GZ01 ] . A GKM manifold is called a j -independent or GKM j ) manifold if k(M)\geq j holds (see [ AMS23 , GS ] . In [ AM23 ] , this is called a j -generality ). If k(M)=m then n=m ; in such case this manifold is called a torus manifold [ MP06 ] , [ HM03 ] . Mikiya Masuda has asked the following question to the first named author: Problem 1 . Let M be an equivariantly formal (i.e., H^{odd}(M)=0 ) 2m -dimensional GKM 4 manifold with a T -action. Is M homeomorphic to a torus manifold, and is the given T -action obtained as the restriction of a half-dimensional torus T^{m} -action to a subtorus T ? The main result of the present paper is given as follows (see Theorem 3.3 ), where by a maximal rank compact symmetric space we mean a compact symmetric space G/H that is a homogeneous space of maximal rank (also referred to as a class 1 symmetric space of inner type ). Theorem 2 . Let G/H be any simply connected, homogeneous space such that T\subset H\subset G is a maximal torus, {\rm rank}\ T=n , and G is a compact, connected, simple Lie group. Then one has k(G/H)=2,3\ \text{or}\ n(=\frac{1}{2}\dim G/H). Furthermore, k(G/H)=3 or n if and only if G/H is a maximal rank compact symmetric space with rank >2 other than B_{n}/D_{i}\times B_{n-i} , C_{n}/A_{n-1}\times T^{1} , F_{4}/C_{3}\times A_{1} . Note that by the classification of the homogeneous torus manifolds in [ Ku09 ] , k(G/H)=n if and only if G/H is equivariantly diffeomorphic to A_{n}/A_{n-1}\times T^{1}(\cong\mathbb{C}P^{n}) or B_{n}/D_{n}(\cong S^{2n}) . Moreover, the independence of M_{1}\times\cdots\times M_{r} coincides with the minimal independence of M_{i} ’s (if it is not a torus manifold, see Proposition 3.1 ), and the independence of any connected homogeneous space G/H is equal to that of its universal covering space G_{1}/H_{1}\times\cdots\times G_{r}/H_{r} . Therefore, together with these facts, Theorem 2 answers Problem 1 positively for the class of homogeneous GKM manifolds. As an application of Theorem 2 , together with the acyclicity results of [ AMS23 ] , [ AM23 ] , we obtain vanishing (in degrees \leq 4 ) for reduced homology with an appropriate coefficients of the orbit spaces for homogeneous GKM manifolds, see Corollary 3.9 . Our results are consistent with those in [ Su19 ] , [ Su21 ] , [ BT19 ] , while providing new insights, particularly for several symmetric spaces. 1.2. Sketch of the proof The proof of Theorem 2 follows by computing k(G/H) for every homogeneous GKM manifold G/H . Here, k(G/H)=k(gH) for every fixed point gH\in(G/H)^{T} (see [ GHZ06 ] ). We prove that k(G/H)\geq 3 implies that G/H is a maximal rank compact symmetric space with rank >2 using the well known characterization of symmetric spaces in terms of the Lie algebra T_{eH}G/H (see Lemma 3.6 ). For maximal rank compact symmetric space G/H such that G is a classical Lie group (i.e., type A – D ), the problem of finding maximal independence for G/H reduces to the study of the matroid formed by the complement \Delta^{+}_{G,H}:=\Delta^{+}_{G}\setminus\Delta^{+}_{H} of positive root systems. The analysis of circuits for such a matroid is possible using weighted signed graphs , which were introduced in [ Ha53 ] and studied in [ Za82 ] (also see [ FGLP00 ] and [ FGGP07 ] ). We apply this technique together with Borel-de Siebenthal theory [ BS49 ] and Cartan’s classification of symmetric spaces [ Ca27 ] to compute k(G/H) for all pairs (G,H) except for G being of type F_{4},E_{6},E_{7},E_{8} . (We remark that signed graphs under the name of Levi graphs or crystallographs were also used in [ Re24 ] to study root subsystems. The difference of our application for signed graphs is that we study complements of root subsystems.) In the remaining cases, i.e. for G being of type F_{4},E_{6},E_{7},E_{8} , the proof is a simple case study. 1.3. Structure of the paper We briefly outline the structure of the present paper. In Section 2 , we recall the isotropy representation for homogeneous GKM manifolds, Borel-de Siebenthal theory, Cartan’s classification of symmetric spaces and homology vanishing of the orbit space for independent GKM manifolds. One of the main contributions of this paper is to clarify the relationship between symmetric spaces and GKM manifolds from the viewpoint of independence. The value of k(G/H) for each homogeneous space G/H is listed in Theorem 3.3 . This result implies \widetilde{H}_{i}(T\backslash G/H;\Bbbk)=0 for all i\leq k(G/H)+1 , where \Bbbk is additionally specified (see Corollary 3.9 for symmetric GKM 3 cases). This homology vanishing is compared with several previously known results in Examples 3.11 and 3.12 . The remainder of the paper is the proof of Theorem 3.3 . In Section 4 , linearly dependent triples are described in several cases in terms of weighted signed graphs (see Lemma 4.2 ). In Sections 5 – 7 , we analyse linear dependencies for root subsystems of classical types, using Borel-de Siebenthal theory and weighted signed graphs. In Sections 8 – 9 , the remaining cases (when G is of type F_{4},E_{6},E_{7},E_{8} ) are studied. Acknowledgements: The authors gratefully acknowledge funding of the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation): Project number 561158824 (the second named author’s Walter Benjamin Fellowship). We thank M. Masuda for many valuable discussions on toric manifolds without whom we would not be able to conduct this project, and to O. Goertsches for the interest to our paper. We also thank M. Nakagawa for giving us a useful information on symmetric spaces. The first author was partially supported by JSPS KAKENHI Grant Number 21K03262. He would like to thank the Philipps University of Marburg for providing him with an excellent environment for research during his stay in August 2025. The second named author is grateful to Okayama University of Science and Osaka Metropolitan University for a very comfortable work atmosphere, where the initial part of the present work was performed during his research visits in 2023. 2. Preparations We first prepare the basic notions and facts to prove our main theorem. 2.1. k -independence Let S:=\{\alpha_{1},\ldots,\alpha_{m}\}\subset\mathbb{Z}^{n}\subset\mathbb{R}^{n} be a set of vectors. We call S a k -independent vectors for some 2\leq k\leq n if every k vectors \alpha_{i_{1}},\ldots,\alpha_{i_{k}}\in S are linearly independent but there exists k+1 vectors \alpha_{j_{1}},\ldots,\alpha_{j_{k+1}}\in S which are not linearly independent. In this case, we also say that the set of vectors S has the property U_{m}^{k} , i.e., the uniform matroid with rank k and with m elements, see [ Ox06 ] . Example 2.1 . Let S:=\{e_{1},e_{2},-e_{1}-e_{2},e_{1}+e_{2}+e_{3}\}\subset\mathbb{R}^{3} be the set of 4 vectors in the 3 -dimensional real vector space \mathbb{R}^{3} , where e_{i} , i=1,2,3 , is the standard basis of \mathbb{R}^{3} . Then, one can easily check that S has the property U_{4}^{2} , i.e., S is 2 -independent. Note that the subset \{e_{1},e_{2},e_{1}+e_{2}+e_{3}\} is linearly independent but \{e_{1},e_{2},-e_{1}-e_{2}\} is not linearly independent. 2.2. Some remarks on k -independence of tangential representations Let M be a 2m -dimensional manifold with an n -dimensional torus T -action. We assume that there is a non-empty fixed point set M^{T} , and the tangential representation at p\in M^{T} is a direct sum of pairwise linearly independent characters of T . Namely, for every p\in M^{T} , the irreducible decomposition (with respect to some complex structure) ( 1.1 ) satisfies that \alpha_{i},\alpha_{j} for every 1\leq i1 embedding of free abelian subgroups (see [ MT79 , Chapter 5 Theorem 4.9] or [ Bu13 , Section 19] ). (Here we identify the map with the induced map for the de Rham cohomology groups H^{1}(T;{\mathbb{R}})\to H^{1}(\widetilde{T};{\mathbb{R}}) , restricted to the respective lattices.) Let \Delta_{G}\subset\Lambda_{G}^{*}\subset\mathfrak{t}^{*} be the root system of G with respect to the maximal torus T . Because the linear relations among the root systems do not change by taking the finite covering of G , that is, one has k(\widetilde{G}/H)=k(G/H) , it is enough to consider the case when G is simply connected for the purpose of this paper. Therefore, we can assume \pi_{1}(G)=0 in this paper without loss of generality. Let G be a simply connected, compact Lie group. The set of the fixed points (G/H)^{T} can be identified with the finite set of the quotient W_{G}/W_{H} of the Weyl groups for the maximal torus T , where W_{G}:=N_{G}(T)/T and W_{H}:=N_{H}(T)/T . Let p_{0}=eH\in(G/H)^{T}\simeq W_{G}/W_{H} be the coset of the identity element e\in G . In other words, W_{G} acts on (G/H)^{T} transitively, where the isotropy subgroup of p_{0} is W_{H} . Then, the tangential representation at wp_{0} is \displaystyle T_{wp_{0}}G/H=\bigoplus_{[\beta]\in\Delta_{G,H}/\{\pm 1\}}V_{[w\beta]}, where \Delta_{G}^{+}\setminus\Delta_{H}^{+}\simeq\Delta_{G,H}/\{\pm 1\} for positive roots \Delta_{H}^{+}\subset\Delta_{G}^{+} . Note that \Delta_{p_{0}} and \Delta_{wp_{0}} for every w\in W_{G}/W_{H} are isomorphic to \Delta_{G,H}/\{\pm 1\} by being in the orbit of the Weyl group transitive action. This implies the following proposition: Proposition 2.4 . Let G be a compact, connected semi-simple Lie group, H be its maximal rank subgroup and T\subseteq H\subset G be a maximal torus. Then k(G/H)=k(gH) for any gH\in(G/H)^{T} . This proposition shows that if we compute the k(G/H) -independence on \Delta_{G}^{+}\setminus\Delta_{H}^{+}\simeq\Delta_{G,H}/\{\pm 1\} , then we have the k(G/H) -independence of all tangent spaces on the fixed points of G/H (Cf. Example 2.2 ). In this paper, we determine the k -independence for all maximal rank homogeneous spaces G/H . 2.4. The Borel-de Siebenthal theory Let G be a compact, connected semi-simple Lie group, H be its maximal rank subgroup. The classification of such pair (G,H) is known by the work of Borel-de Siebenthal [ BS49 ] . In this subsection, we briefly recall the Borel-de Siebenthal theory. Borel-de Siebenthal theory [ BS49 ] describes maximal subgroups of maximal rank in a simple Lie group G as normalizers of certain elements or circle subroups in G . The classification of the respective root subsystems is given in terms of (ordinary or extended) Dynkin diagram of G as follows. (Below we follow notation from [ MT79 ] and tables from [ Yo25 ] .) Recall that \alpha_{1},\dots,\alpha_{n}\in\mathfrak{g}^{*} are simple roots of G , and the dominant root \widetilde{\alpha} can be written as (2.1) \displaystyle\widetilde{\alpha}=\sum_{i=1}^{n}m_{i}\alpha_{i} for some non-negative integers m_{1},\ldots,m_{n}(\geq 0) . We set the additional notation by denoting by X_{i} the vertex opposed to the facet with normal vector \alpha_{i} of the simplex S(G):=\{X\in\mathfrak{g}\mid\alpha_{i}(X)\geq 0,\ i=1,\dots,n,\ \widetilde{\alpha}(X)\leq 1\}\subset\mathfrak{g}. Also let x_{i}:=\exp(X_{i})\in G , and let N(x) be the normalizer of x\in G . Theorem 2.5 ( [ MT79 , Thm 7.16] ) . The Dynkin diagram of a maximal subgroup of maximal rank for a simple Lie group G is obtained from that of G as follows: (1) If m_{i} in ( 2.1 ) is prime, then the Dynkin diagram of G_{i}=N(x_{i}) is obtained by deleting \alpha_{i} from the extended Dynkin diagram, which is the Dynkin diagram of G with -\widetilde{\alpha} added; (2) If m_{i}=1 in ( 2.1 ), then N(\exp X_{i}/2) is locally a direct product of S^{1} and a semisimple subgroup H of corank 1 the Dynkin diagram of H is obtained by deleting \alpha_{i} from the Dynkin diagram of G . Remark 2.6 . By using the Borel-de Siebenthal theory, on the level of Dynkin diagrams, a maximal rank subgroup H(\subset G) is determined by a sequence of certain added or removed vertices. The isomorphism class of a subgroup does not depend on the choice of such a sequence, though the embedding depends on the the order of such operations. However, it turns out that the choice of embedding does not affect independence. 2.5. Symmetric spaces In order to state our main theorem, we also need the notion of symmetric spaces. Here, we quickly recall it. Definition 2.7 ( [ MT79 , Chapter III, Section 6] ) . A homogeneous space G/H admitting an involution \sigma\in\Aut G is called a symmetric space , namely, G^{\sigma}_{e}\subseteq H\subset G^{\sigma}=\{g\in G\mid\sigma g=g\} holds, where G^{\sigma}_{e} is the identity component of G^{\sigma} . There is a well-known criterion for G/H to be a symmetric space in terms of Lie algebras, which we recall next. Let \mathfrak{g} be a Lie algebra of a simply connected, compact, simple Lie group G . It has a decomposition into the Cartan subalgebra C(\mathfrak{g}) of \mathfrak{g} and characters \mathfrak{g}_{\alpha} of C(\mathfrak{g}) as follows: (2.2) \mathfrak{g}=C(\mathfrak{g})\oplus\bigoplus\limits_{\alpha\in\Delta^{+}_{G}}\mathfrak{g}_{\alpha}. Any Lie subalgebra \mathfrak{h} in \mathfrak{g} is the direct sub-summand of ( 2.2 ) given by the embedded root system \Delta_{H}\subseteq\Delta_{G} . Since \mathfrak{h} is a Lie subalgebra, one has [\mathfrak{h},\mathfrak{h}]\subset\mathfrak{h}. Since \mathfrak{g} is semi-simple (and therefore reductive Lie algebra), the Killing form on \mathfrak{g} is nondegenerate, and the respective orthogonal complement to \mathfrak{h} in \mathfrak{g} , say \mathfrak{m} , is \mathfrak{h} -invariant. Therefore, one has [\mathfrak{h},\mathfrak{m}]\subset\mathfrak{m}. Recall the following proposition. Proposition 2.8 ( [ Ba06 , §B.1] ) . Let \mathfrak{h} be any Lie subalgebra \mathfrak{g} of any semisimple Lie algebra. Then the following are equivalent: (1) [\mathfrak{m},\mathfrak{m}]\subset\mathfrak{h} , where \mathfrak{m} is the orthogonal complement to \mathfrak{h} with respect to the Killing form; (2) The Lie algebra \mathfrak{g} has an involution \sigma to which \mathfrak{h} is the eigenspace with value 1 and \mathfrak{m} is the eigenspace with value -1 in the direct sum \mathfrak{h}\oplus\mathfrak{m} ; (3) The homogeneous space G/H is a symmetric space, where G is a Lie group and H is its closed subgroup such that the Lie algebras of G and H are \mathfrak{g} and \mathfrak{h} , respectively. The classification of symmetric spaces was obtained by E. Cartan [ Ca27 ] . In this paper, we restrict our attention to the symmetric spaces that are compact and have nonzero Euler characteristic with {\rm rank}(G)={\rm rank}(H)>2 , i.e., maximal rank compact symmetric spaces with rank >2 . Thus, by Cartan’s classification, we may only consider the following symmetric spaces (see e.g. [ He73 , Ch.X, §6, Table V] for classical types, and [ MT79 , Ch.V, §7, Remark 7.18–7.19] and [ Yo25 ] for exceptional types). Theorem 2.9 . Let G/H be a maximal rank compact symmetric spaces with rank >2 . Then, the pair (G,H) corresponds to one of the following list. label pair G H AIII (A_{n},A_{i}\times A_{n-1-i}\times T^{1}) SU(n+1) S(U(i+1)\times U(n-i)) 0\leq i\leq n-1 BDI (B_{n},D_{i}\times B_{n-i}) SO(2n+1) SO(2i)\times SO(2n-2i+1) 1\leq i\leq n (D_{n},D_{i}\times D_{n-i}) SO(2n) SO(2i)\times SO(2n-2i) 1\leq i\leq n-1 DIII (D_{n},A_{n-1}\times T^{1}) SO(2n) U(n) CI (C_{n},A_{n-1}\times T^{1}) Sp(n) U(n) CII (C_{n},C_{i}\times C_{n-i}) Sp(n) Sp(i)\times Sp(n-i) 1\leq i\leq n-1 EII (E_{6},A_{5}\times A_{1}) E_{6} (SU(6)\times SU(2))/\mathbb{Z}_{2} EIII (E_{6},D_{5}\times T^{1}) E_{6} (Spin(10)\times T^{1})/\mathbb{Z}_{4} EV (E_{7},A_{7}) E_{7} SU(8)/\mathbb{Z}_{2} EVI (E_{7},D_{6}\times A_{1}) E_{7} (Spin(12)\times SU(2))/\mathbb{Z}_{2} EVII (E_{7},E_{6}\times T^{1}) E_{7} (E_{6}\times T^{1})/\mathbb{Z}_{3} EVIII (E_{8},D_{8}) E_{8} Ss(16) EIX (E_{8},E_{7}\times A_{1}) E_{8} (E_{7}\times SU(2))/\mathbb{Z}_{2} FI (F_{4},C_{3}\times A_{1}) F_{4} (Sp(3)\times Sp(1))/\mathbb{Z}_{2} FII (F_{4},B_{4}) F_{4} Spin(9) Here, n\geq 3 , Ss(16)\simeq Spin(16)/\mathbb{Z}_{2} is the semi-spinor group (see [ IT77 , Theorem 2.1] ), and we identify A_{1}\simeq B_{1}\simeq C_{1} , D_{1}\simeq T^{1} , B_{2}\simeq C_{2} , D_{2}\simeq A_{1}\times A_{1} and D_{3}\simeq A_{3} . We will often use the list in Theorem 2.9 , in particular, to state the main theorem, Theorem 3.3 . Remark 2.10 . It is important for our computation that any symmetric space G/H from the list in Theorem 2.9 satisfies the property that H is a maximal proper connected closed subgroup in G . The respective root system embedding can be described by the Borel-de Siebenthal classification [ BS49 ] . One can see from it that a homogeneous space G/H (of maximal rank) is a symmetric space of type 1 in the classification list by Cartan with inner involution if and only if the respective maximal subgroup H in G corresponds to removing a vertex in the (extended) Dynkin diagram of G with multiplicity m_{i}=1 or 2 (see Theorem 2.5 ). 3. Main theorem and its application In this section, we state our main theorem. We also give its proof (modulo root system independence computations in later sections) and an application to the topology of the orbit spaces. 3.1. Main theorem If we assume G is simply connected, then the standard T -action on the simply connected homogeneous space G/H is equivariantly diffeomorphic to the following product of the homogeneous spaces (see e.g. [ Ku10 , Section 2.2] ): (3.1) \displaystyle G/H\cong G_{1}/H_{1}\times\cdots\times G_{r}/H_{r}, where G_{i} is a compact, simply connected simple Lie group, H_{i} be its maximal rank subgroup and there is an isomorphism T\cong T_{1}\times\cdots\times T_{r} such that T_{i}\subset H_{i}\subset G_{i} is a maximal torus for i=1,\ldots,r . Moreover, we can easily show the following proposition. (Recall that a torus manifold is equipped with a half-dimensional torus action.) Proposition 3.1 . Assume that G/H is not a torus manifold. Then, there is the following equality for the splitting ( 3.1 ): \displaystyle k(G/H)=\min\{k(G_{i}/H_{i})\ |\ 1\leq i\leq r\}. Remark 3.2 . For the case of a product for torus manifolds, the independencies equate half dimensions of the respective manifolds. Therefore, one has to replace minimum with the sum in the right hand side of the above formula. Proposition 3.1 shows that to compute the independence of G/H , it is enough to compute the independence of the simple factors G_{i}/H_{i} . Any connected subgroup of G fits into a saturated chain of maximal subgroups, where at every step the respective embedding is that of maximal proper subgroup. The main theorem of this paper is as follows. Theorem 3.3 . Assume that G is simply connected, simple Lie group and H be its maximal rank subgroup such that {\rm rank}(G)={\rm rank}(H)=\dim T=n . Then, k(G/H)=2,3,n and the following holds: (1) k(G/H)=n={\rm rank}(G)={\rm rank}(H) if and only if (G,H) is one of the following types: Type A_{n} : (A_{n},A_{n-1}\times T^{1}) ; Type B_{n} : (B_{n},D_{n}) . (2) k(G/H)=3(\not=n) if and only if (G,H) is one of the pairs appered in Theorem 2.9 except for the above two cases, (B_{n},D_{i}\times B_{n-i}) , (C_{n},A_{n-1}\times T^{1}) and (F_{4},C_{3}\times A_{1}) . (3) k(G/H)=2 , otherwise (all of the other cases). Remark 3.4 . Using the Borel-Hirzebruch result [ BH58 ] , we have that there are exactly two Sp(n) -invariant almost complex structures on G/H=Sp(n)/Sp(n-1)\times T^{1}(\cong\mathbb{C}P^{2n-1}) . Since one of them is the standard complex structure on \mathbb{C}P^{2n-1} , the T^{n} -action on Sp(n)/Sp(n-1)\times T^{1} extends to the T^{2n-1} -action. Namely, Sp(n)/Sp(n-1)\times T^{1} is isomorphic to SU(2n)/S(U(2n-1)\times U(1)) ; therefore, this is a torus manifold. Note that, by Theorem 3.3 , the homogeneous GKM manifold of type (C_{n},C_{n}\times T^{1}) , i.e., G/H=Sp(n)/Sp(n-1)\times T^{1} with the T^{n} -action, satisfies k(G/H)=2 . This implies that a torus manifold structure on a given homogeneous manifold G/H may exist when k(G/H)\leq 3 . However, among the spaces G/H , since there is no case with 4\leq k(G/H)\leq n-1 by Theorem 3.3 , Problem 1 is affirmatively solved for the maximal rank compact homogeneous spaces G/H . Remark 3.5 . We also note that, by [ Ku09 ] , the homogeneous space G/H can be a torus manifold only if it is isomorphic to A_{n}/A_{n-1}\times T^{1} or B_{n}/D_{n} . Therefore, it is easy to check that if k(G/H)\leq 3 and G/H is a torus manifold, then its GKM graph must coincide with that of A_{n}/A_{n-1}\times T^{1} or B_{n}/D_{n} equipped with the non-standard connection (see [ GZ01 , Ku19 ] for the notion of connection). If k(G/H)=3 , then the connection is unique by [ GZ01 ] . Moreover, by using [ Ku19 ] , if such a GKM graph extends to that of A_{n}/A_{n-1}\times T^{1} or B_{n}/D_{n} , then the connection must be the standard one. Consequently, if n\geq 4 , there exists no G/H with k(G/H)=3 that is isomorphic to a torus manifold. We also remark that, in the case G/H=Sp(n)/Sp(n-1)\times T^{1} considered in Remark 3.4 , the standard complex structure corresponds to the standard connection on its GKM graph, while the non-standard almost complex structure corresponds to the non-standard connection. This phenomenon can occur precisely because k(G/H)=2 . 3.2. Proof of the main theorem Here, we prove Theorem 3.3 , assuming several facts that will be established in later sections. The following key lemma significantly reduces the number of cases we need to check. Lemma 3.6 . Under the hypothesis in Theorem 3.3 , if k(G/H)\geq 3 , then G/H is a symmetric space given in Theorem 2.9 . Proof. Suppose that G/H is not a symmetric space, i.e., [\mathfrak{m},\mathfrak{m}]\not\subset\mathfrak{h} by Proposition 2.8 . Then there exist roots a,b\in\Delta^{+}_{G,H} such that a+b is also in \Delta^{+}_{G,H} , by the property (3.2) [\mathfrak{g}_{a},\mathfrak{g}_{b}]=\mathfrak{g}_{a+b}, of a Lie algebra (see [ He73 , Ch III, Theorem 4.3 (iv)] ). Therefore, \Delta^{+}_{G,H} is not 3 -independent, i.e. k(G/H)=2 . This shows that if k(G/H)\geq 3 , then G/H is a symmetric space. ∎ Now we may prove Theorem 3.3 . Proof of Theorem 3.3 . Using Lemma 3.6 , if G/H is not a symmetric space in Theorem 2.9 , then k(G/H)=2 . When G/H is a symmetric space in Theorem 2.9 , in Sections 5 – 7 we find k(G/H) in all cases except for G of type F_{4},E_{6},E_{7},E_{8} using weighted signed graphs. In Sections 8 and 9 , we find k(G/H) for G of type F_{4},E_{6},E_{7},E_{8} by another method. This finishes the proof. ∎ As a corollary of Theorem 3.3 , we have the following description of GKM 4 homogeneous spaces by using the independence of tangential weights (also see the classification of the homogeneous tours manifolds in [ Ku10 ] ). This gives the partial answer to Problem 1 . Corollary 3.7 . If G/H is a simply connected homogeneous GKM manifold with k(G/H)\geq 4 , i.e., GKM 4 homogeneous space, then G/H is a torus manifold which is equivariantly diffeomorphic to a product of \mathbb{C}P^{n}=SU(n+1)/S(U(n)\times U(1)) or S^{2n}=SO(2n+1)/SO(2n) . 3.3. An application to the topology of orbit spaces of homogeneous spaces On the other hand, Ayzenberg and Masuda in [ AM23 ] studied topological properties of GKM k manifolds and proved the following theorem concerning the vanishing of homology of their orbit spaces. Theorem 3.8 . [ AM23 , Theorem 2] Let \Bbbk be a PID. If M is a GKM k manifold with T -action satisfying H^{odd}(M;\Bbbk)=0 , then the orbit space M/T is (k+1) -acyclic, i.e. \widetilde{H}_{i}(M/T;\Bbbk)=0 for any i\leq k+1 . In order to apply Theorem 3.8 for 3 -independent cases (i.e., GKM 3 cases) in Theorem 3.3 , in what follows we need to know H^{*}(G/H,\Bbbk)=0 for particular \Bbbk coefficients. If \Bbbk is a field of \chr\Bbbk=0 , this always holds [ Bor53 , Prop. 30.2] . More precisely, combining with the known results for cohomology, we have the following corollary: Corollary 3.9 . If G/H is a symmetric GKM 3 space (see Theorem 2.9 and (1), (2) of Theorem 3.3 ) with the following coefficient \Bbbk for the reduced homology, then \widetilde{H}_{i}(T\backslash G/H;\Bbbk)=0 for i\leq 4 . If \Bbbk=\mathbb{Z} , the the list is as follows: label pair G H AIII (A_{n},A_{i}\times A_{n-1-i}\times T^{1}) SU(n+1) S(U(i+1)\times U(n-i)) 0\leq i\leq n-1 BDI (D_{n},D_{n-1}\times T^{1}) SO(2n) SO(2n-2)\times SO(2) DIII (D_{n},A_{n-1}\times T^{1}) SO(2n) U(n) CII (C_{n},C_{i}\times C_{n-i}) Sp(n) Sp(i)\times Sp(n-i) 1\leq i\leq n-1 EIII (E_{6},D_{5}\times T^{1}) E_{6} (Spin(10)\times T^{1})/\mathbb{Z}_{4} EVII (E_{7},E_{6}\times T^{1}) E_{7} (E_{6}\times T^{1})/\mathbb{Z}_{3} FII (F_{4},B_{4}) F_{4} Spin(9) If \Bbbk=\mathbb{Z}[\frac{1}{2}] , the the list is as follows: label pair G H BDI (D_{n},D_{i}\times D_{n-i}) SO(2n) SO(2i)\times SO(2n-2i) i>2 , n-i>2 EII (E_{6},A_{5}\times A_{1}) E_{6} (SU(6)\times SU(2))/\mathbb{Z}_{2} EV (E_{7},A_{7}) E_{7} SU(8)/\mathbb{Z}_{2} EVI (E_{7},D_{6}\times A_{1}) E_{7} (Spin(12)\times SU(2))/\mathbb{Z}_{2} EIX (E_{8},E_{7}\times A_{1}) E_{8} (E_{7}\times SU(2))/\mathbb{Z}_{2} Proof. To prove the statement, we only check when the respective results for 3 -independent cases (see Theorem 3.3 ) has the property H^{*}(G/H,\Bbbk)=0 for the corresponding coefficient rings \Bbbk . If m_{i}=1 holds (this corresponds to Theorem 2.5 (2) , i.e., there is a T^{1} -factor in H ), the homogeneous space G/H is a Hermitian symmetric space. It admits an even-dimensional Schubert cell decomposition by existence of a Bott-Morse function [ Bot56 , Theorem A] ; therefore, H^{odd}(G/H;\mathbb{Z})=0 . In this case, the GKM 3 spaces appeared in Theorem 3.3 are as follows [ Wo11 , Corollary 8.11.7, p.293] : \displaystyle(A_{n},A_{i}\times A_{n-1-i}\times T^{1}),\ (D_{n},A_{n-1}\times T^{1}),\ (D_{n},D_{n-1}\times T^{1}),\ (E_{6},D_{5}\times T^{1}),\ (E_{7},E_{6}\times T^{1}). Moreover, it is well-known that the integral cohomology ring for Cayley projective space F_{4}/Spin(9) , i.e., (F_{4},B_{4}) , is generated by the degree 8 element and also satisfies H^{odd}(G/H;\mathbb{Z})=0 . We next claim that H^{odd}(G/H;\mathbb{Z}) is torsion if {\rm rank}\ G={\rm rank}\ H . Indeed, suppose that \displaystyle H^{2i-1}(G/H;\mathbb{Z})\simeq\mathbb{Z}^{m}\oplus F, where \mathbb{Z}^{m} is a free part and F is the torsion subgroup. Then, tensoring \mathbb{Q} gives \displaystyle H^{2i-1}(G/H;\mathbb{Z})\otimes\mathbb{Q}\simeq H^{2i-1}(G/H;\mathbb{Q})\simeq\mathbb{Q}^{m}. Since H^{odd}(G/H;\mathbb{Q})=0 , we must have m=0 . Hence, H^{2i-1}(G/H;\mathbb{Z}) is torsion. Using this claim, we shall prove the statement for the other cases. For the other cases, we refer the known facts from the other literatures. By the computations of torsions for the oriented partial flag varieties in [ MW , Theorem 1.2] , the integer cohomology of the oriented Grassmann (includes (D_{n},D_{i}\times D_{n-i}) for i>2,n-i>2 ) has only 2 or 4 torsions. Therefore, by the above claim, H^{odd}(G/H;\mathbb{Z}[\frac{1}{2}])=0 for G/H=SO(2n)/(SO(2i)\times SO(2n-2i)) . For the exceptional cases EII, EV, EVI, EIX, it satisfies \pi_{1}(H)=\mathbb{Z}_{2} . Therefore, by using [ IT77 , Proposition 3.3] , we have that H^{odd}(G/H;\mathbb{Z}[\frac{1}{2}])=0 . This establishes the lists when \Bbbk=\mathbb{Z}[\frac{1}{2}] . ∎ Remark 3.10 . The integer cohomology ring for EII is determined by generators and relations (see Ishitoya [ Is77 ] ). Some of them for BDI are also determined (e.g. [ Jo24 ] ). However, the integer cohomology rings for all cases of BDI and the cases EV, EVI, EIX are not determined (also see [ Na01 ] ). Example 3.11 . Consider the homogeneous manifold Gr^{+}(2,n)=SO(n)/SO(n-2)\times SO(2)=Q_{n}, with the natural effective T -action. Here, if n=2r then T=T^{r}/{\pm 1} , and if n=2r+1 then T=T^{r} . This oriented Grassmanian is a connected double cover of the real Grassmanian Gr(2,n) . This implies Gr^{+}(2,3)=S^{3} . It is not difficult to show that Gr^{+}(2,4) is homeomorphic to S^{2}\times S^{2} . (Furthermore, Gr^{+}(2,6) is homeomorphic to the complex Grassmanian Gr_{{\mathbb{C}}}(2,4) .) The orbit space is homeomorphic to \Sigma^{\lfloor n/2\rfloor}{\mathbb{C}}P^{\lfloor n/2\rfloor-2} by [ Su19 ] . In particular, the orbit space is {\lfloor n/2\rfloor} -connected (where the suspension of an empty set is by definition a point). In particular, it is contractible for n=3,4 . This homogeneous manifold has type D_{r}/D_{r-1}\times T^{1} for n=2r and B_{r}/B_{r-1}\times T^{1} for n=2r+1 , where r\geq 2 . Here, B_{1}=A_{1} , D_{1}=T^{1} , and D_{2}=(A_{1})^{2} . Since Gr^{+}(2,2r)=Q_{2r} and Gr^{+}(2,2r+1)=Q_{2r+1} , It is known that H^{odd}(Q_{n};\mathbb{Z})=0 . Therefore, Theorem 3.8 implies that the orbit space is 4 -acyclic for n=2r and 3 -acyclic for n=2r+1 . Therefore, our result agrees with previous known results, in particular, in dimensions n=6,7,8 . Example 3.12 . Consider the homogeneous manifold Gr(2,5)=SU(5)/S(U(2)\times U(3)), with the natural effective T^{4} -action. This is the complex Grassmanian of two-planes in {\mathbb{C}}^{5} . The fixed-point data at the identity is a root system of type A_{4}/A_{1}\times A_{2}\times T^{1} . By Theorems 3.3 and Corollary 3.9 , this GKM-manifold has 4 -acyclic orbit space Gr(2,5)/T^{4} . On the other hand, by [ BT19 ] or [ Su21 ] , \widetilde{H}_{*}(Gr(2,5)/T^{4};{\mathbb{Z}})\simeq\begin{cases}{\mathbb{Z}}/2{\mathbb{Z}},\ n=5,\\ {\mathbb{Z}},\ n=8,\\ 0,\ \mbox{else}.\end{cases} This agrees with the above acyclicity claim. Remark 3.13 . The second named author finds the negative answer to Problem 1 for the abstract GKM graphs which are the combinatorial counterpart of GKM manifolds in [ So23 ] . Furthermore, this problem is positively solved for GKM graphs of GKM manifolds that are either Hamiltonian or have complexity 1 by [ GS ] . However, the geometric problem is still open. 4. The weighted signed graph associated from positive root systems In this section, we first recall the signed graph, introduced by Harary [ Ha53 ] and have been extensively studied by Zaslavsky from a matroid theoretic viewpoint (see e.g. [ Za82 ] from one of his series of papers) and also studied by Fern-Gordon-Leasure-Pronchik or Fried-Gerek-Gordon-Perunčić for an automorphism of matroid induced from the root systems of classical types and F_{4} , see [ FGLP00 , FGGP07 ] . Let \Gamma:=(V,E) be a graph (with loops and multiple edges allowed). A signed graph is a graph \Gamma in which each edge e\in E is labeled with a + or - sign, say s(e) . In this paper, we mainly consider the following three types of signed graphs (see Figure 1 ): Slim: the complete graph of n vertices such that all edges have the same sign, say K_{n} ; Puff: the complete graph of n vertices such that every two vertices are connected by exactly two edges (i.e., a multiple-edge) whose signs are + and - respectively, say \pm K_{n} ; Superpuff: the signed graph \pm K_{n} with n loops on each vertex, say K_{n}^{\circ} . 1 2 3 4 1 2 3 4 1 2 3 4 Figure 1 . The signed graph K_{4} , \pm K_{4} and K_{4}^{\circ} (from left). Edges with s(e)=- (resp. s(e)=+ ) are drawn as solid (resp. dashed) edges. Moreover, in this paper, we give a weight on each vertex by a vector in \mathbb{R}^{n} , say w_{V}:V\to\mathbb{R}^{n} (see Figure 2 ). In this situation, we may also define weights on edges and loops, say w_{E}:E\to\mathbb{R}^{n} , by • if l\in E is a loop on the vertex v\in V , then w_{E}(l):=w_{V}(v) ; • if e\in E is an edge connecting two vertices p,q\in V , then for some non-zero r^{+}_{pq},r^{-}_{pq}\in\mathbb{R} , w_{E}(e):=r^{+}_{pq}(w_{V}(p)-w_{V}(q)) if s(e)=+ and w_{E}(e):=r^{-}_{pq}(w_{V}(p)+w_{V}(q)) if s(e)=- . We call such a signed graph equipped with a weight w_{V},w_{E} a weighted signed graph . The positive roots of certain compact Lie groups can be regarded as a weighted signed graph as follows, where e_{i}\in\mathbb{R}^{N} is the standard basis for i=1,\ldots,N and the edge e=ij\in E connecting vertices i , j\in V . • for A_{n} ( n\geq 1 ), the weighted signed graph is K_{n+1} with the weights w_{V}(i):=e_{i} and r^{+}_{ij}=-r^{+}_{ji}=1 , where i,j\in V=[n+1] and 1\leq i