SSPHERE GEOMETRY AND INVARIANTS
OLIVER KNILL
Abstract.
A finite abstract simplicial complex G defines two fi-nite simple graphs: the Barycentric refinement G , connecting twosimplices if one is a subset of the other and the connection graph G (cid:48) , connecting two simplices if they intersect. We prove that thePoincar´e-Hopf value i ( x ) = 1 − χ ( S ( x )), where χ ( S ( x )) is the Eulercharacteristics of the unit sphere S ( x ) of a vertex x in G , agreeswith the Green function g ( x, x ) = (1 + A (cid:48) ) − xx , where A (cid:48) is the adja-cency matrix of the connection graph G (cid:48) of the complex G . By uni-modularity ψ ( G ) = det(1 + A (cid:48) ) = (cid:81) x ( − dim(x) = φ ( G ), the Fred-holm matrix 1 + A (cid:48) is in GL( n, Z ), where n is the number of sim-plices in G . We show that the set of possible unit sphere topologiesin G are combinatorial invariants of the complex G , and establishso that also the Green function range of G is a combinatorial invari-ant. The unit sphere character formula g ( x, x ) = i ( x ) applies espe-cially for the prime graph G ( n ) and prime connection graph H ( n )on square free integers in { , . . . , n } playing the role of simplices.In G ( n ), integers a, b are connected if a | b or b | a and where in H ( n )two a, b are connected if gcd( a, b ) >
1. The Green function g ( x, x )in H ( n ) relate there to the index values i ( x ) in G ( n ). To prove theinvariance of the unit sphere topology we use that all unit spheresin G decompose as S − ( x ) + S + ( x ), where + is the join and S − isa sphere. The join renders the category X of simplicial complexesinto a monoid, where the empty complex is the 0 element andthe cone construction adds 1. The augmented Grothendieck group( X, + ,
0) contains the graph and sphere monoids (Graphs , + , , + , G → i ( G ) =1 − χ ( G ) or G → i G ( x ) = 1 − χ ( S G ( x )) as well as the volumeare multiplicative functions on ( X, +). For the sphere group, both i ( G ) as well as ψ ( G ) are characters. The join + can be augmentedby a product · so that we have a commutative ring ( X, + , , · , ± K i isomorphic to the integers ( Z, + , , · , · , the question of unique prime factorization appearsopen. Date : February 12, 2017.1991
Mathematics Subject Classification.
Key words and phrases.
Graph theory, simplicial complexes, arithmetic. a r X i v : . [ m a t h . GN ] F e b OLIVER KNILL Preface
A finite abstract simplicial complex G has a Barycentric refinement G which is the Whitney complex of a graph. If χ denotes the Eulercharacteristic functional, we identify the values i ( x ) = 1 − χ ( S ( x ))of unit spheres S ( x ) in G as Green function values g ( x, x ) = (1 + A ( G (cid:48) )) − xx , where A ( G (cid:48) ) is the adjacency matrix of the connection graph G (cid:48) of G . The graph G (cid:48) has like G the set of simplices in G as vertex setbut connects two if they intersect; in G , two simplices are connectedif one is contained in the other. Having established that 1 + A ( G (cid:48) ) isunimodular [24], the matrix entries g ( x, y ) became interesting. We cannow look at the connection graph G (cid:48) of G . Its Green function valuesare again related to unit spheres in the Barycentric refinement G of G . We compare the unit sphere topologies in G with the unit spheretopologies of G . We show here that in G , no new topologies appear.It follows that the sphere index spectrum, the set of Green functionvalues is a combinatorial invariant of a simplicial complex Figure 1.
We see the triangle G and two of itsBarycentric refinements G and G . A choice of unitspheres in G and G are marked. Figure 2.
For the windmill graph G , unit spheres in G can have topologies different from unit sphere topolo-gies in G . While all unit spheres in G are contractible,there is a unit sphere S ( x ) in G which as a suspen-sion P + P of the 3-point graph P has χ ( S ( x )) = χ ( P ) + χ ( P ) − χ ( P ) χ ( P ) = −
1. But in G , no newtopologies appear in unit spheres. PHERE GEOMETRY AND INVARIANTS 3 Introduction
The quest to find invariants of topological spaces is particularly con-crete when searching for combinatorial invariants in abstract finite sim-plicial complexes. These are quantities which do not change when ap-plying a Barycentric refinement [2]. Abstract simplicial complexes havea surprisingly rich geometry despite the fact that they are one of thesimplest objects mathematics knows. Every partially ordered set forexample defines a complex, the order complex. Besides matroids alsographs are a source for complexes, like subcomplexes of the clique com-plex or then graphic matroids. Other type of simplicial complexes wereintroduced by Jonsson in [10]. Graph theory enters naturally as anyBarycentric refinement G of an arbitrary abstract simplicial complexis already is the Whitney clique complex of a graph. The faces of G are the vertices of G and two faces connected if one is contained inthe other. The complex G also defines the connection graph G (cid:48) on thesame vertex set, where two simplices connect if they intersect. Both G and G (cid:48) produce aspects of the geometry on G which are well accessibleas graphs are not only intuitive, they also serve well as data structures.Examples of combinatorial invariants of simplicial complexes are sim-plicial cohomology, Euler characteristic, homotopy groups (discrete no-tions of spheres and discrete notions of homotopy allow them purelycombinatorically), the Bott invariants [2], the clique number (which is1 plus the maximal dimension), minimal entries of a f -vector [26, 27],Wu characteristic, as well as connection cohomology attached to Wucharacteristic [35, 23] or minimal possible entries in the f -matrix. Ex-amples of quantities which are not combinatorial invariants are the f -vector itself (as it gets multiplied by a fixed upper triangular matrixwhen applying a refinement), the f -matrix, telling about the cardinal-ities pair intersections, the chromatic number (it can decrease underrefinements but will stabilize already after one step to the clique num-ber) or dimension (it can increase under refinements and will convergeto the maximal dimension when iterating the Barycentric refinementprocess), neither is the Fredholm characteristic ψ (as it stabilizes to 1already after one Barycentric refinement).Focusing on finite simplicial complexes G allows to do geometry onfinite sets but still be close to classical topology: the topology of asmooth compact manifold M for example is completely determined bythe topology of the simplicial complex G defined by a triangulation OLIVER KNILL - - - - - - - - - -
200 0 00 - - - - - - - - - -
20 0000 - - - - - - -
200 0 0 00 0 0 - - - - - - -
20 00000
Figure 3.
Adenine, Guanine, Cytosine and Thyminegraphs with sphere spectrum. These are all 1-dimensional graphs. The Fredholm Characteristic ψ is1 for Adenin and − χ is − σ dependonly on vertex degrees as for all 1-dimensional graphs.It is σ = {− , − , } for the first three. Thymine alonehas a different sphere spectrum σ = {− , − , } . Thevalues ψ, φ, σ are different for all. Also the f -vectors(15 , , (16 , , (13 , , (15 ,
15) differ. All four graphsare prime in the Zykov monoid of graphs.of M in the form of a d -graph and the abstract Whitney complex de-scribing it contains all topological information about M . While it ispossible to look at topological invariants by embedding the complexinto a continuum or by realizing it as a polytop in an Euclidean space,we prefer here to look at invariants defined and computed in a finitecombinatorial manner. This is pragmatic computer science point of PHERE GEOMETRY AND INVARIANTS 5 view.Both the set of abstract finite simplicial complexes as well as the set offinite simple graphs form a Boolean lattice: one can intersect and takeunions. They both do not form a Boolean ring however as the Booleanaddition, the symmetric difference operation, A ∆ B = A ∪ B \ A ∩ B is no more in the same category. A valuation is a numerical quan-tity X satisfying X ( G ∪ H ) + X ( G ∩ H ) = X ( G ) + X ( H ). If wehad a Boolean ring, a valuation would satisfy the linearity condition X ( G ∆ H ) = X ( G ) + X ( H ). By discrete Hadwiger, any valuation isof the form X ( G ) = f ( G ) · χ , where χ is a fixed vector and f ( G ) isthe f -vector of G . Since the only eigenvector of the Barycentric op-erator f ( G ) → f ( G ) is (1 , − , , − , . . . ) Euler characteristic is theonly combinatorial invariant which is also a valuation. While Eulercharacteristic χ ( G ) = (cid:80) x ω ( x ) with ω ( x ) = ( − dim( x ) is pretty uniqueamong valuations, there are other functionals. One is the Fermi charac-teristic ψ ( G ) = (cid:81) x ω ( x ), a multiplicative cousin of Euler characteristic.Then there are multi-variate versions like the Wu characteristic ω ( G ) = χ · F ( G ) χ , where F ( G ) is the f -matrix F ( G ) ij counting the number ofintersections of i and j -dimenensional simplices in the complex. Alsothe Wu characteristic as well as higher order versions are combina-torial invariants. A quadratic valuation X ( G, H ) is a map for which G → X ( G, H ) and H → X ( G, H ) are both valuations. Related to thishad been a question of Gruenbaum [6], whether higher order Dehn-Sommerville relations exist. We answered this affirmatively in [23].The analysis there shows that Wu characteristic is the only combinato-rial invariant among multi-linear valuations. Connection calculus leadsto other combinatorial invariants like the Betti numbers of connectioncohomology. That this is useful has already been demonstrated in ex-amples. It allows to distinguish the cylinder and the Moebius strip forexample.Quantities satisfying the multiplicative valuation property X ( G ∪ H ) X ( G ∩ H ) = X ( G ) X ( H ) must be additive valuations after taking logsand indeed, the Fredholm determinant is an example. It is related tothe valuation f ( G ) counting the number of odd-dimensional simplicesin G .We should point out that cohomology with real-valued cocycles is welldefined also for a finitist who does not accept infinity. The reason isthat the cohomology groups are determined by Hodge as the nullity OLIVER KNILL dim(ker( L k )) of finite integer-valued matrices L k , the form Laplacians L k . The dimension of the kernel as well as the kernel basis itself canbe obtained by row reduction, which is a finite process and done oninteger matrices. So, cohomology is an acceptable concept for a fini-tist. For more about the complexity, see [11]. This happens alreadywhen following the lead of Kirchhoff, Betti and Poincar´e. In topologyhowever still, Euclidean realizations are used even for definitions likeBarycentric refinements. But one perfectly can stay within the realm ofsimplicial complexes or better in the realm of much the more accessiblegraph category, on which the complex is a subcomplex of the Whitneyclique complex.How do simplicial complexes relate to traditional topology? The usualapproach is to realize the complex in some sufficiently high dimensionalEuclidean space and call two complexes G, H topologically equivalentif their topological realizations
G, H are homeomorphic in the classi-cal point set topology sense. A topological invariant of the complexis then a quantity produces homeomorphic geometric realizations. Forabstract finite simplicial complexes, the notion of topological invari-ant is equivalent to the notion of combinatorial invariant but the laterworks in the more restricted axiom system of finite mathematics.The quest to find a complete finite set of invariants which allow to dis-tinguish an arbitrary pair of simplicial complexes has long shown to beunattainable: it is not possible to build a Turing machine which candistinguish any pair of abstract finite simplicial complexes. Markovused in 1958 the unsolvability of the word problem by Novikov to showthat the homeomorphism problem is not solvable. For surveys, see[34, 1, 29].A constructivist would insist on giving a concrete homeomorphismwhich establishes the topological equivalence, but this is not so easy,as the Euclidean realizations are piecewise linear pieces of simplices,glued together in some high dimensional space. Also the complexity todecide whether a given pair of finite simplicial complexes are topologi-cally equivalent can be tough. Given two concrete complexes, how dowe (by just dealing with finite sets) decide whether they are topologi-cally equivalent? We need other morphism for simplicial complexes inorder to mirror topological equivalence. One answer is to use a finiteZariski type topologies which is purely combinatorial. It is conceiv-able however that we have to go through an exponentially large set ofpossible admissible finite topologies and check in each case the graph
PHERE GEOMETRY AND INVARIANTS 7 isomorphism problem.Let us assume that the complex is the Whitney complex of a graph. Asmentioned already, this is not much of a loss of generality as a Barycen-tric refinement of an abstract simplicial complex is always the Whitneycomplex of a graph. Since a graph has a natural metric, the geodesicdistance, one could see such a complex as an example of a finite metricspace. But since every singleton set { x } is both open and closed in thatmetric, the topology is the discrete topology and not that interesting.A weaker topology is Zariski like: it the topology on the vertex set of G defined in such a way that the vertex set of H of any subcomplex H of G is closed. This defines a finite topology on the vertex set of G , (which is in general not Hausdorff similarly as Zariski topology),the homeomorphisms are still the graph isomorphisms. (The analogyis that subgraphs of G play the role of varieties and homeomorphismthe role of regular maps. The algebraic nature of the “varieties” comesfrom the fact that we only allow subgraphs of G and not subgraphs of G . The later would give the discrete topology.)The Zariski idea is more flexible as we can use the same definition toform much weaker topologies by starting with a smaller set of sub-graphs of G defining a sub-base of the topology. For example, we canget topologies on the second refinement G by taking the set of graphs H as closed, where H is a subgraph of H . We have proposed [19] toinsist that a finite topology on a graph should be given by a subbaseand have the property that the nerve graph of a sub-base should behomotopic to the graph and that a dimension condition is satisfied forintersections. The reasons are that we want any reasonable notion ofcohomology to agree with the natural topology of the graph and in-sist that any notion of homeomorphism should honor the concept ofdimension.Unlike the quest to find a complete set of invariants for a simplicialcomplex, the task to find concrete combinatorial invariants of an ab-stract simplicial complex is more accessible. It is part of what wedo here. An example of quantities which are interesting are valua-tions satisfying X ( G ∪ H ) + X ( G ∩ H ) = X ( G ) + X ( H ). The dis-crete Hadwiger theorem [13] has classified all valuations on a complexand identified Euler characteristic as the only combinatorial invariantamong them. It states that if the maximal dimension of G is d , thespace of valuations is d + 1 dimensional. More invariants can be ob-tained by looking a multi-linear valuations or multiplicative invariants OLIVER KNILL X ( G ∪ H ) X ( G ∩ H ) = X ( G ) X ( H ) like Fredholm characteristic.Since Fredholm characteristic det(1 + A ( G (cid:48) )) is {− , } -valued for sim-plicial complexes we can look at the Green function values g ( x, y ) =(1 + A (cid:48) ) − xy . Investigating these integers led to the current paper. Itis here related to locally defined combinatorial invariants, the sphereEuler characteristic spectrum, which is defined by the collection of in-dices i ( S ( x )) = 1 − χ ( S ( x )) which unit spheres S ( x ) in the complex canhave. One could look at higher order versions like the quadratic spherespectrum 1 − χ ( S ( x ) ∩ S ( y )), where x, y run both over all the simplicesin G . This however does not match the off-diagonal Green function val-ues in general and the off-diagonal values (1 + A (cid:48) ) − xy with x (cid:54) = y remainat the moment still unidentified. Analogies from physics suggest thatall Green function values should have some natural interpretation andare possibly of a dynamical nature. Work like [3] show how close thediscrete case can be to the continuum. In the shifted Fredholm case, itis exciting that the Green function values are integers and quantized.There would be other numbers to consider like (1 + A (cid:48) ) kxy but we seemonly get topological invariants for k = − G + 1 is the coneconstruction.Unlike for the arithmetic of numbers, where 1 is the only additive prime,there are more additive primes in the join monoid. The 0-dimensionalsphere P for example is prime. Adding such a sphere to a graph isthe suspension. The maximal dimension of a sum G + H is the sumof the maximal dimensions plus 1. But there are still mysteries aboutthe Zykov monoid. Unlike in the continuum, we suspect (but do notknow nor dare to conjecture) that in the discrete, the join operation isa unique factorization monoid, where the empty graph plays the role PHERE GEOMETRY AND INVARIANTS 9 of the 0 element and where the 1-point graph K is the smallest ex-ample of an additive prime. The complete graph K n decomposes as K + · · · + K and C can be written as P + P . We can how forexample that a triangulation of a 2d-surface of positive genus is prime,as the only 2d-surfaces which can be factored are of the form C n + P which are all 2-spheres which are prism graphs for n ≥ G factors additively into primes G = p + p · · · + p k ,but we don’t know yet whether the factorization is unique. [I couldnot get hold yet of the Zykov article [37]. The MathSciNet review ofTutte mentions a unique prime factorization result there. ] We haveinfinitely many prime graphs, like graphs P n without edges, circulargraphs C n with n > G + H is the product of the number of simplices of G and H . A spherewith a prime number of maximal simplices therefore is an additiveprime in the sphere monoid. In general, the Euler generating functions f G ( x ) = 1 + (cid:80) k =0 v k x k +1 for the join satisfies f G + H = f G f H , whichhas as a corollary χ ( G + H ) = χ ( G ) + χ ( H ) − χ ( G ) χ ( H ) which followsfrom χ ( G ) = 1 − f G ( − i ( G ) = 1 − χ ( G ) ismultiplicative and that graphs with zero Euler characteristic, or graphswith even Euler characteristic and graphs with odd Euler characteristicall form sub-monoids.At the end we point out that there is a multiplication on simplicialcomplexes which is compatible with the Zykov join addition in thesense that distributivity holds. If the Zykov join monoid is extended tobecome a group we get so a commutative ring of simplicial complexes.This ring extends naturally the ring of integers as it contains the ringof integers in the form of complete graphs: K n + K m = K n + m and K n K m = K nm . The empty graph 0 is the zero element and the graph K is the 1-element. We originally looked for a ring structure in orderto proof the unimodularity theorem. A purely algebraic proof has notworked yet and might not even exist. Indeed, the Fredholm functional ψ is only multiplicative on the additive subgroup of signed complexes which have even Euler characteristic. But ψ is a character on thesphere subgroup of the additive group similarly as the Poincar´e-Hopffunctional i ( G ) = 1 − χ ( G ) which is a character on the sphere groupand which plays here an important role.3. Simplicial complexes
A finite abstract simplicial complex is a finite set G of non-empty sets,called simplices, so that G is invariant under the operation of takingnon-empty subsets. The dimension of a simplex x in G is defined as thecardinality of x minus 1. In graph settings, the cardinality is called theclique number. The maximal dimension of G is the maximal dimen-sion, which a simplex in G can have. An abstract simplicial complex issometimes also called hereditary collection [30]. Early references usingfinite abstract simplicial complexes are [32, 12] not using Euclidean re-alizations exclusively as in [36].Having only the subset axiom, simplicial complexes are one of the sim-plest geometrical structures imaginable, simpler even than algebraicstructures, a topology or measure theoretical structures. Indeed, anabstract simplicial complex is an order structure on a set and everypartial ordered set defines a simplicial complex, its order complex. Ab-stract finite simplicial complexes are therefore closely related to finiteposets. In the rest of the paper, we often just say “complex” meaningfinite abstract simplicial complex.The Barycentric refinement G of a complex G is the set of subsets ofthe simplex sets A which have the property that for any pair ( a, b ) in A , either a is a subset of b or b is a subset of a . This defines a graph G = ( V , E ) where V is the set of simplices in G and E the set ofpairs ( a, b ) such that a is contained in b or b is contained in b . TheWhitney complex of this graph is then the complex G . One can alsolook at the connection complex G (cid:48) of G which is the Whitney complexof the graph where two simplices ( a, b ) are connected, if one is con-tained in the other. It can be easier to work with graphs G and G (cid:48) however rather than the intrinsic simplicial complexes. Remarks.1)
There is a more sophisticated dimension of a complex: the inductivedimension dim( x ) of a simplex x . It is defined as 1 + dim( S ( x )), where S ( x ) is the Whitney complex in the unit sphere S ( x ) of x in G , anddim( G ) = (1 / | V ( G ) | ) (cid:80) x dim( x ). Unlike the maximal dimension, the PHERE GEOMETRY AND INVARIANTS 11 inductive dimension is a rational number in general. It has some niceproperties like satisfying the same dimension inequality than Hausdorffdimension or having computable expectation on Erd˝os R´enyi probabil-ity spaces [14]. The Barycentric refinement of G can be written as a product G × K [21]: this simplicial product graph G × H of two complexes is the graphfor which the vertices are the ordered pairs ( x, y ) of simplices x in G and y in H for which two ( a, b ) , ( c, d ) are connected if either a ⊂ c, b ⊂ d or c ⊂ a, d ⊂ b . Even if G is a one-dimensional graph, it has no relationwith the Cartesian product or tensor product for graphs which bothare graphs with vertex set V ( G ) × V ( G ). The simplicial product graph G × H has a vertex set the product of the simplex sets of G and H .The simplicial product graph can be rewritten as the product in theStanley-Reisner ring. It is a natural product as the Euler characteristic χ ( G × H ) = χ ( G ) × χ ( H ) is multiplicative and the K¨unneth formulafor cohomology applies [21].Given a simplicial complex G , we can look both at the Barycentric re-finement G of G as well as the connection graph G (cid:48) of G . Both graphswill play a role here. The two graphs G (cid:48) and G have the same vertexset as this is the set of simplices of G . The graph G is a subgraph of G (cid:48) . In the theory of simplicial complexes one looks also at the Hassediagram, which is a subgraph of G .The unit spheres S ( x ) in G have topologies which are interesting. Itturns out that the possible topologies which appear as unit spheres in G form a combinatorial invariant of G . While G can feature new unitsphere topologies, there are no new sphere topologies appearing in G .Note that we work with general simplicial complexes and do not as-sume any Euclidean structure. Still, there is intuition which leads tothe invariance of the unit sphere topology. A Barycentric refinementscan produce new sphere types but after having done one refinement,the local sphere geometry does not change any more in the interior.The inside of refined simplices is now Euclidean. The unit spheres inthe interior are discrete spheres and space looks there like Euclideanspace. It is getting more interesting at singularities, where unit spheresare no topological spheres any more. An example is the figure 8 graph.Most points in that graph have a zero-dimensional sphere as a unitsphere but there is a point, the singularity, where a sufficiently smallsphere consists of 4 points. The value of the Euler characteristic is 4. It is a value does not change under Barycentric refinement. - - - - - - - - - - - - - -
10 0 - - - - - - - - - - - - - - - -
10 0 00 00 0000 1 10 1000 10 01 10 00 0 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Figure 4.
Examples of graphs and their Barycentricrefinement. In each case, we add the sphere spectralvalues at each vertex.Let us call the number i ( x ) = 1 − χ ( S ( x ))the Poincar´e-Hopf index of the vertex x , where S ( x ) is the unit spherein the Barycentric refinement G of the complex G . The name has beenchosen because if f is a function taking the maximum at x , then this isthe usual Poincar´e-Hopf index i f ( x ) = 1 − χ ( S − ( x )) [15], where S − ( x ) isgenerated by the set of vertices y , where f ( y ) < f ( x ). The name char-acter is adequate since on the subgroup of complexes with χ ( G ) (cid:54) = 1, itis a multiplicative character on a group constructed by the join opera-tion on complexes. The set of possible values { i ( x ) | x ∈ V ( G ) } is thePoincar´e-Hopf spectrum of the graph. Since i G + H ( x, y ) = i G ( x ) i H ( y ),they form generalized characters. The set of unit sphere topologies isthe sphere topology spectrum. We can quantify this numerically forexample by attaching to each vertex x the Betti numbers b k ( S ( x )) ofthe unit sphere. The set { b k ( S ( x )) | x ∈ V ( G ) } for example could becalled the k ’th Betti spectrum of the complex G .The invariance of the sphere topology allows to check whether a graphis a Barycentric refinement of a complex. If the unit sphere topologychanges under Barycentric refinement, then G is not the Barycentric PHERE GEOMETRY AND INVARIANTS 13 refinement of a complex. An example is the windmill graph obtainedby gluing three triangles along a common edge. While all unit spheresof G are connected, there is a vertex in G with a disconnected unitsphere. We have seen before that if ψ ( G ) is not in {− , } , then G isnot the connection graph of a complex. An example is a triangle whichhas ψ ( G ) = 0. 4. Sphere topology
In this section we look at discrete spheres and unit spheres S ( x ) ofBarycentric refinements of finite abstract simplicial complexes. Notethat since we look at rather general graphs, the unit spheres are rarelyspheres but rather general graphs.Barycentric refined graphs are natural as they are Eulerian [20]. Thismeans that they can be colored with a minimal number of colors, thecolor being the dimension which the point has when it was a simplex ofthe complex. This is useful for graph chromatology [18]: one approachto 4-color a planar graph is is to embed it first into a 2-sphere, thenfill out the interior 3-ball by cut it up using edge refinements until it isEulerian and so 4-colorable. Having colored the interior and not cut-ting anything at the boundary, this colors the boundary with 4 colors.The unit sphere S ( x ) of a simplicial complex is a sub-complex con-sisting of all simplices in G which are either contained in x or whichcontain x . This is better seen in the Barycentric refinement G , forwhich the simplices of G are now vertices and the unit sphere is thegeodesic unit sphere with respect to the graph distance.Given a simplex x , the set of subsets of x is the complex S − ( x ), thestable or negative part of S ( x ). The set of simplices which contain x define the sub-complex S + ( x ), the unstable or positive part of S ( x ).More intuitive is the graph theoretical reformulation in the graph G :the unit sphere S ( x ) consists of all vertices in G which are connectedto x . The function f ( x ) = dim( x ) which gives the dimension of x whenwe look at it as a set in the simplicial complex G . It defines a coloringin G . The dimension functional dim resembles a Morse function asit allows to partition the unit sphere S ( x ) into a stable part S − ( x )and unstable part S + ( x ). As in the continuum, the join of S − ( x ) and S + ( x ) is S ( x ), this relation will also hold here in full generality for anysimplicial complex. The property S − ( X ) + S + ( x ) = S ( x ) could serveas a Morse condition without assuming S + ( x ) or S ( x ) are spheres. But here, we don’t make any such assumption.So here is again a formal definition for S ± ( x ) in the case of the function f ( x ) = dim(x): S − ( x ) = { y ∈ S ( x ) | f ( y ) < f ( x ) } and S + ( x ) = { y ∈ S ( x ) | f ( y ) > f ( x ) } . Despite the fact that the definitions for S − and S + look similar, thereis a difference as we do not have any kind of Poincar´e duality in thecase of a general complex. The stable sphere S − ( x ) is always a discretesphere. It only deals with the inside of a simplex which after Barycen-tric refinement is of Euclidean structure. The possibly crazy outsideworld, deals with connections of simplices which are not Euclidean evenfor very simple examples. For a one-dimensional star graph for exam-ple the central vertex x is 0-dimensional and S + ( x ) consists of all edgescontaining x .The Zykov join of two graphs G, H is is the graph G + H which hasas vertex set the union of the vertices of G and H and where two ver-tices are either connected if they were already connected in G or H orthen if they belong to different graphs. We simply call G + H the joinof G and H . The join renders the category of graphs into a monoidwhich can be augmented to an Abelian group on which functionals like j ( G ) = − ψ ( G ) or i ( G ) = 1 − χ ( G ) are multiplicative.A d -sphere or shortly d -sphere is a finite simple graph which has theproperty that every unit sphere is a ( d − − d -graph is a finite simple graph for which all unit spheres are ( d − d -complex is a finite abstract simplicial complex for whichits Barycentric refinement is a d -graph. Remark . The notion of Evako d-sphere and d-graph is equivalent tothe notion of d-sphere within combinatorial d-manifolds in the theoryof simplicial complexes. Notions as such have been put forward in[9, 4]. Related is the notion of a combinatorial d-manifold which is aconnected, pure (every facet has dimension d) finite abstract simplicialcomplex of maximal dimension d for which each ridge (codimension-1-face) has two facets as a boundary and such that all vertex links are PHERE GEOMETRY AND INVARIANTS 15
PL-homeomorphic to the boundary of a d-simplex. A combinatorial d-manifold is then a d-sphere if it is PL homeomorphic to the boundaryof a d-simplex. These notions rely on Euclidean embddings. Forman’sdiscrete Morse theory remains in the discrete: a combinatorial d-sphere G is PL-homeomorphic to a standard PL-sphere if and only if it is aReeb sphere: some Barycentric subdivision G n admits a discrete Morsefunction with exactly two critical points [5]. Beeing a Reeb sphere isequivalent to being an Evako sphere (removing one vertex renders thecomplex contractible). Starting with the notion of Evako sphere hasthe advantage of avoiding Morse theory. In any case, detecting spherescan be difficult [28].We start with a lemma which is motivated by an analogue fact in thecontinuum, where a Morse function at a critical point x defines twoparts S ± ( x ), the stable and unstable part of the sphere. They areobtained by intersecting a small geodesic sphere S ( x ) with the stableor unstable manifolds W ± ( x ). The sphere S ( x ) itself is then S − + S + .This is the picture for the gradient flow of a Morse function f . In thediscrete, in full generality, the dimension function f ( x ) = dim( x ) canserve as a Morse function as we have a stable and unstable part at everypoint. As pointed [25], the Morse cohomology in G in this setting isdirectly equivalent to the simplicial cohomology of G . This gives hopefor a more general Morse setting for functions with possibly much fewercritical points allowing to compute cohomology more efficiently thanwith incidence matrices defined by the simplicial complex. Lemma 1 (Unit sphere decomposition) . If G is the Barycentric re-finement of a finite abstract simplicial complex G , then for any vertex x in G , the unit sphere S ( x ) is the join of S − ( x ) and S + ( x ) .Proof. Both S ± ( x ) are subgraphs of the unit sphere S ( x ). The unionof the vertex sets of S − ( x ) and S + ( x ) is the vertex set of S ( x ). If a ∈ S − ( x ) and b ∈ S + ( x ), then ( a, b ) is an edge in S ( x ) because assimplices, a is a sub simplex of x and x is a sub simplex of b . (cid:3) We can rephrase that also by saying that the unit sphere of a vertex x for which the simplex x is not prime as a complex in G (which is alwaysthe case if the dimension is positive) is not prime in G but splits as S ( x ) = S − ( x ) + S + ( x ).Lets now look at the structure of the stable unit sphere: Lemma 2.
The stable unit sphere S − ( x ) of a vertex x in the Barycen-tric refinement G of an abstract simplicial complex is always a d -spherefor d = dim( x ) − . Proof. S − ( x ) is the boundary of the Barycentric refinement of the sim-plex X in G which belongs to the vertex x in G . We can show byinduction in the dimension d that this is a sphere. When looking atthe intersection with a sphere S ( y ) of a vertex y in S − ( x ) the dimen-sion is reduced by 1, so that by induction, the unit sphere of a vertex y ∈ S − ( x ) is a ( d − d -sphere, the collapsibility is satisfied by induction. Take away a vertex y of S − ( x ). Then every intersection of S − ( x ) with an other sphere S ( z ) which has contained y becomes contractible so that S − ( x ) \ y iscontractible. (cid:3) Examples : If x belongs to a 0-dimensional vertex { x } in G , then S − ( x ) isempty so that it is a d = ( − If x belongs to a 1-dimensional simplex ( x , x ) in G , then S − ( x )consists of two isolated points { x , x } which is d = 0-dimensionalsphere. If x belongs to a two-dimensional simplex ( x , x , x ) in G , then S − ( x ) is a cyclic graph C with 6 elements: { x , x x , x , x x , x , x x , x , x x , x , x x } .We will need a generalization of this which we can see as a higher orderversion of the just said. Given a vertex x in the second Barycentricrefinement G , this corresponds to a simplex ( x , x , . . . , x d ) in G . Wecan assume without loss of generality that the vertices are ordered sothat x ⊂ x ⊂ · · · ⊂ x k ), where x ⊂ y means that as a simplex in G , the simplex x is contained in the simplex y . Now define the secondstable sphere S − ( x ) = { z ⊂ x | z (cid:54) = x , z (cid:54) = x . . . , z (cid:54) = x k } . These second stable spheres are of interest because they appear as unitspheres in the interior of the second Barycentric refinement G of G . Lemma 3.
The second stable unit sphere S − ( x ) of a vertex x = ( x , . . . , x k ) in the Barycentric refinement G of an abstract simplicial complex G is always a ( l − k ) -sphere if dim( x k ) = l .Proof. We can see this in two different ways: first we can see it as anintersection of k unit spheres S ( x j ) in the stable sphere S − ( x k ). Eachintersection reduces the dimension by 1.If we mark the set of dimensions k j = dim( x j ) in the integer interval { , . . . , l } , then the set of simplices which belong to a gap of dimensionvalues [ a, b ] which are omitted, then this is a sphere of dimension b − PHERE GEOMETRY AND INVARIANTS 17 a −
1. We call this the gap sphere belonging to [ a, b ]. The second stableunit sphere is now the join of all these gap spheres: S − ( x ) = S [ a ,b ] + · · · + S [ a m ,b m ] . (cid:3) Examples : The set of simplices in the complex of all complete subgraphs of K which are between a vertex p and the tetrahedron x itself can beseen as the virtual stable sphere S − ( { x, p } ) = S ( x ) ∩ S ( p ) which is astable sphere in the Barycentric refinement. These are all the edgesand triangles which contain p are contained in x . This is a cyclic graph S . In this case, there is only one gap [ a, b ] = [1 , The set of simplices in the complex of all complete subgraphs of K which are different from a fixed vertex p and a fixed triangle t . In thiscase there are two gaps [ a , b ] = [1 ,
1] and [ a , b ] = [3 , P and their join is C = P + P .Again this can be seen as a stable sphere S − ( { x, p } ) of a refinement,where the edge { x, p } has become a vertex.5. The sphere group
In the continuum the monoid of spheres with join operation is notthat interesting as two spheres of the same dimension are topologicallyequivalent. The map dim from (Topologicalspheres , + ,
0) to N is anisomorphism of monoids.This changes completely in a combinatorial setting. The sphere groupis now interesting as there is a countable set of non-isomorophic spheresof each positive dimension. First of all we have to establish that theproperty of being a sphere is invariant under addition given by theZykov join operation.The join operation has the same properties as in the continuum: it isassociative, commutative and preserves spheres. For associativity, notethat the graph A + B + C has the union of the vertex sets as vertices.Two vertices x, y are connected if they were connected in one of thethree components A, B, C or then are in different components. Thejoin of two simplices of dimension n and m is a simplex of dimension n + m + 1: shortly K n + K m = K n + m . The join of two spheres is againa sphere: Lemma 4 (Joins preserve Spheres) . The set (Spheres , +) is a monoid,a groupoid with a neutral element given by the ( − -sphere . The joinof n -sphere and a m -sphere is a n + m + 1 -sphere.Proof. Start that the join of two ( − − n + m of the dimensions.We have to show that every unit sphere of G + H is a sphere and thatremoving a vertex of G + H renders the remaining graph contractible.Given a vertex x in G + H . Assume it is in G . Its unit sphere is the joinof S G ( x )+ H which by induction assumption is a sphere. The case when x is in H , is analog. If we remove the vertex x in G , then by definitionthe graph G \ x is contractible because the join of a contractible graphwith any other unit sphere graph is contractible. (cid:3) The Grothendieck construction produces then a group from this monoid.Lets call it the sphere group. Elements in this groups are equivalenceclasses of pairs A − B of spheres, where A − B ∼ C − D if there existsa sphere K such that A + D + K = B + C + K . The group element O − C with octahedron O for example is P . One can see this alsobecause the octahedron is O = 3 P and the cyclic graph C is 2 P sothat 3 P − P = P .A d -graph is a graph for which every unit sphere is a ( d − d -complex is a finite abstract simplicial complex G for which itsBarycentric refinement G is a d -graph. Corollary 1 (Geometric invariance) . The Barycentric refinement G of a d -complex G is a d -complex.Proof. Every unit sphere S ( x ) is the join of two spheres S − ( x ) and S + ( x ). S − ( x ) is always a sphere. The fact that S ( x ) is a sphere followsin the case of an original vertex by the assumption and for a simplex( x , . . . , x d ) from the fact that S + ( x ) is the intersection of the spheres S ( x k ) which by definition is a sphere. By the previous corollary, also S ( x ) is a sphere. (cid:3) While S − ( x ) is always a sphere as we have seen before, the unstablesphere S + ( x ) is not a sphere in general. For the central vertex p in thestar graph G with n spikes for example, the stable sphere S + ( p ) is P n ,the graph with n vertices and no edges, which is only a sphere if n = 2.In the case of a d -complex G , all unstable spheres S + ( x ) are spherestoo. The unit sphere decomposition S ( x ) = S − ( x ) + S + ( x ) is then theaddition of two spheres. PHERE GEOMETRY AND INVARIANTS 19
But in general, the structure of the stable sphere S + ( x ) also can bedescribed by unit spheres. Lemma 5 (The positive sphere lemma) . If G is the Barycentric re-finement of a simplicial complex G and if x = ( x . . . x k ) with G -vertices x i is a vertex in G , then S + ( x ) = (cid:84) kj =0 S ( x j ) .Proof. We only have to show S + ( x ) = (cid:84) j S ( x j ). Given a vertex y ∈ S + ( x ), then y = x . . . x k y . . . y l . But this is in the intersection. Onother other hand, if we are in the intersection, then we have to be asuper simplex of x , . . . , x k . In other words, we have to be in S + ( x ). (cid:3) Example:
Let G be the octahedron and G its Barycentric refinement.Now, if x = x is a vertex in G which was an original vertex, then S ( x ) = 0 + S ( x ) where 0 is the empty graph. If x = ( x x ) belongedto an edge, then S − ( x ) is a 0-sphere and S + ( x ) is a 0-sphere, as anintersection of two spheres. Now S + ( x ) = S ( x ) ∩ S ( x ).Given a vertex x in G we define the second positive sphere S + ( x ) aspart of the unit sphere S ( x ) consisting of vertices z in G which whenwritten as a simplex Z in G contains the simplex X corresponding to x . Lemma 6 (Second positive sphere) . Given a vertex x = ( x , . . . , x k ) in G with x ≤ · · · ≤ x k ) . The second positive sphere S + ( x ) in G is graph isomorphic to the Barymetric refinement of S + ( y ) in G with y = x k .Proof. Given z ∈ S + ( x ), then x k ≤ z and z = ( x , . . . k , x k +1 , . . . x l ).This point z is in S + ( y ) = (cid:84) j S ( x j ).On the other hand, given w ∈ S + ( y ), then this is a simplex containing y so that ( x , . . . , x k , w ) this defines a new vertex z in G which isconnected to y . It belongs therefore to S ( x ). (cid:3) Let x be a vertex in G with x = ( x , . . . , x k ) in G with dim( x ) = k so that S ( x ) is the join of a ( k − S − ( x ) and S + ( x ). Wehave defined S + ( x ) and S − ( x ) for vertices x . Assume the vertex x in G belongs to the simplex x = ( x . . . x k ) in G . Define S − ( x ) as thesimplex ( x . . . x k ) equipped with the ( k − S + ( x ) = (cid:84) j S ( { x j } ) in G . The graph S − ( x ) + S +1 ( x ) is the virtualunit sphere of the simplex x in G . It is topologically equivalent to S ( x ) in G . We will see in a moment that it is equivalent to an actualunit sphere S ( y ), where y = x k . First a simple property of Barycentric refinement:
Lemma 7 (Intersection complexes) . Let
H, K be subcomplexes of G sothat H , K are subgraphs of G . Then ( H ∩ K ) = H ∩ K .Proof. Given a simplex x in H ∩ K . This means x = { x , ..., x k } is a setwhich is both in H and K and all are connected. Now, also ( x , ..., x k )is a simplex in H as well as in K . The simplex is in the intersection.on the other hand, given a vertex x in H ∩ K . Then x is both in H and K , meaning that x = { x , . . . , x k } is a face in H and in K . (cid:3) For every vertex x in G corresponding to a vertex ( x , . . . , x k ) ∈ G ,each of the elements x j are vertices in G . The following lemma makesa step to relate ( x ) with a new unit sphere S ( y ). Lemma 8. S + ( x ) with G is the Barycentric refinement of a virtualpositive sphere in G which agrees with the positive sphere S + ( y ) in G .Proof. ( x , x , . . . , x k ) ⊂ G . The positive sphere S +2 ( x ) is the Barycen-tric refinement of (cid:84) S ( x k ) in G . Each x k corresponds to a sim-plex X k in G . We have now to find a vertex y ∈ G such that (cid:84) S ( x j ) is equivalent to S + ( y ). Such an y corresponds to a simplex( y , y , . . . , y k ) in G . We can construct that k -simplex from the sim-plices X , . . . , X k inside G . The simplices X k are all contained in eachother since the x j are all connected. Assume without loss of generalitythat things are numbered so that X ⊂ X · · · ⊂ X k . We can nowpick y ∈ X , y ∈ X \ X , y ∈ X \ X , . . . X k \ X k − . Now S + ( y ) isa subgraph of G containing all vertices which belong to simplices in G containing all the y j . S + ( y ) is (cid:84) S ( { y j } ) which is the Barycentricrefinement of (cid:84) j S ( y j ). S + ( x ) is (cid:84) S ( x j ) whose Barycentric refinementis S + ( x ) in G . (cid:3) Examples.1) If x = ( x ) is 0-dimensional and ( x ) is the vertex in G whichbelongs to x , then we can take y = x . The sphere S ( y ) = S + ( y ) hasas a Barycentric refinement the sphere S ( x ) = S + ( x ). If dim( x ) = 1 and x = ( x , x ) then we get the virtual unit sphere S (( x , x )) = P + S + ( x ) in G . The graph S + ( x ) in G is isomorphicto the refinement of S ( x ) ∩ S ( x ).6. The twin propositions
Given a connection graph G (cid:48) of a complex G , we can either remove avertex x in V ( G (cid:48) ) = V ( G ) or we can add a vertex x . We have seenalready that adding a vertex changes the Fredholm determinant by a PHERE GEOMETRY AND INVARIANTS 21 factor i ( x ). It is a bit surprising but in both cases, the Fredholm char-acteristic ψ ( G ) = det(1 + A (cid:48) ( G )) gets multiplied by an integer.Here is again the case when adding a cell. It is a bit more generalthan before as we now do not insist that the new cell has a unitsphere smaller dimensional cells only. The proof is the same as be-fore and uses the concept of a valuation on simplicial complexes. Thisis a functional X on subcomplexes, which satisfies the linearity con-dition X ( H ∪ K ) + X ( H ∩ K ) = X ( K ) + X ( H ) (we can not write X ( H ∆ K ) = X ( K ) + X ( H ) as H ∆ K = H + K is not a complex anymore but it explains the name linearity as the symmetric difference ∆is the addition in a larger Boolean ring of structures). The prototypeof a valuation is the Euler characteristic. As in the continuum, wherevaluations are defined on convex subsets of Euclidean space (which nat-urally come with a simplicial complex structure). It is important to seethat a valuation is not a functional on subsets like in measure theoryand indeed valuations do not extend to the measure theory level ex-cept in geometric measure theory, but where we deal with a Fermionicanalogue of valuations.In the discrete also, it is not just a functional on subsets. The Eulercharacteristic for example depends on the choice of simplicial complexstructure. The Euler characteristic is the prototype of a valuation.Actually there is a (dim( G ) + 1)-dimensional space of valuations bydiscrete Hadwiger and the Euler characteristic is up to a scaling factorthe only which is a combinatorial invariant: the Barycentric operatoron f -vectors has only one eigenvalue 1 and eigenvector (1 , − , , . . . ).Repeating the Barycentric refinement leads to a central theorem [22].In the proof, we again make use of the fact that the only valuation whichassumes the value 1 on all simplices must be the Euler characteristic. Proposition 1 (Attaching cells) . Let G be a simplicial complex and G (cid:48) its connection graph. If we chose a sub-complex H of G leading toa subgraph H (cid:48) of G (cid:48) and attach a new cell x to it, leading to a newcomplex G (cid:48) ∪ H x , then the Fredholm characteristic ψ of G + x getsmultiplied by a factor i ( x ) = 1 − χ ( H ) : ψ ( G ∪ H x ) = (1 − χ ( H )) ψ ( G ) . Proof. (i) The map H → X ( H ) = ψ ( G ∪ H x ) − ψ ( G ) is a valuation.The reason is geometric: as this is a super count of the number of newpaths which are allowed when adding the vertex x . Since x is a singlevertex and only one path can occupy it, the linearity condition is sat-isfied. (ii) Also the map H → Y ( H ) = X ( H ) / ( − ψ ( G )) is an integer valuedvaluation. The reason is that by unimodularity (or induction assump-tion if we want to use this proposition as a proof), ψ ( G ) is {− , } valued and because when looking at a function of H , the complex G isfixed.(iii) The map H → Y ( H ) takes the value 1 on simplices H . The rea-son is that H ∪ { x } is again a simplex and that we look now at thecollection of all Leibniz paths in G (cid:48) ∪ H { x } which go through x . Now Y ( H ) = 1 is equivalent to X ( H ) = − ψ ( G ) and so ψ ( G ∪ H x ) = 0.The reason why this is zero is because ψ ( K n ) is zero for every completegraph K n with n >
1. every path in G ∪ H x can now be paired withan other path of different parity.(iv) Because all valuations which take the value 1 on simplices must bethe Euler characteristic, we have Y ( H ) = χ ( H ). (cid:3) In the new CW complex, S ( x ) is the unit sphere in the Barycentricrefinement graph of G + x .Now lets start with a complex and write G − x for the complex, wherethe cell x has been removed. Now what matters is the sphere S ( x ) in G and not the sphere in G (cid:48) . The reason is that if we remove the cellx, we remove all links to sub-simplices and super simplices. The proofis very similar. We write it again down so that we see the difference. Proposition 2 (Removing cells) . If G is a simplicial complex and G (cid:48) its connection graph. If we chose a vertex x in G (cid:48) and remove x , leadingto the new complex G (cid:48) − x = ( G − x ) (cid:48) , then the Fredholm characteristicof G − x is the Fredholm characteristic of G multiplied by a factor i ( x ) = 1 − χ ( S ( x )) . ψ ( G − x ) = (1 − χ ( S ( x ))) ψ ( G ) . Proof. (i) The map H → X ( H ) = ψ ( G ) − ψ ( G − x ) is a valuation. (ii)Scale it to Y ( H ) = X ( H ) /ψ ( G ). Again this is integer valued completesubgraphs.(iii) The valuation Y takes the value 1 on complete subgraphs.(iv) As any valuation taking the value 1 on complete subgraphs is Eulercharacteristic, we get Y = χ ( H ). This implies χ ( S ( x )) = X ( S ( x )) /ψ ( G ) =( ψ ( G ) − ψ ( G − x )) /ψ ( G ) = 1 − ψ ( G − x ) /ψ ( G ) and so χ ( S ( x )) − − ψ ( G − x ) /ψ ( G ) which is equivalent to the claim. (cid:3) PHERE GEOMETRY AND INVARIANTS 23 Unit sphere topology
We establish now that the set of unit sphere topologies is stable from G on. Further Barycentric refinements do not produce new spheretopologies: Theorem 1. a) Every unit sphere S ( y ) in G is topologically equivalentto a sphere S ( x ) in G .b) Every unit sphere S ( x ) in G is topologically equivalent to a sphere S ( y ) in G .Proof. a) The Barycentric refinement S ( x ) is isomorphic to S (( x )) in G , if ( x ) is the vertex in G which corresponds to the singleton vertex( x ).b) Given now a unit sphere S ( x ) in G . It can be decomposed as S − ( x ) + S + ( x ). If x = ( x , . . . , x k ) is the simplex in G which belongsto the vertex x in G , take y = x k . We have seen that S − ( y ) is asphere of the same dimension than S − ( x ). It is therefore topologicallyequivalent. Furthermore, we have seen that S + ( x ) is the Barycentricrefinement of S − ( x ). It is also topologically equivalent. (cid:3) Examples:1) If G is a one dimensional graph, then the set of unit spheres of G and the set of unit spheres of G agree as we have S ( x ) = P d ( x ) with d ( x ) meaning the vertex degree. But this is special as the Barycentricrefinement of a graph without edges is the graph itself. If G is a graph with only one dimensional spheres, then this is truetoo for G . For an octahedron or icosahedron from example, the unitspheres are cyclic graphs. After parametric refinement, this remains so.The next lemma assures that the join operation honors topologicalequivalence. Now this is obvious if we look at topological realizationsin some Euclidean space. Since we don’t want to use the functor oftopological realizations of abstract simplicial complexes to actual sim-plicial complexes, and since we don’t need topological equivalence inthe sense of topology [19], lets for the following just take a weaker no-tion of equivalence. We only look at the Euler characteristic becausethis is the only thing we really need for establishing the Barycentricinvariance of the Green function diagonal values.So, we say that two finite abstract simplicial complexes are topologi-cally equivalent if their Euler characteristic agree: Lemma 9.
Given two pairs of simplicial complexes
A, B and
C, D . If χ ( A ) = χ ( C ) and χ ( B ) = χ ( D ) , then χ ( A + C ) = χ ( B + D ) . Proof.
This follows from fact that G → − χ ( G ) is a group homomor-phism. 1 − χ ( A + C ) = (1 − χ ( A ))(1 − χ ( C )). (cid:3) Note however that the join does not commute with the Barycentricrefinement:
Examples.1)
The Barycentric refinement ( C + C ) of the join C + C is agraph with 80 vertices. It is larger than the join of the refinements C + C = ( C ) + ( C ) which is a graph with 16 vertices. Still, bothare 3-spheres. The Barycentric refinement ( P + P ) of the utility graph P + P is a graph with 15 vertices and 18 edges. The join of the Barycentricrefinements is the utility graph itself: ( P ) + ( P ) = P + P . Remarks:1)
More generally if H ∼ H (cid:48) and K ∼ K (cid:48) , then H ∼ K is equivalent to H (cid:48) ∼ K (cid:48) . This could be shown most conveniently using an embedding. It is not true that H + K = ( H + K ) as the example of H = C , K = P shows. In this case both are equivalent 2-spheres.8. Prime and prime connection graphs
Given an integer n ≥
2, we can look at two graphs G n and H n . Bothhave the vertex set set { , · · · , n } . In the prime graph G n case, ver-tices are connected, if one divides the other. In the prime connectiongraph H n , two vertices are connected, if they have a common divisorlarger than 1. The graph G n is part of the Barycentric refinement ofthe complete graph on the set of primes and the graph H n is part ofthe connection graph of the complete graph on the set of primes.We can now express the sphere spectrum of G n in terms of the ad-jacency matrix A (cid:48) of H n . It is an application of the connection wehave built between the Green function values, the diagonal values ofthe inverse of the Fredholm matrix, and the Euler characteristic of thecorresponding sphere. Corollary 2. − χ G n ( S ( x )) = (1 + A H n ) − xx .Proof. Given n , let P = { p , . . . , p k } be the primes which appear asfactors of square free integers in V n = { , . . . , n } . This defines now aabstract finite simplicial complex G . The sets are the subsets of G n such that their product is in V n . Because for any given A and any non-empty subset B of A also the product of the elements in B is in V n , PHERE GEOMETRY AND INVARIANTS 25 this is indeed an abstract finite simplicial complex. The Barycentricrefinement G of G is now the graph G n (there should be no confusionsince n > G to be mixed upwith G n ). The connection graph G (cid:48) is the graph H n . (cid:3)
235 67 1011 13141517 1921 22 2326 2930 3133 3435 373839 4142 4346 47515355 57 5859 61 6265 6667 6970 71 73 7477 78 79 828385 8687 8991 93 9495 97 - - - - - - - -
20 0 - - - -
10 0 -
10 01 0 - - - - -
11 1 - - -
11 1 1 - - - - - - - - - - Figure 5.
We see the prime graph G , first withthe vertex labels given by square free integers. In thepicture to the right we see the sphere spectrum i ( x ) = 1 − χ ( S ( x )). For primes x , we have i ( x ) = 1. The unit sphereof x = 2 is a graph with vertex set { , , , , , , , , , , , , , , , , , , } , which has v = 10 connectivity components. There is furthermorea nontrivial loop 6 , , , , , , b = 1.There are no triangles in S ( x ) so that χ ( S ( x )) = 1 − (10 −
1) = − G n .9. Graph arithmetic
The multiplicative monoid ( N , · ) = ( { , , , . . . , } , · ) has the unique fac-torization property, telling that every non-unit can be factored uniquelyinto a product of prime numbers. It is a unique factorization monoid.More generally, the multiplicative group of any unique factorizationdomain is a unique factorization monoid. The additive monoid of { , , , . . . , } also has unique prime factorization but it is trivial sinceonly 1 is an additive prime and any number n naturally can be decom-posed as n = 1 + 1 + 1 + · · · + 1. The question of prime factorization can be extended to networks. It turns out that there is both an ad-dition and multiplication which are compatible by distributivity. Wework here with graphs even so one can take more general simplicialcomplexes.The set G of finite simple graphs G = ( V, E ) becomes an additivemonoid with the join operation G + H . The definition is as follows:we have V ( G + H ) = V ( G ) ∪ V ( H ) and E ( G + H ) = E ( G ) ∪ E ( H ) ∪{ ( a, b ) | a ∈ V ( G ) , b ∈ V ( H ) } . The 0 element is the empty graph( ∅ , ∅ ). Adding one to G is the cone over G and adding the 0-sphere P to G is the suspension of G .The maximal dimension dim( G ) of a graph G is the dimension of thelargest complete subgraph K d +1 which appears in G . The numberdim( G )+1 is also called the clique number. The relation dim( H + K ) =dim( H ) + dim( H ) + 1 follows from K n + K m = K n + m .(Less obvious is the same inequality for the inductive dimension dim( G ) = (cid:80) v ∈ V ( G ) [1 + dim( S ( x ))] / | V | ):dim( G + H ) ≥ dim( G ) + dim( H ) + 1 . Lets call a graph G an additive prime graph if it can not be writtenas G = H + K with H, K different from the 0 element. An exampleof an additive prime graph is K . All the K n can be factored as K n = K + K + K = K n and are therefore not prime. A star graph S n canbe factored as P n + K where P n is the graph with n vertices and noedges. Lemma 10.
Every finite simple graph G can be decomposed into ad-ditive prime factors G = p + · · · + p n , where p i are prime graphs.Proof. If there is a factor H , then its maximal dimension has to besmaller. By induction, we can then break up every factor into smallerfactors or leave it if it if it is an additive prime. (cid:3) Lemma 11.
There are many additive prime graphs: every disconnectedgraph is an additive prime, every non-contractible graph with a primevolume is an additive prime.Proof.
The join of two graphs is connected so that a disconnected graphmust be prime. Every circular graph C n with n > P k is prime. For the circular graph, one can see that if C n = K + H , then both K, H have to have maximal dimension 0. But wecan list P + P = K , P + P = L , P + P = C . Any other P k + P l PHERE GEOMETRY AND INVARIANTS 27 contains a star graph S k or S l . If both k, l >
2, then this is incompatiblewith C n . (cid:3) As we can list all one-dimensional graphs which are not prime we seethat most one dimensional graphs are prime. For example, every treewhich is not a star graph is prime.As in the case N , the question about unique prime factorization is moredifficult than the question of the existence of a factorization. Exper-iments indicate that graphs form a unique factorization monoid: stillthe question remains: is it true that every graph has a unique factor-ization G = p + · + p n into prime graphs p k .Note that this is not true for the join in the standard topology case as ifwe take the join of a k -sphere and a m -sphere, we get a k + m +1 sphere.So, we can for example take a 3-sphere and add a 4-sphere, which givestopologically the same than adding a 2-sphere to a 5 sphere. In eachcase, we get a 8-sphere. In the discrete C = P + P but C can notbe written as a sum.Here is a formula which gives the f -vector for the join of two complexes.Given a simplicial complex with v k simplices of dimension k . Define theextended f -vector generating function = extended Euler polynomial of G as f G ( x ) = 1 + (cid:88) k =0 v k x k +1 . The empty graph has the generating function f G ( x ) = 1. (Originally,I had defined f G ( x ) = 1 /x + (cid:80) v k x k and defining the multiplication as xf G ( x ) g G ( x ) but June-Hou Fung pointed out to me that a multiplica-tion with x simplifies the generating function.) Proposition 3.
The join G + H of two complexes has the f -generatingfunction of the form f G + H = f G f H . Proof.
Given a simplex x in G and a simplex y in H , this produces asimplex x + y of dimension dim( x ) + dim( y ) + 1. (cid:3) This formula can shed light on the unique prime factorization problem.If G + H can be factored, then its rational function can be factored inthe above sense. We have then for example f p + p + p = f p f p f p . But even if we have a unique factorization algebraically in the polyno-mials with multiplication f g ( x ) = f ( x ) g ( x ), we still have the problemto see whether the factors can be realized as simplicial complexes andfurthermore require that the factors have positive entries.Since χ ( G ) = 1+ f ( − χ ( G + H ) = 1+( − f G ( − f H ( −
1) =1 − ( χ ( G ) − χ ( H ) −
1) = χ ( G ) + χ ( H ) − χ ( G ) χ ( H ): Corollary 3.
The Euler characteristic of the join of two complexessatisfies χ ( G + H ) = χ ( G ) + χ ( H ) − χ ( G ) χ ( H ) . Of especial interest is the Euler-Poincar´e functional i ( G ) = − f ( −
1) =1 − χ ( G ). (Again, it was June-Hou Fung who pointed out to me thatthe above corollary implies that 1 − χ ( G ) is multiplicative). From theabove formula we see that it behaves like a multiplicative character onthe group of graphs. However, since we have to define i ( − G ) = 1 /i ( G )to extend it to the entire group, we can not define i ( − G ) for graphs G which have i ( G ) = 0. We see especially that i is not defined onthe “integers” Z given by the graphs K n and their negatives − K n as i ( K n ) = 0. Corollary 4.
For any two simplicial complexes
G, H , the functional i ( G ) = 1 − χ ( G ) satissfies i ( G + H ) = i ( G ) i ( H ) . While i ( G ) can be zero, we can look at the subgroup of the join groupfor which i ( G ) (cid:54) = 0. This is still a group. Corollary 5.
The monoid of graphs with join + has the following sub-monoids:a) the submonoid of graphs with even χ ( G ) b) the submonoid of graphs with odd χ ( G ) c) the submonoid of graphs with zero χ ( G ) d) the submonoid of graphs with χ ( G ) ∈ { , } e) the submonoid of sphere graphsf ) the submonoid of graphs with i ( G ) (cid:54) = 0 .Proof. This all follows from the sum formulas for χ and i as well as thefact that the join leaves spheres invariant. (cid:3) Corollary 6. i ( G ) can be extended to a multiplicative character on theadditive group of spheres. PHERE GEOMETRY AND INVARIANTS 29
Proof.
It takes there the values ± i ( − G ) = i ( G ). Now i ( G − H ) = i ( G ) /i ( H ). The functional i takes values in {− , } . (cid:3) Examples.1) If F = P n , G = P m are both 0-dimensional point graphs, then f ( P n ) f ( P m ) = (1 + nx )(1 + mx ) = 1 + nx + mx + nmx showingthat the graph F + G has nm edges and n + m vertices and Eulercharacteristic n + m − nm . Looking at the number of edges, we seeimmediately that a 1-dimensional graph with a prime number of edgesmust be prime. More generally, any graph with prime volume must beprime. If G + H is 2-dimensional, then one of the graphs, say G has to be0-dimensional and so f G ( x ) = 1 + nx . If f H ( x ) = 1 + mx + kx is theextended Euler polynomial of a 1-dimensional graph. There are f = kn triangles and e = k + mn edges and v = m + n vertices. We see also theEuler formula confirmed as χ ( G + H ) = χ ( G ) + χ ( H ) − χ ( G ) χ ( H ) = n + ( m − k ) − n ( m − k ) which agrees with v − e + f . If G, H are two spheres, where at least one has Euler characteristic 2,then their product has Euler characteristic 2. Only if both have Eulercharacteristic 0, it is possible that the product has Euler characteristic2. We can define the monoid of all graphs with Euler characteristicin { , } . We could also look at the sub-monoid of all graphs witheven Euler characteristic, or the monoid with odd Euler characteristic,or the set of graphs with 0 Euler characteristic. These graphs do notchange the Euler characteristic of the graph it is multiplied with. Lets look at two 1-spheres C n and C m . Their product is a 3-spherewith f C n + C m ( x ) = (1 + nx + nx )(1 + mx + mx ). The new f -vectortherefore is ( m + n, m + n + mn, mn, mn ) which has zero Euler char-acteristic as (cid:80) k ( − k v k = 0.A special case of the unique prime factorization problem in the spheremonoid is the question whether any 3-sphere is either prime or thena unique sphere of two 1-spheres. In this simple case, we can answerunique factorization: it boils down to the question whether a + b = m + n, ab = mn implies m = a, n = b or m = b, n = a which is true.But this could be a case for the Guy law of small numbers. We couldimagine for example that there would be a 7-sphere which can be writ-ten as p + p + p with prime 2-spheres p i but also be able to write itas q + q as a sum of two prime 3-spheres. We have just not seen suchan example yet and it probably does not exist. As class field theory in rings of integers like Z [ √−
5] shows, manymonoid belonging to rings of integers in function fields do not havea unique prime factorization and this could be the case here too.
Remarks.1)
The join G + H can be obtained within an extended Stanley-Reisnerring as follows. If V = { x , . . . , x n } and W = { y , . . . , y n } represent thevertices of the graph G and H and f = 1 + (cid:80) I a I x I , g = 1 + (cid:80) J b J x J represent the complex for G and H , then f g represents the complexfor G + H . For example, if f = 1 + x + y + z + xy + yz + xz + xyz represents the Whitney complex of G = K and g = 1 + a + b + ab represents the complex for H = K , then f g = 1 + x + y + z + a + b + xa + xb + ya + yb + za + zb + xya + xyb + xza + xzb + yza + yzb + xyza represents the complex G + H . Trying to generalize a modular arithmetic, one could try to modifythe join operation to subgraphs of a given background graph P anddefine the graph with V ( G + H ) = V ( G )∆ V ( H ) and E ( G + H ) con-taining all edges contained in V ( G )∆ V ( H ) or then connecting two dif-ferent vertices a ∈ V ( G ) ∩ V ( G )∆ V ( H ) with b ∈ V ( H ) ∩ V ( G )∆ V ( H ).Mathematically we would look at a Stanley-Reisner ring with Z valuedcoefficients over a fixed number of variables. We have not succeeded toget finite rings like that. Any bipartite graph is a subgraph of P n + P m as by definition, wecan split the vertices into two disjoint subsets. The graph P n + P m isthe most extreme case. The utility graph P + P is the most famousexample.Finally, lets look at the relation between the Fredholm characteristic: Corollary 7.
On the sub monoid of graphs with even Euler character-istic, the Fredholm characteristic is multiplicative: Then ψ ( G + H ) = ψ ( G ) ψ ( H ) . This happens especially on spheres so that on that group ψ becomes a character when extended to negative graphs with ψ ( − G ) =1 /ψ ( G ) .Proof. Check cases:If ψ ( G ) = ψ ( H ) = 1, then ψ ( G ∗ H ) = 1.If ψ ( G ) = ψ ( H ) = −
1, then ψ ( G ∗ H ) = 1.If ψ ( G ) = − ψ ( H ) = 1, then ψ ( G ∗ H ) = − (cid:3) Corollary 8.
In the sub-monoid of graphs with odd Euler character-istic, then ψ ( G + H ) = − . This implies that if ψ ( G ) = 1 , then G isprime in that monoid. PHERE GEOMETRY AND INVARIANTS 31
Figure 6.
We see the addition of a star graph S andan octahedron graph. Since the star graph is a cone S = P + K = P + 1 and the Octahedron graph factorsinto Oct = K + C = K + K + K = K · K , thegraph to the right has the additive prime decomposition P + K + K + K + K . In this case, the additive primedecomposition is unique. In all cases we have seen so far,the composition is unique, indicating that ( G , +) couldbe a unique factorization monoid. Proof.
Check cases:If ψ ( G ) = ψ ( H ) = 1, then ψ ( G + H ) = − ψ ( G ) = ψ ( H ) = −
1, then ψ ( G + H ) = − ψ ( G ) = − ψ ( H ) = 1, then ψ ( G + H ) = − (cid:3) Examples.1)
Among complete graphs only K has Fredholm characteristic 1. Allothers have Fredholm characteristic − There is a two dimensional graph G with the topology of the pro-jective plane. Its f -vector is (15 , ,
28) so that ψ ( G ) = 1. Its Eulercharacteristic is 1 too. Therefore, G must be prime. A tree with an even number of edges must be prime in the monoidof odd Euler characteristic graphs.When mentioning the arithmetic of graphs to An Huang, he asked meabout relations of the spectrum of the join and the individual compo-nents. Here are three remarks:
Lemma 12. If G = H + K then there is an eigenvalue | V ( H ) | + | V ( K ) | of the scalar Laplacian L . Consequently, any graph G for which thereis no eigenvalue v ( G ) for L is prime.Proof. If m = | V ( H ) | and n = | V ( K ) | , then the vector which is con-stant n on H and constant m on K is an eigenvector to the Laplacian L with eigenvalue n + m . The eigenvector is perpendicular to the con-stant. (cid:3) Corollary 9.
The graph nG = G + G + · · · + G has the eigenvalue n | V ( G ) | with multiplicity at least n − .Proof. The multiplicity follows because we can write the sum in n − A + B . The proof above determinethen the eigenvector. (cid:3) The example K n = K + K + . . . K , where n appears with multiplic-ity n − n − nP = P + ... + P which has the eigenvalue 2 n withmultiplicity n − λ of the Laplacian. Lemma 13. λ ( G + H ) = min( | V ( H ) | , | V ( K ) | ) + min( λ ( G ) , λ ( H ) .Proof. The Courant-Fischer formula tells λ = inf v · ( v, Lv ) / ( v, v ).Let v ( H ) be the eigenvector of L ( H ) with eigenvalue λ ( H ) and v ( G )the eigenvector of L ( H ) with eigenvalue λ ( G ). Assume they are nor-malized. Extend v ( G ) and v ( H ) onto G + H by putting 0 on the otherpart. They still have norm 1 on the product space. They are also per-pendicular to the constant vector. Now ( v ( H ) , Lv ( H )) = λ ( H ) + m and ( v ( G ) , Lv ( G )) = λ ( G ) + n . This shows λ ( G + H ) ≤ min( | V ( H ) | , | V ( K ) | ) + min( λ ( G ) , λ ( H ) . On the other hand, since G + H can be obtained from the disjointunion of G and P n (the graph without edges), by adding edges, wehave λ i ( G ) + n ≤ λ i ( G + H ). Similarly λ i ( H ) + m ≤ λ i ( G + H ). SeeCorollary 4.4.2. in [33]. (cid:3) Here is an other spectral result. It deals with the highest form Lapla-cian. Let D = d + d ∗ be the Dirac operator of a simplicial complex. Theform Laplacian L = ( d + d ∗ ) splits into blocks called form Laplacians L k , the restriction of L on discrete k -forms. The nullity of L k is the k ’th Betti number b k [17, 16]. We have already seen that if we have twographs G, H , then the volume of G + H is the product of the volumesof G and volumes of H . Lets write L v ( G ) for L dim( G ) ( G ) for the volumeLaplacian, the Laplacian belonging to the largest dimension. Now if G has volume m and H has volume n then we can look at the volumeeigenvalues λ , . . . , λ n of L v ( G ) and the volume eigenvalues µ , . . . , µ m of L v ( H ). What are the volume eigenvalues of H + G ? Lemma 14.
For two arbitrary simplicial complexes
G, H , the volumeeigenvalues of G + H are given by λ + µ , where λ runs over the volumeeigenvalues of G and µ runs over the volume eigenvalues of H . PHERE GEOMETRY AND INVARIANTS 33
Proof. If U is the volume Laplacian of G and K is the volume Laplacianof H . Given U v = λv and V w = µw . The volume Laplacian W of G + H is a ( nm ) × ( nm ) matrix because the volumes of G and K havemultiplied. Define on the facet xy of G + H joining the facets x and y of G and H the value f xy = v x w y . Now W f = λf + µf . (cid:3) A ring of networks
The group ( X, + ,
0) with Zykov addition + (graph join) has a compati-ble multiplication. We formulate it only in the graph case, meaning thecase of simplicial complexes which are Whitney complexes of graphs,even so it would also work for general simplicial complexes.Given two graphs G = ( V, E ) , H = ( W, F ), define the graph G · H =( V × W, Q ), where the vertex set V × W is the Cartesian product ofthe two vertex sets and where the edge set Q consists of all pairs { (( a, b ) , ( c, d ) | ( a, c ) ∈ E or ( b, d ) ∈ F } . This means that we connect ( a, b ) with any ( c, x ) if ( a, c ) ∈ E and withany ( a, b ) with ( x, d ) if ( b, d ) ∈ F . This product is obviously associativeand commutative. Lemma 15.
This product is compatible with the join addition in thesense that the distributivity law G · ( H + K ) = G · H + G · K holds.Proof. Like in the case of the ring of integers, spacial geometric insightcan help by placing the graphs H,K and H+K into one coordinate axesand G onto the other. Now both G · H + G · KandG · ( H + K ) haveas the vertex set the product sets of the vertices. As connections ina sum H+K consist of three types, connections within H, connectionswithin K and any possible connection between H and K, two points( a, b ) , ( c, d ) are connected if either ( a, c ) is an edge in H , or ( b, d ) is anedge in K or then if either a, c or b, d are in different graphs. (cid:3) Examples :All equality signs mean here “graph isomorphic”. a) P n · P m = P nm b) K n + K m = K n + m c) K n · K m = K nm d) G is cone over G e) C n = W n wheel graph f ) P + G suspension of G g) P + C = P + P + P = K · P octahedron graph: Oct h) K , = P + P = K · P utility graph i) K n,m = P n + P m complete bipartite graph j) P n = S n star graph k) S = 1 + 1 + P n = K + P n windmill graph i) K · C = C + C three sphere j) K · Oct = Oct + Oct + Oct is an 7-dimensional sphere k) K n,m + K n,m = K · K n,m = K ( P n + P m ) = K n,n + K m,m l) W + P = 1 + C + 1 + C = K + C + C . + = Figure 7.
Working in the field generated by the ring ofnetworks is analogues to work with fractions. In schoolarithmetic, we start with the monoid N = { , , , , . . . } ,then introduce the integers Z , then build multiplicationleading to the ring of integers and finally produce thefield of fractions. Here we see the identity 1 / / / Q . + = Figure 8.
While the field of networks contains thefield Q as K n + K m = K n + m and K n · K m = K nm , wecan work with more general graphs. Here we see theresult of the addition of two fractions: 1 /C + S /K ,where C is the circular graph and S is the star graph.The addition is done as in school arithmetic. We get K · K + S · C = K + S · C in the nominator and C · K in the denominator.One of the consequences of the fact that we know the maximal eigen-value of G + H is that we can get the maximal eigenvalue of somegraphs immediately. Corollary 10.
For n > , the maximal eigenvalue of K n · G is n | V ( G ) | . PHERE GEOMETRY AND INVARIANTS 35
This implies for example that the maximal eigenvalue of K n is n for n > K n,m is n + m for n, m ≥ λ k ≤ | V ( G ) | for all eigenvalue. It also shows thatbecause K n P is the n − n . Indeed, the octahedron K · P for examplehas maximal eigenvalue 6 and the 16-cell K · P has maximal eigenvalue8 with multiplicity at least 3. Figure 9.
The graph 1 + P + P = S + P is thesuspension of the star graph S = 1 + P . As a non-prime graph, its maximal eigenvalue is the number ofvertices, which is 7. Since P has only 0 eigenvalues, thegraph S = 1 + P has λ = 1. According to the spectrallemma for λ above, we know that P + S has eigenvalue2 + 0 + 1 = 3. 11. Questions
Lets finish with some questions. A) We still have not identified the off diagonal terms (1 + A ( G (cid:48) )) − xy in topological terms, where G (cid:48) is the connection graph of a complex.Having the diagonal terms related to natural sphere index functionals,which are very closely related to Morse theory, it is likely that also the Figure 10.
The graph K · L = K · (1 + P ) = K + K · P = K + Oct = 1 + 1 + 1 + Oct is thesum of the triangle and the octahedron but it is also3 times the linear graph L of length 2. It is also thesuspension of the cone of the octahedron or the cone ofa 3 sphere. As the graph is not an additive prime, itsmaximal eigenvalue is the number of vertices, which is 9.The multiplicity result above only assures the maximaleigenvalue to appear with multiplicity 2. The eigenvalue9 actually appears with multiplicity 5.off diagonal terms refer to some interesting topological or dynamicalnotion. B) Since we have identified two characters i ( G ) = 1 − χ ( G ) and ψ ( G )on the sphere group (a homomorphism from the sphere group to thegroup {| z | = 1 }| z ∈ C } ), it is natural to ask whether there is morerepresentation theory of the sphere group for which the character areclassical characters, represented by a trace of a matrix associated to asphere. We especially would like to know the collection of all characters.Since the sphere group is discrete, one could take any distance (likethe minimal number of vertices or edges to be modified to get fromone to an isomorphic image of the other). This is then a topological PHERE GEOMETRY AND INVARIANTS 37 group which has a Haar measure as discreteness assures that it is locallycompact and such that the Haar measure, the counting measure. Thereare many characters on the sphere group like assigning real numbers k ( x ) to every vertex x ∈ V ( G ) of a graph G and − k ( x ) to vertices of − G . A character is then defined as X ( G − H ) = (cid:89) x ∈ V ( G ) e πik ( x ) / (cid:89) x ∈ V ( H ) e πik ( x ) . Since adding two graphs takes the union of the vertices, this is multi-plicative for the additive group of spheres. But as both i ( G ) and ψ ( G )tap into the structure of the simplices in the graph, it is likely thatthere are more interesting examples of characters. C) The join and multiplication in which the addition is not extendedyet, forms a commutative semiring of type (2 , , , a · D) We would also like to be able to compute in the field of frac-tions defined by the ring and then do a completion solve equationslike G + G = C leading to G = C / G · G = K with so-lution √ K . Both equations can not be solved in the ring because C is an additive prime (every cyclic graph C n with n > K is a multiplicative prime because every graph K p with p vertices is a multiplicative prime if p is a rational prime. E) There are interesting questions related to the number theoreticallydefined prime graphs G n and prime connection graphs H n . One canfor example try to estimate the Green function values from above. Thetopology of G n is linked the Riemann hypothesis as it is directly linkedto the growth of the Euler characteristic of G n by the Mertens con-nection [25]. The Fredholm characteristic of truncations of truncatedprime graphs on the other hand gives a signature, comparing the par-ity of square free integers ≤ n which have an even number of primefactors. Such integers correspond to odd-dimensional simplices in thecomplex. The Green function values for prime graphs, the matrix en-tries of (1 + A ( H n )) − xy could contain some interesting number theory. References [1] U. Andrews. Undecidable problems in topology. Online, May 10, 2015.[2] R. Bott. Two new combinatorial invariants for polyhedra.
Portugaliae Math. ,11:35–40, 1952.[3] F. Chung and S.-Y. Yau. Discrete Green functions.
Journal of CombinatorialTheory (A) , 91:191–214, 2000.[4] A.V. Evako. Dimension on discrete spaces.
Internat. J. Theoret. Phys. ,33(7):1553–1568, 1994.[5] R. Forman. Morse theory for cell complexes.
Adv. Math. , page 90, 1998.[6] B. Gr¨unbaum. Polytopes, graphs, and complexes.
Bull. Amer. Math. Soc. ,76:1131–1201, 1970.[7] F. Harary.
Graph Theory . Addison-Wesley Publishing Company, 1969.[8] A. Hatcher.
Algebraic Topology . Cambridge University Press, 2002.[9] A. Ivashchenko. Contractible transformations do not change the homologygroups of graphs.
Discrete Math. , 126(1-3):159–170, 1994.[10] J. Jonsson.
Simplicial Complexes of Graphs , volume 1928 of
Lecture Notes inMathematics . Springer, 2008.[11] M. Joswig. Computing invariants of simplicial manifolds.https://arxiv.org/abs/math/0401176, 2004.[12] J. Kahn, M. Saks, and D. Sturtevant. A topological approach to evasiveness.
Combinatorica , 4(4):297–306, 1984.[13] D.A. Klain and G-C. Rota.
Introduction to geometric probability . LezioniLincee. Accademia nazionale dei lincei, 1997.[14] O. Knill. The dimension and Euler characteristic of random graphs.http://arxiv.org/abs/1112.5749, 2011.[15] O. Knill. A graph theoretical Poincar´e-Hopf theorem.http://arxiv.org/abs/1201.1162, 2012.[16] O. Knill. The McKean-Singer Formula in Graph Theory.http://arxiv.org/abs/1301.1408, 2012.[17] O. Knill. The Dirac operator of a graph.http://http://arxiv.org/abs/1306.2166, 2013.[18] O. Knill. Coloring graphs using topology.http://arxiv.org/abs/1410.3173, 2014.[19] O. Knill. A notion of graph homeomorphism.http://arxiv.org/abs/1401.2819, 2014.[20] O. Knill. Graphs with Eulerian unit spheres.http://arxiv.org/abs/1501.03116, 2015.[21] O. Knill. The K¨unneth formula for graphs.http://arxiv.org/abs/1505.07518, 2015.[22] O. Knill. Universality for barycentric subdivision.http://arxiv.org/abs/1509.06092, 2015.[23] O. Knill. Gauss-Bonnet for multi-linear valuations.http://arxiv.org/abs/1601.04533, 2016.[24] O. Knill. On Fredholm determinants in topology.https://arxiv.org/abs/1612.08229, 2016.[25] O. Knill. On primes, graphs and cohomology.https://arxiv.org/abs/1608.06877, 2016.
PHERE GEOMETRY AND INVARIANTS 39 [26] F.H. Lutz. Triangulated manifolds with few vertices and vertex-transitivegroup actions. Dissertation, Technische Universitaet Berlin, 1999.[27] F.H. Lutz. Triangulated manifolds with few vertices and combinatorial mani-folds.https://arxiv.org/abs/math/0506372, 2005.[28] M. Tsuruga M. Joswig, F.H.Lutz. Sphere recognition: Heuristics and examples.https://arxiv.org/abs/1405.3848, 2015.[29] B. Poonen. Undecidable problems: a sampler. In J. Kennedy, editor,
Inter-preting G¨odel: Critical Essays , pages 211–241. Cambridge University Press,2014.[30] J. Rhodes and P.V. Silva.
Boolean Representations of Simplicial Complexesand Matroids . Springer Monographs in Mathematics, 2015.[31] C.P. Rourke and B.J. Sanderson.
Introduction to Piecewise-Linear Topology .Springer Verlag, Berlin, 1982.[32] E.H. Spanier.
Algebraic Topology
Bul-letin (New Series) of the AMS , 6, 1982.[35] Wu W-T. Topological invariants of new type of finite polyhedrons.
Acta Math.Sinica , 3:261–290, 1953.[36] A.H. Wallace.
An introduction to Algebraic Topology . Pergamon Press, 1957.[37] A.A. Zykov. On some properties of linear complexes. (russian).
Mat. SbornikN.S. , 24(66):163–188, 1949., 24(66):163–188, 1949.