Modular Cohomology Rings of Finite Groups¶

AUTHORS:

Simon King <simon.king@uni-jena.de> (Cython and Python code, porting, maintainance)
David Green <david.green@uni-jena.de> (underlying C code)

IMPLEMENTATION:

Our package is formed by Singular- and Gap-functions, as well as Python-, Cython- and C-code. It links against SharedMeatAxe, which is a fork of MeatAxe, and uses the Small Groups library of Hans Ulrich Besche, Bettina Eick and Eamonn O’Brien at run time.

Modular cohomology rings of a finite group \(G\) are available via the constructor CohomologyRing(). The real work behind the scenes is done in the modules factory, cohomology, modular_cohomology, barcode, cochain, resolution and dickson.

In the first section of our introduction, we demonstrate the usage of our package. After outlining the functionality, we go into more detail and illustrate various ways of defining a cohomology ring, computing the ring structure, and working in the cohomology ring. Last, we discuss how to save and load a cohomology ring on a local file system and how to export and import it to a different platform.

In the second section, we give an overview on the theoretical background of our package. The emphasis is on the construction of parameters for the cohomology ring, which can then be used to prove completeness of the ring structure by means of completeness criteria that are due to different authors ([Benson], [GreenKing], [King], [Symonds]). We illustrate this with step-by-step examples.

An overview of the various options that can be used is given in the documentation of CohomologyRing().

In a nutshell

One starts with a finite group G (either a group in GAP, a pair of numbers denoting an address in the SmallGroups library, or a string defining the group) and a prime number p
One configures the cohomology ring by H = CohomologyRing(G, prime=p). The prime can be ommitted if G is a prime power group.
One makes the cohomology ring by H.make().

In most cases, the computation of H doesn’t involve further action. However, sometimes particular options help with the computation. This documentation describes the underlying theory, the available options, reveals the underlying computational steps, and of course documents further data that this package can compute.

EXAMPLES:

1. Synopsis¶

Our package is devoted to the cohomology rings of finite groups with coefficients in \(\mathbb F_p\) (the finite field with \(p\) elements, where \(p\) is a prime less than 255 that divides the group order). Note that the first computation of the modular cohomology rings of all groups of order 128 and the first computation of the mod-2 cohomology ring of the third Conway group where obtained with our package; see [GreenKing] and [KingGreenEllis] for more details. We computed the mod-\(p\) cohomology rings for various interesting finite non prime power groups and different primes \(p\). Our results are available here. The non prime power groups are not part of our web repository yet.

We expect the groups to be given either by their address in the Small Groups Library, or as a group in Sage’s Gap C library interface. In the latter case and if \(G\) happens to be a prime power group, it is required that the list of generators of \(G\) starts with a minimal generating set. If this is not the case, a ValueError is raised. Internally, the group will be transformed into a permutation group, which, in some cases, is very difficult. So, it is recommended to start with a permutation group right away.

The package is shipped with the cohomology rings for all groups of order 64; the data are stored in folders and can simply be retrieved by calling CohomologyRing() with appropriate keys. These data are provided to all users of the Sage installation and we refer to it as the local sources.

In addition, we provide the cohomology rings for all groups of order 128 and for all but six groups of order 243 in a web repository. Unless the user requires a recomputation from scratch, these data will be automatically downloaded. In future versions, this will also be possible for various non prime power groups.

Any user can construct new databases; we refer to them as a workspace. At each time, there only is a single workspace, and it is used to store all data created in cohomology computations. It is possible to switch between workspaces. There is a default location in a sub-directory of DOT_SAGE. But any location may be chosen as well, provided that the user has write permission.

Cohomology rings are parent structures and are cached. That means, if two groups are essentially equal then their cohomology rings will be identic (and not only equal). Note for two groups being “essentially equal”, it does not suffice that they are isomorphic. All groups are supposed to have a fixed list of generators. Two groups are “essentially equal”, if there exists a group isomorphism that sends the list of generators of one group to an initial segment of the list of generators of the other group. Moreover, two cohomology rings can only be identic if they have the same location of data storage.

When a cohomology ring is created, the following locations are searched, in order:

It is first tried to find the result in the cache.
It is tried to load it from the current workspace.
If it is not there, the default is to try to find it in the local sources.
Finally, by default it is attempted to download the result from a web repository, which we refer to as remote sources.

Data from either local or remote sources are mirrored in the workspace, which is because the user does not necessarily have write permissions to alter the local sources. To be precise, symbolic links are created in the workspace that link to data in the local sources. If further computation requires more data, they are created and stored in the workspace. If all this fails, a new ring is initialized in the current workspace.

The data belonging to the cohomology ring of a prime power group (e.g., a minimal free resolution, information on embedding of elementary abelian subgroups, standard monomials etc.) are distributed in a folder tree, one tree for each group.

In the non prime power case, the situation is easier. It is not needed to store small data pieces during the computation, and eventually the main data are saved in a single file. Its default location is in the current workspace. The computation does, however, rely on the cohomology rings of various subgroups, e.g., a Sylow subgroup. So, again the data are distributed.

Creating a cohomology ring¶

In the following subsections, we explain here how to describe the group whose modular cohomology ring shall be computed.

There are also various global options that influence the algorithm and the use of databases. For example, it might be instructive to use logging; the log gives detailed information on how data are computed (the “debug” logging level is of course even more verbous). Global options can be changed at any time using the method global_options().

Usually, the package attempts to make use of existing data. However, if the optional parameter from_scratch is set to True then this cohomology ring is not loaded from local or remote sources. Note, however, that this option only takes effect on the ring itself, but not for cohomology rings of certain subgroups that might be created during the computation.

Further options are described in the documentation of CohomologyRing().

… using the SmallGroups library¶

In our examples, we put the workspace into a temporary directory. Of course, in reality, a permanent directory would be chosen instead. We do so in order to avoid a conflict with an existing workspace.

sage: from pGroupCohomology import CohomologyRing
sage: CohomologyRing.doctest_setup()       # reset, block web access, use temporary workspace
sage: H0 = CohomologyRing(8,3)

Since the cohomology ring of SmallGroup(8,3), which is the Dihedral Group of order 8, is in the local sources shipped with our package, it is simply loaded and already completely computed. It is not needed to specify a prime to determine the coefficients, since the group is of prime power order. Note that we also have a list of group names and group descriptions – hence, H0 knows that it belongs to the Dihedral Group of order 8:

sage: H0
H^*(D8; GF(2))
sage: print(H0)
Cohomology ring of Dihedral group of order 8 with coefficients in GF(2)
<BLANKLINE>
Computation complete
Minimal list of generators:
[c_2_2: 2-Cocycle in H^*(D8; GF(2)),
 b_1_0: 1-Cocycle in H^*(D8; GF(2)),
 b_1_1: 1-Cocycle in H^*(D8; GF(2))]
Minimal list of algebraic relations:
[b_1_0*b_1_1]

The generators are of type COCH. It is also possible to convert the cohomology ring into a quotient ring in the Singular interface:

sage: singular(H0)
polynomial ring, over a field, global ordering
// coefficients: ZZ/2
// number of vars : 3
//        block   1 : ordering M
//                  : names    c_2_2 b_1_0 b_1_1
//                  : weights      2     1     1
//                  : weights     -1     0     0
//                  : weights      0    -1     0
//        block   2 : ordering C
// quotient ring from ideal
_[1]=b_1_0*b_1_1

In Sage, parent structures should be unique. And indeed, cohomology rings of the same group in the same workspace are unique:

sage: H0 is CohomologyRing(8,3)
True
sage: H0 is loads(dumps(H0))
True

We set up a different workspace in a temporary directory, and show that indeed two cohomology rings are only identical if their data are stored in the same place. Moreover, we use an optional argument to avoid that the ring is loaded from the local or remote sources:

sage: tmp = tmp_dir()
sage: CohomologyRing.set_workspace(tmp)
sage: H1 = CohomologyRing(8,3,from_scratch=True)
sage: H1 is H0
False

Since we used the option from_scratch=True, it was not attempted to download the cohomology ring from local or remote sources. Hence, beyond the basic setup, H1 is unknown and is thus not even equal to H0:

sage: H0 == H1
False

We now compute H1 (see below for the theoretical background) and verify that then H0 is equal (though not identical) to H1:

sage: H1.make()
sage: H0 == H1
True

Here is a similar example, using a non prime power group. In this case, we need to provide the modulus by the optional parameter prime.

sage: HS6a = CohomologyRing(720,763,prime=2)
sage: HS6a.make()
sage: print(HS6a)
Cohomology ring of SmallGroup(720,763) with coefficients in GF(2)
<BLANKLINE>
Computation complete
Minimal list of generators:
[c_2_1: 2-Cocycle in H^*(SmallGroup(720,763); GF(2)),
 c_1_0: 1-Cocycle in H^*(SmallGroup(720,763); GF(2)),
 b_3_3: 3-Cocycle in H^*(SmallGroup(720,763); GF(2)),
 c_3_2: 3-Cocycle in H^*(SmallGroup(720,763); GF(2))]
Minimal list of algebraic relations:
[b_3_3*c_3_2+c_2_1*c_1_0*b_3_3]
sage: singular(HS6a)
polynomial ring, over a field, global ordering
// coefficients: ZZ/2
// number of vars : 4
//        block   1 : ordering M
//                  : names    c_2_1 c_1_0 b_3_3 c_3_2
//                  : weights      2     1     3     3
//                  : weights     -1    -1     0    -1
//                  : weights     -1     0     0     0
//                  : weights      0    -1     0     0
//        block   2 : ordering C
// quotient ring from ideal
_[1]=b_3_3*c_3_2+c_2_1*c_1_0*b_3_3

… using a group in the GAP interface¶

As of version 3.1, the old pexpect GAP interface has been replaced by the C-library interface libgap. So, in the following, GAP interface refers to libgap.

Above, we mentioned that two cohomology rings are identic only if the corresponding groups are essentially equal. Indeed, if we represent the Dihedral Group of order 8 in a different way then we obtain a different ring presentation. Note that, when referring to a group in the Gap interface, we need to provide a name for that group, in one or the other way. That name also determines the location of data storage, and thus if two different names are provided then the corresponding cohomology rings are not the same:

sage: G = libgap.DihedralGroup(8)
sage: H2 = CohomologyRing(G,GroupName='DihedralA')
sage: G.SetName("DihedralB")
sage: H3 = CohomologyRing(G)
sage: H2 is H3
False

Of course, when H2 and H3 are completely computed, they are isomorphic to H1, since they belong to isomorphic groups. However, they do not have the same ring presentation. This is because the presentation of the cohomology ring depends on the choice of a minimal generating set of the group; and the minimal generating sets of SmallGroup(8,3) and of DihedralGroup(8) are essentially different:

sage: H2.make()
sage: H3.make()
sage: H2 == H3
True
sage: H2 == H1
False

Indeed, the minimal relations of H1 and H2 look different:

sage: print(H1)
Cohomology ring of Dihedral group of order 8 with coefficients in GF(2)
<BLANKLINE>
Computation complete
Minimal list of generators:
[c_2_2: 2-Cocycle in H^*(D8; GF(2)),
 b_1_0: 1-Cocycle in H^*(D8; GF(2)),
 b_1_1: 1-Cocycle in H^*(D8; GF(2))]
Minimal list of algebraic relations:
[b_1_0*b_1_1]
sage: print(H2)
Cohomology ring of DihedralA with coefficients in GF(2)
<BLANKLINE>
Computation complete
Minimal list of generators:
[c_2_2: 2-Cocycle in H^*(DihedralA; GF(2)),
 b_1_0: 1-Cocycle in H^*(DihedralA; GF(2)),
 b_1_1: 1-Cocycle in H^*(DihedralA; GF(2))]
Minimal list of algebraic relations:
[b_1_1^2+b_1_0*b_1_1]

But of course, the rings are isomorphic.

Above, we computed the cohomology of some non prime power group. It is in fact a symmetric group:

sage: G = libgap.SymmetricGroup(6)
sage: HS6b = CohomologyRing(G, prime=2, GroupName='SymmetricGroup(6)', from_scratch=True)
sage: HS6b.make()
sage: print(HS6b)
Cohomology ring of SymmetricGroup(6) with coefficients in GF(2)
<BLANKLINE>
Computation complete
Minimal list of generators:
[c_2_1: 2-Cocycle in H^*(SymmetricGroup(6); GF(2)),
 c_1_0: 1-Cocycle in H^*(SymmetricGroup(6); GF(2)),
 b_3_2: 3-Cocycle in H^*(SymmetricGroup(6); GF(2)),
 b_3_3: 3-Cocycle in H^*(SymmetricGroup(6); GF(2))]
Minimal list of algebraic relations:
[b_3_2*b_3_3]

Obviously the two ring presentations for the mod-2 cohomology of the symmetric group of rank 6 are different. This is since the ring presentation will essentially depend on the choice of generators of the Sylow subgroup (which here is chosen by the program).

Computations in a cohomology ring¶

Ring theoretic invariants¶

Up to now, we implemented the computation of four ring theoretic invariants, namely the dimension, the depth, the \(a\)-invariants and the Poincaré series:

sage: H2.dimension()
2
sage: H2.depth()
2
sage: H2.a_invariants()
[-Infinity, -Infinity, -2]
sage: H2.poincare_series()
1/(t^2 - 2*t + 1)

Of course, they are the same for both presentations of the cohomology ring of the dihedral group of order 8:

sage: H1.dimension()
2
sage: H1.depth()
2
sage: H1.a_invariants()
[-Infinity, -Infinity, -2]
sage: H1.poincare_series()
1/(t^2 - 2*t + 1)

Actually we implemented two essentially different ways of computating the Poincaré series. The method _poincare_without_parameters() is used if the filter degree type hasn’t been computed. Of course, one can invoke it directly:

sage: H2._poincare_without_parameters()
1/(t^2 - 2*t + 1)

After computing the ring structure, generators and relations of the cohomology ring are present. Generator number zero is the multiplicative unit of the underlying field. The other generators are represented by cochains (COCH, or MODCOCH for non prime power groups):

sage: H1.gens()
[1, c_2_2: 2-Cocycle in H^*(D8; GF(2)), b_1_0: 1-Cocycle in H^*(D8; GF(2)), b_1_1: 1-Cocycle in H^*(D8; GF(2))]

The relations are given as strings:

sage: H1.rels()
['b_1_0*b_1_1']

The relations can be interpreted as elements of the cohomology ring, and indeed they vanish:

sage: print(H1(H1.rels()[0]))
2-Cocycle in H^*(D8; GF(2)),
represented by
[0 0 0]

We can also compute the nil radical, i.e., the ideal formed by all nilpotent elements. It is given in the Singular interface, as an ideal in a free graded commututive ring. So, modulo the relation ideal, the returned ideal canonically maps onto the nil radical. For the dihedral group of order 8, the nil radical vanishes:

sage: H1.nil_radical()
0
sage: H2.nil_radical()
0

Here are the data for the mod-2 cohomology of the symmetric group of degree 6:

sage: HS6a.dimension()
3
sage: HS6a.depth()
3
sage: HS6a.a_invariants()
[-Infinity, -Infinity, -Infinity, -3]
sage: HS6a.poincare_series()
(-t^2 + t - 1)/(t^5 - 2*t^4 + t^3 - t^2 + 2*t - 1)

Up to normalisation, we get the same Poinvaré series when computing it with a different method:

sage: HS6a._poincare_without_parameters()
(t^2 - t + 1)/(-t^5 + 2*t^4 - t^3 + t^2 - 2*t + 1)
sage: HS6a.nil_radical()
0

Induced maps¶

As we have seen before, the presentation of the cohomology ring depends on the choice of a minimal generating set of the underlying finite \(p\)-group (resp. Sylow \(p\)-subgroup). For our purposes, we always assume that a finite group has a fixed set of generators, and we call two groups equivalent, if there is a group isomorphism that maps the given list of generators of one group to an initial segment of the list of generators of the other group.

group() returns a permutation group in the gap interface equivalent to the one that was used to set up the cohomology ring:

sage: G1 = H1.group()
sage: G1
Group([ (1,2)(3,8)(4,6)(5,7), (1,3)(2,5)(4,7)(6,8) ])
sage: G2 = H2.group()
sage: G2
Group([ (1,2)(3,8)(4,6)(5,7), (1,3,4,7)(2,5,6,8) ])

Apparently, G1 and G2 are different permutation groups. But in fact they are isomorphic. Note that we could get a group isomorphism by Gap’s function IsomorphismGroups. But the result is not unique. In order to have reproducible doc tests, we provide an explicit isomorphism:

sage: phi = libgap.eval('GroupHomomorphismByImages(Group([(1,2)(3,8)(4,6)(5,7),(1,3)(2,5)(4,7)(6,8)]), Group([(1,2)(3,8)(4,6)(5,7),(1,3,4,7)(2,5,6,8),(1,4)(2,6)(3,7)(5,8) ]), [(1,2)(3,8)(4,6)(5,7), (1,3)(2,5)(4,7)(6,8)], [(1,6)(2,4)(3,5)(7,8), (1,8)(2,7)(3,6)(4,5)])')
sage: phi.IsInjective()
true
sage: phi.IsSurjective()
true

Of course, the group isomorphism induces an isomorphism of the corresponding cohomology rings. After ensuring that phi is printed nicely, we obtain the induced map:

sage: phi.SetName("phi")
sage: phi_star = H2.hom(phi,H1)
sage: phi_star
phi^*

Indeed, we can verify that the image of the relation of H2 is zero in H1:

sage: H2.rels()
['b_1_1^2+b_1_0*b_1_1']
sage: phi_star(H2('b_1_1'))^2 + phi_star(H2('b_1_0'))*phi_star(H2('b_1_1'))
(phi^*(b_1_1))**2+(phi^*(b_1_0))*(phi^*(b_1_1)): 2-Cocycle in H^*(D8; GF(2))
sage: print(phi_star(H2('b_1_1'))^2 + phi_star(H2('b_1_0'))*phi_star(H2('b_1_1')))
2-Cocycle in H^*(D8; GF(2)),
represented by
[0 0 0]

We can also compute the inverse of phi_star, compose the two maps, and see that it is the identity map:

sage: phi_star_inv = phi_star^(-1)
sage: phi_star_inv
(phi^*)^(-1)
sage: [(phi_star_inv*phi_star)(X) == X for X in H2.gens()]
[True, True, True, True]
sage: [(phi_star*phi_star_inv)(X) == X for X in H1.gens()]
[True, True, True, True]

It is possible to convert an induced homomorphism into a map of quotient rings in the Singular interface. This allows for working with cohomology ring elements of very high degrees. However, it is always needed to take care of the basering in Singular:

sage: S_phi_star = singular(phi_star)
sage: singular(H2).set_ring()
sage: I = singular.ideal(['c_2_2^50','b_1_0^50','b_1_1^50'])
sage: singular(H1).set_ring()
sage: S_phi_star(I)
b_1_1^100+c_2_2^2*b_1_1^96+c_2_2^16*b_1_1^68+c_2_2^18*b_1_1^64+c_2_2^32*b_1_1^36+c_2_2^34*b_1_1^32+c_2_2^48*b_1_1^4+c_2_2^50,
b_1_1^50+b_1_0^2*b_1_1^48+b_1_0^16*b_1_1^34+b_1_0^18*b_1_1^32+b_1_0^32*b_1_1^18+b_1_0^34*b_1_1^16+b_1_0^48*b_1_1^2+b_1_0^50,
b_1_1^50

Note that Singular does not do automatic reduction in quotient rings. So, eventually we do the reductions explicitly, in two ways, with the same result:

sage: S_phi_star(I).reduce(singular('ideal(0)'))
b_1_1^100+c_2_2^2*b_1_1^96+c_2_2^16*b_1_1^68+c_2_2^18*b_1_1^64+c_2_2^32*b_1_1^36+c_2_2^34*b_1_1^32+c_2_2^48*b_1_1^4+c_2_2^50,
b_1_1^50+b_1_0^50,
b_1_1^50
sage: singular(H2).set_ring()
sage: I = I.reduce(singular('ideal(0)'))
sage: singular(H1).set_ring()
sage: S_phi_star(I)
b_1_1^100+c_2_2^2*b_1_1^96+c_2_2^16*b_1_1^68+c_2_2^18*b_1_1^64+c_2_2^32*b_1_1^36+c_2_2^34*b_1_1^32+c_2_2^48*b_1_1^4+c_2_2^50,
b_1_1^50+b_1_0^2*b_1_1^48+b_1_0^16*b_1_1^34+b_1_0^18*b_1_1^32+b_1_0^32*b_1_1^18+b_1_0^34*b_1_1^16+b_1_0^48*b_1_1^2+b_1_0^50,
b_1_1^50+b_1_0*b_1_1^49+b_1_0^16*b_1_1^34+b_1_0^17*b_1_1^33+b_1_0^32*b_1_1^18+b_1_0^33*b_1_1^17+b_1_0^48*b_1_1^2+b_1_0^49*b_1_1
sage: S_phi_star(I).reduce(singular('ideal(0)'))
b_1_1^100+c_2_2^2*b_1_1^96+c_2_2^16*b_1_1^68+c_2_2^18*b_1_1^64+c_2_2^32*b_1_1^36+c_2_2^34*b_1_1^32+c_2_2^48*b_1_1^4+c_2_2^50,
b_1_1^50+b_1_0^50,
b_1_1^50

For a non prime power group, a permutation representation equivalent to the given group, a subgroup which was used for the stable element method and a Sylow \(p\)-subgroup are available as well:

sage: HS6a.group()
Group([ (1,2), (1,2,3,4,5,6) ])
sage: HS6a.subgroup()
Group([ (1,4)(2,5), (1,4)(2,3)(5,6), (1,4)(2,5)(3,6), (2,5)(3,6) ])
sage: HS6a.sylow_subgroup()
Group([ (1,4)(2,5), (1,4)(2,3)(5,6), (1,4)(2,5)(3,6) ])

The dihedral group of order 8 is a subgroup of the symmetric group of rank 6. We compute the ring homomorphism that is induced by the embedding. In order to get a reproducible result, we define the group homomorphism explicitly:

sage: phi = libgap.eval('GroupHomomorphismByImages( Group( [ (1,2)(3,8)(4,6)(5,7), (1,3)(2,5)(4,7)(6,8) ] ), Group( [ (1,2), (1,2,3,4,5,6) ] ), [ (1,2)(3,8)(4,6)(5,7), (1,3)(2,5)(4,7)(6,8) ], [ (1,3), (1,2)(3,4) ] )')
sage: phi_star = HS6a.hom(phi,H0)
sage: [H0.element_as_polynomial(phi_star(x)) for x in HS6a.gens()]
[1: 0-Cocycle in H^*(D8; GF(2)),
 b_1_1^2+b_1_0^2+c_2_2: 2-Cocycle in H^*(D8; GF(2)),
 b_1_0: 1-Cocycle in H^*(D8; GF(2)),
 c_2_2*b_1_0: 3-Cocycle in H^*(D8; GF(2)),
 b_1_0^3+c_2_2*b_1_1+c_2_2*b_1_0: 3-Cocycle in H^*(D8; GF(2))]
sage: singular(H0).set_ring()
sage: singular(phi_star)
s...[1]=b_1_1^2+b_1_0^2+c_2_2
s...[2]=b_1_0
s...[3]=c_2_2*b_1_0
s...[4]=b_1_0^3+c_2_2*b_1_1+c_2_2*b_1_0

Here is the kernel of the induced map:

sage: phi_star.preimage()
b_3_3+c_2_1*c_1_0+c_1_0^3,
c_1_0*c_3_2+c_2_1*c_1_0^2
sage: phi_star(HS6a('b_3_3+c_2_1*c_1_0+c_1_0^3')).as_polynomial()
'0'
sage: phi_star(HS6a('c_1_0*c_3_2+c_2_1*c_1_0^2')).as_polynomial()
'0'

Massey products¶

We can compute (set valued) higher Massey products and the (single valued) restricted Massey powers introduced by [Kraines] for cocycles of prime power groups. For non prime power groups, we have so far only an experimental implementation for restricted Massey powers.

We first study some examples on H0 (cohomology of the Dihedral Group of order 8) that we have defined above. Note that we are transforming the returned sets into sorted lists, in order to have reproducible doc tests:

sage: sorted(list(H0.massey_products(H0.2,H0.3,H0.2)))
[0: 2-Cocycle in H^*(D8; GF(2)), b_1_0^2: 2-Cocycle in H^*(D8; GF(2))]
sage: sorted(list(H0.massey_products(H0.2,H0.3,H0.2,H0.3)))
[c_2_2: 2-Cocycle in H^*(D8; GF(2)),
 b_1_1^2+c_2_2: 2-Cocycle in H^*(D8; GF(2)),
 b_1_0^2+c_2_2: 2-Cocycle in H^*(D8; GF(2)),
 b_1_1^2+b_1_0^2+c_2_2: 2-Cocycle in H^*(D8; GF(2))]
sage: sorted(list(H0.massey_products(H0.2*H0.1,H0.3*H0.1,H0.2*H0.1)))
[0: 8-Cocycle in H^*(D8; GF(2)),
 c_2_2^3*b_1_0^2: 8-Cocycle in H^*(D8; GF(2)),
 c_2_2^2*b_1_0^4: 8-Cocycle in H^*(D8; GF(2)),
 c_2_2^2*b_1_0^4+c_2_2^3*b_1_0^2: 8-Cocycle in H^*(D8; GF(2)),
 c_2_2*b_1_0^6: 8-Cocycle in H^*(D8; GF(2)),
 c_2_2*b_1_0^6+c_2_2^3*b_1_0^2: 8-Cocycle in H^*(D8; GF(2)),
 c_2_2*b_1_0^6+c_2_2^2*b_1_0^4: 8-Cocycle in H^*(D8; GF(2)),
 c_2_2*b_1_0^6+c_2_2^2*b_1_0^4+c_2_2^3*b_1_0^2: 8-Cocycle in H^*(D8; GF(2))]

If one is interested in just a single element of the Massey products, one can use the option all=False:

sage: H0.massey_products(H0.2*H0.1,H0.3*H0.1,H0.2*H0.1,H0.3*H0.1, all=False)
{c_2_2^5: 10-Cocycle in H^*(D8; GF(2))}

By the \(i\)-th restricted Massey power of a cocycle \(C\) in a cohomology ring with coefficients \(\mathbb F_p\), we mean the restricted \(p^i\)-fold Massey product of \(C\) as introduced by D. Kraines. If it is defined, then it is a single cocycle:

sage: H = CohomologyRing(27,3)
sage: H.make()
sage: H.8
a_3_4: 3-Cocycle in H^*(E27; GF(3))
sage: H.8.massey_power()
<a_3_4; 1>: 8-Cocycle in H^*(E27; GF(3))
sage: H.element_as_polynomial(_)
b_2_0^2*a_1_1*a_3_5+b_2_0^2*a_1_0*a_3_4+b_2_0*c_6_8: 8-Cocycle in H^*(E27; GF(3))

It is known that in odd degree and for \(p>2\) it can be expressed as the negative of the composition of the first Steenrod power with the Bockstein operator. This allows for a verification of the above result, see massey_power() for details.

Massey products allow to distinguish certain isomorphic cohomology rings, so, they contain information that go beyond the ring structure. An easy example is provided by the cohomology rings of the cyclic groups of order 3 and order 9. The rings are isomorphic.

sage: HC3 = CohomologyRing(3,1)
sage: HC3.make()
sage: HC9 = CohomologyRing(9,1)
sage: HC9.make()
sage: print(HC3)
Cohomology ring of Small Group number 1 of order 3 with coefficients in GF(3)
<BLANKLINE>
Computation complete
Minimal list of generators:
[c_2_0: 2-Cocycle in H^*(SmallGroup(3,1); GF(3)),
 a_1_0: 1-Cocycle in H^*(SmallGroup(3,1); GF(3))]
Minimal list of algebraic relations:
[]
sage: print(HC9)
Cohomology ring of Small Group number 1 of order 9 with coefficients in GF(3)
<BLANKLINE>
Computation complete
Minimal list of generators:
[c_2_0: 2-Cocycle in H^*(SmallGroup(9,1); GF(3)),
 a_1_0: 1-Cocycle in H^*(SmallGroup(9,1); GF(3))]
Minimal list of algebraic relations:
[]

Note that the relation a_1_0^2 is not listed, since the relation \(x^2=0\) holds for any cocycle \(x\) of odd degree, for \(p>2\), and is thus somehow redundant. The 1st restricted Massey power of the degree one generator is non-trivial for the cyclic group of order 3, but only the 2nd restricted Massey power of the degree one generator of the cohomology ring of the cyclic group of order 9 is non-trivial:

sage: HC3.element_as_polynomial(HC3.2.massey_power())
-c_2_0: 2-Cocycle in H^*(SmallGroup(3,1); GF(3))
sage: HC9.element_as_polynomial(HC9.2.massey_power())
0: 2-Cocycle in H^*(SmallGroup(9,1); GF(3))
sage: HC9.element_as_polynomial(HC9.2.massey_power(2))
-c_2_0: 2-Cocycle in H^*(SmallGroup(9,1); GF(3))

Bar Codes¶

Cohomology rings contain much information on a group. Unfortunately, it is, in general, quite difficult to test whether two graded commutative rings are isomorphic. Hence, it is of interest to consider invariants that can be more easily compared, such as the Poincaré series or the depth. Bar codes are not ring theoretic invariants, but are defined in terms of the finite group whose cohomology is being studied.

Any sequence of group homomorphisms \(\phi_i: G_i \to G_{i+1}\) with \(i=1,...,n\) gives rise to a series of induced homomorphisms of cohomology rings. Note that we insist that the sequence starts with the group and ends with the trivial group. It is easy to get such sequences in the case of prime power groups (e.g., upper central series, derived series, \(p\)-central series). For non prime power groups, this may be more difficult.

Persistent group cohomology, introduced in [EllisKing], asks how long cocycles in a given degree “survive” being mapped by the induced homomorphisms. In our paper, we observe a certain correspondence to periodicity of coclass trees, and suggest that persistent group homology may be an interesting theoretical tool. In addition, it can be seen as an invariant of groups.

For a given degree \(d\), the persistent cohomology can be described by an upper triangular matrix of non-negative integers, where entry \(i,j\) (\(i\le j\)) is the dimension of the image of the degree \(d\) part of \(H^*(G_j)\) in \(H^*(G_i)\) under the induced homomorphism given by the composition of \(\phi_i,\phi_{i+1},...,\phi_{j-1}\), including the case \(i=j\) in which the matrix entry simply gives the dimension of the degree \(d\) part of \(H^*(G_i)\).

As usual, the sequence of dimensions sorted by degree gives rise to a Poincaré series. Hence, an upper triangular matrix of rational functions results and this provides information for all degrees.

We like to emphasize that the persistence matrices are very easy to compare and provide a quite strong tool for distinguishing groups. While the cohomology ring only depends on the group algebra, the bar codes may contain additional information on the group itself, by taking into account the normal series. However, the following example shows that bar codes contain information that even goes beyond the isomorphism types of the normal subgroups and quotient groups.

sage: H158 = CohomologyRing(64,158)
sage: H160 = CohomologyRing(64,160)

These cohomology rings are part of our database, hence, they are already completely known. The Poincaré series, the \(a\)-invariants, the degrees of generators and of relations of the cohomology rings coincide:

sage: H158.poincare_series()
(t^4 + t^3 + t^2 + t + 1)/(t^6 - 2*t^5 + 3*t^4 - 4*t^3 + 3*t^2 - 2*t + 1)
sage: H160.poincare_series()
(t^4 + t^3 + t^2 + t + 1)/(t^6 - 2*t^5 + 3*t^4 - 4*t^3 + 3*t^2 - 2*t + 1)
sage: H158.degvec
[4, 4, 1, 1, 1, 3, 3]
sage: H160.degvec
[4, 4, 1, 1, 1, 3, 3]
sage: H158.set_ring()
sage: [singular('deg(%s)'%r) for r in H158.rels()]
[2, 2, 3, 3, 4, 4, 5, 6, 6, 6]
sage: H160.set_ring()
sage: [singular('deg(%s)'%r) for r in H160.rels()]
[2, 2, 3, 3, 4, 4, 5, 6, 6, 6]
sage: H158.a_invariants()
[-Infinity, -Infinity, -2]
sage: H160.a_invariants()
[-Infinity, -Infinity, -2]

We consider here the bar codes associated with the upper central series. It turns out that the non-trivial terms of the upper central series and the resulting factor groups are isomorphic:

sage: G158 = libgap.SmallGroup(64,158)
sage: G160 = libgap.SmallGroup(64,160)
sage: [(G.IdGroup(), (G158/G).IdGroup()) for G in G158.UpperCentralSeries()]
[([ 64, 158 ], [ 1, 1 ]),
 ([ 16, 2 ], [ 4, 2 ]),
 ([ 4, 2 ], [ 16, 11 ]),
 ([ 1, 1 ], [ 64, 158 ])]
sage: [(G.IdGroup(), (G160/G).IdGroup()) for G in G160.UpperCentralSeries()]
[([ 64, 160 ], [ 1, 1 ]),
 ([ 16, 2 ], [ 4, 2 ]),
 ([ 4, 2 ], [ 16, 11 ]),
 ([ 1, 1 ], [ 64, 160 ])]

Nevertheless, the groups can be distinguished using the bar codes associated with the upper central series:

sage: B158 = H158.bar_code('UpperCentralSeries')
sage: B160 = H160.bar_code('UpperCentralSeries')
sage: B158 == B160
False
sage: B158[1]==B160[1]
True
sage: B158[2]==B160[2]
True
sage: B158[3]==B160[3]
False

Indeed, the bar codes differ in degree 3; graphically:

sage: ascii_art(B158[3])
        *
        *
      *-*
      *-*
      *
      *
      *
      *
      *
      *
    *-*
    *-*
    *
    *
  *
  *
  *
  *
*
*
*
*
sage: ascii_art(B160[3])
        *
        *
      *-*
      *
      *
      *
      *
      *
      *
      *
    *-*-*
    *-*
    *
  *-*
  *
  *
  *
*
*
*
*

These pictures (bar codes) can be interpreted as follows. Let \(G\) be a finite \(p\)-group and let \(G=G_0 > G_1 > ... > G_n > 1\) be a normal series; in our example, we have \(n=2\). The first \(n+1\) columns of the bar code correspond to the normal subgroups groups \(G_n, G_{n-1},..., G_0\), while the last \(n\) columns correspond to the factor groups \(G/G_n, G/G_{n-1},..., G/G_1\). We consider the sequence of group homomorphisms given by inclusions and quotients. The stars in the bar code correspond to basis vectors of the degree \(d\) part of the cohomology rings of the respective groups. Stars are connected by a line (i.e., a hyphen) if the corresponding basis vectors are mapped to each other by the induced maps (which of course go from right to left).

In terms of the persistence matrix:

sage: B158[3].matrix()
[ 4  0  0  0  0]
[ 0  4  0  0  0]
[ 0  0  4  2  0]
[ 0  0  0 10  2]
[ 0  0  0  0  4]
sage: B160[3].matrix()
[ 4  0  0  0  0]
[ 0  4  1  0  0]
[ 0  0  4  2  1]
[ 0  0  0 10  2]
[ 0  0  0  0  4]

The bar codes in degree 3 can also be computed directly, and of course the result coincides with the bar codes obtained from the “full” bar codes (based on Poincaré series):

sage: B158[3] == H158.bar_code('UpperCentralSeries',degree=3)
True
sage: B160[3] == H160.bar_code('UpperCentralSeries',degree=3)
True

Since apparently the (2,4) entries of the persistence matrices differ, we show here how the Poincaré series in that position look like:

sage: B158.matrix()[2,4]
2*t^2 + 2*t + 1
sage: B160.matrix()[2,4]
t^3 + 2*t^2 + 2*t + 1

Hence, the iamge of the map \(H^*(G/G_2) \to H^*(G)\) induced by the quotient map contains only finitely many cocycles, where \(G\) is group number 158 or 160 of order 64, and \(G_2\) is the second term of the upper central series.

Essential and Depth Essential Ideals¶

The essential ideal of a cohomology ring is formed by all those elements whose restrictions to all subgroups vanish. The depth essential ideal at rank \(r\) is formed by all those elements that vanish on the centralisers of all \(p\)-elementary abelian subgroups of rank \(r\). These ideals can be computed using essential_ideal() and depth_essential_ideal().

If \(r\) is at most the depth of the cohomology ring, then the depth essential ideal vanishes. It is conjectured by J. Carlson that it is non-zero whenever \(r\) exceeds the depth.

sage: D = CohomologyRing(8,3)
sage: D.make()
sage: Q = CohomologyRing(8,4)
sage: Q.make()
sage: SD = CohomologyRing(16,8)
sage: SD.make()
sage: S6 = CohomologyRing(720,763,prime=2)
sage: S6.make()

The quaternion group is the smallest group that has a non-vanishing essential ideal:

sage: Q.essential_ideal()
a_1_0^2,
a_1_0*a_1_1

The dihedral group and the group of order 720 (which is the symmetric group on 6 elements) have no essential classes, for different reasons:

sage: CohomologyRing.global_options('info')
sage: S6.essential_ideal()
H^*(SmallGroup(720,763); GF(2)):
    Compute essential_ideal
    The group is not of prime power order -- there is no essential ideal
0
sage: D.essential_ideal()
H^*(D8; GF(2)):
    Compute essential_ideal
    The depth exceeds the Duflot bound -- there is no essential ideal
0
sage: CohomologyRing.global_options('warn')

It used to be a conjecture that the product of any two essential classes vanishes. But then it was found that the cohomology of the Sylow 2-subgroup of \(U_3(4)\), which is of order 64, has exactly one essential class that decomposes into essential classes of smaller degrees. See the doc tests of essential_ideal().

Until recently, that was the only known counterexample to the conjecture. However, using this package, it was found that the cohomology ring of the direct product of the cyclic group of order two and the Sylow 2-subgroup of \(U_3(4)\) has the same property.

Last, we illustrate Carlson’s depth essential conjecture with some group of order 64:

sage: H = CohomologyRing(64,23)
sage: H.CenterRk
2
sage: H.depth()
3
sage: H.dimension()
4
sage: H.depth_essential_ideal(2)
0
sage: H.depth_essential_ideal(3)
0
sage: H.depth_essential_ideal(4)
a_1_0,
a_1_1,
a_2_0

Data storage and data relocation¶

Modular cohomology rings of finite groups can be very complicated. When computing the ring structure, many data pieces are needed — but not all pieces at the same time. Therefore we save the various pieces in separate files, and reload them when they are needed. In some examples, the memory and time consumption of saving all data in a single file would not be reasonable. Hence, ‘saving’ a cohomology ring to some file f means that f only contains the information needed to find the files in which the various data pieces are stored.

This method is fast, but has a clear disadvantage: If you wish to transfer your cohomology computations to a different computer, it does not suffice to copy one file for one cohomology ring. All data pieces must be copied. Moreover, when the data are in a different location, then the information stored in f would point to non-existing paths. We tried to address these problems in the least inconvenient way, as explained in the following sub-sections.

Data storage¶

We first describe the situation for prime power groups. We do some new computations from scratch, using a temporary directory.

sage: from pGroupCohomology import CohomologyRing
sage: CohomologyRing.doctest_setup()       # reset, block web access, use temporary workspace
sage: tmp = tmp_dir()
sage: CohomologyRing.set_workspace(tmp)
sage: H = CohomologyRing(8,3,from_scratch=True)
sage: H.make()

By default, when computing a cohomology ring H, the result is automatically saved in a file H.autosave_name(). Cohomology rings are unique parent structures, and therefore reloading from that file only creates a new reference to H:

sage: K = load(H.autosave_name())
sage: K is H
True
sage: loads(dumps(H)) is H
True

By the command CohomologyRing.set_workspace(tmp), all data in the user database are stored in sub-folders of the folder tmp. The names of these folders are composed as follows.

The basic folder tmp is known to H by the attribute root.

sage: os.path.realpath(H.root) == os.path.realpath(tmp)
True

Any cohomology ring has been assigned a “stem name” that determines the names of sub-folders and file names associated with the ring. If the cohomology ring belongs to a group in the Small Groups Library, its stem name is formed using the address in the Small Groups Library. Otherwise, it is obtained from the group name used in the definition of the cohomology ring:
```
sage: H = CohomologyRing(8,3)
sage: G = libgap.DihedralGroup(8)
sage: K = CohomologyRing(G, GroupName = 'DihedralA')
sage: G.SetName("DihedralB")
sage: L = CohomologyRing(G)
sage: H.GStem
'8gp3'
sage: K.GStem
'DihedralA'
sage: L.GStem
'DihedralB'
```
Data belonging to a cohomology ring can be found in a folder that is stored as the attribute “gps_folder”, and is composed by the root and the stem name:
```
sage: os.path.realpath(H.gps_folder) == os.path.realpath(os.path.join(H.root,H.GStem))
True
```

The situation for non prime power groups is different. Here, the data structure for the ring itself is somewhat easier, so, main data are saved in just one file. On the other hand, the computations rely on the stable element method and thus involve the cohomology rings of various subgroups. When reloading, these are sought in the cache or in the workspace or local sources or remote sources, and are re-computed if they can nowhere be found. Here, we show that the rings are cached:

sage: HS6a = CohomologyRing(720,763,prime=2)
sage: HS6a.make()
sage: HS6a is loads(dumps(HS6a))
True
sage: save(HS6a, HS6a.autosave_name())
sage: HS6a is load(HS6a.autosave_name())
True

Data relocation¶

If you wish to transfer data of a cohomology ring H to a different location, you may do so. But please make sure that (in the case of prime power groups) H.gps_folder keeps its name (only the parent directory is allowed to change!), and that all files in H.gps_folder and its sub-folders are preserved. Then, by set_workspace(), one can set up the workspace in the new location, and reload the ring. In order to demonstrate that it really is the same ring, we provide H with an additional attribute before saving and moving the data:

sage: H.setprop('info', 'The cohomology data were moved')
sage: save(H,H.autosave_name())
sage: tmp2 = tmp_dir()
sage: import os
sage: os.rename(tmp,tmp2)
sage: CohomologyRing.set_workspace(tmp2)
sage: M = CohomologyRing(8,3)
sage: M is H
False
sage: print(M)
Cohomology ring of Dihedral group of order 8 with coefficients in GF(2)
<BLANKLINE>
Computation complete
Minimal list of generators:
[c_2_2: 2-Cocycle in H^*(D8; GF(2)),
 b_1_0: 1-Cocycle in H^*(D8; GF(2)),
 b_1_1: 1-Cocycle in H^*(D8; GF(2))]
Minimal list of algebraic relations:
[b_1_0*b_1_1]
sage: M.info
'The cohomology data were moved'

2. The Underlying Theory¶

Let \(R\) be a graded algebra over a field \(K\). An approximation of \(R\) is

a sequence of graded commutative \(K\)-algebras \(\tau_kR\) presentable in degree at most \(k\) (for any positive integer \(k\)), together with
a sequence of \(K\)-algebra homomorphisms \(\alpha_k: \tau_kR\to R\) for \(k=1,2,...\) whose restriction on degree \(d\) is an isomorphism of vector spaces, for any \(d\le k\), and
a \(K\)-algebra homomorphism \(\lambda_k: \tau_kR\to \tau_{k+1}R\) such that \(\alpha_k= \alpha_{k+1}\circ \lambda_k, for \).

We refer to \(\tau_kR\) as an approximation of \(R\) out to degree \(k\). Given a minimal presentation of \(R\) as a graded \(K\) algebra, we obtain an approximation of \(R\) out to degree \(k\) by the quotient of a free graded-commutative algebra whose generators correspond to the generators of \(R\) out to degree \(k\), modulo the two-sided ideal generated by the relations of \(R\) out to degree \(k\).

Modular cohomology rings of finite groups are finitely presentable. Hence, for all sufficiently large \(k\), an approximation of the cohomology ring out to degree \(k\) is actually isomorphic to the cohomology ring. The general scheme of our computations of a cohomology ring is:

Compute an approximation of the cohomology ring in increasing degree.
Use a completeness criterion to detect whether the approximation is isomorphic to the cohomology ring.

Ring approximation in the case of prime power groups¶

Computing a minimal projective resolution¶

[Green] provides an algorithm for the computation of initial segments of a minimal projective resolutions for modular group algebras of finite \(p\)-groups, using non-commutative Groebner bases. He implemented it in C-programs, and we provide a Cython wrapper in pGroupCohomology.resolution.

Chosing generators¶

Knowing a resolution out to degree \(d\) allows for computing a minimal set of cohomology generators out to degree \(d\) and a minimal set of algebraic relations of degree up to \(d\) for these generators. It pays off to choose the generators in a special way, by taking into account the restriction to certain special subgroups \(S_1,...,S_n\). Here, \(S_1\) denotes the greatest central elementary abelian subgroup of \(P\), and \(S_2,...,S_n\) denotes representatives for the conjugacy classes of maximal elementary abelian subgroups of \(P\).

A cohomology class of \(P\) is nilpotent if and only if its restrictions to \(S_2,...,S_n\) are nilpotent. Since the cohomology rings of elementary abelian groups are either polynomial rings or the tensor product of a polynomial ring with an outer algebra (if \(p\) is odd), it is easy to test whether a cohomology class of \(P\) is nilpotent, even if the whole cohomology ring is not known yet. We choose as many as possible new nilpotent generators in degree \(d\).
After step 1, continue with choosing as many as possible new generators in degree \(d\) with nilpotent restriction to \(S_1\).
After step 2, include the remaining new generators in degree \(d\). We call these Duflot generators, referring to [Duflot]

If \(z\) is the rank of the centre of \(P\) then there are exactly \(z\) Duflot generators, which was proved by [Kuhn], and they form a not necessarily maximal regular sequence. David Green introduced a monomial ordering on the cohomology ring of \(P\) that, in the first place, depends on the degree of a monomial, but also depends on the number of nilpotent respectively Duflot regular generators occuring in the monomial. It turns out that this monomial ordering magically simplifies many computations.

Note that the choice of generators and relations depends on how the group has originally been defined.

Ring approximation in the case of non prime power groups¶

The Stable Element Method¶

For the theoretical background of the Stable Element Method, we refer to [AdemMilgram]. It is known that the mod-\(p\) cohomology ring of a finite group \(G\) is a sub-ring of the mod-\(p\) cohomology of any subgroup \(P\) whose index in \(G\) is co-prime to \(p\). This sub-ring is formed by those cocycles of \(P\) that are stable under conjugation by elements of \(G\). That means, a cocycle \(c\in H^\ast(P; \mathbb F_p)\) is stable if and only if for all \(g\in G\) holds \(r_{P, P\cap P^g}(c) = r_{P^g, P\cap P^g}(\phi_g(c))\), where \(\phi_g\) denotes the homomorphism induced by conjugation with \(g^{-1}\), and \(r_{P, P\cap P^g}, r_{P^g, P\cap P^g}\) are restriction maps.

It suffices to test stability for representatives of all double cosets of \(P\) in \(G\). While this is, of course, much better than testing all elements of \(G\), the number of double cosets can still be rather large. It is thus often better to compute the cohomology of \(G\) in two steps. By default, we proceed as follows.

Let \(S\) be a Sylow \(p\)-subgroup of \(G\). Compute its mod-\(p\) cohomology, HS.

Let \(N = N_G(Z(S))\). Compute its mod-\(p\) cohomology HN as a sub-ring of HS.

If \(N\not=G\): Compute the mod-\(p\) cohomology of \(G\) as a sub-ring of HN.

The choice of generators¶

This is mainly as in the case of prime power groups, with exception of the Duflot generators. In the prime power case, we call a cohomology class Duflot, if it has non-trivial restriction to the greatest central elementary abelian subgroup. This makes sense, because this subgroup is non-trivial in the prime power case. If \(G\) is a finite group that is not of prime power order, we call a cohomology class Duflot, if it has non-trivial restriction to the greatest central elementary abelian subgroup of a Sylow `p`-subgroup \(S\) of \(G\).

In particular, if the Sylow \(p\)-subgroup of \(G\) is elementary abelian, then all elements in the mod-\(p\) cohomology ring of \(G\) are Duflot. Let \(z\) be the rank of the greatest central elementary abelian subgroup of \(S\). It is still true that one can always find \(z\) Duflot elements in the mod-\(p\) cohomology ring of \(G\) whose restrictions to the greatest central elementary abelian subgroup of \(S\) forms a regular sequence, which means that these Duflot elements form a regular sequence in the mod-\(p\) cohomology ring of \(G\). We refer to this as a Duflot regular sequence. However, it is not always possible to choose generators so that a Duflot regular sequence can be found among the generators.

Completeness Criteria¶

Let \(R=H^*(G;\mathbb F_p)\). We can compute ring approximation \(\tau_k R\) out to arbitrary finite degree \(k\). Since \(R\) has a finite presentation as an \(\mathbb F_p\)-algebra, there remains to find a bound such that \(\tau_k R\) is isomorphic to \(R\) for all \(k\) exceeding the bound.

In the case of an abelian group, it is known that a computation out to degree 2 will always suffice. In the case of generalized quaternion groups, it is known that a computation out to degree 4 will always suffice.

In all other cases, we need to use a completeness criterion. A completeness criterion is an algorithmic procedure that can decide whether an approximation of \(R\) out to degree \(k\) actually is isomorphic to \(R\), as graded \(\mathbb F_p\)-algebra. Such a criterion must be

effective (there is some \(N_0\) such that the criterion asserts isomorphy of \(\tau_kR\) and \(R\) for all \(k\ge N_0\)) and
correct (if \(n_0\) is the smallest number such that \(\tau_{n_0}R\) is
isomorphic to \(R\), then \(N_0\ge n_0\))

For practical computations, it is important that \(N_0-n_0\) is small. And of course, it is also important that the computations, which the completeness criterion relies on, are not too difficult.

This spkg uses a criterion introduced by [Benson], improved by [GreenKing], a criterion introduced by [Symonds], and a criterion introduced by [King] for finite groups that are not of prime power order.

Homogeneous systems of parameters¶

All three criteria require some knowledge of parameters of the mod-\(p\) cohomology ring, \(R\), of a finite group \(G\). A finite sequence of homogeneous elements \(p_1,...,p_t\) of \(R\) is called homogeneous system of parameters (or hsop), if they generate a sub-algebra of \(R\) over which \(R\) is finite. In this generality, we do not assume that the parameters are algebraically independent.

A complication arises from the fact that we need to compute the parameters of \(R\) in approximations of \(R\). However, while computing the approximations in increasing degrees, we can control the restrictions to the above-mentioned maximal elementary abelian subgroups \(S_2,...,S_n\) of (a Sylow \(p\)-subgroup of) \(G\). If the cohomology of \(S_2,...,S_n\) is finite over the restrictions of the elements \(p_1,...,p_t\) of some approximation of \(R\), then these elements give rise to a hsop of \(R\).

For Benson’s criterion, one needs to construct \(p_1,...,p_t\) such that they form a filter-regular sequence in \(R\). Filter-regular means that the multiplication map by parameter number \(k\) on the quotient of \(H^*(G)\) modulo the first \(k-1\) parameters has finite dimensional kernel. Since they are parameters, the quotient of \(H^*(G)\) by all parameters is finite dimensional. The parameters are used to compute the filter degree type of the ring approximation, and if this succeeds, degree bound \(N_0\) is obtained from the degree sum of the parameters.

In the modified Benson criterion, one still needs to construct \(p_1,...,p_t\) such that they form a filter-regular sequence in \(R\). But then, one can try to show the existence of parameters in smaller degrees over a finite extension field of \(\mathbb F_p\). The degree bound \(N_0\) is obtained from the degree sum of the smaller parameters, which do not require an explicit construction.

In the Symonds criterion, the parameters need to be constructed explicitly, and the degree bound \(N_0\) is obtained from the degree sum of the parameters. However, it is not needed that the parameters are algebraically independent. This additional freedom often allows for a construction of parameters in fairly small degrees.

In the next paragraphs, we explain how we construct (or show the existence of) the parameters in the different flavours.

Filter regular parameters—Dickson classes and the Benson criterion¶

Let \(G\) be a finite group and \(P\) be a Sylow \(p\)-subgroup of \(G\). Let \(r\) be the \(p\)-rank of \(P\), and let \(z\) be the rank of the centre of \(P\). Using Dickson invariants, we define \(r-z\) cohomology classes \(d_{1,j},...,d_{r-z,j}\) of \(S_j\) (\(j=2,...,n\)) in the polynomial part of the cohomology ring of \(S_j\), restricting to zero on \(S_1\) (recall that \(S_2,...S_n\) are maximal elementary abelian subgroups of \(G\), and \(S_1\) is a common subgroup, namely the greatest elementary abelian subgoup in the centre of \(P\)).

If \(G=P\) is a prime power group, [GreenKing] show that there exist cohomology classes \(D_1,...,D_{r-z}\) of \(G\) such that \(D_i\) simultaneously restrict to \(d_{i,2},...,d_{i,n}\) or to their \(p^{k_i}\)-th power for some \(k_1,...,k_{r-z}\). We briefly refer to these classes as restricted Dickson classes or restricted Dickson elements for \(G\) respectively for \(S_j\).

If \(G\) is not of prime power order, then it may be that we won’t find restricted Dickson elements. If they exist, we can proceed as in the prime power case: Based on [Benson], [GreenKing] show that a Duflot regular sequence of \(H^*(G)\) together with the restricted Dickson classes for \(G\) form a filter-regular homogeneous system of parameters (HSOP) for \(H^*(G)\).

If the restricted Dickson elements do not exist, one can use the original construction exposed in [Benson], which uses Dickson invariants in the cohomology rings of \(S_2,...S_n\), but not in a complement of \(S_1\). These unrestricted Dickson elements can be constructed in ring approximations, and they are guaranteed to provided a filter regular hsop of the cohomology ring. Note that unrestricted Dickson elements are of much higher degree than restricted Dickson elements.

Given a filter-regular hsop, one is allowed to try and factorise the parameters, possibly after modification by nilpotent elements. Let \(D_i\) be the smallest resulting non-constant factors. [GreenKing] show that the result is still a filter-regular hsop. And finally, the last parameter may be replaced by any other parameter, since filter-regularity is automatic for the last parameter. The motivation for all these refinements is to keep the parameter degrees low.

From the dimensions of annihilators of the filter regular parameters, one can compute the so-called filter degree type, which is a sequence of numbers whose maximum gives a bound for the Castelnuovo–Mumford regularity of the ring approximation. The filter degree type together with the degrees of the filter regular system of parameters and the rank of the centre of \(P\) give rise to a degree bound of [Benson]:

Theorem 1

Let \(G\) be a finite group. Assume that \(R_n\) is an approximation of \(H^\ast(G;\mathbb F_p)\) out to degree \(n\) and assume that there is a filter regular hsop for \(R_n\) in degrees at least two that gives rise to a filter regular hsop of \(H^\ast(G;\mathbb F_p)\). Let \(f_0,...,f_r\) be the filter degree type and let \(d_1,...,d_r\) be the degrees of the filter regular hsop. If

\(n > \max(0, f_0+0, f_1+1,...,f_{r-1}+r-1) - r + d_1+ ... +d_r\)

then \(R_n\) is isomorphic to the cohomology ring.

If the depth of \(H^\ast(G;\mathbb F_p)\) is at least two, then the preceding statement holds with \(n \ge ...\) instead of \(n > ...\). This is the case, e.g., if the centre of a Sylow \(p\)-subgroup \(P\) of \(G\) has \(p\)-rank at least two, or if there is a subgroup \(P<U<G\) such that \(H^\ast(U;\mathbb F_p)\) has depth at least two.

Benson conjectures that the filter degree type for any modular cohomology ring of a finite p-group is of the form \(-1,-2,...,-r,-r\). [Symonds] has shown that the regularity of a modular cohomology ring is zero. But of course neither of these statements is guaranteed to hold for an incomplete approximation of a cohomology ring. That’s why it is needed to compute the filter degree type in Benson’s criterion. This can be a very difficult task.

According to [GreenKing], it suffices to have one filter regular hsop of the ring approximation that gives rise to a hsop of the cohomology ring, and to prove the existence of a filter regular hsop of smaller degrees by changing the coefficients to a finite extension field—then, one can use Benson’s criterion using the degrees of the smaller hsop. The underlying result for an existence proof is this:

Lemma 2 [GreenKing]

Let \(R=H^\ast(G;\mathbb F_p)\). Assume that \(p_1,...,p_r\) is a filter-regular hsop of \(R\). If there is \(1\le i< r\) and some integer \(d\) such that \(R\) modulo the two-sided ideal generated by \(p_1,...,p_i\) together with the degree-\(d\) part of \(R\) is a finite dimensional algebra, then there is some finite field extension \(K\) of \(\mathbb F_p\) such that \(H^\ast(G; K)\) has a filter-regular hsop formed by elements in degrees \(|p_1|, ..., |p_i|\) together with \(r-i\) elements of degree \(d\).

Since the filter degree type is independent of the choice of a filter-regular system of parameters, we can use the Dickson classes for computing the filter-degree type. Then, we prove the existence of a filter regular hsop in smaller degrees. We do not need this hsop explicitly—we only need the degrees (which comes out of the existence proof) and the filter-degree type (which can be computed using the Dickson classes).

Arbitrary parameters—The Symonds Criterion¶

Theorem 3 [Symonds]

Let \(R_n\) be the approximation of \(H^\ast(G;\mathbb F_p)\) out to degree \(n\). Assume that there is a hsop for \(R_n\) in degrees \(d_1,...,d_r\) that gives rise to a hsop of \(H^\ast(G;\mathbb F_p)\). Let \(N\) be the maximal degree of generators of \(R_n\) as a module over the parameters. If

\(n \ge \max(N, d_1+ ... +d_r - r +1)\)

then \(R_n\) is isomorphic to the cohomology ring.

In order to compute the maximal degree \(N\) of module generators over the parametes, it is needed to explicitly know the parameters. Hence, in contrast to the Benson criterion, it seems impossible to improve the Symonds criterion by an existence result for small parameters over an extension field.

However, for verifying the Symonds criterion, it is not needed to compute the filter degree type of the hsop. Actually, it is not even needed to have algebraically independent parameters: It suffices to have elements that generate an ideal of Krull dimension zero.

We use two methods to compute a suitable hsop:

Start with a filter-regular hsop. Then try to replace each of the given parameters by an element of smaller degree, in a ring approximation. The restrictions to maximal \(p\)-elemenentary abelian subgroups allow to verify whether the result gives still rise to parameters of the cohomology ring (but not necessarily of the current ring approximation!).

Since filter-regular hsops are algebraically independent, the result of the replacements is still an algebraically independent hsop, though not necessarily filter-regular.
By considering the restrictions to maximal \(p\)-elementary abelian subgroups, find a minimal subset of the set of generators of the current ring approximation that forms a hsop of the cohomology ring. Actually, one is free to add all generators of degree 1, since degree 1 parameters do not contribute to the degree bound in Symonds’ result.

The resulting hsop is, of course, in general not even algebraically independent.

We are free to choose the hsop that gives the best degree bound in the Symonds criterion.

Arbitrary parameters over a field extension—The Hilbert–Poincaré criterion¶

Given an arbitrary hsop (for example, the one obtained by the first method explained in the previous paragraph), one can apply an existence result for parameters over a finite field extension, similar to the one used for the modified Benson criterion:

Lemma 4 [King]

Let \(R=H^\ast(G;\mathbb F_p)\) and let \(Q\) be a hsop of \(R\). Let \(S\subset Q\), and let \(d\) be some integer such that \(R\) modulo the two-sided ideal generated by the elements of \(S\) together with the degree-\(d\) part of \(R\) is a finite dimensional algebra. Let \(r\) be the \(p\)-rank of \(G\) (which is equal to the Krull dimension of \(R\)), and let \(r'\) be the Krull dimension of \(R\) modulo the ideal generated by \(S\). Then there is some finite field extension \(K\) of \(\mathbb F_p\) such that \(H^\ast(G; K)\) has a filter-regular hsop formed by elements in degrees \(|q|\) for all \(q\in S\), together with \(r-r'\) elements of degree \(d\).

The difference to the existence result used for the modified Benson criterion is that the subset \(S\) is not necessarily formed by an initial segment of \(Q\), which gives more possibilities to find parameters in small degrees. The resulting hsop over some field extension is not guaranteed to be filter-regular, if \(S\) is not an initial segment of the hsop.

Provided that all generators of the cohomology ring have already been found, one can then deduce completeness of the current ring approximation, by studying its Hilbert–Poincaré series. The following result arose in a discussion of Simon King with Peter Symonds.

Theorem 5 [King]

Let \(R=H^\ast(G;\mathbb F_p)\) and let \(\alpha_n: \tau_nR\to R\) be a ring approximation out to degree \(n\). Assume that \(\alpha_n\) is known to be surjective. Let \(d\) be a lower bound for the depth of \(R\). Assume that there is some finite field extension \(K\) of \(\mathbb F_p\) such that \(H^\ast(G; K)\) has a hsop formed by elements of degree \(d_1,...,d_r\). Denote \(N = \left(\sum_{i=1}^rd_i \right) - d\), and assume \(n\ge N\). Let \(f_n\in \mathbb Z[t]\) be the Hilbert–Poincaré series of \(\tau_nR\).

Then, \(\alpha_n\) is an isomorphism if and only if \(f_n\cdot \prod_{i=1}^r (1-t^{d_i})\) is a polynomial of degree at most \(N\).

Proving surjectivity of a ring approximation in the non prime power case¶

For applying the previous Theorem, there remains the problem of finding a lower bound for the depth, and the problem of deciding whether \(\alpha_n\) is surjective. Here, we need to restrict to the case that \(G\) is a finite group that is not of prime power order. Actually, the criterion is designed for being used in the context of the stable element method explained earlier.

Let \(U<G\) be a subgroup whose index in \(G\) is co-prime to \(p\). Recall that the stable element method describes \(R=H^\ast(G;\mathbb F_p)\) as a sub-algebra of \(H^\ast(U;\mathbb F_p)\), where we assume that the latter cohomology ring has already been computed. In particular, we can compute the depth of \(H^\ast(U;\mathbb F_p)\), which is a lower bound for the depth of \(R\) used in the Theorem above.

Surjectivity of the ring approximation can be proved using the following result.

Theorem 6 [King]

Let \(R=H^\ast(G;\mathbb F_p)\), let \(\alpha_n: \tau_nR\to R\) be a ring approximation out to degree \(n\), denote \(R_U=H^\ast(U;\mathbb F_p)\), and let \(\iota: R\to R_U\) be the inclusion used in the stable element method.

If \(n\) is sufficiently large, then \(R_U\) is finite over \(A_n=\iota\left(\alpha_n(\tau_nR)\right)\). If \(R_U\) is generated in degree at most \(D\) as an \(A_n\) module, and if \(n\ge D\), then \(\alpha_n\) is surjective.

There are many examples in which the Symonds criterion detects completeness of the cohomology ring approximation in lower degrees than the Hilbert–Poincaré criterion. This can be the case, if \(D\) is small. But even in that case, the preceding Theorem is extremely useful to speed up the stable element method. Namely, if surjectivity of the ring approximation is known, then all stable elements of \(R_U\) are decomposable in terms of the generators that have already been found. Hence, there is no need to solve large systems of equations for computing the space of stable elements in higher degree. This often saves a considerable amount of resources.

Comparison of the Criteria¶

If there is a nilpotent generator in degree \(d\), then there will be a relation at least in degree \(2d\). If there are generators in degrees \(d_1\) and \(d_2\) that are neither nilpotent nor Duflot (such generators are called boring), then there will be a relation at least in degree \(d_1+d_2\). Hence, before attempting to prove completeness of a cohomology ring approximation by a computationally expensive criterion, we consider the previously found generators and compute out to the degree in which we expect the last relations, before we attempt any of the completeness criteria.

When this degree is attained, we first try the Symonds criterion (both for prime power and finite non prime power groups). The Symonds criterion requires the explicit construction of a hsop for the cohomology ring, but it does not require algebraic independence. This gives some freedom in the choice of parameters, and very often one gets a very good degree bound. Moreover, the criterion is very easy to use—one only needs to find out in what degree the ring approximation is generated as a module over the sub-algebra generated by the chosen parameters.

If the Symonds criterion detects completeness, the computation can be finished. If it detects that the ring approximation is incomplete (this is the case, if the hsop of the cohomology ring is not a hsop of the ring approximation), then we continue the computation, until the next generator or relation is found.

If the Symonds criterion is inconclusive (this is the case, if the degrees of the explicitly constructed hsop are too large to detect completeness of the current ring approximation), then we try a different criterion, namely the modified Benson criterion in the case of a prime power group, or the Hilbert–Poincaré criterion for finite groups that are not of prime power order. This may be successful, if the existence result for small degree parameters over an extension field succeeds.

3. A step-by-step example¶

The purpose of this section is to illustrate the theoretical results formulated in the previous section. In our example, we compute the mod-2 cohomology ring of the symmetric group of rank 8. Normally, this would be done by defining G = gap.SymmetricGroup(8) and H = CohomologyRing(G, GroupName="S8", prime = 2), simply followed by H.make(). But here, we show what is happening behind the scenes.

The Sylow subgroup¶

Since we are using the stable element method, we first need to get the cohomology ring of a Sylow 2-subgroup. It is of order 128:

sage: G = libgap.SymmetricGroup(8)
sage: G.SylowSubgroup(2).IdGroup()
[ 128, 928 ]

We want to test this example as if it was in a new Sage session and therefore clear all previous computations:

sage: from pGroupCohomology import CohomologyRing
sage: CohomologyRing.doctest_setup()       # reset, block web access, use temporary workspace
sage: HSyl = CohomologyRing(128, 928, from_scratch=True)

Note that the program has a list of names of interesting groups, and so group number 928 of order 128 in the Small Groups library is correctly identified as the Sylow 2-subgroup of the symmetric group of rank 8:

sage: HSyl
H^*(Syl2(S8); GF(2))

At this point, the ring structure is not computed, but only the basic setup is present: The lowest term of a minimal projective resolution, together with the restriction maps on the special subgroups. The cohomology rings of the special subgroups have been computed while setting up HSyl.

sage: print(HSyl)
<BLANKLINE>
Cohomology ring of Sylow 2-subgroup of Symmetric Group S_8 with coefficients in GF(2)
<BLANKLINE>
Computed up to degree 0
Minimal list of generators:
[]
Minimal list of algebraic relations:
[]
<BLANKLINE>
sage: Res = HSyl.resolution(); Res
Resolution of GF(2)[Syl2(S8)]
sage: sorted(HSyl.restriction_maps().items())
[(1,
  [(2, 1),
   Induced homomorphism of degree 0 from H^*(Syl2(S8); GF(2)) to H^*(SmallGroup(2,1); GF(2))]),
 (2,
  [(8, 5),
   Induced homomorphism of degree 0 from H^*(Syl2(S8); GF(2)) to H^*(SmallGroup(8,5); GF(2))]),
 (3,
  [(8, 5),
   Induced homomorphism of degree 0 from H^*(Syl2(S8); GF(2)) to H^*(SmallGroup(8,5); GF(2))]),
 (4,
  [(16, 14),
   Induced homomorphism of degree 0 from H^*(Syl2(S8); GF(2)) to H^*(SmallGroup(16,14); GF(2))]),
 (5,
  [(16, 14),
   Induced homomorphism of degree 0 from H^*(Syl2(S8); GF(2)) to H^*(SmallGroup(16,14); GF(2))]),
 (6,
  [(16, 14),
   Induced homomorphism of degree 0 from H^*(Syl2(S8); GF(2)) to H^*(SmallGroup(16,14); GF(2))])]
sage: sorted(HSyl.subgroups().items())
[((2, 1), H^*(SmallGroup(2,1); GF(2))),
 ((8, 5), H^*(SmallGroup(8,5); GF(2))),
 ((16, 14), H^*(SmallGroup(16,14); GF(2)))]

As we can see, the Sylow 2-subgroup of G contains 2 conjugacy classes of maximal elementary abelian subgroups of order 8 and three conjugacy classes of maximal elementary abelian subgroups of order 16. The maximal central elementary abelian subgroup has order 2.

We could explicitly request the computation of higher terms of the resolution,

sage: Res.nextDiff()
sage: print(Res)
Resolution:
0 <- GF(2) <- GF(2)[Syl2(S8)] <- rank 3

but this is done anyway when we compute an approximation of the cohomology ring in increasing degrees. We compute the ring out to degree 3 (which could also be achieved by HSyl.make(3)):

sage: HSyl.next()
sage: HSyl.next()
sage: HSyl.next()
sage: print(Res)
Resolution:
0 <- GF(2) <- GF(2)[Syl2(S8)] <- rank 3 <- rank 7 <- rank 13 <- rank 22
sage: print(HSyl)
<BLANKLINE>
Cohomology ring of Sylow 2-subgroup of Symmetric Group S_8 with coefficients in GF(2)
<BLANKLINE>
Computed up to degree 3
Minimal list of generators:
[b_2_4: 2-Cocycle in H^*(Syl2(S8); GF(2)),
 b_2_5: 2-Cocycle in H^*(Syl2(S8); GF(2)),
 b_2_6: 2-Cocycle in H^*(Syl2(S8); GF(2)),
 b_1_0: 1-Cocycle in H^*(Syl2(S8); GF(2)),
 b_1_1: 1-Cocycle in H^*(Syl2(S8); GF(2)),
 b_1_2: 1-Cocycle in H^*(Syl2(S8); GF(2)),
 b_3_11: 3-Cocycle in H^*(Syl2(S8); GF(2)),
 b_3_12: 3-Cocycle in H^*(Syl2(S8); GF(2))]
Minimal list of algebraic relations:
[b_1_0*b_1_1,
 b_1_0*b_1_2,
 b_2_4*b_1_2,
 b_2_5*b_1_1,
 b_2_6*b_1_0]
<BLANKLINE>

There is a convention for the names of the cohomology ring generators: If the name starts with the letter “a”, then it is nilpotent. If it starts with the letter “c”, then it is Duflot. Otherwise, it starts with the letter “b”. Hence, all 8 generators found in degree at most 3 are not Duflot and not nilpotent. In other words, they have trivial restriction to the central elementary abelian subgroup, but non-trivial restriction to at least one maximal elementary abelian subgroup. Let us verify this for two of the generators. There are different methods to get the restrictions. First, we call the restriction map to the greatest central elementary abelian subgroup on some degree two generator.

sage: HSyl.1
b_2_4: 2-Cocycle in H^*(Syl2(S8); GF(2))
sage: HSyl.restriction_maps()[1][1]
Induced homomorphism of degree 0 from H^*(Syl2(S8); GF(2)) to H^*(SmallGroup(2,1); GF(2))
sage: HSyl.restriction_maps()[1][1](HSyl.1)
(b_2_4)_: 2-Cocycle in H^*(SmallGroup(2,1); GF(2))

The name of the image is obtained from the name of the preimage. We can verify that the image indeed vanishes:

sage: HSyl.restriction_maps()[1][1](HSyl.1).is_zero()
True

There is a convenient method that computes the restriction of all generators of the current ring approximation to the \(n\)-th special subgroup and expresses the images in terms of the cohomology ring generators of the subgroup. So, we can easily see that all generators restrict to zero on the greatest central elementary abelian subgroup:

sage: HSyl.restrict_to_subgroup(1)
[0: 2-Cocycle in H^*(SmallGroup(2,1); GF(2)),
2-Cocycle in H^*(SmallGroup(2,1); GF(2)),
2-Cocycle in H^*(SmallGroup(2,1); GF(2)),
1-Cocycle in H^*(SmallGroup(2,1); GF(2)),
1-Cocycle in H^*(SmallGroup(2,1); GF(2)),
1-Cocycle in H^*(SmallGroup(2,1); GF(2)),
3-Cocycle in H^*(SmallGroup(2,1); GF(2)),
3-Cocycle in H^*(SmallGroup(2,1); GF(2))]

We verify that each generator has non-trivial restriction to at least one maximal elementary abelian subgroup and is thus not nilpotent:

sage: HSyl.restrict_to_subgroup(2)
[c_1_1*c_1_2: 2-Cocycle in H^*(SmallGroup(8,5); GF(2)),
 0: 2-Cocycle in H^*(SmallGroup(8,5); GF(2)),
 0: 2-Cocycle in H^*(SmallGroup(8,5); GF(2)),
 c_1_2+c_1_1: 1-Cocycle in H^*(SmallGroup(8,5); GF(2)),
 0: 1-Cocycle in H^*(SmallGroup(8,5); GF(2)),
 0: 1-Cocycle in H^*(SmallGroup(8,5); GF(2)),
 0: 3-Cocycle in H^*(SmallGroup(8,5); GF(2)),
 0: 3-Cocycle in H^*(SmallGroup(8,5); GF(2))]
sage: HSyl.restrict_to_subgroup(3)
[0: 2-Cocycle in H^*(SmallGroup(8,5); GF(2)),
 c_1_1*c_1_2: 2-Cocycle in H^*(SmallGroup(8,5); GF(2)),
 0: 2-Cocycle in H^*(SmallGroup(8,5); GF(2)),
 c_1_2+c_1_1: 1-Cocycle in H^*(SmallGroup(8,5); GF(2)),
 0: 1-Cocycle in H^*(SmallGroup(8,5); GF(2)),
 0: 1-Cocycle in H^*(SmallGroup(8,5); GF(2)),
 0: 3-Cocycle in H^*(SmallGroup(8,5); GF(2)),
 0: 3-Cocycle in H^*(SmallGroup(8,5); GF(2))]
sage: HSyl.restrict_to_subgroup(4)
[0: 2-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 c_1_3^2+c_1_2*c_1_3: 2-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 c_1_1*c_1_3+c_1_1*c_1_2+c_1_1^2+c_1_0*c_1_2: 2-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 0: 1-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 0: 1-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 c_1_2: 1-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 0: 3-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 c_1_1*c_1_3^2+c_1_1*c_1_2^2+c_1_1^2*c_1_3+c_1_1^2*c_1_2+c_1_0*c_1_2^2+c_1_0^2*c_1_2: 3-Cocycle in H^*(SmallGroup(16,14); GF(2))]
sage: HSyl.restrict_to_subgroup(5)
[0: 2-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 0: 2-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 c_1_1*c_1_2+c_1_1^2+c_1_0*c_1_3+c_1_0*c_1_2: 2-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 0: 1-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 c_1_3: 1-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 c_1_2: 1-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 c_1_1*c_1_2*c_1_3+c_1_1^2*c_1_3+c_1_0*c_1_2*c_1_3+c_1_0^2*c_1_3: 3-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 c_1_1*c_1_2^2+c_1_1^2*c_1_2+c_1_0*c_1_2^2+c_1_0^2*c_1_2: 3-Cocycle in H^*(SmallGroup(16,14); GF(2))]
sage: HSyl.restrict_to_subgroup(6)
[c_1_3^2+c_1_2*c_1_3: 2-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 0: 2-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 c_1_1*c_1_3+c_1_1^2+c_1_0*c_1_2: 2-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 0: 1-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 c_1_2: 1-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 0: 1-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 c_1_1*c_1_3^2+c_1_1*c_1_2*c_1_3+c_1_1^2*c_1_3+c_1_1^2*c_1_2+c_1_0^2*c_1_2: 3-Cocycle in H^*(SmallGroup(16,14); GF(2)),
 0: 3-Cocycle in H^*(SmallGroup(16,14); GF(2))]

Parameters¶

Note that none of the generators is Duflot. Hence, the current ring approximation does not contain parameters for the cohomology ring and is thus incomplete:

sage: print(HSyl.parameters())
None

In the next degree, we finally find a Duflot generator (there is only one, since the greatest central elementary abelian subgroup is of rank one):

sage: HSyl.next()
sage: HSyl.duflot_regular_sequence()
['c_4_21']

However, the program did not bother to try and compute parameters yet:

sage: HSyl.filter_regular_parameters()
sage: HSyl.parameters()

The first reason is that we expect a relation in degree six:

sage: HSyl.expect_last_relation()
6

The second reason is that the \(p\)-rank of this group (which is equal to the dimension of the cohomology ring) is four and the rank of the centre is one, which means that the restricted Dickson elements are of degrees 4, 6 and 7 and are thus found in higher degrees than the current ring approximation. The default method of finding restricted Dickson elements uses linear algebra and only works in those degrees that already are available:

sage: HSyl.Dickson
['b_1_2*b_3_12+b_1_2^4+b_1_1*b_3_11+b_1_1^2*b_1_2^2+b_1_1^4+b_1_0^4+b_2_6*b_1_2^2+b_2_6*b_1_1*b_1_2+b_2_6^2+b_2_5^2+b_2_4^2',
 None,
 None]

Note that the unrestricted Dickson elements originally considered by [Benson] are of much higher degrees, namely 8, 12, 14 and 15. In any case, it is possible to try to compute the restricted Dickson elements by a method involving elimination. This is potentially very expensive and is here not done by default. Let’s call this method:

sage: HSyl.find_dickson()
True
sage: HSyl.Dickson
['b_1_2*b_3_12+b_1_2^4+b_1_1*b_3_11+b_1_1^2*b_1_2^2+b_1_1^4+b_1_0^4+b_2_6*b_1_2^2+b_2_6*b_1_1*b_1_2+b_2_6^2+b_2_5^2+b_2_4^2',
 'b_3_12^2+b_1_2^3*b_3_12+b_1_1*b_1_2^2*b_3_12+b_1_1^2*b_1_2^4+b_1_1^3*b_3_12+b_1_1^3*b_3_11+b_1_1^4*b_1_2^2+b_2_6*b_1_2*b_3_12+b_2_6*b_1_2^4+b_2_6*b_1_1^3*b_1_2+b_2_6^2*b_1_2^2+b_2_6^2*b_1_1*b_1_2+b_2_6^2*b_1_1^2+b_2_5*b_1_2*b_3_12+b_2_5*b_2_6*b_1_2^2+b_2_5^2*b_1_2^2+b_2_5^2*b_1_0^2+b_2_4*b_1_1*b_3_11+b_2_4*b_2_6^2+b_2_4^2*b_1_1^2+b_2_4^2*b_1_0^2+c_4_21*b_1_2^2',
 'b_1_2*b_3_12^2+b_1_1*b_1_2^3*b_3_12+b_1_1^2*b_1_2^2*b_3_12+b_1_1^3*b_1_2*b_3_12+b_1_1^4*b_3_12+b_2_6*b_1_2^2*b_3_12+b_2_6*b_1_1*b_1_2^4+b_2_6*b_1_1^3*b_1_2^2+b_2_6^2*b_1_1*b_1_2^2+b_2_6^2*b_1_1^2*b_1_2+b_2_5*b_1_2^2*b_3_12+b_2_5*b_2_6*b_1_2^3+b_2_4*b_1_1^2*b_3_11+b_2_4*b_2_6^2*b_1_1+c_4_21*b_1_2^3']

Together with the Duflot generator, we obtain a filter regular hsop. We find that we could replace the last Dickson element by a parameter in degree one (the resulting hsop would still be filter-regular):

sage: HSyl.find_small_last_parameter(HSyl.duflot_regular_sequence()+HSyl.Dickson)
'b_1_2+b_1_1'

Of course, there is no need to do the preceding computation explicitly, as it is done internally anyway when computing a filter-regular hsop:

sage: HSyl.filter_regular_parameters()
['c_4_21',
 'b_1_2*b_3_12+b_1_2^4+b_1_1*b_3_11+b_1_1^2*b_1_2^2+b_1_1^4+b_1_0^4+b_2_6*b_1_2^2+b_2_6*b_1_1*b_1_2+b_2_6^2+b_2_5^2+b_2_4^2',
 'b_3_12^2+b_1_2^3*b_3_12+b_1_1*b_1_2^2*b_3_12+b_1_1^2*b_1_2^4+b_1_1^3*b_3_12+b_1_1^3*b_3_11+b_1_1^4*b_1_2^2+b_2_6*b_1_2*b_3_12+b_2_6*b_1_2^4+b_2_6*b_1_1^3*b_1_2+b_2_6^2*b_1_2^2+b_2_6^2*b_1_1*b_1_2+b_2_6^2*b_1_1^2+b_2_5*b_1_2*b_3_12+b_2_5*b_2_6*b_1_2^2+b_2_5^2*b_1_2^2+b_2_5^2*b_1_0^2+b_2_4*b_1_1*b_3_11+b_2_4*b_2_6^2+b_2_4^2*b_1_1^2+b_2_4^2*b_1_0^2+c_4_21*b_1_2^2',
 'b_1_2*b_3_12^2+b_1_1*b_1_2^3*b_3_12+b_1_1^2*b_1_2^2*b_3_12+b_1_1^3*b_1_2*b_3_12+b_1_1^4*b_3_12+b_2_6*b_1_2^2*b_3_12+b_2_6*b_1_1*b_1_2^4+b_2_6*b_1_1^3*b_1_2^2+b_2_6^2*b_1_1*b_1_2^2+b_2_6^2*b_1_1^2*b_1_2+b_2_5*b_1_2^2*b_3_12+b_2_5*b_2_6*b_1_2^3+b_2_4*b_1_1^2*b_3_11+b_2_4*b_2_6^2*b_1_1+c_4_21*b_1_2^3']

We can also replace other parameters of this filter-regular hsop by elements of smaller degree (here, this fails). The result is an algebraically independent hsop, but there is no guarantee that it is filter-regular:

sage: HSyl.parameters()
['c_4_21',
 'b_1_2*b_3_12+b_1_2^4+b_1_1*b_3_11+b_1_1^2*b_1_2^2+b_1_1^4+b_1_0^4+b_2_6*b_1_2^2+b_2_6*b_1_1*b_1_2+b_2_6^2+b_2_5^2+b_2_4^2',
 'b_3_12^2+b_1_2^3*b_3_12+b_1_1*b_1_2^2*b_3_12+b_1_1^2*b_1_2^4+b_1_1^3*b_3_12+b_1_1^3*b_3_11+b_1_1^4*b_1_2^2+b_2_6*b_1_2*b_3_12+b_2_6*b_1_2^4+b_2_6*b_1_1^3*b_1_2+b_2_6^2*b_1_2^2+b_2_6^2*b_1_1*b_1_2+b_2_6^2*b_1_1^2+b_2_5*b_1_2*b_3_12+b_2_5*b_2_6*b_1_2^2+b_2_5^2*b_1_2^2+b_2_5^2*b_1_0^2+b_2_4*b_1_1*b_3_11+b_2_4*b_2_6^2+b_2_4^2*b_1_1^2+b_2_4^2*b_1_0^2+c_4_21*b_1_2^2',
 'b_1_2+b_1_1']

When we try to find a subset of the generators of the current ring approximation that is a hsop of the cohomology ring, we obtain:

sage: HSyl.dependent_parameters()
['b_1_0', 'b_1_1', 'b_1_2', 'c_4_21', 'b_2_4', 'b_2_5', 'b_2_6']

All these hsops are for the cohomology ring, but they are no hsops for the current ring approximation:

sage: HSyl.set_ring()
sage: I = HSyl.relation_ideal()
sage: singular.eval('degBound=0')
''
sage: (I + singular.ideal(HSyl.filter_regular_parameters())).std().dim()
2
sage: (I + singular.ideal(HSyl.parameters())).std().dim()
2
sage: (I + singular.ideal(HSyl.dependent_parameters())).std().dim()
2

But this is no surprise, since the current ring approximation is up to degree 4, but we expect a relation in degree six. So, let us compute two degrees further:

sage: HSyl.next()
sage: HSyl.next()

Now, the cohomology parameters are parameters for the ring approximation, too:

sage: HSyl.set_ring()
sage: singular.eval('degBound=0')
''
sage: I = HSyl.relation_ideal()
sage: (I + singular.ideal(HSyl.filter_regular_parameters())).std().dim()
0
sage: (I + singular.ideal(HSyl.parameters())).std().dim()
0
sage: (I + singular.ideal(HSyl.dependent_parameters())).std().dim()
0

Completion tests¶

In what degrees would it make sense to try the Symonds criterion? We found algebraically independent parameters in degrees 4, 2, 3, 1, and we found dependent parameters in degrees 1, 1, 1, 4, 2, 2, 2, which means that the Symonds criterion can only apply in degree strictly greater than \(3+1+2+0=6\) or strictly greater than \(0+0+0+3+1+1+1=6\). Hence, it does not apply yet.

What about the modified Benson criterion? It turns out that we can not prove the existence of smaller parameters over a finite extension field, so, we have to work with the explicitly computed filter-regular hsop in degrees 4, 2, 3, 1 (the last parameter needs to be squared, because the parameter degrees must be at least two for Benson’s criterion). The \(p\)-rank of the centre of our Sylow 2-subgroup is one, which means that we only get 1 as lower bound for the depth of the cohomology ring. Hence, the Benson criterion can only apply in degree strictly greater than \(3+1+2+1=7\). Note, however, that the filter degree type (which is stored as an attribute) already is as expected:

sage: HSyl.raw_filter_degree_type(HSyl.filter_regular_parameters())
([-1, -1, -1, 10, 17],
 [[0],
  [0],
  [0],
  [0, 1, 1, 3, 3, 4, 4, 3, 3, 1, 1],
  [1, 3, 7, 13, 20, 28, 35, 40, 43, 41, 38, 31, 24, 17, 10, 6, 2, 1]],
 [4, 4, 6, 7])
sage: HSyl.fdt
[-1, -2, -3, -4, -4]

However, we need to compute the ring approximation out to degree 7 (not finding any new generator or relation), before we succeed with proving completeness with the Symonds criterion.

sage: HSyl.next()
sage: HSyl.last_interesting_degree()
6
sage: HSyl.test_for_completion()
True
sage: HSyl._method
'Symonds'

Indeed, the ring approximation can be generated in degree \(3\le7\) as a module over one of the previously obtained hsops:

sage: HSyl._max_module_deg
3
sage: HSyl.set_ring()
sage: singular.eval('degBound=0')
''
sage: I = HSyl.relation_ideal()
sage: HSyl._parameters_for_criterion
['b_1_0', 'b_1_1', 'b_1_2', 'c_4_21', 'b_2_4', 'b_2_5', 'b_2_6']
sage: (I + singular.ideal(HSyl._parameters_for_criterion)).std().kbase()
b_3_12,
b_3_11,
1

An intermediate subgroup¶

Often it is easier to not directly apply the stable element method to a Sylow subgroup, but first compute the cohomology of an intermediate group between \(G\) and its Sylow subgroup. By default, we use the normaliser in \(G\) of the centre of the Sylow subgroup of \(G\):

sage: G.Normalizer(G.SylowSubgroup(2).Centre()).IdGroup()
[ 384, 5602 ]

So, we create the basic setup of its cohomology ring:

sage: HU = CohomologyRing(384, 5602, prime=2)

Stable elements¶

The stable element method for this intermediate subgroup works on the Sylow subgroup. We find that one stability condition is enough. The stability conditions are expressed in terms of a two maps that either restrict directly from the Sylow subgroup to its intersection with a conjugate, or restrict after twisting with the map that is induced by conjugation:

sage: HU.sylow_cohomology() is HSyl
True
sage: HU._PtoPcapCPdirect
[Induced homomorphism of degree 0 from H^*(Syl2(S8); GF(2)) to H^*(SmallGroup(64,138); GF(2))]
sage: HU._PtoPcapCPtwist
[Induced homomorphism of degree 0 from H^*(Syl2(S8); GF(2)) to H^*(SmallGroup(64,138); GF(2))]

Let us compute subspace of stable elements of degree one in the cohomology ring of the Sylow subgroup:

sage: HU.stable_space(1)
([Gen2: 1-Cocycle in H^*(SmallGroup(384,5602); GF(2)),
  Gen1: 1-Cocycle in H^*(SmallGroup(384,5602); GF(2))],
 ['b_1_2', 'b_1_1', 'b_1_0'],
 ['b_1_2', 'b_1_1'])

This means: The cohomology of our intermediate subgroup has two generators in degree one, whose restrictions to the Sylow subgroup have leading monomials b_1_2 or b_1_1, and apart from this the cohomology of the Sylow subgroup has another degree one monomial b_1_0 that does not occur as leading monomial of a stable element. Of course, there are no relations in degree one. So, in degree one, we obtain:

sage: HU.next()
sage: HU.gens()
[1,
 b_1_0: 1-Cocycle in H^*(SmallGroup(384,5602); GF(2)),
 b_1_1: 1-Cocycle in H^*(SmallGroup(384,5602); GF(2))]

We find that the stable subspace in degree two is of dimension 5. The decomposable elements form a sub-space of dimension 3 (that is the first output of find_relations()), and there are no relations in degree two (that is the second output):

sage: HU.stable_space(2)
([Gen5: 2-Cocycle in H^*(SmallGroup(384,5602); GF(2)),
  Gen4: 2-Cocycle in H^*(SmallGroup(384,5602); GF(2)),
  Gen3: 2-Cocycle in H^*(SmallGroup(384,5602); GF(2)),
  Gen2: 2-Cocycle in H^*(SmallGroup(384,5602); GF(2)),
  Gen1: 2-Cocycle in H^*(SmallGroup(384,5602); GF(2))],
 ['b_1_2^2', 'b_1_1*b_1_2', 'b_1_1^2', 'b_1_0^2', 'b_2_6', 'b_2_5', 'b_2_4'],
 ['b_1_2^2', 'b_1_1*b_1_2', 'b_1_1^2', 'b_1_0^2', 'b_2_6'])
sage: HU.find_relations(2)
([b_1_1^2: 2-Cocycle in H^*(SmallGroup(384,5602); GF(2)),
  b_1_0*b_1_1: 2-Cocycle in H^*(SmallGroup(384,5602); GF(2)),
  b_1_0^2: 2-Cocycle in H^*(SmallGroup(384,5602); GF(2))],
 [])

Hence, we find two cohomology generators in degree two, and obtain

sage: HU.next()
sage: print(HU)
<BLANKLINE>
Cohomology ring of SmallGroup(384,5602) with coefficients in GF(2)
<BLANKLINE>
Computed up to degree 2
Minimal list of generators:
[b_2_3: 2-Cocycle in H^*(SmallGroup(384,5602); GF(2)),
 b_2_4: 2-Cocycle in H^*(SmallGroup(384,5602); GF(2)),
 b_1_0: 1-Cocycle in H^*(SmallGroup(384,5602); GF(2)),
 b_1_1: 1-Cocycle in H^*(SmallGroup(384,5602); GF(2))]
Minimal list of algebraic relations:
[]
<BLANKLINE>

By the way, if one wants to verify that the stability conditions hold for the elements of \(HU\), then one can proceed as follows. Each element knows its restriction to the subgroup used in the stable element method:

sage: g = HU.1; g
b_2_3: 2-Cocycle in H^*(SmallGroup(384,5602); GF(2))
sage: g.val_str()
'b_1_0^2+b_2_5'

We create the corresponding element of HSyl and verify that the two induced maps involved in the stability conditions evaluate equal on this element:

sage: gS = HSyl(g.val_str())
sage: HU._PtoPcapCPdirect[0](gS) == HU._PtoPcapCPtwist[0](gS)
True

In degree 3, we find the first relation:

sage: HU.find_relations(3)[1]
['b_2_3*b_1_1']

The Hilbert–Poincaré test¶

Let us compute the cohomology ring approximation out to degree six:

sage: HU.make(6)
sage: HU.last_interesting_degree()
6
sage: HU.expect_last_relation()
6

Hence, it seems possible that the ring approximation is complete in degree six. What parameters do we find? The filter regular ones are rather lengthy, we only show one term so that the degrees are visible:

sage: HU.parameters_from_sylow_subgroup()
sage: HU.filter_regular_parameters()
['b_1_1^2*b_3_1^2+...',
 'b_3_9^4+b_3_1^4+...',
 'b_1_1^2*b_3_1^4+...',
 'c_4_15']
sage: HU.parameters()
['b_1_1^4+b_1_0^3*b_1_1+b_1_0^4+b_2_4*b_1_1^2+b_2_4*b_1_0^2+b_2_4^2+b_2_3^2',
 'b_3_9^2+b_3_1^2+b_1_0*b_1_1^2*b_3_1+b_1_0^6+b_2_4*b_1_0*b_3_1+b_2_4*b_1_0*b_3_0+b_2_4*b_1_0*b_1_1^3+b_2_4*b_1_0^4+b_2_4^2*b_1_1^2+b_2_4^3+b_2_3*b_2_4^2+b_2_3^2*b_2_4',
 'b_1_1+b_1_0',
 'c_4_15']
 sage: HU.dependent_parameters()
 ['b_1_0', 'b_1_1', 'c_4_15', 'b_3_1', 'b_3_9', 'b_2_3', 'b_2_4']

Let us determine in what degrees our three completion tests have a theoretical chance to apply. We found a filter-regular hsop only in degrees 8, 12, 14 and 1 (the latter by replacing a degree 15 reduced Dickson element by some parameter in degree 1). Hence, the Benson test in its original form could not be applied before degree \(7+11+13+1=32\). But it is possible to prove the existence of a filter-regular hsop over a finite extension field, namely as follows:

sage: HU.potential_degree_bound(HU.filter_regular_parameters(), [8, 12, 14, 1])
18
sage: HU.WhatFRS
(['c_4_15'], 6)

This answer means: There is some finite field extension for which the cohomology ring has a filter-regular hsop formed by the Duflot element c_4_15, together with three elements of degree 6. Let us verify the existence result:

sage: I = HU.relation_ideal()
sage: HU.set_ring()
sage: singular.eval('degBound=0')
''
sage: (I + singular.ideal(['c_4_15'] + HU.standard_monomials(6))).std().dim()
0

Since the lower bound for the depth of this cohomology ring is greater than one,

sage: HU.sylow_cohomology().depth()
3

the modified Benson criterion could apply in degree \(3+5+5+5=18\). But this is not good enough. We could use the Symonds criterion with the algebraically independent hsop in degrees 4, 6, 2, 1, so that the criterion could only apply in degrees strictly greater than \(3+5+1+0=9\), or with an algebraically dependent hsop in degrees 1, 1, 4, 3, 3, 2, 2, so that again the criterion could only apply in degrees strictly greater than \(0+0+3+2+2+1+1=9\). Nevertheless, let us test for completion now:

sage: CohomologyRing.global_options('info')
sage: HU.test_for_completion()
H^*(SmallGroup(384,5602); GF(2)):
    We found parameters, but they would not allow for an application of Symonds' criterion.
    Trying the Hilbert-Poincare criterion
    We expect that the Hilbert-Poincare criterion will not apply before degree 9
    No conclusion on the completeness of this cohomology ring.
sage: CohomologyRing.global_options('warn')

Since it seems possible to apply the Hilbert–Poincaré criterion in degree 9, we compute the ring out to the next degree (but we do not find new relations or generators), and succeed with applying the Hilbert–Poincaré test:

sage: HU.next()
sage: HU.next()
sage: HU.next()
sage: HU.last_interesting_degree()
6
sage: HU.test_for_completion()
True
sage: HU._method
'Hilbert-Poincar&eacute;'

So, in this example, the Hilbert–Poincaré criterion is better than the Symonds criterion, which is better than the modified Benson criterion. But there are examples in which the relative strength of the criteria is different.

Last, we verify the completeness test manually. In this example, we can show that there is a finite field extension of \(\mathbb F_2\) over which there is an hsop of the cohomology ring in degrees 4, 2, 1, 3, by applying our existence result. We can replace the degree 12 parameter in our previously found algebraically independent hsop by an element in degree 3, when extending the base field:

sage: HU.set_ring()
sage: I = HU.relation_ideal()
sage: (I + singular.ideal([HU.parameters()[0]] + HU.parameters()[2:] + HU.standard_monomials(3))).std().dim()
0

This computation is done internally in the following method:

sage: HU.parameter_degrees_over_field_extension()
((4, 1, 4, 3), 3, True)

The first output gives the parameter degrees, the second gives a lower bound for the depths, and the third tells that the existence result has actually been applied. Since \(4+2+1+3-3=7\), the Hilbert–Poincaré criterion can be applied, and it proves completeness, since

sage: p = HU.poincare_series()
sage: t = p.parent().gen()
sage: p*(1-t^4)*(1-t^2)*(1-t)*(1-t^3)
t^6 + t^5 + t^4 + 3*t^3 + 2*t^2 + t + 1

is a polynomial of degree at most 7.

Stable element method in two steps¶

Now, as we have computed the cohomology ring of an intermediate subgroup, we aim at computing the cohomology ring of \(G\), the symmetric group of rank 8. Since this group is not part of the small groups library, we need to assign a name to it when defining the cohomology ring:

sage: H = CohomologyRing(G, GroupName="SymmetricGroup(8)", prime = 2)
sage: H
H^*(SymmetricGroup(8); GF(2))

As we have announced above, the group studied in the previous subsection is used as intermediate subgroup.

sage: H.subgroup_cohomology() is HU
True
sage: H.sylow_cohomology() is HSyl
True

There are three stability conditions, and look a bit more complicated, as there occur three different intersections of our subgroup with its conjugates:

sage: H._PtoPcapCPdirect
[Induced homomorphism of degree 0 from H^*(SmallGroup(384,5602); GF(2)) to H^*(D8xV4; GF(2)),
 Induced homomorphism of degree 0 from H^*(SmallGroup(384,5602); GF(2)) to H^*(SmallGroup(32,27); GF(2)),
 Induced homomorphism of degree 0 from H^*(SmallGroup(384,5602); GF(2)) to H^*(D8; GF(2))]

The computation of an approximation of the cohomology ring works as for the cohomology of the intermediate subgroup. We have not demonstrated yet how to prove that all cohomology generators have been found. So, let us compute out to degree four:

sage: H.make(4)
sage: H.duflot_regular_sequence()
['c_4_3']
sage: H.generator_degbound() # this has no output, but may set some attributes
sage: print(H.degbound_for_gens)
None

So, a Duflot regular sequence is available, but we can’t determine a bound for the maximal degree of a minimal generating set. Let us proceed until degree 7:

sage: H.make(7)
sage: H.generator_degbound()
sage: H.degbound_for_gens
8

So, we will find more generators at most in degree 8. Why is that? There will be generators at most in the degree in which the cohomology ring of the subgroup used in the stable element method can be generated as a module over the image of the current ring approximation. Hence:

sage: HU.set_ring()
sage: singular.eval('degBound=0')
''
sage: I = HU.relation_ideal()
sage: (I + singular.ideal([x.val_str() for x in H.gens()[1:]])).std().kbase().sort()[1]
1,
b_1_0,
b_2_3,
b_2_4,
b_2_3*b_1_0,
b_2_4*b_1_0,
b_3_9,
c_4_15,
b_2_3^2,
b_2_3*b_2_4,
c_4_15*b_1_0,
b_2_3^2*b_1_0,
b_2_3*b_3_9,
b_2_3*c_4_15,
b_2_4*c_4_15,
b_2_3^3,
b_2_4*c_4_15*b_1_0,
c_4_15^2,
b_2_3^2*c_4_15

The maximal degree of a module generator is 8 (which is the degree of the last five elements in the above list of monomials), as we have claimed. By consequence, when we compute degree 8, we still compute the stable subspace.

sage: CohomologyRing.global_options('info')
sage: H.make(8)
H^*(SymmetricGroup(8); GF(2)):
    We have no degree bound yet
    Computing ring approximation in degree 8
    Determining stable subspace in degree 8
H^*(SmallGroup(384,5602); GF(2)):
    Determine degree 8 standard monomials
H^*(D8xV4; GF(2)):
    Determine degree 8 standard monomials
H^*(SmallGroup(32,27); GF(2)):
    Determine degree 8 standard monomials
H^*(D8; GF(2)):
    Determine degree 8 standard monomials
H^*(SymmetricGroup(8); GF(2)):
    Setting up conditions to determine stable elements
    Solving equations
    Computing Groebner basis up to degree 8
    Exploring relations in degree 8
    Determine degree 8 standard monomials
    Express 27 standard monomials as cocycles
    Found 2 relations in degree 8
    There is no new generator in degree 8
    Try to lift 1st power of 0th Dickson invariant
    Exploring relations in degree 8
    Determine degree 8 standard monomials
    Express 25 standard monomials as cocycles
    Found 0 relations in degree 8
    Factorising an element; it can be interrupted with Ctrl-c
    Degree 8 of the visible ring structure is computed!
    Storing ring approximation
    We expect a relation in degree at least 14
    Computation of the ring approximation is finished
<BLANKLINE>
    Storing ring approximation

Hence, we did actually not find another generator in degree 8, and we have just proved that there will be no generators in higher degrees. So, starting with degree 9, the computation of the ring approximation will be considerably simplified, as there is no need to compute the stable subspace:

sage: H.all_generators_found
True
sage: H.make(9)
    We have no degree bound yet
    Computing ring approximation in degree 9
    All generators are known
    Computing Groebner basis up to degree 9
    Exploring relations in degree 9
    Determine degree 9 standard monomials
    Express 35 standard monomials as cocycles
    Found 2 relations in degree 9
    There is no new generator in degree 9
    Degree 9 of the visible ring structure is computed!
    Storing ring approximation
    We expect a relation in degree at least 14
    Computation of the ring approximation is finished
<BLANKLINE>
    Storing ring approximation

As we see, the last relations are expected in degree 14, which is actually too high (the last relation is in degree 12), but let us compute out to degree 14 anyway.

sage: CohomologyRing.global_options('warn')
sage: H.make(14)
sage: H.knownDeg
14
sage: H.last_interesting_degree()
12

It now makes sense to study parameters. We find filter regular parameters in degrees 8, 12, 14 and 6 (we truncate the output),

sage: H.filter_regular_parameters()
['b_3_1*b_5_8+...',
 'b_3_1^4+...',
 'b_7_17^2+...',
 'b_1_0^6+b_6_0']

algebraically independent parameters in degrees 4, 6, 7 and 6

sage: H.parameters()
['b_2_1^2+c_4_3',
 'b_3_1^2+b_3_0^2+b_2_1*c_4_3',
 'b_7_17+b_2_1*b_5_8+c_4_3*b_3_1+c_4_3*b_3_0',
 'b_1_0^6+b_6_0']

and algebraically dependent parameters in degrees 1, 7, 6, 4, 3, 3 and 2.

sage: H.dependent_parameters()
['b_1_0', 'b_7_17', 'b_6_0', 'c_4_3', 'b_3_0', 'b_3_1', 'b_2_1']

Since \(7+11+13+5\), \(3+5+6+5\) and \(0+6+5+3+2+2+1\) are all greater than the current degree of approximation, we can only hope for using an existence proof of parameters in smaller degrees. But alas, this fails for the Benson criterion

sage: H.potential_degree_bound(H.filter_regular_parameters(),[8, 12, 14,6])
36

and for the Hilbert–Poincaré criterion:

sage: H.parameter_degrees_over_field_extension()
((4, 6, 7, 6), 3, False)
sage: H.subgroup_cohomology().depth()
3
sage: 4+6+7+6-3
20

We finish the computation and find that completion has been proven using the Symonds criterion, in degree 20 (so, the same degree could also be obtained with the Hilbert–Poincaré criterion), using an algebraically dependent hsop:

sage: H.make()
sage: H._method
'Symonds'
sage: H.knownDeg
20
sage: H._parameters_for_criterion
['b_1_0', 'b_7_17', 'b_6_0', 'c_4_3', 'b_3_0', 'b_3_1', 'b_2_1']

References¶

AdemMilgram: Alejandro Adem, R. James Milgram: Cohomology of finite groups. Second edition. Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences], 309. Springer-Verlag, Berlin, 2004.
Benson(1,2,3,4,5,6): Dave Benson: Dickson invariants, regularity and computation in group cohomology. Illinois J. Math. 48 (2004), 171–197.
BensonCarlson: Dave Benson, Jon F. Carlson: Projective resolutions and Poincaré duality complexes. Trans. Amer. Math. Soc. 342 (1994), 447–488.
CartanEilenberg: Henri Cartan and Samuel Eilenberg: Homological algebra. Princeton University Press, Princeton, N. J., 1956.
Duflot: Jeanne Duflot: Depth and equivariant cohomology, Comment. Math. Helv. 56 (1981), 627–637.
EllisKing: Graham Ellis and Simon King: Persistent homology of groups., J. Group Theory 14 (2011), 575–587.
Green: David Green: Gr”obner bases and the computation of group cohomology. Lecture Notes in Mathematics, 1828. Springer-Verlag, Berlin, 2003.
GreenKing(1,2,3,4,5,6,7,8): David Green, Simon King: The computation of the cohomology rings of all groups of order 128., J. Algebra 325 (2011), 352–363.
King(1,2,3,4,5): Simon King: Comparing completeness criteria for modular cohomology rings of finite groups., J. of Algebra 374 (2013), 247–256.
KingGreenEllis: Simon King, David Green, Graham Ellis: The Mod-2 Cohomology Ring of the Third Conway Group is Cohen-Macaulay., Algebraic & Geometric Topology 11 (2011), 719–734.
Kraines: David Kraines: Massey Higher Products. Trans. AMS 124 (1966), 431–449.
Kuhn: Nicholas J. Kuhn: Primitives and central detection numbers in group cohomology, Adv. Math. 216 (2007), 387–442.
Pham: Pham Anh Minh: Modular invariant theory and cohomology algebras of extra-special \(p\)-groups. Pacific J. Math. 124 (1986), 345–363.
Ravenel: Douglas Ravenel, Complex cobordism and stable homotopy groups of spheres, Pure and Applied Mathematics, 121. Academic Press, Inc., Orlando, FL, (1986).
Symonds(1,2,3,4): Peter Symonds: On the Castelnuovo-Mumford Regularity of the Cohomology Ring of a Group. J. American Math. Soc., 23 (2010), 1159–1173.
Wilkerson: Clarence Wilkerson: A primer on the Dickson invariants, Proceedings of the Northwestern Homotopy Theory Conference (Evanston, Ill., 1982), Contemp. Math., 19 (1983), 421–434.

pGroupCohomology.CohomologyRing(*args, **kwds)¶

Constructor for modular cohomology rings of finite groups.

This constructor is an instance of CohomologyRingFactory. See there and see pGroupCohomology for more examples.

The constructor can be called directly. Then, it is first checked whether the completely computed cohomology ring of the given group is part of some database, or whether it can be downloaded. If this is not the case, a new cohomology ring is created, being part of the user’s workspace.

Using set_workspace(), the location of the user’s workspace can be changed. By from_workspace(), one can explicitly ask for taking data from the workspace. The input formats for calling CohomologyRing() and for calling from_workspace() or from_local_sources() are the same.

INPUT:

Parameters describing the group

A finite group \(G\) , either
- given by an integer q and a positive number n, determining an entry of the SmallGroups library, or
- given as an object in the libgap interface
GroupName (optional string): a name for the group. If the group \(G\) is given in the Gap interface and if it is not provided with a custom name (using libGap’s SetName) then GroupName must be provided.
GroupDescr (optional string): a description of the group. This can be any string, and is used when printing the cohomology ring or creating a web-site for it.
If the group is not of prime power order, the optional parameter prime must be set to a prime number.

Parameters describing the database

websource: If it is False, it is not attempted to download data from a web repository. If it is a URL (string) or tuple of URLs providing the location(s) of a database in the web, then it is attempted to download the data from there. If websource is not given then first it is tried to look up data in the local file system, and if this fails then it is attempted to download the data from some default location in the web.
from_scratch (default False): If it is True, this cohomology ring may be taken from the cache or from the workspace, but will not be copied from local or remote sources. Note that this will only take effect on this single ring; cohomology rings of subgroups, occuring during the computation, will still be loaded.

Parameters modifying the algorithm

useElimination (optional, default None): If True, the elimination method is used for trying to construct the Dickson classes. If False, the linear algebra method is used for that purpose. By default, the linear algebra method is chosen, unless there is a Dickson class in degree greater than 18 (for prime power groups) or 20 (for non prime power groups).
DicksonExp (optional integer, default = 3): If the elimination method for finding the Dickson classes is used, it is needed to set a bound for the power to which the Dickson classes might be raised: If \(G\) is a \(p\)-group and \(n\) is the given DicksonExp, then the Dickson classes of the elementary abelian sub-groups of \(G\) are raised to the power of \(p^0,p^1,...,p^n\) before trying to simultaneously lift them to \(G\). We do not know any example in which the default value would not suffice.
useFactorization (optional boolean, default True): Try to replace the Dickson classes of \(G\) by their minimal non-constant factors. This may simplify some computations, but there are rare examples in which the factorisation is a bigger problem than a high degree bound.
auto (optional integer, default = 2 for abelian groups, and = 4 otherwise): Only applies to the case of prime power groups. A quick but potentially memory consuming method for lifting chain maps will be used in degree at most auto. For prime power groups up to orders around 500, the default value seems to be heuristically best.
useSlimgb / useStd (optinal boolean): Use Singular’s slimgb (resp. std) for computing the Groebner basis of the relation ideal. If not specified, Singular’s groebner method is chosen, which uses a heuristics to find the best algorithm for the computation of the Groebner basis.

Global options

options (optional string or list of strings): set/unset global options, or a dictionary that the global options are updated with.

There are various global options—they apply to all cohomology rings. Each option is set by a string, and unset by prepending 'no' to that string.

Available options

'warn' [default], 'info', 'debug', logging level

'useMTX' [default], use Matrix_gfpn_dense matrices for linear algebra over finite fields, which rely on SharedMeatAxe. Note that the resolutions will always be computed using the SharedMeatAxe. By consequence, if useMTX is turned off, time is wasted for conversions between different matrix types.

'save' [default], automatically save ring approximations, which comes in very handy when a long computation needs to be interrupted at some point; that’s why it is the default. Note that many data, including a minimal projective resolution for prime power groups, will always be stored on disk.

'sparse' [not default], remove temporarily unneeded data on the resolution from memory. With that option, the computation of very large examples becomes more feasible.

Further options have a numerical value:

autolift [default=1]: Degree up to which cochains are lifted using the autolift (as opposed to the Urbild Gröbner basis) method. Only applies to groups that are not elementary abelian.

autoliftElAb [default=0]: The same as autolift, but for elementary abelian groups.

SingularCutoff [default=70]: This determines how commands for Singular are cut into pieces, in order to reduce the overhead of the pexpect interface.

NrCandidates [default=1000]: Maximal number of candidates that are considered when trying to find special elements (e.g., parameters) by enumeration.

In experiments with unit_test_64(), the different options had the following effect:

With options="nouseMTX", the computation time slightly increases.
With options="sparse", the computation time increases.
With options="nosave", the computation time decreases.

The options can also be (un)set later, by using the method global_options().