[Up] [Previous] [Next] [Index]

3 A basic example

Sections

  1. First Step: Integers instead of group elements
  2. Signatures: An important tool
  3. Change of coset vs. brute force

This chapter shows a basic example of how to use RDS. Some of the functions used here make choices which might not be optimal but should suffice for most ``everyday'' situations. If you plan to do more involved computations, you should also see the other chapters to learn about the concepts behind these high-level functions.

Here we will construct relative difference sets of Dembowski-Piper type ``b'' and order 9 as an example. We will take the elementary abelian group as an example. The general idea is as follows: Find a ``nice'' normal subgroup U and generate relative difference sets coset by coset. The normal subgroup has to be chosen such that we know how many elements to choose from each coset modulo U.

The calculations here are very easy, a more demanding example can be found in chapter RDS:An Example Program.

3.1 First Step: Integers instead of group elements

Difference sets are represented by lists of integers. Every difference set is assumed to contain 1. This is assumed implicitly. So the lists representing difference sets must not contain 1 (a partial difference set of length n is hence represented by a list of length n-1). If a partial difference set contains 1, many functions will produce errors.

To find Difference sets in a group, say G, begin with generating the group (and forbidden subgroup) and defining the parameters. Like this:

gap> LoadPackage("rds");
----------------------------------------------------------------
Loading  RDS 1.2
by Marc Roeder (marc_roeder@web.de)
----------------------------------------------------------------
true
gap> k:=9;;lambda:=1;;groupOrder:=81;;
gap> forbiddenGroupOrder:=9;;
gap> G:=ElementaryAbelianGroup(groupOrder);
<pc group of size 81 with 4 generators>
gap> Gdata:=PermutationRepForDiffsetCalculations(G);; 
gap> N:=Subgroup(G,GeneratorsOfGroup(G){[1,2]});
Group([ f1, f2 ])
gap> Size(N)=forbiddenGroupOrder;   #just a test...
true

Once we have calculated Gdata, this will be used very often to represent the group G as it contains much more information.

3.2 Signatures: An important tool

The ``signature'' of a subset SsubseteqG of a group relative to a normal subgroup U is the multiset of numbers of elements S contains from each coset modulo U. Possible values of these numbers can be calculated a priori for relative difference sets.

gap> sigdat:=SignatureData(Gdata,N,k,lambda,10^5);;

The argument 105 depends on your degree of impatience. Larger numbers take more time in this step, but give better results for later reduction steps.

Now we will look for a ``nice'' normal subgroup. A normal subgroup is ``nice'', if it has only few signatures and the number of different entries in each signature is low. If you have different choices here do some experiments, to see what works. Let's see what we have:

gap> NormalSgsHavingAtMostNSigs(sigdat,1,[1..7]); 
[ rec( sigs := [ [ 3, 3, 3 ] ], subgroup := Group([ f1, f2, f3 ]) ), 
  rec( sigs := [ [ 3, 3, 3 ] ], subgroup := Group([ f1, f2, f4 ]) ), 
  rec( sigs := [ [ 3, 3, 3 ] ], subgroup := Group([ f1, f2, f3*f4 ]) ), 
  rec( sigs := [ [ 3, 3, 3 ] ], subgroup := Group([ f1, f2, f3*f4^2 ]) ) ]

The second parameter of NormalSgsHavingAtMostNSigs is the maximal number of signatures the subgroup may have. The third parameter gives the desired lengths of the signatures (the index of the normal subgroup).

So in this example we have no real choice. Let's take the first group for U. The signature means that we have to get 3 elements from each coset modulo U. So we generate startsets of length 2 in the trivial coset U (representing partial relative difference sets of length 3). This could be done using AllDiffsets, of course. But here we will use another method. The function StartsetsInCoset generates startsets in U by generating an initial set of startsets and then raising the length of each startset by 1. Then a reduction using signatures and automorphism is performed. This is done until all startsets have the desired length or no startset remains (in which case there is no relative difference set). For the reduction, a suitable set of automorphisms must be chosen. This is done by the function SuitableAutomorphismsForReduction:

gap> U:=last[1].subgroup;
Group([ f1, f2, f3 ])
gap> auts:=SuitableAutomorphismsForReduction(Gdata,U);
[ <permutation group of size 303264 with 8 generators> ]
gap> startsets:=StartsetsInCoset([],U,N,2,auts,sigdat,Gdata,lambda);
#I  Size 18
#I  1/ 0 @ 0:00:00.071
#I  Size 8
#I  1/ 0 @ 0:00:00.038
#I  -->1 @ 0:00:00.042
[ [ 4, 22 ] ]

For larger examples, this takes a wile. Taking 106 (or even more) for the generation of sigdat can save some time here. A few remarks about the parameters of StartsetsInCoset. The first parameter [] is the set of startsets which we start with (as we just started, this is empty). The second parameter is the coset we use to generate startsets and third parameter is the forbidden subgroup. The fourth parameter is the length of the startsets we want to generate (remember that 1 is assumed to be in every startset without being listed. So we want startsets of size 3 represented by lists of length 2. Hence the 2 in this place). Instead of auts a suitable list of groups of automorphisms of G in permutation representation may be inserted. These are used for the reduction of startsets. For large groups auts[1] it is a good idea to add some subgroups of auts[1] to the list (ascending in order) auts, as the reduction is done using the first group in the list and then reducing the already reduced list again using the next group.

3.3 Change of coset vs. brute force

Now we have startsets of length 2 in U and there are two possibilities:


(1) Find 3 more elements from another coset like this:
gap> cosets:=RightCosets(G,U);
[ RightCoset(Group( [ f1, f2, f3 ] ),<identity> of ...), 
  RightCoset(Group( [ f1, f2, f3 ] ),f4), 
  RightCoset(Group( [ f1, f2, f3 ] ),f4^2) ]
gap> startsets:=StartsetsInCoset(startsets,cosets[2],N,5,auts,sigdat,Gdata,lambda);
#I  Size 27
#I  1/ 0 @ 0:00:00.127
#I  Size 11
#I  1/ 0 @ 0:00:00.058
#I  -->1 @ 0:00:03.311
#I  Size 2
#I  2/ 2 @ 0:00:00.015
#I  -->2 @ 0:00:00.015
[ [ 4, 22, 5, 28, 73 ], [ 4, 22, 5, 28, 77 ] ]
And 3 more from the last one (of course, we could also change to force, but it seems to work this way...).
gap> startsets:=StartsetsInCoset(startsets,cosets[3],N,8,auts,sigdat,Gdata,lambda);
#I  Size 9
#I  1/ 0 @ 0:00:00.056
#I  Size 1
#I  1/ 1 @ 0:00:00.006
#I  -->1 @ 0:00:00.009
#I  Size 1
#I  1/ 1 @ 0:00:00.006
#I  -->1 @ 0:00:00.006
[ [ 4, 22, 5, 28, 73, 37, 66, 78 ] ]

So we found one difference set of order 9 in the elementary abelian group of order 81. To get the difference set containing 1 explicitly and as a subset of G, say

gap> PermList2GroupList(Concatenation(startsets[1],[1]),Gdata); 
[ f3, f1*f3^2, f4, f2*f3*f4, f1*f2^2*f3^2*f4, f1^2*f4^2, f2*f3^2*f4^2, 
  f1^2*f2^2*f3*f4^2, <identity> of ... ]


(2) Do a brute force search. Here we have to convert the forbidden group N into a list of integers Np. And we have to raise the length of the startsets by one before we can start. This is due to the ordering we chose (which is not necessarily compatible with the cosets modulo U).

gap> Np:=GroupList2PermList(Set(N),Gdata);
[ 1, 2, 3, 6, 7, 10, 16, 19, 32 ]
gap> startsets:=ExtendedStartsetsNoSort(startsets,[1..groupOrder],Np,8,Gdata,lambda);;
gap> Size(startsets);
54
gap> foundsets:=[];;     
gap> for set in startsets
>  do
>   Append(foundsets,AllDiffsets(set,[1..groupOrder],k-1,Np,Gdata,lambda));
> od;
gap> Size(foundsets);
162

Now foundsets contains 162 relative (9,9,9,1)-difference sets (represented by lists of length 8). These are all equivalent (as seen above). Equivalence can be tested like this:

gap> ReducedStartsets(foundsets,[Gdata.Aac],i->true,Gdata);
#I  Size 162
#I  1/ 0 @ 0:00:00.001
[ [ 4, 22, 36, 39, 49, 50, 60, 61 ] ]

[Up] [Previous] [Next] [Index]

RDS manual
February 2012