StarAlgebras
The package implements *
-algebras with basis. The prime example use is group/monoid algebras (or rings) (or their finite dimensional subspaces). An example usage can be as follows.
julia> using StarAlgebras
julia> using PermutationGroups
julia> G = PermGroup(perm"(1,2)", perm"(1,2,3)")
Permutation group on 2 generators generated by
(1,2)
(1,2,3)
julia> b = StarAlgebras.Basis{UInt8}(collect(G))
6-element StarAlgebras.Basis{Permutation{Int64, …}, UInt8, Vector{Permutation{Int64, …}}}:
()
(2,3)
(1,2)
(1,3,2)
(1,3)
(1,2,3)
julia> RG = StarAlgebra(G, b)
*-algebra of Permutation group on 2 generators of order 6
This creates the group algebra of the symmetric group. How do we compute inside the group algebra? There are a few ways to comstruct elements:
julia> zero(RG)
0·()
julia> one(RG) # the canonical unit
1·()
julia> RG(1) # the same
1·()
julia> RG(-5.0) # coerce a scalar to the ring
-5.0·()
julia> RG(rand(G)) # the indicator function on a random element of G
1·(1,3,2)
julia> f = AlgebraElement(rand(-3:3, length(b)), RG) # an element given by vectors of coefficients in the basis
1·() -1·(2,3) +3·(1,2) -1·(1,3,2) -3·(1,3) +3·(1,2,3)
One may work with such element using the following functions:
julia> StarAlgebras.coeffs(f)
6-element Vector{Int64}:
1
-1
3
-1
-3
3
julia> StarAlgebras.star(p::PermutationGroups.AbstractPerm) = inv(p); star(f) # the star involution
1·() -1·(2,3) +3·(1,2) +3·(1,3,2) -3·(1,3) -1·(1,2,3)
julia> f' # the same
1·() -1·(2,3) +3·(1,2) +3·(1,3,2) -3·(1,3) -1·(1,2,3)
julia> g = rand(G); g
(1,2,3)
julia> StarAlgebras.coeffs(RG(g)) # note the type of coefficients
6-element SparseArrays.SparseVector{Int64, UInt8} with 1 stored entry:
[6] = 1
julia> x = one(RG) - 3RG(g); supp(x) # support of the funtion
2-element Vector{Permutation{...}:
()
(1,2,3)
julia> x(g) # value of x at g
-3
julia> x[g] += 3; x # modification of x in-place
1·()
julia> aug(f) # sum of coefficients
2
julia> using LinearAlgebra; norm(f, 2) # 2-norm
5.477225575051661
Using this we can define e.g. a few projections in RG
and check their orthogonality:
julia> using Test
julia> Base.sign(p::PermutationGroups.Permutation) = sign(p.perm);
julia> l = length(b)
6
julia> P = sum(RG(g) for g in b) // l # projection to the subspace fixed by G
1//6·() +1//6·(2,3) +1//6·(1,2) +1//6·(1,3,2) +1//6·(1,3) +1//6·(1,2,3)
julia> @test P * P == P
Test Passed
julia> P3 = 2 * sum(RG(g) for g in b if sign(g) > 0) // l # projection to the subspace fixed by Alt(3) = C₃
1//3·() +1//3·(1,3,2) +1//3·(1,2,3)
julia> @test P3 * P3 == P3
Test Passed
julia> P2 = (RG(1) + RG(b[2])) // 2 # projection to the C₂-fixed subspace
1//2·() +1//2·(2,3)
julia> @test P2 * P2 == P2
Test Passed
julia> @test P2 * P3 == P3 * P2 == P # their intersection is precisely the same as the one for G
Test Passed
julia> P2m = (RG(1) - RG(b[2])) // 2 # orthogonal C₂-fixed subspace
1//2·() -1//2·(2,3)
julia> @test P2m * P2m == P2m
Test Passed
julia> @test iszero(P2m * P2) # indeed P2 and P2m are orthogonal
Test Passed
This package originated as a tool to compute sum of hermitian squares in *
-algebras. These consist not of standard f*f
summands, but rather star(f)*f
. You may think of semi-definite matrices: their Cholesky decomposition determines P = Q'·Q
, where Q'
denotes transpose. Algebra of matrices with transpose is an (the?) example of a *
-algebra. To compute such sums of squares one may either sprinkle the code with star
s, or '
(aka Base.adjoint
postfix symbol):
julia> x = RG(G(perm"(1,2,3)"))
1·(1,2,3)
julia> X = one(RG) - x
1·() -1·(1,2,3)
julia> X'
1·() -1·(1,3,2)
julia> X'*X
2·() -1·(1,3,2) -1·(1,2,3)
julia> @test X'*X == star(X)*X == 2one(X) - x - star(x)
Test Passed
More advanced use
RG = StarAlgebra(G, b)
creates the algebra with TrivialMStructure
, i.e. a multiplicative structure which computes product of basis elements every time it needs it. This of course may be wastefull, e.g. the computed products could be stored in a cache for future use. There are two options here:
julia> mt = StarAlgebras.MTable(b, table_size=(length(b), length(b)))
6×6 StarAlgebras.MTable{UInt8, false, Matrix{UInt8}}:
0x01 0x02 0x03 0x04 0x05 0x06
0x02 0x01 0x06 0x05 0x04 0x03
0x03 0x04 0x01 0x02 0x06 0x05
0x04 0x03 0x05 0x06 0x02 0x01
0x05 0x06 0x04 0x03 0x01 0x02
0x06 0x05 0x02 0x01 0x03 0x04
creates an eagerly computed multiplication table on elements of b
. Keyword table_size
is used to specify the table size (above: it's the whole multiplication table). Since MTable<:AbstractMatrix
, one can use the indexing syntax mt[i,j]
to compute the index of the product of i
-th and j
-th elements of the basis. For example
julia> g = G(perm"(1,2,3)"); h = G(perm"(2,3)");
julia> i, j = b[g], b[h] # indices of g and h in basis b
(0x06, 0x02)
julia> k = mt[i,j] # the index of the product
0x05
julia> @test b[k] == g*h
Test Passed
The second option is
julia> cmt = StarAlgebras.CachedMTable(b, table_size=(length(b), length(b)));
This multiplication table is lazy, i.e. products will be computed and stored only when actually needed. Additionally, one may call
julia> using SparseArrays
julia> StarAlgebras.CachedMTable(b, spzeros(UInt8, length(b), length(b)));
to specify storage type of the matrix (by default it's a simple dense Matrix
).
This may be advisable when a few products are computed repeatedly on a quite large basis.
julia> RGc = StarAlgebra(G, b, cmt)
*-algebra of Permutation group on 2 generators of order 6
should be functinally equivalent to RG
above, however it will cache computation of products lazily. A word of caution is needed here though. Even though RGc
and RG
are functionally equivalent, they are not comparable in the sense that e.g.
julia> @test one(RGc) != one(RG)
Test Passed
This is a conscious decision on our part, as comparing algebraic structures is easier said than done ;) To avoid solving this conundrum (are bases equal? are multiplicative structures equal? are these permuted by a compatible permutation? or maybe a linear transformation was applied to the basis, resulting in a different, but equivalent multiplicative structure?), elements could be mixed together only if their parents are identically (i.e. ===
) equal.
Finally, if the group is infinite (or just too large), but we need specific products, we may reduce the table_size to the required size (it doesn't have to be length(b) × length(b)
). Note that in such case asking for a product outside of multiplication table will rise ProductNotDefined
exception.
Even more advanced use (for experts only)
For low-level usage MultiplicativeStructures
follow the sign convention:
julia> mt = StarAlgebras.CachedMTable(b, table_size=(length(b), length(b)));
```julia
julia> k = mt[-i,j]
0x06
julia> @test star(b[i])*b[j] == b[k]
Test Passed
Note that this (minus-twisted) "product" is no longer associative! Observe:
julia> @test mt[mt[3, 5], 4] == mt[3, mt[5, 4]] # (b[3]*b[4])*b[5] == b[3]*(b[4]*b[5])
Test Passed
julia> @test mt[-signed(mt[-3, 5]), 4] == 0x06 # star(star(b[3])*b[5])*b[4] = star(b[5])*b[3]*b[4]
Test Passed
julia> @test mt[-3, mt[-5, 4]] == 0x01 # star(b[3])*star(b[5])*b[4]
Test Passed
If you happen to use this package please cite either 1712.07167 or 1812.03456. This package superseeds GroupRings.jl which was developed and used there. It served its purpose well. Let it rest peacefully.