Primitivity of the truncated log-signature #
A series b is primitive for the shuffle coproduct when b([]) = 0 and
Σ_{x ∈ u ⧢ v} b(x) = 0 for all nonempty u, v (the boundary sums
against [] are automatic, so this is Δ_⧢ b = b ⊗ 1 + 1 ⊗ b in
coefficient form). Main theorem: the truncated logarithm of a
group-like series is primitive up to the truncation degree — the
first-order shadow of "the log-signature lies in the free Lie algebra".
Proof: shuffle sums of (a-1)^{⊗k} expand via the
shuffle–deconcatenation compatibility Word.shuffle_splits_perm as
Σ_{i,j} N(k,i,j)·(a-1)^{⊗i}(u)·(a-1)^{⊗j}(v), where N(k,i,j) is the
coefficient of xⁱyʲ in (x+y+xy)^k, realised in ℤ[y][x]. The
required vanishing Σ_k (-1)^k/(k+1)·N(k+1,i,j) = 0 for i,j ≥ 1 is the
coefficient identity of log((1+x)(1+y)) = log(1+x) + log(1+y): after
d/dx the sum becomes geometric, (1+x+y+xy)·W = (1-(-x-y-xy)^n)(1+y)
pins down the low-degree coefficients of W by a bidegree recursion.
Primitive series #
A series is primitive for the shuffle coproduct: the empty coefficient vanishes and all mixed shuffle sums over nonempty word pairs vanish.
Equations
Instances For
The bilinear pairing form of shuffle–deconcatenation #
compatibility
Shuffle sums of a tensor product expand over the splittings of
both words — the pairing form of Word.shuffle_splits_perm.
List-sum utilities #
Splitting sums against the unit, and short-word vanishing #
Bidegree coefficients of (x + y + xy)^k in ℤ[y][x] #
The shuffle expansion of tensor powers of the augmentation part #
The main theorem #
The truncated log-signature of a group-like series is primitive up to the truncation degree. This is the coefficient-level statement that the logarithm of a shuffle character is an infinitesimal character — the first-order form of "log-signatures are Lie elements".
The truncated log-signature of a bundled group-like signature is primitive up to the truncation degree.