Skip to content

[fix] BlockArrays multiplication with non-Array base types and CI fixes #71

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
lkdvos merged 1 commit into from
Aug 25, 2026
Merged

[fix] BlockArrays multiplication with non-Array base types and CI fixes #71

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
26 changes: 10 additions & 16 deletions src/linalg/factorizations.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -74,22 +74,16 @@ for f in [
MAK.$f!(t, out, MAK.select_algorithm(MAK.$f!, t, nothing; alg.kwargs...))
end

# specializations until fixes in base package
function MAK.is_left_isometric(A::BlockMatrix; atol::Real = 0, rtol::Real = MAK.defaulttol(A), norm = LinearAlgebra.norm)
P = A' * A
nP = norm(P) # isapprox would use `rtol * max(norm(P), norm(I))`
for I in MAK.diagind(P)
P[I] -= 1
end
return norm(P) <= max(atol, rtol * nP) # assume that the norm of I is `sqrt(n)`
end
function MAK.is_right_isometric(A::BlockMatrix; atol::Real = 0, rtol::Real = MAK.defaulttol(A), norm = LinearAlgebra.norm)
P = A * A'
nP = norm(P) # isapprox would use `rtol * max(norm(P), norm(I))`
for I in MAK.diagind(P)
P[I] -= 1
end
return norm(P) <= max(atol, rtol * nP) # assume that the norm of I is `sqrt(n)`
# specializations until fixes in base packages:
# - `similar(::MulAdd{<:AbstractBlockLayout,...})` hardcodes `Array` blocks, so `A' * A` on
# non-CPU blocks mixes host and device storage (JuliaArrays/BlockArrays.jl#215).
# - `MAK.diagview` silently does not alias into a `BlockMatrix`, so subtracting the identity there is a no-op.
# Densifying sidesteps both, and preserves the storage type of the blocks.
function MAK.is_left_isometric(A::BlockMatrix; kwargs...)
return MAK.is_left_isometric(copy_dense!(similar_dense(A), A); kwargs...)
end
function MAK.is_right_isometric(A::BlockMatrix; kwargs...)
return MAK.is_right_isometric(copy_dense!(similar_dense(A), A); kwargs...)
end

# Make sure sparse blocktensormaps have dense outputs
Expand Down
4 changes: 2 additions & 2 deletions test/abstracttensor/braidingtensor.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,10 +39,10 @@ end
@testset "Issue #68" begin
x = randn(ComplexF64, (⊞(ℙ^1) ⊗ ⊞(ℙ^1)) ← ⊞(ℙ^1))
O = randn(ComplexF64, ((ℙ^1 ⊞ ℙ^4 ⊞ ℙ^16 ⊞ ℙ^4) ⊗ ⊞(ℙ^2)) ← (⊞(ℙ^2) ⊗ ⊞(ℙ^1)))
A = randn(ComplexF64, (ℙ^4 ⊗ F^2 ⊗ (ℙ^2)') ← ℙ^1)
A = randn(ComplexF64, (ℙ^4 ⊗ ℙ^2 ⊗ (ℙ^2)') ← ℙ^1)
Ab = randn(ComplexF64, (ℙ^4 ⊗ ℙ^2 ⊗ (ℙ^2)') ← ℙ^1)

@plansor y[-1 -2; -3] ≔ A[-1 4 2; 1] * O[-2 6; 4 5] * τ[5 7; 2 3] *
conj(Ab[-3 6 7; 8]) * x[1 3; 8]
@test space(y) == (⊞(ℙ^4) ⊗ ⊞(ℙ^1)) ← ⊞(ℙ^4)
@test space(y) == ((⊞(ℙ^4) ⊗ (ℙ^1 ⊞ ℙ^4 ⊞ ℙ^16 ⊞ ℙ^4)) ← ⊞(ℙ^4))
end
Loading