Small fixes and changelog updates (#331)

* fix #314 (again)

* avoid warning in precompilation

* update changelog

* add LinearAlgebra.svd overload

Author: lkdvos

docs/src/Changelog.md

@@ -20,6 +20,18 @@ When releasing a new version, move the "Unreleased" changes to a new version sec
 
 ## [Unreleased](https://github.com/QuantumKitHub/TensorKit.jl/compare/v0.16.0...HEAD)
 
+### Added
+
+- Extended support for selecting storage types in the `TensorMap` constructors ([#327](https://github.com/QuantumKitHub/TensorKit.jl/pull/327))
+- `similar_diagonal` to handle storage types when constructing diagonals ([#330](https://github.com/QuantumKitHub/TensorKit.jl/pull/330))
+- `LinearAlgebra.svd` overloads
+
+### Fixed
+
+- Issue with using relative tolerances in truncation schemes ([#314](https://github.com/QuantumKitHub/TensorKit.jl/issues/314))
+- Using `scalartype` instead of `eltype` in BLAS contraction ([#326](https://github.com/QuantumKitHub/TensorKit.jl/pull/326))
+- Divide by zero error in `show` for empty tensors ([#329](https://github.com/QuantumKitHub/TensorKit.jl/pull/329))
+
 ## [0.16.0](https://github.com/QuantumKitHub/TensorKit.jl/releases/tag/v0.16.0) - 2025-12-08
 
 ### Added

src/factorizations/factorizations.jl

@@ -12,7 +12,7 @@ using ..TensorKit: AdjointTensorMap, SectorDict, SectorVector,
 
 using LinearAlgebra: LinearAlgebra, BlasFloat, Diagonal,
     svdvals, svdvals!, eigen, eigen!,
-    isposdef, isposdef!, ishermitian
+    isposdef, isposdef!
 
 using TensorOperations: Index2Tuple
 
@@ -37,10 +37,10 @@ TensorKit.one!(A::AbstractMatrix) = MatrixAlgebraKit.one!(A)
 #------------------------------#
 
 function LinearAlgebra.eigen(t::AbstractTensorMap; kwargs...)
-    return ishermitian(t) ? eigh_full(t; kwargs...) : eig_full(t; kwargs...)
+    return LinearAlgebra.ishermitian(t) ? eigh_full(t; kwargs...) : eig_full(t; kwargs...)
 end
 function LinearAlgebra.eigen!(t::AbstractTensorMap; kwargs...)
-    return ishermitian(t) ? eigh_full!(t; kwargs...) : eig_full!(t; kwargs...)
+    return LinearAlgebra.ishermitian(t) ? eigh_full!(t; kwargs...) : eig_full!(t; kwargs...)
 end
 
 function LinearAlgebra.eigvals(t::AbstractTensorMap; kwargs...)
@@ -49,6 +49,11 @@ function LinearAlgebra.eigvals(t::AbstractTensorMap; kwargs...)
 end
 LinearAlgebra.eigvals!(t::AbstractTensorMap; kwargs...) = eig_vals!(t)
 
+LinearAlgebra.svd(t::AbstractTensorMap; full::Bool = false) =
+    full ? svd_full(t) : svd_compact(t)
+LinearAlgebra.svd!(t::AbstractTensorMap; full::Bool = false) =
+    full ? svd_full!(t) : svd_compact!(t)
+
 function LinearAlgebra.svdvals(t::AbstractTensorMap)
     tcopy = copy_oftype(t, factorisation_scalartype(svd_vals!, t))
     return LinearAlgebra.svdvals!(tcopy)

src/factorizations/truncation.jl

@@ -145,8 +145,7 @@ end
 # Find truncation
 # ---------------
 # auxiliary functions
-rtol_to_atol(S, p, atol, rtol) =
-    rtol == 0 ? atol : max(atol, TensorKit._norm(S, p, norm(zero(scalartype(valtype(S))))) * rtol)
+rtol_to_atol(S, p, atol, rtol) = rtol == 0 ? atol : max(atol, norm(S, p) * rtol)
 
 function _compute_truncerr(Σdata, truncdim, p = 2)
     I = keytype(Σdata)

test/tensors/factorizations.jl

@@ -282,13 +282,14 @@ for V in spacelist
                 @test norm(t - U3 * S3 * Vᴴ3) ≈ ϵ3 atol = eps(real(T))^(4 / 5)
                 @test space(S3, 1) ≾ space(S2, 1)
 
-                trunc = truncerror(; atol = ϵ2)
-                U4, S4, Vᴴ4, ϵ4 = @constinferred svd_trunc(t; trunc)
-                @test t * Vᴴ4' ≈ U4 * S4
-                @test isisometric(U4)
-                @test isisometric(Vᴴ4; side = :right)
-                @test norm(t - U4 * S4 * Vᴴ4) ≈ ϵ4 atol = eps(real(T))^(4 / 5)
-                @test ϵ4 ≤ ϵ2
+                for trunc in (truncerror(; atol = ϵ2), truncerror(; rtol = ϵ2 / norm(t)))
+                    U4, S4, Vᴴ4, ϵ4 = @constinferred svd_trunc(t; trunc)
+                    @test t * Vᴴ4' ≈ U4 * S4
+                    @test isisometric(U4)
+                    @test isisometric(Vᴴ4; side = :right)
+                    @test norm(t - U4 * S4 * Vᴴ4) ≈ ϵ4 atol = eps(real(T))^(4 / 5)
+                    @test ϵ4 ≤ ϵ2
+                end
 
                 trunc = truncrank(nvals) & trunctol(; atol = λ - 10eps(λ))
                 U5, S5, Vᴴ5, ϵ5 = @constinferred svd_trunc(t; trunc)