Unary operations on adjoint fail

[andreasvarga]

I encoutered the following errors:

julia> L = LinearOperator(rand(2,2))
Linear operator
  nrow: 2
  ncol: 2
  eltype: Float64
  symmetric: false
  hermitian: false

julia> -L'
ERROR: type AdjointLinearOperator has no field nrow
Stacktrace:
 [1] getproperty(::Any, ::Symbol) at .\Base.jl:20
 [2] -(::AdjointLinearOperator{Float64,getfield(LinearOperators, Symbol("##10#13")){Array{Float64,2}},getfield(LinearOperators, Symbol("##11#14")){Array{Float64,2}},getfield(LinearOperators, Symbol("##12#15")){Array{Float64,2}}}) at C:\Users\Andreas\.julia\packages\LinearOperators\La2Ey\src\LinearOperators.jl:260
 [3] top-level scope at REPL[56]:1

julia> 2*L'
ERROR: type AdjointLinearOperator has no field nrow
Stacktrace:
 [1] getproperty(::Any, ::Symbol) at .\Base.jl:20
 [2] *(::Int64, ::AdjointLinearOperator{Float64,getfield(LinearOperators, Symbol("##10#13")){Array{Float64,2}},getfield(LinearOperators, Symbol("##11#14")){Array{Float64,2}},getfield(LinearOperators, Symbol("##12#15")){Array{Float64,2}}}) at C:\Users\Andreas\.julia\packages\LinearOperators\La2Ey\src\LinearOperators.jl:311
 [3] top-level scope at REPL[57]:1

[abelsiqueira]

Thanks for the report @andreasvarga.

@dpo, define -A' as (-A)'?

[dpo]

Yes

[andreasvarga]

Just a comment: Implementing -A' as (-A)' is OK. However, if you intend to implement 2A' as (2A)' , this may raise the question which operation should have priority: the multiplication with 2 or the adjoint. In the LinearAlgebra package of Julia, the multiplication defines the result, and therefore 2*Adjoint(a) does not lead to and Adjoint type, see below:

julia> 2*rand(2,2)'
2×2 Array{Float64,2}:
 1.25212  0.0511282
 1.77288  1.844

This approach has sense, because it is always possible to enforce the Adjoint type by computing (2*A)'.

[dpo]

Do you mean that you're interested in cases where multiplication and transpose don't commute?

[andreasvarga]

No. I just observed the focus in thr LinearAlgebra package. Dominique <notifications@github.com> schrieb am So., 3. Nov. 2019, 15:17:

…

Do you mean that you're interested in cases where multiplication and transpose don't commute? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#107?email_source=notifications&email_token=ALJDHEHNLC5GHESUXCNHQPLQR3FO5A5CNFSM4JIE2342YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEC5TSUY#issuecomment-549140819>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ALJDHEFDRJQHY4BS5YSRT7TQR3FO5ANCNFSM4JIE234Q> .

[dpo]

Because adjoint and transpose operators don't have explicit [ct]prod methods, I'm not sure how else it can be implemented.