Retaining the property of Symmetric?

General Usage
, , ,
1

It doesn’t keep the Symmetric property

import JuMP, Gurobi, LinearAlgebra
model = JuMP.Model(Gurobi.Optimizer)
JuMP.@variable(model, X[1:2, 1:2] >= 0, Symmetric)
JuMP.optimize!(model)
Xt = JuMP.value.(X) # ⚠️ Can this behavior be improved?

# julia> typeof(X)
# LinearAlgebra.Symmetric{JuMP.VariableRef, Matrix{JuMP.VariableRef}}

# julia> typeof(Xt)
# Matrix{Float64} (alias for Array{Float64, 2})

Xt = LinearAlgebra.Symmetric(Xt) # Since JuMP do not support, I have to do this myself

Here is another example

import JuMP, Gurobi, LinearAlgebra
model = JuMP.Model(Gurobi.Optimizer)
JuMP.@variable(model, x); JuMP.@variable(model, y)
trivial_value = 999
JuMP.@expression(model, X, LinearAlgebra.Symmetric([0 x; trivial_value y]))
JuMP.optimize!(model)
Xt = JuMP.value.(X) # ⚠️ Can this behavior be improved?

julia> typeof(X)
LinearAlgebra.Symmetric{JuMP.AffExpr, Matrix{JuMP.AffExpr}}

julia> typeof(Xt)
Matrix{Float64} (alias for Array{Float64, 2})
2

This is a property of Julia, not JuMP:

julia> using LinearAlgebra

julia> x = LinearAlgebra.Symmetric([1 2; 2 3])
2×2 Symmetric{Int64, Matrix{Int64}}:
 1  2
 2  3

julia> x.^2
2×2 Matrix{Int64}:
 1  4
 4  9
2 Likes
3

Related issue on redundant operations in elementwise operations: map and broadcast on symmetric matrices is inefficient · Issue #643 · JuliaLang/LinearAlgebra.jl. However, would it be breaking to go farther and change the return type to Symmetric, even if most people would welcome it?

1 Like
4

Should I tell about it to LinearAlgebra.jl, or I just write Xt = LinearAlgebra.Symmetric(Xt) myself?