Pre-allocated eigen value decomposition

Specific Domains Numerics
1

Is there a way to pre-allocate an array to hold the returned eigenvalues and vectors from eigen or eigen! ? Currently both functions return a factorization which does not seem to allow a pre-allocation. The following code shows what I would like to do but does not work:

julia> A=rand(100,100);

julia> evals=zeros(CF64,100);

julia> evecs=zeros(CF64,100,100);

julia> julia> esys=Eigen(evals,evecs);

julia> esys.values[1]
0.0 + 0.0im

julia> evals[1]=1;

julia> esys.values[1]
1.0 + 0.0im

julia> esys = eigen(A)

julia> esys.values[1]
-3.0686209886339606 - 1.1725958631992774im   

julia> evals[1]
1.0 + 0.0im

Here CF64=ComplexF64. I would hope that eigen would write to evals and evecs but apparently it does not. I can also do something like this:

pre-allocated-eigen!(A,evals,evecs);

I haven’t found such a function though. Any suggestions? Or is there a reason why this is not making sense?

1 Like
2

If there is no support for preallocated results I’d assume the cost of allocating is negligible in comparison to the actual computation?

3

You might have to call the lower-level LAPACK routines directly. (Preallocation shouldn’t make much difference unless your matrices are fairly small.)