qibo.hamiltonian.models built starting from their dense form

[BrunoLiegiBastonLiegi]

Currently, many qibo.hamiltonians.models implement their construction both starting from the dense form, i.e. the matrix representation, and their symbolic form, i.e. starting from qibo symbols (introduced in #1548). Since, the latter is a more flexible approach from which it is relatively easy to recover the former (a SymbolicHamiltonian.dense() call should be totally equivalent), I wonder whether it makes sense to keep the explicit dense construction as well.

https://github.com/qiboteam/qibo/blob/09cba741e391c07910daad0faccf9d5e212b61c2/src/qibo/hamiltonians/models.py#L83C1-L89C58

[alecandido]

Since, the latter is a more flexible

Is it sufficiently general to fully cover the other case? To the best of your understanding

[BrunoLiegiBastonLiegi]

I would say so, yes, the only counter argument I would think of is performance wise: it may be that since the dense construction is targeting a specific hamiltonian, it may be better optimized than the general symbolic to dense implementation (which is thought for any generic hamiltonian).

[alecandido]

I would say so, yes, the only counter argument I would think of is performance wise: it may be that since the dense construction is targeting a specific hamiltonian, it may be better optimized than the general symbolic to dense implementation (which is thought for any generic hamiltonian).

I guess that working with large Hamiltonians is terrible anyhow, because of the exponential scaling. And arrays have limited applications, since it's hard (or impossible) to use them beyond state-vector simulation.
For small ones, efficiency is not a concern anyhow.

So, if performances are the only issue, it doesn't seem to be very limiting...