Saturation solves fail moving from 0.5.7 to 0.5.8

[lucpaoli]

Hey all,

With the update from 0.5.7 to 0.5.8, the saturation solver as we're using it seems to have become less robust, frequently breaking during iterations.

If you look on the SAFT_ML github page (which I think you all have access to), the difference can be seen in the output log between the runs in 22_tests vs the runs in 23_updated_Clapeyron, where "updated_Clapeyron" refers to us downgrading from 0.5.8. to 0.5.7.

I'm afraid I don't have any time to try to validate a minimum working example right now, but an example case where it was failing on for us is:
Sat solver failed for heptacosane = [380.44, 2.0, 3.0, 6.0, 20.01, 200.51] at T=94.33, Tc=266.06, muting...
where those parameters are input into SAFT VR Mie.

[lucpaoli]

It's worth noting we're not using the default initial guess during our iterations as well

[longemen3000]

Hmm, there was an update to increase accuracy on x0_sat_pure, but if it causes this, it needs to be reverted

[longemen3000]

@lucpaoli hmmm, i'm afraid i would need that initial point, we did add an stricter check here:

Clapeyron.jl/src/methods/property_solvers/singlecomponent/saturation/saturation.jl

Lines 77 to 88 in 758d5ce

	function check_valid_sat_pure(model,P_sat,V_l,V_v,T,ε0 = 5e7)
	ε = abs(V_l-V_v)/(eps(typeof(V_l-V_v)))
	ε <= ε0 && return false
	pl,dpdvl = p∂p∂V(model,V_l,T,SA[1.0])
	pv,dpdvv = p∂p∂V(model,V_v,T,SA[1.0])
	ε1 = ε0*eps(P_sat)
	return (dpdvl <= 0) && #mechanical stability of the liquid phase
	(dpdvv <= 0) && #mechanical stability of the vapour phase
	(1 - pl/P_sat < ε1) && #calculated pressure for liquid phase is aproximately the same
	(1 - pv/P_sat < ε1) #calculated pressure for vapour phase is aproximately the same
	#if ΔV > ε then Vl and Vv are different values
	end

so if pressures deviate significantly, then it would fail instead of giving an incorrect result

[longemen3000]

this would be solved on 0.5.10 (the check was done at the solver level instead)