Integration of reactor with water fails below Temperature of 328 K

[ulrich1a]

Cantera version 2.3

Operating System: Windows 10 64-bit

Python 3.5

Expected Behavior: reactor should integrate to next time step

Actual Behavior:

CanteraError thrown by Substance::set_T:
illegal temperature: 272.5

Steps to reproduce: run the following python script

import cantera as ct
myWater = ct.Water()
myWater.TP = 327.0, 1.0*ct.one_atm
r2 = ct.Reactor(contents=myWater)
sim = ct.ReactorNet([r2])
sim.advance(4.e-4)

[bryanwweber]

This is expected behavior. The minimum temperature allowed for the Water phase is 273.16 K, which you can check with

>>> myWater.min_temp
273.16

[bryanwweber]

On second thought, the Reactor shouldn't be doing anything. There aren't any energy transfers, so the temperature shouldn't decrease, so I'm not sure what's happening here. This might require looking into the solver.

You can get this to advance if you turn off the energy equation:

r2.energy_enabled = False
sim = ct.ReactorNet([r2])
sim.advance(4.0E-4)

Can you please give some more detail of your use case that raises this issue?

[ulrich1a]

I am interested in the temperature of a reactive gas pressurized in a reactor with water. The inflow of water is modeled with a valve. Also the heat transfer to the water is needed in this case.
This worked for me with Cantera 2.2. So it looks like a regression.

[skyreflectedinmirrors]

@bryanwweber could this be due to the FD Jacobian?
If there's any non-zero (potentially spurious) derivative of the temperature, you would end up evaluating the RHS at a temperature ~ T + dT/dt * epsilon, which could account for the "drop"

edit: on second thought, that looks like a pretty large deltaT to be evaluating at

[bryanwweber]

@ulrich1a Can you provide a complete script that worked with Cantera 2.2?

[ulrich1a]

Here is a version, that worked with water, although I used here
myWater = ct.Solution('liquidvapor.xml', 'water')
gas_compression.zip

[bryanwweber]

With the script you provided and Cantera 2.2.1 (and 2.2.0), I get an updateState error in the Reactor on macOS, so the script doesn't work for me. Can you confirm that it works for you?

[speth]

I also couldn't get this to work with Cantera 2.2, but the changes that generate the current behavior are the ones introduced in 1dc0ced which were implemented to resolve #257.

I think the issue here is that while the bracket_and_solve_root function is generally more robust than the Newton's method algorithm that was previously implemented, it does need some work so that it doesn't run into trouble trying to evaluate properties outside the valid temperature range for the state.

[ulrich1a]

The code in 1dc0ced contains the line:

+            TT = bmt::bracket_and_solve_root(
+                u_err, T, 1.2, true, bmt::eps_tolerance<double>(48), maxiter);

The reported temperature in the error message is just by a factor 1/1.2 smaller than the starting temperature. So for me it seems, that the factor 1.2 hardcoded in the function call has to be adjusted in case of pure liquids to avoid a temperature outside of the limits.

The following script was running without error in version 2.2.1:

# -*- coding: utf-8 -*-

import cantera as ct

#myWater = ct.Water()
myWater = ct.Solution('liquidvapor.xml', 'water')
#myWater.TP = 328.0, 7.0*ct.one_atm 
#myWater.TX = 328.0, 1.0 
myWater.TPX = (300.0, 7.0*ct.one_atm, 'H2O:1.0') # Temperatur 35°C
r2 = ct.Reactor(contents=myWater)
#r2.volume = 0.000001
r2.syncState()

print('Temperatur 1:',r2.T)
print(r2.thermo.P)
sim = ct.ReactorNet([r2])
sim.verbose = True
#print(r2.get_state())
sim.advance(4.e-4)
print('Temperatur 2:',r2.T)
print(r2.thermo.P)