Unsuccessful calculation of chemical equilibrium (a question).

[Dmitriy-Filatov-A]

Hello!
I use the Cantera 2.5 in C++. When I try to calculate the chemical equilibrium using the method "MultiPhaseEquil::equilibrate", the Cantera can finish failing (the Cantera doesn't send messages). I tried to find a mistake and it seems to me that it can depend on the calculation of the parameter "MultiPhaseEquil::error()"

cantera/src/equil/MultiPhaseEquil.cpp

Line 635 in 4570f3e

doublereal MultiPhaseEquil::error()

. To be more precise, this parameter can take a large constant value during each iteration.
I have a question. Why are you using this parameter (m_deltaG_RT[j]) as an "error"

cantera/src/equil/MultiPhaseEquil.cpp

Line 655 in 4570f3e

err = fabs(m_deltaG_RT[j]);

?
I tried to use the parameter "max(m_dxi)" as an error and the Cantera finished successfully. Can I use this parameter?
The parameter m_dxi is calculated in method "MultiPhaseEquil::computeReactionSteps(vector_fp& dxi)"

cantera/src/equil/MultiPhaseEquil.cpp

Line 516 in 4570f3e

doublereal MultiPhaseEquil::computeReactionSteps(vector_fp& dxi)

in line

cantera/src/equil/MultiPhaseEquil.cpp

Line 586 in 4570f3e

dxi[j] = -fctr*dg_rt;

P.S. The Cantera performs most calculations successfully without modifying the code but the Cantera does not perform some calculations (very few) for the reason I indicated.

Thank you in advance for your help.

[Dmitriy-Filatov-A]

P.S.
The method "MultiPhaseEquil::equilibrate" return the parameter "error" that was discussed.

cantera/src/equil/MultiPhaseEquil.cpp

Line 178 in 4570f3e

return error();

[ischoegl]

Hi @Dmitriy-Filatov-A ... Thank you for reporting this issue. Could you provide an example with numerical input so this can be reproduced?

[Dmitriy-Filatov-A]

Hello! Yes, of course. How can I send my project? (it has several files)

[Dmitriy-Filatov-A]

Before I submit the project, I write about problems that I found for some calculation points.
The Cantera does not return a result when I call the method "MultiPhaseEquil::equilibrate()"

cantera/src/equil/MultiPhaseEquil.cpp

Line 161 in 4570f3e

doublereal MultiPhaseEquil::equilibrate(int XY, doublereal err,

I think this is because the value of "error" from the method "MultiPhaseEquil::error()" stops changing at some iterative step and takes on a large value.

cantera/src/equil/MultiPhaseEquil.cpp

Line 635 in 4570f3e

doublereal MultiPhaseEquil::error()

I also noticed that when calling method "MultiPhaseEquil::computeReactionSteps(vector_fp& dxi)" in line

cantera/src/equil/MultiPhaseEquil.cpp

Line 436 in 4570f3e

doublereal grad0 = computeReactionSteps(m_dxi);

, the values of the vectors "m_dxi" and "m_moles" do not change but the values of the vector "m_work" aren't zero

cantera/src/equil/MultiPhaseEquil.cpp

Line 440 in 4570f3e

multiply(m_N, m_dxi.data(), m_work.data());

and vector "m_works" uses to calculate vector "m_moles" in the method "MultiPhaseEquil::step()"

cantera/src/equil/MultiPhaseEquil.cpp

Line 417 in 4570f3e

m_moles[k] += omega * deltaN[k];

cantera/src/equil/MultiPhaseEquil.cpp

Line 424 in 4570f3e

m_moles[k] += omega * deltaN[k];

cantera/src/equil/MultiPhaseEquil.cpp

Line 426 in 4570f3e

m_moles[k] = fabs(m_moles[k])*std::min(10.0,

So I think that the parameter "omega" tends to zero.

cantera/src/equil/MultiPhaseEquil.cpp

Line 407 in 4570f3e

void MultiPhaseEquil::step(doublereal omega, vector_fp& deltaN,

[ischoegl]

Thank you for providing more details here, @Dmitriy-Filatov-A ... the equilibrium solver is - to my knowledge - one of the older parts of Cantera (meaning that the original authors may not be available). So providing these insights is appreciated.

Regarding the example, is there a way to reproduce the behavior with a simple Python example script that specifies conditions (or C++ or Matlab if those are your preferences), together with a YAML (or CTI) input file that provides the species definitiions? That way, the behavior could be easily reproduced and would document a clear starting point.

[Dmitriy-Filatov-A]

I am sending a link with files.
https://github.com/Dmitriy-Filatov-A/test_Cantera.git

[Dmitriy-Filatov-A]

P.S. There are indicated temperatures for which the program does not work correctly (main.cpp)

[speth]

Thanks for reporting this issue. I've put together a somewhat simpler version based on built-in input files that can be run in Python, and involves fewer species and elements. Input file:

phases:
- name: gas
  thermo: ideal-gas
  species: [{nasa_gas.yaml/species: [O2, O, Fe, FeO]}]
  state: {T: 300, P: 1 atm}
- name: Fe(a)
  thermo: fixed-stoichiometry
  species: [{nasa_condensed.yaml/species: [Fe(a)]}]
  state: {T: 300, P: 1 atm}
- name: FeO(s)
  thermo: fixed-stoichiometry
  species: [{nasa_condensed.yaml/species: [FeO(s)]}]
  state: {T: 300, P: 1 atm}
- name: Fe2O3(s)
  thermo: fixed-stoichiometry
  species: [{nasa_condensed.yaml/species: [Fe2O3(s)]}]
  state: {T: 300, P: 1 atm}
- name: Fe3O4(s)
  thermo: fixed-stoichiometry
  species: [{nasa_condensed.yaml/species: [Fe3O4(s)]}]
  state: {T: 300, P: 1 atm}

code:

import cantera as ct

phase_names = [
	"Fe(a)", "FeO(s)", "Fe2O3(s)", "Fe3O4(s)", "gas"
]

phases = [ct.Solution("all_phases.yaml", name) for name in phase_names]
gas = phases[-1]
successes = 0
failures = 0

for T in range(300, 1000):
    gas.TPX = None, None, "O2: 2"
    moles = {"Fe(a)": 20, "gas": 50}
    mix = ct.Mixture([(p, moles.get(p.name, 0.0)) for p in phases])

    mix.T = T
    mix.P = 5 * ct.one_atm
    try:
        mix.equilibrate("TP", "gibbs")
        successes += 1
    except ct.CanteraError as err:
        print(f"Failed at T = {T}")
        failures += 1
        print(err)

print(f"{successes} successes, {failures} failures")

For this code, I get 688 successes and 12 failures, though I wouldn't be too surprised if this varies a bit on different platforms or with different compilers.

I don't know the logic behind basing the error condition on m_deltaG_RT, but I didn't find that changing it to use m_dxi instead had any benefit in terms of reducing the number of cases where the solver fails.