Improve Initialization in dfChemistryModel

[xiao312]

Summary

This pull request addresses the issue of repetitive initialization in the Cantera Reactor by calling syncState().

Performance Test

Versions Tested

• DeepFlame Version:

  • Commit 0d4e5ee
  • Commit v1.3 94c068d

• Test Environment:

  • Local machine

• Example Used:

  • deepflame-dev/examples/dfLowMachFoam/notorch/twoD_HIT_flame/CH4
  • Dynamic load balancing disabled

Logs

• 0d4e5ee Log:

Time = 1e-06
========Time Spent in diffenet parts========
loop Time                    = 21.7963 s
other Time                   = 1e-06 s
rho Equations                = 0.010316 s
U Equations                  = 0.194424 s
Y Equations                  = 2.65603 s
E Equations                  = 0.18264 s
p Equations                  = 0.531852 s
chemistry correctThermo      = 0.907214 s
turbulence correct           = 2.6e-05 s
combustion correct(in Y)     = 17.2706 s
percentage of chemistry      = 79.2364 %
percentage of rho/U/Y/E      = 13.963 %

• v1.3 94c068d Log:

Time = 1e-06
========Time Spent in diffenet parts========
loop Time                    = 22.8885 s
other Time                   = 0 s
rho Equations                = 0.010008 s
U Equations                  = 0.184964 s
Y Equations                  = 2.60104 s
E Equations                  = 0.159949 s
p Equations                  = 0.495936 s
chemistry correctThermo      = 0.802762 s
turbulence correct           = 2e-05 s
combustion correct(in Y)     = 18.6158 s
percentage of chemistry      = 81.3325 %
percentage of rho/U/Y/E      = 12.9146 %

Python

import cantera as ct
import numpy as np
import time

# Assuming mech and data are defined
gas = ct.Solution(mech)
# Create Reactor and ReactorNet
r = ct.Reactor(gas, name='R1', energy='off')
sim = ct.ReactorNet([r])
sim.rtol, sim.atol = 1e-6, 1e-10

time_steps = []  # List to store time for each iteration

start_time = time.time()  # Start timer
# Repeat the entire process 100 times
for _ in range(10000):
    # Initialize the state
    npstate = data[0, :11]  # Assuming data is defined and has sufficient rows
    T_old, P_old, Y_old = npstate[0], npstate[1], npstate[2:]
    
    gas.TPY = T_old, P_old, Y_old
    res_1st = [T_old, P_old] + list(gas.Y) + list(gas.partial_molar_enthalpies / gas.molecular_weights)

    # Measure time for advancing the simulation
    r.syncState()
    sim.reinitialize()    
    sim.advance(time_step)
    sim.initial_time = 0.0
    

    # Get new state
    new_TPY = [gas.T, gas.P] + list(gas.Y) + list(gas.partial_molar_enthalpies / gas.molecular_weights)
    res_1st += new_TPY
    
elapsed_time = time.time() - start_time  # Calculate elapsed time
# Print average time after all iterations
print("Elapsed times for each iteration:", elapsed_time)
# print("Average time per iteration:", np.mean(time_steps))



# Assuming mech and data are defined
gas = ct.Solution(mech)

time_steps = []  # List to store time for each iteration

start_time = time.time()  # Start timer
# Repeat the entire process 100 times
for _ in range(10000):
    # Initialize the state
    npstate = data[0, :11]  # Assuming data is defined and has sufficient rows
    T_old, P_old, Y_old = npstate[0], npstate[1], npstate[2:]
    
    gas.TPY = T_old, P_old, Y_old
    res_1st = [T_old, P_old] + list(gas.Y) + list(gas.partial_molar_enthalpies / gas.molecular_weights)

    # Create Reactor and ReactorNet
    r = ct.Reactor(gas, name='R1', energy='off')
    sim = ct.ReactorNet([r])
    sim.rtol, sim.atol = 1e-6, 1e-10

    # Measure time for advancing the simulation
    
    sim.advance(time_step)
    

    # Get new state
    new_TPY = [gas.T, gas.P] + list(gas.Y) + list(gas.partial_molar_enthalpies / gas.molecular_weights)
    res_1st += new_TPY
    
elapsed_time = time.time() - start_time  # Calculate elapsed time
# Print average time after all iterations
print("Elapsed times for each iteration:", elapsed_time)
# print("Average time per iteration:", np.mean(time_steps))

Results:

Elapsed times for each iteration: 1.2987749576568604
Elapsed times for each iteration: 1.4612696170806885

Conclusions

An approximate 8% to 10% speedup could be expected, although this improvement may diminish with larger mesh sizes.