3. Brief Introduction to Inputs

The dictionary CanteraTorchProperties is the original dictionay of DeepFlame. It read in network related parameters and configurations. It typically looks like:

chemistry           on;
CanteraMechanismFile "ES80_H2-7-16.yaml";
transportModel "Mix";
odeCoeffs
{
    "relTol"   1e-15;
    "absTol"   1e-24;
}
inertSpecie        "N2";
zeroDReactor
{
    constantProperty "pressure";
}

splittingStretagy false;

TorchSettings
{
    torch on;
    GPU   off;
    log  on;
    torchModel1 "ESH2-sub1.pt";
    torchModel2 "ESH2-sub2.pt";
    torchModel3 "ESH2-sub3.pt";
    coresPerNode 4;

}
loadbalancing
{
        active  false;
        //log   true;
}

In the above example, the meanings of the parameters are:

  • CanteraMechanismFile: the name of the reaction mechanism file.

  • transportModel: the default model is Mix, but other models including UnityLewis and Multi are also availabile.

  • constantProperty: property set to be constant during reaction. It can be set to pressure or volume.

  • odeCoeffs: the ode torlerance. 1e-15 and 1e-24 are used for network training, so it should keep the same when comparing results with and without DNN. Default values are 1e-6 and 1e-10.

  • TorchSettings: all paramenters regarding the usage of DNN. This section will not be read in CVODE cases.

  • torch: the switch used to control the on and off of DNN. If users are running CVODE, this needs to be switched off.

  • GPU: the switch used to control whether GPU or CPU is used to carry out inference.

  • torchModel: name of network.

  • coresPerNode: If you are using one node on a cluster or using your own PC, set this parameter to the actual number of cores used to run the task. If you are using more than one node on a cluster, set this parameter the total number of cores on one node. The number of GPUs used is auto-detected.