Incorrect indication of absence on analytic jacobian when using daeMode?

[Francesco Casella]

Please run ScalablePowerGrids.Tests.GridBusSimpleControl_N_1_M_2, which uses -daeMode=new because of the vendor annotation, and also includes -d=forceNLSanalyticJacobian. OMC issues the following warning:

Iteration variables with default zero start attribute in equation system w/o analytic Jacobian:
         linesHorizontal[1,1].vz.im:VARIABLE(unit = "V" nominal = 230940.1076758503 )  "Real part of complex voltage".ScalablePowerGrids.Tests.GridBusSimpleControl_N_1_M_2, .PowerGrids.Electrical.Branches.LineConstantImpedance$linesHorizontal, .PowerGrids.Types.ComplexVoltage type: Real  [1,1]
         linesHorizontal[1,1].vz.re:VARIABLE(unit = "V" nominal = 230940.1076758503 )  "Imaginary part of complex voltage".ScalablePowerGrids.Tests.GridBusSimpleControl_N_1_M_2, .PowerGrids.Electrical.Branches.LineConstantImpedance$linesHorizontal, .PowerGrids.Types.ComplexVoltage type: Real  [1,1]

I assume the problem is supposed to be in the initial equations, which is where the start attributes really matter.

The last optdaedump output shows a set of initial equations which seems easy to handle - no function calls or record equations.

In fact, if I run the model with the debugger, I get a nonlinear system of 52 initial equations in 52 scalar variables, which appears to be also computing the jacobian analytically, since there are 56 assignments concering the jacobian in the system. BTW, I don't understand why $csevariables are used to avoid repeated sin and cos function computation for the jacobian but not for the residual, but that's a separate issue and it's not critical.

It is not clear whether there is some issue that ultimately prevents the computation of the analytical jacobian at runtime, or if the presence of the analytical jacobian is ignored due to some bug, or if the issue is simply a wrong error message.

Please clarify that and either fix the symbolic jacobian or the error message

[Willi Braun]

Replying to casella:

Please run ScalablePowerGrids.Tests.GridBusSimpleControl_N_1_M_2, which uses -daeMode=new because of the vendor annotation, and also includes -d=forceNLSanalyticJacobian. OMC issues the following warning:

Iteration variables with default zero start attribute in equation system w/o analytic Jacobian:
         linesHorizontal[1,1].vz.im:VARIABLE(unit = "V" nominal = 230940.1076758503 )  "Real part of complex voltage".ScalablePowerGrids.Tests.GridBusSimpleControl_N_1_M_2, .PowerGrids.Electrical.Branches.LineConstantImpedance$linesHorizontal, .PowerGrids.Types.ComplexVoltage type: Real  [1,1]
         linesHorizontal[1,1].vz.re:VARIABLE(unit = "V" nominal = 230940.1076758503 )  "Imaginary part of complex voltage".ScalablePowerGrids.Tests.GridBusSimpleControl_N_1_M_2, .PowerGrids.Electrical.Branches.LineConstantImpedance$linesHorizontal, .PowerGrids.Types.ComplexVoltage type: Real  [1,1]

I assume the problem is supposed to be in the initial equations, which is where the start attributes really matter.

This is a false message and have been fixed in ​PR2582.

The last optdaedump output shows a set of initial equations which seems easy to handle - no function calls or record equations.

True, so there is actually no need for -d=forceNLSanalyticJacobian.

In fact, if I run the model with the debugger, I get a nonlinear system of 52 initial equations in 52 scalar variables, which appears to be also computing the jacobian analytically, since there are 56 assignments concering the jacobian in the system. BTW, I don't understand why $csevariables are used to avoid repeated sin and cos function computation for the jacobian but not for the residual, but that's a separate issue and it's not critical.

This because the wrapFunctionCalls is not used for the initialization system.

It is not clear whether there is some issue that ultimately prevents the computation of the analytical jacobian at runtime, or if the presence of the analytical jacobian is ignored due to some bug, or if the issue is simply a wrong error message.

Please clarify that and either fix the symbolic jacobian or the error message