How to solve the convergence problem in Saber

The questions from netizens are as follows: Teacher Domono, the most common and complicated question encountered in Saber simulation is ---

No convergence. This is the biggest difficulty for us beginners. I encountered it again today. The following error occurred when doing DC analysis: *** ERROR "ALG_SINGULAR_JACOBIAN" *** Singular Jacobian matrix. Some possible causes for a singular Jacobian matrix: - Model operating outside of its intended region of operation. - Node/subsystem with no connection to the reference (floating). - Current loops (voltage sources/inductors in parallel). - Missing equation and/or missing variable reference.

Is the non-convergence caused by grounding or interface problems? Or other reasons? Can you explain it in detail? Thank you! In fact, the error message contains the possible reasons for non-convergence. Let's take a look at them one by one. - Model operating outside of its intended region of operation The model is out of bounds, that is, the result of the calculation makes the model exceed its normal operating range. For the simulator, the calculation results of one or some nodes or variables have singular points at a certain moment (usually when the simulation is interrupted). These points are very different from the original calculation results (usually on the order of 10 to the power of n), which makes it impossible for the simulator to obtain a system solution within the allowable error range, so it does not converge. - Node/subsystem with no connection to the reference (floating).

The node or subsystem is missing a reference point. This means that a node in the target system is in a floating state and the simulator cannot get the correct solution for the node. In this case, you need to carefully search the system to see if this situation occurs. If so, you need to add a pull-up or pull-down bias to the node.

Current loops (voltage sources/inductors in parallel). Current loops appear in the system. Current sources can only be connected in parallel and not in series. If there are more than two current sources in a loop, the simulator cannot get a normal solution and will not converge. Interestingly, another situation is listed in the parentheses above, that is, the voltage source and the inductor are directly connected in parallel. In this case, the DC analysis cannot converge, because the impedance of the inductor is regarded as 0 in DC analysis, and the voltage source is directly connected in parallel with it, which will cause the inductor current to be infinite, causing the DC analysis to not converge. - Missing equation and/or missing variable reference. It is indeed a system equation or a system variable reference. This situation is easy to occur when writing MAST language modeling by yourself (the models that come with Saber are all verified and will not occur in this situation). It mainly refers to the fact that in the self-built model, the defined system variables have no related system equations, which makes the simulator unable to calculate the system variables, resulting in simulation non-convergence.

Among the four situations prompted, the fourth one is mainly related to the self-built model. If you don't have your own MAST program, you don't have to consider it. The second and third ones are prone to non-convergence at the beginning of DC analysis or TR analysis, and they are relatively easy to diagnose. The first one may occur at any time during TR analysis, which will be reflected in the simulation results. Open the simulation result file and carefully check the variable waveforms of each node to find the node variables with sudden waveform simulation changes. At this time, it is necessary to study the model related to the variable and avoid this convergence problem by adjusting the system structure or model parameters.