balkirs@eecs.nwu.edu (Sina Balkir) (06/27/91)
I got a rather technical question regarding transient analysis section of the simulator. Although I am currently dealing with the version 2G6, I guess this also applies to the higher versions too. As known to everyone, spice is either using trapezoidal or variable order gear up to 6 th order for its numerical integration algorithms. Once the time point is computed, the validity of this point is checked using the Local Truncation Error (LTE) estimates. Since the time step sizes are non uniform throughout the analysis, the integration coefficients are computed for every time step. However, when it comes to the LTE test, the coefficients used for the estimates are all constants which are formulated for uniform (fixed-step size) time-step case. Also the mechanism of the transient analysis makes sure that the time step from previous to the next step is allowed to double at most. Although this fact may seem too obvious to be overlooked, can anybodysuggest a justification for using constants for LTEs although the integration is carried out with non-uniform step sizes. I appreciate your comments in advance, thanks ... --sina balkir: balkirs@delta.eecs.nwu.edu