herbert@AEROSPACE.AERO.ORG (04/22/87)
------- Forwarded Message Return-Path: glick Received: by aero.ARPA (4.12/6.0.GT) id AA17331; Thu, 27 Feb 86 08:15:42 pst Posted-Date: 27 Feb 86 08:15:25 PST (Thu) Received-Date: Thu, 27 Feb 86 08:15:42 pst Message-Id: <8602271615.AA17331@aero.ARPA> To: herbert@aero Date: 27 Feb 86 08:15:25 PST (Thu) From: glick@aero - ------- Forwarded Message Return-Path: herbert@aero Received: by aero.ARPA (4.12/6.0.GT) id AA26480; Tue, 14 Jan 86 11:04:04 pst Posted-Date: 14 Jan 86 11:01:19 PST (Tue) Received-Date: Tue, 14 Jan 86 11:04:04 pst Message-Id: <8601141904.AA26480@aero.ARPA> To: Jim Gouge <jgouge@mitre-gateway.arpa> Cc: glick@aero Subject: Re: SPICE problem inquiry In-Reply-To: Your message of 3 Jan 1986 15:52:32 EST (Friday). <8601032054.AA16435@aero.ARPA> Date: 14 Jan 86 11:01:19 PST (Tue) From: herbert@aero In response to your message requesting help with a SPICE problem relating to too small an internal timestep, I have a few comments which I believe will be helpful. SPICE2 has an automatic timestep feature which varies the solution time- step size in an effort to obtain accuracy and convergence throughout the transient solution process. This means that at times when a circuit response is characterized by rapid changes, SPICE will typically reduce the internal timestep in order to converge to a solution. However, when a circuit response is changing slowly, larger internal timesteps are permitted. The internal timestep too small error may result from nonconvergence; that is, the internal timestep is reduced excessively in an effort to converge to a solution. The error also may result from reducing the internal timestep in an effort to reduce the roundoff error (truncation error) resulting from the use fo second-order Taylor expansions for charge-control elements. The minimum internal timestep that SPICE will use is 1E-9*TMAX if TMAX is specified on the .TRAN card or, if TMAX is not specified, SPICE will use the smaller of 1E-9*TSTEP or 1E-9*(TSTOP-TSTART)/50. If the error is the result of nonconvergence, a solution may often be obtained by relaxing the convergence criteria (RELTOL, ABSTOL, VNTOL, CHGTOL) with the .OPTIONS card and/or by increasing the transient analysis timepoint iteration limit (ITL4). If the error is due to roundoff error, it may help to increase the transient error tolerance (TRTOL) and ITL4. You also can try running with just iteration timestep control (LVLTIM=1); in this mode SPICE will not test for roundoff error. However, if you obtain solution by applying any of these techniques, you must assess the accuracy of the results; that is, let the user beware! You may ask why it should be necessary to play around with the options in order to get SPICE to solve your problem. It is, I believe, because the option default values were determined empirically to work with a large class of representative test circuits. However, it is impossible to tune the options so that they will work with all problems. Therefore, the options are provided so that they may be conveniently modified for any circuit which proves to be difficult to solve. With the aid of the options and a few other tricks, I can't remember a problem I couldn't get to run. Cheers, Don Herbert - ------- End of Forwarded Message send ------- End of Forwarded Message