Re: [qucs:discussion] ERROR: TR1: Jacobian singular at t = 0.000e+000, aborting transient analysis

[al davis]

On Fri, 30 Dec 2022 21:23:59 +0100
Felix Salfelder via Gnucap-devel <gnucap-devel@gnu.org> wrote:
> [1] https://sourceforge.net/p/qucs/discussion/311050
> 


I am still in the moving mode.  I will try to make a more detailed
reply to Francisco's questions in a few days.

First ..  The person promoting LTspice is a troll.  He obviously knows
nothing about what he is saying, and made no attempt to understand the
original poster.

In this case, the "Jacobian singular" messages (From, Qucs) say that
the circuit doesn't work.  It goes up in smoke.  Gnucap does not give
that error message.  Instead, it says something like "open circuit",
which hopefully is more meaningful.  Apparently (and confirmed in a
private conversation with Mike E. a bunch of years ago) LTspice will
modify the circuit to get a good looking result, even though it is
likely to be incorrect.  Gnucap takes the opposite stance ...  It is
better to "fail" than to give a believable incorrect result.

Convergence failure means there is something wrong with your circuit or
your models.

In this case, we have an "idealized" schematic, with "idealzed" models.

Think about a circuit that has two current sources in series.  Let's
say one is 1 milliamp, the other is two milliamps.  What is the voltage
at the junction?  The answer is ....  You can't do that.  Circuit goes
up in smoke.

OK .. now make both 1 millamp.  Does it work now?   Maybe, but I still
have no way to  determine the voltage.

So this is what is wrong with the original exercise here.  With
idealized models of transistors, you have this situation.  The models
need to incorporate the non-ideal parts, otherwise the circuit doesn't
work.

As is typical of IC designs, there are lots of transistors.  Resistors
and capacitors are kept to a minimum.  Some of those transistors are
acting as resistors and capacitors.  With incomplete model parameters,
that part is missing, and you will not get a correct simulation.

There are no parameters on the transistors.  You might be tempted to
assume they are all the same.  They are not.  They are all based on the
same "process" but they could be different sizes.  This is why Spice
(and Spice format simulators such as Gnucap) use the model card, and
also some parameters for each instance, usually something like length,
width, and area.  You can specify things like this one is 10 times the
size as that one.