How to reversely perform transient simulation?

 Dear Baiyuzhong,

 >  wish to perform backward transient simulation, i.e., for a given final condition for each node,

> wish to guess/calculte the initial node voltage.

What is the motivation for performing a "backwards transient simulation"? I can't imagine why this is needed. There cannot be a unique solution without knowing some starting point. I have not heard of such a thing.

Maybe he's invented a time machine?

I've never heard of anyone wanting to do this before.  

Andrew 

The time stepping algorithms used for transient simulation always run forward in time.  I don't think there is a way to do what you propose.

smlogan said:

 Dear Baiyuzhong,

 >  wish to perform backward transient simulation, i.e., for a given final condition for each node,

> wish to guess/calculte the initial node voltage.

What is the motivation for performing a "backwards transient simulation"? I can't imagine why this is needed. There cannot be a unique solution without knowing some starting point. I have not heard of such a thing.

Andrew Beckett said:

Maybe he's invented a time machine?

I've never heard of anyone wanting to do this before.  

Andrew 

 Wow, I wish I could invent a time machine. That would be great!!!

Here I just want to explore the convergence problem. To see whether any initial condition can converge to the same final.

Tom Volden said:

The time stepping algorithms used for transient simulation always run forward in time.  I don't think there is a way to do what you propose.

 

You're right.  The transient simulation is basically solving ODE eqations, right? Then if somehow I can change the time variable  't' to '-t', it would be reversely solving the ODE equations (In fact, I tried to use matlab to solve ODE like this, it works). The things I don't know how to access the the ODE model and change the variable. 

Well, this is not possible, and I'm sure we wouldn't implement such a capability. For a start, there are too many assumptions that time goes forward, and some components are not modelled via ODE (eg delays, nport, transmission lines). I suspect with a huge amount of work it might be feasible but one big problem would be how you would solve the end point that you are starting from. Right now we can do that with a dc operating point which works by finding a solution with all the capacitances and inductances removed, but for an arbitrary end point that you'd want to go backwards from, you'd have to solve for potentially part charged capacitances and inductances without being able to work out how you got there. I think the elements involving delay would be a problem too. 

Regardless of this, I seriously doubt it would help you understand anything about convergence.

So no, you can't do this and there is no prospect of this being implemented.

Regards,

Andrew 

Andrew Beckett said:

Well, this is not possible, and I'm sure we wouldn't implement such a capability. For a start, there are too many assumptions that time goes forward, and some components are not modelled via ODE (eg delays, nport, transmission lines). I suspect with a huge amount of work it might be feasible but one big problem would be how you would solve the end point that you are starting from. Right now we can do that with a dc operating point which works by finding a solution with all the capacitances and inductances removed, but for an arbitrary end point that you'd want to go backwards from, you'd have to solve for potentially part charged capacitances and inductances without being able to work out how you got there. I think the elements involving delay would be a problem too. 

Regardless of this, I seriously doubt it would help you understand anything about convergence.

So no, you can't do this and there is no prospect of this being implemented.

Regards,

Andrew 

Thanks for your reply. I still got some questions about this.

(1) what if all the components in the circuit that I'm interested can be modeled by ODE, I think it can be solved reversely by setting 't' to '-t' or setting all capacitances to be minus. Right? 

(2) for the end point, I can use DC simulation to obtain the final state and slightly change this final state to get certain number of end points, (so it's not arbitrary). Then start the backwards from these end points; in each backward, when any node voltage reaches 'vdd' or 'gnd', terminate  and get a solution. Finally obeserve how these solutions distribute.

(3) I didn't quite understand  the statement "you'd have to solve for potentially part charged capacitances and inductances without being able to work out how you got there“. What did you mean by "potentially part charged capacitances and inductances"

I already explained that there is no way of doing this, so I'm not quite sure what the point of persisting with these questions is. If you are starting from a DC operating point (my assumption was that you wanted to do this from some arbitrary condition), I fail to see how doing a reverse time simulation would tell you anything useful, even if it was possible, which it isn't.