Potential solver artifact / integration query at steady-state (GEAR, 1e-6)

[Ben Black]

Hello Xyce team,

I have a netlist where the results are exactly what I mathematically sought, but they are being dismissed by reviewers as a solver error.

Specifically, the simulation shows that the internal current is substantially and consistently greater than the source power at steady-state. Because of this outcome, I am being told this is impossible and must be a failure in the Xyce solver.

To try and break the simulation or smooth out artifacts, I have already forced GEAR integration and tightened tolerances to 1e-6, but the results remain rock solid. The console output confirms the matrix isn't just failing to converge; it is executing cleanly:

• Number Successful Steps Taken: 160,102

• Number Failed Linear Solves: 0

• Number Nonlinear Convergence Failures: 0

I need to know: is Xyce accurately calculating the mathematics of this netlist, or did I genuinely find a bug/artifact in how the engine handles this specific configuration?

I have attached the .cir netlist and .mt0.

I would greatly appreciate any insight into whether the solver is executing this accurately or failing.

Regards, Ben

[Mehmet Cirit]

If your design includes capacitors, they can store energy for later use, even after disconnecting the power source. That is expected behavior.

--
You received this message because you are subscribed to the Google Groups "xyce-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to xyce-users+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/xyce-users/0c3ef17c-3f33-4c6d-b96e-6f597821109cn%40googlegroups.com.

--

http://www.linkedin.com/in/mehmetcirit

Dr. Mehmet A. Cirit                    Phone:  (408) 647-6025

Library Technologies, Inc.        Cell:       (408) 647-6025

19959 Lanark Lane                   http://www.libtech.com

Saratoga, CA 95070                 Email: m...@libtech.com

[Ben Black]

- Dr. Cirit: Thank you for your observation regarding standard capacitor transient discharge characteristics. However, the netlist in question evaluates a continuous 60-Hz AC steady-state equilibrium, not a transient discharge event.

- XUG Team: As I noted previously, this netlist resolves the exact localized apparent power virtualization and thermodynamic parity I mathematically sought, yet external reviewers continue to casually dismiss these outputs as a solver error.

Xyce is and has been a valuable tool. To help put this to rest, can the team confirm if this matrix output is mathematically valid as simulated, or if the solver is producing an undocumented integration artifact at these phase boundaries? Thank you.

For quick reference, the embedded dashboard visualizes the solver's exact steady-state outputs, mapping the continuous current magnification against absolute thermodynamic parity.  

Regards,
Ben
(Running XyceNF Release 7.10.0)

[Ben Black]

[Mehmet Cirit]

Try simulating with other simulators first to see if the response differs. 

On Thu, Aug 13, 2026 at 9:48 PM Ben Black <black...@gmail.com> wrote:

--
You received this message because you are subscribed to the Google Groups "xyce-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to xyce-users+...@googlegroups.com.

To view this discussion visit https://groups.google.com/d/msgid/xyce-users/3396bb0a-3e42-4246-9ec9-d65319017b71n%40googlegroups.com.

[Ben Black]

Dr. Cirit: Cross-platform verification has already been successfully conducted in LTspice and Siemens AMESIM, yielding the identical steady-state equilibrium and power parity shown above.

XUG Team: Because these steady-state metrics fundamentally challenge conventional textbook assumptions, they are routinely dismissed by external observers as a "simulator artifact" or a numerical solver error. I brought this specific netlist directly to this group because Xyce is recognized as the apex standard for high-assurance DAE matrix evaluation.

The core query is not whether other simulators agree, but whether the Xyce KLU solver is accurately resolving this boundary equilibrium without undocumented integration artifacts. I am requesting the core team's attestation on the mathematical integrity of the solver in this specific topology.

The complete dataset, comparative netlists, and open documentation are cataloged on Zenodo for independent review,
(DOI: 10.5281/zenodo.18719728).

Should the matrix stability of this baseline prove of interest, the dataset also contains the RR-GPI_5D-LCFVR_Resilient-Resonance_Netlist. This evaluates the identical phase boundary under significantly higher computational stress, demonstrating the configuration's ability to maintain thermodynamic parity even when a non-linear Jiles-Atherton inductive branch is subjected to severe, near-critical kinetic impedance (20,500 Ω).

Thank you for your time and for maintaining an invaluable tool.

[Kevin Cameron]

Just FYI, I did a wrapper script for xyce to emulate LTspice and Qspice on Windows using WSL.

https://github.com/kev-cam/ltz

It's using an alternative compiler for the Verilog-A, so YMMV.

Kev.

--

You received this message because you are subscribed to the Google Groups "xyce-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to xyce-users+...@googlegroups.com.

To view this discussion visit https://groups.google.com/d/msgid/xyce-users/CAMq64UG5fBn8bayT87OgeJqUc9r-CKY%3DWLKyN9h_Z0AkTwteAg%40mail.gmail.com.

[xyce-users]

Hello Ben,

This is an interesting query, and I'll address it as soon as I can.   I'm out on leave at the moment, so I can't do much from here right now.

I haven't looked at the netlist itself yet, but in circuits where the currents and/or voltages far exceed typical values, there is an option in Xyce to use different error analysis on the current or voltage in question.  So, that might be a thing to try.   But, usually the problem in that case is that the solver fails altogether as it cannot reach convergence.   So, that may not be the solution.

Anyway, I'll try to get back to you as soon as I can.

thanks,

Eric

[xyce-users]

Hi Ben,

I can report that the simulator is accurately calculating the mathematics described in this netlist.  The results are not impacted by the choice of linear solver, KLU and KSparse both generate the same results and do not report any solver failures.  Furthermore, as mentioned by Eric above, the time integrator offers different options that select the reference value used to compute the local truncation error (LTE) in the adaptive time integrator.  These are discussed in the "Error Controls" section of the Xyce User Guide.  Using any of the time integration LTE options, I am seeing the same results as you are.  There is nothing that gives me a reason to think Xyce is inaccurate in performing the simulation described in the provided netlist.

Hope that helps!

Cheers,

Heidi

[Ben Black]

Thank you Heidi and the XUG team for the guidance on this.

With this confirmation, I am now finalizing my documentation on the Tri-Branch-GPI 'Hydra' NEMA 0.5HP Equilibrium Netlist—which builds directly on these validated mechanics to sustain steady-state phase isolation under continuous load—in my upcoming v15 Zenodo update (DOI: 10.5281/zenodo.18719728) for rigorous scientific evaluation.

I will forward a courtesy copy of that finalized dataset directly to the core development team's official contact email strictly for their internal reference.

I truly appreciate the support in helping verify the baseline. We can consider this modeling inquiry fully resolved and closed.

Regards,

Ben