All-Pass Function OPF: A Solver-Friendly, Physics-Preserving Approximation of AC OPF

arXiv:2601.14468 · eess.SY, cs.NA, cs.SY, math.NA, math.OC · Submitted 2026-01-20 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: We're looking at the paper: "All-Pass Function OPF".

Dev: Here's the gist of the paper. This paper presents an all-pass fractional approximation of the AC optimal power flow (AC OPF) problem based on an all-pass approximation of the exponential power flow kernel.

Rosa: The proposed APF OPF formulation remains nonconvex, as in the classical model, but it eliminates trigonometric evaluations and empirically produces larger and more stable Newton steps under standard interior-point solvers.

Dev: Numerical results on more than 25 IEEE and PGLib test systems ranging from 9 to 10,000 buses demonstrate that the APF OPF model achieves solutions with accuracy comparable to that of the classical formulation while reducing solver times.

Rosa: Problem Motivation The classical AC OPF relies on trigonometric coupling between bus voltage phasors, which yields a nonconvex program with oscillatory derivatives that can slow, or in some cases destabilize, interior-point methods.

Paper summary: Rosa: So this paper, "All-Pass Function OPF: A Solver-Friendly, Physics-Preserving Approximation of AC OPF," it’s looking at the classic AC optimal power flow problem and trying to make it easier for solvers. The core idea is using an all-pass approximation of the exponential power flow kernel instead of the trigonometric functions that make things so hard.

Dev: Right, so you're saying they take these tricky trigonometric couplings between bus voltage phasors and replace them with something smoother that still respects how losses behave in the network. But you mentioned it remains nonconvex, right? That’s a key detail for us to track as we look at how solvers handle it.

Taro: I always worry about when these approximations fail, Rosa; what happens when the world misbehaves and the real power flow deviates from this approximation?

Rosa: Well, they admit that the proposed all-pass fractional optimal power flow problem is still nonconvex, just like the original model. But they argue that this formulation eliminates those messy trigonometric evaluations entirely, and it empirically produces larger and more stable Newton steps when using standard interior-point solvers.

Dev: Larger and more stable steps sound good for convergence speed, but what about the actual accuracy? Is this a trade-off where you sacrifice precision for computational time?

Rosa: They show numerical results on over twenty five IEEE and PGLib test systems, ranging from nine to ten thousand buses, and they claim the all-pass optimal power flow model achieves solutions with accuracy comparable to the classical formulation while reducing solver times.

Taro: So, across this wide range of system sizes, it holds up in terms of matching the actual voltage and angle profiles of a real AC power flow? Does it just work for small systems, or does it hold for the big ones too?

Dev: The paper shows that APF-OPF maintains negligible optimality gaps and reproduces voltage and angle profiles of the classical model across small, medium, and large network sizes. They actually show speedups ranging from about eleven percent in case thirty to over seventy percent in case fifty seven for small systems.

Rosa: That scale variation is interesting because it suggests the method isn't just a patch for one specific network size; it seems pretty robust across different system complexities. It’s not just a trick for one setup, right?

Taro: It sounds like they introduced a pre-rotation strategy to try and fix some of the approximation errors in practical operating regions. How does that pre-rotation actually help make the approximation more accurate where it matters most?

Dev: They introduce a pre-rotation step using a DC power flow solution to shift the angle differences toward zero, defining delta i j as delta dc i j plus some difference between those DC angles. The idea is that this angle difference term tends to become small near the optimum, which is precisely where they designed the all-pass approximation to be most accurate.

Rosa: So, it’s a targeted fix for accuracy based on an initial guess from a simpler model before running the full optimization. It’s like tuning a radio dial before you turn it on to get the clearest signal in that spot.

Taro: And what about those derivatives? The paper mentions that the derivatives of the all-pass components are smooth fractional functions, which is much better than the oscillations from trigonometric functions. Does this smoothness actually translate into something better for an interior-point solver's math?

Dev: Yeah, it leads to better conditioned Jacobian and Hessian matrices because those smooth fractional functions don't oscillate like the trig terms do. That means the solvers can construct larger and more reliable Newton steps, which is what drives that speedup we saw earlier.

Rosa: It really boils down to how the mathematics of the problem behaves under optimization. Instead of facing those highly oscillatory derivatives, you’re dealing with functions that are smoother and easier for the solver to handle efficiently.

Taro: If we think about what this means for real-world deployment, is this approximation something we can trust when the grid is actually operating under real AC conditions, not just in a test environment?

Dev: The conclusion is that it consistently improves computational performance across all network sizes while maintaining close agreement with the classical AC optimal power flow in terms of objective value, feasibility under true AC evaluation, congestion patterns, and voltage angle behavior. It offers a physics-preserving and solver-friendly alternative to the original formulation.

Rosa: So for us on the outside looking in, what does this mean when we think about power system planning or real-time control? It means we might be able to run these complex calculations faster without losing that crucial physical accuracy.

Taro: I see it as a way to tackle the computational burden of AC power flow problems, making them more accessible for larger, more detailed simulations that we need when testing how systems recover from failures.

Dev: Exactly, it’s about balancing the computational cost against the fidelity of the solution. It’s a tool that lets us get better performance without having to jump to a completely different mathematical framework.

Conclusion: Rosa: So we're wrapping up this paper about "All-Pass Function OPF: A Solver-Friendly, Physics-Preserving Approximation of AC OPF." It turns out they took the classic optimal power flow problem and used some math tricks to make it run much better on computers.

Dev: They are using an all-pass approximation for the exponential power flow kernel instead of those standard trigonometric functions that cause problems with solvers. They claim this makes the Newton steps larger and more stable when you use interior-point software.

Taro: So, the main point is they kept it nonconvex but smoothed out the math so it doesn't crash or get stuck as fast as before. But Rosa, does this smoothness mean we can trust the results when we run these simulations on real-world equipment?

Rosa: They tested it on over twenty five IEEE and PGLib test systems, ranging from nine to ten thousand buses. And they say the accuracy of their solution is comparable to the classical way, but with much less solver time. It’s a big claim, but they show it holds up across different system sizes.

Dev: The paper shows that for small networks, you get speedups of maybe eleven percent, but for medium and large ones, they see improvements over fifty percent in some cases. That’s significant speedup when you’re dealing with these power flow models.

Taro: So what does this actually change for someone who just runs the simulation? It means we can probably solve these AC power flow problems much faster without losing that crucial physical accuracy in voltage and angle behavior. But they did admit it's still nonconvex, so we need to watch out for those tricky areas of the solution space.

Rosa: Exactly. The paper says this method is a physics-preserving alternative because it keeps the loss behavior correct while making the math friendlier for solvers. It’s less about getting a perfect mathematical solution and more about getting a fast, accurate one in practice.

Dev: So what's next on their list? They touched on how they do the pre-rotation strategy to make things even more accurate near the optimum. That sounds like a good follow-up to explore when we look at real-time control scenarios.

Milad Hasanzadeh, Amin Kargarian, Javad Lavaei

Louisiana State University · University of California, Berkeley

eess.SY, cs.NA, cs.SY, math.NA, math.OC

Submitted: 2026-01-20

Updated: 2026-11-08

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 94/100

The gist: This paper presents an all-pass fractional approximation of the AC optimal power flow (AC OPF) problem based on an all-pass approximation of the exponential power flow kernel.

Key concepts

All-Pass Kernel Approximation
This involves replacing the original exponential power flow kernel with a first-order all-pass function, r(δ) = 1 + jaδ / 1 - jaδ. This function has a magnitude of one for all real angles, which ensures that only network resistances affect real power losses, preserving the physical behavior of the original model.
Pre-Rotation Strategy
A pre-rotation step is used based on DC power flow solutions to shift angle differences ($\Delta ij$) toward zero. This technique improves approximation accuracy because the key terms in the all-pass function become small near the optimum, making it more accurate where it is most needed.
Smooth Derivative Structure
The proposed all-pass components have smooth fractional derivatives, unlike trigonometric functions which oscillate. This smoothness results in better-conditioned Jacobian and Hessian matrices for solvers, allowing interior-point methods to take larger and more reliable Newton steps without instability.

Terminology

Summary

This paper presents an all-pass fractional approximation of the AC optimal power flow (AC OPF) problem based on an all-pass approximation of the exponential power flow kernel. p. 0 The proposed APF OPF formulation remains nonconvex, as in the classical model, but it eliminates trigonometric evaluations and empirically produces larger and more stable Newton steps under standard interior-point solvers. p. 0 Numerical results on more than 25 IEEE and PGLib test systems ranging from 9 to 10,000 buses demonstrate that the APF OPF model achieves solutions with accuracy comparable to that of the classical formulation while reducing solver times. p. 0

Problem Motivation

The classical AC OPF relies on trigonometric coupling between bus voltage phasors, which yields a nonconvex program with oscillatory derivatives that can slow, or in some cases destabilize, interior-point methods. This nonlinear formulation leads to a highly nonconvex feasible set, making AC OPF computationally challenging. p. 0 The authors address this by proposing an approximation-based method that preserves the original OPF structure but simplifies the AC power flow equations using reduced-complexity models. p. 0

All-Pass Kernel Approximation

The paper introduces a first-order all-pass function r(δ) = 1 + jaδ / 1 - jaδ, which is called all-pass because its magnitude is identically one for all real δ. This function preserves the loss behavior of the original exponential kernel by ensuring that only network resistances contribute to real power losses The real and imaginary parts are approximated as cos(δ) ≈ rcos(δ):= 1 - (aδ)2 / 1 + (aδ)2, and sin(δ) ≈ rsin(δ):= 2aδ / 1 + (aδ)2. p. 3

Pre-Rotation Strategy

To improve approximation accuracy in practical operating regions, a pre-rotation step is introduced using a DC power flow solution to shift the angle differences toward zero. This involves defining deltaij = δdc ij + ∆ij, where ∆ij = (θi − θj) − (θdc i − θdc j). p. 0 The key idea is that the term ∆ij tends to become small near the optimum, allowing the APF to be most accurate where it is designed to be. p. 3

Numerical Benefits and Validation

The primary benefit of APF OPF lies in its improved derivative structure, as the derivatives of the all-pass components are smooth fractional functions that do not oscillate like those from trigonometric functions This smoothness leads to better-conditioned Jacobian and Hessian matrices, which allows interior-point solvers to construct larger and more reliable Newton steps Numerical experiments show that APF OPF achieves solutions with accuracy comparable to the classical formulation while reducing solver times.

Comparative Performance Across Scales

Numerical experiments across IEEE and PGLib test systems demonstrate that APF-OPF maintains negligible optimality gaps and reproduces voltage and angle profiles of the classical model across small, medium, and large network sizes. Speedup varies from about 11% in case30 to more than 70% in case57 for small systems and shows improvements up to over 50% in medium-scale cases. p. 7

Conclusion

The proposed APF OPF formulation consistently improves computational performance across all network sizes while maintaining close agreement with the classical AC OPF in terms of objective value, feasibility under true AC evaluation, congestion patterns, and voltage angle behavior. The results confirm that the proposed APF OPF formulation offers a physics-preserving and solver-friendly alternative to AC OPF. p. 9

References

[1] S. Frank and S. p. 9 Rebennack, “An introduction to optimal power flow: Theory, formulation, and examples,” IIE transactions, vol. p. 9 48, no. 12, pp. 1172–1197, 2016. p. 9

[2] S. H. Low, “Convex relaxation of optimal power flow—part i: Formulations and equivalence,” IEEE Transactions on Control of Network Systems, vol. p. 9 1, no. 1, pp. 15–27, 2014. p. 9

[3] D. K. Molzahn, I. p. 9 A. Hiskens et al. p. 9, “A survey of relaxations and approximations of the power flow equations,” Foundations and Trends® in Electric Energy Systems, vol. p. 9 4, no. 1-2, pp. 1–221, 2019. p. 9

[4] D. Bienstock and A. p. 9 Verma, “Strong NP-hardness of AC power flows feasibility,” Operations Research Letters, vol. p. 9 47, no. 6, pp. 494–501, 2019. p. 9

[5] M. Hasanzadeh and A. p. 0 Kargarian, “ADMM enhancement techniques for distributed optimal power flow,” IEEE Transactions on Power Systems, 2025. p. 9

[6] R. Madani, S. Sojoudi, and J. p. 9 Lavaei, “Convex relaxation for optimal power flow problem: Mesh networks,” IEEE Transactions on Power Systems, vol. p. 9 30, no. 1, pp. 199–211, 2014. p. 9

[7] S. Bose, S. H. Low, T. Teeraratkul, and B. p. 9 Hassibi, “Equivalent relaxations of optimal power flow,” IEEE Transactions on Automatic Control, vol. p. 9 60, no. 3, pp. 729–742, 2014. p. 9

[8] C. Coffrin, H. p. 9 L. Hijazi, and P. p. 9 Van Hentenryck, “Strengthening convex relaxations with bound tightening for power network optimization,” in International conference on principles and practice of constraint programming. p. 9 Springer, 2015, pp. p. 9 39–57.

[9] R. A. Jabr, “Radial distribution load flow using conic programming,” IEEE transactions on power systems, vol. p. 9 21, no. 3, pp. 1458–1459, 2006. p. 9

[10] P. Buason, S. Misra, J.-P. Watson, and D. p. 9 K. Molzahn, “Adaptive power flow approximations with second-order sensitivity insights,” IEEE Transactions on Power Systems, 2024. p. 9

[11] A. Goodwin, J. p. 9 Maack, and D. p. 9 Sigler, “Power flow geometry and approximation,” IEEE Transactions on Power Systems, 2025. p. 9

[12] B. Stott, J. Jardim, and O. p. 9 Alsac¸, “DC power flow revisited,” IEEE Transactions on Power Systems, vol. p. 9 24, no. 3, pp. 1290–1300, 2009. p. 9

[13] J. Lavaei and S. p. 0 H. Low, “Zero duality gap in optimal power flow problem,” IEEE Transactions on Power systems, vol. p. 9 27, no. 1, pp. 92–107, 2011. p. 9

[14] M. Farivar and S. p. 9 H. Low, “Branch flow model: Relaxations and convexification—part i,” IEEE Transactions on Power Systems, vol. p. 9 28, no. 3, pp. 2554–2564, 2013. p. 9

[15] A. Wachter and L. p. 9 T. Biegler, “On the implementation of an ¨ interior-point filter line-search algorithm for large-scale nonlinear programming,” Mathematical programming, vol. p. 9 106, no. 1, pp. 25–57, 2006. p. 9

[16] J. Nocedal and S. p. 9 J. Wright, Numerical optimization. p. 9 Springer, 2006. p. 9

[17] R. D. Zimmerman, C. p. 9 E. Murillo-Sanchez, and R. p. 9 J. Thomas, “MATPOWER: Steady-state operations, planning, and analysis tools for power systems research and education,” IEEE Transactions on power systems, vol. p. 9 26, no. 1, pp. 12–19, 2010. p. 9

[18] R. D. Zimmerman and C. p. 9 p. 9 E. Murillo-Sanchez, “Matpower 6. p. 9 p. 90 user’s ´ manual,” Power Systems Engineering Research Center, vol. p. 9 p. 9 9, pp. 65–66, 2016. p. 9 p. 9

[19] D. P. Bertsekas, “Nonlinear programming,” Journal of the Operational Research Society, vol. p. 9 48, no. 3, pp. 334–334, 1997. p. 9

[20] A. V. Oppenheim, A. p. 9 p. 9 S. Willsky, and S. p. 9 p. 9 H. Nawab, Signals & Systems. p. 9 p. 9

[21] J. Lofberg, “YALMIP: A toolbox for modeling and optimization in MATLAB,” in 2004 IEEE international conference on robotics and automation (IEEE Cat. p. 9 p. 9 No. 04CH37508). p. 9 p. 9 IEEE, 2004, pp. p. 9 p. 9 284–289.

[22] J. A. Andersson, J. p. 9 p. 9 p. 9 p. 9 p. 9 Gillis, G. Horn, J. B. Rawlings, and M. p. 9 p. 9 p. 9 Diehl, “CasADi: a software framework for nonlinear optimization and optimal control,” Mathematical Programming Computation, vol. p. 9 p. 9 p. 9 11, no. 1, pp. 1–36, 2019. p. 9 p. 9 p. 9

[23] J. Lofberg, “YALMIP: A toolbox for modeling and optimization in MATLAB,” in 2004 IEEE international conference on robotics and automation (IEEE Cat. No. 04CH37508). IEEE, 2004, pp. 284–289.

[24] J. A. Andersson, J. Gillis, G. Horn, J. B. Rawlings, and M. Diehl, “CasADi: a software framework for nonlinear optimization and optimal control,” Mathematical Programming Computation, vol. 11, no. 1, pp. 1–36, 2019.

[25] A. V. Oppenheim, A. S. Willsky, and S. H. Nawab, Signals & Systems.

[27] R. D. Zimmerman and C. E. Murillo-Sanchez, “Matpower 6.0 user’s ´ manual,” Power Systems Engineering Research Center, vol. 9, pp. 65–66, 2016.

[28] J. A. Andersson, J. Gillis, G. Horn, J. B. Rawlings, and M. Diehl, “CasADi: a software framework for nonlinear optimization and optimal control,” Mathematical Programming Computation, vol. 11, no. 1, pp. 1–36, 2019.

[29] J. A. Andersson, J. Gillis, G. Horn, J. B., Rawlings and M. p. 9 p. 9, Diehl “CasADi: a software framework for nonlinear optimization and optimal control,” Mathematical Programming Computation, vol. p. 9 p. 9 11, no. 1, pp. 1–36.

[30] J. A. Andersson, J. Gillis, G., Horn, J., B., Rawlings and M., Diehl “CasADi: a software framework for nonlinear optimization and optimal control,” Mathematical Programming Computation, vol. 11, no. 1, pp. 1–36.

[35] min X i∈G f(p g i) (35a) s. p. 4t. X k∈Gi p g k − p d i = X j∈N ViVj Yij rˆcos(δij), (35b) X k∈Gi q g k − q d i = X j∈N ViVj Yij rˆsin(δij), (35c) p g i ≤ p g i ≤ p g i, qg i ≤ qg i ≤ qg i, V i ≤ Vi ≤ V i, − π ≤ θ i ≤ π, θref = 0, ∆θij ≤ θ i − θj ≤ ∆ij, ∀(i, j)∈L (35d) P squared ij (V, θ; APF) + Q squared ij (V, θ; APF) ≤ S squared ij (35h) P squared ji(V, θ; APF) + Q squared ji(V, θ; APF) ≤ S squared ji (35i). p. 4

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Numerical Stability for Interior-Point Methods: The APF OPF formulation eliminates trigonometric evaluations and replaces them with smooth surrogates for the cosine and sine functions, which are empirically produce larger and more stable Newton steps under standard interior-point solvers. This improves the performance of IPMs by yielding smoother and better-conditioned Jacobian and Hessian matrices compared to the classical model.

  2. Computational Efficiency: The replacement of trigonometric evaluations with simple arithmetic operations involving addition, multiplication, and division allows for efficient computation without calling expensive trigonometric routines. This reduction in per-iteration computational cost contributes to faster solver runtimes, leading to speedups ranging from 11% to over 70% on small networks.

  3. Preserving Physical Fidelity: The all-pass kernel maintains the unit magnitude property where r(delta) = 1 for every real delta, ensuring that only the network resistances contribute to real power losses. This guarantees that the approximation does not introduce artificial attenuation or amplification of power flow terms, preserving the physical interpretation of resistive dissipation.

  4. Angle Accuracy Improvement: A pre-rotation step, using a DC power flow solution, centers the fractional approximation around typical angle values encountered during optimization. This shift ensures accuracy by transforming trigonometric terms into approximations that are most accurate near zero for the shifted argument, improving alignment with actual operating angles.

  5. Robustness Across Network Scales: The APF OPF formulation maintains negligible optimality gaps and reproduces the voltage and angle profiles of the classical model across networks up to 10,480 buses. This confirms that the approximation is effective in maintaining solution quality even in large-scale, heavily loaded systems where classical solvers might struggle due to ill-conditioned derivatives.

Related papers