Graph-Based Modeling and Decomposition of Energy Infrastructures

Carleton Coffrin; Kaarthik Sundar; Sungho Shin; Victor M. Zavala

arxiv: 2010.02404 · v3 · submitted 2020-10-06 · 🧮 math.OC

Graph-Based Modeling and Decomposition of Energy Infrastructures

Sungho Shin , Carleton Coffrin , Kaarthik Sundar , Victor M. Zavala This is my paper

Pith reviewed 2026-05-24 14:44 UTC · model grok-4.3

classification 🧮 math.OC

keywords graph-based modelingnonlinear optimizationenergy infrastructuresdecomposition methodsinterior-point algorithmsparallel computinggas networksoptimal power flow

0 comments

The pith

Modeling energy infrastructures as graphs and decomposing via restricted additive Schwarz inside interior-point methods cuts solution times by over 300%.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proposes representing nonlinear optimization problems in energy infrastructures as graph-structured problems. This structure enables parallelization of function and derivative computations at the modeling level and parallel linear algebra operations at the solver level through a restricted additive Schwarz scheme. The scheme is embedded in an interior-point algorithm and implemented in the MadNLP.jl solver interfaced with Plasmo.jl. Tests on transient gas network optimization and multi-period AC optimal power flow show solution times dropping from 72.36 to 23.84 seconds and from 515.81 to 149.45 seconds. A reader would care because these problems underpin real-time operations of critical infrastructures.

Core claim

The paper claims that energy infrastructure problems admit graph-structured representations that can be decomposed with a restricted additive Schwarz scheme inside an interior-point method, enabling parallel computations that accelerate solutions by over 300 percent relative to off-the-shelf tools.

What carries the argument

The restricted additive Schwarz scheme for flexible decomposition of graph structures within an interior-point algorithm.

If this is right

Function and derivative evaluations can be parallelized using the graph structure at the modeling level.
Linear algebra operations can be parallelized using the decomposition scheme at the solver level.
Solution times for transient gas network optimization drop from 72.36 seconds to 23.84 seconds.
Solution times for multi-period AC optimal power flow drop from 515.81 seconds to 149.45 seconds.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same graph modeling and decomposition approach could apply to other dynamic infrastructure systems such as water networks.
The speedups might make previously intractable large-scale instances solvable in operational time frames.
Embedding the solver in online control loops could support real-time decision making for infrastructure operators.

Load-bearing premise

Energy infrastructure problems admit accurate graph-structured representations whose physical models remain faithful after decomposition.

What would settle it

If the solutions obtained from the decomposed graph-based approach deviate in objective value or feasibility from the full non-decomposed solutions on the gas network or power flow test cases.

Figures

Figures reproduced from arXiv: 2010.02404 by Carleton Coffrin, Kaarthik Sundar, Sungho Shin, Victor M. Zavala.

**Figure 2.** Figure 2: Graph-based decomposition in MadNLP.jl applied to a transient gas network problem. The RAS scheme (6) involves the following steps. We first obtain the residual at the current step `; then, for each overlapping subdomains {V ω k } K k=1, the associated residual is extracted as R> k r (`) . The k-th subsystem is then solved by applying M−1 k . In MadNLP.jl, a factorization of Mk is computed with a direct so… view at source ↗

**Figure 3.** Figure 3: The problem is modeled as an OptiGraph object (the core modeling object of Plasmo.jl). The OptiGraph [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

**Figure 3.** Figure 3: Schematics of graph-based modeling and solution (top) and conventional modeling and solution (bottom). [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

**Figure 4.** Figure 4: Solution time (top), linear solver time (middle), function evaluation time (bottom) for transient gas network [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

read the original abstract

Nonlinear optimization problems are found at the heart of real-time operations of critical infrastructures. These problems are computationally challenging because they embed complex physical models that exhibit space-time dynamics. We propose modeling these problems as graph-structured optimization problems, and illustrate how their structure can be exploited at the modeling level (for parallelizing function/derivative computations) and at the solver level (for parallelizing linear algebra operations). Specifically, we present a restricted additive Schwarz scheme that enables flexible decomposition of complex graph structures within an interior-point algorithm. The proposed approach is implemented as a general-purpose nonlinear programming solver that we call MadNLP.jl; this Julia-based solver is interfaced to the graph-based modeling package Plasmo.jl. The efficiency of this framework is demonstrated via problems arising in transient gas network optimization and multi-period AC optimal power flow. We show that our framework accelerates the solution (compared to off-the-shelf tools) by over 300%; specifically, solution times are reduced from 72.36 sec to 23.84 sec for the gas problem and from 515.81 sec to 149.45 sec for the power flow problem.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper gives a working restricted additive Schwarz scheme inside an interior-point solver for graph energy problems and reports 3x speedups on two standard test cases.

read the letter

The paper's main point is that a restricted additive Schwarz decomposition can be embedded in an interior-point method to speed up solves on graph-structured nonlinear programs from energy networks. The implementation in MadNLP.jl and Plasmo.jl handles both parallel derivative computations and linear algebra, and the timings on transient gas optimization and multi-period AC power flow show reductions from 72s to 24s and 516s to 149s respectively. That is the concrete result the authors deliver. The approach is new in its specific tailoring of the Schwarz scheme to the IPM setting for these graph models rather than a routine extension of prior decomposition work. The modeling step in Plasmo.jl is a natural fit for infrastructure networks, and the Julia packages make the method usable without starting from scratch. The numerical evidence is grounded in standard benchmark problems and measured against off-the-shelf tools. The central assumption is that the graph decomposition preserves the original problem's solution quality. The reported numbers focus on wall-clock time without accompanying comparisons of objective values, primal-dual infeasibilities, or KKT residuals between the monolithic and decomposed runs. For nonlinear models this matters because interface coupling can shift the effective system; if the paper includes those checks internally then the speedup claim holds cleanly, but the abstract alone leaves that detail open. This work is aimed at researchers who already work on large-scale energy optimization and need faster real-time solves while keeping the physical models intact. It shows honest engagement with the structure of the problems and ships actual code, so it deserves a serious referee even if the equivalence checks need to be made explicit in revision.

Referee Report

1 major / 2 minor

Summary. The paper claims that nonlinear optimization problems for energy infrastructures can be modeled as graph-structured NLPs and solved more efficiently by exploiting this structure both at the modeling level (parallel function/derivative evaluation via Plasmo.jl) and at the solver level (restricted additive Schwarz decomposition inside an interior-point method in MadNLP.jl). It reports concrete speedups on two benchmark problems: transient gas network optimization (72.36 s reduced to 23.84 s) and multi-period AC optimal power flow (515.81 s reduced to 149.45 s), for an overall acceleration exceeding 300%.

Significance. If the decomposition preserves solution quality, the framework supplies a practical, general-purpose Julia implementation that accelerates large-scale energy NLPs by combining graph modeling with domain-decomposition linear algebra inside IPOPT-style solvers. The concrete implementation in MadNLP.jl/Plasmo.jl together with wall-clock timings on standard energy problems constitutes a reproducible strength.

major comments (1)

[Numerical results] Numerical results section: the reported wall-clock reductions are presented without accompanying data on optimal objective values, primal/dual infeasibilities, or KKT residuals for the monolithic versus decomposed runs. Because the restricted additive Schwarz scheme modifies the effective KKT system at the interfaces, equivalence of the computed solutions must be verified to establish that the observed speedups reflect faithful acceleration rather than solution of an altered problem.

minor comments (2)

[Section 4] The description of the restricted additive Schwarz scheme would benefit from an explicit statement of the overlap parameter and its effect on the convergence tolerance of the inner linear solves.
[Figures 5-6] Figure captions for the timing plots should include the number of MPI processes or threads used in each configuration.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive comment on verifying solution equivalence. We will revise the manuscript to address this point directly.

read point-by-point responses

Referee: [Numerical results] Numerical results section: the reported wall-clock reductions are presented without accompanying data on optimal objective values, primal/dual infeasibilities, or KKT residuals for the monolithic versus decomposed runs. Because the restricted additive Schwarz scheme modifies the effective KKT system at the interfaces, equivalence of the computed solutions must be verified to establish that the observed speedups reflect faithful acceleration rather than solution of an altered problem.

Authors: We agree that this verification is necessary to confirm that the observed speedups do not come at the expense of solution quality. In the revised manuscript we will add a dedicated subsection (or table) in the Numerical Results section that reports, for both the transient gas network and multi-period AC OPF problems: (i) the optimal objective values, (ii) primal and dual infeasibility norms, and (iii) the final KKT residual for the monolithic solve versus the decomposed RAS solve. These quantities will be computed with the same convergence tolerances so that any differences attributable to the interface treatment of the restricted additive Schwarz scheme can be quantified. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical speedups are independent measurements.

full rationale

The paper presents a graph-based modeling approach and restricted additive Schwarz decomposition inside an interior-point solver, with performance shown via direct wall-clock timings on gas and power-flow instances versus off-the-shelf tools. No derivation chain reduces a claimed result to a fitted parameter, self-definition, or load-bearing self-citation. The reported accelerations (72.36 s → 23.84 s; 515.81 s → 149.45 s) are external benchmark measurements, not constructed quantities. Standard solver theory and graph partitioning supply the independent foundation.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The work relies on standard interior-point convergence theory and domain assumptions about graph representability of physical networks; no free parameters or invented entities are introduced.

axioms (2)

standard math Interior-point methods converge on the decomposed subproblems
Invoked when embedding the Schwarz scheme inside the IP algorithm.
domain assumption Graph models preserve the essential dynamics of gas and power networks
Central to the modeling step for parallelization.

pith-pipeline@v0.9.0 · 5733 in / 1045 out tokens · 29616 ms · 2026-05-24T14:44:53.503834+00:00 · methodology

discussion (0)

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

Constrained Policy Optimization for Stochastic Optimal Control under Nonstationary Uncertainties
math.OC 2022-09 unverdicted novelty 6.0

Constrained policy optimization for stochastic optimal control under nonstationary uncertainties via Markov embeddability and finite approximation.
A Feasible Reduced Space Method for Real-Time Optimal Power Flow
math.OC 2021-10 unverdicted novelty 5.0

A novel feasible-path method solves optimal power flow in reduced space by directly enforcing power flow equations and softly penalizing operational constraints via Augmented Lagrangian, with GPU acceleration for the ...
On the Convergence of Overlapping Schwarz Decomposition for Nonlinear Optimal Control
math.OC 2020-05 unverdicted novelty 5.0

Overlapping Schwarz decomposition for nonlinear OCPs achieves local linear convergence with rate improving exponentially with overlap size, based on exponential decay of sensitivity for primal and dual solutions.

Reference graph

Works this paper leans on

23 extracted references · 23 canonical work pages · cited by 3 Pith papers · 1 internal anchor

[1]

Balay, S., Abhyankar, S., Adams, M., Brown, J., Brune, P., Buschelman, K., Dalcin, L., Dener, A., Eijkhout, V., Gropp, W., et al. (2019). P E T S c users manual

work page 2019
[2]

Bent, R., Sundar, K., and Fobes, D. (2020). G as M odels.jl. https://github.com/lanl-ansi/GasModels.jl

work page 2020
[3]

Boyd, S., Parikh, N., and Chu, E. (2011). Distributed optimization and statistical learning via the alternating direction method of multipliers. Now Publishers Inc

work page 2011
[4]

and Saad, Y

Cai, X.C. and Saad, Y. (1996). Overlapping domain decomposition algorithms for general sparse matrices. Numerical linear algebra with applications, 3(3), 221--237

work page 1996
[5]

and Sarkis, M

Cai, X.C. and Sarkis, M. (1999). A restricted additive schwarz preconditioner for general sparse linear systems. SIAM Journal on Scientific Computing, 21(2), 792--797

work page 1999
[6]

and Zavala, V.M

Chiang, N.Y. and Zavala, V.M. (2016). An inertia-free filter line-search algorithm for large-scale nonlinear programming. Computational Optimization and Applications, 64(2), 327--354

work page 2016
[7]

Chiang, N., Petra, C.G., and Zavala, V.M. (2014). Structured nonconvex optimization of large-scale energy systems using P I P S - N L P . In 2014 Power Systems Computation Conference, 1--7. IEEE

work page 2014
[8]

Coffrin, C., Bent, R., Sundar, K., Ng, Y., and Lubin, M. (2018). P ower M odels.jl: An open-source framework for exploring power flow formulations. In 2018 Power Systems Computation Conference (PSCC), 1--8. doi:10.23919/PSCC.2018.8442948

work page doi:10.23919/pscc.2018.8442948 2018
[9]

Curtis, F.E., Huber, J., Schenk, O., and W \"a chter, A. (2012). A note on the implementation of an interior-point algorithm for nonlinear optimization with inexact step computations. Mathematical programming, 136(1), 209--227

work page 2012
[10]

Dunning, I., Huchette, J., and Lubin, M. (2017). J u M P : A modeling language for mathematical optimization. SIAM Review, 59(2), 295--320

work page 2017
[11]

Gerstner, P., Schick, M., Heuveline, V., Meyer-H \"u bner, N., Suriyah, M., Leibfried, T., Slednev, V., Fichtner, W., and Bertsch, V.V. (2016). A domain decomposition approach for solving dynamic optimal power flow problems in parallel with application to the german transmission grid. Preprint Series of the Engineering Mathematics and Computing Lab, (1)

work page 2016
[12]

Geth, F., Coffrin, C., and Fobes, D.M. (2020). A flexible storage model for power network optimization. arXiv preprint arXiv:2004.14768

work page arXiv 2020
[13]

HSL, A. (2007). collection of fortran codes for large-scale scientific computation. See http://www. hsl. rl. ac. uk

work page 2007
[14]

Jalving, J., Cao, Y., and Zavala, V.M. (2019). Graph-based modeling and simulation of complex systems. Computers & Chemical Engineering, 125, 134--154

work page 2019
[15]

Jalving, J., Shin, S., and Zavala, V.M. (2020). A graph-based modeling abstraction for optimization: Concepts and implementation in plasmo. jl. arXiv preprint arXiv:2006.05378

work page internal anchor Pith review Pith/arXiv arXiv 2020
[16]

and Anitescu, M

Kim, Y. and Anitescu, M. (2020). A real-time optimization with warm-start of multiperiod ac optimal power flows. Electric Power Systems Research, 189, 106721

work page 2020
[17]

Rodriguez, J.S., Laird, C.D., and Zavala, V.M. (2020). Scalable preconditioning of block-structured linear algebra systems using admm. Computers & Chemical Engineering, 133, 106478

work page 2020
[18]

(2020 a )

Shin, S., Anitescu, M., and Zavala, V.M. (2020 a ). Overlapping schwarz decomposition for constrained quadratic programs. arXiv preprint arXiv:2003.07502

work page arXiv 2020
[19]

(2020 b )

Shin, S., Zavala, V.M., and Anitescu, M. (2020 b ). Decentralized schemes with overlap for solving graph-structured optimization problems. IEEE Transactions on Control of Network Systems

work page 2020
[20]

and Zlotnik, A

Sundar, K. and Zlotnik, A. (2018). State and parameter estimation for natural gas pipeline networks using transient state data. IEEE Transactions on Control Systems Technology, 27(5), 2110--2124

work page 2018
[21]

and Biegler, L.T

W \"a chter, A. and Biegler, L.T. (2006). On the implementation of an interior-point filter line-search algorithm for large-scale nonlinear programming. Mathematical programming, 106(1), 25--57

work page 2006
[22]

, " * write output.state after.block = add.period write newline

ENTRY address author booktitle chapter doi edition editor eid howpublished institution journal key month note number organization pages publisher school series title type url volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.all := #1 'mid.sent...

work page
[23]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

work page

[1] [1]

Balay, S., Abhyankar, S., Adams, M., Brown, J., Brune, P., Buschelman, K., Dalcin, L., Dener, A., Eijkhout, V., Gropp, W., et al. (2019). P E T S c users manual

work page 2019

[2] [2]

Bent, R., Sundar, K., and Fobes, D. (2020). G as M odels.jl. https://github.com/lanl-ansi/GasModels.jl

work page 2020

[3] [3]

Boyd, S., Parikh, N., and Chu, E. (2011). Distributed optimization and statistical learning via the alternating direction method of multipliers. Now Publishers Inc

work page 2011

[4] [4]

and Saad, Y

Cai, X.C. and Saad, Y. (1996). Overlapping domain decomposition algorithms for general sparse matrices. Numerical linear algebra with applications, 3(3), 221--237

work page 1996

[5] [5]

and Sarkis, M

Cai, X.C. and Sarkis, M. (1999). A restricted additive schwarz preconditioner for general sparse linear systems. SIAM Journal on Scientific Computing, 21(2), 792--797

work page 1999

[6] [6]

and Zavala, V.M

Chiang, N.Y. and Zavala, V.M. (2016). An inertia-free filter line-search algorithm for large-scale nonlinear programming. Computational Optimization and Applications, 64(2), 327--354

work page 2016

[7] [7]

Chiang, N., Petra, C.G., and Zavala, V.M. (2014). Structured nonconvex optimization of large-scale energy systems using P I P S - N L P . In 2014 Power Systems Computation Conference, 1--7. IEEE

work page 2014

[8] [8]

Coffrin, C., Bent, R., Sundar, K., Ng, Y., and Lubin, M. (2018). P ower M odels.jl: An open-source framework for exploring power flow formulations. In 2018 Power Systems Computation Conference (PSCC), 1--8. doi:10.23919/PSCC.2018.8442948

work page doi:10.23919/pscc.2018.8442948 2018

[9] [9]

Curtis, F.E., Huber, J., Schenk, O., and W \"a chter, A. (2012). A note on the implementation of an interior-point algorithm for nonlinear optimization with inexact step computations. Mathematical programming, 136(1), 209--227

work page 2012

[10] [10]

Dunning, I., Huchette, J., and Lubin, M. (2017). J u M P : A modeling language for mathematical optimization. SIAM Review, 59(2), 295--320

work page 2017

[11] [11]

Gerstner, P., Schick, M., Heuveline, V., Meyer-H \"u bner, N., Suriyah, M., Leibfried, T., Slednev, V., Fichtner, W., and Bertsch, V.V. (2016). A domain decomposition approach for solving dynamic optimal power flow problems in parallel with application to the german transmission grid. Preprint Series of the Engineering Mathematics and Computing Lab, (1)

work page 2016

[12] [12]

Geth, F., Coffrin, C., and Fobes, D.M. (2020). A flexible storage model for power network optimization. arXiv preprint arXiv:2004.14768

work page arXiv 2020

[13] [13]

HSL, A. (2007). collection of fortran codes for large-scale scientific computation. See http://www. hsl. rl. ac. uk

work page 2007

[14] [14]

Jalving, J., Cao, Y., and Zavala, V.M. (2019). Graph-based modeling and simulation of complex systems. Computers & Chemical Engineering, 125, 134--154

work page 2019

[15] [15]

Jalving, J., Shin, S., and Zavala, V.M. (2020). A graph-based modeling abstraction for optimization: Concepts and implementation in plasmo. jl. arXiv preprint arXiv:2006.05378

work page internal anchor Pith review Pith/arXiv arXiv 2020

[16] [16]

and Anitescu, M

Kim, Y. and Anitescu, M. (2020). A real-time optimization with warm-start of multiperiod ac optimal power flows. Electric Power Systems Research, 189, 106721

work page 2020

[17] [17]

Rodriguez, J.S., Laird, C.D., and Zavala, V.M. (2020). Scalable preconditioning of block-structured linear algebra systems using admm. Computers & Chemical Engineering, 133, 106478

work page 2020

[18] [18]

(2020 a )

Shin, S., Anitescu, M., and Zavala, V.M. (2020 a ). Overlapping schwarz decomposition for constrained quadratic programs. arXiv preprint arXiv:2003.07502

work page arXiv 2020

[19] [19]

(2020 b )

Shin, S., Zavala, V.M., and Anitescu, M. (2020 b ). Decentralized schemes with overlap for solving graph-structured optimization problems. IEEE Transactions on Control of Network Systems

work page 2020

[20] [20]

and Zlotnik, A

Sundar, K. and Zlotnik, A. (2018). State and parameter estimation for natural gas pipeline networks using transient state data. IEEE Transactions on Control Systems Technology, 27(5), 2110--2124

work page 2018

[21] [21]

and Biegler, L.T

W \"a chter, A. and Biegler, L.T. (2006). On the implementation of an interior-point filter line-search algorithm for large-scale nonlinear programming. Mathematical programming, 106(1), 25--57

work page 2006

[22] [22]

, " * write output.state after.block = add.period write newline

ENTRY address author booktitle chapter doi edition editor eid howpublished institution journal key month note number organization pages publisher school series title type url volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.all := #1 'mid.sent...

work page

[23] [23]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

work page