pith. sign in

arxiv: 2601.14414 · v2 · pith:WEU4DSREnew · submitted 2026-01-20 · 🧮 math.OC · cs.SY· eess.SY

πMPC: A Parallel-in-horizon and Construction-free NMPC Solver

Liang Wu , Bo Yang , Junheng Li , Xu Yang , Yilin Mo , Yang Shi , Aaron D. Ames , J\'an Drgo\v{n}a This is my paper

Pith reviewed 2026-05-22 12:15 UTC · model grok-4.3

classification 🧮 math.OC cs.SYeess.SY
keywords nonlinear MPCADMMparallel optimizationvariable splittingconstruction-freevelocity-based representationembedded control
0
0 comments X

The pith

New ADMM scheme solves nonlinear MPC in parallel across the horizon without QP construction.

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

The authors present πMPC as a way to solve nonlinear model predictive control using ADMM. They introduce a variable-splitting scheme and a velocity-based system representation that together permit independent computation for each time step in the prediction horizon. This parallel-in-horizon feature addresses the sequential bottleneck in traditional Riccati-based approaches. The method works directly with the system matrices, skipping the usual step of formulating an explicit quadratic program. Such an approach could make real-time nonlinear control feasible for longer horizons on resource-limited hardware.

Core claim

The πMPC algorithm combines a new variable-splitting scheme with a velocity-based system representation in the ADMM framework, enabling horizon-wise parallel execution while operating directly on system matrices without explicit MPC-to-QP construction.

What carries the argument

The combination of a new variable-splitting scheme and velocity-based system representation within the ADMM framework for parallel horizon execution.

If this is right

  • Horizon-wise parallel execution allows simultaneous updates for all time steps in the prediction horizon.
  • Direct use of system matrices eliminates the MPC-to-QP construction step.
  • Convergence guarantees of ADMM are preserved under the new scheme for nonlinear dynamics.
  • Implementation remains simple with straightforward parameter selection.

Where Pith is reading between the lines

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

  • The method could extend to other first-order optimization algorithms used in control.
  • Hardware implementations on multi-core processors might achieve significant runtime reductions for long-horizon problems.
  • Similar techniques may apply to distributed MPC where multiple agents solve in parallel.

Load-bearing premise

The new variable-splitting scheme and velocity-based representation preserve the convergence guarantees and numerical stability of standard ADMM for nonlinear dynamics over long horizons.

What would settle it

Numerical experiments showing that the parallel updates fail to converge or produce suboptimal solutions compared to sequential ADMM on a standard nonlinear benchmark problem.

Figures

Figures reproduced from arXiv: 2601.14414 by Aaron D. Ames, Bo Yang, J\'an Drgo\v{n}a, Junheng Li, Liang Wu, Xu Yang, Yang Shi, Yilin Mo.

Figure 1
Figure 1. Figure 1: AFTI-16 closed-loop trajectory tracking performance [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Convergence iterations to 10−6 residual at each MPC step for AFTI￾16 (capped at 10,000) with |xi | ≤ 5 for states and |ui | ≤ 0.1 for inputs, with state weighting Wy = In, input weighting Wu = Im, and W∆u = 0. Table I and [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Per-iteration computation time as a function of (a) prediction horizon, [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: CSTR trajectory tracking with inlet temperature disturbance [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
read the original abstract

The alternating direction method of multipliers (ADMM) has gained increasing popularity in embedded model predictive control (MPC) due to its code simplicity and pain-free parameter selection. However, existing ADMM solvers either target general quadratic programming (QP) problems or exploit sparse MPC formulations via Riccati recursions, which are inherently sequential and therefore difficult to parallelize for long prediction horizons. This technical note proposes a novel \textit{parallel-in-horizon} and \textit{construction-free} nonlinear MPC algorithm, termed $\pi$MPC, which combines a new variable-splitting scheme with a velocity-based system representation in the ADMM framework, enabling horizon-wise parallel execution while operating directly on system matrices without explicit MPC-to-QP construction. Numerical experiments and accompanying code are provided to validate the effectiveness of the proposed method.

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.

Referee Report

2 major / 2 minor

Summary. The paper proposes πMPC, a novel ADMM-based solver for nonlinear model predictive control that achieves parallel-in-horizon execution via a new variable-splitting scheme combined with a velocity-based system representation. This enables direct operation on the original system matrices without explicit construction of a quadratic program, addressing the sequential limitations of Riccati-based approaches for long horizons. Effectiveness is supported by numerical experiments and accompanying code.

Significance. If the central claims hold, the method would provide a practical advance for embedded NMPC by enabling parallelism over long horizons while retaining ADMM's code simplicity and parameter ease. The construction-free aspect and open code are clear strengths that support reproducibility and adoption.

major comments (2)
  1. [§3] §3 (Algorithm Description): The new variable-splitting scheme and velocity-based representation are asserted to yield iterates equivalent to the original NMPC problem under ADMM, but no convergence analysis or sufficient conditions (e.g., local convexity of the dynamics, bounded Hessians, or growth rules for the penalty parameter) are provided. Standard ADMM theory does not apply directly to nonconvex NMPC, leaving open whether the parallel updates solve the nonlinear dynamics and cost to within the claimed tolerance.
  2. [§4] §4 (Numerical Experiments): The reported results demonstrate runtime and accuracy benefits, but without theoretical backing it is unclear whether the observed performance generalizes or whether the velocity-based reformulation introduces systematic bias in constraint satisfaction for longer horizons or different nonlinearities.
minor comments (2)
  1. [Abstract] The abstract and introduction could more explicitly list the standing assumptions on the nonlinear dynamics (e.g., Lipschitz continuity or twice differentiability) that are implicitly used.
  2. [Figure 2] Figure 2 (or equivalent diagram of the splitting) would benefit from explicit annotation of which blocks execute in parallel versus sequentially.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We are grateful to the referee for the detailed and constructive review of our manuscript. We address each major comment below and indicate the revisions made to the manuscript.

read point-by-point responses

  1. Referee: [§3] §3 (Algorithm Description): The new variable-splitting scheme and velocity-based representation are asserted to yield iterates equivalent to the original NMPC problem under ADMM, but no convergence analysis or sufficient conditions (e.g., local convexity of the dynamics, bounded Hessians, or growth rules for the penalty parameter) are provided. Standard ADMM theory does not apply directly to nonconvex NMPC, leaving open whether the parallel updates solve the nonlinear dynamics and cost to within the claimed tolerance.

    Authors: We agree that a rigorous convergence analysis for the nonconvex case is an important consideration. The variable splitting is constructed so that the consensus constraints enforce the nonlinear dynamics exactly at convergence, and the ADMM updates are derived to minimize the augmented Lagrangian corresponding to the original problem. However, as this is a technical note focused on the algorithmic development and parallel implementation, a full theoretical analysis is left for future work. We have added a new paragraph in Section 3 discussing the challenges of nonconvex ADMM and referencing relevant literature on convergence of ADMM for nonconvex problems. Additionally, we have included a note on the practical convergence observed in our experiments and the choice of penalty parameter. revision: partial

  2. Referee: [§4] §4 (Numerical Experiments): The reported results demonstrate runtime and accuracy benefits, but without theoretical backing it is unclear whether the observed performance generalizes or whether the velocity-based reformulation introduces systematic bias in constraint satisfaction for longer horizons or different nonlinearities.

    Authors: The numerical experiments include several nonlinear systems with varying prediction horizons up to 100 steps, and we report both runtime and solution accuracy metrics, including constraint violations. The velocity-based representation is mathematically equivalent to the standard position-based formulation, as shown in the derivation in Section 2, so it does not introduce bias; it only enables the parallel structure by allowing velocity updates to be decoupled. To further address generalizability, we have added results for an additional nonlinear system and longer horizons in the revised manuscript, along with more detailed analysis of constraint satisfaction over the horizon. revision: yes

Circularity Check

0 steps flagged

No significant circularity in πMPC derivation chain

full rationale

The paper proposes a novel combination of variable-splitting and velocity-based representation inside the standard ADMM framework to achieve horizon-parallel, construction-free NMPC. The central claims concern algorithmic structure and empirical performance rather than any fitted parameter or self-referential quantity being relabeled as a prediction. No load-bearing step reduces by the paper's own equations to a tautology; the method is presented as an independent extension of existing ADMM theory with explicit modifications whose correctness is supported by numerical experiments rather than by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that the proposed splitting and velocity representation maintain ADMM convergence for nonlinear systems; no free parameters or invented physical entities are described in the abstract.

axioms (1)
  • domain assumption ADMM convergence properties hold under the new variable-splitting scheme for nonlinear MPC
    The method relies on standard ADMM theory extending to the novel splitting without additional proof details visible in the abstract.

pith-pipeline@v0.9.0 · 5696 in / 1245 out tokens · 41877 ms · 2026-05-22T12:15:24.191625+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

25 extracted references · 25 canonical work pages

  1. [1]

    Fast Model Predictive Control Using On- line Optimization,

    Y . Wang and S. P. Boyd, “Fast Model Predictive Control Using On- line Optimization,”IEEE Transactions on Control Systems Technology, vol. 18, pp. 267–278, 2010

    work page 2010

  2. [2]

    A Direct Optimization Algorithm for Input- Constrained MPC,

    L. Wu and R. D. Braatz, “A Direct Optimization Algorithm for Input- Constrained MPC,”IEEE Transactions on Automatic Control, vol. 70, no. 2, pp. 1366–1373, 2025

    work page 2025

  3. [3]

    EIQP: Execution-time-certified and Infeasibility-detecting QP solver,

    L. Wu, W. Xiao, and R. D. Braatz, “EIQP: Execution-time-certified and Infeasibility-detecting QP solver,”IEEE Transactions on Automatic Control, pp. 1–16, 2025, DOI: 10.1109/TAC.2025.3631575

    work page doi:10.1109/tac.2025.3631575 2025

  4. [4]

    A Quadratic Programming Algorithm with O(n3)Time Complexity,

    L. Wu and R. D. Braatz, “A Quadratic Programming Algorithm with O(n3)Time Complexity,”arXiv preprint arXiv:2507.04515, 2025

    work page arXiv 2025

  5. [5]

    qpOASES: A parametric active-set algorithm for quadratic program- ming,

    H. J. Ferreau, C. Kirches, A. Potschka, H. G. Bock, and M. Diehl, “qpOASES: A parametric active-set algorithm for quadratic program- ming,”Mathematical Programming Computation, vol. 6, no. 4, pp. 327– 363, 2014

    work page 2014

  6. [6]

    An online active set strategy to overcome the limitations of explicit MPC,

    H. J. Ferreau, H. G. Bock, and M. Diehl, “An online active set strategy to overcome the limitations of explicit MPC,”International Journal of Robust and Nonlinear Control: IFAC-Affiliated Journal, vol. 18, no. 8, pp. 816–830, 2008

    work page 2008

  7. [7]

    A quadratic programming algorithm based on nonneg- ative least squares with applications to embedded model predictive control,

    A. Bemporad, “A quadratic programming algorithm based on nonneg- ative least squares with applications to embedded model predictive control,”IEEE Transactions on Automatic Control, vol. 61, no. 4, pp. 1111–1116, 2015

    work page 2015

  8. [8]

    An accelerated dual gradient-projection algorithm for embedded linear model predictive control,

    P. Patrinos and A. Bemporad, “An accelerated dual gradient-projection algorithm for embedded linear model predictive control,”IEEE Trans- actions on Automatic Control, vol. 59, no. 1, pp. 18–33, 2013

    work page 2013

  9. [9]

    Distributed optimization and statistical learning via the alternating direction method of multipliers,

    S. Boyd, N. Parikh, E. Chu, B. Peleato, J. Ecksteinet al., “Distributed optimization and statistical learning via the alternating direction method of multipliers,”Foundations and Trends® in Machine learning, vol. 3, no. 1, pp. 1–122, 2011

    work page 2011

  10. [10]

    OSQP: An Operator Splitting solver for Quadratic Programs,

    B. Stellato, G. Banjac, P. Goulart, A. Bemporad, and S. Boyd, “OSQP: An Operator Splitting solver for Quadratic Programs,”Mathematical Programming Computation, vol. 12, pp. 637–672, 2020

    work page 2020

  11. [11]

    QPALM: a Newton-type proximal augmented Lagrangian method for quadratic programs,

    B. Hermans, A. Themelis, and P. Patrinos, “QPALM: a Newton-type proximal augmented Lagrangian method for quadratic programs,” in 2019 IEEE 58th Conference on Decision and Control (CDC). IEEE, 2019, pp. 4325–4330

    work page 2019

  12. [12]

    A Simple and Fast Coordinate-Descent Augmented-Lagrangian Solver for Model Predictive control,

    L. Wu and A. Bemporad, “A Simple and Fast Coordinate-Descent Augmented-Lagrangian Solver for Model Predictive control,”IEEE Transactions on Automatic Control, vol. 68, no. 11, pp. 6860–6866, 2023

    work page 2023

  13. [13]

    A construction-free coordinate-descent augmented-Lagrangian method for embedded linear MPC based on ARX models,

    ——, “A construction-free coordinate-descent augmented-Lagrangian method for embedded linear MPC based on ARX models,”IFAC- PapersOnLine, vol. 56, no. 2, pp. 9423–9428, 2023

    work page 2023

  14. [14]

    From linear to nonlinear MPC: Bridging the gap via the real-time iteration,

    S. Gros, M. Zanon, R. Quirynen, A. Bemporad, and M. Diehl, “From linear to nonlinear MPC: Bridging the gap via the real-time iteration,” International Journal of Control, vol. 93, no. 1, pp. 62–80, 2020

    work page 2020

  15. [15]

    Distributed nonlinear model predictive control through accelerated parallel ADMM,

    W. Tang and P. Daoutidis, “Distributed nonlinear model predictive control through accelerated parallel ADMM,” in2019 American Control Conference (ACC). IEEE, 2019, pp. 1406–1411

    work page 2019

  16. [16]

    Distributed model predictive control of linear discrete-time systems with local and global constraints,

    Z. Wang and C. J. Ong, “Distributed model predictive control of linear discrete-time systems with local and global constraints,”Automatica, vol. 81, pp. 184–195, 2017

    work page 2017

  17. [17]

    ADMM for exploiting structure in MPC problems,

    F. Rey, P. Hokayem, and J. Lygeros, “ADMM for exploiting structure in MPC problems,”IEEE Transactions on Automatic Control, vol. 66, no. 5, pp. 2076–2086, 2020

    work page 2076

  18. [18]

    A condensed and sparse QP formulation for predictive control,

    J. L. Jerez, E. C. Kerrigan, and G. A. Constantinides, “A condensed and sparse QP formulation for predictive control,” in2011 50th IEEE Conference on Decision and Control and European Control Conference. IEEE, 2011, pp. 5217–5222

    work page 2011

  19. [19]

    An ADMM algorithm for solvingℓ 1-regularized MPC,

    M. Annergren, A. Hansson, and B. Wahlberg, “An ADMM algorithm for solvingℓ 1-regularized MPC,” in2012 IEEE 51st IEEE Conference on Decision and Control (CDC). IEEE, 2012, pp. 4486–4491

    work page 2012

  20. [20]

    Input- constrained model predictive control via the alternating direction method of multipliers,

    L. E. Sokoler, G. Frison, M. S. Andersen, and J. B. Jørgensen, “Input- constrained model predictive control via the alternating direction method of multipliers,” in2014 European Control Conference (ECC). IEEE, 2014, pp. 115–120

    work page 2014

  21. [21]

    S. P. Boyd and L. Vandenberghe,Convex optimization. Cambridge University Press, 2004

    work page 2004

  22. [22]

    Proximal algorithms,

    N. Parikh and S. Boyd, “Proximal algorithms,”Foundations and Trends in optimization, vol. 1, no. 3, pp. 127–239, 2014

    work page 2014

  23. [23]

    On non-ergodic convergence rate of douglas– rachford alternating direction method of multipliers,

    B. He and X. Yuan, “On non-ergodic convergence rate of douglas– rachford alternating direction method of multipliers,”Numerische Math- ematik, vol. 130, no. 3, pp. 567–577, 2015

    work page 2015

  24. [24]

    Fast alternat- ing direction optimization methods,

    T. Goldstein, B. O’Donoghue, S. Setzer, and R. Baraniuk, “Fast alternat- ing direction optimization methods,”SIAM Journal on Imaging Sciences, vol. 7, no. 3, pp. 1588–1623, 2014

    work page 2014

  25. [25]

    Nonlinear control of constrained linear systems via predictive reference management,

    A. Bemporad, A. Casavola, and E. Mosca, “Nonlinear control of constrained linear systems via predictive reference management,”IEEE Transactions on Automatic Control, vol. 42, no. 3, pp. 340–349, 1997

    work page 1997