Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

A Million-Point Fast Trajectory Optimization Solver

T0 review · 3 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper presents a fast solver that solves a nonlinear, nonconvex orbit-transfer trajectory optimization over a million grid points on an ordinary laptop, with O(N log N) Birkhoff matrix-vector products and mesh-independent iteration cou

desk verdict A potentially important solver that, as written, does not demonstrate its headline result: the linear algebra is derived for a simplified fixed-endpoint problem, the actual KKT system is absent, and the printed preconditioner algorithm is inconsistent with its own definition. read the letter →

arxiv 2509.01855 v1 pith:JUJXUAA3 submitted 2025-09-02 math.NA cs.MScs.NAcs.SYeess.SYmath.OC

A. Javeed , D. P. Kouri , D. Ridzal , J. D. Steinman , I. M. Ross This is my paper · ORCID
classification math.NAcs.MScs.NAcs.SYeess.SYmath.OC MSC 49M1565F0865T50
keywords trajectoryoptimizationBirkhoffpseudospectralmethodsmatrix-freelinearalgebrafastFouriertransformpreconditioningoptimalcontrolorbittransfermillion-pointdiscretization
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Trajectory optimization normally cannot reach a million grid points: pseudospectral differentiation matrices are dense and ill-conditioned, and off-the-shelf optimizers require explicit Jacobians. This paper claims to break that barrier for a nontrivial, nonlinear, nonconvex orbit-transfer problem by discretizing with a Birkhoff-Chebyshev method whose matrices are well-conditioned and never formed. The key algorithmic step is computing Birkhoff matrix-vector products in O(N log N) time with a discrete cosine transform, and pairing them with a lower-triangular Birkhoff-like preconditioner built from Clenshaw-Curtis weights. On grids from 2^10 to 2^20+1 points, the reported SQP iteration count is constant at 4 and the number of matrix-free KKT solves is constant at 37, yielding a solution on a small laptop in practical time. If correct, this makes dense, high-accuracy pseudospectral grids a realistic tool for optimal control problems that previously forced low-order knotting methods or infeasible memory use.

What carries the argument

The Birkhoff-Chebyshev matrix Ba maps virtual derivative values to state values at Chebyshev-Gauss-Lobatto nodes. Instead of storing Ba, FastBV evaluates Ba V through a DCT-I in O(N log N), using the Chebyshev aliasing formula TN+1 = TN-1 on the grid to keep the mode count consistent. The companion object is the lower-triangular ~Ba defined by Clenshaw-Curtis weights, with diagonal entries half the weight and subdiagonal entries the full weight, which closely approximates Ba and yields the preconditioner P = I - ~Ba ∂X f; P^-1 is applied in O(N) via forward substitution. Together these give an O(N log N) matrix-free linear solver whose Krylov iteration count stays flat as N grows.

What would settle it

Write down the exact KKT system for the orbit-transfer problem including the free final time and the endpoint constraints, form the preconditioned operator AP^{-1} used by FastLinSol, and count Krylov iterations at N = 1024 and N = 2^20. If the iteration counts grow with N, or if the number of matrix-free KKT solves in the actual run differs from the reported 37, the central mesh-independence claim collapses. A second check: run the same solver on a problem whose dynamics couple adjacent grid points so that ∂X f is not diagonal; if FastPinvY cannot be applied or iterations grow, the method is

Watch

Extended reading notes

Core claim

The central discovery is that the Birkhoff-Chebyshev discretization, long known for its O(1) condition numbers, can be made matrix-free at a scale no pseudospectral trajectory solver has reached. The paper constructs algorithms FastBV, FastAX, FastPinvY, and FastLinSol: FastBV applies the dense Birkhoff matrix Ba to a vector by converting to Chebyshev modal coefficients with a DCT-I, using the Chebyshev aliasing formula to absorb the extra mode, for O(N log N) work. FastAX applies the Newton operator I - Ba ∂X f in the same complexity. The preconditioner replaces Ba by a lower-triangular matrix ~Ba whose entries are Clenshaw-Curtis quadrature weights, so P = I - ~Ba ∂X f can be inverted in O

Load-bearing premise

The fast solver is derived for a simplified fixed-endpoint version of Newton's equation, and the paper then assumes, without showing the KKT system, that the same O(N log N) matrix-free products and the 37-solve mesh-independence carry over to the full free-final-time, endpoint-constrained orbit-transfer problem.

Editorial extensions

If this is right

  • A million-point grid is no longer a memory barrier: instead of storing an ~10^12-entry differentiation matrix, the solver stores only O(N) vectors, so high-resolution pseudospectral solutions fit on laptops.
  • Total solve time is dominated by O(N log N) transforms per Krylov iteration rather than O(N^2) or O(N^3) linear algebra, so the practical cost advantage over direct solvers grows with N.
  • If iteration counts are genuinely mesh-independent, trajectory optimization cost decouples from grid resolution: refining from 10^3 to 10^6 points does not change the number of expensive optimizer iterations.
  • The combination of exponential convergence and O(1) conditioning removes the traditional reason for hp/knotting methods, potentially simplifying adaptive mesh refinement in optimal control.
  • The same matrix-free preconditioning pattern should transfer to other Birkhoff variants and to differential-equation-constrained optimization beyond trajectory problems.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper derives the fast linear solver for a simplified fixed-endpoint Newton system, then applies it to a free-final-time, endpoint-constrained problem; the actual KKT system is never written. A careful reader should verify that the reported 37 solves are for that full KKT system and not just for the simplified equation.
  • Because FastAX and FastPinvY rely on ∂X f being diagonal, the O(N log N) guarantee is tied to dynamics whose Jacobian is pointwise diagonal. Problems with path constraints or nonlocal coupling would require block preconditioning, and the O(1) iteration count may degrade.
  • A natural stress test is a bang-bang control problem with many switches: the near-perfect approximation of Ba by ~Ba relies on smoothness, so hard discontinuities could reveal N-dependent iteration growth that the orbit-transfer example does not.
  • If the mesh-independence survives those tests, the solver becomes an enabling block for control-oriented orienteering and traveling-salesperson problems that motivated the million-point challenge.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper claims a first-of-a-kind fast trajectory optimization solver capable of solving a nontrivial, nonlinear, nonconvex orbit-transfer problem over more than a million CGL grid points on an ordinary laptop. The enabling components are: (i) a Birkhoff-Chebyshev discretization, (ii) an FFT-based O(N log N) algorithm (FastBV) for the Birkhoff matrix-vector product, (iii) a lower-triangular Birkhoff-like preconditioner built from Clenshaw-Curtis weights, and (iv) a matrix-free preconditioned iterative solver (FastLinSol) for a simplified Newton system [I - Ba ∂X f]δX = xa b. The paper also reports mesh-independent SQP/KKT iteration counts (Table 1) for a 4-state, 1-control free-time orbit transfer problem, and displays a solution over N+1 = 1,048,577 points. The central advertised claims are O(N log N) per matrix-vector product, O(1) iteration counts in N, and the first million-point solution of a nonlinear nonconvex trajectory optimization problem.

Significance. If all claims held and were fully substantiated, the contribution would be significant: an O(N log N) matrix-free Birkhoff product with a mesh-independent preconditioner would remove a major bottleneck in pseudospectral optimal control and make million-point solutions practical. The FastBV derivation appears plausible and internally consistent, and the idea of a lower-triangular Birkhoff-like preconditioner is attractive. However, the paper as written does not deliver a verifiable version of those claims for the demonstrated problem: the linear solver is derived only for a scalar, fixed-endpoint system, while the reported KKT-solve counts are for a free-time, endpoint-constrained multi-state problem whose KKT system is never written down. In addition, the printed FastPinvY algorithm does not compute the inverse of the advertised preconditioner. These are load-bearing gaps in the central claim. The numerical demonstration would be impressive if the missing derivations, corrected algorithm, and verification metrics were supplied and ideally accompanied by reproducible code.

major comments (3)
  1. [FastPinvY (Section 'A Preconditioned Algorithm for Solving Eq. (28)')] The printed FastPinvY algorithm is inconsistent with the preconditioner P := I - Btilde^a ∂X f defined in Eq. (38). Since Btilde^a is lower triangular with diagonal entries w_k/2 and subdiagonal entries w_j, forward substitution for P ξ = Υ requires s_k = Σ_{j<k} w_j d_j ξ_j and ξ_k = (1 - w_k d_k/2)^{-1}(Υ_k + s_k), followed by s_{k+1} = s_k + w_k d_k ξ_k. The algorithm as printed uses (Υ_k - s) and updates s with w_k (∂x_k f) Υ_k, not with w_k d_k ξ_k. It therefore does not compute P^{-1}Υ. This makes Eq. (33) and all subsequently reported preconditioned iteration counts unverifiable. Please correct the algorithm and, if possible, provide a small reproducible test confirming that P^{-1} is computed correctly.
  2. [Section 'An Illustrative Nonlinear Nonconvex Astrodynamics Problem' and Table 1] The KKT system for the demonstrated orbit-transfer problem (39) is never stated. Problem (39) has four states, one control, free final time T, and endpoint constraints; its KKT system will contain state/control variation blocks, adjoint variables, endpoint multipliers, and dT coupling. Table 1 reports 'Matrix-free KKT solves' with counts 37 for every N, but the system being solved is not defined, and no algorithm is given for applying the FastAX/FastPinvY/FastLinSol machinery to that system. The scalar diagonal argument for ∂X f in Eq. (27) does not automatically extend to the 4N×4N dynamics Jacobian plus control/endpoint/free-time blocks. The O(1) iteration claim and the million-point result therefore rest on an unspecified computation. The authors should write out the KKT system, specify the SQP iteration, and report per-iteration linear-solve counts for that system.
  3. [Verification paragraph in Section 'An Illustrative Nonlinear Nonconvex Astrodynamics Problem'] The text states that the candidate solution was 'verified and validated using Pontryagin's Principle using the method described in Reference [22]', but no numerical verification results are reported. For a paper whose central claim is a numerically computed solution at unprecedented scale, it is essential to give quantitative verification metrics: e.g., KKT residuals, costate error, or convergence of the solution under mesh refinement. Without such metrics, the reader cannot judge whether the million-point plot is a converged optimum or a numerical artifact.
minor comments (4)
  1. [Table 1] The text says the statistics are shown for '21 different values of N', but the table lists only N = 2^10, ..., 2^20, which is 11 values. Please correct the wording or the table.
  2. [Figure 6 and Section 'CONSTRUCTION OF A LOW-MEMORY FAST MILLION-POINT LINEAR SOLVER'] The run times are normalized but no absolute timings or hardware details are given beyond 'MacBook Air'. Absolute timings would help the reader calibrate the speed claim. Also, the stopping tolerance for GMRES and for the outer SQP iteration is never stated.
  3. [Section 'Development of a Special Birkhoff-Centric Preconditioner'] The approximation (35) is justified by visual inspection of a 10-point matrix (34). This is a heuristic; the paper would be stronger if it reported the spectrum of AP^{-1} for representative ∂X f, or a bound on the condition number of the preconditioned system.
  4. [Throughout] There are minor typographical issues, e.g., 'Optimizaiton' in Reference [15], 'Bikrhoff' in the Conclusion, and 'T able' in Table 1. A careful proofreading pass is needed.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the O(1) iteration counts are measured, the preconditioner is a structural heuristic rather than a fit, and the same-author citations are corroborated by in-paper experiments; the KKT-system gap is a rigor issue, not a circular reduction.

full rationale

The derivation chain is largely self-contained. FastBV is derived in Eqs. (9)-(25) from Chebyshev/DCT identities, and FastAX follows from the diagonal Jacobian in Eq. (27). The preconditioner P = I - Btilde_a ∂X f (Eq. 38) is constructed from Clenshaw-Curtis weights as a structural approximation to Ba, not fitted to the solution; the constant 37 table entries are empirical observations over N, not imposed values. The flat condition-number premise is imported from same-author Ref. [8], but Figure 6 and Table 1 provide in-paper empirical corroboration, so the citation is not the sole load-bearing support. The Pontryagin verification cites same-author Ref. [22], but it is a standard residual-check method, not an assumption of the result. The significant limitation is non-circular: FastAX/FastPinvY/FastLinSol are derived only for the scalar fixed-endpoint Newton system Eq. (28), while Table 1 reports 'Matrix-free KKT solves' for the free-time endpoint-constrained orbit transfer (39), whose KKT system is never written; additionally FastPinvY's printed update (ξ_k = (1 - w_k ∂x_k f/2)^{-1}(Υ_k - s); s ← w_k ∂x_k f Υ_k + s) does not match the forward substitution for P^{-1}. These are completeness/correctness gaps, not reductions of a prediction to its inputs, so under the hard rules they do not raise the circularity score.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the authors' prior Birkhoff theory, a hand-constructed triangular approximation to the Birkhoff matrix, and an unstated extension from a simplified Newton system to the full KKT system of the demonstrated problem. There are no fitted constants that define the physical solution, but the preconditioner weights are chosen heuristically.

free parameters (1)
  • Preconditioner off-diagonal/diagonal weights (w_j and w_j/2) = w_j^CC and w_j^CC/2, where w_j^CC are Clenshaw-Curtis weights
    The lower-triangular approximation Btilde is defined by inspection of one 10x10 Birkhoff matrix (Eqs. 34-36), not derived from an error bound. It is a hand-chosen structural approximation of B^a and is load-bearing for the preconditioner.
assumptions (4)
  • domain assumption Universal Birkhoff theory from Ref. [8] gives O(1) condition numbers and exponential convergence for Birkhoff-Chebyshev discretizations.
    Invoked in the introduction and used to justify the entire approach; the paper does not derive this theory.
  • ad hoc to paper The lower-triangular Btilde built from Clenshaw-Curtis weights is close enough to B^a to make P = I - Btilde dX f an effective preconditioner.
    Eq. (35) is motivated by visual inspection of a 10x10 matrix; no norm bound or theory of closeness is given. The claimed O(1) Krylov iterations rest on this heuristic.
  • domain assumption The linear solver derived for the fixed-endpoint Newton system (28) extends to the full KKT system of the free-time orbit-transfer problem (39).
    Table 1 reports KKT solves for problem (39), but the KKT system, coupling to final time T, and endpoint constraints are never written down.
  • standard math Chebyshev-Fourier identities and the DCT-I aliasing rule are valid on the CGL grid.
    Used in Eqs. (12)-(16), (43), and (52)-(54); standard spectral material, correctly applied.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Million-Point Fast Trajectory Optimization Solver." pith.science (2026). https://pith.science/paper/JUJXUAA3

@misc{pith2026250901855,
  author       = {Pith},
  title        = {Pith review of: A Million-Point Fast Trajectory Optimization Solver},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JUJXUAA3}},
  note         = {Machine review of arXiv:2509.01855}
}
abstract

One might argue that solving a trajectory optimization problem over a million grid points is preposterous. How about solving such a problem at an incredibly fast computational time? On a small form-factor processor? Algorithmic details that make possible this trifecta of breakthroughs are presented in this paper. The computational mathematics that deliver these advancements are: (i) a Birkhoff-theoretic discretization of optimal control problems, (ii) matrix-free linear algebra leveraging Krylov-subspace methods, and (iii) a near-perfect Birkhoff preconditioner that helps achieve $\mathcal{O}(1)$ iteration speed with respect to the grid size,~$N$. A key enabler of this high performance is the computation of Birkhoff matrix-vector products at $\mathcal{O}(N\log(N))$ time using fast Fourier transform techniques that eliminate traditional computational bottlenecks. A numerical demonstration of this unprecedented scale and speed is illustrated for a practical astrodynamics problem.

Figures

Figures reproduced from arXiv: 2509.01855 by the authors.

Figure 1
Figure 1. A Schematic for the State-of-Practice in Trajecto [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. ). That is, the condition number of a D-matrix-based PS method (i.e., a Lagrange interpolation-based 101 102 103 104 nodes [N] 100 102 104 106 108 condition number [2-norm] Legendre-Gauss-Lobatto Chebyshev-Gauss-Lobatto Chebyshev-Gauss-Radau Legendre-Gauss-Radau Lagrange PS Methods [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Condition Numbers for Birkhoff-Theoretic Method [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Variants of Birkhoff Discretization Methods for ( [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Relative Run Times for Computing B a V Using FastBV (blue) and as Ordinary Matrix Multiplication (red). Furthermore, to remove other practical factors in compute time,8 the run times were averaged over 100 ap￾plications of the operations. An important point to note in …
Figure 6
Figure 6. Figure 6: shows the results of a scaling experiment using FastLinSol using a GMRES implementation for Step 4. The run times in [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Candidate optimal trajectory and control solutio [PITH_FULL_IMAGE:figures/full_fig_p016_7.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Universal Birkhoff Method for Computing Extremals in the Elliptic Restricted Three-Body Problem

    math.OC 2026-08 conditional novelty 5.0 of 10

    The authors compute and verify optimal finite-thrust transfers between an L1 Lyapunov orbit and an L2 southern near-rectilinear halo orbit in the elliptic restricted three-body problem using a guess-free Birkhoff spec...

Reference graph

Works this paper leans on

44 extracted references · 38 canonical work pages · cited by 1 Pith paper

  1. [22]

    Implementations of the Univ ersal Birkhoff Theory for Fast Trajectory Optimization,

    R. J. Proulx and I. M. Ross, “Implementations of the Univ ersal Birkhoff Theory for Fast Trajectory Optimization,” Journal of Guidance, Control and Dynamics , vol. 47, no. 12, pp. 2482–2496, Dec. 2024. https://doi.org/10.2514/1.G007738

    work page doi:10.2514/1.g007738 2024

  2. [1]

    The Million Point Computational Optimal Con trol Challenge,

    I. M. Ross, “The Million Point Computational Optimal Con trol Challenge,” SIAM Conference on Control and its Applications , MS24, July 10–13, 2017, Pittsburgh, P A

    work page 2017

  3. [2]

    A Nonsmooth Calc ulus for Solving Some Graph- Theoretic Control Problems,

    I. M. Ross, M. Karpenko and R. J. Proulx, “A Nonsmooth Calc ulus for Solving Some Graph- Theoretic Control Problems,” I F AC-PapersOnLine, V olume 49, Issue 18, 2016, Pages 462–467, https://doi.org/10.1016/j.ifacol.2016.10.208

    work page doi:10.1016/j.ifacol.2016.10.208 2016

  4. [3]

    An Optimal Control Theory for the Traveling Salesman Problem and Its Variants

    I. M. Ross and R. J. Proulx and M. Karpenko, “An Optimal Con trol Theory for the Traveling Salesman Problem and Its V ariants,” arXiv preprint, arXiv:2005.03186, 2020, https://arxiv.org/abs/2005.03186

    work page Pith review arXiv 2005

  5. [4]

    Arora and B

    S. Arora and B. Barak, Computational Complexity: A Modern Approach , Cambridge Univerity Press,

    work page

  6. [5]

    A Perspective on Methods for Tra jectory Optimization,

    I. M. Ross and F. Fahroo, “A Perspective on Methods for Tra jectory Optimization,” AIAA/AAS As- trodynamics Specialist Conference and Exhibit , 5-8 August, 2002, Monterey, CA. AIAA 2002-4727. https://doi.org/10.2514/6.2002-4727

    work page doi:10.2514/6.2002-4727 2002

  7. [6]

    A Survey of Methods Available for the Numer ical Optimization of Continuous Dy- namic Systems,

    B. A. Conway, “A Survey of Methods Available for the Numer ical Optimization of Continuous Dy- namic Systems,” Journal of Optimization Theory and Applications , V ol. 152, 2012, pp. 271–306. https://doi.org/10.1007/s10957-011-9918-z

    work page doi:10.1007/s10957-011-9918-z 2012

  8. [7]

    AnAlgorithmic Guide for Finite-Dimensional Optimal Control Problems,

    J-B. Caillau, R. Ferretti, E. Tr´ elat, and H. Zidani, “AnAlgorithmic Guide for Finite-Dimensional Optimal Control Problems,” Chapter 15, Handbook of Numerical Analysis , Elsevier, V olume 24, 2023, Pages 559–626, https://doi.org/10.1016/bs.hna.2022.11.006

    doi:10.1016/bs.hna.2022.11.006 2023

Show all 44 references
  1. [8]

    A Universal Birkhoff Theory for Fast Traject ory Optimization,

    I. M. Ross, “A Universal Birkhoff Theory for Fast Traject ory Optimization,” Journal of Guidance, Con- trol and Dynamics, vol. 47, no. 12, pp. 2468–2481, Dec. 2024. https://doi.org /10.2514/1.G007737

    2024 doi

  2. [9]

    J. M. Longuski, J. J. Guzm´ an and J. E. Prussing, Optimal Control with Aerospace Applications, Springer, New Y ork, N.Y ., 2014

    2014

  3. [10]

    I. M. Ross, A Primer on Pontryagin’s Principle in Optimal Control , Second Edition, Collegiate Pub- lishers, San Francisco, CA, 2015

    2015

  4. [11]

    A Roadmap for Optimal Control: The Right Way to Commute,

    I. M. Ross, “A Roadmap for Optimal Control: The Right Way to Commute,” Annals of the New Y ork Academy of Sciences , 1065/1, 2005, 210–231. https://doi.org/10.1196/annals .1370.015

    2005 doi

  5. [12]

    SNOPT: An SQP Al gorithm for Large-Scale Constrained Optimization,

    P . E. Gill, W . Murray and M. A. Saunders, “SNOPT: An SQP Al gorithm for Large-Scale Constrained Optimization,” SIAM Journal on Optimization , V ol. 12, No. 4, 2002, pp. 979–1006

    2002

  6. [13]

    On the implementation ofan interior-point filter line-search algorithm for large-scale nonlinear programming,

    A. W¨ achter and L. T. Biegler, “On the implementation ofan interior-point filter line-search algorithm for large-scale nonlinear programming,” Mathematical Programming, V ol. 106, No. 1, Mar 2006, pp. 25–57

    2006

  7. [14]

    A Matrix-Free Trust- Region SQP Method for Equality Constrained Optimization,

    M. Heinkenschloss and D. Ridzal, “A Matrix-Free Trust- Region SQP Method for Equality Constrained Optimization,” SIAM Journal on Optimization , V ol. 24, No. 3, 2014, pp. 1507–1541

    2014

  8. [15]

    Hessians in Birkhoff-Theoretic Trajector y Optimizaiton,

    I. M. Ross, “Hessians in Birkhoff-Theoretic Trajector y Optimizaiton,” Journal of Guidance, Control and Dynamics, V ol. 48, No. 9, 2025, pp. 2105–2112. https://doi.org/10.2514/1.G008778

    2025 doi

  9. [16]

    A Review of Pseudospectral O ptimal Control: From Theory to Flight,

    I. M. Ross and M. Karpenko, “A Review of Pseudospectral O ptimal Control: From Theory to Flight,” Annual Reviews in Control, V ol.36, No.2, pp.182–197, 2012. https://doi.org/10.1016/j.arcontrol.2012.09.002

    2012 doi

  10. [17]

    Boyd, Chebyshev and F ourier Spectral Methods, Dover Publications, Inc., Minola, New Y ork, 2001

    J. Boyd, Chebyshev and F ourier Spectral Methods, Dover Publications, Inc., Minola, New Y ork, 2001

    2001

  11. [18]

    L. N. Trefethen, Spectral Methods in MATLAB, SIAM, Philadelphia, P A, 2000

    2000

  12. [19]

    Direct Trajectory Optimizati on by a Chebyshev Pseudospectral Method,

    F. Fahroo and I. M. Ross, “Direct Trajectory Optimizati on by a Chebyshev Pseudospectral Method,” Journal of Guidance, Control and Dynamics , V ol. 25, No. 1, 2002, pp.160-166. https://doi.org/10.2514/2.4862 19

    2002 doi

  13. [20]

    Costate Computation b y a Chebyshev Pseudospectral Method,

    Q. Gong, I. M. Ross and F. Fahroo, “Costate Computation b y a Chebyshev Pseudospectral Method,” Journal of Guidance, Control and Dynamics , V ol. 33, No. 2, pp. 623-628, 2010. https://doi.org/10.2514/1.45154

    2010 doi

  14. [21]

    Fas t Mesh Refinement in Pseudospectral Optimal Control,

    N. Koeppen, I. M. Ross, L. C. Wilcox and R. J. Proulx, “Fas t Mesh Refinement in Pseudospectral Optimal Control,” Journal of Guidance, Control, and Dynamics , vol. 42 no. 4, pp. 711-722, 2018. http://dx.doi.org/10.2514/1.G003904

    2018 doi

  15. [23]

    Pseudospectral Knotting Meth ods for Solving Optimal Control Problems,

    I. M. Ross and F. Fahroo, “Pseudospectral Knotting Meth ods for Solving Optimal Control Problems,” Journal of Guidance, Control and Dynamics , V ol. 27, No. 3, pp. 397-405, 2004. https://doi.org/10.2514/1.3426

    2004 doi

  16. [24]

    Autonomous Pseudospectral Knot ting Methods for Space Mission Optimiza- tion,

    Q. Gong and I. M. Ross, “Autonomous Pseudospectral Knot ting Methods for Space Mission Optimiza- tion,” Advances in the Astronatuical Sciences , V ol. 124, 2006, AAS 06-151, pp. 779–794

    2006

  17. [25]

    Spectral Algorithm fo r Pseudospectral Methods in Optimal Con- trol,

    Q. Gong, F. Fahroo and I. M. Ross, “Spectral Algorithm fo r Pseudospectral Methods in Optimal Con- trol,” Journal of Guidance, Control, and Dynamics , vol. 31 no. 3, pp. 460-471, 2008. http://dx.doi.org/10.2514/1.32908

    2008 doi

  18. [26]

    hp-Adaptive Discontinuous Galerkin Finite Element Methods for First-Order Hyperbolic Problems,

    P . Houston and E. S¨ uli, “hp-Adaptive Discontinuous Galerkin Finite Element Methods for First-Order Hyperbolic Problems,” SIAM Journal on Scientific Computing , vol.23, no. 4, pp. 1226–1252, 2001

    2001

  19. [27]

    Spec- tral/hp Element Methods: Recent Developments, Applicatio ns, and Perspectives,

    Xu, H., Cantwell, C. D., Monteserin, C., Eskilsson, C., Engsig-Karup, A. P . and Sherwin, S. J., “Spec- tral/hp Element Methods: Recent Developments, Applicatio ns, and Perspectives,” Journal of Hydrody- namics, vol. 30, no. 1, pp. 1–22, 2018

    2018

  20. [28]

    Pseudospectral Optimal Control and its Convergence Theorems,

    W . Kang, I. M. Ross and Q. Gong, “Pseudospectral Optimal Control and its Convergence Theorems,” Analysis and Design of Nonlinear Control Systems , Springer-V erlag, Berlin Heidelberg, 2008, pp. 109–

    2008

  21. [29]

    Rate of Convergence for a Legendre Pseudospec tral Optimal Control of Feedback Lin- earizable Systems,

    W . Kang, “Rate of Convergence for a Legendre Pseudospec tral Optimal Control of Feedback Lin- earizable Systems,” Journal of Control Theory and Applications , V ol. 8, No. 4, pp. 391–405, 2010. http://dx.doi.org/10.1007/s11768-010-9104-0

    2010 doi

  22. [30]

    Solving Sparse Linear System s Faster than Matrix Multiplication,

    R. Peng and S. S. V empala, “Solving Sparse Linear System s Faster than Matrix Multiplication,” Com- munications of the ACM , 67/7, pp. 79–86. https://doi.org/10.1145/3615679

    doi

  23. [31]

    Solve for x: Technical Pespective,

    Spielman, D, “Solve for x: Technical Pespective,” Communications of the ACM , 67/7, pp. 78. https://doi.org/10.1145/3643835

    doi

  24. [32]

    L. N. Trefethen and D. Bau, III, Numerical Linear Algebra , SIAM, Philadelphia, P A, 1997. Chs. 1, 3, 4, 6

    1997

  25. [33]

    Golub and C

    G. Golub and C. F. V an Loan, Matrix Computations, 4th Edition, The Johns Hopkins University Press, Baltimore, MD, 2013. Chs. 1, 3, 4, 11

    2013

  26. [34]

    Saad, Y ., Iterative Methods for Sparse Linear Systems , Society for Industrial and Applied Mathematics, 2nd ed., 2003

    2003

  27. [35]

    C. T. Kelley, Iterative Methods for Optimization, Society for Industrial and Applied Mathematics, 1999

    1999

  28. [36]

    gmres: Solve System of Linear Equa tions – Generalized Minimimum Residual Method,

    The MathWorks Inc., “gmres: Solve System of Linear Equa tions – Generalized Minimimum Residual Method,” The MathW orks Inc., Natic, MA., 2024. https://www.mathworks.com/help/matlab/ref/gmres.html

    2024

  29. [37]

    A Universal Bir khoff Pseudospectral Method for Solv- ing Boundary V alue Problems,

    I. M. Ross, R. J. Proulx and C. F. Borges, “A Universal Bir khoff Pseudospectral Method for Solv- ing Boundary V alue Problems,” Applied Mathematics and Computation , V ol. 454, 128101, 2023. https://doi.org/10.1016/j.amc.2023.128101

    2023

  30. [38]

    Clarke, Functional Analysis, Calculus of V ariations and Optimal Co ntrol, Springer-V erlag, London, 2013

    F. Clarke, Functional Analysis, Calculus of V ariations and Optimal Co ntrol, Springer-V erlag, London, 2013

    2013

  31. [39]

    R. B. Vinter, Optimal Control, Birkh¨ auser, Boston, MA, 2000

    2000

  32. [40]

    Enhancements to the DIDO Optimal Control To olbox,

    I. M. Ross, “Enhancements to the DIDO Optimal Control To olbox,” arXiv preprint, arXiv:2004.13112,

    2004 arXiv

  33. [41]

    MATLAB Documentation: mldivide,\, 2024

    2024

  34. [42]

    Algorithm 832: UMFP ACK V4.3—An Unsymmetric-Pattern Multifrontal Method,

    T. A. Davis, “Algorithm 832: UMFP ACK V4.3—An Unsymmetric-Pattern Multifrontal Method,” ACM Trans. Math. Softw., V ol. 30, No. 2, Jun 2004, pp. 196–199. 20

    2004

  35. [126]

    http://dx.doi.org/10.1007/978-3-540-74358-3 8

    doi

  36. [2020]

    https://doi.org/10.48550/arXiv.2004.13112

    2004 doi arXiv

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.