Pith. sign in

REVIEW 4 major objections 5 minor 19 references

Route Optimization Over Scheduled Services For Large-Scale Package Delivery Networks

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A pruning algorithm that keeps exactly the schedule legs lying on feasible paths reduces trailer routing over scheduled services to a tractable scale, and a stabilized column-generation heuristic solves industrial instances with up to 74…

desk verdict Useful applied OR paper with a genuinely new problem and an exact network reduction, but the real-time claims outrun the reported evidence. read the letter →

arxiv 2507.20844 v1 pith:TFN6SB6P submitted 2025-07-28 math.OC

Mohammed Faisal Ahmed , Pascal Van Hentenryck , Ahmed El Nashar This is my paper · ORCID
classification math.OC MSC 90B0690C1190C3590C39
keywords packagedeliverynetworkscheduledservicestrailerpathoptimizationcolumngenerationLagrangiandualreductiontime-dependentshortesttimewindows
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

The paper addresses a daily problem for large package-delivery networks: given a fixed set of driver schedules, each made of timed legs between hubs, how should trailer requests be routed over those legs under tractor capacity and time-window constraints, while minimizing used schedules and total miles? It claims two things. First, a simple polynomial-time pruning rule can identify, for every request, exactly the set of scheduled legs that appear on some time-feasible path, so discarding all other legs loses no optimal solution. Second, a stabilized column-generation heuristic whose pricing subproblem is a time-dependent shortest path can solve the resulting model at industrial scale: up to 74 billion request-leg pairs in planning, reaching within 3.7% to 5.7% of a Lagrangian lower bound in 1.7 to 10.3 hours, and within 3% of a lower bound in 75% of real-time test cases in under a minute. The paper argues this makes automated trailer routing practical for weekly re-planning and same-day insertion of new requests, with claimed cost savings of 2.3% to 3.2% over the current manual process.

What carries the argument

The key machinery is the Earliest-and-Latest-Start-Time Pruning algorithm, EALSP. It computes two labels per hub for each request: an earliest-arrival label propagated forward from the request's origin, and a latest-start label propagated backward from its destination, both by Dijkstra-like dynamic programming over the scheduled-leg network. A leg survives if its departure is no earlier than the earliest-arrival time at its origin hub and its arrival is no later than the latest-start time at its destination hub. This reduction is exact: the surviving legs are exactly those on some feasible path, so the pruned instance contains an optimal solution whenever one exists. The companion mechanism is a column-generation heuristic whose pricing subproblem is a time-dependent shortest path with waiting allowed at hubs; a stabilized variant of the decomposition keeps dual values from oscillating, and a Lagrangian relaxation provides the lower bound against which solution gaps are measured.

What would settle it

Construct a small TPOSSP instance with a handful of hubs, schedules, and requests where the true optimum is known by exhaustive enumeration over all feasible path sets, and check both that every leg on every feasible path survives the EALSP condition and that the heuristic's final solution is within the paper's claimed gap. A single leg on a feasible path that fails the condition would directly refute Theorem 1, and a single instance where the heuristic misses the claimed gap would refute the performance claims.

Watch

Extended reading notes

Core claim

The central discovery is the EALSP characterization of an optimal sub-network for each request. For a request with an earliest pickup time and a latest arrival time, the algorithm computes, for every hub, the earliest arrival time of any feasible path from the request's origin to that hub, and the latest start time of any feasible path from that hub to the destination. A scheduled leg lies on some feasible path, by the paper's Theorem 1, exactly when its departure time is no earlier than the earliest arrival time at its origin hub and its arrival time is no later than the latest start time at its destination hub. The union of such legs is therefore precisely the sub-network of legs that can appear on a feasible path, and pruning everything else is lossless. On top of this, the paper develops a path-based decomposition whose pricing problem is a time-dependent shortest path on that sub-network, and claims that the stabilized column-generation heuristic reaches the reported solution gaps at scales where the arc-based mixed-integer program is intractable.

Load-bearing premise

The load-bearing premise is that the base plan used for the 2.3% to 3.2% cost comparison is a realistic representation of what the logistics provider actually does today; the paper does not document how that base plan is generated or updated, and if it is not a realistic baseline the headline savings and the tens-of-millions-of-dollars figure would be overstated.

Editorial extensions

If this is right

  • The EALSP reduction is lossless, so solving the pruned network is equivalent to solving the full network and all reported solution-quality claims apply to the original instances.
  • In planning, the heuristic reaches within 3.7% to 5.7% of the Lagrangian lower bound on instances with up to 74 billion request-leg pairs, and improves on the current base plan by 2.3% to 3.2% in total cost and 8.5% to 10.3% in empty miles.
  • In real-time operations, the same heuristic returns solutions within 3% of the lower bound in 75% of the test cases, all within one minute, so new requests can be inserted on the day of operations.
  • Because each request's sub-network can be cached and updated only when schedules or requests change, repeated planning runs can avoid recomputing the pruning from scratch.
  • The pruning step cuts runtimes by 53% to 85% for the heuristic and by 83% to 99% for the exact arc-based mixed-integer program on the small real-time instances.

Reading between the lines

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

  • The savings percentages are only as credible as the undocumented base plan used for comparison; if that base plan is not a realistic model of the manual process, the economic claims would shrink. This is an inference because the paper does not describe how the base plan was built.
  • The EALSP condition is not specific to trailers: any scheduled-service network where time feasibility is the main coupling between legs, such as rail, air, or intermodal freight, could use the same two-label pruning, provided capacity and schedule-selection constraints are handled downstream.
  • Because the planning gaps are measured against a Lagrangian lower bound rather than a proven optimum, the true deviation from optimality could be smaller than the reported 3.7% to 5.7% if that bound is loose; a fair reader should treat the bound as an anchor, not as the optimum itself.
  • A direct test of the approach would be to run the same pipeline on synthetic instances with known optima, or on historical data after the fact, to ground-truth both the optimality gap and the savings fractions.
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

4 major / 5 minor

Summary. This paper formalizes the Trailer Path Optimization with Scheduled Services Problem (TPOSSP), in which each request, representing a trailer movement with a time window, must be assigned a path over existing scheduled legs subject to tractor capacity and schedule-utilization costs. The authors propose an arc-based MIP, prove that a polynomial-time pruning algorithm (EALSP) computes the optimal sub-network of legs that can appear on any feasible path for each request, and then propose a stabilized column-generation heuristic whose pricing subproblem is a time-dependent shortest path. The approach is evaluated on industrial instances: planning instances with up to 74 billion request-leg pairs are reported to be solved within 3.7%-5.7% of a Lagrangian lower bound in 1.7-10.3 hours, and real-time instances are reported to be solved within 3% of optimality in under one minute in 75% of cases, with claimed cost savings of 2.3%-3.2% over a base plan.

Significance. If the results hold, the paper contributes a new, industrially motivated problem formulation and a scalable decomposition method, with the EALSP sub-network theorem being a clean and potentially transferable theoretical contribution. The reported scale of 74 billion request-leg pairs is impressive, and the time-dependent pricing and stabilization ideas are appropriate for the problem. However, several load-bearing issues prevent the strongest claims from being accepted as stated: the final master problem is written with continuous path variables while only y variables are declared integer, the planning gaps are measured against a Lagrangian bound rather than a certified optimum, the real-time latency figures appear to exclude EALSP preprocessing, and the baseline used for the business-savings claims is undocumented. These issues need to be resolved before the empirical conclusions are reliable.

major comments (4)
  1. [Section 6.3 / Figure 6] The restricted master problem in Figure 6 is written with λ ∈ [0,∞), and Section 6.3 states that integer solutions are found by enforcing integrality on the y variables only. In TPOSSP, each request is a single trailer that must follow exactly one path; a fractional convex combination of two candidate paths for the same request is not a feasible route. Unless the implementation actually imposes integrality on the λ variables, or a rounding/post-processing procedure is described, the reported objective values in Section 7 are not upper bounds on the TPOSSP optimum, and the savings versus the base plan are not necessarily achievable. This issue is load-bearing for all empirical claims and must be clarified and, if λ is in fact continuous, corrected with new experiments.
  2. [Section 7.4 / Table 3] Figure 10 and the Section 7.4 discussion report only CG-heuristic runtimes, while Table 3 lists EALSP preprocessing times of 6-60 seconds for the real-time instances (60s for instance 2.5, 54s for 2.14, and 42s for 2.15). For newly arriving requests, the sub-network must be computed after the request arrives, so the end-to-end latency is at least the EALSP time plus the CG time. The paper does not state whether the Section 7.4 runtimes include EALSP. As written, the claim that the method meets the '1 minute or so' real-time requirement stated in Section 1.1 is unsupported for several instances. The authors should report per-instance end-to-end wall-clock times, or explicitly justify why EALSP is not on the critical path.
  3. [Sections 7.2, 1.3, and Abstract] The planning optimality gaps of 3.7%-5.7% are computed against a Lagrangian dual solved by the Surrogate Lagrangian Relaxation with a stopping criterion of no improvement for 50 iterations. This yields a lower-bound gap, not a proven gap to the true optimum; the contribution statement in Section 1.3 that the CG-heuristic reaches 'within 3.7%-5.7% of the true optimal solution' overstates what is measured. The abstract's phrase 'based on a MIP relaxation' is also inconsistent with the Lagrangian relaxation described in Section 6.4. The authors should present these as gaps against a lower-bound estimate, state whether the terminal multipliers were evaluated to certify a valid Lagrangian bound, and revise all 'true optimal' wording accordingly.
  4. [Section 7.3] The business-benefit comparison is made against an 'available base solution' described only as the current manual process. The paper does not specify how this base plan is generated, how its schedule and mile costs are computed, which requests it covers, or how it is updated for each of the six planning instances. Since the 2.3%-3.2% cost reduction and the tens-of-millions-of-dollars claim are computed as differences from this baseline, the economic claim cannot be audited or reproduced. The authors should provide a precise description of the base plan construction and cost accounting, or report the comparison in a way that does not depend on an opaque benchmark.
minor comments (5)
  1. [Lemma 1, Section 5.1] The lemma's stated condition eat_i ≤ ts_l ≤ lst_j appears to contain a typo: the feasibility of leg l from i to j requires te_l ≤ lst_j, not ts_l ≤ lst_j. The proof in Appendix A uses the correct condition te_l ≤ lst_j, so the theorem remains sound, but the inconsistent statement should be corrected.
  2. [Table 2] The verbal definition of E_r in Table 2 says legs whose 'start time is less than eat_i', which is the reverse of the correct condition eat_i ≤ ts_l used in Equation (5a). The table entry should be rewritten for consistency.
  3. [Introduction and Section 7.5] The reported speedup percentages are inconsistent across the abstract, Section 1.2, Section 1.3, and Section 7.5: the CG speedup is given variously as 85%, 53%-85%, and 53%-82%, while the Arc-MIP speedup is given as 85%-99% in Section 1.3 and 83%-99% in Section 7.5. These numbers should be reconciled.
  4. [Throughout] There are several grammatical slips, e.g., 'with translates into savings' in the abstract and 'returns solution within 3% of optimality' in Section 1.3. A careful proofread would improve the presentation.
  5. [Reproducibility] No data or code availability is mentioned. Given the proprietary nature of the instances, a full release may not be possible, but the authors should at least state the availability status and consider sharing anonymized or synthetic instances to allow independent verification of the algorithmic claims.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity; the EALSP correctness proof and CG lower-bound comparisons are self-contained, with only minor non-load-bearing self-citations.

full rationale

I walked the main derivation chain. The EALSP result (Section 5, Definition 5.1, Lemma 1, Theorem 1) is a genuine correctness proof: the optimal sub-network is defined as the union of legs lying on some feasible path, and Theorem 1 proves that the EALSP set Er = {l : eat_dl <= ts_l, te_l <= lstr_el} equals that union by constructing feasible prefix and suffix paths from the definitions of earliest-arrival and latest-start times. The result is not assumed by construction; it is a substantive characterization with a proof. Lemma 2's critical-leg pruning is likewise a standard dominance argument, not a restatement of the claim. The CG-heuristic's optimality claims are measured against Arc-MIP and Lagrangian lower bounds, not against a fitted parameter or against the heuristic's own output. The parameter choices (Paths, NumIterations, MaxCost, Mode) are algorithmic tuning choices, not predictions, and the comparison to the manual base plan is an external benchmark, however under-documented. The only self-citations are Mahéo et al. (2017) and Guan et al. (2024) in the literature review, and neither is load-bearing for the EALSP algorithm, the column-generation method, or the computational claims. Concerns about omitted EALSP preprocessing time in real-time runtimes and the undocumented base plan are correctness/validity issues, not circularity.

Assumptions & free parameters 5 free parameters · 6 assumptions · 1 invented entities

The model inputs (costs, capacities, time windows) come from the industrial partner and are not fitted by the paper. The algorithmic parameters Paths, NumIterations, MaxCost, and Mode are chosen by hand and tuned on the test instances. The proofs rely on standard shortest-path and Lagrangian duality results. No new physical entities are introduced; the dummy schedule is a modeling device to guarantee feasibility.

free parameters (5)
  • Paths (paths per request per CG iteration) = 5 to 50; 50 used for final results
    Section 7.2 and 7.6; chosen by hand based on experiments. Higher values improve standard CG convergence but increase runtime per iteration.
  • NumIterations (CG iteration limit) = 50
    Section 7.2; fixed to 50 iterations. No evidence that 50 is optimal or sufficient for all instances.
  • MaxCost (maximum path cost in TDSPP pricing) = 0
    Section 7.2; set to 0, meaning only optimal-cost paths are generated. This restricts the column pool.
  • Mode (standard vs stabilized CG) = Stabilized for planning, Standard for real-time
    Section 7.2; method selected based on experiments. The mode affects runtimes and solution quality.
  • Lagrangian stopping criterion (no improvement for 50 iterations) = 50 iterations
    Section 7.2; used to terminate the surrogate Lagrangian bound computation. The tightness of the resulting lower bound is not measured.
assumptions (6)
  • domain assumption A tractor can carry up to three short trailer equivalents; trailer volumes are normalized (28'=1, 45'=1.5, 48'=1.9, 53'=2.5) and capacities are multiples of short trailers.
    Section 3; this capacity model is taken from industrial practice, not derived in the paper.
  • domain assumption Leg start and end times include the time to attach and detach trailers.
    Section 3; simplifies the time feasibility constraints and avoids modeling handling times explicitly.
  • domain assumption Requests with no feasible path in the given schedules are removed and replaced by direct dummy schedules from origin to destination and back, with cost sigma^dummy_s, to guarantee model feasibility.
    Section 3; this changes the problem by adding artificial schedules whose cost is an input parameter.
  • standard math The time-dependent shortest path problem with nonnegative leg costs and waiting at nodes can be solved with a Dijkstra-like label-setting algorithm, and the critical-legs Lemma 2 ensures pruning non-critical departures is optimal.
    Section 6.1 and Lemma 2; relies on established shortest path theory with step-wise cost functions.
  • standard math The Lagrangian relaxation with nonnegative multipliers provides a valid lower bound on the optimal MIP value; the surrogate Lagrangian method of Bragin et al. (2015) is used to find dual values.
    Section 6.4; standard Lagrangian duality. The stopping rule of 50 non-improving iterations is ad hoc but any fixed multiplier set gives a valid lower bound.
  • ad hoc to paper The optimality gap for planning instances is measured against the Lagrangian lower bound, which the paper assumes is a meaningful proxy for the true optimum.
    Sections 6.4 and 7.2; the tightness of the bound is not verified, so the reported 3.7-5.7% gap is only as good as the bound.
invented entities (1)
  • Dummy direct schedules
    purpose: Ensure the model is feasible for every request by providing a direct origin-destination schedule when no feasible path exists in the regular schedule set.
    Section 3; a modeling construct with an input cost sigma^dummy_s. It is not a real operational schedule and has no external validation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Route Optimization Over Scheduled Services For Large-Scale Package Delivery Networks." pith.science (2026). https://pith.science/paper/TFN6SB6P

@misc{pith2026250720844,
  author       = {Pith},
  title        = {Pith review of: Route Optimization Over Scheduled Services For Large-Scale Package Delivery Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TFN6SB6P}},
  note         = {Machine review of arXiv:2507.20844}
}
read the original abstract

This paper introduces the Trailer Path Optimization with Schedule Services Problem (TPOSSP) and proposes a column-generation heuristic (CG-heuristic) to find high-quality solutions to large-scale instances. The TPOSSP aims at determining trailer routes over a time-dependent network using existing scheduled services, while considering tractor capacity constraints and time windows for trailer pickups and deliveries. The objective is to minimize both the number of schedules used and the total miles traveled. To address the large scale of industrial instances, the paper proposes a network reduction technique that identifies the set of feasible schedule-legs for each requests. Moreover, to address the resulting MIP models, that still contains hundred of millions variables, the paper proposes a stabilized column-generation, whose pricing problem is a time-dependent shortest path. The approach is evaluated on industrial instances both for tactical planning where requests for the entire network are re-optimized and for real-time operations where new requests are inserted. In the tactical planning setting, the column-generation heuristic returns solutions with a 3.7%-5.7% optimality gap (based on a MIP relaxation) in under 1.7-10.3 hours, and improves the current practice by 2.3-3.2%, with translates into savings of tens of millions of dollars a year. In the real-time setting, the column-generation heuristic returns solution within 3% of optimality in under 1 minute, which makes it adequate for real-time deployment. The results also show that the network reduction decreases run times by 85% for the column-generation heuristic.

Figures

Figures reproduced from arXiv: 2507.20844 by the authors.

Figure 1
Figure 1. Illustrating a Potential Path of a Trailer, [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. A TPOSSP Instance with three schedules and three inputs and three hubs. Each schedule is shown with colored [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. A Solution to the TPOSSP Instance: The solution contains the path of each request, and schedule 2 was eliminated. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: The Path Feasibility Constraints G(x). min x,y X s∈S σs ∗ ys + X l∈L X r∈R θl ∗ ml ∗ vr ∗ xlr s.t. {x, y} ∈ G(x) ∩ C(x, y) ∩ E(x, y) (4a) x ∈ {0, 1} |L|X|R| , y ∈ {0, 1} |S| [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]
Figure 5
Figure 5. Figure 5: The Arc-Based MIP Formulation for the TPOSSP (Arc-MIP). [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: The Restricted Master Problem using Path Sets [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: The Sub-Problem Formulation. negative leg cost, where the time-dependency comes from the fact that the possible outbound leg options from a hub depend on the arrival time to the hub. The cost of the leg is given by αlr = θl ∗ml ∗ vr −(π c l ∗ vr +π s lr). Note that αlr…
Figure 8
Figure 8. Figure 8: The Lagrangian Relaxation Formulation LR(κ C , κS ). It is important to highlight three important facts. First, as mentioned earlier, even after the proposed network reduction, the Arc-MIP still contains hundreds of millions variables and is essentially computationally…
Figure 9
Figure 9. Figure 9: Planning Results: the figure on the left shows the reduction in cost, total miles, schedules and empty miles compared [PITH_FULL_IMAGE:figures/full_fig_p020_9.png]
Figure 10
Figure 10. Figure 10: Real-time results: the figure on the left is the optimality gap reached by CG-heuristic and the figure on the right is [PITH_FULL_IMAGE:figures/full_fig_p021_10.png]
Figure 11
Figure 11. Figure 11: The Impact of stabilization and varying paths generated per request per iteration of column generation [PITH_FULL_IMAGE:figures/full_fig_p022_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

19 extracted references · 19 canonical work pages

  1. [1]

    An intermodal multicommodity routing problem with scheduled services, volume 53

    Burak Ayar and Hande Yaman. An intermodal multicommodity routing problem with scheduled services, volume 53. 2012

    work page 2012

  2. [2]

    Johnson, George L

    Cynthia Barnhart, Ellis L. Johnson, George L. Nemhauser, Martin W. P. Savelsbergh, and Pamela H. Vance. Branch-and-price: Column generation for solving huge integer programs. Operations Research, 46 0 (3): 0 293--432, 1998

    work page 1998

  3. [3]

    Hane, and Pamela H

    Cynthia Barnhart, Christopher A. Hane, and Pamela H. Vance. Using branch-and-price-and-cut to solve origin-destination integer multicommodity flow problems. Operations Research, 48 0 (2): 0 318--326, 2000

    work page 2000

  4. [4]

    The continuous-time service network design problem

    Natashia Boland, Mike Hewitt, Luke Marshall, and Martin Savelsbergh. The continuous-time service network design problem. Operations Research, 65 0 (5): 0 1303--1321, 2017

    work page 2017

  5. [5]

    Bragin, Peter B

    Mikhail A. Bragin, Peter B. Luh, Joseph H. Yan, Nanpeng Yu, and Gary A. Stern k. Convergence of the surrogate lagrangian relaxation method. Journal of Optimization Theory and Applications, 164: 0 173--201, 2015

    work page 2015

  6. [6]

    Package delivery statistics: Capital one research, 2024

    CapitalOne. Package delivery statistics: Capital one research, 2024. URL https://capitaloneshopping.com/research/package-delivery-statistics/. [Online; accessed 14-Dec-2024]

    work page 2024

  7. [7]

    Service network design in freight transportation

    Teodor Gabriel Crainic. Service network design in freight transportation. European Journal of Operational Research, 122 0 (2): 0 272--288, 2000

    work page 2000

  8. [8]

    Scheduled service network design with resource acquisition and management

    Teodor Gabriel Crainic, Mike Hewitt, Michel Toulouse, and Duc Minh Vu. Scheduled service network design with resource acquisition and management. EURO J. Transp. Logist., 7: 0 277–309, 2017

    work page 2017

Show all 19 references
  1. [9]

    Dantzig and Philip Wolfe

    George B. Dantzig and Philip Wolfe. Decomposition principle for linear programs. Operations Research, 8 0 (1): 0 101--111, 1960

    1960

  2. [10]

    Path-based formulations for the design of on-demand multimodal transit systems with adoption awareness

    Hongzhao Guan, Beste Basciftci, and Pascal Van Hentenryck. Path-based formulations for the design of on-demand multimodal transit systems with adoption awareness. INFORMS JOURNAL ON COMPUTING, 0 0 (0), 2024

    2024

  3. [11]

    Dynamic discretization discovery algorithms for time-dependent shortest path problems

    Edward Yuhang He, Natashia Boland, George Nemhauser, and Martin Savelsbergh. Dynamic discretization discovery algorithms for time-dependent shortest path problems. INFORMS JOURNAL ON COMPUTING, 34 0 (2): 0 1086--1114, 2022

    2022

  4. [12]

    Pardalos

    Dukwon Kim and Panos M. Pardalos. A solution approach to the fixed charge network flow problem using a dynamic slope scaling procedure. Operations Research Letters, 24 0 (4): 0 195--203, 1999

    1999

  5. [13]

    Column generation for solving large scale multi-commodity flow problems for passenger transportation

    Benedikt Lienkamp and Maximilian Schiffer. Column generation for solving large scale multi-commodity flow problems for passenger transportation. European Journal of Operational Research, 314 0 (2): 0 703--717, 2024

    2024

  6. [14]

    Lübbecke and Jacques Desrosiers

    Marco E. Lübbecke and Jacques Desrosiers. Selected topics in column generation. Operations Research, 53 0 (6): 0 1007--1023, 2005

    2005

  7. [15]

    Benders decomposition for the design of a hub and shuttle public transit system

    Arthur Mahéo, Philip Kilby, and Pascal Van Hentenryck. Benders decomposition for the design of a hub and shuttle public transit system. Transportation Science, 53 0 (1): 0 77--88, 2017

    2017

  8. [16]

    The multicommodity network flow problem: state of the art classification, applications, and solution methods

    Khodakaram Salimifard and Sara Bigharaz. The multicommodity network flow problem: state of the art classification, applications, and solution methods. Operational Research, 22 0 (4): 0 1--47, 2022

    2022

  9. [17]

    Weighted dantzig-wolfe decomposition for linear mixed-integer programming

    Paul Wentges. Weighted dantzig-wolfe decomposition for linear mixed-integer programming. Int. Trans. Opl Res., 4 0 (2): 0 151--162, 1997

    1997

  10. [18]

    Service network design for freight transportation: a review

    Nicole Wieberneit. Service network design for freight transportation: a review. OR Spectrum, 30: 0 77--112, 2008

    2008

  11. [19]

    Scheduled service network design for freight rail transportation

    Endong Zhu, Teodor Gabriel Crainic, and Michel Gendreau. Scheduled service network design for freight rail transportation. Operations Research, 62 0 (2): 0 383--400, 2014

    2014

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.