Pith. sign in

REVIEW 3 major objections 6 minor 33 references

Approximation Algorithms for the Traveling Thief Problem

T0 review · 3 major / 6 minor · reviewed 2026-07-08 · glm-5.2

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

Pith's one-line read First Constant-Factor Approximation for the Traveling Thief Problem

desk verdict First constant-factor bicriteria approximation for the full Traveling Thief Problem (joint tour + packing), with a (9+ε, 9+ε) guarantee and a (2e+ε) approximation for Weighted TSP. read the letter →

arxiv 2607.05164 v1 pith:6APEY7JC submitted 2026-07-06 cs.DS

Jan Eube , Kelin Luo , Heiko Röglin , Sarah Sturm This is my paper
classification cs.DS
keywords problemtimeagentalphaitemsprofitalgorithmsapproximation
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 Traveling Thief Problem (TTP) fuses two classic NP-hard problems: the Traveling Salesperson Problem (deciding which cities to visit and in what order) and the Knapsack Problem (deciding which items to collect under a weight limit). The twist is that as the agent picks up heavier items, they move more slowly, so the routing and packing decisions interact in a way that makes neither subproblem independent of the other. Prior to this paper, only heuristics and exact (exponential-time) algorithms existed; no polynomial-time approximation guarantee of any kind was known for any variant combining both decisions. The authors introduce a variant called the Constrained TTP (CTTP), where one is given a target profit P and must find a tour collecting at least that much profit while minimizing travel time. They prove that for any epsilon > 0, there is a polynomial-time bicriteria (9+epsilon, 9+epsilon)-approximation: the algorithm finds a tour collecting at least 1/(9+epsilon) of the target profit while requiring travel time at most (9+epsilon) times the optimal travel time for that profit level. This extends to a (9+epsilon, 9+epsilon)-approximate Pareto set for the Bi-objective TTP, meaning the algorithm produces a small family of solutions such that for any possible trade-off between profit and travel time, one of the solutions in the family is within a factor of 9+epsilon on both objectives. As a stepping stone, the authors also give a (2e+epsilon)-approximation for the Weighted TSP subproblem, where the item set is fixed and only the tour must be optimized.

What carries the argument

The algorithm combines three subroutines: (1) a Quota TSP approximation for finding short tours collecting a target weight, (2) a (3+epsilon)-approximation for the Capacitated Orienteering Problem to find profitable subtours within weight-budgeted length limits, and (3) an FPTAS for the Multiperiod Binary Knapsack Problem to select which boundary items to collect as independent round-trips. The weight-range partition into speed-doubling intervals is the structural device that makes the decomposition possible.

What would settle it

If the k-TSP approximation does not extend to integer weights without modification, the Weighted TSP approximation factor is unsupported, and the Quota TSP subroutine used in the CTTP algorithm may not achieve its stated cost bound, which would break the approximation guarantee for the main result.

Watch

Extended reading notes

Core claim

The central mechanism is a two-algorithm approach that decomposes the difficulty of jointly optimizing routing and packing. The first algorithm (Algorithm 2) partitions the weight range into intervals where the agent's speed varies by at most a factor of two, then uses a capacitated orienteering subroutine to find profitable subtours within each interval. The second algorithm (Algorithm 3) handles items that sit at weight-interval boundaries by treating each such item as an independent round-trip from the depot, formulated as a Multiperiod Binary Knapsack Problem. The key insight is that for any optimal solution, either the bulk of the profit comes from items cleanly inside weight intervals—

Load-bearing premise

The extension of a known k-TSP approximation algorithm from unit-weight nodes to polynomially-bounded integer-weight nodes is stated informally—the authors say they examined the original algorithm and are confident it carries over without substantial modification, but do not provide a formal proof. This subroutine underpins both the Weighted TSP result and the CTTP algorithm.

Editorial extensions

If this is right

  • Any application modeled as a multi-component optimization problem where routing and packing interact through a speed-weight coupling now has a provable polynomial-time guarantee, whereas previously only heuristic or exponential-time methods were available.
  • The bicriteria approximation framework can serve as a benchmark for evaluating practical heuristics: if a heuristic solution is worse than the (9+epsilon, 9+epsilon) guarantee, the algorithm provides a certified better solution.
  • The Weighted TSP result (2e+epsilon approximation) is independently useful for any delivery or collection problem where the cost of travel depends on accumulated load, such as fuel-consumption-aware vehicle routing.
  • The extension to approximate Pareto sets means decision-makers can enumerate a polynomial-size family of trade-off solutions rather than facing an exponentially large Pareto frontier.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. This paper presents the first polynomial-time approximation algorithms for variants of the Traveling Thief Problem (TTP). The main result is a bicriteria (9+ε, 9+ε)-approximation for the Constrained TTP (CTTP), which extends to a (9+ε, 9+ε)-approximate Pareto set for the Bi-objective TTP. A secondary result is a (2e+ε)-approximation for Weighted TSP. The CTTP algorithm combines several subroutines—Capacitated Orienteering (COP), Quota TSP (QTSP), and Multiperiod Binary Knapsack (MPBKP)—with an interval decomposition of the weight range to handle the non-decreasing inverse-speed function. The proof structure is detailed: Lemma 10 lower-bounds OPT cost via interval decomposition, Lemmas 13–17 bound the profit and cost of the first solution π(1), Lemmas 21–24 bound the second solution π(2), and Theorem 25 combines them.

Significance. This is the first work to provide approximation algorithms for any variant of the TTP involving both routing and packing decisions, a notable advance for this multi-component optimization problem. The algorithmic framework—combining COP, QTSP, and MPBKP subroutines with weight-interval decomposition—is original and could inform future work on related problems. The (9+ε)-bicriteria guarantee is a concrete, falsifiable bound. The extension to a (2e+ε)-approximation for Weighted TSP via derandomization of the doubling algorithm is a clean secondary contribution. The parameter-free derivation (guessing D by geometric enumeration) is standard and sound.

major comments (3)
  1. Section 3, Theorem 6: The extension of the k-TSP / QTSP approximation from unit weights to polynomially bounded integer weights is stated informally: 'their arguments carry over to this generalized case.' Theorem 6 is load-bearing for both the Weighted TSP result (Theorem 1/9) and the CTTP algorithm (via Algorithm 1). The authors write they 'have examined the algorithm from [1] carefully and are confident that it also applies, without substantial modifications, to the case of general integer weights.' This is insufficient for a rigorous proof. Either a formal reduction (e.g., node-splitting with polynomial blowup) should be given explicitly, or a proof sketch showing why the Chaudhuri et al. analysis extends to integer weights should be provided. Without this, Theorems 1 and 9 rest on an unverified subroutine claim.
  2. Appendix B.2, Lemma 30: The stronger lower bound cost(π*) ≥ (1/2)·D_{π*}(0) + Σ 2^{i-1}·D_{π*}(T_i+1) is stated but not proven—the authors write 'One can get an intuition why also this stronger lemma is true by having a look at Figure 2' and then say 'The exact details of this analysis are omitted.' This lemma is load-bearing for the claim that the approximation factor does not increase when requiring all locations to be visited (a standard TTP model variant). The omission should be addressed with a full proof.
  3. Appendix B.4: The analysis for multiple items at the starting location (when items must be collected at the beginning) involves guessing a set S' from O(l²) candidates and arguing that p(S') ≥ p(S*)/2. The argument is sketched but not fully formalized—several steps ('we are then also guaranteed to collect a 1/(9+3ε) fraction') rely on the main CTTP analysis carrying through with modified f and w_max, but the interaction between the guessed initial weight and the interval decomposition (T_i thresholds) is not verified. Since this is a standard model variant, a more careful treatment should be provided.
minor comments (6)
  1. Section 5, Definition 12: The set A_i is defined as items visited on π* where both before and after the visit the collected weight lies in [T_i+1, T_{i+1}]. It would help to clarify that items where the weight crosses a threshold during the visit are excluded from all A_i, and to state how their profit is accounted for in the analysis.
  2. Lemma 13: The phrase 'We directly interpret the result of the subroutine as a subtour on which the agent travels from r to the start of the path and then follows the path back to r' could be clearer. The COP subroutine returns a path ending at r; the subtour construction (r → path start → ... → r) should be stated more explicitly, as the cost analysis depends on this interpretation.
  3. Algorithm 2, line 3: The threshold t_1 = P/((9+3ε)·(6+6ε)·D) uses D, the estimate of OPT cost. It would improve readability to remind the reader here that D is the guessed value satisfying cost(π*) ≤ D ≤ (1+ε)·cost(π*).
  4. Lemma 17: The bound uses (1+ε)·D/2^i as an upper bound on the QTSP path length, citing the guess j = ⌈log_{1+ε}(D/2^i)⌉. The inequality (1+ε)^j ≤ (1+ε)·D/2^i should be stated explicitly to make this step self-contained.
  5. The paper uses both 'ϵ' (U+03F5) and 'ε' (U+03B5) for epsilon. Standardizing on one would improve presentation.
  6. Reference [8] is cited both for Chaudhuri et al. (paths, trees, minimum latency tours) and for Christofides' algorithm (Appendix B.2). These are different works; the Christofides reference should be corrected.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the careful reading and the constructive feedback. The referee correctly identifies that the three major comments all concern proofs or proof sketches that are currently stated too informally in the manuscript. We agree with all three points and will revise the manuscript accordingly in the next version.

read point-by-point responses

  1. Referee: Section 3, Theorem 6: The extension of the k-TSP / QTSP approximation from unit weights to polynomially bounded integer weights is stated informally: 'their arguments carry over to this generalized case.' ... Either a formal reduction (e.g., node-splitting with polynomial blowup) should be given explicitly, or a proof sketch showing why the Chaudhuri et al. analysis extends to integer weights should be provided.

    Authors: The referee is correct that the current treatment of Theorem 6 is insufficient. The statement 'their arguments carry over to this generalized case' is not an adequate justification for a load-bearing subroutine claim. In the revised manuscript, we will provide a formal proof. Our plan is to give an explicit reduction via node-splitting: each vertex v with weight w(v) is replaced by w(v) copies at distance 0 from each other, and the unit-weight algorithm of Chaudhuri et al. is applied to the resulting instance. We will verify that this reduction preserves the approximation guarantee and incurs only polynomial blowup when weights are polynomially bounded integers. We will also include a proof sketch showing directly why the key steps of the Chaudhuri et al. analysis (in particular, the pruning and density arguments) extend to integer weights, so that the result rests on a verified argument rather than an assertion of confidence. revision: yes

  2. Referee: Appendix B.2, Lemma 30: The stronger lower bound cost(π*) ≥ (1/2)·D_{π*}(0) + Σ 2^{i-1}·D_{π*}(T_i+1) is stated but not proven—the authors write 'One can get an intuition why also this stronger lemma is true by having a look at Figure 2' and then say 'The exact details of this analysis are omitted.' This lemma is load-bearing for the claim that the approximation factor does not increase when requiring all locations to be visited.

    Authors: The referee is right to flag this. Lemma 30 is load-bearing for the 'visit every location' variant, and referring the reader to Figure 2 without a proof is not acceptable. In the revised manuscript, we will provide a full proof of Lemma 30. The proof extends the argument already given for Lemma 10: the additional term (1/2)·D_{π*}(0) accounts for the portion of the optimum tour traveled at weight 0 (i.e., before any item is collected), which is lower-bounded by half the total tour length times f(0). We will make the decomposition into weight-0 and weight-positive segments explicit and show how the summation over intervals captures both contributions. We will also complete the subsequent analysis (the argument that the approximation factor does not increase) that currently ends with 'The exact details of this analysis are omitted.' revision: yes

  3. Referee: Appendix B.4: The analysis for multiple items at the starting location (when items must be collected at the beginning) involves guessing a set S' from O(l²) candidates and arguing that p(S') ≥ p(S*)/2. The argument is sketched but not fully formalized—several steps ('we are then also guaranteed to collect a 1/(9+3ε) fraction') rely on the main CTTP analysis carrying through with modified f and w_max, but the interaction between the guessed initial weight and the interval decomposition (T_i thresholds) is not verified.

    Authors: We agree that the argument in Appendix B.4 is not sufficiently formalized. The key issue the referee identifies—the interaction between the guessed initial weight w(S') and the interval decomposition thresholds T_i—is a real subtlety that the current sketch does not address. When the agent starts with initial weight w(S'), the function f is effectively shifted, and the thresholds T_i (which are defined relative to f starting from 0) must be recomputed for the modified function. In the revised manuscript, we will provide a detailed treatment: (1) we will explicitly define the modified inverse-speed function f'(W) = f(W + w(S')) and the corresponding recomputed thresholds T_i', (2) we will verify that the interval decomposition and the efficiency definitions carry over to this shifted setting, and (3) we will confirm that the COP and MPBKP subroutines interact correctly with the shifted thresholds, so that the 1/(9+3ε) profit guarantee and the (9+9ε)D cost bound both hold. We will also formalize the argument that p(S') ≥ p(S*)/2 for the guessed set, including the case analysis over the candidate sets S_{i,j}. revision: yes

Circularity Check

2 steps flagged · score 2.0 of 10

No significant circularity; one minor self-citation that is not load-bearing for the main approximation guarantees.

  1. self citation load bearing [Section 1, paragraph on Weighted TSP background; reference [11]]
    "For more general cost functions, Eube et al. [11] show that the Weighted TSP is polynomially solvable if the metric is a line metric. They also show that the problem is NP-hard even on metrics induced by star-graphs and provide an 8-approximation for this case."

    Reference [11] (Eube et al., arXiv:2604.19271) shares authors with the present paper and is cited for background on Weighted TSP hardness and special-case approximations. However, this citation is not load-bearing for the paper's main results (Theorems 1, 2, 25). The (9+ε,9+ε) bicriteria approximation for Cttp relies on externally-authored subroutines: the k-TSP approximation from Arora and Karakostas [1], the capacitated orienteering approximation from Bock and Sanità [4], and the multiperiod binary knapsack FPTAS from Gao et al. [12]. The doubling framework is adapted from Epstein et al. [10]. The self-citation [11] provides motivational context only. The main derivation chain is parameter-free: D is guessed by enumeration over powers of (1+ε), thresholds t1 and t2 are derived from P and

  2. fitted input called prediction [Section 3, Theorem 6 and surrounding text (Qtsp subroutine extension from [8])]
    "While Chaudhuri et al. do not address the case in which vertices have integer weights, their arguments carry over to this generalized case. Thus, one can also calculate for a desired weight W* a tree containing r that spans weight at least W* with length bounded by (1+ε) times the length of any path ending in r that picks up weight at least W*."

    This is a correctness risk (an unverified claim that a cited algorithm extends to a new setting), not a circularity. The subroutine from Chaudhuri et al. [8] is externally authored and independently verified. The claim that 'their arguments carry over' is a mathematical assertion about the generality of an existing result, not a self-citation that defines the output in terms of the input. If the extension fails, the result is unsupported but not circular.

full rationale

The paper's derivation chain is self-contained against external benchmarks. The main approximation guarantees (Theorems 1, 2, 25) are derived from externally-authored, independently-published subroutines: k-TSP from Arora-Karakostas [1], capacitated orienteering from Bock-Sanità [4], multiperiod binary knapsack from Gao et al. [12], and the doubling framework from Epstein et al. [10]. No parameters are fitted to data and then presented as predictions. The value D is estimated by standard enumeration over powers of (1+ε), not by circular fitting. The thresholds t1 and t2 are algebraically derived from P and D. The one self-citation [11] (Eube et al.) provides background context on Weighted TSP hardness but does not bear the load of any approximation guarantee in the paper. The informal extension of Theorem 6 from unit to integer weights is a correctness concern, not a circularity: it claims an existing external result generalizes, rather than defining a quantity in terms of itself. Score 2 reflects the minor self-citation that is not load-bearing.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

No new mathematical entities, particles, forces, or dimensions are introduced. The Constrained TTP (Cttp) is a new problem variant name but is a straightforward restriction of the bi-objective TTP to a profit-constrained setting, analogous to Quota TSP. The algorithms compose existing subroutines.

free parameters (2)
  • ε (epsilon) = arbitrary constant > 0
    Standard approximation parameter controlling the trade-off between solution quality and running time. Not fitted to data.
  • D (cost estimate) = guessed via enumeration of (1+ε)^j
    An estimate of the optimal tour cost, enumerated over polynomially many values. Not a fitted parameter in the empirical sense; standard technique.
assumptions (5)
  • domain assumption The k-TSP (2+ε)-approximation of Arora and Karakostas [1] extends from unit node weights to polynomially bounded integer node weights.
    Stated in Section 3 and Theorem 6: 'their arguments carry over to this generalized case.' This is load-bearing for the Qtsp subroutine used in Algorithm 1 and for Theorem 1.
  • domain assumption The Cop (3+ε)-approximation of Bock and Sanità [4] returns a path ending at r that can be used as a subtour with the stated cost bounds.
    Used in Section 5.1 (Algorithm 2). The cost analysis in Lemma 13 assumes the path can be traversed from r and back within the derived bounds.
  • domain assumption The Mpbkp FPTAS of Gao et al. [12] applies to the deadline-knapsack formulation in Definition 18.
    Used in Algorithm 3 for computing π(2). The authors note their definition 'differs from the one presented by Gao et al. but it is easy to verify that both definitions are equivalent.'
  • domain assumption The inverse speed function f is non-decreasing and takes integer weight arguments.
    Stated in Section 2. The interval decomposition (T_i values) and Lemma 4's summation over integer weights depend on this.
  • domain assumption All item weights are polynomially bounded integers.
    Stated in Theorems 1 and 2. Required for the binary search in Algorithm 1 and the polynomial running time of the Qtsp subroutine.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Approximation Algorithms for the Traveling Thief Problem." pith.science (2026). https://pith.science/paper/6APEY7JC

@misc{pith2026260705164,
  author       = {Pith},
  title        = {Pith review of: Approximation Algorithms for the Traveling Thief Problem},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6APEY7JC}},
  note         = {Machine review of arXiv:2607.05164}
}
abstract

The Traveling Thief Problem (TTP) combines the Traveling Salesperson Problem with the Knapsack Problem. In this problem, a finite metric space is given, and at each location an item with some profit and weight is placed. An agent seeks to collect a subset of the items. To do so, the agent must decide which items to collect and to determine a cyclic tour visiting the corresponding locations. While collecting an item yields its profit as a reward, the agent's speed decreases as more weight is picked up. The problem involves two competing objectives: maximizing the total profit of the collected items and minimizing the travel time of the tour. While many heuristics and exact algorithms (with a non-polynomial running time) have been developed, no approximation algorithms are known for any variant of the TTP. We aim at computing an $(\alpha_1,\alpha_2)$-approximate Pareto set that, for every solution, contains another solution collecting at least a $\frac{1}{\alpha_1}$ fraction of its profit while requiring at most $\alpha_2$ times its travel time. Our main result is an algorithm that calculates a $(9 + \epsilon,9 + \epsilon)$-approximate Pareto set in polynomial time. We also consider the setting in which the set of items to be collected is given in advance, so that the agent only has to compute a tour through the corresponding locations that minimizes the total travel time. This is the so-called Weighted TSP. For this setting, we present a $(2e + \epsilon)$-approximation algorithm.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

33 extracted references · 33 canonical work pages

  1. [1]

    International Conference on Integer Programming and Combinatorial Optimization , pages=

    Universal sequencing on a single machine , author=. International Conference on Integer Programming and Combinatorial Optimization , pages=. 2010 , organization=

    work page 2010

  2. [2]

    Birge and Varun Gupta , title =

    Zuguang Gao and John R. Birge and Varun Gupta , title =. Computer Science - Theory and Applications - 16th International Computer Science Symposium in Russia,

    work page

  3. [3]

    Handbook of Approximation Algorithms and Metaheuristics , pages=

    Prize collecting traveling salesman and related problems , author=. Handbook of Approximation Algorithms and Metaheuristics , pages=. 2018 , publisher=

    work page 2018

  4. [4]

    Operations Research Forum , volume=

    Worst-case analysis of a new heuristic for the travelling salesman problem , author=. Operations Research Forum , volume=. 2022 , organization=

    work page 2022

  5. [5]

    Improved algorithms for orienteering and related problems , year =

    Chekuri, Chandra and Korula, Nitish and P\'. Improved algorithms for orienteering and related problems , year =. ACM Transactions on Algorithms (TALG) , volume =

    work page

  6. [6]

    SIAM Journal on Computing , volume =

    Chen, Ke and Har-Peled, Sariel , title =. SIAM Journal on Computing , volume =. 2008 , doi =

    work page 2008

  7. [7]

    Approximation Algorithms for the Team Orienteering Problem

    Wenzheng Xu and Zichuan Xu and Jian Peng and Weifa Liang and Tang Liu and Xiaohua Jia and Das, \ Sajal K.\. Approximation Algorithms for the Team Orienteering Problem. Proceedings - IEEE INFOCOM. 2020. doi:10.1109/INFOCOM41043.2020.9155343

    work page doi:10.1109/infocom41043.2020.9155343 2020

  8. [8]

    Discrete Applied Mathematics , volume=

    The capacitated orienteering problem , author=. Discrete Applied Mathematics , volume=. 2015 , publisher=

    work page 2015

Show all 33 references
  1. [9]

    Proceedings of the thirty-seventh annual ACM symposium on Theory of computing , pages=

    Saving an epsilon: a 2-approximation for the k-MST problem in graphs , author=. Proceedings of the thirty-seventh annual ACM symposium on Theory of computing , pages=

  2. [10]

    44th Annual IEEE Symposium on Foundations of Computer Science, 2003

    Paths, trees, and minimum latency tours , author=. 44th Annual IEEE Symposium on Foundations of Computer Science, 2003. Proceedings. , pages=. 2003 , organization=

    2003

  3. [11]

    A 2+ approximation algorithm for the k-

    Arora, Sanjeev and Karakostas, George , journal=. A 2+ approximation algorithm for the k-. 2006 , publisher=

    2006

  4. [12]

    International Conference on Evolutionary Multi-Criterion Optimization , pages=

    Solving the bi-objective traveling thief problem with multi-objective evolutionary algorithms , author=. International Conference on Evolutionary Multi-Criterion Optimization , pages=. 2017 , organization=

    2017

  5. [13]

    2022 , doi =

    A weighted-sum method for solving the bi-objective traveling thief problem , journal =. 2022 , doi =

    2022

  6. [14]

    Proceedings of the Genetic and Evolutionary Computation Conference,

    Hayden Faulkner and Sergey Polyakovskiy and Tom Schultz and Markus Wagner , title =. Proceedings of the Genetic and Evolutionary Computation Conference,. 2015 , url =. doi:10.1145/2739480.2754716 , timestamp =

    2015 doi

  7. [15]

    Genetic and Evolutionary Computation Conference,

    Sergey Polyakovskiy and Mohammad Reza Bonyadi and Markus Wagner and Zbigniew Michalewicz and Frank Neumann , title =. Genetic and Evolutionary Computation Conference,. 2014 , url =. doi:10.1145/2576768.2598249 , timestamp =

    2014 doi

  8. [16]

    Przybylek and Adam Wierzbicki and Zbigniew Michalewicz , title =

    Michal R. Przybylek and Adam Wierzbicki and Zbigniew Michalewicz , title =. Evol. Comput. , volume =

  9. [17]

    Sergey Polyakovskiy and Alexander Makarowsky and Rym M'Hallah , title =

  10. [18]

    Swarm Evol

    Daniel Herring and Michael Kirley and Xin Yao , title =. Swarm Evol. Comput. , volume =

  11. [19]

    Papadimitriou and Mihalis Yannakakis , title =

    Christos H. Papadimitriou and Mihalis Yannakakis , title =. Math. Oper. Res. , volume =

  12. [20]

    Pulleyblank and Prabhakar Raghavan and Madhu Sudan , title =

    Avrim Blum and Prasad Chalasani and Don Coppersmith and William R. Pulleyblank and Prabhakar Raghavan and Madhu Sudan , title =

  13. [21]

    Adel Nikfarjam and Aneta Neumann and Frank Neumann , title =

  14. [22]

    Optimization in Industry , pages =

    Mohammad Reza Bonyadi and Zbigniew Michalewicz and Markus Wagner and Frank Neumann , title =. Optimization in Industry , pages =

  15. [23]

    Frank Neumann and Sergey Polyakovskiy and Martin Skutella and Leen Stougie and Junhua Wu , title =

  16. [24]

    Sergey Polyakovskiy and Frank Neumann , title =. Eur. J. Oper. Res. , volume =

  17. [25]

    Genetic and Evolutionary Computation Conference,

    Jakob Bossek and Katrin Casel and Pascal Kerschke and Frank Neumann , title =. Genetic and Evolutionary Computation Conference,

  18. [26]

    Jakob Bossek and Aneta Neumann and Frank Neumann , title =

  19. [27]

    Proceedings of the Genetic and Evolutionary Computation Conference Companion , pages=

    Multi-objectiveness in the single-objective traveling thief problem , author=. Proceedings of the Genetic and Evolutionary Computation Conference Companion , pages=

  20. [28]

    Proceedings of the genetic and evolutionary computation conference , pages=

    Evolutionary computation plus dynamic programming for the bi-objective travelling thief problem , author=. Proceedings of the genetic and evolutionary computation conference , pages=

  21. [29]

    M. R. 2013 IEEE Congress on Evolutionary Computation , title=. 2013 , volume=. doi:10.1109/CEC.2013.6557681 , ISSN=

    2013 doi

  22. [30]

    2007 , publisher=

    Combinatorial problems and exercises , author=. 2007 , publisher=

    2007

  23. [31]

    M. R. Garey and David S. Johnson , title =. 1979 , isbn =

    1979

  24. [32]

    Ibarra and Chul E

    Oscar H. Ibarra and Chul E. Kim , title =. J

  25. [33]

    2026 , eprint=

    Effective Traveling for Metric Instances of the Traveling Thief Problem , author=. 2026 , eprint=

    2026

Pith tools

Reviewed July 8, 2026 · model on record in the stance chip above.