{"id":"3896f3d7-91bb-4dab-a5ef-5a79b468c801","arxiv_id":"2501.14775","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":3.0,"correctness_risk":"high","formal_verification":"none","parameter_count":10,"one_line_summary":"A hybrid of the Firefly Algorithm and Genetic Algorithm is tested on benchmarks, design problems, and 0-1 knapsack instances, with several reported results matching known optima.","lead":"This paper combines the Firefly Algorithm and a Genetic Algorithm into one hybrid optimizer and applies it to math benchmark functions, engineering design tasks, and 0-1 knapsack problems. The authors report competitive or optimal results on many test cases, though the paper contains internal data inconsistencies.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. 6.2.4 multiplies the penalty by profit, so overweight high-profit solutions receive the highest fitness; this inverts constraint handling and undercuts all SKP/MKP results.","rationale":"The reader's weakest assumption pinpoints Eq. 6.2.4, and that is also the most load-bearing issue I find: it is not a minor typo but a sign error in the core objective for the knapsack domain. If the printed equation is executed, the optimizer maximizes an infeasibility reward rather than penalizing it; if the executed code silently corrects the equation, then the paper's formal description is unreliable and the reported tables cannot be independently checked. Either disjunct undermines the paper's own evidence for its central claim. I considered the Table 7/Table 19 inconsistencies and the missing code and parameters; they add to the lack of trustworthiness but do not identify a single decisive flaw as directly as Eq. 6.2.4. The proposed test is deliberately narrow: it checks whether the published formulation actually produces feasible solutions on the smallest listed instance, which is the minimal condition for the knapsack results to be meaningful. No ad hominem is intended; the concern is about the printed mathematics and its consequences for the reproducibility of the stated results.","tokens_in":35160,"tokens_out":3259,"duration_ms":31813,"concrete_test":"Re-implement the exact printed FAGA for the Appendix f1 instance (N=10, W=269) using Eq. 6.2.4 as written. Run 30 trials and record the best fitness found. If the optimizer returns a solution with Σw_i x_i > 269 (or no feasible solution is ever selected), the printed penalty is confirmed to be non-penalizing. Then re-run with the corrected additive penalty f(x)=Σ p_i x_i − θ·max(0, Σ w_i x_i − W) and compare the best/mean values to Table 16; if the values shift, the published SKP tables do not reflect the described algorithm.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that FAGA solves 0-1 knapsack problems to optimality rests on the fitness function in Eq. 6.2.4: f(x) = Σ p_i x_i × (θ × max(0, Σ w_i x_i − W)). As printed, the penalty is a factor multiplying profit, not an additive subtraction. For any overweight solution the term θ×max(...) is positive, so the fitness is profit times that positive number; a high-profit overweight selection will always score higher than a feasible selection with the same profit profile, and the algorithm is actively driven to violate capacity. The surrounding threshold rule (Eqs. 6.2.2-6.2.3) also conflates the scalar normalized fitness with the binary item variables, so it is unclear how item selections are actually formed. Since no code or parameter set is provided, there is no way to determine whether the implementation followed Eq. 6.2.4 or some corrected version; either way, the published equations do not support Tables 16-20, and those tables are a core part of the claimed general-purpose superiority. This is the most load-bearing weak point because it is a mathematical error in the one part of the paper that defines the evaluation of every knapsack candidate solution.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes FAGA, a hybrid of the Firefly Algorithm and the Genetic Algorithm, and claims it solves unconstrained benchmark functions, five engineering design problems, and single and multidimensional 0-1 knapsack problems with better or competitive accuracy and computational efficiency compared with standalone FA, GA, and several published binary optimizers. The manuscript presents the algorithm structure, a static penalty method, numerical tables comparing FAGA with other methods over 30 runs, and convergence plots. The central claim is that FAGA is a general-purpose constrained optimizer that outperforms or matches existing algorithms on the tested problems.","tokens_in":35348,"tokens_out":4374,"duration_ms":42188,"significance":"If the reported results were reliable, the paper would offer a modest but useful engineering contribution: a hybrid metaheuristic that combines FA's attraction-based search with GA's crossover and mutation, tested on standard benchmarks and knapsack instances. The paper reports many comparisons against external known optima and includes 30-run statistics, which is a reasonable experimental discipline. However, the significance is substantially undercut by a load-bearing error in the knapsack fitness function, by inconsistencies in the reported tables, and by a lack of reproducibility (no code, no exact parameter values). As printed, the paper does not provide a trustworthy basis for its main performance claims on knapsack problems.","major_comments":[{"comment":"The fitness function as printed is f(x) = Σ p_i x_i × (θ × max(0, Σ w_i x_i − W)). This is not a penalty; it is a multiplicative factor. For any feasible solution (Σ w_i x_i ≤ W), the max term is zero, so the entire fitness is zero. For any overweight solution, the factor is positive, so the fitness equals the total profit times a positive number. Consequently, every feasible solution has identical fitness 0, and an overweight solution with higher profit receives higher fitness than any feasible solution. This inverts constraint handling and would actively drive the search toward capacity violations. Since Tables 16, 17, and 18 report near-optimal results for SKP and MKP, those results cannot be supported by the published equation. The authors must provide the correct (presumably subtractive) penalty formulation and re-run the experiments, or clarify whether the implementation differs from Eq. (6.2.4).","section":"Section 6.2, Eq. (6.2.4)"},{"comment":"The binarization procedure is conceptually unclear and unreproducible. Equation (6.2.2) normalizes a scalar fitness f(x_i), but Eq. (6.2.3) then writes x_i = 0 if x_i < 0.5 and x_i = 1 otherwise, as if the scalar fitness value were the binary decision variable. The text in Step 2 also refers to sorting items by profit-to-weight ratio and then thresholding 'each item's value,' which conflates item-level sorting and value normalization with individual-level fitness normalization. Furthermore, the firefly movement in Step 4 uses the continuous-position update of Eq. (2.2), while each individual is defined as a binary vector; no binarization rule for the updated continuous position is given. As a result, a reader cannot determine how binary solutions are actually formed or how the reported tables were produced.","section":"Section 6.2, Eqs. (6.2.2)-(6.2.3) and Step 4"},{"comment":"The statistical results for the pressure vessel problem are internally inconsistent. In Table 7, the FA column reports Best 6090.92 but Mean 2.66, Worst 2.67, and Std. Dev 0.194e-2, which are clearly copied from the helical spring results in Table 5. The text then states that FAGA has 'slightly higher average CPU time compared to GA,' yet Table 7 lists FAGA average CPU time as 77.991 s versus GA's 14.206 s, which is more than five times larger. Additionally, Table 8 reports FAGA function evaluations as 162,460 while Table 7 reports average function evaluations as 183,079 for the same problem. These inconsistencies call into question the accuracy of the experimental tables and the computational-efficiency claims.","section":"Table 7 and Table 8"},{"comment":"The claim that FAGA 'consistently outperforms most algorithms' is not supported by Table 3 for the Rastrigin function. For f4, HFA reports a mean of 3.39E-08 and FAGA reports 9.42E-01, so FAGA is roughly seven orders of magnitude worse. The text mentions that HFA 'at times delivers better results,' but the conclusion and abstract still state that FAGA delivers improved accuracy and outperforms conventional algorithms. This overstates the evidence and should be corrected with a precise, per-function summary of where FAGA is better, tied, or worse.","section":"Table 3 and Section 7.1"},{"comment":"The multidimensional knapsack problem is never formally defined. Section 6.1 gives only the single-constraint formulation (6.1.1)-(6.1.2), and Section 7.3.2 directly applies FAGA to the WEISH dataset without stating the multi-constraint equations, the number of constraints per instance, or how the penalty function (5.4) is adapted to multiple constraints. Since the paper's title and central claim concern multidimensional 0-1 knapsack, the missing formulation is a significant omission. The authors should present the full MKP model and explain how FAGA handles multiple capacity constraints.","section":"Section 6.1 and Section 7.3.2"}],"minor_comments":[{"comment":"The sentence 'if the item's value after sorting is less than 0.5, it is set to 0' is confusing because 'value' is not defined; presumably this refers to the normalized continuous variable, but the connection to the fitness normalization in Eq. (6.2.2) is unclear.","section":"Section 6.2, Step 2"},{"comment":"The flowchart in Fig. 4 says 'If i < j, move firefly i towards j,' but the comparison should be based on fitness values (e.g., f_i < f_j), not on indices. This appears to be a typo but makes the flowchart formally wrong.","section":"Section 6.2, Step 4 and Fig. 4"},{"comment":"The column header 'Optimal Solution f(v)' is ambiguous: for several entries the listed values are the algorithms' best found values, not the known true optimum. For example, Weish22 lists MS 8790 and FAGA 8947, where 8947 is the known optimum, but other entries such as Weish03 list MS 4106 and FAGA 4115. Please label the column as 'Best f(v)' and separately state the true optimum for each instance.","section":"Table 20"},{"comment":"The manuscript contains numerous typos, erroneous equation numbers, and duplicated table entries (e.g., Table 7's FA mean/worst/std as noted above). A careful proofread and a consistency check across tables are needed before any resubmission.","section":"Throughout"},{"comment":"No exact parameter values are reported for alpha, beta0, gamma, crossover rate, mutation rate, tournament size, sigma, or the penalty parameter theta; only ranges (e.g., crossover 60-90%) are given. Since the paper emphasizes solution quality and convergence speed, exact settings and a parameter-sensitivity study would be needed for reproducibility.","section":"Section 4"}],"recommendation":"reject","confidential_remarks":"The paper shows signs of careless data handling (e.g., Table 7 copying values from Table 5) and the central knapsack fitness equation is mathematically inverted, which alone invalidates the SKP/MKP experimental claims as printed. Even if the implementation used a corrected penalty, the absence of code and exact parameters makes verification impossible. In my view this goes beyond minor revision: the authors would need to correct the formulation, re-run all knapsack experiments, and carefully recheck every table. The journal's standards would be better served by a fresh submission after these corrections."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: this is a routine hybrid metaheuristic paper with one genuine scheduling change and a broad set of experiments, but the knapsack evaluation is undermined by a wrong penalty equation. The benchmark-function and design-optimization sections are less harmed, but they have their own carelessness.\n\nThe new piece is the replacement of the split-phase FAGA of Nand and Sharma (2019) with simultaneous integration of GA operators into every FA iteration. That is a legitimate, if incremental, scheduling change. The paper merits credit for the empirical workload: four continuous benchmarks, five engineering design problems, the f1-f20 SKP set, and the Weish MKP set, each run 30 times with standard deviation, function evaluation counts, and CPU times. That is real effort.\n\nThe soft spot is load-bearing. Equation (6.2.4) writes the penalty as a multiplicative factor: f(x) = (Σ p_i x_i) × (θ × max(0, Σ w_i x_i − W)). As printed, an overweight solution with high profit receives a larger score than a feasible one, so the algorithm is actively pushed to overfill the knapsack. The normalization-and-threshold step in Eqs. (6.2.2)-(6.2.3) is also conceptually confused: it normalizes the scalar fitness of the whole individual and then applies a 0.5 cutoff to each item bit, mixing solution-level and variable-level scales. There is no code or exact parameter set, so it is impossible to tell whether the implementation followed the printed equations or a corrected version. That makes Tables 16-20, the core knapsack evidence, untrustworthy as printed.\n\nOther issues are less fatal but real. Table 7 lists mean 2.66 and worst 2.67 for the pressure vessel problem; those are the helical spring values, a copy-paste slip. Table 19, meant to show the new FAGA beating the 2019 split-phase version, shows the old version with a mean closer to the optimum in at least Weish01 and Weish02; the text's claim that the new version is closer \"in most problems\" does not hold up against its own table. Parameter values are only given as ranges (crossover 60-90%, mutation 1-10%), and the paper does not report hardware or implementation details.\n\nWho is this for? Someone looking for an off-the-shelf binary optimizer might find the SKP/MKP tables interesting if they were reproducible, but they are not. The design-optimization tables are also sloppy. I would not cite this work in its current form. For peer review: I would send it to a referee rather than desk reject, because the experimental corpus is substantial and the mathematical errors are identifiable and potentially correctable. But the referee's verdict should clearly be major revision, not acceptance. The authors need to fix Eq. (6.2.4), clarify or abandon the threshold-binarization, release code with parameter settings, and correct the copied table. As it stands, the central knapsack claims are not supported.","headline":"Routine hybrid metaheuristic with a genuine scheduling tweak, but the knapsack penalty equation is wrong, the threshold binarization is confused, and tables have copy-paste errors; needs major revision.","tokens_in":35996,"tokens_out":4838,"would_cite":false,"duration_ms":45050,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A simultaneous firefly-genetic hybrid is presented as a general-purpose constrained optimizer that beats standalone FA and GA and matches published binary optimizers across benchmark, design, and knapsack problems.","keywords":["hybrid firefly-genetic algorithm","0-1 knapsack problem","multidimensional knapsack problem","constrained optimization","static penalty function","metaheuristics","benchmark functions","engineering design optimization"],"falsifier":"Re-implement the printed equations and compute Eq. (6.2.4) for every subset of a small instance such as f3 (4 items, capacity 20, weights 6,5,9,7, profits 9,11,13,15); if any overweight subset scores above the best feasible subset, the reported knapsack numbers cannot be produced by the printed fitness. The same re-implementation should recover the paper's f15 best of 2440, and if the true optimum is 2444 as the comparison table states, the claim that FAGA consistently reaches the optimum would already be settled.","tokens_in":34835,"feed_emoji":"🎒","tokens_out":9273,"duration_ms":83286,"temperature":0.7,"pith_summary":"This paper sets out to show that a hybrid running the Firefly Algorithm (FA) and the Genetic Algorithm (GA) at the same time, rather than one after the other, is a general-purpose solver for constrained optimization. The reported evidence spans four benchmark functions, five engineering design problems, twenty single-constraint 0-1 knapsack instances, and thirty multidimensional knapsack instances. If the evidence is correct, FAGA matches or beats standalone FA, standalone GA, and several published binary optimizers on solution quality, convergence speed, and function evaluations. The practical interest is that the hybrid is a simple combination of two well-known metaheuristics, so a working version would be an easy upgrade for logistics, resource-allocation, and design tasks.","feed_headline":"Firefly + genetic hybrid matches or beats published knapsack optima","feed_subtitle":"Concurrent FA-GA loop reaches optimal or near-optimal profits with fewer function evaluations than standalone parent algorithms.","key_machinery":"The central object is the simultaneous FAGA loop. In each iteration, fireflies are ranked by fitness and moved toward brighter solutions using the standard FA update $X_i^{\\mathrm{new}} = X_i + \\beta_0 e^{-\\gamma r_{ij}^2}(X_j - X_i) + \\alpha(\\mathrm{rand} - 0.5)$; then the updated population goes through tournament selection, crossover, and mutation (Gaussian for continuous problems, bit-flip for binary knapsack), and the worst individual is replaced if the mutated child is fitter. For constrained problems, a static penalty $P_F = \\theta(\\sum_i g_i(X) + \\sum_j h_j(X))$ is added to the objective. For the 0-1 knapsack, solutions are binary vectors; items are sorted by profit-to-weight ratio, a normalized fitness value is thresholded at 0.5 to set each item bit, and fitness is total profit minus a capacity-violation penalty. This machinery couples global search (FA attraction plus GA mutation) and local refinement (crossover, elitist replacement) in one loop.","core_discovery":"The central claim is that integrating FA's brightness-driven movement with GA's tournament selection, one-point crossover, and mutation in a single continuous loop yields a solver that consistently reaches optimal or near-optimal results. On the benchmark functions, the paper reports best values of $4.06\\times10^{-117}$ for Sphere, $1.27\\times10^{-16}$ for Ackley, $2.04\\times10^{-15}$ for Rosenbrock, and $0.845$ for Rastrigin. On the design problems, FAGA matches the best published objectives: spring volume $2.6586$, pressure-vessel cost $6059.71$, cantilever weight $1.3399$, gear-ratio error $2.7\\times10^{-12}$, and I-beam deflection $0.006625$. On the knapsack problems, FAGA matches the best reported profit in nineteen of the twenty single-knapsack cases and reaches the listed true optimum in all thirty WEISH multidimensional instances, with mean profits closer to the optimum than the earlier two-phase FAGA.","pith_inferences":["Our inference, not the paper's claim: if the gains come from synchrony rather than from the operators themselves, an ablation comparing FA-only, GA-only, sequential FA-then-GA, and simultaneous FAGA with identical evaluation budgets on the WEISH set would isolate the mechanism.","The binarization step that thresholds one normalized scalar fitness at 0.5 is not a standard way to map real-valued optimization to the 0-1 knapsack; a per-bit probability or a repair operator would be a more natural bridge, and testing whether the reported optima survive that change would show whether the result depends on the threshold rule.","If the penalty function is corrected and the results still hold, the same simultaneous FA-GA loop is a plausible template for other binary selection problems, such as feature selection or portfolio selection, where the paper's threshold trick would need replacement."],"forward_implications":["A working FAGA gives practitioners a single hybrid that can be applied to continuous, mixed-variable, and binary constrained problems without changing the core loop.","On the 20 single-knapsack instances, FAGA matches the best reported profit in 19 cases and handles large instances (up to 75 items) with runtimes that scale slowly enough for routine use.","On the 30 WEISH multidimensional knapsack instances, FAGA reaches the listed true optimum in every case, with average fitness closer to the optimum than the earlier sequential FAGA, suggesting the simultaneous coupling is the source of improvement.","On five engineering design problems, FAGA matches the best known objective values while using fewer function evaluations than standalone FA or GA, which matters when each evaluation is expensive.","Convergence curves show rapid early progress on unimodal benchmarks and steady progress on multimodal benchmarks, supporting the claim that the hybrid avoids premature convergence."],"supporting_citations":[{"why":"Supplies the Firefly Algorithm equations and attractiveness model that FAGA builds on.","marker":"Yang, 2010"},{"why":"Supplies the Genetic Algorithm selection, crossover, and mutation framework that FAGA merges with FA.","marker":"Holland, 1975"},{"why":"Defines the earlier two-phase FAGA that the proposed simultaneous version is compared against on multidimensional knapsack problems.","marker":"Nand and Sharma, 2019"},{"why":"Provides the 20 single-knapsack test cases $f_1$-$f_{20}$ used in Section 7.3.1.","marker":"Kulkarni and Shabir, 2016"},{"why":"Supplies the OR-Library WEISH benchmark datasets used for the multidimensional knapsack tests.","marker":"Beasley and John, 1990"},{"why":"Provides the BKOA baseline that FAGA is compared against on single-knapsack instances.","marker":"Abdel-Basset et al., 2023"},{"why":"Provides the BIWOA and BMMVO baselines compared on WEISH multidimensional knapsack instances.","marker":"Abdel-Basset et al., 2019"},{"why":"Provides the BHHA baseline used in the multidimensional knapsack comparisons.","marker":"Heidari et al., 2019"},{"why":"Frames the 0-1 knapsack formulation and combinatorial context referenced by the methodology section.","marker":"Martello and Toth, 1990"}],"fun_headline_variants":["Firefly + genetic hybrid hits optimal knapsack profits","Firefly-genetic hybrid matches best knapsack optima","Hybrid firefly-genetic solver achieves optimal knapsack profits","FAGA hybrid nails 30/30 multi-knapsack optima","Firefly + GA hybrid matches optimum in almost all knapsack cases"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the capacity penalty in Eq. (6.2.4) actually lowers the fitness of overweight solutions; as printed, the penalty is multiplied by total profit, so an overweight, high-profit solution could score higher than a feasible one, and separately the 0.5 threshold rule maps one normalized scalar onto many binary item choices.","fun_headline_variants_meta":{"raw":{"variants":["Firefly + genetic hybrid hits optimal knapsack profits","Firefly-genetic hybrid matches best knapsack optima","Hybrid firefly-genetic solver achieves optimal knapsack profits","FAGA hybrid nails 30/30 multi-knapsack optima","Firefly + GA hybrid matches optimum in almost all knapsack cases"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00072,"raw_usage":{"total_tokens":3204,"prompt_tokens":892,"completion_tokens":2312,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":508,"completion_tokens_details":{"reasoning_tokens":2235}},"tokens_in":508,"tokens_out":2312,"duration_ms":16454,"temperature":1.0,"reasoning_tokens":2235,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T22:53:06.404797+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-implement the printed equations and compute Eq. (6.2.4) for every subset of a small instance such as f3 (4 items, capacity 20, weights 6,5,9,7, profits 9,11,13,15); if any overweight subset scores above the best feasible subset, the reported knapsack numbers cannot be produced by the printed fitness. The same re-implementation should recover the paper's f15 best of 2440, and if the true optimum is 2444 as the comparison table states, the claim that FAGA consistently reaches the optimum would already be settled.","supporting_citations":[],"review_version":1}