Integrity Review Report: 'Optimization Method for Multi-Controller Deployment in Software-Defined Networks' (Guo & Hu, 2023)

Academic fraud report · Geng Detector

Summary

Verdict: Strong evidence of serious logical, mathematical, and statistical irregularities indicative of either fabrication or severely flawed scholarship. Key issues include (1) a nonsensical 'optimal' upper-bound formula of 1.15×n that contradicts the authors' own tabulated results (e.g., claiming 115 controllers for a 100-switch network), (2) a genetic-algorithm selection rule that rewards higher latency, inverting survival-of-the-fittest, (3) a dimensional mismatch in the objective function (1×c vector dotted with n×1 vector), (4) counter-intuitive monotonically increasing latency as controller count rises in Table 1, with suspiciously arithmetic-progression-like increments, and (5) explicit discarding of outliers to force-fit curves. Image duplication could not be assessed because only extracted text was provided. Confidence in the logical/mathematical findings is high; conclusions regarding intent (fraud vs. incompetence) cannot be definitively drawn from the text alone.

Verdict

🔴 Severe issues confirmed (text-based analysis). Multiple mathematical and logical contradictions render the central conclusions untenable. Image-level checks were not possible.

Key findings

Implausible scaling formula: The stated optimal interval upper bound 1.15 × n contradicts Table 2 (e.g., n=50 ⇒ claimed 57.5 controllers, n=100 ⇒ 115 controllers), which is physically nonsensical for a deployment optimization paper.
Inverted selection pressure in GA: Algorithm 1, line 10 defines selection probability proportional to l_sum, while the stated objective (Section 1.2.1) is to minimize l_sum. Higher-latency chromosomes are therefore preferred — a reverse-elimination mechanism.
Dimensional mismatch in Equation (6): x(i,:) is 1×c and D(:,i) is n×1; the element-wise product requires c=n, which contradicts the multi-subnet premise.
Counter-intuitive Table 1 trends: As c increases from 2 to 6, latency l increases monotonically across all three test networks, with suspiciously regular differences (e.g., 9.80, 4.90, 5.88) resembling arithmetic progressions rather than stochastic simulation outputs.
Curve-fitting by outlier removal: The authors openly state they "discarded outliers to smooth the two data groups into curves" without justifying why those points are invalid.
Image reuse: Not assessable from extracted text only; original PDF required for pixel-level inspection.

Evidence highlights

Table 2 (n=50 ⇒ c=3; n=100 ⇒ c=10; n=150 ⇒ c=13) vs. formula 1.15×n ⇒ {57.5, 115, 172.5}.
Table 1, Network 1 l values: 298.04 → 307.84 → 313.73 → 323.53 → 327.45; deltas include 9.80 appearing twice consecutively.
Algorithm 1 selection rule: P{select y} = l_y_sum^m / Σ(l_y_i_sum) — numerator scales with the metric to be minimized.
Equation (6): l_sum = Σ_{i=0}^{n-1} x(i,:) * D(:,i), with x ∈ {0,1}^{n×c}, D ∈ ℝ^{n×n}.

Notes

DOI: 10.12018/j.issn.2097-0730.20220831001
Source PDF: 软件定义网络多控制器部署优化方法_郭一平.pdf
Journal: Journal of Army Engineering University of PLA (陆军工程大学学报); received 2022-08-31.
Image-analysis findings (Figures 3, 4) require the original PDF; no pixel evidence is available in the extracted text.
Distinction between fabrication and severe incompetence cannot be resolved from the text alone; institutional investigation is advised.
All numeric values preserved exactly from the source report.