Qyxo-Zhaenivoth v1.0.0: prior-art boundary for a dynamic exact support engine
Research target: a major breakthrough. Review conclusion: the proposed dynamic engine can be a substantial improvement over rescanning on sparse-update workloads, but the sources and mechanism examined do not support calling it a major mathematical breakthrough or superiority over all algorithms.
This review is independent of the runtime implementation. Sources were searched and inspected on 2026-10-03. No benchmark result is asserted here.
Mechanism examined
The problem is a robust normalized weighted average with fixed value intervals and changing nonnegative interval weights. For one upper-value coordinate, write its fixed values as x_i, lower weights as L_i, upper weights as U_i, and delta_i = U_i - L_i. Assume every feasible denominator is positive. Define
[ A_0=N_0+\sum_i L_i x_i,\qquad B_0=D_0+\sum_iL_i. ]
The maximum threshold support is
[ g_+(T)=A_0-TB_0+\sum_i\delta_i(x_i-T)_+. ]
After sorting x_i, the last term is the weighted-value suffix sum minus T times its weight suffix sum. A balanced ordered tree or Fenwick tree that stores the two additive summaries (delta_i, delta_i x_i), plus global (A_0,B_0), therefore supports an exact sign query with O(log M) arithmetic operations. Point reweighting changes those summaries and globals along O(log M) locations. Multiple coordinates require their own sorted index maps; a shared point-weight change costs O(P log M).
For a lower-value coordinate, the corresponding minimum support is
[ g_-(T)=A_0-TB_0+\sum_i\delta_i\min(x_i-T,0), ]
using a prefix instead of a suffix. These are weighted hinge sums, not a new algebraic operation. Binary64 inputs can be represented as dyadic integers so neither tree aggregation nor comparison must rely on an uncertified floating-point reduction.
The root of each support gives the corresponding sharp endpoint. Because any feasible denominator is at least d_min > 0, for T_2 > T_1 both supports decrease by at least d_min(T_2-T_1). Each therefore has one root. A tree descent can compare support at the splitting value while carrying the already resolved prefix and suffix aggregates. It chooses one child at each level and obtains the active affine root at the final interval. This uses O(log M) node visits rather than invoking an O(log M) tree query at each step of a separate O(log M) binary search.
This derivation is the reviewer's mathematical assessment of the proposed data structure; it is not attributed verbatim to one paper. Arithmetic-operation bounds are not constant-bit-cost running-time bounds. Integer operand length includes binary64 exponent range, products, and log M bits from summation.
Primary sources and implications
1. Interval weighted averages and switching structure already exist
Karnik and Mendel, Centroid of a type-2 fuzzy set, Information Sciences 132 (2001), 195–220.
- DOI: https://doi.org/10.1016/S0020-0255(01)00069-X
- Author-hosted paper: https://sipi.usc.edu/~mendel/publications/Karnik%26Mendel%20IS%202001.pdf
The primary PDF was accessible but its text extraction was corrupted. Its established switching structure is independently visible in the official MathWorks equations below. The robust average is the same finite optimization object after the application-specific labels are removed; fuzzy membership bounds correspond to interval weights. An explicit fixed numerator and denominator contribution does not by itself create a new optimization principle.
2. Fractional optimization and root-finding equivalence already exist
Tufan Kumbasar, Revisiting Karnik–Mendel Algorithms in the framework of Linear Fractional Programming, International Journal of Approximate Reasoning 82 (2017), 1–21.
- DOI: https://doi.org/10.1016/j.ijar.2016.11.019
- Author institution record and full abstract: https://research.itu.edu.tr/en/publications/revisiting-karnikmendel-algorithms-in-the-framework-of-linear-fra/
The inspected institutional abstract explicitly links KM switching solutions to linear fractional programming, Charnes–Cooper, Dinkelbach, and discrete root finding. Consequently, describing the centroid as the zero of threshold support does not establish a new mathematical principle.
3. Binary switch search already exists
Xianliang Liu, Yicheng Lin, and Shu-Ping Wan, New efficient algorithms for the centroid of an interval type-2 fuzzy set, Information Sciences 570 (2021), 1–19.
- DOI: https://doi.org/10.1016/j.ins.2021.04.032
- Publisher article: https://www.sciencedirect.com/science/article/pii/S0020025521003595
Publisher search extracts report two binary algorithms, proof of optimal outputs, and complexity comparisons with KM, EKM, and EIASC. Direct full-page retrieval was blocked with HTTP 403, so the exact pseudocode and complexity theorem were not independently inspected in full. This is enough to reject novelty of binary switch search as a general claim. It is not enough to claim that this paper contains the exact proposed mutable-tree implementation. That distinction remains open.
4. Direct switch computation already has published proofs and public software
Chao Chen, Robert John, Jamie Twycross, and Jonathan Garibaldi, A Direct Approach for Determining the Switch Points in the Karnik–Mendel Algorithm, IEEE Transactions on Fuzzy Systems 26 (2018), 1079–1085.
- DOI: https://doi.org/10.1109/TFUZZ.2017.2699168
- Institutional publication record: https://nottingham-repository.worktribe.com/output/929379/a-direct-approach-for-determining-the-switch-points-in-the-karnik-mendel-algorithm
- Public author-maintained R interface: https://search.r-project.org/CRAN/refmans/FuzzyR/html/km.da.html
- Related comment DOI: https://doi.org/10.1109/TFUZZ.2018.2865134
FuzzyR km.da takes lower and upper weights and primary values, supports already sorted input, and can return the selected switch point or membership choices. This makes it a serious published endpoint comparator. It is floating-point R software and is not an exact sparse-update data structure; that is a legitimate contract difference, not evidence that direct switch computation is new.
5. Dynamic additive prefix summaries already exist
Peter M. Fenwick, A new data structure for cumulative frequency tables, Software: Practice and Experience 24 (1994), 327–336.
- DOI: https://doi.org/10.1002/spe.4380240306
- Publisher page and inspected abstract: https://onlinelibrary.wiley.com/doi/10.1002/spe.4380240306
The original abstract describes binary indexed trees with operations constant or logarithmic in table size. Applying the structure to two or four exact additive moments is ordinary augmentation. Tree topology, logarithmic point updates, and logarithmic prefix queries must be attributed to established data structures.
Fenwick also published A New Data Structure for Cumulative Probability Tables: An Improved Frequency-to-Symbol Algorithm, Software: Practice and Experience 26 (1996), 489–490.
- Publisher record: https://onlinelibrary.wiley.com/doi/abs/10.1002/%28SICI%291097-024X%28199604%2926%3A4%3C489%3A%3AAID-SPE22%3E3.0.CO%3B2-S
Inverse cumulative search is historical prior art for tree descent. The proposed support-at-split comparison differs from a fixed cumulative-mass threshold, but exploiting carried subtree sums to select one child is an inherited algorithmic pattern. I did not inspect this two-page paper in full.
6. Current mainstream methods are an active and optimized family
Official MathWorks Type-2 Fuzzy Inference Systems documentation (retrieved version R2026b):
It gives lower/upper centroid equations with opposite upper/lower weight choices around a switch point, and implements KM, EKM, IASC, and EIASC, as well as custom reducers. Its statement about approximating continuous fuzzy-set centroids should not be copied as impossibility of exact finite discrete arithmetic. Qyxo's exactness concerns finite supplied binary64 numbers; it does not make an unknown continuous membership function exact.
Yang Chen, Design of weighted based divided-search enhanced Karnik–Mendel algorithms for type reduction of general type-2 fuzzy logic systems, Complex & Intelligent Systems 11 (2025), 182.
- DOI and inspected full open-access article: https://doi.org/10.1007/s40747-025-01798-9
- https://link.springer.com/article/10.1007/s40747-025-01798-9
This recent paper subdivides switch-search spaces and compares weighted enhanced approaches. Its general type-2 quadrature setting is broader than a finite robust predicate, so it is not a like-for-like exact dynamic implementation. It demonstrates why plain original KM should not be presented as the strongest contemporary algorithmic family.
7. Serious public implementation choices
PyIT2FLS is a maintained open-source Python toolkit, with KM, EKM, WEKM, TWEKM, and EIASC interfaces documented for an n-by-4 interval array. Its repository also advertises a faster typereduction package. This review did not retrieve or execute its optimized backend, so no relative runtime claim is supported.
- Official repository: https://github.com/Haghrah/PyIT2FLS
- Official algorithm interface documentation: https://pyit2fls.readthedocs.io/en/latest/algorithms.html
- Author paper: https://arxiv.org/abs/1909.10051
- Published SoftwareX paper DOI: https://doi.org/10.1016/j.softx.2025.102146
FuzzyR 2.3.2 is another current public implementation, with direct-approach code and a documented sorted-input option:
- https://cran.r-project.org/web/packages/FuzzyR/FuzzyR.pdf
- https://search.r-project.org/CRAN/refmans/FuzzyR/html/km.da.html
These packages produce floating-point endpoints and accept broad fuzzy-system workloads. Qyxo's predicate should be compared to equivalent endpoint-derived decisions; approximate crisp defuzzifiers such as NT are not equivalent exact certificates.
Component-by-component novelty boundary
| Component | Finding | Allowed claim |
|---|---|---|
| Robust interval-weight average and sharp switching endpoints | Established KM/type-reduction problem | Application mapping and contract specialization |
| Threshold support and monotone root | Established fractional/root-finding connection | Explicit exact implementation of the connection |
| Binary switch search | Published at least by 2021 | No standalone novelty claim |
| Point update/prefix-suffix tree sums | Established Fenwick/augmented tree operations | Integration for this workload |
| One-path support-dependent endpoint descent | Specific formula not found in inspected sources; general descent pattern inherited | Derived implementation feature, novelty unestablished |
| Exact dyadic accumulators throughout dynamic updates | No equivalent specific implementation retrieved | Strong useful capability if independently verified; absence of a match is not proof of invention |
| Optional exact fallback filter | Established robust-predicate technique | Engineering/contract contribution |
| Stable API, witnesses, CLI, qualification | Practical release engineering | Adoption contribution |
Introduced in this release: a concrete maintained-state engine, exact arithmetic contract, application API, tested integration, and reproducible workload evidence, subject to the implementation and tests actually completing.
Inherited: linear-fractional robust optimization, switching extrema, hinge support, binary search, augmented prefix-sum structures, and exact/filtered numerical predicates.
Not established: first invention of the dynamic composition, a new general-purpose computational principle, or a complexity improvement over every competent existing tree implementation.
Required strong comparison
Benchmark at least four alternatives on exactly the same mutation/query sequence:
- Existing v0.2.1 exact direct-support rescanning, keeping static preparation out of the hot path fairly.
- Existing v0.2.1 NumPy filter with exact fallback, including actual sparse changes to its required complete current weight arrays and all validation overhead.
- An independently implemented exact Fenwick/tree threshold engine using the same fixed sorted values and the same dyadic arithmetic contract. It has O(log M) point updates and O(log M) threshold queries. This comparator separates useful dynamic acceleration from a new data-structure claim.
- Exact sorted prefix endpoint rebuild plus binary switch search, with sorted values reused. Also compare to a known floating-point EKM/EIASC or FuzzyR direct reducer where the additional language/dependency cost is acceptable, clearly labeling its different numerical contract.
Time preparation, update, query, optional endpoint computation, witness materialization, and end-to-end totals separately. Vary changed rows k, query count Q, dimensions P, M, ties, exponent range, and exact equality. Sparse updating can plausibly win by a large factor over O(MP) scans; changing all rows forces O(MP) input work and can favor vectorization. Witness output requires O(MP) output size and cannot be claimed logarithmic. A static unchanged-box cached enclosure can answer threshold queries in O(P), and should remain superior when no weights change.
The strongest defensible prospective claim is: after fixed-value preparation, exact robust threshold decisions can be maintained through k sparse weight changes using O(kP log M) update operations and O(P log M) query operations, while preserving the supplied-number contract. This is a derived algorithmic capability, not yet a measured speedup or an assertion of new mathematical priority.
Limits of the prior-art search
Queries covered dynamic interval weighted averages, Karnik–Mendel with segment/Fenwick trees, incremental type reduction, binary switch algorithms, fractional-programming equivalence, direct switch calculation, and current public implementations. No directly matching exact mutable-tree implementation was found. Full retrieval of the 2021 binary paper and Nottingham institutional full text was blocked, some PDF extraction was poor, and optimized public backends were not executed. These gaps must remain visible. A negative search cannot certify novelty.
The major-breakthrough target survives only as a research target. It should not become a release label without both stronger prior-art separation and decisive empirical evidence against optimized peers on economically meaningful workloads.