Snail Homing and Mating Search (SHMS)

• SHMS is a population-based metaheuristic algorithm inspired by snail homing, trail following, and mating behaviors for robust continuous optimization.
• It utilizes adaptive operators such as the love dart for step-size control and roulette wheel selection based on fecundity, balancing exploration and exploitation.
• Applied in engineering, SHMS has optimized stepped-transmission shaft weight by addressing fatigue, geometric, and deflection constraints effectively.

Searching arXiv for the specified SHMS papers to ground the article in the cited sources. Snail Homing and Mating Search (SHMS) is a population-based, bio-inspired metaheuristic algorithm derived from terrestrial snail behaviour, especially homing, trail following, fecundity-guided mate search, and the “love dart” mechanism. In the optimization literature, SHMS models candidate solutions as snails and organizing reference points as homes; search dynamics arise from neighbourhood sampling around homes, fecundity-based partner selection, adaptive step sizing through the love dart operator, and trail-following updates toward fecund snails. The method was introduced as a general continuous optimization algorithm and later applied to constrained engineering design, including the weight optimization of a stepped-transmission shaft under fatigue, geometric, and deflection constraints (Kulkarni et al., 2023, Saha et al., 17 Sep 2025).

1. Biological origin and algorithmic abstraction

SHMS is explicitly inspired by several behavioural traits of snails. The source description emphasizes homing or shelter seeking, in which snails search for homes with “better” environmental conditions such as cooler, more humid, and shadier locations, compare a new location to the current one, and either remain or retrace their path back. It also emphasizes mating search, where snails follow mucous trails carrying chemical markers and pheromones from fecund individuals, with preference for fresher and stronger trails. Some species additionally employ a calcareous “love dart,” whose stronger effects are associated biologically with more successful mating and more offspring (Saha et al., 17 Sep 2025).

In SHMS, these biological motifs are mapped onto canonical search operators. Homes correspond to regions in the decision space, and each home acts as a reference point around which a set of snails, i.e. candidate solutions, is generated. Mucous trail quality is represented by fitness and by a fecundity index, so that better snails are more attractive and exert greater influence on subsequent search. Trail following becomes a local exploitation mechanism around promising solutions, while the love dart becomes an adaptive step-size control that depends on the fitness difference between a snail and its selected fecund partner (Saha et al., 17 Sep 2025).

The earlier formulation also frames SHMS as a dual-level population system consisting of homes and snails per home, with biased movements toward better solutions and occasional reassignment across homes. In that interpretation, homing corresponds to exploitation near a good region, whereas following foreign trails and cues induces broader search and reallocation across promising regions (Kulkarni et al., 2023). This places SHMS in the same broad class as swarm- and population-based metaheuristics such as GA, PSO, and WOA, while distinguishing it through fecundity-based mating and homing dynamics (Saha et al., 17 Sep 2025).

2. Core mathematical structure

SHMS is defined for a general constrained minimization problem of the form

$$\begin{aligned} \min_{\mathbf{X}} \quad & f(\mathbf{X}) = f(X_1, X_2, \dots, X_N) \ \text{s.t.} \quad & g_i(\mathbf{X}) \le 0, \quad i = 1,\dots,n \ & h_j(\mathbf{X}) = 0, \quad j = 1,\dots,m \ & X_i^{L} \le X_i \le X_i^{U}, \quad i = 1,\dots,N . \end{aligned}$$

Within this formulation, $H$ denotes the number of homes, $S$ the number of snails per iteration, and $$\mathbf{X}_H = (\mathbf{X}^1, \dots, \mathbf{X}^h, \dots, \mathbf{X}^H)$$ the home positions (Saha et al., 17 Sep 2025).

For a home $h$, an interval $U_s$ is defined around $\mathbf{X}^h$ by

$U_s = \mathbf{X}^h + c,$

where $c$ is a sampling interval parameter controlling the neighbourhood around each home. Each associated snail is initialized randomly in this interval with position

$$\mathbf{X}^s = \big( X_1^s, X_2^s, \dots, X_N^s \big)^T,$$

and objective value $H$0 is then evaluated (Saha et al., 17 Sep 2025).

Constraint handling in the shaft study uses a Static Penalty Function (SPF). The penalty is given as

$H$1

with $H$2 the penalty parameter, and the pseudo-objective becomes

$H$3

For the stepped-transmission shaft case, the penalty factor is set to $H$4 (Saha et al., 17 Sep 2025).

The fecundity index encodes improvement over iterations and readiness to mate. In the shaft paper it is written as

$H$5

with random initialization in $H$6 for the first two iterations and random fallback when the expression is not well-defined (Saha et al., 17 Sep 2025). The earlier SHMS paper states the exact printed fecundity equation is somewhat unclear in the published text, but preserves the operational interpretation: positive fecundity indicates an improving, attractive snail; negative fecundity indicates deterioration (Kulkarni et al., 2023).

Mating-partner selection is probabilistic. One SHMS formulation gives

$H$7

paired with roulette wheel selection, while the earlier paper uses

$H$8

before normalization, also under roulette wheel selection (Saha et al., 17 Sep 2025, Kulkarni et al., 2023). The shaft paper notes that, for minimization, inverse fitness would be more intuitive, but keeps the stated form while conceptually treating better snails as occupying more roulette-wheel space (Saha et al., 17 Sep 2025).

The love dart operator is defined as

$H$9

with normalized form

$S$0

Position updates then follow

$S$1

$S$2

Homes are updated by keeping the best $S$3 solutions each iteration as new homes, an elitist homing strategy (Saha et al., 17 Sep 2025).

3. Search dynamics, selection pressure, and population organization

The operational cycle of SHMS consists of home and snail generation in neighbourhoods, constraint evaluation and penalty assignment, fecundity computation, roulette wheel selection, love dart calculation and normalization, trail-following position update, and best-home replacement. Iteration continues until a maximum iteration count is reached or convergence criteria are satisfied, such as negligible improvement over several iterations (Saha et al., 17 Sep 2025).

A central structural feature of SHMS is the coexistence of exploration and exploitation through two interacting levels of organization. Homes preserve region-level memory, while snails provide local stochastic sampling. Better homes correspond to better fitness regions, and better snails leave higher-quality trails through the fecundity index and selection probabilities (Saha et al., 17 Sep 2025). The earlier algorithm paper likewise characterizes the method as a multi-home, multi-snail process in which local search occurs near current homes and global search emerges when snails follow trails or cues to other homes (Kulkarni et al., 2023).

The love dart operator is the principal adaptive movement mechanism. Large fitness differences between a snail and its fecund partner produce larger movements toward better regions, while small differences produce finer local moves. In the source interpretation, this yields strong mating or learning when performance differentials are large and intensification when they are small (Saha et al., 17 Sep 2025). This suggests that SHMS encodes a state-dependent step-size rule rather than a fixed displacement schedule.

Selection pressure is explicitly tied to fecundity and mating probability. Fecund snails are more attractive, attract more followers, and thereby dominate the search. The biological analogy is not merely descriptive: it becomes a mechanism for directing population mass toward improving regions without introducing a velocity state, a crossover operator, or a gradient term (Saha et al., 17 Sep 2025). In comparison with GA, PSO, DE, AVOA, GWO, WOA, SSA, TLBO, FFA, GSA, IPO, and related methods, SHMS is therefore characterized by homes, fecundity, trail following, and love-dart-biased movement rather than explicit leader hierarchies, velocity equations, or differential mutation (Kulkarni et al., 2023).

4. Application to stepped-transmission shaft optimization

In the stepped-transmission shaft study, SHMS is used for weight minimization of a shaft with three cylindrical sections of lengths $S$4, $S$5, and $S$6, using AISI 1040 CD steel with density $S$7. The decision variables are the section diameters $S$8, and the objective is

$S$9

so the optimization problem is $$\mathbf{X}_H = (\mathbf{X}^1, \dots, \mathbf{X}^h, \dots, \mathbf{X}^H)$$0 (Saha et al., 17 Sep 2025).

The mechanical model incorporates fatigue loading, combined bending, torsion loads, and the Modified Goodman criterion. The shaft experiences fluctuating bending moment $$\mathbf{X}_H = (\mathbf{X}^1, \dots, \mathbf{X}^h, \dots, \mathbf{X}^H)$$1 and torque $$\mathbf{X}_H = (\mathbf{X}^1, \dots, \mathbf{X}^h, \dots, \mathbf{X}^H)$$2, decomposed into alternating and mean components. The alternating and mean normal and shear stresses are

$$\mathbf{X}_H = (\mathbf{X}^1, \dots, \mathbf{X}^h, \dots, \mathbf{X}^H)$$3

$$\mathbf{X}_H = (\mathbf{X}^1, \dots, \mathbf{X}^h, \dots, \mathbf{X}^H)$$4

and the corresponding von Mises equivalents are

$$\mathbf{X}_H = (\mathbf{X}^1, \dots, \mathbf{X}^h, \dots, \mathbf{X}^H)$$5

The Modified Goodman criterion is

$$\mathbf{X}_H = (\mathbf{X}^1, \dots, \mathbf{X}^h, \dots, \mathbf{X}^H)$$6

The factor of safety is $$\mathbf{X}_H = (\mathbf{X}^1, \dots, \mathbf{X}^h, \dots, \mathbf{X}^H)$$7, with $$\mathbf{X}_H = (\mathbf{X}^1, \dots, \mathbf{X}^h, \dots, \mathbf{X}^H)$$8, $$\mathbf{X}_H = (\mathbf{X}^1, \dots, \mathbf{X}^h, \dots, \mathbf{X}^H)$$9, and $h$0 (Saha et al., 17 Sep 2025).

The paper provides the diameter expression

$h$1

and formulates the corresponding fatigue constraints as $h$2, $h$3, ensuring that the chosen diameters are at least the minimum required by fatigue strength (Saha et al., 17 Sep 2025).

The geometric and manufacturing constraints require sufficient differences between adjacent diameters. In the integrated optimization encoding these are written as

$h$4

A deflection constraint is enforced at the gear location $h$5:

$h$6

Deflection is computed using Macaulay’s method from

$h$7

with

$h$8

and the paper writes the general deflection expression as

$h$9

subject to $U_s$0 and $U_s$1 (Saha et al., 17 Sep 2025).

The case-study workflow computes torques from fluctuating power with $U_s$2, $U_s$3, and shaft speed $U_s$4. Using

$U_s$5

the study obtains $U_s$6, $U_s$7, $U_s$8, and $U_s$9. Bending moments are extracted from force and bending moment diagrams obtained using MDSOLIDS software, after resolving pulley and gear forces and computing bearing reactions (Saha et al., 17 Sep 2025).

5. Constraint handling, implementation, and computational workflow

For the stepped-shaft encoding, each snail position is

$\mathbf{X}^h$0

and candidate solutions are generated by perturbing home diameters within interval $\mathbf{X}^h$1:

$\mathbf{X}^h$2

Lengths and material properties remain fixed (Saha et al., 17 Sep 2025).

Fitness evaluation combines objective and constraints. For each snail, the algorithm computes the shaft weight, evaluates fatigue diameter constraints $\mathbf{X}^h$3, step-difference constraints $\mathbf{X}^h$4, and the deflection constraint $\mathbf{X}^h$5, then forms the penalty

$\mathbf{X}^h$6

which yields the pseudo-objective

$\mathbf{X}^h$7

Feasible solutions satisfy $\mathbf{X}^h$8 and compete purely on weight, while violated constraints contribute proportionally through the static penalty (Saha et al., 17 Sep 2025).

The reported implementation platform is Python in Google Colab on a Windows machine with Intel i5-6200U and 8 GB RAM. The study performs 30 independent runs to assess robustness. The stopping criterion is a fixed number of iterations, though the paper states that the average function evaluations are approximately 937 rather than listing the iteration count explicitly (Saha et al., 17 Sep 2025).

The textual workflow given for the shaft case is concise and sequential: choose $\mathbf{X}^h$9, $U_s = \mathbf{X}^h + c,$0, and $U_s = \mathbf{X}^h + c,$1; initialize homes within bounds; generate snail positions around homes; evaluate objective and constraints; compute penalty and pseudo-objective; select the best $U_s = \mathbf{X}^h + c,$2 snails as new homes; compute fecundity and selection probabilities; assign fecund mates by roulette wheel; compute love dart and normalize; update positions via trail following; enforce variable bounds; and terminate after maximum iterations or convergence (Saha et al., 17 Sep 2025).

A common misconception is that SHMS contains an intrinsic, problem-independent constraint-handling mechanism. The source materials do not support that interpretation. The general SHMS paper notes that constrained problems are handled externally through penalty approaches and explicitly identifies the development of a constrained SHMS version as future work (Kulkarni et al., 2023). The shaft paper therefore represents a concrete constrained instantiation through SPF rather than a universally embedded feasibility framework (Saha et al., 17 Sep 2025).

6. Empirical results, validation, and limitations

For the stepped-transmission shaft problem, SHMS produced the following representative solutions across 30 runs (Saha et al., 17 Sep 2025):

Case	Weight $U_s = \mathbf{X}^h + c,$3 (lb)	$U_s = \mathbf{X}^h + c,$4 (in)	$U_s = \mathbf{X}^h + c,$5 (in)	$U_s = \mathbf{X}^h + c,$6 (in)
Best	16.5139	1.6822	1.8105	1.5721
Mean	17.0795	1.7086	1.8291	1.6118
Worst	19.6995	1.7654	1.9289	1.8139

The reported analytical deflections at the gear position in the horizontal plane are $U_s = \mathbf{X}^h + c,$7, $U_s = \mathbf{X}^h + c,$8, and $U_s = \mathbf{X}^h + c,$9 for the best, mean, and worst cases respectively, all below the allowable $c$0 limit (Saha et al., 17 Sep 2025).

Validation was carried out in ANSYS SpaceClaim and ANSYS Workbench using Static Structural analysis. The model used AISI 1040 CD steel properties from MatWeb, input torque $c$1 anticlockwise, counter torque $c$2 clockwise, rotational velocity $c$3, cylindrical supports at $c$4 and $c$5, and tetrahedral elements with characteristic size $c$6. The resulting total deformation at the gear region was $c$7, $c$8, and $c$9 for the best, mean, and worst cases, corresponding to relative errors of approximately $$\mathbf{X}^s = \big( X_1^s, X_2^s, \dots, X_N^s \big)^T,$$0, $$\mathbf{X}^s = \big( X_1^s, X_2^s, \dots, X_N^s \big)^T,$$1, and $$\mathbf{X}^s = \big( X_1^s, X_2^s, \dots, X_N^s \big)^T,$$2 when compared with the SHMS analytical deflection values (Saha et al., 17 Sep 2025).

The paper attributes these discrepancies to modelling differences: the SHMS deflection calculations were based primarily on horizontal bending, whereas ANSYS used the resultant of vertical and horizontal loads. It also notes that the worst-weight case has the lowest deflection because larger diameters increase stiffness and reduce bending at the expense of weight. This directly illustrates the trade-off between weight and deflection within the feasible design space (Saha et al., 17 Sep 2025).

The reported robustness metrics for the shaft study are an average of 937 function evaluations, average CPU time per run of 2.11 s, and standard deviation of weight across runs of $$\mathbf{X}^s = \big( X_1^s, X_2^s, \dots, X_N^s \big)^T,$$3. The convergence curve is described as steadily improving the objective to reach $$\mathbf{X}^s = \big( X_1^s, X_2^s, \dots, X_N^s \big)^T,$$4 in the best run (Saha et al., 17 Sep 2025).

In the broader SHMS literature, the original algorithm paper reports strong performance on unimodal and many multimodal benchmark functions, validated with two-sided and pairwise signed rank Wilcoxon tests and Friedman rank tests, and demonstrates an engineering application to shell-and-tube heat exchanger design (Kulkarni et al., 2023). At the same time, that study identifies weaker performance on fixed-dimension multimodal functions $$\mathbf{X}^s = \big( X_1^s, X_2^s, \dots, X_N^s \big)^T,$$5–$$\mathbf{X}^s = \big( X_1^s, X_2^s, \dots, X_N^s \big)^T,$$6, where other algorithms statistically outperform SHMS, and the shaft study likewise notes parameter sensitivity, penalty tuning requirements, local-minima risk, and modelling error relative to ANSYS in the worst case (Kulkarni et al., 2023, Saha et al., 17 Sep 2025).

The stated strengths of SHMS in the shaft context are biologically rich operators, effective constraint handling through SPF, robustness and consistency across multiple runs, engineering suitability for continuous variables and physically grounded constraints, and low computational cost. The stated limitations are the need to tune the static penalty parameter $$\mathbf{X}^s = \big( X_1^s, X_2^s, \dots, X_N^s \big)^T,$$7, sensitivity to the number of homes, snails, and the sampling interval $$\mathbf{X}^s = \big( X_1^s, X_2^s, \dots, X_N^s \big)^T,$$8, susceptibility to premature convergence in rugged landscapes, and simplified analytical modelling relative to full finite-element analysis (Saha et al., 17 Sep 2025). The authors suggest hybridization with algorithms such as TLBO, CI, and PSO, as well as adaptive penalty control or dynamic fecundity scaling, and identify applications including composite structures, trusses and beams, multi-objective design, and fatigue-critical rotating machinery (Saha et al., 17 Sep 2025). A plausible implication is that SHMS is most naturally suited to continuous engineering design problems in which neighbourhood sampling, elitist region retention, and penalty-based feasibility control are acceptable modelling choices.