Does each moving point improve on every iteration?

No. The update is geometric, and the objective need not decrease for every point.

Why does the best value remain available in the basic deterministic formulation?

The center has zero displacement and stays in the candidate set during the update.

Can it operate in more than two dimensions?

Yes. The published extension constructs rotations from coordinate-plane rotation matrices, although the visual spiral analogy becomes less direct.

AI Glossary / Mathematics and optimization / Advanced

Spiral optimization algorithm

Definition

The spiral optimization algorithm is a multipoint search method for continuous optimization that moves candidate solutions through rotating, contracting steps around the best current candidate. Kenichi Tamura and Keiichiro Yasuda developed the spiral-dynamics approach. Its geometric inspiration is a logarithmic spiral. The method searches for low objective values without requiring derivatives, but its spiral trajectories do not by themselves certify global optimality.

What it means

The algorithm maintains several points in a numerical search space. A rotation changes the direction from which a candidate approaches the current center, while contraction reduces its distance from that center. After evaluating the moved points, the best point becomes the next center. This creates a changing balance between searching around a region and concentrating on a promising solution. The center is the best solution found so far within the algorithm's retained population, not an already known true optimum.

How it works

Choose initial feasible points, a contraction factor r between zero and one, a rotation setting and a stopping budget. In two dimensions, let R be the chosen rotation matrix and S = rR. With current center c, update each point by xnew = c + S(x − c). Evaluate the updated points and select the best as the next center. The center point itself has zero displacement, so the basic update retains it. Higher-dimensional versions compose rotations in coordinate planes. Bounds and additional constraints need explicit treatment: rotating around a feasible center can still send another candidate outside the allowed domain.

Three practical examples

A numerical-methods study can compare spiral search with other derivative-free optimizers on continuous benchmark functions under equal evaluation budgets.
A simulation-based design experiment can investigate spiral updates for a modest number of continuous parameters, provided infeasible candidates are handled consistently.
A teaching visualization can show how rotation angle and contraction factor affect trajectories around a fixed center before introducing center updates and objective evaluations.

Think of it this way

Think of circling a promising location while drawing closer, then shifting attention when a better location is discovered. The analogy captures rotation, contraction and recentering. The search takes discrete numerical steps, however, and the objective landscape may contain many poor local regions. Circling a promising spot is not evidence that every other part of the landscape has been explored.

Worked example

Minimize f(x, y) = x² + y² using two illustrative points c = (1, 1) and p = (3, 1). Their scores are 2 and 10, so c is initially the center. Set r = 1/2 and use a counterclockwise 90-degree rotation, R = [[0, −1], [1, 0]]. The displacement p − c = (2, 0) rotates to (0, 2) and contracts to (0, 1). Thus pnew = (1, 2), with score 5. The center remains (1, 1), whose score is still lower. On the next step, displacement (0, 1) rotates to (−1, 0) and contracts to (−1/2, 0), giving the candidate (1/2, 1). Its score is 1/4 + 1 = 1.25, so it becomes the new center. Relative to the old center, the moving point's distance shrank from 2 to 1 to 1/2. That geometric contraction is separate from distance to the true optimum (0, 0), and the center must change when the new candidate is selected. The example traces two updates only; it is not a benchmark run or a convergence proof.

Illustrative case study

In a fictional project, Alder Dynamics tests spiral search on a deterministic calibration simulator. Engineers scale parameters to comparable numerical ranges, compare several initial populations and record objective evaluations rather than only iteration counts. They inspect runs that converge tightly around a poor result and compare them with a simple baseline. The team presents the best value found and its variability, without claiming that the algorithm has established the global optimum or guaranteed a practical improvement.

Common misconceptions and mistakes

Spiral optimization is not the same as scanning every point along a continuous spiral. With a fixed center and fixed parameters, its updates are deterministic once initial points are given. Random initialization does not make every subsequent step a fresh random sample. A contracting trajectory approaches its fixed center, but that fact alone does not prove the center is globally optimal. The rotation angle matters: special choices can repeatedly visit a small set of directions. Contraction that is too rapid can limit exploration, while slower contraction can require more evaluations to concentrate.

Business use

Use the method as an experimentally assessed optimizer when derivative-free continuous search is appropriate. Record parameter scaling, initial points, contraction and rotation settings, constraints and stopping conditions. Compare results over multiple initializations with relevant alternatives using the same objective-evaluation budget. Preserve a trace of best objective and feasibility so premature concentration is visible. For a noisy simulator, determine whether an apparent new best is reliable before allowing noise to redirect the entire population. Any modification, such as restarts, adaptive parameters or projection onto bounds, should be identified because it changes the original procedure.

Check your understanding

A point halves its distance to the current center. Must its objective value halve?

Answer: No. The objective is an independent function of position. Even for the worked quadratic, contraction around a nonoptimal center does not translate into a fixed proportional reduction in objective.

Related terms

Derivative-free optimization
rotation matrix
contraction mapping
population-based search
metaheuristic
exploration and exploitation

Learn next

reproduce the two worked updates and then continue using the newly selected center. Compare trajectories for 45-degree and 90-degree rotations while holding contraction fixed. Track both distance to the current center and best objective so geometric convergence is not confused with optimization success.

Sources

Sources: Kenichi Tamura and Keiichiro Yasuda, Spiral Dynamics Inspired Optimization, Journal of Advanced Computational Intelligence and Intelligent Informatics 15(8), 1116–1122, https://doi.org/10.20965/jaciii.2011.p1116 ; publisher article and full paper, https://www.fujipress.jp/jaciii/jc/jacii001500081116/ . The two-point calculation and fictional calibration study are original; the source's benchmark results are not generalized to every problem.

Previous: Spiral hashing Next: SPL notation