Non-parametric structural shape optimization of piecewise developable surfaces using discrete differential geometry

Jingyao Zhang; Kentaro Hayakawa; Makoto Ohsaki

REVIEW 3 major objections 6 minor 11 references

Non-parametric structural shape optimization of piecewise developable surfaces using discrete differential geometry

T0 review · 3 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A two-level optimization scheme generates approximately stiffest piecewise developable shell surfaces, with internal creases emerging automatically from the optimization rather than being prescribed in advance.

desk verdict A competent two-level method for approximately piecewise-developable shell shapes, but the developability claim rests on a diagnostic that is the same as the minimized objective and is never independently checked. read the letter →

arxiv 2411.13843 v1 pith:QM7A5GZI submitted 2024-11-21 math.OC

Makoto Ohsaki , Kentaro Hayakawa , Jingyao Zhang This is my paper · ORCID

classification math.OC MSC 49Q1065D1790C26

keywords shapeoptimizationpiecewisedevelopablesurfacediscretedifferentialgeometryGaussmapcompliancesimulatedannealingshellstructureshyperbolictangentfilter

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper proposes a two-level optimization method for finding an approximately optimal shell shape that is piecewise developable—made of patches that can be produced by bending a flat sheet without stretching or in-plane shear. Its central claim is that the stiffest such surface can be found without the designer pre-specifying where the patches meet: internal crease lines emerge automatically from the optimization. This matters because developable panels are much cheaper to fabricate, and automating crease placement removes a major modeling bottleneck for free-form architectural roofs. In numerical examples with square and rectangular plans, the method reduces structural compliance by about 32% and 55%, respectively, relative to the initial non-developable surfaces, while keeping per-vertex developability error small away from the creases.

What carries the argument

The central object is the discrete local Gauss map at a vertex: translate the unit normals of the faces around a vertex to a common origin; the area of the spherical polygon they trace vanishes exactly when the surface is locally developable, that is, flattenable without stretching. The lower-level objective sums the squares of these areas over interior vertices, then applies a hyperbolic-tangent filter $\tanh(c A_i)$ so that a large local error contributes less than its square, letting large errors accumulate at a few vertices instead of spreading. Those vertices become the internal boundaries between developable patches. A stochastic optimizer varies the heights of selected points to minimize structural compliance, with the developable shape for each height found by the filtered least-squares problem.

What would settle it

For the optimized rectangular shell, compute the intrinsic (pairwise) distances between vertices within each detected patch; if any patch with near-zero local Gauss-map area cannot be flattened to the plane without changing those distances, the developability claim fails. A simpler check: plot the per-vertex Gauss-map area for the Case 2 optimum and verify that the high-error vertices form a thin connected curve; the paper already notes these points are 'not distinctly distributed,' so such a plot would show whether a true internal boundary exists.

Watch

Extended reading notes

Core claim

The central discovery is that stiffness optimization can be carried out inside the class of piecewise developable surfaces by separating the problem into two levels. The lower level drives the area of the discrete local Gauss map at every interior vertex toward zero, using a hyperbolic-tangent filter that underestimates large errors so that they concentrate as internal boundaries instead of spreading; the upper level then varies the heights of selected points to minimize compliance. On square and rectangular test shells, the resulting surfaces are stiffer than the initial non-developable shapes while remaining locally developable off the creases. The authors describe the outcome as approximate: in the rectangular case they state that the residual Gauss-map error for the optimal solution is 'not distinctly distributed,' and for the larger filter parameter convergence near some design-variable points is imperfect.

Load-bearing premise

The method assumes that the area of the discrete local Gauss map at a vertex is a valid and sufficient measure of local developability, and that minimizing its filtered sum makes the remaining errors concentrate as clean internal boundaries rather than spreading through the patches.

Editorial extensions

If this is right

Shell roofs can be optimized within the fabricable class of piecewise developable surfaces without a priori placement of crease lines, because the tanh filter lets the optimizer choose where developability is relaxed.
The resulting surfaces can be built from flat panels bent along straight or curved creases, reducing fabrication and construction cost compared with general doubly curved shells.
The two-level decomposition lets each level use a different solver—gradient-based local search for developability and global stochastic search for stiffness—so the approach can be adapted to other loads, supports, and plan geometries.
Because no symmetry or periodic grid is assumed, the internal boundaries are not aligned with a preset mesh direction and can appear at various orientations.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

If the local Gauss-map measure were replaced by a strict flat-unfolding check, the same two-level scheme could certify the fabricability of each patch; the paper stops at the local measure and observes imperfect localization in one case.
The filter parameter $c$ is effectively a knob controlling crease sharpness: small values blur boundaries, large values sharpen them; an automatic schedule for $c$ could make the lower-level problem less prone to poor local minima.
The upper level currently searches only a small set of height variables; introducing a gradient-based upper level with compliance sensitivities would let the method scale to many more design variables and finer meshes.
A natural testable extension is to optimize with additional constraints such as maximum crease curvature or minimum patch area, which would make the numerically optimized surfaces directly manufacturable as physical panels.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The genuinely new thing here is the two-level coupling: the lower level generates a piecewise developable surface with creases that emerge from a tanh-filtered Gauss-map error, and the upper level adjusts selected heights to minimize compliance. That combination, plus the two worked examples, is not in the authors' prior work. The paper is clearly written, the optimization setup is sensible, and the reported compliance reductions (from 4.998 to 3.386 kNm in Case 1, from 0.148 to 0.067 kNm in Case 2) are plausible for the described geometry changes. I also give them credit for flagging their own trouble spots: Case 2's error distribution is admitted to be "not distinctly distributed," and convergence near the design variables is called "not good." That is honest reporting.

The soft spot is exactly the one a skeptical reader should worry about. The developability measure in Eq. (1) is the same local Gauss-map area that is plotted in the figures, so the color plots are not an independent check of flattenability. There is no computation of angle defect at interior vertices, no attempt to unfold the patches isometrically, and no quantitative residual reported. The tanh filter deliberately underweights large errors, so the optimizer has weak incentive to concentrate all error into narrow creases; the Case 2 admission suggests the errors are genuinely diffuse. That means "piecewise developable" is a characterization the paper has not actually earned. It may still be that the final surfaces are close enough for practical purposes—this is an engineering approximation paper, and the word "approximately" appears in the abstract—but the load-bearing premise is unverified.

Two smaller issues: the stochastic optimization (dual annealing, perturbed grid points) is run once, with no seeds, restarts, or error bars, so the robustness of the reported numbers is unknown. And the paper gives no code or data, which makes independent verification harder than it should be for a method built on standard SciPy/OpenSeesPy components.

These are fixable in revision, not fatal. The method is real, the direction is sensible, and the honest limitations section makes me trust the authors more than the average paper. I would send this to peer review, and I would require: a quantitative developability check that is not the objective (for example, discrete angle defects per vertex, or an actual flattening error per patch), multiple restarts with variance, and a paragraph explaining what "approximate developability" means for fabrication. Without that, the central claim stays unsupported. With it, this becomes a useful paper for shell and tensile-structure engineers who care about forming cost.

Referee Report

3 major / 6 minor

Summary. The paper proposes a two-level optimization method for generating approximately stiffest piecewise developable surfaces (PDSs) without prescribing internal boundaries. In the lower level (Problem 1), the vertical coordinates of non-fixed grid points are optimized by minimizing a hyperbolic-tangent-filtered sum of squared areas of local discrete Gauss-map triangles at interior vertices, with the aim of concentrating large developability errors along emergent internal creases. In the upper level (Problem 2), the heights of selected points are optimized by dual-annealing simulated annealing to minimize structural compliance under vertical loading, using shell finite element analysis. Two numerical examples are presented: a square plan (Case 1) and a rectangular plan (Case 2). In both cases, compliance is reduced relative to the initial non-developable shape after the two-level procedure (e.g., Case 1: 4.998 to 3.386 kNm; Case 2: 0.1480 to 0.06668 kNm), while the paper claims the final surfaces remain approximately piecewise developable.

Significance. If the central claim is valid, the paper offers a practical, fully discretized pipeline for architectural shell design in which developable patches are generated without a priori specification of crease locations, thereby combining fabrication-friendly geometry with stiffness optimization. The strengths are the clear two-level formulation, the use of discrete differential geometry to avoid parametric surface restrictions, and the use of standard, reproducible computational tools (SLSQP, dual_annealing, OpenSeesPy). However, the significance is conditional on two unverified premises: that vanishing local Gauss-map area is a faithful and sufficient measure of developability at a vertex, and that the tanh-filtered objective actually drives errors to concentrate into narrow internal boundaries rather than diffuse over large regions. The paper's own Case 2 report that the violating points are 'not distinctly distributed' indicates that the second premise is not always met, which weakens the piecewise-developable characterization of the optimized surfaces.

major comments (3)

[Section 2, Eq. (1)] The lower-level objective is exactly the tanh-filtered sum of squared local Gauss-map triangle areas. The paper states that the area of the local Gauss map vanishes if the surface is developable at the grid point, but it provides no proof or independent verification of this equivalence. For a triangulated surface, the standard local developability condition is zero angle defect at interior vertices, and global flattenability requires an isometric unfolding of each patch. Because the diagnostic plotted in Figs. 4(c)-11(c) is the same quantity that is minimized in Problem 1, a small plotted error only restates that the objective is small; it is not an independent check. The authors should compute and report the vertex angle defect distribution and/or perform an actual unfolding of the final patches, and show that the patches have negligible distortion away from the detected internal boundaries.
[Section 2, Eq. (1), and Section 3.2] The hyperbolic tangent filter deliberately underestimates large errors, so the optimizer has little incentive to concentrate all error on a narrow crease; it may instead spread errors diffusely. The paper's own statement in Section 3.2 that for the optimal solution of Problem 2 in Case 2 the violating points 'are not distinctly distributed' directly undercuts the claim that the result is a piecewise developable surface with identifiable internal boundaries. The authors should quantify the spatial concentration of developability error (e.g., fraction of total error contained in a small fraction of vertices) and, if necessary, revise the objective or post-processing to enforce or detect crease localization. Without this, the phrase 'while keeping developability of each surface patch' is not supported.
[Section 3.2, Figure 11] The paper reports that for Case 2 with c=100 the convergence near the variable points of Problem 2 is 'not good'. This raises doubt about the reliability of the upper-level optimum and about the comparison among the c=10, c=100, and c=200 results. The authors should provide quantitative convergence information (e.g., multiple runs of dual-annealing, history of the best compliance, sensitivity to random seed) and, ideally, report the final objective value and developability error for each parameter setting rather than only the shapes.

minor comments (6)

[Section 3] The section numbering is inconsistent: Section 3.1 is used for both 'Description of shell model and problem setting' and 'Case 1: Square plan'. Please renumber the subsections.
[Figures 4-11] The color scale for the area of the local Gauss map is not defined in the figures or captions. Providing a common color bar with numeric values would help the reader interpret the magnitude of developability errors.
[Section 2, Eq. (1)] The notation tanh(c A_i) is ambiguous regarding the argument; consider writing tanh(c A_i) with parentheses or defining the function explicitly as a mapping on nonnegative reals.
[Introduction and Section 2] The term 'meshless' is used, but the method relies on grid points and auxiliary edges; clarifying the sense in which the approach is meshless would avoid confusion.
[References] Reference [2] appears to contain a typo: 'Bobenco' should likely be 'Bobenko'.
[Section 3.1] For completeness, please specify the boundary conditions used in the structural analysis (the figures show supports but the precise constraints and load direction are not fully described in the text).

Circularity Check

1 steps flagged · score 2.0 of 10

The developability check is the same filtered Gauss-map error that is minimized, but the compliance reduction is independently evaluated, so the circularity is minor.

self definitional [Section 2, Eq. (1); Section 4 Conclusions; Figs. 6(c), 10(c)]
"The condition for developability of a polyhedral surface onto a plane is formulated using the area of discrete Gauss map formed by unit normal vectors at the faces adjacent to each vertex. The objective function of the lower-level optimization problem is the sum of square errors for developability at all interior vertices. ..."

The developability of the optimized surface is demonstrated only by plotting the area of the local Gauss map, which is exactly the quantity whose filtered sum is minimized in Problem 1 (Eq. (1)). Thus 'the final surface keeps developability' is essentially a restatement that the lower-level objective was made small; it is not an independent flattenability check, and no angle-defect or isometric-unfolding test is reported. This circularity is partial: the upper-level compliance values are computed with an independent OpenSeesPy shell model, so the main structural conclusion does not reduce to the objective.

full rationale

The paper's principal structural claim, that compliance is reduced, is evaluated with a separate finite-element model (OpenSeesPy shellMITC4) and is therefore independent of the optimized developability objective. The only circular element is the validation of 'keeping developability': the diagnostic used (local Gauss-map area) is the same quantity minimized in Problem 1, and the paper itself notes in Case 2 that the violating points are 'not distinctly distributed', weakening the claim that creases emerged as well-localized internal boundaries. The method is largely assembled from the authors' prior works (Refs. [6-8]), but those are used as building blocks rather than as a substitute for the structural analysis, so the self-citation is not the main driver of the result. Overall, the circularity is minor and confined to the developability verification, not to the compliance optimization.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a hand-tuned filter parameter c, a random grid perturbation whose seed is unreported, and the equivalence of the local Gauss map area to developability. No new physical entity is introduced. The structural model (MITC4) and loads are standard engineering inputs.

free parameters (4)

c (hyperbolic tangent steepness) = 10, 100, 200 (Case 1); 10, 100 (Case 2)
Controls how strongly large developability errors are underestimated; the paper reports no clear internal boundary for c=10 and little change for c>100, so results depend on this hand-chosen value.
grid perturbation amplitude = u,v varied by +/-0.015 (Case 1) and +/-0.01 (Case 2) from uniform values
Introduced to make internal boundaries appear in unspecified directions; the seed and exact sampling are not given, so the demonstrated shapes depend on an unreported random realization.
selected upper-level design points = 5 points (Case 1), 11 points (Case 2)
The z-coordinates of these points are the only structural design variables; the choice of which points to release is made by the authors and is not optimized.
z-coordinate bounds = +/-18 m lower-level, +/-21 m upper-level relative to initial z
Algorithm-specific bounds that constrain the feasible region and therefore influence the reported compliance reductions.

assumptions (4)

domain assumption Vanishing area of the discrete local Gauss map at a vertex is equivalent to local developability of the polyhedral surface at that vertex.
Invoked in Section 2 (Problem 1 formulation) and cited to Refs [6,7]; no derivation or counterexample check is given in this paper, and sufficiency for flattenability is not established.
ad hoc to paper Applying tanh(c A_i) to the developability error makes large errors concentrate along emergent internal boundaries rather than destroying the patch developability.
This is the central mechanism of Ref. [8] reused here; the paper observes it in figures but provides no convergence or localization guarantee.
domain assumption The four-node MITC4 shell finite element model in OpenSeesPy gives compliance values accurate enough to rank candidate shell shapes.
Used for every W(Z) evaluation in Section 3; mesh refinement and verification against analytical solutions are not reported.
domain assumption The randomly perturbed NURBS-derived initial grid spans the useful design space for both plans.
Initial vertical coordinates and grid connectivity are set from a NURBS surface and random parameter perturbation; no evidence is given that the final shapes are independent of this initialization.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Non-parametric structural shape optimization of piecewise developable surfaces using discrete differential geometry." pith.science (2026). https://pith.science/paper/QM7A5GZI

@misc{pith2026241113843,
  author       = {Pith},
  title        = {Pith review of: Non-parametric structural shape optimization of piecewise developable surfaces using discrete differential geometry},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QM7A5GZI}},
  note         = {Machine review of arXiv:2411.13843}
}

read the original abstract

We propose a two-level structural optimization method for obtaining an approximate optimal shape of piecewise developable surface without specifying internal boundaries between surface patches. The condition for developability of a polyhedral surface onto a plane is formulated using the area of discrete Gauss map formed by unit normal vectors at the faces adjacent to each vertex. The objective function of the lower-level optimization problem is the sum of square errors for developability at all interior vertices. The contribution of large error to the objective function is underestimated by filtering with hyperbolic tangent function so that the internal boundary between the surface patches can naturally emerge as a result of optimization. Vertices are located non-periodically to generate the internal boundaries in various unspecified directions. Simulated annealing is used for the upper-level optimization problem for maximizing stiffness evaluated by the compliance under the specified vertical loads. The design variables are the heights of the specified points. It is shown in the numerical examples that the compliance values of the surfaces with a square and a rectangular plan are successfully reduced by the proposed method while keeping the developability of each surface patch. Thus, a new class of structural shape optimization problem of shell surfaces is proposed by limiting the feasible surface to piecewise developable surfaces which have desirable geometrical characteristics in view of fabrication and construction.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

11 extracted references · 11 canonical work pages

[1]

Farin, G. (2001). Curves and Surfaces for CAGD: A Practical Guide. Morgan Kaufmann

work page 2001
[2]

Bobenco, A. I. , Schröder, P., Sullivan, J. M. and Ziegler, G. M. (Eds.). (2008). Discrete Differential Geometry. Birkhäuser

work page 2008
[3]

Cui, J., Ohsaki, M. (2018). Shape design of curved surface of membrane structure using developable surface. J. Int. Assoc. Shell Spatial. Struct., 59(3), 199−214

work page 2018
[4]

and Yokosuka, Y

Hayashi, K., Jikumaru, Y., Ohsaki, N., Kagaya, T. and Yokosuka, Y. (2021). Discrete Gaussian curvature flow for piecewise constant Gaussian curvature surface. Computer-Aided Design, 134, 102992

work page 2021
[5]

and Ohsaki, M

Hayakawa, K. and Ohsaki, M. (2021). Form generation of rigid origami by multiobjective optimization for approximating curved surface. Proc. IASS Symposium 2021, Surrey, UK, 2743−2754

work page 2021
[6]

Hayakawa, K., Ohsaki, M., Zhang, J. Y. (2024). Meshless non -parametric shape design of piecewise approximately developable surfaces using discretized local Gauss map . J. Int. Assoc. Shell Spatial. Struct., 65, published online

work page 2024
[7]

Ohsaki, M., Hayakawa. K. and Zhang, J. Y. (2023). Nonparametric shape optimization of piecewise developable surfaces for maximum stiffness using discrete differential geometry. Proc. IASS Symposium 2023, Melbourne, Australia, 1452−1456

work page 2023
[8]

and Ohsaki, M

Hayakawa, K. and Ohsaki, M. (2023). Form generation of discrete piecewise developable surface and its interior boundaries using local gauss map, Summaries of Technical Papers of Annual Meeting, Architectural Institute of Japan, Structures-1, 997–998. (in Japanese)

work page 2023

Show all 11 references

[9]

SLSQP Reference Manual, https://docs.scipy.org/doc/scipy/reference/optimize.minimize-slsqp.html
[10]

Dual_annealing Reference Manual, https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.dual_annealing.html
[11]

OpenSeesPy Reference Manual, https://openseespydoc.readthedocs.io/en/latest/index.html

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [1]

Farin, G. (2001). Curves and Surfaces for CAGD: A Practical Guide. Morgan Kaufmann

work page 2001

[2] [2]

Bobenco, A. I. , Schröder, P., Sullivan, J. M. and Ziegler, G. M. (Eds.). (2008). Discrete Differential Geometry. Birkhäuser

work page 2008

[3] [3]

Cui, J., Ohsaki, M. (2018). Shape design of curved surface of membrane structure using developable surface. J. Int. Assoc. Shell Spatial. Struct., 59(3), 199−214

work page 2018

[4] [4]

and Yokosuka, Y

Hayashi, K., Jikumaru, Y., Ohsaki, N., Kagaya, T. and Yokosuka, Y. (2021). Discrete Gaussian curvature flow for piecewise constant Gaussian curvature surface. Computer-Aided Design, 134, 102992

work page 2021

[5] [5]

and Ohsaki, M

Hayakawa, K. and Ohsaki, M. (2021). Form generation of rigid origami by multiobjective optimization for approximating curved surface. Proc. IASS Symposium 2021, Surrey, UK, 2743−2754

work page 2021

[6] [6]

Hayakawa, K., Ohsaki, M., Zhang, J. Y. (2024). Meshless non -parametric shape design of piecewise approximately developable surfaces using discretized local Gauss map . J. Int. Assoc. Shell Spatial. Struct., 65, published online

work page 2024

[7] [7]

Ohsaki, M., Hayakawa. K. and Zhang, J. Y. (2023). Nonparametric shape optimization of piecewise developable surfaces for maximum stiffness using discrete differential geometry. Proc. IASS Symposium 2023, Melbourne, Australia, 1452−1456

work page 2023

[8] [8]

and Ohsaki, M

Hayakawa, K. and Ohsaki, M. (2023). Form generation of discrete piecewise developable surface and its interior boundaries using local gauss map, Summaries of Technical Papers of Annual Meeting, Architectural Institute of Japan, Structures-1, 997–998. (in Japanese)

work page 2023