Deblurring structural edges in variable thickness topology optimization via density-gradient-informed projection Authors: Gabriel Stankiewicz, Chaitanya Dev, Paul Steinmann Published: 2026 ABSTRACT Variable thickness topology optimization (VTTO) is a potent methodology for designing high-performance, high-stiffness sheet structures. However, this method frequently encounters two primary challenges: (1) the formation of undesirable low-thickness regions, which present manufacturing difficulties, and (2) the blurring of structural edges. This blurring is an artifact inherent to the regularization filters required for well-posedness. This paper proposes solutions to address both challenges. First, to mitigate low-thickness regions, we introduce a robust, combined approach. This strategy utilizes a SIMP-based penalization and an updated projection method, which effectively suppresses nearly all low-thickness domains. Second, the main contribution of this work is a novel method to deblur structural edges, termed the density-gradient-informed (DGI) projection. This projection utilizes local density gradient information. It selectively applies a strong projection in high-gradient regions (i.e., structural edges) to restore sharpness, while minimally affecting low-gradient regions within the structure’s interior. Numerical examples demonstrate that the DGI projection successfully deblurs the structural edges, restoring a distinct solid-void transition, while preserving the internal form. Most importantly, this significant improvement in edge definition is achieved with a negligible impact on the final structural compliance. This establishes the DGI projection as a non-invasive and effective regularization tool for enhancing VTTO designs. Introduction Since Michell’s classical theory (Michell 1904) of optimal frame structures, structural optimization and, in particular, topology optimization (TO) has developed into a powerful tool for creating structures with high performance and manufacturing feasibility. Early frameworks such as homogenization-based TO (Bendsøe and Kikuchi 1988), density-based methods (Bendsøe 1989), and level-set techniques (Wang et al. 2003) shaped the field, with the density-based SIMP (Solid Isotropic Material with Penalization) approach becoming widely chosen because of its simple formulation and integration into commercial FEM tools. By penalizing intermediate densities (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p > 1$$\end{document}p>1), SIMP favors black-and-white designs and prevents non-physical gray regions. On the contrary, as early as in the 1970s, Rossow and Taylor (1973) proposed a two-dimensional optimization framework with variable thickness sheet (VTS) that bears a close relation to density-based methods. When no penalization is applied in the SIMP approach (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p = 1$$\end{document}p=1), the interpretation of the design variables changes: intermediate values can be understood as thicknesses, as in the VTS optimization of Rossow and Taylor (1973). Variable thickness topology optimization (VTTO), a specific case of the VTS method representing a density-based or homogenization-based approach with possible topological changes, received little interest in its early phase, while immense developments in (penalized) TO methods were achieved. Characteristic of the penalized TO methods is the formation of visually appealing truss-like features. The obtained structures often mimic biological tissues, suggesting a valid comparison with nature’s tendency to evolve and optimize its creations, further encouraging scientific developments in penalized TO methods. On the other hand, the VTTO approach renders visually less appealing sheet-like structures. However, the lack of penalization of intermediate densities enlarges the design space and can yield solutions of higher stiffness compared to penalized approaches. Only recently, Sigmund et al. (2016) demonstrated a comparative study between the (thin) truss-like and sheet-like structures, clearly indicating the superior stiffness of sheet-like structures. The renewed interest in VTTO is also linked to the dehomogenization method, where coarse variable thickness fields are post-processed into high-resolution truss-like structures (Groen and Sigmund 2018; Larsen et al. 2018), or multi-patch microstructures in a level-set setting (Li et al. 2018). Moreover, other VTS approaches have been extended to plates and shells (Zhao et al. 2017; Meng et al. 2022), multimaterial designs (Banh and Lee 2019; Nguyen and Bui 2022), and composites (Stegmann and Lund 2005; Sørensen et al. 2014). Advanced formulations couple thickness with shape or material optimization (Meng et al. 2022; Sjølund et al. 2018; Kashanian and Kim 2021). The resurgence of VTTO highlighted both the potential and the challenges of such formulations. Kandemir et al. (2018) compare a variety of approaches to obtain VTTO structures including the application of penalized TO, while Yarlagadda et al. (2022) proposed the SIMTP extension with nodal thickness variables. However, one of the major challenges arising in VTTO is the formation of extremely thin sheets, which are sensitive to buckling, hard to interpret, and difficult to manufacture. For instance, Groen and Sigmund (2018) and Larsen et al. (2018) introduced a specialized smoothed Heaviside projection function that suppresses low-thickness regions for the dehomogenization step. Giele et al. (2021) introduced two strategies inspired by the cut element method, in which an additional level-set field is introduced that truncates low-thickness regions, Geoffroy-Donders et al. (2022) and Pozo et al. (2023) suggested alternative penalization laws to avoid thin parts, while Endress and Zimmermann (2023) applied a SIMP-like rule to suppress thicknesses below a threshold. All these contributions aimed at stable, production-ready designs without fragile sheet-like parts. In this work, we propose a generalized approach to avoid low-thickness regions that is highly robust, easy to control, and exhibits high tolerance with respect to the choice of parameters. As its main contribution, this work addresses another challenge inherent to VTTO, namely, the formation of blurred structural edges. Blurred edges are a natural consequence of employing any type of blurring filter, i.e., a radius or Helmholtz-type PDE filter (Lazarov and Sigmund 2011). The aforementioned low-thickness projection methods partially mitigate the issue, forming sharp structural edges, however, only of the thickness \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _\textrm{low}$$\end{document}ρlow. Sharp structural edges of thickness \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho > \rho _\textrm{low}$$\end{document}ρ>ρlow naturally occur in VTTO problems, as seen in the unfiltered density field. However, they are entirely eliminated when a blurring filter, necessary for well-posedness of TO, is applied. In standard TO, where black-and-white designs are desired, this issue is resolved by applying the smoothed Heaviside projection which simply pushes each filtered density either towards 0 or 1 (Wang et al. 2011). In VTTO, where any density in the range \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho \in [0,1]$$\end{document}ρ∈[0,1] is equally permitted, this form of Heaviside projection cannot be employed. Therefore, in this work we propose a projection method that restores the original sharpness of structural edges in VTTO problems and enables fully robust VTS designs with clear structural edges of any thickness in the range \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho \in [\rho _\textrm{low},1]$$\end{document}ρ∈[ρlow,1]. The proposed projection method utilizes information about the density gradient to target the structural edges specifically while minimally affecting the structure’s interior. On top of that, an improved strategy to handle low-thickness regions is proposed which allows a selection of more aggressive continuation parameters and to obtain stable solutions each time. This article is organized as follows. Section 2 briefly introduces the concept of VTTO in the context of density-based TO. In Sect. 3 we propose and test the improved and highly robust approach to eliminate undesired low-thickness regions. The main contribution of this work, that is, the density-gradient-informed (DGI) projection for sharp structural edges, is introduced and tested in Sect. 4. Final conclusions and outlook are given in Sect. 5. Variable thickness topology optimization Rossow and Taylor (1973) introduced the VTS problem by assigning a thickness variable \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau _i$$\end{document}τi to each finite element. Bounded by the thickness constraints \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau _i^L \le \tau _i \le \tau _i^U$$\end{document}τiL≤τi≤τiU, the method inherently avoided low-thickness regions. However, no topological changes nor formation of intra-domain structural edges were possible. The original VTS problem is, however, closely related to conventional density-based TO, in which the penalization factor is omitted from the SIMP method. By adopting the densities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _e$$\end{document}ρe as design variables, the VTTO dealing with a volume-constrained compliance minimization problem is formulated as follows:1where is the compliance objective and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {G}_{vol }\left( \rho \right) $$\end{document}Gvolρ is the volume fraction constraint. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E(\rho _e)$$\end{document}E(ρe) is the interpolated element Young’s modulus, for which the element density \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _e$$\end{document}ρe is applied without the penalty exponent p—recall the SIMP formula \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E(\rho _e) = \rho _e^p E_0$$\end{document}E(ρe)=ρepE0. The minimum value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _\textrm{min} = 10^{-9}$$\end{document}ρmin=10-9 is understood as a void, hence permitting the formation of intra-domain void regions. Unlike in conventional TO, intermediate values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _e \in (0,1)$$\end{document}ρe∈(0,1) are physically meaningful and desired, since they represent the out-of-plane thickness of an extruded structure. Since the penalty exponent p is a restrictive measure, enabling only black-and-white structures, it effectively shrinks the design space. As a consequence, inferior designs in terms of the performance objective are obtained, as shown by Sigmund et al. (2016), with compliance approximately \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$30\%$$\end{document}30% larger as compared to sheet-like structures. For all numerical tests in this work, we use the common setup for the cantilever benchmark, as shown in Fig. 1.Fig. 1Standard setup for the cantilever benchmark used throughout this work Standard setup for the cantilever benchmark used throughout this work The parameter setup, common for all examples used throughout this work—unless otherwise specified—is summarized in Table. 1.Table 1General parameter setup for the numerical testsGeneral\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _\textrm{init}$$\end{document}ρinit0.3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{V}_\textrm{frac}$$\end{document}V¯frac0.3Material TypeLinear elasticE1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu $$\end{document}μ0.3Optimizer TypeGOCMInitial step length0.05Step length decay0.98Min. step length\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10^{-4}$$\end{document}10-4Design updateExponentialFilter TypePDERadius0.375 (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.5h_2$$\end{document}1.5h2)Stopping criterion Type\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \rho _\textrm{mean}$$\end{document}ΔρmeanTolerance\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10^{-4}$$\end{document}10-4Adaptive meshInitial ref. level2Refinement critCoarsening critMax ref. level4Min ref. level0Level 0 elem. size\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h_0 = 1$$\end{document}h0=1 Table 1General parameter setup for the numerical testsGeneral\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _\textrm{init}$$\end{document}ρinit0.3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{V}_\textrm{frac}$$\end{document}V¯frac0.3Material TypeLinear elasticE1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu $$\end{document}μ0.3Optimizer TypeGOCMInitial step length0.05Step length decay0.98Min. step length\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10^{-4}$$\end{document}10-4Design updateExponentialFilter TypePDERadius0.375 (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.5h_2$$\end{document}1.5h2)Stopping criterion Type\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \rho _\textrm{mean}$$\end{document}ΔρmeanTolerance\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10^{-4}$$\end{document}10-4Adaptive meshInitial ref. level2Refinement critCoarsening critMax ref. level4Min ref. level0Level 0 elem. size\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h_0 = 1$$\end{document}h0=1 General parameter setup for the numerical tests The implemented optimizer is the Generalized Optimality Criteria Method (GOCM), as proposed by Kim et al. (2021). In our experience, GOCM deals excellently with both pure compliance minimization problems and multi-constraint problems, i.e., including stress constraints, while maintaining the simplicity level of the original OCM method. The step length is multiplied by the decay parameter in every iteration to aid convergence. Helmholtz-type PDE filter is employed for regularization (Lazarov and Sigmund 2011). Since we use adaptive coarsening and refinement, we use the concept of refinement levels, where the coarsest allowable level 0 corresponds to an element size of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h_0 = 1$$\end{document}h0=1 and the finest allowable level 4 corresponds to an element size of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h_4 = 1 \cdot 0.5^4 = 0.06125$$\end{document}h4=1·0.54=0.06125. The initial mesh level is 2, corresponding to a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$80 \times 40$$\end{document}80×40 grid and an element size of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h_2 = 1 \cdot 0.5^2 = 0.25$$\end{document}h2=1·0.52=0.25. For a detailed description of how mesh adaptivity is handled, refer to Stankiewicz et al. (2025) and implementation details of the deal.II FEM library Bangerth et al. (2007); Arndt et al. (2021). The criterion for mesh adaptation is the density jump . That is, for each element e, we find the maximum density difference between the element e and its neighbors, as introduced in Stankiewicz et al. (2026). Remark 1 We emphasize that mesh adaptivity is neither required for the proposed methodology, nor it affects the main findings in this work. While beneficial for computational efficiency, adaptivity introduces implementation complexity. A fixed grid of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$320 \times 160$$\end{document}320×160 cells achieves the same resolution for the cantilever benchmark and remains feasible on standard hardware. For the stopping criterion, we use the mean density change, accounting for varying element sizes, defined as2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Delta \rho _\textrm{mean} = \frac{\sum _e v_e \Delta \rho _e}{\sum _e v_e}, \end{aligned}$$\end{document}Δρmean=∑eveΔρe∑eve,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \rho _e$$\end{document}Δρe is the density change between optimization iterations for the element e and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_e$$\end{document}ve is the volume of the element e. In Fig. 2 a comparison between standard and VTTO for the cantilever benchmark is shown. Penalization of intermediate densities results in an around \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$17.2\%$$\end{document}17.2% increase in the compliance objective.Fig. 2The common benchmark problem—the cantilever—is compared using standard SIMP-based TO and VTTO. The final compliance values are \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {F}_{c } = 7.61 \cdot 10^{-5}$$\end{document}Fc=7.61·10-5 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {F}_{c } = 8.92 \cdot 10^{-5}$$\end{document}Fc=8.92·10-5, respectively, indicating a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$17.2\%$$\end{document}17.2% increase as a result of density penalization The common benchmark problem—the cantilever—is compared using standard SIMP-based TO and VTTO. The final compliance values are \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {F}_{c } = 7.61 \cdot 10^{-5}$$\end{document}Fc=7.61·10-5 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {F}_{c } = 8.92 \cdot 10^{-5}$$\end{document}Fc=8.92·10-5, respectively, indicating a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$17.2\%$$\end{document}17.2% increase as a result of density penalization While VTTO enables superior performance due to an enlarged design space, it faces several issues, like very thin thickness regions and blurred structural edges. These issues are addressed in the following sections. Generalized treatment of low-thickness regions VTTO inherently produces very thin sheets, which are not only difficult to interpret and manufacture, but also do not contribute significantly to the objective value, as pointed out by Giele et al. (2021). Hence, penalization of low-thickness densities is necessary to obtain meaningful structures, see Fig. 3.Fig. 3Penalization of low-thickness densities (here \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho < 0.1$$\end{document}ρ<0.1) helps generate manufacturable structures with a negligible cost to the performance objective. In this case, the penalization caused merely a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0.26\%$$\end{document}0.26% increase in compliance Penalization of low-thickness densities (here \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho < 0.1$$\end{document}ρ<0.1) helps generate manufacturable structures with a negligible cost to the performance objective. In this case, the penalization caused merely a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0.26\%$$\end{document}0.26% increase in compliance We classify density values in the range \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho \in (0, \rho _\textrm{low})$$\end{document}ρ∈(0,ρlow) as unacceptably thin sheets, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _\textrm{low}$$\end{document}ρlow is a critical value below which we deem the resulting thickness undesirable. There are several approaches to mitigate low-thickness regions in VTTO, either focusing on introducing a selective, SIMP-based penalization (Endress and Zimmermann 2023), projection-based approaches in the regularization step (Groen and Sigmund 2018; Larsen et al. 2018), or utilizing an additional design field (Giele et al. 2021). In the following, we combine the SIMP-based and the projection-based approaches, closely following the logic of standard TO, where sharp black-and-white structures are obtained using both penalization and projection. This is also motivated by findings that using only either the penalization-based or projection-based approaches either does not strictly eliminate low-thickness densities or does not converge to meaningful structures, exhibiting high sensitivity to continuation schemes. Our numerical studies have shown that using a combination of SIMP-based and projection-based approaches leads to stable and effective suppression of low-thickness regions, even with aggressive parameter continuation strategies. SIMP-based suppression As the first component of the strategy for treating low-thickness regions, we consider the penalization-based approach as proposed by Endress and Zimmermann (2023):3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \rho _e^\textrm{VT} = {\left\{ \begin{array}{ll} \rho _e& \text {if } \rho _e \ge \rho _\textrm{low},\\ \left( \frac{\rho _e}{\rho _\textrm{low}}\right) ^{p}\rho _\textrm{low}& \text {else}, \end{array}\right. } \end{aligned}$$\end{document}ρeVT=ρeifρe≥ρlow,ρeρlowpρlowelse,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _e^\textrm{VT}$$\end{document}ρeVT is the selectively penalized density, with the penalty exponent p as in standard TO. The per-element Young’s modulus is then computed as4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} E(\rho _e) = \left[ \rho _e^\textrm{VT}\left[ 1 - \rho _\textrm{min}\right] + \rho _\textrm{min}\right] E_{0}. \end{aligned}$$\end{document}E(ρe)=ρeVT1-ρmin+ρminE0.The penalty factor is chosen to be \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p = 3$$\end{document}p=3. In order to allow for a full design freedom in the initial iterations of optimization, a simple continuation approach is implemented Start with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p = p_\textrm{init} = 1$$\end{document}p=pinit=1.Perform in every iteration: 5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} p^{I+1} = \min (c_p p^I, p_\textrm{max}), \end{aligned}$$\end{document}pI+1=min(cppI,pmax), with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_\textrm{max} = 3$$\end{document}pmax=3.The evolution of the penalized density \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _e^\textrm{VT}$$\end{document}ρeVT during the continuation is visualized in Fig. 4.Fig. 4SIMP-based penalization for the suppression of low-thickness regions for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _\textrm{low} = 0.1$$\end{document}ρlow=0.1, including the continuation for the penalty exponent: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p = 1 \rightarrow 3$$\end{document}p=1→3 Start with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p = p_\textrm{init} = 1$$\end{document}p=pinit=1. Start with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p = p_\textrm{init} = 1$$\end{document}p=pinit=1. Perform in every iteration: 5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} p^{I+1} = \min (c_p p^I, p_\textrm{max}), \end{aligned}$$\end{document}pI+1=min(cppI,pmax), with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_\textrm{max} = 3$$\end{document}pmax=3. Perform in every iteration: 5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} p^{I+1} = \min (c_p p^I, p_\textrm{max}), \end{aligned}$$\end{document}pI+1=min(cppI,pmax), with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_\textrm{max} = 3$$\end{document}pmax=3. SIMP-based penalization for the suppression of low-thickness regions for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _\textrm{low} = 0.1$$\end{document}ρlow=0.1, including the continuation for the penalty exponent: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p = 1 \rightarrow 3$$\end{document}p=1→3 This penalization approach contains a discontinuity at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _e = \rho _\textrm{low}$$\end{document}ρe=ρlow, hence, the derivative of Eq. 3 is not available at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _\textrm{low}$$\end{document}ρlow. Consequently, this approach is mathematically inconsistent regarding gradient computation. However, considering that the numerical treatment involves floating-point arithmetic, the elemental densities never strictly fulfill the condition \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _e = \rho _\textrm{low}$$\end{document}ρe=ρlow. Hence, for any numerical value of the density \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _e$$\end{document}ρe, an inequality condition is sufficient to determine whether the sensitivity should be computed for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _e < \rho _\textrm{low}$$\end{document}ρe<ρlow or otherwise. A continuous version of Eq. 3 can be formulated using, for instance, a sigmoid-type switching function. However, to keep the formulation simple, the current, discontinuous version of Eq. 3 is chosen. Updated projection-based suppression The second component of the strategy to treat low-thickness regions involves a more common approach, namely, a projection function in the regularization step. The projection step employs a form of sigmoid-type smoothed Heaviside projection, as first introduced by Wang et al. (2011). Adapted functions to target only low-thickness regions were proposed in Groen and Sigmund (2018) and Larsen et al. (2018). These approaches, however, use a continuation strategy for both the sharpness \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\beta }$$\end{document}β¯ and threshold \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\eta }$$\end{document}η¯ using predetermined pairs of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{\bar{\beta },\bar{\eta }\}$$\end{document}{β¯,η¯} values. The version of the projection of Larsen et al. (2018) is shown in Fig. 5. The evolution of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\beta }$$\end{document}β¯ parameter is non-monotonic, making it difficult to adjust for various problems. For different low-thickness thresholds, a new set of predetermined pairs of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{\bar{\beta },\bar{\eta }\}$$\end{document}{β¯,η¯} needs to be defined.Fig. 5Projection-based suppression function for low-thickness regions as proposed by Larsen et al. (2018), plotted for each pair of parameters \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{\bar{\beta },\bar{\eta }\}$$\end{document}{β¯,η¯} used in the continuation scheme. The gray region marks the impact of the projection on densities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho > \rho _\textrm{low}$$\end{document}ρ>ρlow Projection-based suppression function for low-thickness regions as proposed by Larsen et al. (2018), plotted for each pair of parameters \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{\bar{\beta },\bar{\eta }\}$$\end{document}{β¯,η¯} used in the continuation scheme. The gray region marks the impact of the projection on densities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho > \rho _\textrm{low}$$\end{document}ρ>ρlow In order to tackle the aforementioned challenges, we adapt the projection function to fulfill the following criteria: (i) the continuation strategy is based purely on changing the projection sharpness \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\beta }$$\end{document}β¯; (ii) the continuation strategy is controlled by a single parameter in the manner \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\beta }^{I+1} = \bar{c}_\beta \bar{\beta }^I$$\end{document}β¯I+1=c¯ββ¯I; (iii) the parameter choice is independent of the low-thickness threshold \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _\textrm{low}$$\end{document}ρlow; (iv) the values \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho > \rho _\textrm{low}$$\end{document}ρ>ρlow should be minimally affected. The criteria (i) and (ii) are formulated in analogy to the smoothed Heaviside projection used for standard TO (Wang et al. 2011). By employing analogous parameter selection and continuation strategies as in standard TO, we point to several practical benefits such as intuitive parameter selection and a clean, consistent code implementation in case both standard TO and VTTO coexist in the same code. Thus, we employ an approach that utilizes two basic functions: the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\beta }$$\end{document}β¯-penalized density \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{\rho }^{\bar{\beta }}$$\end{document}ρ~β¯ and the identity \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{\rho }$$\end{document}ρ~ which are interpolated using a switching function \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S(\tilde{\rho },\bar{\beta },\rho _\textrm{low})$$\end{document}S(ρ~,β¯,ρlow) in the following manner:6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} H^\textrm{LT}(\tilde{\rho },\bar{\beta }) = [1 - S(\tilde{\rho },\bar{\beta },\rho _\textrm{low})] \tilde{\rho }^{\bar{\beta }} + S(\tilde{\rho },\bar{\beta },\rho _\textrm{low}) \tilde{\rho }, \end{aligned}$$\end{document}HLT(ρ~,β¯)=[1-S(ρ~,β¯,ρlow)]ρ~β¯+S(ρ~,β¯,ρlow)ρ~,where the switching function \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S(\tilde{\rho },\bar{\beta },\rho _\textrm{low})$$\end{document}S(ρ~,β¯,ρlow) is a sigmoid function given by7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} S(\rho ,\beta ,\eta ) = \frac{1}{2} \left[ 1 + \tanh \left( \beta \left[ \frac{\rho }{\eta } - \eta ^{1/\beta } \right] \right) \right] . \end{aligned}$$\end{document}S(ρ,β,η)=121+tanhβρη-η1/β.The switching function \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S(\rho ,\beta ,\eta )$$\end{document}S(ρ,β,η) is similar to the known smoothed Heaviside projection introduced by Wang et al. (2011) with the difference that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S(\rho ,\beta ,\eta )$$\end{document}S(ρ,β,η) is a rather standard sigmoid-type function with a projection range of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\rho }\in (0,1)$$\end{document}ρ¯∈(0,1), as opposed to one exactly passing through points \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{0,0\}$$\end{document}{0,0} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{1,1\}$$\end{document}{1,1}. The characteristic modification to the sigmoid function is the threshold shifting part: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left[ \rho /\eta - \eta ^{1/\beta } \right] $$\end{document}ρ/η-η1/β, where incorporation of the sharpness parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta $$\end{document}β enables threshold shifting behavior similar to the approach of Larsen et al. (2018), without the need to explicitly define the continuation values for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\eta $$\end{document}η. In Fig. 6, the updated projection function is plotted for a selection of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta $$\end{document}β values.Fig. 6Updated version of the projection-based suppression function for low-thickness regions. The new version exploits only the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\beta }$$\end{document}β¯ parameter for the continuation in a robust, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _\textrm{low}$$\end{document}ρlow-independent way. Densities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho > \rho _\textrm{low}$$\end{document}ρ>ρlow are minimally affected Updated version of the projection-based suppression function for low-thickness regions. The new version exploits only the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\beta }$$\end{document}β¯ parameter for the continuation in a robust, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _\textrm{low}$$\end{document}ρlow-independent way. Densities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho > \rho _\textrm{low}$$\end{document}ρ>ρlow are minimally affected Starting with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\beta }= 1$$\end{document}β¯=1, the projection \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H^\textrm{LT}(\tilde{\rho },\bar{\beta })$$\end{document}HLT(ρ~,β¯) is an identity function \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H^\textrm{LT}(\tilde{\rho },1) = \tilde{\rho }$$\end{document}HLT(ρ~,1)=ρ~, since both basic functions \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{\rho }^1$$\end{document}ρ~1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{\rho }$$\end{document}ρ~ are identity functions. An increase in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\beta }$$\end{document}β¯ gradually penalizes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{\rho }^{\bar{\beta }}$$\end{document}ρ~β¯ while at the same time the switching function \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S(\tilde{\rho },\bar{\beta },\rho _\textrm{low})$$\end{document}S(ρ~,β¯,ρlow) becomes sharper. Most notably, densities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{\rho }> \rho _\textrm{low}$$\end{document}ρ~>ρlow are only marginally affected while the behavior for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{\rho }< \rho _\textrm{low}$$\end{document}ρ~<ρlow is similar to the original projection in Fig. 5, with only \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta $$\end{document}β being updated. As for the SIMP-based suppression, we employ a continuation strategy in an analogous manner: Start with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\beta }= \bar{\beta }_\textrm{init} = 1$$\end{document}β¯=β¯init=1.Perform in every iteration: 8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \bar{\beta }^{I+1} = \min (\bar{c}_\beta \bar{\beta }^I, \bar{\beta }_\textrm{max}). \end{aligned}$$\end{document}β¯I+1=min(c¯ββ¯I,β¯max). Start with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\beta }= \bar{\beta }_\textrm{init} = 1$$\end{document}β¯=β¯init=1. Start with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\beta }= \bar{\beta }_\textrm{init} = 1$$\end{document}β¯=β¯init=1. Perform in every iteration: 8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \bar{\beta }^{I+1} = \min (\bar{c}_\beta \bar{\beta }^I, \bar{\beta }_\textrm{max}). \end{aligned}$$\end{document}β¯I+1=min(c¯ββ¯I,β¯max). Perform in every iteration: 8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \bar{\beta }^{I+1} = \min (\bar{c}_\beta \bar{\beta }^I, \bar{\beta }_\textrm{max}). \end{aligned}$$\end{document}β¯I+1=min(c¯ββ¯I,β¯max). Combined suppression Ultimately, we aim to use the SIMP-based and the projection-based approaches to suppress low-thickness regions together. In order to arrange for this, a joint continuation scheme is used in a manner analogous to multi-thickness topology optimization Stankiewicz et al. (2026), where the continuation of p and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\beta }$$\end{document}β¯ is arranged sequentially. This order of sequence has worked very well for various increase rates in multi-thickness TO and, similarly, works well in the proposed low-thickness suppression approach in this work. The continuation scheme is arranged as follows: Start with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p = p_\textrm{init} = 1$$\end{document}p=pinit=1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\beta }= \bar{\beta }_\textrm{init} = 1$$\end{document}β¯=β¯init=1.Perform in every iteration: 9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} p^{I+1} = \min (c_p p^I, p_\textrm{max}), \end{aligned}$$\end{document}pI+1=min(cppI,pmax), where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_\textrm{max} = 3$$\end{document}pmax=3.Once \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p = p_\textrm{max} = 3$$\end{document}p=pmax=3, perform in every iteration: 10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \bar{\beta }^{I+1} = \min (\bar{c}_\beta \bar{\beta }^I, \bar{\beta }_\textrm{max}). \end{aligned}$$\end{document}β¯I+1=min(c¯ββ¯I,β¯max).Although sequential suppression is preferred due to higher stability during optimization, our additional studies have shown that a simultaneous continuation of p and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\beta }$$\end{document}β¯ can arrive at satisfactory designs as well. However, more attention is required for the appropriate selection of the increase rates. Ultimately, in combination with the DGI projection, which is introduced in Sect. 4, the sequential continuation approach is strongly advised. Start with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p = p_\textrm{init} = 1$$\end{document}p=pinit=1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\beta }= \bar{\beta }_\textrm{init} = 1$$\end{document}β¯=β¯init=1. Start with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p = p_\textrm{init} = 1$$\end{document}p=pinit=1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\beta }= \bar{\beta }_\textrm{init} = 1$$\end{document}β¯=β¯init=1. Perform in every iteration: 9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} p^{I+1} = \min (c_p p^I, p_\textrm{max}), \end{aligned}$$\end{document}pI+1=min(cppI,pmax), where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_\textrm{max} = 3$$\end{document}pmax=3. Perform in every iteration: 9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} p^{I+1} = \min (c_p p^I, p_\textrm{max}), \end{aligned}$$\end{document}pI+1=min(cppI,pmax), where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_\textrm{max} = 3$$\end{document}pmax=3. Once \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p = p_\textrm{max} = 3$$\end{document}p=pmax=3, perform in every iteration: 10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \bar{\beta }^{I+1} = \min (\bar{c}_\beta \bar{\beta }^I, \bar{\beta }_\textrm{max}). \end{aligned}$$\end{document}β¯I+1=min(c¯ββ¯I,β¯max). Once \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p = p_\textrm{max} = 3$$\end{document}p=pmax=3, perform in every iteration: 10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \bar{\beta }^{I+1} = \min (\bar{c}_\beta \bar{\beta }^I, \bar{\beta }_\textrm{max}). \end{aligned}$$\end{document}β¯I+1=min(c¯ββ¯I,β¯max). Numerical example In the following, we test the effectiveness of the introduced strategies involving the SIMP-based and the projection-based low-thickness suppression. First, we are interested in how the SIMP-based and projection-based approaches perform alone using our continuation approach. In particular, for the projection-based approach, the proposed continuation strategy from Sect. 3.2, using the increase rates of our choice, is more aggressive compared to the reference method from Fig. 5, for which an interval of 50 iterations between each continuation step is chosen. Ultimately, the combined suppression approach is tested and compared against the pure SIMP-based and projection-based cases. The basis for assessment is the presence of undesired low-thickness densities in the range \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho \in (0.001, \rho _\textrm{low})$$\end{document}ρ∈(0.001,ρlow). The values \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho < 0.001$$\end{document}ρ<0.001 are considered acceptable void. For visualization, we employ a reversed red-to-transparent scale for the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho \in (0.001, \rho _\textrm{low})$$\end{document}ρ∈(0.001,ρlow) range to draw attention to the most undesired densities, namely the values close to the void threshold. The red-to-transparent scale is only applied to the bottom half of the structures, exploiting the symmetry of the designs. In Fig. 7 the resulting structures are shown.Fig. 7Numerical tests for the studied low-thickness suppression approaches. From top to bottom: pure SIMP-based, pure projection-based, and combined suppression. Undesired low-thickness densities in the range \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho \in (0.001, \rho _\textrm{low})$$\end{document}ρ∈(0.001,ρlow) are highlighted using a red-to-transparent scale. Using the combined approach, the amount of low-thickness densities is negligible Numerical tests for the studied low-thickness suppression approaches. From top to bottom: pure SIMP-based, pure projection-based, and combined suppression. Undesired low-thickness densities in the range \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho \in (0.001, \rho _\textrm{low})$$\end{document}ρ∈(0.001,ρlow) are highlighted using a red-to-transparent scale. Using the combined approach, the amount of low-thickness densities is negligible The first row in Fig. 7 depicts the resulting structures for the pure SIMP-based suppression for varying low-thickness thresholds \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _\textrm{low} = \{0.05,0.1,0.25\}$$\end{document}ρlow={0.05,0.1,0.25}. The maximum penalty value \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_{\max }$$\end{document}pmax and the increase rate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$c_p$$\end{document}cp are kept constant, as their variation revealed no significant design changes. SIMP-based suppression effectively forms structures that respect the low-thickness threshold, similarly to standard TO, due to penalized stiffness and, therefore, undesirably high compliance in the low-thickness regime. However, due to the blurring filter, low-thickness transition regions of width proportional to the filter radius occur between solid and void. This indicates the need for a projection-based treatment after applying the filter. In the second row, the resulting structures for the pure projection-based approach are shown. This approach, using the function from Fig. 5, is successfully employed in literature, however, using a less aggressive continuation strategy that spans approximately 300–400 optimization iterations. In this work, we test a more aggressive continuation scheme that usually spans over 100–150 iterations, which naturally risks decreased optimization stability. The results revealed just that. Even for the low-thickness threshold of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _\textrm{low} = 0.05$$\end{document}ρlow=0.05, the final designs contain large regions with undesired low-thickness densities. Hence, for the pure projection-based study, we focused on the influence of the final projection sharpness \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\beta }_{\max }$$\end{document}β¯max and the increase rate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{c}_\beta $$\end{document}c¯β to find out whether an improvement can be achieved. Unfortunately, variations of these parameters did not yield significant changes. Testing the original approach from Fig. 5, but using our continuation setup, revealed similar behavior. The large presence of undesired low-thickness densities can be justified by the fact that low-thickness densities are actually still desired by the optimizer. In the physical sense, they do not come at an additional cost in terms of compliance as in the case of the SIMP-based approach. Moreover, the projection does not approach the near-perfect sharpness of the Heaviside function, as would be the case for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\beta }_{\max } \rightarrow \infty $$\end{document}β¯max→∞. Therefore, there is a small range of densities for which low-thickness densities are possible, which pushes the optimizer to find that range if it minimizes the objective function. The higher the projection sharpness, the more noisy the low-thickness regions, while still being largely present. The observations from the pure SIMP-based and projection-based approaches naturally suggest a combined approach to fully eliminate low-thickness regions in a stable manner. While the SIMP-based approach informs the optimizer that low-thickness densities are undesired to minimize the objective, the projection-based approach can assist with eliminating the filter-induced low-thickness solid-to-void transitions. The last row in Fig. 7 shows the resulting structures employing the combined approach. We chose to showcase the designs for various low-thickness thresholds, as the influence of various increase rates, penalties, and sharpness parameters was minimal. The resulting designs are satisfactory with an almost complete absence of low-thickness densities. By using relatively aggressive continuation parameters, the robustness of the combined approach was proven. Moreover, the projection sharpness of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\beta }_{\max } = 25$$\end{document}β¯max=25 was sufficient for effective elimination of the low-thickness densities in the solid-to-void transition. The remaining challenge: low-thickness edges The combined low-thickness suppression approach has proven to be a reliable solution to eliminate low-thickness regions. However, a characteristic occurrence in such designs is the formation of structural edges with the exact thickness of the threshold parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _\textrm{low}$$\end{document}ρlow. This is a natural consequence of the combined regularization using a blurring filter and low-thickness projection, necessary to deal with checkerboards and mesh dependencies. Such low-thickness edges are closely inspected in Fig. 8.Fig. 8Influence of the common regularization scheme, including the blurring filter and the low-thickness projection, on the formation of structural edges. Three cases with various filter radii \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r \in \{h_2, 1.5h_2, 2h_2\}$$\end{document}r∈{h2,1.5h2,2h2} are compared Influence of the common regularization scheme, including the blurring filter and the low-thickness projection, on the formation of structural edges. Three cases with various filter radii \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r \in \{h_2, 1.5h_2, 2h_2\}$$\end{document}r∈{h2,1.5h2,2h2} are compared To understand the influence of the PDE filter, in particular the choice of the filtering radius, we compare three cases: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r \in \{h_2, 1.5h_2, 2h_2\}$$\end{document}r∈{h2,1.5h2,2h2}. Naturally, the optimizer tries to form structural edges of thicknesses also above \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _\textrm{low}$$\end{document}ρlow, as can be seen in the raw (unfiltered) density field, see the first column in Fig. 8. We determined two critical locations where the influence of density regularization on the formation of structural edges is clearly visible. Close-up views of these regions for both the raw densities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho $$\end{document}ρ and regularized densities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\rho }$$\end{document}ρ¯ reveal a substantial difference in the sharpness of the structural edges. The optimizer attempts to form a sharp structural edge of thickness \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho = 1$$\end{document}ρ=1 near the Dirichlet boundary condition. Interestingly, as the filter radius increases, the structural edge in the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho $$\end{document}ρ field gets sharper, suggesting that the optimizer more strongly prevents the formation of a blurred low-thickness edge. A better intuition for the low-thickness edges can be obtained with the help of 3D visualization by a simple extrusion of the finite elements in the out-of-plane direction, proportionally to their thicknesses \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\rho }$$\end{document}ρ¯. The structural edges are clearly formed by density values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$~ \rho _\textrm{low}$$\end{document}ρlow, with a decreasing density slope for an increasing filter radius. Clearly, low-thickness edges are a filtering artifact. The current regularization strategy prevents the formation of natural structural edges of any thickness. While this occurrence is not a major hindrance for simple compliance minimization problems, it is highly undesired for problems that contain a singularity-like feature, for instance, the L-beam. The reason for this is the accumulation of stresses in the vicinity of the singularity, which will be amplified if the structural edges can only be of low-thickness \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _\textrm{low}$$\end{document}ρlow. While L-beam-type problems and stress-constrained optimization are beyond the scope of this paper, it is important to consider the bigger picture of VTTO when addressing the formation of structural edges. Density-gradient-informed projection for sharp edges The formation of blurred, low-thickness structural edges in VTTO is inherent to the common regularization scheme involving a blurring filter, like the Helmholtz-type PDE filter. Thus, to address this issue, we propose a projection method to deblur (resharpen) structural edges as an additional step in the regularization workflow. The projection needs to fulfill the following requirements: (i) restore the sharpness of structural edges as in the raw density field \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho $$\end{document}ρ; (ii) target structural edges and minimally affect the interior of the structure; (iii) use an intuitive \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta $$\end{document}β-like parameter to control the intensity of deblurring. A targeted effect on the structural edges and not the interior of the structure is essential to preserve the “variable thickness” character of the method. In the following, we first introduce the terms used in the proposed projection. The neighborhood of an element e encompasses elements within the distance of the filter radius r:11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} N_r(e) = \{ e' \in \Omega _e \mid \Vert \textbf{x}_{e} - \textbf{x}_{e'} \Vert \le r \}, \end{aligned}$$\end{document}Nr(e)={e′∈Ωe∣‖xe-xe′‖≤r},where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e'$$\end{document}e′ indicates the elements that are neighbors of e and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\textbf{x}_{e}$$\end{document}xe is the center of element e. As the selected filter radius for our numerical tests of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r = 1.5h_2$$\end{document}r=1.5h2 is based on the initial mesh size of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h_2 = 0.25$$\end{document}h2=0.25, and we utilize adaptive meshing, in Fig. 9 we show the neighborhood \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_r(e)$$\end{document}Nr(e) for a case with adaptive meshing of up to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h_4$$\end{document}h4 and with a uniform mesh, using \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h = h_2$$\end{document}h=h2. For uniform meshes, it is recommended to use a minimum neighborhood radius of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r_{\min } = 1.5h$$\end{document}rmin=1.5h to ensure the immediate neighborhood of e exists, even when the filter radius is smaller. For \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r < h$$\end{document}r