Rigorous foundations of the automatic hp-adaptive finite element method for thin-walled structures: convergence, goal-oriented reliability, and locking-free guarantees

The fully automatic hp-adaptive finite element method — which selects element-size refinements and polynomial-order enrichments by maximizing the predicted error reduction per added degree of freedom, with the unknown exact solution replaced by a surrogate from a globally refined mesh — achieves near-exponential convergence on thin-walled structures, yet rests on unproven heuristics. Tews and Rachowicz (2009), who drove this strategy with energy-norm and goal-oriented adaptivity through three-dimensional elasticity of shells, plates, and stiffened frames, explicitly flagged as open: the multidimensional generalization of the one-dimensional optimality principle behind the refinement rule; the replacement of the exact solution by the refined surrogate $u_{h/2,p+1}$; the neglect of the term $G_{hp}-\Pi_{hp}G$ in the goal-oriented error estimate; and the empirical, unproven claim that tangential polynomial order $p \ge 3$ prevents membrane/shear locking. This problem asks for the rigorous foundation of these heuristics: convergence and optimality guarantees (or demonstrated failure modes) for the practical algorithm in multidimensional anisotropic settings, reliability conditions for the neglected-term goal-oriented estimate, and thickness-uniform (locking-free) error conditions for displacement-based high-order shell discretizations. Existing theory covers only model problems — purpose-built hp algorithms in one and two dimensions, contraction results under additional assumptions for elliptic problems, and locking analyses for cylindrical-shell models — while the practical two-grid algorithm on general shells remains without theory.

Background

The finite element method (FEM) approximates boundary-value problems by piecewise polynomial functions on a mesh. Its two refinement levers are the element size $h$ (h-refinement, subdividing elements) and the polynomial degree $p$ (p-enrichment). The hp-version combines both: for solutions with isolated singularities or analytic data — the typical situation in elastic shells, where stress boundary layers of width proportional to the thickness decorate smooth interiors — hp discretizations can converge exponentially in the number of degrees of freedom, whereas any fixed-degree h-method converges only algebraically. In three-dimensional (3D) solid elasticity of thin-walled structures (shells, plates, stiffened frames and their junctions) the hp idea is especially attractive because a single high-order element through the wall thickness can represent the thickness behavior while high tangential orders resolve layers and joints; the enabling technology, including arbitrary-order shape functions and hanging nodes (mesh irregularities where a refined element meets an unrefined neighbor, kept conforming by constraining the shape functions), is developed in the two monographs of Demkowicz and coworkers.

Making hp refinement automatic is harder. The fully automatic hp strategy — implemented for thin-walled structures by Tews and Rachowicz (2009) within that framework — runs a two-grid ("second mesh") loop: from the solution on the current mesh it predicts, for every candidate local h- or p-modification, the reduction of the squared (energy- or goal-) error per added degree of freedom, selects a near-optimal sequence of modifications on element edges, faces, and interiors (with an hp-reconciliation step enforcing mesh regularity), and repeats. Two heuristics carry the whole construction. First, the true error being reduced is unknown, so the exact solution $u$ is replaced by a surrogate $u_{h/2,p+1}$ computed on a globally refined mesh (halved elements, enriched orders); trusting it is a saturation assumption — that the refined space reduces the error by a definite factor — whose validity and robustness in $p$ is unproven in 3D. Second, the selection rule generalizes a genuine optimality principle from one dimension: classical one-dimensional hp theory (Babuška and Guo) shows that optimal hp meshes exist and can be built greedily; nothing comparable is known for multidimensional meshes with hanging nodes, where the rule is a heuristic.

The method is also used goal-orientedly. Engineering quantities of interest (q.o.i.) — Tews and Rachowicz use averaged displacements, membrane forces, and bending moments on small mid-surface patches, written as $H^1$-continuous linear functionals of the form $F(u)=\int_{\Delta S} n_i(x)u_i(x)w(x)\,dS$ with a smooth weight $w$ — are controlled through a dual (adjoint) problem whose solution $G$ weights the residual. Their practical estimate neglects the term $G_{hp}-\Pi_{hp}G$, the difference between the finite-element dual solution and its coarse-space interpolation; whether this neglect is safe is assumed, not proven.

Finally, thin shells threaten any displacement-based discretization with locking. A shell's bending response carries far less strain energy than its membrane/shear response, so under-resolved polynomial spaces stiffen the model unphysically (membrane/shear locking) and the discretization error need not stay bounded as the thickness $d \to 0$. Classical analyses show that the p-version escapes locking in model plate problems (Suri, Babuška and Schwab) and in specific cylindrical-shell models (Leino and Pitkäranta; Gerdes, Matache and Schwab), that reduced-constraint hp variants mitigate locking (Suri), and that a priori hp estimates hold for hierarchical plate/shell models (Cho and Oden); but no general theorem identifies the polynomial orders, element proportions, or mesh alignments that make displacement-based 3D hp shell discretizations locking-free. Empirically, Tews and Rachowicz observed on the cylindrical, spherical, and conical shell problems they tested that tangential order $p \ge 3$ removed visible locking in their tests.

For pure h-refinement the convergence and optimality theory of adaptive FEM (AFEM) is now essentially complete, and partial hp results exist: a purpose-built hp algorithm was proved convergent and instance-optimal (i.e., optimal for each particular solution, not only in an averaged or classwise sense) in one and two space dimensions (Canuto, Nochetto, Stevenson and Verani), and guaranteed contraction was proved for equilibrated-flux-driven hp refinement under additional refinement assumptions (Daniel, Ern, Smears and Vohralík; Daniel and Vohralík). None of this covers the practical surrogate-driven, per-degree-of-freedom automatic loop — least of all in 3D, on anisotropic thin geometry, with goal-oriented error control. The question below asks precisely for that missing foundation.

Problem Statement

Provide the rigorous foundation that Tews and Rachowicz (2009) explicitly left open for the fully automatic hp-adaptive finite element method for thin-walled (shell/plate) structures — or demonstrate its limits by concrete failure cases. The target is the algorithm as actually implemented in that work: refinement candidates are scored by comparing the coarse-space solution with the globally refined surrogate $u_{h/2,p+1}$; edge, face, and element h- and p-modifications are selected to maximize the predicted reduction of the squared energy- or goal-error per added degree of freedom; mesh regularity is enforced by hp-reconciliation; and the goal-oriented estimate neglects the term $G_{hp}-\Pi_{hp}G$.

1. Convergence and optimality of the practical refinement loop: prove that the loop converges (error contraction per adaptive step, or convergence to the exact solution) for 3D linear-elasticity problems with thin-walled geometry of the kind the method targets, and characterize its optimality (error versus total degrees of freedom relative to best hp approximation) — thereby justifying the multidimensional generalization of the one-dimensional optimality principle on which the selection rule rests — and establish the validity and $p$-robustness, in three dimensions, of the saturation assumption underlying the practical replacement of the exact solution by the refined surrogate $u_{h/2,p+1}$.

2. Reliability of the goal-oriented estimate: establish conditions — on the quantity of interest, dual-solution regularity, mesh, and polynomial orders — under which the neglected term $G_{hp}-\Pi_{hp}G$ is bounded by a computable factor times the estimated error, so that the goal-oriented estimate is reliable.

3. Locking-free conditions: determine the conditions on tangential polynomial order $p$, through-thickness discretization, element aspect ratio, and shell geometry/load class under which the displacement-based 3D hp discretization error is bounded uniformly in the thickness — in particular, examine whether the empirical observation that tangential order $p \ge 3$ removed visible membrane/shear locking on the tested cylindrical, spherical, and conical shell benchmarks admits a rigorous thickness-uniform justification.

These three targets cover the theoretical questions the source explicitly flagged as open. They are independently posed: resolving one does not require resolving the others, and a concrete failure case refutes the corresponding claim. Target 1 is conjunctive, asking together for convergence, optimality, and the saturation assumption underlying the surrogate $u_{h/2,p+1}$; target 3 likewise asks for the general thickness-uniform conditions, of which the $p \ge 3$ observation is the specific benchmark instance. Solver scalability and automatic mesh generation, also flagged in the source's outlook, are engineering development directions and are outside the scope of this problem.

References

1. R. Tews, W. Rachowicz (2009). Application of an automatic hp adaptive Finite Element Method for thin-walled structures. Computer Methods in Applied Mechanics and Engineering 198(21-26):1967-1984. DOI: 10.1016/j.cma.2008.12.041, https://doi.org/10.1016/j.cma.2008.12.041
2. L. Demkowicz (2006). Computing with hp-Adaptive Finite Elements, Volume 1: One and Two Dimensional Elliptic and Maxwell Problems. Chapman & Hall/CRC. DOI: 10.1201/9781420011685, https://doi.org/10.1201/9781420011685
3. L. Demkowicz, J. Kurtz, D. Pardo, M. Paszyński, W. Rachowicz, A. Zdunek (2007). Computing with hp-Adaptive Finite Elements, Volume 2: Frontiers: Three Dimensional Elliptic and Maxwell Problems with Applications. Chapman & Hall/CRC. DOI: 10.1201/9781420011692, https://doi.org/10.1201/9781420011692
4. I. Babuška, B. Q. Guo (1992). The H, P and H-P Version of the Finite Element Method: Basic Theory and Applications. ICASE/DTIC report. DOI: 10.21236/ada260197, https://doi.org/10.21236/ada260197
5. M. Suri, I. Babuška, C. Schwab (1995). Locking effects in the finite element approximation of plate models. Mathematics of Computation 64:461-482. DOI: 10.1090/S0025-5718-1995-1277772-6, https://doi.org/10.1090/S0025-5718-1995-1277772-6
6. Y. Leino, J. Pitkäranta (1994). On the membrane locking of h-p finite elements in a cylindrical shell problem. International Journal for Numerical Methods in Engineering 37:1053-1070. DOI: 10.1002/nme.1620370611, https://doi.org/10.1002/nme.1620370611
7. K. Gerdes, A. M. Matache, C. Schwab (1998). Analysis of Membrane Locking in hp FEM for a Cylindrical Shell. ZAMM — Journal of Applied Mathematics and Mechanics 78(10):663-686. DOI: 10.1002/(SICI)1521-4001(199810)78:10<663::AID-ZAMM663>3.0.CO;2-G, https://doi.org/10.1002/%28SICI%291521-4001%28199810%2978%3A10%3C663::AID-ZAMM663%3E3.0.CO;2-G
8. M. Suri (1997). A reduced constraint hp finite element method for shell problems. Mathematics of Computation 66:15-30. DOI: 10.1090/S0025-5718-97-00806-5, https://doi.org/10.1090/S0025-5718-97-00806-5
9. J. R. Cho, J. T. Oden (1996). A priori error estimations of hp-finite element approximations for hierarchical models of plate- and shell-like structures. Computer Methods in Applied Mechanics and Engineering 132:135-177. DOI: 10.1016/0045-7825(95)00985-X, https://doi.org/10.1016/0045-7825(95)00985-X
10. H. Hakula (2014). hp-boundary layer mesh sequences with applications to shell problems. Computers & Mathematics with Applications 67:899-917. DOI: 10.1016/j.camwa.2013.03.007, https://doi.org/10.1016/j.camwa.2013.03.007
11. C. Canuto, R. H. Nochetto, R. Stevenson, M. Verani (2017). Convergence and optimality of hp-AFEM. Numerische Mathematik 135:1073-1119. DOI: 10.1007/s00211-016-0826-x, https://doi.org/10.1007/s00211-016-0826-x (preprint arXiv:1503.03996)
12. C. Canuto, R. H. Nochetto, R. Stevenson, M. Verani (2017). On p-robust saturation for hp-AFEM. Computers & Mathematics with Applications 73:2004-2022. DOI: 10.1016/j.camwa.2017.02.035, https://doi.org/10.1016/j.camwa.2017.02.035
13. P. Antonietti, C. Canuto, M. Verani (2019). An Adaptive hp-DG-FE Method for Elliptic Problems: Convergence and Optimality in the 1D Case. Communications on Applied Mathematics and Computation 1:309-331. DOI: 10.1007/s42967-019-00026-9, https://doi.org/10.1007/s42967-019-00026-9
14. P. Daniel, A. Ern, I. Smears, M. Vohralík (2018). An adaptive hp-refinement strategy with computable guaranteed bound on the error reduction factor. Computers & Mathematics with Applications 76:967-983. DOI: 10.1016/j.camwa.2018.05.034, https://doi.org/10.1016/j.camwa.2018.05.034
15. P. Daniel, M. Vohralík (2023). Guaranteed contraction of adaptive inexact hp-refinement strategies with realistic stopping criteria. ESAIM: Mathematical Modelling and Numerical Analysis 57:329-366. DOI: 10.1051/m2an/2022082, https://doi.org/10.1051/m2an/2022082
16. M. S. Mommer, R. Stevenson (2009). A Goal-Oriented Adaptive Finite Element Method with Convergence Rates. SIAM Journal on Numerical Analysis 47(2):861-886. DOI: 10.1137/060675666, https://doi.org/10.1137/060675666
17. V. Helml, M. Innerberger, D. Praetorius (2023). Plain convergence of goal-oriented adaptive FEM. Computers & Mathematics with Applications 147:130-149. DOI: 10.1016/j.camwa.2023.07.022, https://doi.org/10.1016/j.camwa.2023.07.022
18. G. Zboiński (2010). Adaptive hpq finite element methods for the analysis of 3D-based models of complex structures. Part 1. Hierarchical modeling and approximations. Computer Methods in Applied Mechanics and Engineering 199:2913-2940. DOI: 10.1016/j.cma.2010.06.003, https://doi.org/10.1016/j.cma.2010.06.003
19. G. Zboiński (2010). Unresolved Problems of Adaptive Hierarchical Modelling and hp-Adaptive Analysis within Computational Solid Mechanics. In: Advanced Structured Materials, Springer, pp. 111-145. DOI: 10.1007/978-3-642-05241-5_7, https://doi.org/10.1007/978-3-642-05241-5_7
20. G. Zboiński (2013). Adaptive hpq finite element methods for the analysis of 3D-based models of complex structures. Part 2. A posteriori error estimation. Computer Methods in Applied Mechanics and Engineering 267:531-565. DOI: 10.1016/j.cma.2013.08.018, https://doi.org/10.1016/j.cma.2013.08.018
21. Ł. Miazio, G. Zboiński (2020). A Posteriori Detection of Numerical Locking in hpq-Adaptive Finite Element Analysis. Applied Sciences 10:8247. DOI: 10.3390/app10228247, https://doi.org/10.3390/app10228247
22. Ł. Miazio, G. Zboiński (2023). A posteriori boundary layer detection and resolution in hpq-adaptive finite element methods for 3D-based hierarchical plate and shell models. Computers & Mathematics with Applications 129:50-68. DOI: 10.1016/j.camwa.2022.11.010, https://doi.org/10.1016/j.camwa.2022.11.010
23. V. Darrigrand, D. Pardo, T. Chaumont-Frelet, I. Gómez-Revuelto, L. E. Garcia-Castillo (2020). A painless automatic hp-adaptive strategy for elliptic problems. Finite Elements in Analysis and Design 178:103424. DOI: 10.1016/j.finel.2020.103424, https://doi.org/10.1016/j.finel.2020.103424
24. F. V. Caro, V. Darrigrand, J. Alvarez-Aramberri, E. Alberdi, D. Pardo (2022). A painless multi-level automatic goal-oriented hp-adaptive coarsening strategy for elliptic and non-elliptic problems. Computer Methods in Applied Mechanics and Engineering 401:115641. DOI: 10.1016/j.cma.2022.115641, https://doi.org/10.1016/j.cma.2022.115641
25. T. Sluzalec, R. Grzeszczuk, S. Rojas, W. Dzwinel, M. Paszyński (2022). Quasi-optimal hp-finite element refinements towards singularities via deep neural network prediction. arXiv:2209.05844, https://arxiv.org/abs/2209.05844
26. T. Chaumont-Frelet, Z. Dong, G. Gantner, M. Vohralík (2026). On p-robust convergence and optimality of adaptive FEM driven by equilibrated-flux estimators. arXiv:2603.08887, https://arxiv.org/abs/2603.08887