Hybrid high-order approximations of div-curl systems on domains with general topology
Jérémy Dalphin, Jean-Pierre Ducreux, Simon Lemaire, Silvano Pitassi · 2025
Read Hybrid high-order approximations of div-curl systems on domains with general topology by Jérémy Dalphin, Jean-Pierre Ducreux, Simon Lemaire, Silvano Pitassi (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-19616.txt.
Introduction
Jeremy Dalphin, Jean-Pierre Ducreux, Simon Lemaire, and Silvano Pitassi build math methods for a class of field problems on complex shaped domains, in a paper from 2025. The problems, called div-curl systems, come from the study of magnetic fields at rest. Real domains can have holes and handles, which break many standard methods.
The authors design hybrid methods that work on many sided cells at any chosen order. Two forms of the problem are covered, one in terms of the field and one in terms of a linked potential. Proofs that the discrete problems are sound rest on new forms of known inequalities that hold for any topology. Computer tests, some on domains that are not simply connected, check the error behavior in practice.
Catalog record
| CREATOR | Jérémy Dalphin, Jean-Pierre Ducreux, Simon Lemaire, Silvano Pitassi |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 125 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-19616
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Hybrid high-order approximations of div-curl systems on domains with general topology Jérémy Dalphin, Jean-Pierre Ducreux, Simon Lemaire, Silvano Pitassi Abstract We devise and analyze hybrid polyhedral methods of arbitrary order for the approximation of div-curl systems on three-dimensional domains featuring non-trivial topology. The div-curl systems we are interested in stem from magnetostatics, and can either be first-order (field formulation) or second-order (vector potential formulation). The well-posedness of the resulting discrete problems essentially hinges on recently established, topologically generic, hybrid versions of the (first and second) Weber inequalities. Our error analysis covers the case of regular solutions. Leveraging (co)homology computation techniques from the literature, we perform an in-depth numerical assessment of our approach, covering, in particular, the case of non-simply-connected domains. Keywords: Div-curl systems; Polyhedral meshes; Hybrid methods; de Rham cohomology; Computational topology. AMS Subject Classification 2020: 65N12, 14F40, 35Q60. 1 Introduction Let \Omega be a domain in \mathbb{R}^{3} , i.e. a bounded and connected Lipschitz open set of \mathbb{R}^{3} . In the L^{2} framework, the standard (primal) de Rham complex reads: \{0\}\xrightarrow{\mathmakebox{0}}H^{1}(\Omega)\xrightarrow{\mathmakebox{\Grad}}\boldsymbol{H}(\Curl;\Omega)\xrightarrow{\mathmakebox{\Curl}}\boldsymbol{H}(\Div;\Omega)\xrightarrow{\mathmakebox{\Div}}L^{2}(\Omega)\xrightarrow{\mathmakebox{0}}\{0\}, (1) with cohomology spaces \displaystyle\mathfrak{H}^{0} \displaystyle\mathrel{\mathop{:}}={\rm Ker}(\Grad)(/{\rm Im}(0)),\hskip 22.99988pt \displaystyle\mathfrak{H}^{1} \displaystyle\mathrel{\mathop{:}}={\rm Ker}(\Curl)/{\rm Im}(\Grad), (2) \displaystyle\mathfrak{H}^{2} \displaystyle\mathrel{\mathop{:}}={\rm Ker}(\Div)/{\rm Im}(\Curl),\hskip 22.99988pt \displaystyle\mathfrak{H}^{3} \displaystyle\mathrel{\mathop{:}}=({\rm Ker}(0)=)L^{2}(\Omega)/{\rm Im}(\Div). The spaces \mathfrak{H}^{n} have respective dimensions equal to the Betti numbers \beta_{n} , with \beta_{0}=1 (number of connected components of \Omega ), \beta_{1}\in\mathbb{N} (number of tunnels crossing through \Omega ), \beta_{2}\in\mathbb{N} (number of voids encapsulated by \Omega ), and \beta_{3}=0 . In turn, the dual de Rham complex reads: \{0\}\xrightarrow{\mathmakebox{0}}H^{1}_{0}(\Omega)\xrightarrow{\mathmakebox{\Grad_{\,0}}}\boldsymbol{H}_{\boldsymbol{0}}(\Curl;\Omega)\xrightarrow{\mathmakebox{\Curl_{\,\boldsymbol{0}}}}\boldsymbol{H}_{0}(\Div;\Omega)\xrightarrow{\mathmakebox{\Div_{\,0}}}L^{2}(\Omega)\xrightarrow{\mathmakebox{0}}\{0\}, (3) where the operators \Grad_{\,0} , \Curl_{\,\boldsymbol{0}} , and \Div_{\,0} are respectively L^{2} -adjoint to the operators -\Div , \Curl , and -\Grad . The zero subscripts aim to recall that the domains of the adjoint differential operators embed zero boundary conditions. The corresponding homology spaces are \displaystyle\mathfrak{H}^{0}_{0} \displaystyle\mathrel{\mathop{:}}={\rm Ker}(\Grad_{\,0})(/{\rm Im}(0)),\hskip 22.99988pt \displaystyle\mathfrak{H}^{1}_{0} \displaystyle\mathrel{\mathop{:}}={\rm Ker}(\Curl_{\,\boldsymbol{0}})/{\rm Im}(\Grad_{\,0}), (4) \displaystyle\mathfrak{H}^{2}_{0} \displaystyle\mathrel{\mathop{:}}={\rm Ker}(\Div_{\,0})/{\rm Im}(\Curl_{\,\boldsymbol{0}}),\hskip 22.99988pt \displaystyle\mathfrak{H}^{3}_{0} \displaystyle\mathrel{\mathop{:}}=({\rm Ker}(0)=)L^{2}(\Omega)/{\rm Im}(\Div_{\,0}). Importantly, there holds \mathfrak{H}^{n}_{0}=\mathfrak{H}^{3-n} (this is the so-called Poincaré–Lefschetz duality ). Let \mu be a possibly varying coefficient, assumed to be uniformly bounded by above and by below by some positive constants. In what follows, the symbols \perp and \perp_{\mu} respectively denote standard and \mu -weighted L^{2} -orthogonality. Let the sequence of differential operators \{\texttt{Grad},\texttt{Curl},\texttt{Div}\} stand either (i) for \{\Grad,\Curl,\Div\} from the primal de Rham complex ( 1 ), or (ii) for \{\Grad_{\,0},\Curl_{\,\boldsymbol{0}},\Div_{\,0}\} from the dual de Rham complex ( 3 ). In the first case, let \mathfrak{h}^{n}\mathrel{\mathop{:}}=\mathfrak{H}^{n} ; in the second, let \mathfrak{h}^{n}\mathrel{\mathop{:}}=\mathfrak{H}^{n}_{0}=\mathfrak{H}^{3-n} . We aim to study problems of the following form. Given some datum \boldsymbol{j}\in{\rm Ker}(\texttt{Div}) , find \boldsymbol{h}\in{\rm Dom}(\texttt{Curl})\cap{\rm Im}(\texttt{Grad})^{\perp_{\mu}} s.t. \texttt{Curl}(\boldsymbol{h})=\boldsymbol{j}. (5) One can easily verify that Curl , seen as a bounded linear mapping from {\rm Dom}(\texttt{Curl})\cap{\rm Im}(\texttt{Grad})^{\perp_{\mu}} to {\rm Ker}(\texttt{Div}) , is a Fredholm operator, with null space {\rm Ker}(\texttt{Curl})\cap{\rm Im}(\texttt{Grad})^{\perp_{\mu}}\cong\mathfrak{h}^{1} , and defect space {\rm Ker}(\texttt{Div})/{\rm Im}(\texttt{Curl})=\mathfrak{h}^{2} . Thus, the following alternative holds true for Problem ( 5 ): \bullet either {\rm dim}(\mathfrak{h}^{1})={\rm dim}(\mathfrak{h}^{2})=0 , then Problem ( 5 ) admits a unique solution; \bullet in the opposite case, for Problem ( 5 ) to admit a solution, it is necessary that \boldsymbol{j}\in{\rm Im}(\texttt{Curl}) , which amounts, for \boldsymbol{j}\in{\rm Ker}(\texttt{Div}) , to additionally satisfying \boldsymbol{j}\perp\mathfrak{h}^{2} ; the uniqueness of the solution is then recovered by imposing {\rm dim}(\mathfrak{h}^{1}) additional constraints to \boldsymbol{h} (which already satisfies \boldsymbol{h}\perp_{\mu}{\rm Im}(\texttt{Grad}) ). Problem ( 5 ) is prototypical of the first-order formulation (also known as field formulation ) of magnetostatics. In practice, \boldsymbol{j}:\Omega\to\mathbb{R}^{3} is a given electric current density, \mu:\Omega\to\mathbb{R}_{>0} is the magnetic permeability of the medium, and \boldsymbol{h}:\Omega\to\mathbb{R}^{3} is the sought magnetic field. The model can be endowed with either normal or tangential boundary conditions, depending on which version (respectively, ( 1 ) or ( 3 )) of the de Rham complex the sequence of differential operators stems from. We refer the reader to Section 4.1.1 for a somewhat more conventional rewriting of Problem ( 5 ) in the case of normal boundary conditions. Let us now introduce \texttt{Curl}^{\star} , L^{2} -adjoint of Curl . Recall that {\rm Ker}(\texttt{Curl})^{\perp}={\rm Im}(\texttt{Curl}^{\star}) . If the solution \boldsymbol{h}\perp_{\mu}{\rm Im}(\texttt{Grad}) to Problem ( 5 ) (assuming it exists) additionally satisfies \boldsymbol{h}\perp_{\mu}\mathfrak{h}^{1} , then it can be written \boldsymbol{h}=\mu^{-1}\texttt{Curl}^{\star}(\boldsymbol{a}) for some vector potential \boldsymbol{a}\in{\rm Dom}(\texttt{Curl}^{\star}) . Obviously, the vector potential \boldsymbol{a} is non-unique; it is only defined up to an element of {\rm Ker}(\texttt{Curl}^{\star}) (gauge). Gauging out to zero, Problem ( 5 ) equivalently rewrites: find \boldsymbol{a}\in{\rm Dom}(\texttt{Curl}^{\star})\cap{\rm Ker}(\texttt{Curl}^{\star})^{\perp} s.t. \texttt{Curl}\big(\mu^{-1}\texttt{Curl}^{\star}(\boldsymbol{a})\big)=\boldsymbol{j}, (6) and the (unique) solution to this problem does exist as soon as \boldsymbol{j}\in{\rm Ker}(\texttt{Div}) additionally satisfies \boldsymbol{j}\perp\mathfrak{h}^{2} . Problem ( 6 ) is prototypical of the second-order formulation (also known as vector potential formulation ) of magnetostatics. In practice, \boldsymbol{a}:\Omega\to\mathbb{R}^{3} is the magnetic vector potential. When Problem ( 5 ) is endowed with normal (resp. tangential) boundary conditions, Problem ( 6 ) is endowed with tangential (resp. normal) boundary conditions. In the second scenario, part of the boundary conditions in Problem ( 6 ) are actually natural, and encode the tangential prescription on \boldsymbol{h} . For a more conventional rewriting of Problem ( 6 ) in the case of tangential boundary conditions (first scenario), we refer to Section 4.1.2 . In this work, we aim at devising arbitrary-order hybrid polyhedral discretizations of the first- and second-order model problems ( 5 ) and ( 6 ). By definition, hybrid methods only attach unknowns to the faces and to the cells of the spatial partition at hand. Examples of such approaches are the Hybridizable Discontinuous Galerkin (HDG) [ 17 ] , the Weak Galerkin (WG) [ 39 ] , the Hybrid High-Order (HHO) [ 22 , 23 ] , or the non-conforming Virtual Element (ncVE) [ 35 , 5 ] methods. As might be expected, and as was first documented in [ 15 , 16 ] , all these technologies are very tightly connected. The devising of hybrid polyhedral methods for div-curl systems has already been addressed to some extent in the HDG and WG literatures. For second-order models, these contributions include [ 37 , 13 , 12 ] (cf. also [ 24 ] ) for HDG, and [ 36 ] for WG, all of them restricted to trivial topologies. Regarding now first-order models, the only contributions we are aware of are [ 34 ] , in which a primal WG method is introduced, and [ 38 , 9 , 10 ] , in which alternative (primal-dual) WG approaches are studied. In all these contributions, full face polynomial spaces are employed to discretize the vectorial variable. Also, either the case of non-trivial topologies is not covered (like in [ 34 ] ), or it is treated but in a non-robust fashion. Recently, the discretization of both first- and second-order models of magnetostatics has been studied in the HHO context [ 11 ] . Therein, the vectorial variable is sought into trimmed face polynomial spaces, reusing and extending ideas introduced in [ 13 ] (cf. also [ 32 ] ). The contribution [ 11 ] , however, solely covers the case of trivial topologies. Our objective in the present work is to fill this gap. We aim at devising, analyzing, and numerically assessing trimmed HHO methods for (first- and second-order) models set in domains with arbitrary topology. With respect to