Abstract
This review paper addresses the development of numerical modeling of electromagnetic fields in geophysics with a focus on recent finite element simulation. It discusses ways of estimating errors of our solutions for a perfectly matched modeling domain and the problems that arise from its insufficient representation. After a brief outline of early methods and modeling approaches, the paper mainly discusses the capabilities of the finite element method formulated on unstructured grids and the advantages of local hrefinement allowing for both a flexible and largely accurate representation of the geometries of the multiscale geomaterial and an accurate evaluation of the underlying functions representing the physical fields. In summary, the accuracy of the solution depends on the geometric mapping, the choice of the mathematical model, and the spatial discretization. Although the available error estimators do not necessarily provide reliable error bounds for our complex geomodels, they are still useful to guide grid refinement. Therefore, an overview of the most common a posteriori error estimators is given. It will be shown that the sensitivity is the most important function in both guiding the geometric mapping and the local refinement.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
Article Highlights

Modern finite element simulations of electromagnetic fields are able to incorporate complex closetoreality geometries in the geomodels

A posteriori error estimators provide efficient tools to guide the adaptive refinement of unstructured meshes

The sensitivity function governs both the level of geometric detailedness and goaloriented mesh refinement
1 Introduction
The goal of electromagnetic (EM) modeling in geophysics is to find mathematical and physical models that represent as well as possible the reality of our perceived geoenvironment and allow us to predict and understand the behavior of EM fields in space and time. As a matter of fact, these models are always an abstract and incomplete representation of this reality. They therefore contain a multitude of sources of error and are subject to a limited range of validity. Particularly, the geometry of our physical models in the Earth sciences is extremely challenging due to its multiscale nature.
On a small scale, we may end up in microscopic, possibly selfsimilar material structures (Pape et al. 1982, 1998a, b; Weiss et al. 2020) where a variety of porescale and innersurface electrochemical processes are usually not covered anymore by our chosen mathematical model. Electric conductivity can become complexvalued, frequency dependent and nonlinear. Moreover, texture or heterogeneity below a certain spatial scale can appear as anisotropy, leading to a matrix representation of electromagnetic parameters in the equations to be discretized (Weidelt 1999; Li and Pek 2008; Yan et al. 2016; Wang et al. 2018; Li et al. 2021). Much of this is subject to current petrophysical research, which gives us new insights into the fundamental properties of these parameters (e.g., Börner et al. 2013, 2017).
On medium and large scales  and this is actually the main focus of this article  there are sources of error associated with the representation of macroscopic geometry and its discretization. Whereas geometry modeling started with homogeneous and layered halfspaces with flat surface topography and only a handful of parameters in the first half of the twentieth century and up to the end of the 1960s (e.g., Wait 1953; Gosh 1971), we were able to approximate rough 2D and 3D geometries in the 1970s (Jones and Pascoe 1971; Schmucker 1971; BrewittTaylor and Weaver 1976; Dey and Morrison 1979) and 1980s (Scriba 1981; Oristaglio and Hohmann 1984) using rectangular building blocks to construct geometric models in a bricklike and conforming fashion mostly using finite difference (FD) methods. These are simple and easy to handle but sometimes also inefficient, particularly when it comes to adopting to the geometric idiosyncrasies of our georeality. Staircaselike structures associated with rectangular Cartesian grids tend to introduce artifacts into the model response once the configuration of sources and receivers generates sensitivity toward the artificial geometric features. Figure 1 shows an example for the DC resistivity case. Note, however, that this applies in general to EM methods, especially but not exclusively, if they apply point sources or receivers. Find another very illustrative example in fig. 7 of Li and Pek (2008) for the marine magnetotelluric method in 2D, where the rectangular discretization of a triangularly shaped sea mount leads to an erroneous oscillating response for the FD method.
Therefore, more elegant and accurate ways of representing geometry evolved during the last 20 years incorporating unstructured and nonconforming grids or, recently, even meshless approaches (Wittke and Tezkan 2014; Wittke 2017).
Before we start discretizing, however, we need to describe the geometry itself which is a major problem to tackle. We need tools to disassemble the computational domain into ‘watertight’ subdomains with arbitrary shapes. These domains may encompass a large range of spatial scales. Boreholes, mining tunnels, bathymetry, or the topography of the Earth’s surface, for example, can all significantly influence the electromagnetic response within a specific frequency range or time interval at certain locations. This ultimately means that we have to deal with a multiscale issue which requires an appropriate level of geometric detailedness due to the particular sensitivity pattern of the experiment carried out to account for the true physical response once a reliable mathematical model has been chosen. In other words, the geometry has to be mapped onto the model in great detail where the sensitivity of the experiment is high and can be less detailed in regions of low sensitivity. See Zehner et al. (2015) for a variety of approaches to geometry modeling and Börner et al. (2015a, 2015b) for a series of virtual electromagnetic experiments using one geometric model described by nonuniform rational Bsplines (NURBS) for different EM methods.
In the following, we consider the case where the geometry is perfectly matched. Even in this case, however, there remain sources of error. Since there are no closedform solutions of Maxwell’s equations with arbitrarily varying coefficients, discrete solutions inevitably lead to approximation errors, the magnitude of which is generally difficult to estimate even when the computational domain is perfectly geometrically consistent. FE error analyses, which began with the important work of Babuška and Rheinboldt (1978), nowadays provide us with apriori and, even better and practically more relevant, a posteriori error estimation that give us certain means to estimate the accuracy of the computed solution or at least to bring the properties of the used discretization to a desired level (Sect. 3). Particularly, we will see that the mesh refinement is driven by goaloriented error estimators that are basically also governed by the sensitivity function (Sect. 3.3.3).
Very instructive and worth reading reviews on EM modeling have been published by Avdeev (2005), Börner (2010), Everett (2012), Newman (2014), Pankratov and Kuvshinov (2016), and Ren and Kalscheuer (2020) focusing on different aspects of the development of electromagnetic modeling including uncertainty issues and computational efficiency. Here, I will focus on the numerical and computational side of modeling mainly during the last decade where finite element (FE) methods became the methods of choice for many research groups in geoelectromagnetics (geoEM) due to their inherent flexibility and power to represent complex geometries, especially when they are formulated on unstructured grids.
Having appropriately solved the forward problem we then would need to address the geometry in the inverse problem by trying to find a meaningful parametrization for the inverse solution. This, however, goes beyond the scope of this article. The paper much more reviews some attempts to gain control over the error sources in the forward process, particularly with respect to error estimators defined within the theory of FE (Sect. 3). Before doing so, Sect. 2 gives a brief overview how the different discretization methods, mainly FD (Sect. 2.1) and FE (Sects. 2.2 and 2.2.1), have evolved in mathematical research over time. I will then quickly browse through the different types of FE (Sect. 2.2.2) used in our approaches to modeling vector electromagnetic fields and scalar potentials, and discuss the most important types of meshes underlying the discretization schemes with respect to their ability to represent arbitrary geometry. Section 2.2.3 reviews the work that has been done particularly in geoEM where I have organized the subsection according to the chronological evolution of the simulation strategies starting with early nodal and edge elements on structured grids and moving on to their implementation on unstructured and nonconforming grids.
I will not address the inverse problem and efforts to quantify uncertainty in a Bayesian sense. Moreover, I will not discuss the idiosyncrasies of quasistatic, frequency or time domain methods and their particular applications. However, the methods discussed range from quasistatic to diffusive. The main focus is on the spatial part of those partial differential equations that are underlying our electromagnetic problems. Particularly, I will address discretization schemes of the partial differential operators of the divergence and the curl that lead to their numerical representations. I will not address specific time integration approaches (e.g., Druskin and Knizhnerman 1988, 1994; Commer 2003; Commer and Newman 2004; Börner et al. 2008, 2015c; Zimmerling et al. 2018), integral equation techniques (e.g., Raiche 1974; Weidelt 1975; Hohmann 1975; Zhdanov 1988; Wannamaker 1991; Xiong and Kirsch 1992; Avdeev et al. 2002; Pankratov and Kuvshinov 2016; Chen et al. 2021), and will rather briefly touch finite difference methods.
The emphasis of this review is on the FE technique in its different variants and its solid mathematical background that leads to the attempt of quantifying errors in the solution which, in turn, drive the adaptive mesh refinement. Consequently, Sect. 3 starts with some remarks on apriori error estimation in FE. After a brief historical outline of the evolution of error estimation in Sect. 3.1, Sect. 3.2 presents the basics of the variational problem including its weak formulation by means of a simple model problem, which is used to introduce three different popular a posteriori error estimators in Sect. 3.3: the residualbased (Sect. 3.3.1), the recoverybased (Sect. 3.3.2), and the goaloriented error estimators (Sect. 3.3.3). The insights from Sects. 3.2 and 3.3 are subsequently transferred from the scalar model problem to geoEM problems using a typical vectorial frequencydomain secondary field approach in the form of the curl–curl equation for the electric field as an example in Sect. 3.4. Sections 3.4.1 to 3.4.3 summarize the main work done in the geoEM community for the different types of error estimation, each in a chronological order with illustrative example figures.
There are mainly three types of refinement that are discussed in the literature: h, p and rrefinement. hrefinement stands for the increase of the spatial resolution of the grid by decreasing the size of the grid cells. Cells are earmarked if the estimated local error of the solution exceeds a certain value relative to the largest estimated error and subsequently refined in a way that is briefly outlined in Sect. 4. prefinement denotes the increase of the order p of the basis functions. Finally, rrefinement is very similar to hrefinement and rearranges a given number of grid nodes for a given order p of the basis functions. The latter is rather uncommon in geoEM research and the exclusive prefinement is also quite rarely used. It leads to the class of spectral FE methods (e.g., Zhu et al. 2020; Weiss et al. 2022), which will not be discussed further here. In the practice of geoEM applications, hrefinement with moderate order basis functions (i.e., linear to second or third order) has become essentially the prevailing method, with quadratic basis functions as the best compromise between accuracy and numerical effort (Schwarzbach et al. 2011; Grayver and Kolev 2015). Examples of joint h and prefinement are described by Pardo et al. (2006) for the 2D simulation of EM loggingwhiledrilling measurements on structured grids and by Pardo et al. (2011) in the context of marine 2.5D CSEM simulation.
Section 4 closes with some thoughts on alternative strategies of representing spatial structures and an appendix gives some more details on the energy norm and the derivation of the variational problem.
2 Discretization Techniques and Meshes
This section is dedicated to the discretization methods and their development over time. Thereby, I restrict myself as far as possible to FE methods. Due to the practical relevance, I briefly browse through FD techniques and the closely related finite volume (FV) methods. Concerning the grids I will address structured (rectangular, Cartesian, tetrahedral, hexahedral), unstructured (tetrahedral) and nonconforming grids (quad and octrees). In general, all methods can be formulated on all grids. Nevertheless, e.g., FD methods are mainly associated with structured rectangular (2D) or hexahedral (3D) grids with only a few exceptions [e.g., triangular grids for DC resistivity by Erdogan et al. (2008), Demirci et al. (2012) and Penz et al. (2013) going back to Liszka and Orkisz (1980)]. FE methods were formulated on both structured and unstructured grids. However, they develop their full strength only on unstructured grids due to the geometric flexibility of the mesh, the solid theoretical foundation and the resulting elaborated error analysis.
This section is divided into a brief outline of FD methods in geoEM (Sect. 2.1) followed by a more comprehensive discussion of the FE technique and its application to geoEM research (Sect. 2.2).
2.1 Finite Differences and Finite Volumes
In FD, a very early reference goes back to a work done by the GermanAmerican mathematician Richard Courant (1888–1972) in Göttingen in which he presented basic FD formulations (Courant et al. 1928). Even earlier Lewis Fry Richardson (1881–1953), a British meteorologist and peace researcher, came up with FD formulations in meteorology (Richardson 1922) but was not successful right away due to the limited computational facilities (i.e., performing calculations using paper and pencil) and the resulting inaccurate results in predicting the weather for a day. This alleged failure was interpreted by his peer group as the impossibility of calculating weather forecasts. A major breakthrough was achieved only decades later when moderate electronic computational facilities became available. Yee (1966) introduced a staggered grid cell to account for the different locations at which the electric and magnetic fields are calculated when a curl operator is discretized (Fig. 2).
Also in geoEM, FD has a long history and important developments were made from the beginning of the 1970s (see the references in Sect. 1). In the 1990s, computing facilities became more powerful and iterative equation solvers like the preconditioned conjugate gradient method were able to solve large systems of equations with moderate memory requirements and reasonable run times (Mackie et al. 1993; Wang and Hohmann 1993; Druskin and Knizhnerman 1994; Spitzer 1995; Newman and Alumbaugh 1995; Smith 1996b; Newman and Alumbaugh 1997; Spitzer and Wurmstich 1999). In the end, millions of rectangular building blocks were used to construct geometric models in a bricklike and conforming fashion. FD methods are still widely used today for practical data interpretation (Xiong et al. 2000; Sasaki 2001; Siripunvaraporn et al. 2002; Davydycheva et al. 2003; Sasaki 2004; Streich 2009; Streich and Becken 2011; Egbert and Kelbert 2012; Grayver et al. 2013). They are implemented in forward and/or inverse modeling schemes and compete with upcoming FE techniques. The problem of incorporating topography using FD is usually solved by either applying appropriate boundary conditions at the staircaseshaped earth’s surface or assigning high resistivities to the air cells. Resolution of both geometry and grid is usually controlled only by the modeler’s experience. Systematic studies on the grid quality are rare (see, e.g., Ingerman et al. 2000), and a rigorous investigation of the influence of the stairlike topography is still missing.
Using a singularity removal procedure according to Lowry et al. (1989), more geometric flexibility on structured grids can be achieved by detaching the source positions from the grid nodes (Spitzer et al. 1999; Spitzer and Chouteau 1997, 2003). Singh and Dehiya (2023) use a coarse mesh to calculate boundary conditions for a smaller fine mesh to improve the condition number of the system matrix.
In addition to these wellknown FD techniques used in EM, relatively few FV approaches were reported, some on structured rectangular grids (Haber et al. 2000; Haber and Ascher 2001; Mulder 2006; Plessix et al. 2007) and others on more flexible meshes like quad/octrees or unstructured tetrahedral meshes. The latter allow for local refinement and will be discussed in more detail later in the context of FE formulations. Haber et al. (2007) presented FV approaches using octrees for forward and inverse problems in EM. Parallel inversion of largescale airborne timedomain EM data on multiple octree meshes for FV or FE were outlined by Haber and Schwarzbach (2014). FV implementations on 3D unstructured tetrahedral/Voronoi grids for a totalfield Aphi Coulombgauged formulation (i.e., the magnetic vector potential \(\varvec{A}\) and the scalar electric potential \(\phi\)) were made by Jahandari and Farquharson (2015).
2.2 Finite Elements
In this section, FE techniques will be illuminated in more detail. For this purpose, we look at the historic development of FE techniques at first and briefly sketch the underlying concepts of variational problems leading to the weak formulation of the problem where equalities are typically defined in the sense of inner products with test functions. Because it usually takes a while for mathematical innovations to find their way into the fields of application I have included the use of FE in geoEM in a separate Sect. 2.2.3.
2.2.1 A Short Historical Overview of the Development of the Finite Element Method
The variational problem dates back to 1696, when Johann Bernoulli (1644–1748) formulated the socalled brachistochrone problem, in which a curve AMB was sought that gives the shortest possible time for a body to slide from A to B under gravity, where A and B are two fixed points on a vertical plane. In 1744, Euler formulated a general solution to the variational problem in the form of a differential equation that was a decade later justified by Joseph Louis de Lagrange (1736–1813). The origins of the FE method date back to the early nineteenth century when Karl Heinrich Schellbach (1805–1892), a German mathematician, solved a minimum surface area problem with some steps typical for FE (Schellbach 1851). Later, in the early twentieth century, Walter Ritz (1878–1909), a Swiss mathematician and physicist in Göttingen, minimized a quadratic functional in a finitedimensional functional space (Ritz 1909). Whereas Ritz used eigen functions, Boris Grigoryevich Galerkin (1871–1945) and Ivan Grigoryevich Bubnov (1872–1919), two Russian mathematicians and engineers, later used polynomials or trigonometric functions to solve engineering problems. Their method is called the method of weighted residuals, Galerkin or GalerkinBubnov method (Galerkin 1915). Again, Richard Courant used ansatz functions with small support (hat functions) for the first time in the midtwentieth century (Courant 1943). From the 1950s onward, the engineering community discovered FE techniques. In mechanical engineering, solids were disassembled into a finite number of elements and the displacements were calculated under given loads in the nodes of the FE. First practical applications came up in the aerospace and automotive industry (Turner et al. 1956; Clough 1960). In the 1960s, there was support coming from the Math community and very important monographs by Zienkiewicz and Cheung (1967) and Strang and Fix (1973) appeared. Furthermore, the computer program NASTRAN, that was originally developed for NASA, was commercialized by the MacNealSchwendler Corporation at the end of the 1960s. The method was extended to vector functions that are particularly important for the solution of electromagnetic problems by Raviart and Thomas (1977) and Nédélec (1980). Very good and comprehensive text books on the FE technique for Maxwell’s equations were published by Jin (2002) and Monk (2003). An excellent historical review is given by Gander and Wanner (2012).
While initially FE methods in geoEM were formulated on structured grids just like FD methods, they were later mostly used in conjunction with unstructured triangular (2D) or tetrahedral (3D) meshes which develop their strengths mainly with respect to the flexible adaptation to arbitrary geometries. Since nodal or Lagrange elements enforce continuity of the scalar function over element boundaries, they were only applicable to those cases where components of the electromagnetic fields or potentials behave in such a way. As we know from EM theory, some components like the normal component of the electric field \(\varvec{E}\) or the tangential component of the magnetic field \(\varvec{H}\) are discontinuous at parameter jumps. In this case, the simulation with Lagrange elements becomes erroneous, which was not remedied until the introduction of vector FE in the late 1970s/early 1980s (Raviart and Thomas 1977; Nédélec 1980).
2.2.2 Types of Finite Elements
In the following, I will introduce several types of FE, but will discuss only three of them in more detail because of their particular importance to EM modeling. These are Lagrangian, Nédélec, and Raviart–Thomas FE. In Table 1, there is a compilation of various FE types including the type of basis functions they are associated with (scalar, vector, matrix).
Lagrange elements (Fig. 3, lefthand subplot) are chosen if a piecewise smooth scalar function v as, e.g., in the case of the scalar electric potential is considered that lives in the Sobolev space \({\mathcal{H}}^{1}(\Omega ):= \left\{ v\in L^{2} (\Omega ):\nabla v\in [L^{2}(\Omega )]^{3}\right\}\) with \(L^{2}(\Omega ):= \left\{ v:\Omega \rightarrow {\mathbb{C}}^{3}:\int _{\Omega } v^{2} dx < \infty \right\}\), i.e., v is squareintegrable and its gradient exists. Lagrange elements are also called nodal elements because their degrees of freedom (dof) live at the nodes of the mesh. They enforce continuity of v across element boundaries. Using 2D triangular meshes and linear basis functions (\(p=1\)), three dof are associated with a Lagrange element (\(n=3\)). Increasing the order of the basis functions to \(p=2\) (quadratic), the number of dof n rises to 6 where the additional dof are associated with the center of each edge. In case of three dimensions and tetrahedral meshes, \(n=4\) for \(p=1\) and \(n=10\) for \(p=2\).
The most important element type for EM modeling in 3D is the Nédélec FE (Nédélec 1980) (Fig. 3, righthand subplot). For a piecewise polynomial to be \({\mathcal{H}} (\text{curl})\)conforming, the tangential component of \({{\varvec{v}}}\) must be continuous. \({{\varvec{v}}}\in {\mathcal{H}} (\text{curl};\Omega ):= \left\{ {\varvec{v}}\in [L^{2} (\Omega )]^{3}:\nabla \times {\varvec{v}}\in [L^{2} (\Omega )]^{3}\right\}\), i.e., \({\varvec{v}}\) is squareintegrable and its curl exists. In case of \(p=1\), they are also called edge elements because their vectorial basis functions \({\varvec{v}}\) are situated along the edges of the elements. Consequently, \(n=3\) in 2D and \(n=6\) in 3D. For \(p=2\), additional dof are associated with the element face centers giving \(n=8\) in 2D and \(n=20\) in 3D.
The third and final type of element introduced here is the mixed FE according to Raviart and Thomas (1977) (Fig. 3, center subplot). It guarantees the normal component of \({\varvec{v}}\) to be continuous and is mainly used for fluid flow and transport modeling. In geoEM, it is important in conjunction with the regularization of the inverse problem (Schwarzbach and Haber 2013; Blechta and Ernst 2022). The corresponding Sobolev space is \({\mathcal{H}} ({\text{div}};\Omega ):= \left\{ {\varvec{v}}\in [L^{2} (\Omega )]^{3}:\nabla \cdot {\varvec{v}}\in L^{2} (\Omega )\right\}\), i.e., \({\varvec{v}}\) is squareintegrable and its divergence exists.
2.2.3 Finite Elements in Geoelectromagnetics
In this section, I will review the course that FE modeling has taken in geoEM starting with nodal elements on structured grids in the 1970s, edge elements on structured grids in the 1990s, nodal elements on unstructured grids in the 2000s and Nédélec elements of different order on unstructured grids from the end of the 2000s until today. Note that all approaches presented in this section are not subject to any kind of quantitative error estimation and use meshes defined by the modeler. Yet, these meshes show regions of refinement usually around sources and/or receivers and/or at specific geometric features. But these refined regions are identified through the experience of the modeler in conjunction with a mesh generator that serves for mesh quality (see Sect. 3.4.4). It is not until Sect. 3 that I introduce different kinds of error estimation and present work that employs these techniques for the adaptive refinement of grids.
2.2.3.1 Structured Grids and Nodal Elements
There were a number of early works based on structured rectangular/triangular/tetrahedral grids using nodal FE. As early as 1971, Coggon (1971) published a secondary field 2D EM/DC/IP approach for a line source on a small structured triangular grid (Fig. 4). Reddy et al. (1977) reported the application of FE to the 3D MT and magnetovariational (MV) methods and already recognized that by enforcing the continuity of the normal component of the electric field, an erroneous transition zone is created in the vicinity of conductivity jumps. Later, Pridmore et al. (1981) used FE on a rectangular grid for the discretization of the 3D equation of continuity in a DC resistivity application. The 2D MT method was tackled by Wannamaker et al. (1987), and Gupta et al. (1989) presented a 3D timedomain (TD) code on a rectangular grid. 2.5D EM simulations for a horizontal electric dipole (HED) were carried out by Unsworth et al. (1993). The same year Livelybrooks (1993) came up with 3D nodal element simulations for a secondary planewave and electric dipole field. He introduced socalled discontinuity terms to correct for the errors resulting from the violation of the continuity conditions. The 3D DC resistivity case was formulated on a structured tetrahedral grid by Sasaki (1994). Everett and Schultz (1996) used a 3D nodal Aphi formulation on tetrahedrals for the simulation of induction phenomena in a heterogeneous conducting sphere excited by an external source current. A 3D secondary field Coulombgauged Aphi approach to induction logging was performed by Badea et al. (2001) for controlledsource EM (CSEM) in cylindrical coordinates using linear nodal basis functions on tetrahedra. Li (2002) developed a 2D simulation code for MT in anisotropic electric media using triangular grids. Shortly thereafter, Li and Spitzer (2002) compared linear FE and 7stencil FD simulations in 3D for the DC case on rectangular hexahedral grids and came up with the conclusion that for this particular setup, the two approaches were comparable in their performance. The accuracy was slightly higher with FE, however, they were also a bit more expensive in numerical terms. The investigated FE approach gives 27 entries per row of the system matrix whereas it is only 7 for FD. Later Li and Spitzer (2005) extended the DC approach to cases with arbitrary anisotropy. Erdogan et al. (2008) presented 2D DC simulations on structured triangular grids with simple topography using both FD and FE.
2.2.3.2 Structured Grids and Edge Elements
Already Sugeng (1998) came up with a 3D edge FE code on structured hexahedral grids. Mitsuhata and Uchida (2004) used Lagrangian and edge elements for discretizing a 3D TOmega formulation (with \({\varvec{T}}\) as the electric vector potential and \(\Omega\) as the magnetic scalar potential) on a hexahedral grid. Isoparametric hexahedral grids were used by Nam et al. (2007) for 3D MT with simple topography. They implemented edge elements and solved the resulting system of equations using a biconjugate gradient method with a Jacobian preconditioner. Farquharson and Miensopust (2011) solved the 3D MT problem on rectangular hexahedral grid and applied a divergence correction scheme according to Smith (1996a) within the stabilized conjugate gradient solver BICGSTAB. For this purpose, a scalar electric potential is calculated from the divergence of the electric current density to correct for numerical inaccuracies of the electric field. This correction potential is repeatedly applied to the electric field after a number of steps within the iterative equation solver. The electric field was discretized with linear Nédélec elements, the scalar electric potential with nodal elements. Another 3D CSEM secondary field approach using edge elements on hexahedral grids solved by a multifrontal direct solver was outlined by Da Silva et al. (2012). Cai et al. (2014) used edge elements on isoparametric hexahedral grids for 3D CSEM in anisotropic environments with simple topography. Finally, Kordy et al. (2016) described a 3D magnetotelluric inversion scheme including topography using deformed hexahedral edge FE and direct solvers parallelized on symmetric multiprocessor computer architectures.
2.2.3.3 Unstructured Grids and Nodal Elements
While FE and FD on structured grids did not seem to differ significantly in performance, the breakthrough came with the introduction of unstructured meshes. A very early work goes back to Rücker et al. (2006) for 3D DC secondary field simulations on unstructured tetrahedral meshes including realistic topography inferred from a digital terrain model (Fig. 5).
Due to the topography, there is no simple analytical solution to the reference field so that it was accurately calculated on a very fine grid. Only Penz et al. (2013) showed later that this was actually not necessary if the noflow boundary condition at the surface is adopted correspondingly. Blome et al. (2009) reported on advances in threedimensional geoelectric forward modeling by calculating the primary field by fast multipolebased boundary element methods and applying infinite elements at boundaries. Udphuay et al. (2011) carried out threedimensional resistivity tomography in extreme coastal terrain using unstructured tetrahedral grids. A Lagrange element secondary field Aphi approach to 3D CSEM modeling on unstructured tetrahedral meshes and anisotropic conductivities was presented by Puzyrev et al. (2013). The inversion of 2D MT and radiomagnetotellurics (RMT) data using FE on unstructured grids was communicated by Özyildirim et al. (2017). An example of an underground mining tunnel that was incorporated into a 3D DC resistivity inversion was given by Pötschke (2017). The tunnel surface was captured with great accuracy using a laser scanner technology. Electrodes were placed along the tunnel wall. It turned out that, in regions of high sensitivity, it is extremely important to match the true locations of the electrodes very accurately and to incorporate the tunnel geometry with a high level of detailedness to avoid artifacts in the inversion results (Fig. 6).
2.2.3.4 Unstructured Grids and Nédélec Elements
Very early work using Nédélec elements in 3D on unstructured tetrahedral grids and secondorder basis functions goes back to Franke et al. (2007a, 2007c) and Kütter et al. (2010). The latter already included the topography and bathymetry around Stromboli volcano, Italy, using a digital elevation model (Fig. 7). Um et al. (2012) addressed a 3D iterative firstorder Nédélec FE timedomain method formulated on unstructured tetrahedral meshes. To account for the ineffective time integration through time stepping, they introduced an adaptive time step doubling. A 3D CSEM modeling algorithm using edge elements on unstructured tetrahedral grids was published by Mukherjee and Everett (2011). They included both the magnetic permeability and electrical conductivity for a secondary field Aphi approach. An adaptation to geometry or topography was, however, not performed. Schwarzbach and Haber (2013) reported a FEbased inversion for timeharmonic electromagnetic problems using unstructured tetrahedral meshes and edge elements. Nédélec elements on 3D unstructured tetrahedral grids were used by Um et al. (2013) for a frequency domain Efield formulation. The system of linear equations was solved by the QMR iterative solver with an ILU preconditioning. A 3D Coulombgauged Aphi formulation on unstructured grids using Nédélec and nodal elements and electric and magnetic sources was reported by Ansari and Farquharson (2013) and Ansari et al. (2017). The latter apply their scheme to a realistic geophysical scenario using a model of the Ovoid ore deposit at Voisey’s Bay, Labrador, Canada, for helicopterborne EM (Fig. 8). Usui (2015) presented a 3D MT inversion code based on linear edge elements and unstructured tetrahedral grids written in C++. Jahandari et al. (2017) compare two FE and three FV schemes on unstructured tetrahedral grids. They conclude that both FE and FV schemes are comparable in terms of accuracy and computational resources. However, the FE schemes are slightly more accurate but also more expensive than the FV schemes. Cai et al. (2017) used an anisotropic total field approach for the CSEM method on a 3D unstructured tetrahedral grid. It was implemented on a parallel computer architecture using linear Nédélec elements and solved by a multifrontal direct solver. Käufl et al. (2018) investigated the influence of 3D topography on the magnetotelluric (MT) response using highorder Nédélec FE. Note that they have formulated the problem on locally refined octrees. A 3D marine CSEM inversion scheme including anisotropic conductivities was presented by Wang et al. (2018). The C++ code uses an unstructured tetrahedral grid Nédélec forward modeling code based on the FEMSTER library (Schwarzbach et al. 2011; Castillo et al. 2005). A 3D secondary field marine CSEM solution with linear Nédélec elements on a tetrahedral unstructured mesh was reported by CastilloReyes et al. (2018). They used the PETSc library to solve the resulting system of linear equations and refined the meshes due to the skin effect. The code was written in Python and implemented on highperformance computing (HPC) facilities. A 3D secondary field frequency domain parallel higherorder Nédélec FE approach for the marine CSEM forward problem with isotropic conductivities discretized on tetrahedral unstructured meshes was reported by CastilloReyes et al. (2019). Their mesh refinement was roughly guided by the skin effect. Rochlitz et al. (2019) presented a further total and secondary field frequency domain approach for CSEM based on the opensource library FEnics using linear and quadratic edge elements on unstructured tetrahedral grids for the E, H, and Aphi formulations. An opensource 3D magnetotelluric forward modeling code on unstructured meshes using higher order edge elements and being optimized for both HPC and nonHPC architectures was outlined by GallardoRomero and RuizAguilar (2022), Finally, Blechta et al. (2022) presented a generic Matlabbased forward modeling and inversion library at the 25th International Workshop on Electromagnetic Induction in the Earth held in Çeşme, Turkey, in 2022 including higher order Lagrange, Nédélec, and Raviart–Thomas elements covering basically all frequency domain geoEM methods including DC resistivity and induced polarization (IP).
3 Error Estimation
In this section, I am going to review the most important a posteriori error estimators used in the geoEM community. Basically, a priori error analysis is useful too because it provides information about the convergence behavior of the solution. However, it is very expensive in terms of numerical work. In general, the FE theory states that an arbitrarily exact solution can be obtained if the discretization width h goes toward zero. There are two very important and fundamental theorems on which this statement is based. According to the theorem of Lax–Milgram (Lax and Milgram 1954), the solution exists and is unique. Céa’s lemma (Céa 1964) states that the Galerkin method yields the best solution with respect to the underlying FE subspace. The error of the numerical solution can theoretically be estimated in terms of the \(L^{2}\)norm \(\left\Vert e_{h}\right\Vert _{L^{2}}=\left\Vert u  u_{h}\right\Vert _{L^{2}}\le C N^\alpha\), where C is a constant depending on the regularity of the solution k, the polynomial degree p of the basis functions, the modeling domain \(\Omega\), and the triangulation \({\mathcal{T}}\) but does not depend on the exact solution u itself and the number N of degrees of freedom. The asymptotic rate of convergence (or simply convergence rate) is then defined by
(Babuška and Aziz 1972; Strang and Fix 1973). To estimate the convergence rate in practice, solutions must be computed for a series of refinements and, with \(h \rightarrow 0\), the numerical effort becomes very large (FrankeBörner 2012). Therefore, while such error analyses are informative, they are of limited use in practice.
In contrast, a posteriori error estimators, which indicate the magnitude of the error and can theoretically estimate error bounds, are very helpful and numerically cheap. A posteriori error estimators are generally based on the convergence theory and require local convergence, which is not always the case as the work of FrankeBörner et al. (2013) shows. This has farreaching consequences, since in cases of highly inhomogeneous discretization and complex highcontrast parameter models (as is usually the case in geoEM modeling), neither the convergence rate nor the error bounds can be reliably predicted. According to Grätsch and Bathe (2005) all the error estimators generally do not provide guaranteed bounds so that the estimation technique most useful is probably the method that works efficiently in general analyses. They actually conclude that the practical error estimation techniques do not provide mathematically proven bounds on the error and need to be used with care. Also, Schwarzbach and Haber (2013) note that the meshdependent convergence rate determines the accuracy of the numerical solution only up to a meshindependent constant factor. This factor is essentially determined by the smoothness of the true solution which is generally difficult to quantify as it varies with problemdependent parameters like, for example, discontinuities in electrical conductivity or magnetic permeability, source type and geometry. Still error estimators are the best option we have for getting an idea of the accuracy of our solutions. They are therefore definitely helpful tools when it comes to adaptive meshing.
After a short historical outline, we are going to look at a simple model problem in the form of Poisson’s equation and subsequently transfer the findings to the induction (curl–curl) problems that we are facing in geoEM research.
3.1 A Short Historical Overview of Error Estimators
The beginning of modern a posteriori error estimation dates back to the end of the 1970s and beginning of the 1980s when Babuška and Rheinboldt (1978) defined the error in the energy norm (see Appendix 1). In the 1980s, element residual methods were established by Demkowicz et al. (1984) and Bank and Weiser (1985) where typically the weak residual was exploited to derive quantitative error measures. At the end of 1980s a new type of error estimator, the recoverybased method, was formulated taking into account the behavior of the gradient of the solution on a patch of elements (Zienkiewicz and Zhu 1987, 1992a, b). In the early 1990s, the basic principles of error estimation were established for scalar functions. The extension to vector FE was achieved in the mid 1990s for elliptic problems and Raviart–Thomas elements by Braess and Verfürth (1996), for parabolic problems by Eriksson and Johnson (1991) and Maxwell’s equations with edge elements by Beck and Hiptmair (1999). Goaloriented error estimation takes into account the adjoint problem using any of the previous error estimators (Prudhomme and Oden 1999; Oden and Prudhomme 2001). An extensive overview was given by Ainsworth and Oden (2000) and an informative review article by Grätsch and Bathe (2005).
3.2 The Variational Problem and the Weak Formulation
I demonstrate the principles of error estimation in FE using a simple model boundary value problem for a scalar function u with appropriate boundary conditions. The socalled strong formulation of Poisson’s equation reads
and represents an elliptic linear boundary value problem on a domain \(\Omega \in {\mathbb{R}}^{3}\) that is Lipschitzbounded, i.e., the data \(f\in L^{2}(\Omega )\) and \(g\in L^{2}(\Gamma _{\textrm{N}})\) are sufficiently smooth and squareintegrable. \({\varvec{n}}\) denotes the outward normal on the boundary \(\Gamma =\Gamma _{\textrm{D}} \cup \Gamma _{\textrm{N}}\) (split into a Dirichlet and a Neumann part). Multiplying both sides with test functions \(v\in {\mathcal{V}} =\left\{ v\in {\mathcal{H}}^{1}(\Omega ):v=0 {\text{ on }} \Gamma _{\textrm{D}}\right\}\) yields, after some rearrangement, the weak formulation
See Appendix 2 for more details. There are infinitely many functions in \({\mathcal{V}}\); therefore, one restricts oneself to finite dimensional spaces \({\mathcal{V}}_{h}\), which are easier to deal with. \({\mathcal{V}}\) contains the exact solution, i.e. any residual becomes 0. \({\mathcal{V}}_{h}\) is contained in \({\mathcal{V}}\) and therefore functions from \({\mathcal{V}}_{h}\) cannot make the residual vanish. Reducing \({\mathcal{V}}\) to \({\mathcal{V}}_{h} \subset {\mathcal{V}}\) we end up with finding \(u_{h}\in {\mathcal{V}}_{h}\) so that
This is the Galerkin equation with \({\mathcal{V}}_{h}\) as the FE subspace of cellwise polynomial functions of order p over the FE partition \({\mathcal{T}}_{h}\), the latter of which denotes a mesh consisting of individual cells. Galerkin or Rayleigh–Ritz methods are a class of methods for converting a continuous problem in terms of a differential equation in its weak formulation, e.g., according to Eq. (2), into a discrete problem yielding a linear system of equations. The Galerkin method uses the basis functions \(\phi \in {\mathcal{V}}^{h}\) as test functions, a finite weighted sum of which approximates the sought solution \(u_{h}=\sum _{i=1}^{n} u_{i}\phi _{i}\). The coefficient values \(u_{i}\) associated with each test function are determined such that the error between the linear combination of the test functions and the solution is minimized in a chosen norm yielding the linear system of equations
with \(A_{ij}=\displaystyle \int _{\Omega }\nabla \phi _{i}\cdot \nabla \phi _{j} \, {\textrm{d}}\Omega\) defining the elements of the system matrix and \({\varvec{f}}\) as the righthand side vector.
3.3 A Posteriori Error Estimation
In this section, using the scalar boundary value problem introduced above, I briefly outline the principles of three important a posteriori error estimators used in the geoEM community. These are the socalled residualbased, recoverybased and goaloriented error estimators. After that, I will explain how error estimation is applied to Maxwell’s equations and used for our geoEM problems.
3.3.1 ResidualBased Error Estimation
Generally, in a posteriori error estimation we want to estimate the discretization error while not knowing the exact solution u. The error is therefore defined as the difference between the unknown solution u and the FE solution \(u_{h}\): \(e_{h}=uu_{h}\). It satisfies the error representation
with \({\mathscr {R}}_{h}\) as the weak residual. The key property of the Galerkin approach is that the error is orthogonal to the chosen subspaces. Since \({\mathcal{V}}_{h} \subset {\mathcal{V}}\), we can use \(v_{h}\) as a test vector in the original equation. Subtracting the two gives
which is called the Galerkin orthogonality.
Explicit error estimates are determined by the direct computation of interior element residuals and jumps at element boundaries (Babuška and Rheinboldt 1978). The global error representation over the whole domain is therefore split up into the sum over elemental integrals
Integration by parts and rearranging gives
with
as the interior element residual R in element K and J as the jump of the gradient across the element face \(\gamma\):
In the first case where \(\gamma\) is not part of an outer boundary (\(\gamma \not \subseteq \Gamma\)), \(\gamma\) separates two elements \(K_{1}\) and \(K_{2}\). Using interpolation theory and the Cauchy–Schwarz inequality elementwise, this directly leads to a local error indicator for element K
with constants \(c_{1},c_{2}\in {\mathbb{R}}\) and \(h_{K}\) as the circumdiameter of element K. For more details see, e.g., Grätsch and Bathe (2005) and Ainsworth and Oden (2000).
3.3.2 RecoveryBased Error Estimation
Recoverybased error estimators consider an ‘improved’, i.e., generally smoothed gradient of the solution \({\mathscr {M}}[ u_{h} ]\) and compare it to the original unsmoothed gradient. Large differences point toward large errors of the solution and provide some estimate of the true error because, in general, gradients across element boundaries tend to be discontinuous. The a posteriori error estimator according to Zienkiewicz and Zhu (1987, 1992a, 1992b) therefore reads
and constitutes a rather heuristic approach. In practical application, the error estimate is calculated with respect to each element
with \(\eta _{K}^{2}=\Vert \nabla u_{h}^{\star }  \nabla u_{h}\Vert ^{2}_{L^{2}(K)}\) and \(\nabla u_{h}^{\star }\) being a ‘recovered’ gradient in element K. It reads
and can be obtained by smoothing over a patch of elements \(\Lambda\) surrounding element K. \({\varvec{P}}\) is a vector polynomial defined on \(\Lambda\), \(\phi _{i}\) are the basis functions, and \(\xi _{i}\) and n are the coordinates and the number of vertices in element K. The assumption that variations in the gradient indicate errors and that smooth solutions stand for accurate solutions is, however, not strictly valid. Therefore, the method might fail in cases of intrinsically smooth solutions.
Another approach of recoverybased error estimation goes back to Beck and Hiptmair (1999) and is formulated directly for timeharmonic Maxwell’s equations. They determine the magnetic field using basis functions from different Sobolev spaces, i.e., \({\hat{\varvec{H}}}_{\textrm{s}}\in {\mathcal{H}}_{\textrm{curl}}\) and \(\tilde{\varvec{H}}_{\textrm{s}}\notin {\mathcal{H}}_{\textrm{curl}}\) giving the error indicator
where \(\tilde{\varvec{H}}_{\textrm{s}}\) serves as the ‘improved’ quantity in the sense above. The asterisk denotes the complex conjugate here.
3.3.3 GoalOriented Error Estimators
The last type of error estimator we consider here is called a goaldirected error estimator, which means that we take into account not only the physical field being modeled, but also the fact that the field is observed exclusively at a limited number of locations. Basically, we are not interested in the solution being accurate in regions where we have no observations. Consequently, the global energy norm or total error is not necessarily the most important quantity we are interested in.
For understanding the concept of goaloriented error estimation we consider the parallels that exist with respect to the solution of the inverse problem. There, we have functions that, in some way, take into account the source and the receiver locations at the same time. The function is called the sensitivity function, and it is defined by the Jacobian matrix consisting of derivatives of the data with respect to the model parameters. There are several ways to determine its spatial behavior (Spitzer 1998). One is called the sensitivity theorem (Lanczos 1961; Geselowitz 1971) that allows for the determination of the sensitivity function \(\Phi _{lmn}\) by forming an inner product of the solution of the actual physical field \({\varvec{F}}^{l}\) excited by a source at location l and a virtual field \({\varvec{F}}^{m}\) excited by a virtual source at the receiver location m integrated over a volume \(\tau _{n}\) at location n. It reads
In mathematics, the virtual problem is called the adjoint or dual problem. It is solved as an additional auxiliary problem using the original system matrix with a different righthand side. The adjoint field \({\varvec{F}}^{m}\) can be viewed as a weighting function.
The underlying idea of using the solution of the adjoint problem as a weighting function is also the basis of goaloriented error estimation (Ainsworth and Oden 2000). In FE terminology the virtual weighting function is called an influence function and serves for a more accurate determination of the socalled quantity of interest. With goaloriented adaptivity, refinements are directed toward a most beneficial increase of the accuracy of this quantity of interest Q(u). It typically yields fine meshes around sources and receivers. Note that the nature of the reciprocity principle is closely related to selfadjointness that finds expression in a symmetric or Hermitian system matrix.
Let again \(u_{h}\in {\mathcal{H}}_{0}^{1}\) denote the FE approximation of the solution of the model problem
and let \(Q:{\mathcal{V}}\rightarrow {\mathbb{R}}\) be a bounded, linear functional with \(w^{Q}\) as the solution of the adjoint problem
Then the error in estimating Q(u) using the FE approximation \(Q(u_{h})\) is
where e and \(e^{Q}\) are the errors of the FE approximation of u and \(w^{Q}\), respectively, and \(\Vert e^{Q}\Vert \le \eta ^{Q}\) and \(\Vert e\Vert \le \eta\). This means the true solution of the quantity of interest is bounded by \(Q(u_{h})\eta \eta ^{Q}\le Q(u)\le Q(u_{h})+\eta \eta ^{Q}\). The errors of the primal and dual problems e and \(e^{Q}\) can be estimated using any of the error estimators presented above or possibly also others not mentioned here.
A very illustrative example is given by Ainsworth and Oden (2000) for our model problem (1) where the function
is defined as the right hand side f and homogeneous boundary conditions are given by
The solution of function (7) displays steep gradients as can be seen in Fig. 9a. Using a global energy norm the original rectangular mesh is strongly refined within the area of these gradients (Fig. 9b) using a nonconforming mesh to achieve the given target accuracy. The goaloriented algorithm refines also around the given point \({\varvec{x}}_{0}=(0.6,0.4)\) (Fig. 9c) referring to the quantity of interest Q(u) to reach the target accuracy with less numerical effort and a steeper convergence rate.
3.4 Error Estimation in Geoelectromagnetics
In this section, we are going to transfer the findings from the general case above to geoEM starting with Maxwell’s classical equations
and discussing a typical geoEM case in terms of the curl–curl equation for the secondary electric field in the frequency domain. We assume a timeharmonic factor \(e^{i\omega t}\) (see, e.g., Schwarzbach et al. 2011; Grayver and Bürg 2014; Grayver 2015) yielding
which is a type of vector Helmholtz equation subject to, e.g., a homogeneous Dirichlet boundary condition as stated in Eq. (14). Equation (13) is derived from Maxwell’s equations by taking the curl of Faraday’s law stated in Eq. (10) and substituting the resulting curl of the magnetic field by Ampère’s law Eq. (9). Furthermore, we have considered the constitutive relations \(\varvec{B}=\mu \varvec{H}\), \(\varvec{D}= \varepsilon \varvec{E}\) and Ohm’s law \({\varvec{j}}=\sigma \varvec{E}\) implying the material to be linear and isotropic. In order, \(\varvec{B}, \varvec{H}, \varvec{D}\), and \(\varvec{E}\) are the complexvalued EM fields in terms of the magnetic flux density, the magnetic field, the dielectric displacement and the electric field. Note that the lower case letters in Eqs. (9) to (12) refer to the corresponding fields in the time domain. \(\varvec{E}^{p}\) is the primary electric field associated with a background conductivity model \(\sigma ^{p}\). \(\varvec{E}^{s}\) is the secondary electric field due to the anomalous conductivity \(\sigma ^{a}=\sigma \sigma ^{p}\). \(\mu\) is the scalar magnetic permeability, \(\varepsilon\) the scalar electric permittivity, \({\varvec{j}}\) the current density, \(\omega\) the angular frequency, and i the imaginary unit. \(\varvec{D}\) is usually neglected in typical induction problems considered in geoEM but has to be taken into account for sufficiently high frequencies especially in resistive environments.
Now, Eq. (13) is multiplied by vectorial test functions \({\varvec{v}} \in {\mathcal{H}}_{0}({\text{curl}},\Omega ):=\left\{ {\varvec{v}}\in {\mathcal{H}}({\text{curl}};\Omega ):n \times {\varvec{v}}=0{\text{ on }}\partial \Omega \right\}\) and integrated over the computational domain \(\Omega\) giving the weak formulation of Eq. (13)
We again apply integration by parts and convert the integrals into sums over the elements and faces. In the manner of Eqs. (3) and (4), the elemental residual and jumpbased terms boil down to
with elemental residuals
and jumps over elemental faces
where \(h_{K}\) is a measure of the size of element K (e.g., the circumdiameter of a cell), e is an interior face of element K with \(e\cap \Gamma =\emptyset\), \(h_{e}\) is the circumdiameter of e, and \({\varvec{n}}_{e}\) is the outward normal on it. Brackets denote the jump of the quantity across element boundaries. The residualbased term \(\eta _{R,K}^{2}\) vanishes if both field equation and its divergence are exactly fulfilled. The jumpbased term \(\eta _{J,K}^{2}\) vanishes if the tangential components of the magnetic field and the normal component of the current density are continuous. The continuity of these quantities is not guaranteed by the weak formulation.
In most cases, error estimation in geoEM has been used in conjunction with unstructured 2D triangular and 3D tetrahedral grids. However, there were also some cases where quad and octrees have been used as we will see below.
3.4.1 ResidualBased Error Estimators in GeoEM
A 2D MT residualbased adaptive FE unstructured triangular grid approach was described by Franke et al. (2007b). The 2D and 3D MT boundary value problem has been investigated by Franke et al. (2007a), FrankeBörner et al. (2011), and FrankeBörner (2012) in terms of (1) the electric field (2) the magnetic field, (3) the magnetic vector potential and the electric scalar potential (\({\varvec{A}}\phi\)), (4) the magnetic vector potential only, and (5) the anomalous magnetic vector potential in a secondary field formulation using a residualbased adaptive unstructured tetrahedral discretization with nodal and vector elements. FrankeBörner (2012) also did extensive research on the apriori convergence behavior of these FE solutions.
3.4.2 RecoveryBased Error Estimators in GeoEM
Li and Key (2007) and Li and Constable (2007) used nodal FE and recoverybased refinement strategies for the 2D secondaryfield marine CSEM case on triangular unstructured grids (Fig. 10). Ren and Tang (2010) introduced recoverybased error estimators for the 3D DC resistivity case. The 3D secondaryfield CSEM case was tackled by Schwarzbach et al. (2011) using a recoverybased adaptive technique according to Beck and Hiptmair (1999) on unstructured tetrahedral grids and higher order Nédélec elements (Fig. 11). The code allows for topography and arbitrary conductivity, magnetic permeability and dielectric permittivity. Another 3D recoverybased adaptive FE modeling code for marine CSEM with seafloor topography on the basis of secondary magnetic vector potentials was outlined by Ye et al. (2018).
3.4.3 GoalOriented Error Estimators in GeoEM
Key and Weiss (2006) used a goaloriented gradient recovery operator for the adaptive refinement of an unstructured triangular grid in 2D MT modeling. Adaptive FE modeling of 2D magnetotelluric fields in general anisotropic media using a goaloriented gradient recovery operator for the refinement of an unstructured triangular grid was published by Li and Pek (2008). The 2D marine CSEM case was treated in a similar manner by Li and Dai (2011). A parallel goaloriented recoverybased adaptive FE method for 2.5D EM was outlined by Key and Ovall (2011). Three different a posteriori goaloriented residualbased error estimators were investigated by Ren et al. (2013) for the 3D MT case (Fig. 12). These are based on the nonzero residuals of the electric field and the continuity of both the normal component of the current density and the tangential component of the magnetic field across interior elemental interfaces. The FE formulation uses linear Nédélec elements on unstructured tetrahedral grids with arbitrary conductivity, magnetic permeability and dielectric permittivity.
Another goaloriented adaptive unstructured tetrahedral grid approach was outlined by Ren and Tang (2014) for the 3D DC resistivity case using secondary fields and boundary conditions according to Penz et al. (2013) but driving the refinement by a superconvergent patch recoverybased a posteriori error estimator. A residualbased a posteriori error estimator for 3D MT on unstructured nonconforming hexahedral meshes (octrees) was presented by Grayver and Bürg (2014) using direct solvers or conjugate gradient type methods preconditioned with an auxiliary space technique based on highly efficient algebraic multigrid methods. The code is based on the opensource software deal.II (Bangerth et al. 2007), and parallelization was established using the message passing interface MPI. Grayver and Kolev (2015) reported a 3D EM adaptive highorder FE method using fully parallel and distributed robust and scalable linear solvers based on optimal blockdiagonal and auxiliary space preconditioners. Adaptivity is implemented through residualbased goaloriented adaptive mesh refinement (Fig. 13).
A 3D inversion for MT based on a Nédélec unstructured grid forward operator and goaloriented residualbased adaptivity was presented by Grayver (2015). Another 2D inversion code for MT and CSEM called MARE2DEM was published by Key (2016) where a recoverybased goaloriented adaptivity scheme was implemented on unstructured triangular grids. Likewise on unstructured triangular grids, adaptive FE modeling of DC resistivity in 2D generally anisotropic structures was described by Yan et al. (2016) using a gradient recoverybased scheme. The 3D airborne EM case was tackled using a goaloriented jumpbased adaptive edge FE algorithm implemented by Yin et al. (2016) on unstructured tetrahedral grids. Anisotropic conductivities were incorporated in the 3D MT problem by Liu et al. (2018) using linear Nédélec FE on unstructured tetrahedral grids. Adaptivity was achieved by a goaloriented residualbased error estimator. The resulting system of linear equation was solved using direct solvers from the MUMPS library (Amestoy et al. 2000). Recently, Rulff et al. (2021) reported a 3D CSEM total electric field forward operator in the frequency domain using firstorder Nedelec elements and a residualbased goaloriented error estimator. And finally, a 3D marine CSEM forward modeling code for general anisotropic structures was presented by Li et al. (2021) using goaloriented residualbased adaptivity and Nédélec FE.
3.4.4 Mesh Generation and Refinement, FE Toolboxes and Equation Solver Libraries
Mesh generators, FE toolboxes, equation solvers and the actual procedure of mesh refinement, referred to in FE terminology as enrichment of the FE space, are beyond the scope of this article. This section will nevertheless present some key features and facilities used in the geoEM community for these purposes. Besides the fineness of the grid, the accuracy of the solution is also influenced by geometric properties of the cells, independent of the function to be modeled. This socalled mesh quality can be steered via different options in the mesh generator. The mesh generator is usually an external tool to set up meshes with appropriate properties and features which affect the mesh quality and, in return, the accuracy of the solution of the PDE. According to Shewchuk (1997), we can generally and very briefly state that

small dihedral angles in cells cause poor conditioning of the system matrix,

large angles cause discretization errors,

fewer elements yield faster solutions, and

more elements yield higher accuracy.
Highquality meshes should not include poorlyshaped elements (needles, wedges, caps, and slivers), should be as small as possible, and should be locally refined with moderate grading from small to large. See Dey et al. (1992), Shewchuk (2002), and Bern et al. (2017) for more details. The mesh refinement itself is a wide field of research of the meshing community and is only perceived as a utility by geoEM modelers. There are a multitude of different criteria and ways to split up elements into smaller ones, each affecting the solution of the PDE in an individual way.
Especially simple is the refinement of rectangular cells in a nonconforming way into quad and octrees. Most widespread for unstructured triangular/tetrahedral meshes is the regular refinement where, for 2D, four new triangles are created from an old one by bisecting each edge (Fig. 14a) or the longestedge bisection that creates two new triangles from an old one (Fig. 14b). The minimum inner angle of the elements can be used as a criterion. According to Bern et al. (2017), another geometric criterion for splitting up a cell is the circumradiustoshortest edge ratio of a triangle or tetrahedron, where the circumradius refers to a circle or sphere through all vertices. Any triangle whose circumradiustoshortest edge ratio is larger than a given value is split up by inserting a new vertex at its circumcenter (Fig. 14c, d). Theoretically, meshing algorithms still cannot guarantee all tetrahedra to include good angles. But, in practice, it is possible to create meshes with sufficient quality even if the process is tedious.
Very often, if the new mesh is not supposed to contain any hanging nodes, it has to be regenerated from scratch, which is timeconsuming. Quad and octrees avoid this circumstance by creating a hierarchical set of meshes but leave the modeler with the treatment of hanging nodes. Hanging nodes appeared as early as the 1930s in a work by Shortley and Weller (1938). There is one approach to refine also unstructured tetrahedral grids by the introduction of hanging edges (the equivalent of hanging nodes) for timedomain EM problems. This technique was presented by Schneider et al. (2022) and is able to rapidly adopt the mesh, e.g., to the temporal change of a decaying electric field.
Sometimes mesh generators are also integrated into FE toolboxes (libraries) that include the assembly of the system matrix. The computation of FE coefficients for unstructured meshes requires a high administrative effort, but is highly automatable once the underlying PDE is identified, the boundary conditions to be implemented are set, and appropriate basis functions as well as the desired mesh type are selected. This is the reason for the increasing number of readytouse and nearly allpurpose FE libraries that can be tailored toward the intended application. 2D and 3D mesh generators or FE toolboxes frequently used in the geoEM community are, e.g., Gmsh (Geuzaine and Remacle 2009), Tetgen (Si 2015), Triangle (Shewchuk 1996), NetGen (Schöberl 1997), libmesh (Kirk et al. 2006), FEniCS (Alnæs et al. 2015), Dune (Blatt et al. 2016), OpenFOAM (Weller et al. 1998), deal.II (Bangerth et al. 2007), and MFEM (Anderson et al. 2021).
Some popular and powerful libraries for the direct or iterative numerical solution of large sparse systems of linear algebraic equations are, e.g., the sharedmemory multiprocessing parallel direct sparse solver library PARDISO (Schenk and Gärtner 2004; PARDISO Solver Project 2023; Bollhöfer et al. 2020), the multifrontal massively parallel sparse direct solver MUMPS (Amestoy et al. 2000), and the portable, extensible toolkit for scientific computation PETSc (Balay et al. 2022).
4 Concluding Remarks
In this paper, I have addressed some important lines of development of geoEM modeling including their historical evolution in mathematics. The key innovations, in my opinion, are the introduction of FE in the context of unstructured grids including their local and adaptive refinement, which allow in an elegant and flexible way to integrate complex multiscale structures, as we find them in the Earth sciences, into our conductivity models. The underlying and comprehensive finite element theory enables us to roughly assess the accuracy of the solution using a posteriori error estimators which, in turn, are mainly employed to adaptively control and optimize the grids, even if they are not able to provide definite and helpful bounds. Fast direct and iterative equation solvers as well as powerful mesh generators are indispensable tools without which we would not be able to obtain solutions for closetoreality models, even if they basically always remain abstract and incomplete.
I have pointed out that the sensitivity function plays an important role when it comes to goaloriented error estimation. It guides the refinement of the meshes with respect to the quantity of interest. The sensitivity is likewise important to determine the level of detailedness of geometric features of the model such as topography, bathymetry or any other interior structure.
There are certainly other strategies to build realistic models, both on the physical and mathematical side. Primarily, it is the technical framework and developments in neighboring scientific fields that initiate and enable streams of development. In our case, computer architectures and new mathematical approaches and algorithms are of the utmost importance. Parallel computer architectures and distributed systems have been with us for a while and we are not yet fully exploiting their possibilities in geoEM.
Moreover, there are alternative approaches to describe structures in a modified form to better represent the multiscale properties. There is, for example, the hierarchical parameterization as introduced by Weiss (2017) and Beskardes and Weiss (2018). It is able to efficiently include one and twodimensional structures such as lines (representing, e.g., boreholes) or surfaces (e.g., for fissures) in the modeling of large volumes via separate integrals over edges and faces of the finite elements. Equivalent materials associated with coordinate transformations allow simplification of complicated structures and mapping to regular grids, which in turn leads to simplification of model handling (Kamm et al. 2020). Wavelet expansions can represent spatial structures with variable accuracy, which can help to reduce the number of free parameters (Nittinger and Becken 2018).
Creating a highquality unstructured tetrahedral mesh that meets all the requirements for solution accuracy and mapping fidelity of the geometry in a practical simulation task takes a significant amount of time and is still always a matter of handson tailoring. On the contrary, FD methods impress by their simplicity and straightforward implementation on conforming grids, which greatly simplifies the overall handling and facilitates techniques such as domain decomposition (see, e.g., Bohlen (2002) for seismic modeling). If we look at the development in consumer electronics, we can, for example, see that higher resolution of screens is not achieved by completely new approaches, but by the huge number of pixels (which are comparable to the parameter cells here). Over the decades, the evolution went from VGA (640\(\times\)480 pixels) over HD (920 \(\times\)1080 pixels) to currently 4K (4096 \(\times\) 2160) without changing the concept of the pixel. 8K and 16K will be reached in the not too distant future. Transferring this to our modeling approaches, one could use simple structured grids for extremely high resolution, if only one can provide enough cells. Of course, this is only possible if there is sufficiently large and adequate computing power. All the techniques we use unfold their strength and power only in connection with the currently available resources. Sometimes moving forward can mean to go back one step first.
References
Ainsworth M, Oden JT (2000) A posteriori error estimation in finite element analysis. In: Pure and applied mathematics. Wiley, New York
Alnæs MS, Blechta J, Hake J et al (2015) The FEniCS project version 1.5. Arch Numer Softw 3(100):9–23
Amestoy P, Duff I, l’Excellent JY (2000) Multifrontal parallel distributed symmetric and unsymmetric solvers. Comput Methods Appl Mech Eng 184(2–4):501–520
Anderson R, Andrej J, Barker A et al (2021) MFEM: a modular finite element methods library. Comput Math Appl 81:42–74
Ansari S, Farquharson CG (2013) 3D finiteelement forward modeling of electromagnetic data using vector and scalar potentials and unstructured grids. Geophysics 79(4):E149–E165
Ansari SM, Farquharson CG, MacLachlan SP (2017) A gauged finiteelement potential formulation for accurate inductive and galvanic modelling of 3D electromagnetic problems. Geophys J Int 210(1):105–129
Avdeev DB (2005) Threedimensional electromagnetic modelling and inversion from theory to application. Surv Geophys 26(6):767–799
Avdeev DB, Kuvshinov AV, Pankratov OV et al (2002) Threedimensional induction logging problems, part I: an integral equation solution and model comparisons. Geophysics 67:413–426
Babuška I, Aziz AK (1972) Survey lectures on the mathematical foundations of the finite element method with applications to partial differential equations. In: Aziz AK (ed) The mathematical foundations of the finite element method with applications to partial differential equations. Academic Press, New York
Babuška I, Rheinboldt WC (1978) A posteriori error estimates for the finite element method. Int J Numer Methods Eng 12(10):1597–1615
Badea EA, Everett ME, Newman GA et al (2001) Finiteelement analysis of controlledsource electromagnetic induction using Coulombgauged potentials. Geophysics 66(3):786–799
Balay S, Abhyankar S, Adams MF et al (2022) PETSc web page. https://petsc.org/. Accessed January 2023
Bangerth W, Hartmann R, Kanschat G (2007) deal.II–a generalpurpose objectoriented finite element library. ACM Trans Math Softw 33(4):24
Bank RE, Weiser A (1985) Some a posteriori error estimators for elliptic partial differential equations. Math Comput 44(170):283–301
Beck R, Hiptmair R (1999) Multilevel solution of the timeharmonic maxwell’s equations based on edge elements. Int J Numer Methods Eng 45(7):901–920
Bern M, Shewchuk JR, Amenta N (2017) Triangulations and mesh generation. In: Goodman JE, O’Rourke J, Toth CD (eds) Handbook of discrete and computational geometry, 3rd edn. Chapman and Hall, London, pp 763–786
Beskardes GD, Weiss CJ (2018) Modelling DC responses of 3D complex fracture networks. Geophys J Int 214(3):1901–1912
Blatt M, Burchardt A, Dedner A et al (2016) The distributed and unified numerics environment, version 2.4. Arch Numer Softw 4(100):13–29
Blechta J, Ernst OG (2022) Efficient solution of parameter identification problems with h^{1} regularization. https://doi.org/10.48550/ARXIV.2209.02815. arXiv:2209.02815
Blechta J, Börner RU, Ernst O et al (2022) A MATLAB FE library for the simulation and inversion of EM problems. In: 25th international workshop on electromagnetic induction in the earth, Çeşme, Turkey, 11–17 September 2022, 1 p
Blome M, Maurer HR, Schmidt K (2009) Advances in threedimensional geoelectric forward solver techniques. Geophys J Int 176(3):740–752
Bohlen T (2002) Parallel 3D viscoelastic finite difference seismic modelling. Comput Geosci 28(8):887–899
Bollhöfer M, Schenk O, Janalik R et al (2020) Stateoftheart sparse direct solvers. In: Grama A, Sameh AH (eds) Parallel algorithms in computational science and engineering. Springer, Cham, pp 3–33
Börner RU (2010) Numerical modelling in geoelectromagnetics: advances and challenges. Surv Geophys 31(2):225–245
Börner RU, Ernst OG, Spitzer K (2008) Fast 3D simulation of transient electromagnetic fields by model reduction in the frequency domain using Krylov subspace projection. Geophys J Int 173:766–780
Börner JH, Herdegen V, Repke JU et al (2013) The impact of CO_{2} on the electrical properties of water bearing porous media—laboratory experiments with respect to carbon capture and storage. Geophys Prospect 61(SUPPL.1):446–460
Börner JH, Bär M, Spitzer K (2015a) Electromagnetic methods for exploration and monitoring of enhanced geothermal systems—a virtual experiment. Geothermics 55:78–87
Börner JH, Wang F, Weißflog J et al (2015b) Multimethod virtual electromagnetic experiments for developing suitable monitoring designs: a fictitious CO_{2} sequestration scenario in Northern Germany. Geophys Prospect 63(6):1430–1449
Börner RU, Ernst O, Güttel S (2015c) Threedimensional transient electromagnetic modelling using rational Krylov methods. Geophys J Int 202:2025–2043
Börner JH, Herdegen V, Repke JU et al (2017) Spectral induced polarization of the threephase system CO_{2}–brine–sand under reservoir conditions. Geophys J Int 208(1):289–305
Braess D, Verfürth R (1996) A posteriori error estimators for the Raviart–Thomas element. SIAM J Numer Anal 33(6):2431–2444
BrewittTaylor CR, Weaver JT (1976) On the finite difference solution of twodimensional induction problems. Geophys J R Astr Soc 47:375–396
Cai H, Xiong B, Han M et al (2014) 3D controlledsource electromagnetic modeling in anisotropic medium using edgebased finite element method. Comput Geosci 73:164–176
Cai H, Hu X, Li J et al (2017) Parallelized 3D CSEM modeling using edgebased finite element with total field formulation and unstructured mesh. Comput Geosci 99:125–134
Castillo P, Rieben R, White D (2005) FEMSTER: an objectoriented class library of highorder discrete differential forms. ACM Trans Math Softw 31(4):425–457
CastilloReyes O, de la Puente J, Cela JM (2018) PETGEM: a parallel code for 3D CSEM forward modeling using edge finite elements. Comput Geosci 119:123–136
CastilloReyes O, De La Puente J, GarcíaCastillo LE et al (2019) Parallel 3D marine controlledsource electromagnetic modelling using highorder tetrahedral Nédélec elements. Geophys J Int 219(1):39–65
Céa J (1964) Approximation variationnelle des problèmes aux limites. Phd thesis Annales de l’Institut Fourier, Grenoble, France, vol 14, no 2, pp 345–444
Chen C, Kruglyakov M, Kuvshinov A (2021) Advanced threedimensional electromagnetic modeling using a nested integral equation approach. Geophys J Int 226(1):114–130
Clough RW (1960) The finite element method in plane stress analysis. In: Proceedings of the 2nd ASCE conference on electronic computation, Pittsburg, PA, USA, pp 345–378
Coggon JH (1971) Electromagnetic and electrical modeling by the finite element method. Geophysics 36:132–155
Commer M (2003) Threedimensional inversion of transient electromagnetic data: a comparative study. Phd thesis, Universität zu Köln, Cologne, Germany
Commer M, Newman G (2004) A parallel finitedifference approach for 3D transient electromagnetic modeling with galvanic sources. Geophysics 69(5):1192–1202
Courant R (1943) Variational methods for the solution of problems of equilibrium and vibrations. Bull Am Math Soc 49(1):1–23
Courant R, Friedrichs K, Lewy H (1928) Über die partiellen Differenzengleichungen der mathematischen Physik. Math Ann 100(1):32–74
Da Silva NV, Morgan JV, MacGregor L et al (2012) A finite element multifrontal method for 3D CSEM modeling in the frequency domain. Geophysics 77(2):E101–E115
Davydycheva S, Druskin V, Habashy T (2003) An efficient finitedifference scheme for electromagnetic logging in 3D anisotropic inhomogeneous media. Geophysics 68(5):1525–1536
Demirci I, Erdoǧan E, Candansayar ME (2012) Twodimensional inversion of direct current resistivity data incorporating topography by using finite difference techniques with triangle cells: investigation of Kera fault zone in western Crete. Geophysics 77(1):E67–E75
Demkowicz L, Oden JT, Strouboulis T (1984) Adaptive finite elements for flow problems with moving boundaries. Part I: Variational principles and a posteriori estimates. Comput Methods Appl Mech Eng 46(2):217–251
Dey A, Morrison HF (1979) Resistivity modeling for arbitrarily shaped threedimensional structures. Geophysics 44(4):753–780
Dey TK, Bajaj CL, Sugihara K (1992) On good triangulations in three dimensions. Int J Comput Geom Appl 2(1):75–95
Druskin VL, Knizhnerman LA (1988) A spectral semidiscrete method for the numerical solution of 3D nonstationary problems in electrical prospecting. Izv Acad Sci USSR Phys Solid Earth 8:63–74
Druskin V, Knizhnerman L (1994) Spectral approach to solving threedimensional Maxwell’s diffusion equations in the time and frequency domains. Radio Sci 29(4):937–953
Egbert GD, Kelbert A (2012) Computational recipes for electromagnetic inverse problems. Geophys J Int 189(1):251–267
Erdogan E, Demirci I, Candansayar ME (2008) Incorporating topography into 2D resistivity modeling using finiteelement and finitedifference approaches. Geophysics 73(3):F135–F142
Eriksson K, Johnson C (1991) Adaptive finite element methods for parabolic problems I. A linear model problem. SIAM J Numer Anal 28(1):43–77
Everett ME (2012) Theoretical developments in electromagnetic induction geophysics with selected applications in the near surface. Surv Geophys 33(1):29–63
Everett ME, Schultz A (1996) Geomagnetic induction in a heterogenous sphere: azimuthally symmetric test computations and the response of an undulating 660km discontinuity. J Geophys Res B Solid Earth 101(2):2765–2783
Farquharson CG, Miensopust MP (2011) Threedimensional finiteelement modelling of magnetotelluric data with a divergence correction. J Appl Geophys 75(4):699–710
Franke A, Börner RU, Spitzer K (2007a) 3D finite element simulation of magnetotelluric fields using unstructured grids. In: 4th international symposium on threedimensional electromagnetics. G. W. Hohmann Memorial Trust for Teaching and Research in Applied Electrical Geophysics, p 15–18
Franke A, Börner RU, Spitzer K (2007b) Adaptive unstructured grid finite element simulation of twodimensional magnetotelluric fields for arbitrary surface and seafloor topography. Geophys J Int 171(1):71–86
Franke A, Kütter S, Börner RU et al (2007c) Numerical simulation of magnetotelluric fields at Stromboli. In: 4th International symposium on threedimensional electromagnetics. G. W. Hohmann Memorial Trust for Teaching and Research in Applied Electrical Geophysics, pp 138–141
FrankeBörner A (2012) Threedimensional finite element simulation of magnetotelluric fields on unstructured grids on the efficient formulation of the boundary value problem. Phd thesis, TU Bergakademie Freiberg, Freiberg, Germany
FrankeBörner A, Börner RU, Spitzer K (2011) Convergence studies for the 2D finite element simulation of the MT boundary value problem. In: Börner RU, Schwalenberg K (eds) Proceedings 24th Schmucker–Weideltcolloquium on deep electromagnetic research, Neustadt/Weinstraße, Sep 26–30, 2011. Deutsche Geophysikalische Gesellschaft, pp 57–74
FrankeBörner A, Börner RU, Spitzer K (2013) Convergence studies for the finite element simulation of the 3D MT boundary value problem. In: 5th International symposium on threedimensional electromagnetics, May 7–9, 2013, Sapporo, Japan, 4p
Galerkin BG (1915) Series solution of some problems of elastic equilibrium of rods and plates. Vestnik Inzhenerov I Tekhnikov 19(7):897–908
GallardoRomero E, RuizAguilar D (2022) High order edgebased finite elements for 3D magnetotelluric modeling with unstructured meshes. Comput Geosci 158:104971
Gander MJ, Wanner G (2012) From Euler, Ritz, and Galerkin to modern computing. SIAM Rev 54(4):627–666
Geselowitz DB (1971) An application of electrocardiographic lead theory to impedance plethysmography. IEEE Trans Biomed Eng BME 18(1):38–41
Geuzaine C, Remacle JF (2009) Gmsh: a 3D finite element mesh generator with builtin pre and postprocessing facilities. Int J Numer Methods Eng 79(11):1309–1331
Gosh DP (1971) The application of linear filter theory to the direct interpretation of geoelectrical resistivity sounding measurements. Geophys Prosp 19:192–217
Grätsch T, Bathe KJ (2005) A posteriori error estimation techniques in practical finite element analysis. Comput Struct 83(4–5):235–265
Grayver AV (2015) Parallel threedimensional magnetotelluric inversion using adaptive finiteelement method. Part I: theory and synthetic study. Geophys J Int 202(1):584–603
Grayver AV, Bürg M (2014) Robust and scalable 3D geoelectromagnetic modelling approach using the finite element method. Geophys J Int 198(1):110–125
Grayver AV, Kolev TV (2015) Largescale 3D geoelectromagnetic modeling using parallel adaptive highorder finite element method. Geophysics 80(6):E277–E291
Grayver AV, Streich R, Ritter O (2013) Threedimensional parallel distributed inversion of CSEM data using a direct forward solver. Geophys J Int 193(3):1432–1446
Gupta PK, Raiche AP, Sugeng F (1989) Threedimensional timedomain electromagnetic modelling using a compact finiteelement frequencystepping method. Geophys J Int 96(3):457–468
Haber E, Ascher UM (2001) Fast finite volume simulation of 3d electromagnetic problems with highly discontinuous coefficients. SIAM J Sci Comput 22(6):1943–1961
Haber E, Schwarzbach C (2014) Parallel inversion of largescale airborne timedomain electromagnetic data with multiple octree meshes. Inverse Probl 30(5):055011
Haber E, Ascher UM, Aruliah DA et al (2000) Fast simulation of 3D electromagnetic problems using potentials. J Comput Phys 163(1):150–171
Haber E, Heldmann S, Ascher U (2007) Adaptive finite volume method for distributed nonsmooth parameter identification. Inverse Probl 23(4):1659–1676
Hohmann GW (1975) Threedimensional induced polarization and electromagnetic modeling. Geophysics 40(2):309–324
Ingerman D, Druskin V, Knizhnerman L (2000) Optimal finite difference grids and rational approximations of the square root I. Elliptic problems. Pure Appl Math 53:1039–1066
Jahandari H, Farquharson CG (2015) Finitevolume modelling of geophysical electromagnetic data on unstructured grids using potentials. Geophys J Int 202(3):1859–1876
Jahandari H, Ansari S, Farquharson CG (2017) Comparison between staggered grid finitevolume and edgebased finiteelement modelling of geophysical electromagnetic data on unstructured grids. J Appl Geophys 138:185–197
Jin J (2002) The finite element method in electromagnetics, 2nd edn. Wiley, New York
Jones FW, Pascoe LJ (1971) A general computer program to determine the perturbation of alternating electric currents in a twodimensional model of a region of uniform conductivity with an embedded inhomogeneity. Geophys J R Astron Soc 24(1):3–30
Kamm J, Becken M, Abreu R (2020) Electromagnetic modelling with topography on regular grids with equivalent materials. Geophys J Int 220:2021–2038
Käufl JS, Grayver AV, Kuvshinov AV (2018) Topographic distortions of magnetotelluric transfer functions: a highresolution 3D modelling study using real elevation data. Geophys J Int 215(3):1943–1961
Key K (2016) MARE2DEM: a 2D inversion code for controlledsource electromagnetic and magnetotelluric data. Geophys J Int 207(1):571–588
Key K, Ovall J (2011) A parallel goaloriented adaptive finite element method for 2.5D electromagnetic modelling. Geophys J Int 186(1):137–154
Key K, Weiss C (2006) Adaptive finiteelement modeling using unstructured grids: the 2D magnetotelluric example. Geophysics 71(6):G291–G299
Kirk BS, Peterson JW, Stogner RH et al (2006) libMesh: a C++ library for parallel adaptive mesh refinement/coarsening simulations. Eng Comput 22(3–4):237–254
Kordy M, Wannamaker P, Maris V et al (2016) 3D magnetotelluric inversion including topography using deformed hexahedral edge finite elements and direct solvers parallelized on SMP computers  Part I: Forward problem and parameter Jacobians. Geophys J Int 204(1):74–93
Kütter S, FrankeBörner A, Börner RU et al (2010) Threedimensional finite element simulation of magnetotelluric fields at Stromboli volcano incorporating digital topography and bathymetry. In: 20th International workshop on electromagnetic induction in the Earth, Giza, Egypt, September 18–24, 2010, 4 p
Lanczos C (1961) Linear differential operators. Van Nostrand, London
Lax PD, Milgram AN (1954) Parabolic equations, contributions to the theory of partial differential equations. Ann Math Stud 33:167–190
Li Y (2002) A finiteelement algorithm for electromagnetic induction in twodimensional anisotropic conductivity structures. Geophys J Int 148(3):389–401
Li Y, Constable S (2007) 2D marine controlledsource electromagnetic modeling: Part 2—the effect of bathymetry. Geophysics 72(2):WA63–WA71
Li Y, Dai S (2011) Finite element modelling of marine controlledsource electromagnetic responses in twodimensional dipping anisotropic conductivity structures. Geophys J Int 185(2):622–636
Li Y, Key K (2007) 2D marine controlledsource electromagnetic modeling: Part 1—an adaptive finiteelement algorithm. Geophysics 72(2):WA51–WA62
Li Y, Pek J (2008) Adaptive finite element modelling of twodimensional magnetotelluric fields in general anisotropic media. Geophys J Int 175(3):942–954
Li Y, Spitzer K (2002) Threedimensional DC resistivity forward modelling using finite elements in comparison with finitedifference solutions. Geophys J Int 151(3):924–934
Li Y, Spitzer K (2005) Finite element resistivity modelling for threedimensional structures with arbitrary anisotropy. Phys Earth Planet Inter 150(1–3 SPEC. ISS.):15–27
Li J, Li Y, Liu Y et al (2021) 3D marine CSEM forward modeling with general anisotropy using an adaptive finiteelement method. IEEE Geosci Remote Sens Lett 18(11):1936–1940
Liszka T, Orkisz J (1980) The finite difference method at arbitrary irregular grids and its application in applied mechanics. Comput Struct 11(1–2):83–95
Liu Y, Xu Z, Li Y (2018) Adaptive finite element modelling of threedimensional magnetotelluric fields in general anisotropic media. J Appl Geophys 151:113–124
Livelybrooks D (1993) Program 3Dfeem: a multidimensional electromagnetic finite element model. Geophys J Int 114(3):443–458
Lowry T, Allen MB, Shive PN (1989) Singularity removal: a refinement of resistivity modeling techniques. Geophysics 54:766–774
Mackie RL, Madden TR, Wannamaker PE (1993) Threedimensional magnetotelluric modeling using difference equations—theory and comparisons to integral equation solutions. Geophysics 58(2):215–226
Mitsuhata Y, Uchida T (2004) 3D magnetotelluric modeling using the T–Ω finiteelement method. Geophysics 69(1):108–119
Monk P (2003) Finite element methods for Maxwell’s equations. Oxford University Press, New York
Mukherjee S, Everett ME (2011) 3D controlledsource electromagnetic edgebased finite element modeling of conductive and permeable heterogeneities. Geophysics 76(4):F215–F226
Mulder WA (2006) A multigrid solver for 3D electromagnetic diffusion. Geophys Prospect 54(5):633–649
Nam MJ, Kim HJ, Song Y et al (2007) 3D magnetotelluric modelling including surface topography. Geophys Prospect 55(2):277–287
Nédélec JC (1980) Mixed finite elements in \({\mathbb{R} }\)3. Numer Math 35(3):315–341
Newman GA (2014) A review of highperformance computational strategies for modeling and imaging of electromagnetic induction data. Surv Geophys 35(1):85–100
Newman GA, Alumbaugh DL (1995) Frequencydomain modelling of airborne electromagnetic responses using staggered finite differences. Geophys Prospect 43(8):1021–1042
Newman GA, Alumbaugh DL (1997) Threedimensional massively parallel electromagnetic inversion—I. Theory. Geophys J Int 128(2):345–354
Nittinger C, Becken M (2018) Compressive sensing approach for twodimensional magnetotelluric inversion using wavelet dictionaries. Geophys Prospect 66(4):664–672
Oden JT, Prudhomme S (2001) Goaloriented error estimation and adaptivity for the finite element method. Comput Math Appl 41(5–6):735–756
Oristaglio ML, Hohmann GW (1984) Diffusion of electromagnetic fields into a twodimensional earth: a finitedifference approach. Geophysics 49(7):870–894
Özyildirim O, Candansayar ME, Demirci I et al (2017) Twodimensional inversion of magnetotelluric/radiomagnetotelluric data by using unstructured mesh. Geophysics 82(4):E197–E210
Pankratov O, Kuvshinov A (2016) Applied mathematics in EM studies with special emphasis on an uncertainty quantification and 3D integral equation modelling. Surv Geophys 37(1):109–147
Pape H, Riepe L, Schopper JR (1982) A pigeonhole model for relating permeability to specific surface. Log Anal 23:5–13
Pape H, Clauser C, Bartels J (1998a) Bestimmung von Porosität und Permeabilität aus Bohrlochmessungen der induzierten Polarisation (IP) oder der nuklearmagnetischen Resonanz (NMR) auf der Grundlage einer fraktalen Porenraumgeometrie—Teil 1: PorositätsPermeabilitätsbeziehung und NMRAnwendungsbeispiel. In: Bahr K, Junge A (eds) Protocol colloquium elektromagnetische tiefenforschung. ISSN 09467467, Deutsche Geophysikalische Gesellschaft, Göttingen, Frankfurt, pp 351–362
Pape H, Grinat M, Clauser C (1998b) Bestimmung von Porosität und Permeabilität aus Bohrlochmessungen der induzierten Polarisation (IP) oder der nuklearmagnetischen Resonanz (NMR) auf der Grundlage einer fraktalen Porenraumgeometrie—Teil 2: Elektrische Ersatzschaltbilder und IPAnwendungsbeispiel. In: Bahr K, Junge A (eds) Protocol colloquium elektromagnetische tiefenforschung. ISSN 09467467, Deutsche Geophysikalische Gesellschaft, Göttingen, Frankfurt, p 363–374
PARDISO Solver Project (2023) http://www.pardisoproject.org. Accessed January 2023
Pardo D, Demkowicz L, TorresVerdín C et al (2006) Twodimensional highaccuracy simulation of resistivity loggingwhiledrilling (LWD) measurements using a selfadaptive goaloriented HP finite element method. SIAM J Appl Math 66(6):2085–2106
Pardo D, Nam MJ, TorresVerdín C et al (2011) Simulation of marine controlled source electromagnetic measurements using a parallel Fourier HPfinite element method. Comput Geosci 15(1):53–67
Penz S, Chauris H, Donno D et al (2013) Resistivity modelling with topography. Geophys J Int 194(3):1486–1497
Plessix RE, Darnet M, Mulder WA (2007) An approach for 3D multisource, multifrequency CSEM modeling. Geophysics 72(5):SM177–SM184
Pötschke D (2017) Geoelektrische Tomographie an einem Erzgang im Bergwerk Reiche Zeche (Freiberg). Master’s thesis, Institute of Geophysics and Geoinformatics, TU Bergakademie Freiberg
Pridmore DF, Hohmann GW, Ward SH et al (1981) An investigation of finiteelement modelling for electrical and electromagnetical data in three dimensions. Geophysics 46:1009–1024
Prudhomme S, Oden JT (1999) On goaloriented error estimation for elliptic problems: application to the control of pointwise errors. Comput Methods Appl Mech Eng 176(1–4):313–331
Puzyrev V, Koldan J, de la Puente J et al (2013) A parallel finiteelement method for threedimensional controlledsource electromagnetic forward modelling. Geophys J Int 193(2):678–693
Raiche AP (1974) An integral equation approach to threedimensional modelling. Geophys J R Astron Soc 36(2):363–376
Raviart PA, Thomas JM (1977) A mixed finite element method for 2nd order elliptic problems. In: Galligani I, Magenes E (eds) Lecture notes in mathematics. Mathematical aspects of finite element methods, vol 606. Springer, Berlin, Heidelberg
Reddy IK, Rankin D, Philips RJ (1977) Threedimensional modeling in magnetotelluric and magnetic variational sounding. Geophys J R astr Soc 51:313–325
Ren Z, Kalscheuer T (2020) Uncertainty and resolution analysis of 2D and 3D inversion models computed from geophysical electromagnetic data. Surv Geophys 41(1):47–112
Ren Z, Tang J (2010) 3D direct current resistivity modeling with unstructured mesh by adaptive finiteelement method. Geophysics 75(1):H7–H17
Ren Z, Tang J (2014) A goaloriented adaptive finiteelement approach for multielectrode resistivity system. Geophys J Int 199(1):136–145
Ren Z, Kalscheuer T, Greenhalgh S et al (2013) A goaloriented adaptive finiteelement approach for plane wave 3D electromagnetic modelling. Geophys J Int 194(2):700–718
Richardson LF (1922) Weather prediction by numerical process. Cambridge University Press, Cambridge
Ritz W (1909) Über eine neue Methode zur Lösung gewisser Variationsprobleme der mathematischen Physik. J Reine Angew Math 1909(135):1–61
Rochlitz R, Skibbe N, Günther T (2019) CustEM: customizable finiteelement simulation of complex controlledsource electromagnetic data. Geophysics 84(2):F17–F33
Rücker C, Günther T, Spitzer K (2006) Threedimensional modelling and inversion of dc resistivity data incorporating topography—I. Modelling. Geophys J Int 166(2):495–505
Rulff P, Buntin L, Kalscheuer T (2021) Efficient goaloriented mesh refinement in 3D finiteelement modelling adapted for controlled source electromagnetic surveys. Geophys J Int 227(3):1624–1645
Sasaki Y (1994) 3D resistivity inversion using the finiteelement method. Geophysics 59(12):1839–1848
Sasaki Y (2001) Full 3D inversion of electromagnetic data on PC. J Appl Geophys 46(1):45–54
Sasaki Y (2004) Threedimensional inversion of staticshifted magnetotelluric data. Earth Planets Space 56(2):239–248
Schellbach KH (1851) Probleme der Variationsrechnung. J Reine Angew Math 41(4):293–363
Schenk O, Gärtner K (2004) Solving unsymmetric sparse systems of linear equations with Pardiso. J Future Gener Comput Syst 20(3):475–487
Schmucker U (1971) Neue rechenmethoden zur tiefensondierung. In: Kolloquium elektromagnetische tiefensondierung, Rothenberge. Deutsche Geophysikalische Gesellschaft, pp 1–39
Schneider C, Spitzer K, Hort M (2022) Timedependent adaptive mesh refinement for 3D forward modelling of transient electromagnetic fields in volcanic environments including topography. In: 25th International workshop on electromagnetic induction in the Earth, Çeşme, Turkey, 11–17 September 2022, 4 p
Schöberl J (1997) An advancing front 2D/3Dmesh generator based on abstract rules. Comput Vis Sci 1(1):41–52
Schwarzbach C, Haber E (2013) Finite element based inversion for timeharmonic electromagnetic problems. Geophys J Int 193:615–634
Schwarzbach C, Börner RU, Spitzer K (2011) Threedimensional adaptive higher order finite element simulation for geoelectromagneticsa marine CSEM example. Geophys J Int 187(1):63–74
Scriba H (1981) Computation of the electric potential in threedimensional structures. Geophys Prospect 29:790–802
Shewchuk JR (1996) Triangle: engineering a 2D quality mesh generator and Delaunay triangulator. In: Lecture notes in computer science (including subseries lecture notes in artificial intelligence and lecture notes in bioinformatics), vol 1148, pp 203–222
Shewchuk JR (1997) Delaunay refinement mesh generation. PhD thesis, School of Computer Science, Computer Science Department, Carnegie Mellon University, Pittsburgh, PA, USA
Shewchuk JR (2002) Delaunay refinement algorithms for triangular mesh generation. Comput Geom 22(1):21–74. 16th ACM symposium on computational geometry
Shortley GH, Weller R (1938) The numerical solution of Laplace’s equation. J Appl Phys 9(5):334–348
Si H (2015) TetGen, a Delaunaybased quality tetrahedral mesh generator. ACM Trans Math Softw 41(2):1–36
Singh A, Dehiya R (2023) An efficient EM modeling scheme for large 3D models—a magnetotelluric case study. IEEE Trans Geosci Remote Sens 61:1–11
Siripunvaraporn W, Egbert G, Lenbury Y (2002) Numerical accuracy of magnetotelluric modeling: a comparison of finite difference approximations. Earth Planets Space 54(6):721–725
Smith JT (1996a) Conservative modeling of 3D electromagnetic fields, Part II: biconjugate gradient solution and an accelerator. Geophysics 61(5):1319–1324
Smith T (1996b) Conservative modeling of 3D electromagnetic fields, Part I: properties and error analysis. Geophysics 61:1308–1318
Spitzer K (1995) A 3D finitedifference algorithm for DC resistivity modelling using conjugate gradient methods. Geophys J Int 123(3):903–914
Spitzer K (1998) The threedimensional DC sensitivity for surface and subsurface sources. Geophys J Int 134:736–746
Spitzer K, Chouteau M (1997) Crosshole resistivity and IP experiment at Casa Berardi gold mine: interpretation using a newly developed 3D forward modeling code. In: Extended abstracts book, 67th annual international meeting of the Society of Exploration Geophysicists, pp 1957–1960
Spitzer K, Chouteau M (2003) A DC resistivity and IP borehole survey at the Casa Berardi gold mine in northwestern Quebec. Geophysics 68(2):453–463
Spitzer K, Wurmstich B (1999) Speed and accuracy in 3D resistivity modeling. In: Oristaglio MJ, Spies BR (eds) Threedimensional electromagnetics. No. 7 in geophysical developments. Society of Exploration Geophysicists, pp 161–176
Spitzer K, Chouteau M, Boulanger O (1999) Gridindependent electrode positioning for 3D DC and IP forward modeling. In: Proceedings II. International symposium on threedimensional electromagnetics in Salt Lake City, UT, Oct 26–29, 1999. University of Utah, pp 189–192
Strang G, Fix J (1973) TAn analysis of the finite element method. PrenticeHall, Englewood Cliffs
Streich R (2009) 3D finitedifference frequencydomain modeling of controlledsource electromagnetic data: direct solution and optimization for high accuracy. Geophysics 74(5):F95–F105
Streich R, Becken M (2011) Electromagnetic fields generated by finitelength wire sources: comparison with point dipole solutions. Geophys Prospect 59(2):361–374
Sugeng F (1998) Modeling the 3D TDEM response using the 3D fulldomain finiteelement method based on the hexahedral edgeelement technique. Explor Geophys 29(4):615–619
Turner MJ, Clough RW, Martin HC et al (1956) Stiffness and deflection analysis of complex structures. J Aeronaut Sci 23(9):805–823
Udphuay S, Günther T, Everett ME et al (2011) Threedimensional resistivity tomography in extreme coastal terrain amidst dense cultural signals: application to cliff stability assessment at the historic DDay site. Geophys J Int 185(1):201–220
Um ES, Harris JM, Alumbaugh DL (2012) An iterative finite element timedomain method for simulating threedimensional electromagnetic diffusion in earth. Geophys J Int 190(2):871–886
Um ES, Commer M, Newman G (2013) Efficient preconditioned iterative solution strategies for the electromagnetic diffusion in the Earth: finiteelement frequencydomain approach. Geophys J Int 193:1460–1473
Unsworth MJ, Travis BJ, Chave AD (1993) Electromagnetic induction by a finite electric dipole source over a 2D Earth. Geophysics 58(2):198–214
Usui Y (2015) 3D inversion of magnetotelluric data using unstructured tetrahedral elements: applicability to data affected by topography. Geophys J Int 202(2):828–849
Wait JR (1953) Propagation of radio waves over a stratified ground. Geophysics 20:416–422
Wang T, Hohmann GW (1993) A finitedifference time domain solution for threedimensional electromagnetic modeling. Geophysics 58:797–809
Wang F, Morten JP, Spitzer K (2018) Anisotropic threedimensional inversion of CSEM data using finiteelement techniques on unstructured grids. Geophys J Int 213(2):1056–1072
Wannamaker PE (1991) Advances in threedimensional magnetotelluric modeling using integral equations. Geophysics 56(11):1716–1728
Wannamaker PE, Stodt JA, Rijo L (1987) A stable finite element solution for twodimensional magnetotelluric modelling. Geophys J R Astron Soc 88(1):277–296
Weidelt P (1975) Electromagnetic induction in threedimensional structures. J Geophys 41:85–109
Weidelt P (1999) 3D conductivity models, implications of electrical anisotropy. In: Oristaglio MJ, Spies BR (eds) Threedimensional electromagnetics. Society of Exploration Geophysicists, Houston, pp 119–137
Weiss CJ (2017) Finiteelement analysis for model parameters distributed on a hierarchy of geometric simplices. Geophysics 82(4):E155–E167
Weiss CJ, van Bloemen Vanders BG, Antil H (2020) Fractional operators applied to geophysical electromagnetics. Geophys J Int 220:1242–1259
Weiss M, Kalscheuer T, Ren Z (2022) Spectral element method for 3D controlledsource electromagnetic forward modelling using unstructured hexahedral meshes. Geophys J Int 232(2):1427–1454
Weller HG, Tabor G, Jasak H et al (1998) A tensorial approach to computational continuum mechanics using objectoriented techniques. Comput Phys 12(6):620–631
Wittke J (2017) Meshless methods for maxwell’s equations with applications to magnetotelluric modelling and inversion. Phd thesis, Universität zu Köln, Cologne, Germany
Wittke J, Tezkan B (2014) Meshfree magnetotelluric modelling. Geophys J Int 198(2):1255–1268
Xiong Z, Kirsch A (1992) Threedimensional earth conductivity inversion. J Comput Appl Math 42(1):109–121
Xiong Z, Raiche A, Sugeng F (2000) Efficient solution of full domain 3D electromagnetic modelling problems. Explor Geophys 31(2):158–161
Yan B, Li Y, Liu Y (2016) Adaptive finite element modeling of direct current resistivity in 2D generally anisotropic structures. J Appl Geophys 130:169–176
Ye Y, Li Y, Li G et al (2018) 3D adaptive finiteelement modeling of marine controlledsource electromagnetics with seafloor topography based on secondary potentials. Pure Appl Geophys 175(12):4449–4463
Yee KS (1966) Numerical solution of initial boundary value problems involving Maxwell’s equations in isotropic media. IEEE Trans Antennas Propag 14(3):302–307
Yin C, Zhang B, Liu Y et al (2016) A goaloriented adaptive finiteelement method for 3D scattered airborne electromagnetic method modeling. Geophysics 81(5):E337–E346
Zehner B, Börner JH, Görz I et al (2015) Workflows for generating tetrahedral meshes for finite element simulations on complex geological structures. Comput Geosci 79:105–117
Zhdanov MS (1988) Integral transforms in geophysics. Springer, New York
Zhu J, Yin C, Liu Y et al (2020) 3D DC resistivity modelling based on spectral element method with unstructured tetrahedral grids. Geophys J Int 220:1748–1761
Zienkiewicz OC, Cheung YK (1967) The finite element method in structural and continuum mechanics. McGraw Hill, London
Zienkiewicz OC, Zhu JZ (1987) A simple error estimator and adaptive procedure for practical engineering analysis. Int J Numer Methods Eng 24(2):337–357
Zienkiewicz OC, Zhu JZ (1992a) The superconvergent patch recovery and a posteriori error estimates. Part 1: the recovery technique. Int J Numer Methods Eng 33(7):1331–1364
Zienkiewicz OC, Zhu JZ (1992b) The superconvergent patch recovery and a posteriori error estimates. Part 2: error estimates and adaptivity. Int J Numer Methods Eng 33(7):1365–1382
Zimmerling J, Druskin V, Zaslavsky M et al (2018) Modelorder reduction of electromagnetic fields in open domains. Geophysics 83(2):WB61–WB70
Acknowledgements
I am grateful to the IAGA Div VI Committee and the Local Organizing Committee of the 25th Workshop of Electromagnetic Induction in the Earth and Planetary Bodies for inviting me to give this review talk in Çeşme, Turkey, on 13 September 2022. Moreover, I would like to thank my colleagues and students at the Institute of Geophysics and Geoinformatics and the Departments of Numerical Analysis of the TU Freiberg and TU Chemnitz for providing such an inspiring environment over the years. And finally, I thank the editors for this special issue, Anne Neska and Ahmet Başokur, as well as Colin Farquharson and an anonymous reviewer for taking the time to read this rather extensive article and suggest meaningful improvements.
Funding
Open Access funding enabled and organized by Projekt DEAL.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The author has no competing interests to declare that are relevant to the content of this article. He certifies that he has no affiliations with or involvement in any organization or entity with any financial interest or nonfinancial interest in the subject matter or materials discussed in this manuscript. The author has no financial or proprietary interests in any material discussed in this article.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Paper presented at the 25th Electromagnetic Induction Workshop, held in Çeşme, Turkey in September 2022.
Appendices
Appendix 1: The Energy Norm
The energy norm was defined by Zienkiewicz and Zhu (1987) and reads
where \(e=uu_{h}\) is the error of the FE solution \(u_{h}\) with respect to the unknown exact solution u. L is a differential operator. For the identity operator, the \(L_{2}\)norm is obtained as a special case of the energy norm
The squared error can be obtained by summing over the N elements of the partition \({\mathcal{T}}_{h}\) giving \(\begin{aligned} e^{2}=\sum _{i=1}^{N} e_{i}^{2}. \end{aligned}\) With this, a relative percentage error can be expressed as \(\eta =\frac{e}{u}*100 \%\) which represents a weighted RMS.
Appendix 2: The Variational Problem
We consider the strong formulation of our model boundary value problem of Poisson’s equation representing an elliptic linear problem
over the domain \(\Omega\) with Lipschitz boundary \(\partial \Omega =\Gamma =\Gamma _{\textrm{D}} \cup \Gamma _{\textrm{N}}\), i.e., the data \(f\in L^{2}(\Omega )\) and \(g\in L^{2}(\Gamma _{\textrm{N}})\) are sufficiently smooth. \({\varvec{n}}\) is the outward normal on \(\Gamma\).
Now we multiply both sides of Eq. (17) with test functions \(v\in {\mathcal{V}}\), where \({\mathcal{V}}=\left\{ v\in {\mathcal{H}}^{1}(\Omega ):v=0 {\text{ on }} \Gamma _{\textrm{D}}\right\}\) is the Sobolev space \({\mathcal{H}}_{0}^{1}\) giving
Applying the product rule
to the equation above and applying Gauß’ theorem yields
With \(v\in {\mathcal{H}}^{1}_{0}\), the boundary integral reduces to \(\displaystyle {\int _{\Gamma _{N}}} gv \; {\textrm{d}}S\) and we obtain
giving a bilinear form
and a linear functional
so that the weak formulation of the problem can be written as
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Spitzer, K. Electromagnetic Modeling Using Adaptive Grids – Error Estimation and Geometry Representation. Surv Geophys 45, 277–314 (2024). https://doi.org/10.1007/s10712023097949
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10712023097949