# Dimensionality, Coordinate System and Reference Frame for Analysis of In-Situ Space Plasma and Field Data

- 761 Downloads

## Abstract

In the analysis of in-situ space plasma and field data, an establishment of the coordinate system and the frame of reference, helps us greatly simplify a given problem and provides the framework that enables a clear understanding of physical processes by ordering the experimental data. For example, one of the most important tasks of space data analysis is to compare the data with simulations and theory, which is facilitated by an appropriate choice of coordinate system and reference frame. While in simulations and theoretical work the establishment of the coordinate system (generally based on the dimensionality or dimension number of the field quantities being studied) and the reference frame (normally moving with the structure of interest) is often straightforward, in space data analysis these are not defined *a priori*, and need to be deduced from an analysis of the data itself. Although various ways of building a dimensionality-based (D-based) coordinate system (i.e., one that takes account of the dimensionality, e.g., 1-D, 2-D, or 3-D, of the observed system/field), and a reference frame moving along with the structure have been used in space plasma data analysis for several decades, in recent years some noteworthy approaches have been proposed. In this paper, we will review the past and recent approaches in space data analysis for the determination of a structure’s dimensionality and the building of D-based coordinate system and a proper moving frame, from which one can directly compare with simulations and theory. Along with the determination of such coordinate systems and proper frame, the variant axis/normal of 1-D (or planar) structures, and the invariant axis of 2-D structures are determined and the proper frame velocity for moving structures is found. These are found either directly or indirectly through the definition of dimensionality. We therefore emphasize that the determination of dimensionality of a structure is crucial for choosing the most appropriate analysis approach, and failure to do so might lead to misinterpretation of the data. Ways of building various kinds of coordinate systems and reference frames are summarized and compared here, to provide a comprehensive understanding of these analysis tools. In addition, the method of building these systems and frames is shown not only to be useful in space data analysis, but also may have the potential ability for simulation/laboratory data analysis and some practical applications.

## Keywords

Dimensionality Dimension number Coordinate system Reference frame Normal Variant/invariant axis Flux rope Current sheet Magnetic reconnection Reconstruction## 1 Introduction

In physics studies, the establishment of two systems is fundamental: one is the reference frame of a system relative to the observer and another is the coordinate system. A coordinate system establishes the orientation of an observed object/field in space, and a reference frame (with defined velocity) establishes its motion. An appropriate reference frame and coordinate system may help us greatly simplify a given problem, perform calculations more easily, make experimental data more ordered and enable a clearer understanding of physical processes. This is especially vital in space plasma theory, simulation and data analysis.

In both theoretical analysis and numerical simulations, the coordinate system and the reference frame are chosen a priori. For example, in the theoretical analyses of Kelvin-Helmholtz waves (e.g., Pu and Kivelson 1983), or tearing mode configurations (e.g., Terasawa 1983), the physical fields are set to be 2-D and then the coordinate system is naturally based on the dimensionality (dimension number, number of spatial degrees of freedom required to describe a field, i.e., whether it is 1-D, 2-D or 3-D) based coordinate system (hereafter we refer to as a ‘D-based coordinate system’). Examples in simulation work include 2-D simulations of magnetic reconnection (e.g., Lin and Swift 1996; Birn and Hesse 2001; Daughton et al. 2009) or 1-D simulations of plasma processes (e.g., Dawson 1983), which have all used a D-based coordinate system. In these studies, the reference frame is just the frame moving along with the structure, e.g., the current sheet or a flux rope.

For the discussion of coordinate systems, in this article we mainly focus on a local Cartesian coordinate system which varies with position. Global coordinate systems can be seen in the review papers by Song and Russell (1999) and Kivelson and Russell (1995). As we have mentioned, in most simulation and theoretical analysis, the natural coordinate system choice is D-based. We should emphasize the definition of dimensionality (dimension number) here. Dimensionality is a basic concept in plasma physics and ordinary fluid dynamics. In most physical problems we only care about the variation of physical fields, and therefore we use the spatial variation of field quantities instead of the field quantity itself to define the dimensionality. For example, say we have a (scalar or vector) field quantity with a given structure in 3 dimensional Cartesian space \(\varphi =\varphi ( x,y,z )\). If the field quantity varies in only one direction (say, \(x\)), such that \(\partial \varphi /\partial y = \partial \varphi /\partial z = 0\) (for each Cartesian component, if it is a vector), then we have \(\varphi = \varphi (x)\) and the structure is one-dimensional (1-D). If there is no change along only one direction (say \(z\)), such that \(\partial \varphi / \partial z = 0\), then \(\varphi = \varphi (x,y)\) (i.e., the physical field varies in the \(x\)–\(y\) plane) and the field/structure is two-dimensional (2-D). In this case the \(z\)-direction is known as the invariant axis of the structure. If one cannot find any invariant directions, the field/structure is three dimensional (3-D). For both 1-D and 2-D cases, we allow the existence of all three components of a field quantity. Based on dimensionality, we can establish a local Cartesian coordinate system. This D-based coordinate system has a very clear physical meaning. For example, a flux rope or a flux transfer event (e.g., Russell 1995) is often a 2-D structure and all fields vary little along its axis. Then, if we can determine its dimension number, the invariant axis, which corresponds to the axis of the flux rope, can be found. Another example is a 1-D current sheet (often seen at the magnetopause and the magnetotail or in a shock front) in which all field quantities vary only along its normal direction. If we have determined its dimension number, then the only variation direction is found and it corresponds to just the normal of the current sheet.

After we have obtained data from space instruments, we often hope to interpret it through some theoretical work or numerical simulations that are expressed in a D-based coordinate system. However, the data are obtained initially in the spacecraft frame. For example, for a spin-stabilized satellite, one axis is the spin axis and the other two axes are in the plane perpendicular to the spin axis. Then, if we know the direction of the spin axis in the Earth’s frame we can transform the data to a global coordinate system such as the Geocentric Solar Ecliptic (GSE) or Geocentric Solar Magnetospheric (GSM) coordinates. This step of axis transformation is usually not very difficult and normally is provided in the scientific data of the satellite mission. However, if we intend to analyze the data in a D-based coordinate system, it is not very straight forward. We require a general method to identify this coordinate system through analyzing the data itself.

For a 1-D structure such as a current sheet, finding its normal forms the basis of building a D-based coordinate system, because the normal is the only variation direction of the 1-D structure. To completely build a Cartesian coordinate system, we need the other two axes, which can be any two orthogonal directions that are in the plane perpendicular to the normal. Over the past years of space data analysis, various attempts to define such coordinates have been made. The first systematic and quantitative method for establishing a Cartesian coordinate system was the minimum variance analysis (MVA) method proposed by Sonnerup and Cahill (1967). It is undoubted that the coordinate system established by using the method of MVA (Sonnerup and Cahill 1967) has played, and will continue to play, a key role in single and multiple satellite data analysis. Methods like the Timing (e.g., Russell et al. 1983) or coplanarity (e.g., Schwartz 1998) can also be used to find the normal for building a D-based coordinate system for 1-D cases. For a 2-D structure such as a flux rope, because its axis is the invariant direction, finding the flux rope axis is the first step to build a D-based coordinate system. After we have determined the axis, the other two axes can be any two orthogonal directions that are in the plane perpendicular to the axis, and we can choose one direction as the projection of the spacecraft path in this plane and the last axis of this D-based coordinate system completes the right hand orthogonal set. D-based coordinate systems are a kind of local coordinate system. Other kinds of local coordinate systems such as the local field aligned coordinate system, used when studying some waves in the magnetosphere (e.g., Hartinger et al. 2011; Shi et al. 2013, 2014) will not be discussed in detail in this article.

For the description of processes taking place in space, one must use a reference frame. A good frame of reference is often the frame that is moving along with the structure in space, within which one can analyze the physical processes (note that the “structure” of interest, e.g. a magnetic flux rope, may not be simply moving with the plasma flow). In theory/simulation work, for example, to study a flux tube, reconnection point, or current sheet characteristics, we often need to study the electromagnetic fields and plasma/particle dynamics in the reference frame moving along with the structure. Then the mass, momentum and energy conservation equations can more easily be solved in this structure rest-frame. When we intend to link the data to the physical parameters obtained from theory/simulation, or to give the observed phenomena a physical explanation, if we do not use the same frame, the explanation will be very difficult. However, the measured plasma and electromagnetic field data are gathered in the spacecraft frame, and in most cases of interest the structure moves with respect to the spacecraft. One simple and direct consideration is that if we could find the structure velocity, we then will be able to obtain the reference frame.

_{str}in (1.1) is the structure velocity relative to the observer. This equation means that the time variation of the observed \(\varphi \) can be caused either by the

*temporal variation*(first term on the right) or the

*spatial variation*(second term on the right), or both. In fact equation (1.1) can be derived from the material derivative expanded in the Euler description of a fluid, $\frac{\mathrm{d}\phi}{\mathrm{d}t}=\frac{\partial \phi}{\partial t}+\stackrel{\rightharpoonup}{V}\cdot \mathrm{\nabla}\phi $, where \(\frac{\mathrm{d}\varphi }{ \mathrm{d}t}\) is the material derivative (also called substantial derivative or particle derivative) that describes the time variation in the frame moving with the material particle, \(\frac{\partial \varphi }{ \partial t}\) is the local derivative representing the time variation in the observer frame and $\stackrel{\rightharpoonup}{V}\cdot \mathrm{\nabla}\phi $ is the convective derivative. Because the local derivative and the convective derivative are much easier to measure/observe than the material derivative, in practice we normally use the former two to describe the physical processes, although many physical laws like momentum conservation are more conveniently described under a Lagrangian description using material derivatives. In space, the structure can be analogous to the material particle and then we obtain equation (1.1). For all these time derivatives, we use partial derivative with subscripts indicating the frame instead of using partial derivative or total derivative, because the only difference between partial derivative or total derivative here is the frame difference and in different points of view the partial derivative and total derivative can be changed into each other. For example, the position of the symbol ‘\(d\)’ and ‘\(\partial \)’ we used in this equation $\frac{\mathrm{d}\phi}{\mathrm{d}t}=\frac{\partial \phi}{\partial t}+{\stackrel{\rightharpoonup}{V}}_{\mathrm{str}}\cdot \mathrm{\nabla}\phi $ are opposite to that in (1.1) in Song and Russell (1999) but the equation we are using is the same, which means the symbol ‘\(d\)’ and ‘\(\partial \)’ themselves have no physical difference here. Therefore to state clearly and avoid unnecessary confusions, we use the ‘\(\partial \)’ to replace the ‘\(d\)’ and use different subscripts (‘obs’ or ‘str’) to distinguish temporal variations in different reference frames. Then, when we intend to use a theory or simulation which is described in the ‘str’ frame to interpret observations that are measured in the spacecraft ‘sc’ frame, equation (1.1) provides a way of frame transformation.

In summary, while in simulations and theoretical work the establishment of the coordinate system and the reference frame is straightforward, the determination of these from space data needs some more analysis. In this article, we will review the methods of establishing the D-based coordinate system and ways of constructing the proper frame of reference that are used in the space data analysis, and their application in the analysis on structures measured in space. Over the past twenty years, especially following the launch of the ESA Cluster constellation, many multi-point methods have been developed and applied to study space physics processes. Applications of these techniques are critically dependent on correct determination of structure dimensionality, principle axis and velocity. Nevertheless, we find that in some cases some techniques were not quite appropriately applied and may have affected their conclusions. Now that the new NASA constellation Magnetospheric Multiscale (MMS) is operational; it is necessary and timely to clarify these problems and further develop new multi spacecraft data analysis techniques.

In this review paper, the determination of dimensionality analysis, principle axis and velocity determination will be discussed, and also other related methods, such as some single satellite methods will be summarized and compared. Applications on reconstruction techniques for magnetic flux ropes, current sheets and other magnetic structures will be introduced, and, we will review the conditions under-which each method should be most appropriately and effectively used. In both Sects. 2 and 3 we will first review single spacecraft-based methods, and then multi-point methods, including some traditional methods and their continuing development, and novel approaches developed very recently. In Sect. 2, we will review six methods for building D-based coordinate systems. While all of these methods can build a D-based coordinate system for 1-D structures, some methods may no longer be D-based when applied to higher dimensional structures. In some of these (but not all) cases, this may be rectified by a simple axis rotation. In Sect. 3 we will review the frame of reference in which the observer resides. We make the argument that finding the observational frame is dependent upon/closely related to finding the D-based coordinate system. For example, the application of the traditional Triangulation/Timing methods (Russell et al. 1983) to 2-D structures obtains both the reference frame and a D-based coordinate system together. In Sect. 4 we discuss some uncertainties and cautions in using some important methods. Since in data analysis different field quantities might have different features, we also discuss the dimensionality for different field quantities. Then we compare all the methods discussed, and show where they can be best applied. We emphasize that different methods will have their best application in different circumstances. Therefore, we advise in many cases to use different methods for the same event to compare and obtain a more reliable coordinate system and reference frame. Lastly we show some potential applications of some gradient methods in simulation and other circumstances.

## 2 D-Based Coordinate Systems

As we have mentioned Sect. 1, dimensionality based coordinate system is very commonly used in the numerical simulation and theoretical analysis of 1-D or 2-D problems. In the data analysis of in situ observations, 2-D or 1-D problem is often much easier to study and compare with numerical or theoretical analysis than a three dimensional (3-D) one. It is important, therefore, to pre-determine the dimensionality (dimension number) and characteristic directions of observed space structures before proceeding with further data analysis. In addition, a reduced dimension number is an assumption of many analysis methods. For example, the widely used MVA method (Sonnerup and Scheible 1998), the multi-spacecraft-timing method and its later revision (e.g., Russell et al. 1983; Zhou et al. 2006a, 2006b; Zhou et al. 2009), and Grad–Shafranov (GS) reconstruction methods (Hau and Sonnerup 1999; Hu and Sonnerup 2002; Sonnerup et al. 2006; Tian et al. 2010, 2014) are all set up for 1-D or 2-D structures. However, even for commonly identified structures the dimension number is not always as expected. For example, the magnetopause current sheet sometimes is not 1-D but has some small 2-D structures embedded (e.g., Sonnerup and Guo 1996). Magnetic flux ropes, which are generally regarded as 2-D, may actually be 3-D. In such situations the GS method is not applicable. Therefore, the examination of the structure dimension number with multi-points data is desirable.

In the past and recent years, the dimensionality based coordinated system in data analysis has been established in various cases. In this Section first we will introduce some traditional and new single spacecraft methods, followed by a review of some multi spacecraft approaches including a method directly through the definition of dimensionality (dimension number). The comparison of the methods will be made in Sect. 4.3.

### 2.1 Sonnerup-Cahill Minimum/Maximum Variance Analysis (MVA) Based Coordinate System

The first systematic and quantitative method for establishing a Cartesian coordinate system is the minimum variance analysis (MVA) method proposed by Sonnerup and Cahill (1967). The MVA based coordinate system is a kind of principle-axes coordinate system. It is the most commonly used method to analyze current layers (e.g., magnetopause, shock, or tail current sheet) and was developed using magnetic field measurements in near-Earth space (Sonnerup and Cahill 1967). This method is easy to understand and implement, and it can always provide a Cartesian coordinate system, which makes it very powerful and useful in the space data analysis community. There is no doubt that coordinate systems established by using the method of MVA (Sonnerup and Cahill 1967) have played and will continue to play an indispensable role in single satellite data analysis and still in multi-satellite data analysis.

For 1-D cases such as a current sheet, there is only a single variation direction (the normal direction) and this can be found using MVA analysis, leading directly to the construction of a D-based coordinate system. For 2-D cases, the construction of a D-based coordinate system is indirect. Here is one approach (see details in Sect. 2.2): first we build a coordinate system using the three eigenvectors, \(L\), \(M\) and \(N\), which indicate maximum, intermediate and minimum variance directions from the MVA method. Then we can rotate any of them to obtain the invariant axis using the method mentioned in Hu and Sonnerup (2002). In principle we do not need these \(L\), \(M\) or \(N\)—we can just use an arbitrary direction as an initial guess and then rotate it to obtain the invariant axis using a minimization procedure. Once we have determined the invariant axis, the construction of the D-based coordinate system is almost complete.

When it was first introduced, the MVA method was based on the assumption that the boundary is 1-D (Sonnerup and Scheible 1998), such that the magnetic field along the normal of the 1-D structure does not vary either temporally or spatially (this requires that both the magnetic and electric fields should be 1-D, i.e., they only vary along one direction). For many 2D or 3D structures it is also very useful to provide a local coordinate system (not D-based), although sometimes the physical interpretation of the original axes may not be very clear. This method gives three orthogonal axes based on the magnetic field measurements not at one time moment, but a time interval between two observational time points, which are selected arbitrarily but sufficiently far apart to use enough sampled data. Then in some cases using different time intervals one may obtain different axes, indicating finer scale structure; for example, when there are some sublayers within a current sheet.

A detailed introduction of the method can be found in (Sonnerup and Scheible 1998). The starting point of the MVA method is this: for a 1-D magnetic structure, the condition that \(\nabla \cdot \vec{B} =0\) implies, for a suitably rotated set of coordinate unit vectors \(( \vec{n}_{1}, \vec{n}_{2}, \vec{n}_{3} )\), that \({\partial B_{n1}} / \partial n_{1} = {\partial B_{n2}} / {\partial n_{2}} = {\partial B_{n3}} / {\partial n_{3}} = 0\). This is because, for a 1-D structure, variations in all components of \(\vec{B}\) must already be zero in two of the three directions (e.g. \(\vec{n}_{1}\) and \(\vec{n}_{2}\); \(\partial / \partial n_{1} = \partial / \partial n_{2} = 0\)), therefore $\mathrm{\nabla}\cdot \stackrel{\rightharpoonup}{B}=\partial {B}_{n3}/\partial {n}_{3}=0$ for the third direction (Actually this is the coordinate system which can be determined by the three eigenvectors of the minimum directional derivative (MDD) method described in Sect. 2.4). This means that \(B_{n3}\) does not change along the direction ${\stackrel{\rightharpoonup}{n}}_{3}$. Similarly, \(\nabla \times \vec{E} =0 \) in the direction of ${\stackrel{\rightharpoonup}{n}}_{3}$, if \(\vec{E}\) is also 1-D (Note that when \(\vec{B}\) is 1-D, it is not always guaranteed that \(\vec{E}\) is also 1-D, as discussed in Sect. 4.2). According to Faraday’s law $\partial \stackrel{\rightharpoonup}{B}/\partial t=-\mathrm{\nabla}\times \stackrel{\rightharpoonup}{E}$, we obtain that \(\partial B_{n3}/\partial t = 0\), namely \(B_{n3}\) does not change with time. Then, for a 1-D structure, \(B_{n3}\) does not vary either in time or space, that is, it is always constant. To find the direction of ${\stackrel{\rightharpoonup}{n}}_{3}$ which makes \(B_{n3}\) nearly constant, we can use a set of (\(N\)) magnetic field measurements over a given time interval to calculate the variance of the magnetic field in a direction \(\vec{n}\), given by: \(\sigma _{n}^{2} = \frac{1}{N} \sum_{i=1}^{N} ( B_{n} ( i ) - \langle B_{n} ( i ) \rangle ) ^{2}\) (where \(B_{n} ( i ) = \vec{B} ( i ) \cdot \vec{n}\)) and find the direction \(\vec{n}\) that minimizes \(\sigma _{n}^{2}\). In practice, as shown in detail by Sonnerup and Scheible (1998), this can be expressed as a conditional minimization/maximization problem and can be solved by calculating the eigenvalues and eigenvectors of a symmetric matrix $\overleftrightarrow{M}=\u3008\stackrel{\rightharpoonup}{B}\stackrel{\rightharpoonup}{B}\u3009-\u3008{\stackrel{\rightharpoonup}{B}}_{i}\u3009\u3008{\stackrel{\rightharpoonup}{B}}_{j}\u3009$. The three eigenvalues of $\overleftrightarrow{M}$ (\(\lambda _{1}\), \(\lambda _{2}\), \(\lambda _{3}\)) are real and the eigenvectors \(L\), \(M\) and \(N\) are perpendicular to each other. Then the three eigenvectors build a new coordinate system. If the satellite passes through a 1-D structure this is indicated by (\(\lambda _{1}\), \(\lambda _{2} \gg \lambda _{3}\) (the converse of this may not be true, see Sonnerup and Scheible (1998) and the discussion below in this section). In this case the normal direction of this 1-D structure is just along the eigenvector corresponding to the minimum eigenvalue, ${\stackrel{\rightharpoonup}{n}}_{3}$. Then, for a 1-D case the three orthogonal eigenvectors derived from $\overleftrightarrow{M}$ build a D-based coordinate system. But if \(\lambda _{1}\), \(\lambda _{2} \gg \lambda _{3}\), the structure is not necessarily 1-D, see Sonnerup and Scheible (1998) and the discussion below in this section.

^{∘}) for MVAB and MVAJ, respectively, only when the impact parameter (IP) is close to zero for flux ropes with non-zero axial fields. Nevertheless, MVAJ has been successfully used in the analysis of data from the Cluster mission (Escoubet et al. 2001) to determine the axis of a flux rope (e.g., Pu et al. 2005) and a discontinuity (Haaland et al. 2004; Rezeau et al. 2018). The results of MVA also depend on the model of the structure (Tian et al. 2019; also see Lepping et al. 1990 for a similar experiment). For example, when the axial field is zero, \(N\) from MVAB or \(L\) from MVAJ can well characterize the axis direction (Fig. 1d and g).

The MVA methods can not only be used to analyze magnetic field and current density, but also can be applied to the electric field (Sonnerup and Scheible 1998), mass flow $\rho \stackrel{\rightharpoonup}{V}$ (e.g., Sonnerup and Scheible 1998; Zhao et al. 2016), velocity vector $\stackrel{\rightharpoonup}{V}$ (e.g., Knetter 2005; Ling et al. 2018), and other vector fields. For a 1-D structure, as we mentioned above, the magnetic field along normal does not vary with both time and space. Theoretically, this is not valid for 2-D or 3-D structures but we can still perform the MVA on a time series of data to obtain a coordinate system which is in many cases better than the original system for the problem we need to analyze.

When one uses the GS reconstruction method (e.g., Sonnerup and Guo 1996; Sonnerup et al. 2006; Hasegawa et al. 2007; Tian et al. 2014, 2019) to reconstruct a flux tube, as it is a requirement to have a sufficiently accurate axis, the minimum or medium variation direction from MVA needs to be rotated to an angle to approach the real axial direction of the flux tube (e.g. Hu and Sonnerup 2002); then a D-based coordinate system is built. In this way, MVA can act as an indirect way to build a D-based coordinate system. A GUI interface for the MVA method can now be accessed in the Space Physics Environment Data Analysis System (SPEDAS).

### 2.2 D-Based Coordinate System for a 2-D Structure Based on Grad–Shafranov Reconstruction Method

The theories of series of Grad–Shafranov (GS) or MHD reconstruction methods are 2-D based in a D-based coordinate system, and the reconstructed plane is chosen to be the plane perpendicular to the invariant axis (e.g. Hau and Sonnerup 1999; Hu and Sonnerup 2002; Hasegawa et al. 2007; Teh et al. 2007; Sonnerup et al. 2006, 2016; Sonnerup and Teh 2008; Hasegawa et al. 2017). This reconstruction method has been applied to 2-D flux ropes (e.g., Hau and Sonnerup 1999; Hu and Sonnerup 2002; Hasegawa et al. 2007), the magnetopause current sheet (e.g., Hasegawa et al. 2004), reconnection structures (Teh et al. 2010), and drift mirror structures (e.g., Tian et al. 2012). The construction of a D-based coordinate as the first step is essential to the whole reconstruction.

### 2.3 Multi-point Timing—Setting a D-Based Coordinate System for a 1-D Structure

For 1-D structures, Timing methods can help to build a D-based coordinate system after finding the normal direction. Since one can get the normal and velocity at the same time from this approach, we will discuss it in detail in Sect. 3.3. There are a number of different versions of this method that differ based on their starting assumptions. For example, assuming the velocity is constant we have the CVA (Constant Velocity Approach: e.g., Russell et al. 1982; Knetter et al. 2004), while assuming the thickness is constant we get CTA (Constant Thickness Approach: Haaland et al. 2004). Other related approaches are DA (Discontinuity Analyzer, Dunlop and Woodward 1998) and MTV (Minimum Thickness Variation, Paschmann et al. 2005). In this review we will mainly discuss the CVA.

### 2.4 Method of Building a D-Based Coordinate System Through Definition of Dimensionality: Minimum Directional Derivative (MDD) Analysis

Shi et al. (2005) proposed a method directly based on the definition of dimensionality. Since this analysis method is derived from looking for the minimum derivative along various directions, it was named as “Minimum Directional Derivative (or Difference)” analysis, or MDD analysis in short. Note that although other ways of building a D-based Coordinate System are not as straightforward as the MDD method from the definition of the dimension number, they are still very necessary, especially when the estimation of field gradient fails, which happens in many cases. A GUI interface for the MDD method can now be accessed in the SPEDAS.

#### 2.4.1 Review of the Analysis Processes

First we discuss the dimension number determination for the magnetic field. For other parameters like electric field or flow field the algebraic manipulations are the same. For a 1-D or 2-D structure, if a certain direction $\stackrel{\rightharpoonup}{n}$ is along the invariant direction, i.e. along which all the parameters remain constant, from the definition of dimensionality we mentioned in Sect. 1, it will certainly satisfy that the directional derivative along $\stackrel{\rightharpoonup}{n}$ for all component of the magnetic field is equal to zero, i.e., \(\partial B_{x}/\partial n=0\), \(\partial B _{y}/\partial n=0\) and \(\partial B_{z}/\partial n=0\), where \(x\), \(y\), and \(z\) are the axes of a certain coordinate system such as GSE, and then one finds ${(\partial \stackrel{\rightharpoonup}{B}/\partial n)}^{2}={(\partial {B}_{x}/\partial n)}^{2}+{(\partial {B}_{y}/\partial n)}^{2}+{(\partial {B}_{z}/\partial n)}^{2}=0$. To find the invariant direction $\stackrel{\rightharpoonup}{n}$, we just need to find the minimum value of ${(\partial \stackrel{\rightharpoonup}{B}/\partial n)}^{2}$. Therefore we must first calculate the gradient of the magnetic field.

It is worth noting here that $\stackrel{\rightharpoonup}{n}$ and $-\stackrel{\rightharpoonup}{n}$ are the same eigenvector (the same situation also appears in the MVA analysis, see the discussion in Sonnerup and Scheible (1998) when calculating eigenvectors. For an ordered visualization of the results, one way is to set arbitrarily the \(x\) (or \(y\), \(z\)) component of $\stackrel{\rightharpoonup}{n}$ to be positive so that we can get a series of directions which can be compared with each other, and one can also calculate the average direction or check for variations of the structure.

Another point we would like to mention here is the attempt of finding the quantitative index of dimension number in order to visualize the effective dimensionality more easily. Rezeau et al. (2018) recently introduced three parameters that may be used as proxies, \(D_{1} = (\lambda _{\max } - \lambda _{\mathrm{mid}} )/\lambda _{\max } \), \(D _{2} = (\lambda _{\mathrm{mid}} - \lambda _{\min } )/\lambda _{ \max } \) and \(D_{3} = \lambda _{\min } /\lambda _{\max } \), which all vary from 0 to 1 and whose sum is 1. When \(\lambda _{\max }\), \(\lambda _{\mathrm{mid}}\ \mbox{and} \lambda _{\min }\) are comparable to each other, one obtains \(D_{1} \approx 0\) and \(D_{2} \approx 0\), while \(D_{3} \approx 1\), indicating a quasi-3-D case. When \(\lambda _{\max } \gg \lambda _{\mathit{mid}}, \lambda _{\min } \), one obtains \(D_{1} \approx 1\), while \(D_{2} \approx 0\) and \(D_{3} \approx 0\), which indicates a quasi-1-D case. When \(\lambda _{\max },\lambda _{\mathrm{mid}}\gg \lambda _{\min } \) one obtains \(D_{1} \approx 0\), \(D_{3} \approx 0\), while \(D_{2} \approx 1\), which indicates a quasi-2-D case. However, the difference between these three cases are not always clear and the three proxies are not always ideal. Considering a flux rope with \(\lambda _{\max } = 5\), \(\lambda _{\mathrm{mid}} = 1\) and \(\lambda _{\min } = 0.1\), for example, we get the dimensionality proxies \(D_{1} = 0.8\), \(D_{2} = 0.18\) and \(D_{3} = 0.02\). The structure can be 1-D but it shows a slightly 2-D character since \(D_{2}\) is not negligible (1-D but much more 2-D than 3-D). The fact that \(D_{1} > D _{2}\) indicates that the tube is strongly flattened in one direction, which shows a transition between 1D (tube flattened) and 2D (circular tube). Such flux rope structures have been shown in Shi et al. (2006) from Cluster data and Tian et al. (2019) from MMS data.

Moreover, direct comparison of eigenvalues may overestimate the difference between spatial gradients, since the eigenvalues are actually the square of the spatial gradients along the corresponding eigenvectors. Since \(\sqrt{\lambda } \) is equivalent to the directional derivative with the same units, we can also use \(\sqrt{\lambda } \) instead of \(\lambda \) in the calculations stated above. Tian et al. (2019) have also introduced some other parameters to indicate the dimension number.

Denton et al. (2010, 2012) have proposed a modified method and tested it using simulation data, which will be discussed in Sects. 4.1 and 4.4. Rezeau et al. (2018) have also proposed generalized MDD methods, which will be mentioned in Sect. 4.2.

#### 2.4.2 Normal of a 1-D Structure and D-Based Coordinate

For a 1-D structure, as discussed in Sect. 1, all the parameters vary only in one direction, i.e., the maximum derivative direction, which is also the normal of the structure. Therefore, we can use the MDD analysis to determine the normal of a quasi-1-D discontinuity and then build a D-based coordinate system.

For a 1-D case, the maximum derivative direction ${\stackrel{\rightharpoonup}{n}}_{max}$ from the MDD analysis is along the gradient of the total magnetic field. This can be demonstrated as follows: in the MDD coordinate system, for a 1-D structure, \(\nabla B = (0,0,\partial B/ \partial n_{\max})\) is just along the ${\stackrel{\rightharpoonup}{n}}_{max}$ direction. In the same way, for 2-D cases one can find that \(\nabla B = (0,\partial B/\partial n_{\mathrm{mid}},\partial B/ \partial n_{\max})\) is in the plane perpendicular to ${\stackrel{\rightharpoonup}{n}}_{min}$, not solely along ${\stackrel{\rightharpoonup}{n}}_{max}$ or ${\stackrel{\rightharpoonup}{n}}_{\mathrm{mid}}$.

Unlike the MVA method, which generally obtains results by using a series of data samples during an interval by a single satellite, using the MDD analysis we can obtain the direction at every observed moment using multipoint measurements. Therefore, MDD can in principle see the time variation of the directions. For example, in Fig. 4 of Shi et al. (2005), the maximum directions at some points in the boundary layer (shading area) have some rotations to the mean normal direction, implying that the layer may not be spatially uniform or has some temporal variations.

In data analysis, Shi et al. (2005, 2009a, 2009b), Sun et al. (2010) and Yao et al. (2016) have calculated the normal direction using Cluster data, and Yao et al. (2017, 2018) and Rezeau et al. (2018) have applied this approach to MMS data.

#### 2.4.3 Invariant Axis and D-Based Coordinate for a 2-D Structure

If the observed flux tube is a quasi-2-D structure, we can determine its invariant axis direction using the MDD analysis method, and then we can obtain a D-based coordinate system by determining the invariant axis. Shi et al. (2005) have applied the analysis on a modeled flux rope by Elphic and Russell (1983) and a flux rope from Cluster observations. Denton et al. (2016) and Hasegawa et al. (2017) have applied the analysis on a magnetic reconnection site using MMS data.

Recent studies using MMS data show that a combination of MDD and MVA provides more reasonable estimates of the \(L\)–\(M\)–\(N\) coordinate systems of approximately 2-D current sheets during ongoing reconnection than MDD or MVA only. Here the \(L\) axis is along the direction of the reconnecting magnetic field component, the \(N\) axis is perpendicular to the current sheet, and the \(M\) axis is along the reconnection line (the \(X\) line) in the 2-D model. Denton et al. (2018) developed a hybrid method in which the normal (\(N\)) is estimated as the maximum directional derivative of the magnetic field and the \(L\) axis is along the maximum variance direction of the magnetic field. Small adjustments are necessary to make the \(L\)–\(M\)–\(N\) axes strictly orthogonal to each other (see Appendix of Denton et al. 2018 for details). Genestreti et al. (2018) showed that the best \(L\)–\(M\)–\(N\) coordinate system for a magnetotail reconnection event can be estimated by a combined MDD and MVAVe (minimum variance analysis of the electron velocity) method. In their study, the \(M\) axis is defined to be along the cross product of the \(N\) axis from MDD and the maximum variance direction of the electron velocity (which turns out to be roughly along the \(L\) axis), and the \(L\) axis completes the right-handed orthogonal system.

For a 3-D structure, if it is not perfectly isotropic, it can still have maximum, medium and minimum derivative directions. Then we can still get its three principle axis from MDD analysis and a D-based coordinate system can still be built.

### 2.5 D-Based Coordinate System for a 2-D Structure Based on MVA of the Magnetic Pressure Gradient

As mentioned above it is found that the MVA on the electric current density, i.e., MVAJ is sometimes valid for finding a flux rope invariant axis. Then the D-based coordinate system can be built when studying a flux rope using MVAJ if we can obtain accurate current observations/estimations inside the flux rope.

Other single or multi-point methods for building a D-based coordinate system for a flux rope have also been developed. Assuming axial symmetry, Rong et al. (2013) developed a method to obtain the invariant axis of a flux rope. Zhang et al. (2013) and Yang et al. (2014) using the methods of Shen et al. (2007)’s curvature determination methods studied some force free flux ropes.

### 2.6 Minimum Gradient Analysis (Local-MVA-Like Method)

Using this method, we cannot directly build a D-based coordinate system. However, the MVAB method can help find the \(L\) direction which is difficult for the MDD analysis. Then in some 1-D cases, as described in (2.7) using the combination of MDD and MGA methods we may find different axes for the D-based coordinate system. Note that, if we use the four satellite data at every moment in time to perform the MVA, one can expect the same results with the MGA.

### 2.7 D-Based Coordinate System for a 2-D Structure Based on Current Density Measurements

The MMS mission has for the first time enabled sufficiently accurate measurements of the electric current density with the plasma instruments (e.g., Eastwood et al. 2016; Phan et al. 2016), and that capability allowed for the development of a new method for the invariant axis orientation of steady, 2-D structures (Hasegawa et al. 2019). The method can be used to estimate the orientation of the X line and flux rope axis from single-spacecraft measurements of the magnetic field and current density.

## 3 Frame of Reference

Here the frame of reference in which the observer resides is called the observational frame. If we can find a reference frame in which the observed magnetic field does not vary with time, then this is a steady magnetic or an electrostatic structure, and this reference frame (which moves with the magnetic field structure) is often called a ‘proper frame’ (e.g., Khrabrov and Sonnerup 1998b; Sonnerup et al. 2013). Obviously, in this frame ${(\frac{\partial \stackrel{\rightharpoonup}{B}}{\partial t})}_{\mathrm{str}}=-\mathrm{\nabla}\times {\stackrel{\rightharpoonup}{E}}_{\mathrm{str}}=0$, where the subscript ‘str’ indicates the field quantities in the magnetic field structure reference frame. To find this frame in which the curl of the electric field vanishes, one easy way is to let the electric field vanish in a frame. Then we have the deHoffmann–Teller (HT) frame, which can be determined by a single satellite method, as discussed in Sect. 3.1. Other single satellite methods will be discussed in Sect. 3.2, followed by some multi spacecraft methods discussed in later subsections.

### 3.1 Frame in Which Electric Field Disappears: deHoffmann–Teller Frame

De Hoffmann and Teller (1950) first introduced the HT reference frame in the study of an MHD shock, where the electric field ${\stackrel{\rightharpoonup}{E}}_{\mathrm{str}}$ disappears in this reference frame. Obviously, if ${\stackrel{\rightharpoonup}{E}}_{\mathrm{str}}=0$, ${(\frac{\partial \stackrel{\rightharpoonup}{B}}{\partial t})}_{\mathrm{str}}=-\mathrm{\nabla}\times {\stackrel{\rightharpoonup}{E}}_{\mathrm{str}}=0$ must be satisfied. If the HT reference frame exists, then the magnetic field versus time observed by a satellite is only caused by the motion of a quasi static magnetic field structure relative to the satellite. The ultimate goal of the deHoffmann–Teller (HT) analysis is to find the velocity of the HT reference frame ${\stackrel{\rightharpoonup}{V}}_{\mathrm{HT}}$, using a set of discretely sampled data points in the practical analysis. This generally involves the use of the least squares method to search for the minimum value of residual electric field in the new reference frame. Details can be seen in the review by Khrabrov and Sonnerup (1998b).

A very prominent advantage of the HT analysis is that one can find some indicators that may be used to estimate the reliability of the analysis results. In short, one can compare the electric field measured by the satellite and the electric field caused by the motion of the HT frame to determine whether the resulting HT frame is reasonable. If they are very close, the electric field in the HT reference frame should be very close to zero. One specific approach is to draw a scatter plot of electric field components in the satellite frame versus the corresponding components of the field in the HT frame. If the slope of the line of best fit and the correlation coefficient is close to 1, the reliability of the HT should be good. Another way is to calculate the ratio of the mean square of the residual electric field in the HT frame and the mean square of the original electric field in the satellite frame. A measure of the reliability of the HT frame is then given by the reciprocal of this ratio. However, we should be careful when using these indicators. If the derived HT velocity is very high, as expected in the solar wind, the correlation coefficient is naturally high (i.e., the ratio is small). So they are not good proxies of a good HT frame. The correlation coefficient or ratio should be calculated not in the spacecraft frame, but in the frame in which the average plasma flow velocity is zero, as has been done in Hasegawa et al. (AG, 2015).

However, the requirement of ${\stackrel{\rightharpoonup}{E}}_{\mathrm{str}}=0$ is too strict (i.e., it is sufficient for a proper frame, but not necessary). In many cases, such as perpendicular shock (with a cross shock electric potential inside the ramp), some magnetic flux ropes (see discussion by Sonnerup and Hasegawa 2005) and other structures possessing a curl-free electric field in the frame moving with the structure, a proper frame can still exist, but cannot be obtained through the HT analysis. In some structures such as shocks and other discontinuities, there are often some intrinsic electric fields within the layer along the normal, which can affect the quality of the determination of the frame velocity from the direct HT analysis, and may be important for the understanding of physical processes within the layer. When performing HT analysis on these structures, we should manually exclude the data points within the layer to obtain a correct proper frame, see reviews in Khrabrov and Sonnerup (1998a, 1998b, ISSI book) and Paschmann and Sonnerup (2008, ISSI book). In these cases, the use of SH method (Sonnerup and Hasegawa 2005) or STD (Sect. 3.4) may be helpful in finding the proper frame.

A revised HT analysis can also find an estimate of the acceleration (Khrabrov and Sonnerup 1998a, 1998b, ISSI book), but this is an average (constant) acceleration over the time-domain of the sampled data. For instantaneous velocity calculations at every time sample (i.e. allowing variable acceleration) one can refer to Sect. 3.3.

### 3.2 Proper Frame Obtained from Single Point Data: Minimum Faraday Residue (MFR) and Sonnerup–Hasegawa (SH) Methods by Assuming a Priori the Dimensionality (Dimension Number) of a Structure

Since multi-point data sources are currently limited to the Cluster and MMS missions, finding a proper frame from a single satellite when the HT analysis fails is still very useful. Several novel attempts have been made previously. Assuming a 1-D structure, minimum Faraday residue analysis (MFR) (Terasawa et al. 1996; Khrabrov and Sonnerup 1998b) and minimum mass flux residue analysis (MMR) (Sonnerup et al. 2004) have been proposed. For a 1-D structure, Faraday’s law requires that the components of the electric field tangential to the layer are constant, and then a least squares method can be performed to obtain the normal and the velocity along the normal. Sonnerup et al. (2006, 2007) have suggested some unified approaches which can be applied to any measured quantity that follows a classical conservation law. See a detailed review in Sonnerup and Teh (2008, ISSI book).

For a time invariant structure, it is required that \(\partial B/ \partial t=0\) in the proper frame we are looking for. According to Faraday’s law, \(\nabla \times E = -\partial B/\partial t\), so \(\nabla \times E = 0\). Further, if the structure is 2-D, the electric component Ez along the invariant axis should be constant across the structure. Note that the components perpendicular to the invariant axis are not necessarily zero, so a HT frame may not exist. Sonnerup and Hasegawa (2005) have proposed a scheme (hereafter referred to as the SH method) to derive the direction along which the electric field component has minimum variance. By this method, the orientation of the invariant direction and the velocity components of the structure perpendicular to the invariant direction can be obtained. For structures of magnetic flux rope type, the SH method can give satisfactory results, consistent with estimates from other methods, e.g., from a multi-spacecraft method based on G–S reconstruction (Hasegawa et al. 2006). However, other attempts show that the SH method does not work for most observed as well as numerically simulated reconnection events (Teh and Sonnerup 2008; Denton et al. 2010, 2012). Sonnerup et al. (2013) theoretically discussed the reasons for such shortcomings, and made clear that a significant, non-removable, non-uniform electric field in the plane transverse to the invariant direction is required for the method to work properly. It is also found that the results are sensitive to deviations from strict two-dimensionality and time stationarity.

If we can combine the MDD and MFR/SH methods for multi-point data analysis, we may obtain more reliable results. For example, we can use the MDD method to find a structure close to 1-D, and then use MFR to get the normal direction and velocity.

### 3.3 Triangulation for 1-D/2-D Structures: Four Spacecraft Timing

Here is another method to find the normal of a 1-D structure, and then build a D-based coordinate system and reference frame. Burlaga and Chao (1971) and Russell et al. (1983) developed and used the Triangulation method, also named Timing method, to study interplanetary discontinuities. It is used for a planar structure crossed by at least four spacecraft. A planar structure is actually a 1-D structure, in which all field quantities vary only in one direction, i.e., its normal direction. If a structure has a finite thickness rather than lying in a plane, the timing method is still valid as long as the structure is one-dimensional. So the traversal of a 1-D structure is the basic assumption of Triangulation method. The original Triangulation method also assumes that the velocity of the 1-D boundary does not vary during the crossing of all spacecraft, and then it is also called the ‘Constant Velocity Approach’ or CVA. If one assumes that the velocity can be changed but the boundary thickness is constant, the approach can be modified to a ‘Constant Thickness Approach’ or CTA, which is summarized in Haaland et al. (2004) and Sonnerup and Teh (2008).

In addition to the 1-D assumption, the structure must be quasi-static, such that when the structure is crossed by all satellites, its normal direction does not change during the interval. Recently Plaschke et al. (2017) performed a time-varying Timing velocity determination, using 3 s long sliding intervals of high time resolution data from four MMS satellites, by computing the cross-correlation functions of each spacecraft pair to obtain \(\Delta t_{ij}\). Knetter (2005), Xiao et al. (2015), and Yao et al. (2016, 2017, 2018) further considered the uncertainties of such a calculation.

Since the traditional Timing method is only applicable to the 1-D case and the MTA method can be used in 1-D or 2-D cases, one may use the MDD analysis (when the satellite separation is small enough so that the gradient calculation is valid) to determine the structure dimension number and then perform the traditional timing or MTA methods. For example, when calculating the velocity of magnetic peaks, Yao et al. (2018) first determined the dimension number of a structure with MDD analysis, found its boundary can be deemed as 1-D, and then used traditional timing methods to calculate the normal direction and propagation velocity. The velocity can also be obtained by the method introduced in Sect. 3.4, and the results from different methods can confirm each other.

The timing velocities are calculated by each magnetic field contour surface. Therefore, according to these series of timing velocities and directions, one can obtain the velocity perpendicular to the axis using similar mathematical procedures to those used in MVA and MDD. From some examples, Zhou et al. (2006b) found that there is little difference between the results of the MTA and STD method introduced in Sect. 3.4 for a 2-D structure.

### 3.4 Proper Frame for a (Quasi-) Stationary Structure: Spatio-Temporal Difference (STD) Frame from Multi-point Data

As mentioned in Sect. 3.1, when performing the HT analysis, we require the electric field to vanish in the frame we want to find, which is too strict because in some cases the electric field does not vanish but we only find that the curl of electric field disappears in a proper frame where ${(\frac{\partial \stackrel{\rightharpoonup}{B}}{\partial t})}_{\mathrm{str}}=0$. In addition, the traditional timing method assumes a 1-D structures—i.e. the dimension number of a structure is assumed before performing the analysis. Trying to avoid these problems, Shi et al. (2006) developed a method of velocity calculation (or frame determination) for any structure dimension number, known as the “Spatiotemporal Difference” (STD) analysis of the magnetic field. A GUI interface for the STD method can now be accessed in the SPEDAS.

#### 3.4.1 Introduction to the Analysis

In space observation, we have measurements of various parameters, such as moments (including density, temperature, and velocity) and vector fields (electric field and magnetic field, each of which contain three components). In principle we just need to pick any three of these quantities (each component of a vector field can be used as one field quantity) to replace \(\phi \) in (3.2) and obtain three linear equations. The calculation of \(\nabla \phi \) needs data from at least four spacecraft (see Sect. 2.4.1 in detail). By utilizing the finite difference approximation, \(\frac{\partial \phi }{\partial t} \big \vert _{\mathrm{sc}}\) at every moment can be calculated using data from one spacecraft or mean values from multi-spacecraft data. Then we can obtain the three components of the vector ${\stackrel{\rightharpoonup}{v}}_{\mathrm{str}}$ by solving the three linear equations.

Here it is worth noting that when calculating \(\partial \vec{B}/\partial t\vert _{\mathrm{sc}}\) we can obtain the result with second order accuracy by using the central finite difference \((\vec{B}_{i+1}-\vec{B}_{i-1})/ (t_{i+1}-t_{i-1})= (\vec{B}_{i+1}-\vec{B}_{i-1})/2 \Delta t \). If we use the mean value measured by the four satellites, that is a linear interpolation of the measured magnetic field to the barycenter of the tetrahedron, as has been demonstrated in Harvey (1998). The time step length \(\Delta t\) should be set according to the characteristic length of the observed structure, neither too short nor too long. If the step length \(\Delta t\) is too short, the measurement data of the magnetic field may often have a lot of short time disturbances which will influence the calculation of \(\partial \vec{B}/\partial t \vert _{\mathrm{sc}}\) and then the velocity of structure. If \(\Delta t\) is too long, the accuracy of the difference will be poor. Empirically, we suggest that \(\Delta t\) can be taken as \({\sim} 1/10\) of the characteristic time scale of the structure. For example, when a current sheet crossing takes ∼1 min, the \(\Delta t\) can be taken as ∼6 seconds.

For a 3-D structure, the calculation is straightforward. The three components of \(\vec{V}_{\mathrm{str}}\) can be directly calculated by solving (3.2), expanded as three linear equations with three unknowns.

However, for 1-D or 2-D structures, there must be at least one direction $\stackrel{\rightharpoonup}{n}$ satisfying \(\partial /\partial n \sim 0\) and then the determination of the magnetic gradient tensor has \(\det (\nabla \vec{B})\sim 0\). This is the reason why directly solving (3.2) will produce inaccurate solutions which may result in apparently turbulent velocity components. This can be shown in Figures in Sect. 3.4.3 and 3.4.4 for 2-D and 1-D structures: the inaccurate solution along the invariant direction will be distributed in all the three components of the velocity data in the GSE coordinate system and makes all the three components contain large uncertainties.

Therefore, we expect that from the magnetic field data, we can only obtain a reliable velocity determination along one direction for a 1-D structure, and along two directions for a 2-D structure. To solve this problem, we need to use the MDD method to find a structure’s dimension number and its characteristic (principal) directions, using multi-point magnetic field measurements, as introduced in Sect. 2.4.

An alternative way is to first calculate the three components of the velocity by solving the difference equations of (3.3), and then project the result to the three directions (\(\lambda _{\max } \), \(\lambda _{\mathrm{mid}} \) and \(\lambda _{\min } \)). The velocities along the maximum direction (for a 1-D structure) or maximum and intermediate directions (for a 2-D structure) can have a relatively reliable accuracy, but the other direction(s) will not. Then from figures in Sect. 3.4.3 and 3.4.4, we find that the velocities along \(\mathit{Nmax}\) and \(\mathit{Nmid}\) are no longer turbulent and the only turbulent velocity is along \(\mathit{Nmin}\) for a 2-D structure, and for a 1-D structure the velocity along \(\mathit{Nmax}\) is no longer turbulent and the turbulent velocities are along \(\mathit{Nmin}\) and \(\mathit{Nmid}\). Since generally the variations are not exactly 1-D or 2-D, the corresponding eigenvalues are not exactly zero (then the \(\mathrm{det}(\nabla \vec{B} )\) is not strictly equal to zero). This point has been shown to be the case over many years of data analysis (see e.g., Shi et al. 2006, 2009a, 2009b, 2013; Sun et al. 2010). In benchmark calculations for a pure 1-D or 2-D case when some eigenvalues are zero in a certain direction, it is found that the calculation seldom overflows because of the limited digits a computer can deal with and there will be a very small deviation from pure 1-D and 2-D. Even when we get strict zero eigenvalues at some calculating points, by adding very small random perturbations (e.g., \(10^{-5}\) of the original values) to the original field, one can still get a very accurate dimension numbers and directions. This is good example that observational errors can play a positive role. This is similar to the positive effect that numerical errors can play in the numerical simulation of the fluids, where dissipation provided by the accumulated error can stabilize the numerical scheme.

The above two solutions are intrinsically identical in the analysis of real data: the transformation to the MDD coordinate system before (i.e., in the 1st way) or after (i.e., in the 2nd way) the calculation can give same results. In practice, we can use both of them and cross-check each other.

1. The MDD analysis is carried out to obtain the dimensionality (dimension number) of the structure.

2. We solve the problem by a method depending on the structure dimension number. For a 3-D structure we can calculate three components of the velocity vector after estimating the magnetic gradient tensor G at every moment and the time variation of the magnetic field, \(\partial \vec{B}/\partial t \vert _{\mathrm{sc}}\); for a 2-D (1-D) structure, we can solve (3.3) in the original coordinate system (e.g., in GSE) and then project the velocity vector onto the eigenvectors calculated from the MDD method, or calculate the velocity along two or one directions (by solving (3.4)) in the coordinate system determined by the eigenvectors of the MDD analysis. We emphasize here that the setting of time step length \(\Delta t\) is sometimes very important, and it should be set according to the characteristic length of the observed structure, neither too short nor too long, as discussed above.

3. Finally, the velocity can be obtained along the normal direction for a 1-D structure, or along the direction perpendicular to the invariant axis for a 2-D structure.

#### 3.4.2 Application to a 3-D Structure

If the three eigenvectors are stable during the crossing of a localized magnetic structure, we can use the three MDD directions to build a new coordinate system which can represent the principle directions of the structure and might also provide some help to analyze the structure.

#### 3.4.3 Application to a 2-D Structure

For a 2-D case, the velocity can be calculated only perpendicular to the axis (invariant axis). Typically we find that \(V_{ \mathrm{min} }\) fluctuates, and \(V_{ \mathrm{max} }\) or \(V_{ \mathrm{mid} }\) are often not stable themselves (mostly because the maximum and minimum direction are not stable), but \(V_{\mathrm{2D}}\) which is a vector composed of \(V_{ \mathrm{max} }\) and \(V_{ \mathrm{mid} }\) is stable and represents the motion perpendicular to the axis.

#### 3.4.4 Application to a 1-D Structure

#### 3.4.5 Application to Determine a Pass-Through Spatial Structure and Enter/Retreat Structure

## 4 Discussion

### 4.1 Uncertainties and Cautions Concerning Various Analysis Methods

The uncertainties of MVA have been discussed by Sonnerup and Scheible (1998). Errors for the Timing method have been discussed by various authors (e.g., Zhou et al. 2009; Vogt et al. 2011; Xiao et al. 2015; Plaschke et al. 2017). Here we mainly discuss the uncertainties and cautions for the MDD and STD analysis. Since these two methods are based on the estimation of gradient of the magnetic field $G=\mathrm{\nabla}\stackrel{\rightharpoonup}{B}$, just as the current estimation method does, a simple guideline is that whenever the current calculation is accurate, the MDD calculation should be accurate. The accuracy of the STD method depends not only on the accuracy of $G=\mathrm{\nabla}\stackrel{\rightharpoonup}{B}$ estimation, but also on two other factors: the accuracy of calculating the \(( \frac{\partial \vec{B}}{\partial t} ) _{\mathrm{sc}}\) and the accuracy of the steady state assumptions \(( \frac{\partial \vec{B}}{\partial t} )_{\mathrm{str}} = 0\).

Robert et al. (1998) have analyzed the current calculation uncertainties in detail. They also proposed an empirical method to measure the magnitude of the current calculation error (or the calculation quality), that is, if the magnitude of \(|\nabla \cdot \vec{B}/\nabla \times \vec{B}|\) is greater than a certain value (like 0.4), we deem the current calculation error too large and not credible. We can also use an error indicator in an MDD plot as in Fig. 6 for the building of a D-based coordinate system. Considering that sometimes the magnetic gradient is large but the current is small (e.g., in the center of a magnetic hole or a magnetic mirror (Tian et al. 2019; Yao et al. 2016, 2017; Shi et al. 2009a)), we also use another index, i.e., $\frac{|\mathrm{\nabla}\cdot \stackrel{\rightharpoonup}{B}|}{max(|\frac{\partial {B}_{i}}{\partial j}|)}\phantom{\rule{0.25em}{0ex}}(i,j=x/y/z)$, where \(\max (|\frac{ \partial B_{i}}{\partial j}|)\) is the maximum absolute value of all components of \(G= \partial \vec{B}\).

Here we summarize the error sources of the MDD method, which come from the measurement error of the magnetic field, the error in determination of the satellite relative position, simultaneity of the measurements made between different spacecraft, and truncation errors in the estimation of the matrix (spacecraft tetrahedron shape parameters, c.f. Robert et al. 1998 as a source of error are included in the truncation error) $G=\mathrm{\nabla}\stackrel{\rightharpoonup}{B}$, which have been discussed extensively before (e.g., Chanteur 1998; Harvey 1998; Robert et al. 1998; Denton et al. 2010, 2012).

This does not always mean, however, that the smaller spacecraft separations, the better the accuracy. As pointed by Robert et al. (1998), the relative measurement errors between every spacecraft pair become large at small separations, because the field measurement errors are nearly constant, and hence there should be an optimal separation distance.

Considering the measurement error \(\delta B\), the smallest gradient of magnetic field should be \({\sim }\delta B/l_{\max } \), where the \(l_{\max } \) is the largest separation among spacecraft. In the panel of eigenvalues, we can add a dashed horizontal line to indicate \(\delta B/l_{\max } \), given \(\delta B = 0.05\) nT and \(l_{\max } \) depending on the real situation. Then if the square of the eigenvalues is lower than this line, we should be careful; see examples in Figs. 8, 16 and 18.

Denton et al. (2010, 2012) discussed the MDD and STD methods and further developed them to study the magnetic reconnection point, considering various errors of the field data which may bring uncertainties in the calculation. First they tested these methods by using a magnetotail reconnection point obtained from numerical simulations assuming four virtual satellites passing through the structure. In these cases, they found that the characteristic directions of the reconnection point and the moving velocity of the reconnection point can be well determined. They considered two kinds of errors in magnetic field measurement that is worth our attention in real satellite data analysis: one is the digitization (noise) errors that randomly vary with respect to time, another is systematic inter-spacecraft calibration errors that are either constant or at least very slowly evolving in time. In general, the former is small and unlikely to be a problem, and the latter is larger and sometimes could reach 0.1 nT, in the cases of Cluster and MMS. To minimize the influence of the calibration error, Denton et al. (2010, 2012) suggested to use the gradient of the perturbed field $\delta (\mathrm{\nabla}\stackrel{\rightharpoonup}{B})=\mathrm{\nabla}\stackrel{\rightharpoonup}{B}-\u3008\mathrm{\nabla}\stackrel{\rightharpoonup}{B}\u3009$ instead of that of the total field $\mathrm{\nabla}\stackrel{\rightharpoonup}{B}$ when carrying out analysis using MDD and STD. In their simulated case, they argue that the calculation can be improved, mainly for the intermediate and minimum directions. Teh et al. (2010) reconstructed a reconnection structure by solving the steady resistive MHD equations in two dimensions, with initial inputs of field and plasma data from a single spacecraft as it passes through it, using the velocity calculation by Denton et al. (2010). However, Denton et al. (2012) also mentioned that if the background has a spatial gradient itself, removing the background may lead to a systematic deviation of the calculated results. Therefore, caution must be made when we use this modified approach.

One type of error in the STD method comes from our stationarity assumption, i.e., (\((\frac{\partial \vec{B}}{\partial t})_{\mathrm{str}}\sim 0\)), which means we can calculate its velocity only if the structure itself changes very slowly on the time scale of the motion. When we calculate ${\stackrel{\rightharpoonup}{V}}_{\mathrm{str}}$ from the STD analysis, if the structure is steady, we have ${(\frac{\partial \stackrel{\rightharpoonup}{B}}{\partial t})}_{\mathrm{str}}=-\mathrm{\nabla}\times {\stackrel{\rightharpoonup}{E}}_{\mathrm{str}}=0$ in the rest frame of the structure. To test this assumption of steady state, we can calculate $\mathrm{\nabla}\times {\stackrel{\rightharpoonup}{E}}_{\mathrm{str}}$ in the structure frame and see if it is close to zero. Substituting ${\stackrel{\rightharpoonup}{E}}_{\mathrm{str}}={\stackrel{\rightharpoonup}{E}}_{\mathrm{sc}}+{\stackrel{\rightharpoonup}{V}}_{\mathrm{str}}\times \stackrel{\rightharpoonup}{B}$, we get $\mathrm{\nabla}\times {\stackrel{\rightharpoonup}{E}}_{\mathrm{str}}=\mathrm{\nabla}\times {\stackrel{\rightharpoonup}{E}}_{\mathrm{sc}}-\stackrel{\rightharpoonup}{B}(\mathrm{\nabla}\cdot {\stackrel{\rightharpoonup}{V}}_{\mathrm{str}})+\stackrel{\rightharpoonup}{B}\cdot \mathrm{\nabla}{\stackrel{\rightharpoonup}{V}}_{\mathrm{str}}-{\stackrel{\rightharpoonup}{V}}_{\mathrm{str}}\cdot \mathrm{\nabla}\stackrel{\rightharpoonup}{B}$. From \(( \frac{\partial \vec{B}}{ \partial t} )_{\mathrm{sc}} + \vec{V}_{\mathrm{str}} \cdot \nabla \vec{B} = 0\) (assume this is valid here) and ${(\frac{\partial \overrightarrow{B}}{\partial t})}_{\mathrm{sc}}=-\mathrm{\nabla}\times {\stackrel{\rightharpoonup}{E}}_{\mathrm{sc}}$, we get $\mathrm{\nabla}\times {\stackrel{\rightharpoonup}{E}}_{\mathrm{str}}=-\stackrel{\rightharpoonup}{B}(\mathrm{\nabla}\cdot {\stackrel{\rightharpoonup}{V}}_{\mathrm{str}})+\stackrel{\rightharpoonup}{B}\cdot \mathrm{\nabla}{\stackrel{\rightharpoonup}{V}}_{\mathrm{str}}$. So if ${\stackrel{\rightharpoonup}{V}}_{\mathrm{str}}$ does not vary spatially, we should get $\mathrm{\nabla}\times {\stackrel{\rightharpoonup}{E}}_{\mathrm{str}}=0$. This result is very reasonable: at a given moment if we can calculate ${\stackrel{\rightharpoonup}{V}}_{\mathrm{str}}$ in many different positions and find it is homogeneous, then it is a coherent and steady structure. On the other hand, if ${\stackrel{\rightharpoonup}{V}}_{\mathrm{str}}$ is inhomogeneous, we will see the expansion, compression or deformation and then it is not steady. How can we test whether ${\stackrel{\rightharpoonup}{V}}_{\mathrm{str}}$ is spatially constant or not in practice? If we have far more than four satellites to calculate the ${\stackrel{\rightharpoonup}{V}}_{\mathrm{str}}$ simultaneously at different places using each combination system of four satellites, it will be easy. However, if we only have one cluster of four satellites, we may need some assumptions. For example, if we observe ${\stackrel{\rightharpoonup}{V}}_{\mathrm{str}}$ does not vary with time during the whole crossing of the structure, it is likely that the ${\stackrel{\rightharpoonup}{V}}_{\mathrm{str}}$ is also spatially constant. Another possible way is to perform a reconstruction. Reconstruction methods have been proposed and applied based mainly on single point measurements, e.g., GS reconstruction as discussed in Sect. 2.2 or on multi-point measurements, e.g., first-order expansion (Wendel and Adrian 2013; Fu et al. 2015). Higher order reconstructions based on multi-point measurements can also be possibly performed even if we only have four satellites. Assuming that the structure is stationary we can obtain its velocity, and then we can deem the whole crossing time of the structure as one time moment so that we will obtain a system with many satellites in different positions of the structure. The relative distance of different positions can be derived from the structure velocity and time difference when the satellites reach different positions. Then we can perform 2nd to 3rd order (or even higher order depending on the point number) fits to get the whole field of the structure. Chanteur (1998) has proposed to use this approach to get the higher order gradient of the field. For 2-D or 3-D cases by this way of reconstruction one may not get accurate results perpendicular to the spacecraft trajectory. However, for the 1-D case, if the variant axis is along the trajectory, we may reconstruct the field of the structure more accurately. Then if we reconstruct the structure from data of two different parts (corresponding to two different time moments) to obtain a whole picture of the structure, we may compare the reconstruction at different times. If the differences between the two results are only minor, then it can be considered stationary, provided the reconstruction is sufficiently accurate. Of course this proposed approach is too idealized. A much better way is to launch tens of satellites and then we can make better reconstruction with higher order.

Other sources of STD calculation error exist in the finite difference equations of (3.3), and generally, the truncation errors dominate, which give us two limitations of the STD method: \(l_{\mathrm{sc}} \ll l_{\mathrm{str}}\) and \(\Delta t V_{\mathrm{str}} \ll l_{\mathrm{str}}\) (see the analysis detail in Shi et al. 2006), where \(l_{\mathrm{sc}}\) is the spacecraft separation scale, \(l_{\mathrm{str}}\) is the structure’s scale, and \(\Delta t\) is the time step which we used in calculating \(( \frac{\partial \vec{B}}{\partial t} )_{\mathrm{sc}}\). Because the accuracy of the magnetic field, \(\delta B_{i}\) (where \(B_{i}\) is one component of the magnetic field), is normally dominated by systematic inter-spacecraft calibration errors, which can reach 0.1 nT for Cluster and MMS as mentioned above, the field variation during the interval \(\Delta t\) should be larger than \(\delta B_{i}\), that is \(\delta B_{i} < \Delta t \vert ( \frac{ \partial B_{i}}{\partial t} )_{\mathrm{sc}} \vert \), so the \(\Delta t\) should satisfy \(\Delta t > \frac{\delta B_{i}}{ \vert ( \frac{\partial B_{i}}{\partial t} )_{\mathrm{sc}} \vert }\). Similarly, \(l_{\mathrm{sc}}\) should satisfy \(l_{\mathrm{sc}} > \frac{\delta B_{i}}{ \vert \frac{\partial B_{i}}{\partial l} \vert }\), where \(l \) is along a certain direction. Thus, the optimal \(\Delta t\) and \(l_{\mathrm{sc}}\) satisfy \(\frac{\delta B_{i}}{ \vert ( \frac{\partial B_{i}}{ \partial t} )_{\mathrm{sc}} \vert } < \Delta t \ll \frac{l_{\mathrm{str}}}{V _{\mathrm{str}}}\) and \(\frac{\delta B_{i}}{ \vert \frac{\partial B_{i}}{ \partial l} \vert } < l_{\mathrm{sc}} \ll l_{\mathrm{str}}\). Therefore, suitable \(\Delta t\) and \(l_{\mathrm{sc}}\) should be taken for different cases.

### 4.2 Dimensionality (Dimension Number) for Different Field Quantities

In previous sections we have only considered the dimensionality of magnetic field structures. However, if we consider other fields such as current density, electric field, or velocity field, we may find that they have different dimensionalities. That is, when the magnetic field is 1D, it does not guarantee that the other parameters are also 1D.

Since the current density is $\stackrel{\rightharpoonup}{J}=\mathrm{\nabla}\times \stackrel{\rightharpoonup}{B}$, if $\partial \stackrel{\rightharpoonup}{B}/\partial n=0$, $\partial \stackrel{\rightharpoonup}{J}/\partial n=0$. So the dimension number of the current density should be equal to or smaller than that of the magnetic field. The dimension number can be smaller (e.g., 2-D for magnetic field but 1-D or constant for current density) because the current density is calculated from the spatial derivative of the magnetic field.

Rezeau et al. (2018) argue that in low beta plasmas when the magnetic field controls all the other plasma parameters, one can deem that if magnetic field is 1-D, then all the other parameters will be 1-D, while in higher beta plasma, pressure effects are important, and it is not certain that the 1-D variations of \(B\) can ensure all the plasma parameters will be 1-D from the fluid momentum equations of both ions and electrons.

We should also note that the dimensionality may be related to the coordinate system we use. For example, for an axially symmetrical structure (like some kinds of flux ropes), in D-based Cartesian coordinates, it is 2-D as \(B=B(x,y)\) if \(z\) is the invariant axis. However, if we use a cylindrical coordinate system, we find that the field only varies in the r direction, i.e., \(B=B(r)\). Then from this point of view, the structure turns out to be 1-D.

### 4.3 Comparison of Various Methods

In this section we will compare and contrast between the methods discussed in previous sections, and try to give a description of where they can be best applied. We emphasize here that there are not ‘better’ or ‘worse’ methods. Different methods will have their best application in different circumstances. In many cases we may use different methods at the same time to compare and obtain a more reliable coordinate system and reference frame.

First we discuss the difference between MVA and MDD. As emphasized by Sonnerup and Scheible (1998), \(\lambda _{3}\ll \lambda _{2}\) from the MVA method does not automatically indicate that a 1-D current layer has been traversed and for a 2-D structure, one cannot necessarily conclude that the minimum variance direction is along with the invariant axis. In other words, the fact that one or two eigenvalues are equal to zero in the MVAB method is not a sufficient condition for one or two-dimensionality (also see Dunlop and Woodward 1998; Dunlop et al. 2002). This means that one may not directly use the MVA method to determine the dimension number or invariant axis orientation of a structure observed. In the MDD analysis, if we find an eigenvalue equal to zero, it follows that ${(\partial \stackrel{\rightharpoonup}{B}/\partial n)}^{2}={(\partial {B}_{x}/\partial n)}^{2}+{(\partial {B}_{y}/\partial n)}^{2}+{(\partial {B}_{z}/\partial n)}^{2}=0$, which means the derivatives of \(B_{x}\), \(B_{y}\), and \(B_{z}\) along this eigenvector direction $\stackrel{\rightharpoonup}{n}$ should all be zero. This shows that the MDD method can provide a sufficient condition for one or two-dimensionality of a structure and then give the invariant axis directions at the same time.

Although MVAB and MDD are both looking for a coordinate system to simplify the problem, the MVA method is seeking for the extreme variation value of \(B_{n}^{2}\), and the MDD approach is seeking for the extremum of \(|\partial \vec{B}/\partial n|^{2}\). So the extreme values and eigenvector directions obtained by the two methods are often different.

Summary of MVAB result of 1D current sheet of (4.2). Random errors on the order of 0.01 nT are added to the magnetic field data to resemble real data and to avoid the singularity in calculating the eigenvalues of the matrix. Three runs have been carried out for two sets of parameters of the model. We can find that in the same model the eigenvalue and eigenvector of the intermediate and minimum directions are quite different every time, while the eigenvalue and eigenvector of maximum direction remain the same

\({B}_{{x0}}=40~\mbox{nT}\) \(B_{{y0}}=0~\mbox{nT}\) \(B_{{z0}}=0~\mbox{nT}\) | \({B}_{{x0}}=40~\mbox{nT}\) \(B_{y0}=2~\mbox{nT}\) \(B_{z0}=10~\mbox{nT}\) | |||||
---|---|---|---|---|---|---|

Run 1 for SC1 | Max | Mid | Min | Max | Mid | Min |

Eigenvalue | 769.53 | 1.23 × 10 | 8.83 × 10 | 769.53 | 1.28 × 10 | 8.01 × 10 |

Eigenvector | [1,0,0] | [0,−0.34,−0.94] | [0,−0.94,0.34] | [1,0,0] | [0,−0.99,−0.06] | [0,−0.06,0.99] |

Run 2 for SC1 | Max | Mid | Min | Max | Mid | Min |

Eigenvalue | 769.53 | 1.37 × 10 | 9.85 × 10 | 759.63 | 1.36 × 10 | 9.84 × 10 |

Eigenvector | [1,0,0] | [0,0.66,−0.75] | [0,0.75,0.66] | [1,0,0] | [0,0.67,−0.74] | [0,0.74,0.0.63] |

Run 3 for SC1 | Max | Mid | Min | Max | Mid | Min |

Eigenvalue | 769.53 | 1.23 × 10 | 7.83 × 10 | 769.53 | 1.05 × 10 | 6.81 × 10 |

Eigenvector | [1,0,0] | [0,0.99,−0.13] | [0,0.13,0.99] | [1,0,0] | [0,−0.53,−0.85] | [0,0.85,0.53] |

Summary of MVAB result of 1D current sheet as in (4.3). For this current sheet which is closer to real data in the magnetopause or magnetotail current sheet, MVA can distinguish between the mid and min directions and then obtain the correct normal

\({L}_{0} =100~\mbox{km}\) \(L_{1}=30~\mbox{km}\) \(B_{x0}=40~\mbox{nT}\) \(B_{y0} =4~\mbox{nT}\) \(B_{z0} =10~\mbox{nT}\) | |||
---|---|---|---|

SC1 | Max | Mid | Min |

Eigenvalue | 769.564 | 1.511 | 0 |

Eigenvector | [0.999,0.007,0] | [−0.007,0.999,0] | [0,0,1] |

SC2 | Max | Mid | Min |

Eigenvalue | 759.685 | 1.520 | 0 |

Eigenvector | [0.999,0.006,0] | [−0.006,0.999,0] | [0,0,1] |

Capabilities and requirements of various methods in solving different issues

MVA | MDD | STD | MGA | Timing | MTA | HT | GS | MFR | SH | Curlometer | |
---|---|---|---|---|---|---|---|---|---|---|---|

Section | |||||||||||

Works for single s/c | yes | no | no | no | no | no | yes | yes | yes | yes | no |

Works for multi s/c | no | yes | yes | yes | yes | yes | no | yes | no | no | yes |

Yields coordinate system | yes | yes | no | yes | yes | yes | no | yes | yes | yes | – |

Yields frame velocity | no | no | yes | no | yes | yes | yes | no | yes | yes | – |

Determine dimensionality | no | yes | no | no | no | no | yes | no | no | no | no |

1-D assumption | yes | no | no | no | yes | no | no | no | yes | no | no |

2-D assumption | no | no | no | no | no | yes | no | yes | no | yes | no |

Works for 1-D | yes | yes | yes | yes | yes | yes | yes | no | yes | yes | yes |

Works for 2-D | yes | yes | yes | yes | no | yes | yes | yes | no | yes | yes |

works for 3-D | yes | yes | yes | yes | no | no | no | no | no | no | yes |

Quasi-static assumption | no | no | yes | no | yes | yes | yes | yes | yes | yes | no |

Works for small separation of s/c | – | yes | yes | yes | yes | yes/no | – | – | – | – | yes |

Works for large separation of s/c | – | no | no | no | yes | yes/no | – | – | – | – | no |

To see time variation | No | yes | yes | yes | no | no | no | no | no | no | yes |

### 4.4 Potential Applications of Gradient Based Methods in Simulation Data Analysis and Other Problems

The MDD and STD methods can be effectively used in numerical simulations. In the analysis of simulated data, we can calculate dimension number, characteristic directions and velocities of simulated magnetic field structures (such as plasmoid, FTE, magnetopause current sheet, bow shock, and magnetotail current sheet). It can also be very convenient to automatically calculate the time variation of velocity and direction for the simulated structure.

In laboratory plasmas, if we have multi point magnetic probes, in principle we can also apply the same techniques to attain the proper frame and coordinate systems. This could be useful in the analysis of transient convective MHD instabilities. By the way, it is well known that many everyday objects in our world produce a magnetic field, such as vehicles, aircraft and so on. When they are moving, they carry magnetic field along with them. So using the STD method we can also measure the movement velocity of magnetic field and then we will know the speed of the object. If the magnetic field of objects does not vary with time, the first term in the right hand side of equation (1.1) is strictly equal to 0. For example, for a bar magnet as discussed in Sect. 3.4.2, its magnetic field is a dipole magnetic field, and the magnetic field on it does not vary with time. If it moves at some given velocity and we calculate the velocity using the STD method, we find our calculation is quite consistent with the given velocity (not shown here). The magnetic field of real magnetic objects may be complicated by including dipole magnetic field, quadrupole magnetic field and higher order terms. In principle, the STD method should work, and the further investigation is still ongoing. More work is required to test if this is practically useful. First we need to put a huge number of magnetometers in space (atmosphere). It might not work if magnetometers are installed on or near vehicles, because they may observe a time-independent field (in the vehicle rest frame) under a steady state condition. There may also be substantial electromagnetic radiation from the vehicle that cannot be neglected. More work should be done to investigate the feasibility of this idea.

## Notes

### Acknowledgements

This work was supported by the National Natural Science Foundation of China (grants 41774153 and 41574157), and project supported by the Specialized Research Fund for State Key Laboratories, and the International Space Science Institute (ISSI). The instrumental teams of MMS and Cluster are greatly appreciated for providing magnetic field, electric field and plasma data. All the data are available from MMS Science Data Center (https://lasp.colorado.edu/mms/sdc/public/) and Cluster Science Archive (https://csa.esac.esa.int/csa-web/). MMS data access and processing was done using Space Physics Environment Data Analysis System (SPEDAS, www.spedas.org) V3.1, see Angelopoulos et al. (2019). GUI interfaces for the MVA, MDD and STD methods can now be accessed in the SPEDAS.

## References

- V. Angelopoulos et al., The Space Physics Environment Data Analysis System (SPEDAS). Space Sci. Rev.
**215**, 9 (2019). https://doi.org/10.1007/s11214-018-0576-4 ADSCrossRefGoogle Scholar - J. Birn, M. Hesse, Geospace Environment Modeling (GEM) magnetic reconnection challenge: resistive tearing, anisotropic pressure and Hall effects. J. Geophys. Res.
**106**(A3), 3737–3750 (2001). https://doi.org/10.1029/1999JA001001 ADSCrossRefGoogle Scholar - J.L. Burch, T.D. Phan, Magnetic reconnection at the dayside magnetopause: advances with MMS. Geophys. Res. Lett.
**43**(16), 8327–8338 (2016). https://doi.org/10.1002/2016gl069787 ADSCrossRefGoogle Scholar - L.F. Burlaga, J.K. Chao, Reverse and forward slow shocks in the solar wind. J. Geophys. Res.
**76**(31), 7516–7521 (1971). https://doi.org/10.1029/JA076i031p07516 ADSCrossRefGoogle Scholar - G. Chanteur, Spatial interpolation for four spacecraft: theory, in
*Analysis Methods for Multi-Spacecraft Data*, ed. by G. Paschmann, P.W. Daly (Int. Space Sci. Inst., Bern, 1998), pp. 349–369 Google Scholar - G. Chanteur, C.C. Harvey, in
*Spatial Interpolation for Four Spacecraft: Application to Magnetic Gradients*, ed. by G. Paschmann, P.W. Daly (Int. Space Sci. Inst., Bern, 1998), pp. 349–369 Google Scholar - W. Daughton, V. Roytershteyn, B.J. Albright, H. Karimabadi, L. Yin, K.J. Bowers, Transition from collisional to kinetic regimes in large-scale reconnection layers. Phys. Rev. Lett.
**103**(6) (2009). https://doi.org/10.1103/physrevlett.103.065004 - J.M. Dawson, Particle simulation of plasmas. Rev. Mod. Phys.
**55**, 403–447 (1983) ADSCrossRefGoogle Scholar - F. De Hoffmann, E. Teller, Magneto-hydrodynamic shocks. Phys. Rev.
**80**(4), 692–703 (1950). https://doi.org/10.1103/physrev.80.692 ADSMathSciNetCrossRefzbMATHGoogle Scholar - R.E. Denton, B.U.Ö. Sonnerup, J. Birn, W.-L. Teh, J.F. Drake, M. Swisdak, M. Hesse, W. Baumjohann, Test of methods to infer the magnetic reconnection geometry from spacecraft data. J. Geophys. Res.
**115**, A10242 (2010). https://doi.org/10.1029/2010JA015420 ADSCrossRefGoogle Scholar - R.E. Denton, B.U.Ö. Sonnerup, M. Swisdak, J. Birn, J.F. Drake, M. Hesse, Test of Shi et al. method to infer the magnetic reconnection geometry from spacecraft data: MHD simulation with guide field and antiparallel kinetic simulation. J. Geophys. Res.
**117**, A09201 (2012). https://doi.org/10.1029/2012JA017877 ADSCrossRefGoogle Scholar - R.E. Denton, B.U.Ö. Sonnerup, H. Hasegawa, T.-D. Phan, C. Russel, R.J. Strangeway, B.L. Giles, D.J. Gershman, R.B. Torbert, Motion of the MMS spacecraft relative to the magnetic reconnection structure observed on 16 Oct. 2015 at 1307 UT. Geophys. Res. Lett.
**43**, 5589–5596 (2016). https://doi.org/10.1002/2016GL069214 ADSCrossRefGoogle Scholar - R.E. Denton, B.U.Ö. Sonnerup, C.T. Russell, H. Hasegawa, T.-D. Phan, R.J. Strangeway, B.L. Giles, R.E. Ergun, P.-A. Lindqvist, R.B. Torbert, J.L. Burch, S.K. Vines, Determining \(L\)–\(M\)–\(N\) current sheet coordinates at the magnetopause from magnetospheric multiscale data. J. Geophys. Res. Space Phys.
**123**, 2274–2295 (2018). https://doi.org/10.1002/2017JA024619 Google Scholar - M.W. Dunlop, T.I. Woodward, Multi-spacecraft discontinuity analysis: orientation and motion, in
*Analysis Methods for Multi-Spacecraft Data*, ed. by G. Paschmann, P.W. Daly (Int. Space Sci. Inst, Bern, 1998), pp. 271–305 Google Scholar - M.W. Dunlop, A. Balogh, K.-H. Glassmeier, Four-point cluster application of magnetic field analysis tools: the discontinuity analyzer. J. Geophys. Res.
**107**(A11), 1385 (2002). https://doi.org/10.1029/2001JA005089 CrossRefGoogle Scholar - J.P. Eastwood, T.D. Phan, P.A. Cassak et al., Ion-scale secondary flux ropes generated by magnetopause reconnection as resolved by MMS. Geophys. Res. Lett.
**43**, 4716–4724 (2016). https://doi.org/10.1002/2016GL068747 ADSCrossRefGoogle Scholar - R.C. Elphic, C.T. Russell, Evidence for helical kink instability in the Venus magnetic flux ropes. Geophys. Res. Lett.
**10**(6), 459–462 (1983). https://doi.org/10.1029/gl010i006p00459 ADSCrossRefGoogle Scholar - C.P. Escoubet, M. Fehringer, M. Goldstein, The cluster mission. Ann. Geophys.
**19**, 1197 (2001) ADSCrossRefGoogle Scholar - H.S. Fu, A. Vaivads, Y.V. Khotyaintsev, V. Olshevsky, M. André, J.B. Cao, S.Y. Huang, A. Retinò, G. Lapenta, How to find magnetic nulls and reconstruct field topology with MMS data? J. Geophys. Res. Space Phys.
**120**(5), 3758–3782 (2015). https://doi.org/10.1002/2015ja021082 ADSCrossRefGoogle Scholar - K.J. Genestreti, T.K.M. Nakamura, R. Nakamura, R.E. Denton, R.B. Torbert, J.L. Burch, F. Plaschke, S.A. Fuselier, R.E. Ergun, B.L. Giles, C.T. Russell, How accurately can we measure the reconnection rate EM for the MMS diffusion region event of 11 July 2017? J. Geophys. Res. Space Phys.
**123**, 9130–9149 (2018). https://doi.org/10.1029/2018JA025711. ADSCrossRefGoogle Scholar - J.T. Gosling, T.D. Phan, Magnetic reconnection in the solar wind at current sheets associated with extremely small field shear angles. Astrophys. J. Lett.
**763**, L39 (2013). https://doi.org/10.1088/2041-8205/763/2/L39 ADSCrossRefGoogle Scholar - S. Haaland, B.U.Ö. Sonnerup, M.W. Dunlop, E. Georgescu, G. Paschmann, B. Klecker, A. Vaivads, Orientation and motion of a discontinuity from cluster curlometer capability: minimum variance of current density. Geophys. Res. Lett.
**31**(31), 377–393 (2004). https://doi.org/10.1029/2004gl020001 CrossRefGoogle Scholar - M. Hartinger, V. Angelopoulos, M.B. Moldwin, K.-H. Glassmeier, Y. Nishimura, Global energy transfer during a magnetospheric field line resonance. Geophys. Res. Lett.
**38**, L12101 (2011). https://doi.org/10.1029/2011GL047846 ADSCrossRefGoogle Scholar - C.C. Harvey, Spatial gradients and the volumetric tensor, in
*Analysis Methods for Multi-Spacecraft Data*, ed. by G. Paschmann, P.W. Daly (Int. Space Sci. Inst., Bern, 1998), pp. 307–322 Google Scholar - H. Hasegawa, B.U.Ö. Sonnerup, M.W. Dunlop, A. Balogh, S.E. Haaland, B. Klecker, G. Paschmann, B. Lavraud, I. Dandouras, H. Rème, Reconstruction of two-dimensional magnetopause structures from cluster observations: verification of method. Ann. Geophys.
**22**, 1251–1266 (2004). https://doi.org/10.5194/angeo-22-1251-2004 ADSCrossRefGoogle Scholar - H. Hasegawa, B.U.Ö. Sonnerup, B. Klecker, G. Paschmann, M.W. Dunlop, H. Rème, Optimal reconstruction of magnetopause structures from cluster data. Ann. Geophys.
**23**(3), 973–982 (2005). https://doi.org/10.5194/angeo-23-973-2005 ADSCrossRefGoogle Scholar - H. Hasegawa, B.U.Ö. Sonnerup, C.J. Owen, B. Klecker, G. Paschmann, A. Balogh, H. Rème, The structure of flux transfer events recovered from cluster data. Ann. Geophys.
**24**, 603–618 (2006). https://doi.org/10.5194/angeo-24-603-2006 ADSCrossRefGoogle Scholar - H. Hasegawa, B.U.Ö. Sonnerup, M. Fujimoto, Y. Saito, T. Mukai, Recovery of streamlines in the flank low-latitude boundary layer. J. Geophys. Res.
**112**, A04213 (2007). https://doi.org/10.1029/2006JA012101 ADSCrossRefGoogle Scholar - H. Hasegawa, B.U.Ö. Sonnerup, S. Eriksson, T.K.M. Nakamura, H. Kawano, Dual-spacecraft reconstruction of a three-dimensional magnetic flux rope at the Earth’s magnetopause. Ann. Geophys.
**33**, 169–184 (2015). https://doi.org/10.5194/angeo-33-169-2015 ADSCrossRefGoogle Scholar - H. Hasegawa, B.U.Ö. Sonnerup, R.E. Denton, T.-D. Phan, T.K.M. Nakamura, B.L. Giles, D.J. Gershman, J.C. Dorelli, J.L. Burch, R.B. Torbert, C.T. Russell, R.J. Strangeway, P.-A. Lindqvist, Y.V. Khotyaintsev, R.E. Ergun, P.A. Cassak, N. Kitamura, Y. Saito, Reconstruction of the electron diffusion region observed by the magnetospheric multiscale spacecraft: first results. Geophys. Res. Lett.
**44**, 4566–4574 (2017). https://doi.org/10.1002/2017GL073163 ADSCrossRefGoogle Scholar - H. Hasegawa, R.E. Denton, R. Nakamura, K.J. Genestreti, T.K.M. Nakamura, K.-J. Hwang et al., Reconstruction of the electron diffusion region of magnetotail reconnection seen by the MMS spacecraft on 11 July 2017. J. Geophys. Res. Space Phys.
**124**, 122–138 (2019). https://doi.org/10.1029/2018JA026051 ADSCrossRefGoogle Scholar - L.-N. Hau, B.U.Ö. Sonnerup, Two-dimensional coherent structures in the magnetopause: recovery of static equilibria from single-spacecraft data. J. Geophys. Res. Space Phys.
**104**(A4), 6899–6917 (1999) ADSCrossRefGoogle Scholar - H. Hietala, T.D. Phan, V. Angelopoulos, M. Oieroset, M.O. Archer, T. Karlsson, F. Plaschke, In situ observations of a magnetosheath high-speed jet triggering magnetopause reconnection. Geophys. Res. Lett.
**45**(4), 1732–1740 (2018). https://doi.org/10.1002/2017gl076525 ADSCrossRefGoogle Scholar - Q. Hu, B.U.Ö. Sonnerup, Reconstruction of magnetic clouds in the solar wind: orientations and configurations. J. Geophys. Res.
**107**(A7) (2002) Google Scholar - A.V. Khrabrov, B.U.Ö. Sonnerup, Orientation and motion of current layers: minimization of the Faraday residue. Geophys. Res. Lett.
**25**, 2373 (1998a) ADSCrossRefGoogle Scholar - A.V. Khrabrov, B.U.Ö. Sonnerup, DeHoffmann–Teller analysis, in
*Analysis Methods for Multi-Spacecraft Data*, vol. SR-001, ed. by G. Paschmann, P. Daly (International Space Science Institute, Bern, 1998b), pp. 221–248, Chap. 9 Google Scholar - M.G. Kivelson, C.T. Russell,
*Introduction to Space Physics*(Cambridge University Press, Cambridge, 1995) Google Scholar - T. Knetter, A new perspective of the solar wind micro-structure due to multi-point observations of discontinuities. Ph.D. thesis (2005) Google Scholar
- T. Knetter, F.M. Neubauer, T. Horbury, A. Balogh, Four-point discontinuity observations using Cluster magnetic field data: a statistical survey. J. Geophys. Res.
**109**(A6), A06102 (2004). https://doi.org/10.1029/2003JA010099 ADSCrossRefGoogle Scholar - R.P. Lepping, J.A. Jones, L.F. Burlaga, Magnetic field structure of interplanetary magnetic clouds at 1 AU. J. Geophys. Res.
**95**, 11,957–11,965 (1990) ADSCrossRefGoogle Scholar - Y. Lin, D.W. Swift, A two-dimensional hybrid simulation of the magnetotail reconnection layer. J. Geophys. Res.
**101**(A9), 19859–19870 (1996). https://doi.org/10.1029/96JA01457 ADSCrossRefGoogle Scholar - Y. Ling, Q.Q. Shi, X.-C. Shen, A.M. Tian, W. Li, B.B. Tang, A.W. Degeling, H. Hasegawa, M. Nowada, H. Zhang, I.J. Rae, Q.-G. Zong, S.Y. Fu, A.N. Fazakerley, Z.Y. Pu, Observations of Kelvin–Helmholtz waves in the Earth’s magnetotail near the lunar orbit. J. Geophys. Res. Space Phys.
**123**, 3836–3847 (2018). https://doi.org/10.1029/2018JA025183 ADSCrossRefGoogle Scholar - V. Olshevsky, A. Divin, E. Eriksson, S. Markidis, G. Lapenta, Energy dissipation in magnetic null points at kinetic scales. Astrophys. J.
**807**(2), 155 (2015). https://doi.org/10.1088/0004-637x/807/2/155 ADSCrossRefGoogle Scholar - G. Paschmann, B.U.Ö. Sonnerup,
*Proper Frame Determination and Walén Test*, ISSI Scientific Rep., vol. 8 (2008), pp. 65–74 Google Scholar - G. Paschmann, S. Haaland, B.U.Ö. Sonnerup, H. Hasegawa, E. Georgescu, B. Klecker, T.D. Phan, H. Rème, A. Vaivads, Characteristics of the near-tail dawn magnetopause and boundary layer. Ann. Geophys.
**23**(4), 1481–1497 (2005). https://doi.org/10.5194/angeo-23-1481-200 ADSCrossRefGoogle Scholar - T.D. Phan, J.P. Eastwood, P.A. Cassak et al., MMS observations of electron-scale filamentary currents in the reconnection exhaust and near the X line. Geophys. Res. Lett.
**43**, 6060–6069 (2016). https://doi.org/10.1002/2016GL069212 ADSCrossRefGoogle Scholar - F. Plaschke, T. Karlsson, H. Hietala, M. Archer, Z. Vörös, R. Nakamura, W. Magnes, W. Baumjohann, R.B. Torbert, C.T. Russell, B.L. Giles, Magnetosheath high-speed jets: internal structure and interaction with ambient plasma. J. Geophys. Res. Space Phys.
**122**, 10,157–10,175 (2017). https://doi.org/10.1002/2017JA024471 CrossRefGoogle Scholar - Z.-Y. Pu, M.G. Kivelson, Kelvin:Helmholtz instability at the magnetopause: solution for compressible plasmas. J. Geophys. Res.
**88**(A2), 841 (1983). https://doi.org/10.1029/ja088ia02p00841 ADSCrossRefGoogle Scholar - Z.Y. Pu et al., Multiple magnetic reconnection events observed by Cluster II, in
*Proceedings on Magnetic Reconnection Meeting*, ed. by R. Lundin, R. McGregor. IRF Sci. Rep., vol. 280 Kiruna, Sweden, September 2002 (2003), pp. 60–64, Swed. Inst. of Space Phys. Google Scholar - Z.Y. Pu, Q.-G. Zong, T.A. Fritz, C.J. Xiao, Z.Y. Huang, S.Y. Fu, Q.Q. Shi, M.W. Dunlop, K.-H. Glassmeier, A. Balogh, P. Daly, H. Reme, J. Dandouras, J.B. Cao, Z.X. Liu, C. Shen, J.K. Shi, Multiple flux rope events at the high-latitude magnetopause: cluster/rapid observation on 26 January, 2001. Surv. Geophys.
**26**(1–3), 193–214 (2005). https://doi.org/10.1007/s10712-005-1878-0 ADSCrossRefGoogle Scholar - L. Rezeau, G. Belmont, R. Manuzzo, N. Aunai, J. Dargent, Analyzing the magnetopause internal structure: new possibilities offered by MMS tested in a case study. J. Geophys. Res. Space Phys.
**123**(1), 227–241 (2018). https://doi.org/10.1002/2017ja024526 ADSCrossRefGoogle Scholar - P. Robert, M.W. Dunlop, A. Roux, G. Chanteur, Accuracy of current density determination, in
*Analysis Methods for Multi-Spacecraft Data*, vol. 398, ed. by G. Paschmann, P.W. Daly (International Space Science Institute, Bern, 1998), pp. 395–418 Google Scholar - Z.J. Rong, W.X. Wan, C. Shen, T.L. Zhang, A.T.Y. Lui, Y. Wang, M.W. Dunlop, Y.C. Zhang, Q.-G. Zong, Method for inferring the axis orientation of cylindrical magnetic flux rope based on single-point measurement. J. Geophys. Res. Space Phys.
**118**, 271–283 (2013). https://doi.org/10.1029/2012JA018079 ADSCrossRefGoogle Scholar - C.T. Russell, A study of flux transfer events at different planets. Adv. Space Res.
**16**(4), 159–163 (1995) ADSCrossRefGoogle Scholar - C.T. Russell, M.M. Hoppe, W.A. Livesey, J.T. Gosling, S.J. Bame, ISEE-1 and -2 observations of laminar bow shocks: velocity and thickness. Geophys. Res. Lett.
**9**, 1171–1174 (1982). https://doi.org/10.1029/GL009i010p01171 ADSCrossRefGoogle Scholar - C.T. Russell, M.M. Mellott, E.J. Smith, J.H. King, Multiple spacecraft observations of interplanetary shocks: four spacecraft determination of shock normals. J. Geophys. Res.
**88**(A6), 4739 (1983) ADSCrossRefGoogle Scholar - C.T. Russell, R.J. Strangeway, C. Zhao, B.J. Anderson, W. Baumjohann, K.R. Bromund, D. Fischer, L. Kepko, G. Le, W. Magnes, R. Nakamura, F. Plaschke, J.A. Slavin, R.B. Torbert, T.E. Moore, W.R. Paterson, C.J. Pollock, J.L. Burch, Structure, force balance, and topology of Earth’s magnetopause. Science
**356**(6341), 960–963 (2017). https://doi.org/10.1126/science.aag3112 ADSCrossRefGoogle Scholar - S.J. Schwartz, Shock and discontinuity normals, mach number, and related parameters, in
*Analysis Methods for Multi-Spacecraft Data*, ed. by G. Paschmann, P.W. Daly (ESA Publications Division, Noordwijk, 1998) Google Scholar - C. Shen, X. Li, M. Dunlop, Q.Q. Shi, Z.X. Liu, E. Lucek, Z.Q. Chen, Magnetic field rotation analysis and the applications. J. Geophys. Res.
**112**, A06211 (2007). https://doi.org/10.1029/2005JA011584 ADSCrossRefGoogle Scholar - Q.Q. Shi, C. Shen, Z.Y. Pu, M.W. Dunlop, Q.-G. Zong, H. Zhang, C.J. Xiao, Z.X. Liu, A. Balogh, Dimensional analysis of observed structures using multi-point magnetic field measurements: application to cluster. Geophys. Res. Lett.
**32**, L12105 (2005). https://doi.org/10.1029/2005GL022454 ADSCrossRefGoogle Scholar - Q.Q. Shi, C. Shen, M.W. Dunlop, Z.Y. Pu, Q.-G. Zong, Z.X. Liu, E. Lucek, A. Balogh, Motion of observed structures calculated from multi-point magnetic field measurements: application to cluster. Geophys. Res. Lett.
**33**, L08109 (2006). https://doi.org/10.1029/2005GL025073 ADSCrossRefGoogle Scholar - Q.Q. Shi, Z.Y. Pu, J. Soucek, Q.-G. Zong, S.Y. Fu, L. Xie, Y. Chen, H. Zhang, L. Li, L.D. Xia, Z.X. Liu, E. Lucek, A.N. Fazakerley, H. Reme, Spatial structures of magnetic depression in the Earth’s high-altitude cusp: cluster multipoint observations. J. Geophys. Res.
**114**, A10202 (2009a). https://doi.org/10.1029/2009JA014283 ADSCrossRefGoogle Scholar - Q.Q. Shi, Q.-G. Zong, H. Zhang, Z.Y. Pu, S.Y. Fu, L. Xie, Y.F. Wang, Y. Chen, L. Li, L.D. Xia, Z.X. Liu, A.N. Fazakerley, H. Reme, E. Lucek, Cluster observations of the entry layer equatorward of the cusp under northward interplanetary magnetic field. J. Geophys. Res.
**114**, A12219 (2009b). https://doi.org/10.1029/2009JA014475 ADSCrossRefGoogle Scholar - Q.Q. Shi, Q.-G. Zong, S.Y. Fu, M.W. Dunlop, Z.Y. Pu, G.K. Parks, Y. Wei, W.H. Li, H. Zhang, M. Nowada, Y.B. Wang, W.J. Sun, T. Xiao, H. Reme, C. Carr, A.N. Fazakerley, E. Lucek, Solar wind entry into the high-latitude terrestrial magnetosphere during geomagnetically quiet times. Nat. Commun.
**4**(1), 1466 (2013). https://doi.org/10.1038/ncomms2476 CrossRefGoogle Scholar - Q.Q. Shi, M.D. Hartinger, V. Angelopoulos, A.M. Tian, S.Y. Fu, Q.-G. Zong, J.M. Weygand, J. Raeder, Z.Y. Pu, X.Z. Zhou, M.W. Dunlop, W.L. Liu, H. Zhang, Z.H. Yao, X.C. Shen, Solar wind pressure pulse-driven magnetospheric vortices and their global consequences. J. Geophys. Res. Space Phys.
**119**(6), 4274–4280 (2014). https://doi.org/10.1002/2013ja019551 ADSCrossRefGoogle Scholar - P. Song, C.T. Russell, Time series data analyses in space physics. Space Sci. Rev.
**87**, 387–463 (1999) ADSCrossRefGoogle Scholar - B.U.Ö. Sonnerup, L.J. Cahill, Magnetopause structure and attitude from explorer 12 observations. J. Geophys. Res.
**72**(1), 171 (1967) ADSCrossRefGoogle Scholar - B.U.Ö. Sonnerup, M. Guo, Magnetopause transects. Geophys. Res. Lett.
**23**, 3679 (1996). https://doi.org/10.1029/96gl03573 ADSCrossRefGoogle Scholar - B.U.Ö. Sonnerup, H. Hasegawa, Orientation and motion of two-dimensional structures in a space plasma. J. Geophys. Res.
**110**, A06208 (2005). https://doi.org/10.1029/2004JA010853 ADSCrossRefGoogle Scholar - B.U.Ö. Sonnerup, M. Scheible, Minimum and maximum variance analysis, in
*Analysis Methods for MultiSpacecraft Data*, ed. by G. Paschmann, P.W. Daly (Int. Space Sci. Inst./Eur. Space Agency, Bern/Paris, 1998), pp. 185–220, Chap. 8 Google Scholar - B.U.Ö. Sonnerup, W.-L. Teh, Reconstruction of two-dimensional coherent MHD structures in a space plasma: the theory. J. Geophys. Res.
**113**, A05202 (2008). https://doi.org/10.1029/2007JA012718 ADSCrossRefGoogle Scholar - B.U.Ö. Sonnerup, S. Haaland, G. Paschmann, B. Lavraud, M.W. Dunlop, H. Rème, A. Balogh, Orientation and motion of a discontinuity from single-spacecraft measurements of plasma velocity and density: minimum mass flux residue. J. Geophys. Res.
**109**, A03221 (2004). https://doi.org/10.1029/2003JA010230 ADSCrossRefGoogle Scholar - B.U.Ö. Sonnerup, H. Hasegawa, W.-L. Teh, L.-N. Hau, Grad–Shafranov reconstruction: an overview. J. Geophys. Res.
**111**(A9) (2006). https://doi.org/10.1029/2006JA011717 - B.U.Ö. Sonnerup, S. Haaland, G. Paschmann, M.W. Dunlop, H. Rème, A. Balogh, Orientation and motion of a plasma discontinuity from single-spacecraft measurements: generic residue analysis of cluster data. J. Geophys. Res.
**111**, A05203 (2007). https://doi.org/10.1029/2005JA011538 ADSCrossRefGoogle Scholar - B.U.Ö. Sonnerup, R.E. Denton, H. Hasegawa, M. Swisdak, Axis and velocity determination for quasi two-dimensional plasma/field structures from Faraday’s law: a second look. J. Geophys. Res. Space Phys.
**118**, 2073–2086 (2013). https://doi.org/10.1002/jgra.50211 ADSCrossRefGoogle Scholar - B.U.Ö. Sonnerup, H. Hasegawa, R.E. Denton, T.K.M. Nakamura, Reconstruction of the electron diffusion region. J. Geophys. Res. Space Phys.
**121**, 4279–4290 (2016). https://doi.org/10.1002/2016JA022430 ADSCrossRefGoogle Scholar - W. Sun, Q. Shi, S. Fu, Q. Zong, Z. Pu, L. Xie, T. Xiao, L. Li, Z. Liu, H. Reme, E. Lucek, Statistical research on the motion properties of the magnetotail current sheet: cluster observations. Sci. China, Technol. Sci.
**53**, 1732–1738 (2010). https://doi.org/10.1007/s11431-010-3153-y CrossRefGoogle Scholar - W.-L. Teh, B.U.Ö. Sonnerup, First results from ideal 2-D MHD reconstruction: magnetopause reconnection event seen by cluster. Ann. Geophys.
**26**(9), 2673–2684 (2008). https://doi.org/10.5194/angeo-26-2673-2008 ADSCrossRefGoogle Scholar - W.-L. Teh, B.U.Ö. Sonnerup, L.-N. Hau, Grad–Shafranov reconstruction with field-aligned flow: first results. Geophys. Res. Lett.
**34**, L05109 (2007). https://doi.org/10.1029/2006GL028802 ADSCrossRefGoogle Scholar - W.-L. Teh, B.U.Ö. Sonnerup, J. Birn, R.E. Denton, Resistive MHD reconstruction of two-dimensional coherent structures in space. Ann. Geophys.
**28**(11), 2113–2125 (2010). https://doi.org/10.5194/angeo-28-2113-2010 ADSCrossRefGoogle Scholar - T. Terasawa, Hall current effect on tearing mode instability. Geophys. Res. Lett.
**10**(6), 475–478 (1983) ADSCrossRefGoogle Scholar - T. Terasawa, H. Kawano, I. Shinohara, T. Mukai, Y. Saito, M. Hoshino, A. Nishida, S. Machida, T. Nagai, T. Yamamoto, S. Kokubun, On the determination of a moving MHD structure: minimization of the residue of integrated Faraday’s equation. J. Geomagn. Geoelectr.
**48**(5–6), 603–614 (1996) ADSCrossRefGoogle Scholar - A.M. Tian, Q.-G. Zong, Y.F. Wang, Q.Q. Shi, S.Y. Fu, Z.Y. Pu, A series of plasma flow vortices in the tail plasma sheet associated with solar wind pressure enhancement. J. Geophys. Res.
**115**, A09204 (2010). https://doi.org/10.1029/2009JA014989 ADSCrossRefGoogle Scholar - A.M. Tian, Q.-G. Zong, Q.Q. Shi, Reconstruction of morningside plasma sheet compressional ULF Pc5 wave. Sci. China, Technol. Sci.
**55**, 1092–1100 (2012). https://doi.org/10.1007/s11431-011-4735-z CrossRefGoogle Scholar - A.M. Tian, Q.Q. Shi, Q.-G. Zong, J. Du, S.Y. Fu, Y.N. Dai, Analysis of magnetotail flux rope events by ARTEMIS observations. Sci. China, Technol. Sci.
**57**, 1010–1019 (2014). https://doi.org/10.1007/s11431-014-5489-1007/s11431-014-1 CrossRefGoogle Scholar - A. Tian, Q. Shi, A.W. Degeling, S. Zhang, Study of magnetic flux ropes by multi-spacecraft analysis method and GS method. Sci. China Tech. Sci. (2019 submitted) Google Scholar
- R.B. Torbert, J.L. Burch, T.D. Phan, et al., Electron-scale dynamics of the diffusion region during symmetric magnetic reconnection in space. Science
**362**, 1391–1395 (2018). https://doi.org/10.1126/science.aat2998 ADSMathSciNetCrossRefGoogle Scholar - J. Vogt, S. Haaland, G. Paschmann, Accuracy of multi-point boundary crossing time analysis. Ann. Geophys.
**29**, 2239–2252 (2011). https://doi.org/10.5194/angeo-29-2239-2011 ADSCrossRefGoogle Scholar - D.E. Wendel, M.L. Adrian, Current structure and nonideal behavior at magnetic null points in the turbulent magnetosheath. J. Geophys. Res. Space Phys.
**118**, 1571–1588 (2013). https://doi.org/10.1002/jgra.50234 ADSCrossRefGoogle Scholar - C.J. Xiao, Z.Y. Pu, Z.W. Ma, S.Y. Fu, Z.Y. Huang, Q.G. Zong, Inferring of flux rope orientation with the minimum variance analysis technique. J. Geophys. Res.
**109**, A11218 (2004). https://doi.org/10.1029/2004JA010594 ADSCrossRefGoogle Scholar - T. Xiao, H. Zhang, Q.Q. Shi, Q.-G. Zong, S.Y. Fu, A.M. Tian, W.J. Sun, S. Wang, G.K. Parks, S.T. Yao, H. Rème, I. Dandouras, Propagation characteristics of young hot flow anomalies near the bow shock: cluster observations. J. Geophys. Res. Space Phys.
**120**, 4142–4154 (2015). https://doi.org/10.1002/2015JA021013 ADSCrossRefGoogle Scholar - Y.Y. Yang, C. Shen, Y.C. Zhang, Z.J. Rong, X. Li, M. Dunlop, Y.H. Ma, Z.X. Liu, C.M. Carr, H. Rème, The force-free configuration of flux ropes in geomagnetotail: cluster observations. J. Geophys. Res. Space Phys.
**119**(8), 6327–6341 (2014). https://doi.org/10.1002/2013ja019642 ADSCrossRefGoogle Scholar - S.T. Yao, Q.Q. Shi, Z.Y. Li, X.G. Wang, A.M. Tian, W.J. Sun, M. Hamrin, M.M. Wang, T. Pitkänen, S.C. Bai, X.C. Shen, X.F. Ji, D. Pokhotelov, Z.H. Yao, T. Xiao, Z.Y. Pu, S.Y. Fu, Q.G. Zong, A. De Spiegeleer, W. Liu, H. Zhang, H. Rème, Propagation of small size magnetic holes in the magnetospheric plasma sheet. J. Geophys. Res. Space Phys.
**121**, 5510–5519 (2016). https://doi.org/10.1002/2016JA022741 ADSCrossRefGoogle Scholar - S.T. Yao, X.G. Wang, Q.Q. Shi, T. Pitkänen, M. Hamrin, Z.H. Yao, Z.Y. Li, X.F. Ji, A. De Spiegeleer, Y.C. Xiao, A.M. Tian, Z.Y. Pu, Q.G. Zong, C.J. Xiao, S.Y. Fu, H. Zhang, C.T. Russell, B.L. Giles, R.L. Guo, W.J. Sun, W.Y. Li, X.Z. Zhou, S.Y. Huang, J. Vaverka, M. Nowada, S.C. Bai, M.M. Wang, J. Liu, Observations of kinetic-size magnetic holes in the magnetosheath. J. Geophys. Res. Space Phys.
**122**, 1990–2000 (2017). https://doi.org/10.1002/2016JA023858 ADSCrossRefGoogle Scholar - S.T. Yao, Q.Q. Shi, R.L. Guo, Z.H. Yao, A.M. Tian, A.W. Degeling, W.J. Sun, J. Liu, X.G. Wang, Q.G. Zong, H. Zhang, Z.Y. Pu, L.H. Wang, S.Y. Fu, C.J. Xiao, C.T. Russell, B.L. Giles, Y.Y. Feng, T. Xiao, S.C. Bai, X.C. Shen, L.L. Zhao, H. Liu, Magnetospheric multiscale observations of electron scale magnetic peak Geophys. Res. Lett.
**45**(2), 527–537 (2018). https://doi.org/10.1002/2017gl075711 ADSCrossRefGoogle Scholar - Yao et al., Kinetic-scale flux rope in the magnetosheath boundary layer. J. Geophys. Res. (2019 submitted) Google Scholar
- Y.C. Zhang, C. Shen, Z.X. Liu, Z.J. Rong, T.L. Zhang, A. Marchaudon, H. Zhang, S.P. Duan, Y.H. Ma, M.W. Dunlop, Y.Y. Yang, C.M. Carr, I. Dandouras, Two different types of plasmoids in the plasma sheet: cluster multisatellite analysis application. J. Geophys. Res. Space Phys.
**118**(9), 5437–5444 (2013). https://doi.org/10.1002/jgra.50542 ADSCrossRefGoogle Scholar - C. Zhao, C.T. Russell, R.J. Strangeway, S.M. Petrinec, W.R. Paterson, M. Zhou, B.J. Anderson, W. Baumjohann, K.R. Bromund, M. Chutter, D. Fischer, G. Le, R. Nakamura, F. Plaschke, J.A. Slavin, R.B. Torbert, H.Y. Wei, Force balance at the magnetopause determined with MMS: application to flux transfer events. Geophys. Res. Lett.
**43**(23), 11,941–11,947 (2016). https://doi.org/10.1002/2016gl071568 CrossRefGoogle Scholar - X.-Z. Zhou, Q.-G. Zong, Z.Y. Pu, T.A. Fritz, M.W. Dunlop, Q.Q. Shi, J. Wang, Y. Wei, Multiple triangulation analysis: another approach to determine the orientation of magnetic flux ropes. Ann. Geophys.
**24**(6), 1759–1765 (2006a). https://doi.org/10.5194/angeo-24-1759-2006 ADSCrossRefGoogle Scholar - X.-Z. Zhou, Q.-G. Zong, J. Wang, Z.Y. Pu, X.G. Zhang, Q.Q. Shi, J.B. Cao, Multiple triangulation analysis: application to determine the velocity of 2-D structures. Ann. Geophys.
**24**(11), 3173–3177 (2006b). https://doi.org/10.5194/angeo-24-3173-2006 ADSCrossRefGoogle Scholar - X.-Z. Zhou, Z.Y. Pu, Q.-G. Zong, P. Song, S.Y. Fu, J. Wang, H. Zhang, On the error estimation of multi-spacecraft timing method. Ann. Geophys.
**27**, 3949–3955 (2009). https://doi.org/10.5194/angeo-27-3949-2009 ADSCrossRefGoogle Scholar

## Copyright information

**Open Access** This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.