---
title: Rank-2 Magnetic Polarizability Tensor
url: https://www.emergentmind.com/topics/rank-2-magnetic-polarizability-tensor
type: topic
---

# Rank-2 Magnetic Polarizability Tensor

Searching arXiv for the cited papers on magnetic polarizability tensors and related formulations.
{"query":"all:magnetic polarizability tensor rank-2 Ledger Lionheart Ammari deuteron twisted electron", "max_results": 10}
Searching arXiv for relevant papers.
The rank-2 magnetic polarizability tensor is, in the standard eddy-current theory of small conducting permeable objects, a complex symmetric second-order tensor that appears as the leading-order coefficient in the asymptotic perturbation of the magnetic field. In Cartesian representation it is a \(3\times 3\) tensor, typically denoted \(\mathcal M\), with at most six independent complex coefficients, and it encodes shape, size, conductivity, permeability contrast, frequency dependence, and orientation through its tensor transformation law rather than through explicit position dependence [1705.00580]. Within this literature, “rank-2” refers to a spatial tensor on \(\mathbb R^3\), not to a relativistic spacetime tensor, and the object serves as a compact descriptor for target characterization, inverse problems, and metal detection [2110.06624].

## 1. Definition and conceptual scope

In the metal-detection and eddy-current asymptotics literature, the magnetic polarizability tensor is the tensor \(\mathcal M\) in the leading perturbation formula
\[
(\boldsymbol H_\alpha-\boldsymbol H_0)(\boldsymbol x)_i
=
(\boldsymbol D_x^2 G(\boldsymbol x,\boldsymbol z))_{ij}
(\mathcal M)_{jk}
(\boldsymbol H_0(\boldsymbol z))_k
+O(\alpha^4),
\]
for a small object \(B_\alpha=\alpha B+\boldsymbol z\) [2001.07629]. This identifies \(\mathcal M\) as the object-dependent factor in a dipole-type response: it maps the local background magnetic field at the object location into the leading perturbation measured away from the object [1906.00382].

The tensor is explicitly treated as complex symmetric and rank 2, equivalently a \(3\times 3\) matrix in an orthonormal basis, with six independent complex coefficients in the generic case [2110.06624]. Its entries scale like \(\alpha^3\), consistent with dipolar volume scaling, and its dependence on frequency is governed by the parameter
\[
\nu=\omega\mu_0\sigma_*\alpha^2,
\]
or the corresponding piecewise definition for inhomogeneous objects [1510.01356]. In this formulation, the tensor depends on shape, conductivity, permeability, size, and excitation frequency, but it is independent of object position; translation enters separately through the Green function and background field evaluation [1809.08916].

A central point in the later literature is that the rank-2 MPT is the lowest-order member of a larger hierarchy of generalised magnetic polarizability tensors. In the complete asymptotic expansion, the classical rank-2 tensor is recovered by taking the lowest-order indices \(m=0\) and \(p=0\), so the familiar MPT is the leading-order truncation of a higher-order object-characterization framework [2207.03791].

## 2. Eddy-current asymptotics and the leading-order field perturbation

The governing regime is the time-harmonic eddy-current approximation of Maxwell’s equations for a small conducting permeable inclusion. The fields satisfy
\[
\nabla \times \boldsymbol H_\alpha = \sigma_\alpha \boldsymbol E_\alpha + \boldsymbol J_0,\qquad
\nabla \times \boldsymbol E_\alpha = i\omega \mu_\alpha \boldsymbol H_\alpha,
\]
with piecewise-constant conductivity and permeability in the object and its exterior [1705.00580]. The asymptotic regime assumes \(\alpha\to 0\) with \(\nu=\omega\mu_0\sigma_*\alpha^2=O(1)\), so the quasi-static eddy-current scaling is retained while the object size shrinks [1510.01356].

The leading-order perturbation formula has the standard dipole structure
\[
(\boldsymbol H_\alpha-\boldsymbol H_0)(\boldsymbol x)_i
=
(\boldsymbol D_x^2 G(\boldsymbol x,\boldsymbol z))_{ij}
(\mathcal M)_{jk}
(\boldsymbol H_0(\boldsymbol z))_k
+(\boldsymbol R(\boldsymbol x))_i,
\]
with
\[
G(\boldsymbol x,\boldsymbol z)=\frac{1}{4\pi |\boldsymbol x-\boldsymbol z|},
\qquad
|\boldsymbol R|\le C\alpha^4\|\boldsymbol H_0\|_{W^{2,\infty}(B_\alpha)}.
\]
This makes the tensor the coefficient of the leading-order magnetic dipole response [1906.00382]. In the notation of the GMPT literature, the same term is obtained from the complete expansion by retaining only the lowest-order contribution in derivatives of both the Green tensor and the background field [2207.03791].

Historically, the engineering literature often used a rank-2 tensor directly for object characterization, whereas earlier rigorous asymptotics had produced a rank-4 object. The reduction to a complex symmetric rank-2 tensor in orthonormal coordinates is one of the major conceptual clarifications in this line of work [1510.01356]. This reduction is not merely notational; it establishes that the object signature relevant to leading-order hidden-target characterization can be represented by a second-order tensor with a standard transformation law under rotations [1705.00580].

## 3. Coefficient formulas, transmission problems, and spectral structure

The rank-2 tensor is commonly written as
\[
\mathcal M=-\mathcal C+\mathcal N,
\]
with \(\mathcal C\) the conductivity-driven part and \(\mathcal N\) the permeability-contrast part [1809.08916]. For a homogeneous object, the coefficients are
\[
( \mathcal C )_{ij}
:=
-\frac{i\alpha^3}{4}\,\boldsymbol e_i\cdot
\int_B
\nu\, \boldsymbol \xi \times
\big(\boldsymbol\theta_j + \boldsymbol e_j\times \boldsymbol \xi\big)\,
d\boldsymbol \xi,
\]
\[
( \mathcal N )_{ij}
:=
\alpha^3 \int_B
\left(1-\frac{\mu_0}{\mu_*}\right)
\left(
\boldsymbol e_i\cdot \boldsymbol e_j
+\frac12 \boldsymbol e_i\cdot \nabla\times \boldsymbol\theta_j
\right)
d\boldsymbol \xi,
\]
where the auxiliary fields \(\boldsymbol\theta_j\) solve a vector transmission problem on the reference domain and its exterior [1906.00382].

Those auxiliary fields satisfy, for \(j=1,2,3\),
\[
\nabla_\xi \times \mu_*^{-1}\nabla_\xi \times \boldsymbol\theta_j
-
i\omega \sigma_*\alpha^2 \boldsymbol\theta_j
=
i\omega \sigma_*\alpha^2 \boldsymbol e_j\times \boldsymbol\xi
\quad \text{in } B,
\]
together with continuity of tangential fields, the appropriate jump condition for \(\mu^{-1}\nabla\times \boldsymbol\theta_j\), a divergence constraint, and decay at infinity [1705.00580]. The tensor is therefore not postulated phenomenologically; it is computed from a well-posed transmission problem tied directly to the eddy-current model.

A further structural decomposition isolates a magnetostatic part and two conductive frequency-dependent pieces:
\[
\mathcal M=\mathcal N^0+\mathcal R^{\sigma_*}+i\,\mathcal I^{\sigma_*},
\]
where \(\mathcal N^0\), \(\mathcal R^{\sigma_*}\), and \(\mathcal I^{\sigma_*}\) are each real symmetric rank-2 tensors [1906.00382]. This decomposition is especially important because \(\mathcal N^0\) is the low-frequency magnetostatic contribution, while \(\mathcal R^{\sigma_*}\) and \(\mathcal I^{\sigma_*}\) encode the dispersive conductive response. In the homogeneous case, \(\mathcal M(0)=\mathcal N^0\), and for homogeneous \(\mu_*\) one has \(\mathcal M(0)=\mathcal T(\mu_r)\), identifying the low-frequency MPT with the Pólya–Szegő tensor [1510.01356].

The spectral analysis of the tensor coefficients is one of the most developed parts of the subject. For homogeneous conductivity, the fields admit an eigenfunction expansion in terms of an auxiliary spectral problem, and the conductive parts satisfy modal formulas of the form
\[
(\mathcal R^{\sigma_*})_{ij}
=
\frac{\alpha^3}{4}\sum_{n=1}^\infty
\Re(\beta_n)\lambda_n
\left\langle \boldsymbol\phi_n,\boldsymbol\Theta^{(0)}(\boldsymbol e_i)\right\rangle
\left\langle \boldsymbol\phi_n,\boldsymbol\Theta^{(0)}(\boldsymbol e_j)\right\rangle,
\]
\[
(\mathcal I^{\sigma_*})_{ij}
=
\frac{\alpha^3}{4}\sum_{n=1}^\infty
\Im(\beta_n)\lambda_n
\left\langle \boldsymbol\phi_n,\boldsymbol\Theta^{(0)}(\boldsymbol e_i)\right\rangle
\left\langle \boldsymbol\phi_n,\boldsymbol\Theta^{(0)}(\boldsymbol e_j)\right\rangle,
\]
with
\[
\beta_n=-\frac{i\nu}{i\nu-\lambda_n}.
\]
These formulas explain why the real part tends to show bounded monotone behavior and the imaginary part tends to exhibit a single peak in the homogeneous case [1906.00382].

Because \(\mathcal N^0\), \(\mathcal R^{\sigma_*}\), and \(\mathcal I^{\sigma_*}\) are real symmetric, each is orthogonally diagonalizable and has real eigenvalues. This makes eigenvalues and principal invariants natural orientation-independent descriptors, whereas the raw tensor entries depend on the lab-frame orientation [2110.06624].

## 4. Generalisations, multiple objects, and computational use

The rank-2 MPT formalism extends beyond a single homogeneous target. For \(N\) sufficiently well-separated homogeneous objects \((B_\alpha)^{(n)}=\alpha^{(n)}B^{(n)}+\mathbf z^{(n)}\), the leading perturbation becomes a sum of \(N\) dipole-type terms,
\[
(\mathbf H_{\boldsymbol\alpha} - \mathbf H_0)(\mathbf x)_i
=
\sum_{n=1}^N
(\mathbf D_x^2 G(\mathbf x,\mathbf z^{(n)}))_{ij}
(\mathcal M[\alpha^{(n)} B^{(n)}])_{jk}
(\mathbf H_0(\mathbf z^{(n)}))_k
+
(\mathbf R(\mathbf x))_i,
\]
so each object contributes its own position-independent complex symmetric rank-2 tensor [1809.08916]. When objects are closely spaced or materially inhomogeneous but scale and translate as a single composite, the cluster is instead characterized by one effective rank-2 tensor computed from a coupled transmission problem over the composite geometry [1809.08916].

The complete asymptotic expansion later developed for generalised MPTs shows precisely when the classical rank-2 description ceases to be sufficient. Higher-order tensors enter when the background field varies appreciably over the object or when greater discrimination power is needed beyond the uniform-field dipole approximation [1705.00580]. This suggests that the rank-2 tensor is the correct first descriptor, but not necessarily the complete one in near-field or strongly nonuniform excitation settings.

Computationally, the tensor is especially attractive because it compresses the forward electromagnetic response into a small number of coefficients. Proper-orthogonal-decomposition reduced-order models were developed to accelerate the evaluation of the spectral signature \(\omega\mapsto \mathcal M(\omega)\), with full-order finite element solutions used offline and reduced projections used online [2001.07629]. That work also established scaling identities,
\[
(\mathcal M[\alpha B,\omega,\mu_r,s\sigma_*])_{ij}
=
(\mathcal M[\alpha B,s\omega,\mu_r,\sigma_*])_{ij},
\]
and
\[
(\mathcal M[s\alpha B,\omega,\mu_r,\sigma_*])_{ij}
=
s^3(\mathcal M[\alpha B,s^2\omega,\mu_r,\sigma_*])_{ij},
\]
which make size and conductivity sweeps inexpensive once one reference signature is known [2001.07629].

For classification, the spectral MPT is treated as a function of frequency,
\[
\omega \mapsto \mathcal M[\alpha B,\omega,\sigma_*,\mu_r],
\]
and machine-learning pipelines typically use rotation-invariant features derived from \(\tilde{\mathcal R}\) and \(\mathcal I\), rather than raw coefficients [2110.06624]. The principal invariants
\[
I_1(\mathcal A)=\operatorname{tr}(\mathcal A),\qquad
I_2(\mathcal A)=\frac12\left(\operatorname{tr}(\mathcal A)^2-\operatorname{tr}(\mathcal A^2)\right),\qquad
I_3(\mathcal A)=\det(\mathcal A)
\]
are preferred because they are orientation invariant, avoid eigenvalue-ordering ambiguity, and are numerically smoother than root-finding-based features [2110.06624].

## 5. Distinct meanings of “rank-2 magnetic polarizability tensor”

The term is not used uniformly across all branches of electromagnetic and particle theory. Three distinct meanings appear in the cited literature.

| Context | Rank-2 object | Role |
|---|---|---|
| Eddy-current metal detection | \(\mathcal M\) | Complex symmetric \(3\times 3\) descriptor of a small conducting permeable object |
| Relativistic polarization theory | \(M^{\mu\nu}\) | Polarization-magnetization tensor, not a standalone magnetic polarizability tensor |
| Spin- or OAM-dependent particle response | \(\beta_T\) via quadratic operators | Tensor magnetic polarizability in spin or orbital degrees of freedom |

In relativistic field theory for structural microparticles, the central second-rank object is the polarization tensor
\[
M^{\mu\nu},
\]
defined through
\[
j_\mu^{(M)}=\partial^\rho M_{\rho\mu},
\]
and reconstructed as
\[
M^{\mu\nu}=d^\mu u^\nu-d^\nu u^\mu+\varepsilon^{\mu\nu\rho\sigma}u_\rho m_\sigma.
\]
In that paper, magnetic polarizability is not a rank-2 tensor; it is a scalar coefficient \(\beta_M\) entering the constitutive relation
\[
m^\mu=4\pi\beta_M\,b^\mu,
\]
and the interaction Lagrangian
\[
\mathcal L_I=-2\pi\left[\alpha_E e^2+\beta_M b^2\right].
\]
The paper is explicit that there is no anisotropic tensor \(\beta_{ij}\) there [2108.09510].

A different meaning appears in the deuteron literature, where tensor magnetic polarizability is a genuine rank-2 spin-dependent response of a spin-1 system. There the interaction is
\[
V=-\frac{\alpha_T}{\gamma}(\mathbf S\cdot\mathbf E')^2
-\frac{\beta_T}{\gamma}(\mathbf S\cdot\mathbf B')^2,
\]
and \(\beta_T\) couples to the quadratic spin operator \(S_z^2\) in the frozen-spin storage-ring geometry [1011.2352]. This is tensorial in spin space rather than a \(3\times 3\) spatial object characterizing a hidden conductor.

A third variant arises for twisted electrons, where an orbital, rather than spin, tensor magnetic polarizability appears through
\[
W=-\beta_T(\mathbf L\cdot\mathbf B)^2,\qquad
\beta_T=\frac{e^2\hbar^2}{8m^3}=5.25\times 10^4\,\mathrm{fm}^3.
\]
Here the rank-2 structure is carried by bilinears in intrinsic orbital angular momentum, not by the eddy-current object descriptor \(\mathcal M\) [1902.06882].

These distinctions are essential because the same phrase can refer either to a spatial object-characterization tensor, a relativistic polarization tensor, or a quadratic spin/OAM interaction coefficient.

## 6. Applications, invariants, and limitations

In metal detection and hidden-object identification, the rank-2 MPT is used as the sole descriptor layer between the electromagnetic forward problem and the classifier. The tensor’s spectral signature can be obtained either from simulated eddy-current solutions or, in practice, from induced-voltage measurements over a frequency band followed by inversion to the tensor coefficients [2110.06624]. Because the tensor is position independent and transforms covariantly under rotation, it is well suited to dictionary-based identification once rotationally invariant features are extracted.

The practical value of the tensor follows from several structural properties. First, it is compact: six complex coefficients in the generic case, fewer under symmetry [2110.06624]. Second, it is interpretable: low-frequency limits connect to the Pólya–Szegő tensor, while high-conductivity limits approach \(\mathcal T(0)\) for simply connected objects and exteriors [1510.01356]. Third, it is computationally accessible: explicit coefficient formulas, finite element transmission problems, reduced-order surrogates, and scaling laws all exist in the literature [2001.07629].

The limitations are equally clear. The rank-2 model assumes a small-object regime and, at leading order, effectively a background field that is approximately uniform over the object [1705.00580]. When the background field varies significantly across the target, the classical tensor may be insufficient and higher-order GMPTs become relevant [2207.03791]. Measurement noise, nonuniform background fields, capacitive coupling, soil and background effects, parasitic voltages, and filtering also perturb recovered coefficients; reported measurement errors for MPT coefficients are typically about \(1\%\) to \(5\%\) in the classification study [2110.06624].

A common misconception is that the rank-2 MPT is a universal notion independent of modeling regime. The literature does not support that view. In the dominant eddy-current usage, it is a complex symmetric \(3\times 3\) tensor associated with small conducting permeable objects. In relativistic polarization theory, the analogous second-rank object is not itself a magnetic polarizability tensor. In nuclear and vortex-beam physics, “tensor magnetic polarizability” usually denotes a quadratic spin or OAM response coefficient rather than a spatial object-characterization tensor [2108.09510].

The mature interpretation, therefore, is domain-specific. In low-frequency induction sensing, the rank-2 magnetic polarizability tensor is the leading-order, orientation-covariant, frequency-dependent object signature of a small conducting permeable target. It forms the classical entry point to a broader hierarchy of generalised tensors, and it remains the foundational descriptor for asymptotic modeling, reduced-order computation, and spectral classification of hidden metallic objects [2207.03791].

Source: https://www.emergentmind.com/topics/rank-2-magnetic-polarizability-tensor