An integration by parts shows that and so a physical interpretation of Belinfante tensor is that it includes the "bound momentum" associated with gradients of the intrinsic angular momentum. In other words, the added term is an analogue of the "bound current" associated with a magnetization density. The curious combination of spin-current components required to make symmetric and yet still conserved seems totally ad hoc, but it was shown by both Rosenfeld and Belinfante that the modified tensor is precisely the symmetric Hilbert energy–momentum tensor that acts as the source of gravity in general relativity. Just as it is the sum of the bound and free currents that acts as a source of the magnetic field, it is the sum of the bound and free energy–momentum that acts as a source of gravity.
The Hilbert energy-momentum tensor is defined by the variation of the action functional with respect to the metric as or equivalently as We may also define an energy-momentum tensor by varying a Minkowski-orthonormal vierbein to get Here is the Minkowski metric for the orthonormal vierbein frame, and are the covectors dual to the vierbeins. With the vierbein variation there is no immediately obvious reason for to be symmetric. However, the action functional should be invariant under an infinitesimal local Lorentz transformation ,, and so should be zero. As is an arbitrary position-dependent skew symmetric matrix, we see that local Lorentz and rotation invariance both requires and implies that . Once we know that is symmetric, it is easy to show that, and so the vierbein-variation energy-momentum tensor is equivalent to the metric-variation Hilbert tensor. We can now understand the origin of the Belinfante-Rosefeld modification of the Noether canonical energy momentum tensor. Take the action to be where is the spin connection that is determined by via the condition of being metric compatible and torsion free. The spin current is then defined by the variation and the "canonical" Noether energy momentum tensor is the part that arises from the variation where we keep the spin connection fixed: Then Now, for a torsion-free and metric-compatible connection, we have that where we are using the notation Using the spin-connection variation, and after an integration by parts, we find Thus we see that corrections to the canonical Noether tensor that appear in the Belinfante-Rosenfeld tensor occur because we need to simultaneously vary the vierbein and the spin connection if we are to preserve local Lorentz invariance. As an example, consider the classical Lagrangian for the Dirac field Here the spinor covariant derivatives are We therefore get There is no contribution from if we use the equations of motion, i.e. we are on shell. Now if are distinct and zero otherwise. As a consequence is totally antisymmetric. Now, using this result, and again the equations of motion, we find that Thus the Belinfante-Rosenfeld tensor becomes The Belinfante-Rosenfeld tensor for the Dirac field is therefore seen to be the symmetrized canonical energy-momentum tensor.
Weinberg's definition
Weinberg defines the Belinfante tensor as where is the Lagrangian density, the set are the fields appearing in the Lagrangian, the non-Belinfante energy momentum tensor is defined by and are a set of matrices satisfying the algebra of the homogeneous Lorentz group
