Let us summarize at this point what has been achieved so far. We have four coupled differential equations for our dynamic field variables . The first of these is the continuity equation (56), which is given, in terms of the present variables, by
The three other differential equations, the evolution equations (73), (74) and the generalized Hamilton-Jacobi equation (79), do not yet possess a definite mathematical form. They contain four unknown functions which are constrained, but not determined, by (75), (78).
The simplest choice, from a formal point of view, is . In this limit the present theory agrees with Schiller’s field-theoretic (Hamilton-Jacobi) version, see , of the equations of motion of a classical dipole. This is a classical (statistical) theory despite the fact that it contains [see (63)] a number . But this classical theory is not realized in nature; at least not in the microscopic domain. The reason is that the simplest choice from a formal point of view is not the simplest choice from a physical point of view. The postulate of maximal simplicity (Ockham’s razor) implies equal probabilities and the principle of maximal entropy in classical statistical physics. A similar principle which is able to ’explain’ the nonexistence of classical physics (in the microscopic domain) is the principle of minimal Fisher information . The relation between the two (classical and quantum-mechanical) principles has been discussed in detail in I.
The mathematical formulation of the principle of minimal Fisher information for the present problem requires a generalization, as compared to I, because we have now several fields with coupled time-evolution equations. As a consequence, the spatial integral (spatial average) over in the variational problem (44) should be replaced by a space-time integral, and the variation should be performed with respect to all four variables. The problem can be written in the formsolutions of (81), (82) for are inserted in the variational problem (81), the four relations (82) become redundant and becomes the Lagrangian density of our problem. Thus, Eqs. (81) and (82) represent a method to construct a Lagrangian.
We assume a functional form , where . This means does not possess an explicit -dependence and does not depend on (this would lead to a modification of the continuity equation). We further assume that does not depend on time-derivatives of (the basic structure of the time-evolution equations should not be affected) and on spatial derivatives higher than second order. These second order derivatives must be taken into account but should not give contributions to the variational equations (a more detailed discussion of the last point has been given in I).
The variation with respect to reproduces the continuity equation which is unimportant for the determination of . Performing the variation with respect to and taking the corresponding conditions (79), (74) (73) into account leads to the following differential equations for and ,
The solutions for may be obtained with the help of the second condition () listed in Eq. (75). The result may be written in the form
Eqs. (88) and (89) show that the first condition listed in (75) is also satisfied. The last condition is also fulfilled: can be written as , where
and fulfills (78). We see that is a quantum-mechanical contribution to the rotational motion while is related to the probability density of the ensemble (as could have been guessed considering the mathematical form of these terms). The last term is the same as in the spinless case [see (51)].
The remaining task is to show that the above solution for does indeed lead to a (appropriately generalized) Fisher functional. This can be done in several ways. The simplest is to use the following result due to :
Summarizing, our assumption, that under certain external conditions four state variables instead of two may be required, led to a nontrivial result, namely the four coupled differential equations (80), (79), (74), (73) with given by (86), (89), (88). The external condition which stimulates this splitting is given by a gauge field; the most important case is a magnetic field but other possibilities do exist (see below). These four differential equations are equivalent to the much simpler differential equation
which is linear in the complex-valued two-component state variable and is referred to as Pauli equation (the components of the vector are the three Pauli matrices and ). To see the equivalence one writes, see , ,
and evaluates the real and imaginary parts of the two scalar equations (93). This leads to the four differential equations (80), (79), (74) (73) and completes the present spin theory.
In terms of the real-valued functions the quantum-mechanical solutions (86), (88), (89) for look complicated in comparison to the classical solutions . In terms of the variable the situation changes to the contrary: The quantum-mechanical equation becomes simple (linear) and the classical equation, which has been derived by , becomes complicated (nonlinear). The simplicity of the underlying physical principle (principle of maximal disorder) leads to a simple mathematical representation of the final basic equation (if a complex-valued state function is introduced). One may also say that the linearity of the equations is a consequence of this principle of maximal disorder. This is the deeper reason why it has been possible, see , to derive Schrödinger’s equation from a set of assumptions including linearity.
Besides the Pauli equation we found, as a second important result of our spin calculation, that the following local force is compatible with the statistical constraint:
Here, the velocity field and the magnetic moment field have been replaced by corresponding particle quantities and ; the dot denotes the inner product between and . The first force in (95), the Lorentz force, has been derived here from first principles without any additional assumptions. The same cannot be said about the second force which takes this particular form as a consequence of some additional assumptions concerning the form of the ’internal force’ [see (71)]. In particular, the field appearing in was arbitrary as well as the proportionality constant (g-factor of the electron) and had to be adjusted by hand. It is well-known that in a relativistic treatment the spin term appears automatically if the potentials are introduced. Interestingly, this unity is not restricted to the relativistic regime. Following  and  we report in the next section an alternative (non-relativistic) derivation of spin, which does not contain any arbitrary fields or constants - but is unable to yield the expression (95) for the macroscopic electromagnetic forces.
In the present treatment spin has been introduced as a property of an ensemble and not of individual particles. Similar views may be found in the literature, see . Of course, it is difficult to imagine the properties of an ensemble as being completely independent from the properties of the particles it is made from. The question whether or not a property ’spin’ can be ascribed to single particles is a subtle one. Formally, we could assign a probability of being in a state () to a particle just as we assign a probability for being at a position . But contrary to position, no classical meaning - and no classical measuring device - can be associated with the discrete degree of freedom . Experimentally, the measurement of the ’spin of a single electron’ is - in contrast to the measurement of its position - a notoriously difficult task. Such experiments, and a number of other interesting questions related to spin, have been discussed by .