Modeling spin magnetization transport
in a spatially varying magnetic field
Abstract
We present a framework for modeling the transport of any number of globally conserved quantities in any spatial configuration and apply it to obtain a model of magnetization transport for spinsystems that is valid in new regimes (including highpolarization). The framework allows an entropy function to define a model that explicitly respects the laws of thermodynamics. Three facets of the model are explored. First, it is expressed as nonlinear partial differential equations that are valid for the new regime of high dipoleenergy and polarization. Second, the nonlinear model is explored in the limit of low dipoleenergy (semilinear), from which is derived a physical parameter characterizing separative magnetization transport (SMT). It is shown that the necessary and sufficient condition for SMT to occur is that the parameter is spatially inhomogeneous. Third, the high spintemperature (linear) limit is shown to be equivalent to the model of nuclear spin transport of Genack and Redfield Genack1975 (). Differences among the three forms of the model are illustrated by numerical solution with parameters corresponding to a magnetic resonance force microscopy (MRFM) experiment Degen2009 (); Kuehn2008 (); Sidles2003 (); Dougherty2000 (). A family of analytic, steadystate solutions to the nonlinear equation is derived and shown to be the spintemperature analog of the Langevin paramagnetic equation and Curie’s law. Finally, we analyze the separative quality of magnetization transport, and a steadystate solution for the magnetization is shown to be compatible with Fenske’s separative mass transport equation Fenske1932 ().
keywords:
magnetization dynamics, spin diffusion, spin magnetization transport, separative transport, hyperpolarization, high magnetic fieldgradient1 Introduction
Magnetization transport for a spinsystem in a spatially varying magnetic field has been studied theoretically Genack1975 () and experimentally Eberhardt2007 (); Budakian2004 (). Hightemperature models of spin magnetization transport, such as that of Genack and Redfield, do not apply for systems with highpolarization. With recent significant enhancements of the technique of dynamic nuclear polarization (DNP) Ni2013 (); Abragam1978 (); Krummenacker2012 (), which has been shown to achieve significant hyperpolarization, models that can describe the highpolarization regime in a spatially varying magnetic field are needed.
We present a framework for modeling the transport of any number of globally conserved quantities in any spatial configuration. We then apply it to obtain a model of magnetization transport for spinsystems that is valid in new regimes (including highpolarization). Finally, we analyze the separative quality of the magnetization transport. A particularly useful feature of the framework is that specifying an entropy density function completely determines the system model. Such a function is presented for a spinsystem, and its validity is demonstrated by deriving classical models from it, which is justification for using it in new regimes (as we do in Section 3).
In Section 2, we introduce the framework. It is general in the sense that it can be applied to systems with any number of globally conserved distributed quantities that evolve over any (smooth) spatial geometry in any number of spatial dimensions. The laws of thermodynamics are included a priori such that any specific model based on the framework will be guaranteed to respect all four laws.
In Section 3, we apply the framework by specifying conserved quantities, an entropy function, and a spatial geometry for a spinsystem and thereby obtain a new model of magnetization transport in a magnetic field gradient. It accommodates previously unmodeled regimes of high energy and high polarization, such as may develop with DNP. The remainder of the section explores the model in various limits and connects them to previous models.
In Section 4, we analyze the separative quality of magnetization transport, highlighting the parallelism between it and the separative mass transport work that began with Fenske Fenske1932 (). Magnetization transport in a magnetic field gradient is both diffusive and separative, and the latter is of particular interest for technologies that may be enhanced by hyperpolarization, such as magnetic resonance imaging (MRI), nuclear magnetic resonance (NMR) spectroscopy, and magnetic resonance force microscopy (MRFM). DNP has been used to achieve significant hyperpolarization through transferring polarization from electronspins to nuclearspins. But DNP is still a slow process, taking tens to thousands of seconds to develop, and it is impeded by magnetic field gradients, which for certain applications (such as MRFM) is undesirable. We consider the feasibility of a different technique in which no polarization is transferred among spinspecies, but in which magnetization is concentrated by the phenomenon of separative magnetization transport (SMT). We develop the necessary and sufficient conditions for the SMT of a single spinspecies — most notably, that the magnetic field must be spatially varying. Taken as a whole, this paper lays the groundwork for an investigation into how the SMTeffect might be enhanced to produce hyperpolarization.
2 Framework for transport analysis
We will proceed in the coordinatefree language of differential geometry, which allows the laws of thermodynamics to be respected explicitly, regardless of spatial geometry or the number of conserved quantities.
What follows is a necessarily extensive list of definitions and remarks. As we will see, the mathematical rigor of these definitions will enable and greatly simplify the subsequent theoretical development.
The following definitions are the elements from which two propositions are constructed that describe a framework for transport analysis and its adherence to the laws of thermodynamics. In any specific application of the framework, defining the spatial geometry, conserved quantities, entropy function, and a spacetime scale will be sufficient to construct a model of transport that respects the laws of thermodynamics from the following definitions (as detailed in Remark 2).
The elements of the framework are defined in the following order:

the spatial manifold, metric, and coordinates;

conserved quantities and their local densities;

the entropy density and thermodynamic potentials;

Onsager’s kinetic coefficients;

the current of local quantity densities;

the continuity equation for local densities; and

a transport rate tensor and an ansatz further specifying the kinetic coefficients.
1 will describe how the laws of thermodynamics are satisfied in the definitions and 2 will assert that the definitions describe a physically valid model of transport. We begin with spatial considerations.
Definition 1 (spatial manifold).
Let be a Riemannian (smooth) manifold, which represents the spatial geometry of a macroscopic thermodynamic system. We call the spatial manifold.
For many applications, a Euclidean space^{1}^{1}1See Bullo2005 (), p. 22 and Lee2012 (), p. 598. is an appropriate choice for .
Remark 1 (spatial coordinates).
Let be some local coordinate map.^{2}^{2}2See Lee2012 (), pp. 1516, 6065. Typically, we will denote component functions of , defined by for some point , as . These are called local spatial coordinates, and typically denoted . (See Figure 1.)
By definition, the Riemannian spatial manifold is endowed with a Riemannian metric, which determines the geometry of .
Definition 2 (spatial metric).
Let be a Riemannian metric^{3}^{3}3See Lee1997 (), p. 23. on . We call the spatial metric.
In local spatial coordinates, the metric is written as
(1) 
Definition 3 (conserved quantities).
Let be the tuple , where represents a conserved quantity.
Definition 4 (standard thermodynamic dual basis).
Let the ordered basis for be
We call this the standard thermodynamic dual basis,^{4}^{4}4Although it is an uncommon practice to introduce a dual basis before a basis, we do so here because the quantities represented by the quantities and are more naturally — from a physical standpoint — considered dual to the potentials . Yet, the quantities naturally arise first in the series of definitions. and it is often denoted .
In the standard basis, with the Einstein summation convention,
(2) 
Definition 5 (local quantity density).
Let be the set of smooth maps from (where represents time) to (i.e. for each point in space and time we assign a vector in ). Given a vector of conserved quantities , let represent the local spatial density of each of the conserved quantities , such that
(3) 
where is a volume element of .
In the standard thermodynamic dual basis , we write
(4) 
where each is a function . The mathematical structure of assigned in the definition is equivalent to a section of the product bundle . Figure 2 illustrates this description, where copies of correspond to each location in .
We now turn to entropic considerations.
Definition 6 (local entropy density).
Let the local entropy volumetric density function be a function that is nonnegative and concave.
The restriction of the local entropy density to nonnegative functions satisfies the third law of thermodynamics. Moreover, we require that be concave to allow the Legendre dual relationship that will now be introduced.
At times it is convenient to work with another set of variables called local thermodynamic potentials. These are significant because their spatial gradients drive the flow of .
Definition 7 (local thermodynamic potentials).
Let be defined by the relation
(5) 
where the exterior derivative is taken with respect to the vector space .^{5}^{5}5The vector space has dual space . Therefore maps to vectors in and maps to covectors in . We call the local thermodynamic potential, and it is the Legendre transform^{6}^{6}6For an excellent article on the Legendre transform and this duality, see Zia2009 (). tuple conjugate of . The dual space of is denoted , and so . The duality gives the standard thermodynamic basis to be such that , where is the Kroneckerdelta.
The standard thermodynamic basis representation of the potential is
(6) 
The convention has been adopted that thermodynamic vectors are represented by uppercase symbols with upper indices on vector components (e.g. ) and thermodynamic covectors are represented by lowercase symbols with lower indices on covector components (e.g. ).
Commonly, inverse temperatures are the thermodynamic potentials of internal energy quantities. In this manner, other thermodynamic potentials can be considered to be analogs of inverse temperature. For instance, magnetic moment quantities have spintemperature thermodynamic potentials. Keeping this in mind can aid the intuition that spatial gradients in drive the flow of , as in the familiar case of heat transfer being driven by gradients in (inverse) temperature.
We now begin to construct the current of conserved quantity densities . First, a discussion of spatially and thermodynamically indexed tensor structures is needed, and this requires calculus on manifolds.^{7}^{7}7See both Lee2012 () and Spivak1965 ().
In order to do calculus on manifolds, the notion of a tangent space is required:^{8}^{8}8See Lee2012 (), p. 54. a tangent space at a point on the spatial manifold is a vector space on which tangent vectors of curves through live (see Figure 4). A chart that includes provides a convenient basis for via its coordinate vectors^{9}^{9}9See Lee2012 (), p. 60. at point , , where is the local coordinate representation of . The tangent bundle is the disjoint union of the tangent spaces at all points on the manifold.^{10}^{10}10See Lee2012 (), pp. 658.
The dual space of the tangent space at , or cotangent space , can be given a convenient basis by dualmapping the tangent space basis to the cotangent space. The cotangent bundle is the disjoint union of the cotangent spaces at all points on the manifold.^{11}^{11}11See Lee2012 (), pp. 272303.
Spatially indexed structures will be expressed in terms of the coordinate vectors and coordinate covectors . Tensors that are indexed by both spatial coordinates and thermodynamic bases we call thermometric structures.
The following convention for describing thermometric structures will be used. As with any tensor or tensor field, the order of their indices is merely conventional, but must be accounted. No convention is established for the ordering of the indices, but we will describe a tensor as being indexed covariantly (acting on covectors) or contravariantly (acting on vectors), first with a thermodynamic pair, say , and second with a spatial pair, say . For instance, a tensor at some point might be , which would have as its standard basis some permutation of the tensor product .
The convention that has been adopted is that spatial coordinate vectors and covectors are indexed by the lowercase Greek alphabet and thermodynamic basis vectors and covectors are indexed by the lowercase Latin alphabet.
Definition 8 (Onsager kinetic coefficients).
Let be defined as a positivedefinite type thermometric contravariant tensor field,^{12}^{12}12Note that a tensor field is a tensor bundle section. which is called Onsager kinetic coefficients tensor field, which serves as a thermodynamic and spatial metric.
To satisfy the second law of thermodynamics, must be positive semidefinite, so Definition 8 satisfies the Second Law.
Typically, a contravariant tensor field is considered to assign a tensor map at each point from the tangent space to a real number. However, considering as a section of a fiberbundle, it is a structure analogous to the tangent bundle , and so is indexed with both the usual cotangent local coordinate vectors and the thermodynamic dual basis . We often use the symmetric basis for , .
Definition 9 (current).
Let be defined as
(7) 
where is the musical isomorphism,^{13}^{13}13The musical isomorphisms are defined by the metric as maps between the tangent bundle and the cotangent bundle (See Lee2012 (), pp. 3413 and Lee1997 (), pp. 279.). The operator maps spatial 1forms to vectors. is the exterior derivative,^{14}^{14}14The exterior derivative is the coordinatefree generalization of the familiar differential of a function. See Lee2012 (), pp. 36272. and the symbol denotes a functional composition. We call the spatial transport current.
Equation 7 asserts that the current is driven by the (generalized) gradient in thermodynamic potential , which is a multipotential statement of the zeroth law of thermodynamics.
When , as is often the case, (7) can be written in terms of the coordinatefree grad operator^{15}^{15}15See Lee2012 (), p. 368. as
(8) 
With the preceding definitions, we can introduce the governing equation for , which is the fundamental equation of the framework of transport.
Definition 10 (continuity).
Let the continuity equation be defined as
(9) 
where is the Hodge codifferential operator.^{16}^{16}16The Hodge codifferential maps forms to forms (Lee2012, , pp. 4389). In (9) it maps a spatial 1form to a 0form. This is similar to the divergence operator, except that it acts on a 1form instead of a vector.
Equation 9 is an expression of the global conservation of local quantities . It states that the local quantities change with the (generalized) divergence of a current. For an energy quantity, this is the first law of thermodynamics.
Proposition 1 (laws of thermodynamics).
Each of the following statements is a necessary and sufficient condition for the adherence of the framework for transport analysis to the corresponding law of thermodynamics. In aggregate, then, they are necessary and sufficient conditions for adherence to all four laws of thermodynamics.
Notice that, while the laws of thermodynamics have narrowed, considerably, the possible forms of the transport equation, two elements remain indefinite, although their general structures have been prescribed: the local entropy density function and the tensor of Onsager’s kinetic coefficients .
The specific form of the entropy function depends on the system, and so it is as yet necessarily unspecified. The final step, then, is to specify the form of the Onsager kinetic coefficients tensor field . Two forms are presented, the first (13) is quite general, and applies to systems that have different transport rates. The second (14) is an ansatz that can be used in certain applications that have a single transport rate, and is applied in Section 3 to model the magnetization transport of a system of one spinspecies.
But, first, two more definitions are required.
Definition 11 (covariance tensor field).
Let be defined as the negativedefinite tensor field^{17}^{17}17See Equation 211 of Onsager1953 ().
(10) 
The tensor field represents quantum mechanical observation processes which are related to the entropy density by the expression.^{18}^{18}18This is related to the Ruppeiner metric (Ruppeiner1979, ; Ruppeiner1995, ). We call the covariance tensor field.
This is a connection between the quantum mechanical and the macroscopic descriptions of transport. It can also be expressed in terms of a freeenergy and local thermodynamic potentials, which are related to the entropy and local quantity densities by the Legendre transform.
Definition 12 (entropy Hessian).
Let be defined as the type thermometric contravariant tensor field
(11) 
We will call the entropy Hessian.
In local coordinates and the standard thermodynamic basis,
(12) 
A symmetric standard basis for is . is positivedefinite because is positivedefinite by its definition as a Riemannian metric and is by definition negativedefinite.
Definition 13 (transport rate tensor field).
Let be a thermometric mixed tensor field called the transport rate tensor, which is defined by the relation
(13) 
The transport rate tensor field sets the spacetime scales for transport. In general, there are transport spacetimes scales, but we often assume many fewer by symmetry and spatial isotropy.
Proposition 2 (framework for transport analysis).
Given a system of thermodynamic quantities, its covariance tensor field,^{19}^{19}19The covariance tensor field can be found by observation, by quantum simulation, or by an entropy density function satisfying Equation 10. The Legendre duality makes it equivalent to have a known, valid free energydensity function. and its transport rate tensor — and if the system has no significant advective transport^{20}^{20}20We have not here considered the case of advective transport. For magnetization transport, this means that only solidstate magnetization samples are considered. — the continuity equation of 10 and its dependent definitions describe the transport of the system over times for which global quantities can be considered substantially conserved.^{21}^{21}21That is, times for which 3 holds for .
For certain systems (e.g. a system of spins of a single species), the following ansatz simplifies the analysis.
Definition 14 (Ozansatz).
Let be a real number called the OnsagerZiegler transport coefficient that specifies a single spacetime scale. Let the OnsagerZiegler ansatz (OZansatz) be the following relation that specifies the Onsager kinetic coefficient tensor field (8):
(14) 
Equation 14 assumes a single spacetime scale for all transport. This ansatz is roughly accurate in many physical systems, but we do not assert that it is generally valid. In nuclear magnetization transport, this ansatz is usually reasonable.^{22}^{22}22See Genack1975 (), p. 83.
Remark 2.
Therefore, to implement the transport model for any specific system, the only elements needed are:

the spatial manifold , metric , and local coordinates ;

the globally conserved thermodynamic quantities that define local quantity densities ;

the local entropy density function ; and

the transport rate tensor .
3 Model of magnetization transport
In this section, we construct a specific model of magnetization transport from the framework of the last section in one spatial dimension, with a single spinspecies, and (therefore) with two conserved quantities. Table 1 summarizes the steps for developing this magnetization transport model. We begin with some definitions.
Definition 15 (spatial manifold).
Let the manifold be defined by . We call the spatial manifold.^{23}^{23}23We consider geometries with transverse isotropy of magnetization, external magnetic field, and sample composition; therefore, only one spatial dimension is of consequence. For this reason, although volumes are considered threedimensional, in all other cases we consider only the single spatial dimension.
An atlas for is given by the chart , where is the identity, which yields the single Cartesian spatial coordinate in the direction normal to the isotropic plane.
Definition 16 (spatial metric).
Let be the (Riemannian) Euclidean metric, .
For the transport of a single spinspecies, there are two globally conserved functions on , so .
Definition 17 (conserved quantities).
Let be a vector with components: representing the total magnetic energy (Zeeman and dipole) and representing the total magnetic moment.^{24}^{24}24Note that any linear transformation of these quantities is also conserved.
Remark 3 (quantity densities).
These definitions, along with the framework of 2, posit a temporally varying local quantity density , the components of which, in the standard basis, represent the energy volumetric density () and the magnetization ().
3.1 Local entropy density
The local entropy density is defined in a coordinatefree manner that requires some discussion. The definition is derived from the entropy of mixing, which is an entropy function that describes the mixing of several nonreactive quantities.
Definition 18 (local entropy density).
Let be the temporally invariant volumetric spin density (spins per unit volume),^{25}^{25}25Since only a single spatial dimension is considered for , volumes will have physical dimensions , or meters in SI units. be the external spatially varying magnetic field, be the average dipole magnetic field, and be the magnetic moment of an individual spin. Additionally, let the vectorvalued functionals be
(15a)  
and  
(15b)  
and the real number be  
(15c) 
Then let the local entropy density be defined as
(16) 
It can be shown that (18) satisfies the constraints of 6. Although this definition makes reference to the standard basis , and can be written in any basis.
Due to the logarithmic definition of entropy density, the minimum of entropy is inconsequential, except in that it is necessarily nonnegative. With another definition of globally conserved functions, say , the minimum entropy would change, but the resulting transport is invariant.
3.2 Model of magnetization transport
We now present the model of magnetization transport. To remain as general as possible, the OZ transport coefficient will not be assigned. This remaining parameter will require selection based on the spin system. Diffusion coefficients are often suitable choices, and for nuclear and electronspin systems, the literature contains theoretical and empirical values Budakian2004 (); Eberhardt2007 (); Genack1975 (); Dougherty2000 ().
Proposition 3 (model of magnetization transport).
Let a magnetization system be such that the entropy of 18 is valid and let the system have a single spinspecies such that 17 holds. Additionally, let the system be nonadvective and let the OZansatz hold for the system. And let the system meet the other criteria of the framework for transport analysis of 2, using Definitions 15 – 18 where appropriate. Then the continuity equation of 10 and its dependent definitions describe the transport of the system of magnetization.
3.3 Magnetization transport equation
3 defines the magnetization transport equation to be the continuity equation (9) with the specific choices of spatial manifold , local quantity densities , and entropy density function as given in Definitions 15 – 18. This is the nonlinear magnetization transport equation. A family of steadystate solutions for this model will be presented in Section 3.3.4.
Two additional forms of this equation will be presented in important limits. These are solved numerically and compared to the nonlinear model. Along the way, various connections to prior art will be discussed.
3.3.1 Basis considerations
The definitions of the preceding section allow any thermodynamic basis for the local quantity density . In a spatially varying magnetic field, it is convenient and common practice to use a thermodynamic basis that is itself spatially varying. This is computationally advantageous and allows the continuity equation to be written in a simple component form. Two classes of basis are now defined.
Definition 19 (homogeneous basis).
Let a homogeneous basis be a thermodynamic ordered vector or covectorbasis that is not spatially varying.
The standard thermodynamic bases are spatially invariant, and so they are homogeneous.
Definition 20 (inhomogeneous basis).
Let an inhomogeneous basis be a thermodynamic ordered vector or covectorbasis that is spatially varying.
Definition 21 (polarization thermodynamic covector basis).
Let be defined as an ordered thermodynamic covector basis by the basis transform from the standard dual basis:
(17) 
where is the thermodynamic mixed tensor with matrix representation
(18) 
Because is a coordinatefree object, the basis vectors of must transform to those of in a compensatory manner. Similarly, the polarization thermodynamic vector basis for , , is easily derived. All such relations are shown in Figure 5.
From the relations of Figure 5, it can be shown that the basis covectors and vectors of the basis must be spatially varying, and so it is an inhomogeneous basis.
The physical interpretations of the components of are no longer the same as those of . Both components are normalized such that they are nondimensional (their basis vectors have assimilated the units originally associated with the components). The second component has become what is typically called “polarization,” and ranges over the interval . The first component is a nondimensional version of the dipoleenergy density (since the Zeeman energy density has been subtracted from the total energy density).
3.3.2 Thermodynamic potentials and spintemperature
Definitions 7 and 18 give the basis representation of the thermodynamic potential
(19)  
where is the Euclidean norm. Figure 6 shows density plots for the components of . These approach as approaches unity.
Spin temperature is closely associated with , which can be considered to be the inverse spintemperature associated with each quantity . In the standard basis, the components of represent the inverse spintemperatures of total energydensity () and total magnetization (). In the basis, they represent the inverse spintemperatures of the normalized dipole energydensity () and polarization ().
This interpretation is consistent with the relationship illustrated in Figure 6 in which as , . This means that high spintemperatures correspond to low energy densities and polarization and low spintemperatures correspond to high energy and polarization.
Equation 19 can be solved over the domain for
(20)  
where the terms are positive for , positive or negative for , and negative otherwise. The density plots of Figure 6 show the functional relationships of the components.
It is simple to show that the following two identities hold for the relationship between and for all time and space:
(21a)  
(21b) 
3.3.3 Magnetization transport equation: method
Although there is no compact expansion of the magnetization transport equation in coordinates, the key operations required to derive it in perhaps the most compact basis, the basis, are here presented.
Let be spatially homogeneous. The first operation is the gradient of the potential that is an element of the expression for (7). In a single spatial dimension with coordinate and thermodynamic basis, this amounts to
(22)  
where the fieldgradient term arises from the inhomogeneity of the basis. Second and similarly, it can be shown that the codifferential of the current is
(23)  
3.3.4 Steadystate solutions
A family of steadystate solutions is developed. Later in the section, certain limits are explored in which familiar theories are shown to be subsets of this family of solutions.
Steadystate means that is temporally invariant. A simple solution can be developed in the case that . Then from (7), it can be shown that must be zero because is positive definite. In the standard basis this implies that the components , i.e. the system evolves toward uniform distributions of inverse spintemperatures. In the polarization basis it implies that the components of (22) are zero. Let denote constants determined by a boundary condition , and let and . The resulting simple system of ordinary differential equations can be solved for
(24)  
If we let be spatially invariant, is unity and is also spatially invariant. The implications of this are interesting, particularly when an attempt is made to map this solution to a solution. It can be shown from (20) and (24) that the spatialdependence of the components of will be contained in . In fact, a search for a dimensionless parameter to represent the spatial and magnetic fielddependence quickly yields that itself is an excellent choice. This means that a family of spatial steadystate solutions are to be found as constant slices of the surfaces of , as shown in Figure 7, and that can be considered to be the dimensionless spatial variable.
For small , (20) yields the expression
(25) 
This corresponds to the slice of Figure 7. In the standard basis (with functional dependencies suppressed), (25) becomes
(26) 
This is a spintemperature analog of the Langevin paramagnetic equation of statistical mechanics, which expresses the steadystate magnetization distribution as a function of external magnetic field and temperature. If (26) is linearized about small (high spintemperature of total energy), a spintemperature analog of Currie’s law results.
3.3.5 Low dipoleenergy magnetization transport equation
In many applications it will be reasonable to assume small dipoleenergy. A semilinear magnetization transport equation can be derived from the continuity equation of 3 by linearizing about . This model will provide insight into the conditions for separative magnetization transport (SMT). The continuity equation and current can be linearized in basis components, with constant, as
(27a)  
(27b)  
where  
(27c)  
is a dimensionless factor and  
(27d)  
is the dimensionless singlespinspecies SMT parameter, written in the basis. It is more physically intuitive in basis components:  
(27e) 
In Section 4, we will see that this parameter is functionally related to the relative volatility parameter of the mass separation literature. After evaluating the codifferential, (27) becomes
(28)  
3.3.6 High spintemperature
magnetization transport equation
In the literature, high spintemperature is often assumed Genack1975 (); Eberhardt2007 (); Ramanathan2008 (). In many applications, especially when there is little separation, this is sufficient. High spintemperature approximations are low and low approximations, and can be derived from the model of magnetization transport (3) by a firstorder powerseries expansion of the continuity equation about .
Proceeding with this approach in the basis, the following component expression is derived:
(29a)  
(29b)  
Equation 23 can be used to compute a component form of the differential equation (29a). If is assumed to be spatially homogeneous, under a change of basis, the transport equations of Genack and Redfield are recovered.^{26}^{26}26See Genack1975 (), p. 83. The equivalence will be discussed in Section 3.4.
Genack and Redfield state that their expression must be altered to account for a nonuniform density of spins per unit volume . Equation 29 is such an alteration.
3.3.7 Comparison of the models
In certain regimes, the nonlinear equation of 3 differs from both the semilinear (27a) and linear (29) versions. For large , only the nonlinear equation is valid. For large but small , only the nonlinear and partially linear equations are valid. All three are valid for small .
Figure 8 numerically compares the three models. The parameters chosen are typical for a magnetic resonance force microscopy (MRFM) experiment Degen2009 (); Kuehn2008 (); Sidles2003 (); Dougherty2000 (). The operating regime was large and field gradient, along with initial conditions of constant and steplike . This models an MRFM experiment in which a method such as dynamic nuclear polarization (DNP) has hyperpolarized a sample and a region of polarization has been inverted, yielding a sharp transition in the polarization from negative to positive. The nonlinear model predicts that, after some time, the spatial distribution of shown in the upper figure would occur. The semilinear and linear approximations are inaccurate in this operating regime, with the semilinear performing better than the linear approximation, as the plot of deviations from the nonlinear model shows. The linear approximation deviates by nearly 30% at the transition, where the semilinear approximation deviates by approximately 20%.
The deviations are most significant near sharp transitions in polarization. In many magnetic resonance applications this is an important regime that occurs when the polarization of a region of the sample is inverted or saturated.
3.4 Equivalence to other models at high spintemperatures
In Section 3.3.6, we claimed that if is assumed to be spatially homogeneous, the linear transport equation developed there is equivalent to the transport equations of Genack and Redfield.^{27}^{27}27Equations (24a,b) of Genack1975 () are the equivalent expression. We believe that (24b) has a typo in the term containing the current (it is missing a negative sign), but it is otherwise equivalent. Here we show the basis transformation that yields this equivalency. Additionally, another commonly encountered form of the linear model—one expressed in inverse spintemperature variables—is shown to be equivalent.
Two methods have been used to verify the equivalency of the equations. The first is to relate their basis to the basis, derive the nonlinear equation in terms of their components via 3, and linearize it for small quantities. The component transformation from the (inhomogeneous) Genack and Redfield basis to the (inhomogeneous) basis is given by the tensor transformation , which has the matrix representation
(30) 
where is the magnetic constant.^{28}^{28}28See nist (), p. 1. The appearance of the magnetic constant is due to Genack and Redfield’s definition of dipole energydensity as the magnetization times (Genack1975, , p. 83).
The second method is to directly transform the Genack and Redfield equations into Equation 29 (with homogeneous ) via (30). Let and denote the magnetic susceptibility and magnetization, respectively.^{29}^{29}29We use the definitions of Genack and Redfield. Both the variables and the equations must be transformed by the relations
(31a)  
(31b) 
Both methods have been used to verify the equivalency. The latter method is valid only because both bases shared the same zerostate, about which each system was linearized.
Eberhardt et al. and others express this high spintemperature model in terms of inverse spintemperatures Genack1975 (); Eberhardt2007 (); Ramanathan2008 (). It is tedious but straightforward to show that these are equivalent to linearized equations in that have been transformed to equations in via a linearized version of (19), under the following choice of basis: is the dipoleenergy density and is the Zeeman energy density. The component transformation from the standard basis to this (inhomogeneous) basis is given by the tensor transformation , which has the matrix representation
(32) 
The linearized map of the components of this basis to the components of its thermodynamic dual basis is given by the tensor transformation , which has the matrix representation
(33) 
The variable equation is transformed to the variable equation by the relations
(34a)  
(34b) 
The variable and variable expressions of Genack and Redfield’s model are both equivalent to the high spintemperature limit of the model of magnetization transport here presented.
4 Separative transport and the Fenske equation
From (27) and (28) it can be determined that a necessary and sufficient condition for SMT at some time and location is that is nonzero. This condition implies several other necessary conditions. First, it implies that is necessarily spatially inhomogeneous. Second, it implies that is necessarily either less than unity or spatially varying. Third, it implies that is necessarily spatially inhomogeneous. Finally, it implies that is necessarily spatially inhomogeneous. If the polarization is nonunity or spatially varying, as is typically the case, the nonuniformity of is a both necessary and sufficient condition for SMT. If any of these conditions is not met in a region, no SMT occurs there.
Fenske developed a set of equations to describe the process of mass separation in the fractional distillation of hydrocarbons Fenske1932 (). These have become the standard for simple models of mass separation.^{30}^{30}30See Kister1992 (), p. 114. In this section, a model will be derived in the manner of Fenske that is equivalent, in some operating regimes, to the model of magnetization transport presented above. This will highlight the separative aspect of magnetization transport.
Consider the discrete system illustrated in Figure 9. It is analogous to a fractional distillation column in which “tray” contains a certain ratio of one substance to another, and tray contains a greater concentration. Figure 9 shows two “spin” trays in which the two “substances” are spinup and spindown, described by the fractions and . Let the total amount of spin in a tray be fixed:
(35) 
Through an exchange process among the trays, tray obtains a higher ratio between up and downspin. In a fractional distillation column, concentration occurs by boiling liquid in a tray, the vapor of which condenses in the tray above and contains a higher concentration of the product, and fluid flows downward for mass balance. In the spin system, the exchange occurs through dipoledipole interactions. Let be the relative volatility (typically is not much larger than unity). The Fenske model describes this process by the relation^{31}^{31}31See Halvorsen2000 (), p. 35.
(36) 
Let and be the spin fractions of some reference tray (0). Equation 36 is a difference equation that can be solved in the steadystate for the Fenske equation^{32}^{32}32This is not the usual form of the Fenske equation, but it is equivalent.^{,}^{33}^{33}33The superscript of is not an index, but an exponent.
(37) 
A continuum solution is found if tray is taken to be spatially small and mapped to the dimensionless spatial coordinate (i.e. ). Polarization can be identified as
(38) 
Equation 37 can be written in terms of polarization. What is more interesting, however, is that for near unity, the following equation is approximate (and exact in the limit):
(39a)  
where  
and  (39b) 
Equation 39 can be written in the standard basis by the simple relation . Comparing this to (26) with a linear field , the following identification can be made:
(40) 
Recalling the definition of the SMT parameter from (27e), we write
(41) 
This result connects the theory of separative magnetization transport with Fenskestyle separation theory. For separation to occur, the relative volatility must be nonunity and the SMT parameter must be nonzero.^{34}^{34}34As discussed in Section 3.3.5, must be spatially inhomogeneous for SMT to occur, and so it must be nonzero over the region. These are two statements of the same rule.
This highlights the separative nature of transport in the spin system. Small concentrations of magnetization develop in the steadystate, and this can be considered separation of upspin and downspin.
Beyond increasing , no mechanism to enhance this natural separation is apparent in this singlespin species system, as it is in a fractional distillation system, but as we will discuss in the conclusions, we believe that an enhancement of this separative effect is possible with the introduction of a second spinspecies. This will require a model that includes largemagnetization regimes and three conserved quantities (e.g. total magnetic energy, nuclearspin magnetic moment, and electronspin magnetic moment). The model of Section 3 satisfies the former requirement, but it is only valid for two quantities. However, because the framework of Section 2 allows any number of conserved quantities, future work will develop from it a threequantity model of magnetization transport.
5 Conclusions and prospects
We presented a framework for modeling the transport of any number of globally conserved quantities in any spatial configuration. We applied it to obtain a model of magnetization transport for spinsystems that is valid in new regimes (including highpolarization). Finally, we analyzed the separative quality of the magnetization transport.
Separative magnetization transport (SMT) was explored, and found to occur in a manner analogous to separative mass transport. The analogy suggests that exploring ways of enhancing the SMTeffect may yield a useful new technique of hyperpolarizing spins.
Much is known about separative mass transport, but much remains unknown about SMT. A distillation column has temperature and gravitational field gradients. Its components have different volatility, and so one is more represented in the vapor phase than the other. Understanding how these factors affect separative mass transport has been the key to harnessing separation. For SMT, we have shown that it is necessary that the magnetic field be spatially varying. Greater fieldgradients induce greater separative transport. Previous results have shown that diffusive transport is suppressed in a high magnetic field gradient Eberhardt2007 (); Budakian2004 (). Conversely, the present results show that separative transport is enhanced. Certain magnetic resonance technologies are currently betterequipped than others to take advantage of this, such as magnetic resonance force microscopy (MRFM) Degen2009 (); Kuehn2008 (); Sidles1991 (), which already operates in very high magnetic field gradients. However, typical concentrations, even in high magnetic field gradients may not be sufficient in many applications.
We conjecture from the preceding considerations that adding a second spinspecies to the system may be advantageous as follows: With two spinspecies (e.g. nuclear and electron), spatial gradients in the magnetization of one will affect the transport of the other through spinspin interactions. Using magnetic resonance techniques to locally induce such gradients in the magnetization of one species, that of the other may be concentrated. This is unlike dynamic nuclear polarization (DNP) Ni2013 (); Abragam1978 (); Krummenacker2012 () in that no polarization is transferred from one species to another: each species’ polarization is separately conserved.
The magnetization transport model presented in Section 3 does not include a second spinspecies, but the transport framework of Section 2 can be used to develop such a model. Once experimentally validated, the model may be used to harness SMT to hyperpolarize and drive magnetic resonance technology development. This work is underway and forthcoming.
6 Acknowledgements
This work was supported by the Army Research Office (ARO) MURI program under contract # W911NF0510403. We thank Professor John M. Lee for many insightful discussions.
References

(1)
A. Z. Genack, A. G. Redfield,
Theory of nuclear spin
diffusion in a spatially varying magnetic field, Phys. Rev. B 12 (1975)
78–87.
doi:10.1103/PhysRevB.12.78.
URL http://link.aps.org/doi/10.1103/PhysRevB.12.78 
(2)
C. L. Degen, M. Poggio, H. J. Mamin, C. T. Rettner, D. Rugar,
Nanoscale magnetic
resonance imaging, Proceedings of the National Academy of Sciences 106 (5)
(2009) 1313–1317.
arXiv:http://www.pnas.org/content/106/5/1313.full.pdf+html, doi:10.1073/pnas.0812068106.
URL http://www.pnas.org/content/106/5/1313.abstract 
(3)
S. Kuehn, S. A. Hickman, J. A. Marohn,
Advances in mechanical
detection of magnetic resonance, The Journal of Chemical Physics 128 (5)
(2008) 052208.
doi:10.1063/1.2834737.
URL http://link.aip.org/link/?JCP/128/052208/1  (4) J. A. Sidles, J. L. Garbini, W. M. Dougherty, S.h. Chao, The classical and quantum theory of thermal magnetic noise, with applications in spintronics and quantum microscopy, Proceedings of the IEEE 91 (5) (2003) 799–816. doi:10.1109/JPROC.2003.811796.
 (5) W. M. Dougherty, K. J. Bruland, S. H. Chao, J. L. Garbini, S. E. Jensen, J. A. Sidles, The Bloch equations in highgradient magnetic resonance force microscopy: Theory and experiment, Journal of Magnetic Resonance 143 (1) (2000) 106 – 119. doi:DOI:10.1006/jmre.1999.1994.

(6)
M. R. Fenske,
Fractionation of
straightrun pennsylvania gasoline, Industrial & Engineering Chemistry
24 (5) (1932) 482–485.
arXiv:http://pubs.acs.org/doi/pdf/10.1021/ie50269a003, doi:10.1021/ie50269a003.
URL http://pubs.acs.org/doi/abs/10.1021/ie50269a003  (7) K. W. Eberhardt, S. Mouaziz, G. Boero, J. Brugger, B. H. Meier, Direct observation of nuclear spin diffusion in real space, Phys. Rev. Lett. 99 (2007) 227603. doi:10.1103/PhysRevLett.99.227603.

(8)
R. Budakian, H. J. Mamin, D. Rugar,
Suppression of
spin diffusion near a micronsize ferromagnet, Phys. Rev. Lett. 92 (2004)
037205.
doi:10.1103/PhysRevLett.92.037205.
URL http://link.aps.org/doi/10.1103/PhysRevLett.92.037205 
(9)
Q. Z. Ni, E. Daviso, T. V. Can, E. Markhasin, S. K. Jawla, T. M. Swager, R. J.
Temkin, J. Herzfeld, R. G. Griffin,
High frequency dynamic
nuclear polarization, Accounts of Chemical Research 46 (9) (2013)
1933–1941.
arXiv:http://pubs.acs.org/doi/pdf/10.1021/ar300348n, doi:10.1021/ar300348n.
URL http://pubs.acs.org/doi/abs/10.1021/ar300348n 
(10)
A. Abragam, M. Goldman,
Principles of dynamic
nuclear polarisation, Reports on Progress in Physics 41 (3) (1978) 395.
URL http://stacks.iop.org/00344885/41/i=3/a=002 
(11)
J. G. Krummenacker, V. P. Denysenkov, M. Terekhov, L. M. Schreiber, T. F.
Prisner,
Dnp
in mri: An inbore approach at 1.5 t, Journal of Magnetic Resonance 215 (0)
(2012) 94 – 99.
doi:http://dx.doi.org/10.1016/j.jmr.2011.12.015.
URL http://www.sciencedirect.com/science/article/pii/S1090780711005945  (12) F. Bullo, A. D. Lewis, Geometric control of mechanical systems: modeling, analysis, and design for simple mechanical control systems, Springer, 2005.
 (13) J. M. Lee, Introduction to Smooth Manifolds, 2nd Edition, Vol. 218 of Graduate Texts in Mathematics, Springer, 2012.
 (14) J. M. Lee, Riemannian Manifolds: an introduction to curvature, 1st Edition, Vol. 176 of Graduate Texts in Mathematics, Springer, 1997.
 (15) R. K. P. Zia, E. F. Redish, S. R. McKay, Making sense of the Legendre transform, American Journal of Physics 77 (2009) 614–622.

(16)
M. Spivak, Calculus On
Manifolds: A Modern Approach To Classical Theorems Of Advanced Calculus,
Mathematics Monograph Series, Westview Press, 1965.
URL http://books.google.com/books?id=g_EXJtkz7PYC  (17) L. Onsager, S. Machlup, Fluctuations and irreversible processes, Physical Review 91 (1953) 1505 – 1512.

(18)
G. Ruppeiner,
Thermodynamics: A
riemannian geometric model, Phys. Rev. A 20 (1979) 1608–1613.
doi:10.1103/PhysRevA.20.1608.
URL http://link.aps.org/doi/10.1103/PhysRevA.20.1608 
(19)
G. Ruppeiner,
Riemannian geometry
in thermodynamic fluctuation theory, Rev. Mod. Phys. 67 (1995) 605–659.
doi:10.1103/RevModPhys.67.605.
URL http://link.aps.org/doi/10.1103/RevModPhys.67.605  (20) C. Ramanathan, Dynamic nuclear polarization and spindiffusion in nonconducting solids, ArXiv eprintsarXiv:0801.2170.
 (21) NIST, Fundamental physical constants–extensive listing, Tech. rep., National Institute of Standards and Technology (2010).
 (22) H. Z. Kister, Distillation design, McGrawHill chemical engineering series, McGrawHill, 1992.

(23)
I. J. Halvorsen, S. Skogestad,
Distillation
Theory, Unknown, 2000, Ch. 2.
URL http://www.nt.ntnu.no/users/skoge/publications/1999/DistillationTheory/original/distillationtheory.pdf 
(24)
J. A. Sidles,
Noninductive
detection of singleâproton magnetic resonance, Applied Physics Letters
58 (24) (1991) 2854–2856.
doi:http://dx.doi.org/10.1063/1.104757.
URL http://scitation.aip.org/content/aip/journal/apl/58/24/10.1063/1.104757
*
Appendix A Symbol reference
sym.  definition 

Hodge star operator (Lee2012, , pp. 4378)  
relative volatility (Eq. 36)  
separative mag. transport coef. (Eq. 27e)  
external spatially varying magnetic field  
dimensionless (Eq. 24)  
spatial derivative of ()  
average dipole magnetic field  
covariance tensor (Def. 11)  
spins per unit volume (Def. 18)  
spatial derivative of ()  
dimensionless (Sec. 3.3.4)  
exterior derivative (Lee2012, , pp. 36272)  
Hodge codifferential (Lee2012, , pp. 4389)  
standard cotangent bundle basis  
partial derivative with respect to  
standard tangent bundle basis  
separative mag. transport factor (Eq. 27c)  
thermodynamic covector basis (Def. 21)  
thermodynamic vector basis (Def. 21)  
thermodynamic covector basis (Def. 4)  
thermodynamic vector basis (Def. 7)  
an element of (Def. 18)  
Onsager’s kinetic coefficients (Def. 8)  
OZansatz kinetic coefficients (Def. 14)  
transport rate tensor (Def. 11)  
OZ transport coefficient (Def. 14)  
spatial metric (Def. 2)  
entropy Hessian (Def. 12)  
spatial transport current (Def. 9)  
magnetic moment of an individual spin  
dimension of (Def. 1)  
basis transformation (Eq. 32) 
number of conserved quantities (Def. 3)  
basis transformation (Eq. 33)  
map composition  
tensor product (Lee2012, , p. 306)  
local thermodynamic potential (Def. 7)  
the set of smooth maps from to  
the set of smooth maps from to  
a point on  
 to basis transform (Def. 21)  
conserved thermocovector (Def. 3)  
local quantity densities (Def. 5)  
standard spatial coordinate (Rem. 1)  
basis transform (Eq. 30)  
sharp operator (Lee2012, , pp. 3413)  
local entropy density (Def. 6)  
an element of 