To implement boundary conditions which distinguish between particles
flowing into and those flowing out of a system, we must reexpress
the Liouville
equation (2.3) in terms of the classical phase space ,
where **q** in this case corresponds to the position **x** and **p** is the
momentum. This is naturally done by the Wigner-Weyl transformation,
which transforms the density operator into the Wigner
distribution function
(Wigner, 1932; Heller, 1976; Berry, 1977; Carruthers and Zachariasen,
1983).
For the present purposes, the
Wigner-Weyl transformation consists of a change of independent
coordinates to the diagonal and cross-diagonal coordinates:

followed by a Fourier transformation with respect to **r**.
The variables **x** and
may be expressed in terms of **q** and **r** by

Thus the Wigner distribution can be expressed as

The Liouville equation becomes

where the kernel of the potential operator is given by

These expressions are derived under the assumption that the domain is unbounded.

Let us consider the interpretation of the terms of the Liouville
equation (4.33). The first term on the right-hand side is
derived from the kinetic energy operator and is of the form known as a
drift, streaming, or advection term (in various nomenclatures).
This term is exactly the same as the corresponding term of the classical
Liouville equation with force **F**:

The correspondence between the classical and quantum drift terms will be exploited in defining the open-system boundary conditions.

Quantum-interference effects enter the Wigner-Weyl representation via the
nonlocal potential term of (4.33). The kernel of this operator
in effect redistributes the Wigner function among
different **p**'s at each position **q**. The extent to which it does so
depends upon the potential at positions remote from **q** (4.34).
This is the way that interference between alternative paths is
incorporated into the equation. Thus, a rough intuitive image of the action
of is that it represents particles which have scattered
off the potential at some point and upon returning,
interfere with the particles propagating over other paths. This image
will be invoked to interpret the effects of cutting off the integral
in (4.34) at some finite value, which is required in practical
computations.

Let us now consider a model in which the domain is bounded by **q = 0**
and **q = l**.
To address the question of boundary conditions, first note that in
the Wigner-Weyl representation, the Liouville equation (4.33) is
of first order with respect to **q** and does not contain derivatives
with respect to **p**. The characteristics of the derivative term are lines of
constant **p**, and we must supply one and only one boundary value at some
point on each characteristic, because the equation is of first order in
**q**. The kinds of
boundary conditions which are appropriate are illustrated in Fig.
7. To implement the picture described above, that the
particles entering the device depend only upon the state of the reservoirs
and that the particles leaving the device depend only upon the state of
the device, we should apply the boundary conditions illustrated in
Fig. 7(c). That is, we set

where is the distribution function of the reservoir to the left of the system and is the distribution function of the reservoir to the right. These boundary conditions are not invariant under time-reversal, because time-reversal would change the problem of Fig. 7(c) into that of Fig. 7(d).

Figure 7. Possible boundary conditions for the Liouville equation (4.33) in phase space. The points at which the boundary values are specified (indicated by a heavy line) can be atq = 0as in (a), atq = las in (b), or divided between the two boundaries, depending upon the sign ofp, as shown in (c) and (d). The boundary conditions (c) are, in fact, the appropriate ones for an open system.

Conceptually, the boundary conditions (4.36) are identical
to those employed in the conventional tunneling theory (see Appendices
9 and 12), in the Landauer approach
(Landauer, 1957, 1970; Büttiker * et al.*, 1985; Stone and Szafer,
1988), and in solutions of the Boltzmann equation for nonuniform
systems (see Appendix 11 and Duderstadt and Martin,
1979). However, some care must be taken in this identification. It is
true that the variable **p** goes over into the classical momentum
appearing in the Boltzmann equation, by the correspondence principle.
However, it is * not* true that **p** is the same quantity as the
operator , or its eigenvalue.
In particular, as
will be discussed in Sec. 6.1, the
traveling-wave boundary conditions actually depend upon the energy of
the state, rather than **p**.
Thus, the boundary conditions (4.36) are conceptually identical
to but mathematically different from those employed in the tunneling and
Landauer approaches.

Let us call the Liouville superoperator which results from the boundary conditions (4.36) (for open-system, irreversible). For the purposes of the present discussion, it will be separated into two terms:

where is the superoperator derived from the kinetic energy term of the Hamiltonian:

and is the superoperator derived from the potential term:

Let us note in passing that can be written in two other forms. One is Groenewold's expression (Groenewold, 1946):

The other is the Wigner-Moyal expansion (Moyal, 1949):

where in the last expression it is understood that only acts upon . The utility of both of these expressions depends upon the existence of a rapidly-converging series expansion for . Such an expansion is not available for the abrupt energy-barrier structures which originally motivated the present study, so the integral form of (4.39) is preferred for practical computations.

Thu Jun 8 17:53:37 CDT 1995