Matrix transfer

We suppose that a measurement signal is a mix of r (unknown) independent sources Sj with variance cf. A detector give p> r measurement signals /W/t obtained with several frequencies. The relations between sources and measurement signals are supposed to be linear, but the transfer matrix T is unknown. If we get n >p>r samples m i) of measurement signals, mix of n... 

This is an important general result which relates the free energy per particle to the largest eigenvalue of the transfer matrix, and the problem reduces to detennining this eigenvalue. 

By a deft application of the transfer matrix teclmique, Onsager showed that the free energy is given by... 

Lattice models have been studied in mean field approximation, by transfer matrix methods and Monte Carlo simulations. Much interest has focused on the occurrence of a microemulsion. Its location in the phase diagram between the oil-rich and the water-rich phases, its structure and its wetting properties have been explored [76]. Lattice models reproduce the reduction of the surface tension upon adsorption of the amphiphiles and the progression of phase equilibria upon increasmg the amphiphile concentration. Spatially periodic (lamellar) phases are also describable by lattice models. Flowever, the structure of the lattice can interfere with the properties of the periodic structures. 

Here ak a ) is the annihilation (creation) operator of an exciton with the momentum k and energy Ek, operator an(a ) annihilates (creates) an exciton at the n-th site, 6,(6lt,) is the annihilation (creation) operator of a phonon with the momentum q and energy u) q), x q) is the exciton-phonon coupling function, N is the total number of crystal molecules. The exciton energy is Ek = fo + tfcj where eo is the change of the energy of a crystal molecule with excitation, and tk is the Fourier transform of the energy transfer matrix elements. 

A natural question is just how big does Mq have to be to see this ordered phase for M > Mq. It was shown in Ref 189 that Mq <27, a very large upper bound. A direct computation on the Bethe lattice (see Fig. 2) with q neighbors [190,191] gives Mq = [q/ q — 2)f, which would suggest Mq 4 for the square lattice. By transfer matrix methods and by Pirogov-Sinai theory asymptotically (M 1) exact formulas were derived [190,191] for the transition lines between the gas and the crystal phase (M 3.1962/z)... 

This review is structured as follows. In the next section we present the theory for adsorbates that remain in quasi-equilibrium throughout the desorption process, in which case a few macroscopic variables, namely the partial coverages 0, and their rate equations are needed. We introduce the lattice gas model and discuss results ranging from non-interacting adsorbates to systems with multiple interactions, treated essentially exactly with the transfer matrix method, in Sec. II. Examples of the accuracy possible in the modehng of experimental data using this theory, from our own work, are presented for such diverse systems as multilayers of alkali metals on metals, competitive desorption of tellurium from tungsten, and dissociative... 

With the availabihty of computers, the transfer matrix method [14] emerged as an alternative and powerful technique for the study of cooperative phenomena of adsorbates resulting from interactions [15-17]. Quantities are calculated exactly on a semi-infinite lattice. Coupled with finite-size scaling towards the infinite lattice, the technique has proved popular for the determination of phase diagrams and critical-point properties of adsorbates [18-23] and magnetic spin systems [24—26], and further references therein. Application to other aspects of adsorbates, e.g., the calculation of desorption rates and heats of adsorption, has been more recent [27-30]. Sufficient accuracy can usually be obtained for the latter without scaling and essentially exact results are possible. In the following, we summarize the elementary but important aspects of the method to emphasize the ease of application. Further details can be found in the above references. 

To introduce the transfer matrix method we repeat some well-known facts for a 1-D lattice gas of sites with nearest neighbor interactions [31]. Its grand canonical partition function is given by... 

To avoid numerical differentiation (which is inherently unstable) one uses the fact that an eigenvalue can be expressed as Ai = v Tvf where are the corresponding normalized left and right eigenvectors. Differentiation of the eigenvalue with respect to any parameter is then equivalent to the differentiation of the transfer matrix, and one finds... 

The transfer matrix method extends rather straightforwardly to more than one dimension, systems with multiple interactions, more than one adsorption site per unit cell, and more than one species, by enlarging the basis in which the transfer matrix is defined. 

For the extension to two dimensions we consider a square lattice with nearest-neighbor interactions on a strip with sites in one direction and M sites in the second so that, with cyclic boundary conditions in the second dimension as well, we get a toroidal lattice with of microstates. The occupation numbers at site i in the 1-D case now become a set = ( ,i, /25 5 /m) of occupation numbers of M sites along the second dimension, and the transfer matrix elements are generalized to... 

We have also given the number of particles, p, and the degeneracy, associated with each equivalence class and the base-10 equivalent, nia, of the binary number associated with state a. Because uniquely determines the occupancies in o a mere (cychc) bitshifting generates all equivalent states. This results in a reduced transfer matrix T, of the form... 

Transfer matrix calculations of the adsorbate chemical potential have been done for up to four sites (ontop, bridge, hollow, etc.) or four states per unit cell, and for 2-, 3-, and 4-body interactions up to fifth neighbor on primitive lattices. Here the various states can correspond to quite different physical systems. Thus a 3-state, 1-site system may be a two-component adsorbate, e.g., atoms and their diatomic molecules on the surface, for which the occupations on a site are no particles, an atom, or a molecule. On the other hand, the three states could correspond to a molecular species with two bond orientations, perpendicular and tilted, with respect to the surface. An -state system could also be an ( - 1) layer system with ontop stacking. The construction of the transfer matrices and associated numerical procedures are essentially the same for these systems, and such calculations are done routinely [33]. If there are two or more non-reacting (but interacting) species on the surface then the partial coverages depend on the chemical potentials specified for each species. 

As an example of a multilayer system we reproduce, in Fig. 3, experimental TPD spectra of Cs/Ru(0001) [34,35] and theoretical spectra [36] calculated from Eq. (4) with 6, T) calculated by the transfer matrix method with M = 6 on a hexagonal lattice. In the lattice gas Hamiltonian we have short-ranged repulsions in the first layer to reproduce the (V X a/3) and p 2 x 2) structures in addition to a long-ranged mean field repulsion. Second and third layers have attractive interactions to account for condensation in layer-by-layer growth. The calculations not only successfully account for the gross features of the TPD spectra but also explain a subtle feature of delayed desorption between third and second layers. As well, the lattice gas parameters obtained by this fit reproduce the bulk sublimation energy of cesium in the third layer. 

Our last example of the utility of the transfer matrix approach concerns the thermodynamics and desorption kinetics of H/Rh(311) obtained from... 

FIG. 5 Schematic of site parameters and interactions employed for the hollow-bridge site model of Te on W(IOO). Also depicted are the six hollow sites (squares) and adjacent bridge sites (small open circles) allowed in one strip in the construction of the transfer matrix. (Reprinted from Ref. 37 with permission from Elsevier Science.)... 

Additional applications of the transfer matrix method to adsorption and desorption kinetics deal with other molecules on low index metal surfaces [40-46], multilayers [47-49], multi-site stepped surfaces [50], and co-adsorbates [51-55]. A similar approach has been used to study electrochemical systems. 

For the equihbrium properties and for the kinetics under quasi-equilibrium conditions for the adsorbate, the transfer matrix technique is a convenient and accurate method to obtain not only the chemical potentials, as a function of coverage and temperature, but all other thermodynamic information, e.g., multiparticle correlators. We emphasize the economy of the computational effort required for the application of the technique. In particular, because it is based on an analytic method it does not suffer from the limitations of time and accuracy inherent in statistical methods such as Monte Carlo simulations. The task of variation of Hamiltonian parameters in the process of fitting a set of experimental data (thermodynamic and... 

Transfer matrix ealeulations ean be done routinely, for instanee, with the ASTEK program paekage for the analysis and simulation of thermal equilibrium and kineties of gases adsorbed on solid surfaees, written by H. J. Kreuzer and S. H. Payne (available from Helix Seienee Applieations, 618 Keteh Harbour Road, Portuguese Cove, N.S., B3V IKl, Canada). 

In the general case there will be n roots which are the eigenvalues of the transfer matrix K. Each of the eigenvalues defines a particular phase of the time course of the contents in the n compartments of the model. The eigenvalues are the hybrid transfer constants which appear in the exponents of the exponential function. For example, for the ith compartment we obtain the general solution ... 

---