A practical method for using bulk averaged values of observables for the characterization and prediction of the molecular population time evolution during isothermal relaxation is presented. In practical applications to realistic examples of vibrational relaxation very few bulk averages were required to accurately predict the population distribution even when the initial population was very strongly inverted. The time dependence of the macroscopic observables which are employed as input is conveniently formulated in terms of sum rules. The bulk average values are used as constraints in a maximal entropy procedure for the determination of the population distribution. It is shown that the procedure is of a variational type. Monotonic convergence of the information theoretic predicted distribution to the exact one is guaranteed upon inclusion of additional macroscopic input. The concept of ’’independent moments’’ is introduced for this purpose. Only independent observables are informative, i.e., provide independent data which are required for convergence. The number of informative observables decreases with time and is typically very much smaller than the number of significantly populated molecular energy states. The method is illustrated by comparing its predictions to the results of a numerical solution of the master equation, with a realistic set of rate constants and for different initial conditions. The application of the surprisal analysis to the interpretation, characterization, and compaction of the population distribution is demonstrated. Turning to predictions (’’surprisal synthesis’’), only strongly inverted initial populations required three independent moments, during the initial stages. Over much of the relaxation a single moment (’’vibrational temperature’’) sufficed for an accurate prediction. The limits where the characteristic vibrational temperature is high or low compared to the temperature of the buffer gas are discussed. Special reference is made to rotational relaxation. The rate of internal entropy production due to the irreversible relaxation and the rate of increase of the global entropy are discussed and shown to be positive.

## REFERENCES

*Reaction Kinetics*(The Chemical Society, London, 1975), Vol. 1, p. 243;

*Molecular Energy Transfer*, edited by R. D. Levine and J. Jortner (Wiley, New York, 1975), p. 67.

*Dynamics of Molecular Collisions*, edited by W. H. Miller (plenum, New York, 1976).

*Molecular Reaction Dynamics*(Clarendon, Oxford, 1975).

*Kinetic Processes in Gases and Plasmas*, edited by A. R. Hochstim (Academic, New York, 1969), p. 281.

*Quantum Mechanics of Molecular Rate Processes*(Clarendon, Oxford, 1969).

*n*by a finite number of transitions. Even when $k(n\u2192\nu )\u2009=\u20090,$ we assume that there is a sequence of intermediate states such that $k(n\u2192n\u2032)k(n\u2032\u2192n\u2033)\u2026k(\nu \u2032\u2192\nu )\u22600.$ For the rate constants employed in Sec. IV, $k(\nu \u2192\nu \u2032)\u22600$ for all ν and $\nu \u2032.$

^{3, 25, 26}the reference distribution is chosen as the “invariant distribution,” i.e., as the distribution which is unchanged under the equations of motion.

*The Principles of Statistical Mechanics*(Clarendon, Oxford, 1938), Sec. 104] and Pauli [W. Pauli,

*Sommerfeld Festschrift*, edited by P. Debye (Herzel Verlag, Leipzig, 1928), p. 30]. Their proofs that $\u2010dH\u033f/dt\u22650$ are thus equally valid for the present definition. [Tolman’s G’s are the equivalent of our $P\xb0$’s cf. his Eqs. (105.3) and (105.4).]

*The Theory of Heat Radiation*(Blakiston, Philadelphia, 1914), part V;

*Elementary Principles in Statistical Mechanics*(Scribner, New York, 1902), Chap. 12–14].

*Information Theory*(Interscience, New York, 1967), p. 8. For further discussion see

*Principles of Statistical Mechanics*(Freeman, San Francisco, 1967), or M. Tribus,

*Thermostatics and. Thermodynamics*(Van Nostrand, New York, 1961).

*n*th observable by its part that is independent of the previous $n\u22121$ observables, $n\u2009=1,2,\u2026$ Alternatively, one can envisage a symmetric procedure, which will treat all observables on equal footing.

*n*th state. This introduces “edge effects” seen, for example, in Fig. 14 where the increase of $(\nu )$ for $\nu \u2009=\u200919$ reflects the neglect of $k(19\u219220),$ etc.