Existing thermodynamic theory cannot throw light on the behaviour of non-equilibrium systems beyond the first bifurcation point as we leave equilibrium behind. We can explore such states theoretically only by considering the dynamics of the systems. To describe the one-way evolution of such non- equilibrium systems, we must construct mathematical models based on equations that show how various observable properties of a system change with time.In agreement with the second law of thermodynamics, such sets of equations describing irreversible processes always contain the arrow of time.
Chemical reactions provide typical examples of how this works. We can describe how fast such chemical reactions go as they are driven in the direction of thermodynamic equilibrium in terms of rate laws written in the form of differential equations. The quantities that we can measure are the concentrations of the chemicals involved and the rates at which they change with time.
We do not expect to see self-organising processes in every chemical reaction maintained far from equilibrium. But we often find that the mechanism underlying a reaction leads to differential equations that are nonlinear. Nonlinearities arise, for example, when a certain chemical present enhances (or suppresses) its own production; and they can generate unexpected complexity. Indeed, such nonlinearities are necessary, but not sufficient, for self-organised structures, including deterministic chaos, to appear. An example is the famous Belousov-Zhabotinski reaction discovered in the 1950s, described by Stephen Scott in his recent article on chemical chaos ( "Clocks and chaos in chemistry", New Scientist, 2 December 1989).
Today, as this series of chaos articles admirably shows, many scientists are using nonlinear dynamics to model a dizzying range of complicated phenomena, from fluid dynamics, through chemical and biochemical processes, to genetic variation, heart beats, population dynamics, evolutionary theory and even into economics. The two universal features of all these different phenomena are their irreversibility and their nonlinearity. Deterministic chaos is only one possible consequence; the other is a more regular self-organisation; indeed, chaos is just a special, but very interesting, form of self-organisation in which there is an overload of order.
We can again ask what is the origin of the irreversibility enshrined within the second law of thermodynamics. The traditional reductionist view is that we should seek the explanation on the basis of the reversible mechanical equations of motion. But, as the physicist Ludwig Boltzmann discovered, it is not possible to base the arrow of time directly on equations that ignore it. His failed attempt to reconcile microscopic mechanics with the second law gave rise to the "irreversibility paradox" that I mentioned at the beginning of this article.
The standard way of attempting to derive the equations employed in non-equilibrium thermodynamics starts from the equations of motion, whether classical or quantum mechanical, of the individual particles making up the system, which might, for example, be a gas. Because we cannot know the exact position and velocity of every particle, we have to turn to probability theory-statistical methods-to relate the average behaviour of each particle to the overall behaviour of the system. This is called statistical mechanics. The approach works very well because of the exceedingly large numbers of particles involved (of the order of 1024). The reason for using probabilistic methods is not, merely the practical difficulty of being unable to measure the initial positions and velocities of the participating particles. Quantum mechanics predicts these restrictions as a consequence of Heisenberg's uncertainty principle. But the same is also true for sufficiently unstable chaotic classical dynamical systems. In Ian Percival's article last year ("Chaos: a science for the real world", New Scientist, 21 October 1989), he explained that one of the characteristic features of a chaotic system is its sensitivity to the initial conditions: the behaviour of systems with different initial conditions, no matter how similar, diverges exponentially as time goes on. To predict the future, you would have to measure the initial conditions with literally infinite precision-a task impossible in principle as well as in practice. Again, this means we have to rely on a probabilistic description even at the microscopic level.