# Time reversibility

**Time reversibility** is an attribute of some stochastic processes and some deterministic processes.

If a stochastic process is time reversible, then it is not possible to determine, given the states at a number of points in time after running the stochastic process, which state came first and which state arrived later.

If a deterministic process is time reversible, then the time-reversed process satisfies the same dynamical equations as the original process (see reversible dynamics); in other words, the equations are invariant or symmetric under a change in the sign of time. Classical mechanics and optics are both time-reversible. Modern physics is not quite time-reversible; instead it exhibits a broader symmetry, CPT symmetry.{{ safesubst:#invoke:Unsubst||date=__DATE__ |$B=
{{#invoke:Category handler|main}}{{#invoke:Category handler|main}}^{[citation needed]}
}}

Time reversibility generally occurs when, within a process, it can be broken up into sub-processes which undo the effects of each other. For example, in phylogenetics, a time-reversible nucleotide substitution model such as the generalised time reversible model has the total overall rate into a certain nucleotide equal to the total rate out of that same nucleotide.

Time Reversal, specifically in the field of acoustics, is a process by which the linearity of sound waves is used to reverse a received signal; this signal is then re-emitted and a temporal compression occurs, resulting in a reverse of the initial excitation waveform being played at the initial source. Mathias Fink is credited with proving Acoustic Time Reversal by experiment.

## Stochastic processes

A formal definition of time-reversibility is stated by Tong^{[1]} in the context of time-series. In general, a Gaussian process is time-reversible. The process defined by a time-series model which represents values as a linear combination of past values and of present and past innovations (see Autoregressive moving average model) is, except for limited special cases, not time-reversible unless the innovations have a normal distribution (in which case the model is a Gaussian process).

A stationary Markov Chain is reversible if the transition matrix {*p _{ij}*} and the stationary distribution {

*π*} satisfy

_{j}for all *i* and *j*.^{[2]} Such Markov Chains provide examples of stochastic processes which are time-reversible but non-Gaussian.

Time reversal of numerous classes of stochastic processes have been studied including Lévy processes^{[3]} stochastic networks (Kelly's lemma)^{[4]} birth and death processes ^{[5]} Markov chains^{[6]} and piecewise deterministic Markov processes.^{[7]}

## See also

## Notes

- ↑ Tong(1990), Section 4.4
- ↑ Isham (1991), p 186
- ↑ Template:Cite doi
- ↑ Template:Cite jstor
- ↑ Template:Cite doi
- ↑ {{#invoke:citation/CS1|citation |CitationClass=book }}
- ↑ Template:Cite doi

## References

- Isham, V. (1991) "Modelling stochastic phenomena". In:
*Stochastic Theory and Modelling*, Hinkley, DV., Reid, N., Snell, E.J. (Eds). Chapman and Hall. ISBN 0-412-30390-9 Template:Please check ISBN. - Tong, H. (1990)
*Non-linear Time Series: A Dynamical System Approach*. Oxford UP. ISBN 0-19-852300-9

- Isham, V. (1991) "Modelling stochastic phenomena". In: