KITS / IoP / CECAM Workshop & Discussion Meeting

Beijing, August 5-7 2018

Date: July 29 - Aug. 5, 2018

Venue: Rm. S102, Kavli ITS, UCAS [View Map]

Organizers

Daan Frenkel, University of Cambridge, UK

Ignacio Pagonabarraga Mora, CECAM, Switzerland

Rudi Podgornik, UCAS CAS, Beijing

Jure Dobnikar, IoP CAS Beijing

Description

The link between entropy and heat is one of the cornerstones of thermodynamics. Yet, thermodynamics provides no intuitive physical picture of entropy. In contrast, Statistical Mechanics provides a framework that highlights the relation between entropy and probability: it provides an unambiguous microscopic interpretation of entropy, although not necessarily one that is intuitively simple. The probabilistic framework of Statistical Mechanics can also be used to describe systems that are not in equilibrium, nor even material systems. For this reason, the term `entropy’ has spread beyond `Boltzmann-Gibbs’ statistical mechanics. Examples of such non-thermal entropies are Information Entropy and Granular Entropy. However, in the absence of a link between entropy and heat, different definitions of these entropies may be inequivalent and there is no a priori criterion to decide which ones (if any) are preferable.

In general, the `predictability’ of a system or, more precisely the minimal information that is needed to characterize it, is a measure for the (negative of the) `information entropy’. Interestingly, the information entropy of many-body systems appears to provide a good estimate of the thermal entropy (at least in systems where this comparison could be made). As this entropy can also be computed for non-thermal systems, this suggests that we now have another quantity (in addition to the probabilistic Gibbs entropy) that works in equilibrium but is also defined for non-thermal systems. But what does this entropy mean?

Moreover, other entropy definitions exist that are also computable, such as the `granular’ entropy and the pair entropy that is directly related to the structure of the system under consideration (provided that an appropriate metric exists). In addition, there are physical properties (such as e.g. structural hyper-uniformity) that seem to correlate with at least some of the entropy definitions. Other information measures may exist, and we will explore to what extent machine learning can be used to identify such entropy descriptors.

The aim of the discussion meeting is to assess the current state-of-the-art in this (still fragmented) field and identify possible fruitful topics for a full-scale workshop.

Schedule (Booklet Download)

Invited participants

Roy Beck-Barkai

Alpha Lee

Erik Luijten

Sri Sastry

Ali Naji

Limei Xu

Masao Doi

Rafi Blumenfeld

Ke Chen

Jiajia Zhou

Mincheng Yang

Xianren Zhang