CPS: Breakthrough: Understanding Sub-Second Instabilities in a Global Cyber-Physical System

Abstract: Future real-world CPS will likely be large, messy and evolve in a decentralized way due to changing market conditions, yielding a system comprising many heterogeneous components that may have incompatible communication protocols (e.g. subsets of actuators and sensors built by different manufacturers and/or operated by competing entities). The number operating at any one time may also fluctuate, creating a time-dependent ecology comprising anywhere from a few hundred to several million distinct pieces of hardware, software or mechanical components. Since much of this ‘CPS ecology’ will be driven by technological advances in optoelectronics, which itself is racing toward the nanosecond operating scale, the overall CPS operation may produce undesirable behaviors on timescales lying beyond human response times, meaning that the CPS cannot be overridden in real time by any human-in-the-loop (see for example, Ref. [1]). In short, the CPS will operate quicker than any human regulator or moderator can respond. Such sudden extremes can also occur on the slower scales of minutes and days, as long as the underlying nucleating activity is sufficiently hidden within the CPS system and hence takes longer for any human-in-the-loop to find than it does to occur.

This raises the following key scientific questions which are of core interest to the safety and security of all large-scale CPS implementations: What is the nature of such emergent instabilities? How can they be described mathematically in a way that gives insight and quantitative results that are applicable across CPS systems, and are also scalable to CPS systems of any size? This project takes the following steps to answering these questions: We focus on developing a rigorous yet practical mathematical description of extreme behaviors in societal CPS ecologies comprising arbitrarily large numbers of heterogeneous components (sensors/actuators). By contrast, most existing theoretical descriptions of CPS systems focus on the optimal behavior of a well-defined CPS system with finite size and a known, fixed number of components. We have already developed (Year 1) a generic, scalable model that is minimal in its structure and amenable to mathematical analysis, yet highly non-trivial in terms of its ability to produce complex dynamics of the type observed empirically across a wide variety of real-world complex systems and networks [1-5]. Our modeling approach considers a dynamical multi-agent game which spontaneously produces large, unexpected (and hence unsafe or insecure) behaviors. The model features an arbitrary number of heterogeneous, adaptive sensor-actuator components which feed in information from the outside (sensor) and take an action (actuator). Though developed in a simplified environment, the findings of the model should hold in far more general situations [1-3] since they depend on a just a few commonly encountered system features. Our methodology is supplemented by the perspective of statistical mechanics from Physics and Chemistry, in which the details of individual nodes become washed out in favor of a focus on the correlations that develop between components – even though these components may be actively competing rather than cooperating. We are currently (Year 2) exploring the conditions under which large changes are generated in this model CPS ecology, and how these can be described and perhaps predicted mathematically in terms of their duration and magnitude.

As a real-world example, we show in the video a recent example from the world’s largest and fastest CPS: the network of networks of electronic financial exchanges (see Ref. [1] for more details). The empirical data show how the build-up of seemingly unimportant millisecond signal transmission delays, can generate a huge unexpected change within the CPS that suddenly arises from out of nowhere and lies b eyond human response times. We show that our scalable multi-agent model described above, is able to reproduce and explain such extremes. The challenge now lies in how to describe this mathematically. We have made some progress along this direction, and will present these init ial mathematical results. We will also discuss ‘CPS ecology management’ safety and security in large, decentralized CPS that have developed in an organic, market- driven way. More generally, this project promises new science not just for applications of CPS, but also for the core question of dynamical behavior of CPS complex systems. Broader impacts also include the promise of enhanced safety and security across a wide variety of societal CPS systems, from financial networks, smart energy buildings and cities, to autonomous vehicle design, both at individual and swarm level.

1. Neil F. Johnson. “To slow or not? Challenges in subsecond networks”, Science 355, 801 (2017)
2. Pedro D. Manrique, Minzhang Zheng, D. Dylan Johnson Restrepo, Pak Ming Hui, Neil F. Johnson. “Impact of delayed information in sub-second complex systems” Results in Physics 7, 3024 (2017)
3. Pedro D. Manrique, D. Dylan Johnson  and  Neil  F.  Johnson.  “Using  Competition  to  Control  Congestion  in  Autonomous  Drone Systems”, Electronics 6, 31 (2017)
4. P. Manrique, M. Zheng, Z. Cao, D. Johnson, P.M. Hui and N.F. Johnson. “Subsecond tsunamis and delays in decentralized electronic systems” Electronics 6, 80 (2017)
5. A. De Mendoza, F. Caycedo-Soler, P. Manrique, L. Quiroga, F. Rodriguez, N.F. Johnson. “Exploiting non-trivial spatio-temporal correlations of thermal radiation for sunlight harvesting”, J. Phys. B: At. Mol. Opt. Phys. 50, 124002 (2017)

Explanation of Demonstration: The demo shows a recent example of the type of extreme behavior of central interest to this project. It is taken from the world’s largest and fastest CPS: the network of networks of electronic financial exchanges [NFJ “To slow or not? Challenges in subsecond networks”, Science 355, 801 (2017)]. The empirical data show how the build-up of seemingly unimportant millisecond signal transmission delays, can generate a huge unexpected change within the CPS that suddenly arises from out of nowhere and lies beyond human response times. The scalable multi-agent CPS model that we are developing in this project, is able to reproduce and explain such extremes – and explain the impact of the implicit latency in the global information (see poster).