<?xml version="1.0" ?> <tei> <teiHeader> <fileDesc xml:id="0"/> </teiHeader> <text xml:lang="en"> <p>An 'optimal' solution to a problem is, in<lb/> some sense, the 'best' solution. It's the<lb/> shortest route to work, or the way of pack-<lb/>ing oranges that takes up the least vol-<lb/>ume. In science, optimality has long been<lb/> an organizing principle. Mathematical<lb/> physics views the Universe as unfolding<lb/> with dynamics that minimize a quantity<lb/> known as the 'action' , whereas economists<lb/> and other social scientists often take opti-<lb/>mality as a guide to human behaviour: we<lb/> act, they say, to maximize our utility, be it<lb/> financial or otherwise.<lb/></p> <p>Looking at our networked and inter-<lb/>connected world, we may wonder<lb/> whether anything is optimal here,<lb/> from the food webs that underlie<lb/> ecosystems, to technological networks<lb/> such as the Internet. And science recog-<lb/>nizes that the real world often falls short of<lb/> optimality. Physicists know of many mate-<lb/>rials in which, as in glass, complex inter-<lb/>actions among the molecules prevent<lb/> the ordered arrangement of lowest<lb/> (free) energy; these materials per-<lb/>sist naturally in disordered, sub-<lb/>optimal confusion. Meanwhile, the<lb/> economists' Homo economicus has been<lb/> replaced by a biologically more plausible<lb/> creature acting on the basis of fast yet fal-<lb/>lible instincts. We solve life's problems the<lb/> way we pack the dishwasher — not opti-<lb/>mally, after long calculation, but quickly<lb/> and more or less efficiently.<lb/></p> <p>What about networks? Is the Internet<lb/> optimal, in any sense? Is there a 'best' way<lb/> to design the connections that link together<lb/> a collection of cells, people or computers?<lb/> This question has many possible answers<lb/> — which always depend on what, if any-<lb/>thing, a network is 'designed' to do.<lb/></p> <p>In 1964, American engineer Paul Baran<lb/> conceptualized a layout for the command<lb/> and control network of the US military<lb/> aiming to withstand an attack by the Soviet<lb/> Union. He visualized it as a meshwork,<lb/> something like a fishnet, with roughly the<lb/> same number of links emanating from each<lb/> element. No one element would be a com-<lb/>munications hub — and primary target —<lb/> and the natural redundancy of paths would<lb/> allow messages to find a route through the<lb/> network even if much of it was destroyed.<lb/></p> <p>A few years later, US researchers had<lb/> Baran's meshwork in mind while devel-<lb/>oping the ARPANET, the seed of today's<lb/> Internet. Yet the Internet itself has turned<lb/> out to be wildly different from Baran's<lb/> hubless ideal. Although decentralized like<lb/> Baran's meshwork, it is in fact conspicuous<lb/> for its communications hubs — network<lb/> routers (or clusters of routers) that accu-<lb/>mulate far more connections than most<lb/> others. The network is especially sensitive<lb/> to any failure of these centres. In terms of<lb/> resilience, it is clearly not optimal.<lb/></p> <p>Of course, the forces driving Inter-<lb/>net growth have never been channelled<lb/> towards resilience, and its architecture<lb/> reflects the independent actions and deci-<lb/>sions of untold millions of people and<lb/> organizations. Quite possibly, its architec-<lb/>ture is not<lb/> optimal for anything.<lb/></p> <p>Yet in other settings the<lb/> notion of network optimality may have<lb/> considerable value. Suppose the popu-<lb/>lation in a certain country has a known<lb/> geographical distribution. How should<lb/> facilities such as hospitals be located most<lb/> conveniently? One recent study suggests<lb/> a relatively simple, if somewhat pecu-<lb/>liar, answer: that the density of facilities<lb/> shouldn't just be proportional to that of<lb/> people, but to that density raised to the<lb/> power of two-thirds. This puts more facili-<lb/>ties where there are more people, but not so<lb/> many more that they become redundant.<lb/> It is unlikely that urban planners currently<lb/> adhere to this principle, but it presents a<lb/> clear target for future policies.<lb/></p> <p>Another problem that a network might<lb/> solve is coordinating the activity of a set of<lb/> elements — by helping them to synchro-<lb/>nize their behaviour, for example. The<lb/> elements could be living cells such as neu-<lb/>rons, or anything else with rhythmic activ-<lb/>ity, and interact through signals of some<lb/> kind — electrical or chemical signals for<lb/> cells, or light for fireflies in a tree. It turns<lb/> out that the way such a network synchro-<lb/>nizes partly reflects its community struc-<lb/>ture, a community being any cluster with<lb/> far more links between its own elements<lb/> than to elements elsewhere in the network.<lb/> Studies show that under broad conditions,<lb/> the smallest and densest communities tend<lb/> to synchronize first, with larger-scale syn-<lb/>chronization arriving later.<lb/></p> <p>This suggests that a modular structure<lb/> of communities within communities may<lb/> allow for rich and varied dynamics — con-<lb/>ditions conducive to efficient information<lb/> processing and storage. In this sense, quite<lb/> plausibly, the wiring of the brain has been<lb/> crafted into optimal or near-optimal form<lb/> by the forces of evolution.<lb/></p> <p>Yet in other cases, even proven mathe-<lb/>matical optimality may lack any exemplars<lb/> in nature. Researchers have identified a<lb/> broad class of networks that conspicuously<lb/> lack any modular structure, yet support<lb/> extremely efficient global coordina-<lb/>tion. These networks — named after<lb/> the renowned Indian mathemati-<lb/>cian Srinivasa Ramanujan — have<lb/> highly uniform architectures,<lb/> with additional long loops<lb/> lending them an 'entangled'<lb/> character. They seem to be<lb/> almost optimal for many pur-<lb/>poses, for example by providing<lb/> networks that can be easily and<lb/> efficiently searched while avoiding<lb/> congestion. Even so, no one as yet has<lb/> found any natural network that adopts<lb/> this style.<lb/></p> <p>What all this suggests is that questions<lb/> of optimality are rarely straightforward. A<lb/> network will probably reveal optimal form<lb/> only if it has a ready means for evolving<lb/> towards it, and time to do so. And the<lb/> nature of such optimality always depends<lb/> on the network's function. In many cases<lb/> today, even that remains at least a partial<lb/> mystery.<lb/></p> ■ </text> </tei>