Category Archives: Quantum Physics

It does not answer the most fundamental mystery: what constitutes the dark energy and dark matter that make up the majority of our universe

The Standard Model evolved as the fundamental model of elementary particle physics in the 2nd half of the 20th century. It is considered today as the best description of the building blocks of universe. It explains how quarks (which form protons and neutrons) and leptons (electrons etc) make up all the known matter. It is also an explanation of how quarks and leptons are influenced via the exchange of intermediating force carriers. It describes three of the four fundamental interactions that sum up the structure of matter down to the measure of 10 raised to -18 metre. It is in short a quantum theory of three basic theories of electromagnetic interaction, strong interaction and weak interaction. The development of the Standard Model was driven both by huge number of experimental and theoretical physicists alike. The mathematical structure or framework for Standard Model is provided by Quantum Field theoryAccording to the Standard Model, all matter is made of three kinds of elementary particles: leptons, quarks, and their mediators. There are six leptons which fall into three generations. There are six anti-leptons as well, so the total number of leptons is 12. Similarly, the number of quarks is six, with each coming in three colours (this colour has no resemblance with the concept of colour in our everyday life), which accounts for 36 quarks in total including anti quarks. Quarks like leptons have three generations. Finally, for every interaction we have a mediator. The carrier of the electromagnetic interaction is a massless photon while carriers of the weak interaction are called intermediate vector bosons, which are two charged Ws and a neutral heavy Z. Finally, for exchange of strong interaction we have 8 gluons.The missing link in the Standard Model, the Higgs Boson theoretically predicted by Peter Higgs in early 1960s which accounts for the mass of elementary particles via Higgs mechanism, involving chiral symmetry breaking, was found in 2012 at the Large Hadron Collider in Geneva Switzerland. The marvellous achievement of the Standard Model can be gauged by the simple fact that it has led to over 50 Nobel Prizes in Physics so far.Loopholes:Even though the Standard Model is currently the best description we have of the sub-atomic world but despite its robust predictions, there is a consensus among physicists that the Standard Model is neither complete nor is it the final theory. There is a degree of ugliness in Standard Model, says Steve Weinberg, one of its prime architects. First, the Standard Model is totally silent on dark energy and dark matter. It does not answer the question of what constitutes the dark energy and dark matter that make up the majority of matter in our universe. Secondly, it does not explain neutrino oscillations and most importantly, it does not incorporate one of the most fundamental interactions, gravity, that accounts for the large-scale structure of the universe. On a more basic level it fails to explain why there are precisely three generations of quarks and leptons. Similarly, the difference in masses of the elementary particles which they gain as a result of their interaction with the Higgs Field via Higgs Boson remains a mystery.Possible way out:In order to account for many of the shortcomings of the Standard Model as listed above, physicists over the passage of time have come up with different theories and approaches. All these theories and approaches fall in the category of Physics Beyond Standard Model. Theories that lie Beyond Standard Model include the various extensions of supersymmetry and entirely novel explanations and theories such as string theory, Loop Quantam Gravity, and extra dimensions. But the theory that has gained most prominence among them is string theory. String theory has captured the imagination of an entire generation of particle physicists in the last 40 years. String theory not only promises the reconciliation of quantum mechanics with Einsteins General Relativity and eliminates the infinities that plague Quantam Field Theory, it also provides a unified theory of everything from which all elementary particle physics, including gravity, would emerge as an inescapable consequence. But the bottleneck of particle physics as string theorist and Nobel laureate David Gross says, is experimental and not theoretical, so in absence of experimental evidence to back up its predictions, string theorys future seems bleak, or at least one has to keep fingers crossed. If string theory meets expectations, which seems unlikely, it will be the ultimate triumph of the human mind.

The writer is a student of Physics[emailprotected]

Facebook Twitter WhatsAppEmailPrint

Related

Here is the original post:

Some reflections on the Standard Model of Particle Physics - Kashmir Reader

CU Boulder Professor Emeritus Karl Gustafson has high expectations for the quantum engineering researcher who will soon hold the faculty chair recently endowed in his name.

I hope that someone might have a fundamental breakthrough in some way other than just writing papers, he said. It's too easy to write papers, but very hard to actually build a piece of hardware, like a quantum computer or do something else in quantum.

Karl Gustafson

Recently created by an anonymous donor, the Karl Gustafson Endowed Chair of Quantum Engineering will be embedded in the Department of Electrical, Computer and Energy Engineering. It is intended for a faculty member with multidisciplinary research and teaching interests, who is focused on the hardware side of quantum computing and devices.

This gift, honoring Professor Gustafson for his distinguished career, will further enable CU Boulder and the College of Engineering and Applied Science to lead the way in quantum discoveries and application, Acting Dean Keith Molenaar said. We are deeply grateful to the donor and pleased that Karls legacy will be tied to the multidisciplinary impact of this faculty chair.

Gustafson said he was initially surprised when the donor a former student of his reached out to him about making a gift in his name.

It is quite an honor. Its an honor for me; it's an honor for the university, he said. It's kind of a very pleasant capstone on my career.

While he worked closely with the donor to outline the terms of the gift, it was Gustafson who insisted on the word multidisciplinary, as its the term that most accurately describes his own career.

Gustafson retired in 2020 after a 52-year career in the Department of Mathematics and holds three degrees from CU Boulder in engineering physics, applied mathematics and business finance. Shortly before a major stroke in 2016 prevented him from traveling and publishing, he gave three significant keynote addresses around the world one in each of those fields.

The Department of Physics is enormously proud that Professor Gustafson is one of our alumni, and deeply grateful to the donor for endowing this chair position in Quantum Engineering in Karls honor, said Michael Ritzwoller, the chair of physics. Karls illustrious career spanned mathematical physics, applied mathematics, and engineering both within and outside academia, and we hope the endowed chair holder can follow in his giant footsteps.

Gustafson advised more than 20 PhD students across science and engineering disciplines, and published more than 300 papers and more than a dozen books in topics ranging from computational fluid dynamics to financial engineering. His ties to CU Engineering include serving as a founding member of the multimillion-dollar NSF Optoelectronic Computing Systems Center from 1988 to 2000.

He has also long been interested in quantum mechanics and computation and keeps up-to-date on developments in the field even in retirement.

My impression is that we may not get quantum computation for a while. It's a very hard problem, Gustafson said. The software exists, the algorithms exist, all kinds of theory, thousands of papers. But no one can really build a quantum computer. You'll read a lot of hype about people claiming they have. But they'll only be able to do 20 or 30 qubits, and that's not a very big computer.

However, Gustafson said quantum is a fascinating field partly because of that search for the unknown. He thinks back to the Einstein-Bohr debates, which pitted a stubborn analytical person against a mystic. Gustafson said Bohr usually defeated Einstein in the thought experiments because he explained what we didnt know and claimed that we may never know.

I tend to hope that Einstein was right, Gustafson said. I tend to believe there's an underlying reality that we can maybe discover.

Here is the original post:

Honoree hopes new endowment will lead to quantum breakthrough - CU Boulder Today

If cats can be simultaneously dead and alive, can space aliens exist and not exist at the same time?

It might sound like the kind of thing youd ask while sitting around a campfire with two kinds of smoke lingering in the air, but that doesnt mean its not an important question.

At the center of our capacity for scientific inquiry lies a simple query: are we alone in the universe?

For millennia weve gazed at the skies with a binary perspective. Either theyre out there and we havent found them, or all life in the universe is confined to Earth.

Schrdingers cat is both dead and alive to an outside observer because, until the box is opened, you cant tell either way. You have to assume its one or another, which makes it essentially both until we know for sure.

The question is whether the universe knows for sure. And, to that, we answer: dont be ridiculous. If the universe has inherent sentience then all bets are off. We could be characters in a turtles dream.

The point being: if we assume for the sake of science that the universe doesnt have a base perspective, or its own unique point of view, then we have to figure that base reality is only made up of that which we can confirm through observation.

If the universe were governed strictly by classical physics, all would be copacetic. But, as we are nearly certain, the universe is actually a construct of quantum mechanics and, thus, it has quantum physics underpinning its apparent classical structure. And that means observation works differently in the quantum world.

Despite how it might feel, a watched pot doesnt technically boil any slower. You can measure the rate at which the temperature changes and predict within a reasonable amount of certainty how long itll take for your pasta to be ready.

However, in quantum terms, when we conduct measurements we fundamentally change reality. A quantum particle can exist in multiple states at once. We force it into one or another when we measure or observe it.

In quantum physics, the cat is both dead and alive until you open the box.

As New Scientists Amanda Gefter wrote in their fantastic article discussing multiverse theory:

When you jump into the cats point of view, it turns out that just as in relativity things have to warp to preserve the laws of physics. The quantumness previously attributed to the cat gets shuffled across the Heisenberg cut. From this perspective, the cat is in a definite state it is the observer outside the box who is in a superposition, entangled with the lab outside. Entanglement was long thought to be an absolute property of reality.

Basically, the universe doesnt just pretend at quantum physics. It lives it. Schrdingerscat is entangled to the observer outside of the box. But, to the cat, youre the entangled one.

Gefters article also references another physics experiment where theres a cat in a box thats dead and alive to the outside observer, and that observer has a friend outside of the laboratory staring at the door.

If the cats entangled inside the box to the first observer, and the first observer is entangled inside of the lab to the second observer, whose perspective is base reality? The outside-iest observer is the friend.

But what if the guy staring at the box with a cat in it hears a gunshot and yells Hey, friend, you okay out there?

Now wheres base reality at? Ill tell you: terrified. One false move in such a reality and bam, it all comes crashing down.

One resolution to the idea that the very fabric of our universe is subject to a ripple effect that could cause it to collapse in a quantum extinction event at any given second, is to imagine reality as a multiverse in and of itself.

The big idea here is that particles are all just tiny balls of potential that exist to be observed. That observation may very well dictate reality and, until that reality is shared by multiple observers, it remains independent of all other realities.

Gefters article explains this in elegantly simple terms: Start with observers sending messages, and you can derive space-time.

Imagine a coin spinning just out of view. If you turn your head far enough to glance at it, it will instantly drop and land on either heads or tails. Now imagine two coins spinning directly next to each other, each attached to an individual observer in a paradigm where whenever one observer glances, the other does too.

Neither observer would know which coin was attached to their observation, thus the significance of their effective control over bespoke reality would be lost on both.

However, if you could observe someone else changing reality with a mere glance, without the coincidence of also affecting reality with yours, the perspective would be much different.

Arguably, this is why the multiverse could be directly under our noses without us ever suspecting. Its possible that every time more than one intelligent observer changes a quantum outcome, theyre forced into a frame of reference that uses the uncertainty of never knowing which of us is actually causing the coins to stop spinning as a foundation for their shared reality.

It was a long walk, but heres the payoff. If this particular multiverse theory is true, then aliens should definitely exist.

Theyre currently entangled and, once we observe them, theyll either exist or they wont. Which means they exist.

However theres some bad news too. If the multiverse theory is true, then its also a certainty that aliens do not exist.

Because, from their perspective, were the entangled ones. Maybe we do exist, maybe we dont. Just like Schrdingerscat.

If were lucky, all this quantum physics stuff is so deeply embedded in the bedrock of reality that none of this actually matters. Just like opening up a box to find out the cats alive and well, maybe well bump into aliens one day and merge our realities.

But what if were not lucky? What if we run into aliens and, because of their observation of us, the coin of our reality stops spinning and lands on the side where we dont exist in any reality?

View original post here:

Why aliens and humans may not share the same reality - The Next Web

Department:Theoretical PhysicsVacancy ID:013861Closing Date:27-Feb-2022

We are seeking to appoint an energetic and enthusiastic Senior Post-doctoral Researcher in Theoretical Physics to work on development of compilers, quantum control protocols and algorithms for quantum information processing in quantum photonic systems as a part of the project Quantum Computing in Ireland: A Software Platform for Multiple Qubit Technologies (QCoIr) funded by Enterprise Ireland within the Disruptive Technology Innovation Fund. QCoIr which is the largest quantum computing project in Ireland is coordinated by IBM and involves a number of academic as well as industrial partners.

This strategically important partnership makes Department of Theoretical Physics at Maynooth University an excellent place to pursue theoretical physics research in quantum computation in collaboration with relevant experimental efforts as well as to explore its technological applications. In the project, Maynooth University provides theoretical physics expertise to multiple work packages which involve collaborations primarily with Tyndall National Institute, Rockley Photonics and IBM.

The successful candidate will join the group of Prof. Jiri Vala and will be working in a close collaboration primarily with IBM.

Salary

Post-Doctoral Researcher:39,132 per annum (with annual increments)

Senior Post-Doctoral Researcher: 46,442 per annum (with annual increments)

Appointment will be made in accordance with the Department of Finance pay guidelines.

See the original post here:

Post Doctoral Researcher, Theoretical Physics job with MAYNOOTH UNIVERSITY | 280363 - Times Higher Education (THE)

Significance

Modern evolutionary theory gives a detailed quantitative description of microevolutionary processes that occur within evolving populations of organisms, but evolutionary transitions and emergence of multiple levels of complexity remain poorly understood. Here, we establish the correspondence among the key features of evolution, learning dynamics, and renormalizability of physical theories to outline a theory of evolution that strives to incorporate all evolutionary processes within a unified mathematical framework of the theory of learning. According to this theory, for example, replication of genetic material and natural selection readily emerge from the learning dynamics, and in sufficiently complex systems, the same learning phenomena occur on multiple levels or on different scales, similar to the case of renormalizable physical theories.

We apply the theory of learning to physically renormalizable systems in an attempt to outline a theory of biological evolution, including the origin of life, as multilevel learning. We formulate seven fundamental principles of evolution that appear to be necessary and sufficient to render a universe observable and show that they entail the major features of biological evolution, including replication and natural selection. It is shown that these cornerstone phenomena of biology emerge from the fundamental features of learning dynamics such as the existence of a loss function, which is minimized during learning. We then sketch the theory of evolution using the mathematical framework of neural networks, which provides for detailed analysis of evolutionary phenomena. To demonstrate the potential of the proposed theoretical framework, we derive a generalized version of the Central Dogma of molecular biology by analyzing the flow of information during learning (back propagation) and predicting (forward propagation) the environment by evolving organisms. The more complex evolutionary phenomena, such as major transitions in evolution (in particular, the origin of life), have to be analyzed in the thermodynamic limit, which is described in detail in the paper by Vanchurin etal. [V. Vanchurin, Y. I. Wolf, E. V. Koonin, M. I. Katsnelson, Proc. Natl. Acad. Sci. U.S.A. 119, 10.1073/pnas.2120042119 (2022)].

What is life? If this question is asked in the scientific rather than in the philosophical context, a satisfactory answer should assume the form of a theoretical model of the origin and evolution of complex systems that are identified with life (1). NASA has operationally defined life as follows: Life is a self-sustaining chemical system capable of Darwinian evolution (2, 3). Apart from the insistence on chemistry, long-term evolution that involves (random) mutation, diversification, and adaptation is, indeed, an intrinsic, essential feature of life that is not apparent in any other natural phenomena. The problem with this definition, however, is that natural (Darwinian) selection itself appears to be a complex rather than an elementary phenomenon (4). In all evolving organisms we are aware of, for natural selection to kick off and to sustain long-term evolution, an essential condition is replication of a complex digital information carrier (a DNA or RNA molecule). The replication fidelity must be sufficiently high to provide for the differential replication of emerging mutants and survival of the fittest ones (this replication fidelity level is often referred to as Eigen threshold) (5). In modern organisms, accurate replication is ensured by elaborate molecular machineries that include not only replication and repair enzymes but also, the entire metabolic network of the cell, which supplies energy and building blocks for replication. Thus, the origin of life is a typical chicken-and-egg problem (or catch-22); accurate replication is essential for evolution, but the mechanisms ensuring replication fidelity are themselves products of complex evolutionary processes (6, 7).

Because genome replication that underlies natural selection is itself a product of evolution, origin of life has to be explained outside of the traditional framework of evolutionary biology. Modern evolutionary theory, steeped in population genetics, gives a detailed and arguably, largely satisfactory account of microevolutionary processes: that is, evolution of allele frequencies in a population of organisms under selection and random genetic drift (8, 9). However, this theory has little to say about the actual history of life, especially the emergence of new levels of biological complexity, and nothing at all about the origin of life.

The crucial feature of biological complexity is its hierarchical organization. Indeed, multilevel hierarchies permeate biology: from small molecules to macromolecules; from macromolecules to functional complexes, subcellular compartments, and cells; from unicellular organisms to communities, consortia, and multicellularity; from simple multicellular organisms to highly complex forms with differentiated tissues; and from organisms to communities and eventually, to eusociality and to complex biocenoses involved in biogeochemical processes on the planetary scale. All these distinct levels jointly constitute the hierarchical organization of the biosphere. Understanding the origin and evolution of this hierarchical complexity, arguably, is one of the principal goals of biology.

In large part, evolution of the multilevel organization of biological systems appears to be driven by solving optimization problems, which entails conflicts or trade-offs between optimization criteria at different levels or scales, leading to frustrated states, in the language of physics (1012). Two notable cases in point are parasitehost arms races that permeate biological evolution and makes major contributions to the diversity and complexity of life-forms (1316) and multicellular organization of complex organisms, where the tendency of individual cells to reproduce at the highest possible rate is countered by the control of cell division imposed at the organismal level (17, 18).

Two tightly linked but distinct fundamental concepts that lie effectively outside the canonical narrative of evolutionary biology address evolution of biological complexity: major transitions in evolution (MTEs) (1921) and multilevel selection (MLS) (2227). Each MTE involves the emergence of a new level of organization, often described as an evolutionary transition in individuality. A clear-cut example is the evolution of multicellularity, whereby a new level of selection emerges, namely selection among ensembles of cells rather than among individual cells. Multicellular life-forms (even counting only complex organisms with multiple cell types) evolved on many independent occasions during the evolution of life (28, 29), implying that emergence of new levels of complexity is a major evolutionary trend rather than a rare, chance event.

The MLS remains a controversial concept, presumably because of the link to the long-debated subject of group selection (27, 30). However, as a defining component of MTE, MLS appears to be indispensable. A proposed general mechanism behind the MTE, formulated by analogy with the physical theory of the origin of patterns (for example, in glass-like systems), involves competing interactions at different levels and the frustrated states, such interactions cause (12). In the physical theory of spin glasses, frustrations result in nonergodicity and enable formation and persistence of long-term memory: that is, history (31, 32). By contrast, ergodic systems have no true history because they reach all possible states during their evolution (at least in the large time limit), and thus, the only content of quasihistory of such systems is the transition from less probable to more probable states for purely combinatorial reasons: that is, entropy increase (33). As emphasized in Schroedingers seminal book (34), even if only in general terms, life is based on negentropic processes, and frustrations at different levels are necessary for these processes to take off and persist (12).

The origin of cells, which can and probably should be equated with the origin of life, was the first and most momentous transition at the onset of biological evolution, and as such, it is outside the purview of evolutionary biology sensu stricto. Arguably, the theoretical investigation of the origin of life can be feasible only within the framework of an envelope theory that would incorporate biological evolution as a special case. It is natural to envisage such a theory as encompassing all nonergodic processes occurring in the universe, of which life is a special case, emerging under conditions that remain to be investigated and defined.

Here, in pursuit of a maximally general theory of evolution, we adopt the formalism of the theory of machine learning (35). Importantly, learning here is perceived in the maximally general sense as an objective process that occurs in all evolving systems, including but not limited to biological ones (36). As such, the analogy between learning and selection appears obvious. Both types of processes involve trial and error and acceptance or rejection of the results based on some formal criteria; in other words, both are optimization processes (22, 37, 38). Here, we assess how far this analogy extends by establishing the correspondence between key features of biological evolution and concepts as well as the mathematical formalism of learning theory. We make the case that loss function, which is central to the learning theory, can be usefully and generally employed as the equivalent of the fitness function in the context of evolution. Our original motivation was to explain major features of biological evolution from more general principles of physics. However, after formulating such principles and embedding them within the mathematical framework of learning, we find that the theory can potentially apply to the entire history of the evolving universe (36), including physical processes that have been taking place since the big bang and chemical processes that directly antedated and set the stage for the origin of life. The central propositions of the evolution theory outlined here include both key physical principles (namely, hierarchy of scale, frequency gaps, and renormalizability) (39, 40) and major features of life (such as MLS, persistence of genetic parasites, and programmed cell death).

We show that learning in a complex environment leads to separation of scales, with trainable variables splitting into at least two classes: faster- and slower-changing ones. Such separation of scales underlies all processes that involve the formation of complex structure in the universe from the scale of an atom to that of clusters of galaxies. We argue that, for the emergence of life, at least three temporal scales, which correspond to environmental, phenotypic, and genotypic variables, are essential. In evolving learning systems, the slowest-changing variables are digitized and acquire the replication capacity, resulting in differential reproduction depending on the loss (fitness) function value, which is necessary and sufficient for the onset of evolution by natural selection. Subsequent evolution of life involves emergence of many additional scales, which correspond to MTE. Hereafter, we use the term evolution to describe temporal changes of living and lifelike and prebiotic systems (organisms), whereas the more general term dynamics refers to temporal processes in other physical systems.

At least since the publication of Schroedingers book, the possibility has been discussed that, although life certainly obeys the laws of physics, a different class of laws unique to biology could exist. Often, this putative physics of life is associated with emergence (4143), but the nature of the involved emergent phenomena, to our knowledge, has not been clarified until very recently (36). Here, we outline a general approach to modeling and studying evolution as multilevel learning, supporting the view that a distinct type of physical theory, namely the theory of learning (35, 36), is necessary to investigate the evolution of complex objects in the universe, of which evolution of life is a specific, even if highly remarkable form.

In this section, we attempt to formulate the minimal universal principles that define an observable universe, in which evolution is possible and perhaps, inevitable. Our analysis started from the major features of biological evolution discussed in the next section and proceeded toward the general principles. However, we begin the discussion with the latter for the sake of transparency and generality.

What are the requirements for a universe to be observable? The possibility to make meaningful observations implies a degree of order and complexity in the observed universe emerging from evolutionary processes, and such evolvability itself seems to be predicated on several fundamental principles. It has to be emphasized that observation and learning here by no means imply mind or consciousness but a far more basic requirement. To learn and survive in an environment, a system (or observer) must predict, with some minimal but sufficient degree of accuracy, the response of that environment to various actions and to be able to choose such actions that are compatible with the observers continued existence in that environment. In this sense, any life-form is an observer, and so are even inanimate systems endowed with the ability of feedback reaction. In this most general sense, observation is a prerequisite for evolution. We first formulate the basic principles underlying observability and evolvability and then, give the pertinent comments and explanations.

P1. Loss function. In any evolving system, there exists a loss function of time-dependent variables that is minimized during evolution.

P2. Hierarchy of scales. Evolving systems encompass multiple dynamical variables that change on different temporal scales (with different characteristic frequencies).

P3. Frequency gaps. Dynamical variables are split among distinct levels of organization separated by sufficiently wide frequency gaps.

P4. Renormalizability. Across the entire range of organization of evolving systems, a statistical description of faster-changing (higher-frequency) variables is feasible through the slower-changing (lower-frequency) variables.

P5. Extension. Evolving systems have the capacity to recruit additional variables that can be utilized to sustain the system and the ability to exclude variables that could destabilize the system.

P6. Replication. In evolving systems, replication and elimination of the corresponding information-processing units (IPUs) can take place on every level of organization.

P7. Information flow. In evolving systems, slower-changing levels absorb information from faster-changing levels during learning and pass information down to the faster levels for prediction of the state of the environment and the system itself.

The first principle (P1) is of special importance as the starting point for a formal description of evolution as a learning process. The very existence of a loss function implies that the dynamical system of the universe or simpler, the universe itself is a learning (evolving) system (36). Effectively, here we assume that stability or survival of any subsystem of the universe is equivalent to solving an optimization or learning problem in the mathematical sense and that there is always something to learn. Crucially, for solving complex optimization problems dependent on many variables, the best and in fact, the only efficient method is selection implemented in various stochastic algorithms (Markov Chain Monte Carlo, stochastic gradient descent, genetic algorithms, and more). All evolution can be perceived as an implementation of a stochastic learning algorithm as well. Put another way, learning is optimization by trial and error, and so is evolution.

The remaining principles P2 to P7 provide sufficient conditions for observers of our type (that is, complex life-forms) to evolve within a learning system. In particular, P2, P3, and P4 comprise the necessary conditions for observability of a universe by any observer, whereas P5, P6, and P7 represent the defining conditions for the origin of life of our type (hereafter, we omit the qualification for brevity). More precisely, P2 and P3 provide for the possibility of at least a simple form of learning of the environment (fast-changing variables) by an observer (slow-changing variables) and hence, the emergence of complex organization of the slow-changing variables. P4 corresponds to the physical concept of renormalizability, or renormalization group (39, 40), whereby the same macroscopic equations, albeit with different parameters, govern processes at different levels or scales, thus limiting the number of relevant variables, constraining the complexity, and allowing for a coarse-grained description. This principle ensures a renormalizable universe capable of evolution and amenable to observation. Together, P2 to P4 define a universe, in which partial or approximate knowledge of the environment (in other words, coarse graining) is both attainable and useful for the survival of evolving systems (observers). In a universe where P4 does not apply (that is, one with nonrenormalizable physical laws), what happens at the macroscopic level will critically depend on the details of the processes at the microlevel. In a universe where P2 and P3 do not apply, the separation of the micro- and macrolevels itself would not be apparent. In such a universe, it would be impossible to survive without first discovering fundamental physical laws, whereas living organisms on our planet have evolved for billions of years before starting to study quantum physics.

Principles P5, P6, and P7 endow evolving systems with the access to more advanced algorithms for learning and predicting the environment, paving the way for the evolution of complex systems, including eventually, life. These principles jointly underlie the emergence of the crucial phenomenon of selection (44, 45). In its simplest form, selection is for stability and persistence of evolving, learning systems (46). Learning and survival are tightly linked because survival is predicated on the systems ability to extract information from the environment, and this ability depends on the stability of the system on timescales required for learning. Roughly, a system cannot survive in a world where the properties of the environment change faster than the evolving system can learn them. According to P5, evolving systems consume resources (such as food), which themselves could be produced by other evolving systems, to be utilized as building blocks and energy sources, which are required for learning. This principle embodies Schroedingers vision that organisms feed on negentropy (34). Under P6, replication of the carriers of slowly changing variables becomes the basis of long-term persistence and memory in evolving systems. This principle can be viewed as a learning algorithm built on P3, whereby the timescales characteristic of an individual organism and of consecutive generations are separated. This principle excludes from consideration certain imaginary forms of life: for example, Stanislav Lems famous Solaris (47). Finally, P7 describes how information flows between different levels in the multilevel learning, giving rise to a generalized Central Dogma of molecular biology, which is discussed in Generalized Central Dogma of Molecular Biology.

In this section, we link the fundamental principles of evolution P1 to P7 formulated above to the basic phenomenological features of life (E1 to E10) and seek equivalencies in the theory of learning. The list below is organized by first formulating a biological feature, and then, it is organized by 1) tracing the connections to the fundamental principles and 2) adding more general comments.

Discrete IPUs (that is, self- vs. nonself-differentiation and discrimination) exist at all levels of organization. All biological systems at all levels of organization, such as genes, cells, organisms, populations, and so on up to the level of the entire biosphere, possess some degree of self-coherence that separates them, first and foremost, from the environment at large and from other similar-level IPUs.

1) The existence of IPUs is predicated on the fundamental principles P1 to P4. The wide range of temporal scales (P2) in dynamical systems and gaps between the scales (P3) naturally enable the separation of slower- and faster-changing components. In particular, renormalizability (P4) applies to the hierarchy of IPUs. The statistical predictability of the higher frequencies allows the IPUs to decrease the loss function of the lower frequencies, despite the much slower reaction times.

2) Separation of (relatively) slow-changing prebiological IPUs from the (typically) fast-changing environment kicked off the most primitive form of prebiological selection: selection for stability and persistence (survivor bias). More stable, slower-changing IPUs win in the competition and accumulate over time, increasing the separation along the temporal axis as the boundary between the IPUs and the environment grows sharper. Additional key phenomena, such as utilization of available environmental resources (P5) and the stimulusresponse mode of information exchange (P7), stem from the flow of matter and information across this boundary and the ensuing separation of internal and external physicochemical processes. Increasing self- vs. nonself-differentiation, combined with replication of the carriers of slow-changing variables (P6), sets the stage for competition between evolving entities and for the onset of the ultimate evolutionary phenomenon, natural selection (E6).

All complex, dynamical systems face multidimensional and multiscale optimization problems, which generically lead to frustration resulting from conflicting objectives at different scales. This is a key, intrinsic feature of all such systems and a major force driving the advent of increasing multilevel complexity (12). Frustration is an extremely general physical phenomenon that is by no account limited to biology but rather, occurs already in much simpler physical systems, such as spin and structural glasses, the behavior of which is determined by competing interactions so that a degree of complexity is attained (31, 32).

1) The multiscale organization of the universe (P2) provides the physical foundation for the ubiquity of frustrated states that typically arises whenever there is a conflict (trade-off) between short- and long-range optimization problems. Frustrated interactions yield multiwell potential landscapes, in which no single state is substantially fitter than numerous other local optima. Multiparameter and multiscale optimization of the loss function on such a landscape involves nonergodic (history-dependent) dynamics, which is characteristic of complex systems.

2) IPUs face conflicting interactions starting from the most primitive prebiological state (12). Indeed, the separation of any system from the environment immediately results in the conflict of permeability; a stronger separation enhances the self- vs. nonself-differentiation and thus, increases the stability of the system, but it compromises information and matter exchange with the environment, limiting the potential for growth. In biology, virtually all aspects of the organismal architecture and operation are subject to such frustrations or trade-offs: the conflict between the fidelity and speed of information transmission at all levels, between specialization and generalism, between the individual- and population-level benefits, and more. The ubiquity of frustrations and the fundamental impossibility of their resolution in a universally optimal manner are perpetual drivers of evolution and give rise to evolutionary transitions, attaining otherwise unreachable levels of complexity.

There are two distinct types of frustrations, spatial and temporal. Spatial frustration is similar to the frustration that is commonly analyzed in condensed matter systems, such as spin glasses (31, 32). In this case, the spatially local and nonlocal interacting terms have opposite signs so that the equilibrium state is determined by the balance between the terms. In neural networks, a neuron (like a single spin) might have a local objective (such as binary classification of incoming signals) but is also a part of a neural network (like a spin network), which has its own global objective (such as predicting its boundary conditions). For a particular neuron, optimization of the local objective can conflict with the global objective, causing spatial frustration. Temporal frustration emerges because in the context of multilevel learning, the same neuron becomes a part of higher-level IPUs that operate at different temporal scales (frequencies). Then, the optimal state of the neuron with respect to an IPU operating at a given timescale can differ from the optimal state of the same neuron with respect to another IPU operating at a different timescale (36). Similarly to the spatial frustrations, temporal frustrations cannot be completely resolved, but an optimal balance between different spatial and temporal scales is achievable and represents a local equilibrium of the learning system.

The hierarchy of multiple levels of organization is an intrinsic, essential feature of evolving biological systems in terms of both the structure of these systems (genes, genomes, cells, organisms, kin groups, populations, species, communities, and more) and the substrate the evolutionary forces act upon.

1) Renormalizability of the universe (P4) implies that there is no inherently preferred level of organization, for which everything above and below would behave as a homogenous ensemble. Even if some levels of organization come into existence before others (for example, organisms before genes or unicellular organisms before multicellular ones), the other levels will necessarily emerge and consolidate subsequently.

2) The hierarchy of the structural organization of biological systems was apparent to scholars from the earliest days of science. However, MLS was and remains a controversial subject in evolutionary biology (23, 26, 27). Intuitively and as implied by the Price equation (48), MLS should emerge in all evolving systems as long as the higher-level agency of selection possesses a sufficient degree of self- vs. nonself-differentiation. In particular, if organisms of a given species form populations that are sufficiently distinct genetically and interact competitively, population-level selection will ensue. Evolution of biological systems is driven by conflicting interactions (E2) that tend to lead to ever-increasing complexity (12). This trend further feeds the propensity of these systems to form new levels of organization and is associated with evolutionary transitions that involve the advent of new units of selection at multiple levels of complexity. Thus, E3 can be considered a major consequence of E2.

Stochastic optimization or the use of stochastic optimization algorithms is the only feasible approach to complex optimization, but it guarantees neither finding the globally optimal solution nor retention of the optimal configuration when and if it is found. Rather, stochastic optimization tends to rapidly find local optima and keeps the system in their vicinity, sustaining the value of the loss function at a near-optimal level.

1) According to P1, the dynamics of a learning (that is, self-optimizing) system is defined by a loss function (35, 36). When there is a steep gradient in the loss function, a system undergoing stochastic optimization rapidly descends in the right direction. However, because of frustrations that inevitably arise from interactions in a complex system, actual local peaks on the landscape are rarely reached, and the global peak is effectively unreachable. Learning systems tend to get stalled near local saddle points where changes along most of the dimensions either lead up or are flat in terms of the loss function, with only a small minority of the available moves decreasing the loss function (49).

2) The extant biological systems (cells, multicellular organisms, and higher-level entities, such as populations and communities) are products of about 4 billion y of the evolution of life, and as such, they are highly, albeit not completely, optimized. As a consequence, the typical distribution of the effects of heritable changes in biological evolution comprises numerous deleterious changes, comparatively rare beneficial changes and common neutral changes, and those with fitness effects below the noise level (50). The preponderance of neutral and slightly deleterious changes provides for evolution by genetic drift whereby a population moves on the same level or even slightly downward on the fitness landscape, potentially reaching another region of the landscape where beneficial mutations are available (51, 52).

Solutions on the loss function landscapes that arise in complex optimization problems span numerous local peaks of comparable heights.

1) The existence of multiple peaks of comparable heights in the loss function landscapes is a fundamental physical property of frustrated systems (E2), whereas the pervasiveness of frustration itself is a consequence of the multiscale and multilevel organization of the universe (P2). Frustrated dynamical systems are nonergodic, which from the biological perspective, means that, once separated, evolutionary trajectories diverge rather than converge. Because most of these trajectories traverse parts of the genotype space with comparable fitness values, competition rarely results in complete dominance of one lineage over the others but rather, generates rich diversity.

2) In terms of evolutionary biology, fitness landscapes are rugged, with multiple adaptive peaks of comparable fitness (53, 54), and a salient trend during evolution is the spread of life-forms across multiple peaks as opposed to concentrating on one or few. Evolution pushes evolving organisms to explore and occupy all available niches and try all possible strategies. In the context of machine learning, identical neural networks can start from the same initial state but for example, under the stochastic gradient descent algorithm, would generically evolve toward different local minima. Thus, the diversity of solutions is a generic property of learning systems. More technically, the diversification is due to the entropy production through the dynamics of the neutral trainable variables (see the next section).

This quintessential feature of life embodies two distinct (albeit inseparable in known organisms) symmetry-breaking phenomena: 1) separation between dedicated digital information storage media (stable, rarely updatable, tending to distributions with discrete values) and mostly analog processing devices and 2) asymmetry of the information flow within the IPUs whereby the genotype provides instructions for the phenotype, whereas the phenotype largely loses the ability to update the genotype directly. The separation between the information storage and processing subsystems is a prerequisite for efficient evolution that probably emerged early on the path from prebiotic entities to the emergence of life.

1) The separation between phenotype and genotype extends the scale separation on the intra-IPU level as follows from the fundamental principles P1 to P4. Intermediate-frequency components of an IPU (phenotype) buffer the slowest components from direct interaction with the environment (the highest-frequency variables), further increasing the stability of the slowest components and making them suitable for long-term information storage. As the temporal scales separate further, the interactions between them change. Asymmetric information flow (P7) stabilizes the system, enabling long-term preservation of information (genotype) while retaining the reactive flexibility of the faster-changing components (phenotype).

2) The emergence of the separation between phenotype and genotype is a crucial event in prebiotic evolution. This separation is prominent in all known as well as hypothetical life-forms. Even when the phenotype and genotype roles are fulfilled by chemically identical molecules, as in the RNA world scenario of primordial evolution (55, 56), their roles as effectors and information storage devices are sharply distinct. In biological terms, the split is between replicators (that is, the digital information carriers [genomes]) and reproducers (5759), the analog devices (cells, organisms) that host the replicators, supply them with building blocks (P5), and themselves reproduce (P6) under the replicators instruction (P7). Although the genotype/phenotype separation is a major staple of life, it is in itself insufficient to qualify an IPU as a life-form (computers and record players, in which the separation between information storage and operational parts is prominent and essential, clearly are not life, even though invented by advanced organisms). The asymmetry of information flow between genotype and phenotype (P7) is the most general form of the phenomenon known as the Central Dogma of molecular biology: the unidirectional flow of information from nucleic acids to proteins as originally formulated by Crick (60). This asymmetry is also prominent in other information-processing systems, in particular computers. Indeed, von Neumann architecture computers have inherently distinct memory and processing units, with the instruction flow from the former to the latter (61, 62). It appears that any advanced information-processing systems are endowed with this property.

Emergence of long-term digital storage devices, that is genomes consisting of RNA or DNA (E6) provides for long-term information preservation, facilitates adaptive reactions to changes in the environment, and promotes the stability of IPUs to the point where (at least in chemical systems) it is limited by the energy of the chemical bonds rather than the energy of thermal fluctuations. Obviously, however, as long as this information is confined to a single IPU, it will disappear with the inevitable eventual destruction of that IPU. Should this be the case, other IPUs of similar architecture would need to accumulate a comparable amount of information from scratch to reach the same level of stability. Thus, copying and sharing information are essential for long-term (effectively, indefinite) persistence of IPUs.

1) The fundamental principle P6 postulates the existence of mechanisms for information copying and elimination. If genomic information can be replicated, even most primitive sharing mechanisms (such as physical splitting of an IPU under forces of surface tension) would result (even if not reliably) in the emergence of distinct IPUs preloaded with information that was amassed by their progenitor(s). This process short circuits learning and allows the information to accumulate at timescales far exceeding the characteristic lifetimes of individual IPUs.

2) Information copying and sharing are beneficial only if the fidelity exceeds a certain threshold, sometimes called Eigen limit in evolutionary biology (57). Nevertheless, in primitive prebiotic systems, the required fidelity level could have been quite low (63). For instance, even a biased chemical composition of a hydrophobic droplet could enhance the stability of the descendant droplets and thus, endow them with an advantage in the selection for persistence. However, once relatively sophisticated mechanisms of information copying and sharing emerge or more precisely, when replicators become information storage devices, the overall stability of the system can increase by orders of magnitude. To wit and astonishingly, the only biosphere known to us represents an unbroken chain of genetic information transmission that spans about 4 billion y, commensurate with the stellar evolution scale.

Evolution by natural selection (Darwinian evolution) arises from the combination of all the principles and phenomena described above. The necessary and sufficient conditions for Darwinian evolution to operate are 1) the existence of IPUs that are distinct from the environment and from each other (E1), 2) the dependence of the stability of an IPU on the information it contains (that is, the phenotypegenotype feedback; E6), and 3) the ability of IPUs to make copies of embedded information and share it with other IPUs (E7). When these three conditions are met, the relative frequencies of the more stable IPUs will increase with time via attrition of the less stable ones (survival of the fittest) and transfer of information among IPUs, both vertically (to progeny) and horizontally. This process engenders the key feature of Darwinian evolution, differential reproduction of genotypes, based on the feedback from the environment transmitted through the phenotype.

1) All seven fundamental principles of life-compatible universes (P1 to P7) are involved in enabling evolution by natural selection. The very existence of units, on which selection can operate, hinges on self- vs. nonself-discrimination of prebiotic IPUs (E1) and the emergence of shareable information storage (E6 and E7). The crucial step to biology is the emergence of the link between the loss function (P1), on the one hand, and the existence of the IPUs (P2, P3, P4, and E1), on the other hand. Consumption of (limited) external resources (P5) entails competition between IPUs that share the same environment and turns mere shifts of the relative frequencies into true survival of the fittest. The ability of the IPUs to replicate (P6) and expand their memory storage (genotype; P7, E6, and E7) provides them with access to hitherto unavailable degrees of freedom, making evolution an open-ended process rather than a quick, limited search for a local optimum.

2) Evolution by natural selection is the central tenet of evolutionary biology and a key part of the NASA definition of life. An important note on definitions is due. We already referred to selection when discussing prebiotic evolution (E1); however, the term natural (Darwinian) selection is here reserved for the efficient form of selection that emerges with the replication of dedicated information storage devices (P6 and E6). Differential reproduction, whereby the environment provides feedback on the fitness of genotypes while acting on phenotypes, turns into Darwinian survival of the fittest in the presence of competition. When IPUs depend on environmental resources, such competition inevitably arises, except in the unrealistic case of unlimited supply (44). With the onset of Darwinian evolution, the system can be considered to cross the threshold from prelife to life (64, 65). The evolutionary process is naturally represented by movement of an evolving IPU in a genotype space, where proximity is defined by similarity between distinct genotypes and transitions correspond to elementary evolutionary events: that is, mutations in the most general sense (66). For any given environment, fitnessthat is, a measure of the ability of a genotype to produce viable offspringcan be defined for each point in the genotype space, forming a multidimensional fitness landscape (53, 54). Selection creates a bias for preferential fixation of mutations that increase fitness, even if the mutations themselves occur completely randomly.

Parasites and hostparasite coevolution are ubiquitous across biological systems at multiple levels of organization and are both intrinsic to and indispensable for the evolution of life.

1) Due to the flexibility of life-compatible systems (P5 and P6) and the symmetry breaking in the information flow (P7) combined with the inherent tendency of life to diversify (E5), parts of the system inevitably settle on a parasitic state: that is, scavenging information and matter from the host without making a positive contribution to its fitness.

2) From the biological perspective, parasites evolve to minimize their direct interface with the environment and conversely, maximize their interaction with the host; in other words, the host replaces most of the environment for the parasite. Parasites inevitably emerge and persist in biological systems because of two reasons. 1) The parasitic state is reachable via an entropy-increasing step and therefore, is highly probable (16), and 2) highly efficient antiparasite immunity is costly (67). The cost of immunity reflects another universal trade-off analogous to the trade-off between information transfer fidelity and energy expenditure; in both cases, an infinite amount of energy is required to reach a zero error rate or a parasite-free state. From a complementary standpoint, parasites inevitably evolve as cheaters in the game of life that exploit the host as a resource, without expending energy on resource production. Short-term, parasites reduce the host fitness by both direct drain on its resources and various indirect effects, including the cost of defense. However, in a longer-term perspective, parasites make up a reservoir for recruitment of new functions (especially, but far from exclusively, for defense) by the hosts (14, 15). The hostparasite relationship can evolve toward transition to a mutually beneficial, symbiotic lifestyle that can further progress to mutualism and in some cases, complete integration as exemplified by the origin of essential endosymbiotic organelles in eukaryotes, mitochondria, and chloroplasts (68, 69). Parasites emerge at similar levels of biological organization (organisms parasitizing other organisms) or across levels (genetic elements parasitizing organismal genomes or cell clones parasitizing multicellular organisms).

Programmed (to various degrees) death is an intrinsic feature of life.

1) Replication and elimination of IPUs (P6) and utilization of additional degrees of freedom (P5) form the foundation for the phenomenon of programmed death. At some levels of organization (for example, intragenomic), the ability to add and eliminate units (such as genes) for the benefit of the higher-level systems (such as organisms) provides an obvious path of optimization. Elimination of units could be, in principle, completely random, but selection (E8) generates a sufficiently strong feedback to facilitate and structure the loss process (for example, purging low-fitness genes via homologous recombination or altruistic suicide of infected or otherwise impaired cells). The same forces operate at least at the cell level and conceivably, at all levels of organization and selection (P4). In particular, if population-level or kin-level selection is sufficiently strong, mechanisms for altruistic death of individual organisms apparently can be fixed in evolution (70, 71).

2) Programmed death is a prominent case of minimization of the higher-level (for example, organismal) loss function at the cost of increasing the lower-level loss function (such as that of individual cells). Although (tightly controlled) programmed cell death was originally discovered in multicellular organisms and has been thought to be limited to these complex life-forms, altruistic cell suicide now appears to be a universal biological phenomenon (7173).

To conclude this section, which we titled fundamental evolutionary phenomena, deliberately omitting biological, it seems important to note that phenomena E1 to E7 are generic, applying to all learning systems, including purely physical and prebiotic ones. However, the onset of natural selection (E8) marks the origin of life, so that the phenomena E8 to E10 belong in the realm of biology.

In the previous sections, we formulated the seven fundamental principles of evolution P1 to P7 and then, argued that the key evolutionary phenomena E1 to E10 can be interpreted and analyzed in the context of these principles and apparently, derived from the latter. The next step is to formulate a mathematical framework that would be consistent with the fundamental principles and thus, would allow us to model evolutionary phenomena analytically or numerically. For concreteness, the proposed framework is based on a mathematical model of artificial neural networks (74, 75), but we first outline a general optimization approach in a form suitable for modeling biological evolution.

We are interested in the broadest class of optimization problems, where the loss (or cost) function H(x,q) is minimized with respect to some trainable variables,q=(q(c),q(a),q(n)),[3.1]for a given training set of nontrainable variables,x=(x(o),x(e)).[3.2]

Near a local minimum, the first derivatives of the average loss function with respect to trainable variables q are small, and the depth of the minimum usually depends on the second derivatives. In particular, the second derivative can be large for the effectively constant degrees of freedom, q(c); small for adaptable degrees of freedom, q(a); or near zero for symmetries or neutral directions q(n). The separation of the neutral directions q(n) into a special class of variables simply means that some of the trainable variables can be changed without affecting the learning outcome: that is, the value of the loss function. Put another way, neutral changes are always possible. The neutral directions q(n) are the fastest changing among the trainable variables because fluctuations resulting in their change are, in general, fully stochastic. On the other end of the spectrum of variables, even minor changes to the effectively constant variables q(c) compromise the entire learning (evolution) process: that is, result in a substantial increase of the loss function value; these variables correspond to deep minima of the loss function. When the basin of attraction of a minimum is deep and narrow, the system stays in its bottom for a long time, and then, to describe such a state, it is sufficient to use discrete information (that is, to indicate that the system stays in a given minimum) rather than to list all specific values of the coordinates in a multidimensional space.

In a generic optimization problem, the dynamics of both trainable and nontrainable variables involves a broad distribution of characteristic timescales , and switching between scales is equivalent to switching between different frequencies or in the context of biological evolution, between different levels of organization. For any fixed , all variables can be partitioned into three classes depending on how fast they change with respect to the specified timescale:

fast-changing nontrainable variables that characterize an organism(x(o)) and its environment x(e) and change on timescales ;

intermediate-changing adaptable variables q(a) or neutral directions q(n) that change on timescales ~; and

slow-changing variables, which are the degrees of freedom q(c) that have already been well trained and are effectively constant (at or near equilibrium), only changing on timescales .

As will become evident shortly, the separation of these three classes of variables and interactions between them are central to the evolution and selection on all levels of organization, resulting in pervasive multilevel learning and selection.

Depending on the considered timescale (or as a result of environmental changes), the same dynamical degree of freedom can be assigned to different classes of variables: that is, x(o), x(e), q(c), q(a), or q(n). For example, on the shortest timescale, which corresponds to the lifetime of an individual organism (one generation), the adaptable variables are the phenotypic traits that quickly respond to environmental changes, whereas the slowest, near-constant variables are the genomic sequences (genotype) that change minimally if at all. On longer timescales, corresponding to thousands or millions of generations, fast-evolving portions of the genome become adaptable variables, whereas the conserved core of the genome remains in the near-constant class (50). Analogously, the neutral directions correspond either to nonconsequential phenotypic changes or to neutral genomic mutations, depending on the timescale. It is well known that the overwhelming majority of the mutations are either deleterious and therefore, eliminated by purifying selection or (nearly) neutral and thus, can be either lost or fixed via drift (76, 77). However, when the environment changes or under the influence of other mutations, some of the neutral mutations can become beneficial [a genetic phenomenon known as epistasis, which is pervasive in evolution (78, 79)], and in their entirety, neutral mutations form the essential reservoir of variation available for adaptive evolution (80). Even which variables are classified as nontrainable (x) depends on the timescale . For example, if a learning system was trained for a sufficiently long time, some of the trainable variables q(a) or q(n) might have already equilibrated and become nontrainable.

Now that we described an optimization problem that is suitable for modeling evolution of organisms (or populations of organisms), we can construct a mathematical framework to solve such optimization problems. For this purpose, we employ a mathematical theory of artificial neural networks (74, 75), which is simple enough to perform calculations while being consistent with all of the fundamental principles (P1 to P7), and thus, it can be used for modeling evolutionary phenomena (E1 to E10). We first recall a general framework of the neural network theory.

Consider a learning system represented as a neural network, with the state vector described by trainable variables q (which describe a collective notation for weight matrix w^ and bias vector b) and nontrainable variables x (which describe the current state vector of individual neurons). In the biological context, x collectively represent the current state of the organism x(o) and of its environment x(e), and q determines how x changes with time, in particular, how the organism reacts to environmental challenges. The nontrainable variables are modeled as changing in discrete time stepsxi(t+1)=fi(jwijxj(t)+bi),[4.1]where fi(y)s are some nonlinear activation functions (for example, hyperbolic tangent or rectifier activation functions). The trainable variables are modeled as changing according to the gradient descent (or stochastic gradient descent) algorithmqi(t+1)=qi(t)H(x(t),q(t))qi,[4.2]

where is the learning rate parameter and H(x,q) is a suitably defined loss function (Eqs. 4.3 and 4.4). In other words, q are gross or main variables, which determine the rules of dynamics, and the dynamics of all other variables x is governed by these rules, per Eq. 4.1. In the biological context, Eq. 4.1 represents fast, often stochastic environmental changes and the corresponding fast reaction of organisms at the phenotype level, whereas [4.2] reflects slower-learning dynamics of evolutionary adaptation via changes in the intermediate, adaptable variables: that is, the variable portion of the genome. The main learning objective is to adjust the trainable variables such that the average loss function is minimized subject to boundary conditions (also known as the training dataset), which in our case, is modeled as a time sequence of the environmental variables.

For example, on a single-generation timescale, the fast-changing variables represent the environment x(e) and nontrainable variables associated with organisms x(o), the intermediate-changing variables represent adaptive q(a) and neutral q(n) phenotype changes, and the slow-changing variables q(c) represent the genotype (SI Appendix, Fig. S1).

The temporal-scale separation in biology is readily apparent in all organisms. Indeed, consequential changes in the environment x(e) often occur on the scale of milliseconds to seconds, triggering physical changes within organisms x(o) at matching timescales. In response, individual organisms respond with phenotypic changes both adaptive q(a) and neutral q(n) on the scale of minutes to hours, exploiting their genetically encoded phenotypic plasticity. A paradigmatic example is induction of bacterial operons in response to a change in the chemical composition of the environment, such as the switch from glucose to galactose as the primary nutrient (81, 82). In contrast, changes in the genome q(c) take much longer. Mutations typically occur at rates of about 1 to 10 per genome replication cycle (83), which for unicellular organisms, is the same as a generation comprising from about an hour to hundreds or even thousands of hours. However, fixation of mutations, which represents an evolutionarily stable change at the genome level, typically takes many generations and thus, always occurs orders of magnitude slower than phenotype changes. Accordingly, on this timescale, any changes in the genome represent the third layer in the network, the slowly changing variables.

To specify a microscopic loss function that would be appropriate for describing evolution and thus, give a specific form to the fundamental principle P1, we first note that adaptation to the environment is more efficient (that is, the loss function value is smaller) for a learning system, such as an organism, that can predict the state of its environment with a smaller error. Then, the relevant quantity is the so-called boundary loss function defined as the sum of squared errors,He(x,q)12iE(xi(e)fi(x(o),q))2,[4.3]where the summation is taken only over the boundary (or environmental) nontrainable variables. It is helpful to think of the boundary loss function as the mismatch between the actual state of the environment and the state that would be predicted by the neural network if the environmental dynamics was switched off. In neuroscience, boundary loss is closely related to the surprise (or prediction error) associated with predictions of sensations, which depend on an internal model of the environment (84). In machine learning, boundary loss functions are most often used in the context of supervised learning (35), and in biological evolution, the supervision comes from the environment, which the evolving system, such as an organism or a population, is learning to predict.

Another possibility for a learning system is to search for the minimum of the bulk loss function, which is defined as the sum of squared errors over all neurons:H(x,q)=12i(xifi(x(o),q))2.[4.4]

The bulk loss function assumes extra cost incurred by changing the states of organismal neurons, x(o): that is, rewarding stationary states. In the limit of a very large number of environmental neurons, the two loss functions are indistinguishable, H(x,q)He(x,q), but bulk loss is easier to handle mathematically (the details of boundary and bulk loss functions are addressed in ref. 35).

More generally, in addition to the kinetic term [4.4], the loss function can include a potential term V(x,q):H(x,q)=12i(xifi(x(o),q))2+V(x,q).[4.5]

The kinetic term in [4.5] reflects the ability of organisms x(o) to predict the changes in the state of the given environment x(e) over time, whereas the potential term reflects their compatibility with a given environment and hence, the capacity to choose among different environments.

In the context of biological evolution, Malthusian fitness is defined as the expected reproductive success of a given genotype: that is, the rate of change of the prevalence of the given genotype in an evolving population (85). However, in the context of the theory of learning, the loss function must be identified with additive fitness: that is,H(x,q)=Tlog(x,q).[4.6]

For a microscopic description of learning, the proportionality constant is unimportant, but as we argue in detail in the accompanying paper (86), in the description of the evolutionary process from the point of view of thermodynamics, T plays the role of evolutionary temperature.

Given a concrete mathematical model of neural networks, one might wonder if all fundamental principles of evolution (P1 to P7) can be derived from this model. Such derivation would comprise additional evidence supporting the claim that the entire universe can be adequately described as a neural network (36). Clearly, the existence of a loss function (P1) follows automatically because learning of any neural network is always described relative to a specified loss function (Eq. 4.4 or 4.5). The other six principles also seem to naturally emerge from the learning dynamics of neural networks. In particular, the hierarchy of scales (P2) and frequency gaps (P3) are generic consequences of the learning dynamics, whereby a system that involves a wide range of variables changing at different rates is attracted toward a self-organized critical state of slow-changing trainable variables (87). Additional gaps between levels of organization are also expected to appear through phase transitions as becomes apparent in the thermodynamic description of evolution we develop in the accompanying paper (86). Renormalizability (P4) is a direct consequence of the second law of learning (35), according to which entropy of a system (and consequently, complexity of neural network or rank of its weight matrix) decreases with learning. This phenomenon was observed in neural network simulations (35) and is the exact type of dynamics that can make the system renormalizable even if it started off as a highly entangled (large rank of weight matrix), nonrenormalizable neural network. The extension (P5) and replication (P6) principles simply indicate that additional variables can lead to either increase or decrease in the value of the loss function (35). It is also important to note that in neural networks, an additional computational advantage (quantum advantage) can be achieved if the number of IPUs can vary (88). Therefore, to achieve such an advantage, a system must learn how to replicate and eliminate its IPUs (P6). Finally, in Generalized Central Dogma of Molecular Biology, we illustrate how Fourier transform (or more generally, wavelet transform) of the environmental degrees of freedom can be used for learning the environment and how the inverse transform can be used for predicting it. Thus, to be able to predict the environment (and hence, to be competitive), any evolving system must learn the mechanism behind such asymmetric information flow (P7).

In the previous sections, we argued that the learning process naturally divides all the dynamical variables into three distinct classes: fast-changing ones, x(o) and x(e); intermediate-changing ones, q(a) and q(n) (q(n) being faster than q(a)); and slow-changing ones, q(c) (SI Appendix, Fig. S1). Evidently, this separation of variables depends on the timescale during which the system is observed, and variables migrate between classes when is increased or decreased (SI Appendix, Fig. S2). The longer the time, the more variables reach equilibrium and therefore, can be modeled as nontrainable and fast changing, x(e), and the fewer variables remain slowly varying and can be modeled as effectively constant q(c). In other words, many variables that are nearly constant at short timescales migrate to the intermediate class at longer timescales, whereas variables from the intermediate class migrate to the fast class.

In biological terms, if we consider learning dynamics on the timescale of one generation, then q(a) and q(n) represent phenotype variables, and q(c) represents genotype variables; however, on much longer timescales of multiple generations, the learning dynamics of populations (or communities) of organisms becomes relevant. On such timescales, the genotype variables acquire dynamics, with purifying and positive selection getting into play, whereas the phenotype variables progressively equilibrate. There is a clear connection between learning dynamics, including that in biological systems, and renormalizability of physical theories (P4). Indeed, from the point of view of an efficient learning algorithm, the parameters controlling learning dynamics, such as effective learning (or information-processing) rate , can vary from one timescale to another (for example, from individual organisms to populations or communities of organisms), but the general principles as well as specific dependencies captured in the equations above that govern the learning dynamics on different timescales remain the same. We refer to this universality of the learning process on different timescales and partitioning of the variables into temporal classes as multilevel learning.

More precisely, multilevel learning is a property of learning systems, which allows for the basic equations of learning, such as [4.4], to remain the same on all levels of organization but for the parameters, which describe the dynamics such as (), to depend on the level or on the timescale . For example, if the effective learning (or information-processing) rate () decreases with timescale , then the local processing time, which depends on (), runs differently for different trainable variables: slower for slow-changing variables (or larger ) and faster for fast-changing ones (or smaller ). For such a system, the concept of global time (that is, the same time for all variables) becomes irrelevant and should be replaced with the proper or local time, which is defined for each scale separately:t()t.[5.1]

This effect closely resembles time dilation phenomena in physics, except that in special and general relativity, time dilation is linked with the possibility of movement between slow and fast clocks (or variables) (89). To illustrate the role time dilation plays in biology, consider only two types of variables: slow changing and fast changing. Then, the slow variables should be able to outsource certain computational tasks to faster variables. Because the local clock for the fast-changing variables runs faster, the slow-changing variables can take advantage of the fast-changing ones to accelerate computation, which would be rewarded by evolution. The flow of information between slow-changing and fast-changing variables in the opposite direction is also beneficial because the fast-changing variables can use the slow-changing variables to store useful information for future retrieval: that is, the slow variables function as long-term memory. In the next section, we show that such cooperation between slow- and fast-changing variables, which is a concrete manifestation of principle P7, corresponds to a crucial biological phenomenon as the Central Dogma of molecular biology (60).

In terms of learning theory, the two directions of the asymmetric information flow (P7) represent learning the state of the environment and predicting the state of the environment from the results of learning. For learning, information is passed from faster variables to slower variables, and for predicting, information flows in the opposite direction from slower variables to the faster ones. A more formal analysis of the asymmetric information flows (or a generalized Central Dogma) can be carried out by forward propagation (from slow variables to fast variables) and back propagation (from fast variables to slow variables) of information within the framework of the mathematical model of neural networks developed in the previous sections (SI Appendix, Fig. S3).

Consider nontrainable environmental variables that change continuously with time x(e)(t), while the learning objective of an organism is to predict x(e)(t) at time t> given that it was observed for time 0, and to do so, it should be able to store and retrieve the values of the Fourier coefficientsqk=10dtx(e)(t)ei2fkt,[6.1]or more generally, wavelet coefficientsqk=10dtWk(t)x(e)(t)ei2fkt,[6.2]

for suitably defined window functions Wi. Then, a prediction could be made by extrapolating x(e)(t) using the inverse transformationx(e)(t+)2k=kminkmaxRe(qkei2fk),[6.3]

for some >0, which is not too large compared with . However, in general, the total number of (Fourier or wavelet) coefficients qk would be countably infinite. Therefore, any finite-size organism has to decide which frequencies to observe (and remember) and which ones to filter out (and forget).

Let us assume that the organism decided to only observe/remember discrete frequenciesfminfkmin,,fkmaxfmax,[6.4]and forget everything else. Then, to predict the state of the environment [6.3] and as a result, minimize the loss function [4.5], the organism should be able to store, retrieve, and adjust information about coefficients qk in some adaptable trainable variables q(a).

Given this simple model, we can study the flow of information between different nontrainable variables of the organism x(o). To this end, it is convenient to organize the variables asx=(x(o),x(e))=(xkmin,,xkmax,x(e)),[6.5]wherexk(t+)2l=kminkRe(qlei2fl),[6.6]

and assume that the relevant information about qis is stored in the adaptable trainable variablesq(a)=(qkmin,,qkmax).[6.7]

In the estimate of xk(t+) in Eq. 6.6, all the higher-frequency modes are assumed to average to zero as is often the case if we are only interested in the timescale fk1. A better estimate can be obtained using, once again, the ideas of the renormalization group flow following the fundamental principle P4. To make learning (and thus, survival) efficient, truncation of the set of variables relevant for learning is crucial. The main point is that the higher-frequency modes can still contribute statistically, and then, an improved estimate of xk(t+) would be obtained by appropriately modifying the values of the coefficients qk. Either way, in order to make an actual prediction, the organism should first calculate xk(t+) for small fk and then, pass the result to the next level to calculate xk+1(t+) for larger fk+1 and so on. Such computations can be described by a simple mappingxk+1(t+)=xk(t+)+2Re(qk+1ei2fk+1),[6.8]which can be interpreted as passage of data from one layer to another in a deep, multilayer neural network (SI Appendix, Fig. S2). Eq. 6.8 implies that, during the predicting phase, relevant information only flows from variables encoding low frequencies to variables encoding high frequencies but not in the reverse direction. In other words, in the process of predicting the environment, information propagates from slower variables to faster variables: that is, from genotype to phenotype or from nucleic acids to proteins (hence, the Central Dogma). Because only the fast variables change in this process, the prediction of the state of the environment is rapid, as it is indeed required to be for the organism survival. Conversely, in the process of learning the environment, information is back propagated in the opposite direction: that is, from faster to slower variables. However, this back propagation is not a microscopic reversal of the forward propagation but a distinct, much slower process (given that changes in slow variables are required) that involves mutation and selection.

Thus, the meaning of the generalized Central Dogma from the point of view of the learning theoryand our theory of evolutionis that slow dynamics (that is, evolution on a long timescale) should be mostly independent of the fast variables. In less formal terms, slow variables determine the rules of the game, and changing these rules depending on the results of some particular games would be detrimental for the organism. Optimization within the space of opportunities constrained by temporally stable rules is advantageous compared with optimization without such constraints. The trade-off between global and local optimization is a general, intrinsic property of frustrated systems (E2). For the system to function efficiently, the impact of local optimization on the global optimization should be restricted. The separation of the long-term and short-term forms of memory through different elemental bases (nucleic acids vs. proteins) serves this objective.

In this work, we outline a theory of evolution on the basis of the theory of learning. The parallel between learning and biological evolution becomes obvious as soon as the mapping between the loss function and the fitness function is established (Eq. 4.6). Indeed, both processes represent movement of an evolving (learning) system on a fitness (loss function) landscape, where adaptive (learning), upward moves are most consequential, although neutral moves are most common, and downward moves also occur occasionally. However, we go beyond the obvious analogy and trace a detailed correspondence between the essential features of the evolutionary and learning processes. Arguably, the most important fundamental commonality between evolution and learning is the stratification of the trainable variables (degrees of freedom) into classes that differ by the rate of change. At least in complex environments, all learning is multilevel, and so is all selection that is relevant for the evolutionary process. The framework of evolution as learning developed here implies that evolution of biological complexity would be impossible without MLS permeating the entire history of life. Under this perspective, emergence of new levels of organization, in learning and in evolution, and in particular, MTE represent genuine phase transitions as previously suggested (41). Such transitions can be analyzed consistently only in the thermodynamic limit, which is addressed in detail in the accompanying paper (86).

The origin of complexity and long-term memory from simple fundamental physical laws is one of the hardest problems in all of science. One popular approach is synergetics pioneered by Haken (91, 92) and the related nonequilibrium thermodynamics founded by Prigogine and Stengers (93) that employ mathematical tools of the theory of dynamical systems, such as theory of bifurcations and analysis of attractors. However, these concepts appear to be too general and oversimplified to usefully analyze biological phenomena, which are far more complex than dissipative structures that are central to nonequilibrium thermodynamics, such as, for example, autowave chemical reactions.

An alternative is the approach based on the theory of spin glasses (43, 94), which employs the mathematical apparatus of statistical physics and seems to provide a deeper insight into the origin of complexity. However, the energy landscape of spin glasses contains too many minima that are too shallow to account for long-term memory that is central to biology (12, 41). Thus, some generalization of the spin glass concept is likely to be required for productive application in evolutionary biology (95).

A popular and promising approach is self-organized criticality (SOC), a concept developed by Bak etal. (96, 97). Although relevant in biological contexts (12), SOC, by definition, implies self-similarity between different levels of organization, whereas biologically relevant complexity is rather associated with distinct emergent phenomena at different spatiotemporal scales (90).

A fundamental shortcoming of all these approaches is that they do not include, at least not as a major component, evolutionary concepts, such as natural selection. The framework of learning theory used here allows us to naturally unify the descriptions of physical and biological phenomena in terms of optimization by trial and error and loss (fitness) functions. Indeed, a key point of the present analysis is that most of our general principles apply to both living and nonliving systems.

The detailed correspondence between the key features of the processes of learning and biological evolution implies that this is not a simple analogy but rather, a reflection of the deep unity of evolutionary processes occurring in the universe. Indeed, separation of the relevant degrees of freedom into multiple temporal classes is ubiquitous in the universe from composite subatomic particles, such as protons, to atoms, molecules, life-forms, planetary systems, and galaxy clusters. If the entire universe is conceptualized as a neural network (36), all these systems can be considered emerging from the learning dynamics. Furthermore, scale separation and renormalizability appear to be essential conditions for a universe to be observable. According to the evolution theory outlined here, any observable universe consists of systems that undergo learning or synonymously, adaptive evolution, and actually, the universe itself is such a system (36). The famous dictum of Dobzhansky (98), thus, can and arguably should be rephrased as [n]othing in the world is comprehensible except in the light of learning.

Within the theory of evolution outlined here, the difference between life and nonliving systems, however important, can be considered as one in the type and degree of optimization, so that all evolutionary phenomena can be described within the same formal framework of the theory of learning. Crucially, any complex optimization problem can be addressed only with a stochastic learning algorithm: hence, the ubiquity of selection. Origin of life can then be conceptualized within the framework of multilevel learning as we explicitly show in the accompanying paper (86). The point when life begins can be naturally associated with the emergence of a distinct class of slowly changing variables that are digitized and thus, can be accurately replicated; these digital variables store and supply information for forward propagation to predict the state of the environment. In biological terms, this focal point corresponds to the advent of replicators (genomes) that carry information on the operation of reproducers within which they reside (99). This is also the point when natural (Darwinian) selection takes off (64). Our theory of evolution implies that this pivotal stage was preceded by evolution of prelife, which comprised reproducers that lacked genomes but nevertheless, were learning systems that were subject to selection for persistence. Self-reproducing micelles that harbor autocatalytic protometabolic reaction networks appear to be plausible models of such primordial reproducers (100). The first replicators (RNA molecules) would evolve within these reproducers, perhaps, initially, as molecular parasites (E9) but subsequently, under selection for the ability to store, express, and share information essential for the entire system. This key step greatly increased the efficiency of evolution/learning and provided for long-term memory that persisted throughout the history of life, enabling the onset of natural selection and the unprecedented diversification of life-forms (E5). It has to be emphasized that, compared with the existing evolutionary models that explore replicator dynamics, the learning approach described here is more microscopic in that the existence of replicators is not initially assumed but rather, appears as an emergent property of multilevel learning dynamics. For learning to be efficient, the capacity of the system to add new adaptable variables is essential. In biological terms, this implies expandability of the genome (that is, the ability to add new genes), which necessitated the transition from RNA to DNA as the genome substrate given the apparent intrinsic size constraints on replicating RNA molecules. Another essential condition for efficient learning is information sharing, which in the biological context, corresponds to horizontal gene transfer. The essentiality of horizontal gene transfer at the earliest stages of life evolution is perceived as the cause of the universality of the translation machinery and genetic code in all known life-forms (101). The conceptual model of the origin of life implied by our learning-based theoretical framework appears to be fully compatible with Gntis chemoton, a model of protocell emergence and evolution based on autocatalytic reaction networks (102104).

Go here to read the rest:

Toward a theory of evolution as multilevel learning - pnas.org

Who was the smartest person in the world? There are certainly many worthy contenders. Today, the very name of Einstein is synonymous with genius. Others may suggest Stephen Hawking. Those who appreciate literature and music may proffer William Shakespeare or Ludwig van Beethoven. Historians may recommend Benjamin Franklin.

Before I submit my own suggestion, we must first discuss what we even mean bysmart. Colloquially, we routinely interchange the wordssmartandintelligent, but they are not necessarily the same thing. There is an ongoing debate among psychologists, neuroscientists, and artificial intelligence experts on what intelligence actually is, but for our purposes here, a simpledictionary definitionwill suffice: capacity for learning, reasoning, understanding, and similar forms of mental activity; aptitude in grasping truths, relationships, facts, meanings, etc.

Implicit in this definition of intelligence isgeneral knowledge. An intelligent person capable of understanding quantum mechanics is useless to society if he is completely ignorant. So, a truly smart person will know a lot of things, preferably about many different topics. He should be apolymath, in other words.

Finally, there is the element ofcreativity. Creative people think in ways in which most other people do not. Where society sees a dead end, a creative person sees an opportunity.

Which person from history was the bodily manifestation of intelligence, knowledge, and creativity? Isaac Newton.

What was Newtons IQ? Its impossible to say. IQ tests didnt exist in the 17th Century, and if they had, Mr. Newton certainly would not have deigned to spend 90 minutes filling out ovals on a multiple choice test. Besides, he likely would have finished the test early and then spent the remaining time correcting errors and devising more difficult questions.

Nobody doubts that Isaac Newton was an intelligent man, but he also exhibited in spades the two other characteristics outlined above: knowledge and creativity.

Newton was a true polymath. Not only did he master physics and mathematics, but he was also a theologian. He was obsessed with eschatology (end-times prophecy), and he calculated based on his interpretation of the Bible thatJesus Christ would return to Earth in 2060. His dedication to religion was so great that,according toNature, more than half of his published writings were on theology.

He also became well versed in alchemy. Do not hold that against him. Many great scientists of his time believed that any metal could be transmuted into gold.The Economistexplainswhy the notion was not entirely unreasonable in Newtons time:

Alchemical theories were not stupid. For instance, lead ore often contains silver and silver ore often contains gold, so the idea that lead ripens into silver, and silver into gold, is certainly worth entertaining. The alchemists also discovered some elements, such as phosphorous.

Furthermore, later in life, Newton dabbled in economics. James Gleick, author of the truly excellent biographyIsaac Newton, wrote that [h]e wrestled with issues of unformed monetary theory and international currency. As Master of the Mint, Newton was tasked with tracking down currency counterfeiters, which he did, as Gleick wrote, with diligence and even ferocity. He showed no pity in his relentless pursuit of justice. When notorious counterfeiter William Chaloner attacked Newtons personal integrity, he doubled down his efforts to catch him.Mental Flossreports:

Acting more the grizzled sheriff than an esteemed scientist, Newton bribed crooks for information. He started making threats. He leaned on the wives and mistresses of Chaloners crooked associates. In short, he became the Dirty Harry of 17th-century London.

Newtons sleuthing worked. Chaloner was caught and hanged.

Impressive as all that, what truly separates Newton from other luminaries was his unparalleled creativity. He created multiple tools that simply never existed before. For example, in order to study acceleration, the change in velocity, a tool beyond basic algebra was required. That tool, called the derivative, is the most basic function in calculus. It didnt exist in the 17th century. Newton invented it.

In order to find the area beneath a curve, another tool beyond basic algebra was needed. That tool, called integration, is the second most basic function in calculus. Like the derivative, it did not exist in the 17th century. So, Newton invented it. He also invented areflecting telescopeand the ridges on coins, which serve as ananti-theft measurethat prevents coin clipping.

Newtons inventiveness is perhaps best summarized by the epigraph to Gleicks biography, which was written by his nieces husband in 1726:

I asked him where he had it made, he said he made it himself, & when I asked him where he got his tools said he made them himself & laughing added if I had staid for other people to make my tools & things for me, I had never made anything

Sadly, despite his fame, Isaac Newton led a very lonely life. His incomparable brilliance came at a hefty cost; his reclusive and anti-social nature strongly suggest that he was autistic, and his obsessive and disagreeable nature suggest mental illness, perhaps obsessive-compulsive disorder.Mental Flossnot-so-charitablydescribesNewton as suffering from everything:

[H]istorians agree he had a lot going on. Newton suffered from huge ups and downs in his moods, indicating bipolar disorder, combined with psychotic tendencies. His inability to connect with people could place him on the autism spectrum. He also had a tendency to write letters filled with mad delusions, which some medical historians feel strongly indicates schizophrenia.

The more I study Isaac Newton, the more fascinating he becomes. In my opinion, the genius of the precocious boy from Woolsthorpe has never been, nor ever will be, surpassed.

This article is adapted from a version originally published on RealClearScience.

View original post here:

Who was the smartest person in the world? - Big Think

Any leap in quantum computing multiplies the potential of a technology capable of performing calculations and simulations that are beyond the scope of current computers while facilitating the study of phenomena that have been only theoretical to date.

Last year, a group of researchers put forward the idea in the journal Nature that an alternative to quantum theory based on real numbers can be experimentally falsified. The original proposal was a challenge that has been taken up by the leading scientist in the field, Jian-Wei Pan, with the participation of physicist Adn Cabello, from the University of Seville. Their combined research has demonstrated the indispensable role of complex numbers [square root of minus one, for example] in standard quantum mechanics. The results allow progress to be made in the development of computers that use this technology and, according to Cabello, to test quantum physics in regions that have previously been inaccessible.

Jian-Wei Pan, 51, a 1987 graduate of the Science and Technology University of China (USTC) and a PhD graduate of Vienna University, leads one of the largest and most successful quantum research teams in the world, and has been described by physics Nobel laureate Frank Wilczek as a force of nature. Jian-Wei Pans thesis supervisor at the University of Vienna, physicist Anton Zeilinger, added: I cannot imagine the emergence of quantum technology without Jian-Wei Pan.

Pans leadership in the research has been fundamental. The experiment can be seen as a game between two players: real-valued quantum mechanics versus complex-valued quantum mechanics, he explains. The game is played on a quantum computer platform with four superconducting circuits. By sending in random measurement bases and measuring the outcome, the game score is obtained which is a mathematical combination of the measurement bases and outcome. The rule of the game is that the real-valued quantum mechanics is ruled out if the game score exceeds 7.66, which is the case in our work.

Covered by the scientific journal Physical Review Letters, the experiment was developed by a team from USTC and the University of Seville to answer a fundamental question: Are complex numbers really necessary for the quantum mechanical description of nature? The results exclude an alternative to standard quantum physics that uses only real numbers.

According to Jian-Wei Pan: Physicists use mathematics to describe nature. In classical physics, a real number appears complete to describe the physical reality in all classical phenomenon, whereas a complex number is only sometimes employed as a convenient mathematical tool. However, whether the complex number is necessary to represent the theory of quantum mechanics is still an open question. Our results disprove the real-number description of nature and establish the indispensable role of a complex number in quantum mechanics.

Its not only of interest regarding excluding a specific alternative, Cabello adds, the importance of the experiment is that it shows how a system of superconducting qubits [those used in quantum computers] allows us to test predictions of quantum physics that are impossible to test with the experiments we have been carrying out until now. This opens up a very interesting range of possibilities, because there are dozens of fascinating predictions that we have never been able to test, since they require firm control over several qubits. Now we will be able to test them.

According to Chao-Yang Lu, of USTC and co-author of the experiment: The most promising near-term application of quantum computers is the testing of quantum mechanics itself and the study of many-body systems.

Thus, the discovery provides not only a way forward in the development of quantum computers, but also a new way of approaching nature to understand the behavior and interactions of particles at the atomic and subatomic level.

But, like any breakthrough, the opening of a new way forward generates uncertainties. However, Jian-Wei Pan prefers to focus on the positive: Building a practically useful fault-tolerant quantum computer is one of the great challenges for human beings, he says. I am more concerned about how and when we will build one. The most formidable challenge for building a large-scale universal quantum computer is the presence of noise and imperfections. We need to use quantum error correction and fault-tolerant operations to overcome the noise and scale up the system. A logical qubit with higher fidelity than a physical qubit will be the next breakthrough in quantum computing and will occur in about five years. In homes, quantum computers would, if realized, be available first through cloud services.

According to Cabello, when quantum computers are sufficiently large and have thousands or millions of qubits, they will make it possible to understand complex chemical reactions that will help to design new drugs and better batteries; perform simulations that lead to the development of new materials and calculations that make it possible to optimize artificial intelligence and machine learning algorithms used in logistics, cybersecurity and finance, or to decipher the codes on which the security of current communications is based.

Quantum computers, he adds, use the properties of quantum physics to perform calculations. Unlike the computers we use, in which the basic unit of information is the bit [which can take two values], in a quantum computer, the basic unit is the quantum bit, or qubit, which has an infinite number of states.

Cabello goes on to say that the quantum computers built by companies such as Google, IBM or Rigetti take advantage of the fact that objects the size of a micron and produced using standard semiconductor-manufacturing techniques can behave like qubits.

The goal of having computers with millions of qubits is still a long way off, since most current quantum computers, according to Cabello, only have a few qubits and not all of them are good enough. However, the results of the Chinese and Spanish teams research make it possible to expand the uses of existing computers and to understand physical phenomena that have puzzled scientists for years.

For example, Google Quantum AI has published the observation of a time crystal through the Sycamore quantum processor for the first time in the Nature journal. A quantum time crystal is similar to a grain of salt composed of sodium and chlorine atoms. However, while the layers of atoms in that grain of salt form a physical structure based on repeating patterns in space, in the time crystal the structure is configured from an oscillating pattern. The Google processor has been able to observe these oscillatory wave patterns of stable time crystals.

This finding, according to Pedram Roushan and Kostyantyn Kechedzhi, shows how quantum processors can be used to study new physical phenomena. Moving from theory to actual observation is a critical leap and is the basis of any scientific discovery. Research like this opens the door to many more experiments, not only in physics, but hopefully inspires future quantum applications in many other fields.

In Spain, a consortium of seven companies Amatech, BBVA bank, DAS Photonics, GMV, Multiverse computing, Qilimanjaro Quantum Tech and Repsol and five research centers Barcelona Supercomputing Center (BSC), Spanish National Research Council (CSIC), Donostia International Physics Center (DIPC), The Institute of Photonic Sciences (ICFO), Tecnalia and the Polytechnic University of Valencia (UPV) have launched a new project called CUCO to apply quantum computing to Spanish strategic industries: energy, finance, space, defense and logistics.

Subsidized by the Center for the Development of Industrial Technology (CDTI) and with the support of the Ministry of Science and Innovation, the CUCO project, is the first major quantum computing initiative in Spain in the business field and aims to advance the scientific and technological knowledge of quantum computing algorithms through public-private collaboration between companies, research centers and universities. The goal is for this technology to be implemented in the medium-term future.

English version by Heather Galloway.

Read more from the original source:

Jian-Wei Pan: The next quantum breakthrough will happen in five years - EL PAS in English

By Amanda Gefter

Mary Iverson

IMAGINE approaching a Renaissance sculpture in a gallery. Even from a distance, it looks impressive. But it is only as you get close and walk around it that you begin to truly appreciate its quality: the angle of the jaw, the aquiline nose, the softness of the hair rendered in marble.

In physics, as in life, it is important to view things from more than one perspective. As we have done that over the past century, we have had plenty of surprises. It started with Albert Einsteins theory of special relativity, which showed us that lengths of space and durations of time vary depending on who is looking. It also painted a wholly unexpected picture of the shared reality underneath one in which space and time were melded together in a four-dimensional union known as space-time.

When quantum theory arrived a few years later, things got even weirder. It seemed to show that by measuring things, we play a part in determining their properties. But in the quantum world, unlike with relativity, there has never been a way to reconcile different perspectives and glimpse the objective reality beneath. A century later, many physicists question whether a single objective reality, shared by all observers, exists at all.

Now, two emerging sets of ideas are changing this story. For the first time, we can jump from one quantum perspective to another. This is already helping us solve tricky practical problems with high-speed communications. It also sheds light on whether any shared reality exists at the quantum level. Intriguingly, the answer seems to be no until we start talking to each other.

When Einstein developed his theory of relativity in the early 20th century, he worked from one fundamental assumption: the laws of physics should be the same for everyone. The trouble was, the laws of electromagnetism demand that light always travels at 299,792 kilometres per second and Einstein realised this creates a problem. If you were to race alongside a light beam in a spaceship, you would expect to see the beam moving far slower than usual just as neighbouring cars dont look to be going so fast when you are zipping along the motorway. Yet if that was the case, the laws of physics in that perspective would be violated.

In the quantum world, there has never been a way to reconcile different perspectives and glimpse the shared reality beneath

Einstein was convinced that couldnt happen, so he was forced to propose that the speed of light is constant for everyone, regardless of how fast they are moving. To compensate, space and time themselves had to change from one perspective to the next. The equations of relativity allowed him to translate from one observers perspective, or reference frame, to another, and in doing so build a picture of the shared world that remains the same from all perspectives.

He went on to develop these ideas into general relativity, which remains our best theory of gravity. But it isnt the whole story. In Einsteins writings, reference frames are always defined by rods and clocks, physical objects against which space and time are measured. These objects are, however, governed by a different theory altogether.

Quantum theory deals with matter and energy and is even more successful than relativity. But it paints a deeply unfamiliar picture of reality, one in which particles dont have definite properties before we measure them, but exist in a superposition of multiple states. It also shows that particles can become entangled, their properties intimately linked even over vast distances. All this puts the definition of a reference frame on shaky ground. How do you measure time with a clock that is entangled, or distance with a ruler that is in multiple places at once?

How do you measure time with a clock that is entangled, or distance with a ruler that is in multiple places at once?

Quantum physicists usually avoid this question by treating measuring instruments as if they obey the classical laws of mechanics developed by Isaac Newton. The particle being measured is quantum; the reference frame isnt. The dividing line between the two is known as the Heisenberg cut. It is arbitrary and it is moveable, but it has to be there so that the measuring device can record a definite result.

Consider Schrdingers cat, the thought experiment in which an unfortunate feline is in a box with a radioactive particle. If the particle decays, it triggers a hammer that breaks a vial that releases a poison that kills the cat. If it doesnt, the cat lives. You are outside the box. From your perspective, the contents are entangled and in a superposition. The particle both has and hasnt decayed; the cat is both dead and alive. But, as in relativity, shouldnt it be possible to describe the situation from the perspective of the cat?

This conundrum has long bothered aslav Brukner at the Institute for Quantum Optics and Quantum Information in Vienna, Austria. He wanted to understand how to see things from multiple points of view in quantum theory. Following Einsteins lead, he started from the assumption that the laws of physics must be the same for everyone, and then developed a way to mathematically switch between quantum reference frames. If we could describe a situation from either side of the Heisenberg cut, Brukner suspected that some truth about a shared quantum world might emerge.

What Brukner and his colleagues found in 2019 was a surprise. When you jump into the cats point of view, it turns out that just as in relativity things have to warp to preserve the laws of physics. The quantumness previously attributed to the cat gets shuffled across the Heisenberg cut. From this perspective, the cat is in a definite state it is the observer outside the box who is in a superposition, entangled with the lab outside. Entanglement was long thought to be an absolute property of reality. But in this new picture, it is all a matter of perspective. What is quantum and what is classical depends on the choice of quantum reference frames, says Brukner.

Jacques Pienaar at the University of Massachusetts says all this allows us to rigorously pose some fascinating questions. Take the well-known double-slit experiment, which showed that a quantum particle can travel through two slits in a grating at once. We see that, relative to the electron, it is the slits themselves that are in a superposition, says Pienaar. To me, thats just wonderful. While that might all sound like mere theorising, one thing that gives Brukners ideas credence is that they have already helped solve an intractable problem relating to quantum communication (see Flying qubits).

Quantum reference frames do have an Achilles heel though, albeit one that might ultimately point us to a deeper appreciation of reality. It comes in the form of Wigners friend, a thought experiment dreamed up in the 1950s by physicist Eugene Wigner. It adds a mind-bending twist to Schrdingers puzzle.

Faced with the usual set-up, Wigners friend opens the box and finds, say, that the cat is alive. But what if Wigner himself stands outside the lab door? In his reference frame, the cat is still in a superposition of alive and dead, only now it is entangled with the friend, who is in a superposition of having-seen-an-alive-cat and having-seen-a-dead-cat. Wigners description of the cat and the friends description of it are mutually exclusive, but according to quantum theory they are both right. It is a deep paradox that seems to reveal a splintered reality.

Brukners rules are no help here. We cant hop from one side of the Heisenberg cut to the other because the two people are using different cuts. The friend has the cut between herself and the box; Wigner has it between himself and the lab. They arent staring at each other from across the classical-quantum divide. They arent looking at one another at all. My colleagues and I were hoping that the Wigners friend situation could be rephrased in quantum reference frames, says Brukner. But so far, that hasnt been possible. I dont know, he sighs. Theres a missing element.

Suhaimi Abdullah/Getty Images

Hints as to what that might be are coming from work by Flavio Mercati at the University of Burgos in Spain and Giovanni Amelino-Camelia at the University of Naples Federico II in Italy. Their research seems to suggest that by exchanging quantum information, observers can create a shared reality, even if it isnt there from the start.

The duo were inspired by research carried out in 2016 by Markus Mller and Philipp Hhn, both then at the Perimeter Institute in Waterloo, Canada, who imagined a scenario in which two people, Alice and Bob, send each other quantum particles in a particular state of spin. Spin is a quantum property that can be likened to an arrow that can point up or down along each of the three spatial axes. Alice sends Bob a particle and Bob has to figure out its spin; then Bob prepares a new particle with the same spin and sends it back to Alice, who confirms that he got it right. The twist is that Alice and Bob dont know the relative orientation of their reference frames: ones x-axis could be the others y-axis.

Alice and Bobs communication may forge the structure of space-time

If Alice sends Bob just one particle, he will never be able to decode the spin. Sometimes in physics, two variables are connected in such a way that if you measure one precisely, the other no longer exists in a definite state. This tricky problem, known as the Heisenberg uncertainty principle, applies to particles spin along different axes. So if Bob wants to measure spin along what he thinks is Alices x-axis, he has to take a wild guess as to which axis that really is if he is wrong, he erases all the information. The pair can get around this, however, if they exchange lots of particles. Alice can tell Bob, Im sending you 100 particles that are all spin up along the x-axis. As Bob measures more and more of them, he can begin to work out the relative orientation of their reference frames.

Here is where it gets interesting. Mller and Hhn realised that, in doing all this, Alice and Bob automatically derive the equations that enable you to translate the view from one perspective to another in Einsteins special relativity. We tend to think of space-time as the pre-existing structure through which observers communicate. But Mller and Hhn flipped the story. Start with observers sending messages, and you can derive space-time.

For Mercati and Amelino-Camelia, who first came across the work a few years ago, that flip was a light-bulb moment. It raised a key question that turns out to have a crucial bearing on Brukners work: are Alice and Bob learning about a pre-existing space-time or is the space-time emerging as they communicate?

There are two ways in which the latter could play out. The first has to do with the trade-off in quantum mechanics between information and energy. To gain information about a quantum system you have to pay energy, says Mercati. Every time Bob chooses the correct axis, he loses a bit of energy; when he chooses wrong and erases Alices information, he gains some. Because the curvature of space-time depends on the energy present, when Bob measures his relative orientation he also ends up changing the orientation a tiny bit.

There could be a more profound sense in which quantum communication creates space-time. This comes into play if space is whats called non-commutative. If you want to arrive at a point on a normal map, it doesnt matter in which order you specify the coordinates. You can go over five and up two; or up two and over five either way you will land on the same spot. But if the laws of quantum mechanics apply to space-time itself, this might not be true. In the same way that knowing a particles position prevents you from measuring its momentum, going over five might prevent you from going up two.

Mercati and Amelino-Camelia say that if space-time does work in this way, Alice and Bobs attempts to find out their relative orientation wouldnt merely uncover the structure of space-time, they would actively forge it. The choices they make as to which axes to measure would alter the very thing their communication was meant to reveal. The pair have also devised a way to test whether this is really the case (see Does space-time commute?).

All this work points towards a startling conclusion: that as people exchange quantum information, they are collaborating to construct their mutual reality. It means that if we simply look at space and time from one perspective, not only do we miss its full beauty, but there might not be any deeper shared reality. For Mercati and Amelino-Camelia, one observer does not a space-time make.

That leads us back to the Wigners friend paradox that flummoxed Brukner. In his work, observers can be treated as having perspectives on the same reality only when they are gazing at one another from across the Heisenberg cut. Or, put another way, only when it is possible for them to communicate, which is precisely what Wigner and his friend cant do. Perhaps this is telling us that until two people interact, they dont share the same reality because it is communication itself that creates it.

Networks of cables that carry quantum information are already being set up around the world as a prototype quantum internet. These networks transport information in the form of qubits, or quantum bits, which can be encoded in the properties of particles typically in a quantum property called spin. One person sends a stream of particles to another, who then measures their spin to decode the message.

Except, not so fast. To be a useful means of communication, these particles must travel at close to the speed of light. At such speeds, a particles spin gets quantum entangled with its momentum in such a way that if the receiver only measures the spin, information will be lost. This is serious, says Flaminia Giacomini at the Perimeter Institute in Canada. The qubit is the basis for quantum information, but for a particle moving at very high velocities, we can no longer identify a qubit. As if that werent enough of a problem, each qubit doesnt move at one definite speed: thanks to quantum mechanics, it is in what is known as a superposition of velocities.

The rules of quantum reference frames developed by aslav Brukner (see main story) could be the answer. Giacomini has shown how the rules can be used to jump into the particles reference frame, even when the particle is in a superposition. From that perspective, it is the rest of reality that is whizzing past in a blurred superposition. Armed with knowledge of how the qubit sees the world, you can then determine the mathematical transformation to perform on the particle to recover the information in the original qubit.

In ordinary space, it isnt the journey that matters so much as the destination. If youre trying to arrive at a given place, it makes no difference whether you head 5 kilometres south and then 3 kilometres west, or vice-versa. That is because the coordinates commute; they get you to the same spot regardless of the order.

At very small scales to which quantum theory applies, this might not be true. In quantum theory, measuring a particles position erases information about its momentum. Similarly, it could be that the order in which movements are made could affect the structure of space. If this is so, it makes no sense to talk about space-time as a fixed arena.

Physicists Flavio Mercati and Giovanni Amelino-Camelia think they have a way to find out whether space-time commutes. They were inspired by research that imagined two people exchanging quantum particles and measuring their properties to deduce their relative orientation (see main story). What would happen, Mercati and Amelino-Camelia asked, if this game were played for real?

As the people exchange more and more particles, their uncertainty about their orientation should decrease. But will it ever get to zero? In ordinary space-time, it will. But if space-time is non-commutative, some uncertainty will always remain, since their orientation is ever so slightly rewritten with each measurement. The pair might have to exchange trillions of particles before we will have an answer but Mercati thinks it is worth a try.

More on these topics:

See the original post here:

Do we create space-time? A new perspective on the fabric of reality - New Scientist

DURHAM, N.C., Feb. 3, 2022 /PRNewswire/ -- February 3, 2022, collaboration agreement was signed between Allosteric Bioscience, a company founded in 2021 integrating Quantum Computing and Artificial Intelligence with Biomedical sciences to create improved treatments for Aging and Longevity and Polaris Quantum Biotech, a company at the vanguard of Quantum Computing for drug discovery. Together, they are utilizing advancements in Quantum Computing and Artificial Intelligence for development of novel pharmaceuticals.

Improved Aging, Longevity and Aging related diseases is a lead program at Allosteric Bioscience and the focus of this agreement, supported by an investment in Polarisqb. This joint program uses Quantum Computing (QC) and artificial intelligence (AI) for creation of an inhibitor of a key protein involved in Aging that could have benefits for health representing a multibillion-dollar market. Allosteric Bioscience is using its "QAB" platform for integrating QC, AI, genetics, genomics, system biology, epigenetics, and proteomics, as well as two Aging platforms: "ALT" - Aging Longevity Targets and "ALM" Aging Longevity Modulators.

Dr. Shahar Keinan, CEO of Polarisqb stated, "Quantum Computing technology is coming of age, allowing us to revolutionize drug discovery timelines, while improving the overall profile of the designed drugs. We are excited about the joint program with Allosteric tackling Aging and Longevity using Polarisqb's Tachyon platform. The application of Quantum Computers to solving these complex questions is extraordinary."

Dr. Arthur P. Bollon, President of Allosteric Bioscience stated, "The agreement between Allosteric Bioscience and Polarisqb represents an important milestone in implementing the Allosteric Bioscience strategy of integrating the Quantum Computer and advanced AI with Biomedical sciences for creation and development of advanced treatments for Improved Aging, Longevity and Aging related diseases."

Story continues

Polaris Quantum Biotech, a leader in Quantum Computing for drug discovery, created the first drug discovery platform built on a Quantum Computer. Founded in 2020 by Shahar Keinan, CEO, and Bill Shipman, CTO, Polarisqb uses cloud, quantum computing, and machine learning to process, evaluate and identify lead molecules 10,000 times faster than alternative solutions. These high-quality drug leads are taken to synthesis, testing, and licensed to partners for development within months, rather than years. Information is available at http://www.Polarisqb.com

Allosteric Bioscience founders, Bruce Meyers, Arthur P. Bollon, Ph.D., and Peter Sordillo, Ph.D., M.D., have decades of expertise in the biotechnology industry as well as biomedical disciplines including genomics, epigenetics, systems biology, proteomics as well as oncology and quantum physics. Bruce Meyers and Dr. Bollon, founded multiple biotechnology companies including Cytoclonal Pharmaceutics (Dr. Bollon served as Chairman and CEO) which merged to create OPKO Health, a NASDAQ company with a market cap of $2 billion. Dr. Sordillo, who has a background in quantum information theory, is a leader in treating sarcomas and other cancers and managed over 50 clinical trials at leading institutions including Sloan Kettering Cancer Center.

For information about Allosteric Bioscience: Dr. Arthur P. Bollon- abollon@allostericbioscience.com or Bruce Meyers- bmeyers@allostericbiocience.com

For information about Polarisqb: Dr. Shahar Keinan - skeinan@polarisqb.com or Will Simpson - wsimpson@polarisqb.com.

Cision

View original content:https://www.prnewswire.com/news-releases/quantum-computing-targets-improved-human-aging-and-longevity-in-new-agreement-between-allosteric-bioscience-and-polaris-quantum-biotech-301474412.html

SOURCE PolarisQB

Read the original here:

Quantum Computing Targets Improved Human Aging and Longevity in new Agreement between Allosteric Bioscience and Polaris Quantum Biotech - Yahoo...

Fundamental physics research in Finland has led to at least six very successful spin-offs that have supplied quantum technology to the global market for several decades.

According to Pertti Hakonen, an academic at Aalto University, it all started with Olli Viktor Lounasmaa, who in 1965 established the low-temperature laboratory at Aalto University, formerly Helsinki University of Technology. He served as lab director for about 30 years, says Pertti Hakonen, professor at Aalto University.

The low-temperature lab was a long-term investment in basic research in low-temperature physics that has paid off nicely. Hakonen, who has been conducting research in the lab since 1979, witnessed the birth and growth of several spin-offs, including Bluefors, a startup that is now by far the market leader in cryostats for quantum computers.

In the beginning, there was a lot of work on different cryostat designs, trying to beat low-temperature records, says Hakonen. Our present record in our lab is 100 pico-kelvin in the nuclei of rhodium atoms. Thats the nuclear spin temperature in the nuclei of rhodium atoms, not in the electrons.

For quantum computing you dont need temperatures this low. You only need 10 milli-kelvin. A dilution refrigerator is enough for that. In the old days, the cryostat had to be in a liquid helium bath. Bluefors was a pioneer in using liquid-free technology, replacing the liquid helium with a pulse tube cooler, which is cheaper in the long run. The resulting system is called a dry dilution refrigerator.

The pulse tube cooler is based on two stages in series. The first stage brings the temperature down to 70 kelvin and the next stage brings it down to 4 kelvin. Gas is pumped down and up continuously, passing through heat exchangers a process that drops the temperature dramatically.

Bluefors started business with the idea of adding closed-loop dilution refrigeration after pulse tube cooling. In 2005 and 2006, pulse tube coolers became more powerful, says David Gunnarsson, CTO at Bluefors. We used pulse tube coolers to pre-cool at the first two stages, which takes you down to around 3 kelvin. We get the pulse tube coolers from an American company called Cryomech.

Bluefors key differentiator is a closed-loop circulation system, the dilution refrigerator stages, where we circulate a mixture of helium 4 and helium 3 gas. At very cold temperatures, this becomes liquid, which we circulate through a series of well-designed heat exchangers. This approach can get the temperature down to below 10 milli-kelvin. This is where our specialty lies going below the 3 kelvin you get from off-the-shelf coolers.

Bluefors has more than 700 units on the market that are used for both research in publicly funded organisations, and for commercial research and development. One big market that has driven the dilution refrigeration is quantum computing. Anyone currently doing quantum computing based on superconducting qubits is most likely to have a Bluefors cryogenic system.

When a customer recognises the need for a cryogenic system, they talk to Bluefors to decide on the size of the refrigerator. This depends on the tasks they want to do and how many qubits they will use. Then they start looking at the control and measurement infrastructure, which must be tightly integrated with the cryogenic system. Some combination of different components and signalling elements might be added, depending on the frequencies being used. If the control and measurement lines are optical, then optical fibres are included.

As soon as Bluefors and the customer reach an agreement, Bluefors begins to produce the cryogenic enclosure, along with a unique set of options tailored to the use case. Bluefors then runs tests to make sure everything works together and that the enclosure reaches and maintains the temperatures required by the application.

The system has evolved since the company first started marketing its products in 2008. To cool down components with a dilution refrigerator, Bluefors uses a cascade approach, with nested structures that drop an order of magnitude in temperature at each level. The typical configuration includes five stages, with the first stage now bringing the temperature down to 50 kelvin. The temperature goes down to about 4 kelvin at the second stage, and reaches 1 kelvin at the third. It then drops to 100 milli-kelvin at the fourth stage, and at the fifth stage gets down to 10 milli-kelvin, or even below.

The enclosure can cool several qubits, depending on the power dissipation and the temperature the customer needs. A challenge here is that the more power dissipates, the higher the temperature is raised, and every interaction can increase the temperature.

Our most powerful model today can probably run a few hundred qubits in one enclosure, says Gunnarsson. IBM has just announced it has a system with 127 qubits. We can handle that many in one enclosure using the most powerful system we have today.

In most architectures, quantum programs work by sending microwave signals to the qubits. The sequence of signals constitutes a program. Then you have to read the outcome at the end.

The user typically has a microwave source at room temperature, says Gunnarsson. Usually, when it reaches the chips, its at power levels of the order of pico-watts, which is all that is needed to drive a qubit. Pico-watts are one trillionth of a watt a very small power requirement.

That is also a power that is very hard to read out at room temperature. So to read the output from a chip, the signal has to be amplified and taken back up to room temperature. A cascade of amplification is required to get the signal to the level you need.

The microwave control signals and the read-out process at the end constitute a cycle that lasts about 100 nanoseconds. Several such cycles occur per second, collectively making up a quantum program.

Another challenge for quantum computing is to get electronics inside the refrigerators. All operations are performed at very low temperatures, but then the result has to be taken up to room temperature to be read out. Wires are needed to start a program and to read results. The problem is that electrical wires generate heat.

This means that quantum computing lends itself only to programs where the results are not read out until the end one of many reasons interactive application such as Microsoft Excel will never be appropriate for the quantum paradigm.

It also means that every qubit needs at least one control line and then one readout line. Multiplexing can be used to reduce the number of readout lines, but there is still a lot of wiring per qubit. The chips themselves are not that large what takes up most space are all the wires and accompanying components. This makes it challenging to scale up refrigeration systems.

Since Bluefors supplies the cryogenic measurement infrastructure, we developed something we call a high-density solution, where we made it possible to have a six-fold increase in the amount of signal lines you can have in our system, says Gunnarsson. Now you can have up to 1,000 signal lines in a Bluefors state-of-the-art system using our current form factor.

One very recent innovation from Bluefors is a modular concept for cryostats, which is used by IBM. The idea is to combine modules and have information exchanged between them. This modular concept is going to be an interesting development, says Aalto Universitys Hakonen, who since the 1970s has enjoyed a front-row view of the development of quantum technology in Finland.

Finland has a very strong tradition in quantum theory in general and specifically, the quantum physics used in superconducting qubits, which is the platform used by IBM and Google. Now a large area of active research is in quantum algorithms.

How one goes about making a program is a key question, says Sabrina Maniscalco, professor of quantum information and logic at the University of Helsinki. Nowadays, the situation is such that programming quantum computing is much more quantum theory-related than any software ever managed or developed. We are not yet at a stage where a programming language exists that is independent of the device on which it runs. At the moment, quantum computers are really physics experiments.

Finland has long been renowned worldwide for its work in theoretical quantum physics, an area of expertise that plays nicely into the industry growing up around quantum computing. Two other factors that contribute to the growing ecosystem in Finland are the willingness of the government to invest in blue-sky research and the famed Finnish education system, which provides an excellent workforce for startups.

The countrys rich ecosystem of research, stable political support and the education system have resulted in the birth and growth of many startups that develop quantum algorithms. This seems like quite an achievement for a country of only five million inhabitants. But in many ways, Finlands small population is an advantage, creating a tight-knit group of experts, some of whom wear several different hats.

Maniscalco is a case in point. In addition to her research into quantum algorithms at the University of Helsinki, she is also CEO of quantum software startup Algorithmiq, which is focused on developing quantum software for life sciences.

We are trying to make quantum computers more like standard computers, but its still at a very preliminary stage Sabrina Maniscalco, University of Helsinki

As a researcher, I am first of all a theorist, she says. I dont get involved in building hardware, but I have a group of several people developing software. Quantum software is as important as hardware nowadays because quantum computers work very differently from classical computers. Classical software doesnt work at all on quantum systems. You have to completely change the way you program computers if you want to use a quantum computer.

We are trying to make quantum computers more like standard computers, but its still at a very preliminary stage. To program a quantum computer, you need quantum physicists who work with computer scientists, and experts in the application domain for example, quantum chemists. You have to start by creating specific instructions that make sense in terms of the physics experiments that quantum computers are today.

Algorithm developers need to take into account the type of quantum computer they are using the two leading types are superconducting qubits and trapped ions. Then they have to look at the quality of the qubits. They also need to know something about quantum information theory, and about the noise and imperfections that affect the qubits the building blocks of quantum computers.

Conventional computers use error correction, says Maniscalco. Thanks to error correction, the results of the computations that are performed inside your laptop or any computer are reliable. Nothing similar currently exists with quantum computers. A lot of people are currently trying to develop a quantum version of these error correction schemes, but they dont exist yet. So you have to find other strategies to counter this noise and the resulting errors.

Overcoming the noisiness of the current generation of qubits is one of many challenges standing in the way of practical quantum computers. Once those barriers are lifted, the work Maniscalco and other researchers in Finland are doing on quantum algorithms will certainly have an impact around the world.

Read this article:

Finland brings cryostats and other cool things to quantum computing - ComputerWeekly.com