Search

In artificial neural networks, the activation dynamics of non-trainable variables is strongly coupled to the learning dynamics of trainable variables. During the activation pass, the boundary neurons (e.g., input neurons) are mapped to the bulk neurons (e.g., hidden neurons), and during the learning pass, both bulk and boundary neurons are mapped to changes in trainable variables (e.g., weights and biases). For example, in feed-forward neural networks, forward propagation is the activation pass and backward propagation is the learning pass. We show that a composition of the two maps establishes a duality map between a subspace of non-trainable boundary variables (e.g., dataset) and a tangent subspace of trainable variables (i.e., learning). In general, the dataset-learning duality is a complex non-linear map between high-dimensional spaces, but in a learning equilibrium, the problem can be linearized and reduced to many weakly coupled one-dimensional problems. We use the duality to study the emergence of criticality, or the power-law distributions of fluctuations of the trainable variables. In particular, we show that criticality can emerge in the learning system even from the dataset in a non-critical state, and that the power-law distribution can be modified by changing either the activation function or the loss function., Comment: 29 pages, 9 figures, 1 table, minor corrections

We develop a macroscopic description of the evolutionary dynamics by following the temporal dynamics of the total Shannon entropy of sequences, denoted by $S$, and the average Hamming distance between them, denoted by $H$. We argue that a biological system can persist in the so-called quasi-equilibrium state for an extended period, characterized by strong correlations between $S$ and $H$, before undergoing a phase transition to another quasi-equilibrium state. To demonstrate the results, we conducted a statistical analysis of SARS-CoV-2 data from the United Kingdom during the period between March, 2020 and December, 2023. From a purely theoretical perspective, this allows us to systematically study various types of phase transitions described by a discontinuous change in the thermodynamic parameters. From a more practical point of view, the analysis can be used, for example, as an early warning system for pandemics., Comment: 20 pages, 10 figures, 2 tables, accepted for publication in Entropy

Consider a reinforcement learning problem where an agent has access to a very large amount of information about the environment, but it can only take very few actions to accomplish its task and to maximize its reward. Evidently, the main problem for the agent is to learn a map from a very high-dimensional space (which represents its environment) to a very low-dimensional space (which represents its actions). The high-to-low dimensional map implies that most of the information about the environment is irrelevant for the actions to be taken, and only a small fraction of information is relevant. In this paper we argue that the relevant information need not be learned by brute force (which is the standard approach), but can be identified from the intrinsic symmetries of the system. We analyze in details a reinforcement learning problem of autonomous driving, where the corresponding symmetry is the Galilean symmetry, and argue that the learning task can be accomplished with very few relevant parameters, or, more precisely, invariants. For a numerical demonstration, we show that the autonomous vehicles (which we call autonomous particles since they describe very primitive vehicles) need only four relevant invariants to learn how to drive very well without colliding with other particles. The simple model can be easily generalized to include different types of particles (e.g. for cars, for pedestrians, for buildings, for road signs, etc.) with different types of relevant invariants describing interactions between them. We also argue that there must exist a field theory description of the learning system where autonomous particles would be described by fermionic degrees of freedom and interactions mediated by the relevant invariants would be described by bosonic degrees of freedom., Comment: 15 pages, 3 figures

We analyze algorithmic and computational aspects of biological phenomena, such as replication and programmed death, in the context of machine learning. We use two different measures of neuron efficiency to develop machine learning algorithms for adding neurons to the system (i.e. replication algorithm) and removing neurons from the system (i.e. programmed death algorithm). We argue that the programmed death algorithm can be used for compression of neural networks and the replication algorithm can be used for improving performance of the already trained neural networks. We also show that a combined algorithm of programmed death and replication can improve the learning efficiency of arbitrary machine learning systems. The computational advantages of the bio-inspired algorithms are demonstrated by training feedforward neural networks on the MNIST dataset of handwritten images., Comment: 17 pages, 5 figures

Neural network is a dynamical system described by two different types of degrees of freedom: fast-changing non-trainable variables (e.g. state of neurons) and slow-changing trainable variables (e.g. weights and biases). We show that the non-equilibrium dynamics of trainable variables can be described by the Madelung equations, if the number of neurons is fixed, and by the Schrodinger equation, if the learning system is capable of adjusting its own parameters such as the number of neurons, step size and mini-batch size. We argue that the Lorentz symmetries and curved space-time can emerge from the interplay between stochastic entropy production and entropy destruction due to learning. We show that the non-equilibrium dynamics of non-trainable variables can be described by the geodesic equation (in the emergent space-time) for localized states of neurons, and by the Einstein equations (with cosmological constant) for the entire network. We conclude that the quantum description of trainable variables and the gravitational description of non-trainable variables are dual in the sense that they provide alternative macroscopic descriptions of the same learning system, defined microscopically as a neural network., Comment: 20 pages, minor corrections

We outline a phenomenological theory of evolution and origin of life by combining the formalism of classical thermodynamics with a statistical description of learning. The maximum entropy principle constrained by the requirement for minimization of the loss function is employed to derive a canonical ensemble of organisms (population), the corresponding partition function (macroscopic counterpart of fitness) and free energy (macroscopic counterpart of additive fitness). We further define the biological counterparts of temperature (biological temperature) as the measure of stochasticity of the evolutionary process and of chemical potential (evolutionary potential) as the amount of evolutionary work required to add a new trainable variable (such as an additional gene) to the evolving system. We then develop a phenomenological approach to the description of evolution, which involves modeling the grand potential as a function of the biological temperature and evolutionary potential. We demonstrate how this phenomenological approach can be used to study the "ideal mutation" model of evolution and its generalizations. Finally, we show that, within this thermodynamics framework, major transitions in evolution, such as the transition from an ensemble of molecules to an ensemble of organisms, that is, the origin of life, can be modeled as a special case of bona fide physical phase transitions that are associated with the emergence of a new type of grand canonical ensemble and the corresponding new level of description, Comment: 23 pages

We apply the theory of learning to physically renormalizable systems in an attempt to develop 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. These principles also follow naturally from the theory of learning. We formulate 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 accompanying paper., Comment: 29 pages, 3 figures

We demonstrate, both analytically and numerically, that learning dynamics of neural networks is generically attracted towards a self-organized critical state. The effect can be modeled with quartic interactions between non-trainable variables (e.g. states of neurons) and trainable variables (e.g. weight matrix). Non-trainable variables are rapidly driven towards stochastic equilibrium and trainable variables are slowly driven towards learning equilibrium described by a scale-invariant distribution on a wide range of scales. Our results suggest that the scale invariance observed in many physical and biological systems might be due to some kind of learning dynamics and support the claim that the universe might be a neural network., Comment: 11 pages, 4 figures

It was recently shown that the Madelung equations, that is, a hydrodynamic form of the Schr\"odinger equation, can be derived from a canonical ensemble of neural networks where the quantum phase was identified with the free energy of hidden variables. We consider instead a grand canonical ensemble of neural networks, by allowing an exchange of neurons with an auxiliary subsystem, to show that the free energy must also be multivalued. By imposing the multivaluedness condition on the free energy we derive the Schr\"odinger equation with "Planck's constant" determined by the chemical potential of hidden variables. This shows that quantum mechanics provides a correct statistical description of the dynamics of the grand canonical ensemble of neural networks at the learning equilibrium. We also discuss implications of the results for machine learning, fundamental physics and, in a more speculative way, evolutionary biology., Comment: 18 pages, added a new section on quantum superpositions

We discuss a possibility that the entire universe on its most fundamental level is a neural network. We identify two different types of dynamical degrees of freedom: "trainable" variables (e.g. bias vector or weight matrix) and "hidden" variables (e.g. state vector of neurons). We first consider stochastic evolution of the trainable variables to argue that near equilibrium their dynamics is well approximated by Madelung equations (with free energy representing the phase) and further away from the equilibrium by Hamilton-Jacobi equations (with free energy representing the Hamilton's principal function). This shows that the trainable variables can indeed exhibit classical and quantum behaviors with the state vector of neurons representing the hidden variables. We then study stochastic evolution of the hidden variables by considering $D$ non-interacting subsystems with average state vectors, $\bar{\bf x}^{1}$, ..., $\bar{\bf x}^{D}$ and an overall average state vector $\bar{\bf x}^{0}$. In the limit when the weight matrix is a permutation matrix, the dynamics of $\bar{\bf x}^{\mu}$ can be described in terms of relativistic strings in an emergent $D+1$ dimensional Minkowski space-time. If the subsystems are minimally interacting, with interactions described by a metric tensor, then the emergent space-time becomes curved. We argue that the entropy production in such a system is a local function of the metric tensor which should be determined by the symmetries of the Onsager tensor. It turns out that a very simple and highly symmetric Onsager tensor leads to the entropy production described by the Einstein-Hilbert term. This shows that the learning dynamics of a neural network can indeed exhibit approximate behaviors described by both quantum mechanics and general relativity. We also discuss a possibility that the two descriptions are holographic duals of each other., Comment: 23 pages

We define a neural network as a septuple consisting of (1) a state vector, (2) an input projection, (3) an output projection, (4) a weight matrix, (5) a bias vector, (6) an activation map and (7) a loss function. We argue that the loss function can be imposed either on the boundary (i.e. input and/or output neurons) or in the bulk (i.e. hidden neurons) for both supervised and unsupervised systems. We apply the principle of maximum entropy to derive a canonical ensemble of the state vectors subject to a constraint imposed on the bulk loss function by a Lagrange multiplier (or an inverse temperature parameter). We show that in an equilibrium the canonical partition function must be a product of two factors: a function of the temperature and a function of the bias vector and weight matrix. Consequently, the total Shannon entropy consists of two terms which represent respectively a thermodynamic entropy and a complexity of the neural network. We derive the first and second laws of learning: during learning the total entropy must decrease until the system reaches an equilibrium (i.e. the second law), and the increment in the loss function must be proportional to the increment in the thermodynamic entropy plus the increment in the complexity (i.e. the first law). We calculate the entropy destruction to show that the efficiency of learning is given by the Laplacian of the total free energy which is to be maximized in an optimal neural architecture, and explain why the optimization condition is better satisfied in a deep network with a large number of hidden layers. The key properties of the model are verified numerically by training a supervised feedforward neural network using the method of stochastic gradient descent. We also discuss a possibility that the entire universe on its most fundamental level is a neural network., Comment: 32 pages, 6 figures, accepted for publication in Machine Learning: Science and Technology

We develop a non-perturbative method for calculating partition functions of strongly coupled quantum mechanical systems with interactions between subsystems described by a path integral of a dual system. The dual path integral is derived starting from non-interacting subsystems at zeroth order and then by introducing couplings of increasing complexity at each order of an iterative procedure. These orders of interactions play the role of a dual time and the full quantum partition function is expressed as a transition amplitude in the dual system. More precisely, it is expressed as a path integral from a deformation-operators dependent initial state at zero time/order to the inverse-temperature dependent final state at later time/order. We provide three examples of strongly coupled systems with first-order, second-order and higher-order interactions and discuss a possible emergence of space-time, quantum field theories and general relativity in context of the dual path integral., Comment: 20 pages, 4 figures, discussion of Ising model added, accepted for publication in European Physical Journal C

We consider finite quantum systems defined by a mixed set of commutation and anti-commutation relations between components of the Hamiltonian operator. These relations are represented by an anti-commutativity graph which contains a necessary and sufficient information for computing the full quantum partition function. We derive a second-order differential equation for an extended partition function $z[\beta, \beta', J]$ which describes a transformation from a ``parent'' partition function $z[0, \beta', J]$ (or anti-commutativity graph) to a ``child'' partition function $z[\beta, 0, J]$ (or anti-commutativity graph). The procedure can be iterated and then one forms a pseudo-forest of duality transformations between quantum systems, i.e. a directed graph in which every vertex (or quantum system) has at most one incoming edge (from its parent system). The pseudo-forest has a single tree connected to a constant partition function, many pseudo-trees connected to self-dual systems and all other pseudo-trees connected to closed cycles of transformations between mutually dual systems. We also show how the differential equation for the extended partition function can be used to study disordered systems., Comment: 18 pages, 7 figures, accepted for publication in the Journal of Mathematical Physics

We consider the quantum partition function for a system of quantum spinors and then derive an equivalent (or dual) classical partition function for some scalar degrees of freedom. The coupling between scalars is non-trivial (e.g. a model on 2-sphere configuration space), but the locality structure of the dual system is preserved, in contrast to the imaginary time formalism. We also show that the measure of integration in the classical partition function can be formally expressed through relativistic Green's functions which suggests a possible mechanism for the emergence of a classical space-time from anti-commutativity of quantum operators., Comment: 17 pages, typos corrected

We consider a stochastic process which is (a) described by a continuous-time Markov chain on only short time-scales and (b) constrained to conserve a number of hidden quantities on long time-scales. We assume that the transition matrix of the Markov chain is given and the conserved quantities are known to exist, but not explicitly given. To study the stochastic dynamics we propose to use the principle of stationary entropy production. Then the problem can be transformed into a variational problem for a suitably defined action and with time-dependent Lagrange multipliers. We show that the stochastic dynamics can be described by a Schrodinger equation, with Lagrange multipliers playing the role of phases, whenever (a) the transition matrix is symmetric or the detailed balance condition is satisfied, (b) the system is not too far from the equilibrium and (c) the number of the conserved quantities is large., Comment: 13 pages, 3 figures, references added

Informational dependence between statistical or quantum subsystems can be described with Fisher matrix or Fubini-Study metric obtained from variations of the sample/configuration space coordinates. Using these non-covariant objects as macroscopic constraints we consider statistical ensembles over the space of classical probability distributions or quantum wave-functions. The ensembles are covariantized using dual field theories with either complex or real scalar fields identified with complex wave-functions or square root of probabilities. We argue that a full space-time covariance on a field theory side is dual to local computations on the information theory side. We define a fully covariant information-computation tensor and show that it must satisfy conservation equations. Then we switch to a thermodynamic description and argue that the (inverse of) space-time metric tensor is a conjugate thermodynamic variable to the ensemble-averaged information-computation tensor. In the equilibrium the entropy production vanishes and the metric is not dynamical, but away from equilibrium the entropy production gives rise to an emergent dynamics of the metric. This dynamics can be described by expanding the entropy production into products of generalized forces (derivatives of metric) and conjugate fluxes. Near equilibrium these fluxes are given by an Onsager tensor contracted with generalized forces and on the grounds of time-reversal symmetry the Onsager tensor is expected to be symmetric. We show that a particularly simple and highly symmetric form of the Onsager tensor gives rise to the Einstein-Hilbert term. This proves that general relativity is equivalent to a theory of non-equilibrium (thermo)dynamics of the metric which is expected to break down far away from equilibrium where the symmetries of the Onsager tensor are to be broken., Comment: 25 pages, 1 figure, 1 table, minor errors corrected, accepted for publication in Int. J. Mod. Phys. A

Given a quantum (or statistical) system with a very large number of degrees of freedom and a preferred tensor product factorization of the Hilbert space (or of a space of distributions) we describe how it can be approximated with a very low-dimensional field theory with geometric degrees of freedom. The geometric approximation procedure consists of three steps. The first step is to construct weighted graphs (we call information graphs) with vertices representing subsystems (e.g. qubits or random variables) and edges representing mutual information (or the flow of information) between subsystems. The second step is to deform the adjacency matrices of the information graphs to that of a (locally) low-dimensional lattice using the graph flow equations introduced in the paper. (Note that the graph flow produces very sparse adjacency matrices and thus might also be used, for example, in machine learning or network science where the task of graph sparsification is of a central importance.) The third step is to define an emergent metric and to derive an effective description of the metric and possibly other degrees of freedom. To illustrate the procedure we analyze (numerically and analytically) two information graph flows with geometric attractors (towards locally one- and two-dimensional lattices) and metric perturbations obeying a geometric flow equation. Our analysis also suggests a possible approach to (a non-perturbative) quantum gravity in which the geometry (a secondary object) emerges directly from a quantum state (a primary object) due to the flow of the information graphs., Comment: 15 pages, 3 figures, 1 table, 1 chart, accepted for publication in Foundations of Physics

We find two families of analytic solutions to the ideal magnetohydrodynamics (iMHD) equations, in a class of 4-dimensional (4D) curved spacetimes. The plasma current is null, and as a result, the stress-energy tensor of the plasma itself can be chosen to take a cosmological-constant-like form. Despite the presence of a plasma, the force-free condition - where the electromagnetic current is orthogonal to the Maxwell tensor - continues to be maintained. Moreover, a special case of one of these two families leads us to a fully self-consistent solution to the Einstein-iMHD equations: we obtain the Vaidya-(anti-)de Sitter metric sourced by the plasma and a null electromagnetic stress tensor. We also provide a Mathematica code that researchers may use to readily verify analytic solutions to these iMHD equations in any curved 4D geometry., Comment: 12 pages. The Mathematica package TensoriaCalc used in this paper, as well as its accompanying MHDSystem Guide, can be found at http://www.stargazing.net/yizen/MHD.html

Given two quantum states of N q-bits we are interested to find the shortest quantum circuit consisting of only one- and two- q-bit gates that would transfer one state into another. We call it the quantum maze problem for the reasons described in the paper. We argue that in a large N limit the quantum maze problem is equivalent to the problem of finding a semiclassical trajectory of some lattice field theory (the dual theory) on an N+1 dimensional space-time with geometrically flat, but topologically compact spatial slices. The spatial fundamental domain is an N dimensional hyper-rhombohedron, and the temporal direction describes transitions from an arbitrary initial state to an arbitrary target state. We first consider a complex Klein-Gordon field theory and argue that it can only be used to study the shortest quantum circuits which do not involve generators composed of tensor products of multiple Pauli Z matrices. Since such situation is not generic we call it the Z-problem. On the dual field theory side the Z-problem corresponds to massless excitations of the phase (Goldstone modes) that we attempt to fix using Higgs mechanism. The simplest dual theory which does not suffer from the massless excitation (or from the Z-problem) is the Abelian-Higgs model which we argue can be used for finding the shortest quantum circuits. Since every trajectory of the field theory is mapped directly to a quantum circuit, the shortest quantum circuits are identified with semiclassical trajectories. We also discuss the complexity of an actual algorithm that uses a dual theory prospective for solving the quantum maze problem and compare it with a geometric approach. We argue that it might be possible to solve the problem in sub-exponential time in 2^N, but for that we must consider the Klein-Gordon theory on curved spatial geometry and/or more complicated (than N-torus) topology., Comment: 19 pages, 1 figure

We initiate a formal study of logical inferences in context of the measure problem in cosmology or what we call cosmic logic. We describe a simple computational model of cosmic logic suitable for analysis of, for example, discretized cosmological systems. The construction is based on a particular model of computation, developed by Alan Turing, with cosmic observers (CO), cosmic measures (CM) and cosmic symmetries (CS) described by Turing machines. CO machines always start with a blank tape and CM machines take CO's Turing number (also known as description number or G{\" o}del number) as input and output the corresponding probability. Similarly, CS machines take CO's Turing number as input, but output either one if the CO machines are in the same equivalence class or zero otherwise. We argue that CS machines are more fundamental than CM machines and, thus, should be used as building blocks in constructing CM machines. We prove the non-computability of a CS machine which discriminates between two classes of CO machines: mortal that halts in finite time and immortal that runs forever. In context of eternal inflation this result implies that it is impossible to construct CM machines to compute probabilities on the set of all CO machines using cut-off prescriptions. The cut-off measures can still be used if the set is reduced to include only machines which halt after a finite and predetermined number of steps., Comment: 13 pages, 2 figures, accepted in JCAP

We consider generating functionals for computing correlators in quantum field theories with random potentials. Examples of such theories include condensed matter systems with quenched disorder (e.g. spin glass) or cosmological systems in context of the string theory landscape (e.g. cosmic inflation). We use the so-called replica trick to define two different generating functionals for calculating correlators of the quantum fields averaged over a given distribution of random potentials. The first generating functional is appropriate for calculating averaged (in-out) amplitudes and involves a single replica of fields, but the replica limit is taken to an (unphysical) negative one number of fields outside of the path integral. When the number of replicas is doubled the generating functional can also be used for calculating averaged probabilities (squared amplitudes) using the in-in construction. The second generating functional involves an infinite number of replicas, but can be used for calculating both in-out and in-in correlators and the replica limits are taken to only a zero number of fields. We discuss the formalism in details for a single real scalar field, but the generalization to more fields or to different types of fields is straightforward. We work out three examples: one where the mass of scalar field is treated as a random variable and two where the functional form of interactions is random, one described by a Gaussian random field and the other by a Euclidean action in the field configuration space., Comment: 21 pages. Made some changes in Introduction, Sec 5 and discussion. An appendix added. Accepted in JHEP

We develop a field theory description of non-dissipative string fluids and construct an explicit mapping between field theory degrees of freedom and hydrodynamic variables. The theory generalizes both a perfect particle fluid and pressureless string fluid to what we call a perfect string fluid. Ideal magnetohydrodynamics is shown to be an example of the perfect string fluid whose equations of motion can be obtained from a particular choice of the Lagrangian. The Lagrangian framework suggests a straightforward extension of the perfect string fluid to more general anisotropic fluids describing higher dimensional branes such as domain walls. Other modifications of the Lagrangian are discussed which may be useful in describing relativistic superfluids and fluids containing additional currents., Comment: 23 pages. v2: Expanded discussion. The variational principle is discussed for both magnetohydrodynamics and a model of relativistic superfluids

We discuss eternal inflation in context of classical probability spaces defined by a triplet: sample space, $\sigma$-algebra and probability measure. We show that the measure problem is caused by the countable additivity axiom applied to the maximal $\sigma$-algebra of countably infinite sample spaces. This is a serious problem if the bulk space-time is treated as a sample space which is thought to be effectively countably infinite due to local quantum uncertainties. However, in semiclassical description of eternal inflation the physical space expands exponentially which makes the sample space of infinite trajectories uncountable and the (future) boundary space effectively continuous. Then the measure problem can be solved by defining a probability measure on the continuum of trajectories or holographically on the future boundary. We argue that the probability measure which is invariant under the symmetries of the tree-like structure of eternal inflation can be generated from the Lebesgue measure on unit interval. According to Vitali theorem the Lebesgue measure leaves some sets without a measure which means that there are certain probabilistic questions in eternal inflation that cannot be answered., Comment: 12 pages

We study the solutions of string fluid equations under assumption of a local equilibrium which was previously obtained in the context of the kinetic theory. We show that the fluid can be foliated into non-interacting submanifolds whose equations of motion are exactly that of the wiggly strings considered previously by Vilenkin and Carter. In a special case of negligible statistical variance in either the left or the right-moving directions of microscopic strings, the submanifolds are described by the action of chiral strings proposed by Witten. When both variances vanish the submanifolds are described by the Nambu-Goto action and the string fluid reduces to the string dust introduced by Stachel., Comment: 16 pages

We consider a covariant approach to coarse-graining a network of interacting Nambu-Goto strings. A transport equation is constructed for a spatially flat Friedmann universe. In Minkowski space and with no spatial dependence this model agrees with a previous model. Thus it likewise converges to an equilibrium with a factorizability property. We present an argument that this property does not depend on a `string chaos' assumption on the correlations between strings. And in contrast to the earlier model, this transport equation agrees with conservation equations for a fluid of strings derived from a different perspective., Comment: 17 pages, 1 figure

We consider conserved currents in an interacting network of one-dimensional objects (or strings). Singular currents localized on a single string are considered in general, and a formal procedure for coarse-graining over many strings is developed. This procedure is applied to strings described by the Nambu-Goto action such as cosmic strings. In addition to conserved currents corresponding to the energy-momentum tensor, we consider an antisymmetric tensor of conserved currents related to the string tangent vector. Under the assumption of local equilibrium we derive a complete set of hydrodynamic equations for strings., Comment: 16 pages

We develop further a kinetic theory of strings and derive a transport equation for a network of cosmic strings with Nambu-Goto evolution, interactions and background gravitational effects taken into account. We prove an H-theorem and obtain necessary and sufficient conditions for a thermodynamic equilibrium. At the lowest order the equilibrium is estimated by the von Mises-Fisher distributions parametrized by mean directions and dispersions of the right- and left-moving tangent vectors. Under assumption of a local equilibrium we derive a complete set of hydrodynamic equations that govern the evolution of strings on large scales. We also argue that on small scales the assumption of a local equilibrium would break down, and non-equilibrium steady states, described by the Sinai-Ruelle-Bowen distributions, should be used instead., Comment: 15 pages, replaced to match version accepted by PRD

There are two main approaches to non-equlibrium statistical mechanics: one using stochastic processes and the other using dynamical systems. To model the dynamics during inflation one usually adopts a stochastic description, which is known to suffer from serious conceptual problems. To overcome the problems and/or to gain more insight, we develop a dynamical systems approach. A key assumption that goes into analysis is the chaotic hypothesis, which is a natural generalization of the ergodic hypothesis to non-Hamiltonian systems. The unfamiliar feature for gravitational systems is that the local phase space trajectories can either reproduce or escape due to the presence of cosmological and black hole horizons. We argue that the effect of horizons can be studied using dynamical systems and apply the so-called thermodynamic formalism to derive the equilibrium (or Sinai-Ruelle-Bowen) measure given by a variational principle. We show that the only physical measure is not the Liouville measure (i.e. no entropy problem), but the equilibrium measure (i.e. no measure problem) defined over local trajectories (i.e. no problem of observables) and supported on only infinite trajectories (i.e. no problem of initial conditions). Phenomenological aspects of the fluctuation theorem are discussed., Comment: 21 pages, replaced to match version accepted by PRD

The definition of probabilities in eternally inflating universes requires a measure to regulate the infinite spacetime volume, and much of the current literature uses a global time cutoff for this purpose. Such measures have been found to lead to paradoxical behavior, and recently Bousso, Freivogel, Leichenauer, and Rosenhaus have argued that, under reasonable assumptions, the only consistent interpretation for such measures is that time must end at the cutoff. Here we argue that there is an alternative, consistent formulation of such measures, in which time extends to infinity. Our formulation begins with a mathematical model of the infinite multiverse, which can be constructed without the use of a measure. Probabilities, which obey all the standard requirements for a probability measure, can then be defined by mathematical limits. They have a peculiar feature, however, which we call time-delay bias: if the outcome of an experiment is reported with a time delay that depends on the outcome, then the observation of the reports will be biased in favor of the shorter time delay. We show how the paradoxes can be resolved in this interpretation of the measure., Comment: 32 pages, 10 figures, 1 table

We study the dynamics of strings by means of a distribution function f(A, B, x, t) defined on a 9+1D phase space, where A and B are the correlation vectors of right- and left-moving waves. We derive a transport equation (an analogous to Boltzmann transport equation for particles) that governs the evolution of long strings with Nambu-Goto dynamics as well as reconnections taken into account. We also derive a system of coupled transport equations (an analogous to BBGKY hierarchy for particles) which can simultaneously describe long strings \tilde{f}(A, B, x, t) as well as simple loops \mathring{f}(A, B, x, t) made out of four correlation vectors. The formalism can be used to study non-linear dynamics of fundamental strings, D-brane strings or field theory strings. For example, the complicated semi-scaling behavior of cosmic strings translates into a simple solution of the transport system at small energy densities., Comment: 11 pages, 5 figures, improved transverse collisions, accepted for publication in Physical Review D

Many probability measures in the multiverse depend exponentially on some observable parameters, giving rise to potential problems such as youngness bias, Q-catastrophe etc. In this paper we explore a possibility that the exponential runaway dependence should be viewed not as a problem, but as a feature that may help us to fix all parameters in the landscape, including the value of the cosmological constant, without using anthropic considerations., Comment: 5 pages

We show that most of cutoff measures of the multiverse violate some of the basic properties of probability theory when applied repeatedly to predict the results of local experiments. Starting from minimal assumptions, such as Markov property, we derive a correspondence between cosmological measures and quantum field theories in one lesser dimension. The correspondence allows us to replace the picture of an infinite multiverse with a finite causally connected region accessible by a given observer in conjunction with a Euclidean theory defined on its past boundary., Comment: 6 pages, many points clarified, discussion of Markov random fields added

We develop a model of string dynamics with back-reaction from both scaling and non-scaling loops taken into account. The evolution of a string network is described by the distribution functions of coherence segments and kinks. We derive two non-linear equations which govern the evolution of the two distributions and solve them analytically in the limit of late times. We also show that the correlation function is an exponential, and solve the dynamics for the corresponding spectrum of scaling loops., Comment: 8 pages, minor corrections, accepted for publication in JCAP

At early stages the dynamics of cosmic string networks is expected to be influenced by an excessive production of small loops at the scales of initial conditions l_{min}. To understand the late time behavior we propose a very simple analytical model of strings with a non-scaling population of loops. The complicated non-linear dynamics is described by only a single parameter N ~ 2/(1-C(l_{min})) where C(l) is a correlation function of the string tangent vectors. The model predicts an appearance of two new length scales: the coherence length \xi ~ t/N^2 and the cross-correlation length \chi ~ t/N. At the onset of evolution N ~ 10 and at late times N is expected to grow logarithmically due to cosmological stretching and emission of small loops. The very late time evolution might be modified further when the gravitational back-reaction scale grows larger than l_{min}., Comment: 5 pages, minor corrections, accepted for publication in Physical Review D

We argue that the total number of distinguishable locally Friedmann universes generated by eternal inflation is proportional to the exponent of the entropy of inflationary perturbations and is limited by e^{e^{3 N}}, where N is the number of e-folds of slow-roll post-eternal inflation. For simplest models of chaotic inflation, N is approximately equal to de Sitter entropy at the end of eternal inflation; it can be exponentially large. However, not all of these universes can be observed by a local observer. In the presence of a cosmological constant \Lambda the number of distinguishable universes is bounded by e^{|\Lambda|^{-3/4}}. In the context of the string theory landscape, the overall number of different universes is expected to be exponentially greater than the total number of vacua in the landscape. We discuss the possibility that the strongest constraint on the number of distinguishable universes may be related not to the properties of the multiverse but to the properties of observers., Comment: 12 pages, minor changes

We analyze statistical properties of the separate multipole moments of the CMB temperature maps and find that the distribution tails are slightly non-Gaussian. Moreover, the deviation from Gaussianity peaks sharply at around $l\sim45\pm10$. If the detected non-Gaussianities should be attributed to the remaining foreground contamination from the galactic plane, an appearance of a similar peak in different frequency bands is unexpected. The presence of the peak was also confirmed by the analysis of weighted multipoles which is less susceptible to the galactic noise. A separate analysis of the multipoles with odd and even $l+m$ shows two opposite types of non-Gaussianities which cannot be explained with only galactic foregrounds. We argue that cosmic strings or correlated gravitational waves could lead to the observed effect., Comment: an important note added

We analyze the behavior of linear perturbations in vector inflation. In contrast to the scalar field inflation, the linearized theory with vector fields contains couplings between scalar, vector and tensor modes. The perturbations decouple only in the ultraviolet limit, which allows us to carry out the canonical quantization. Superhorizon perturbations can be approximately analyzed due to suppressed mixing between different modes in the small fields models. We find that the vector perturbations of the metric decay exponentially, but the scalar and tensor modes could remain weakly coupled throughout the evolution. As a result, the vector inflation can produce significant correlations of the scalar and tensor modes in the CMB. For the realistic models the effect is rather small, but not negligible., Comment: minor changes, some references added; accepted for publication in Physical Review D

We study the recently proposed "stationary measure" in the context of the string landscape scenario. We show that it suffers neither from the "Boltzmann brain" problem nor from the "youngness" paradox that makes some other measures predict a high CMB temperature at present. We also demonstrate a satisfactory performance of this measure in predicting the results of local experiments, such as proton decay., Comment: 12 pages

We discuss the gravitational waves (GW) in the context of vector inflation. We derive the action for tensor perturbations and find that tachyonic instabilities are present in most (but not all) of the inflationary models with large fields. In contrast, the stability of the small field inflation ($A_{\mu}A^{\mu}\ll\frac{1}{N}$) is ensured by the usual slow-roll conditions, where $N$ is the total number of fields. For example, the Coleman-Weinberg potential and the power-law inflation are always stable in the small fields limit with an approximately flat spectrum of GW. We also provide some examples which lead to a rapid decay of GW and predict the absence of tensor modes in the CMB., Comment: 6 pages

We propose a scenario where inflation is driven by non-minimally coupled massive vector fields. In an isotropic homogeneous universe these fields behave in presicely the same way as a massive minimally coupled scalar field. Therefore our model is very similar to the model of chaotic inflation with scalar field. For vector fields the isotropy of expansion is achived either by considering a triplet of orthogonal vector fields or for the expense of $N$ randomly oriented vector fields. In the last case the substantial anisotropy of the expansion of order $1/\sqrt{N}$ survives until the end of inflation. The lightest vector fields might also force the late time acceleration of the Universe., Comment: 6 pages; minor changes

We develop an analytical model to study the production spectrum of loops in the cosmic string network. In the scaling regime, we find two different scales corresponding to large (one order below horizon) and small (few orders below horizon) loops. The very small (tiny) loops at the gravitational back reaction scale are absent, and thus, our model has no ultra-violet divergences. We calculate the spectrum of loops and derive analytical expressions for the positions and magnitudes of the small and large scale peaks. The small loops are produced by large bursts of similar loops moving with very high velocities in the same direction. We describe the shape of large loops, which would usually consist of few kinks and few cusps per oscillation cycle. We also argue that the typical size of large loops is set by the correlation length, which does not depend on the intercommutation probability p, while the interstring distance scales as p^{1/3}., Comment: 6 pages, 1 figure, power-law approximation is replaced with exponential

Cosmological scenarios with k-essence are invoked in order to explain the observed late-time acceleration of the universe. These scenarios avoid the need for fine-tuned initial conditions (the "coincidence problem") because of the attractor-like dynamics of the k-essence field \phi. It was recently shown that all k-essence scenarios with Lagrangians p=L(X)/\phi^2, necessarily involve an epoch where perturbations of \phi propagate faster than light (the "no-go theorem"). We carry out a comprehensive study of attractor-like cosmological solutions ("trackers") involving a k-essence scalar field \phi and another matter component. The result of this study is a complete classification of k-essence Lagrangians that admit asymptotically stable tracking solutions, among all Lagrangians of the form p=K(\phi)L(X) . Using this classification, we select the class of models that describe the late-time acceleration and avoid the coincidence problem through the tracking mechanism. An analogous "no-go theorem" still holds for this class of models, indicating the existence of a superluminal epoch. In the context of k-essence cosmology, the superluminal epoch does not lead to causality violations. We discuss the implications of superluminal signal propagation for possible causality violations in Lorentz-invariant field theories., Comment: 27 pages, RevTeX4. Minor cosmetic changes, references added

We propose a numerical test of fundamental physics based on the complexity measure of a general set of functions, which is directly related to the Kolmogorov (or algorithmic) complexity studied in mathematics and computer science. The analysis can be carried out for any scientific experiment and might lead to a better understanding of the underlying theory. From a cosmological perspective, the anthropic description of fundamental constants can be explicitly tested by our procedure. We perform a simple numerical search by analyzing two fundamental constants: the weak coupling constant and the Weinberg angle, and find that their values are rather atypical., Comment: 6 pages, 3 figures, RevTeX, expansion and clarification, references added

We study the landscape models of eternal inflation with an arbitrary number of different vacua states, both recyclable and terminal. We calculate the abundances of bubbles following different geodesics. We show that the results obtained from generic time-like geodesics have undesirable dependence on initial conditions. In contrast, the predictions extracted from ``eternal'' geodesics, which never enter terminal vacua, do not suffer from this problem. We derive measure equations for ensembles of geodesics and discuss possible interpretations of initial conditions in eternal inflation., Comment: 7 pages, 4 figures

We study the production of loops in the cosmic string network in the expanding background by means of a numerical simulation exact in the flat-spacetime limit and first-order in the expansion rate. We find an initial regime characterized by production of small loops at the scale of the initial correlation length, but later we see the emergence of a scaling regime of loop production. This qualitatively agrees with earlier expectations derived from the results of flat-spacetime simulations. In the final scaling regime we find that the characteristic length of loops scales as $\sim 0.1 t$ in both radiation and matter eras., Comment: 9 pages, 5 figures, RevTeX; v2: expansion and clarification, improved figures; v3: reference updated, typos fixed

We study a class of ``landscape'' models in which all vacua have positive energy density, so that inflation never ends and bubbles of different vacua are endlessly ``recycled''. In such models, each geodesic observer passes through an infinite sequence of bubbles, visiting all possible kinds of vacua. The bubble abundance $p_j$ can then be defined as the frequency at which bubbles of type $j$ are visited along the worldline of an observer. We compare this definition with the recently proposed general prescription for $p_j$ and show that they give identical results.