Filename
stringlengths 22
64
| Paragraph
stringlengths 8
5.57k
|
|---|---|
Processed_Self-consistent_description_of_$Λ$_hypernuclei_in_.txt | the Pauli principle at the quark level by Hungerford and Biedenharn [34]. Investigations of this repulsive core in the Λ-nucleus system have been made by Takeuchi and Shimizu, and others [35, 36], based on a nonrelativistic quark model. |
Processed_Self-consistent_description_of_$Λ$_hypernuclei_in_.txt | In the light of these earlier investigations, it seems appropriate to investigate the (heav- ier) hypernuclear systems quantitatively using a microscopic model based on quark degrees of freedom. For this purpose, the QMC model (which is built explicitly on quark degrees of freedom) seems ideally suited, because it has already been shown to describe the properties of finite nuclei quantitatively. |
Processed_Self-consistent_description_of_$Λ$_hypernuclei_in_.txt | Here, zY , B, x, xs and mu,d,s are the parameters for the sum of the c.m. and gluon fluctuation effects, bag pressure, lowest eigenvalues for the (u, d) and s quarks, respectively, and the corresponding current quark masses with mu = md. zN and B (zY ) are fixed by fitting the nucleon (hyperon) mass in free space. |
Processed_Self-consistent_description_of_$Λ$_hypernuclei_in_.txt | This correction to the spin-orbit force, which appears naturally in the QMC model, may also be modelled at the hadronic level of the Dirac equation by adding a tensor interaction. |
Processed_Self-consistent_description_of_$Λ$_hypernuclei_in_.txt | We are now in a position to discuss the results. As already mentioned, the calculations are fully self-consistent for all fields appearing in Eqs. (4)-(9). In principle, the existence of the Λ outside of the nuclear core breaks spherical symmetry, and one should include this in a truly rigorous treatment. We neglect this effect, since it is expected to be of little importance for spectroscopic calculations [37, 38]. However, we do include the response of the nuclear core arising from the self-consistent calculation, which is significant for a description of the baryon currents and magnetic moments, and a pure relativistic effect [38, 39]. We will discuss later the self-consistency effects between the nuclear core and the Λ. |
Processed_Self-consistent_description_of_$Λ$_hypernuclei_in_.txt | In Fig. 1, we show the effective masses of the nucleon and Λ as well as the baryon densities calculated for (a): 17 Λ Pb. The results are for the 1s1/2 Λ state, where effects of the Λ on the whole system are expected to be the largest. The effective masses in the hypernuclei, 17 Λ Pb behave in a similar manner as the distance r from the center of each nucleus increases (the baryon density decreases). The decrease of the baryon densities around the center in 17 Λ O in Fig. 1(a), which one might think a shortcoming of the present treatment, is also expected in nonrelativistic potential model calculation [40]. |
Processed_Self-consistent_description_of_$Λ$_hypernuclei_in_.txt | The calculated Λ single-particle energies for the closed-shell nuclear core plus one Λ con- figuration are listed in Table 1, with the experimental data [29, 32]. At a glance, one can easily see that the spin-orbit splittings in the present calculation are very small for all hyper- nuclei. These small spin-orbit splittings, tend to be smaller as the baryon density increases, or the atomic number increases. This smallness of the spin-orbit splitting is a very promising achievement in the present approach. Concerning the single-particle energy levels, although a direct comparison with the data is not precise due to the different configurations, the cal- culated results seem to lead to overestimates. In order to make an estimate of the difference due to this different configuration, we calculated the following quantity. By removing one 1p3/2 neutron in 16O, and putting a Λ as experimentally observed, we calculated in the same way as for 17 Λ O-which means the nuclear core was still treated as spherical. In this case the calculated energy for the 1s1/2 Λ is -19.9 MeV, to be compared with the value -20.5 MeV of the present calculation. For larger atomic numbers, the difference is smaller. (See also Table 1.) We also attempted the calculation using the scaled coupling constant, 0.93 × gΛ σ (σ = 0), which repro- duces the empirical single-particle energy for the 1s1/2 in 41 Λ Ca, -20.0 MeV [29]. The results obtained using this scaled coupling constant are also listed in Table 1, denoted by 41 Λ Ca⋆ and 209 Λ Pb⋆. Then, one can easily understand that the QMC model does not require a large SU(6) (SU(3)) breaking effect in this respect (7 %) to reproduce the empirical single-particle energies. Furthermore, the model still achieves the very small spin-orbit interaction for Λ in hypernu- clei, without introducing the tensor coupling of the ω meson, which may be contrasted with the QHD type calculations [22]. We note that the spin-orbit splittings for the Λ (or hyperon) single-particle energies in hypernuclei are not well determined by the experiments [28]-[32]. |
Processed_Self-consistent_description_of_$Λ$_hypernuclei_in_.txt | In Table 2, we list the calculated binding energy per baryon, −E/A, rms charge radius, rch, and rms radii of the Λ and the neutron and proton distributions (rΛ, rn and rp, respectively), for the 1s1/2 and 1p3/2 Λ configurations. The rms charge radius is calculated by convolution with a proton form factor [2]. For comparison, we also give these quantities without a Λ-i.e., for normal finite nuclei. The differences in values for finite nuclei and hypernuclei listed in Table 2 reflect the effects of the Λ, through the self-consistency procedure. One can easily see that the effects of the Λ become weaker as the atomic number becomes larger, and the Λ binding energy becomes smaller. |
Processed_Can_Quantum_Markov_Evolutions_Ever_Be_Dynamically_.txt | We consider the class of quantum stochastic evolutions (SLH-models) leading to a quantum dynamical semigroup over a fixed quantum mechan- ical system (taken to be finite-dimensional). We show that if the semi- group is dissipative, that is, the coupling operators are non-zero, then a dynamical decoupling scheme based on unitary rotations on the system space cannot suppress decoherence even in the limit where the period between pulses vanishes. We emphasize the role of the Fock space di- lation used here to construct a quantum stochastic model, as there are often dilations of the same semigroup using an environmental noise model of lower level of chaoticity for which dynamical decoupling is effective. We show that the Chebotarev-Gregoratti Hamiltonian behind a quantum stochastic evolution is an example of a Hamiltonian dynamics on a joint system-environment that cannot be dynamically decoupled in this way. |
Processed_Can_Quantum_Markov_Evolutions_Ever_Be_Dynamically_.txt | scheme will be effective. The question turns out to be misstated - a quantum dynamical semigroup may have several inequivalent dilations, some of which are amenable to dynamical decoupling and some of which are not. The cor- rect question is whether a given dilation is amenable to dynamical decoupling. Physically we have no choice but to work with the actual environment, and the restriction is that the decoupling scheme can only be applied to the system (not the environment). We shall focus on environments which allow a quantum stochastic description - so called SLH-models [14]. |
Processed_Can_Quantum_Markov_Evolutions_Ever_Be_Dynamically_.txt | For instance, if V is a group then the group average is a conditional expectation onto the group centralizer, so it suffices to choose the group to have trivial centralizer. Such a group is called a decoupling group. For the quantum memory scenario described above, applying a DD sequence based on a decoupling group one gets that ˜Uv(tN , t0) = I N ), where N = #V and Z(t) an Z(t) + O(t2 arbitrary bath operator. So DD suppresses the effect of the Hamiltonian H to first order in tN . |
Processed_Can_Quantum_Markov_Evolutions_Ever_Be_Dynamically_.txt | We shall assume that for large N the sequences v have the property that each element v It is often sufficient to assume that the sequence cycles deterministically through the elements of V in some fixed order. Alternatively, we may assume that the vk form an i.i.d. sequence of random variables taking values in V , with a common distribution being the uniform distribution. |
Processed_Can_Quantum_Markov_Evolutions_Ever_Be_Dynamically_.txt | Our main point however is that there is a longer time-scale Tdd = Nddτ at which the averaging takes place and dynamical decoupling starts to become effective. |
Processed_Can_Quantum_Markov_Evolutions_Ever_Be_Dynamically_.txt | tk−1 t1 ≤ where 0 (k = 2, , n). This is known as the “quantum regression” formula in the physics literature [18]. The rationale for considering only multi-time correlations with these specific time-orderings is well known in quantum physics - see for instance [20, Section 2.3]. |
Processed_Can_Quantum_Markov_Evolutions_Ever_Be_Dynamically_.txt | In this case we find E[jt(X)] = Et[X] as given in (14) above. Technically the quantization of the random variable Λ is unnecessary, but allows us to frame the model in the language of quantum dilations. |
Processed_Can_Quantum_Markov_Evolutions_Ever_Be_Dynamically_.txt | We see from (15) that these models give the correct one-point function for averages associated with the quantum dynamical semigroup in (16), but the higher-order (n > 1) time-ordered correlation kernels arte not those of a Markov dilation. |
Processed_Can_Quantum_Markov_Evolutions_Ever_Be_Dynamically_.txt | which is typically different from that the original . This opens up the possi- bility that the dynamical decoupling may result in extending the dissipation to observables that were previously invariant under the original dynamics. |
Processed_Can_Quantum_Markov_Evolutions_Ever_Be_Dynamically_.txt | L−X = 1 2 γσ+σ−X. Without dynamical decoupling, this would γσ+Xσ− − imply a master equation where the state converges to the ground state (south pole of the Bloch sphere). If we applied DD with Pauli group VPauli, then we would in fact find ¯ 2 L+ + 1 L+ is the generator corresponding 2 L−, where to L = √γσ+. This is because conjugation with either σz change σ− I or σ+. This to DD therefore results in the convergence to a different state - the minimum information state (centre of the Bloch sphere). The damping rates are same in both cases: the decaying terms have factors e−γt on the diagonal, and e−γt/2 off-diagonal. Arguably the DD scheme is making things worse in this case. |
Processed_Can_Quantum_Markov_Evolutions_Ever_Be_Dynamically_.txt | This is, of course, consistent with the dynamical decoupling scheme for Hamil- tonian evolution coupling the system to the Fock bath via the Chebotarev- Gregoratti Hamiltonian HC.G.. |
Processed_Can_Quantum_Markov_Evolutions_Ever_Be_Dynamically_.txt | Gregoratti Hamiltonian HC.G. cannot be dynamically decoupled using schemes acting solely on the system space. |
Processed_Can_Quantum_Markov_Evolutions_Ever_Be_Dynamically_.txt | In this paper we have studied dynamical decoupling associated with dilations of quantum dynamical semigroups. A fixed semigroup in general will possess several inequivalent dilations, and we have seen instances where a dissipative semigroup has a dilation where dynamical decoupling works, and another where it does not. |
Processed_Can_Quantum_Markov_Evolutions_Ever_Be_Dynamically_.txt | To better understand this, it is worth recalling that there are chaotic hier- archies of noise models. This was first articulated by Accardi [27] for quantum systems: at the top are independent increment processes; then Markov processes with an expected past filtration; and so on down to processes with deterministic algebraic filtration. |
Processed_Can_Quantum_Markov_Evolutions_Ever_Be_Dynamically_.txt | If we look at the minimal stochastic model (coupling our system to an un- stable particle), then the “noise” is effectively just a single real random variable Λ. This type of model would lie close to the bottom of the Accardi hierarchy of quantum chaoticity. Effectively it is just a Hamiltonian randomized by a single random variable (which we could just as easily take to be classical). |
Processed_Can_Quantum_Markov_Evolutions_Ever_Be_Dynamically_.txt | The quantum stochastic models considered in Section 3 in contrast requires a Fock space (Bose second quantization) of time-pulse wavefunctions. We are now at the maximum end of chaoticity as quantum noise goes. This level is necessary if we are to realize Markovianity. |
Processed_Can_Quantum_Markov_Evolutions_Ever_Be_Dynamically_.txt | The implications of this paper are that complete dynamical decoupling does not work for quantum stochastic evolutions. To be clear, the goals of DD are more realistic than complete decoupling in practical situations, often seeking to slow down the rate of decoherence, and maybe only for a particular subalgebra of operators. Our results however are general enough: the question of how applica- ble they are ultimately comes down to modelling issues. If one is working with these type of models, which is typically the case in quantum optics situations, then one must be working at a time scale where the bath auto-correlation time is negligible - and the while the dynamical decoupling may be considered as periodic and rapid, the time steps are nevertheless large compared to the bath auto-correlation time. This sets the various time scale regimes. In principle, as the decoupling operations are applied at longer time scales that the bath auto- correlation, we should not be too surprised to see that the dynamical decoupling is no longer effective. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | This paper shows that conditional independence reasoning can be applied to optimize epistemic model checking, in which one verifies that a model for a number of agents operating with imperfect information satisfies a formula expressed in a modal multi-agent logic of knowledge. The optimiza- tion has been implemented in the epistemic model checker MCK. The paper reports experimental results demonstrating that it can yield multiple orders of magnitude performance improvements. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Epistemic model checking [12] is a technique for the verification of information theoretic properties, stated in terms of a modal logic of knowledge, in systems in which multiple agents operate with imperfect information of their environment. It has been applied to settings that include diagnosis [6], and reasoning in game-like settings [13, 14, 10], concurrent hardware protocols [3] and security protocols [1, 5, 21]. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | The contribution of the present paper is to demonstrate that conditional independence techniques from the Bayesian Net literature [17, 23, 8] can be applied in the context of epistemic model checking. We develop a generalization of these techniques for a multi-agent modal logic of knowledge, that enables model checking computations for this logic to be optimized by reducing the number of variables that need to be included in data structures used by the computation. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | We have implemented the technique in the epistemic model checker MCK [12]. The technique de- veloped can be applied for other semantics and algorithms, but we focus here on agents with synchronous perfect recall and model check the reduced representation using binary decision diagram techniques. The synchronous perfect recall semantics presents the most significant challenges to the computational cost of epistemic model checking, since it leads to a rapid blowup in the number of variables that need to be handled by the symbolic model checking algorithms. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | The paper presents experimental results that demonstrate that the conditional independence opti- mization yields very significant gains in the performance of epistemic model checking. Depending on the example, the optimization yields a speedup as large as four orders of magnitude. Indeed, it can yield linear growth rates in computation time on examples that otherwise display an exponential growth rate. It adds significantly to the scale of the examples that can be analyzed in reasonable time, increasing both the number of agents that can be handled, the length of their protocols, and the size of messages they communicate. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | We begin by recalling some basic definitions from epistemic logic and epistemic model checking. We use epistemic variable structures, a particular concrete representation of Kripke structures (it can be shown that there is no loss of generality). We show how these structures arise in a multi-agent setting in which each agent’s behaviour is described by a program. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | where p ∈ V and i ∈ Agts. That is, the language is a modal propositional logic with a set of modalities Ki, such that Kiφ means, intuitively, that the agent i knows that φ . We freely use common abbreviations from propositional logic, e.g., we write φ1 ∨ φ2 for ¬(¬φ1 ∧ ¬φ2) and φ1 ⇒ φ2 for ¬φ1 ∨ φ2 and φ1 ⇔ φ2 for (φ1 ⇒ φ2) ∧ (φ2 ⇒ φ1). We write vars(φ ) for the set of variables occurring in the formula φ . |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Define an epistemic variable structure over a set of variables V to be a tuple M = (A, O,V ) where A ⊆ assgt(V ) and O = {Oi}i∈Agts is a collection of sets of variables Oi ⊆ V , one for each agent i. Intu- itively, such a structure is an alternate representation of an epistemic Kripke structure, where the indis- tinguishability relation for an agent is specified by means of a set of variables observable to the agent. The elements of A correspond to the worlds of this Kripke structure. The relation ∼i on worlds for agent i is defined by u ∼i v when u (cid:22) Oi = v (cid:22) Oi. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Intuitively, the clause for the operator Ki says that Kiφ holds when φ is true at all worlds that the agent considers to be possible. We write M |= φ when M, w |= φ for all worlds w ∈ W . |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | In the context of model checking, one is interested in analyzing a model represented as a program. We now show how programs generate a Kripke structure that serves as their semantics. We work with a very simple straightline programming language in which a multi-agent scenario is represented by each of the agents running a protocol in the context of an environment. The syntax and operational semantics of this language is shown in Figure 1. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | terminated agent does nothing while other agents are still running. A joint protocol J, is represented by a statement of the form P1 || . . . || Pn ∆ CE, and consists of a number of agent protocols P1, . . . , Pn, running in the context of an environment represented by code CE. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | There are two relations in the operational semantics. States s are assignments of boolean variables to boolean values, and we write e(s) for the value of boolean expression expression e in state s. The binary relation →0 on configurations of type (s,C) represents zero-time state transitions, which do not change the system clock. The binary relation →1 on configurations of type (s, J) represents state transitions corresponding to a single clock tick. Thus, C →∗ 0 ε represents that code C runs to termination in time 0. In a single tick transition represented by →1, we take the next atomic action ai = (cid:104)Ci(cid:105) from each of the agents, and compose the code Ci in these actions with the code from the environment CE to form the code C = C1; . . .Cn;CE. The single step transition is obtained as the result of running this code C to termination in zero-time. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | A system is represented using this programming language by means of a tuple I = (J, I, Q), where J is a joint protocol for n agents, I is a boolean formula expressing the initial condition, and Q is a tuple of n sets of variables, with Qi representing the variables observable to agent i. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Given a maximum running time n, a system I = (J, I, Q) is associated to an epistemic variable structure Mn(I ) = (cid:104)A, O,V (cid:105) as follows. A run of length n of the system is a sequence of states r = s0, s2, . . . , sn, where s0 satisfies the initial condition I and (s0, J) →1 (s1, J1) →1 . . . →1 (sn, Jn) for some J1, . . . , Jn. If U is the set of variables appearing in J, we define V to be the set of timed variables, i.e., the set of variables vt where 0 ≤ t ≤ n. We take A to be the set of assignments αr to variables V derived from runs r by αr(vt) = st(v) when v ∈ U and 0 ≤ t ≤ n. For the perfect recall semantics, which is our focus in this paper, we define the observable variables Oi for agent i to be the set of timed variables vt where v ∈ Qi and 0 ≤ t ≤ n. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Three cryptographers are sitting down to dinner at their favourite restaurant. Their waiter informs them that arrangements have been made with the maitre d’hotel for the bill to be paid anonymously. One of the cryptographers might be paying for the dinner, or it might have been NSA (U.S. National Security Agency). The three cryptographers respect each other’s right to make an anonymous payment, but they wonder if NSA is paying. They resolve their uncertainty fairly by carrying out the following protocol: Each cryptographer flips an unbiased coin behind his menu, between him and the cryptogra- pher on his right, so that only the two of them can see the outcome. Each cryptographer then states aloud whether the two coins he can see–the one he flipped and the one his left-hand neighbor flipped–fell on the same side or on different sides. If one of the cryptographers is the payer, he states the opposite of what he sees. An odd number of differences uttered at the table indicates that a cryptographer is paying; an even number indicates that NSA is paying (assuming that the dinner was paid for only once). Yet if a cryptographer is paying, neither of the other two learns anything from the utterances about which cryptographer it is. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | The solution generalizes to any number n of cryptographers C0, . . . ,Cn−1 at the table. We may rep- resent the protocol by means of the following program for cryptographer i, who is assumed to have a boolean variable paidi that indicates whether (s)he is the payer. (The program starts running from an initial state in which the constraint (cid:87) 0≤i< j≤n−1 ¬(paidi ∧ paid j) is satisfied.) We write ⊕ for the exclusive-or. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | All variables take boolean values. Each cryptographer is associated with a set of variables, whose values they are able to observe at each moment of time. Note that a cryptographer may write to a variable that they are not able to observe. In particular, Ci writes to the variable lefti+1 mod n that is observed only by Ci+1 mod n. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | We will work with dependency networks that show how the values of variables change over time. The DC protocol runs for 4 ticks of the clock, (time 0 plus one tick for each step in the protocol), so we have instances v0 . . . v3 of each variable v. Figure 2 shows the dependencies between these instances. The figure is to be understood as follows: a variable vt takes a value that directly depends on the values of the variables ut−1 to vt. Additionally, there is a dependency 1 between the initial values paid0 i captured using a special variable pinit. (We give a more formal presen- tation of such dependency structures below.) The observable variables for agent C0 have been indicated by rectangles: timed variables inside these rectangles are observable to C0. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | and applies also in other contexts such as Spohn’s theory of ordinal conditional functions [28]. There is a close connection also to ideas in database query optimization [20] and operations research [4]. We show here that Shenoy and Shafer’s general axiomatic framework applies to epistemic model checking. This will enable us to apply the variable elimination algorithm to derive techniques for optimizing epistemic model checking. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | We begin by presenting Shenoy and Shafer’s framework, following [16]. Let Vars be a set of vari- ables, with each v ∈ Vars taking values in a set Ωv. For a set X of variables, the set ΩX = Πx∈X Ωx is called the frame of X. Elements of ΩX are called configurations of X. In case X = /0, the set ΩX is interpreted as {(cid:104)(cid:105)}, i.e., the set containing just the empty tuple. We write D for P(Vars). |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | A valuation algebra is a tuple (cid:104)Φ, dom, e, ⊗, ↓(cid:105), with components as follows. A state of information is represented in valuation algebra by a primitive notion called a valuation. Component Φ is a set, the set of all valuations, and dom is function from Φ to D. Intuitively, for each valuation s ∈ Φ, the domain dom(s) is the set of variables that the information is about. For a set of variables X, we write ΦX for the set of valuations s with dom(s) = X. Component e gives an element eX ∈ ΦX for each X ∈ D. A valuation algebra also has two operations ⊗ : Φ × Φ → Φ (combination) and ↓: Φ × D → Φ (marginalization), with ⊗, intuitively, representing the combination of two pieces of information, and ↓ used to restrict a piece of information to a given set of variables. Both are written as infix operators. From marginalization, another operator − : Φ × Vars → Φ called variable elimination can be defined, by s−x = s ↓ (dom(s) \ {x}). |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | VA2. Domain of combination. For all s,t ∈ Φ, dom(s ⊗ t) = dom(s) ∪ dom(t). |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | A key result that follows from these axioms, called the Fusion Algorithm [26], exploits Distributivity of Elimination over Combination to give a way of computing the result of a marginalization operation applied to a sequence of combinations, by pushing in variable eliminations over elements of the combi- nation that do not contain the variable. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | For a finite set S = {s1, s2, . . . , sk} ⊆ Φ, write ⊗S for s1 ⊗ s2 ⊗ . . . ⊗ sk. We define the fusion of S via x ∈ Vars to be the set Fusx(S) = {(⊗S+)−x} ∪ S− where we have partitioned S as S+ ∪ S−, such that S+ is the set of s ∈ S with x ∈ dom(s), and S− is the set of s ∈ S with x (cid:54)∈ dom(s). That is, in the fusion of the set S with respect to x, we combine all the valuations with x in their domain, and then eliminate x, and preserve all valuations with x not in their domain. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Theorem 1 ([26]) Let S be a finite set of valuations, and X ⊆ Vars. Suppose dom(S)\X = {x1, x2, . . . , xn}. Then (⊗S) ↓ X = ⊗Fusxn(. . . (Fusx1(S))) . |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Each ordering of the variables x1 . . . xn gives a different way to compute (⊗S) ↓ X. A well chosen order can yield a significant optimization of the computation, by keeping the domains of the intermediate valuations in the sequence of fusions small. Finding an optimal order may be computationally complex, but there exist heuristics that produce good orders in practice [22, 18]. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | We now show that the relational structures that underly Kripke structures are associated with alge- braic operations that satisfy the conditions VA1-VA6. It will follow from this that the Fusion algorithm can be applied to these structures. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | eX = (assgt(X), X), • s1 ⊗ s2 = (A,V ) where V = V1 ∪ V2, and A ⊆ assgt(V ) is defined by α ∈ A iff α (cid:22) V1 ∈ A1 and α (cid:22) V2 ∈ A2. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | s1 ↓ X = (A,V ) where V = V1 ∩ X, and A = {α (cid:22) X | α ∈ A1}. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | To use terminology from relational databases, s1 ⊗ s2 is the join of relations and s ↓ X is the projection of the relation s onto attributes X. The following result is straightforward; these properties are well-known for relational algebra. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | We may extend the operation of marginalization in this valuation algebra to epistemic variable struc- tures as follows. If M = (A, O,V ) is an epistemic variable structure and X ⊆ V , we define M ↓ X = (A(cid:48), O(cid:48),V (cid:48)) where A(cid:48) = {α (cid:22) X | α ∈ A} and O(cid:48) i = Oi ∩ X for all i ∈ Agts and V (cid:48) = V ∩ X. In general, this operation results in agents losing information, since their knowledge is based on the observation of fewer variables. Below, we identify conditions where knowledge is preserved by this operation. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Let X,Y, Z ⊆ V be sets of variables. The notion of conditional independence expresses a generalized type of independency relation. Variables X are said to be conditionally independent of Y , given Z, if, intuitively, once the values of Z are known, the values of Y are unrelated to the values of X, so that neither X not Y gives any information about the other. This intuition can be formalized for both probabilistic and discrete models. The following definition gives a discrete interpretation, related to the notion of embedded multivalued dependencies from database theory [11]. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Definition 1 Let A ⊆ assgt(V ) be a set of assignments over variables V and let X,Y, Z ⊆ V . We say that A satisfies the conditional independency X⊥Y |Z, and write A |= X⊥Y |Z, if for every pair of worlds u, v ∈ A with u (cid:22) Z = v (cid:22) Z, there exists w ∈ A with w (cid:22) X ∪ Z = u (cid:22) X ∪ Z and w (cid:22) Y ∪ Z = v (cid:22) Y ∪ Z. For an epistemic variable structure M = (A, O,V ), we write M |= X⊥Y |Z if A |= X⊥Y |Z. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Conditional independencies can be deduced from graphical representations of models. Such rep- resentations have been used in the literature on Bayesian Nets [23, 17], and have also been applied in propositional reasoning [8, 9]. The following presentation is similar to [8] except that we work with relations over arbitrary domains rather than propositional formulas. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | The notion of d-separation [23] provides a way to derive a set of independency statements from a directed graph G. We present here an equivalent formulation from [19], that uses the notion of the moralized graph Gm of a directed graph G. The graph Gm is defined to be the undirected graph obtained from G by first adding an edge u − v for each pair u, v of vertices that have a common child (i.e. such that there exists w with u → w and v → w), and then replacing all directed edges with undirected edges. The set of parents of a node u is defined to be the set pa(v) = {u ∈ V | u → v}. For a set of vertices X of the directed graph G, we write An(X) for the set of all vertices v that are ancestors of some vertex x in X (i.e., such that there exists a directed path from v to x). For a subset X of the set of vertices of graph G = (V, E), we define the restriction of G to X to be the graph GX = (V ∩ X, {(u, v) ∈ E | u, v ∈ X}). For disjoint sets X,Y, Z, we then have that X is d-separated from Y by Z if all paths from X to Y in (GAn(X∪Y ∪Z))m include a vertex in Z. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Intuitively, the second constraint says that the relation sv does not constrain the parents of v: for each assignment of values to the parents of v, there is at least one value of v that is consistent. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Proposition 2 Suppose that M = (cid:104)V, E, S (cid:105) is a structured model and X,Y, Z are disjoint subsets of the vertices V of the directed graph G = (V, E). If X is d-separated from Y by Z, then ⊗S |= X⊥Y |Z. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Structured models have an additional property that provides an optimization when eliminating vari- ables: if a leaf node is one of the variables eliminated from the combination of the nodes of the graph, then it can be removed from the model without changing the result. This is captured in the following result. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Proposition 3 Suppose that M = (cid:104)V, E, S (cid:105) is a structured model, let X ⊆ V and let v ∈ V \ X be a leaf node. Then ⊗S ↓ X = ⊗(S \ {sv}) ↓ X. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | To apply these results for structured models to model checking epistemic logic, we use the following definition. We say that a structured model M = (cid:104)V, E, S (cid:105) represents the worlds of an epistemic variable structure M = (A, O,U) if V = U and A = ⊗S . That is, the structured model captures the set of assignments making up the epistemic variable structure. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Consider the following formulation of the model checking problem: for an epistemic formula φ , we wish to verify M |= φ where M = (A, O,V ) is an epistemic variable structure with observable variables O, with worlds represented by a structured model M = (cid:104)V, E, S (cid:105). |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | A first idea for how to optimize this verification problem is to reduce the structure M to the set of variables vars(φ ), together with the sets Oi for any operator Ki in φ . In fact, using the notion of conditional dependence, it is often possible to identify a smaller set of variables that suffices to verify the formula. The intuition for this is that some of the observed variables in Oi may be independent of the variables in the formula, and moreover, information may be redundantly encoded in the observable variables. The following definitions strengthen the idea of restricting to vars(φ ) ∪ O by exploiting a sufficient condition for the removal of observable variables. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Theorem 2 Suppose that κ is a relevance function for φ with respect to epistemic variable structure M and that X is a set of variables with κ(φ ) ⊆ X ⊆ dom(M ). Then for all worlds w of M , we have M , w |= φ iff M ↓ X, w (cid:22) X |= φ . |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | We work with the more complicated version because the algorithm for d-separation assumes disjoint sets. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | stronger conditions that κ(φ ) ∩ Oi ⊆ Ui and Ui d-separates κ(φ ) \Ui from Oi \Ui in the directed graph G associated with M. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Note (κ(φ ) \Ui) ∪ (Oi \Ui) ∪Ui = κ(φ ) ∪ Oi for any set Ui. Thus, the d-separation properties we are interested in are computed in the moralized graph H = (GAn(Oi∪κ(φ )))m, which is independent of Ui. Let U be the set of vertices v ∈ Oi such that there exists a path in H from a vertex u ∈ κ(φ ) \ Oi to v, with v the first vertex on that path that is in Oi. The set U can be constructed in linear time by a depth first search from κ(φ ) \ Oi. Take W = U ∪ (κ(φ ) ∩ Oi). |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Unfolding a program into a structured model tends to create a large number of timed variable in- stances whose associated value represents an equality between two variables. Such instances can be eliminated by a simple transformation of the structured model. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | For an assignment α with domain V , define α[y/x] to be the assignment α (cid:48) with domain (V \ {x}) ∪ {y} with α(y) = α (cid:48)(x) and α (cid:22) (dom(s)\{x}) = α (cid:48) (cid:22) (dom(s)\{x}). For a relational value s and variables x, y with x ∈ dom(s) and y (cid:54)∈ dom(s), define s[y/x] to be the relational value t with dom(t) = (dom(s) \ {x}) ∪ {y}, consisting of all assignments α[y/x] for α ∈ s. Intuitively, this is simply the relation s with variable x renamed to y. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | M[y/x] = (V (cid:48), E(cid:48), S (cid:48)) with V (cid:48) = V \ {x}, and E(cid:48) = E ∩ (V (cid:48) × V (cid:48)), and S = {s(cid:48) sv[y/x]. α(x) = α(y). |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Proposition 5 Suppose that M = (V, E, S ) is a structured relational model with x, y ∈ V , and Ωx = Ωy, and pa(y) = {x} and sy = δx,y. Let M[y/x] = (V (cid:48), E(cid:48), S (cid:48)). Then ⊗S (cid:48) = (⊗S ) ↓ V (cid:48). |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | The definition furthermore extends to epistemic models M = (A, O,V ) with worlds represented by a structured relational model M = (V, E, S ). Let M[y/x] = (V (cid:48), E(cid:48), S (cid:48)). We define M [x/y] = (A(cid:48), O(cid:48),V (cid:48)) i = Oi ∪ {y | x ∈ Oi} for each i ∈ Agts, and A(cid:48) = ⊗S (cid:48). Note that O(cid:48) where O(cid:48) = {O(cid:48) i additionally makes variable y visible to agent i if x was visible to i, in case this variable was not originally visible. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Proposition 6 If Ωx = Ωy, and pa(y) = {x} and sy = {(x : a, y : a) | a ∈ Ωx} then M , α |= φ iff M [y/x], α[y/x] |= φ [y/x]. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | with the existing implementation in MCK. (Since MCK remains the only symbolic epistemic model checker that deals with perfect recall knowledge, there are no other systems to compare to.) All experi- ments were conducted on an Intel 2.8 GHz Intel Core i5 processor with 8 GB 1600 MHz DDR3 memory running Mac OSX 10.10. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Except where indicated, the unoptimized model checking algorithm against which we compare is that invoked by the construct spec_spr_xn in the MCK scripting language, which implements the algorithm of van der Meyden and Su [21]. (We refer to this algorithm as xn in legends, and the algorithm using conditional independence optimization is referenced as ci.) The results demonstrate both significant speedups of as large as four orders of magnitude, as well as a significant increase in the scale of problem that can be handled in a given amount of time. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Dining Cryptographers: Our first example is the Dining Cryptographers protocol [7], discussed above. It was first model checked using epistemic logic in [21]. This example scales by the number n of agents; the number of state variables is O(n), and the protocol runs for 3 steps. The initial condition needs to say that at most one of the agents paid – this is done by means of a formula of size O(n2). The rest of the script scales linearly. The formula in all instances states that at time 3, agent C0 either knows that no agent pays, knows that C0 is the payer, or knows that one of the other agents is the payer, but does not know which. This involves O(n) atomic propositions, and is of linear size in n. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | case of 12 agents is handled in 0.05 seconds, and 100 agents are handled in 9.69 seconds. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | One-time Pad: The next example concerns message transmission using one-time pad encryption in the presence of an eavesdropper. Each instance has three agents (Alice, who sends an encrypted message to Bob, and Eve, who taps the wire). We scale the example by the length of the message, which is sent one bit at a time. For a message of length n, states have O(n) variables. The protocol runs 2n steps, two for each bit. The formula is evaluated at time 2n, and says that Eve does not learn the value of the first bit. For this example, we found that the best performance for the unoptimized version was obtained using MCK version 0.5.1, which used a different symbolic encoding of the transition relation from more recent versions. Performance of model checking is shown in Table 1. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | The running times for the optimized version grow very slowly. Intuitively, the conditional indepen- dence optimization detects in this example that the first bit and the others are independent, and uses this to optimize the model checking computation. This means that for all n, the ultimate BDD model check- ing computation is performed on the same model for all n, and the primary running time cost lies in the generation of the dependence graph, and its analysis, that precedes the BDD computation. On the other hand, the unoptimized (xn) model checking running times show significant growth, with a large spike towards the end, where the speedup obtained from the optimization is over 10,000 times. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Oblivious Transfer: The next example concerns an oblivious transfer protocol due to Rivest [24], which allows Bob to learn exactly one of Alice’s two messages m0, m1, of his choice, without Alice knowing which message was chosen by Bob. Each instance has two agents, and we scale by the length of the message. For a message of length n, states have O(n) variables. We consider two formulas for this protocol. Both are evaluated at time 3 in all instances. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | optimization allows us to handle significantly larger instances: up to 97 agents can be handled in under 200 seconds, compared with 19 agents in 170 seconds unoptimized. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | An example in which the optimization does not always yield a performance improvement arises when we change the formula model checked in this example to one that states that if Bob chose to receive m1, then he does not learn the value of any bit of m0. The running times are shown in Table 2(b). Here, the optimization initially gives a speedup of roughly one order of magnitude, but on the three largest examples, the performance of the unoptimized algorithm is better by a factor of two. The lower size of the initial speedup, compared to the first formula, can be explained from the fact that there are obviously fewer variables that are independent of the second formula, since the formula itself contains more variables. (The “all bits” formula contains O(n) rather than just one variable explicitly, but recall that knowledge operators implicitly introduce more variables, so the “first bit” formula implicitly has O(n) variables.) It is not immediately clear exactly what accounts for the switchover. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Message Transmission: The next example concerns the transmission of a single bit message across a channel that is guaranteed to deliver it, but with uncertain delay. This example has two agents Alice and Bob , and runs for n + 1 steps, where n is the maximum delay. States have O(n) variables. The formula considered asserts at time n + 1 that Alice knows that Bob knows ... (nested five levels) that the message has arrived. Because of the nesting, the algorithm used in the unoptimized case is that invoked by the MCK construct spec_spr_nested – this essentially performs BDD-based model checking in a structure in which the worlds are runs of length equal to the maximum time relevant to the formula. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Table 3(a) compares the performance of the conditional independence optimization with this algo- rithm. The degree of optimization obtained is significant, increasing to over four orders of magnitude. Running just the optimization for larger instances, we find that the optimization enables significantly larger instances to be handled in a given amount of time: as many as 65 agents in 342 seconds, compared to just 16 agents in 360 seconds for the unoptimized version. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | is trying to send bit 1. The protocol is more complex than the others considered above. An initial set of n “booking” rounds of the Dining Cryptographers protocol is used to anonymously attempt to book one of n slots, and this is followed by n “slot” rounds of the Dining Cryptographers protocol, in which an agent who has booked a slot without detecting a collision with another agent’s booking, uses that slot to attempt to broadcast a message. Because undetected booking collisions remain possible, collisions might also be detected in the second phase. Because of the complexity of the protocol, this example can only be model checked on small instances in reasonable time, even with the optimization. Table 3(b) shows the running times obtained: for the unoptimized version, we again used MCK-0.5.1. The running time of the unoptimized computation explodes at n = 5 as we increase the number of agents. The optimized computation takes significantly less time, but also eventually explodes, at n = 9. Thus, the optimization has doubled the size of the problem that can be handled in reasonable time. |
Processed_Optimizing_Epistemic_Model_Checking_Using_Conditio.txt | Wilson and Mengin [30] have previously related modal logic to valuation algebra, but their definition requires that the marginalization of a Kripke structure have exactly the same set of worlds and equiv- alence relation, and merely restricts the assignment at each world, so their approach does not give the optimization that we have developed, and a model checking approach based on it would be less efficient than that developed in the present paper. They do not discuss conditional independence, which is a key part of our approach. |
Processed_A_Note_on_the_Overall_Magnification_of_a_Gravitati.txt | Publications of the Astronomical Society of Australia c(cid:13) Astronomical Society of Australia 2021; published by Cambridge University Press. |
Processed_A_Note_on_the_Overall_Magnification_of_a_Gravitati.txt | Abstract The total magnification due to a point lens has been of particular interest as the theorem that gravitational lensing results in light amplification for all observers appears to contradict the conservation of photon number. This has been discussed several times, and various resolutions have been offered. In this note, we use a kinematic approach to provide a formula for the magnification factor for the primary image accurate to first order and valid for rays leaving the source at any trajectory. We thus determine the magnification over a sphere surrounding the system. A new result found is that while the magnification dips below unity far from the optical axis as noted by others, it returns to unity directly behind the source. |
Processed_A_Note_on_the_Overall_Magnification_of_a_Gravitati.txt | The magnification formula for the increase of light from a distant source when ‘lensed’ by an intervening massive object, was given eighty years ago in a small article by Einstein (1936). The derivation of this formula was later clarified by Refsdal (1964) and others. An interesting feature of the formula is that the magnification factor is arbitrarily large near the optical axis (the line on which the source and lens are situated), and falls away as the observer moves far from that line, but never falls below unity. This may appear to create a paradox in that if the magnification is always greater than one, there must be an increase in photon number due to the presence of the lensing object, but a lensing object such as a black hole does not create photons. For discussions of this paradox see for example, Schneider (1984) and Jaroszy´nski & Paczy´nski (1996). We note that Einstein and Refsdal make no mention of a paradox. Rather, they were only considering the region near the optical axis and so developed this formula as an appropriate approximation for that region only. |
Processed_A_Note_on_the_Overall_Magnification_of_a_Gravitati.txt | are largely in agreement with the results of Wucknitz (2008). However, we also show that very far from the optical axis, the magnification factor reaches a mini- mum, and then returns to unity as the observer moves to the point (on a sphere surrounding the source) fur- thest from the lens. |
Processed_A_Note_on_the_Overall_Magnification_of_a_Gravitati.txt | Figure 1. Magnification factor plotted against position on the observer’s plane. Specific values of the three constants have been chosen as follows: L1 = 10, L2 = 15, rs = 0.2. A horizontal line has been drawn where the image is neither magnified nor de- magnified (M = 1). The curve passing through the horizontal line is plotted using the full magnification formula. The curve staying above the line uses the small angle approximation. |
Processed_A_Note_on_the_Overall_Magnification_of_a_Gravitati.txt | Re-arrangement of this equation allows us to replace rs by LL1φ4/(2L2) in the full equation for the magnifi- cation, and confirm that at φ = φunity the magnification M = 1 + O(φ6). |
Processed_A_Note_on_the_Overall_Magnification_of_a_Gravitati.txt | In order to find the total magnification due to the PSPL system, we follow the approach of Wucknitz (2008), by extending the magnification map from an ob- server’s plane to a sphere surrounding the source. We will use kinematical ray equations, accurate to first or- der in the appropriate small parameter to derive the magnification on the surface of the sphere due to the single lensing mass. |
Processed_A_Note_on_the_Overall_Magnification_of_a_Gravitati.txt | Note that for an initial trajectory directly away from the lens mass φ = π, that is C1 = −1, both the terms above become equal to one, giving no change in magnification. The complete magnification over the sphere has been plotted in Fig. 2. The system is set up with an observer on the surface of a sphere with radius 200 centred on the source. There is a lensing object on the x-axis at x = 20 with ‘mass’ rs = 0.5. On the left side, magni- fication is plotted against the observer’s x-position. It can be seen that directly behind the source (that is, on the opposite side to the lens mass), the magnification is unchanged by the presence of the lens (M = 1). As the observer moves around the sphere, the source is slightly de-magnified to a minimum value, and then rises arbi- trarily high as the observer approaches the (positive) optical axis. A ‘side on’ view of the magnification over the sphere is shown in the right side of the figure. The parameters used are arbitrary and have been chosen to exaggerate the magnification so as to make it clearly vis- ible. In addition, since the magnification on the optical axis is infinite, here it has been truncated to a maximum of 1.01 in order to increase the level of contrast over the remainder of the sphere. In more realistic systems (with larger distances and a smaller lensing mass) the slight de-magnification, while real, would be so small as to be unmeasurable. |
Processed_A_Note_on_the_Overall_Magnification_of_a_Gravitati.txt | In this note, we have considered the magnification of a source by a lens having the property that it deflects light by an amount which is an approximation of the relativistic deflection of light. We have considered only the primary image. We briefly discussed the thin lens approximation, including the full magnification formula due to such an approach. It was seen that the tradi- tional formula is a small angle approximation to a more general formula. Extending the use of the small angle approximation to large angles gives rise to a paradox which does not exist if the more general result is used. That is, even the thin lens approach need not give rise to the paradox. |
Processed_A_Note_on_the_Overall_Magnification_of_a_Gravitati.txt | Figure 2. Magnification factor plotted against x-position on the observer’s sphere of radius 200, centred on a source at the origin with a single lens with rs = 0.5 located on the positive x-axis at x = 20. For the point-source-point-lens system it may be observed that the magnification is equal to unity directly behind the source, stays below one for most observer positions, but rises to infinity as observers approach the positive x-axis. The magnification value near the optical axis has been truncated to a maximum of 1.01. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.