components of simultaneous interpreting

Simultaneous interpretationThe simultaneous interpretation means translating the speaker’s speech almost simultaneously, which is used for international conferences, seminars and so on. Simultaneous interpretation makes a feature of professional and academic. Some people may think simultaneous interpretation is just a simple thing when you are proficient in the language and are quick-minded. However, if you can’t grasp the real meaning of the speech and just do simple conversion, you may fail.I think, to be a good simultaneous interpreter, you should have the following abilities.To begin with, memory ability is very important. Memory ability includes short-term memory and long-term memory. And short-term memory creates the probability of simultaneous interpretation. The interpreter should memory the sentences in the limited time, or the short-term memory will disappear quickly. But the interpreter needs long-term memory to get rid of simple conversion. And also, long-term memory can help interpreters grasp the spokesman’s meaning well and truly.Then, interpreters should have the ability to predict the sentences and understand the speech in advance. Although the interpreter speaks after the spokesman, they also can catch up with the speed of spokesman. The decisive factors of prediction ability are language skills and the scope of knowledge.Furthermore, strain capacity is necessary in simultaneous interpretation. The interpreter can’t avoid suffering unexpected difficulties. For example, when missing some words, the interpreter should complement the sentence according to the specific circumstance. When the prediction is opposite to the original sentences, the interpreter should not translate again but add an explanation in time.Finally, we talk about storage capacity. In order to insure the fluency of simultaneous interpretation, the interpreter should store one or more conception that spokesman refers to. Because of there are many differences between Chinese and English, interpreters usually need adjust the structure of sentences. As a result, the storage capacity is very important for interpreters.In conclusion, to make simultaneous interpretation accurate, clear and integrated, the interpreter should cultivate the four abilities above. But these abilities are the foundation of simultaneous interpretation. Also, how to practice is very important. There are two model of practicing simultaneous interpretation. The first is Jill Model, which is SI=L+M+P+C. It means simultaneous interpretation depends on listening and analysis, short-term memory effort, speech production and coordination. Another model is Xiamen University Model. It is more difficult than Jill Model. And I want to talk about it in detail.Xiamen University Model says the analysis of discourse and cross-cultural communication is important. Only when you understand the differences of culture, you can understand the meaning exactly. And you should have good comprehension in source language and the knowledge. Moreover, skill and professional standard play an important role. The interpreter should abide by the criteria of interpreting. The lastbut not the least, interpreters should reconstruct the target language to complete the task. According to this model, the practice of interpreting should base primarily on the training of relevant skills. Whatever methods you use to practice, the perseverance is indispensable.。

Synthesis and Simulation of Digital Systems Containing Interacting Hardware and Software ComponentsRajesh K.Gupta Claudionor Nunes Coelho,Jr.Giovanni De MicheliCenter for Integrated SystemsStanford University,Stanford,CA94305.AbstractSynthesis of systems containing application-specific as well as re-programmable components,such as off-the-shelf microprocessors, provides a promising approach to realization of complex systems using a minimal amount of application-specific hardware while still meeting the required performance constraints.We describe an approach to synthesis of such hardware-software systems starting from a behavioral description as input.The input system model is partitioned into hardware and software components based on im-posed performance constraints.Synchronization between various elements of a mixed system design is one of the key issues that any synthesis system must address.In this paper,we consider software and interface synchronization schemes that facilitate communica-tion between system components.We present tools to perform synthesis and simulation of a system description into hardware and software components.In particular, we describe a program,Poseidon,that performs concurrent event-driven simulation of multiple functional modules implemented ei-ther as a program or as behavioral or structural hardware models. Input to Poseidon consists of description of interacting functional models with their respective clock cycle times and the interface synchronization scheme chosen by the partitioner.The resulting software component is assumed to be implemented for the DLX machine,a load/store microprocessor.We present simulation ex-amples and design of a graphics controller demonstrate the fea-sibility of mixed system synthesis.1IntroductionIn this paper we consider an approach to extend the high-level synthesis techniques to synthesize system designs using application-specific and reprogrammable components.We refer to the application-specific component as the hardware component, while the program running on the reprogrammable component as the software component.Our approach to system synthesis us-ing hardware and software components is inspired by the fact that in practice most systems used in embedded control and telecom-munication applications consist of application-specific hardware components as well as reprogrammable components.While most 0digital functions can be implemented by software programs,a ma-jor reason for building dedicated ASIC hardware is satisfaction of performance constraints.The performance constraints can be on the overall time(latency)to perform a given task or on the in-put/output data rates.Typically,the pure software implementations of a system design are often too slow to meet the imposed per-formance constraints.Therefore,specialized hardware chips are often needed to complement or assist the reprogrammable compo-nent on certain performance-critical tasks.Further,mixed system designs reduce the size of the synthesis task by reducing the num-ber of application-specific chips required while at the same time achieving theflexibility of software reprogramming to alter system behavior.Such aflexibility is also important in achieving rapid prototyping of complex system designs where non performance-critical and unconstrained operations are shifted into a program running on an off-the-shelf microprocessor.Whereas the focus of high-level synthesis techniques thus far has been to generate a purely hardware implementation of a system design either as a single chip or as an interconnection of multiple chips,each of which is individually synthesized[1][2][3][4], attempts at system synthesis using both hardware and software components have been rare and limited to developing frameworks for facilitating the design process[5].The problem of synthesis of mixed systems is fairly complex. There are many subproblems that must be solved before an ef-fective synthesis system can be developed.Among the important issues are the problems of modeling of system functionality and constraints,determination of the boundary between hardware and software components in the system model,specification and syn-thesis of the hardware-software interface,and implementation of hardware and software components.In this paper,we summarize a systematic approach to automatic synthesis of mixed systems and focus on the issue of interface design and the synchroniza-tion mechanisms that are used to facilitate communication between system models and ensure correctness of system functionality.Figure1shows organization of the CAD design system used for synthesis of mixed system designs.The input to our synthe-sis system is an algorithmic description of system functionality. We model system behavior using the HardwareC[6]language that has a C-like syntax and supports timing and resource con-straints.HardwareC supports specification of unknown delay op-erations that can arise from data dependent decisions and external synchronizations.The HardwareC description is compiled into a system graph model based on data-flow graphs[6].The sys-tem graph model consists of vertices representing operations,and edges which represent either a data dependency or a control depen-dency.Overall the system graph model is composed of concurrent data-flow sections which are ordered by the system controlflow. The data-flow sections simplify the tasks of identification of con-currency in the system model,while use of control constructs suchFigure 1:System Synthesis Procedureas conditionals and loops obviate the need for a separate descrip-tion of control flow.Associated with input/output statements,we specify corresponding constraints on input/output data rates.The input (output)rate constraints refer to the rates at which the data is required to be consumed (produced).The system graph model is input to Vulcan-II which partitions the system graph model into portions to be implemented either as dedicated hardware modules or as a sequence of instructions on a reprogrammable processor based on feasibility of satisfaction of externally imposed data-rate constraints.System partitioning constitutes an important phase of the system synthesis process.However,it is not the intent of this paper to delve into the partitioning issues.For an approach to system partitioning the reader is referred to [7].Hardware synthe-sis of the models identified by Vulcan-II is performed by program Hebe [6].For synthesis of the software component,we generate a corresponding C-description after generating a total order of oper-ations in accordance with the partial order imposed by the graph model.The C-code is then compiled into assembly code for the target processor using existing software compilers.The interface synthesis is performed by Vulcan-II under timing constraints im-posed on the system model.At the present time,Vulcan-II is a framework to carry out various synthesistasks,and it is not yet an automated tool.Figure 2:Target System ArchitectureTarget System ArchitectureFigure 2illustrates the broad features of the system architecture that is the target of our system synthesis approach.The target architecture consists of a general-purpose processor assisted by application-specific hardware components.The memory used for program and data-storage may be on-board the processor.How-ever,the interface buffer memory needs to be accessible to the hardware modules directly.Because of the complexities asso-ciated with modeling hierarchical memory design,in this paper we consider the case where all memory accesses are to a single level memory,i.e.,outside the reprogrammable component.The hardware modules are connected to the system address and data busses.Thus all the communication between the processor and different hardware modules takes place over a shared medium.Further,the mechanisms of data transfer between processor and application-specific components are constrained to those supported by the processor.Concurrently executing hardware modules generate data that is consumed by the program(s)running on the processor.Due to the inherent serialization of operations implemented on the proces-sor,any implementation must ensure that the data transfer across components takes place in an efficient manner that reduces the associated area and time overheads.The data transfer between the program(s)and the hardware modules is facilitated by the hardware and software synchronization mechanisms used.The choice of a data transfer scheme is determined by the individual execution rates of different models and control model used for scheduling various components.For example,a blocking transfer protocol may ensure correctness of the data-transfer but it may impose undue overheads on speed of execution of hardware mod-ules.Whereas a non-blocking transfer may starve or overrun the interface buffers.In the following sections,we outline our approach to system synthesis,and discuss different synchronization mechanisms used for synthesis for software and hardware components.The issue of synchronization between operations in a system model is closely related to the issue of communication between operations.For this reason,a choice for synchronization scheme is influenced by the selection of the communication scheme between system compo-nents.We first present the communication model supported in our synthesis system and then describe the synchronization schemes in context.We then address issues related to design of the hardware-software interface.2System SynthesisSynthesis of application-specific hardware components under tim-ing and resource constraints requires generation of a schedule of operations which satisfies the imposed timing constraints,an allo-cation of hardware resources that satisfies the resource constraints and finally the construction of a suitable control to facilitate hard-ware execution.Our model of hardware supports specification of data-dependent operations.Since data-dependent operations may offer unbounded delays it becomes necessary to schedule these operations dynamically.Therefore,we refer to data-dependent de-lay operations as points of synchronization in the system model.Our approach to synthesis of hardware under relative scheduling formulation has been described in detail elsewhere [6].Briefly,the relative scheduling formulation makes it possible to achieve a data-driven dynamic schedule of operations with respect to a set of synchronization points (also referred to as anchors in [6]).Here we focus on the problem of synthesis of the software component of the target system design.The software component is implemented as a program running on the onboard processor,i.e.,the reprogrammable component.We assume that this pro-gram is small enough that it can be mapped to real memory,sothat the issues related to virtual memory management can be ig-nored.As indicated in Figure1,we start with a partition of the system graph model.System partitioning into hardware and soft-ware components is performed under the constraint that specified system input/output data rates can be supported by thefinal system implementation.One such partitioning approach relies on identi-fying and partitioning unbounded delay operations[7].As a result of system partitioning we essentially have a set of concurrently ex-ecuting hardware and software models.The software component consists of a set of concurrently executing routines,called threads. All threads begin with a point of synchronization and as such these are scheduled dynamically.However,within each thread of exe-cution all the operations are statically scheduled.Therefore,for a given reprogrammable component the latency of each thread is known statically.As an example,data-dependent loops in soft-ware are implemented as a single thread with a data-dependent repeat count.The problem of concurrent multi-thread implementation is well known[8].In general,the program threads may be implemented either as a subroutines to a global task scheduler or as coroutines. However,in the context of mixed system designs where the pro-cessor is completely dedicated to the implementation of the system model and all software tasks are known statically,it is possible to use simpler and more relevant schemes to implement the soft-ware component.In the following,we present two schemes for implementation of the software component.Software Implementation as CoroutinesCoroutines provide an attractive means of achieving concurrency between various program threads by reducing the cost of switch-ing execution from one thread to another[9].In this scheme, the reprogrammable component runs a task scheduler based on a priority assigned to various routines which are maintained in a co-operative,rather than hierarchical,relationship to each other. Each coroutine maintains a local state and willingly relinquishes control of the execution machine at points of synchronization. Coroutines provide a limited form of message passing via follow-ing two primitive operations:resume and detach.A coroutine switch consists in saving the current machine status and restoring the machine status of the next process to be executed.In the most general case,where any interruptions or exceptions may cause a context switch,all machine registers andflags should be saved. In case of an R/M processor,that is a processor that provides in-structions with a register and memory operands such as8086,the code for a coroutine based scheduler amounts to34instructions taking about100bytes.The coroutine switch takes364cycles when implemented for8086processor.By contrast,implementa-tion of a global task scheduler using subroutines takes728clock cycles for the8086processor[10].It is possible to reduce the overhead due to context switch if all the coroutine switches are explicit and known at the compile time. By making sure that during code optimization,variable lifetimes do not cross the coroutine boundaries,then the only register that needs to be saved is the program counter of the current corou-tine and also only register that should be restored is the program counter of the next coroutine to be executed.The code for a such a scheduler on8086processor takes103cycles for each context switch.By comparison,on an load/store(L/S)machine,such as DLX[11],the code for task scheduler is reduced to17instructions (19machine cycles),as opposed to the general case when all64 registers would have to be saved requiring192instructions.Software Implementation using Case DescriptionsIn this approach,we merge different routines and describe all op-erations in a single routine using a method of description by cases [12].This scheme is simpler than the coroutine scheme presented above.Here we construct a single program which has a unique case assignment for each point of synchronization.Thus each thread now corresponds to a case description of a rather large conditional in thefinal program.A global state register is used to store the state of execution of a thread.This method is restrictive since it precludes use of nested routines and requires description as a single switch statement,which in cases of particularly large software descriptions,may be too cumbersome.Overhead due to state save and restore amounts to85clock cycles for every point of synchronization when implemented on a8086processor.Con-sequently this scheme entails smaller overheads when compared to the general coroutine scheme described earlier.Corresponding overheads for the DLX processor amounts to35clock cycles for every point of synchronization.Figure3:System Synthesis ExampleIn order to illustrate our system synthesis approach we con-sider synthesis of a graphics controller that provides for drawing of lines and circles given the end coordinates(and radius in case of a circle).Figure3illustrates some of the steps in synthesis of the graphics controller.The HardwareC description consisting of 457lines of code is input to the behavioral synthesis phase.The resulting system graph model is input to Vulcan-II.As a result of system partitioning and program threads generation in Vulcan-II, the system design at this stage consists of interacting hardware modules modeled by the hardware graph models and a software component modeled by program threads.Next step is to synthe-size the interface circuitry that would facilitate synchronization and communication between heterogeneous system components. Synthesis of interface circuitry is driven by the requirements im-posed by system synchronization.We shall revisit this example in Section4to show how multiple program threads are synchronized with the concurrently operating hardware portions.3System SynchronizationA system design consists of various components which carry out operations in response to input data.An event refers to the ex-ecution of a data input/output operation.Synchronization in a system design refers to constraints on system design that ensure the partial ordering of events and operation executions imposed by the system model must be observed in any execution trace of the system model.Some synchronization constraints are needed to ensure correctness of the execution model,for example,all the data generated within the system model must be consumed in the time order in which it was generated.Typically this is guar-anteed by appropriate choice of the execution semantics for the system model.Additional constraints may be needed to ensure correctness of a set of concurrently executing models.Further, some synchronization conditions may be externally imposed.Forexample,a certain precedence or simultaneity condition between execution of two operations imposed by the system controlflow.Communication ModelIn the system graph model,communication between two opera-tions is indicated by presence of an edge between respective op-eration vertices.When considering hardware synthesis,an edge between two operations may translate into either a physical wire connection,or it may be buffered and/or blocked to facilitate asyn-chronous communication.Final selection of data-transfer mecha-nism is made based on the data transfer requirement and how in-dividual communicating models are implemented.However,note that in a mixed system implementation,due to inherently different rates of computation between hardware and software modules,it is necessary to allow multiple executions of individual models in order to achieve high system throughput.However,in presence of variation in rates of communication across different models ap-propriate buffering and handshake mechanisms may be required.3.1Software SynchronizationOur model of software component relies on the sequential execu-tion of different threads of execution.Due to this serialization of the input system model,software synchronization is needed to en-sure correct ordering of operations within the program threads and between different threads.A thread of execution already main-tains an order of execution of its instructions,so a schedule of the operations is implicit to the sequential execution model of the instructions in a reprogrammable component.This solves the problem when a single thread of execution can be found for an entire description or among operations implemented in software belonging to the same thread-synchronization is only needed in points of synchronization and where the control is transferred be-tween software and hardware.When data-dependent loops,and asynchronous message passing are present in the code,it may not always be possible tofind a static schedule of the operations. If the order of execution can still be found,a single thread of execution could be determined that preserves the order in which the operations are executed.In case no such thread of execution can be determined,multiple threads of execution are required.In presence of multiple threads of executions(whether implemented as multiple programs or a single program using case descriptions described before)software synchronization consists of a mecha-nism to transfer control from one thread to another.In case of small number of threads,such a transfer can be done based on a statically defined priority of threads.For example,in case of two threads,control would simply switch from one thread to the other.In the general case,however,due to unbounded delay op-erations,we look for a dynamic scheduling of different threads of execution.Such a scheduling is done based on availability of data.Suppose we were to time stamp each data generated and also for each data request.Then the next thread of execution to be scheduled would be the one with the smallest request time stamp. Further,in order to maintain the correct order of data production and consumption,at any time the data being consumed is the one with the smallest time stamp.Such an scheme is implemented using a control FIFO that contains pointer to the next thread to be scheduled for execution[7].Data transfer between two threads of execution can be implemented with shared memory or mes-sage passing.Shared memory can be facilitated by maintaining read and write pointers on each data-transfer.Such an scheme would add the overhead of maintaining and updating the read and write pointer for each data transfer across the program threads. Non-register based data-transfers(or data transfers which could culminate in control transfer)are well suited to be implemented as a queue connected with the control FIFO.On the other hand, register based transfers have the characteristic that once something is written,the data may be read many times.It is possible to use processor registers to transfer information between threads.How-ever,such a scheme requires global register assignments which are not available for reassignment by the compiler.A limited form of message passing can be achieved by using co-routine model of implementation described before.3.2Hardware-Software Synchronization Synchronization between hardware and software components is determined by the data transfer requirements between the sender and the receiver.A data transfer between two models can be either blocking or non-blocking.A blocking transfer protocol requires the sender(receiver)to block transfer until the corresponding re-ceiver(sender)is ready to receive(send)data.Blocking can also be made conditional so as to reduce the associated timing penalties due to blocking[13].With respect to their overheads,a non-blocking transfer consumes the system bus bandwidth,whereas a blocking transfer costs not only system bus bandwidth but also additional control lines that are needed to implement the required handshake.Therefore,for lower system costs,it is necessary to implement blocking only when absolutely necessary.A blocking transfer protocol can be thought of as a non-blocking transfer with an infinitely deep queue buffer.The queue size may be bounded by addition of handshake signals that treat queue as the sender or receiver of data.Alternatively,in presence of specific constraints on rates of data transfer,the queues can be sized algorithmically [14].For a given data-transfer edge in the system graph model,we first attempt to determine the rates of data production and con-sumption associated with the sender and receiver models.Such a rate determination requires specification of data rates for ex-ternal inputs and outputs.In case of unknown or varying data rates,a blocking protocol for both sending and receiving ends is selected.Either sender or receiver end of a transfer can be made non-blocking if it can be determined that the corresponding oper-ations are always slower.In case of perfectly matched data-rates a synchronous non-blocking protocol is selected.After selecting transfer protocols for different data-transfers across the hardware and software models,the interface circuitry can be synthesized us-ing asynchronous and synchronous logic synthesis techniques[15] [16].For a description of the interface architecture the reader is referred to[7].SLIF Netlistb. event-driven simulation of multiple modelsc multiple clocks and clock rates between modelsFigure4:Event-driven simulation of a mixed system design 4Simulation of Hardware-Software SystemsWe have developed an event-driven simulator,named Poseidon, that performs concurrent simulation of multiple functional modelsInRqOutAkoutPortFigure5:Simulation Example1implemented either as a program or as application-specific hard-ware.The software component is compiled into the assembly code of the target microprocessor.Poseidon currently supports simulation of assembly code for the DLX microprocessor,a RISC oriented load/store processor[11].The hardware component of system design can be simulated either before or after the struc-tural synthesis phase.The graph model before structural synthesis is simulated using program Ariadne.A gate-level description of the hardware component of system design is generated using struc-tural synthesis techniques in program Hebe and simulated using program Mercury.Thus,Poseidon supports simulation of partially synthesized hardware modules along with the software component of the system design.Poseidon maintains an event queue which stores all simulation models sorted by their activation times.After simulating an event,the event is enqueued in the event queue.A system specification in Poseidon consists of following parts:1.Model declarations:consists of declarations of the concur-rently executing simulation models.Models can be eithersoftware or hardware models.Each model has an associatedclock signal and clock cycle-time used for its simulation.Itis assumed that the clock cycle-times are a rational multipleof each other.Further it is assumed that different modelssupply(latch)data at the interface usingflip-flops at the in-terface edge-triggered by their respective clock signals.2.Model interconnections:The interface between differentsystem components is specified by connections among mod-els.A connection between two models may be either a directconnection through a wire,or a port connection through aregister or a queue.Queues can have multiple fanins andfanouts.Signal assignments indicate direct connections be-tween respective models.For connections such as queuesthat require existence of additional control signals for syn-chronization,it is possible to group signals having identicalsynchronization requirements together for a given set of syn-chronization signals.munication protocols:Interface protocol for data-transfer is specified via guarded commands[17].A guardedcommand is executed only when some precondition is true.Each precondition is specified as a logic equation of signalvalues and transitions.There are four commands recognizedby the connection types.Enqueue and dequeue are usedfor queues port connections and load and store are used forregister port connections.4.System outputs:Outputs to be observed during simulationruns may be indicated by direct connections to the internalsignals in the system model.For illustration purposes,we consider a simple example of two models,Producer and Consumer connected by means of a finitely sized queue as shown in Figure5.We consider two cases: one in which the producer model is implemented in software and consumer in hardware and the other in which producer,consumer implementations are reversed.Example1shows system specifi-cation for this example for thefirst case.The threefirst lines of the specification declare the models to be simulated.Model io models the external system inputs and outputs.The following pa-rameter specifies the clock period of the clock signal associated with the respective model.A value of3.0for the consumer modelPROCESSOR ASIC Hardwareint lastPC[MAXCRS]={scheduler,circle,line,main);int current=MAIN;int *controlFIFO = (int *) 0xaa0000;int *controlFIFO_rq = (int *) 0xaa0004;main(){resume(SCHEDULER);};int nextCoroutine;scheduler(){resume(LINE);resume(CIRCLE);while (!RESET){do {nextCoroutine = *controlFIFO;} while ((nextCoroutine & 0x4) != 0x4);resume(nextCoroutine & 0x3); }}Figure6:Example2:Graphics Controller Designindicates that consumer is implemented in an ASIC technologythat uses a clock signal that is three times slower than the clockused by the reprogrammable component,which is usually a cus-tom designed component.The system input/outputs are sampledhere at the same rate as the consumer.The last two parameters specify the directory location where the model description canbe found and the model name.The queue statement declaresa queue named,comm,which is4bits wide and3words deep.We use rq and ak signals to implement a blocking communi-cation protocol as indicated by the guarded commands.A‘+’suffix indicates rising edge transition of the corresponding signal.A‘-’suffix indicates falling edge transition.Symbols‘&’and‘!’indicate the boolean and and not operations.Example1:Specification of a producer-consumer pair(Figure5).#Modelsmodel IO io 1.0/local/ioDir IO;model P dlx 1.0/local/ProducerDir Producer;model C mercury 3.0/local/ConsumerDir Consumer;#Connectionsqueue[4]comm[3];C.RESET=IO.RESET;C.r[0:0]=IO.r[0:0];#Communication protocolP.0xff004[0:0]=!comm.full;C.b_rq=!comm.empty;when(P.0xff000_wr+&!comm.full)do comm[0:3]enqueue P.0xff000[0:3];when(C.b_ak+&!comm.empty)do comm[0:3]dequeue C.b[0:3];#OutputsIO.inChannel[0:3]=P.0xff000[0:3];IO.outPort[0:3]= C.c[0:3];IO.InRq=P.0xff000_wr;IO.OutAk= C.b_ak;In order to illustrate the effect of software and hardware-software synchronization mechanisms we now consider the de-sign of the graphics controller introduced in Figure3.Figure6 shows thefinal implementation of the system design.The design consists of application-specific portions containing initial coordi-nate generators and control logic for controlFIFO and a software portion implemented on the DLX processor.The software com-ponent consists two threads of execution corresponding to the lineand circle drawing routines.Both program threads generate coor-dinates that are used by the dedicated hardware.Input to Poseidon consists of gate-level description of the ASIC hardware,assemblycode of the coroutines,and a description of the interface.Exam-ple2shows the Poseidon interface specification of the graphics controller design.。

1NUMERICAL SIMULATION FOR THE SIMPLE SHEAR TEST INAMORPHOUS GLASSY POLYMERSHu Ping, Wang Demin and Liu Yuqi(Institute of Automobile Panel Forming Technology,Department of Applied Mechanics,Jilin University of Technology,Changchun, 130025)Abstract In this paper, a driving stress finite element method of elastic-plastic large deformation based on implicit time integrating algorithm and an eight-chain molecular network model are used for the numerical simulation of the simple shear test of polycarbonate materials. The simulated results are compared with experimental ones. The strain localization propagation for the shear band deformation for simple shear deformation is investigated numerically. The effects of microstructure parameters in the model on strain softening and orientation hardening of the PC are discussed in detail.Key Words : Driving stress FEM; Eight-chain molecular network model; Polycarbonate; Plane strain simple shear.1. IntroductionDeformation behavior of polymeric materials is obviously different as compared with that of metallic materials due to the dramatic stiffening of polymers at large strain, which is associated with molecular chains orienting and stretching during the deformation process. It is well known that the plastic instability like necks or shear bands in metals almost invariably tends to localized failure. However, the necks and macroscopic shear bands in polymers do not develop to localize but tend to propagate along the specimen [1,2]. The propagation of localized deformation is fundamental to the processing of polymers since the permanently oriented state generally has enhanced properties, but can also affect subsequent rupturing.From the physical point of view, the typical deformation behavior of amorphous glassy polymeric materials can be explained by the alignment of molecular chain. However, the quantitative understanding of the macro viscoplastic localization in polymers is still in its infancy, since the fact that appropriate constitutive theories and numerical simulation technique have been developed only fairly recently.In this paper, the Boyce’s 3-D constitutive model of large inelastic deformation of amorphous glassy polymers [3] is first briefly recapitulated ; and then the Eight-chain BPA model is introduced into the driving stress (DS) finite element method [4,5]. Finally, the shear banding propagation in simple shear test of plane strain block is simulated numerically for the Polycarbonate(PC) materials. Numerical results predicted by the present DS finite element method are partly compared with experimental results found in the literature in order to examine the validity of the numerical scheme. Here, attention will be focused on the Influence of microstructure parameters on nondimensional shear and normal stress responses2. Constitutive Equations and Finite Element ModelThe intermolecular resistance to plastic flow is considered to be due to the impedance imposed by neighbouring chains on the ability of a chain segment to rotate either individually or in a cluster. Based on this above assumption, Boyce et al. [3] extended Argon’s expression [6]for the2plastic shear strain rate &γP&&ex p [~{(~)}]γγτp A S T S=−−0561 (1)where &γ0 is constant deformation rate factor, τ is the applied shear stress, A is a material constant that is proportional to the activation volume/Boltzmann’s constant, T is the absolute temperature, and ~S S p =+α (2) wher e p is pressure, αis a pressure-dependent coefficient; and S is the shear stress which evolves with the plastic strain from S 0 to a stable value S ss , and depends on temperature and strain rate. S 0(= 0.077G/(1－µ)), is the athermal shear strength, G is the elastic shear modulus, and µ is Poisson’s ratio. Furthermore, S is assumed to evolve with plastic strain via&()&S h S S ssp =−1γ (3)here, h is the rate of resistance drop with respect to the plastic strain. In the above micro-material parameters, an important parameter is S ss , which, in general, evidently affects deformation behavior and softening character of polymeric materials. Once the polymeric material is stressed to exceed its intermolecular resistance to chain motion, the molecular chains align along the principal direction of plastic stretch. This alignment generates an internal network stress very similar to that of rubber network. Haward and Thackray [7] suggested to describe this stress by means of a back stress determined through a Langevin spring; and then Boyce et al.[3] extended this approach to general 3-D plastic deformations by introducing the back stress B , whose principal components for the eight-chain model are B i C R NV i p c h c h c h N =−−3221λλψλ() (4)and λch V p V p V p 213122232=++() (5)ψβββλ()coth =−=1chN (6)where V i P is principal plastic stretch, C R is a micro constant,N means the locking stretch in tension, ψλβ−=1(c h Nis the inverse function of the Langevin function . The deformation gradient F from the initially isotropic state of the polymer can be multiplicatively decomposed into elastic and plastic parts. According to Lee [8], the elastic part F e is taken to be symmetric, and the plastic part F P presents the relaxed configuration obtained by unloading without rotation and permanent orientation of molecular alignment. The magnitude ofthe plastic deformation rate tensor &dp is assumed to be given by representative plastic shear strain rate &γp [9], and the direction is specified by the normalized deviatoric part of the driving stress T ∗, that is,&&*'d p T =γτp 2 (7)3T T F e F BF *det()=−−11e e (8)τ=[()]**1212tr T T (9)where, T is Cauchy stress tensor, and B is back stress tensor with respect to principal values B i in Eqn.(4). The shear stress τin Eqn.(1) is estimated by τin Eqn.(9). Assuming that the elastic strains remain small and neglecting the all geometrical changes associated with the elastic part of deformation, the elastic plastic constitutive equation with rate form can be established by introducing Eqn.(7) into the hypo-elastic rate formS L d L d d L d D ∇==−=−e e e p e p &(&&)&& (10)whereS ∇ is the Jaumann rate of Kirchhoff stress S ，and L e is the fourth-order isotropic elasticmodulus tensor, and &&DL p e p =γT *'/2τacts as an instantaneous stress rate term that represents the visco-plastic contribution. Introducing eqn.(10) into the driving stress finite element method of elastic-plastic large deformation [10] , we obtain the following finite element equation based on implicit time integrating algorithmKv f f &&&=+t P (11)in which, the expresses of matrix K and vectors &f t and &f P can be found in Ref.[5].3. Comparison Between Experimental Results and Computational Predictions In order to examine the validity of the present finite element method, we have first finished a direct comparison between the current simulated results for the plane strain simple shear and the experimental data [11] for polycarbonate (PC). In the experiments done by Gopez [11] and G’Sell [12] ,specimens were designed to create a uniform state of plane simple shear in the gauge section.However it was found that due to some inhomogeneity in the material, a highly localized shear band appeared near the center of the specimen, which subsequently grew and widened in the direction perpendicular to the shear direction. In order to reveal the propagation of such a macroscopic shear band observed in the experiments, the following initial imperfection ∆S 0of the shear strength S o was assumed :{}∆S x L L H y H L H 002002020202001=−−−+×−×+ξexp [()(/)()][.()] (12)The specimen geometry is illustrated in Fig.1, in which the specimen’s width (2H 0) is equal to 4mm, length (2L 0) 60mm and the specimen is divided into a regular mesh with 10*100 elements,by the homogeneous relative glide of parallel planes along the shear direction. The amount of shear is commonly expressed by the ratio Γ=u H /()20,where u is the relative displacement of the opposite faces of the parallelepiped. The loading condition is set to take a constant velocity&./u m s =×−12105,here &&u H =×20Γ, corresponding to an applied shear rate of &Γ=×−−31031s .According to Gopez and G’Sell experimental data, the material parameters in eqn. (1)~(6) are taken as follows:T=294K , S 0=97MPa, µ=03., E=2300MPa, h/S 0=5.15, S ss =89.29, A=240K/MPa,N=6.3, C R =5.7MPa, α=0.08,ξ=0005.，&.γ02010151=×−s .The deformation mode is of a 2-D plane strain type in the present numerical simulation, because variations of the thickness during the deformation are relevant only to second-order terms in the strain tensor. Fig. 2 shows the predicted stress responses to the plane strain simple shear. The numerical results are presented in terms of the nondimensional shear stress F s/L0, here F s means the applied shear force. It can be found from Fig.2 that the predicted shear stress is in a good agreement with the experimental data. The six deformation configurations corresponding to the points A~F in Fig.2 and the contour curves of the normalized effective plastic shear strain, which means Γ/&)（2γp, at the six deformation stages have been presented in Fig.3. It was clearly seen that at the B stage, there are two high shear strain distribution regions appeared near the center of the specimen with respect to the initial imperfection ∆Sand near the two freeedges with respect to free surfaces. As the deformation develops, the two high shear strain regions tend to coupling, and finally form an obvious shear band. The predicted initial and subsequent propagation of the shear band is found to be in good agreement with experimental observations[12]. Therefore it can be concluded that the present constitutive model and the numerical method used here are capable of capturing the major characteristics of the large simple shear behavior in a qualitatively reasonable way.4. Simulation and Discuss of Softening and Hardening ParametersIt is generally admitted that the yield and post-yield deformation behavior of glassy polymers show orientation hardening and true strain softening. In the Eight-chain BPA model, the influence of the strain softening and the orientation hardening can be described in terms of the parameters S ss and C R , respectively.Considering three kinds of cases with S ss/S0=0.85、0.94 and 1.0, where S ss/S0= 1.0 means the limiting case of no softening. The influence of softening on the predicted nondimensional shear stress and normal stress (F n/L0, here F n means the applied normal force) is numerically simulated and the simulated results are shown in Fig. 4, from which it is observed that as excepted, the parameter S ss/S0 determines the extent of softening. It should be mentioned that in the limiting case of no softening, the simulated results of plastic shear strain shows no initiation and propagation of a shear band.From Eqn.(4) it can be seen that the parameter C R governs the orientation hardening through the non-Gaussian network model for rubber elasticity. Therefore, the simulation and study of the influence of the orientation hardening are schemed by taking three kinds of cases with C R/S0=0.0, 0.1 and 0.2, respectively, in which C R/S0=0.0 means the limit case of no orientation hardening. In all these simulations, the intensity of softening is assumed to be S ss/S0=0.85. The simulated results are shown in Fig.5. It is found from the predicted shear stresses that a larger value of C R/S0increases the stiffness of the network and therefore increases the orientation hardening. The value of C R/S0 is found to affect the apparent overall yield stress as well. The reason of this is that at the yield strainΓ=0143., obvious plastic deformation has occurred at some locations due to the inhomogeneity of deformation. If C R is large, orientation hardening is already significant, resulting in a noticeable increment of the shear stress responses. The normal stress responses with different values of C R/S0 ,in fact, give the identical predictions up to Γ=022., where they approach the maximum compressive stress. From that point on, the normal stress for no hardening increases monotonically and becomes tensile at an applied shear strain of 0.267. As opposed to this result, the normal stress response for large values of C R/S0=0.1 or 0.2 is not4monotonic but gradually reaches a maximum and then drops monotonically again. In this case of C R/S0=0.2 , the normal stress response maintains compressive during the entire deformation process. The effects of orientation hardening on initiation and propagation of shear band have also been studied. In the limiting case of no orientation hardening, the fully elongated shear band will localized and the applied strain will concentrate entirely in the band.ConclusionsIn this paper, a 2-D driving stress finite element method combining the microstructure and macrostructure character of amorphous glassy polymers has been used to simulate large simple shear tests of polycarbonate. The shear band correspondence of simple shear simulation with experimental ones shows the validity of the present finite element formulation. The focusing attention is on the effects of strain softening and orientation hardening .It is found from the present numerical simulation results that the parameters S ss/S0 and C R/S0 determines the extent of softening and orientation hardening. A larger value of S ss/S0 decreases the softening in PC, and there is no evidence of initiation and propagation of a shear band in the limiting case of no intrinsic softening. On the other hand, a larger value of C R/S0increases the stiffness of the network and therefore increases the orientation hardening. Acknowledgement:The authors acknowledge the financial support of National Natural Science Foundation of China.REFERENCES[1]. G’sell C, Boni S and Shrivastava S. Application of the plane simple shear test for determination of theplastic behavior of solid polymers at large strains. J Mat Sci, 1983,18:903.[2]Neale K W and Tugcu P. Analysis of necking and neck propagation in polymeric materials. J MechPhys Solids, 1985,33:323.[3]Arruda E M and Boyce M C. Evolution of plastic anisotropy in amorphous polymers during finitestraining. Int J Plasticity, 1993, 9: 697.[4]Hu P and Tomita Y. Deformation localization for plane strain tension of polymers. Acta Mechanica Sinica,1996, 28(5):564.[5]Hu P, Liu X Y and Na J X. A finite element analysis of the large plastic deformation behavior ofamorphous glassy circular polymeric bars. Acto Mechanica Sinica Solida, 1997,10(2):139.[6]Argon A S. A theory for the low-temperature plastic deformation of glassy polymers. Phil Mag,1973,28:839.[7]Haward R N and Thackray G. The use of a mathematical model to describe isothermal stress-strain curvesin glassy thermoplastics. Proc R Soc Lond, 1968,302A:453.[8]Lee E H. Elastic-plastic deformation at finite strain. J Appl Mech, 1969,36:1.[9]Boyce M C, Parks D M and Argan A S. Large inelastic deformation of glassy polymers-Part I. Ratedependent constitutive model. Mech Mater, 1988,7:15.[10]Hu P and Tomita Y. Deformation localization for plane strain tension of polymers, Acta MechanicaSinica, 1996, 28(5):564.[11]G opez A J, Etude de la Deformation du Polycarbonate en Cisaillement Simple, These d’Ingenieur,CNAM, Nancy, France(1983).[12]G’Sell C, Plastic deformation of glassy polymer: Constitutive equations and macromolecularmechanisms. In Strengths of Metals and Alloys (Edited by McQueen H J et al.) Pergamon Press, Oxford,1986:1943.[13]G’Sell C and Gopez A J, Plastic banding in glassy polycarbonate under plane simple shear, J Mat Sci,1985, 20:3462.[14]B oyce M C and Arruda E M. An experimental and analytical investigation of the large straincompression and tensile response, Poly Eng & Sci, 1990, 30(20): 128856。

2Simulating without Negation2DefinitionsNegation-free,-formulas are built up from propositional variables,,...,and the con-stants and,using Boolean conjunction and disjunction,and the unary modal oper-ators(diamond)and(box).We use to denote this language,and to denote with Boolean negation.A transition system(or model)for is a triple,where is a non-empty set of states,is a binary relation on,and is a valuation on,that is:a function assigning a subset of to every proposition letter.We sometimes write to denote the domain of.The satisfaction relation is defined in the familiar way for the atomic case and for the Boolean connectives and;observe that we can always interpret Boolean negation on our models,even when it is not present in our language,For the modal connectives we put iff there exists such that and;and iff for all such that,.The(negation-free)modal theory of a state is the set nf-tp. If we want to emphasize the transition system in which lives,we write nf-tp. Modal logic is just one of many possible description languages for specifying and con-straining transition systems.We will encounter several languages in this paper,and we relate them all tofirst-order logic.To be precise,let be thefirst-order language with unary pred-icate symbols corresponding to the proposition letters in,and with one binary relation symbol.is called the correspondence language for.denotes the set of all -formulas having one free variable.To view transition systems as-structures in the usualfirst-order sense,we use to interpret the unary predicate symbol that corresponds to.The standard translation takes modal formulas to equivalent formulas of.It maps proposition letters onto unary predicate symbols,it commutes with the Booleans,and the modal cases areandFor all transition systems and states we have iff,where the latter denotesfirst-order satisfaction of under the assignment of to the free variable of.A modal formula is said to correspond to afirst-order formulaif.3Simulations forIn this section we adapt the notion of bisimulation to the setting of negation-free modal for-mulas.The resulting notion of directed simulations is then used to analyse the expressive power of negation-free formulas in three different ways:in terms of preservation,safety,and definability.D EFINITION3.1(Directed modal simulations)Let be a non-empty binary relation between two transition systems and,that is, .Then is called a directed(modal)simulation between and if it satisfies the following clauses:1.If and is a proposition letter such that,then.2.If and,then there exists in such that and(back).Simulating without Negation3 3.If and,then there exists in such that and(forth).We write)to indicate that is a directed simulation between and(that links to).A(strong)bisimulation is a directed simulation for which clause1above is an equivalence: if then iff.We write,then implies.P e induction on formulas in.The back and forth clauses in Definition3.1 were introduced especially to deal with the two modal cases.Here’s a proof for the case(the case is similar).Assume,.As,we get,and asThus,the existence of a directed simulation between and guarantees that nf-tp nf-tp.Clearly,if in addition,there is a directed simulation going in the opposite direction,from to,then nf-tp nf-tp.The obvious question, then,is:does imply that4Simulating without NegationThen is a directed simulation between and that links to,and is a directed simulation that links to.However,there is no ordinary bisimulation linking to—there is no state in to which can be linked in a bisimulation.To formulate a converse to Proposition3.2we define a transition system to be image-finite if for every state the set of its successors isfinite.P ROPOSITION3.4Let and be image-finite models with,.Then the following are equivalent:1.nf-tp nf-tp2.we have that implies.T HEOREM3.5(Preservation Theorem)Let be an-formula.Then is equivalent to the standard translation of a(negation-free)modal formula iff it is preserved under directed simulations.P ROOF.The proof uses some basicfirst-order model theory;we refer the reader to Hodges [10]for background material.The right-to-left implication is immediate from Proposition3.2. For the other direction,assume that is preserved under directed simulations.Consider the set of negation-free consequences of imply:NF-Mod-Cons and(3.1) By a compactness argument it suffices to show that NF-Mod-Cons itself implies;for then afinite subset of NF-Mod-Cons will already imply,and will be equivalent to the conjunction of the formulas in thisfinite subset.To prove that NF-Mod-Cons implies,assume that NF-Mod-Cons;we have to show that.Consider the following set of-formulas:andThat is:consists of negations of negation-free modal formulas that are refuted at.Simulating without Negation5 C LAIM1.The set is satisfiable.P ROOF.Assume that it is not.Then there exist formulas,...,such that orBy definition,,...,are negation-free,so,and hencefor some with.But then—a contradiction.This proves Claim1. Now,to‘lift’from to we make a detour via two other transition systemsas follows.Take-saturated elementary extensions and of and, respectively.And define a relation by puttingiff nf-tp nf-tpNotefirst that is non-empty:by Claim2we have nf-tp nf-tp,and so nf-tp nf-tp,as and are elementary extensions of and, respectively.Next,clause1of Definition3.1is trivially fulfilled.To see that clause2is satisfied,suppose that and;we need tofind a such that and.Putnf-tpWe will show that anyfinite subset of nf-tp is refutable in an-successor of.Let, ...,nf-tp.Then,so,as,. This implies that for some both and hold.Now,by -saturation of,all of nf-tp can be refuted at an-successor of.For this we have and nf-tp nf-tp,that is:,as required.For clause3we argue as follows.Suppose that and.We need tofind a such that and;we achieve this by showing that everyfinite subset of nf-tp is satisfiable in a successor of.Let,...,nf-tp.Then,and hence.So there exists a in with and. By-saturation all of nf-tp can be satisfied in a successor of.For this we have and nf-tp nf-tp,that is:,as required.Putting things together,wefind that implies by elementary ex-tension.As6Simulating without NegationE XAMPLE3.6Let be afirst-order formula that is preserved under strong bisimulations.Thenis equivalent to a modal formula in that may include Boolean negation.By testing whether is preserved under directed simulations we canfind out whether is in fact equivalent to a negation-free modal formula.An easy example is thefirst-order formula.This formula is thefirst-order translation of,and it is certainly preserved under strong bisimulation —but here’s an example showing that it is not preserved under directed simulations:take,and,where and are such that all,,verify all proposition letters,and such that all proposition letter but are true in.Clearly,there exists a directed simulation linking and,but, whereas.By Theorem3.5directed simulations uniquely identify a certain fragment offirst-order logic,namely the‘negation-free modal fragment.’By identifying and comparing fragments offirst-order logic that correspond to modal languages in this manner,we have a method for comparing the expressive power of(negation-free)modal languages.We proceed with three corollaries to Theorem3.5and its proof.Thefirst of these concerns a‘dual’to preservation under directed simulations:afirst-order formula is said to be anti-preserved under directed simulations if for all transition systems,,all states ,and all directed simulationsThe following corollary characterizes the relation‘nf-tp nf-tp’between states, in terms of directed simulations.We refer the reader to Hodges[10]for the notion of an ultrapower.C OROLLARY3.8Let and be two transition systems,and let and.Then nf-tpnf-tp iff for some ultrapowers of,and of,we have that,starting from the assumption that nf-tpnf-tp.By a result infirst-order model theory,these-saturated extensions may be ob-tained as suitable ultrapowers of the original models and;see Chang and Keisler [6,Theorem6.1.1]for details.Simulating without Negation7 P ROOF.The left-to-right implication is Proposition3.2.For the right-to-left implication,use Theorem3.5plus the fact that modal formulas are equivalent to theirfirst-order translations under.3.2SafetyIn this subsection we take a different perspective on the expressive power of negation-free modal languages by considering the notion of safety recently introduced by van Benthem[3]. Let denote afirst-order formula with at most two free variables.Then is called safe for bisimulation if whenever8Simulating without Negationthe question for safety can be understood as asking whether the back-and-forth conditions of Definition3.1hold for whenever they hold for the relation symbols in.The definition of safety depends in an essential way on the symmetric character of bisimulations:if the operation expressed by is performed in,then it can be matched by an step in,and vice versa.What is the appropriate notion of safety for directed simulations?Their non-symmetric character causes a split in the notion of safety,depending on whether the operation is per-formed on the left-hand side or on the right-hand side of a pair of directedly similar transition systems and.To be precise,afirst-order formula is left safe for directed simu-lations if wheneverwith and then there exists a such that and.For example,atomic tests,whose semantics are given by,are left safe, but not right safe.On the other hand,tests on negated atoms are right safe,but not left safe.More generally,all negation-rich formulas are right safe.Even though left and right safety have been defined independently,one may be character-ized in terms of the other.To this end we need the following definition.The dynamic negation of a relation is the relation.C LAIM3.12Let be afirst-order formula in.Then is right safe for directed simu-lations iff is left safe for directed simulations.The proof of the above claim is immediate from the definitions.As a consequence to the above claim it suffices to characterize just one of left and right safety;below we characterize the former.T HEOREM3.13(Safety)Let be afirst-order formula in.Then is left safe for directed sim-ulations iff it can be defined from the atomic relation and tests on negation-free modal formulas using only and.Our proof of the above result is tailored after similar results in[3];it requires a careful anal-ysis of so-called continuous negation-free formulas,which we have included in an appendix. Here are the relevant definition and lemma.D EFINITION3.14A modal formula is continuous in if the following holds for every transition system:for each family of subsets such that:iff, for some,,where and for.E XAMPLE3.15The formula is not continuous in,but is.And in fact the latter format typical for safety,as is shown by the following lemma.L EMMA3.16A negation-free formula is continuous in iff it is equivalent to a disjunction of formulas of the form,where each of the formulas is negation-free and-free in the sense that they don’t contain occurrences of.Simulating without Negation9 A proof of the above lemma may be found in Appendix A.P ROOF.[of Theorem3.13]Wefirst prove part1of Theorem3.13.To see that the construc-tions mentioned are indeed left safe,argue as follows.It is clear that the atomic relation and tests on negation-free formulas are left safe.To see that composition is left safe,assume that There is a natural follow-up to Theorem3.13:what are thefirst-order operationsthat are doubly safe for directed simulations,i.e.formulas that are both left and right safe. The following result combines our characterizations of left and right safety to characterize the doubly safe operations.T HEOREM3.17Let be a formula in.Then is doubly safe for directed simulations iff it can be defined from the atomic relation and tests on negation-free modal formulas without occurrences of proposition letters using only and.P ROOF.The right-to-left implication is easily verified.For the converse,assume thatis doubly safe.By Theorem3.13is equivalent to a formulawhere each is negation-free.In addition,is equivalent to a formulawhere each is negation-free.Observe that every proposition letter occurs only negatively in.10Simulating without NegationLet us write to denote the result of substituting for all occurrences of all propo-sition letters in.We will show that,and we will use the fact that formulas in which all(translations of)proposition letters occur only positively(negatively)are up-ward(downward)monotone.If is any transition system,then we write to denote the transition system that is just like except that it assigns to every proposition letter. Observing that all proposition letters in occur only positively in,and that all proposition letters in occur only negatively in,we have,for any transition system,andThis proves,and the latter is of the required form.to denote the class of pointed transition systems that are not in.We say that is closed under ultraproducts(ultrapowers) if any ultraproduct(ultrapower)of transitions systems in is itself in.Likewise,is closed under directed simulations if andis closed ultrapowers.2.is negation-free definable by a single formula iff is closed under directed simulationsand ultraproducts,whileP ROOF.The left-to-right implications are left to the reader.For the right-to-left implication of item1,argue as follows.If andare also closed under bisimulations,and hence,by[18,Theorem6.3]they are definable by a set of modal formulas.Now,as is closed under directed simulations,each formula in must be preserved under directed simulations,and hence equivalent to a negation-free modal formula by Corollary3.9.This shows that is negation-free definable.Next,for the right-to-left implication of item2we use a similar argument.If andThe characterization of definability given in Theorem3.18is hard to use in practice as ultraproducts are rather abstract objects.The following gives a more manageable Fra¨ıss´e-type characterization.Let and be pointed transition systems.We define directed similarity up to between and()by requiring that there exists a sequence of binary relations ,...,such that1.and;2.for each,if and,then;3.for the back-and-forth properties of Definition3.1are satisfied relative to theindices:(a)if and in,then there exists such that and,(b)if and in,then there exists such that and.We write.To see this,define relations for by putting iff every negation-free modal formula of degree at most that is true at,is also true at.Then,...,is a directedsimulation up to that links to.As is closed under directed simulations up to ,this implies,and we are done.Syntax Modalconcept nametopbottomconjunctiondisjunction()negation()univ.quantificationguages.By way of example,we consider the hierarchy of-languages specified in Ta-ble1.Here,we use,to denote atomic concept names(‘proposition letters’),and, to denote complex concepts(‘modal formulas’);and we use to denote roles(‘binary relations’).Terminological expressions are interpreted using an interpretation functionon transition systems.The various languages differ in the constructions they admit;denotes the language with universal quantification,conjunction and unqualified existential quantification. Superlanguages of are identified by strings of the form.We will assume that contains and.Clearly,coincides with(a multi-modal version)of our negation-free modal lan-guage,and hence Theorems3.5,3.13,3.18,and3.19all carry over without effort to .Likewise,the analogous results on expressivity for the standard modal language carry over to the corresponding terminological language.(Further details on the latter connection may be found in[21,11].)Thus,two of the languages in the hi-erarchy have been equipped with model-theoretic tools for analyzing their expressive power. The remaining terminological languages in Table1call for further non-standard notions of (bi-)simulation;coming up with such notions and using them to arrive at a model-theoretic analysis of the remaining languages in Table1is part of our ongoing work.4.2Since and UntilOur next example concerns directed simulations for a negation-free fragment of Since,Un-til logic.To simplify matters we restrict ourselves to the forward looking fragment of the language that only contains the Until operator.Recall its truth definition on transition systems:iff there exists with andfor all,if then. Recently,a notion of bisimulations for Since and Until has been introduced that allows for a complete development of the model theory of the full Since,Until language(see[15]). Building on this,we define the following simulations for the negation-free forward looking fragment.A directed-simulation from to is a pair,where and such that1.and implies;2.if and then there exists such that,and;3.if and then there exists with and. Thefirst of the above clauses is the same as before;the second records transitions in simu-lating pairs of states,and the third clause makes sure that if two pairs of states simulate each other,then they‘agree’on intermediate states.R EMARK4.1In the definition of directed simulation for we had back-and-forth clauses to be able to transfer true formulas involving the diamond operator and its dual the box operator from one model to another.To simplify matters,we have left out a dual for the-operator from our negation-free fragment of the Since,Until language.As a consequence we can make do withname,is a subset of.(Feature systems are simply labeled transition systems of a special kind.)Various logical systems have been proposed to constrain feature structures.Each takesa slightly different view of its models,but often they are non-Boolean fragments of modal logics.In this paper we consider a single example of a feature logic;see Rounds[20]for asurvey of feature languages.The logic we consider is called Kasper-Rounds logic.Assuming and as above,the atomic expressions of are the following:proposition letters(),and so-calledpath equations,for,.Complex formulas are built up using conjunction, disjunction and modalities and,for.The only novel aspect in the interpretationof is the interpretation of the path equations and of the indexed modal operators.Path equations are meant to express that two sequences of transitions lead to the same state; for convenience,we will assume that everyfinite sequence of feature names comes with its own transition relation.if there exists with.(If it exists,this will be unique.)if there exists with and.(Again,if it exists,this will be unique.)Our next aim is to state an analogue of Theorem3.5for.First wefix afirst-orderlanguage into which we translate Kasper-Rounds formulas.has binary relationsymbols for,and unary predicate symbols for all sort names.The novel clauses for the standard translation are the following:where and,and all the and are‘atomic’feature names in ;and,and similarly for.What kind of simulations are we to use to identify as a fragment of thefirst-order language?Note that path equations are essentially intersections of compositions of‘atomic’transition relations.Their intersective character calls for a special kind of simulation in which we ensure that intersecting paths are preserved.The definition below achieves this by relating states to states and pairs of states to pairs of states;it is based on[5] and[13].We write for the reflexive,transitive closure of.A directed KR-sim-ulation from to is a triple where,, and such that1.and implies.2.(a)and implies,(b)and implies.3.(a)if and then there exists with,(b)if and then there exists with.4.implies and;similarly for.5.(a)if and,then there exists such that bothand,(b)if and,then there exists such that bothand.Clause1is familiar.The back-and-forth conditions in clauses2and3ensure that transitions are recorded in simulating pairs of states;together with clause5they allow us to simulate intersecting paths in one transition system with intersecting paths in the other.Finally,clause 4is a bookkeeping clause that relates the behaviour of on pairs of states to its behaviour on single states.With this notion of directed KR-simulation one can proceed to prove analogs of Theorems 3.5,3.13,3.18and3.19for by combining the techniques and results of Section3and [13].The details would take us to far astray from the main points of the present paper to be included here;instead,we refer the reader to[16].5Non-classical negationAlthough in many application areas Boolean negation is unwanted,some form of negation is often called for.This motivates the introduction of non-classical negations.Thefirst ex-ample that comes to mind is probably intuitionistic negation.In this section we show how our directed simulations have to be amended for the results of Section3to carry over to intuitionistic logic.Recall that a transition system is called an intuitionistic model if is a partial order,and is a valuation that assigns-closed subsets of to proposition letters. We assume that the language of intuitionistic logic has,,,and.Conjunction and disjunction are interpreted in the Boolean manner,while is false at all states,and if for all,and implies.As usual,negation is introduced as an abbreviation for.Let,be two intuitionistic models.A directed intuitionistic bisimulation is a pairwith and such that1.(a)if and,then,(b)if and,then.2.(a)if and,then there exists such that,and,(b)if and,then there exists such that,and.We useP ROPOSITION5.1Intuitionistic formulas are preserved under directed intuitionistic bisimulations:ifUsing the notion of directed intuitionistic bisimulation,one can establish counterparts of Theorems3.5,3.13,3.18and3.19.To prove a preservation result along the lines of Theo-rem3.5,we need to define a translation of intuitionistic formulas in tofirst-order formulas. The intuitionistic standard translation takes intuitionistic formulas to-formulas as follows:;commutes with and;andT HEOREM5.2(Preservation Theorem)Let be an-formula.Then is equivalent(on intuitionistic models)to the translation of an intuitionistic formula iff it is preserved under directed intuitionistic bisimu-lations.P ROOF.The left-to-right implication is Proposition5.1.The converse is proved along the lines of Theorem3.5.There are a few things to take into account:we need infinitely many axioms to express the-closedness of the interpretation of proposition letters(unary predicates);we need axioms to express that intuitionistic models are partial orders.As these axioms are allfirst-order axioms,we can use the techniques of Theorem3.5as before.Hence,we only sketch the main steps here.By a compactness argument it suffices to show that show that is itself a consequence of the set of its intuitionistic consequencesInt-Cons intuitionisticSo,consider a model Int-Cons;we have to show that.We achieve this by showing that there exists a model foritp itpwhere itp is the set of(translations of)intuitionistic formulas satisfied by,and itpis the set of Boolean negations of(translations of)intuitionistic formulas false ing this,we move to two-saturated elementary extensions of and and show that there must be a directed intuitionistic bisimulation relating and between those two models.The latter allows us to conclude that.C OROLLARY5.3A modal formula in is equivalent to an intuitionistic formula iff it is preserved under directed intuitionistic bisimulations.E XAMPLE5.4The modal formulas and are preserved under directed intuitionistic bisimula-tions(between intuitionistic models),and,so,on intuitionistic models they are equivalent to intuitionistic formulas.We leave it to the reader to show that,more generally,every modal formula which contains negation only in the scope of modal operators is preserved under directed intuitionistic bisimulations,and therefore equivalent to an intuitionistic formula. Our next goal is to state a safety result for intuitionistic logic along the lines of Theo-rem3.13.As with directed simulations,we get two versions of safety for directed intuitionis-tic bisimulations.We call afirst-order formula in left safe for directed intu-itionistic bisimulations if wheneverwith and, then there exists a with and.We leave it to the reader to verify that is right safe iff is left safe(here,is the dynamic negation of as defined in Section4:).T HEOREM5.5Let be afirst-order formula in.Then is left safe for directed intu-itionistic bisimulations iff it can be defined from the the atomic relation and tests on intuitionistic formulas,using only composition,and choice.P ROOF.The proof is similar to the proof of Theorem3.13,but the required analysis of in-tuitionistic continuity(as in Lemma3.16)requires the use of binary an existential operation that is dual to intuitionistic implication.is closed ultrapowers.2.is definable by a single intuitionistic formula iff is closed under directed intuitionisticbisimulations and ultraproducts,whileWe leave it to the reader to introduce the notion of a directed intuitionistic bisimulation up to,and to formulate an intuitionistic analogue of Theorem3.19.6ConclusionIn this paper we have introduced the notion of a directed simulation to analyse the expressive power of a number of negation-free description languages for transition systems.Our results concerned preservation,safety and definability aspects of negation-free modal logic and some extensions,and we established similar results for intuitionistic logic.Moreover,our results can also be applied to full modal languages with Boolean negation.For example,if afirst-order formula is preserved under strong bisimulations,but not under directed simulations, then we know that its modal equivalent must contain negation in an essential way.To conclude we mention some possibilities for building on the work reported here.The paper is part of a general enterprise that aims to give model-theoretic characterizations of logic-based description formalisms.A lot of work remains to be done,even on arbitrary sub-Boolean fragments offirst-order logic.More concretely,as mentioned in Section4there are several hierarchies of terminological languages waiting to be analyzed using the tools of this paper,A second example concerns the study of expressiveness of feature logics touched upon in Section3.2;this theme is developed in a separate paper[16].Third,one can build on recent work on general fragments offirst-order logic,includingfinite-variable fragments (see[1]),and develop the theory of their negation-free fragments.And a fourth line concerns negation-free substructural logics;there is a close relation there between notions of directed simulation and generative capacities of various formal languages(see[14]),and we plan to report on this in a future paper.AcknowledgmentMaarten de Rijke was partially supported by the Research and Teaching Innovation Fund at the University of Warwick.An earlier version of this paper was presented at Computer Science Logic‘96[17].References[1]H.Andr´e ka,J.van Benthem and I.N´e meti.Back and forth between modal logic and classical logic.Journal ofthe IGPL,3,685–720,1995.[2]F.Baader.A formal definition for the expressive power of terminological knowledge representation languages.Journal of Logic and Computation,6,33–54,1996.[3]J.van Benthem.Exploring Logical Dynamics.Studies in Logic,Language and Information.CSLI Publications,Stanford,1996.[4]J.van Benthem and J.Bergstra.Logic of transition systems.Journal of Logic,Language and Information,3,247–283,1994.[5]J.van Benthem,J.van Eijck and V.Stebletsova.Modal logic,transition systems and processes.Journal ofLogic and Computation,4,811–855,1994.[6]C.C.Chang and H.J.Keisler.Model Theory.North-Holland,Amsterdam,1973.[7]F.M.Donini,M.Lenzerini,D.Nardi and A.Schaerf.Reasoning in Description Logics.In Principles of Knowl-edge Representation.G.Brewka,ed.Studies in Logic,Language and Information.CSLI Publications,Stanford, 1996.[8]J.M.Dunn.Positive modal logic.Studia Logica,55,301–317,1995.[9]M.Hennessy and ner.Algebraic laws for indeterminism and concurrency.Journal of the ACM,32,137–162,1985.[10]W.Hodges.Model Theory.Cambridge University Press,1993.[11]W.van der Hoek and M.de Rijke.Counting objects.Journal of Logic and Computation,5,325–345,1995.[12]M.Hollenberg.Hennessy-Milner Classes and Process Algebra.In Modal Logic and Process Algebra,A.Ponse,M.de Rijke and Y.Venema,eds.pp.187–216.CSLI Publications,1995.。

2007German e-Science Available online at http://www.ges2007.de This document is under the terms of theCC-BY-NC-ND Creative Commons AttributionUser-Centric Monitoring and Steering of theExecution of Large Job SetsRalph M ¨u ller-Pfefferkorn 1,Reinhard Neumann 1,Thomas William 1,Stefan Borovac 2,Torsten Harenberg 2,Matthias H ¨u sken 2,Peter M ¨a ttig 2,Markus Mechtel 2,David Meder-Marouelli 2,Peer Ueberholz 3,Peter Buchholz 4,Daniel Lorenz 4,ChristianUebing 4,Wolfgang Walkowiak 4,Roland Wism ¨u ller 41Center for Information Services and High Performance ComputingTechnische Universit ¨a t Dresden,D-01062Dresden,Germany 2Bergische Universit ¨a t Wuppertal,Gaußstraße 20D-42119Wuppertal,Germany 3Hochschule Niederrhein,Reinarzstraße 49D-47805Krefeld,Germany 4Universit ¨a t Siegen,D-57068Siegen,GermanyAbstractProcessing of large data sets with high through put is one of the majorfocus of Grid computing today.If possible,data are split up into smallchunks that are processed independently.Thus,job sets of hundreds >or even thousands of individual jobs are possible.For the job submitteror the resource providers such a scenario is a nightmare currently,asit is hard to keep track of such an amount of jobs or to identify failurereasons.We present a system that will support gLite users to track and monitortheir jobs and their resource usage,to find and identify failure reasonsand even to steer running applications.1IntroductionToday,one of the major challenges in science is the processing of large datasets.Experiments or simulations can produce an enormous amount of re-sults that are stored in databases or files.Processing these data is usually done by splitting the analysis process into a large number of small jobs that read only chunks of the data.By running these jobs in parallel on a Grid the processing time can be decreased significantly.Examples of such a scenario are the pro-cessing of images in medicine or the analysis of high energy physics data.The Large Hadron Collider (LHC)at CERN will provide particle physicists with several Petabytes of experiment data every year.Additionally,simulations of the physics processes and the detector response are needed to understand the experiment.With the LHC Computing Grid (LCG)the High Energy Physics community wants to provide a Grid environment to enable such data processing capabilities.The gLite middleware stack is the base of this Grid effort.To monitor the hundreds or thousands of jobs a physicist usually submits for an analysis intelligent tools are needed to support the user.The existing monitoring tools of the LCG/gLite environment currently provide only limited functionality.Either they focus on the underlying fabric,i.e.hardware,in-frastructure(e.g.the LCG Real Time Monitor or GridIce)or are only simple command line toolsflooding the user with textual information.The High Energy Particle Physics Community Grid project1(HEPCG)[1] of the German D-Grid Initiative[2]wants to contribute to the functionality the LCG provides to the users.In this paper the user monitoring tools that have been developed in HEPCG are presented.Section2describes the job and resource usage monitoring system,that supports users in handling the large number of jobs.Section3presents the job execution monitor to track down problems in single steps of a job.Finally,a tool for online steering of jobs in LCG is introduced in section4.2User-centric monitoring of jobs and their resources AMon,the user-centric monitoring tool delivers the user information about the status and the resource usage of the submitted jobs.The information are answers to questions like:•Is the job still running,successfully done or did it fail?•What is the CPU usage over time?What is the memory consumption on the machine where the job is running?What about the I/O of my program?•Are there any critical usage parameters of my jobs like a directoryfilling up or a job hanging?Such information can indicate a untroubled run or problems of the users applica-tion.In this context,the term”user”comprises both the submitters of the jobs, who want to know what is going on with their jobs,and the resource providers who want information about the usage of their resources.Due to the large number of jobs and thus the large amount of monitoring data there are several constraints for the monitoring to be beneficial for the scientists:1.Easy access and handling-only limited knowledge about monitoring shouldbe needed by the user2.Support the users with graphical representations of the pre-analysed infor-mation,that allow interaction to get further and detailed information3.Authentication,authorisation and secure data transmission for data pri-vacy reasonsThefirst constraint led to the decision to realise a web browser based inter-face.Due to the usage of web browsers in daily work people know how to handle and to work with them.A new tool would need new and additional knowledge. Therefore,the monitoring is integrated into a Web portal using the GridSphereFigure1:The monitoring is integrated into GridSphere.Here an example graph of the temporal development of an information(the CPU usage of the jobs). data is denoted by the scrollbars of the display.portal technology[3].It already provides features like user management and user login(with password or credentials).After login the user obtains the monitoring data by a click.Java applets-running inside the portlets-handle the visualisa-tion of the data.The data are provided in a diversity of displays.Examples for displays are statistical summaries(e.g.pie charts)that give a general overview. Timelines(the temporal development of a information)inform about the dy-namic behaviour of the jobs regarding a metric(a measurement,e.g.the used CPU time).An example screenshot is given infigure1.Clicking into the charts will display more detailed information or allow the user to zoom into the data.The distributed monitoring data are gathered and pre-analysed by a Web Service that the GridSphere portlet is contacting.Currently,it collects infor-mation from tables in R-GMA,the Relational Grid Monitoring Architecture[4]. R-GMA is a kind of a distributed relational data base that is used in LCG to store monitoring data.As the Web Service provides a generic interface access to any monitoring system could be plugged in.The job monitoring information are measured and collected on the worker nodes(computers)where the jobs are running.The existing LCG worker node monitoring[5]was extended to collect a variety of useful information(see ta-ble1).These data are stored in variable and configurable time intervals into R-GMA.The default,e.g.collects information every3minutes for thefirst20 minutes of the runtime of the job,every10minutes for the next40minutes and then every half an our for the rest of the runtime.The overall architecture with the described four components-information collection on the nodes where the jobs run,information storage,information analysis and the user interface-is sketched infigure2.3Figure2:Architecture of the Job and Resource Usage Monitoring SystemFuture versions will extend the user interface with an authorisation frame-work using VOMS[6].This will allow different levels of access rights to the monitoring data(a user,a site administrator,a VO manager).Furthermore,filters will pre-analyse the data to give the users direct hints and help infinding problems.MetricsGeneralWallClockTime;UsedCPUTime;load averagesMemoryfree space on home,temporary and work directory;summary offile system propertiesFile I/Oreceived and transmitted networkbased software which monitors the execution of script files within the user jobs.Its purpose is to watch the execution of every command the user job wants to execute on a compute node and to provide information about its success or failure to the user on the user interface.3.1Structure of the Job ExecutionMonitorFigure 3:Job Execution Monitor,global principle There are two main components ofthe JEM,which both run on the com-pute node:the Job Wrapper and theWatchdog.Furthermore,the Wrap-per makes use of the script parser,which is a third independent module.The Script Wrapper can bethought of as an interpreter for thesupported script languages.Cur-rently,shell scripts for the sh andbash shell and the python program-ming language are supported,but itis planed to provide a generic frame-work for other languages.With this it will be possible to write separate mod-ules/plugins,so that more languagesmay be interpreted as well.With this kind of generic interpreter for script files,it is possible to do other things in between the execution of the actual commands.Especially,the success-ful execution of every command can be checked,monitored and logged,so that the user is informed about the current state of his program,at all time.But as the most important (and most difficult)requirement,such a system must handle a given script file as if it were executed without this monitoring framework.3.2Script WrapperThe Script Wrapper is the main component for monitoring the execution and the current state of the user job.It has a modular structure,which makes it easy to add support for other script languages.3.2.1Bash WrapperStepwise execution of bash scripts takes place in two steps.First,a given bash script is parsed and a modified script file is then written to disk.As a second step,the Bash Wrapper launches the modified script file created by the Bash Parser,which then runs in parallel to the wrapper.With this concept it is possible to track all input and output of all commands and the success of every single command can be checked.This mechanism is described in more detail below.Even more detailed information can be found in [7]and [8].5Bash Parser The purpose of this component is to analyse a given shell script,find every single command and put an escape sequence in front of it.This escape sequence is a call to a python module,which gets the actual command as a shell parameter.This intermediate python program can then doanything Figure 4:Bash Wrapper it likes with the commandstring.Of course,in the end,the command should be exe-cuted.But with this technique,it is possible to do other thingsbetween the call of a commandand its execution.The bash parser does hiswork in three steps:it startswith a lexical analysis of thescript,then parses the script toidentify where the commandsare and finally it inserts theescape sequences to insert themonitoring code into the script.Execution Shell The execu-tion shell is the place,whereall commands are actually pro-cessed.After the script wrap-per has gotten the information,it writes an entry to R-GMA,that it is going to start the next command.After that,the wrapper executes the command in the execution shell and waits for its completion.It is not possible to execute each command in an independent shell,that is closed after the command has finished.There has to be a shell kept open and running as long as the modified script is running.All commands have to be executed in the same shell.This is necessary,because the script may modify the environment,on which subsequent commands depend.If the shell is closed,the values of these variables are lost.3.2.2Python WrapperSupporting the Python programming language is quite important for users in high energy physics,as most software environments of the LHC experiments are Python based.The Python Wrapper offers the same functionality as the bash wrapper.However,its implementation is much easier,as Python already offers debug methods,which are used by the Python Wrapper to examine the result of every command line.The success or failure of each command line is determined like in the bash and sh case.63.3Watchdogs and Pre-execution TestsThe watchdog component is responsible for monitoring the system resources, while the user script is being executed by the script wrapper.In a configurable regular time interval the watchdog publishes the current values of these resources to a table in R-GMA.Currently,the status of thefile systems and the current memory usage as well as network statistics are monitored.Before the script wrapper is started,several tests are performed on the com-pute node to check whether some needed services are available.Among these are tests of the R-GMA interface or accessability of some directories.Furthermore, several tests on the numerical quality[9,10]of the results are performed.4Interactive steering of jobsWhile the job monitoring supports the user with information about the en-vironment of the job,this section describes the support given to the user for interactive monitoring and steering of the running job itself.Once the job has started,it may produce useless results due to configuration faults of the job,or because the software installation has bugs.In other cases, the user may want to change parameters in order to improve performance or to explore a parameter realm.Without interactive steering,the user has to wait until the job hasfinished and then evaluate the results.Interactive steering can accelerate the research process,by providing earlier access to intermediate results and the possibility to modify parameters of running jobs.The Result Monitoring and Online Steering Tool(RMOST)[11,12]is build upon a model of distributed shared memory,to reduce modification requirements of existing applications and visualisation software.The goal is to provide a middleware which connects a local visualisation tool to remote Grid jobs.A general solution requires that the steering system does not need to know the type of the data it transports and synchronises.The marshaling problem is exported from the steering system to the application,or to a special data access layer.Most applications have already a serialisation method for their data for storing them on disk.4.1The FunctionalityThe functionality of RMOST is demonstrated by steering Grid jobs of the High Energy Physics(HEP)experiment ATLAS.In the ATLAS experiment a huge amount of data has to be evaluated and computed,which requires the use of a Grid.There exists an experiment software framework Athena[13]for computing the results.Athena is a modular framework which contains different kinds of components.The user creates a job descriptionfile(job options)for his job,where he specifies the components and parameters.Numerous scientists around the world developed Athena components,thus modifications of core components for steering are hardly accepted by the HEP7community.A solution for the integration of steering into the Athena framework is to add additional components to the framework,that are sent with the job.The intermediate results for visualisation are stored in so called ROOTfiles, which can reach a size of some GB.For the steering,thesefiles must be accessed remotely.The user adjustable parameters are contained in the job optionsfile. Thus,RMOST allows the modification of the job optionsfile.The ROOT toolkit[14]is used for visualisation of the physics results.Many scientists create own tools and components with ROOT,again changes to core components will be hardly accepted by the HEP community.To integrate steer-ing into ROOT,the possibility of ROOT to extend its functionality with dy-namically loaded libraries,was used.For the steering the user must be able to create an interactive communication channel to the remote Grid job.This meansfirstly tofind the job in the Grid. Then an interactive connection to the job has to be established,andfinally,this connection must be secured.Hereby,some problems may occur: After job submission the user retrieves a string,which identifies his job,but he does not know the target host,where the job runs.The host that executes the job,may be protected byfirewalls or may be located in a private IP network, which inhibits a direct connection to the job.In addition the communication system must not compromise the target site’s security nor allow unauthorised persons access to the site,and must ensure that only the submitter of the job can steer it.In the ATLAS experiment the user typically submits hundreds of jobs in parallel.With this large number of jobs it is impossible to inspect each single job to control the execution of all jobs.Thus,a notification mechanism is provided, which automatically evaluates a condition and informs the user about suspicious jobs or on remarkable events.Furthermore,when a failure occurs and Athena terminates smoothly,the termination can be suspended for some time,the user is notified,and has the chance to inspect intermediate results or restart the computation with modified parameters.4.2RMOST ComponentsRMOST connects the user interface and the Grid job at runtime.Thus, RMOST provides functionality to access the data from the Athena framework and synchronises it with local data copies used by the user interface.To fulfil this task,a set of libraries and services are provided.In Fig.5an overview of the different components is shown.On the side of the remote Grid job,the integration is realized by additional Athena components.Thefirst component is called RMSpy the following functionality is provided:•Intermediate results stored in ROOTfiles can be accessed for visualisation.•Online access to the logfiles.•The number of executed events can be monitored.•The job optionsfile can be modified or replaced.The changes in the job options are applied by a restart of the job without resubmission.8Figure5:The components of RMOST•The execution of the job can be suspended,terminated,or continued.•The job can be executed stepwise.•The job can be restarted without changing the job options.•Optional notification of start and end of the execution.•Optional notification on failure with delayed termination of the job for user interaction during the waiting period,enabling in-situ inspection.The second component is RMconnection service,which is needed to connect to Grid jobs in spite offirewalls and private IP networks[11,15].5SummaryWith the tools presented we hope to give the scientists in LCG and-as gLite gets more and more widespread-also in other communities helpful and valuable support in their daily use of the Grid.References1.HEPCG.High Energy Physics Community Grid,2005..2.D-Grid.The D-Grid Initiative,2005.http://www.dgrid.de.3.J.Novotny,M.Russell,and O.Wehrens.Gridsphere:A portal framework for building collaborations,2005.:80/gridsphere/gridsphere?cid=publications.4. A.J.Wilson et rmation and monitoring services within a Grid environ-ment.In CHEP2004,September2004.5.L.Field, F.Naz,et er level tools documentation,2006..tw/gocwiki/User。

67th IFLA Council and GeneralConferenceAugust 16-25, 2001Code Number:068-122-EDivision Number:IProfessional Group:Library and Research Services for ParliamentsJoint Meeting with:-Meeting Number:122Simultaneous Interpretation:YesTowards the electronic parliamentary library in the context of the European UnionDick ToornstraEuropean Parliament1Brussels, BelgiumAbstract:The electronic parliamentary library at the European Union level is developing to meet the needs of a changing political context. It is emerging in multiple forms: cooperation amongst EU institutions, cooperation amongst EU parliaments, and the service of the European Parliament itself. The development of electronic services presents a management challenge for libraries, and in the context of the European Parliament there is a challenge to differentiate the library's service in a highly competitive environment. A practical framework for service differentiation is outlined. Working with 'communities of practice' is proposed as a key strategy.1.IntroductionThis paper concerns the development of electronic library services for the European Parliament (EP), rather than for all parliaments in the European Union (EU), although it does address the issue of EU wide inter-parliamentary cooperation2. It takes 'electronic library' to be synonymous with 'digital' or 'virtual' libraries, meaning broadly a library service offering electronic materials3.2.Context2.1.The European ParliamentThe European Parliament is directly elected by the 370 million citizens of the EU and its role includes participation in the EU legislative process and scrutiny and control of the EU executive. Its origins lie in the 1950's but it became of major significance following the introduction of direct elections (1979) and increased responsibilities brought by subsequent treaties4. An 'electronic parliamentary library' is in the course of development in the EP. The context for such a library comprises three main elements: the political context; institutional factors; and managerial issues.2.2.Political contextThe broad political situation facing EU institutions is positive but includes some questioning of their legitimacy5. Specific EU issues are compounded by the reduced confidence in politicians and political institutions which seems a global phenomenon. This crisis of legitimacy is being addressed in part through efforts to develop greater 'transparency': making citizens aware of what has been achieved, explaining decisions, increasing accountability and the scope for meaningful participation6. There are also efforts to increase legitimacy through improved quality of legislation, with electronic information a key resource and medium for cooperation in the process. Finally, the EU institutions seek legitimacy through improving the efficiency and effectiveness of their operations. For an institution such as the Parliament whose main operations span three sites in three different countries, and whose Members work in fifteen countries, electronic communications are essential. For all these reasons, developing intranet and internet communications have therefore been a priority for the Parliament.2.3.Institutional factorsThe treaties of Amsterdam (1997) and of Nice (2000) have given Parliament a much stronger role in the legislative process: It has also been given an impetus towards even closer relations with national parliaments. Parliament needs to understand national contexts and to draw on national experience in its work. The national parliaments, in turn, may find it helpful to understand both the context of the European legislation and draw on experience in other national arenas. It follows that there is a developing need for EU-wide inter-parliamentary information exchange. This can most effectively and efficiently be supported by electronic media.The new treaties increased the Parliament's role in legislation through the 'co-decision' procedure. The reforms have also tightened-up legislative procedures with formal deadlines and closer management by all the institutions involved. It has become more important for the Parliament to be informed, and information to the Parliament must be delivered against tight deadlines. Information about Parliament's activities and positions is also of increased interest to the other EU institutions. Again, fast and effective internal and external communication is essential.2.4.Managerial issuesIn institutions with a substantial legacy of traditional library services the development of electronic services can pose problems of transition7. In so far as the Parliament offers a traditional library service then it is vulnerable to the 'new economics of information'8 in which businesses with assets "that traditionally offered competitive advantages and served as barriers to entry will become liabilities" [especially where they involve information content that can be more effectively and efficiently communicated electronically]. In such cases"the loss of even a small portion of customers to new distribution channels or the migration of a high-margin product to the electronic domain can throw a business with high fixed costs into a downward spiral.It may be easy to grasp this point intellectually, but it is much harder for managers to act on its implications. In many businesses, the assets in question are integral to a company's core competence. It is not easy psychologically to withdraw from assets so central to a company's identity. It is not easy strategically to downsize assets that have high fixed costs when so many customers still prefer the current business model."9The need for a transition in library resource allocation may not be recognised by clients or other stakeholders, at least in the short term. The 'political' management of the transition is a key management task in the development of the electronic library.rmation services of the European Parliament3.1.IntroductionThe Parliament has three information services: one deals with the external audience, another provides day-to-day support to parliamentary committees, and a third provides longer term research and the main library & documentation service. All MEPs have internet access and personal research assistance, and often that provides as much supplementary information as the MEPs require. Many organisations target their information services on the MEPs for lobbying purposes. The various party groups in the Parliament support their own internal information services, and of course national party structures also provide information. Information overload rather than information delivery is seen as the MEPs' problem.3.2.Directorate General (DG) for Research and DocumentationThis DG provides what would be recognised by professional peers as the library, documentation and research service of the Parliament. Its intranet site combines access to research products, external databases, the library catalogue, selected internet links, service information, staff directory etc. In so far as there is an emerging 'electronic library' for the Parliament, then it is this site, but it exists in a competitive and slightly chaotic environment.4.Towards a common library of EU institutions?4.1.EUROLIBThere is voluntary cooperation at the practical level between the libraries of EU institutions and related agencies through the EUROLIB organisation10. While most of the libraries have distinct subject specialisations and clienteles there are both information sources and processes which are common. In these areas there is scope for sharing information electronically, consortium negotiation and purchasing, sharing expertise and cooperative work. In this practical cooperation it is possible to see an emergent electronic library at the European level, consisting of some common infrastructure and content.4.2.An inter-institutional library?Recently, the more ambitious concept of an integrated 'inter-institutional library' has been the subject of a high-level proposal. Current thinking is that this would be a physical library based in Brussels serving the three main institutions (Parliament, Commission and Council of Ministers) but open to the staff of all EU institutions and perhaps, in time, the public. An electronic inter-institutional library could meet some of the basic electronic library requirements of the institutions as a consortium; it may also have some value as an interinstitutional resource which could be accessed by citizens from across the EU11. The emphasis on aphysical library in the new proposal underlines the fact that professional thinking on electronic services could be out of step with that of clients - a problem identified by Evans and Wurster.5.Towards an electronic library for European parliaments?Information exchange between Parliaments has already been highlighted as an emerging requirement. Following the conclusions of a conference of Speakers of national parliaments in 2000, the ECPRD12 is elaborating an electronic information service concept. This is intended to facilitate the tracking of EU legislation and the exchange of research and documentation on current national and international issues. A basis for information exchange is being laid by the common use of the Eurovoc thesaurus and the ongoing development of ParlML (based on XML) to structure parliamentary documents. The service will be placed on the ECPRD site as a central place in the European arena: to act as the switching point between parliaments, to make use of work already done at the EU level, and to undertake or collate added-value services such as translation and indexing.6.Developing the electronic library of the European ParliamentThe EP has a profusion of electronic information sources both internal and external. The end result is not wholly positive: it includes information overload, complexity and inefficiency. It underlines the difference between simply providing electronic resources and an 'electronic library'. The elements which make a 'library' distinct are traditional, obvious, ones, even if they take new forms13.6.1.Differentiating the libraryThe key clients we wish to reach have access to good information from other sources - in most cases, too much information and too many alternative sources. On a case-by-case basis these other sources may have advantages in being more specialised, closer to the action, tailored to a political view etc. To succeed in such a competitive and 'noisy' environment, the Parliament's electronic library must offer a package which is clearly differentiated from alternative sources.The most obvious 'Unique Selling Point' (USP) of parliamentary services should be that they are non-partisan and dedicated to the primary work of the Parliament. Other information sources are partisan and/or address multiple audiences. This means for the MEPs that their electronic library should be a safe information source which is uncompromisingly dedicated to their requirements. This is a message that should be pervasive in service communications and in the development of staff.The service must offer information which is relevant. This requires1.front-line staff getting close to clients and understanding their precise needs, working co-operativelywith clients and other information providers2.'environmental scanning' at a strategic level to identify emerging/future issues3.the conversion of such knowledge into information products (produced in-house or identified fromexternal sources)4.added value by sifting, summarising and analysing information - pre-processing and screening forthe client.5.the minimum of editorial/bureaucratic processing - the link between client knowledge/environmentalawareness and content should be as direct as possible.6.ideally, facilities to allow clients to shape the content and organisation of the service. This can meanin terms of personalised 'views' of selected site elements and/or in terms of clients contributing information and/or in terms of normal user feedback mechanisms.Currency of information is key and can be achieved through1.understanding of client requirements and environmental awareness2.investment in current awareness of staff (the web site itself can be part of this)3.management commitment to maintaining current information on the site4. a production system for the web site which allows for rapid publication.The service must offer convenience to clients, which can be delivered through:1.simplicity of use through design and ease of information retrieval2.customised or semi-customised access to information3.rapidity of loading (simplicity of design)4.ready access to human assistance, hard copy information and referral to specialist services5.access off-site (i.e. on the Internet or through external access to the network).6.bringing together of key information sources, directly and through search facilities.The ultimate objective in this last case could be an integrated search mechanism which allows a single search procedure to produce a unified result from all recommended sources. This appears one of the key issues in developing an electronic library. New electronic information sources and substitutes for labour-intensive in-house indexing and cataloguing can be bought in, but the various sources need to be integrated for clients if they are to offer the same convenience as a traditional library, and if the 'electronic library' is to be distinguishable from the electronic jungle. Integration poses technical problems but also ones which relate to commercial and intellectual property issues e.g. combining access to open sources with access to those with password protection or specific licensing requirements. Interestingly, Akeroyd reports research which suggests that clients may shun what appears the obvious convenience of integrated searching14.The service should develop a sense of ownership and community in its users, through:anising the site to reflect and support existing and potential 'communities of practice' (seesection 6.2 below).2.an active policy of user feedback which results in visible and rapid action.3.making users visible on the site (e.g. publishing comments, allocating space for 'newsgroup' typefeatures, creating knowledge directories, profiles of interesting users etc.).4.off-line activities for users, bringing users and staff togethermunities of practiceThe concept of 'communities of practice' is one of the most interesting in the field of knowledge management. Briefly, it is suggested that "collective practice leads to forms of collective knowledge, shared sensemaking, and distributed understanding that doesn't reduce to the content of individual heads….[A community of practice is a] group across which such know-how and sensemaking are shared - [a] group which needs to work together for its dispositional know-how to be put into practice"15. The subject-specialised committees of the Parliament can be regarded as the focus for communities of practice extending through the politicians, the committee administrators, experts in the political groups, external experts, in-house researchers, documentalists etc. Equally, such a community might exist at the inter-parliamentary level. The strength of these communities in the EP, and the extent of library/documentation service participation, varies a great deal. Communities of practice in their nature are informal and cannot be constituted by administrative action. However, the facilities of the electronic library can be directed to supporting communications within the group, creating a forum where existing members can easily exchange information and new/potential members can learn and effect an introduction to the community. The attractions of this community forum can be enhanced by using the same facility to provide access to formal knowledge (internet resources, databases, books etc) and to specialist news. Orienting electronic library services to support communities of practice has important benefits:Øthe library aligns itself with the core work and key players of the institution, in a practical and visible way. This means that the service is relevant and demonstrates it is relevant.Øthe library enhances the operation of the institution, in as far as it enriches the dialogue of its communities and helps to build the communities. (This last is important in a multinational organisation with a relatively transient population).Ølibrary staff can access and be involved in the communities and so can maintain the relevance and visibility of library services in general.Parliament has sought to support communities of practice through the creation of joint researcher/PDC subject teams, and through the development of new electronic services designed around the subject specialisations of Parliamentary committees. At the inter-parliamentary level the national parliaments, notably through the ECPRD also intend to offer a service which will support actual and potential communities.7.ConclusionThe EU has perhaps been slow in taking up the idea but the momentum for creating an electronic EU library is now strong. If the electronic library is in its infancy (Akeroyd) then we are conceptually at an earlier stage: will it be twins (parliamentary and inter-institutional) or triplets (add inter-parliamentary)? Will the inter-institutional consume its siblings? It is clear in any event that there are unprecedented opportunities for parliamentary libraries to benefit their parent institutions. These benefits will be based on traditional library skills and virtues such as co-operation, but require a break from the past in terms of resource allocation and the rigorous application of a strategy of differentiation.ReferencesAkeroyd, John 'The management of change in electronic libraries', IFLA, 66th General Conference, 2000.(/IV/ifla66/papers/037-110e.htm)Brown, John Seely & Duguid, Paul 'Organising knowledge', California Management Review, Vol. 40, No.3, reprinted in 'Managing Knowledge: Readings Part 1', Open University, Milton Keynes, 1999.'Eurobarometer', No. 35, June 1991 and No. 53, October 2000, European Commission, Brussels. Evans, Philip B. & Wurster, Thomas S., 'Strategy and the new economics of information', Harvard Business Review, Sep-Oct 1997, pp. 71-82.1 The opinions expressed in this paper are the sole responsibility of the author and do not necessarily reflect the official position of the European Parliament.2 An impression of developments across European was conveyed at the "The electronic library for parliament" seminar held by the European Centre for Parliamentary Research & Documentation in November 2000 (no proceedings published to date). This included a report of a questionnaire completed by 36 European parliaments/assemblies. Electronic services are ubiquitous - 100% had an internet site and 86% an intranet.3 For general background on electronic library issues, see article by Akeroyd.4 An introduction to the Parliament can be found at http://www.europarl.eu.int/presentation/default_en.htm5 Regular polls of EU citizens show a peak in support in 1991 (72%+ thought membership a 'good thing'), and it has fallen fairly steadily since then (to 49% in 2000 - around the level it was in 1981). Amongst 25-39 year olds it has fallen from 75% in 1991 to 50% in 2000, and amongst 15-24 year olds from 73% to 55%. (Eurobarometer, Report No. 35 p. A12 & No. 53 p. 53, p.B12).6 See, for example, Article 255 EU Treaty, EP Decision 10.7.97; Council Decision 93/731/CE7 See Akeroyd for a practical discussion of some of the issues.8 See article by Evans & Wurster.9 Evans & Wurster, p. 82.10 EUROLIB currently has 21 members, drawn from EU institutions, EU agencies and other organisations which operate at the European level and deal with European information.11 The concept of a common electronic library as a support to an existing network of conventional libraries can be seen in the UK with the pilot electronic library of the National Health Service /12 The European Centre for Parliamentary Research and Documentation is a cooperative body under the aegis of the European Parliament and the Parliamentary Assembly of the Council of Europe, set up in 1977. It is primarily a network of research departments and parliamentary libraries but also involves other officials responsible for information gathering and dissemination. Its aim is to facilitate contacts and exchanges between the officials of member parliaments.13 "if we argue that traditional libraries comprise more than just data, if we define them as the sum not only of information sources, but also navigational tools, metadata systems such as catalogues, human support systems and a suitable environment within which information is delivered, then we can say that the digital library is still in its infancy" Akeroyd, p. 1.14 Akeroyd, p. 215 Brown & Duguid, p. 8.。

Importance of quantum interferencein molecular-scale devicesKamil Walczak 1Institute of Physics, Adam Mickiewicz UniversityUmultowska 85, 61-614 Poznań, PolandElectron transport is theoretically investigated in a molecular device made of anthracene molecule attached to the electrodes by thiol end groups in two different configurations (para and meta, respectively). Molecular system is described by a simple Hückel-like model (with non-orthogonal basis set of atomic orbitals), while the coupling to the electrodes is treated through the use of Newns-Anderson chemisorption theory (constant density of states within energy bandwidth). Transport characteristics (current-voltage and conductance-voltage) are calculated from the transmission function in the standard Landauer formulation. The essential question of quantum interference is discussed in detail. The results have shown a striking variation of transport properties of the device depending on the character of molecular binding to the electrodes.Key words: molecular device, quantum interference, electronic transport, molecular electronicsPACS numbers: 85.65.+h , 73.23.-bI. IntroductionMolecular junctions are promising candidates as future electronic devices because of their small size and self-assembly features. Such junctions are usually composed of two metallic electrodes (source and drain) joined by individual molecule (bridge). The charge is transferred under the bias voltage and current-voltage (I-V) characteristics are measured experimentally [1]. In general, transport properties of such structures are dominated by some effects of quantum origin, such as: tunneling, quantization of molecular energy levels and discreteness of electron charge and spin. However, recently it was pointed out that also quantum interference effects can lead to substantial variation in the conductance of molecule-scale devices [2-9].The main purpose of this work is to show some theoretical aspects of interference phenomena in anthracene molecule connecting two identical electrodes by thiol (–SH) end groups (see fig.1). These end groups (or more precisely sulfur terminal atoms, since hydrogen atom seems to be lost in the chemisorption process) ensure readily attachment to metal surfaces [10]. It is shown that the molecule acts not only as a scattering impurity between two reservoirs of electrons (electrodes), but simultaneously as an “electronic interferometer”. Interference itself reveals the wave nature of the electrons passing from the source to drain through the molecule. Here the variation of interference conditions is achieved by changing the connection between anthracene molecule and electrodes.Fig.1 A schematic model of analyzed samples.II. Theoretical treatmentMolecular device is defined as anthracene molecule joined to two metallic surfaces with the help of thiol end groups in two different configurations – para (A) and meta (B), respectively. In both cases we have different interference conditions and so we expect to observe changes in transport characteristics. Problem of electronic conduction between two continuum reservoirs of states via a molecular bridge with discrete energy levels can be solved within transfer matrix technique of scattering theory [11,12]. The current flowing through the device is obtained from the transmission function T through the integration procedure [12]: []dE )E (f )E (f )E (T h e 2)V (I D S m m ---=ò+¥¥-, (1)where: f denotes Fermi distribution function for room temperature (293 K) with chemical potentials 2/eV E F D /S ±=m referred to the source and drain, respectively. In this type of non-self-consistent calculations, one must postulate voltage distribution along the molecular bridge. For the sake of simplicity we assume that voltage drop is limited to the electrodes only [13], shifting their Fermi level located in the middle of the HOMO-LUMO gap [14]. However, other choices of the voltage distribution have only a small effect on our final results and general conclusions. The differential conductance is then calculated as the derivative of the current with respect to the voltage [15]:[])(T )(T G G D S 021m m +=, (2) where 5.77h /e 2G 20»= [μS] is the quantum of conductance.Formula for the transmission probability can be expressed in the convenient matrix form[12]:[]+++--=G )(G )(tr )E (T D D S S S S S S , (3)where D /S S and are self-energy terms of the source/drain electrode and the Green ’s function of the molecule is expressed as follows:1D S ]H ES [G ----=S S . (4)Here S denotes overlapping matrix (where the overlap between the nearest-neighbor sites is assumed to be equal to 0.25). Since only delocalized π-electrons dictate the transport properties of organic molecules, the electronic structure of the molecule is described by a simple H ückel Hamiltonian H with one π-orbital per site (atom) [16], where overlapping is explicitly included (using non-orthogonal basis set of atomic orbitals). Throughout this work we take the standard energy parameters for organic conjugated systems: on-site energy is 6.6-=a eV and nearest-neighbor hopping integral is 7.2-=b eV. In the H ückel π-bond picture, all carbon and sulfur atoms are treated equivalently (because of their electronegativity). In our simplified model, the coupling to the electrodes is treated through the use of Newns-Anderson chemisorption theory [11], where ideal electrodes are described by constant density of states within energy bandwidth [17-20]. So self-energy matrices (S ) take the diagonal form with elements equal to i 05.0- [eV].Fig.2 Transmission as a function of electron energy (with respect to Fermi energy level)for devices in configuration A (solid curve) and B (broken curve), respectively.Fig.3 Comparison of conductance spectra for devices in configurationA (solid curve) andB (broken curve), respectively. III. Results and discussionNow we proceed to analyze our results from the point of view of quantum interference effects. The geometry of the molecule is taken to be that of anthracene with sulfur atoms on either end of the molecule, binding it to the electrodes in two different configurations – para(A) and meta (B), respectively. For isolated anthracene the HOMO is at 614.7- eV and the LUMO is at 352.5- eV. Because of our simplification that Fermi level is arbitrarily chosen to be located in the middle of the HOMO-LUMO gap, 483.6E F -= eV. The HOMO-LUMO gap for molecular system in para configuration is reduced from 262.2 eV for anthracene to the value of 667.0 eV, but for molecular system in meta configuration it is reduced to zero.Figure 2 shows the transmission dependence on the electron energy for anthracene in para (A) and meta (B) connections with identical electrodes. For transparency we plot it in the logarithmic scale. Asymmetry of the transmission function (with respect to the Fermi energy level) is due to non-orthogonality of atomic orbitals used to describe molecular system. The existence of resonances in the transmission probability is associated with resonant tunneling through molecular eigenstates. Such resonance peaks are shifted and broadened by the fact of the coupling with the electrodes (just like discrete energy levels of the molecule). A change in the configuration of connection between anthracene and two electrodes results in variation of the interference conditions and obvious changes in the transmission function. It manifests itself as shifts in the resonance peaks and in reduction of their height. Well-separated energy levels give rise to distinct peaks in the spectrum, while molecular levels close in energy can overlap and eventually interfere (reduction of resonance peaks is due to destructive interference).Fig.4 Comparison of current-voltage characteristics for devices in configurationA (solid curve) andB (broken curve), respectively.Another remarkable feature of the transmission spectrum is the appearance of antiresonances, which are defined as transmittance zeros and correspond to the physical situation for incident electron being perfectly reflected by a molecule. There are two different mechanisms (well-known in literature) responsible for the origin of antiresonances. One of these is associated with interference between the different molecular orbitals through which the electron propagates [2,21]. The second mechanism is due entirely to the non-orthogonality of atomic orbitals on different atoms [17]. In principle, transport problem in which a non-orthogonal basis set of states is used can be solved by a method proposed recently by Emberly and Kirczenow [5], where condition for antiresonances was analytically demonstrated. However, in this work we perform numerical evaluations of energies at which incoming electron has no chance to leave the source electrode. There are six antiresonances for device in configuration A (F E 821.2E +-=, F E 160.2+-, F E 622.1+-, F E 320.2+,F E 600.3+, F E 907.5+) and only one for device in configuration B (F E E =). Antiresonance is predicted to manifest itself by producing a drop in the differential conductance [5]. Moreover, the fact that it is generated exactly at the Fermi energy of metallic electrodes has important consequences for the conductance spectrum in which antiresonance can be observed (as shown in fig.3). However, in practice this unusual phenomenon can be blurred by some neglected factors which are present in realistic systems, such as: Stark effect, σ states, σ-π hybridization or many-body effects.In figure 4 we plot the current-voltage (I-V) characteristics for both analyzed structures (in para – A and meta – B connections, respectively). The current steps are attributed to the discreteness of molecular energy levels as modified by the coupling with the electrodes [12]. Because this coupling is assumed to be small (bad contacts are suggested by experimental data [1]), the transmission peaks are very narrow and therefore the I-V dependence has a step-like character. In particular, the height of the step in the I-V curve is directly proportional to the area of the corresponding peak in the transmission spectrum. Since quantum interference is important in determining the magnitudes of the resonance peaks, it is also crucial for calculations of the tunneling current. Indeed, the magnitude of the current flowing through the device is very sensitive on the manner of attachment between anthracene molecule and metal surfaces. Large values of the current are predicted for device of configuration A, while reduction of the current by orders of magnitude is expected for device of configuration B (although the shape of the I-V curve is similar in both cases). Such reduction is caused by destructive interference.IV. SummaryIn this paper we have examined the possibility that quantum interference can substantially affect the conductance in molecular-scale devices. The results have shown a striking variation of all the transport characteristics depending on the geometry of the molecular system (its connection with the electrodes). Anyway, the quantum effect of destructive interference may be used within the molecular device to switch its conductivity on and off [8,9]. The existence of interference effects in molecular devices open the question of their control. The phase shift of molecular orbitals could be controlled by a transverse magnetic field or a longitudinal electric field. However, magnetic field seems to be too large to produce significant phase shift (according to our simulations – hundreds of Teslas). AcknowledgmentsAuthor is very grateful to B. Bułka, T. Kostyrko and B. Tobijaszewska for illuminating discussions. Special thanks are addressed to S. Robaszkiewicz for his stimulating suggestions.References1E-mail address: walczak@.pl[1] M. A. Reed, Proc. IEEE 87, 625 (1999) and references therein.[2] P. Sautet, C. Joachim, Chem. Phys. Lett. 153, 511 (1988).[3] V. Marvaud, J. P. Launay, C. Joachim, Chem. Phys. 177, 23 (1993).[4] M. N. Paddon-Row, K. D. Jordan, J. Am. Chem. Soc. 115, 2952 (1993).[5] E. Emberly, G. Kirczenow, J. Phys.: Condens. Matter 11, 6911 (1999).[6] M. Magoga, C. Joachim, Phys. Rev. B 59, 16011 (1999).[7] C. Untiedt, G. Rubio Bollinger, S. Vieira, N. Agraït, Phys. Rev. B 62, 9962 (2000).[8] R. Baer, D. Neuhauser, J. Am. Chem. Soc. 124, 4200 (2002).[9] R. Baer, D. Neuhauser, Chem. Phys. 281, 353 (2002).[10] H. Sellers, A. Ulman, Y. Shnidman, J. E. Eilers, J. Am. Chem. Soc. 115, 9389 (1993).[11] V. Mujica, M. Kemp, M. A. Ratner, J. Chem. Phys. 101, 6849 (1994);ibid. 101, 6856 (1994); ibid. 104, 7296 (1996).[12] S. Datta, Electronic transport in mesoscopic systems, Cambridge University Press,Cambridge 1995.[13] S. Datta, W. Tian, S. Hong, R. Reifenberger, J. I. Henderson, C. P. Kubiak,Phys. Rev. Lett. 79, 2530 (1997).[14] S. N. Yaliraki, A. E. Roitberg, C. Gonzalez, V. Mujica, M. A. Ratner,J. Chem. Phys. 111, 6997 (1999).[15] W. Tian, S. Datta, S. Hong, R. Reifenberger, J. I. Henderson, C. P. Kubiak,J. Chem. Phys. 109, 2874 (1998).[16] E. G. Emberly, G. Kirczenow, Nanotechnology 10, 285 (1999).[17] M. Kemp, A. Roitberg, V. Mujica, T. Wanta, M. A. Ratner,J. Phys. Chem. 100, 8349 (1996).[18] L. E. Hall, J. R. Reimers, N. S. Hush, K. Silverbrook, J. Chem. Phys. 112, 1510 (2000).[19] J. E. Han, V. H. Crespi, Appl. Phys. Lett. 79, 2829 (2001).[20] S. T. Pantelides, M. Di Ventra, N. D. Lang, Physica B 296, 72 (2001).[21] A. Cheong, A. E. Roitberg, V. Mujica, M. A. Ratner,J. Photochem. Photobiol. A 82, 81 (1994).。

G.GottsteinInstitut f¨u r Metallkunde und Metallphysik, RWTH Aachen,GermanyPhysical Foundations of Materials ScienceJune30,2005SpringerBerlin Heidelberg NewYorkHong Kong LondonMilan Paris TokyoContents1Recovery,Recrystallization,Grain Growth (1)1.1Phenomena and Terminology (1)1.2Energetics of Recrystallization (6)1.3Deformation Microstructure (10)1.4Recovery (14)1.5Nucleation (19)1.6Grain Boundary Migration (25)1.7Kinetics of Primary Recrystallization (28)1.8The Recrystallization Diagram (33)1.9Recrystallization in Homogeneous Alloys (35)1.10Recrystallization in Multiphase Alloys (36)1.11Normal Grain Growth (39)1.12Discontinuous Grain Growth(Secondary Recrystallization) (45)1.13Dynamic Recrystallization (46)1.14Recrystallization Textures (48)1.15Recrystallization in Nonmetallic Materials (54)References (55)1Recovery,Recrystallization,Grain Growth1.1Phenomena and TerminologyMany materials undergo a heat treatment during their processing that strongly impacts properties.During heat treatment subsequent to plastic deformation in particular the mechanical properties and the microstructure are strongly affected,while other physical properties (e.g.electrical resistivity)are scarcely influenced (Fig.7.1).0204060801001207505002502550751000200400600800grain size [µm]electrical conductivity [%]ductility [%]strain [%]annealing temperature [°C]recrystallization recoverygrain growthFig.1.1.The effect of cold forming and annealing on the properties of a Cu35%Zn alloy.121Recovery,Recrystallization,Grain GrowthIn the course of plastic deformation the strength of a material increases (strain hardening),while the strain remaining until fracture decreases.Con-versely,during heat treatment the strength decreases and ductility improves. By successive deformation and heat treatment,large degrees of deformation can be imposed on a material.The underlying physical reason for these phenomena are the dislocations, which are stored during plastic deformation and cause strain hardening.Their rearrangement and removal during annealing softens the material.In princi-ple,there are two different reasons for the loss of strength during annealing, recovery and recrystallization.Recrystallization is referred to as a reconstruction of the grain structure during annealing of deformed metals.It proceeds by generation and motion of high-angle grain boundaries that concurrently removes the deformed mi-crostructure.In contrast,recovery comprises all phenomena that are associ-ated with the rearrangement or annihilation of dislocations.More specifically, the term recrystallization as defined here relates to the most important among many recrystallization processes,namely primary”static”recrystallization. The term recrystallization,however,is commonly used in a much broader sense,by including all kinds of phenomena associated with grain boundary migration that lead to a lower free energy of the crystalline aggregate.These phenomena comprise in particular all processes of grain growth,granular re-structuring during deformation,and special cases of strong recovery.Generally,one distinguishes whether the processes occur during deforma-tion(dynamic recrystallization,dynamic recovery)or subsequent to cold form-ing during annealing treatment(static recrystallization,static recovery).If recrystallization occurs during heat treatment of a deformed material,at first the generation of small new grains is observed that grow at the expense of the deformed microstructure until they impinge and,eventually,completely1replace the deformed microstructure(Fig.7.2).This process—characterized by nucleation and nucleus growth—is referred to as primary recrystallization.Since the dislocation density in the material is not removed homogeneously,but discontinuously at moving grain boundaries it is also termed discontinuous recrystallization in association with the terminology of phase transformations.Although primary recrystallization is by far the most important annealing phenomenon,sometimes quite different microstructural changes are observed during annealing of a cold deformed mi-crostructure.In particular,after large cold deformation or in case that grain boundary migration is strongly impeded,for instance by dispersion of a sec-ond phase,strong recovery can occur.During this type of recovery even high-angle grain boundaries can be generated besides low-angle grain boundaries. In this case the new microstructure has been formed without the migration of high-angle grain boundaries,so this phenomenon is referred to as in-situ recrystallization.This process—as any recovery process—occurs homoge-neously throughout the microstructure and is,therefore,also referred to as continuous recrystallization to discriminate it from discontinuous(primary)1.1Phenomena and Terminology3(a)(b)(c)(d)(e)(f)Fig.1.2.Microstructural change during recrystallization of cold rolled Armco iron.recrystallization.If the degree of prior cold forming is small,it is observed that nucleation of new grains is suppressed.Specifically,existing grain boundaries migrate locally and remove the dislocation structure in the swept volumes (SIBM:strain-induced grain boundary motion).Figure 7.3gives an example of this process in aluminum.At low degrees of deformation not all grains are de-formed equally.During SIBM a less deformed crystal grows into the adjacent crystal of higher energy and removes the deformed microstructure of the con-sumed grain.The energetic reason for this grain boundary migration is the difference of stored deformation energy (i.e.dislocation density)in adjacent grains.If the heat treatment is continued after complete primary recrystalliza-tion —or in other essentially dislocation-free microstructures,for instance in cast structures —the grain size usually increases.The respective phenomena are subsumed under the term grain growth,which occurs in two modes.Either the average grain size continuously increases,which is referred to as continu-ous or normal grain growth (Fig.7.4),or only a few grains grow rapidly,while the other grains grow slowly or not at all.The latter process is called discon-tinuous or abnormal grain growth (Fig.7.5).Because of the similarity of its microscopic appearance to primary recrystallization (nucleation and nucleus growth),discontinuous grain growth is also termed secondary recrystalliza-tion.Note that only discontinuous grain growth is referred to as secondary recrystallization.It leads to large grains and its occurrence is usually unde-sirable in commercial materials.During grain growth not only the average grain size but the entire grain 11141Recovery,Recrystallization,GrainGrowthFig.1.3.SIBM (c.f.text)of aluminum annealed up to 130min.at 350◦Cafter 12%rolling deformation.The original position of the grain boundary is still visible [7.1].200µm 200µm 200µm(a)(b)(c)Fig.1.4.Microstructural change during continuous grain growth of Al-0.1%Mn after 95%cold rolling.Annealing temperature 450◦C.Fig.1.5.Discontinuous grain growth of high purity zinc at 240◦C in a hot stage of a microscope after 40%deformation.Annealed 25s (a),79min.(b),92min.(c),and 135min.(d).11.1Phenomena and Terminology 5size distribution changes,and this is characteristically different for continu-ous and discontinuous grain growth.During normal grain growth,the average (logarithmic)grain size lnD m is shifted to larger values,but the height of the maximum and the standard deviation remain unchanged (Fig.7.6a).Such behavior of the distribution is also referred to as self-similarity,i.e.,if the dis-tribution is plotted versus the normalized logarithmic grain size ln (D/D m ),the distribution does not change during normal grain growth.Of course,this requires that the distribution is normalized as any probability distribution.Normalization means in this context that the integral of the distribution has a constant value,for instance unity.If this were not the case,the maximum of the distribution would decrease,since fewer and fewer grains remain during grain growth.ln D ln Dln D m,aln D m(a)(b)Fig.1.6.Time dependence of grain size distribution for normal (a)and abnormal (b)grain growth(schematic).During discontinuous grain growth,however,the grain size distribution does not remain self-similar.Rather,during incomplete secondary recrystal-lization a bimodal distribution develops;where one distribution relates to the slowly-growing grains and the other distribution represents the rapidly-growing grains.The distribution of the remaining slowly-growing grains be-comes smaller and will disappear eventually,but its average grain size does not change.In contrast,the distribution of the few abnormally growing grains significantly changes,because the respective average grain size lnD m,a and the maximum frequency f max increase with increasing annealing time until ab-normal grain growth has gone to completion (Fig.7.6b).Normal grain growth usually ceases,if the grain size becomes compara-ble to the smallest specimen dimension,for instance the sheet thickness.In some cases,especially for thin sheet,discontinuous growth of a few grains is observed after continuous growth has come to an end.By choice of an appropriate gas atmosphere during annealing this process can be promoted,1161Recovery,Recrystallization,Grain Growthc1c0p cFig.1.7.Schematic representation of discontinuous precipitation.The supersatu-rated solution of concentration c0acts on the grain boundary as a chemical driving force p c.suppressed or even inverted.Because of its discontinuous appearance,and to distinguish it from discontinuous grain growth owing to different energetic reasons,this is referred to as tertiary recrystallization.A particular phenomenon of recrystallization can be observed during re-crystallization in supersaturated solid solutions if recrystallization occurs con-currently with the phase transformation.The otherwise impeded precipitation processes are accelerated by grain boundary diffusion in the moving grain boundary,and a two-phase microstructure appears behind the moving bound-1ary(Fig.7.7).This phenomenon is called discontinuous precipitation although according to its physical nature it is also a recrystallization process.The high driving forces for phase transformation can lead to high recrystallization rates.1.2Energetics of RecrystallizationIn contrast to the atomistic details of recrystallization,the energetic reasons for recrystallization are well understood.Basically,there is always a driving force on a grain boundary if Gibbs free energy,G,of a crystal is reduced during motion of the boundary.If an area element dA of a grain boundary is displaced by an infinitesimal distance,dx,the free energy will be changed bydG=−pdAdx=−pdV(1.1) where dV is the volume swept by the moving grain boundary.The termp=−dG/dV(1.2) is referred to as driving force;it has the dimension of gained free energy per unit volume(J/m3),but it can also be considered as a force acting per unit area on the grain boundary(N/m2),i.e.,as pressure on the grain boundary.The driving force for primary recrystallization is the stored energy of the dislocations.If a recrystallized grain grows into the deformed microstructure, most of the dislocations in the swept volume are consumed by the boundary, and a volume with a substantially lower dislocation density is left behind1.2Energetics of Recrystallization 7(about 1010[m −2]compared to 1016[m −2]in heavily deformed metals).The energy of a dislocation per unit length is given by (see Chapter 6)E d =12Gb 2(1.3)(G -shear modulus,b -Burgers vector).For a dislocation density,ρ,the driving force for primary recrystallization may be formulated (assuming that the remaining dislocation density is so small that it can be neglected)p =ρE d =12ρGb 2(1.4)For ρ∼=1016m −2,G ∼=5·104MPa,and b ∼=2·10−10m,the driving force amounts to about p =10MPa (107J/m 3≈2cal/cm 3),and compares well to values of stored energy measured by calorimetry.The driving force for grain growth results from the boundaries themselves because the total grain boundary area of the crystalline aggregate is reduced during grain growth.If a large grain grows into an environment of small grains with stable grain size,i.e.for discontinuous grain growth (Fig.7.8),the calculation of the driving force becomes simple.Assume that the consumed grains have a cuboidal shape.If their diameter is d and their grain boundary energy γ[J/m 2],then the grain boundary energy per unit volume and thus the driving force on a boundary sweeping such a volume equals p =3d 2γd 3=3γd(1.5)The factor 3results from the fact that each of the 6faces of a cube is shared by two adjacent grains.From a quantitative assessment we obtain p ∼=0.03MPa (=3·104J/m 3)for a grain size of the consumed grains d =10−4m typ-ical for recrystallized microstructures and γ≈1J/m 2.Obviously,the driving force even for discontinuous grain growth is smaller by orders of magnitude than for primary recrystallization.Therefore,it can be expected that grain growth phenomena proceed much more slowly or become noticeable only at much higher temperatures.The derivation of Eq.(7.5)is based on the assumption that a large crystal grows into a fine-grained microstructure and thus liberates the energy of the consumed grain boundaries,i.e.the driving force is the same for any element of the moving grain boundaries.An arbitrary area element of a moving bound-ary,in general,does not ”sense”the existence of the remote grain boundaries that provide the driving force.The force on this area element results from the fact that at the grain boundary junctions mechanical equilibrium has to be established,which always leads to a curvature of the boundary.A curved boundary,however,responds to a force by moving towards the center of cur-vature in order to become straighter and thus reduce its area.Therefore,the driving force on a boundary segment is given by the pressure on a curved 1181Recovery,Recrystallization,Grain Growth1035064Fig.1.8.Schematic representation of a primary recrystallized structure with differ-ent grain sizes.The numerals indicate the number of nearest neighbors of a grain.(Grain 50is growing abnormally,grain 10continuously,grain 3is shrinking.)surface.To calculate the magnitude of this driving force,let us consider the change of surface and volume of a spherical cap,or simply of an entire spherep =8πRγdR 4πR 2dR =2γR(1.6)The driving force in Eq.(7.6)resembles the one derived in Eq.(7.5)if the radius of curvature R is approximately the grain ually,however,the curvature of a grain boundary is small,and correspondingly the radius of curvature is substantially larger than the grain size (by factors of 5-to-10).Therefore,the driving force for continuous grain growth (Eq.(7.6))is about 5to 10times smaller than for discontinuous grain growth (Eq.(7.5))so thatcontinuous grain growth proceeds much more slowly than discontinuous grain growth or secondary recrystallization.The driving force for tertiary recrystallization is caused by the orientation dependence of the free surface energy.A grain exposed to the surface is bound to grow at the expense of its neighbors if its surface energy γ0(because of the crystallography of its surface)is smaller than that of its neighbors.For a thin sheet of width B and a grain size that is large compared to the thickness of the sheet h ,i.e.the grain boundaries extend through the entire sheet thickness,and they run perpendicular to the sheet plane (Fig.7.9),one obtains for the driving force p =2(γ02−γ01)Bdx Bhdx =2∆γ0h (1.7)Again this driving force is transmitted to any boundary element in the interior by a curvature of the grain boundary due to its motion on the surface.For ∆γ0∼=0.1J/m 2and a sheet thickness h ∼=10−4m,the driving force amounts to p ∼=2·10−3MPa (=2·103J/m 3).Since the surface energy depends on the ambient atmosphere,∆γ0can be made larger,smaller,or even change111111.2Energetics of Recrystallization 9dxBhγγFig.1.9.Calculating the driving force of tertiary recrystallization for γ01<γ02.sign by choice of an appropriate annealing atmosphere,and,therefore,ter-tiary recrystallization can be influenced correspondingly (see section 7.12).During discontinuous precipitation,recrystallization proceeds in a supersatu-rated solid solution with concurrent phase transformation.Therefore,besides the stored energy of cold work (Eq.(7.4))the chemical driving force for phase transformation also contributes to the total driving force.Let us assume that the concentration of the supersaturated solid solution is c 0,the correspondingsolvus temperature T 0(Fig.7.10)and the solubility limit is c 1at temperature T 1.For a regular solution the chemical driving force at temperature T 1results from the free energy of mixingp c =Q v Ωc 0(1−c 0)+kT 1Ω[c 0ln c 0+(1−c 0)ln (1−c 0)]− Q v Ωc 1(1−c 1)+kT 1Ω[c 1ln c 1+(1−c 1)ln (1−c 1)] (1.8)where Q v is the atomic heat of solution and Ωthe atomic volume.Since Q v can be obtained from the solvus lineTT 1T 0c 0c 1cαα + βFig.1.10.Detail of a binary phase diagram with limited solubility.111101Recovery,Recrystallization,Grain Growthc=exp(−Q v/kT)(1.9) we obtain for small concentrations and c1 c0p c∼=kΩ(T1−T0)c0ln c0(1.10)For5at%Ag in Cu a solid solution is obtained above780◦C.If this solution is quenched and annealed at300◦C a driving force of6·102MPa is obtained, i.e.,more than10times as much as for primary recrystallization.In addition to the cases represented here,there are abundant examples of driving forces acting on grain boundaries.Any energy state that depends on orientation can be used to exert a driving force on the grain boundary and to force it to move.Examples are magnetic and elastic energy density ow-ing to the orientational dependence of the magnetic susceptibility or Young’s modulus.The respective driving forces,however,are much smaller than those for primary recrystallization and grain growth(Table7.1),hence they play a minor role for the process of recrystallization.1.3Deformation MicrostructureRecrystallization always proceeds from a deformed microstructure by the for-mation of nuclei and their growth.Plastic deformation of metals is mainly caused by the motion of dislocations(see Chapter6).The mechanisms of deformation and the obtained deformation microstructures depend on the availability and mobility of the dislocations in the microstructure.The mo-bility of dislocations is markedly influenced by obstacles that hinder their motion,such as solute atoms,precipitates,other dislocations,and,in par-ticular,dissociation of the dislocation cores.The normalized stacking fault energyγSF/Gb≈˜γSF controls the dissociation width of the dislocations. The smaller˜γSF of a material,the larger is the dissociation.With increas-ing width of dissociation cross slip of screw dislocations and climb of edge dislocations become more difficult.Therefore,obstacles can not be as easily circumvented and eventually work hardening increases.In materials with low stacking fault energy theflow stress can even reach the stress level required to activate mechanical twinning,and if twinning constitutes a major defor-mation mechanism,it will markedly impact deformation microstructure.Besides internal obstacles and dislocation core structure,the deforma-tion temperature also exerts a major influence on the deformation behavior, as both cross slip and climb of dislocations are thermally activated processes. At low temperatures,therefore,twinning can become favored over dislocation motion.In essence,the deformation mechanisms of a material can be differ-ent in different temperature regimes.Important examples are the deformation behavior of copper and its alloys.After large degrees of deformation there are two major types of defor-mation microstructures in fcc metals.Depending on the magnitude of˜γSF11.3Deformation Microstructure11 Table1.1.The driving force of grain boundary migration.Estimated Source Equation Approximate value of parameters driving forcein[MP a] Storedρ=dislocation density∼1015/m2deformation p=12ρGb210energy Gb22=dislocation energy∼10−8J/mGrain boundary p=2γRγ=grain boundary energy∼0.5N/menergy R=grain boundary curvature radius∼10−4m1012d=sample thickness∼10−5mSurface energy p=2∆γ0d∆γ0=surface energy difference of two adjacent2·10−2grains∼0.1N/mChemical p=R(T1−T0)c0=concentration b=max.solubility at T06·102driving force ·c0ln c0T1(<T0)annealing temperature(5%Ag inCu at300◦C)material:BismuthMagnetic field p=µ0H22(χ1−χ2)H=magneticfield strength(107A/m)3·10−5χ1,χ2=magnetic susceptibilities of adjacent grainsmaterial:BismuthElastic energy p=σ22“1E1−1E2”H=elastic moduli of adjacent grains 2.5·10−4∼105MP a∆S=(difference in entropy between grainboundary and crystal(approx.equivalentto melting entropy)∼8·103J/K·molTemperature gradient p=∆S·2agradTV m2a=grain boundary thickness∼5·10−10m4·10−5 gradT=temperature gradient∼104K/mV m=molar volume∼10cm3/moland/or deformation temperature they are characterized by the appearance or absence of deformation twins.Even at low degrees of deformation dislocations are not distributed homo-geneously in the material.Rather dislocations cluster and eventually form a so-called cell structure with a distribution of cell sizes(Fig.7.11).Such a cell structure is characterized by cell walls with high dislocation density that enclose cell interiors of relatively low dislocation density.The character and appearance of a cell structure depends on the material and is mainly determined by the normalized stacking fault energy(˜γSF),degree of deforma-tion,and deformation temperature.With increasing temperature,respectively larger˜γSF,the thickness of the cell walls decreases until eventually sharp subgrain boundaries are formed,and concurrently the cell interior becomes1121Recovery,Recrystallization,GrainGrowthFig.1.11.Electron microscope image of the structure of a 10%rolled {112} 111 copper single crystal with irregular cell size distribution.The image plane is per-pendicular to the transversedirection.Fig.1.12.Electron microscope image of a ”brass-type”shear band in copper after 50%rolling deformation in liquid nitrogen.Twinning planes are parallel to the rolling plane.depleted of dislocations.The degree of deformation affects the cell size and the misorientation between adjacent cells.With increasing degree of deformation the average cell size decreases,and the orientation difference between adjacent cells increases.At large strains deformation inhomogeneities tend to form in a globular cell structure.They are referred to as bands,for instance kink bands during tensile deformation of single crystals,or shear bands during rolling (inclined35◦to the rolling plane)(Fig.7.12,7.13)and deformation bands (parallel to the rolling plane).These deformation inhomogeneities can contain orienta-tions quite different from the matrix orientation,and frequently an orientation111.3Deformation Microstructure13Fig.1.13.Electron microscope image of a ”copper-type”shear band in Cu 0.6%Cr after 95%rolling deformation.5102015123456789101112131415[]111x [µm]Fig.1.14.Dislocation structure and orientation dependence in a kink band of a tensile deformed 451 copper single crystal.141Recovery,Recrystallization,Grain Growthgradient is observed at the transition from the matrix to the bands (see kink band,Fig.7.14).1.4RecoveryThe deformed state of a material is principally unstable,because the disloca-tion structure generated during deformation is not in thermodynamic equi-librium.At low deformation temperatures,however,the deformed state is conserved after deformation because the structure is mechanically stable,i.e.,the dislocations of the microstructure are in a state of mechanical equilibrium.Upon increase of the temperature,however,this mechanical stability can be overcome by thermally activated processes,i.e.,cross slip of screw dislocations and climb of edge dislocations by which the dislocations unlock and move to interact with other dislocations.Through climb,dislocations can leave their glide planes to arrange in energetically more favorable patterns,can mutu-ally annihilate,or leave the crystal altogether.These processes are subsumed under the term recovery,which is always associated with a decrease of the dislocation density and the formation of special dislocation patterns,i.e.,net-works of low-angle grain boundaries,also referred to as polygonization.Recovery is caused by the interaction of dislocations via their long-range stress fields.For instance,the interaction of an edge dislocation with Burg-ers vector b 1with another parallel edge dislocation of Burgers vector b 2(see Chapter 6.4)is given by the interaction forceF =τb 2=Gb 1b 22πr d (1−ν)cos Φcos 2Φ(1.11)where r d and Φdenotes the position of dislocation 2with respect to disloca-tion 1(r d -distance between the dislocations,Φ-angular coordinate with Φ=0◦on the slip plane)and νis the Poisson ratio.If both dislocations are of the same sign (parallel dislocations)and lie on the same slip plane (Φ=0◦)the force is always positive,i.e.,the dislocations repel each other.If the dislocations have opposite sign,the force is negative,and the dislocations attract.If such antiparallel dislocations meet they re-combine and annihilate (Fig.7.15a).The same holds for antiparallel screw dislocations.Correspondingly,over time the dislocation density decreases.If antiparallel dislocations are on an adjacent slip plane,they do not annihilate,but rather form a dislocation dipole (Fig.7.15b)that corresponds to a chain of vacancies.Such a dipole has a much lower energy than the sum of the energies of two separate dislocations.By climb of a dislocation by one lattice spacing such a dipole can be annihilated,as observed in the electron microscope.Even if the dislocations are several lattice spacings apart,they will still interact and,in case of attraction,they can annihilate by climb.If the angu-lar coordinate Φ>45◦,the sign of the interaction force changes (Eq.(7.11)).Now antiparallel dislocations repel each other,but parallel dislocations attract11111.4Recovery 15(a)(b)vacancyFig.1.15.Illustration of the principle of annihilation (a)and dipole generation (b)by edge dislocations.each other along the glide plane.The equilibrium arrangement of two such parallel dislocations is an arrangement of one above the other.In this case Φ=90◦,and according to Eq.(7.11)F =0.Each displacement from this posi-tion results in a force to restore the equilibrium position.This arrangement is energetically most favorable.A large gain of energy is obtained if many dis-locations polygonize to align in a periodic pattern in the plane Φ=90◦,thus forming a low-angle symmetrical tilt grain boundary (LATB).The periodic arrangement of dislocations causes a superposition of the long-range stress fields in such a way that the range of the stress field is reduced to about the dislocation spacing r d .This is associated with a substantial decrease of the energy of each dislocation in the arrangement.If there are Z d dislocations per unit length in this arrangement,the energy per unit area isγLATB =Z d Gb 24π(1−ν)ln r d 2b+E C (1.12)(E C -energy of dislocation core).Such dislocation arrangement corresponds to a LATB as shown in Fig.7.16,and γLAT B in Eq.(7.12)reflects the specific energy of this low-angle grain boundary.Since the orientation difference Θof the adjacent grains is related to the dislocation spacing in the boundaryΘ=br d (1.13)and 1/r d =Θ/b =Z d is the number of dislocations per unit length in theLATB,Eq.(7.12)can be rearranged to yield the specific energy of the LATB,namely1111。