Paper to be presented at the Summer Conference 2009

on

CBS - Copenhagen Business SchoolSolbjerg Plads 3

DK2000 FrederiksbergDENMARK, June 17 - 19, 2009

THE EMERGENCE OF ARCHITECTURE IN MODULAR SYSTEMS: COORDINATIONACROSS BOUNDARIES AT ATLAS, CERN

Philipp TuertscherVienna University of Economics and Business

philipp.tuertscher@wu-wien.ac.at

Raghu GarudPennsylvania State University

rgarud@psu.edu

Markus NordbergCERN

markus.nordberg@cern.ch

Abstract:The development of complex technological systems can be coordinated by pre-specifying system architecture.However, what are the origins of architecture? To address this question, we explore the development of ATLAS,a complex technological system that is being developed at CERN, the European Organization for NuclearResearch. Our study suggests that the emergence of technological architecture is characterized by an ongoingprocess of negotiations across actors with different perspectives who have to justify and explain their designrationales to one another. Such a process results in what we call as interlaced knowledge partial overlappingof knowledge across actors in an overall knowledge network. Such interlaced knowledge generates a deepersystemic understanding of the requirements of the various components, making it possible for multiple groupsto anticipate latent interdependencies and coordinate interdependent contributions even as the design emerges.

JEL - codes: O32, O33, -

1

Please do not circulate as this is a first draft.

THE EMERGENCE OF ARCHITECTURE IN MODULAR SYSTEMS: COORDINATION ACROSS BOUNDARIES AT ATLAS, CERN1

Philipp Tuertscher Vienna University of Economics and Business

philipp.tuertscher@wu-wien.ac.at

Raghu Garud Pennsylvania State University

rgarud@psu.edu

Markus Nordberg CERN

markus.nordberg@cern.ch

February 28, 2009

Paper submitted for consideration for the DRUID conference at Copenhagen, 2009

1 We thank the many scientists at ATLAS, CERN who have contributed generously of their time. We also thank Carlyss Baldwin, Paul David, Simone Ferriani, Oliver Gasmann, Simon Grand, Barbara Gray, and Georg von Krogh for their valuable comments which helped to improve this paper.

2

THE EMERGENCE OF ARCHITECTURE IN MODULAR SYSTEMS: COORDINATION ACROSS BOUNDARIES AT ATLAS, CERN

ABSTRACT

The development of complex technological systems can be coordinated by pre-specifying

system architecture. However, what are the origins of architecture? To address this question, we

explore the development of ATLAS, a complex technological system that is being developed at

CERN, the European Organization for Nuclear Research. Our study suggests that the emergence of

technological architecture is characterized by an ongoing process of negotiations across actors with

different perspectives who have to justify and explain their design rationales to one another. Such a

process results in what we call as ‘interlaced knowledge’ – partial overlapping of knowledge across

actors in an overall knowledge network. Such interlaced knowledge generates a deeper systemic

understanding of the requirements of the various components, making it possible for multiple groups

to anticipate latent interdependencies and coordinate interdependent contributions even as the design

emerges.

3

Many technological systems are organized to benefit from the contributions of many different

organizations. To reduce transactions costs, the principle of decomposability (Simon, 1962) underlies

the division of labor across organizations within such technological systems. Decomposability

involves the division of complex problems into smaller subtasks so as to facilitate distributed problem

solving and reduce complexity. These distributed efforts must be integrated for the system to work, a

task that requires coordination across boundaries (Clark & Fujimoto, 1990; Ulrich, 2003). Such

coordination may start even before the development of a technological system. Typically, this involves

the specification of the functionality of the various components and their functionality as well as the

allocation of subtasks to different contributors (Baldwin & Clark, 2003; Langlois, 2006; Simon, 1996).

Given the uncertainties involved with innovation, however, such coordination can be

problematic. For instance, it may be difficult, if not impossible, to find the optimal decomposition of

the technological system ex ante when many system requirements and features have yet to be

identified and developed. Complicating matters, developments that occur with one component may

change the specifications of another. As a result, many design decisions may be only provisional

during the early stages of innovation (Mihm, Loch, & Huchzermeier, 2003).

How, then, are the various parts of a system coordinated during innovation? A way to provide

such coordination is to pre-specify architectures. However, pre-specification can both enable and

constrain actors in specific ways (Garud & Jain, 1996). Realizing this, the engaged actors actively try

to shape the emerging architectures based on their own frames of reference such that the emerging

architecture enables their vision of the project. The process of emergence, therefore, is likely to be

fraught with negotiations and potential conflict. In other words, not only are traditional coordination

mechanisms through architectures not available, but the very process of trying to institutionalize such

mechanisms generates dynamics that require coordination across a distributed set of actors.

These discussions point to two central questions that drive this paper. First, how do

architectures emerge? Specifically, what are the processes and underlying mechanisms that shape the

emergence of an architecture? A related question concerns a 'coordination of coordination' problem.

Specifically, what are the mechanisms that can engender coordination among actors trying to agree

upon an architecture that, only when it has been institutionalized, an provides embedded coordination?

4

To address these issues, we studied the development of ATLAS2, a complex technological

system that has been developed at CERN, the European Organization for Nuclear Research. The

development of ATLAS is revelatory with respect to several dimensions. First, the project is a

collaborative effort amongst approximately 2000 scientists and engineers from more than 165

institutions located in 34 countries. The different groups contributing to the project are distributed

around the globe with few opportunities to meet physically and engage in collocated collaboration.

Second, ATLAS involves many specialized scientists and engineers from a variety of backgrounds,

such as high-energy physics (HEP), semi-conductor technology, cryogenics, superconductivity,

material science, electronics, opto-electronics, electrical engineering, mechanical engineering, and

computer sciences, to name just the most important disciplines. Third, the scientists and engineers who

are involved in this decentralized project have been collaborating in the absence of traditional

organizational structures or hierarchy. Finally, the ATLAS detector has not evolved from a dominant

design but represents a one-of-a-kind technological system, making it an excellent case for studying

the emergence of technological architectures.3

We offer a perspective that acknowledges ongoing interaction between both the social and

technical facets of a given system. Actors are embedded within such systems, mobilizing its resources

towards their own individual and collective purposes. We use this broader perspective to motivate a

longitudinal study of the development of a complex innovation project, the ATLAS experiment at

CERN. Whereas prior literature on modularity focuses on specifications that are clearly understood a

priori, we suggest that there is value in understanding the role of ambiguity in the emergence of

architectures. In particular, we explore the role of different perspectives that arise from alternative

technological solutions and interpretations. We study how these diverse views are sorted out through a

process of confrontation and justification. Such engagement generates what we call ‘interlaced

knowledge’ – a partial overlap of knowledge across actors and groups. Such interlaced knowledge, we

suggest, facilitates coordination of the distributed development efforts as the architectures emerge.

2 ATLAS is an acronym for A Toroidal Large Hadron Collider Apparatus, but is now used simply as a name. 3 It should be noted that, ex ante, there was no inevitability in the emergence of a specific architecture – there are three other experiments that are unfolding, each with its own architecture. In fact, one of the experiments, CMS, also has been designed to identify the Higgs Boson particle just as the Atlas experiment has been, but, the architectures of the two detectors are very different.

5

MODULAR ARCHITECTURES

The existing literature on innovation points to the role of modularity and architecture in the

coordination of innovation systems (Baldwin & Clark, 2000; Henderson & Clark, 1990). A modular

architecture has certain advantages over other designs with respect to coordination (Sanchez &

Mahoney, 1996). This type of architecture consists of modules that can be used as self-contained

building blocks for high-level systems. Each module interacts with another only through standardized

interfaces that define functional, spatial, and other relationships among the components (Garud &

Kumaraswamy, 1995). In this way, modular architecture allows for a range of variations in its loosely

coupled components without having to change the designs of other components (Sanchez, 1995). As

long as components conform to standardized interface specifications, there is no need to directly

manage or monitor the processes of the individual component developers.

While modular components can be developed autonomously by different groups, coordination

of the overall development process is generated because of the presence of interface specifications

embedded in the architecture, thereby shifting the burden of coordination away from managerial

authority (Sanchez, 1995). Less need for managerial coordination makes it possible to engage in

parallel and distributed processes even while reducing costs (Garud & Kotha, 1994). In addition,

modularity increases the range of manageable complexities because the scope of interaction between

elements or tasks is limited by agreed-upon standard interface specifications (Baldwin & Clark, 2000).

Moreover, modularity is a means to accommodate uncertainty. Modular architecture is flexible

because standardized interfaces allow for variations in components (Ulrich & Eppinger, 2000) that

may be substituted in response to market and technological changes (Garud & Kumaraswamy, 1995;

Sanchez & Mahoney, 1994).

However, these benefits are obtained when interfaces have been specified and understood by

the actors, a situation that is typically encountered after the emergence of a dominant design. In many

innovation systems, however, architecture needs to be created in the first place (Wood & Brown,

1998). While several scholars have studied how dominant architectures are selected in an evolutionary

process on an industry level (Tushman & Murmann, 1998, 2003), there is known about the emergence

of a specific architecture on the organizational level. The puzzle is that, on the one hand, architecture

6

is required for coordination, but on the other hand, coordination is required for the architecture to

emerge.

How do architectures emerge?

Current research on modularity focuses on the natural partitioning of complex technological

systems. For instance, Baldwin and Clark (2003) suggest that technological systems have “natural

transfer bottlenecks” that represent optimal locations for the decomposition of complex technological

systems into components. Recent contributions provide tools and heuristics, such as the “design

structure matrix” (Baldwin & Clark, 2005; Eppinger, Whitney, Smith, & Gebala, 1994), to identify

natural boundaries between partitions of complex systems. The intent of such tools is to offer “design

rules” (Baldwin & Clark, 2000) that are useful for organizing and improving the architecture of

existing systems. Similarly, recent work on task decomposition by Ethiraj and Levinthal (2004)

suggests that the problem of designing a complex system can be represented as the search for the

optimal modular architecture. Common to these approaches is that they presuppose the existence of an

optimal architecture, determined by properties of the technology, that needs only to be discovered.

The partitioning of novel systems is not as simple as the literature suggests. Because many

interdependencies between components of emerging technological systems are latent (Alexander,

1964) and become visible only with developmental work, it is not feasible to identify natural

boundaries within novel systems even if they may seem obvious post hoc. Moreover, it is difficult to

discover architecture that is inherent when a system and its components are co-emerging (Garud &

Munir, 2008). Even if emerging architectures build upon existing technologies, novel ways of linking

the existing components may cause interference that may not have been observed in prior systems

(Barry & Rerup, 2006).

Consequently, rather than assume that designers can identify architectures inherent in

technological systems, we would like to draw attention to additional social factors that influence the

emergence of architecture. For instance, Bijker, Hughes and Pinch (1987) illustrated how “interpretive

flexibility” surrounds the emergence of technologies as different social groups, through their

engagement, shape the path through which technological systems develop. Indeed, strategic actions of

firms involved are critical in shaping these technological paths (Tushman & Rosenkopf, 1992). As

7

these paths emerge, institutional pressures in the form of standards (Tushman & Murmann, 2003) can

generate path dependencies (David, 1985).

The presence of these various technological, institutional and social contexts of the actors

involved in complex projects results in a situation where coordination is not automatically guaranteed

by the system architecture as it has yet to emerge. In determining the architecture, differences of

opinion are bound to arise as members from different communities with different perspectives interact.

Key design decisions revolve around issues such as detailing the boundaries of each component and

establishing stable interface specifications to ensure smooth functioning between modules. A number

of interdependencies may exist among components that make it difficult to identify the “one best”

(i.e., optimal) solution. For example, more physical space for one system could imply less for another.

If each group tries to optimize the performance of its own component, conflicting claims are bound to

arise as to ways the geometrical boundaries of the components should be specified. Similar differences

are likely to emerge around other specifications, such as, in the case of ATLAS, the thermal or

electromagnetic properties of interdependent components.

Consequently, a central task is to understand how a disparate group of actors with different

perspectives are able to coordinate their activities to develop an architecture. It is to this task that we

focus our attention. Specifically, we study the development of a complex technological system during

its early stages when its architecture had yet to emerge and stabilize. In our study, we consider

alternative interpretations of the architecture as it emerged, the controversies that were generated, and

how they were resolved (see Galison (1997) for disagreement that unleash social dynamics of

controversy particularly in scientific settings). This provides us with a more in-depth understanding of

the micro-processes of coordination across emergent boundaries in the emergence of a technological

system.

RESEARCH DESIGN

This research investigates the development and construction of a complex technological

system, in this case a detector that measures subatomic particles in a high energy physics (HEP)

experiment. Specifically, we study the ATLAS experiment at the European Organization for Nuclear

Research, known as CERN, in Geneva, Switzerland. The development, construction and running of

8

the actual experiment is organized in the form of a collaboration, i.e., a distributed problem-solving

network. ATLAS is an archetypal example of this type of scientific project; similar endeavors can be

found in basic physics, astronomy, and even life sciences (Galison & Stump, 1996).

The development of ATLAS is a revelatory case on several dimensions. First, the project is a

collaborative effort amongst approximately 2000 scientists and engineers from more than 165

institutions located in 34 countries. The different groups contributing to the project are distributed

around the globe with few opportunities to meet physically. Second, ATLAS involves many

specialized scientists and engineers from a variety of backgrounds, such as HEP, semi-conductor

technology, cryogenics, superconductivity, material science, electronics, opto-electronics, electrical

engineering, mechanical engineering, and computer sciences, to name just the most important

disciplines. Third, the scientists and engineers who are involved in this decentralized project have

been collaborating in the absence of traditional organizational structures and hierarchies. Finally, the

ATLAS detector has not evolved from a previous design thereby making it an excellent case for

studying the emergence of technological architectures.

Data Sources

We had the opportunity to collect different types of data from a variety of sources. Archival

records, meeting minutes, electronic mailing list archives and written reports were used as the main

sources of data for the longitudinal study that explored the emergence of the ATLAS detector.

Moreover, we were fortunate to obtain the right to use CERN archives that contain internal memos,

technical reviews and other similar material. In addition to documents from the CERN archives,

personal notes written by participants of the meetings offered insights into the micro details of

emergence that were not readily apparent in the meeting minutes.

Documents of the discourse that unfolded in co-located meetings were complemented by

electronic mailing list archives that captured real time ongoing conversations amongst development

groups who were globally dispersed. As a result, a significant amount of information regarding the

events, the participants, and the context within which the architecture emerged was readily available

and easily retrievable. The availability of this searchable mailing list archive, which contains tens of

9

thousands of messages dating back to 1993, proved very useful for reconstructing events in the

emergence of the ATLAS architecture.

During field visits, we conducted semi-structured interviews with 32 scientists and engineers

involved with the collaboration throughout the genesis of ATLAS. Some of these actors were

interviewed on multiple occasions. These respondents helped us make sense of the many technical

details. They also helped us identify controversies and differences in opinion that existed during the

emergence of the ATLAS architecture. Indeed, respondents had no difficulty remembering

controversial issues, such as around the ‘air core toroid decision’ or the ‘inner detector cooling

review’. In addition, during eight field visits conducted over a period of two years, each typically

lasting one week, we had the opportunity to observe conferences, participate in meetings, and engage

in many informal conversations with members of the ATLAS collaboration.

Data Analysis

We adopted a process perspective (Mohr, 1982) to explore the dynamics of emergent

architectures. In order to generate a longitudinal study, we relied on archival records that were created

from the very beginning of the collaboration (Van de Ven, 1992). In many cases, the same event had

been discussed in various types of meeting, such as working groups, review panels, or plenary

meetings. Although the descriptions of the events were often recounted from different perspectives,

they were mutually confirming, thereby generating confidence in the quality and depth of the data.

Overall, we analyzed several hundred pages of meeting minutes and correspondence that

covered the development of ATLAS from 1991 to 2003. We also studied documents such as the Letter

of Intent, Technical Proposal and the Technical Design Reports of the ATLAS components. These

detailed descriptions of the technological concepts and design considerations represented snapshots of

the ATLAS architecture for the years 1992, 1994, and 1997. These data were complemented by semi-

structured interviews and many informal conversations we had with scientists and engineers during

eight field visits over a period of two years.

We constructed a database that contains the controversial issues identified by the respondents

as well as other controversies that were captured in the archival data. The database includes the events

around these controversies as well as our interpretations (Miles & Huberman, 1984). We captured how

10

the controversial issues were resolved and tracked changes in the technical architecture before, during

and after each event.

Consistent with a process perspective, we considered each event as an important occurrence

within a larger flow of events (Van de Ven, 1992); an approach that helped us generate a deep and

consistent understanding of the unfolding processes. We focused on the controversial issues that were

identified and paid particular attention to the motivations and justifications of the actors involved. In

particular, we studied if and how these controversies were resolved as the advocates of the different

technological components and the various groups confronted one another in meetings and review

panels. As Knorr-Cetina (1999: 190) noted in her work on HEP experiments, “discourse is … the

activity through which much of the construction work is not only exhibited, but done.”

Beyond the historical reconstruction of the emergence of the ATLAS detector through the

analysis of archival records, we also used additional approaches to data analysis to complement the

historical account. Specifically, we used bibliometrics and social network analysis to infer the

collaborative structure over time and identify the locus of problem-solving activity throughout the

emergence of ATLAS. We used Qualitative data from archival sources, interviews, and electronic

databases to identify instances of justification in different subsystem communities. For instance, we

used ‘latent semantic analysis’ (Deerwester, Dumais, Furnas, Landauer, & Harshman, 1990) to

explore discourse around controversies that we had identified through our qualitative probe of the

data. To understand the structure of the distributed yet collective knowledge that emerged over time,

we used co-word analysis. In combination, latent semantic analysis and co-word analysis helped us to

generate graphs of different subsystem’s knowledge structures using network analysis techniques. We

present some of these results in this paper.

THE EMERGENCE OF ATLAS

ATLAS is one of four particle detector experiments unfolding at the Large Hadron Collider

(LHC), a new particle accelerator at CERN. The detector, completed in 2008, is 45 meters long and 25

meters in diameter and weighs about 7,000 tons. By measuring the reactions of massive particles not

measurable by earlier low-energy accelerators, ATLAS is expected to shed light on new theories of

11

particle physics that go beyond the Standard Model, the most comprehensive model of particle

interactions available today.

The ATLAS project, now involving around 2,000 scientists and engineers from 167

institutions from 37 countries – has its roots in UA1 and UA2, two very successful HEP experiments

conducted at CERN in the 1980s. These experiments became well known for the discovery of the W

and Z bosons that led to a Nobel Prize awarded to CERN physicists in 1984. While these two

experiments were running, a core group initiated the development of future particle detector concepts.4

Members of these “proto-collaborations”, as they are called by Knorr-Cetina (1995), met during HEP

conferences where they discussed new concepts. In 1989, CERN started funding institutions to

conduct R&D and to develop technologies for new generation detectors. Four independent

collaboration clusters emerged from this initiative in the early 1990s. Each cluster was comprised of

multiple R&D projects (please see Table 1 for an overview) and each was built around a new detector

concept.

--- Table 1 here ---

A critical incident occurred in 1992 when CERN decided that it was not feasible to continue

with four large experiments. As a consequence, the collaborators tried to join forces to develop and

construct two independent detectors – ATLAS and CMS.5 The various proto-collaborations had

already developed idiosyncratic technological paths. That is, although the functional requirements for

each particle detector were the same, each group had opted for different technological solutions that

were manifest in the form of different architectures. Consequently, the independent groups were

forced to think about ways that their design concepts could be merged into a new architecture.

Besides seeking to integrate the technological approaches, it was also important for the

institutions to be organizationally aligned. Two of the collaborations, Experiment for Accurate

Gamma, Lepton and Energy measurement (EAGLE) and Apparatus with Superconducting Toroids

4 This parallel development of upgrades and future generations of experiments is typical of HEP experiments because development and construction of a new detector requires 10 – 15 years of work. While ATLAS was being developed, some groups within the collaboration were already working on upgrades for 2015 and beyond. 5 According to the scientists we interviewed, they always have checks and balances by conducting at least two experiments. There are two other smaller experiments as well besides ATLAS and CMS.

12

(ASCOT), seemed to be the most similar; both had opted for a complex configuration consisting of

two magnets. Consequently, the two groups decided to merge, and formed ATLAS.

The emergence of controversies

EAGLE and ASCOT included four key components: an inner tracker, a calorimeter system, a

muon spectrometer and a magnet system (see Figure 1 and Table 2). The two collaborations had

placed different emphases on the different components. The core feature of the EAGLE concept was

an elaborate calorimeter (a hybrid calorimeter consisting of two independent subsystems) combined

with a powerful inner tracker. By contrast, the ASCOT architecture was dominated by a

superconducting magnet system required for its powerful muon spectrometer.

--- Figure 1 and Table 2 here ---

A reason for these differences between EAGLE and ASCOT is the fundamental uncertainty

that underlies high energy physics. For each of the four components of the detector, several different

technologies were considered as potential candidates for the emerging design. Indeed, options had to

be developed before a confirmation could be sought as to whether or not a particular approach would

meet the requirements of a challenging environment. With such irreducible uncertainty, preferences

and prior experiences of the EAGLE and ASCOT groups played an important role in the choices that

they had made. Each group focused on the components with which they were most familiar and had

most experience. Because ASCOT’s strength was in muon detection, its design was constrained by the

characteristics of a muon system designed around a very large superconducting magnet system.

EAGLE, on the other hand, was spearheaded by a group with a strong background in calorimetry.

Smaller groups, which had strong interests in inner tracker systems using semiconductor technologies,

complemented the EAGLE community.

Scientists of both EAGLE and ASCOT clearly were closely attached to their respective

proposal. Therefore, members of the ATLAS collaboration agreed upon a convention that competing

options “…can and should be analyzed in a scientific (and not emotional) way.”6 The objective was to

ensure that controversies, if any, would largely be of a technical nature. Beyond performance issues,

competing technologies would be evaluated in terms of cost, construction time and technical risks, as

6 “Emotional way” refers to political controversies between competing groups at that time.

13

well as on the basis of the impact that choices would have on other subsystems. Working groups

identified areas of consensus and tried to find common solutions. A series of ‘concept optimization

meetings’ were held between members of different groups to discuss the strengths and weaknesses of

each approach. Based on these insights, representatives of EAGLE and ASCOT started to negotiate

ways that their technologies and personnel could potentially be integrated into a single experiment.

Divergent opinions emerged when decisions were to be made for and against specific

technologies. A major concern was that a vote against a particular technology could alienate some

members and that ATLAS could thereby potentially lose necessary critical knowledge, manpower and

funding as a result. Consequently, a process was embraced that could accommodate negotiations

between the two groups.

Going beyond the controversies generated by differences in technological perspectives, the

actual composition of each of the two groups also began to play a role. ASCOT was dominated by two

large French and German research institutes. EAGLE, by contrast, was a larger community, but

consisted of small to medium sized groups. Whereas big national laboratories have a tendency to work

on large systems, such as muon spectrometry, small institutes often focus on areas that do not require

large infrastructural investments, such as chip design for the inner tracker (please see Table 3 for an

overview of the countries with research interests per subsystem). These smaller institutes with a strong

stake in the EAGLE inner detector were afraid that the large institutes could force adoption of

ASCOT’s superconducting toroid into the collaboration, an eventuality that would result in the

domination of the design by larger components.

--- Table 3 here ---

It became apparent during the course of negotiations that there were two fundamentally

different sets of assumptions at play. Both EAGLE and ASCOT had proposed a large toroid magnet to

create an independent measuring capability for muons7 and other highly energetic particles. The

EAGLE group wanted to increase the reliability of the experiment with this additional measurement

capability, which would become a significant advantage once the collaboration made new discoveries.

7 These muon particles were considered to be very important for the experiment because theories predict that they are related to the Higgs mechanism that physicists need to discover in order to explain physics beyond the Standard Model.

14

For example, when the detector identifies a new particle, the measurements taken by the muon

spectrometer can be compared with observations from more central detectors that are located closer to

the collision, such as the inner detector. The ASCOT group, on the other hand, feared a failure of the

inner detector because of the unprecedented energy that would be released – 700% higher than in

previous experiments. Components close to the collisions, such as the inner detector, are especially

subject to aging and radiation damage. For the ASCOT group, a separate muon spectrometer would

provide the redundancy required to overcome any damage that might occur to the inner detector.

The two groups had different perceptions of this risk based on their assumptions. EAGLE

expected to operate at low to moderate luminosity levels during the first phase of the experiment

because it usually takes several years for accelerators to reach their full potential. Therefore, EAGLE

emphasized a balanced detector concept featuring an elaborate inner tracker for optimal particle

detection at low luminosity levels. The inner detector would sustain this first phase and could be

upgraded at a later stage if necessary. ASCOT, on the other hand, expected early operation of the

accelerator at high luminosity levels. In its scenario, which was driven by expected competition from

the Superconducting Super Collider planned in the United States,8 the risk of damage to the inner

detector was much higher than that anticipated by EAGLE. ASCOT, therefore, optimized the overall

detector layout for the muon spectrometer while the inner tracker was a relatively simple device,

because the inner tracker was expected to fail at high luminosity levels in any case.

In short, the different socio-cognitive bases of the two groups created a collaborative

challenge. Specifically, individuals from EAGLE and ASCOT were driven by different sets of beliefs,

artifacts and evaluation routines. Although a convergence of ideas and practices within each

collaboration facilitated exchange of information and ideas, it was difficult to generate an

understanding across the boundaries of the collaboration because each group was locked into its own

ways of thinking.

Unfolding of Architecture

In the Letter of Intent signed in October 1992, the 88 institutions already involved in the

collaboration explicitly agreed on a shared schema that included provisional specifications of the key

8 The Superconducting Super Collider was a particle accelerator which was planned to be built in Texas. During the design and first construction phase, Congress cancelled the project in 1993 by due to rising cost estimates.

15

properties and functionalities of the various ATLAS components and the ways that these would

interact with one another. Although the overall detector concept outlined the basic design

consideration, the choice of the specific technologies for the different components was left open. The

R&D projects conducted by participating institutes since the late 1980s resulted in various

technologies that potentially could be used for parts of detectors. Development of multiple

technological options was considered to be important during the early phases as it was unclear which

of the new technologies would be capable of operating under the harsh radiation conditions of the

LHC.

The Letter of Intent listed several competing technological options for each subsystem. In

most cases, a preferred technology was selected as a “baseline design” with alternative options that

would be chosen if the baseline turned out to be untenable. However, in some cases, several R&D

projects were continued even when it was clear that the baseline technology was the most appropriate

one to pursue. Such continuation of some of the R&D projects allowed technological decisions to be

postponed in the hope of retaining members who favored options that had not been chosen. The

ATLAS community hoped that, by continuing such R&D projects and testing prototypes, the

superiority of a particular approach over others would become apparent to all participants and would

encourage the proponents of paths not chosen to nevertheless remain engaged with ATLAS (cf. Knorr-

Cetina, 1995).

Another reason to defer decision-making was to gain the time required to set up review panels

that could evaluate the various technologies. The review panels did not have any mandate to select a

particular technology; rather, they were set up to investigate the potential of the competing

technologies and to make recommendations to the collaboration board, which consisted of all ATLAS

members. The collaboration nominated senior scientists, highly respected by those in the HEP

community, to serve on these panels. The aim of the review panels was to reduce the role of politics

by allowing obvious decisions to emerge. As a physicist described, the review panel operated in ways

similar to a tribunal:

The panelists who are chosen by the management are the judges, and you have two parts against each other. In our case, for the hadronic part [of the calorimeter], we were against the liquid argon people, and we were looking for problems of their approach and they were looking for problems with our approach. The whole thing was relatively formal. We would

16

present our results and our calculations, and they would present their results and their calculations. And then we would ask them nasty questions in writing, and they would ask us nasty questions in writing. And then, at the next meeting, there would be the answers to these questions. The processes that unfolded in these review panels challenged both the proponents and the

opposition of a given technology. Many assumptions that had been taken for granted by the

proponents of a certain technology were now challenged by the members of the review panel or by

scientists who proposed competing approaches. Each group had to justify their assumptions – through

analytical reasoning, evidence from simulation studies and tests of prototypes, or support from

external experts. Although the scientists worked hard to justify their technologies, even more

demanding was the effort to identify potential shortcomings of competing technologies and to ask

probing questions, as this effort required a deep understanding of the concepts proposed by the

competing groups.

To some extent, the competing technologies could be exchanged with little impact on the

overall architecture. Some solutions, however, required an adaptation of the architecture, thereby

critically shaping the developmental path of the overall detector. Therefore, mastery of the question

and answer procedure in the review process required knowledge not only of the components that the

groups were creating, but also the interdependencies with other systems and the reasons that such

interdependencies were critical. For example, during the review for the calorimeter, it was essential

that the groups understood the implications of the proposed specifications for neighboring systems,

such as the inner detector. As an ATLAS scientist explained:

(The) liquid argon calorimeter has a lot of advantages and so on, but it is a very expensive technology. It has a lot of channels and so on and so forth. This is kind of forcing you to a small radius, and therefore has certain implications on the inner detector. The liquid argon calorimeter group prepared to defend against this possible critique and

worked on solutions for this conflict between the two interdependent subsystems. Specifically, the

solenoid magnet, a part required for the inner detector to identify charged particles, was integrated into

the cryostat of the calorimeter. In this way, material and space constraints for the inner detector were

mitigated, and the proponents of the liquid argon calorimeter could justify the viability of their

proposal.

17

The ATLAS scientists had come from different backgrounds and, therefore, had different

knowledge bases and experiences with these technologies. As a result, they also had varying

judgments as to which technological option would be most appropriate for a particular component of

ATLAS. Specifically, the scientists tended to have great confidence in their technologies because they

had worked with them for a long time. From their experiences, these scientists could anticipate

potential problems and knew the options that could offer a potential fix. These scientists were

therefore convinced that their own proposal was preferable in terms of feasibility and reliability and

risk and cost because of this familiarity. However, when confronted with alternative perspectives, they

were prompted to scrutinize their own proposals and rethink some of the assumptions they had earlier

taken for granted.

The confrontation process was facilitated by the eagerness of competing groups to justify the

superiority of their own proposals and was mediated by very experienced panel members who often

had a strong background in related ATLAS components. From this review process, a collective

knowledge structure began to emerge that provided scientists of one system with an understanding of

ways that related components, and ATLAS as a whole, worked. We label this partially overlapping

knowledge structure that emerged as ‘interlaced knowledge’, a knowledge structure that has both

individual and collective properties.

The knowledge that emerged from the review panels benefited not only the immediate

participants but also others involved in the collaboration as a whole. Part of the motivation behind the

panel sessions was to generate a consensus. Although a few private panel meetings took place, for the

most part, a group of scientists from each detector would be present, and some of the meetings were

completely public. In any case, the collaboration was marked by constant discourse and

communication, which were comprised of talk threads, emails, meeting presentations and transparency

exchanges. As Knorr-Cetina (1995) describes, HEP experiments were “mapped into a fine grid of

discourse spaces created by intersections between participants.” This narration and account of the

groups’ activities and experiences with equipment, data sets, physics calculations and the like rendered

the emergent interlaced knowledge both collective and dispersed.

18

The creation of interlaced knowledge throughout the collaboration, combined with the deferral

of decision-making to the dispersed yet collective knowledge structure, was characteristic of the

selection process of relevant options that regulated work at ATLAS. Rather than deliberate decision

making driving the emergence of the ATLAS architecture, it was left to emerge. Routinely, the

outcomes of review panels and other discourses pointed the system towards particular directions that

built upon the challenges and events that had already transpired, such as the problems that were

encountered in real time as well as future possibilities that were yet unknown. Through the constant

unraveling of the features of technical components, the proponents of competing technologies tried to

create an understanding throughout the collaboration as to the reasons that their design was the

solution for the overall detector in terms of performance, cost and risk. Decisions at ATLAS were left

to emerge as the obvious, reasonable or unavoidable options. As a senior scientist involved in the

muon spectrometer expressed, “When decision-making was an item on the agenda, this often meant

that something which was already agreed upon and clear for everyone in the collaboration was made

plausible and formally approved.”

This process, which deferred decision making and was based on interlaced knowledge,

minimized the dysfunctional facets of politics in the decision making process. Consequently, the

collaboration did not lose members, even those whose proposals had been turned down. If their

opponents had made a convincing argument, the scientists were usually willing to give in and even to

suggest that the better design should be implemented.

Changes in the Architecture as the Design Emerged

The properties of many of the technologies used for ATLAS components could only be

estimated due to the lack of sufficient prior experience. The performance of these technologies and

their impact on other parts of ATLAS could be determined only by educated guesses and complex

simulations. As a consequence, to determine an overall architecture that would remain stable during

the development of the overall experiment was extremely difficult. The design, therefore, was left

incomplete; many specifications remained preliminary on purpose (Garud, Jain, & Tuertscher, 2008) .

A specification of ATLAS that generated a considerable amount of uncertainty concerned the

size of the superconducting air core toroid and the related muon spectrometer. Originally designed to

19

consist of 12 coils and an inner radius of 5 meters, the costs and risks of designing and deploying this

gigantic magnet system turned out to be much greater than had been anticipated. While scientists were

already working on the designs of other ATLAS components based on the original specifications, the

ATLAS toroid was scaled down to have 8 coils with an inner radius of only 4.3 meters, a change in

specifications that triggered controversies over other components.

For instance, the downscaling of the magnet system benefited the muon spectrometer because

a smaller radius meant that the muon spectrometer had to cover a smaller area and, thus, required

fewer expensive muon chambers. The calorimeter could adapt to this change by also reducing its inner

radius, while maintaining the thickness required to effectively measure the energy of the particles.

However, the situation was different for the inner detector because surrounding components had

already squeezed its design. This further reduction in size meant that the space envisaged for the inner

detector electronics would no longer be available. Therefore, these components had to be housed

elsewhere, a step that required other components to sacrifice space in order to host the inner detector

electronics. As one person recounted:

What I negotiated with the muons was a gap at the end of the barrel calorimeter, so the services9 came out and went straight out, or 50 percent of them did, between the inner layer and the second layer of the muon chambers where there is more space to put a panel. And that has had an impact on the muons. They had to have a gap. We also have to cool the inner detector services in that region so that it does not warm up the muon electronics and the muon chambers themselves, which is quite critical. Moreover, cables to the inner detector electronics added additional materials which, in turn,

generated heat from their millions of readouts. Both of these were negative side effects for the liquid

argon calorimeter. To change the location of the electronics was not a trivial task; cables would have

to be routed next to power supplies and through strong magnetic fields. To protect the electronic

cables from picking up noise, effective shielding was required, which, in turn, required additional

space and material.

Eventually, the need for more space inside the inner detector triggered additional architectural

changes. For example, the gap between the barrel and the end-cap calorimeter was widened on two

occasions to allow for additional space for shielded cables and cooling. Although services such as

9 In ATLAS jargon, services are the pipes and cables for the various detector systems, such as power supplies, cooling lines, cryogenics and read-out cables.

20

cooling and cables posed less of a problem for peripheral components (as they could be routed outside

the detector), services for the inner detector began to invade the space of both the calorimeter and

muon spectrometer. However, it was not too difficult for the inner detector group to justify their need

for extra space for cables and cooling as the increased need for services was partly due to constraints

imposed by the outside detectors.

Besides space limitations, the goal to minimize materials led to controversy between members

working on the inner detector and those working on surrounding components. For example, whereas

the inner detector engineers preferred additional material for cooling and shielding, the calorimeter

group preferred less material in front of the calorimeter because the amount of material used in the

inner detector would determine particle reactions before they could enter the calorimeter, thereby

biasing measurements.

The calorimeter group, therefore, decided to lobby for an inner detector design with fewer

components. Specifically, there was discussion that the inner detector should have been built without a

transition radiation tracker (TRT). However, the inner detector scientists could demonstrate in

complex simulations that the TRT was critical for achieving the physics performance specified by

ATLAS and could convince members of the review panel that the amount of material was justified.

Consequently, it was decided to adapt the design of the calorimeter to account for the increased

amount of material. The design of the liquid argon calorimeter was changed to introduce an additional

measurement layer for the detection of electron and photon showers caused by material located in

front of the calorimeter.

The interlaced knowledge that was created and emerged through such justification processes

became more important as the architecture continued to change. Indeed, interlaced knowledge enabled

adaptive coordination when problems occurred that could not be solved locally by members of a

module but instead required changes across components. Interlaced knowledge enabled scientists of

the different components to interrelate carefully and with consideration and to anticipate interference

with other parts of ATLAS.

For example, it was a review of the cooling system that identified potential risks and

technological problems inherent to the inner detector. The cooling system was based on binary ice, a

21

coolant consisting of ice crystals in a cooling liquid, which was pumped through a complex system of

pipes. Scientists of other components, such as the liquid argon calorimeter, who tried to avoid large

amounts of binary ice, pointed to problems associated with water-based cooling, such as water

leakages that could destroy parts of the detector. Although this risk was highest for the inner detector,

it was not recognized as a threat by the inner detector team. Like all other ATLAS components, the

inner detector was designed by scientists and engineers who possessed local knowledge and

incentives. Of greater concern to these engineers working on the inner detector was the problem of

extricating heat from the densely packed inner detector. It is for this reason that they had focused on

the superior performance of binary ice cooling.

Interestingly, groups other than the inner detector designers perceived the risk of water

leakage more readily. The better choice from the perspective of those working on the inner detector

was considered to be an inferior choice from the perspective of those working on the calorimeter.

Scientists and engineers working on other components were less focused on the cooling performance

of binary ice. Instead, they were concerned about the amount of material that was introduced by binary

ice cooling. This concern – and a difference in perspective – enabled this group to draw attention to

the risks involved in using binary ice, and prompted the group to propose an evaporative cooling

system instead. The resolution of this controversy resulted in a design that not only used less material

but also minimized the risk of water leaks in the inner detector.

The complex task of designing the inner detector, given difficult space and material

constraints, resulted in further changes in the ATLAS architecture. The pressure to use more efficient

electronics forced the inner detector group to change the design of the pixel system and adopt a new

type of semiconductor. Controversy arose over conflicting schedules due to this change; because this

technology was new, further testing and R&D work were required, thus resulting in a delay of the

pixel system of the inner detector. An inner detector scientist explained:

We thought having three systems, the pixels, the silicon strip detector and TRT, to have them all ready at the same time, it is going to be very difficult because the pixel changed…. Then we changed the system. We put the silicon strip detector together with the TRT and then also put in the end-cap. We then put in the pixel system in the end. This means that the schedule of the pixel system was decoupled from the schedule of the two other systems.

22

As this example illustrates, conflicting schedules resulted in the decoupling of elements of the

inner detector that were to be integrated into other components of ATLAS. To implement such a

change in the architecture of ATLAS, agreed-upon interfaces had to be renegotiated at a very late

stage of development. A change at this stage was only possible with the support of the other groups

who recognized that a delay of the inner detector development would create problems. Such an

awareness of the status of the various components was possible due to the reports and justifications

that had been presented in the numerous review panels, working groups and plenary meetings.

These negotiations occurred not only during early phases of ATLAS, but were ongoing across

the collaboration throughout development and construction phases as well. A fresh round of

negotiations was triggered when the collaboration encountered situations where earlier specifications

were found to be unfeasible. Even basic dimensions such as radii, lengths and allowances of materials

in each subsystem had to be adapted over time.

These observations suggest that coordination, rather than being generated by an architecture

that embeds pre-specified interfaces as proposed by the modularity literature, can nevertheless be

accomplished through a process of negotiations. Indeed, the need for managerial intervention is not

embedded in a clearly specified, stable modular architecture. Rather, self-coordination of the

collaboration is enabled by interlaced knowledge that allows for mutual adjustments of the

interdependent actors and groups involved in the ATLAS collaboration.

DISCUSSION

The longitudinal analysis of ATLAS suggests that during the development of complex

technological systems, architectures are neither clearly understood by participants nor are they stable.

For instance, in the case of ATLAS, although a very rudimentary architecture of the experiment

existed at the very beginning of the experiment, this structure was far from being understood by all

participants. Instead, this architecture experienced significant changes over time. Almost on a daily

basis, geometrical and functional boundaries of interdependent components were challenged, resulting

in a renegotiation of specifications that continued even during the construction phase.

Non-sequential Search for Design

23

Problem solving in the development of ATLAS did not progress sequentially from top level

problems to design problems at the lower levels of the architecture, an observation that is contrary to

predictions made by prior research (e.g., Clark, 1985; Murmann & Frenken, 2006; Tushman &

Murmann, 1998). Rather, early design tasks focused on medium and low level problems that were

solved in a decentralized manner, whereas integration and top level problem solving often became an

issue only later in the process. A bibliometric analysis of ATLAS publication data clearly visualizes

this pattern (please see Figure 2). The result of this analysis indicates when and where groups formed

and how they interacted. The locus of activity is seen at the periphery in the early design phase (blue

lines) and gradually shifted towards the center as the design of various components crystallized (green

lines). Most of the activity in the center, however, is represented by yellow and red lines, indicating

that activities regarding integration occurred mostly in the later phases of ATLAS development.

--- Figure 2 here ---

As a consequence, the design of ATLAS emerged as a non-sequential search process in which

many choices were only preliminary. This decentralized approach, and its many iterations, gave rise to

technological controversy over the basic architecture, controversy among technologies in competition

for a subsystem, and controversy within a subsystem and across subsystems as unforeseen problems

required changes in the specified design. In sum, the coordination of the dispersed development and

construction work of the more than 150 institutions could not be provided by simply pre-specifying

system boundaries and interfaces.

Given the absence of a central organizational structure, the coordination of the decentralized

activities of 2,000 scientists and engineers to create a functional design required a different solution. A

striking facet of the ATLAS case, in this regard, was the presence of ongoing negotiations and the

creation of compromises and consensus. In ATLAS, political issues were harnessed to generate

collaboration. The architecture of the experiment to be built was not defined only by some scientists

and engineers (the actors in this case) ‘winning out’ or taking control, but through a process of

confrontation and justification of the positions held by the various scientists and engineers, including

physicists who had a big stake and interest in these matters. By the public summation of the

24

knowledge that emerged, the collaboration was able to overcome political strategies by pursuing a

strategy that allowed the obvious decisions to emerge.

Creation of interlaced knowledge

In particular, a process of justification was the hallmark of the negotiations associated with the

review process. Specifically, during negotiations, it was essential for proponents of a particular

technology to explain the reasons that their choice was the superior solution (see also Galison (1987)

for how demonstration plays an important role as to how experiments end). Figure 3 indicates that not

only were such justifications important during the initial specification of the architecture, but also

when controversies arose as the architecture emerged. If a group wanted to change a technological

path, it had to mobilize the support of other groups throughout the collaboration. In order to convince

the collaboration that the new proposal would result in a better outcome, it was essential to present

reasonable evidence to justify the group’s claims.

--- Figure 3 here ---

Because of the openness of the ATLAS collaboration and the involvement of individuals from

multiple groups, justifications were scrutinized and were always available for inspection. In this

interactive process, many different points of view and component-specific perspectives were

articulated. These observations are in line with current research on epistemology, which argues that

justification is a key process for creating organizational knowledge (Boltanski & Thévenot, 2006;

Garud, 1997; Nonaka, 1994). Justification essentially determines whether claims of individual groups

are rejected by others in the organization, or believed to be valid and incorporated into their

knowledge bases (von Krogh & Grand, 2000).

While negotiations can be seen as reaching a compromise, this process also contributes to the

building of consensus at the knowledge level. At ATLAS, such ongoing negotiations resulted in a

reconfiguration and recontextualization of available knowledge. By the time differences had been

resolved, competing groups had developed a shared understanding of each other’s components and the

roles that these components play within the overall technological architecture. As the various groups

created such overlap across their local knowledge bases, they developed interlaced knowledge across

interdependent components.

25

By choosing the term interlaced knowledge, we differentiate this concept from others, such as

common knowledge (Grant, 1996). We would like to emphasize that overlaps between and among the

various knowledge bases do not imply a common knowledge base shared by all members in the

collaboration. On the contrary, interlaced knowledge was created at various levels, reflecting a fine

“grid of discourse” (Knorr-Cetina, 1999) within the ATLAS collaboration. The content of the

knowledge and the degree to which it was shared varied throughout the collaboration, depending on

the controversies within local working groups, review panels and plenary meetings that involved the

collaboration at large. The three panels in Figure 4, for example, visualize the interlaced knowledge

structure of the ATLAS muon spectrometer community for the early design phase (1994), the

advanced stage of detailed development (1998) and the late stage of construction and installation of

the detector components (2006). Figure 5 suggests a similar temporal pattern for the interlaced

knowledge structure in all major ATLAS subsystems.

--- Figure 4, Figure 5 ---

Interlaced knowledge at ATLAS comprised various levels of detail. At the overall

collaboration level, interlaced knowledge included basic knowledge about the progress and status of

detector components. This knowledge gave individuals an appropriate representation of current issues,

the challenges facing related components, and pockets where controversy was likely to emerge. Such

discourse took place in plenary meetings, which played an important role as an attention-driving

mechanism. Regularly scheduled updates created an internal rhythm and an internal time pacing in the

development process (Brown & Eisenhardt, 1997). At this level, interlaced knowledge helped groups

of interdependent systems synchronize and schedule their work and keep track of changes in the

technological system.

The discourse that unfolded within the working groups and the review panels resulted in more

detailed knowledge. Differences usually involved justification across boundaries, for example, during

negotiations of interface specifications between two interdependent components. The interlaced

knowledge that emerged at this level offered the different groups a deeper understanding of each

other’s context and requirements. This understanding, in turn, enabled the developers of the

interdependent components to be mindful of each other’s positions and interrelate with one another

26

when unforeseen changes occurred. Differences over the materials budget could anticipate the

difficulties of the inner detector to build a TRT within the material constraints. The calorimeter

group’s decision to introduce an additional layer of detectors enabled the inner detector to develop a

working TRT, which exceeded the amount of material originally specified.

Interlaced knowledge that emerged from such controversies was articulated and shared in a

variety of documents, such as meeting minutes, technical reviews, and design reports. These

documents were presented on many occasions and circulated via electronic mailing lists. Overall

interaction was polymorphic, with every scientist or engineer enjoying free access to all the

information generated within ATLAS; thus, the knowledge percolated throughout the ATLAS

collaboration at large.

Coordination through interlaced knowledge

The modularity literature emphasizes the importance of black-boxing and information hiding

(Parnas, 1972). In contrast, the findings of this study suggest the importance of interlaced knowledge

that emerged from the resolution of controversies that helped the coordination of the development and

construction of ATLAS. Such interlaced knowledge provided redundancy through overlapping

knowledge, thus enabling the collaboration to employ a “shared division of labor” (Nonaka &

Takeuchi, 1995). This type of redundancy is not associated with inefficiency, which is the usual

meaning associated with this term. Rather, such redundancy allows an accommodation of irreducible

uncertainty associated with the emergence of complex technological systems and allows a system to

adapt to new contingencies.

Whereas the rigid information structure embedded in a modular architecture is likely to lock a

system into a pre-specified development path, interlaced knowledge enables collaborators to

renegotiate specifications and to shape the transformation of paths in real time. In this sense, interlaced

knowledge is generative in nature. That is, it enables developers to respond to emerging problems by

creating novel solutions rather than merely adhering to pre-determined specifications. We observed

that groups in ATLAS were able to coordinate their actions effectively even in the presence of

uncertainty when the architecture itself was not stable and when specifications were ambiguous. As

unforeseen events occurred, such as the delay of the ATLAS pixel detector, interlaced knowledge

27

enabled the participants to find a workaround in collaboration with interdependent groups. Some

solutions even involved the renegotiation of previously specified interfaces, as was the case with the

integration of the pixel detector into the end-cap, thus resulting in a change in the architecture.

Modular architecture as a boundary infrastructure

Similar to research on modularity (Baldwin & Clark, 2000; Sanchez & Mahoney, 1996; Ulrich

& Eppinger, 1995), we too found that the structures that facilitated coordination became embedded

into the ATLAS architecture. At the same time, the ATLAS case reveals a nuanced understanding of

the relationship between agency and structure. Specifically, we saw a process of adaptive structuration

(one where the rules built into structures and agency co-emerge) unfold (DeSanctis & Poole, 1994).

As we have explained earlier, it is important to preserve the rationale as to why particular

specifications were chosen in order for such interlaced knowledge to provide necessary coordination.

Agreed-upon specifications were merely the tip of the iceberg of interlaced knowledge that emerged

from ongoing processes of negotiation and ordering. While mere specifications – a reduced form of

knowledge – may be sufficient to coordinate a stable system, the origins of this knowledge as well as

important interdependencies are often forgotten or overlooked. Consequently, pre-specified interfaces

are not sufficient to coordinate the development of an emergent architecture; they may even result in

dysfunctional actions among the groups.

A preliminary specification of the architecture nevertheless plays an important role,

specifically that of a boundary infrastructure that connects heterogeneous components (Bowker &

Stars, 1999) rather than a role that predetermines the development of interrelated components. As a

boundary infrastructure, a modular architecture can be interpreted from different perspectives, yet

many points of connections enable the different groups to structure their conversations when

negotiating the design (Star & Griesemer, 1989). Different groups can use the architecture as a “means

of representing, learning about, and transforming knowledge to resolve the consequences that exist at

a given boundary” (Carlile, 2002: 442).

The predominant view on modularity underestimates the architecture’s role as serving as the

boundary infrastructures by assuming that specifications of components and interfaces are clearly

28

articulated and understood.10 Indeed, recent research suggests that justifications across boundaries

decrease as standards become taken for granted (Green, 2004). When this happens, a system may

begin to lose some of its emergent properties. In this regard, research on high reliability organizations

suggests that justification processes are required on an ongoing basis in order to create ongoing rich

representations of the overall system so that mindful interrelationships can unfold when organizations

confront novel situations (Weick & Roberts, 1993; Weick & Sutcliffe, 2001).

Similarly, studies on organizational learning have found that, under conditions of uncertainty,

organizations create more knowledge about the complex interdependencies when people from

different perspectives engage in sense-making processes to generate representations of ways that the

overall system works (March, Sproull, & Tamuz, 1991). An openness to a variety of possibly

interpretations is often more valuable than a predefined model, however clearly the model might be

defined ex ante (March, 1987).

Of course, interlaced knowledge may be a luxury if such flexibility is not required. From an

information processing perspective, coordination is more efficient if people can rely on taken-for-

granted specifications. This is the perspective that modularity research offers: a pre-specified

architecture that has evolved historically or by convention, allowing for distributed information

processing to occur (Baldwin & Clark, 2003). The architecture provides embedded coordination by

splitting up complex tasks and then integrating them into an overall solution. This perspective seems

appropriate if a mature architecture already exists. However, other mechanisms come into play when

uncertainty is involved during the emergence of architectures.

To the extent that specifications are perceived as boundary architectures that can be

interpreted from different perspectives, architectures do provide adaptive coordination in dynamic

contexts. The interlaced knowledge that is created through ongoing negotiations across boundaries

provides a deeper understanding of other perspectives including the requirements for other

components. Groups can anticipate latent interdependencies and mindfully interrelate with one another

as they encounter novel events; this enables dispersed groups to coordinate their actions, even as the

architectures and specifications are still emerging.

10 A notable exception is recent work by Brusoni, Prencipe and Pavitt (2001) that challenges mainstream scholars of modularity by suggesting that systems integrators may need to know more than they do.

29

CONCLUSION

We have explored the ATLAS case to sketch out the processes that are involved in the

emergence of technological architectures. Prior negotiations and the codification of specifications into

a modular architecture can result in a situation where complex interdependencies are hidden behind a

small number of pre-specified interfaces. Such a process supports a division of cognitive labor among

the different component developers. However, whereas designers in established technological systems

can take for granted a common definition of interfaces, actors in emergent systems such as ATLAS

cannot build on standards and conventions that have been institutionalized. For emergent

technological systems, an architecture needs to be created in the first place.

The ATLAS case illustrates that there is ambiguity around specifications surrounding the

emergence of architectures and that these specifications are interpreted differently by different groups

that are involved. Of particular interest is the way in which controversies that result from conflicting

interpretations are resolved as the different technological components and groups confront one

another. Our analysis suggests that justifications across boundaries create interlaced knowledge – a

structure of local knowledge bases that overlap. Such interlaced knowledge provides a collective yet

distributed understanding and appreciation of the requirements posed by advocates of other

components, thereby allowing multiple groups to anticipate latent interdependencies and to mindfully

interrelate with one another. As a consequence, distributed groups are able to coordinate their

activities effectively, even as the architecture and specifications emerge.

30

REFERENCES

Alexander, C. 1964. Notes on the Synthesis of Form. Cambridge, MA: Harvard University Press.

Baldwin, C. Y., & Clark, K. B. 2000. Design rules, Vol. 1: The power of modularity. Cambridge:

MIT Press.

Baldwin, C. Y., & Clark, K. B. 2003. Where Do Transactions Come From? A Perspective from

Engineering Design. Cambridge, MA: Harvard Business School.

Baldwin, C. Y., & Clark, K. B. 2005. Between "Knowledge" and "the Economy": Notes on the

Scientific Study of Designs. Cambridge, MA: Harvard Business School.

Barry, D., & Rerup, C. 2006. Going Mobile: Aesthetic Design Considerations from Calder and the

Constructivists. Organization Science, 17(2): 262-276.

Bijker, W. E., Hughes, T. P., & Pinch, T. J. (Eds.). 1987. The Social Construction of Technological

Systems. Cambridge, MA: MIT Press.

Boltanski, L., & Thévenot, C. P. 2006. On Justification: Economies of Worth. Princeton, NJ:

Princeton University Press.

Bowker, G. C., & Stars, S. L. 1999. Sorting things out: classification and its consequences.

Cambridge: MIT Press.

Brown, S. L., & Eisenhardt, K. M. 1997. The Art of Continuous Change: Linking Complexity Theory

and Time-paced Evolution in Relentlessly Shifting Organizations. Administrative Science

Quarterly, 42: 1-34.

Brusoni, S., Prencipe, A., & Pavitt, K. 2001. Knowledge Specialization, Organizational Coupling, and

the Boundaries of the Firm: Why Do Firms Know More Than They Make? . Administrative

Science Quarterly, 46(4): 597-621.

Carlile, P. R. 2002. A Pragmatic View of Knowledge and Boundaries: Boundary Objects in New

Product Development. Organization Science, 13: 442-455.

Clark, K. B. 1985. The interaction of design hierarchies and market concepts in technological

evolution. Research Policy, 14(5): 235-251.

Clark, K. B., & Fujimoto, T. 1990. The power of product integrity. Harvard Business Review, 68(6):

107-118.

David, P. A. 1985. Clio and the Economics of QWERTY. The American Economic Review, 75(2):

332-337.

Deerwester, S., Dumais, S., Furnas, G. W., Landauer, T. K., & Harshman, R. 1990. Indexing by Latent

Semantic Analysis. Journal of the American Society for Information Science, 41(6): 391-

407.

DeSanctis, G., & Poole, M. S. 1994. Capturing the Complexity in Advanced Technology Use:

Adaptive Structuration Theory. organization Science, 5(2): 121-147.

Eppinger, S. D., Whitney, D. E., Smith, R. P., & Gebala, D. A. 1994. A model-based method for

organizing tasks in product development. Research in Engineering Design, 6: 1-13.

31

Ethiraj, S. K., & Levinthal, D. 2004. Modularity and innovation in complex systems. Management

Science, 50(2): 159-173.

Galison, P. L. 1987. How Experiments end. Chicago, IL: University of Chicago Press.

Galison, P. L. 1997. Image and Logic: A material Culture of Microphysics. Chicago, IL: University

of Chicago Press.

Galison, P. L., & Stump, D. J. (Eds.). 1996. The Disunity of Science: Boundaries, Contexts, and

Power. Stanford, CA: Stanford University Press.

Garud, R. 1997. On the distinction between know-how, know-why and know-what in technological

systems. In J. P. Walsh, & A. S. Huff (Eds.), Advances in Strategic Management, Vol. 14:

81-101. Greenwich, CT: JAI Press.

Garud, R., & Jain, S. 1996. Technology Embeddedness. In J. Baum, & J. E. Dutton (Eds.), Advances

in Strategic Management, Vol. 13: 389-408. Greenwich, CT: JAI Press.

Garud, R., Jain, S., & Tuertscher, P. 2008. Designing for incompleteness and incomplete by design.

Organization Studies, 29(3): 351-371.

Garud, R., & Kotha, S. 1994. Using the brain as metaphor to model flexible production systems.

Academy of Management Review, 19: 671-698.

Garud, R., & Kumaraswamy, A. 1995. Technological and organizational designs to achieve

economies of substitution. Strategic Management Journal, 16: 93-110.

Garud, R., & Munir, K. 2008. From transactions to transformation costs: The case of Polaroid's SX-70

Camera. Research Policy, 37(4): 690-705.

Grant, R. M. 1996. Toward a knowledge-based theory of the firm. Strategic Management Journal,

17: 109-122.

Green, S. E., Jr. 2004. A rhetorical theory of diffusion. Academy of Management Review, 29: 653-

669.

Henderson, R. M., & Clark, K. B. 1990. Architectural innovation: the reconfiguration of existing

product technologies and the failure of established firms. Administrative Science Quarterly,

35(9-30).

Knorr-Cetina, K. 1995. How Superorganisms Change: Consensus Formation and the Social Ontology

of High-Energy Physics Experiments. Social Studies of Science, 25: 119-147.

Knorr-Cetina, K. 1999. Epistemic Cultures: How the Sciences Make Knowledge. Cambridge:

Harvard University Press.

Langlois, R. N. 2006. The Secret Life of Mundane Transaction Costs. Organization Studies, 27(9):

1389-1410.

March, J. G. 1987. Ambiguity and Accounting: The Elusive Link between Information and Decision

Making. AOS, 12: 153-168.

March, J. G., Sproull, L. S., & Tamuz, M. 1991. Learning from Samples of One or Fewer.

Organization Science, 2: 1-13.

32

Mihm, J., Loch, C., & Huchzermeier, A. 2003. Problem-solving oscillations in complex engineering

projects. Management Science, 49(6): 733-750.

Miles, M. B., & Huberman, A. M. 1984. Analyzing qualitative data: a source book for new methods.

Beverly Hills: Sage.

Mohr, L. B. 1982. Explaining organizational behavior: the limits and possibilities of theory and

research. San Francisco: Jossey-Bass.

Murmann, J. P., & Frenken, K. 2006. Toward a systematic framework for research on dominant

designs, technological innovation, and industrial change. Research Policy, 35(7): 925-952.

Nonaka, I. 1994. A dynamic theory of organizational knowledge creation. Organization Science, 5:

14-37.

Nonaka, I., & Takeuchi, H. 1995. The knowledge-creating company. Oxford: Oxford University

Press.

Parnas, D. L. 1972. On the criteria to be used in decomposing systems into modules. Communications

of the ACM, 15: 1053-1058.

Sanchez, R. 1995. Strategic Flexibility in Product Competition. Strategic Management Journal, 16:

135-159.

Sanchez, R., & Mahoney, J. T. 1994. The modularity principle in product and organization design:

achieving flexibility in the fusion of intended and emergent strategies in hypercompetitive

product markets, Office of Research working paper. University of Illinois at Urbana-

Champaign.

Sanchez, R., & Mahoney, J. T. 1996. Modularity, flexibility, and knowledge management in product

and organization design. Strategic Management Journal, 17: 63-76.

Simon, H. A. 1962. The architecture of complexity. Proceedings of the American Philosophical

Society, 106: 467-482.

Simon, H. A. 1996. The Sciences of the Artificial. Cambridge, MA: MIT Press.

Star, S. L., & Griesemer, J. R. 1989. Institutional Ecology, `Translations' and Boundary Objects:

Amateurs and Professionals in Berkeley's Museum of Vertebrate Zoology, 1907-39. Social

Studies of Science, 19(3): 387-420.

Tushman, M. L., & Murmann, J. P. 1998. Dominant designs, technology cycles, and organizational

outcomes Research in Organizational Behavior, 20: 231-266.

Tushman, M. L., & Murmann, J. P. 2003. Dominant designs, technology cycles, and organizational

outcomes. In R. Garud, A. Kumaraswamy, & R. N. Langlois (Eds.), Managing in the

modular age: architectures, networks, and organizations. Oxford: Blackwell Publishers.

Tushman, M. L., & Rosenkopf, L. 1992. Organizational Determinants of Technological Change:

Toward a Sociology of Technological Evolution. In B. M. Staw, & L. L. Cummings (Eds.),

Research in Organizational Behavior, Vol. 14: 311-347. Greenwich, CT: JAI Press.

33

Ulrich, K. 2003. The role of product architecture in the manufacturing firm. In R. Garud, A.

Kumaraswamy, & R. N. Langlois (Eds.), Managing in the modular age: architectures,

networks, and organizations: 117-145. Oxford: Blackwell Publishers.

Ulrich, K., & Eppinger, S. D. 1995. Product design and development. New York: McGraw-Hill.

Ulrich, K., & Eppinger, S. D. 2000. Product design and development. New York: Irwin McGraw-

Hill.

Van de Ven, A. H. 1992. Suggestions for Studying Strategy Process: A Research Note. Strategic

Management Journal, 13: 169-188.

von Krogh, G., & Grand, S. 2000. Justification in knowledge creation: open issues and research

questions. In G. von Krogh, I. Nonaka, & T. Nishiguchi (Eds.), Knowledge creation: a source

of value. Houndmills: Macmillan Press.

Weick, K. E., & Roberts, K. H. 1993. Collective mind in organizations: heedful interrelating on flight

decks. Administrative Science Quarterly, 38: 357-381.

Weick, K. E., & Sutcliffe, K. M. 2001. Managing the Unexpected: Assuring High Performance in

an Age of Complexity. San Francisco: Jossey-Bass.

Wood, S. C., & Brown, G. S. 1998. Commercializing nascent technology: The case of laser diodes at

Sony. Journal of Product Innovation Management, 15(2): 167-183.

34

Table 1: R&D projects with ATLAS involvement

ATLAS Subsystem R&D Project Comments on Progress of R&D

Inner detector • Vertexing and

innermost tracking • Outer tracking and

electron identif.

RD19 Si pixel detectors RD20 Si micro strip detectors RD8 GaAs detectors RD2 Si strip and pad detectors RD6 TRD straw detectors RD7 scintillating fibers

All part of baseline design; R&D was needed to integrate concepts All part of baseline design; R&D was needed to integrate concepts Alternative option; R&D was needed to confirm feasibility of concept

EM calorimeter and preshower detector

RD3 liquid argon accordion P44 liquid argon TGT RD1 scintillating fibers

Baseline for barrel, baseline end-cap; R&D was needed to optimize design Alternative for barrel, baseline end-cap; R&D needed to demonstrate feasibility Alternative; only limited R&D was required

Hadronic calorimeter RD1 scintillating fibers RD3 liquid argon accordion P44 liqud argon TGT Scintillating tiles pre-prototype

All considered baseline options; R&D required for decision before Technical Proposal

Forward calorimeter Liquid scintillator and high pressure gas pre-prototypes

Both baseline options, R&D required for decision before Technical Proposal

Muon system • Tracking detectors

• Trigger detectors • General aspects

RD5 honeycomb strip chambers High pressure drift tubes Jet cell drift chambers RD5 resistive plate chambers RD5 punch through, em showers, etc.

All considered baseline options; R&D required for decision before Technical Proposal

Trigger • Level 1

• Level 2 • Level 3

RD5 muon triggers RD27 calorimeter triggers, system aspects RD2 and RD6 electron track triggers RD11 EAST general architectures RD13 general architectures

Front end electronics RD12 general read-out systems RD16 FERMI digital calorimeter FE RD29 DMILL radiation hard electronics

Detector-specific FE electronics; R&D was included in the corresponding R&D projects

DAQ system RD13 general DAQ and readout RD23 optoelectronic signal transfer

35

Table 2: Description of Major ATLAS subsystems

Subsystems Description

Inner detector • Pixel detector • Semi-conductor

tracker (SCT) • Transition radiation

tracker (TRT)

• The ATLAS inner detector combines high-resolution detectors at the inner radii with continuous tracking elements at the outer radii, all contained in the central solenoid magnet. The outer radius of the inner detector is 1.15 m, and the total length is 7 m.

• The pixel detector provides a high precision set of three measurements as close to the interaction point as possible, and mostly determines the impact parameter resolution and the ability of the inner detector to find short-lived particles such as B-Hadrons.

• The SCT system is designed to provide a set of eight precision measurements per track in the intermediate radial range, contributing to the measurement of momentum, impact parameter and vertex position

• At larger radii, typically 36 tracking points are provided by the TRT. The Transition radiation Tracker (TRT) is based on the use of straw detectors, which can operate at the expected high rates due to their small diameter and the isolation of the sense wires within individual gas volumes. Electron identification capability is added by employing Xenon gas to detect transition radiation photons created in a radiator between the straws.

Calorimeter system • Liquid argon

calorimeter • Tile calorimeter

• The calorimeter measures the energy of charged and neutral particles. It consists of metal plates (absorbers) and sensing elements. Interactions in the absorbers transform the incident energy into a "shower" of particles that are detected by the sensing elements.

• In the inner sections of the calorimeter, the liquid argon calorimeter, the sensing element is liquid argon. The showers in the argon liberate electrons that are collected and recorded.

• In the outer sections, the sensors are tiles of scintillating plastic, i.e., the tile calorimeter. The showers cause the plastic to emit light that is detected and recorded.

Muon spectrometer • Precision chambers • Trigger chambers

• Muons are particles, like electrons, but 200 times heavier. They are the only detectable particles that can traverse all the calorimeter absorbers without being stopped. The muon spectrometer surrounds the calorimeter and measures muon paths to determine their momentum with high precision.

• In precision chambers, gas-filled metal tubes with wires running down their axes are used as sensors. High voltage between the wire and the tube wall allows detection of the traversing muons by the electrical pulses they produce. With careful timing of the pulses, muon positions can be measured to an accuracy of 0.1 mm. The reconstructed muon path determines its momentum and sign of charge.

• Trigger chambers are based on a similar principle; however, high time resolution rather than precision is the key feature of trigger chambers. Using thin plates or multiple wires as sensors, trigger chambers have a time resolution better than 25 ns.

Magnet system • Toroid magnet • Solenoid magnet

• The ATLAS toroid magnet system consists of eight barrel coils housed in separate cryostats and eight coils housed in each of two end-cap cryostats. The end-cap coil systems are rotated by 22.5° with respect to the barrel toroids in order to provide radial overlap and to optimize the bending power in the interface regions of both coil systems.

• The central ATLAS solenoid has a length of 5.3 m with a bore of 2.4 m. The conductor is a composite that consists of a flat superconducting cable located in the center of an aluminum stabilizer with a rectangular cross-section. The solenoid is designed to provide a field of 2 T with a peak magnetic field of 2.6 T. The total assembly weight is 5.7 tons.

36

Table 3: Institutions Participating in Subsystems, by Country of origin

Inner detector Liquid argon calorimeter

Tile calorimeter Muon spectrometer

Armenia 1

Australia 2

Belarus 2

Brazil 1

Canada 9

China 1 1

Czech Republic 3 2

Denmark 1

France 1 6 1 1

Georgia 1

Germany 6 4 2

Greece 1

Italy 3 1 1

Japan 5 7

Morocco 1

Netherlands 1 2

Norway 2

Poland 2

Portugal 1

Romania 1

Russia 6 3

Slovenia 1

Spain 1

Sweden 2 1 1

Switzerland 1 1

Taipei 1 1

Turkey 1

United Kingdom 11

USA 17 6 5 12

CERN 1 1 1 1

68 32 19 29

37

Figure 1: Overview of the Atlas Detector

38

Figure 2: ATLAS Co-Authorship Network over Time

1990 2006

This figure indicates when and where groups formed and how they interacted. The locus of activity is seen at the periphery in the early design phase (blue lines) and gradually shifted towards the center as the design of various components crystallized (green lines). Most of the activity in the center, however, is represented by yellow and red lines, indicating that activities regarding integration occurred mostly in the later phases of ATLAS development.

39

Figure 3: Justification in Three Subsystems

This figure plots the justification factor for the 3 major subsystem groups of the ATLAS detector over time. The justification factor was obtained as a result from the latent semantic analysis and indicates to what extent the subsystem groups engaged in justification. Figure 3 clearly indicates that the level of justification increased during the early design phase and reached the highest level during the detailed development phase between 1996 and 1998. The level of justification decreased as the designs of the various components materialized the subsystem groups started to construct components in the late 1990s.

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1990 1993 1996 1999 2002 2005 2008

Ju

sti

ficati

on

Facto

r

Justification in ATLAS Notes

Calorimeter

Muon

Inner Detector

40

Figu

re 4

: Evo

lutio

n of

Inte

rlac

ed K

now

ledg

e St

ruct

ure

of A

TL

AS

Muo

n Sp

ectr

omet

er C

omm

unity

a) 1

994

b) 1

998

c) 2

006

The

node

s in

the

net

wor

k gr

aph

repr

esen

t di

stin

ct e

xper

tise

area

s th

at w

ere

extra

cted

fro

m m

ore

than

250

0 te

chni

cal

docu

men

ts u

sing

lat

ent

sem

antic

an

alys

is. T

he c

olor

cod

e in

dica

tes

how

cen

tral a

par

ticul

ar e

xper

tise

area

is in

the

com

mun

ity’s

kno

wle

dge

stru

ctur

e (r

ed in

dica

ting

high

est c

entra

lity)

. The

ed

ges

betw

een

the

node

s in

dica

te c

onne

ctio

ns b

etw

een

expe

rtise

are

as,

i.e.

peop

le h

avin

g ex

perti

se i

n tw

o or

mor

e in

terd

epen

dent

are

as o

f kn

owle

dge

(yel

low

edg

es in

dica

ting

wea

k an

d gr

een

edge

s in

dica

ting

stro

ng c

onne

ctio

ns).

Figu

re 4

sug

gest

s th

at th

e kn

owle

dge

stru

ctur

e ha

s be

com

e m

ore

inte

rlace

d as

th

e de

velo

pmen

t of

the

dete

ctor

pro

gres

sed.

The

inte

rlace

d kn

owle

dge

stru

ctur

e, h

owev

er, s

eem

s to

hav

e de

caye

d du

ring

the

cons

truct

ion

and

inst

alla

tion

phas

e. S

imila

r res

ults

wer

e ob

tain

ed fo

r the

two

othe

r mai

n su

bsys

tem

s.

!"#$%&#'()*+,*)($-$.'/,0

!"#$%&#'()*+,*)($-$.'/,0

%'/

%'/

%&1

%&1

22

33

3232

322

322

323

323

324

324

325

325

326

326

327

327

328

328

329

329

32:

32:

32;

32;

3333

332

332

333

333

334

334

335

335

336

336

337

337

338

338

339

339

33:

33:

33;

33;

3434

342

342

343

343

344

344

345

345

346

346

347

347

348

348

349

349

34:

34:

34;

34;

3535

352

352

353

353

354

354

355

355

356

356

357

357

358

358

359

359

35:

35:

35;

35;

3636

362

362

363

363

364

364

365

365

366

366

367

367

368

368 369

369

36:

36:

36;

36;

3737

372

372

373

373

374

374

375

375

376

376

377

377

378

378

379

379

37:

37:

37;

37;

3838

382

382

383

383

384

384

385

385

386

386

387

387

388

388

389

389

38:

38:

38;

38;

3939

392

392

393

393

394

394

395

395

396

396

397

397

398

398

399

399

39:

39:

39;

39;

3:3:

3:2

3:2

3:3

3:3

3:4

3:4

3:5

3:5

3:6

3:6

3:7

3:7

3:8

3:8

3:9

3:9

3::

3::

3:;

3:;

3;3;

3;2

3;2

3;3

3;3

3;4

3;4

3;5

3;5

3;6

3;6

3;7

3;7

3;8

3;8

3;9

3;9

3;:

3;:

3;;

3;;

44

4242

422

422

423

423

424

424

425

425

426

426

427

427

428

428

429

429

42:

42:

42;

42;

4343

432

432

433

433

434

434

435

435

436

436

437

437

438

438

439

439

43:

43:

43;

43;

4444

442

442

443

443

444

444

445

445

446

446

447

447

448

448

449

449

44:

44:

44;

44;

4545

452

452

453

453

454

454

455

455

456

456

457

457

458

458

459

459

45:

45:

45;

45;

4646

462

462

463

463

464

464

465

465

466

466

467

467

468

468

469

469

46:

46:

46;

46;

4747

472

472

473

473

474

474

475

475

476

476

477

477

478

478

479

479

47:

47:

47;

47;

4848

482

482483

483

484

484

485

485

486

486

487

487

488

488

489

489

48:

48:

48;

48;

4949

492

492

493

493

494

494

495

495

496

496

497

497

498

498

499

499

49:

49:

49;

49;

4:4:

4:2

4:2

4:3

4:3

4:4

4:4

4:5

4:5

4:6

4:6

4:7

4:7

4:8

4:8

4:9

4:9

4::

4::

4:;

4:;

4;4;

4;2

4;2

4;3

4;3

4;4

4;4

4;5

4;5

4;6

4;6

4;7

4;7

4;8

4;8

4;9

4;9

4;:

4;:

4;;

4;;

55

5252

522

522

523

523

524

524

525

525

526

526

527

527

528

528

529

529

52:

52:

52;

52;

5353

532

532

533

533

534

534

535

535

536

536

537

537

538

538

539

539

53:

53:

53;

53;

5454

542

542

543

543

544

544

545

545

546

546

547

547

548

548

549

549

54:

54:

54;

54;

5555

552

552

553

553

554

554

555

555

556

556

557

557

558

558

559

559

55:

55:

55;

55;

5656

562

562

563

563

564

564

565

565

566

566

567

567

568

568

569

569

56:

56:

56;

56;

5757

572

572

573

573

574

574

575

575

576

576

577

577

578

578

579

579

57:

57:

57;

57;

5858

582

582

583

583

584

584

585

585

586

586

587

587

588

588

589

589

58:

58:

5959

592

592

593

593

594

594

595

595

596

596

597

597

598

598

599

599

59:

59:

59;

59;

5:5:

5:2

5:2

5:3

5:3

5:4

5:4

5:5

5:5

5:6

5:6

5:7

5:7

5:8

5:8

5:9

5:9

5::

5::

5:;

5:;

5;5;

5;2

5;2

5;3

5;3

5;4

5;4

5;6

5;6

5;7

5;7

5;9

5;9

5;:

5;:

5;;

5;;

6262

623

623

624

624

625

625

626

626

627

627

628

628

629

629

62:

62:

62;

62;

6363

634

634

635

635

636

636

637

637

638

638

63:

63:

6464

642

642

643

643

644

644

645

645

646

646

647

647

648

648

649

649

64:

64:

64;

64;

6565

652

652

653

653

654

654

655

655

656

656

657

657

658

658

659

659

65:

65:

65;

65;

6666

662

662

664

664

665

665

666

666

667

667

668

668

669

669

66:

66:

66;

66;

6767

672

672

673

673

674

674

675

675

676

676

677

677

6868 6969

6:6:

6;6;

77

7272 7373

7474

7575

7676

7777

7878

7979

7:7:

7;7;

88

8282

8383

8484

8585

8686

8787

8888

8989

8:8:

8;8;

99

9292

9393

9494

9595

9696

9797

9898

9999

9:9:

9;9;

::

:2:2

:3:3

:4:4

:5:5

:6:6

:7:7

:8:8

:9:9

::::

:;:;

;;

;2;2

;3;3

;4;4

;5;5

;6;6

;7;7

;8;8

;9;9

;:;:

;;;;

!"#$%&#'()*+,*)($-$.'/,0

!"#$%&#'()*+,*)($-$.'/,0

%'/

%'/

%&1

%&1

2233

3232

322

322

323

323

324

324

325

325

326

326

327

327

328

328

329

329

32:

32:

32;

32;

3333

332

332

333

333

334

334

335

335

336

336337

337

338

338

339

339

33:

33:

33;

33;

3434

342

342

343

343

344

344

345

345

346

346

347

347

348

348

349

349

34:

34:

34;

34;

3535

352

352

353

353

354

354

355

355

356

356

357

357

358

358

359

359

35:

35:

35;

35;

3636

362

362

363

363

364

364

365

365

366

366

367

367

368

368

369

369

36:

36:

36;

36;

3737

372

372

373

373

374

374

375

375376

376

377

377

378

378

379

379

37:

37:

37;

37;

3838

382

382

383

383

384

384

385

385

386

386

387

387

388

388

389

389

38:

38:

38;

38;

3939

392

392

393

393

394

394

395

395

396

396

397

397

398

398

399

399

39:

39:

39;

39;

3:3:

3:2

3:2

3:3

3:3

3:4

3:4

3:5

3:5

3:6

3:6

3:7

3:7

3:8

3:8

3:9

3:9

3::

3::

3:;

3:;

3;3;

3;2

3;2

3;3

3;3

3;4

3;4

3;5

3;5

3;6

3;6

3;7

3;7

3;8

3;8

3;9

3;9

3;:

3;:

3;;

3;;

44

4242

422

422423

423

424

424

425

425

426

426

427

427

428

428

429

429

42:

42:

42;

42;

4343

432

432

433

433

434

434

435

435

436

436

437

437

438

438

439

439

43:

43:

43;

43;

4444

442

442

443

443

444

444

445

445

446

446

447

447

448

448

449

449

44:

44:

44;

44;

4545

452

452

453

453

454

454

455

455

456

456

457

457

458

458

459

459

45:

45:

45;

45;

4646

462

462

463

463

464

464

465

465

466

466

467

467

468

468

469

469

46:

46:

46;

46;

4747

472

472

473

473

474

474

475

475

476

476

477

477

478

478

479

479

47:

47:

47;

47;

4848

482

482

483

483

484

484

485

485

486

486

487

487

488

488

489

489

48:

48:

48;

48;

4949

492

492

493

493

494

494

495

495

496

496

497

497

498

498

499

499

49:

49:

49;

49;

4:4:

4:2

4:2

4:3

4:3

4:4

4:4

4:5

4:5

4:6

4:6

4:7

4:7

4:8

4:8

4:9

4:9

4::

4::

4:;

4:;

4;4;

4;2

4;2

4;3

4;3

4;4

4;4

4;5

4;5

4;6

4;6

4;7

4;7

4;8

4;8

4;9

4;9

4;:

4;:

4;;

4;;

55

5252

522

522

523

523

524

524

525

525

526

526

527

527

528

528

529

529

52:

52:

52;

52;

5353

532

532

533

533

534

534

535

535

536

536

537

537

538

538

539

539

53:

53:

53;

53;

5454

542

542

543

543

544

544

545

545

546

546

547

547

548

548

549

549

54:

54:

54;

54;

5555

552

552

553

553

554

554

555

555

556

556

557

557

558

558

559

559

55:

55:

55;

55;

5656

562

562

563

563

564

564

565

565

566

566

567

567

568

568

569

569

56:

56:

56;

56;

5757

572

572

573

573

574

574

575

575

576

576

577

577

578

578

579

579

57:

57:

57;

57;

5858

582

582

583

583

584

584

585

585

586

586

587

587

588

588

589

589

58:

58:

58;

58;

5959

592

592

593

593

594

594

595

595

596

596

597

597

598

598

599

599

59:

59:

59;

59;

5:5:

5:2

5:2

5:3

5:3

5:4

5:4

5:5

5:5

5:6

5:6

5:7

5:7

5:8

5:8

5:9

5:9

5::

5::

5:;

5:;

5;5;

5;2

5;2

5;3

5;3

5;4

5;4

5;5

5;5

5;6

5;6

5;7

5;7

5;8

5;8

5;9

5;9

5;:

5;:

5;;

5;;

6262

622

622

623

623

624

624

625

625

626

626

627

627

628

628

629

629

62:

62:

62;

62;

6363

632

632

633

633

634

634

635

635

636

636

637

637

638

638

639

639

63:

63:

63;

63;

6464

642

642

643

643

644

644

645

645

646

646

647

647

648

648

649

649

64:

64:

64;

64;

6565

652

652

653

653

654

654

655

655

656

656

657

657

658

658

659

659

65:

65:

65;

65;

6666

662

662

663

663

664

664

665

665

666

666

667

667 668

668

669

669

66:

66:

66;

66;

6767

672

672

673

673

674

674

675

675

676

676

677

677

6868

6969

6:6:

6;6;

77

7272

7373

7474

7575

7676

7777

7878

7979

7:7:

7;7;

88

8282

8383

8484

8585

8686

8787 8888

8989

8:8:

8;8;

99

9292

9393

9494

9595

9696

9797

9898

9999

9:9:

9;9;

::

:2:2

:3:3

:4:4

:5:5

:6:6

:7:7

:8:8

:9:9

::::

:;:;

;;

;2;2

;3;3;4;4

;5;5

;6;6

;7;7

;8;8

;9;9

;:;:

;;;;

!"#$%&#'()*+,*)($-$.'/,0

!"#$%&#'()*+,*)($-$.'/,0

%'/

%'/

%&1

%&1

22

33

3232

322

322

323

323

324

324

325

325

326

326

327

327

328

328

329

329

32:

32:

32;

32;

3333

332

332

333

333

334

334

335

335

336

336

337

337

338

338

339

339

33:

33:

33;

33;

3434

342

342

343

343

344

344

345

345

346

346

347

347

348

348

349

349

34:

34:

34;

34;

3535

352

352

353

353

354

354

355

355

356

356

357

357

358

358

359

359

35:

35:

35;

35;

3636

362

362

363

363

364

364

365

365

366

366

367

367

368

368

369

369

36:

36:

36;

36;

3737

372

372

373

373

374

374

375

375

376

376

377

377

378

378

379

379

37:

37:

37;

37;

3838

382

382

383

383

384

384

385

385

386

386

387

387

388

388

389

389

38:

38:

38;

38;

3939

392

392

393

393

394

394

395

395

396

396

397

397

398

398

399

399

39:

39:

39;

39;

3:3:

3:2

3:2

3:3

3:3

3:4

3:4

3:5

3:5

3:6

3:6

3:7

3:7

3:8

3:8

3:9

3:9

3::

3::

3:;

3:;

3;3;

3;2

3;2

3;3

3;3

3;4

3;4

3;5

3;5

3;6

3;6

3;7

3;7

3;8

3;8

3;9

3;9

3;:

3;:

3;;

3;;

44

4242

422

422

423

423

424

424

425

425

426

426

427

427

428

428

429

429

42:

42:

42;

42;

4343

432

432

433

433

434

434

435

435

436

436

437

437

438

438

439

439

43:

43:

43;

43;

4444

442

442

443

443

444

444

445

445

446

446447

447

448

448

449

449

44:

44:

44;

44;

4545

452

452

453

453

454

454

455

455

456

456

457

457

458

458

459

459

45:

45:

45;

45;

4646

462

462

463

463 464

464

465

465

466

466

467

467

468

468

469

469

46:

46:

46;

46;

4747

472

472

473

473

474

474

475

475

476

476

477

477

478

478

479

479

47:

47:

47;

47;

4848

482

482

483

483

484

484

485

485

486

486

487

487

488

488

489

489

48:

48:

48;

48;

4949

492

492

493

493

494

494

495

495

496

496

497

497

498

498

499

499

49:

49:

49;

49;

4:4:

4:2

4:2

4:3

4:3

4:4

4:4

4:5

4:5

4:6

4:6

4:7

4:7

4:8

4:8

4:9

4:9

4::

4::

4:;

4:;

4;4;

4;2

4;2

4;3

4;3

4;4

4;4

4;5

4;5

4;6

4;6

4;7

4;7

4;8

4;8

4;9

4;9

4;:

4;:4;;

4;;

55

5252

522

522

523

523

524

524

525

525

526

526

527

527

528

528

529

529

52:

52:

52;

52;

5353

532

532

533

533

534

534

535

535

536

536

537

537

538

538

539

539

53:

53:

53;

53;

5454

542

542

543

543

544

544

545

545

546

546

547

547

548

548

549

549

54:

54:

54;

54;

5555

552

552

553

553

554

554

555

555

556

556

557

557

558

558

559

559

55:

55:

55;

55;

5656

562

562

563

563

564

564

565

565

566

566

567

567

568

568

569

569

56:

56:

56;

56;

5757

572

572

573

573

574

574

575

575

576

576

577

577

578

578

579

579

57:

57:

57;

57;

5858

582

582

583

583

584

584

585

585

586

586

587

587

588

588

589

589

58:

58:

58;

58;

5959

592

592

593

593

594

594

595

595

596

596

597

597

598

598

599

599

59:

59:

59;

59;

5:5:

5:2

5:2

5:3

5:3

5:4

5:4

5:5

5:5

5:6

5:6

5:7

5:7

5:8

5:8

5:9

5:9

5::

5::

5:;

5:;

5;5;

5;2

5;2

5;3

5;3

5;4

5;4

5;5

5;5

5;6

5;6

5;7

5;7

5;8

5;8

5;9

5;9

5;:

5;:

5;;

5;;

6262

622

622

623

623

624

624

625

625

626

626

627

627

628

628

629

629

62:

62:

62;

62;

6363

632

632

633

633

634

634

635

635

636

636

637

637

638

638

639

639

63:

63:

63;

63;

6464

642

642

643

643

644

644

645

645

646

646

647

647

648

648

649

649

64:

64:

64;

64;

6565

652

652

653

653

654

654

655

655

656

656

657

657

658

658

659

65965:

65:

65;

65;

6666

662

662

663

663

664

664

665

665

666

666

667

667

668

668

669

669

66:

66:

66;

66;

6767

672

672

673

673

674

674

675

675

676

676

677

677

6868

6969

6:6:

6;6;

77

7272

7373

7474

7575

7676

7777

7878

7979

7:7:

7;7;

88

8282

8383

8484

8585

8686

8787

8888

8989

8:8:

8;8;

99

9292

9393

9494

9595

9696

9797

9898

9999

9:9:

9;9;

::

:2:2

:3:3

:4:4

:5:5

:6:6

:7:7

:8:8

:9:9

::::

:;:;

;;

;2;2

;3;3

;4;4

;5;5

;6;6

;7;7

;8;8

;9;9

;:;:

;;;;

41

Figure 5: Closeness Centrality of Interlaced Knowledge in Three Sub-systems

This figure plots the closeness centrality of the three subsystem communities’ knowledge structures over time. Closeness centrality is a measure of how densely interlaced a community’s knowledge structure is. The green line for the muon spectrometer, for example, indicates how the knowledge structures shown in Figure 4.a-c became more interlaced first and decayed during the construction and installation phase.

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

1990 1993 1996 1999 2002 2005 2008

Closeness of Interlaced Knowledge

Calorimeter

Muon

Inner Detector