Overcoming knowledge network failures: evidence from a French interregional scientific alliance in green chemistry

ABSTRACT Competition in research has led to the emergence of new regional spaces and interorganizational arrangements, designed to improve their participants’ visibility, at the crossroads of regional systems and global exchanges. This article investigates the determinants of the formation of interregional scientific alliances through the in-depth case study of a federation of research laboratories in green chemistry in western France. Working from bibliometric and qualitative data and using a network analysis approach to evidence interurban network failures in this scientific field, we examine the federation’s geographical scope. We highlight the decisive role of (1) the spatial organization of the field (structure effect), (2) interpersonal ties between consortium members (network effect) and (3) national scientific guidelines and incentives (policy effect). We propose a novel mixed-method approach to understand interregional alliances and defend its potential to uncover the importance of interregional institutional arrangements in prompting interurban knowledge exchanges.


INTRODUCTION
These last years a growing number of institutional arrangements (as ventures, alliances, etc., according to the level of integration and to national contexts) have been developed between higher education, research and innovation (HERI) entities (Youtie et al., 2017;Yudkevich et al., 2016). Whether the purpose is to create an academic community, a thematic network or a local ecosystem, HERI stakeholders are encouraged to join forces, without necessarily having to comply with the administrative boundaries of their respective territories. Focusing on university alliances in the UK, Harrison et al. (2016) observed 'a move from promoting collaborative activity within a predetermined regional templatedefined by the state and where participation is predetermined by an institution's geographic locationtowards allowing universities to actively (re)invent regions based on their own collaborative practices' (p. 9). These opportunities lead to the emergence of new regional spaces and imaginaries that merit further analysis. This is the intention of this article, which proposes an in-depth analysis of the genesis and geography of an interregional scientific alliance in green chemistry in western France. By focusing on a specific public policy instrument, it contributes to overcoming one of the limits pointed by Kogler (2015) in a review of theoretical and empirical progress in Evolutionary Economic Geography, which is the lack of studies taking regional innovation policy as their focal point of interest. Our contribution addresses this point by offering to examine empirically the implementation of a coordination tool between laboratories located in neighbouring but administratively distinct regions and underpins the role that research policies can play for the development of such transregional knowledge networks.
By adopting a spatial approach, we highlight the potential of interregional scientific alliances to fix 'network failures' (Lucena Piquero & Vicente, 2019). Such failures can be the product of long-term spatial processes, or path dependencies, that shape and restrain the organization and evolution of scientific networks. Here, the challenge is to go beyond the regional level, at which it is traditionally easier to establish relationships, and to bring together thematically complementary laboratories.
The culture of networks as a public policy tool is traditional in the field of innovation and has been the subject of many developments, both theoretical and empirical (Gordon & McCann, 2005;Yu & Jackson, 2011). Cluster policies design research and development (R&D) collaborative incentives to strengthen knowledge network and better benefit from complementarities between actors: while innovation policy has long been centred on individual incentives to address traditional market failures, cluster policies have emerged with the growing awareness from academics and policymakers that network failures exist (Vicente, 2017;Woolthuis et al., 2005). In the context of HERI deconcentration and territorial planning (Huggins & Kitagawa, 2012;Kitagawa, 2010), coordination mechanisms have become necessary to overcome failures resulting from local strategies and cluster policies developed by competing regions (Charles et al., 2014).
In this article we tackle the importance of supporting interregional networks between research units from peripheral cities as an alternative to 'monopolistic concentrations' which nowadays constitute a standard of research policies in their structural incentive schemes .
Focusing on the case of a French scientific alliance, we investigate the roles of the following effects on the alliance's genesis and geographical scope: . Structure effects: the spatial organization of chemistry and the impact of past research activities on the geography of collaboration. . Network effects: the pre-existing interpersonal relationships between the alliance's members. . Policy effect: guidelines and incentives of national research institutions.
Analysing the articulation of these three dimensions is made possible using both quantitative data (from scientific collaborations) and qualitative material (interviews with researchers and research agencies). This article contributes to the literature about scientific network formation by demonstrating the interest and effectiveness of mixing interviews, bibliometric data and network analysis: this mixed-method approach proves operational for capturing the diversity of the determinants of the genesis and the geographical scope of scientific alliances.
Practically, the alliance under study is a green chemistry federation named INCREASE. Led by the CNRS, 1 it comprises six research laboratories and a club of industrial businesses located in cities of different sizes, belonging to neighbouring regions in western France, and coordinated from Poitiers, a French medium-sized city.
The remainder of the paper is structured as follows. Section 2 provides some literature background as well as the analytical framework used for this research. Section 3 presents the context of the case study and the data used, including both bibliometric data as well as information collected through qualitative interviews. Section 4 presents the research results obtained from the mixed-method approach. The findings discussed in Section 5 provide the grounds for public policy recommendations.

LITERATURE BACKGROUND AND ANALYTICAL FRAMEWORK
2.1. Knowledge networks as science and innovation policy instruments To describe territorial trajectories, examine knowledge spillovers and explain the uneven geographical distribution of innovative sectors, the literature on science and innovation in economic geography traditionally puts a lot of emphasis on the following: . Synergy effects between stakeholders on a city-wide or regional scale (Asheim & Coenen, 2005;Catini et al., 2015;Cooke et al., 1997). . Complementarity effects achieved through the networking of clusters, cities and regions on a continental or global scale (Bathelt et al., 2004;Morrison et al., 2013).
Although the reticular dimension of knowledge exchange and the links between places have attracted increasing attention in this field since the mid-2000s (Van Egeraat et al., 2015), this literature tends to categorize exchanges of knowledge that are not intra-local or intra-regional as 'non-local links' (Fassio et al., 2019;Meyer et al., 2011). Their geography is often approached through the binary opposition between 'distance effects' and 'network effects' (Bergman & Maier, 2009;D'Amore et al., 2013;Ter Wal & Boschma, 2009). When using social network analysis (SNA) indicators, the main focus of research on interregional knowledge networks is the benefit that regions can derive from their connectivity and the effect of a region's network position (approached via its centrality or betweenness) on science and innovation productivity (Bergé et al., 2017;Mitze & Strotebeck, 2018;Sebestyén & Varga, 2013). Considering what assemblages or 'constellations' of regions/urban areas emerge from innovation actors' networkinghow to characterize their structural properties and efficiencyremains rarely looked compared to assessing regions' or urban centres' capabilities. As pointed out by Kogler (2015), regional policies rarely are at the centre of regional innovation studies, excepted some branches of work related to Smart Specialisation policies (Balland & Boschma, 2021;Dosso & Lebert, 2020), European Framework Programs (Breschi & Cusmano, 2004;Scherngell & Barber, 2011) and national schemes such as cluster policies, for example, the competitiveness clusters in France (Gardes et al., 2015;Longhi, 2016). The structures formed by the latter, in their capacity to bring together actors from different cities and regions, are quite similar to the instrument that interests us in this article. In the case of cluster policies, what is at stake is typically to overcome network failures relying on an 'insufficient level of reciprocal absorptive capabilities of knowledge between public research organizations and firms' (Lucena Piquero & Vicente, 2019).
Regarding the research federation studied in this article, the need for linkage occurs between several R&D laboratories located in distinct urban areas. The aim is to create a visible pool of expertise in green chemistry that the partner firms can call on directly without having to contract each laboratory separately. In this context, we use the notion of 'network failure' not to refer to a market failure but to a collaboration failure, that is, a situation when intercity collaboration in a specific topic/discipline (meso) are inferior to what could be expected given the structure of the knowledge network at an upper level (macro). From this point on the research federation offers a means to address the failure at the micro-level. If the facilitation of relationships by this instrument proves to be efficient, an embeddedness phenomenon may then occur and modify the relational structure and more especially the geography of scientific collaboration observed at the meso-level.

Improving the understanding of knowledge networks: methodological issues
The literature describes two main mechanisms behind the formation of knowledge networks: links caused by a need for complementarity (between partners who have never worked together) and links building on previously established relationships (embedded in the network of preexisting relationships). The study of these mechanisms is well documented at the level of innovation actors, but it is less common to study this networking process at the aggregate level of urban areas. Some exceptions exist, however, such as the work of Ó hUallacháin and Lee (2014). Given the embeddedness of social relationships at the local level, it is interesting to consider the interurban level as revealing latent relationships that may be related to a structural embedding that has more to do with organizational or cognitive proximity than with geographical proximity. At the non-local level, geographical proximity nevertheless intervenes and remains a potential facilitator for the establishment and maintenance of relationships. Owen-Smith and Powell (2006) considered that very few works have yet studied the earlier moment leading to the formation of innovation networks (i.e., before they become visible and observable as such), and that 'new attention is being turned to both the origins of regional networks … with an eye toward understanding the necessary inputs for cluster formation, as well as the initial conditions that shape their trajectories'. As we follow this line, we note that little attention has been paid to the methodological strategy to capture the genesis of clusters and networks. Existing works are mainly based either on quantitative strategies using econometrics and/ or SNA, or qualitative case studies using interviews and resource materials. Both methodologies have their own advantages: quantitative studies are relevant to describe the roles of actors in networks and how networks properties evolve while qualitative studies are considered as appropriate for the evaluation of processes and contribute to enhancing information about the institutional environment and to a better understanding of the complexity of the formation process. The combination of qualitative and quantitative approaches through a mixedmethod therefore appears promising for an in-depth analysis of scientific network formation (Bernela et al., 2019). Balland et al. (2013) suggest not only to implement a SNA based on collaborative innovation data, but also to conduct 'a more qualitative approach … that could deepen our understanding of the motives behind networking and the role of more informal personal ties'. Data availability is therefore one of the main challenges since, in addition to quantitative data that allow to analyse networks' structure and composition, we need qualitative relational data to uncover the determinants of observed relationships.
Our research on the interurban cooperation supported by a French green chemistry alliance thus offers to combine spatial scientometrics approaches to interurban knowledge networks (Frenken et al., 2009), with the more traditional approach of inter-organizational networks in economic geography, which focuses on social embeddedness and emergence process. Applied to our case, the mixed-method implementation relies on the collection of bibliometric data (co-publication networks) combined with historical and qualitative data. We first carry out a SNA from bibliometric data, and then complement the analysis with a qualitative approach to enter the black box of networks. These data serve to take the context into account, to bring content to the SNA and to explore reasons for change (D'Angelo et al., 2016).

Analytical framework
The interregional scientific alliance, which we present in detail in the next section, needs to be conceptually defined as a research object at this point. We build on and adapt the definition of International University Research Venture (IURV) from Youtie et al. (2017), that is, research arrangements established by a university to partner with a university or other HERI entity in another country. The nominal characteristics of an IURV include the involvement of senior researchers from both home and host institutions, joint research projects, a formal agreement describing responsibilities of partners, and a public name for the venture. According to the efforts made to move beyond an informal researcher-to-researcher crossnational research collaboration mode, other characteristics include an appointed director of the IURV, administrative support, multi-year research commitments, agreed targets for IURV participant research projects, mechanisms for convening or exchanging research personnel, and performance review. The research federation INCREASE covers most of these characteristics. Youtie et al. (2017) investigate which specific features raise a collaboration from an informal international research relationship to an IURV, and what kinds of supportive environments lead to their development. One could consider that these scientific alliances arise from several motivations such as individual researcher collaborations that institutionalized over time, shared use of Overcoming knowledge network failures: evidence from a French interregional scientific alliance in green chemistry specific scientific equipment, funding opportunities, emerging research problems, etc. (Chompalov et al., 2002). Bozeman and Rogers (2002) highlight the 'knowledge value collective' in which researchers working on the same topic without knowing one another transition into a research collaboration network with explicit agreements. Georghiou (1998) links the growth of research collaboration to mechanisms such as research exchange, workshops, cooperative networks, and large-scale scientific equipment and instruments, and national-level initiatives. The author also notes the importance of indirect drivers of collaboration such as national economic development considerations.
Based on this literature, we propose an analytical framework to examine the scientific cohesion of the territory formed by the INCREASE federation and to understand the determinants that shaped its emergence. To investigate the specific case of this cross-regional research and innovation alliance, our framework considers three levels of explanation ( Figure 1).
First, to understand the context in which INCREASE emerged, we examine the French map of scientific collaboration in chemistry. Drawing on the concept of 'embeddedness' (Granovetter, 1985;Grossetti, 2008;Gulati & Gargiulo, 1999), we consider that the structure of scientific collaboration in chemistry that existed prior to the creation of the federation partly explains the decision of federating laboratories in western France, as well as the fact that this decision originated from Poitiers. Using a SNA approach, we assess the need for a better coordination of chemistry laboratories in this part of France, and the need to overcome a 'network failure'. For this purpose, we carry out a bibliometric analysis of scientific collaboration between French cities. Along with Katz (1994), we consider that scientific collaboration can be assessed from co-authorship data.
Second, we focus on interpersonal ties that existed prior to the creation of the consortium and favoured its emergence. The presumed need for cross-regional coordination can only be fulfilled if there is at least some level of acquaintance between the parties involved in the emerging network. To assess these relations, we use data collected through interviews with federation members and through data on these members' collaboration networks, which we extrapolated from a corpus of publication data from the Web of Science (the content of this corpus is detailed below).
Finally, although the federation is a 'bottom-up' construct, we show that its existence depends on it being signed off by public authorities. In fact the consortium would not have any official existence without the CNRS's moral and financial backing. This means that its scientific agenda is in line with the CNRS research policy at the national level. The consortium benefits from a context that made it possible, acceptable, and even desirable in the eyes of CNRS leadership. These circumstances are not just the product of the CNRS's internal policy, but also of a range of political, scientific, and socio-environmental priorities on various scales (local, regional, national, European). To examine the policy level, we use qualitative materials including interviews, reports and grey literature.
By considering these various levels of causality, as represented in Figure 1, we can identify the social, scientific, and historical factors that led to the integration of specific members into the consortium and the choice of its final geographical scope.

CASE STUDY AND DATA COLLECTION
3.1. Green chemistry research in France and its spatial organization Unlike nano-technology research, which developed in a limited number of scientific hubs due to the spatial concentration of dedicated funding (Bozeman et al., 2007;Meyer et al., 2011;Robinson et al., 2007), green chemistry in France has benefitted from support mechanisms that encouraged the creation of territorial alliances. The fact that this sector tends to develop through the networking of regional and provincial territories in such a way as to make several territorial alliances emerge on a national scale appears to offer an interesting counterpoint. This present study offers to explore a very distinct and less documented territorial development logic in economic geography than that of making innovation hotspots emerge in the most central city-regions. In so doing, we follow the avenue opened by Harrison et al. (2020) arguing for greater consideration and attention in regional economics for territorial development occurring in small and medium-sized cities, and the way innovation takes place in peripheral regions.
Along with nano-chemistry, green chemistry is one of the ground-breaking sectors that have transformed the way chemistry is taught, organized and practiced over the past 15 years (Moiseev, 2016;Morris, 2011). In France, in green chemistry no fewer than four structuring programmes support the clustering of researchers and industries on a geographical basis: . Instituts pour la Transition Energétique ( and renewable resources, with the aim of developing green chemistry by using biomass as a raw material. It involves firms from different sectors (cosmetics, agri-food and detergents) as well as eight research centres based in western France. Another of INCREASE's objectives is to promote the training of young researchers and the dissemination of knowledge in the field of green chemistry through the organization of the International Symposium on Green Chemistry (ISGC), which takes place every two years in La Rochelle. The INCREASE network is the latest addition to a range of institutional programmes. According to its participants, the consortium aims to fill a gap in the geography delineated by prior programmes. These programmes produced the impression that laboratories in the south, east and north of the country were more visible thanks to their integration to large networks, while laboratories based in the west were more isolated. To fill this gap, INCREASE connects six cities located across four neighbouring regions of western France: Occitanie, Nouvelle-Aquitaine, Pays de la Loire and Bretagne. This 'assemblage' has no administrative existence as such, but it more or less coincides with the area known as the 'Atlantic Arc'. Its territorial cohesion has been the object of historical debates about France territorial development: the question of whether it would be useful to support the structuring of the Atlantic Arc has been discussed at length by geographers and territorial planners in the early 1990s (Brunet, 1993). At the time it appeared that France's polarization around Paris was detrimental to the development of transversal relations along the western and southern coastal corridors.
This coastline running from Brittany to the greater south-west region is dotted with medium-sized cities of limited influence, including Poitiers, which hosts the headquarters of the INCREASE Federation and has the region's oldest university (Soumagne, 1993). Poitiers is located near several other medium-sized cities with more recently established universities: in the 1960 s for Angers, Limoges and Orléans, and in the 1990s for La Rochelle (Milard & Grossetti, 2019). The resulting network of HERI institutions is relatively dense, with a positive momentum in chemistry in terms of output, diversity and visibility (Milard, 2012).
While the importance of chemistry in this area is unquestionably an asset, this does not suffice to conclude that the territory delineated by INCREASE is characterized by an outstanding scientific cohesion in chemistry, or for that matter in green chemistry.

Case study research design
Our research is carried out through the application of a single case study design. We emphasize that the mode of inference is the one typical of qualitative studies. The key condition is the selection of the right kind of case and not the right number of cases since the latter will offer no measurable improvement in the probability of the conclusion (Yin, 2009). The units of analysis are the interregional scientific alliance (meso), the actors involved in this arrangement (micro) and the environment it is embedded in (macro).
Our approach is comprehensive, as we aim to understand the mechanisms at work behind the development of INCREASE. We are therefore involved in a unique case study of a qualitative nature, which is justified when one wishes to study empirically a complex contemporary phenomenon. The phenomenon must be examined in its organic context and aims to disclose the real dispositions and entanglements of actors' decisions in their environment (Benbasat et al., 1987). This research strategy should thus open the black box of the alliance building process. According to Yin (2009), a qualitative case study allows for an in-depth understanding of the process and longitudinal dimensions of the study object. Here, we move to a mixed method, as we have the potential to seize the scientific environment and the structure of relationships through publication data. The goal of our research is to provide a thorough and comprehensive description of the phenomenon we wish to study, in this case the genesis of an interregional scientific alliance. We aim to write 'a good story' in the sense of Dyer and Wilkins (1991), thus joining the position of Eisenhardt (1991) who Overcoming knowledge network failures: evidence from a French interregional scientific alliance in green chemistry considers that it is less the number of cases that is crucial than the researcher's capacity to generate new knowledge with the case(s) at his/her disposal.

Capturing prior networks through a bibliometric analysis
The bibliometric corpuses analysed in this article are from the Science Citation Index (SCI) Expanded. We consider three corpuses of publications (Table 1; and see Appendix A in the supplemental data online): . The first corpus (corpus A: 313,165 publications; see Figure A1 online) includes all French contributions to journals listed in the SCI Expanded, regardless of their discipline. . The second corpus (corpus B: 37,790 publications; see Figure A2 online) includes all French contributions to publications listed as chemistry journals in the SCI Expanded. . The third corpus (corpus C: 1813 publications; see Figure A3 online) includes all contributions published by consortium members. To delineate this corpus we asked the representatives of each laboratory to give us a list of the members whose current work is consistent with the research agenda of INCREASE (Schultz, 2016). Those include the launch of a call for projects by the National Research Agency (ANR), and the creation by the CNRS of a national network of green chemistry researchers named Chemistry and Processes for Sustainable Development (CPDD).
To obtain a global overview of these datasets, we used a spatial bibliometric framework, that is, the quantitative geographical analysis of publication activity, developed to process the entire contents of the Web of Science at the scale of urban areas . Urban areas were chosen as our resolution level to model knowledge networks because the administrative boundaries of cities are not comparable, and scopes of universities and labs are very heterogeneous from a place to the next. Knowing that the majority of co-authoring links happen within a single urban area, we place our focus on coauthoring links that develop between different urban areas. These links reflect collaborations that contribute to structuring scientific communities beyond the local scale (regional, cross-regional, national, macro-regional and global scale). We fraction the links' weight according to the number of distinct urban areas that contribute to each co-publication: this method is referred to as 'whole normalized counting' (Gauffriau et al., 2008).
Furthermore, in order to compare the different networks and to highlight significant links at a given level (e.g., that of the discipline) in relation to a higher level (e.g., that of all the scientific disciplines combined), we propose a statistical method of normalizing the links in order to highlight the most significant links.
Suppose that x ij is the number of co-authored papers between urban areas i and j in chemistry (corpus B), and y ij is the number of co-authored papers between urban areas i and j across all disciplines (corpus A). In order to compare x ij with y ij , we normalize them according to the following formulae: 3 X ij = x ij /(x i· + x ·j ) for corpus B; and Y ij = y ij /(y i· + y ·j ) for corpus A. We then measure the deviation between X ij and Y ij using a measure of deviation from independence, which indicates the degree of representativeness of co-authoring links between two cities in chemistry compared to in all disciplines.
This indicator of representativeness' formula is: If p ij > 0, then the intensity of relations observed in the field of chemistry is greater than expected in light of scientific collaborations between the two cities across all disciplines.

Understanding the emergence of the federation through stakeholder interviews
To understand the set-up of the INCREASE federation and to qualify the relations between INCREASE members as well as their integration into national and international green chemistry networks, we conducted two waves of interviews. Details about interviews are summarized in Appendix B in the supplemental data online.
The first wave focused on INCREASE stakeholders: seven research lab representatives, three industry leaders as well as three institutional partners (University of Poitiers, Poitiers' Regional Council and the CNRS's chemistry department). A total of 13 interviews were completed in 2016, with an average duration of 90 min. Their objectives were to question the stakeholders on the consortium's genesis, history and objectives, as well as on the benefits derived by each organization from participating in the network.
The second wave took place during the ISCG congress organized by INCREASE in La Rochelle in May 2017. At the end of this week-long conference which brought together more than 800 participants from 48 different countries, we completed 20 interviews of an average duration of 20 min, including seven with academic members of INCREASE (who had not been interviewed during stage 1), and eight with researchers or industrials who had co-authored a publication with an INCREASE member. We conducted these interviews with the help of data from the bibliometric analysis of corpus C. Interviewees were asked questions on their co-authorship networks and on their participation in national and international green chemistry networks.
These interviews led us to refer to grey literature on the history of green chemistry and on the organization of chemistry in France. For instance, we studied the creation and positioning of the other French green chemistry networks mentioned above in section 3.1, as well as the other networks and activities in which INCREASE researchers had taken part in the past.
Interview data were recorded and transcribed. These textual data are then the subject of a content analysis allowing to explore the determinants of scientific network formation; they also and above all allow to interpret elements observed quantitatively with SNA of scientific co-publications (i.e., verification and clarification) thanks to the identification of discourse elements in the interview data.

The organization of the chemistry field in France
For many stakeholders in the French chemistry sector, the choice of federating laboratories based in the west of the country is coherent with the discipline's national structure. While laboratories in eastern France appeared well connected and inserted into diverse networks, there was the feeling that those based in the west had little interaction between them despite their visibility and notoriety in the field of green chemistry.
According to the founder of INCREASE located in Poitiers, who was interviewed shortly after the launch of the federation in 2016, the French map of green chemistry networks was organized as followed: to the north-east, 'a giant banana-shaped area' spanning from Lille to Reims (the PIVERT institute); to the south-east, 'Lyon's chemical industry hub, with the AXELERA competitiveness cluster'; and further south, Montpellier with the 3BCAR Carnot Institute, and Toulouse with the Toulouse White Biotechnologies (TWB); 'but the whole of the Greater West was left out, with isolated laboratories'. For this reason, the idea of structuring this field in the 'Atlantic Arc' emerged, which led to the creation of INCREASE.
To test this interpretation and understand the national context in which this research federation emerged, we studied empirical data on the co-authorship of chemistry Overcoming knowledge network failures: evidence from a French interregional scientific alliance in green chemistry papers. Our analysis embraced urban areas, measuring the number of co-publications associating two different urban areas over the period 2007-14, just before the launch of the INCREASE consortium. Concerning the spatial distribution of French collaborations in chemistry, the cities with the greatest number of collaborations are Paris and Lyon. It also appears that the French map of chemistry is multi-centred, and that the two leading cities do not polarize most relations. Each French region has a relatively dense network of cities that contributes to the national network. Participating cities are not necessarily metropolises: they also include small and medium-sized cities such as Pau in the south-west, Le Bourget-du-Lac near Grenoble, Lorient in Brittany, Le Mans in the Pays de la Loire region, Limoges, Avignon, etc. Reversely, some university cities with a high ranking in other disciplines appear to lag behind in chemistry, such as Dijon and Marseille.
In order to qualify the existing relations between the six cities involved in the INCREASE chemistry consortium, we use SNA to compare the spatial distribution of cross-urban collaborations in chemistry and across all disciplines. Figure 2 shows overrepresented relations (using the normalization method presented in section 3). This map indicates that the chemical industry hub around Lyon accounts for a relatively small share of relations with INCREASE cities. Only researchers from Poitiers and La Rochelle have a special relationship with Lyon in chemistry. Aside from a few exceptions, the main collaborations of the six INCREASE cities in chemistry are mostly located in the north and west of France. In particular, INCREASE cities share strong links with small and medium-size western cities: Pau, Orléans, Le Mans, Angoulême, Albi and Lorient. Figure 3 thus suggests an east-west divide confirmed by the interviews.
According to our interviewees, aside from their relative independence from the 'chemistry valley' (south-east France), western cities are traditionally isolated from each other. To ascertain this second fact, we apply a community detection algorithm to the network of collaborations between French cities in chemistry. This algorithm partitions the network into several groups characterized by a greater density of interactions between their members than with external groups. By applying the 'Louvain' method, 4 we identify seven collaboration areas in chemistry represented in Figure 3. We observe that INCREASE cities belong to different groups. It confirms and makes it possible to objectify the feeling of disconnection and the need for coordination described by the initiators of the federation. Rennes, Nantes and La Rochelle connect with the Brittany group as well as with a few medium-sized cities in the centre of France (green group). Bordeaux and Toulouse are part of a group that includes the greater south-west (purple group). Located at the intersection of Brittany and greater south-west, Poitiers is isolated and connected to the central network formed by Paris and Lyon (light blue group). Figure 3 shows that Poitiers sits in a unique position: the choice of creating a Poitiers-based consortium to develop links between western French cities thus appears as an ambitious one, as this initiative could transform established networks.
Rather than being a satellite of Paris and Lyon, Poitiers aspires to become a bridge in the Atlantic Arc. According to the consortium's founder, Poitiers indeed holds a strategic position in this cross-regional network of six cities: When you look at the map, Poitiers is at the center, so when we have meetings, everyone is at about two hours' distance, which is pretty good for our operations. Proximity is importantyou can always use videoconferences but they're not the same.
This bottom-up decision, signed off by national CNRS leaders, intends to 'fix' the lack of connections between the north and south of the Atlantic Arc in chemistry, while deeply transforming Poitiers' position in the overall structure. As such, the emergence of this new alliance is a way to overcome a 'network failure', which is both felt by actors and empirically measurable with a replicable approach based on SNA: the detection of overrepresented links combined with the detection of strongly connected subgroups.

The role of interpersonal relationships
In the absence of much prior intensive scientific collaboration between network members (only Toulouse-Rennes; Nantes-La Rochelle, Nantes-Poitiers and Bordeaux-Toulouse had some prior co-authorship links according to corpus C, that is, only a quarter of the possible connections), other types of relations played a crucial role in the design and roll-out of the INCREASE network. In our interviews, we noted that several members of the consortium knew each other, or were at least aware of each other, before the federation's launch. These acquaintance ties were decisive in the partners agreeing to join INCREASE when contacted by Poitiers' laboratory. The interviews allow us to distinguish three types of ties, complementing typologies documented by Bernela and Milard (2016): . Type 1: ties formed through researchers' professional mobility experiences, 'job-to-job mobility'. . Type 2: ties resulting from teaching and training functions (lectureships, thesis juries, etc.), 'academic ties'. . Type 3: ties formed between researchers contributing to common institutions, conferences, and research programmes, 'organisational research linkages'.
Type 1 refers to the job-to-job mobility of scientists and its impact on network formation over time, and types 2 and 3 refer to the temporary circulation of scientists and the importance of networking in the academia organized as a professional system. Relationships can form over the course of an individual's career, including between laboratories that do not have a record of scientific collaboration.
By way of illustration of type 1, the Rennes and Poitiers INCREASE representatives met when they were both doing a doctorate in Dijon on subjects unrelated to green chemistry. Furthermore, the federation's founder briefly worked in Rennes before taking up his job position in Poitiers. For this reason, although the Rennes and Poitiers laboratories have never yet collaborated on green chemistry projects, they might be able to do so in the future thanks to their relation of trust and to the consortium's incentive mechanisms.   Overcoming knowledge network failures: evidence from a French interregional scientific alliance in green chemistry 1629 laboratory]. But I do keep in touch with them: I take part in their recruitment days, I teach in a training course we run in common and we were in contact at the time of the creation of the Green Chemistry Institute.
The Poitiers laboratory has indeed long specialized in green chemistry, and has done so even before the arrival of the INCREASE coordinator in Poitiers. An initiative led by the previous generation of researchers had helped position the Poitiers lab in the national CPDD programme (2007-09), and federate local researchers with support from the Region. And finally, to illustrate type 3: aside from those who get to know each other through working or meeting in the same laboratory at a point in their career, or through contributing to common training programmes or thesis juries, others are well known because of their prominence in the field. For instance, two INCREASE researchers explained knowing each other before because they had been members of the same regional branch of the Société Chimique de France. This is also the case with the Bordeaux and Toulouse INCREASE representatives, known for having sat on the CNRS's National Committee and on the ANR, the agency in charge of the evaluation and selection of calls for national projects. From 2006, some programmes led by the CNRS and the ANR contributed to the national structuring of the 'sustainable' chemistry field. These programmes played an essential part on three levels: they helped the teams working on this subject to become aware of each other; they helped foster links with industrials; and they contributed to setting a scientific agenda.
4.3. Policies supporting the emergence of a regional multi-skill network 4.3.1. Incentives for a multi-skill network The networking of research institutions can be motivated by purely scientific motives: the network is often seen as a way of solving research problems by cross-pollinating skills and knowledge from fields with little communication between them (Rafols, 2007). The development of green chemistry is a good illustration of this need for cross-pollination: today's challenges require connections between chemistry (reaction, catalysis), processes (chemical engineering), biomass issues (biology) and life cycle analysis (environmental impact). The main barrier to the pooling of skills is the scientific communities' poor awareness of each other, due to their respective institutional processes (recruitment, conferences, journals, etc.) which keep them isolated from each other: several interviewees reminded us that chemistry itself is deeply divided into silos according to the type of catalysis the researchers specialize in (homogeneous vs. heterogeneous vs. enzymatic catalysis). Public policy instruments can be used to foster the development of hybridization between specialties and thus promote the emergence of new interdisciplinary fields such as bio nanotechnology studied by Rafols (2007) and green chemistry, which we examine here.
As mentioned above (sections 3.3. and 4.2), from 2006, the CNRS and the ANR contributed to promoting sustainable chemistry's development and organization in France (Schultz, 2016). We can distinguish three steps in this historical process: . In 2006, the CNRS launched a cross-disciplinary programme (CPDD) aimed at drawing an inventory of sustainable chemistry teams in France. The programme was led by three laboratories that later formed the core of the INCREASE consortium: Poitiers, Nantes and Toulouse. This initiative offered little funding but created a momentum amongst CNRS researchers. The Toulouse INCREASE representative reported that this programme had given him the opportunity to launch a thematic summer school and an international conference on sustainable process engineering. . Shortly after this the ANR launched a large-scale call for projects, 'Chemistry and Processes for Sustainable Development', targeting the entire French community beyond CNRS researchers. This time, some funding was available to support selected projects and the programme offered strong incentives for multi-partner projects. The first phase was successful but its outcomes in terms of building links with the industry were disappointing. . A second phase of the ANR call for projects was launched in 2010 under the title 'Sustainable Chemistry -Industries, Innovation'. The programme supported research projects with the ability to produce technology transfers that could benefit the industry and focused on two aspects: improving communication between chemistry, processes and biotechnology; and assessing the costs and environmental impacts associated with resources, reactions and processes (Schultz, 2016).
These last incentives significantly oriented the selection of labs contacted for the creation of the INCREASE consortium. In particular, the issue of environmental impacts was first built into the agenda of the ISGC international congress, organized bi-annually by the Poitiers lab. Individuals who expressed their interest in this aspect of the congress were then invited by the founder of INCREASE to join the consortium. As explained by the La Rochelle INCREASE representative: At the congress, he [the creator of INCREASE] wanted to have ethical and environmental impacts on the agenda. In parallel, two teams working on these aspects were invited to join INCREASE. The Bordeaux team (Molecular Science Institute), which specialises in lifecycle analysis, and us, who are working on eco-toxicology and on the impact of pollutants on ecosystems. He was really keen to include these aspects.
Other interactions also played a central part in the creation of the consortiumincluding the relation between chemistry and processes. As summed up by the La Rochelle INCREASE representative, 'I hope the federation really helps build bridges between process and synthesis experts, and also between lifecycle analysis and environmental impact experts.' In addition, by involving chemists from Rennes, Nantes and Poitiers, the consortium brings together specialists on all three types of catalysis, which are hardly ever associated (enzymatic, homogeneous and heterogeneous). This diversity of research subjects is the reason why the various teams had never had the opportunity to collaborate much before. In the rationale of the Federation leader, combining diverse specialties has both scientific and economic benefits: the joint objectives are to further sustainable chemistry and to create a 'toolbox' that industrials can access by directly contracting with the consortium rather than having to negotiate with each separate team.

Incentives for a cross-regional network between neighbouring regions
Interestingly, the interviews produce the impression that the INCREASE consortium's geographical coherence is more the result of multiple decisions than that of a clear intention to stay within the boundaries of a predefined geographical perimeter. According to the consortium's leader, geographical proximity was not initially perceived as a must. It was even envisaged for the consortium to be 'international' from the outset and to include foreign laboratories. In addition, the geographical scope of the INCREASE consortium appears to be only partially the product of pre-existing relations between its members. As explained by Harrison (2016) for the matter of UK university alliances, it is interesting to understand how the founders set the boundaries of such 'constellations', and to identify the criteria that determine the fact that one laboratory ends up being in the network and another is not.
In the case of INCREASE, some interviewees explained that the Lille team could also have been part of the consortium due to its prior connections with Poitiers and Rennes, to its scientific specialism in characterization and to its international reputation. Similarly, Montpellier has a significant collaboration record with Bordeaux in the field of polymers and could legitimately have been part of the consortium. However, geography appears to have gradually become a resource on different levels. This is evidenced by the arguments put forward and examined above (section 4.1) related to the need for structuring and connecting the Atlantic Arc laboratories.
Indeed, the consortium's geographical coherence has proven to be a strong asset to rally support from national and international institutions and from industry partners. This coherence now makes the French map of green chemistry more readable. It also contributes to the international visibility of western France. In a context of growing international competition with the emergence of new congresses on this theme in Germany and New Zealand, it contributes to anchor the Poitiers group into a structured local network with an established reputation for its diverse skill base. In terms of science-industry transfers, this geographical coherence is also an asset. The industry was consulted during the creation of the consortium, which presents itself as a means of facilitating interactions and cutting down the red tape attached to contracting with each laboratory. From the perspective of large industrial partners, the consortium ensures that their negotiations with their usual contacts in western France take place on a level where skills and resources have already been pooled.
Overall, the emergence of INCREASE responds to a trend that is broadly encouraged by today's public authorities and policies: clustering and/or networking research institutions located within one same area that offer a combination of diverse resources and skills, even if this area goes beyond the regional administrative boundaries to give rise to a transregional configuration extended to the whole of western France. This response to existing needs is driven by the founder of INCREASE, who wants to develop his leadership in a context where the CNRS is keen to train professional research executives. Ultimately, the creation of INCREASE owes as much to his skills as a researcher as it does to those as an academic entrepreneur (Jain et al., 2009;Perkmann et al., 2013): an entrepreneur who took advantage of geography to build a consortium that could in turn transform the geography of knowledge connections.

DISCUSSION AND CONCLUSIONS
Green chemistry refers to a turn in chemistry towards adopting more environmentally friendly practices (Marion et al., 2017). These practices, which can be expected in term to spread across all of chemistry, are today becoming structured as a specialismor at least, as a cross-disciplinary area of work. Its development parallels that of nanochemistry, which has also contributed to transforming the discipline. In both areas, change does not only affect major international hubs. The emergence of new perspectives in chemistry makes it necessary to move beyond old collaboration habits and to build links between teams that had not been accustomed to collaborating in the past. Coordination mechanisms, such as national or European calls for research projects, were created for this purpose. Another structuring initiative in France has been the opportunity to form consortia of laboratories around specific themes. This opportunity, although rarely taken up, is one of the mechanisms developed by the CNRS to structure the national map of scientific activities. It can facilitate the emergence of cross-regional networks whose boundaries are original in that they do not match existing administrative unitssuch as for instance the scientific federation in green chemistry INCREASE.
The main objective of this article was to examine the determinants of the creation of the cross-regional thematic alliance, and its scientific and geographical delineation. For this purpose, we examined the spatial organization of the chemistry field and the impact of past research activities on the geography of collaborations (structure effect), the various types of pre-existing relations between consortium members (network effect) as well as the impact of policy incentives, scientific and socio-economic issues Overcoming knowledge network failures: evidence from a French interregional scientific alliance in green chemistry on the scope of this inter-organizational creation (policy effects). These three levels of causality shed an interesting light on the structuration processes that affected this cluster of western France cities in green chemistry.
1. Structure effect: The state of scientific collaborations in chemistry between French cities highlights the existence of a network failure in the 'Atlantic arc', which legitimizes the federation's creation Working from bibliographic data and interviews, we demonstrated how the need for the creation of this consortium stemmed from prior collaboration habits in chemistry. Following an assessment of the disconnection between cities in the north and south of the Atlantic Arc, and of Poitiers's relative isolation from these other cities, it was felt that a support programme was needed to enable large cities of the Atlantic Arc to pool their skills in sustainable chemistry. Geography appears more as a resource than a constraint for the coordination of scientists: the laboratories that joined the consortium did not do so because of their shared belonging to the Atlantic Arc, but rather because of their thematic diversity and of the complementarity they can see between their respective skills. Once the consortium is in place, geographical coherence helps define its thematic positioning, considering local natural resources and industrial networks; geography also consolidates the consortium's legitimacy and visibility in the global arena.
2. Network effect: The federation's geographical scope derives from prior relations of acquaintance and from projects of future collaborations between its members Since the laboratories involved had few records of scientific collaboration, relations of acquaintance played a key role in the implementation of the network: those included interpersonal ties deriving from career mobility; professional ties formed through joint participation in training sessions, thesis juries or scientific events, which all involve a degree of thematic proximity; and finally, ties associated with belonging to the same discipline, and with taking part in the discipline's national and regional institutions. It is consistent with the seminal work of Agrawal et al. (2006, p. 573), arguing that: geography is likely to be less important in mediating social relationships between individuals in the same field since they have various alternative mechanisms through which to establish relationships. For example, individuals in the same community of practice or invisible college attend conferences together, belong to common associations, and have other institutional settings in which to fraternise and share ideas.
From there, geographical and institutional proximity are not so decisive for networking. Temporary geographical mobility opportunities characterized by a combination of organized proximity and temporary geographical proximity between non-co-located partners (visiting scholar positions, conferences, juries, etc.) offer linkage opportunities and foster network building (Bernela & Milard, 2016;Torre, 2008).
3. Policy effect: The consortium's emergence was made possible by its alignment with a number of scientific and political priorities Finally, the context in which the consortium's shape and scope were determined also played an important part, as the founders had to make sure that INCREASE did not overlap with the scope and specialism of other existing alliances, and that it took in existing policy and scientific incentives towards more cross-pollination between chemistry, processes and economic and environmental impact. Unlike the alliances studied by Harrison et al. (2016), which are initiated by the government to facilitate the allocation of HERI funding and let alliance members distribute the funds between themselves in proportion with their weight or evaluation scores, the instrument we studied is a flexible one, born from the scientists' desire to federate around a common theme. Member laboratories retain their full autonomy and freedom to work with any other laboratory. 5 Some members may question the consortium's boundaries and wish that such or such laboratory had been included; others are not very convinced by the need for geographical coherence (as discussed above, this coherence is a resource more than a necessity); but all understand the benefits of a consortium of this type. These results, although obtained from one case study, allow us to discuss the role of geography in the design of public support schemes for scientific collaboration and innovation. Geographically speaking, the consortium was designed and rolled out from Poitiers. This laboratory is a pioneer in green chemistry, but in a context of growing competition, Poitiers researchers led by a CNRS manager found it particularly useful to identify close allies in other neighbouring labs, in order to preserve their leadership and visibility in their field. While according to Harrison's (2016) vision, small universities tend to benefit less from university alliances, in this case the consortium is driven by a laboratory based in a medium-sized university town, which occupies a central place, both geographically and scientifically, in this network.
If a lesson had to be drawn from this study, it would be that it is not for geography to constrain public policies, but instead for public policies to facilitate the activation of connections on different geographical scales. No single level should be privileged above othersfor instance through pooling all efforts into the development of local synergies (as was the case with the development of 'technopoles'), or into the consolidation of international links (Laranja et al., 2008). On the contrary, the INCREASE case shows that cross-regional links can also be wise. As shown by recent studies on network failures (Lucena Piquero & Vicente, 2019), scientific collaboration policies should not set constraints but instead ambition to support researchers and industrials. This involves giving them the instruments they need to facilitate connections that appear to have potential but that are challenging to implement for historical reasons to do with collaboration habits and institutional boundaries (such as the administrative unit of the region).