See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/282426900

A century of physics

Article  in  Nature Physics · October 2015

DOI: 10.1038/nphys3494

CITATIONS READS

62 1,178

5 authors, including:

Roberta Sinatra Pierre Deville

Northeastern University Université Catholique de Louvain - UCLouvain
39 PUBLICATIONS   810 CITATIONS    7 PUBLICATIONS   604 CITATIONS   

SEE PROFILE SEE PROFILE

Albert-Laszlo Barabasi
Northeastern University
502 PUBLICATIONS   137,207 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Privacy-preserving data-mining View project

All content following this page was uploaded by Roberta Sinatra on 24 November 2016.

The user has requested enhancement of the downloaded file.

 perspective

A century of physics
Roberta Sinatra, Pierre Deville, Michael Szell, Dashun Wang and Albert-László Barabási
An analysis of Web of Science data spanning more than 100 years reveals the rapid growth and
increasing multidisciplinarity of physics — as well its internal map of subdisciplines.

T
he conventional narrative of physics physics journals and documented in Web of Taken together, we find that the literature
is one of paradigm shifts1: from the Science (WoS) between 1900 and 2012. of direct interest to the physics community
Copernican Revolution to Einstein’s The problem with the above definition is more than twice that published by
annus mirabilis2. And for many, the stories is that many influential physics papers, physics journals: on top of the 2.4 million
would seem to involve genius in isolation — and an increasing fraction associated core physics papers (Box 1c, blue), there
the lone physicist divorced from other with Nobel prizes, are published in are 3.2 million interdisciplinary papers
sciences, unperturbed by societal beliefs. interdisciplinary journals, such as Nature published in non-physics journals (Box 1c,
But the reality is quite different: physics or Science. Furthermore, many papers red), that, based on their referencing and
has always been in a constant dialogue of interest to physicists are published in citation patterns, are indistinguishable
with other disciplines, be it mathematics, journals of other disciplines. Take for from papers published in physics journals.
chemistry or theology. This dialogue is example the founding paper of chaos We identified six physics Nobel winning
largely driven by methodology: what theory, a thriving subfield of statistical publications8 in this interdisciplinary set,
traverses disciplinary boundaries is the physics, which was published in Journal of and many other highly influential physics
idea that complex phenomena can be the Atmospheric Sciences 5. This and many papers, such as Hubbard’s 1963 model of
understood in terms of a small number of similar examples force us to address an interacting particles9 and Hopfield’s 1982
universal laws3. important question: how do we identify paper on neural networks10.
In this era of interdisciplinary science, papers that are not published in the physics
of biological physics, network science and literature, but given their subject matter and The growth of physics
econophysics, defining physics as the science their impact on the evolution of physics, Throughout the history of physics, major
of the properties of matter and energy 3 is could or should have been? paradigm shifts, such as the development
increasingly outdated and inaccurate. We To map out the complete physics of quantum physics, have spurred
are therefore prompted to ask anew: what is literature, we compared the references and significant new research, resulting in a
physics? When two engineers accidentally citations of all papers in WoS to a null model burst of publications and giving birth to
discover cosmic microwave background in which each paper’s citations are assigned new and enduring subfields, from nuclear
radiation, is that physics or engineering? randomly, regardless of a paper’s journal or to condensed-matter physics11. The very
When a physicist uncovers the structure research area6. A paper is a potential physics existence of this growth is supported by the
of DNA, is that biology or physics? Is the publication if its references and citations to number of physics papers published each year
interdisciplinary role of physics something the core physics literature are significantly (Fig. 1a), which has been increasing roughly
new — a potential fad — or has it always higher than expected by chance6,7. Our exponentially for the past 110 years, an
been an integral part of the field? Is physics algorithm recursively scans the 40 million expansion that was halted temporarily only
dying or thriving, becoming more insular papers published between 1900 and 2012 and by the two World Wars. Note, however, that
or more interdisciplinary? To answer documented in WoS, identifying ~5.1 million the growth rate of physics is indistinguishable
these questions, we will rely on the very papers of potential interest to the physics from the growth of science in general12.
framework physics pioneered: collecting community outside the core (Box 1). Hence, the field’s exponential growth is not
data from which to draw our conclusions. This corpus contains two classes of driven by paradigm changes, but by societal
papers: the first class consists of 4.5 million needs, and capped by access to resources.
What is physics? papers whose references are significantly This growth was particularly remarkable
The late Cambridge physicist Sam Edwards biased towards core physics papers; this is following World War II, when the physics
once remarked that “Physics is what the body of literature within the physics literature doubled every 6.5 years. And yet,
physicists do.”4 Following in his footsteps, we influence sphere. The second class contains after 1970 this growth slowed, settling on
define physics not from an epistemological 3.8 million papers heavily cited by core its current rate of doubling every 18.7 years.
point of view 3, but look instead at what physics papers, representing papers of direct Once again, the recent slowdown is not
physicists do. We do so by focusing on interest to the physics community. The unique to physics, but characterizes the whole
the research papers through which we intersection of these two classes consists of scientific literature contained in WoS. Finally,
communicate our basic discoveries — 3.2 million papers, distinguished by the fact whereas pre-1910 physics literature was
forcing us to ask: what exactly is a physics that they reference the physics literature limited to physics journals, since the 1920s
paper? A simplistic answer would be: it and are also cited by it in a statistically the growth of the core and interdisciplinary
is a paper published in a physics journal. significant fashion. Hence, these papers are physics literature have been indistinguishable,
This narrow, yet obvious definition allows indistinguishable from the physics core, apart indicating that publishing outside the physics
us to construct a core physics dataset from their place of publication, prompting us core has been integral to the development of
of ~2.4 million papers published in 242 to call them interdisciplinary physics papers. physics throughout the last century.

NATURE PHYSICS | VOL 11 | OCTOBER 2015 | www.nature.com/naturephysics 791

© 2015 Macmillan Publishers Limited. All rights reserved

perspective

Box 1 | Defining physics

The key unit of communication within of journals being extracted by combining citations were randomly reassigned to
science is a research publication, information from Wikipedia, Scopus and any paper in the WoS dataset, regardless
whose references and citations contain Scimago (www.scimagojr.com). These core of the discipline. If the observed fraction
considerable contextual information about physics papers are shown as blue nodes and of references and citations to core physics
the topic and the discipline of the paper marked with P in panel b. papers was significantly larger than in
(panel a). We mined this information To identify papers that are not in the the null model, we labelled the paper an
to understand the nature and evolution core, but nevertheless belong to the physics interdisciplinary physics paper, shown as
of physics. literature, for each paper connected to red nodes and marked with IP in panel
To identify the whole corpus of core the core by a reference or a citation, we b. The first three steps of this procedure
physics papers, we started from 2.4 million measured the fraction of its links to the depicted in panel b were repeated until the
papers published in 242 physics journals core physics literature. Then we estimated algorithm converged and no new papers
indexed in Web of Science (WoS), the list the expected fraction if references and were added.
In this way, our algorithm identified
3.2 million papers published outside the
a b c Web of Science, 27 million papers
? ? main physics literature that, with respect
? Not physics to their reference and citation patterns, are
Citations
indistinguishable from the core physics
P
? papers (panel c).
P ? Interdisc. physics
P 3.2 million papers To validate our algorithm, we measured
its ability to detect a set of papers published

0.6 million papers

?
1.3 million papers
Citing physics

? outside the physics core that are classified

Physics-cited
Core physics
2.4 million papers as physics in Science and Proceedings of the
? ? 242 journals National Academy of Sciences USA (true
? positives). We also tested the algorithm’s
? P ability to minimize the inclusion of papers
Publication IP that are known to have no relation to
P
P physics (false positives). The algorithm
? identifies 92.4% of the 1,178 physics papers
?
as true positives, and includes only 0.5%
? ? d Of 1,178 Of 3,715 of the 3,715 non-physics papers as false
IP physics papers... non-physics papers... positives (panel d). To check that the
...90 identified ...3,698 limited coverage of WoS (for example, due
? P
References as non-physics correctly to ceased journals before the database was
P IP identified as
P non-physics
created in the 1960s) did not influence our
? results, we ran all analyses on the dataset of
? American Physical Society papers, which
...17 identified
...1,088 correctly as physics contains a complete set of physics papers
identified as physics and citations from 1893 to 2010. We found
no qualitative differences in the results.

The literature published in physics American Physical Society (ref. 13 and collaborative effects: although on average
journals went from representing around 4% Sinatra et al., manuscript in preparation), each physicist continues to write fewer than
of the scientific literature in 1945 to about finding that the number of authors has one paper (Fig. 1a), she/he ends up as co-
10% after 1980, and has been approximately increased at the same rate as the number of author of multiple publications. In other
steady since then (Fig. 1b). Interdisciplinary papers (Fig. 1c). This leads us to conclude words, during the past decade, collaborative
physics has followed a similar pattern, that the growth of physics literature is driven work has significantly boosted individual
growing from 6% in 1945 to a maximum solely by the increasing number of authors. productivity when measured as whole paper
of 18% in 1964, and stabilizing at 12% after We do observe, however, nontrivial shifts counts14,15, while leaving the field’s overall
1980. The wider physics literature represents in productivity. Indeed, whereas before productivity unchanged.
around 22% of all scientific literature 2000, a typical physicist co-authored fewer
since the 1980s, a remarkable fraction than one publication per year, in the past Impact and its variations
that documents the profound role and 15 years, the number of papers co-authored Is the steady growth of physics accompanied
embeddedness of physics within the larger by each physicist jumped above one for by a steady growth in impact? To address
scientific enterprise. the first time (Fig. 1d, black curve). Yet, this question we used the average number
Is the exponential growth of the physics this remarkable growth in productivity did of citations each paper acquires within
literature driven by an exponential growth not boost the field’s overall productivity 10 years of its publication, 〈c10〉 (Fig. 2a) as
in the number of physicists, or by gradually (Fig. 1c), as the total papers-per-author ratio a proxy for impact 16,17. We find a puzzling
increasing productivity? To answer this dropped slightly (Fig. 1d, red curve). This peak in impact for papers published between
question, we used the disambiguated indicates that the observed singular growth 1955 and 1965, hinting at the existence
authorships of papers published by the in individual productivity has its origins in of a ‘golden age of physics’ around 1960.

792 NATURE PHYSICS | VOL 11 | OCTOBER 2015 | www.nature.com/naturephysics

© 2015 Macmillan Publishers Limited. All rights reserved

 perspective

Interdisciplinary physics papers also had a a

small peak in the 60s, but in the 1990s their 106
Web of Science
impact overcame their 60s peak. These peaks Interdisciplinary + core
prompt us to ask: did core physics really Interdisciplinary
105
peak in the 1960s? Core

Number of papers
A steady growth of citations is typically
fuelled by two effects: a growing number of 104
papers, and a growing number of references
per paper. None of these curves shows
discontinuities around the 1960s; however, 103
the exponential growth in the literature was
unabated in this period (Fig. 1a) and the
average number of references per paper has 102
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
been growing steadily from two references in
Year
1900 to approximately 15 in 2000 (Fig. 2b).
There was, however, a clear discontinuity in b c d
106 2.0
the way we cite papers, occurring during the 1.0 Author yearly
1960s. Indeed, in the late 1930s, physics was 105 productivity
the most myopic it has ever been: citations 0.8 1.5 Papers per author
104
Fraction in WoS

were largely to papers published within 0.6

the previous four years (Fig. 2c,d). After 103 Cumulative # 1.0
World War II, the average age of references 0.4
102
new authors
Non-physics
dramatically increased, probably because Cumulative # 0.5
0.2 101 new papers
there were too few papers published during Interdisciplinary
the war period18,19. The scientific community 0 Core 100 0
simply continued where it left off before the

00
00

00
00
00

40
40

80
60
80
60

20
20
00

40

60

80
20

20
20

20
war, citing pre-war papers.

19

19

19

19

19
19

19

19

19

19
19

19

19

19

19

Year Year Year

As the literature recovered, the myopic
citation style set in again — the age of Figure 1 | The evolution of physics. a, Changes in the number of papers in physics and the Web of Science
references reaching a new low in 1960. (WoS) data over the past century. The plot shows that the literature has been growing exponentially
This was followed, however, by a gradual over time. The growth was steepest between 1950 and 1970, doubling every 6.5 years. After 1970, the
but sustained change in citation habits doubling time increased to 18.7 years. Growth was only disrupted around 1915 and 1945 due to the World
that persists even today: physicists are Wars. b, The fraction of physics core papers grew slowly over time, from 4% in 1920 to about 10% in
systematically reaching deeper into the 2000 (blue). The fraction of interdisciplinary physics papers (red) showed a dip during World War II, but
literature, citing older and older papers recovered to a stable fraction of about 12% of the whole scientific literature. c, During the past century
(Fig. 2c,d). Our hypothesis is that this the number of authors increased at the same rate as the number of papers. d, Overall productivity,
dramatic change is a consequence of peer representing the number of papers per author, the ratio of the curves shown in panel c, has been
review. Indeed, before the 1960s, papers fluctuating between 1900 and 1950 in the vicinity of one. In contrast, the number of papers co-authored
were accepted solely on editorial discretion20. by each physicist has been less than one during the past 100 years, but increased sharply in the past
The increasing specialization of science has 15 years. This growth is a direct consequence of collaborative effects: individual productivity is boosted
created the need for expert opinion, which because physicists end up on many more papers as co-authors.
was difficult to implement until photocopying
became available in 1959 (ref. 21).
Peer review became common in the necessarily represent evidence of a golden was strongly coupled to core physics. In
1960s — Nature officially introduced it in age of physics; it is likely an unintended 2000, however, the self-referential nature of
1967. The reviewer’s ability to point out consequence of the emergence of deeper core physics significantly decreased, while
relevant work not only boosted references referencing, probably induced by the gradual that of interdisciplinary physics slightly
to the older literature, but also induced a introduction of peer review. increased. In other words, over the past
behavioural change, prompting the authors Figure 2a also documents a systematic four decades, core physics has lost some of
to more carefully credit earlier work. This gap in impact between core and its tribal nature, becoming more open to
change in referencing habits can explain interdisciplinary papers. To understand its the vast non-physics literature. In contrast,
the 1960s peak in impact. Indeed, whereas origin, we explored the citation patterns of interdisciplinary physics started to act as
papers published in the late 1950s were core and interdisciplinary physics in 1960, a separate discipline, tightly coupled with
still highly cited by the papers immediately when the impact gap was the largest, and physics, but also increasingly tribal.
following them, they also got an additional in 2000, when it was close to vanishing
boost from later papers. And yet, papers given the steady rise in the impact of The inner structure of physics
published from the late 1960s suddenly had interdisciplinary physics papers. As Fig. 2e Physics is not monolithic, but consists of
to compete for citations with much earlier indicates, in the decade 1950–1960, core a number of subfields, each with its own
papers, and in doing so they lost citations. physics was extremely self-referential, intellectual challenges, methodologies
Indeed, a null model indicates that a change sending 11.1 times more citations to papers and culture. We used the Physics and
in the referencing habits documented in in physics journals than expected by chance. Astronomy Classification Scheme (PACS)
Fig. 2d does induce a peak in citations for Interdisciplinary physics was much less by the American Physical Society to classify
papers published in the 1960s (Fig. 2a). self-referential, and with a fourfold increase the entire physics literature (core and
Hence, the 1960s peak in impact may not in references towards physics journals, it interdisciplinary) into ten major subfields.

NATURE PHYSICS | VOL 11 | OCTOBER 2015 | www.nature.com/naturephysics 793

© 2015 Macmillan Publishers Limited. All rights reserved

perspective

a b c
20 12
15 Deep referencing

Average age of citations (years)

Average number of references
Core 11
Average number of citations

Interdisciplinary
15 1946
10
after 10 years 〈c 10〉

10
9
10
8 1960
1919
5 7
5
Core
Interdisciplinary 6
1931 Myopic referencing
Null model
0 0 5

0
0

00

40

60
20
00

40

60
20

80

00

20

40

60

80

0
80

0
0

0
20
20

19

19

19

19
19

19

19

19

19
19

19

19
19
19

20
19

Year Year Year

d 0.25 e
1930 1950-1960 1990-2000
1940
Fraction of citations given

0.2
1950
1.2 1.3
1960
NP NP
0.15 1970
1980
1990
0.1 2000 0.04 0.2 0.03 0.1
0.08 0.1 0.08 0.1
11.1 4.5 7.0 4.8
1.7 2.1
0.05
CP IP CP IP
3.9 2.8
0
-10 -8 −6 −4 -2 0
Years back

Figure 2 | The evolution of impact. a, We used the number of citations collected over 10 years, 〈c10〉, as a proxy of a publication’s long-term impact. The average
impact of physics papers has grown from 1900 to 1950, and papers published between 1950 and 1960 received the most citations. The dotted line corresponds
to a random null model, obtained by using the growth in number of papers (Fig. 1a), the increasing number of references (panel b) and different citation age
distribution (panel d). These three time-dependent factors together can reproduce the observed impact peak in the 1960s. b, The average number of references
per paper has been increasing steadily over time. c, The average age of the references of papers published in American Physical Society journals, documenting
large variations in the depth of referencing over the past century. d, The probability of citing past papers for different years shows again remarkable shifts in
citation patterns from myopic referencing in the 1930s–1950s to the current, increasingly deep referencing. e, The flow of citations between the core physics
(CP), interdisciplinary physics (IP) and the non-physics (NP) literature in the decade before 1960 and before 2000. Node sizes are proportional to the logarithm
of the number of papers published in each area. The weight of each link corresponds to the number of citations divided by expected citations, for instance, a
weight 2 indicates that citations are twice the number of expected citations in a null model, where citations are randomly redistributed between areas.

We find that these subfields naturally self-referential it is. Indeed, we find that a PACS code, and fitted a citation model17 to
cluster into three domains (Fig. 3a)22. The subfield’s degree of self-referencing decreases their citation trajectories, predicting their
first and largest domain corresponds to as N–1 with the number of papers, N (Fig. 3b). average number of citations over the next
condensed matter and interdisciplinary For example, papers in plasma, nuclear and 20 years (Fig. 3d). These fits provide two
physics and related areas (IPR), where astrophysics, the smallest subfields, are 23 key parameters: the ultimate impact, c∞,
IPR corresponds to the subset of papers times more likely to reference themselves representing the total number of citations
pertaining to PACS code 80, distinct from than expected by chance. In contrast, a typical paper will collect during its
the interdisciplinary physics corpus defined condensed matter, IPR and general physics, lifetime, and impact time, T*, representing
in Box 1. The second domain incorporates the four largest subfields, cite themselves the characteristic time over which a
topics from electromagnetism to atomic and only about four times more than expected paper remains cited. As Fig. 3e indicates,
plasma physics, whereas the third domain by chance. This self-referential nature results papers published in electromagnetism and
contains papers on particle, nuclear and in systematic citation barriers between some interdisciplinary physics have the largest
astrophysics. The subfields within each subfields (Fig. 3c). Nuclear physics suffers ultimate impact, a typical paper in these
domain cite each other in a statistically the most from this, having up to seven times areas collecting more than 50 citations
significant fashion, and tend not to cite fewer citations than expected by chance from over its lifetime. They also have a much
subfields in other domains. These three the four largest subfields, and displaying a longer impact time than other subfields
domains are held together by general 23-fold self-referencing. General physics is (T* = 11.74 and 14 years, respectively),
physics, which is significantly co-cited by absent in Fig. 3c because it is not undercited indicating that high-impact papers require a
multiple subfields from each domain. by any other subfield, affirming its central long time to acquire their citations.
The loops in Fig. 3a indicate the degree role in Fig. 3a. On the other extreme of the spectrum
to which papers in each subfield cite papers To understand the long-term impact are papers published in particle and nuclear
in their own area, uncovering a remarkable of papers in each subfield, we selected physics that burn out after 6–7 years,
pattern: the smaller a subfield, the more all papers published in 2000 within each consequently collecting much fewer citations

794 NATURE PHYSICS | VOL 11 | OCTOBER 2015 | www.nature.com/naturephysics

© 2015 Macmillan Publishers Limited. All rights reserved

 perspective

a b 30
PACS 70 (808,000) Observed
Condensed matter: electronic Expected

Observed/expected self-references
structure, electrical, magnetic, 25
and optical properties 20×
12× 20
8×
PACS 80 (473,000) CM:EMO 4× 15
Interdisciplinary physics and
2× N–1
related areas of science and
technology 10
PACS 60 (676,000)
Condensed matter:
IPR CM:SMT structural, mechanical 5
and thermal properties
0
0 2 4 6 8
PACS 10 (284,000) Number of papers, N (= 105 )
PACS 40 (344,000)
The physics of
Electromagnetism, c
PACS 00 (538,000) elementary
optics, acoustics,
General physics particles and
heat transfer, classical
fields EOA
mechanics, and AM GPE
fluid dynamics HCF
GP
EOA
HCF EPF

GAA NP EPF

AM GPE GAA NP

CM:EMO CM:SMT IPR

PACS 30 (238,000) PACS 50 (106,000) PACS 90 (110,000) PACS 20 (154,000)

Atomic and Physics of gases, plasmas, Geophysics, astronomy, Nuclear physics
molecular physics and electric discharges and astrophysics

d e
3.0 60
EPF EOAHCF
55 IPR
2.5
50
2.0
Ultimate impact, cg

IPR 45
Yearly impact, c

GP
GAA CM:SMT
1.5 40 AM
EOAHCF
NP 35
1.0 CM:EMO EPF
GPE
30 CM:EMO
0.5
25
NP
20
0
0 5 10 15 20 5 6 7 8 9 10 11 12 13 14 15
Year T * (years)

Figure 3 | The anatomy of physics. We classified each physics paper from the core and interdisciplinary literature into one of the ten major Physics and
Astronomy Classification Scheme (PACS)-based subfields of physics, available after 1975. Since only 5% of these papers have PACS numbers, 3.7 million of them
do not show subfield-specific citation bias, hence could not be assigned a subfield by our algorithm. The remaining 1.9 million papers were successfully assigned
a subfield, allowing us to explore the inner structure of the physics literature. Note that 1.1 million papers have multiple PACS numbers, hence they may belong
to multiple subfields. a, The co-citation patterns between the different physics subfields. Node sizes are proportional to the number of papers published in each
subfield. Two subfields are connected if the number of citations between them significantly exceeds the expected citations, the line widths for all links, shown in
the key, correspond to how many times the observed reference exceeds the random expectation. We find that core physics naturally clusters into three domains,
as indicated by the colours of the clusters. General physics plays a central role, linking the three domains together. The loops on each node show the degree of
self-referencing of the corresponding field. b, The smaller a subfield, the more its reference list is biased towards papers in the same subfield. c, A link between
two subfields is evidence of citation barriers, where two areas have a significantly smaller number of citations than expected by chance. The highest citation
barriers, up to nine times fewer citations than expected by chance, are between GAA and CM:EMO, and between NP and IPR. d, The yearly citations, c, of papers
published in 2000 in selected subfields. Dotted curves indicate predictions. e, The ultimate impact, c∞, representing the total number of citations a typical paper
receives (that is, the area under the curves shown in panel d), versus impact time, T*, for papers published in 2000 in each subfield, representing the typical time
over which a paper collects its citations. The symbol size is proportional to the number of papers published in each subfield.

NATURE PHYSICS | VOL 11 | OCTOBER 2015 | www.nature.com/naturephysics 795

© 2015 Macmillan Publishers Limited. All rights reserved

perspective

during their lifetime. Their low impact brings significant impact penalties: papers catholique de Louvain, Louvain-la-Neuve 1348,
is partly explained by the insular nature in those areas burn out very fast and have Belgium. A.L.B. is also at the Center for Cancer
of these subfields: both fields are about much lower ultimate impact than their Systems Biology, Dana-Farber Cancer Institute, Boston,
sevenfold undercited by the largest domain peers in other subdisciplines. These patterns Massachusetts 02115, USA, and in the Department of
of physics (Fig. 3c), and show the highest hide important vulnerabilities: research in Medicine, Brigham and Women’s Hospital, Harvard
degree of self-referencing. Taken together, ecology and technology adoption indicates Medical School, Boston, Massachusetts 02115, USA,
Fig. 3 documents the rather heterogeneous that such monocultures, unable to link and at the Center for Network Science, Central
nature of physics, consisting of subfields to their environment, are often destined European University, Budapest 1051, Hungary.
with widely different impact, longevity, to extinction23. *e-mail: r.sinatra@neu.edu; alb@neu.edu
internal culture and ability to interact with Our understanding of the dialogue that
the rest of physics. physics shares with other disciplines is still References
in its infancy. Stumbling blocks include a 1. Kuhn, T. The Structure of Scientific Revolutions (Univ. Chicago
Press, 1996).
Where to from here? lack of accurate records for the publications 2. Barabasi, A-L. Nature Phys. 8, 14–16 (2011).
Many of us involved in hiring committees of each scientist, and the paucity of tools 3. Main, P. & Tracy, C. Phys. World 26 (4), 17–18 (2013).
or thesis defences in physics have been capable of mining the text of research 4. Donald, A. Phys. Biol. 11, 053008 (2014).
5. Lorenz, E. N. J. Atmos. Sci. 20, 130–141 (1963).
confronted with the question: can a papers. But given the rapid rate at which 6. Fortunato, S. Phys. Rep. 486, 75–174 (2010).
particular body of work be considered such tools are developed, we are advancing 7. Zhu, X. Semi-Supervised Learning Literature Survey Tech.
physics, or is a particular scientist a physicist? on an era in which hypotheses about the Rep. 1530 (Univ. Wisconsin-Madison, 2005).
8. Shen, H-W. & Barabasi, A-L. Proc. Natl Acad. Sci. USA
The futility of the debate is often unnerving, nature and evolution of a discipline or 111, 12325–12330 (2014).
and the analysis of the physics literature discovery will be testable in minute detail. 9. Hubbard, J. Proc. R. Soc. Lond. A 276, 238–257 (1963).
shows us why: there is not a single standard Such advances offer a chance to go beyond 10. Hopfield, J. J. Proc. Natl Acad. Sci. USA 79, 2554–2558 (1982).
11. Mehra, J. The Golden Age of Theoretical Physics (World
of what physics is. On one end there is the sweeping generalizations such as the theory Scientific, 2001).
considerable corpus of work published in of paradigms1, and ask deeper questions 12. Bornmann, L. & Mutz, R. J. Assoc. Inf. Sci. Technol. (2015).
physics journals that, by the virtue of their about the way physics changes, and how 13. Deville, P. et al. Sci. Rep. 4, 4770 (2014).
publication venue, tends to automatically get a certain discovery influences subsequent 14. Wuchty, S., Jones, B. & Uzzi, B. Science 316, 1036–1039 (2007).
15. Pavlidis, I., Petersen, A. M. & Semendeferi, I. Nature Phys.
a stamp of approval by most physicists. Yet, work18, within and outside of physics. Given 10, 700–702 (2014).
as we showed here, there exists a much larger that the most fruitful discoveries come from 16. Radicchi, F. & Castellano, C. Phys. Rev. E 83, 046116 (2011).
physics literature, that in its subject matter the cross-pollination of different fields — 17. Wang, D., Song, C. & Barabási, A-L. Science 342, 127–132 (2013).
18. Redner, S. Phys. Today 58 (6), 49–54 (2005).
and referencing patterns is indistinguishable recombining previously disconnected 19. Nakamoto, H. Synchronous and Diachronous Citation Distribution
from the core physics literature, yet it is ideas and resources24 — such quantitative (Elsevier, 1988).
published outside of physics journals. It understanding could help us identify as 20. Burnham, J. C. J. Am. Med. Assoc. 263, 1323–1329 (1990).
21. Spier, R. Trends Biotechnol. 20, 357–358 (2002).
contains many papers without which physics, yet unexplored research areas. A better 22. Pan, R. K., Sinha, S., Kaski, K. & Saramaki, J. Sci. Rep. 2, 551 (2012).
as we know it, could not flourish. understanding of the optimal paths of 23. Arthur, W. B. The Nature of Technology: What it is and How it
The analysis of this extended corpus innovation may also change the way we fund Evolves (Simon and Schuster, 2009).
24. Uzzi, B., Mukherjee, S., Stringer, M. & Jones, B. Science
offers a treasure trove of quantitative scientists and institutions to unlock their 342, 468–472 (2013).
information, unveiling the anatomy of the creative potential and enhance their long-
discipline. It demonstrates, for example, term impact. ❐ Acknowledgements
that our ability to define a field such as R.S., M.S. and A.L.B. are supported by the AFOSR grant
physics using sets of journals is long gone, Roberta Sinatra*, Pierre Deville, Michael Szell FA9550-15-1-0077. P.D. is supported by the National
Foundation for Scientific Research (FNRS) and by the
and exposes the tribal nature of the different and Albert-László Barabási* are at the Center Research Department of the Communaute Francaise de
subdisciplines of physics: the smaller the for Complex Network Research and Physics Belgique (Large Graph Concerted Research Action). D.W.
subfield, the more self-referential it becomes. Department, Northeastern University, Boston, is supported by the AFOSR, grant FA9550-15-1-0162. R.S.
Nuclear and particle physics stand out in Massachusetts 02115, USA. Dashun Wang is at is also supported by the James S. McDonnell Foundation.
this context: they are not only the most self- the College of Information Sciences and Technology, A.L.B. is also supported by the Future and Emerging
Technologies Project 317 532 ‘Multiplex’ financed by the
referential subfields of physics, but are also Pennsylvania State University, University Park, European Commission. We wish to thank I. Georgescu for
separated by significant citation barriers Pennsylvania 16801, USA. P.D is also in the many fruitful discussions and ideas throughout this project
from most other subfields. This isolation Department of Applied Mathematics, Université and J. De Nicolo for support with handling the WoS data.

796 NATURE PHYSICS | VOL 11 | OCTOBER 2015 | www.nature.com/naturephysics

© 2015 Macmillan Publishers Limited. All rights reserved

 perspective

Correction
In the Perspective ‘A century of physics’
(Nature Physics 11, 791–796; 2015), the text in
the legend in Fig. 3a was obscured. This error
has now been corrected online 5 October 2015.

NATURE PHYSICS | VOL 11 | OCTOBER 2015 | www.nature.com/naturephysics 797

View publication stats © 2015 Macmillan Publishers Limited. All rights reserved