An analytical approach to binary populations in globular clusters Journal: ApJ Cluster Monte Carlo ( K. J. Joshi et al. 2000 ; B. Pattabiraman et al. 2013 ; C. L. Rodriguez et al. 2022 ) , COSMIC ( K. Breivik et al. 2020 ) , Matplotlib ( J. D. Hunter 2007 ) , NumPy ( C. R. Harris et al. 2020 ) Christopher E. O’Connor, Kyle Kremer, Frederic A. Rasio Abstract Globular clusters (GCs) display much lower binary fractions than found among main-sequence stars in the solar neighborhood. The physical cause of this difference is debatable: does it reflect different star formation outcomes at low metallicity and/or high density, the dynamical processing of primordial binaries over cluster lifetimes, or a combination of the two? Starting from the assumption that the initial binary distribution (IBD) in GCs is the same as the binary distribution observed in the solar neighborhood, we show with straightforward analytical calculations that the dynamical dissolution of “soft” primordial binaries can fully explain the main-sequence binary fractions in present-day GCs. We validate our estimates against a detailed N N -body simulation with the Cluster Monte Carlo code. Adopting the view that the observed binary fraction in a given cluster constrains the location of the hard/soft boundary at birth, we infer that surviving Milky Way GCs had a similar distribution of birth radii to young massive clusters in the local universe. Our findings underscore the crucial role of stellar black holes (through “black hole burning”) in sculpting GC binary populations and reinforce the need for realistic initial conditions in theoretical modeling of GC dynamics. show]christopher.oconnor@northwestern.edu I Introduction Stellar multiplicity is the norm across most galactic environments. A major exception to this rule occurs among low-mass main-sequence stars and giants in old stellar systems such as globular clusters (GCs). Observational studies spanning several decades, and employing various techniques, have consistently found much lower binary fractions within well-studied GCs than in the Galactic field and halo, typically \sim 1\mbox{--}10\% in the former ( M. D. Albrow et al. 2001 ; A. Sollima et al. 2007 ; A. P. Milone et al. 2012 ; J. Ji & J. N. Bregman 2015 ; J. Müller-Horn et al. 2025 , e.g.,) versus \sim 50\% in the latter ( A. Duquennoy & M. Mayor 1991 ; D. Raghavan et al. 2010 ; M. Moe et al. 2019 ; S. S. R. Offner et al. 2023 , e.g.,) . A similar paucity of low-mass binaries may exist in GC stellar streams ( S. A. Usman et al. 2024 ; A. Phillips et al. 2026 ) . At face value, the intuitive explanation for this trend is that dynamical interactions within GCs reduce the total binary fraction from its primordial value ( P. Hut et al. 1992a ; N. Ivanova et al. 2005 ; J. M. Fregeau et al. 2009 ; N. W. C. Leigh et al. 2013 ; N. W. C. Leigh et al. 2015 ; L. Wang et al. 2022 ; A. Phillips et al. 2026 , e.g.,) . On the other hand, many early studies were interested in how to form long-lived binaries in GCs, especially to explain the presence of low-mass X-ray binaries and millisecond pulsars in these environments ( A. C. Fabian et al. 1975 ; S. L. W. McMillan et al. 1987 ; P. Hut et al. 1992a , e.g.,) . Clearly, demonstrating end-to-end consistency for any version of binary evolution in GCs is a tall order. Several aspects of the problem call for re-examination in light of recent developments. The gold standard for testing theories of long-term cluster evolution is set by comparison between direct-summation N -body simulations and observations of Milky Way GCs ( H. Baumgardt & M. Hilker 2018 , e.g.,) . However, full-lifetime direct integrations of dense clusters with large primordial binary populations have only recently become computationally feasible (e.g., L. Wang et al. 2020 ; L. Wang et al. 2022 ; A. W. H. Kamlah et al. 2022 ; see R. Spurzem & A. Kamlah 2023 for a recent review of direct N -body methods), and direct integrations of clusters with \gtrsim 10^{6} stars (such as typical GCs) remain costly and rare ( K. Wu et al. 2025 ) . Monte Carlo methods can be deployed to study in-cluster binary evolution in higher-mass GCs ( J. M. Fregeau et al. 2009 ; N. W. C. Leigh et al. 2015 ; J. Hong et al. 2015 ; J. Hong et al. 2017 ; J. Hong et al. 2018 ; E. González Prieto et al. 2021 ; E. González Prieto et al. 2022 ; E. González Prieto et al. 2024 ; F. Kıroğlu et al. 2025 , e.g.,) . Even so, observations of stellar binaries in the nearest analogs of GC progenitors – young massive clusters (YMCs) and super star clusters (SSCs) – are sensitive only to the most massive systems ( H. Sana et al. 2012 ; H. Sana et al. 2013 ; H. Sana et al. 2025 ; J. S. Clark et al. 2020 ; J. S. Clark et al. 2023 ; B. W. Ritchie et al. 2022 , e.g.,) . Options for direct observational studies of intermediate- and low-mass binaries in high-density environments are scant. Additionally, the demographics of primordial binaries as a function of stellar mass is not known a priori. Like the stellar initial mass function (IMF), the initial binary distribution (IBD) might depend in a complicated manner on the physics of star formation in an environment of a given density and metallicity ( M. Y. Grudić et al. 2021 ; D. Guszejnov et al. 2022 ; D. Guszejnov et al. 2023 ; C. D. Matzner 2024 , e.g.,) . Binary interactions during the pre-MS phase may likewise complicate definitions of “primordial”/“initial” binary properties ( P. Kroupa 1995 , e.g.,) . Only recently has there been sufficient data on binary demographics at all separations to provide a realistic empirical template for the IBD of the Galactic field population ( M. Moe & R. Di Stefano 2017 ; M. Moe et al. 2019 ; K. El-Badry & H.-W. Rix 2019 ; H.-C. Hwang et al. 2021 ; S. S. R. Offner et al. 2023 , e.g.,) . Faced with these uncertainties, there is value in the old-fashioned exercise of postulating a universal IBD and then assessing the extent to which observations require variations between different environments ( N. W. C. Leigh et al. 2015 ; D. Belloni et al. 2017 , e.g.,) . This channels the spirit in which earlier works have interrogated the hypothesis of a universal stellar IMF ( P. Kroupa 2001 ; G. Chabrier 2003 , e.g.,) and, in some instances, inferred variations of the IMF in certain environments ( H. Bartko et al. 2010 ; J. R. Lu et al. 2013 ; T. Jeřábková et al. 2017 ; M. W. Hosek et al. 2019 ; P. van Dokkum & C. Conroy 2024 , e.g.,) . In this study, we advance the case for a universal IBD between the solar neighborhood and Galactic GCs. In Section II , we summarize pertinent observational constraints on field and cluster binary populations from the literature. In Section III , we present a concise analytical formalism to account for the effects of cluster dynamics on primordial binaries; we predict the correct range of observed binary fractions among low-mass stars in well-studied GCs by invoking the dissolution of dynamically “soft” binaries – those with orbital velocities below the cluster’s velocity dispersion at birth. In Section IV , we quantify the rate and energy cost of soft binary dissolution with simple analytical estimates, confirm the overall picture with a detailed N -body simulation, and infer birth conditions for a set of Milky Way GCs based on our results. In Section V , we discuss how our results can inform future theoretical and observational investigations of binaries in dense clusters, as well as how they relate to ideas developed in previous works. In Section VI , we summarize our takeaway messages. II Pertinent Observations Here we give a brief synopsis of current knowledge about observed binary populations in GCs and the field of the solar neighborhood, and use these data to formulate a fiducial universal IBD for the Milky Way. Early searches for stellar binaries in GCs using traditional spectroscopic or photometric monitoring were rather limited in the scope, but nonetheless established that binaries are much rarer in GCs than in the field ( M. D. Albrow et al. 2001 , e.g.,) . In recent years, the most widely employed method of searching for binaries in a GC is to measure the broadening of the cluster’s main sequence on a color–magnitude diagram ( A. Sollima et al. 2007 ; A. P. Milone et al. 2012 ; J. Ji & J. N. Bregman 2015 ; J. Müller-Horn et al. 2025 , e.g.,) . This technique is sensitive to the frequency of unresolved, near-equal-mass binaries (mass ratios q\gtrsim 0.5 ), but is subject to contamination in crowded fields near the center of a cluster and to modeling uncertainties related to how one corrects for the low- q binaries to which it is not sensitive. Bearing these limitations in mind, the photometric-broadening method has yielded consistent estimates for well-studied GCs over the past few decades, with total binary fractions of \sim 1\mbox{--}10\% estimated near the half-mass or half-light radius ( A. P. Milone et al. 2012 ; J. Ji & J. N. Bregman 2015 ; J. Müller-Horn et al. 2025 , e.g.,) . A GC’s binary fraction tends to increase towards its center, typically by a factor of \sim 2\mbox{--}5 above the half-mass value ( A. Sollima et al. 2007 ; A. P. Milone et al. 2012 ; J. Ji & J. N. Bregman 2015 , e.g.,) . This is the expected result of mass segregation via two-body relaxation ( P. Hut et al. 1992a ; D. Heggie & P. Hut 2003 ) . The minority of GCs for which larger total binary fractions ( \sim 15\mbox{--}30\% ) are measured stand out as unusually diffuse, which will prove consistent with the calculations of Section III . A few other lines of evidence have opened up in recent years, thanks to Gaia . Spectroscopic observations with Gaia ’s Radial Velocity Spectrometer have yielded independent estimates of binary fractions for a subset of nearby GCs, which largely agree with the photometrically derived values ( D. Bashi & V. Belokurov 2025 ) . Meanwhile, Gaia astrometry has facilitated the identification of GC stellar streams, whose binary fractions also appear to be low ( S. A. Usman et al. 2024 ) . The binary population in the solar neighborhood, including nearby resolved star-forming regions, is much better understood. The fraction of primary stars with at least one bound stellar companion, or “multiplicity fraction” (MF) rises monotonically from \sim 20\% at the lowest masses to virtually 100\% at the highest (see S. S. R. Offner et al. 2023 and references therein). We adopt the bias-corrected MF as compiled in Table 1 of S. S. R. Offner et al. 2023 for primary stellar masses of 0.075\mbox{--}50M_{\odot} ; these data are described well by the following sigmoid-like fitting formula: F_{\rm b,tot}(m_{1})=F_{\rm lo}+\frac{F_{\rm hi}-F_{\rm lo}}{1+(m_{1}/m_{\rm mid})^{-\beta}}, (1) with best-fitting parameters F_{\rm lo}=0.18 , F_{\rm hi}=0.99 , m_{\rm mid}=1.6M_{\odot} , and \beta=3.1 . We use this function as an Ansatz for the total primordial binary fraction as a function of stellar-mass. The second step in building the IBD is to decide the separation (i.e., period) distribution of primordial binaries. This determines the predicted proportions of hard and soft binaries in a given cluster. The observed separation distribution of solar-neighborhood binaries varies with mass (and also with the occurrence of triple and higher-order multiple systems, which we do not take into account). Low-mass binaries follow a roughly log-normal separation distribution peaked at \sim 10\mbox{--}30\,\mathrm{AU} and with a dispersion of \sim 1\mbox{--}1.5\,{\rm dex} ( A. Duquennoy & M. Mayor 1991 ; D. Raghavan et al. 2010 , e.g.,) . Companions to high-mass stars are closer to being flat in the log of separation overall ( H. Sana et al. 2012 ; S. S. R. Offner et al. 2023 , e.g.,) , though there is also an overabundance of close companions due to the prevalence of triple systems in observed massive-star samples. We consider both log-flat and log-normal separation distributions in our calculations in Section III , finding that the result depends moderately on the form of the distribution for high-mass systems but only weakly for low-mass. Observational input on the mass-ratio and eccentricity distributions in multi-star systems is also required to fully realize a model binary population. Both distributions have been characterized in some detail as functions of primary mass and separation ( M. Moe & R. Di Stefano 2017 , e.g.,) . In a dense cluster, strong dynamical interactions efficiently thermalize binary eccentricities, and exchange interactions skew the mass-ratio distribution toward near-equal-mass systems ( D. C. Heggie 1975 ; P. Hut et al. 1992a , e.g.,) . Only in a minority of very close and/or very massive binaries, where tidal interactions operate on a timescale that competes with the rate of strong dynamical encounters, might the initial mass-ratio or eccentricity distribution make an observationally relevant impact on the population. For purposes of this work, it is sufficient to consider only the total binary fraction and separation distribution as functions of primary stellar mass. III Analytical formalism III.1 Preliminaries Consider a young star cluster immediately after the expulsion of its residual gas, modeled as a Plummer sphere of mass M_{\rm cl} and scale length b . The cluster’s gravitational potential \Phi , mass density \rho , and enclosed mass M are given as functions of radius r by \displaystyle\Phi(r) \displaystyle=-\frac{GM_{\rm cl}}{\sqrt{r^{2}+b^{2}}}, (2a) \displaystyle\rho(r) \displaystyle=\frac{3M_{\rm cl}}{4\pi b^{3}}\left(1+\frac{r^{2}}{b^{2}}\right)^{-5/2}, (2b) \displaystyle M(r) \displaystyle=M_{\rm cl}\left(1+\frac{b^{2}}{r^{2}}\right)^{-3/2}. (2c) The Plummer length b is related to two other convenient reference lengths: the half-mass radius r_{\rm h}\equiv[2^{2/3}-1]^{-1/2}b\approx 1.3b and the virial radius r_{\rm v}\equiv 16b/(3\pi)\approx 1.7b . The 3D velocity dispersion is given by \sigma^{2}(r)=-\frac{\Phi(r)}{6}. (2d) Unless otherwise specified, we evaluate the cluster’s properties at r=r_{\rm h} and use r_{\rm v} as our default reference length. Let the cluster contain a population of primordial binaries whose semi-major axes a are distributed over many orders of magnitude. We choose physically motivated cutoffs for this distribution as follows. Let the lower cutoff be given by the binary’s near-contact limit during the pre-MS stage, approximated as: a_{\rm min}\approx 5(R_{1}+R_{2}), (3) where R_{1,2} are the component radii on the zero-age main sequence. We adopt the following approximate ZAMS mass–radius relation ( C. A. Tout et al. 1996 , cf.) : \frac{R(m)}{R_{\odot}}=\left\{\begin{array}[]{ll}(m/M_{\odot})^{0.9},&m\leq M_{\odot};\\ (m/M_{\odot})^{0.6},&m\geq M_{\odot}.\end{array}\right. (4) Meanwhile, let the upper cutoff be determined by the tidal (Hill) radius of a circular binary within the cluster potential: a_{\rm max}\sim\left(\frac{m_{\rm b}}{\rho(r)}\right)^{1/3}=b\left(\frac{4\pi m_{\rm b}}{3M_{\rm cl}}\right)^{1/3}\left(1+\frac{r^{2}}{b^{2}}\right)^{5/6} (5) where m_{\rm b}=m_{1}+m_{2} is the mass of a given binary. At r=r_{\rm h} , this yields \displaystyle a_{\rm max} \displaystyle=2.17r_{\rm v}\left(\frac{m_{\rm b}}{M_{\rm cl}}\right)^{1/3} (6a) \displaystyle=5600\,\mathrm{AU}\left(\frac{r_{\rm v}}{1\,\mathrm{pc}}\right)\left(\frac{m_{\rm b}}{M_{\odot}}\right)^{1/3}\left(\frac{M_{\rm cl}}{10^{5}M_{\odot}}\right)^{-1/3}. (6b) Binaries with separations \gtrsim a_{\rm max} would dissolve within an orbital period simply as a result of the cluster’s tidal field, even if stellar encounters were somehow suppressed. For low-mass stars ( m_{\rm b}\lesssim M_{\odot} ), a_{\rm max} is of the order of the average interparticle distance within a region of locally uniform density. As a binary moves through the cluster, it undergoes numerous weak scattering encounters with background stars, as well as occasional strong encounters. The cumulative effect of these encounters is determined by the binary’s hardness, the ratio of its characteristic orbital velocity to the background velocity dispersion: \eta=\frac{G(m_{1}+m_{2})}{\sigma^{2}a}. (7) Binaries with \eta>1 and \eta<1 are called hard and soft, respectively. 1 1 1 Traditionally, the hardness of a binary is defined as \tilde{\eta}=Gm_{1}m_{2}/(\bar{m}\sigma^{2}a) , where \bar{m} is the average mass of a background star. This definition amounts to a comparison of the dynamical energy scales of the binary and third body, rather than their velocities . Numerical scattering experiments show that the velocity-based definition (Eq. 7 , also called the fast/slow boundary) better predicts the statistics of binary–single scattering events with arbitrary mass ratios. Within a Plummer sphere, the hard/soft boundary is given as a function of radius by setting \eta=1 : a_{\rm hs}(r)=\frac{Gm_{\rm b}}{\sigma^{2}(r)}=6(r^{2}+b^{2})^{1/2}\frac{m_{\rm b}}{M_{\rm cl}}. (8) At the half-mass radius, we have \displaystyle a_{\rm hs} \displaystyle=5.8r_{\rm v}\frac{m_{\rm b}}{M_{\rm cl}} (9a) \displaystyle\approx 12\,\mathrm{AU}\left(\frac{r_{\rm v}}{1\,\mathrm{pc}}\right)\left(\frac{m_{\rm b}}{M_{\odot}}\right)\left(\frac{M_{\rm cl}}{10^{5}M_{\odot}}\right)^{-1}. (9b) This expression readily generalizes to arbitrary spherical cluster potentials; the numerical coefficient varies only by a factor of a few as long as we do not stray too far into the cluster’s outer reaches. It is well understood that, on average, hard binaries harden (i.e., become more tightly bound) as a result of encounters, while soft binaries soften and eventually dissolve ( D. C. Heggie 1975 ) . The remainder of this work follows from the realization that, all else being equal, a_{\rm hs} is directly proportional to the binary mass scale m_{\rm b} (which is of the same order as the primary mass m_{1} ) and inversely proportional to the velocity dispsersion \sigma^{2}\propto M_{\rm cl}/r_{\rm v} . This implies (i) that in a given cluster, massive binaries ( m_{\rm b}\gg M_{\odot} ) are dynamically hard over a much wider range of separations than low-mass binaries ( m_{\rm b}\lesssim M_{\odot} ), and likewise (ii) that less compact clusters host more hard binaries of a given mass. Figure 1: Map of the primary mass–semimajor axis parameter space for binary systems embedded in fiducial Plummer-sphere clusters. Solid lines show the hard/soft boundary as a function of m_{1} in each cluster for binaries with mass ratio q=0.5 ; the color of each line indicates the cluster parameters described in the text: blue for model A, amber for B, green for C, and red for D. Dashed lines of the corresponding colors shows a_{\rm max} in the same clusters. The dotted black curve shows a_{\rm min} . Binaries cannot exist in the shaded regions under our assumptions. Figure 1 illustrates this point by showing a_{\rm min} , a_{\rm hs} , and a_{\rm max} as functions of primary mass m_{1} for a sequence of binaries with mass ratio q=m_{2}/m_{1}=1/2 . We consider four fiducial cluster models with masses M_{\rm cl}=(10,3,1,0.3)\times 10^{5}M_{\odot} and r_{\rm v}=(0.5,1.0,2.0,4.0)\,\mathrm{pc} , respectively. For brevity, we refer to these as models ‘A’ through ‘D’ from greatest density to least. Spanning factors of \approx 300 in compactness and \approx 2\times 10^{4} in density, these models are intended to be representative for the range of typical GC progenitors in that detailed model grids covering a similar range of M_{\rm cl} and r_{\rm v} result in a similar present-day properties to the Milky Way GCs ( J. Hong et al. 2018 ; K. Kremer et al. 2020 ; A. Hypki et al. 2025 ; M. Giersz et al. 2025 , e.g.,) . The lower cutoff a_{\rm min} is determined by the ZAMS mass–radius relation and is independent of host cluster properties. Plotted in Fig. 1 as a dotted black curve, a_{\rm min} ranges from 0.005\,\mathrm{AU} at the H-burning limit to 0.8\,\mathrm{AU} at 150M_{\odot} . The hard/soft boundary a_{\rm hs} (solid curves) and the tidal radius a_{\rm max} (dotted curves) depend on both binary and cluster properties, but to differing degrees. At a fixed m_{1} , a_{\rm hs} spans a factor of \approx 300 across the four models in inverse proportion to velocity dispersion (Eq. 8 ), while the tidal radius a_{\rm max} spans a factor of \approx 25 due to the scaling with mean cluster density (Eq. 5 ). For fixed cluster properties, a_{\rm hs}\propto m_{\rm b}\sim m_{1} increases by a factor of 150/0.08\approx 1800 between the least and most massive MS stars; a_{\rm max} scales as m_{\rm b}^{1/3}\sim m_{1}^{1/3} and thus varies by a comparatively modest factor of (150/0.08)^{1/3}\approx 12 . We conclude that the relative proportions of hard and soft binaries, both within a single cluster and across a diverse population of dense clusters, is driven predominantly by the location of the hard/soft boundary; a_{\rm min} and a_{\rm max} are comparatively consistent across different environments. III.2 Hard and soft binary fractions Figure 2: Binary fraction as a function of primary mass. The black points are the measured MF among MS stars in the solar neighborhood as compiled in Table 1 of S. S. R. Offner et al. 2023 , with the mass bins and quoted uncertainties shown as horizontal and vertical error bars. The black curve is given by Eq. 1 . Dashed colored curves show the predicted primordial hard binary fractions at the half-mass radii of our three fiducial Plummer-sphere clusters, using the same color scheme as in Fig. 1 . The shaded regions highlight areas of observational relevance: light blue shows the range of MS stellar masses and observed binary fractions in GCs, while grey shows the approximate mass range for BH progenitor stars. Figure 3: Like Fig. 2 , but for a log-normal distribution of semi-major axes with median \mu_{a}=30\,\mathrm{AU} and dispersion \sigma_{a}=1.5\,{\rm dex} . Given a model for the primordial binary distribution function and the cluster’s initial properties, one can compute the hard and soft binary fractions as a function of m_{\rm b}\sim m_{1} . We now enact a simplified example of such a calculation. We consider only single stars and binaries as possible configurations, neglecting triples and higher-order multiplicities. For a star of mass m_{1} , let the probability of having a companion with semi-major axis between a and a+\mathrm{d}a , and whose mass ratio is between q and q+\mathrm{d}q , be \mathrm{d}F_{\rm b}=f_{\rm b}(a,q;m_{1})\,\mathrm{d}a\,\mathrm{d}q, (10) where f_{\rm b} is the IBD. The binary fraction in any finite region of parameter space is then given by the integral of equation ( 10 ) over a and q . By equation (2), the hard binary fraction is determined as a function of primary mass by F_{\rm b,hard}(m_{1})=\int_{0}^{1}\int_{a_{\rm min}(m_{1},q)}^{a_{\rm hs}(m_{1},q)}f_{\rm b}(a,q;m_{1})\,\mathrm{d}a\,\mathrm{d}q (11) where we set m_{\rm b}=m_{1}(1+q) when evaluating a_{\rm hs} . The double integral can be evaluated numerically without difficulty for general f_{\rm b} . The soft binary fraction is given by the analogous integral over [a_{\rm hs},a_{\rm max}] . Consider now the special case where the mass-ratio dependency is separable and independent of mass, i.e. f_{\rm b}=f_{\rm b}(a;m_{1})\xi_{q}(q) with \xi_{q}(q) a probability density function supported over 0