Chemical fingerprints of binary mass transfer in massive stars

The majority of massive stars are born in close binary systems. As stars expand when they age, mass transfer or even a merger with their companion is inevitable. However, most binary interaction products appear as single stars, such that the main evidence of their exciting past is lost. In a comprehensive grid of detailed massive binary evolution models we find systematic trends in chemical surface abundances that allow identifying the past mass gainers. We develop an analytic framework which is independent of specific evolutionary models, to constrain the amount and composition of the accreted material from their observed surface abundances. This yields tight constraints on the uncertain mass transfer physics in massive binary stars and allows us to reconstruct the past evolutionary history of the progenitor binary system. This method, which is shown to also constrain binary mergers (for example, SN 1987A), is applied to some of the best-studied OB stars so far. For γ Columbae, suggested to be an envelope-stripped star, we show that it is a mass gainer instead, whose companion star likely formed a stripped-envelope supernova. Our results highlight surface abundance measurements as a powerful tool to improve our understanding of massive binary systems evolving towards supernovae and compact object binaries.

The fraction of unevolved massive stars with a stellar companion so close that mass exchange becomes inevitable is ∼70%, independent of the heavy element fraction of the star forming environment. With less than ∼0.1% of their lifetime, the main episode of mass exchange is typically so fast that it is very rarely caught by observations. At the same time, it takes far too long for hydrodynamic models to assess it entirely. Therefore, the stability of the mass transfer process, as well as its efficiency (the fraction of mass that will be accreted by the companion star), remain weakly constrained so far, e.g. through population models of post-interaction stars. Binary mass transfer changes the evolution and fate of the involved stars drastically, even often leading to a merger of both components, or changing whether and how the stars explode, and what final remnants they produce. This affects our understanding of the chemical, mechanical, and radiative feedback of massive stars, which shapes the evolution of star-forming galaxies across cosmic time.

Empirical constraints for the mass transfer process have been worked out for individual cases where the binary nature of the post-mass transfer system could be identified. However, this concerns small samples, which do not allow a comprehensive assessment of the dependence of the mass transfer physics on the binary parameters. Observationally, the vast majority of mass gainers and stellar merger products are not recognized as such — they mostly appear single because they have lost their companion, merged with it, or the companion is faint and of low mass. Up to ∼30% of all core hydrogen burning and ∼70% of core helium burning stars are expected to have a binary interaction history. At the same time, a large fraction of massive stars are found to show thermonuclear processed matter at their surface, which provides a large reservoir of information. However, so far it has proven difficult to distinguish between the different possible sources of enrichment, which are stellar wind mass loss, envelope mixing in single stars, and binary interaction.

Results from detailed binary evolution models: It has recently become possible to compute comprehensive grids of massive binary evolution models, which densely cover the whole initial binary parameter space, and include a detailed treatment of internal mixing and nucleosynthesis. Here, we employ such a grid computed with mass and angular momentum transfer, differential rotation, and tides using MESA. The models were computed with an extended nuclear network through hydrogen burning which follows the time evolution of all stable CNO isotopes, which is necessary to accurately compute deviations from CN- and CNO-equilibrium. We scan through the parameter space of the mass gainer models in search for distinctive surface abundance patterns of binary interaction products. We focus on the elemental abundances altered by the CNO-cycle, i.e. helium, carbon, nitrogen, and oxygen. Heavier elements (e.g. sodium) and isotopic ratios are also affected but are harder to observe. We focus on the mass gainers that experience accretion during their main sequence stage, become the brighter of the two stars, and still live for a long time. The binary model grid does not include mergers, but we discuss their expected surface abundances based on simplified models below.

Figure 1 shows the CNO surface abundances of the mass gainer models during core helium burning, which reflects their enrichment at core hydrogen exhaustion, together with those of mass donors and single stars. Since the CN- and CNO-cycles conserve the number of CNO nuclei, our models cannot populate the region outside of the area framed by the two equilibrium lines, unless primary CNO products caused by α-captures appear at the surface when helium burning products are revealed. Surprisingly, Fig. 1 also shows that most of the allowed area remains empty, as the models populate only two branches, one close to the “CNO-eq. + dilution”-line and the other close to the “CN-eq. + dilution”-line.

Finding models on the lower of the two branches is expected, since envelope stripping, by winds or by mass transfer, reveals matter in the former convective core of the stripped stars, for which the high central temperature of the core ensures CNO-equilibrium. This branch is therefore populated by mass donors of any initial mass, and by the mass gainers and single stars that are so massive that they can self-strip their envelopes by winds and evolve to become hydrogen-poor, so called Wolf-Rayet stars.

Strikingly, the branch near the “CN-eq. + dilution”-line is exclusively populated by mass gainers and is avoided by mass donors and single stars. This is surprising at first because, next to pristine matter from the donor’s envelopes, the mass gainers accrete matter containing CNO-equilibrium abundances from the mass donors. However, this can be understood as follows. At low enough temperatures, CN-equilibrium may be reached while CNO-equilibrium is not, as in the core of our Sun (∼15 MK). In massive core-hydrogen burning stars, CNO-equilibrium is quickly obtained in their cores (∼30... 40 MK). In addition, in a small region just above the convective core, temperatures resemble those in the Solar center (∼15... 17 MK) and allow for the conversion of carbon to nitrogen via the CN cycle, but CNO-equilibrium can not be achieved. In single stars, this usually remains without observable consequences.

In our mass gainers, matter in CNO-equilibrium with and without helium enrichment is added to the envelope, besides pristine matter. The most enriched matter is added last, and the surface helium- and nitrogen enrichment is very high at this time. The helium enrichment of the outer layers imposes fast thermohaline mixing which evenly distributes all chemical components in the mass gainer’s envelope on the thermal timescale (“fast mixing”; τth ≃ 104 yr). The timescale of this fast mixing is much shorter than the timescale of CN-processing at the bottom of the envelope (τCN ≃ 0.5 Myr), such that the envelope is not enriched in CN-processed matter at this stage. The star will therefore be close to the “CNO-eq. + dilution”-line in Fig. 1 at this time. After accretion and fast envelope mixing, thermohaline mixing persists during the further nuclear timescale evolution, but as the inverted chemical composition profile is nearly smoothed out, the mixing timescale eventually exceeds the timescale of CN-processing at the bottom of the envelope (“slow mixing”). Assisted by rotational mixing in the spun-up envelope, this slow mixing brings unprocessed carbon into the CN-cycling layer at the bottom of the envelope, and CN-processed matter towards the surface, where the N/C-ratio increases. If CN-processing and mixing were efficient enough, the star could reach the top black line in Fig. 1, but not higher N/C-values.

The described slow mixing elevates many of our mass gainer models in Fig. 1 significantly above the “CNO-eq. + dilution”-line, while their envelope mixing remains incomplete. Their evolution in Fig. 1 during this phase can be well approximated by the dashed black lines. They start on the “CNO-eq. + dilution”-line at a point which reflects the amount of accreted CNO-equilibrium matter, and then follow the analytic prescription for converting the available carbon to nitrogen towards CN-equilibrium. For an observed star located above the “CNO-eq.+dilution” line, the black dashed lines can trace its evolution back to its foot point on the “CNO-eq.+dilution” line, to determine the accreted fraction of CNO-equilibrium material.

Notably, the efficiency of thermohaline mixing in massive stars has so far not been calibrated empirically, nor has its interplay with rotational mixing been fully understood, such that the elevation of our gainer models above the “CNO-eq. + dilution”-line must be considered uncertain. While the described mechanism has been discussed before, it has so far not been evaluated in a population of binary models, nor was its unique CNO-signature worked out. Figure 1 indicates that the CNO signature can be used as a fingerprint to identify past mass gainers observationally. Below we show that it allows to quantitatively constraint their past accretion history, independent of uncertainties in the efficiency of the slow mixing in the envelope of stellar models.

Analytic framework: We set up a simple, analytic model to exploit the quantitative information contained in the CNO diagram. We consider mass transfer to occur when the initially more massive star (primary) attempts to expand right after core hydrogen exhaustion (Case B). At this stage, the companion (secondary) is burning hydrogen in its core, and the interiors of both stars can be divided into a helium-rich core, a near pristine hydrogen-rich envelope (with a small CN-processing region at the bottom), and in-between a layer containing the hydrogen/helium gradient left by the receding convective core in the course of hydrogen burning. Notably, the composition in the inner two regions is in CNO-equilibrium in both stars.

We then assume that the mass gainer accretes parts of the pristine envelope and the H/He-gradient layer of the mass donor, which is instantly mixed with the gainer’s envelope. This alone would place the mass gainers on the “CNO-eq. + dilution”-line in Fig. 1, with their location set by the fraction of the H/He gradient layer incorporated into the envelope (fCNO). In the final step, we assume slow mixing of the envelope and an increase of the N/C surface ratio due to conversion of the available carbon to nitrogen by the CN-cycle, along a line parallel to the dashed lines in Fig. 1. The efficiency of this last process determines the final N/C-ratio. As we explain in Methods, all steps can be worked out analytically.

Before applying this model to an observed star, it is interesting to consider the location of the stellar models in the helium-nitrogen (HeN) diagram to unlock the potential of helium surface abundance measurements. Figure 3 shows that, also here, the mass gainers are found in a distinct sub-space of the diagram, although rotating single star models in which core material is mixed efficiently into the envelope may populate a similar area. The strength of this diagram is that it allows to determine the average helium abundance in the fraction of the H/He-gradient layer of the donor that is accreted by the mass gainer (YCNO). However, since surface nitrogen is enhanced not only by the accretion of CNO-equilibrium matter, but also by subsequent CN-cycling and slow mixing in the mass gainer’s envelope, only upper limits on fCNO can be derived from this diagram alone. Only in combination with the CNO diagram, the exact amounts can be determined, and the ambiguity with single star models is lifted.

We emphasize that our simple model is motivated by the findings from our numerical binary evolution models, but the inferred amounts of accreted CNO-equilibrium material and helium are determined independently of those models. The physics which sets the accretion efficiency in our numerical MESA models had to be fixed before the binary model grid was computed. Accordingly, the accretion efficiency derived for an obtained progenitor configuration of observed accretors via our analytic framework generally differs from the accretion efficiency of the corresponding binary evolution models in the binary model grid. These differences will be useful as empirical benchmarks for future numerical binary evolution models.

With three measured quantities, we cannot directly invert the equations, as we have five unknowns (the two initial masses, the masses of pristine and CNO-cycled accreted matter, and the average helium abundance in the latter). However, since different initial masses produce different amounts of CNO-processed material in the H/He gradient layer, this provides additional constraints on the initial mass of the donor star. Similarly, the current mass of the gainer, combined with the amount of accreted CNO-equilibrium material, constrains the initial mass of the mass gainer. Further constraints arise from considering the timing of the donor’s supernova explosion, and the requirement of stable mass transfer, which depends on the initial mass ratios of the binary.

We test our simple model by applying it to one of our post-mass transfer MESA models that initially consisted of 22.4 M⊙ + 7.8 M⊙ stars with an initial orbital period of 10 days. This model leads to an early B-type Mock star which is characterized by relatively high surface abundance ratios, with N/O≈1 and N/C≈4, after Case B mass transfer. Our method successfully reproduces the key properties of the accreted material such as Macc,CNO, and the initial binary configuration.

Application to observed stars: To demonstrate the benefits of our method, we use it for some of the best studied OB stars, which were found to exhibit significant surface abundance enrichment. A well observed case is the naked eye star γ Columbae of spectral type B2.5IV, an apparent single star of ∼6.0 M⊙. In a recent spectral analysis, a helium surface mass fraction of 0.38 ± 0.06 and a nitrogen enhancement factor of ∼7 were found. Interpreting it as a former binary mass gainer, we find that, based on its CNO-fingerprint, its envelope material displays nearly undiluted CN-equilibrium abundances (Fig. 1). With a current envelope mass of 4.4 M⊙, its N/O-value indicates that about 0.8 M⊙ (fCNO ∼17%) of matter containing CNO-equilibrium matter had been accreted. Shifting the observational point for γ Columbae down to the corresponding fCNO-value of 17% reveals that the average helium mass fraction in this matter was ∼90% (Fig. 3).

The initial parameters of the progenitor binary of γ Columbae can now be constrained as follows. With an evolutionary mass of 6.0 M⊙ and having accreted at least 0.8 M⊙, the initial mass of γ Columbae must have been smaller than 5.2 M⊙ (see Methods for an additional constraint giving < 5.0 M⊙). The mass donor must have contained a region of 0.8 M⊙ in its H/He-gradient region with an average helium mass fraction of 90% prior to mass transfer. Investigating single star models at core hydrogen exhaustion reveals that this excludes initial donor masses below ∼14 M⊙. This implies that the initial mass ratio of the binary was below 0.35, and that the mass transfer was highly non-conservative.

The initial binary parameter space for γ Columbae can be further constrained since at a given mass accretion rate, mass transfer becomes unstable for too small initial accretor masses, as the accretor would swell and the binary would merge. This process has recently been extensively studied. Using their results allows us to check for each point in the initial masses-diagram — for which we can compute the corr…

Improvements to AI Systems:

1. Physics-Informed Abundance Classification Model

• Train a classifier on the CNO-diagram branches (CN-eq. + dilution vs. CNO-eq. + dilution) to automatically distinguish mass gainers, mass donors, single stars, and merger products from observed surface abundances.

• The improved AI can label stars in large spectroscopic surveys (e.g., Gaia, LAMOST) without needing full binary evolution modeling, enabling rapid population-level identification of binary interaction products.

2. Analytic Inversion Engine for Binary Parameters

• Implement the paper’s analytic framework as a differentiable module that takes observed (N/C, N/O, He/H) and stellar mass as inputs, and outputs posterior distributions for initial masses, accreted mass fractions (fCNO), and mass transfer efficiency.

• The improved AI can reconstruct the progenitor binary configuration for any enriched OB star, providing empirical constraints on mass transfer physics across a wide parameter space.

3. Uncertainty-Aware Merger vs. Mass Gainer Discriminator

• Build a Bayesian model that uses the location in the CNO and HeN diagrams to separate stable mass transfer products (which lie above the CNO-eq. line due to slow mixing) from merger products (which lie on the CNO-eq. line with high helium).

• The improved AI can reclassify ambiguous objects like SN 1987A’s progenitor, and predict whether a given enriched star is a merger remnant or a mass gainer, even when only photospheric abundances are available.

4. Forward-Model Surrogate for Binary Evolution Grids

• Train a neural network surrogate on the MESA binary grid to map initial binary parameters (masses, period, metallicity) to final surface abundance patterns and remnant properties.

• The improved AI can instantly explore the full initial parameter space, replacing expensive hydrodynamic simulations, and enable real-time fitting of observed stars to constrain accretion efficiency and thermohaline mixing efficiency.

5. Self-Consistent Mass Transfer Efficiency Estimator

• Develop a regression model that, given the derived fCNO and helium enrichment, predicts the mass accretion efficiency (β) and stability boundary for Case B mass transfer, using the paper’s constraints on donor mass and accretor swelling.

• The improved AI can provide empirical β values for many systems, directly feeding into population synthesis codes for gravitational wave progenitors and supernova rate predictions.

6. Automated Survey Pipeline for Binary Interaction Products

• Create an end-to-end pipeline that ingests stellar spectra, extracts CNO abundances, applies the analytic inversion, and outputs a catalog of likely mass gainers, mergers, and their progenitor properties.

• The improved AI can process thousands of stars from upcoming surveys (e.g., 4MOST, WEAVE), producing a statistically robust map of mass transfer physics across metallicity and mass ranges.

7. Hybrid Single/Binary Evolution Classifier

• Use the paper’s finding that accretion raises N/C more than rotational mixing for moderate N/O to train a discriminator that separates single-star rotational mixing from binary accretion.

• The improved AI can identify which enriched stars are truly single (fast rotators) versus binary products, correcting biases in stellar evolution models and improving estimates of the binary fraction among massive stars.

The majority of massive stars are born in close binary systems. As stars expand when they age, mass transfer or even a merger with their companion is inevitable. However, most binary interaction products appear as single stars, such that the main evidence of their exciting past is lost. In a comprehensive grid of detailed massive binary evolution models we find systematic trends in chemical surface abundances that allow identifying the past mass gainers. We develop an analytic framework which is independent of specific evolutionary models, to constrain the amount and composition of the accreted material from their observed surface abundances. This yields tight constraints on the uncertain mass transfer physics in massive binary stars and allows us to reconstruct the past evolutionary history of the progenitor binary system. This method, which is shown to also constrain binary mergers (for example, SN 1987A), is applied to some of the best-studied OB stars so far. For gamma Columbae, suggested to be an envelope-stripped star, we show that it is a mass gainer instead, whose companion star likely formed a stripped-envelope supernova. Our results highlight surface abundance measurements as a powerful tool to improve our understanding of massive binary systems evolving towards supernovae and compact object binaries.