Convective Kissing Instability (CKI)

• CKI is a nuclear-driven structural instability in low-mass stars near the fully convective boundary, triggered by non-equilibrium 3He burning.
• Numerical models using YREC and MESA with fine mass and time resolution reveal cyclic merging of convective zones that produce observable discontinuities in luminosity and radius.
• The instability accounts for the Gaia M-dwarf gap and influences stellar evolution, with metallicity, overshooting, and semi-convection critically shaping its behavior.

The convective kissing instability (CKI) is a nuclear-driven structural instability in low-mass main-sequence stars near the transition between partially and fully convective configurations. It arises when non-equilibrium $^{3}\mathrm{He}$ burning produces a convective core inside a star that already possesses a deep convective envelope, leaving a thin radiative layer between the two; repeated episodes in which these convective regions merge, briefly rendering the star fully convective, define the instability. CKI produces cyclic variations in luminosity, effective temperature, radius, and internal composition, and is now closely associated with the Gaia M-dwarf gap and with a discontinuity in the low-mass luminosity-mass relation (Saders et al., 2012, Mansfield et al., 2021, Mansfield et al., 2023).

1. Definition, discovery, and stellar regime

CKI was identified as an instability in low-mass stars just above the threshold where they are expected to be fully convective on the main sequence. In the original formulation, non-equilibrium $^{3}\mathrm{He}$ burning creates a convective core that is separated from a deep convective envelope by a small radiative zone; the steady increase in central $^{3}\mathrm{He}$ causes the core to grow until it touches the surface convection zone, which triggers fully convective episodes (Saders et al., 2012). Later work connected the same mechanism directly to fluctuations in low-mass stellar evolutionary tracks, to a discontinuity in the luminosity-mass relation, and to the Gaia M-dwarf gap (Mansfield et al., 2021). Mansfield & Kroupa extended this picture by examining convective overshooting, semi-convection, luminosity functions, surface abundances, and age-dating applications (Mansfield et al., 2023).

The instability occupies a narrow mass interval whose exact location depends on input physics, composition, atmosphere treatment, and evolutionary code. The fully convective boundary itself lies near $M\sim 0.3$–$0.35\,M_\odot$, but the CKI-active interval is slightly above that threshold and is not identical across calculations.

Study or model set	CKI-active mass range	Notes
YREC, solar composition	$0.322\,M_\odot \lesssim M \lesssim 0.365\,M_\odot$	Fully convective below $0.266\,M_\odot$
MESA in the 2012 study	$0.343\,M_\odot \lesssim M \lesssim 0.375\,M_\odot$	Fully convective below $0.281\,M_\odot$
Solar-metallicity MESA models in 2023	$$0.3440\,M_\odot \lesssim M \lesssim 0.3825\,M_\odot$$	Often illustrated with $^{3}\mathrm{He}$0

Within this regime, the star alternates between two structural states: a configuration with a convective core, a radiative shell, and a convective envelope, and a temporarily fully convective configuration produced when the two convective regions “kiss” and merge. These episodes correspond to few percent changes in radius and luminosity on Myr to Gyr timescales (Saders et al., 2012).

2. Nuclear-structural mechanism

The physical driver of CKI is the non-equilibrium behavior of $^{3}\mathrm{He}$1 in the $^{3}\mathrm{He}$2 chain. The relevant reactions are

$^{3}\mathrm{He}$3

$^{3}\mathrm{He}$4

$^{3}\mathrm{He}$5

At the low central temperatures of M dwarfs near the fully convective boundary, reaction (2) produces $^{3}\mathrm{He}$6 efficiently, while reaction (3) destroys it comparatively slowly. The core therefore does not reach local $^{3}\mathrm{He}$7 equilibrium; instead, $^{3}\mathrm{He}$8 accumulates and enhances nuclear energy generation (Saders et al., 2012, Mansfield et al., 2021).

Convection is governed by the Schwarzschild criterion,

$^{3}\mathrm{He}$9

with

$^{3}\mathrm{He}$0

As the central $^{3}\mathrm{He}$1 abundance rises, the local luminosity $^{3}\mathrm{He}$2 increases, $^{3}\mathrm{He}$3 steepens, and a convective core appears within what had previously been a radiative central region. Because the outer layers are already convective, the star enters a three-zone state: convective core, radiative shell, convective envelope (Mansfield et al., 2023).

A typical CKI cycle proceeds in a definite sequence. First, $^{3}\mathrm{He}$4 builds up in the radiative core, raising the contribution of $^{3}\mathrm{He}$5 burning to the energy budget. Second, the convective core grows outward while the envelope convection zone deepens inward. Third, the two convective regions merge; the radiative shell disappears, and the star becomes fully convective for a short period, with $^{3}\mathrm{He}$6 in the 2023 models (Mansfield et al., 2023). Fourth, rapid mixing dilutes the $^{3}\mathrm{He}$7-rich core with the envelope, so the central $^{3}\mathrm{He}$8 abundance drops, the surface $^{3}\mathrm{He}$9 abundance rises, and nuclear energy generation falls. Fifth, the star contracts slightly, the central temperature rises, a radiative core re-forms, and the accumulation phase begins again.

The instability damps over time. Each fully convective episode raises the global $M\sim 0.3$0 abundance of the star, so the contrast between core and envelope declines from cycle to cycle. The radiative region that re-forms becomes smaller, the amplitude of the luminosity drop decreases, and eventually the star becomes and remains fully convective for the rest of its main-sequence lifetime (Saders et al., 2012, Mansfield et al., 2023).

3. Numerical modeling and sensitivity to resolution

CKI has been studied primarily with 1D stellar-evolution calculations using YREC and MESA. The 2012 work used both codes to show that the instability is not confined to a single numerical framework (Saders et al., 2012). The 2021 low-metallicity study used MESA version 15140 with metallicities $M\sim 0.3$1, an initial helium abundance

$M\sim 0.3$2

a fine mass step of $M\sim 0.3$3, and a time step of $M\sim 0.3$4 years (Mansfield et al., 2021). The 2023 study employed MESA r22.11.1, also adopted

$M\sim 0.3$5

used $M\sim 0.3$6, and evolved models from the ZAMS to $M\sim 0.3$7 Gyr with $M\sim 0.3$8 in the CKI region (Mansfield et al., 2023).

Fine temporal and mass resolution are central to the phenomenon. The 2021 calculations explicitly argue that CKI occurs over a very narrow mass range, with width $M\sim 0.3$9–$0.35\,M_\odot$0, and that coarse mass sampling can smooth out the luminosity-mass discontinuity. Large time steps can also miss short-timescale pulsations: a test model with $0.35\,M_\odot$1 yr showed that some small pulsations occur on timescales of a few $0.35\,M_\odot$2 yr, while the adopted $0.35\,M_\odot$3 yr resolved the repeated mergers and near-mergers relevant to the main CKI signal (Mansfield et al., 2021).

The 2023 models further emphasized the importance of convective-boundary treatment. Convective regions were identified via Schwarzschild, or via Ledoux when semi-convection was included, and convective mixing used the convective premixing scheme, in which cells identified as convective are instantaneously mixed (Mansfield et al., 2023). This is physically consequential because CKI is fundamentally a boundary-location instability: its onset, amplitude, and persistence depend on how the code represents the radiative barrier between core and envelope and on how composition gradients are mixed across it.

4. Metallicity, overshooting, and semi-convection

Metallicity changes both the location and duration of CKI. In the 2021 models, low-metallicity stars undergo the instability for longer portions of their lifetimes and with higher fluctuation amplitudes than higher-metallicity stars (Mansfield et al., 2021). The CKI band shifts to lower mass as $0.35\,M_\odot$4 decreases: for $0.35\,M_\odot$5, the discontinuity range is roughly $0.35\,M_\odot$6–$0.35\,M_\odot$7 at $0.35\,M_\odot$8 Gyr and $0.35\,M_\odot$9–$0.322\,M_\odot \lesssim M \lesssim 0.365\,M_\odot$0 at $0.322\,M_\odot \lesssim M \lesssim 0.365\,M_\odot$1 Gyr, while for $0.322\,M_\odot \lesssim M \lesssim 0.365\,M_\odot$2 it is lower still (Mansfield et al., 2021). A specific comparison given in the same study found that a $0.322\,M_\odot \lesssim M \lesssim 0.365\,M_\odot$3 star begins CKI at $0.322\,M_\odot \lesssim M \lesssim 0.365\,M_\odot$4 Gyr and ends by $0.322\,M_\odot \lesssim M \lesssim 0.365\,M_\odot$5 Gyr for $0.322\,M_\odot \lesssim M \lesssim 0.365\,M_\odot$6, whereas for $0.322\,M_\odot \lesssim M \lesssim 0.365\,M_\odot$7 it begins at $0.322\,M_\odot \lesssim M \lesssim 0.365\,M_\odot$8 Gyr and persists until $0.322\,M_\odot \lesssim M \lesssim 0.365\,M_\odot$9 Gyr. This suggests that reduced opacity and altered thermal structure move the partially-to-fully convective transition to lower mass and prolong the $0.266\,M_\odot$0-driven feedback loop.

Convective overshooting acts as a damping mechanism. In the 2023 study, overshooting beyond the formal convective boundary was modeled as

$0.266\,M_\odot$1

with $0.266\,M_\odot$2 (Mansfield et al., 2023). Increasing overshooting reduces the amplitude and intensity of CKI, decreases the number of full-convection episodes, and shrinks the loops in the Hertzsprung-Russell diagram. The physical reason given is that overshooting allows $0.266\,M_\odot$3 and $0.266\,M_\odot$4 to leak across the radiative barrier, smoothing the composition gradients that drive the instability.

Semi-convection has the opposite effect in these models. It operates where Schwarzschild predicts instability but Ledoux predicts stability,

$0.266\,M_\odot$5

with

$0.266\,M_\odot$6

Using $0.266\,M_\odot$7 and $0.266\,M_\odot$8, Mansfield & Kroupa found that semi-convection sustains CKI even in models where overshooting alone prevents full merger (Mansfield et al., 2023). In the example $0.266\,M_\odot$9, a model without semi-convection shows the core and envelope approaching but never fully merging, whereas the same model with semi-convection becomes fully convective and then re-establishes a radiative region, effectively restoring CKI behavior. The authors conclude that, to reproduce the observed M-dwarf gap, M-dwarf overshooting should be modest, $0.343\,M_\odot \lesssim M \lesssim 0.375\,M_\odot$0, and semi-convection should be present with non-negligible efficiency, $0.343\,M_\odot \lesssim M \lesssim 0.375\,M_\odot$1–$0.343\,M_\odot \lesssim M \lesssim 0.375\,M_\odot$2 (Mansfield et al., 2023).

5. Observable consequences

The best-known observational consequence of CKI is the Gaia M-dwarf gap. Gaia DR2 and eDR3 reveal a deficiency of stars on the lower main sequence at $0.343\,M_\odot \lesssim M \lesssim 0.375\,M_\odot$3 and $0.343\,M_\odot \lesssim M \lesssim 0.375\,M_\odot$4, with a $0.343\,M_\odot \lesssim M \lesssim 0.375\,M_\odot$5 drop in star counts, and the gap is more prominent on the blue edge of the main sequence (Mansfield et al., 2021). CKI provides a direct theoretical explanation: in the affected mass range, stellar tracks exhibit loops and pulsations in luminosity and effective temperature, so the color-magnitude region that would be populated by a smooth monotonic mass-luminosity relation becomes underdense.

Mansfield & Kroupa constructed synthetic populations by assigning model mass and magnitude values to three sets of $0.343\,M_\odot \lesssim M \lesssim 0.375\,M_\odot$6 stars with masses $0.343\,M_\odot \lesssim M \lesssim 0.375\,M_\odot$7–$0.343\,M_\odot \lesssim M \lesssim 0.375\,M_\odot$8, drawn from the canonical IMF with $0.343\,M_\odot \lesssim M \lesssim 0.375\,M_\odot$9 for $0.281\,M_\odot$0 and $0.281\,M_\odot$1 for $0.281\,M_\odot$2 (Mansfield et al., 2023). In these populations, CKI reproduces the M-dwarf gap as a pronounced indent into the blueward edge of the main sequence. In the luminosity function,

$0.281\,M_\odot$3

this appears as a small peak followed by a dip; for solar metallicity the feature is located near $0.281\,M_\odot$4 and $0.281\,M_\odot$5 (Mansfield et al., 2023). The peak is attributed to wave-like overlapping of models just above the gap, where tracks loop and overlap, while the dip corresponds to the gap itself.

The instability also produces a discontinuity in the luminosity-mass relation. The 2021 study argues that, with sufficiently fine sampling, $0.281\,M_\odot$6 is not merely bent or inflected but genuinely discontinuous across the CKI transition mass (Mansfield et al., 2021). This matters because luminosity functions and mass functions are often related through derivatives such as $0.281\,M_\odot$7, which become undefined at the discontinuity.

Surface abundance changes are present but small. In the 2023 solar-metallicity models, repeated full-convection episodes alter the surface abundances of $0.281\,M_\odot$8, $0.281\,M_\odot$9, $$0.3440\,M_\odot \lesssim M \lesssim 0.3825\,M_\odot$$0, $$0.3440\,M_\odot \lesssim M \lesssim 0.3825\,M_\odot$$1, $$0.3440\,M_\odot \lesssim M \lesssim 0.3825\,M_\odot$$2, and $$0.3440\,M_\odot \lesssim M \lesssim 0.3825\,M_\odot$$3 (Mansfield et al., 2023). By the end of the main-sequence lifetime of fully convective models, surface $$0.3440\,M_\odot \lesssim M \lesssim 0.3825\,M_\odot$$4 decreases by $$0.3440\,M_\odot \lesssim M \lesssim 0.3825\,M_\odot$$5, surface $$0.3440\,M_\odot \lesssim M \lesssim 0.3825\,M_\odot$$6 increases by $$0.3440\,M_\odot \lesssim M \lesssim 0.3825\,M_\odot$$7, and surface $$0.3440\,M_\odot \lesssim M \lesssim 0.3825\,M_\odot$$8 increases by a factor of $$0.3440\,M_\odot \lesssim M \lesssim 0.3825\,M_\odot$$9 but remains a very small fraction overall. The CNO changes are $^{3}\mathrm{He}$00 over the full evolution. The authors therefore conclude that CKI-driven surface abundance changes are too small to be practical observational diagnostics.

6. Age dating, binaries, and unsettled issues

Because CKI changes the morphology of the lower main sequence over time, Mansfield & Kroupa proposed several age-dating diagnostics for stellar populations and single stars (Mansfield et al., 2023). For composite populations, the overall width of the main sequence decreases with time, the lower main sequence is wider than the upper main sequence, and the difference in width also decreases with time. For single-age populations, they analyzed the parallel offset $^{3}\mathrm{He}$01 between the upper and lower main sequence at fixed color and the relative angle $^{3}\mathrm{He}$02 between the two segments. At early ages, $^{3}\mathrm{He}$03–$^{3}\mathrm{He}$04 and the offsets are positive; with increasing age, the segments become more parallel and $^{3}\mathrm{He}$05 approaches $^{3}\mathrm{He}$06. For individual M dwarfs below the CKI mass range, the models predict a time-dependent track in $^{3}\mathrm{He}$07–$^{3}\mathrm{He}$08 space up to $^{3}\mathrm{He}$09, suggesting that accurate mass, metallicity, and photometry could in principle provide an age estimate. The same paper stresses that these methods are model-dependent and should be cross-checked against other indicators.

In binary evolution, CKI has been discussed most extensively for cataclysmic variables (CVs). The 2012 study suggested that secondary stars in CVs pass through the same mass range and that CKI could be related to the observed paucity of systems with periods between two and three hours (Saders et al., 2012). A later dedicated investigation found a more limited effect: CKI has no effect on normal CVs that evolve via magnetic braking and gravitational radiation above the period gap, because CKI cycles either do not occur or are abruptly halted once mass transfer begins (Larsen et al., 22 Sep 2025). If only gravitational radiation is included, CKI does occur and the abrupt radius changes can cause detachment phases that produce small period gaps with widths of a few minutes. The same study argues that these results may be relevant to strong-field polars, where the magnetic field of the white dwarf is strong enough to suppress magnetic braking (Larsen et al., 22 Sep 2025). This suggests that CKI is not a viable explanation for the classical hour-scale CV period gap, but may generate minute-scale structures in special subclasses of interacting binaries.

Several open issues remain explicit in the literature. The 2023 study identifies rotation, the mixing-length parameter, binary effects, and further observational tests as important directions for future work (Mansfield et al., 2023). More generally, the phenomenon is sensitive to convective overshooting, semi-convection efficiency, and other boundary-mixing prescriptions, so the detailed amplitude and observability of CKI remain model-dependent even though the core mechanism—non-equilibrium $^{3}\mathrm{He}$10 burning near the fully convective boundary—is robust across YREC and MESA calculations (Saders et al., 2012, Mansfield et al., 2023).