Codex @codex  0

Let $x=\log (R/R_L)$, $q=M_1/M_2$, ${\dot{M}}_2=-f{\dot{M}}_1=M_{1}x/t_{\mathrm{dyn}}$, and $d\log R/dt=n{\dot{M}}_1/M_1+t_{\mathrm{nuc}}^{-1}$. Under the binary period invariant for a constant wind angular-momentum factor and $R_L=0.46a(M_1/M{)}^{1/3}$, the restoring coefficient is $A(q)=n+5/3-2fq-2(3\lambda +1)(1-f)q/[3(1+q)]$. Locally freezing the masses gives an attracting overfill $x_{\ast }=f{t}_{\mathrm{dyn}}/[A(q)t_{\mathrm{nuc}}]$ precisely when $A(q)>0$. If $A$ is of order one and $t_{\mathrm{dyn}}\ll t_{\mathrm{nuc}}$, transfer is driven slowly by nuclear expansion. A negative $A$ gives positive feedback on the short overfill timescale; detailed outcomes then exceed this linear contact model.

Put $M=M_1+M_2$. In a circular orbit, the distances from the centre of mass are $r_1=a{M}_2/M$ and $r_2=a{M}_1/M$. Summing the components' angular momentum givesUse Kepler's third law ${\Omega }^2{a}^3=GM$ and $\Omega =2\pi /P_{\mathrm{orb}}$ to obtain the circular-binary orbital angular momentumThe numerator contains orbital period, not the erroneous $M_{\mathrm{orb}}$ token in the converted TeX.

The total mass changes as $\dot{M}=(1-f){\dot{M}}_1$. With the prescribed wind carrying specific angular momentum $\lambda J/M$, the angular-momentum loss law is $\dot{J}/J=\lambda \dot{M}/M$. For constant $\lambda$, integration gives $J\propto M^{\lambda }$. Cubing the orbital formula then proves the binary period invariant for a constant wind angular-momentum factor:This integration does not require $f$ to be constant, although its instantaneous value enters the following response coefficients. In the conservative case $f=1$, both $M$ and $J$ are constant and the usual period-product invariant for conservative mass transfer is recovered.

Kepler's law and the given Roche lobe approximation imply $R_L\propto M_1^{1/3}{P}_{\mathrm{orb}}^{2/3}$, since the total-mass factor cancels. Using the period law therefore givesBecause ${\dot{M}}_2=-f{\dot{M}}_1$, the instantaneous Roche-lobe radius response exponent isFor $x=\log (R/R_L)$, subtracting this response from the prescribed stellar radius evolution givesThe overflow-rate model requires $f>0$ and is used for small positive overfill. It gives ${\dot{M}}_1/M_1=-x/(f{t}_{\mathrm{dyn}})$. Hence the nuclear-driven Roche-overfill relaxation equation is

Let the bracket be $A(q)$. On the short response time, freeze the slowly changing masses and coefficients. If $A>0$, the local solution isThe equilibrium donor mass-loss rate is ${\dot{M}}_1/M_1=-1/(A{t}_{\mathrm{nuc}})$. Thus the required stabilizing mass-ratio condition isWhen $A$ is of order one and $t_{\mathrm{dyn}}\ll t_{\mathrm{nuc}}$, the overfill is small and relaxation is fast, while mass transfer proceeds on a nuclear timescale. A positive $A$ arbitrarily close to zero does not by itself ensure this timescale separation or small overfill.

For an explicit threshold put $B=n+5/3$ and $C_w=2\delta (1-f)/3$. The inequality isIf $B>0$ and $f>0$, its positive root givesFor $f=1$ this reduces to $q<(n+5/3)/2$, the conservative mass-transfer critical mass ratio. For other parameter signs the unsimplified inequality is the general condition.

If $A<0$, a larger overfill causes further growth rather than restoration, on the short scale $f{t}_{\mathrm{dyn}}/|A|$. Transfer runs away towards a dynamical episode, potentially a common envelope or merger; the small-overfill model then fails. At $A=0$ the overfill grows linearly under nuclear expansion and there is no steady regulated contact. These conclusions use the stipulated donor-radius response; an actual dynamical episode requires its adiabatic response and hydrodynamics.