A Bonded-Particle Discrete-Element Model with Self-Gravity for the Collision and Reaccumulation of Rubble-Pile Asteroids: Formulation and Numerical Protocol

The fundamental degree of freedom of the model is the rigid sphere. Particle $k$ is fully specified by a position $\bm{x}_{k}$, a velocity $\bm{v}_{k}$, a radius $r_{k}$, and a density $\rho_{k}$, from which the mass follows as $m_{k}=\tfrac{4}{3}\pi\rho_{k}r_{k}^{3}$. Each particle additionally carries an integer label $b_{k}$ identifying the composite body to which it belongs. Two particles are only ever candidates for a permanent cohesive bond (Sec II.3) if they share the same $b_{k}$, but they interact through contact (Sec II.2) and through gravity (Sec II.4) irrespective of body membership. Here, a composite asteroid is nothing more than a set of particles sharing a label and a network of cohesive bonds. No rigid-body or continuum degrees of freedom are introduced at the composite level. The composite’s translational and any effective rotational or elastic, behavior is entirely emergent from the sum of its grains’ interactions. Rotational degrees of freedom of the individual grains themselves are not modeled (only translational motion is integrated) so that contact and bond moments are absent from the formulation. I address to this limitation in Sec II.3.

Two spheres $i,j$ that overlap, $\delta_{n}\equiv r_{i}+r_{j}-|\bm{x}_{j}-\bm{x}_{i}|>0$, interact through a Cundall-Strack linear model [2], illustrated in Fig 1(a), irrespective of whether they belong to the same or to different composite bodies: once a cohesive bond between two grains has failed (Sec II.3), this is precisely the interaction that takes over when the two grains remain in geometric contact. Let $\bm{n}_{ij}=(\bm{x}_{j}-\bm{x}_{i})/|\bm{x}_{j}-\bm{x}_{i}|$ be the unit normal, $\bm{v}_{ij}=\bm{v}_{j}-\bm{v}_{i}$ the relative velocity, and $v_{n}=\bm{v}_{ij}\cdot\bm{n}_{ij}$, $\bm{v}_{t}=\bm{v}_{ij}-v_{n}\bm{n}_{ij}$ its normal and tangential components. The normal contact force is

with $k_{n}$ a normal contact stiffness and $\gamma_{n}$ a normal damping coefficient. The clipping at zero enforces that contact is purely repulsive (i.e. two grains that are not bonded exert no adhesive force on one another once they separate).

The tangential response uses an incremental elastic-predictor, Coulomb-corrector scheme with memory. A per-contact tangential elastic displacement $\bm{u}_{t}$ is carried from one time step to the next, re-projected onto the current tangential plane to remove any drift along the (possibly rotating) normal direction, and advanced according to $\bm{u}_{t}\leftarrow(\bm{u}_{t}-(\bm{u}_{t}\!\cdot\!\bm{n}_{ij})\bm{n}_{ij})+\bm{v}_{t}\,\Delta t$. A trial tangential force,

is then clipped to satisfy the Coulomb inequality $|\bm{F}_{t}|\leq\mu F_{n}$: if $|\bm{F}_{t}^{\text{trial}}|>\mu F_{n}$ the contact is sliding and the transmitted force is rescaled, $\bm{F}_{t}=\bm{F}_{t}^{\text{trial}}\,\mu F_{n}/|\bm{F}_{t}^{\text{trial}}|$, with the stored displacement $\bm{u}_{t}$ simultaneously corrected so that re-inserting it through the elastic law above reproduces the clipped force exactly, keeping the memory variable consistent with what was actually applied. The pair $(\bm{F}_{n},\bm{F}_{t})$ act along $\bm{n}_{ij}$ and within the tangential plane respectively and are applied as an action-reaction pair, $\bm{F}_{i\to j}=-\bm{F}_{j\to i}$. Contacts that stop to overlap have their tangential memory discarded. Because this short-range interaction must in principle be evaluated for every pair of grains at every time step, its cost is controlled by restricting the candidate pairs to a Verlet neighbor list (Sec II.5) rather than by rescanning all $O(N^{2})$ pairs.

Cohesion between grains of a common parent body is introduced through a bonded-particle model sketched in Fig 1(b). At sample-generation time (detailed in Sec III.1), pairs of grains in the same body that are sufficiently close are glued by an elastic link of cross-sectional area:

recording the pair’s initial center-to-center distance $r_{0}$ as its natural, unstressed length. Unlike the contact law of Sec II.2, an intact bond transmits both tensile and compressive normal load: with $\delta_{n}\equiv r-r_{0}$ the elongation at the current separation $r=|\bm{x}_{j}-\bm{x}_{i}|$,

positive $\delta_{n}$ (stretching) pulling the two grains together and negative $\delta_{n}$ (compression) pushing them apart. The bond behaves as a simple linear spring around $r_{0}$, with no rectification at $\delta_{n}=0$. The shear response accumulates a tangential elastic displacement $\bm{u}_{t}^{\text{bond}}$ exactly as in Sec II.2, giving a shear force $\bm{F}_{t}^{\text{bond}}=k_{t}^{\text{bond}}\,\bm{u}_{t}^{\text{bond}}$ with no Coulomb cap. So while intact, a bond is a permanent elastic link, not a frictional contact.

The bond carries a normal and a shear stress,

evaluated at every time step and compared against critical thresholds $\sigma_{c}$ (tensile) and $\tau_{c}$ (shear). The bond is intact as long as $\sigma\leq\sigma_{c}$ and $\tau\leq\tau_{c}$. As soon as either threshold is exceeded, the bond is marked broken and thereafter transmits no force for the remainder of the simulation. Failure is an irreversible event and a fixed compressive threshold is used symmetrically for $\sigma<0$ since the criterion in Eq 5 is applied to the signed stress. Once broken, the pair of grains it used to connect reverts to ordinary, tensionless contact (Eq 1) if they remain in geometric overlap. This is the microscopic mechanism by which a composite body absorbs a disturbance elastically but sheds discrete bonds and/or fragments, under a sufficiently violent one. This is a stress, rather than an energy-based, failure criterion, applied (not combined, e.g., in a Mohr-Coulomb or elliptical envelope) to the normal and shear channels.

Every pair of particles in the simulation, whether or not they belong to the same composite body, attracts every other pair through Newtonian gravity, softened at short range to regularize near-coincident grains:

where $\epsilon$ is a softening length (set to zero unless stated otherwise) and $G$ is the gravitational constant. The sum runs over all particles in the simulation without regard to body membership: it is this term, evaluated intra-body, that (together with cohesion) holds a rubble pile together in the first place and, inter-body, that lets two initially separate composites attract each other and, potentially, remain gravitationally bound after a gentle encounter, whether or not any residual bond survives between them.

Equation 6 is evaluated by direct pairwise summation. It is an unavoidable $O(N^{2})$ cost per time step for a genuinely long-range interaction with no natural notion of a neighbor cutoff. The double loop over $i$ and $j$ is distributed over threads by particle index $i$, each thread privately accumulating $\bm{F}_{i}^{\text{grav}}$, which avoids any write conflict between threads at the cost of a factor of two in floating-point operations relative to a half loop with explicit action-reaction bookkeeping.

Rescanning all pairs at every time step merely to determine which grains are close enough for the short-range interactions of Secs II.2–II.3 to matter would be wasteful, since the great majority of pairs are not in contact at any given instant. The neighbor list retains, at intervals, every pair $(i,j)$, $i<j$, whose center-to-center distance is below $r_{i}+r_{j}+s$, where $s$ pads the true contact threshold $r_{i}+r_{j}$ with a safety margin. Once built, the list remains valid as long as no single particle has moved by more than $s/2$ from the position it held at the last rebuild: two particles absent from the list can approach each other by at most $2\times(s/2)=s$ before their true separation could possibly have dropped below $r_{i}+r_{j}$, which is exactly the margin built into the cutoff.

The rebuild criterion is therefore

tracked against the reference configuration $\bm{x_{k}^{\text{ref}}}$ recorded at the previous rebuild. Between rebuilds, the contact model of Sec II.2 only tests true overlap among the candidate pairs returned by the list, rather than over all $O(N^{2})$ pairs. The list narrows the candidate set but does not itself decide contact. The $s$ parameter is a pure performance parameter (by default I use $0.3$ times the mean grain radius) trading rebuild frequency against candidate-list size, with no effect on the converged physical trajectory in the small time-step limit. As noted above, this acceleration structure is specific to the short-range channels: self-gravity is evaluated over the full particle set at every step regardless.

At a given instant, the total force on particle $i$ is the sum of three contributions,

with the restriction that a pair of grains still linked by an intact bond is excluded from the non-bonded contact sum of Sec II.2 to avoid double-counting their normal interaction.

The equations of motion, $m_{k}\ddot{\bm{x}}_{k}=\bm{F}_{k}$, are integrated with a Verlet scheme [14] at fixed time step $\Delta t$,

where $\bm{F}_{k}^{n+1}$ is evaluated at the updated positions $\bm{x}_{k}^{n+1}$ using the half-step velocities $\bm{v}_{k}^{n+1/2}$ for every velocity-dependent term. Only two force evaluations per step are needed: one at the start-of-step configuration for the half-step kick, and one at the predicted end-of-step configuration, which simultaneously serves as the corrector force and, on steps where output is due, supplies the detailed per-contact and per-bond records, so that no additional force pass is required purely for reporting purposes.

The neighbor list of Sec II.5 is checked, and rebuilt (if necessary) once per step, immediately before the first evaluation. The time step is chosen from the classical critical time of a linear spring-mass contact,

with $m_{\min}$ the mass of the lightest grain in the sample and a safety factor $\eta=1/30$.

Figure 2 summarizes the resulting per-step control flow.

Table 1 collects the free parameters of the model with the default value chosen. No intrinsic unit system is imposed by the code itself, and the rescaling of $G$ used in the reaccumulation sweep of Sec III.3 is a property of the numerical simulation, not of the model.

Each composite asteroid is built independently by a random-sequential addition (RSA) packing procedure [16]. Grain radii are chosen once from a mean value, $r_{k}=\bar{r}\,(1+\xi\,(2u_{k}-1))$ with $u_{k}\sim\mathcal{U}(0,1)$ and $\xi$ a relative fraction and grains are then inserted one at a time at a uniformly random direction and a radially biased distance inside a sphere of radius $R$, subject to non-overlap with every previously placed grain. Numerically, I have done a criterion that if an insertion fails after a fixed number of trials, it aborts the packing at its current occupancy rather than looping indefinitely. The confining radius $R$ is set from a target volume filling fraction $\phi=0.35$ (something like a loose granular packing)

with $N$ the number of grains per body. Once a body’s grains are placed, any pair within a tolerance of first contact, $|\bm{x}_{i}-\bm{x}_{j}|\leq\beta\,(r_{i}+r_{j})$ with $\beta$ a bond-tolerance factor slightly above unity (the default value is $1.05$, absorbing the residual gaps of the RSA packing), is recorded as a permanent cohesive bond (Sec II.3). Each asteroid is generated independently with its own random realization of this procedure (same $N$, $\bar{r}$, $\zeta$, $\phi$, $\beta$ for both bodies in every run reported here), then rigidly translated to a prescribed center and given a common translational velocity, without otherwise perturbing its internal configuration.

The two composite bodies are placed symmetrically about the origin along the $x$ axis at a prescribed center-to-center separation $d$ (by default $6R$, comfortably outside any initial overlap) and launched toward each other at a prescribed relative approach speed $v_{0}$ split symmetrically between the two centers of mass, as depicted in Fig 3.

A distinct, complementary numerical experiment addresses the opposite regime: two bodies released from rest, $v_{0}=0$, so that the only force bringing them together is their mutual self-gravity (Eq 6, evaluated inter-body), and the question of interest is not whether a collision occurs, but whether repeated low-speed contact under self-gravity leads, within a finite simulated time, to a single merged remnant, or instead to one or more bounces that have not yet settled.

Two points of orbital mechanics fix the shape of this simulation. First, at $v_{0}=0$ the total angular momentum of the two-body system about their common center of mass is identically zero, $\bm{L}=\sum_{b}\bm{r}_{b}\times m_{b}\bm{v}_{b}=\bm{0}$ at $t=0$, and remains so for all time by conservation, since the mutual gravitational force is central: an initial transverse offset (the impact parameter $\ell$ of Sec III.2) therefore has no effect whatsoever on the ensuing trajectory when $v_{0}=0$ The impact parameter is accordingly dropped from this sweep entirely because it only becomes a physically meaningful control variable once $v_{0}\neq 0$, as in Sec III.2. Second, the total mechanical energy at release, $E=0+U_{\text{grav}}(d)<0$, is negative for any finite starting separation $d$, so the pair is gravitationally bound in the wide sense from the outset. Since contact damping $\gamma_{n}$ can only remove mechanical energy, never add it, the two bodies can never escape to infinity.

So, the physical question is not whether the two bodies remain associated (obviously… they always do) but how quickly the encounter converges to a single aggregate as opposed to bouncing without yet having merged by the end of the simulated window. This distinction motivates the choice of the three parameters actually varied in the sweep:

• •

  the grain density $\rho$, which at fixed geometry sets each body’s mass and hence the depth of the mutual gravitational potential well.

• •

  the initial separation $d$, which sets the gravitational potential energy released before first contact, and so the impact speed.

• •

  the normal contact damping $\gamma_{n}$, which sets how each individual bounce is damped (the range is in accordance with [17]).

For each of the $5\times 5\times 5=125$ combinations drawn from five logarithmically spaced densities spanning four decades, $\rho\in\{10^{1},10^{2},10^{3},10^{4},10^{5}\}\,$\mathrm{k}\mathrm{g}\mathrm{/}\mathrm{m}^{3}$$, five logarithmically spaced initial separations spanning two decades, $d\in\{10^{1.5},10^{2},10^{2.5},10^{3},10^{3.5}\}\,$\mathrm{m}$\approx\{31.6,\,100,\,316,\,1000,\,3162\}\,$\mathrm{m}$$, and five logarithmically spaced contact damping coefficients spanning four decades, $\gamma_{n}\in\{10^{2},10^{3},10^{4},10^{5},10^{6}\}\,$\mathrm{N}\!\cdot\!\mathrm{s}\mathrm{/}\mathrm{m}$$ (grain number, radius, contact and bond stiffness, and the critical bond stress are held fixed across the sweep at a target $N=500$ grains per body, $\bar{r}=$1\text{\,}\mathrm{m}$$, $k_{n}=$1\text{\times}{10}^{6}\text{\,}\mathrm{N}\mathrm{/}\mathrm{m}$$, $\sigma_{c}=$1\text{\times}{10}^{5}\text{\,}\mathrm{P}\mathrm{a}$$ the generator of Sec III.1 is invoked with $v_{0}=0$ and the corresponding $(\rho,d,\gamma_{n})$ triple, followed by a full run of the code of Sec II on the resulting sample, in its own, independent output subdirectory. The logarithmic spacing of $\rho$ and $\gamma_{n}$ is deliberate since the critical contact damping of a single grain-grain contact, $\gamma_{n}^{\text{crit}}=2\sqrt{k_{n}m_{\text{eff}}}$ with $m_{\text{eff}}$ the reduced mass of the two grains, itself scales as $\sqrt{\rho}$, a linear grid in either variable alone would sample the physically relevant transition $\gamma_{n}/\gamma_{n}^{\text{crit}}\sim 1$ only over a narrow, density-dependent corner of the grid.

Each of the $125$ grid points is generated from a single random packing realization. The reported scatter therefore convolves the physical response with the stochastic variability of the generator, which is not separately quantified here.

Two practical points complete the protocol. First, the gravitational constant used in this study, $G_{\text{sweep}}=$2\text{\times}{10}^{-5}\text{\,}$$, is deliberately amplified by six orders of magnitude above the physical value $G=$6.674\text{\times}{10}^{-11}\text{\,}$$. At the toy length and mass scale adopted here (meter-sized grains, few-tonne bodies), the physical free-fall time between the two bodies would run to simulated weeks, i.e., far beyond a tractable number of time steps at the resolution of Eq II.6.

Second, the number of integration steps allotted to each run is set from an estimate of the two-point-mass radial free-fall time (the degenerate, zero-angular-momentum limit of the Kepler problem),

with $M_{\text{tot}}$ the combined mass of the two bodies at the sampled density, multiplied by a safety factor of $4$ to allow for the initial fall, the first collision, and a few subsequent bounces, and capped at a fixed maximum of $5\times 10^{6}$ steps regardless of how large the free-fall estimate turns out to be for the most extreme corner of the parameter grid.

Before assessing the outcome of the sweep, its self-gravity and integrator can first be checked against Sec III.3’s own analytic reference: the zero-angular-momentum Kepler solution $x(\tau)$, $\dot{x}(\tau)$ parameterized by Eq II.6’s companion angle variable (Sec III.3), universal and free of any run parameter once separation and time are rescaled by $d_{0}$ and $t_{\text{ff}}(d_{0})$ respectively.

Figure 4 shows this parameter-free curve with the measured $d(t)/d_{0}$ trajectory of every one of the $125$ runs, colored by $\zeta$. Every trajectory collapses onto the analytic curve to within plotting accuracy from $t=0$ up to the geometrically predicted first-contact point $(\tau_{c},x_{c})$ (open circles, one per sampled $d_{0}$, computed from $d_{\text{contact}}/d_{0}$ alone), independent of $\rho$, $\gamma_{n}$, or the eventual outcome. This is a direct argument demonstrating that the model accurately reproduces two-body free fall before any contact occurs, and it determines the point at which the point-mass approximation must be abandoned in favor of full dynamics. Past that marked point, trajectories immediately fan out according to $\zeta$, as expected once contact damping starts to compete with the kinetic energy delivered by the fall.

On the $125$ runs, $71$ ($56.8\%$) are classified merged, $42$ ($33.6\%$) infalling, and $12$ ($9.6\%$) bouncing. The purely geometric ratio $v_{\text{imp}}/v_{\text{esc}}=\sqrt{1-d_{\text{contact}}/d_{0}}$ takes exactly the five values dictated by the five sampled separations (Table 2), independent of density, since both velocities scale identically with $\sqrt{M_{\text{tot}}}$.

This trend, however, is still not primarily a statement about reaccumulation dynamics: it remains, even at the enlarged step budget, overwhelmingly a statement about how much of the fixed-step-budget window (Sec III.3) is spent in the pre-contact free fall of Sec IV.1. Of the $42$ infalling runs, $40$ never reach a separation below half of $d_{0}$ at any point in their recorded history, and at the largest sampled separation, none of the $16$ still infalling runs there ever comes within $25\text{\,}\mathrm{m}$ (essentially $d_{\text{contact}}$) of contact, while every one of the $9$ that do reach that band’s merger is (unsurprisingly) counted as merged, not infalling. This is corroborated independently by the bounce count of Sec IV.5: every single one of the $42$infalling runs records exactly zero sign changes of the approach velocity. It is consistent with a run that never gets far enough to complete even its first contact, let alone bounce.

Four runs in the entire grid (density ${10}^{5}\text{\,}\mathrm{k}\mathrm{g}\mathrm{/}\mathrm{m}^{3}$, $d_{0}=$100\text{\,}\mathrm{m}$$) are caught genuinely near, though not quite past, first contact ($46$–$54\,$\mathrm{m}$$, against a contact distance of $22.7\text{\,}\mathrm{m}$) at the moment their window closes. These four belong to the density corner flagged as energetically unreliable below, so their proximity to contact should be read as geometrically indicative rather than quantitatively trustworthy. The infalling label remains predominantly an artifact of a step budget sized from the point-mass free-fall estimate $t_{\text{ff}}(d_{0})$ rather than from the substantially longer time needed to also resolve the ensuing contact sequence at the lowest sampled densities and the largest sampled separations.

A second, opposite effect operates at the single grid corner of highest density and closest separation ($\rho=${10}^{5}\text{\,}\mathrm{k}\mathrm{g}\mathrm{/}\mathrm{m}^{3}$$, $d_{0}=$31.6\text{\,}\mathrm{m}$$): there, the same step-budget heuristic collapses to its floor of $200$ steps for all five sampled $\gamma_{n}$, because the point-mass estimate $4\,t_{\text{ff}}(d_{0})$ is itself already shorter than that threshold. These five runs record the most violent encounters of the entire sweep on purely geometric grounds and are still separating, at up to $8.7\text{\,}\mathrm{m}\mathrm{/}\mathrm{s}$, when the run ends. However, it is precisely in this area of density that the code’s own energy balance loses its reliability. Therefore, from a quantitative standpoint, one should not attach too much importance to either the damage indicated or the rate at which it propagates.

Fig 5 shows the full particle-level final state of four representative runs, one per broad outcome, rendered directly from each run’s own final particle positions, radii, body labels, and intact-bond list. Grain shading is produced exclusively by angularly-clipped cross-hatching under a single fixed light direction, never by a continuous tone. Intact cohesive bonds are drawn as solid links; a handful of the fastest-moving grains are marked by short directional strokes rather than by an instantaneous velocity field. And the mutual gravitational field of Eq 6 is superposed as a bundle of regularly dashed lines connecting the two bodies’ centers of mass, drawn without regard to occlusion by the grains themselves.

Fig 5.a) ($\rho=$10\text{\,}\mathrm{k}\mathrm{g}\mathrm{/}\mathrm{m}^{3}$$, $d_{0}=$31.6\text{\,}\mathrm{m}$$, $\gamma_{n}=$100\text{\,}\mathrm{N}\,\mathrm{s}\mathrm{/}\mathrm{m}$$) is merged: the two composites have fallen into, and remain in a single aggregate. Fig 5.c) (same $\rho$ and $\gamma_{n}$ but $d_{0}=$316\text{\,}\mathrm{m}$$) is infalling: by the end of the run the separation has fallen from $316$ to $229\text{\,}\mathrm{m}$, but the two bodies remain fully apart and intact. Fig 5.d) ($\rho=${10}^{4}\text{\,}\mathrm{k}\mathrm{g}\mathrm{/}\mathrm{m}^{3}$$, $d_{0}=$3162\text{\,}\mathrm{m}$$, $\gamma_{n}=${10}^{6}\text{\,}\mathrm{N}\,\mathrm{s}\mathrm{/}\mathrm{m}$$) is merged as well ($d_{\text{final}}\approx$6\text{\,}\mathrm{m}$$) but with $D_{\text{final}}=1$ ($D$ is described in Sec IV.4): no bond survives, and the directional strokes mark a population of fast-moving grains within the reassembled pile consistent with a large kinetic energy. Fig 5.b) ($\rho=${10}^{3}\text{\,}\mathrm{k}\mathrm{g}\mathrm{/}\mathrm{m}^{3}$$, $d_{0}=$100\text{\,}\mathrm{m}$$, $\gamma_{n}=${10}^{6}\text{\,}\mathrm{N}\,\mathrm{s}\mathrm{/}\mathrm{m}$$) is bouncing, but at the resolution of individual grains this label does not mean two intact bodies caught mid-separation: extending the integration this far instead reveals a slowly evaporating swarm, a minority of grains still clustered near a common center while the rest have individually drifted, under self-gravity alone, out to distances far beyond the pair’s own nominal radius. This is a direct particle-level illustration of the sub-percolation-threshold bond network: with a mean bond count below one per grain, a sizable fraction of each composite was never cohesively attached to anything in the first place, and is consequently free to disperse under even the modest gravity of this toy system once given the several-$\mathrm{k}\mathrm{s}$ window this run’s step budget now allows.

For the subset of simulations that do reach contact, the relevant impact severity is not $v_{\text{imp}}$ alone but its ratio to the bond network’s own force scale. An unbonded grain-grain contact closing at speed $v_{\text{imp}}$ loads a linear contact spring of stiffness $k_{n}$ and reduced mass $m_{\text{eff}}$ up to a peak elastic force $v_{\text{imp}}\sqrt{k_{n}m_{\text{eff}}}$ with $m_{\text{eff}}=m_{i}m_{j}/(m_{i}+m_{j})$ the reduced mass of the two colliding grains.

Comparing this force scale to the bond-breaking force $\sigma_{c}A$ of Eq 5 (where $A=\pi\bar{r}^{2}$) defines the dimensionless impact-stress number

which spans four decades over the study, from $1.9\times 10^{-3}$ to $35.9$. However, as established in section IV.2, all simulations greater than $\Pi_{\sigma}\approx 3.6$ belong to a single sampling density whose energy balance is unreliable. Consequently, the reliable subset of $90$ simulations only ever samples values of $\Pi_{\sigma}$ up to $3.59$.

Figure 7 plots the final damage $D_{\text{final}}$ and largest-fragment mass fraction $f_{\text{largest}}$ of every run against $\Pi_{\sigma}$, including the $35$ excluded ones (identifiable by their label). The damage variable, determined such as:

tracks the fraction of the initial cohesive network that has failed.

To define $f_{\text{largest}}$, I define a fragment as a connected component of a graph whose vertices are the grains and whose edges are the bonds still intact at that instant. Two grains merely touching, without an intact bond between them, are not considered part of the same fragment, since an unbonded contact can slip apart at any subsequent step. Connected components are extracted with a union-find (disjoint-set) structure with path compression, run over the current set of intact bonds, and the mass fraction carried by the single largest component,

is reported alongside the total fragment count $N_{\text{frag}}$. This mass fraction of the largest fragment is shown schematically in Figure 6.

The interpretation below applies only to the $90$ retained simulations. No damage is recorded below $\Pi_{\sigma}\approx 0.036$. $D_{\text{final}}$ becomes systematically positive by $\Pi_{\sigma}\approx 0.19$ and within this trustworthy subset, already reaches complete failure ($D_{\text{final}}=1$, $f_{\text{largest}}=0.0016$, one surviving grain) by $\Pi_{\sigma}\approx 3.59$. The transition to a complete breakage therefore lies within (and not beyond) the explored range. The collapse in $\Pi_{\sigma}$ is not exact: at fixed $\Pi_{\sigma}\approx 1.92$, for instance, $D_{\text{final}}$ still ranges from $0.69$ to $0.96$ across the five sampled $\gamma_{n}$. The residual dispersion is attributable to $\zeta$, which determines the number of loading cycles (Section IV.5) to which the surviving network is subjected before the end of the test, rather than to $\Pi_{\sigma}$, which incorrectly measures the severity of a single impact. $\Pi_{\sigma}$ therefore determines the overall extent of damage and fragmentation, but does not by itself determine the exact value in a given simulation.

Figure 8 reports the number of sign changes of the radial approach velocity recorded before the end of each simulation as a function of $\zeta$, for the $90$ simulations unaffected by either pathology of Sec IV.2. Here, $\zeta\equiv\gamma_{n}/\gamma_{n}^{\text{crit}}$ denotes the dimensionless contact damping ratio. Two regimes stand out. For $\zeta\gtrsim 3$ ($20$ runs), only $60\%$ merge (or are caught mid-fall) within zero to three bounces. The others require between $4$ and $53$ bounces despite their high damping ratio. A single sufficiently overdamped contact removing most of the kinetic energy delivered by the fall, consistent with the analytic single-contact restitution coefficient $e(\zeta)=\exp(-\pi\zeta/\sqrt{1-\zeta^{2}})$ falling rapidly toward $0$ over the same range, is therefore at best a partial description even at $\zeta\gtrsim 3$, and the qualitative claim that high damping alone guarantees fast capture does not survive the longer integration. For $\zeta\lesssim 1$, merged runs require anywhere from a handful to as many as $69$ bounces, with substantial scatter at fixed $\zeta$ rather than a single well-defined count, before enough energy has been removed for the pair to settle. This remains the qualitative behavior expected once each bounce only removes a fraction $1-e(\zeta)^{2}\ll 1$ of the orbital kinetic energy, but, as before, no quantitative fit to $e(\zeta)$ is attempted, since $e(\zeta)$ is a single-grain-contact quantity, whereas the number of bounces of a two-body, many-grain composite additionally depends on the evolving bond damage (Sec IV.4). The three bouncing simulations not otherwise flagged in Sec IV.2 sit at $\zeta\approx 10.9$, with only one to three bounces recorded and a small residual outward velocity ($<$0.2\text{\,}\mathrm{m}\mathrm{/}\mathrm{s}$$) at the end of the run: these remain consistent with a merger simply not yet complete within the allotted window, rather than with either censoring mechanism identified above.