---
title: Edge-Based Interface Tracking (EBIT)
url: https://www.emergentmind.com/topics/edge-based-interface-tracking-ebit
type: topic
---

# Edge-Based Interface Tracking (EBIT)

Edge-Based Interface Tracking (EBIT) is a class of front-tracking methods for multiphase flows in which the interface is represented by Lagrangian marker points constrained to lie on the edges of the computational grid, while marker connectivity is reconstructed locally rather than stored as a global graph. In the EBIT literature, this edge-constrained representation has been developed from a simple two-dimensional Cartesian, piecewise-linear kinematic method into a broader framework that includes circle-fit reconstruction, the Color Vertex topological classifier, coupling to incompressible Navier–Stokes solvers with surface tension and phase change, an unsplit advection scheme, and a simplified three-dimensional extension in Basilisk [2202.06704][2309.00338][2402.13677][2509.06661][2509.26361]. A related method on unstructured triangle meshes, based on triangle edge cuts and polygonal area mapping, has been described as a close geometric cousin of EBIT because it likewise reconstructs the interface from edge intersections rather than from a diffuse scalar field or an explicit global connectivity structure [2410.11073].

## 1. Conceptual origin and development

EBIT was introduced as a “novel class” of interface-tracking methods for advecting interfaces in multiphase flows. The distinctive design choice is that interface markers live on the grid skeleton: in 2D on cell sides or grid lines, and in later formulations more generally on Eulerian mesh edges. This gives EBIT the explicitness of front tracking while avoiding permanent storage of global connectivity data, a burden that classical front-tracking methods typically carry [2202.06704].

The published EBIT line then evolves in stages. The 2022 paper presented a deliberately simple two-dimensional Cartesian instantiation with linear interface representation and a split geometric advection procedure. The 2023 surface-tension paper and the 2024 phase-change paper completed the method as a multiphase solver by adding circle-fit reconstruction, the Color Vertex, computation of volume fractions from markers, coupling to Basilisk’s Navier–Stokes solver, and automatic topology change. The 2025 unsplit paper replaced directional operator splitting by a true unsplit advection step to enable higher-order time integration and improved multiscale front-tracking correspondence. A later 2025 paper extended EBIT to three dimensions by reducing each 3D directional advection step to two 2D face problems and reusing the 2D reconstruction machinery [2309.00338][2402.13677][2509.06661][2509.26361].

| Paper | Main addition | Setting |
|---|---|---|
| "An Edge-Based Interface-Tracking Method for Multiphase Flows" [2202.06704] | Novel EBIT class; 2D Cartesian; linear interface representation | Kinematic |
| "An Edge-based Interface Tracking (EBIT) Method for Multiphase-flows Simulation with Surface Tension" [2309.00338] | Circle fit, Color Vertex, surface tension, automatic topology change | 2D Basilisk |
| "An Edge-based Interface Tracking (EBIT) Method for Multiphase Flows with Phase Change" [2402.13677] | Phase change, ghost-fluid treatment, thermal coupling | 2D Basilisk |
| "An Unsplit Scheme for Interface Advection in the Edge-based Interface Tracking (EBIT) Method" [2509.06661] | Unsplit advection, PC/RK4, modified color update | 2D Basilisk |
| "A Three-dimensional Edge-Based Interface Tracking (EBIT) Method for Multiphase-flow Simulations" [2509.26361] | Simplified 3D extension, Front2VOF coupling | 3D Basilisk |
| "An Interface Tracking Method with Triangle Edge Cuts" [2410.11073] | Triangle-edge-cut cousin with compact local encoding | Unstructured triangles |

## 2. Edge markers, local reconstruction, and the Color Vertex

The core EBIT representation is explicit but localized. Marker points represent interface intersections with grid edges, yet the interface is not stored as a globally connected polyline or spline. Instead, connectivity is reconstructed locally cell by cell from boundary markers and a binary color field attached to nearby vertices. This architecture is the central reason EBIT is described as easier to parallelize, easier to adapt under refinement and coarsening, and simpler under topology changes than classical front tracking [2509.06661].

In the original two-dimensional square-grid formulation, the interface was reconstructed as straight line segments between edge markers, and the method allowed at most two intersections per square edge. In the later 2D multiphase-flow implementation, the interface remained grid-bound but the paper states that only one marker per edge is allowed. Both descriptions are paper-specific formulations of the same edge-based idea: the front is represented by its intersections with the grid skeleton rather than by a free-standing front mesh [2202.06704][2402.13677].

The Color Vertex is the local topological classifier that removes ambiguity. It is a binary phase label, written in the phase-change paper as
$$
f_C = 1 \quad \text{in liquid}, \qquad f_C = 0 \quad \text{in vapor}.
$$
In 2D, five color vertices are sufficient for local topology identification: the four cell corners and the cell center. This local coloring is used to identify the reference phase, reconstruct the correct interface segments without ambiguity, distinguish ambiguous topological configurations, and drive automatic topology change [2402.13677].

A standard ambiguous case is a cell with four boundary markers, for which two different interface connectivities are possible. The Color Vertex selects the admissible connectivity. Because connectivity is implicit rather than explicitly stored, EBIT can update local interface topology through local color changes rather than through manual reconnection logic. In the surface-tension and phase-change formulations, a marker exists on a cell edge only if the two edge endpoints have different Color Vertex values; if two markers move into the same edge, they are eliminated automatically, and the remaining markers are reconnected according to the local Color Vertex pattern [2309.00338].

## 3. Advection algorithms and geometric reconstruction

The marker motion in EBIT follows the standard interface advection law
$$
\frac{d\mathbf{x}_i}{dt} = \mathbf{u}_i,
$$
with the original implementations using first-order explicit Euler time integration,
$$
\mathbf{x}_i^{n+1} = \mathbf{x}_i^n + \mathbf{u}_i^n \Delta t.
$$
In the earliest 2D Cartesian EBIT, advection was performed by dimension splitting. During a 1D sweep, markers on grid lines aligned with the advection direction were treated as aligned points and moved directly, while unaligned points were first fictitiously advected and then reconstructed geometrically by intersecting reconstructed interface segments with grid lines [2202.06704].

The 2023 and 2024 multiphase-flow papers retained the split strategy but replaced the earlier straight-line projection of unaligned markers by a circle fit. The new position of an unaligned marker is obtained by fitting a circle through surrounding marker points and intersecting that circle with the relevant grid line. This change was introduced specifically to improve mass conservation and geometric accuracy during reconstruction, and the papers state that the earlier straight-line approximation led to significant mass loss at low resolution [2309.00338][2402.13677].

The 2025 unsplit paper altered the time advancement fundamentally. Instead of repeated directional substeps with reconstruction after each sweep, markers are advected in all directions together over one full timestep, and reconstruction is performed once at the end of the timestep. The paper considers Euler, Predictor-Corrector (Heun), and classical RK4. Its stated motivations are that high-order time integration is difficult under operator splitting because intermediate marker positions are lost, and that multiscale front-tracking models with physical quantities attached to interface elements require better correspondence between pre- and post-advection segments [2509.06661].

The reconstruction itself remained geometric. The unsplit paper retained circle-fit reconstruction because it “significantly improves mass conservation,” but it also modified the earlier averaging of two candidate circle fits. Near the tip of a thin ligament, averaging could produce an artificial bulbous shape. The new ad hoc rule computes the radii of the two candidate circles and, if
$$
\frac{r_{\max}}{r_{\min}} > 10,
$$
discards the smaller-radius fit and uses only the larger-radius circle. The threshold \(10\) is reported as empirical and chosen from kinematic tests [2509.06661].

The unsplit paper also revised the Color Vertex update. A color vertex changes phase if and only if it is swept by an interface segment during the timestep, and this is tested by constructing the polygon traced by the segment endpoints before and after advection and applying a ray-casting point-in-polygon test. The implementation uses a ray in the positive horizontal direction with a tiny slope angle \(\theta = 10^{-32}\) to avoid degeneracies. To restore consistency near cell corners, the paper includes reconstructed intersection points on grid edges in the swept polygon; using only the four advected endpoints could otherwise misclassify a corner [2509.06661].

## 4. Coupling to multiphase flow, surface tension, and phase change

Once the interface geometry is known, EBIT-derived volume fractions provide the bridge to Eulerian multiphase solvers. In the surface-tension and phase-change papers, the workflow is to compute volume fractions from marker positions and the Color Vertex, then derive density and viscosity fields from those fractions, and finally compute surface-tension stresses with the Height-Function method. EBIT therefore does not transport a cell-centered color function directly; it reconstructs geometry first and then derives the volume fraction from that geometry [2309.00338][2402.13677].

In the one-fluid surface-tension formulation, the governing equations are written as incompressible Navier–Stokes with material properties obtained from the phase indicator \(H\):
$$
\rho = H\rho_1 + (1-H)\rho_2, \qquad \mu = H\mu_1 + (1-H)\mu_2.
$$
Surface tension is introduced through the standard continuous-surface-force form
$$
\mathbf{f}_\sigma = \sigma \kappa \mathbf{n}\,\delta_S.
$$
The volume fraction field reconstructed from EBIT data is then used in the same way as in Basilisk’s VOF framework for property interpolation and Height-Function curvature estimation [2309.00338].

The phase-change extension retains the EBIT kinematics but adds a sharp thermal and momentum coupling. The incompressible formulation includes a phase-change source in the divergence constraint,
$$
\nabla \cdot \mathbf{u} = S_{pc},
$$
while the interface temperature is imposed as
$$
T_\Gamma = T_{\text{sat}}.
$$
The interfacial mass flux is obtained from the jump in heat flux, and this mass flux produces a velocity jump across the interface. To handle the resulting discontinuity without oscillations, the paper uses a ghost fluid method rather than a one-fluid source term in the Poisson equation. The momentum equations are solved separately for liquid and vapor with ghost velocities set from the jump condition, and the paper emphasizes this as a key stabilization choice for approximately projected cell-centered velocity fields [2402.13677].

The phase-change paper also states an overall algorithmic loop: solve the advective part of the energy equation; advect the interface with EBIT using the interfacial velocity; solve the diffusive part of the energy equation; compute the heat-flux jump and mass flux and set ghost velocities; then solve the mass and momentum equations with ghost-fluid coupling. This places EBIT as the sharp geometric front representation inside a broader multiphysics solver rather than as a purely kinematic method [2402.13677].

## 5. Benchmark behavior, error metrics, and observed accuracy

Across the EBIT papers, accuracy is assessed with geometric rather than solely field-based diagnostics. The unsplit paper defines the three principal error measures as
$$
E_{\text{area}} = \frac{|A(T)-A(0)|}{A(0)}, \qquad
E_{\text{shape}} = \max_i |\mathrm{dist}(\mathbf{x}_i)|, \qquad
E_{\text{sym}} = |A \triangle B|,
$$
with \(\mathrm{dist}(\mathbf{x}_i)\) taken relative to the analytical interface in circular tests [2509.06661].

The 2022 kinematic paper validated EBIT on translation with uniform velocity, single vortex rotation, and Zalesak’s disk rotation. It reported roughly first-order convergence in shape and area error in the translation test, about first-order convergence in the single-vortex test, and approximately first-order behavior again for area error in Zalesak’s disk. That paper explicitly presented the 2D Cartesian linear method as a “bare bones” EBIT instantiation whose accuracy was only first order in the tested benchmarks [2202.06704].

The 2023 surface-tension paper and the 2024 phase-change paper greatly expanded the benchmark set. In 2D these papers reported validation on stagnation flow, translation with uniform velocity, single vortex, Zalesak’s disk, capillary wave, Rayleigh–Taylor instability, and rising bubble; the phase-change paper added phase-change-specific tests beyond that canonical set. In the translation benchmark, the circle fit improved area preservation substantially over the earlier straight-line version, and the paper states that EBIT’s area error was extremely small and showed second-order convergence, while shape and symmetric-difference errors were roughly first order. In the capillary-wave problem, EBIT coupled with Height-Function curvature matched the expected damped oscillation and showed convergence close to second order in the \(L_2\) error, though the Basilisk VOF method was reported as somewhat more accurate [2309.00338][2402.13677].

The 2025 unsplit paper refined the interpretation of EBIT accuracy. In translation with uniform velocity, split and unsplit-Euler produced nearly identical interface shapes, with second-order area convergence and first-order shape and symmetric-difference convergence. In solid-body rotation, unsplit-Euler was much worse than split, but unsplit-PC and unsplit-RK4 were far more accurate; the paper reports first order for unsplit-Euler, area approximately third order and shape/symmetric second order for unsplit-PC, and fourth order for all errors for unsplit-RK4 until machine precision. In Zalesak’s disk, however, unsplit-PC and unsplit-RK4 did not significantly outperform the split method because sharp-corner reconstruction error dominated. In the single-vortex test, unsplit-PC and unsplit-RK4 recovered the circle much better than split or unsplit-Euler, but the paper states that RK4 was not noticeably better than PC in the end because the dominant error was the circle-fit reconstruction rather than the time integrator [2509.06661].

The 3D EBIT paper added translation with uniform velocity, a 3D deformation test, oscillating drop, rising bubble, and bubble merging. In 3D translation of a sphere, it reports second-order convergence in mass and shape errors, with shape error about one order of magnitude smaller than Basilisk’s PLIC-VOF at all resolutions considered. In the oscillating-drop benchmark, both EBIT and VOF matched the more accurate Prosperetti initial-value solution better than Lamb’s asymptotic solution, but EBIT showed slightly stronger numerical damping. In bubble merging, EBIT reproduced the overall coalescence process, while the onset of merging remained mesh dependent, as expected for topology changes triggered when interface segments enter the same cell [2509.26361].

## 6. Three-dimensional extension and triangle-edge-cut relatives

The 3D extension of EBIT was designed to avoid a wholly new volumetric reconstruction algorithm. The 2025 3D paper generalized the directional split scheme directly: advection within a cubic cell along one direction is decomposed into two 2D advection problems on the two cube faces normal to that direction. This allows reuse of the 2D circle fit and the 2D Color Vertex evolution rules. The paper describes this as a simplified strategy that reuses the 2D interface reconstruction algorithms “fully” in the 3D code path [2509.26361].

In 3D, facewise ambiguity is handled by the same color-vertex logic used in 2D, but the paper also notes an additional internal ambiguity: even when every face is uniquely reconstructed, the cell interior may still admit either two separated interface sheets or a thin ligament-like diagonal connection. The method does not introduce a new 3D interior criterion to resolve this. Instead, it chooses the configuration with separated interface segments and notes that a complete cure would require extra interior information, such as a center vertex or more markers [2509.26361].

For Navier–Stokes coupling in 3D, the paper computes volume fractions using the Front2VOF geometrical method. Given marker positions and color-vertex connectivity, the method groups face-reconstructed interface lines into closed loops, computes the loop centroid
$$
\mathbf{x}_c = \frac{1}{N_m}\sum_{i=1}^{N_m}\mathbf{x}_i,
$$
triangulates the loop by connecting the centroid to successive marker pairs, and then applies the geometrical Front2VOF algorithm via Gauss’s theorem. As in the 2D dynamic formulations, the resulting volume fraction is then used for material properties and Height-Function curvature [2509.26361].

A distinct but closely related development appears in the unstructured-mesh paper “An Interface Tracking Method with Triangle Edge Cuts” [2410.11073]. That method represents the interface in each triangle by edge intersections, stores the interface “encoded compactly with 6 numbers per triangle,” infers polygonal material regions implicitly rather than storing polygon connectivity, advects the interface through geometric queries without an optimization process, and applies an area-correction step to enforce volume conservation. The paper explicitly frames this strategy as a close geometric cousin of EBIT: both are edge-centric, local, and geometry-driven, with topology emerging from how edge cuts connect rather than from a global front graph [2410.11073].

## 7. Relation to VOF, MOF, and principal limitations

EBIT has repeatedly been characterized as methodologically intermediate. The 2022 paper states that it sits conceptually between Front-Tracking, Level-Set, and Volume-of-Fluid ideas: like Front-Tracking, it explicitly follows marker locations; like PLIC-VOF, it reconstructs a piecewise linear interface from those markers; and like Level-Set, marker positions implicitly encode distance information relative to grid vertices. The triangle-edge-cut paper sharpens that positioning further by describing the edge-cut strategy as more geometric than classical VOF but more compact and topology-light than full marker-based front tracking [2202.06704][2410.11073].

The relation to VOF is especially close in the dynamic EBIT papers. EBIT computes cell volume fractions from reconstructed interface geometry and then uses those fractions to obtain density, viscosity, and curvature-related quantities in a VOF-style one-fluid solver. The conservation philosophy is therefore shared, but the stored state differs: VOF stores a cell-centered fraction and reconstructs an interface from it, whereas EBIT stores interface position through markers or edge intersections and derives the fraction from geometry afterward. The triangle-edge-cut paper makes the same contrast on unstructured triangles, noting that its method is closer to a geometrical Lagrangian interface tracker than to a purely Eulerian VOF scheme [2309.00338][2410.11073].

The relation to MOF is stated more narrowly in the triangle-edge-cut work. MOF reconstructs interfaces from volume fraction plus higher moments, typically centroid information. The triangle-edge-cut method does not reconstruct from moments; it uses edge-based geometry directly, which the paper presents as simpler in data requirements and closer in spirit to EBIT/front tracking. A plausible implication is that EBIT-like methods trade moment-rich reconstruction for direct geometric manipulability and local storage [2410.11073].

The limitations reported in the EBIT literature are also consistent across papers. The early 2D Cartesian method was only first-order accurate in the tested benchmarks [2202.06704]. In the later 2D and unsplit formulations, the circle-fit reconstruction improved mass conservation but also became the dominant bottleneck once higher-order time integration was introduced; the unsplit paper explicitly states that RK4 often did not outperform Predictor-Corrector because reconstruction error dominated [2509.06661]. Sharp corners remain challenging, as Zalesak’s disk shows; smoothing near corners persists even at higher order [2309.00338][2509.06661]. Automatic topology change is a major EBIT advantage, but the 2D surface-tension paper notes that the current rule is anisotropic, acting for interfaces approaching parallel to grid lines while diagonal approaches may behave differently, and the 3D paper likewise reports orientation-dependent artifacts near thin skirts and thin films [2309.00338][2509.26361]. The one-marker-per-edge restriction in the later 2D implementation and the unresolved internal ambiguity in the 3D implementation further delimit the current scope of the method [2402.13677][2509.26361].

Taken together, these papers define EBIT as an edge-centric front-tracking framework in which the principal state variables are local edge intersections and local phase labels, not a global front mesh and not a transported diffuse scalar. Its published trajectory shows a consistent attempt to preserve explicit geometric interface information while reducing bookkeeping, enabling almost automatic parallelization in Basilisk, and accommodating topology change through local reconstruction rules rather than global reconnection algorithms [2309.00338][2509.06661][2509.26361].

Source: https://www.emergentmind.com/topics/edge-based-interface-tracking-ebit