WorksPhysics and Astronomy
Newtonian fluid invasion into a Yield-Stress Fluid-Saturated Medium
Abstract
We investigate pressure imposed two-phase flows in porous media where a Newtonian fluid displaces an immiscible Bingham yield-stress fluid. Using a pore-network model, we identify invasion patterns governed by three dimensionless parameters: the pressure-based capillary number , the viscosity ratio , and a number , the ratio of yield stress to capillary forces. Beyond the classical Lenormand regimes (viscous fingering, capillary fingering, stable displacement), we discover two new regimes: a tree-like pattern at low viscosity ratios and a progressively widening column-like pattern at high viscosity ratios. We characterize all regime transitions and establish a breakthrough condition for the invading fluid.
Introduction
The displacement of yield-stress fluids, fluids that require a minimum amount of stress to flow, in porous media in the presence of a second Newtonian phase, such as air or water, is encountered in a wide range of industrial and environmental applications. Notable examples include enhanced oil recovery (EOR) through gas injection, where the mobilization of yield-stress fluids plays a critical role in improving recovery efficiency [1, 2]. Similarly, soil consolidation processes often rely on the injection of cement or polymer slurries to stabilize porous structures [3, 4]. Foam and clay suspensions are also used to create subsurface barriers for pollution containment [5] and to engineer dynamic seals in hydrogen storage reservoirs [6]. In the latter case, Laponite suspensions are employed due to their ability to transition from low-viscosity states during injection to high-viscosity, yield-stress states that effectively trap hydrogen [7]. These processes are governed by complex interactions between capillary, viscous, and yield-stress forces, which determine the flow patterns and displacement efficiency in porous media. Understanding and controlling these mechanisms is essential for optimizing applications ranging from energy storage to environmental remediation. This study aims to investigate the displacement of a yield-stress fluid by an immiscible Newtonian one. More specifically, we aim to characterise the distinct flow patterns that appear depending on the properties of the fluids.
Two-phase flows involving Newtonian fluids in porous media have been the subject of extensive research [8, 9, 10]. In particular, since the seminal work of Zarcone and Lenormand [11], it is now well established that distinct flow regimes emerge depending on the relative dominance of viscous or capillary effects. Notably, they proposed a phase diagram that maps the various invasion patterns as a function of two key dimensionless numbers: the viscosity ratio and the capillary number.
Three distinct flow regimes have been identified in two-phase Newtonian displacements in porous media. When viscous dissipation dominates and the invading fluid is significantly less viscous than the defending fluid, a viscous fingering regime emerges. In this regime, the displacement front destabilizes, resulting in the formation of multiple preferential paths [12, 13, 14]. The resulting growth patterns have been shown to belong to the same universality class as diffusion-limited aggregation [?, 16]. Conversely, when the invading fluid is more viscous than the displaced fluid, the front remains stable, leading to a flat, compact displacement pattern [13, 17, 18]. At low capillary numbers, capillary effects dominate, producing capillary fingering patterns that can be effectively described using invasion percolation models [19, 20, 21]. Recent studies have extended the classical Lenormand phase diagram to account for additional physical effects. Primkulov et al. [22] examined the influence of the contact angle at the fluid-solid interface on flow regime transitions. Geistlinger et al. [23] experimentally characterized how heterogeneity and surface roughness affect flow behavior.
In this study, we propose extending the classical phase diagram to include visco-plastic effects for the invaded fluids. Flows in porous media involving yield-stress fluids have received comparatively little attention in the literature. For instance, in the single-phase case, the existence of a minimum pressure drop, , required to initiate flow is a first consequence of the yield stress. Furthermore, at low applied pressure drops, regions where the fluid remains at rest may coexist with flowing paths. Due to the inherent disorder of the porous medium, each potential flow path has a different activation pressure drop. This results in a hierarchy of preferential flow channels [24, 25, 26, 27, 28].
The two-phase flow case has received even less attention than the single-phase scenario. The most closely related study is that of Chen and Yortsos [29], who employed pore-network simulations to investigate the invasion of a Newtonian fluid into a porous medium initially saturated with a Bingham fluid. However, their analysis relied on two key assumptions. First, they considered the zero-viscosity limit for the invading fluid. Second, they neglected capillary forces. Additionally, due to the complexity of the non-linear equation system and the computational constraints of the time, both the accessible system sizes and the range of physical variables explored were limited. Despite these limitations, Chen and Yortsos identified a novel flow regime in which yield stress dominates viscous forces. This regime is characterized by the formation of a single invading path, similar to the first flowing path observed in the single-phase case. Consequently, they proposed a new region in the phase diagram; however, its boundaries remained poorly resolved due to the aforementioned limitations. More recently, Pourzahedi and Frigaard [30] developed a pore-network model for a system closely related to the present study. Their model incorporates the presence of liquid films and examines the influence of several key parameters, such as pore-size distribution and system size, on breakthrough time and displacement efficiency. However, their approach neglects capillary forces and imposes a fixed flow direction within the network. The invading fluid is also assumed viscous-less, similarly to Chen and Yortsos. Another contribution is the work of Li et al. [31], who investigated invasion patterns using both experimental observations and two-dimensional Lattice Boltzmann simulations under mixed boundary conditions (imposed flow rate at the inlet and fixed pressure at the outlet). Their findings demonstrate that the yield stress modifies the transition between invasion percolation and viscous fingering regimes. Additionally, they observed flow patterns characterized by fewer branches and reduced tortuosity. They propose a phase diagram based on the sweeping efficiency, but do not report any new displacement regimes beyond those already identified in Newtonian fluids. It is important to note, however, that their use of the full Stokes equation in the Lattice Boltzmann framework inherently limits the system size, the number of pores, and the range of physical parameters that can be explored. We also note studies on the displacement of yield-stress fluids in Hele-Shaw cells, which can be analogous to porous media. These studies analyze experimentally and theoretically viscous fingering that can emerge when a yield-stress fluid is displaced by a Newtonian fluid [32, 33, 34]. The competition between capillary and yield forces has also been studied theoretically in [35], by adopting a capillary bundle model for a two-phase non-Newtonian flow and assuming steady-state conditions, new relations for the flow rate-pressure drop characteristic were found.
To the best of our knowledge, a comprehensive understanding of the mechanisms governing invasion patterns in the presence of yield-stress fluids remains elusive, and clear criteria for distinguishing between the different flow regimes are still lacking. The objective of this study is to extend the Lenormand phase diagram to incorporate yield stress behavior in the defending phase. To this end, we employ a dynamic pore-network model, based on the algorithm introduced in Ref. [36], where interfaces between the two phases are represented by menisci that influence the flow. Our results reveal that when yield-stress effects dominate, the system exhibits invasion patterns that deviate from the purely Newtonian case, with the morphology depending on both fluid viscosities and the imposed pressure. Specifically, we observe a tree-like pattern for low-viscosity invading fluids and a column-like pattern for high-viscosity invading fluids, both of which we characterize in detail as functions of the system parameters. The paper is organized as follows. In Section II, we present the pore-network model. Section III derives a breakthrough criterion and describes the flow regimes arising when the invading fluid is much less viscous than the defending fluid. Section IV addresses the opposite limit, where the Newtonian invading fluid is much more viscous than the defending fluid. Finally, Section V summarizes our findings in the form of a phase diagram.
Model/Numerical Method
In this study, we investigate the invasion of a porous medium, initially saturated with a yield-stress fluid, by a less wetting Newtonian fluid. Our approach employs a pore network model, which simplifies the porous structure by representing pores as nodes and pore throats as links. Specifically, we consider a two-dimensional diamond network of width and length , where each node is associated with a pressure (see Figure 1).
Figure 1. Illustration of the pore-network model. The system consists of a diamond lattice of length and width . The invading Newtonian fluid is shown in red, while the defending yield-stress fluid is shown in blue (light blue when immobile, dark blue when mobilized). Each link is assumed to have a hour-glass geometry, and can contain up to five menisci. A pressure is imposed at the bottom boundary, while is imposed at the top boundary, driving flow from bottom to top.
The Newtonian fluid is injected at the bottom boundary of the network. Each link is modeled as a hour-glass shaped conduit with a length and a radius . The radii are randomly sampled from a uniform distribution ranging from to . A global pressure drop is applied along the longitudinal direction , with at the inlet nodes and at the outlet nodes. Mass conservation at each node is governed by the Kirchhoff condition:
where represents the set of neighboring nodes of .
For each link, the presence of two immiscible phases implies the formation of menisci between the yield-stress fluid and the invading Newtonian fluid (see Figure 1). Both the menisci and the yield stress determine the onset flow condition in the link. A creeping laminar-flow regime is assumed, for which inertial terms can be neglected. Following [26, 29, 30], the flow rate of the yield-stress fluid is modeled using a piecewise-linear approximation of Bingham rheology, characterized by a yield stress .
The flow rate -pressure relationship is given by:
where is the hydraulic conductivity of the link, where denotes the Newtonian saturation in the link, while and represent the plastic viscosities of the Newtonian fluid and the yield-stress fluid, respectively. The pressure difference is , and the local yield-stress pressure threshold is given by:
The capillary pressure drop is determined from the location of every meniscus . Following [36], the link radius is assumed to be sinusoidal, thus for a link with menisci:
where , with the surface tension between the two fluids and the contact angle between the capillary wall and the fluids interfaced, assumed to be constant in the whole network. The sign in the sum reflects the orientation of the meniscus. In our simulation, we used for the link geometry. Eq. (3) assumes a straight-tube geometry. However, for sinusoidal or more complex geometries [37, 38], the corresponding pressure threshold differs only by a constant prefactor. This factor can be absorbed into a rescaling of the yield stress without loss of generality.
The algorithm proceeds through the following three steps. First, the pressure and flow fields are computed to satisfy Eq. (1) and Eq. (2) using the Augmented Lagrangian method, as outlined in [39]. The velocity in each link is then used to displace the menisci. The time step, , is determined based on the largest meniscus speed, , which is calculated as follows: . Subsequently, in each node, the volume of wetting and non-wetting fluid from the ingoing links is redistributed proportionally to the flow rates of the outgoing links, following the algorithm described in [36]. Finally, menisci that are close to each other (at a distance less than ) are merged, and a maximum number of menisci (five in our simulations) is enforced.
In the following, we denote average quantities by . We nondimensionalize all quantities using the characteristic length , the characteristic (yield-stress) pressure threshold , and the yield-stress fluid viscosity . This leads to the introduction of several dimensionless numbers:
The viscosity ratio, defined as the ratio of the Newtonian viscosity to the yield-stress (plastic) viscosity:
The pressure-imposed Bingham number:
The ratio of the yield stress to the capillary forces:
For convenience, we also define a capillary number. However, unlike the standard capillary number, which is based on the total flow rate imposed at the inlet [29], we define it in terms of the imposed pressure difference:
where is the minimal pressure difference required to initiate flow.
As discussed in the Introduction, the present model is a continuation of the works of Chen and Yortsos [29], and Pourzahedi and Frigaard [30], which describe the two-phase flow invasion of a Newtonian fluid into a medium saturated with a yield-stress fluid. We note that, although both studies neglect capillary forces, Pourzahedi and Frigaard account for the deposition of films of Bingham fluid on the throat walls when displaced by invading gas. In contrast, our model includes capillary forces at the menisci for partially wetting fluids, but does not account for liquid films. Furthermore, in contrast to the latter work, the flow in our model is not constrained to be directed, a factor which can significantly impact the flow properties, as previously emphasized in [40].
Low viscosity ratio
We first consider the limiting case where the viscosity of the invading Newtonian fluid is much lower than that of the defending yield-stress fluid ().
Critical pressure for breakthrough
The first step in determining the flow regimes is to establish the conditions for flow and breakthrough. This scenario differs from both the single-phase yield-stress case and the two-phase Newtonian case.
In the case of single-phase flow with a yield-stress fluid, the critical pressure corresponds to the onset of the first flowing path. This is determined by minimizing the sum of local pressure thresholds over all possible paths [25, 41].
By contrast, for immiscible fluids, the breakthrough condition is governed by a percolation criterion due to capillary forces [42]. This approach involves identifying the first breakthrough path, where each path becomes accessible once the applied pressure exceeds the largest capillary barrier along that path.
To incorporate both effects, we propose the following argument to determine the minimum breakthrough pressure. We observe that, for a link in the network to be blocked by both capillarity and yield stress, it must first be connected to the inlet by the Newtonian fluid.
Consequently, the key idea is to identify, for each possible flow path , the most restrictive (or blocking) link due to the combined effects of capillarity and the yield stress fluid. Specifically, if the invasion front has reached the link , the minimal condition for further advancement corresponds to an imposed pressure:
Here, represents the set of all possible flow paths in the yield-stress fluid connecting to the outlet (illustrated as the blue paths in Figure 2), excluding any overlap with the already invaded path. For this path, the invasion by the Newtonian fluid proceeds iteratively from the inlet.
Figure 2. Illustration of the procedure for searching for the optimal path, see (10). When a path of Newtonian fluid reaches link , it needs to overcome both the local capillary barrier and the remaining yield stress constraint (path ).
For each path , the critical pressure is identified as the maximum value of along that path. The first path to open is then determined as the one with the smallest critical pressure among all possible paths. This minimal critical pressure, , is thus defined as:
To compute , we employed an adapted version of the min-max algorithm proposed in [42], with the key modification that, at each iteration, the optimal path from a given link to the outlet is determined using Dijkstra’s algorithm [43] while explicitly excluding the previously explored optimal path.
We observe that this formula accurately accounts for the two expected limiting cases. First, if both fluids are Newtonian (i.e., ), the expression reduces to the min-max percolation criterion [42]. Second, if capillary effects are neglected, the maximum threshold along any path will necessarily occur at the first link (connected to the inlet), since all are positive. As a result, the selected path coincides with that of the single yield-stress fluid case, and the critical pressure simplifies to . For example, we recover the MTP expression proposed by Chen and Yortsos [29], since they neglected capillary forces.
In Figure 3 (left), we illustrate the evolution of the non-dimensional critical pressure, , as a function of . Two distinct asymptotic regimes emerge:
Figure 3. Left: Average over 100 realizations of the normalized critical pressure threshold, , as a function of for a system of size . Right: Relative difference between the exact value of , computed using (10), and the upper bound . Note that due the minimization over all possible paths.
When , the critical pressure converges to the percolation threshold, , which is independent of and scales with (orange line).
Conversely, as , the critical pressure approaches , a value proportional to (blue line).
We note that the formulation given in (10) is computationally intensive for large systems. However, an upper bound for the critical pressure can be derived by decoupling the two contributions:
with . In Figure 3, we plot this upper bound (green line) along with (red line) and observe that the latter quantity provides a very good approximation of the true critical pressure , with an error of less than 6% for the distribution shown (Figure 3, right). We emphasize that this result applies to a relatively weakly disordered medium; we expect the approximation error to increase for systems with higher heterogeneity.
The critical pressure threshold defines a region in the plane where breakthrough cannot occur. This prediction is confirmed numerically in Appendix A.
directed tree-like regime
We investigate fluid displacement in the low viscosity ratio limit for pressures exceeding the critical threshold.
Figure 4 presents invasion patterns observed at a fixed low capillary number for varying values of . As increases, the flow morphology evolves significantly.

Figure 4. Snapshots of simulations on a network with and system size , taken at breakthrough for multiple values. For low capillary numbers, the displacement shows a viscous fingering pattern. When yield stress effects become significant, the invasion takes the form of a directed, tree-like pattern.
At low values of , the patterns closely resemble classical viscous fingering, which is consistent with observations in Newtonian fluids at low viscosity ratios (see, for example, [44]). In this regime, multiple heterogeneous fingers develop, with one dominant finger inhibiting the growth of the others.
As increases, the flow undergoes a transition towards a tree-like structure characterised by a primary invasion path. This path progressively branches as it advances from the inlet, with the branches appearing straighter and more directed towards the outlet than in the viscous fingering regime. At high , a significant proportion of the displaced fluid near the inlet remains immobile and is trapped below the yield stress threshold.
Tortuosity: To quantitatively distinguish between these regimes, we propose measuring the directness of the invasion tree by evaluating the tortuosity of the shortest path from the leaves to the root. This tortuosity is determined using a breadth-first search algorithm and is defined as the ratio of the shortest path length from the tip to the root, , to the vertical height of the tip, .
In Figure 5, we plot the evolution of the average branch tortuosity as a function of saturation for various values of . For all cases, the tortuosity initially starts near unity but evolves differently depending on . At low values of , the tortuosity increases continuously with saturation, reflecting the complex branching typical of viscous fingering. In contrast, at high values of , the tortuosity remains close to unity, indicating straighter, more directed invasion paths. We note that this tortuosity property of the invasion pattern is used to distinguish the two phases in the diagram in Section V. We observe that selecting the shortest path makes the tree’s morphology dependent on the structure of the underlying network. In our network, for example, the tree takes the form of a diamond.
Figure 5. Average tortuosity of the Newtonian fluid paths connecting each tip of the invading fluid to the inlet as a function of Newtonian saturation , for and various values of . The system size is .
Qualitative mechanism: To gain insight into the structure of the branching pattern, qualitative reasoning can be employed, as illustrated schematically in Figure 6. Assuming an extremely low viscosity ratio, i.e. that the viscosity of the Newtonian fluid is negligible, the pressure within the Newtonian fluid remains constant and equal to the imposed inlet pressure. The progress of the Newtonian fluid requires flow paths to exist within the yield-stress fluid. In a disordered medium, it is however known that the number of such flow paths depends on the difference between the mean pressure gradient and a threshold pressure gradient (see [26]). Since the pressure is uniform in the Newtonian fluid, the number of paths is thus determined by the remaining distance between the tip and the outlet. Indeed, the mean pressure gradient in the yield-stress fluid scales as , where denotes the height of the tip. As the front advances, the pressure gradient increases, thereby increasing the number of possible flow paths. In particular, at very low flow rates, the initial front consists of a single flow path. However, as increases, the number of flow paths grows, leading to the observed tree-like structure. Furthermore, we note that the flow condition in a yield-stress fluid strongly disfavors tortuous pathways.
Figure 6. Illustration of the branching process typical of the directed tree regime, for three different time steps . The mobilized yield stress fluid is represented is blue, the Newtonian fluid in red. As the Newtonian fluid invades the medium, the pressure is conserved in the fingers but the yield stress to overcome is reduced. The pressure gradient thus increases, causing more paths to flow in the defending phase.
Flow rate-saturation relationship: We now proceed by examining the temporal evolution of both the flow rate and the displacement efficiency during the invasion process. To this end, we first present in Figure 7 (left) the dimensionless flow rate as a function of saturation. The flow rate is non-dimensionalized by scaling with a characteristic flow rate , which corresponds to the typical flow of a homogeneous front in a homogeneous medium (the radii fluctuations are neglected) at very short times. Assuming that the flow initiates only when the pressure difference reaches , it is written :
Figure 7. Left: Normalized flow rate as a function of Newtonian saturation, for , and . When the yield stress effects dominate (), we observe the emergence of a power law behaviour : . This scaling is lost once capillary effects dominate. Middle: Saturation as a function of the most advanced tip’s position . A scaling behaviour is observed. Right: Normalized flow rate as a function of the tree height .
Our observations reveal that the evolution of the flow rate depends strongly on the value of . For low values of , the flow rate remains relatively modest during the early stages of the invasion and then increases gradually until breakthrough. A pronounced acceleration occurs just before breakthrough, which can be attributed to the divergence of the pressure gradient in the remaining defending fluid.
Conversely, for higher values of , the flow rate increases from the onset of the invasion. The flow rate appears to scale with saturation according to a power law. Furthermore, at a given saturation, the flow rate is significantly higher than in cases where is much less than 1. This behavior stems from the directed, tree-like structure of the flow: for instance, at high , the invasion initially proceeds along a single flow path, resulting in a more advanced tip height for a given saturation compared to the low regime. Consequently, this leads to a higher pressure gradient in the remaining fluid domain.
This behaviour is corroborated in Figure 7 (middle), which plots saturation as a function of the most advanced tip position.
Depending on the value of , distinct trends are observed. For high values of , the saturation follows a non-trivial power-law scaling with the height of the tree-like structure, with , suggesting a fractal morphology.
In Figure 7 (right), the evolution of the flow rate is presented as a function of the height of the most advanced tip. Due to the two scaling behaviours described above, the flow rate evolves differently in the high and low regimes. Specifically, the flow rate scales as at high , whereas it remains nearly constant at low values. To qualitatively account for the non-trivial power-law dependence of the flow rate on saturation, we propose a heuristic argument. We assume that the overall flow rate is governed by the flow within each individual tip, denoted as . This quantity is determined by the pressure gradient in the yield-stress fluid over the remaining distance to be invaded. By taking into account both capillary forces and the remaining yield stress, it can be expressed as:
where we introduced the dimensionless numbers defined in Eqs. (8) and (7) with the approximation . Within a specific range of , such that (early stage of the invasion, near the onset of mobilization), the flow rate per tip further reduces to
Moreover, the tree-like structure of the system implies a proportionality between the number of tips and the height of the tree, . Assuming that all tips carrying flow are located at roughly the same height, and combining Eq. (14) with the fractal scaling relation , we derive the scaling law for the total flow rate:
The resulting exponent, , is consistent with the value measured in Figure 7 (left).
Foam transition: Similarly to the observations reported by Lanza et al. [45] in pressure-driven two-phase flows, our experiments also suggest the emergence of a foam phase beyond a certain distance from the inlet. While a detailed investigation of this foam formation is beyond the scope of the present study, we note its occurrence. However, as illustrated in Figure 8, the distance from the inlet at which foam appears seems to decrease with increasing .

Figure 8. Snapshots of simulations performed on systems of size and with (a), (b), (c), and (d).
High viscosity ratio
We now consider the complementary configuration, in which the invading Newtonian fluid has a higher viscosity than the yield-stress fluid (i.e. ).
Figure 9 shows the front displacement for a low capillary number and varying values of . At low and high , the displacement exhibits the classical relatively stable Newtonian front, albeit with small regions of trapped defending fluid due to the presence of capillary forces. As increases, the overall front displacement remains qualitatively similar, although the flow field upstream in the yield-stress region becomes increasingly channelised. However, for sufficiently high values of , the invasion pattern undergoes a radical transition: the invading fluid concentrates into a single column, leaving an entire column of yield-stress fluid at rest.
Figure 9. Snapshots of the simulation at breakthrough, obtained for , and . When the yield stress becomes dominant () and the capillary number is sufficiently low, the displacement front destabilizes due to the formation of preferential flow paths. This process gives rise to a characteristic “column” invasion pattern.
Stable front
In the stable front regime, although the displacement front appears qualitatively similar to the Newtonian case, in the presence of yield stress in the defending fluid, the dynamics of the front undergoes an important difference. Figure 10 (left) shows the front speed as a function of normalized time for various values of . At low , the front velocity decays as a power law, , which is reminiscent of the Washburn dynamics [46] observed for Newtonian fluids. In contrast, at higher , the velocity decreases much more slowly, or even remains approximately constant. The difference in dynamics is further confirmed in Figure 10 (right), which plots the front speed as a function of the front position, using a rescaling that will be discussed later. At low or low , the front speed follows a power-law decay, , but it reaches a plateau at higher values of .
Figure 10. Speed of the invasion front for and various values of , shown as a function of dimensionless time (left) and rescaled interface height (right). The dashed black line is the expected plateau value from Eq. (19). The system size is .
To understand the evolution of the front speed, we adopt an approach analogous to Washburn’s theory, but incorporate the effects of yield stress. We consider a stable, compact invasion front progressing through a homogeneous medium, as illustrated in Figure 11.
Figure 11. Sketch of a homogeneous porous medium: the radii fluctuations are ignored. The Newtonian fluid of viscosity invades a medium of size initially saturated with a Bingham fluid of viscosity . The interface between the two fluids is located at height and is assumed flat.
Given the flatness and compactness of the invasion front, the interface speed is related to the flow rate by . This leads to the following expression for the front velocity:
In the limit where the Newtonian fluid is highly viscous () and the interface height is non-zero, the interface speed can be approximated as:
We define the effective pressure threshold as . The front speed in Eq. (17) comprises two terms: a constant term and a term proportional to , leading to two distinct regimes:
The constant speed regimes occurs when:
The first regime corresponds to Washburn-like dynamics, as observed in Newtonian fluids, while the second regime is governed by the yield stress contribution. Notably, the transition depends on the front height , and for high , it occurs closer to the inlet.
We observe that the power-law exponent (0.6) of the first regime in our results differ slightly from the theoretical prediction 1/2, obtained by solving the ODE in (17) after setting . This discrepancy may arise from the non-compact nature of the displaced fluid. Indeed, a fractal structure would imply with . Nonetheless, this model reasonably captures the transition height and the plateau value observed in the simulations.
Column regime
As described above, for high , the displacement front undergoes a significant change in morphology. Instead of maintaining a relatively flat interface, the flow becomes highly channelized, with a single column of invading fluid advancing alongside a column of undisplaced fluid.
To better understand the physical processes at work, we illustrate the temporal evolution of this regime in Figure 12. At the very early stage, the invasion is confined to only a few links. This behavior arises due to the limited number of flowing channels in the yield-stress fluid, a consequence of the combined effects of high and low . As the displacement progresses, unlike in the tree-like or viscous fingering regimes, the column widens rather than splitting into multiple fingers. This widening is driven by a branching process within the column, where new branches initiate either from the inlet or from the column itself.
Figure 12. Snapshots of the simulation at successive time steps for , and . The system size is . Initially, a single Newtonian column penetrates the medium, after which secondary branches emerge, originating either from the inlet or from previously invaded paths. As a result, the invading column progressively widens over time.
To qualitatively describe the branching process within the column (rather than tip splitting), one may adopt a reasoning analogous to that presented in Section III. In this case however, given that is large, the pressure gradient increases significantly along the column, therefore promoting the emergence of branching flow paths. To illustrate this effect, we consider the pressure gradient along a single invasion path, as depicted in Figure 13. Assuming a homogeneous medium, the pressure difference between the base and the tip of a single flow path is given by:
Figure 13. Sketch of a column formation and widening. A single channel of Newtonian fluid reaches height , until the pressure drop is large enough to activate a new flowing path.
It is assumed here that, beyond the tip, the flow continues through a single mobilized channel within the Bingham fluid. From this expression, it is evident that the pressure gradient is negligible for small , but increases as the invasion progresses thus increasing the probability for emergence of a new flowing branch. A first approximation condition for the onset of branching in a yield-stress fluid can be estimated as
This leads to a threshold distance for the emergence of new branches:
Notably, this reasoning also applies to any two points within the column. However, we note that this prediction results in an overestimation of the branching height. For instance, using the parameters of Figure 12, it predicts that branching would begin only when . The reason for this discrepancy is that the condition given in (22) is overly restrictive, as it neglects the optimization of the selected flow path.
Indeed, it is well established that the activation of new flow paths occurs below the average threshold, such that , where [26, 40, 47]. This is due to the combined effects of disorder and path optimization. Accounting for this correction, we derive the critical height for branching as:
This can be simplified to:
We emphasize that the position of is highly sensitive to the value of , due to the very small slope of near , as illustrated in Figure 14 (left). Using the value , measured for infinite system sizes (see also [40]), yields a significantly smaller , in agreement with our observations. In Figure 14, we plot the value of using the estimation from (25). We observe that the branching height converges to a constant value in the limit of low (high yield stress with respect viscous forces), where this constant increases with decreasing . This is consistent with the previous limit, where for very low , no branching occurs in the invading flow path. As increases, the value of decreases toward zero, implying that the advancement of a single link can trigger flow in a neighboring link. This scenario is thus compatible with a stable front regime. Using this argument, we can estimate the transition between the stable and columnar regimes by setting , which leads to the scaling relation:

Figure 14. Left: Normalized pressure gradient between the inlet and the tip of the invaded channel as a function of the tip’s height from Eq. (21), with , and . The pressure gradient attains the values (green line) and (red line) at radically different heights. Right: Height predicted for the branching processes, using Eq. (25), as a function of the ratio and for multiple values of .
This scaling will be confirmed in the next section.
Phase diagram
In this section, we summarize the different observed regimes as a function of the control parameters by locating them in a three-dimensional phase diagram.
Low viscosity ratio: Figure 15 presents the three distinct regimes identified for across a range of parameters. We observe three transitional regimes. At low , there is a transition between viscous fingering and foam generation, which has been studied in [45] for Newtonian fluids, and occurs at a critical capillary number . Moreover, according to this study, this critical value is expected to depend on the system size.

Figure 15. Phase diagram for . We observe the viscous fingering (I), directed tree (II) and foam (III) regimes.
Another transition occurs at low between viscous fingering and the directed tree regime. The vertical nature of this transition line suggests that it is governed by the competition between yield stress and capillary effects, which dictates the morphology of the invasion pattern. The criterion used to determine the morphology is based on the tortuosity described in section III.
is based on the tortuosity described in section III. We note that this criterion is not sufficiently sensitive to distinguish the ”soft” transition between viscous fingering and invasion percolation. Indeed, at very low , the morphology appears to be slightly closer to invasion percolation. More generally, the transitions between regimes are relatively continuous, making the corresponding transition lines difficult to locate precisely. In particular, simulations performed near the boundary between two regimes often exhibit characteristics of both neighboring invasion patterns. The last transition occurs at high and high , where we observe a transition between the foam and the tree region. This occurs approximately at , which suggests that it is governed by the competition between viscous stress and the yield stress. A possible interpretation would be as follows. Since it is the merging of finger tips that creates the foam [45], foam generation requires a large number of Newtonian finger tips. On the other hand, the yield stress controls the number of tips, as described in Section III, through the ratio .
High viscosity ratio: We now examine the phase diagram for a very high viscosity ratio (), as presented in Figure 16. As discussed previously, three distinct regimes are observed: invasion percolation at low and low , the column regime at low and high , and a stable front at high . The transition between the invasion percolation and column regimes occurs at a constant , highlighting the competition between capillary forces and yield stress. Although the transition between the stable front and invasion percolation regimes is relatively smooth, the transition line is horizontal, indicating that it occurs at a constant (in agreement with Lenormand diagram [11]). Finally, the transition between the stable front and the column regime occurs at a constant ratio, reflecting the competition between yield stress and viscosity. Moreover, the estimation from (26) in the previous section predicts a transition at , which aligns well with our simulation results.
Figure 16. Phase diagram for . We observe the stable front (IV), invasion percolation (V) and column (VI) regimes.
Neutral viscosity ratio: The previous diagrams were constructed for extreme viscosity ratios, where the flow regimes are relatively well defined. Here, we consider the case of equal viscosity for both fluids (), as shown in Figure 17. In this scenario, we still observe the stable front and invasion percolation regimes. However, at high , a hybrid flow regime emerges. Initially resembling the column regime, with a primary path forming and widening at the bottom, this regime then transitions into directed branching at a certain distance from the inlet. This behaviour arises from the combined effect of
Figure 17. Phase diagram for . We observe the invasion percolation regime (IV), an hybrid behavior (II/V) and a stable front regime (VI).
the two previously discussed mechanisms: the presence of a pressure drop within the column and an increasing pressure gradient in the yield-stress fluid region as the front propagates. Similar invasion patterns are observed when considering other intermediate values of the viscosity ratio, see Appendix B for the cases and . Finally, as shown in Appendix C, the invasion patterns reported here do not seem to depend on the specific distribution the radii of the network are sampled from.
Conclusion
In this work, we have investigated the displacement patterns arising when a Newtonian fluid invades a porous medium saturated with an immiscible yield-stress fluid, by imposing a constant pressure drop between the inlet and the outlet. Our analysis reveals two previously uncharacterized flow regimes, governed by the dominance of yield stress over capillary and viscous effects. At low viscosity ratios, a directed tree regime emerges, where the invasion proceeds via successive branching from a primary path. This behavior stems from the assumption of nearly uniform pressure within the invading fluid, with the pattern dictated by the flow paths in the yield-stress fluid ahead of the front. At high viscosity ratios, a column-like regime appears, marked by progressive widening of the invasion path. Here, the mechanism is driven by viscous dissipation-induced pressure drops within the invading fluid, leading to the branching of new flow paths from the inlet. Using a homogeneous model, we derived a critical height for the invading path at which branching occurs.
We further explored the transitions between these new regimes and the established Newtonian regimes, viscous fingering and foam (low viscosity ratio), as well as stable front and invasion percolation (high viscosity ratio). At intermediate viscosity ratios, hybrid regimes arise, where the front may initially resemble a column before transitioning into a tree-like structure. Additionally, we established a breakthrough criterion for the Newtonian fluid as a function of system parameters. To categorize these regimes, we introduced three dimensionless numbers: the capillary number , the viscosity ratio , and , which quantifies the ratio of yield stress to capillary forces.
This study opens up a number of possibilities for further research. For example, fluid configuration could be extended to cases where a yield-stress fluid invades a Newtonian fluid or another yield-stress fluid. While we focused on drainage (non-wetting invading fluid), exploring imbibition (wetting invading fluid) would provide complementary insights. This study has been conducted using a constant pressure drop imposed at the inlet. Other inlet boundary conditions could be explored in future work. This is particularly important for non-Newtonian fluids, since the flow rate-pressure drop relationship cannot generally be inverted as for Newtonian flows. Consequently, different injection conditions may lead to different flow behaviors and possibly to different regimes [48]. Model refinements could incorporate additional physical mechanisms, such as shear-thinning or shear-thickening effects in the invading/defending fluid, liquid films at the walls [30], or visco-elastic effects [34]. Investigating more realistic network topologies may also be important, since the exact shape of the invading structure in the tree regime may depend on the connectivity of the underlying network. Finally, experimental validation of these predictions would be valuable. Two-dimensional drainage experiments could, for example, be performed using a 3D-printed porous Hele-Shaw cell as the porous medium and a Carbopol-water solution as the Bingham fluid.
acknowledgements
This work was partly supported by the Research Council of Norway through the INTPART cell program (project number 309139) and its Centers of Excellence funding scheme (project number 262644). AH also acknowledges funding from the European Research Council (Grant Agreement 101141323 AGIPORE).
References
Validation of the critical pressure
In this appendix, we numerically verify that the critical pressure threshold , defined in (10), provides the correct criterion for the onset of breakthrough. To this end, we perform simulations for both limiting cases of the viscosity ratio , over a range of values of , and for negative values of the capillary number , defined in (8), ranging from to (the negative values are due to the fact we impose ). The resulting invasion patterns are represented in the plane, where the Bingham number
is recalled here for convenience. From the computed value of , we define the corresponding critical Bingham number,
above which breakthrough is not expected to occur. It can be expressed in terms of the other dimensionless numbers as
The results are presented in Figure 18. The critical Bingham number, shown as the solid black line, accurately separates breakthrough from no-breakthrough events over the entire range of parameters investigated, including cases lying very close to the predicted threshold.

Figure 18. Phase diagram for (left) and (right). The black solid line is the prediction for the critical Bingham using , computed from (10). The black dots represent the simulations where we imposed , for which no breakthrough occurs.
Phase diagrams for intermediate viscosity ratios
In this appendix, we briefly examine how the invasion patterns observed in the limiting cases of low and high viscosity ratios evolve for intermediate values of . Figure 19 shows the corresponding phase diagrams for (left) and (right). For , we observe essentially the same regimes as for the case discussed in Section V. The main difference with the diagram is the gradual transition from a stable-front regime towards a foam regime for the high , low region. When yield-stress effects become dominant, we again observe the hybrid directed tree/column regime identified for .
Figure 19. Phase diagram for (top) and (bottom).
For , the invasion patterns are qualitatively similar to those observed in the limiting case . The main difference is that the column regime persists up to larger values of the capillary number . This is consistent with the stabilizing effect of increasing the viscosity ratio, as discussed in Section IV. However, the transition line predicted by (26) no longer provides a quantitatively accurate description, overestimating the transition by approximately one order of magnitude. We attribute this discrepancy to the fact that, for intermediate viscosity ratios, the different invasion regimes are less clearly separated than in the asymptotic limits. The simplified arguments used to derive the transition criterion thus become less accurate.
Rayleigh distribution of radii
In the main text, a uniform distribution for the radii was assumed, with values constrained between two strictly positive bounds. In this appendix, we investigate the sensitivity of the invasion patterns, presented in Section V, to the choice of radius distribution. Specifically, we assess whether the observed patterns are robust to alternative distributions by repeating our simulations using a Rayleigh distribution for the radii:
where . Unlike the uniform distribution, the Rayleigh distribution permits radii arbitrarily close to zero, thereby introducing greater disorder into the medium. The resulting phase diagrams for and are presented in Figure 20.
Figure 20. Phase diagram for (top) and (bottom) with a Rayleigh distribution of radii. The same displacement patterns are observed.
Overall, the phase diagrams remain largely consistent with those obtained under the uniform distribution. However, subtle differences emerge in the boundaries between the observed regimes. Notably, the occurrence of column and tree regimes is more pronounced within their respective parameter spaces. In the low-viscosity limit, this behavior can be attributed to the presence of very small radii in the medium, which increases the pressure cost associated with tortuous flow paths, thereby favoring the formation of directed trees. Additionally, the greater variability in the pressure threshold required to activate new flow paths inhibits the stabilization of a coherent front in the high-viscosity limit.
References
- [1]M.T. Ghannam, S. W. Hasan, B. Abu-Jdayil, and N. Esmail, “Rheological properties of heavy & light crude oil mixtures for improving flowability,” Journal of petroleum science and engineering 81, 122–128 (2012).DOI
- [2]James J. Sheng, “Enhanced oil recovery in shale reservoirs by gas injection,” Journal of Natural Gas Science and Engineering 22, 252–259 (2015).
- [3]A. Seiphoori and M. Zamanian, “Improving mechanical behaviour of collapsible soils by grouting clay nanoparticles,” Engineering Geology 298, 106538 (2022).
- [4]Farhad Zeighami, Alessandro Lenci, and Vittorio Di Federico, “Drainage of power-law fluids from fractured or porous finite domains,” Journal of Non-Newtonian Fluid Mechanics 305, 104832 (2022).
- [5]C. Portois, E. Essouayed, M. D. Annable, N. Guiserix, A. Joubert, and O. Atteia, “Field demonstration of foam injection to confine a chlorinated solvent source zone,” Journal of Contaminant Hydrology 214, 16–23 (2018).DOI
- [6]Behbood Abedi, Alirza Orujov, Ehsan Dabbaghi, Kam Ng, John Ackerman, and Saman A. Aryana, “Containment strategy for subsurface h2 storage based on time-dependent soft solids,” International Journal of Hydrogen Energy 82, 1001–1014 (2024).DOI
- [7]Alirza Orujov, Pablo Godoy, Behbood Abedi, Reza Barati, and Saman A. Aryana, “Engineered barriers for subsurface hydrogen storage: A containment strategy,” Journal of Energy Storage 142, 119525 (2026).DOI
- [8]J. Bear, Dynamics of Fluids in Porous Media, edited by Dover (Elsevier, New York, 1988).
- [9]Francis AL Dullien, Porous media: fluid transport and pore structure (Academic press, 1991).
- [10]M Sahimi, Flow and Transport in Porous Media and Fractured Rock (VCH Verlagsgesellschaft, 1995).
- [11]R. Lenormand and C. Zarcone, “Invasion percolation in an etched network: Measurement of a fractal dimension,” Phys. Rev. Lett. 54, 2226–2229 (1985).
- [12]P. G. Saffman and G. I. Taylor, “The penetration of a fluid into a porous medium or hele-shaw cell containing a more viscous liquid,” Proceedings of the Royal Society of London. A. Mathematical and Physical Sciences 245, 312–329 (1958).
- [13]G M Homsy, “Viscous fingering in porous media,” Annu. Rev. Fluid Mech. 19, 271–311 (1987).
- [14]Santanu Sinha, Yves Méheust, Hursanay Fyhn, Subhadeep Roy, and Alex Hansen, “Disorder-induced nonlinear growth of fingers in immiscible two-phase flow in porous media,” Physics of Fluids 36, 033309 (2024). [15] Knut Jørgen Måløy, Jens Feder, and Torstein Jøssang, “Viscous fingering fractals in porous media,” Phys. Rev. Lett. 55, 2688–2691 (1985).
- [16]J. Mathiesen, I. Proccacia, H. L. Swinney, and M. Thrasher, “The universality class of diffusion-limited aggregation and viscous fingering,” Europhys. Lett. 76, 257–263 (2006).DOI
- [17]Christophe Cottin, Hugues Bodiguiel, and Annie Colin, “Drainage in two-dimensional porous media: From capillary fingering to viscous flow,” Physical Review E 82, 046315 (2010).
- [18]D.W. van Batenburg, S. Berg, S. Oedai, L. L. David, A. O. Siemens, and K. Elewaut, “Visualisation of light oil mobilisation in asp core floods using x-ray ct imaging,” SPE Asia Pacific Enhanced Oil Recovery Conference (2015), 10.2118/174660-MS.DOI
- [19]B. Xu, Y. C. Yortsos, and D. Salin, “Invasion percolation with viscous forces,” Phys. Rev. E 57, 739–751 (1998).
- [20]T. Vincent-Dospital, M. Moura, R. Toussaint, and K. J. Måløy, “Stable and unstable capillary fingering in porous media wit a gradient in grain size,” Commun. Phys. 5 (2022).
- [21]E. Clément, C. Baudet, E. Charlaix, S. Roux, and J-P. Hulin, “Demonstrating invasion percolation with a tabletop experiment,” Eur. Phys. J. Spec. Top. (2025).DOI
- [22]Bauyrzhan K. Primkulov, Amir A. Pahlavan, Xiaojiao Fu, Benzhong Zhao, Christopher W. MacMinn, and Ruben Juanes, “Wettability and lenormand’s diagram,” Journal of Fluid Mechanics 923, A34 (2021).DOI
- [23]H. Geistlinger, S. Golmohammadi, B. Zulfiqar, M. Kuechler, D. Reuter, S. Schlueter, E. Segre, R. Holtzman, and M. Amro, “A new phase diagram for fluid invasion patterns as a function of pore-scale heterogeneity, surface roughness, and wettability,” Water Resources Research 60 (2024).DOI
- [24]L. Talon, H. Araudou, M. Pessel, and A. Hansen, “Geometry of optimal path hierarchies,” EPL (Europhysics Letters) 103, 30003 (2013).
- [25]L. Talon and D. Bauer, “On the determination of a generalized Darcy equation for yield-stress fluid in porous media using a lattice-Boltzmann TRT scheme,” Eur. Phys. J. E 36, 139 (2013).DOI
- [26]C. Liu, A. De Luca, A. Rosso, and L. Talon, “Darcy’s law for yield stress fluids,” Phys. Rev. Lett. 122, 245502 (2019).
- [27]N. Waisbord, N. Stoop, D. M. Walkama, J. Dunkel, and J. S. Guasto, “Anomalous percolation flow transition of yield stress fluids in porous media,” Phys. Rev. Fluids 4, 063303 (2019).
- [28]Vincenzo Maria Schimmenti, Federico Lanza, Alex Hansen, Silvio Franz, Alberto Rosso, Laurent Talon, and Andrea De Luca, “Darcy’s law of yield stress fluids on a treelike network,” Phys. Rev. E 108, L023102 (2023).DOI
- [29]M. Chen, W. Rossen, and Y. C. Yortsos, “The flow and displacement in porous media of fluids with yield stress,” Chem. Eng. Sci. 60, 4183 – 4202 (2005).
- [30]A Pourzahedi and I. Frigaard, “A network model for gas invasion into porous media filled with yield-stress fluid,” Journal of Non-Newtonian Fluid Mechanics 323, 105155 (2024).DOI
- [31]Yu Li, Huijing Liu, Chao Peng, and Qing Wang, “Investigation into invasive pattern evolution in immiscible two-phase flow with bingham behavior in porous media,” Fuel (Lond.) 380, 133243 (2025).DOI
- [32]P. Coussot, “Saffman-Taylor instability in yield-stress fluids,” Journal of Fluid Mechanics 380, 363–376 (1999).
- [33]Anke Lindner, Philippe Coussot, and Daniel Bonn, “Viscous fingering in a yield stress fluid,” Physical Review Letters 85, 314–317 (2000).
- [34]A. Eslami and S.M. Taghavi, “Viscous fingering regimes in elasto-visco-plastic fluids,” Journal of Non-Newtonian Fluid Mechanics 243, 79–94 (2017).
- [35]Federico Lanza, Alberto Rosso, Laurent Talon, and Alex Hansen, “Non-newtonian rheology in a capillary tube with varying radius,” Transp. Porous Media 145, 245–269 (2022).
- [36]Santanu Sinha, Magnus Aa. Gjennestad, Morten Vassvik, and Alex Hansen, “Fluid meniscus algorithms for dynamic pore-network modeling of immiscible two-phase flow in porous media,” Frontiers in Physics 8 (2021), 10.3389/fphy.2020.548497.DOI
- [37]A. Roustaei and I.A. Frigaard, “The occurrence of fouling layers in the flow of a yield stress fluid along a wavy-walled channel,” J. Non-Newtonian Fluid Mech. 198, 109–124 (2013).
- [38]A. Roustaei, T. Chevalier, L. Talon, and I. A. Frigaard, “Non-Darcy effects in fracture flows of a yield stress fluid,” J. Fluid Mech. 805, 222–261 (2016).DOI
- [39]L. Talon and A. Hansen, “Effective rheology of bi-viscous non-Newtonian fluids in porous media,” Front Phys 7, 225 (2020).
- [40]Nathan Abitbol, Alex Hansen, Alberto Rosso, and Laurent Talon, “Flow of yield stress fluid in a percolating network,” Physical Review E 113 (2026), 10.1103/qgv2-3f7r.
- [41]S. Roux and H. J. Herrmann, “Disorder-induced nonlinear conductivity,” Europhys. Lett. 4, 1227 (1987).DOI
- [42]A. Hansen and S. Roux, “Application of ’logical transport’ to determine the directed and isotropic percolation thresholds,” Journal of Physics A: Mathematical and General 20, L873 (1987).DOI
- [43]E. W. Dijkstra, “A note on two problems in connexion with graphs,” Numerische Mathematik 1, 269–271 (1959).
- [44]R. Lenormand, E. Touboul, and C. Zarcone, “Numerical models and experiments on immiscible displacements in porous media.” J. Fluid Mech. 189, 165–187 (1988).DOI
- [45]F. Lanza, S. Sinha, A. Hansen, A. Rosso, and L. Talon, “Transition from viscous fingers to foam during drainage in heterogeneous porous media,” Physics of Fluids 35, 103119 (2023).arxiv.org/abs/2307.13451
- [46]Edward W Washburn, “The dynamics of capillary flow,” Physical review 17, 273 (1921).
- [47]M. Kardar and Y.-C. Zhang, “Scaling of directed polymers in random media,” Phys. Rev. Lett. 58, 2087–2090 (1987).
- [48]L. Talon, A. A. Hennig, A. Hansen, and A. Rosso, “Influence of the imposed flow rate boundary condition on the flow of bingham fluid in porous media,” Phys. Rev. Fluids 9, 063302 (2024).














