betaFeedback

WorksPhysics and Astronomy

Modeling Ostwald Ripening Dynamics in Porous Microstructures

Abstract

Partially miscible ganglia trapped in a porous medium evolve through Ostwald ripening, driven by differences in interfacial curvature. In practice, ganglia can span multiple pores and undergo discrete capillary events - invasion, snap-off, retraction, fragmentation, coalescence, and dislocation - that alter their topology and induce local flow. Existing pore-network models (PNMs) for ripening are limited to single-pore ganglia, assume idealized pore shapes, and operate under quasi-static conditions that preclude flow. We present an image-based pore-network model (iPNM) that removes these limitations. Unlike existing PNMs, iPNM requires no idealization of pore shapes, as the effect on capillarity is encoded locally in curvature-saturation curves computed via the pore-morphology method. iPNM couples two-phase flow, solute transport, and Ostwald ripening within a unified framework. We first verify iPNM against a prior quasi-static PNM, then validate it against recent high-resolution microfluidic experiments of hydrogen ripening in a sandstone-patterned micromodel over 15-24 days at 40C and 80C. Good agreement is obtained without adjustable parameters. Comparison with a continuum model shows that while macroscopic saturation is captured by both approaches, iPNM uniquely resolves population statistics, individual ganglion curvatures, and pre-equilibrium ripening dynamics within a representative elementary volume. Its computational efficiency over direct numerical simulation makes it suitable for guiding the development of improved theories of ripening in confined geometries.

Introduction

Partially miscible ganglia trapped within a liquid-saturated porous medium arise in several energy and environmental applications. In underground hydrogen storage (UHS), hydrogen is cyclically injected into and withdrawn from an aquifer to buffer the intermittency of renewable energy supply. Each injection-withdrawal cycle leaves behind a population of disconnected ganglia whose long-term fate determines storage efficiency [1]. In geological carbon storage, residual trapping of CO2_2 ganglia is a primary mechanism to render the injected CO2_2 hydrodynamically immobile [2, 3]. In polymer electrolyte membrane fuel cells and water electrolyzers, oxygen and hydrogen produced at catalytic sites must traverse porous diffusion layers, and managing trapped gas is central to device performance [4]. In all these settings, trapped ganglia exchange mass and evolve through a process known as Ostwald ripening [5].

Ripening is driven by differences in capillary pressure PcP_c between ganglia. Because PcP_c is proportional to interfacial curvature κ\kappa, and κ\kappa sets the dissolved concentration at a ganglion’s interface by the Kelvin equation, high-κ\kappa ganglia dissolve more rapidly and feed into low-κ\kappa ganglia, causing the latter to grow. In bulk fluids, κ\kappa is inversely proportional to the bubble radius, which drives smaller bubbles to dissolve while feeding the growth of larger bubbles. At equilibrium, only one large bubble remains [6, 7, 8]. Porous confinement fundamentally alters this picture. For a bubble confined to a single pore, κ\kappa first decreases as the bubble grows until it touches the pore’s walls (sub-critical regime), then increases as it deforms under the confining geometry (super-critical regime) [9]. This non-monotonicity admits stable equilibrium of bubble populations that have a multitude of sizes and shapes but equal curvature, impossible in

Email addresses: [email protected] (Md Zahidul Islam Laku), [email protected] (Mohammad Salehpour), [email protected] (Tian Lan), [email protected] (Benzhong Zhao), [email protected] (Yashar Mehmani) free space [10, 11]. In most applications of practical interest, however, ganglia span multiple pores, as evidenced by numerous X-ray imaging studies of rocks undergoing cyclic displacement [3, 12, 13, 14]. Here, the relationship between κ\kappa and ganglion size is far more complex, characterized by oscillatory behavior and multiple discontinuities [15, 16]. Each time a ganglion invades a pore through a narrow throat, κ\kappa drops abruptly since the ganglion occupies a larger pore volume. As the ganglion grows further, κ\kappa climbs again. Growth and shrinkage are therefore punctuated by discrete capillary events such as invasion, snap-off, retraction, fragmentation, coalescence, and dislocation, which alter the ganglion’s topology [16]. Capturing this physics is key to any predictive pore-scale model of Ostwald ripening.

These rich dynamics have motivated a range of theoretical descriptions at different scales. At Darcy scale, continuum models describe ganglia with macroscopic field variables like saturation [17, 18, 19], while equilibrium [11, 20] and kinetic theories [21, 22, 23] describe how the statistics of bubble populations evolve. Pore-scale models play a vital role in formulating and closing such theories. Among them, direct numerical simulation (DNS) provides detailed mechanistic insight by resolving the fluid-fluid interface in full [24, 25]. However, the cost of DNS is prohibitive for studying sufficiently large ganglion populations that yield statistically meaningful information. Pore-network models (PNMs) remain the only viable option, which represent the void space by a computational graph comprised of pore bodies connected by throats. Recently for ripening, a series of PNMs with increasing levels of sophistication have been proposed [10, 11, 16], whose validation has relied primarily on controlled microfluidic experiments [9, 26].

A major gap in the above PNMs for ripening is that they restrict ganglia to a single pore, and the microfluidic experiments used to validate them meet this condition. However as discussed, multi-pore spanning ganglia dominate in practice, and the only PNM to have relaxed this assumption is [16]. But this PNM assumes instantaneous capillary equilibration within each ganglion, does not capture fluid flow, and requires all pores to share the same shape to be tractable. The PNM of [16] also consists of a delicate and path-dependent root-finding procedure to compute each ganglion’s PcP_c, which makes generalization to heterogeneous pore shapes difficult. A key ingredient in PNMs for ripening is a κb\kappa_b–SbS_b curve that relates bubble curvature κb\kappa_b to saturation SbS_b within each pore. The curve is also vital to most dynamic PNMs of two-phase flow [27, 28, 29], used to study fluid-fluid displacement. In these PNMs, κb\kappa_b–SbS_b curves are based on idealized pore geometries (cube, polyhedron) that are amenable to an analytical form. Moreover, this form is held fixed regardless of the connectivity of the non-wetting phase in the pore, an assumption relaxed only in [16]. Another simplification in displacement-centric PNMs is that the sub-critical branch of the κb\kappa_b–SbS_b curve is regularized for numerical robustness. This is inconsequential for forced displacement problems, where viscous forces dominate. However for ripening, where the driving force consists of differences in κb\kappa_b, such regularization can prove catastrophic. For example, if κb\kappa_b is assumed constant on the sub-critical branch [28], spherical bubbles of different sizes would cease to ripen, contrary to experimental evidence [9]. Nevertheless, similar simplifications about capillary pressure within pores have been made in older PNMs used to study ganglion dissolution and pressure dissolution [30, 31, 32].

We present an image-based pore-network model (iPNM) that addresses all the above limitations. iPNM operates directly on binarized images of a porous microstructure, extracting its network and computing distinct κb\kappa_b–SbS_b curves for each pore via the pore-morphology method [33, 34] (modified for ripening [18]). These κb\kappa_b–SbS_b curves apply to arbitrary pore shapes, obviating any geometric simplification. Moreover, iPNM classifies bubble-occupied pores into three types (singleton, terminal, and bridge), each with a distinct κb\kappa_b–SbS_b curve that dynamically adapts to the local connectivity of the non-wetting phase. This divide-and-conquer strategy avoids the analytically intractable task of constructing highly oscillatory capillary pressure versus bubble-size relations for multi-pore ganglia [15], their effects instead emerging implicitly from the pore-wise κb\kappa_b–SbS_b curves. iPNM couples two-phase flow, solute transport, and Ostwald ripening while handling all capillary events a ganglion may undergo. Unlike the quasi-static PNM of [16], no intricate root finding is needed. Finally, iPNM establishes a one-to-one mapping between simulated pore saturations on the network’s graph and the microstructure image, allowing fast calculations on the graph to be transformed into detailed images of phase distribution similar to DNS. We validate iPNM against recent microfluidic experiments [19] of controlled and long-term (15–24 days) measurements of Ostwald ripening in a realistic sandstone microstructure, containing multi-pore H2_2 ganglia. Very good agreement is observed without any adjustable parameters.

The paper’s outline follows: Section 2 describes the problem and governing physics. Section 3 formulates iPNM. Section 4 verifies iPNM against a previous quasi-static PNM [16] in capturing key capillary events. Section 5 validates iPNM against microfluidic experiments [19]. We discuss results in Section 6 and summarize findings in Section 7.

Problem description

We consider a rigid porous medium whose void space is partially occupied by trapped non-wetting ganglia, as shown in Figure 1a. Black represents the solid grains, gray the void space filled with the wetting phase, and white the trapped ganglia. The wetting phase is a liquid in which the gaseous non-wetting phase is partially miscible. Ganglia vary in size and span one or more pores. We assume local thermodynamic equilibrium between each ganglion and the immediately surrounding liquid. The concentration of dissolved non-wetting species at a ganglion’s interface is governed by the Kelvin equation. We also assume bubble composition is uniform and the vapor pressure of the wetting component within each ganglion is constant. Differences in interfacial curvature κ\kappa between ganglia engender concentration gradients in the wetting phase, driving molecular diffusion from regions of high κ\kappa to low κ\kappa. This process is called Ostwald ripening, and it causes high-κ\kappa ganglia to shrink and low-κ\kappa ganglia to grow. We also assume that the capillary pressure variations among ganglia are small relative to the ambient pressure magnitude, rendering the non-wetting density approximately spatially uniform. We justify this assumption later for the experiments in § 5.

Schematic of a porous medium, distance map, and watershed segmentation

Figure 1. (a) Schematic of a porous medium, where solid grains are colored black, trapped ganglia white, and the wetting phase gray. (b) Distance map of the void space with the extracted pore network superposed, represented as a computational graph of nodes (pores) and links (throats). (c) Watershed segmentation partitioning the void space into pores, shown as colored regions. Each node in the network corresponds to a region. The boxed region in (c) illustrates a local pore subimage used to compute the pore-wise κb\kappa_b–SbS_b curve depicted in Figure 3.

Porous confinement fundamentally alters the relationship between κ\kappa and ganglion size compared to bubbles in a bulk fluid, as discussed in § 1. As ganglia evolve during ripening, they undergo discrete capillary events that alter their topology. They include invasion (entering a new pore), retraction (withdrawal from a pore), snap-off (pinching of a meniscus in a throat), fragmentation (splitting into disjoint parts), coalescence (merging of disjoint parts), and dislocation (migration into an adjacent pore). Such topological changes redistribute the non-wetting phase and drive local flow within the wetting phase, even in the absence of an external pressure gradient. Capturing this ripening-induced flow is essential, as it governs the timescale and fate of mass redistribution among ganglia.

Image-based pore network model (iPNM)

Network extraction

iPNM operates on a binarized image of a porous microstructure to extract a pore network: a computational graph made up of nodes (pores) and links (throats) representing geometric enlargements and constrictions of the void space, respectively (Figure 1b). First, a distance map of the void space is computed, assigning to each void pixel its distance to the nearest solid pixel (Figure 1b). Local maxima of this distance map are then identified and used as markers for watershed segmentation [35, 36], which partitions the void into distinct pores shown by colored regions in Figure 1c. The volume of each pore VpV_p is the sum of its constituent pixel volumes, and the maximum inscribed radius of each pore RpR_p is the largest distance-map value in the region. In iPNM, the local subimage of each pore (e.g., boxed region in Figure 1c) is also retained for computing pore-wise κb\kappa_b–SbS_b curves, as described in § 3.4. Throats linking adjacent pores are approximated as prisms with rectangular cross sections. This is appropriate for the topologically 2D micromodel analyzed in § 5, and shown to be an accurate representation in prior work [37]. The in-plane half-width aa of each throat’s cross section (Figure 1c and Figure 2) is the maximum inscribed radius at the constriction, i.e., pixels along the interface shared between two colored regions (Figure 1c). The out-of-plane half-width bb (Figure 2) equals the uniform gap thickness gg of the micromodel divided by two. The length of each throat LtL_t is the sum of the Euclidean distances from the centroid of the constriction to the centroids of the two neighboring pores. We note that iPNM applies equally to networks with 3D topologies as well as other throat geometries. Throats carry no volume in iPNM, i.e., all void volume is assigned to pores. Throats provide the connectivity between pores and hydraulic conductance for flow.

Cross-section of a rectangular throat with half-widths $a$ and $b$

Figure 2. Cross-section of a rectangular throat with half-widths aa and bb. Non-wetting phase (orange) occupies the interior, and wetting phase resides in the corners. The radius of curvature of the corner arc menisci (AMs) is rr and the distance from the contact line to the corner apex is ℓ\ell.

Balance equations

The image-based pore network model (iPNM) is built upon three balance equations formulated for each pore ii: a wetting-phase molar balance, a non-wetting (bubble) phase molar balance, and a dissolved-species molar balance. In the following, we present these equations in their most general form, then simplify them by exploiting inherent timescale separations and order-of-magnitude approximations valid in the dilute limit. The result is a set of numerically tractable equations that can be solved sequentially via operator splitting.

Conservation of the wetting phase in pore ii reads:

ρwVidSw,idt=ρw∑j=1ziqw,ij(1)\rho_w V_i \frac{dS_{w,i}}{dt} = \rho_w \sum_{j=1}^{z_i} q_{w,ij} \tag*{(1)}

where ρw\rho_w is the molar density of the wetting phase, ViV_i is the pore volume, Sw,iS_{w,i} is the wetting-phase saturation, ziz_i is the coordination number, and qw,ijq_{w,ij} is the volumetric flow rate of wetting phase through throat ijij (positive into pore ii). No mass-transfer source appears as the partial pressure of the wetting component in each bubble is assumed constant.

Conservation of the non-wetting component residing in the bubble phase reads:

ρbVidSb,idt=ρb∑j=1ziqb,ij+m˙i(2)\rho_b V_i \frac{dS_{b,i}}{dt} = \rho_b \sum_{j=1}^{z_i} q_{b,ij} + \dot{m}_i \tag*{(2)}

where ρb\rho_b is the molar density of the bubble phase (treated as constant under the incompressibility assumption; see § 2), Sb,i=1−Sw,iS_{b,i}=1-S_{w,i} is the bubble saturation, qb,ijq_{b,ij} is the volumetric flow rate of the bubble phase through throat ijij, and m˙i\dot{m}_i [mol/s] is the interphase mass-transfer rate (positive for exsolution, negative for dissolution).

Lastly, conservation of the dissolved non-wetting species in the wetting phase reads:

d(ρwCiSw,iVi)dt=ρw∑j=1ziwJij−m˙i(3)\frac{d(\rho_w C_i S_{w,i} V_i)}{dt} = \rho_w \sum_{j=1}^{z_i^w} J_{ij} - \dot{m}_i \tag*{(3)}

where CiC_i is the mole fraction of the dissolved non-wetting species and JijJ_{ij} is the solute flux through throat ijij (defined in Section 3.3). The summation runs over the ziw=zi−zibz_i^w = z_i - z_i^b unoccupied throats (filled with wetting phase) connected to pore ii, where zibz_i^b denotes the number of bubble-occupied throats (zib≤ziz_i^b \le z_i). We neglect dissolved-species transport through occupied throats because the wetting phase there is confined to thin corner films whose cross-sectional area is much smaller than that of unoccupied throats. Furthermore, hydrodynamic transport of the non-wetting component through occupied throats via the bubble phase redistributes mass within a ganglion much faster than the transport of dissolved species through corner films. This intra-ganglion redistribution is captured by the flow equations (Eqs. (1)–(2)).

The right-hand side of Eq. (2) contains two processes operating on distinct timescales: bubble-phase flow (qb,ijq_{b,ij}; fast) and interphase mass transfer (m˙i\dot{m}_i; slow). We decouple them via Lie–Trotter operator splitting. Setting m˙i=0\dot{m}_i = 0 in Eq. (2):

ρbVidSb,idt=ρb∑j=1ziqb,ij(4)\rho_b V_i \frac{dS_{b,i}}{dt} = \rho_b \sum_{j=1}^{z_i} q_{b,ij} \tag*{(4)}

Eqs. (1) and (4), combined with the constraint Sw+Sb=1S_w + S_b = 1, constitute the standard incompressible two-phase flow problem, which we solve for Pw,iP_{w,i} and Sb,iS_{b,i} while holding CiC_i frozen.

Setting qb,ij=0q_{b,ij} = 0 in Eq. (2) yields:

ρbVidSb,idt=m˙i.(5)\rho_b V_i \frac{dS_{b,i}}{dt} = \dot{m}_i. \tag*{(5)}

By adding Eqs. (5) and (3), we can eliminate m˙i\dot{m}_i and obtain the following total non-wetting component balance:

ρbVidSb,idt+d(ρwCiSw,iVi)dt=ρw∑j=1ziwJij.(6)\rho_b V_i \frac{dS_{b,i}}{dt} + \frac{d(\rho_w C_i S_{w,i} V_i)}{dt} = \rho_w \sum_{j=1}^{z_i^w} J_{ij}. \tag*{(6)}

For a vacant pore (Sb,i=0S_{b,i} = 0), the first term on the left-hand side of Eq. (6) vanishes, yielding the standard advection–diffusion equation for CiC_i:

ρwVidCidt=ρw∑j=1ziwIij(7)\rho_w V_i \frac{dC_i}{dt} = \rho_w \sum_{j=1}^{z_i^w} I_{ij} \tag*{(7)}

For an occupied pore (Sb,i>0S_{b,i} > 0), we simplify Eq. (6) by first expanding the second term on the left-hand side:

d(ρwCiSw,iVi)dt=ρwVi(Sw,idCidt+CidSw,idt).(8)\frac{d(\rho_w C_i S_{w,i} V_i)}{dt} = \rho_w V_i \left(S_{w,i} \frac{dC_i}{dt} + C_i \frac{dS_{w,i}}{dt}\right). \tag*{(8)}

Because the dissolved molar concentration ρwCi\rho_w C_i is much smaller than the molar density in the bubble phase ρb\rho_b, we neglect the second term on the right-hand side of Eq. (8). Moreover, the concentration CiC_i in an occupied pore is determined by local thermodynamic equilibrium: CiC_i depends on capillary pressure Pc,iP_{c,i}, which in turn depends on the bubble saturation Sb,iS_{b,i} (these functional relationships are presented in § 3.4). Applying the chain rule gives:

dCidt=dCdPcdPcdSbdSb,idt.(9)\frac{dC_i}{dt} = \frac{dC}{dP_c} \frac{dP_c}{dS_b} \frac{dS_{b,i}}{dt}. \tag*{(9)}

Substituting Eq. (9) into Eq. (6), the prefactor ρwSw(dC/dPc)(dPc/dSb)\rho_w S_w (dC/dP_c)(dP_c/dS_b) multiplying the first term on the right-hand side of Eq. (8) becomes negligible compared to ρb\rho_b for our H2_2–H2_2O system. This follows from an order-of-magnitude approximation detailed in Remark 1 of § 3.4. Hence, we can drop the foregoing term, reducing Eq. (6) to:

ρbVidSb,idt=ρw∑j=1ziwJij(10)\rho_b V_i \frac{dS_{b,i}}{dt} = \rho_w \sum_{j=1}^{z_i^w} J_{ij} \tag*{(10)}

Eq. (10) states that the rate of bubble growth is driven entirely by the net solute flux JijJ_{ij} through unoccupied throats. In summary, the balance equations solved in iPNM include Eq. (1) for wetting-phase conservation, Eq. (2) for non-wetting phase conservation, and Eq. (3) for dissolved-species conservation. The flow problem consists of Eqs. (1) and (4) and is solved first for Pw,iP_{w,i} and Sb,iS_{b,i} while holding CiC_i frozen. The transport problem consists of Eqs. (10) and (7) and is solved second for CiC_i and Sb,iS_{b,i} while holding qw,ijq_{w,ij} frozen (computed in the flow problem). A capillary stability check—encompassing invasion, snap-off, and retraction events—is performed after each problem as described in § 3.5.

Throat constitutive equations

We next present constitutive equations for throats with rectangular cross sections. The geometric parameters aa, bb, and LtL_t were defined in § 3.1. Recall that for a micromodel with uniform gap thickness gg, we have b=g/2b = g/2.

Flux laws. For Newtonian fluids at low Reynolds number, the flow rate of phase α∈{w,b}\alpha\in\{w,b\} through throat ijij is:

qα,ij=Kα,ijμα(Pα,j−Pα,i),(11)q_{\alpha,ij} = \frac{K_{\alpha,ij}}{\mu_\alpha}(P_{\alpha,j}-P_{\alpha,i}), \tag*{(11)}

where Kα,ijK_{\alpha,ij} is the hydraulic conductivity of phase α\alpha and μα\mu_\alpha is its dynamic viscosity. The net solute flux through an unoccupied throat ijij, which appears in Eq. (3), comprises advective transport and Fickian diffusion:

Jij=qw,ijCijup+DmAijLij(Cj−Ci),(12)J_{ij} = q_{w,ij}C_{ij}^{\mathrm{up}} + D_m\frac{A_{ij}}{L_{ij}}(C_j-C_i), \tag*{(12)}

where CijupC_{ij}^{\mathrm{up}} is the upwind pore’s mole fraction, DmD_m is the molecular diffusion coefficient, and AijA_{ij} and LijL_{ij} are the cross-sectional area and length of the throat. Below, we replace the ijij subscript with tt for throat properties (e.g., AtA_t, LtL_t).

Single-phase conductivity. Throats that are not invaded by the non-wetting phase are fully occupied by the wetting phase. For an unoccupied throat, Kb=0K_b = 0 and Kw=KtK_w = K_t, where KtK_t is the single-phase conductivity. For a rectangular cross-section, the analytical solution to the Stokes equations yields [38, 39]:

Kt=g~(ε)At2Lt,(13)K_t = \widetilde{g}(\varepsilon)\frac{A_t^2}{L_t}, \tag*{(13)}

where At=4abA_t = 4ab is the throat’s cross-sectional area, LtL_t its length, and:

g~(ε)=(1+1ε)2[13−64επ5∑n=0∞tanh⁡[(2n+1)πε/2](2n+1)5]Gt(14)\widetilde{g}(\varepsilon) = \left(1+\frac{1}{\varepsilon}\right)^2\left[\frac{1}{3}-\frac{64}{\varepsilon\pi^5}\sum_{n=0}^{\infty}\frac{\tanh[(2n+1)\pi\varepsilon/2]}{(2n+1)^5}\right]G_t \tag*{(14)}

the dimensionless conductivity. The aspect ratio and shape factor of the throat are computed via ε=a/b\varepsilon= a/b and Gt=At/Pt2G_t = A_t/P_t^2, respectively, where Pt=4(a+b)P_t = 4(a+b) is the throat’s perimeter. Eqs. (13)–(14) are exact, but approximate closed-form expressions are also available [40]. For a square cross-section, Eq. (14) reduces to g~=0.5623Gt\widetilde{g} = 0.5623G_t. Mehmani and Tchelepi [37] found that Eq. (13) yields network permeability predictions within 10% of direct numerical simulation.

Two-phase wetting conductivity. Throats invaded by the non-wetting phase contain both a central bubble region and wetting films occupying the four corners. The wetting-phase conductivity KwK_w (Eq. (11)) is computed following [41]:

Kw=8g~wℓ4Lt,(15)K_w = \frac{8\widetilde{g}_w\ell^4}{L_t}, \tag*{(15)}

where ℓ=rcos⁡(θ+β)/sin⁡β\ell= r\cos(\theta+\beta)/\sin\beta is the distance from the corner arc meniscus (AM) to the throat corner (Figure 2), β=π/4\beta= \pi/4 is the corner half-angle, and r=σ/Pcr = \sigma/P_c is the AM’s radius of curvature, with σ\sigma denoting the interfacial tension. The:

g~w=exp⁡[a1G~w2+a2G~w+a3+0.02sin⁡(β−π/6)(1/4π−G~w)e1cos⁡2(β−π/6)+2ln⁡A~w],(16)\widetilde{g}_w = \exp\left[\frac{a_1\widetilde{G}_w^2+a_2\widetilde{G}_w+a_3+0.02\sin(\beta-\pi/6)}{(1/4\pi-\widetilde{G}_w)^{e_1}\cos^2(\beta-\pi/6)}+2\ln\widetilde{A}_w\right], \tag*{(16)}

is the dimensionless conductivity of one half-corner film, with dimensionless cross-sectional area:

A~w=(sin⁡βcos⁡(θ+β))2[cos⁡θcos⁡(θ+β)sin⁡β+θ+β−π2].(17)\widetilde{A}_w = \left(\frac{\sin\beta}{\cos(\theta+\beta)}\right)^2\left[\frac{\cos\theta\cos(\theta+\beta)}{\sin\beta}+\theta+\beta-\frac{\pi}{2}\right]. \tag*{(17)}

and corresponding shape factor:

G~w=A~w4[1−(θ+β−π2)sin⁡β/cos⁡(θ+β)]2.(18)\widetilde{G}_w = \frac{\widetilde{A}_w}{4\left[1-\left(\theta+\beta-\frac{\pi}{2}\right)\sin\beta/\cos(\theta+\beta)\right]^2}. \tag*{(18)}

The e1e_1, e2e_2, a1a_1, a2a_2, and a3a_3 are fitting coefficients. In [41], these are provided for no-slip and perfect-slip conditions at the bubble-water interface. We use the perfect-slip coefficients (e1,e2,a1,a2,a3)=(1,0,−18.2066,5.88287,−0.351809)(e_1,e_2,a_1,a_2,a_3)=(1,0,-18.2066,5.88287,-0.351809), appropriate for the gaseous H2_2 bubbles immersed in water as considered in the experiments of § 5.

Two-phase non-wetting conductivity. The non-wetting phase conductivity KbK_b (Eq. (11)) for a rectangular throat is not available in the literature. We derive one here using the hydraulic diameter concept [38], ensuring that Kb→KtK_b \to K_t as Sb→1S_b \to1. The derivation is given in Appendix B. Here we present the final result:

Kb=Dh,b2Ab2PoLt,(19)K_b = \frac{D_{h,b}^{2} A_b}{2 Po L_t}, \tag*{(19)}

where Ab=At−AwA_b=A_t-A_w is the cross-sectional area occupied by the non-wetting phase, with Aw=4ℓ2A~wA_w=4\ell^2\widetilde{A}_w, the total wetting-phase area in the four corners (ℓ\ell and A~w\widetilde{A}_w were defined previously). The hydraulic diameter of the bubble region is Dh,b=4Ab/CbD_{h,b}=4A_b/C_b, where the wetted perimeter, invoking again perfect slip at the bubble-water interface, is:

Cb=4(a−ℓ)+4(b−ℓ).(20)C_b=4(a-\ell)+4(b-\ell). \tag*{(20)}

The Poiseuille number PoPo is computed from the single-phase conductivity:

Po=Dh2At2KtLt,(21)Po=\frac{D_h^2 A_t}{2K_tL_t}, \tag*{(21)}

and assumed to be independent of saturation. Here, Dh=4At/PtD_h=4A_t/P_t is the hydraulic diameter of the throat itself.

Capillary entry pressure. We apply the Mayer-Stowe-Princen (MS-P) theory [42, 43] to derive the capillary entry pressure of throats with rectangular cross-sections. MS-P equates the curvature of the AMs in the corners of the throat to the curvature of the main terminal meniscus (MTM)—the advancing part of the interface. The entry pressure is:

Pce=σr,(22)P_{ce}=\frac{\sigma}{r}, \tag*{(22)}

where rr is the AM radius of curvature. The latter is obtained by solving:

1r=Pe(r)Ae(r),(23)\frac{1}{r}=\frac{P_e(r)}{A_e(r)}, \tag*{(23)}

where PeP_e is the effective perimeter and AeA_e the effective cross-sectional area of the non-wetting phase, both functions of rr. Explicit expressions for Pe(r)P_e(r) and Ae(r)A_e(r) are derived in Appendix § 8 for a rectangular throat with side lengths 2a2a and 2b2b. Eq. (23) is solved numerically using the bisection method. For the case of a throat with square cross-section (a=ba=b), the derived expressions reduce to the analytical solution by [44]. The proof is given in Appendix § 8.

Snap-off pressure. Snap-off occurs when the AMs in adjacent corners of a throat coalesce, at which point any infinitesimal decrease in capillary pressure becomes unsustainable, causing the wetting phase to fill the throat. For a rectangular throat with half-widths aa and bb, snap-off occurs when the distance from each AM’s contact line to the corner of the throat equals min⁡(a,b)\min(a,b) (see Appendix § 8), giving the snap-off pressure:

Pcs=σsin⁡(π/4−θ)min⁡(a,b)sin⁡(π/4).(24)P_{cs}=\frac{\sigma\sin(\pi/4-\theta)}{\min(a,b)\sin(\pi/4)}. \tag*{(24)}

For a square cross-section, this reduces to Pcs=σ(cos⁡θ−sin⁡θ)/aP_{cs}=\sigma(\cos\theta-\sin\theta)/a, as proposed by Ma et al. [44]. For zero contact angle in a rectangular throat, Eq. (24) reduces to Pcs=σ/min⁡(a,b)P_{cs}=\sigma/\min(a,b), which was proposed by Lenormand et al. [45].

Pore constitutive equations

We next present constitutive equations for pores. The pore volume VpV_p and maximum inscribed radius RpR_p were defined in Section 3.1. Beyond these scalar quantities, iPNM requires the local subimage of each pore to compute its κb\kappa_b–SbS_b curve via the pore morphology method (PMM). Each subimage corresponds to a colored region in fig:1c.

Capillary-pressure curve. The capillary pressure PcP_c of a bubble-occupied pore depends on the mean interfacial curvature κb\kappa_b of the non-wetting phase within that pore through the Young-Laplace equation:

Pc=σκb(25)P_c = \sigma\kappa_b \tag*{(25)}

For micromodels with uniform gap thickness gg and zero contact angle, as those in Section 5, (25) reduces to:

κb=κbip+κgap,κgap=2/g(26)\kappa_b = \kappa_b^{ip} + \kappa_{gap}, \qquad\kappa_{gap} = 2/g \tag*{(26)}

where κbip\kappa_b^{ip} is the in-plane curvature and κgap\kappa_{gap} is the constant out-of-plane curvature. Mean curvature κb\kappa_b (or κbip\kappa_b^{ip} in 2D) is a direct function of the pore’s saturation Sb=Vb/VpS_b = V_b/V_p, where VbV_b is the non-wetting phase volume in the pore. The specific form of this κb\kappa_b–SbS_b relation depends on the pore type, which we classify into three categories on the basis of how many invaded throats they connect to (zbz^b): singleton (zb=0z^b = 0), terminal (zb=1z^b = 1), and bridge (zb>1z^b > 1) [16]. These are depicted in Figure 3 along with their corresponding κb\kappa_b–SbS_b curves, which we discuss next for each pore type.

Bubble curvature–saturation relationships for different pore types

Figure 3. Bubble curvature–saturation (κb\kappa_b–SbS_b) relationships for different pore types. (a) Singleton pore: the sub-critical branch (green) corresponds to a spherical bubble that does not touch the pore walls; the super-critical branch (red) is fitted to PMM data (black circles); the yellow square marks the critical point (ScS_c, κb,min⁡\kappa_{b,\min}). Insets show the bubble configuration at different saturations obtained from PMM. (b) Terminal pore: the sub-critical branch (green) intersects the κb\kappa_b-axis at Pce/σP_{ce}/\sigma, corresponding to the entry pressure of the connected invaded throat. Insets show the bubble at the moment of invasion and shortly after. (c) Bridge pore: the interface emerges from one throat and terminates at another, allowing curvatures below κb,min⁡\kappa_{b,\min}. The linear sub-critical branch ensures κb→0\kappa_b \to0 as Sb→0S_b \to0 (flat interface). Insets show possible interface configurations.

Singleton. For a pore containing a single-pore spanning bubble (Figure 3a):

κb={κb,min⁡(ScSb)1/D,Sb<Sc,κb,min⁡(1+λ1Sb−Sc1−Sc+λ2Sb−Sc1−Sb),Sb≥Sc(27)\kappa_b = \begin{cases} \kappa_{b,\min}\left(\dfrac{S_c}{S_b}\right)^{1/D}, & S_b < S_c, \\ \kappa_{b,\min}\left(1+\lambda_1\dfrac{S_b-S_c}{1-S_c}+\lambda_2\dfrac{S_b-S_c}{1-S_b}\right), & S_b \ge S_c \end{cases} \tag*{(27)}

where DD is the network’s dimensionality (= 2 for micromodels). The super-critical branch (Sb≥ScS_b \ge S_c) corresponds to bubbles that are deformed by and conform to the pore’s walls. The sub-critical branch (Sb<ScS_b < S_c) corresponds to bubbles that do not touch the walls and remain spherical (or circular in 2D). The minimum curvature κb,min⁡=2/Rp\kappa_{b,\min}=2/R_p (or 1/Rp1/R_p in 2D) occurs at the critical saturation ScS_c, where RpR_p is the maximum inscribed radius. The critical saturation is Sc=Vc/VpS_c=V_c/V_p, where Vc=4πRp3/3V_c=4\pi R_p^3/3 in 3D (or Vc=πRp2V_c=\pi R_p^2 in 2D). The parameters λ1\lambda_1 and λ2\lambda_2 depend on the pore shape.

Terminal. For a pore that is part of a multi-pore ganglion, with one invaded throat of entry pressure PceP_{ce} (Figure 3b):

κb={(Pceσ−κb,min⁡)(Sc−SbSc)n+κb,min⁡,Sb<Scκb,min⁡(1+λ1Sb−Sc1−Sc+λ2Sb−Sc1−Sb),Sb≥Sc(28)\kappa_b = \begin{cases} \left(\dfrac{P_{ce}}{\sigma}-\kappa_{b,\min}\right)\left(\dfrac{S_c-S_b}{S_c}\right)^n+\kappa_{b,\min}, & S_b<S_c \\ \kappa_{b,\min}\left(1+\lambda_1\dfrac{S_b-S_c}{1-S_c}+\lambda_2\dfrac{S_b-S_c}{1-S_b}\right), & S_b\ge S_c \end{cases} \tag*{(28)}

The super-critical branch is identical to Eq. (27), but the sub-critical branch differs. This is because at the instant when the non-wetting phase invades a terminal pore (i.e., the terminal pore is “born”), capillary pressure equals PceP_{ce} (Figure 3b). Hence, as Sb→0S_b\to0, we must have κb→Pce/σ\kappa_b\to P_{ce}/\sigma. The parameter nn is empirical, which we set to n=2n=2 for smoothness.

Bridge. For a pore that is part of multi-pore ganglion, and connected to multiple invaded throats (Figure 3c):

κb={κb,cSbcSc,Sb<cScκb,min⁡(1+λ1Sb−Sc1−Sc+λ2Sb−Sc1−Sb),Sb≥cSc(29)\kappa_b = \begin{cases} \kappa_{b,c}\dfrac{S_b}{cS_c}, & S_b<cS_c \\ \kappa_{b,\min}\left(1+\lambda_1\dfrac{S_b-S_c}{1-S_c}+\lambda_2\dfrac{S_b-S_c}{1-S_b}\right), & S_b\ge cS_c \end{cases} \tag*{(29)}

The second case (Sb≥cScS_b\ge cS_c) is identical to the super-critical branch of Eq. (27). The first case is a line crossing the origin (Sb,κbS_b,\kappa_b) = (0, 0). The rationale is that in a bridge pore, the interface emanates from one throat and terminates at another, forming a curved surface through the pore whose curvature can fall well below κb,min⁡\kappa_{b,\min} (Figure 3c). Complete flattening of the interface, however, never occurs because one or more of the connected throats would snap off first, transforming the bridge pore into a terminal or singleton. Eq. (29) therefore extrapolates the super-critical formula from ScS_c to cSccS_c, then connects it to the origin to ensure κb→0\kappa_b\to0 as Sb→0S_b\to0. The parameter cc is evaluated as follows:

c={1,c∗>0.10.1,c∗≤0.1(30)c= \begin{cases} 1, & c^*>0.1 \\ 0.1, & c^*\le0.1 \end{cases} \tag*{(30)}

where κb,c=κ(cSc)\kappa_{b,c}=\kappa(cS_c) and c∗c^* solves κb(c∗Sc)=0\kappa_b(c^*S_c)=0, with κb(⋅)\kappa_b(\cdot) corresponding to the super-critical formula. For most pore shapes, c∗≤0.1c^*\le0.1 and c=0.1c=0.1 suffices. Rarely c∗>0.1c^*>0.1, implying κb\kappa_b is zero at Sb>0.1ScS_b>0.1S_c, so we anchor the sub-critical branch at ScS_c by setting c=1c=1. While Eq. (29) for Sb<ScS_b<S_c is empirical, its exact form plays a minor role in simulations.

The super-critical branches of Eqs. 27–29 are identical and depend on pore shape through the parameters λ1\lambda_1 and λ2\lambda_2. A key novelty of iPNM is to use the pore morphology method (PMM), adapted for Ostwald ripening by [18, 19], to compute λ1\lambda_1 and λ2\lambda_2. Doing so avoids making simplifying assumptions about pore shape, allowing accurate upscaling of its impact on capillary pressure. In applying PMM, each pore’s subimage IpI_p is subjected to morphological opening Ip∘BrI_p\circ B_r by a spherical (or circular in 2D) structuring element BrB_r of radius rr. By varying rr from zero to RpR_p, we get a sequence of images that show the ganglion’s configuration in the pore (insets in Figure 3a). Associated with each image is a curvature κb=2/r\kappa_b=2/r (or κbp=1/r\kappa_b^p=1/r in 2D) and a saturation SbS_b, which constitutes the super-critical branch of the κb\kappa_b–SbS_b curve (black circles in Figure 3a). The parameters λ1\lambda_1 and λ2\lambda_2 are fitted to this data (red line in Figure 3a). Notice ScS_c and κb,min⁡\kappa_{b,\min} correspond to r=Rpr=R_p. The super-critical branch (red line in Figure 3a) has a vertical asymptote: as Sb→1S_b\to1, κb→∞\kappa_b\to\infty. This limit is never approached in practice because the ganglion invades a vacant throat in its periphery if κb\kappa_b is sufficiently high. As ganglia evolve during simulation, a pore’s type and thus κb\kappa_b–SbS_b curve changes dynamically.

Equilibrium concentration. The dissolved concentration of the non-wetting species in the wetting phase of a pore is governed by local thermodynamic equilibrium with the bubble phase. For a dilute solution, Henry’s law reads:

C=Pb−PvH=Pw+Pc(Sb)−PvH,(31)C = \frac{P_b-P_v}{H} = \frac{P_w+P_c(S_b)-P_v}{H}, \tag*{(31)}

where CC is the equilibrium mole fraction of the dissolved solute, PbP_b is the bubble-phase pressure, PwP_w is the wetting-phase pressure, and PcP_c is the capillary pressure in the pore. Moreover, PvP_v is the spatially constant vapor pressure of the wetting phase in the bubble, and HH is Henry’s constant. PcP_c depends on bubble saturation SbS_b through Eqs. (25)–(29).

Remark 1. We are now in a position to justify neglecting the term ρwSw(dC/dPc)(dPc/dSb)\rho_w S_w(\mathrm{d}C/\mathrm{d}P_c)(\mathrm{d}P_c/\mathrm{d}S_b) compared to ρb\rho_b in Eq. (10). For the H2\mathrm{H}_2–H2O\mathrm{H}_2\mathrm{O} system of interest here, dC/dPc=1/H\mathrm{d}C/\mathrm{d}P_c = 1/H from Eq. (31). The slope dPc/dSb\mathrm{d}P_c/\mathrm{d}S_b is on the order of the capillary pressure Pc∼σ/RpP_c \sim\sigma/R_p. For typical micromodel pore sizes (Rp∼50R_p \sim50–100 μm100\ \mu\mathrm{m}) and σ≈0.07 N/m\sigma\approx0.07\ \mathrm{N}/\mathrm{m}, this gives dPc/dSb∼103\mathrm{d}P_c/\mathrm{d}S_b \sim10^3–104 Pa10^4\ \mathrm{Pa}. With H≈7.8×109 PaH \approx7.8 \times10^9\ \mathrm{Pa} for H2\mathrm{H}_2 at 25∘C25^\circ\mathrm{C} and ρw≈55.5×103 mol/m3\rho_w \approx55.5 \times10^3\ \mathrm{mol}/\mathrm{m}^3, the term evaluates to ρwSw(dC/dPc)(dPc/dSb)∼10−2\rho_w S_w(\mathrm{d}C/\mathrm{d}P_c)(\mathrm{d}P_c/\mathrm{d}S_b) \sim10^{-2}–10−1 mol/m310^{-1}\ \mathrm{mol}/\mathrm{m}^3, which is orders of magnitude smaller than ρb≈40 mol/m3\rho_b \approx40\ \mathrm{mol}/\mathrm{m}^3 at 1 atm.

Capillary stability

As the balance equations of § 3.2 evolve Sb,iS_{b,i} during a simulation, the occupancy of pores and throats by the non-wetting phase may become unstable, inducing capillary events such as snap-off, invasion, retraction, and ganglion coalescence and fragmentation. At each time step, we check whether the current configuration is stable and, if not, trigger the appropriate capillary events and update the occupancy until a stable configuration is found. Let χi∈{0,1}\chi_i \in\{0,1\} and χij∈{0,1}\chi_{ij} \in\{0,1\} be indicator variables for the occupancy of pore ii and throat ijij by the non-wetting phase, respectively. To evaluate the capillary stability of a throat, we require the throat’s capillary pressure, defined here as:

Pc,ij=Pc,i+Pc,j2,(32)P_{c,ij} = \frac{P_{c,i}+P_{c,j}}{2}, \tag*{(32)}

where Pc,iP_{c,i} and Pc,jP_{c,j} are computed from the κb\kappa_b–SbS_b curves (Eqs. (27)–(29)). Both pores ii and jj straddling the occupied throat are themselves occupied, so Eq. (32) is always well-defined. We check for three types of capillary events:

Snap-off. An occupied throat snaps off when its capillary pressure drops below the snap-off pressure (Eq. (24)):

Pc,ij<Pcs,ijandχij=1⟹χij=0.(33)P_{c,ij} < P_{cs,ij} \quad\text{and}\quad\chi_{ij}=1 \quad\Longrightarrow\quad\chi_{ij}=0. \tag*{(33)}

Snap-off fills the throat with wetting phase but leaves the occupancy and saturation of neighboring pores intact.

Retraction. The non-wetting phase in an occupied pore retracts when its saturation reaches zero:

Sb,i<ϵandχi=1⟹χi=0, Sb,i=0, χij=0 ∀j∈Nib,(34)S_{b,i}<\epsilon\quad\text{and}\quad\chi_i=1 \quad\Longrightarrow\quad\chi_i=0,\ S_{b,i}=0,\ \chi_{ij}=0\ \forall j\in\mathcal{N}_i^b, \tag*{(34)}

where ϵ≈10−16\epsilon\approx10^{-16} is used to avoid excessively small adaptive time steps (discussed in § 3.6). Nib\mathcal{N}_i^b is the set of occupied throats connected to pore ii. Upon retraction, all occupied throats connected to pore ii also become vacant.

Invasion. An unoccupied throat is invaded when the capillary pressure of a neighboring pore at super-critical state (Sb≥ScS_b \ge S_c) exceeds the entry pressure of the throat (Eq. (22)):

max⁡(Pc,i,Pc,j)>Pce,ijandχij=0⟹χij=1.(35)\max(P_{c,i},P_{c,j}) > P_{ce,ij} \quad\text{and}\quad\chi_{ij}=0 \quad\Longrightarrow\quad\chi_{ij}=1. \tag*{(35)}

Only super-critical pores contribute to the maximum in Eq. (35). A sub-critical bubble does not contact the walls of its confining pore and, therefore, cannot invade a throat. If the invaded throat connects to a previously vacant pore jj, that pore also becomes occupied (χj=1\chi_j=1) and is initialized to Sb,j=10−6S_{b,j}=10^{-6}. Similar to retraction, the initialization avoids excessively small time steps. To conserve mass, the corresponding non-wetting-phase volume (10−6Vj10^{-6}V_j) is subtracted from the source pore ii that drove the invasion. This subtraction is bounded so that Sw,iS_{w,i} remains in [0,1][0,1].

Capillary events can trigger one another by altering the occupied coordination number zibz_i^b of pores. For example, snap-off reduces zibz_i^b, which can transform a bridge pore (zib>1z_i^b>1) into a terminal (zib=1z_i^b=1) or a terminal into a singleton (zib=0z_i^b=0). Since each pore type has a distinct κb\kappa_b–SbS_b curve (§ 3.4), the pore’s capillary pressure changes, which Algorithm 1. Capillary stability check

Input: Sb,iS_{b,i}, χi\chi_i, χij\chi_{ij}

Output: Sb,iS_{b,i}, χi\chi_i, χij\chi_{ij}, Pc,iP_{c,i}, Pc,ijP_{c,ij}

Mark all pores and throats as unprocessed

Repeat

Compute pore capillary pressures Pc,iP_{c,i} from Eqs. (27)–(29)

Compute throat capillary pressures Pc,ijP_{c,ij} from Eq. (32)

Snap-off: For each unprocessed throat with χij=1\chi_{ij}=1,

if Pc,ij<Pcs,ijP_{c,ij}<P_{cs,ij}, set χij=0\chi_{ij}=0 and mark as processed

Retraction: For each unprocessed pore with χi=1\chi_i=1,

if Sb,i<εS_{b,i}<\varepsilon, set χi=0\chi_i=0, Sb,i=0S_{b,i}=0, χij=0 ∀j∈Nib\chi_{ij}=0\ \forall j\in\mathcal{N}_i^b, and mark pore and connected throats as processed

Invasion: For each unprocessed throat with χij=0\chi_{ij}=0,

set Pc,i∗=Pc,iP_{c,i}^{*}=P_{c,i} if Sb,i≥Sc,iS_{b,i}\geq S_{c,i}, else Pc,i∗=0P_{c,i}^{*}=0 (same for pore jj)

if max⁡(Pc,i∗,Pc,j∗)>Pce,ij\max(P_{c,i}^{*},P_{c,j}^{*})>P_{ce,ij}, set χij=1\chi_{ij}=1. If a neighboring pore (say jj) is vacant, set it to occupied with Sb,j=10−6S_{b,j}=10^{-6} and subtract 10−6Vj10^{-6}V_j from the source pore. Mark the invaded throat and pore as processed

Until χi\chi_i and χij\chi_{ij} are unchanged

Algorithm 2. Image-based pore network model (iPNM)

Input: Pw,iP_{w,i}, Sb,iS_{b,i}, CiC_i, χi\chi_i, χij\chi_{ij}, TT (simulation time)

Output: Pw,iP_{w,i}, Sb,iS_{b,i}, CiC_i, χi\chi_i, χij\chi_{ij}

time = 0

While time <T<T

[Pw,i,Sb,i,χi,χij,qw,ij]=Flow⁡(Pw,i,Sb,i,χi,χij,Δt)[P_{w,i}, S_{b,i}, \chi_i, \chi_{ij}, q_{w,ij}] = \operatorname{Flow}(P_{w,i}, S_{b,i}, \chi_i, \chi_{ij}, \Delta t) [Algorithm 3]

[Ci,Sb,i,χi,χij]=Transport⁡(Ci,Sb,i,qw,ij,χi,χij,Δt)[C_i, S_{b,i}, \chi_i, \chi_{ij}] = \operatorname{Transport}(C_i, S_{b,i}, q_{w,ij}, \chi_i, \chi_{ij}, \Delta t) [Algorithm 4]

Adapt Δt\Delta t

time = time +Δt+\Delta t

End While

may cause other throats to undergo snap-off or retraction. Conversely, invasion increases zibz_i^b, which can transform a singleton into a terminal or a terminal into a bridge, thereby altering the local capillary pressure and triggering other events. Finding a stable configuration is therefore iterative. Here, we sweep through all pores and throats, check for all three events, then update χi\chi_i, χij\chi_{ij}, and Sb,iS_{b,i} accordingly. Once a pore or throat has been processed in a given sweep, it is not re-examined in subsequent sweeps within the same call. This prevents stagnation due to two or more events undoing one another (e.g., a newly invaded pore immediately flagged for retraction). Iterations terminate when no occupancy changes occur between consecutive sweeps. Algorithm 1 summarizes the steps described above.

Discretization and algorithm

In this section, we describe how the balance equations of § 3.2 are discretized in time and solved algorithmically. We employ backward Euler for time integration and Lie–Trotter operator splitting (introduced in § 3.2) to decouple the flow problem (Eqs. (1) and (4)), solved for Pw,iP_{w,i} and Sb,iS_{b,i} while holding CiC_i frozen, and the transport problem (Eqs. (7) and (10)), solved for CiC_i and Sb,iS_{b,i} while holding qw,ijq_{w,ij} frozen. The two sub-problems are solved sequentially, not iteratively, and accuracy is controlled by an adaptive time step Δt\Delta t. Within each outer time step, the flow and transport solvers may independently progress in sub-steps with smaller increments δt≤Δt\delta t\leq\Delta t. A capillary stability check (Algorithm 1) is performed after each sub-step of both solvers. Algorithm 2 summarizes the overall procedure.

Using Sb,i=1−Sw,iS_{b,i}=1-S_{w,i} and Pb,i=Pw,i+Pc(Sb,i)P_{b,i}=P_{w,i}+P_c(S_{b,i}), the flow problem (Eqs. (1) and (4)) is expressed in terms of two primary unknowns per pore: Pw,iP_{w,i} and Sb,iS_{b,i}. The flow rates qw,ijq_{w,ij} and qb,ijq_{b,ij} are computed from Eq. (11) using the conductivities in

Algorithm 3. Flow solver

Input: Pw,iP_{w,i}, Sb,iS_{b,i}, χi\chi_i, χij\chi_{ij}, Δt\Delta t

Output: Pw,iP_{w,i}, Sb,iS_{b,i}, χi\chi_i, χij\chi_{ij}, qw,ijq_{w,ij}

time = 0; δt=Δt\delta t = \Delta t

While time <Δt< \Delta t

xn,0=[Pw,in;Sb,in]\mathbf{x}^{n,0} = [P_{w,i}^{n}; S_{b,i}^{n}]

For k=1,…,kmax⁡k = 1, \ldots, k_{\max}

Compute F(xn,k)\mathbf{F}(\mathbf{x}^{n,k}) and J(xn,k)\mathbf{J}(\mathbf{x}^{n,k}) from Eq. (36)

Δx=−J−1F\Delta\mathbf{x} = -\mathbf{J}^{-1}\mathbf{F}; xn,k+1=xn,k+Δx\mathbf{x}^{n,k+1} = \mathbf{x}^{n,k} + \Delta\mathbf{x}

If converged, exit

End For

[Pw,in+1,Sb,in+1]=xn,k+1[P_{w,i}^{n+1}, S_{b,i}^{n+1}] = \mathbf{x}^{n,k+1}

Adapt δt\delta t

[Sb,i,χi,χij]=Capillary Stability(Sb,in+1,χi,χij)[S_{b,i}, \chi_i, \chi_{ij}] = \text{Capillary Stability}(S_{b,i}^{n+1}, \chi_i, \chi_{ij}) [Algorithm 1]

time = time + δt\delta t

End While

Compute qw,ijq_{w,ij} from Pw,iP_{w,i}, Sb,iS_{b,i} via Eq. (11)

Applying backward Euler yields the nonlinear system:

F(xn+1)=[Vi(Sw,in+1−Sw,in)−δt∑j=1ziqw,ijn+1Vi(Sb,in+1−Sb,in)−δt∑j=1ziqb,ijn+1]=0(36)\mathbf{F}(\mathbf{x}^{n+1}) = \begin{bmatrix} V_i(S_{w,i}^{n+1}-S_{w,i}^{n})-\delta t\sum_{j=1}^{z_i}q_{w,ij}^{n+1}\\ V_i(S_{b,i}^{n+1}-S_{b,i}^{n})-\delta t\sum_{j=1}^{z_i}q_{b,ij}^{n+1} \end{bmatrix} = \mathbf{0} \tag*{(36)}

where x=[Pw,i;Sb,i]\mathbf{x}=[P_{w,i}; S_{b,i}] for all pores ii. The nonlinearity is due to conductivities KwK_w and KbK_b (Eqs. (15) and (19)) and capillary pressure PcP_c (Eqs. (27)–(29)) being nonlinear functions of Sb,iS_{b,i}. We solve Eq. (36) via Newton’s method with the Jacobian computed by automatic differentiation. Iterations terminate when ∥F∥\lVert\mathbf{F}\rVert, ∥F∥/∥F0∥\lVert\mathbf{F}\rVert/\lVert\mathbf{F}^{0}\rVert, or ∥Δx∥/∥x∥\lVert\Delta\mathbf{x}\rVert/\lVert\mathbf{x}\rVert falls below a tolerance. After each sub-step, a capillary stability check (Algorithm 1) is performed. Algorithm 3 summarizes the flow solver.

The solute flux JijJ_{ij} (Eq. (12)) is a linear function of CC, so the transport equations reduce to a linear system that is solved in a single step without Newton iteration. At each sub-step, the concentration in occupied pores is first set from local thermodynamic equilibrium (Eq. (31)). These values serve as internal boundary conditions for the vacant pores, following the approach of [16]. Applying backward Euler to Eq. (7) for each vacant pore ii gives:

Vi(Cin+1−Cin)=δt∑j=1ziwJijn+1(37)V_i(C_i^{n+1}-C_i^n)=\delta t\sum_{j=1}^{z_i^w}J_{ij}^{n+1} \tag*{(37)}

The saturation in occupied pores is updated explicitly from Eq. (10) using the concentration field at time level nn:

Sb,in+1=Sb,in+ρwρbVi∑j=1ziwJijnδt(38)S_{b,i}^{n+1}=S_{b,i}^{n}+\frac{\rho_w}{\rho_bV_i}\sum_{j=1}^{z_i^w}J_{ij}^{n}\delta t \tag*{(38)}

After each sub-step, a capillary stability check (Algorithm 1) is performed. Algorithm 4 outlines the transport solver. The flow solver (Algorithm 3) adapts δt\delta t based on the bubble saturation Sb,iS_{b,i} in each occupied pore (terminal or bridge). For each pore, the nearest event saturation Sb,ixS_{b,i}^{x} in the direction of Sb,iS_{b,i} change is identified. Events include the pore filling to its capacity Sb,imax⁡S_{b,i}^{\max} (computed as the saturation at which Pc,i=100max⁡(Pce,ij)P_{c,i}=100\max(P_{ce,ij}), where Pce,ijP_{ce,ij} are the entry pressures of the pore’s unoccupied throats), the pore emptying (Sb,i→−10−6S_{b,i}\to-10^{-6}; the slightly negative bound ensures retraction is triggered in Algorithm 1 restoring Sb,i=0S_{b,i}=0), and crossings of critical saturation (Sb,i=ScS_{b,i}=S_c), invasion, and snap-off thresholds. The value of Sb,imax⁡S_{b,i}^{\max} is chosen large enough that invasion into adjacent throats occurs well before

Algorithm 4. Transport solver

Input: CiC_i, Sb,iS_{b,i}, qw,ijq_{w,ij}, χi\chi_i, χij\chi_{ij}, Δt\Delta t

Output: CiC_i, Sb,iS_{b,i}, χi\chi_i, χij\chi_{ij}

time =0= 0; δt=Δt\delta t = \Delta t

While time <Δt< \Delta t

Set Cin+1C_i^{n+1} from Eq. (31) for occupied pores

Compute (dSb,i/dt)n(dS_{b,i}/dt)^n from Eq. (38)

Adapt δt\delta t

Solve Eq. (37) for Cin+1C_i^{n+1} at vacant pores

Update Sb,in+1S_{b,i}^{n+1} at occupied pores from Eq. (38)

[Sb,i,χi,χij]=CapillaryStability⁡(Sb,in+1,χi,χij)[S_{b,i}, \chi_i, \chi_{ij}] = \operatorname{Capillary Stability}(S_{b,i}^{n+1}, \chi_i, \chi_{ij}) [Algorithm 1]

time == time +δt+ \delta t

End While

Sb,iS_{b,i} reaches it, while staying far from Sb,i=1S_{b,i}=1 where the κb\kappa_b–SbS_b curve has a vertical asymptote. End-point saturations Sb,max⁡xS_{b,\max}^{\phantom{x}} and −10−6-10^{-6} are hard limits, and the sub-step is rejected if these limits are violated. The intermediate Sb,ixS_{b,i}^{x} are allowed to overshoot by up to 10−310^{-3}, ensuring that the solver can cross them rather than asymptotically approaching them. If Newton fails to converge, the step is rejected, δt\delta t is halved, and the iterations are restarted. The next sub-step is estimated by the linear extrapolation δtnew=(∣Sb,ix−Sb,in+1∣/∣Sb,in+1−Sb,in∣)δt\delta t_{\mathrm{new}}=(|S_{b,i}^{x}-S_{b,i}^{n+1}|/|S_{b,i}^{n+1}-S_{b,i}^{n}|)\delta t, followed by taking the minimum over all occupied pores. While we check for all event saturations in the simulations of Section 5, we have found empirically that checking for only the end-points is sufficient and accelerates computations with negligible impact on accuracy.

The transport solver (Algorithm 4) adapts δt\delta t somewhat differently, by controlling the saturation increment per sub-step. Concretely, a trial increment ΔSb,i=(dSb,i/dt)nδt\Delta S_{b,i}=(dS_{b,i}/dt)^n\delta t is first computed for each occupied pore, then bounded to lie within prescribed limits [ΔSb,min⁡,ΔSb,max⁡][\Delta S_{b,\min},\Delta S_{b,\max}] (here [10−3,10−1][10^{-3},10^{-1}]). If the trial updated saturation Sb,i∗=Sb,in+ΔSb,iS_{b,i}^{*}=S_{b,i}^{n}+\Delta S_{b,i} falls outside the interval [−10−6,Sb,max⁡][-10^{-6},S_{b,\max}], the increment is further reduced. The sub-step is then back-calculated from δt=min⁡i(ΔSb,i/(dSb,i/dt)n)\delta t=\min_i(\Delta S_{b,i}/(dS_{b,i}/dt)^n) over all occupied pores that exchange solute through at least one unoccupied throat (Eq. (10)). The outer coupling loop (Algorithm 2) employs the same adaptive protocol. If either returns δt=0\delta t=0, indicating stagnation, the outer time step Δt\Delta t is halved and both the flow and transport solvers are restarted.

Visualization

A key feature of iPNM is that the bubble saturations Sb,iS_{b,i} computed for each pore can be mapped directly onto the original pore-space image. For each occupied pore, the PMM subimage corresponding to Sb,iS_{b,i} provides the spatial distribution of the non-wetting phase within that pore (§ 3.4). Because the κb\kappa_b–SbS_b curves are constructed from the actual pore geometry, this mapping is faithful to the underlying microstructure. This is in contrast to traditional PNMs, which approximate pore shapes with idealized geometries (e.g., spheres, cubes) and cannot reconstruct a high-fidelity phase distribution. A naive per-pore mapping, however, produces disconnected fragments for ganglia that span multiple pores (Figure 4a; see red ganglion occupying three adjacent pores). To restore visual connectivity, we first identify which pores belong to the same ganglion via connected-component labeling of the subgraph defined by χi\chi_i and χij\chi_{ij} (§ 3.5). Each ganglion is assigned a unique color (Figure 4). The medial axis (skeleton) of the pore space is then used to bridge fragments across adjacent pores within each ganglion, preserving local curvature and matching the ganglion’s total volume (Figure 4b). Such a connectivity-corrected visualization is used in Section 5. While iPNM does not impose uniform PcP_c within a ganglion, the balance equations naturally drive it toward uniformity, consistent with experimental evidence [19]. Transient deviations arise during capillary events, which induce intra-ganglion flow.

Phase mapping before and after connectivity fix

Figure 4. Visualization of the non-wetting phase distribution mapped onto the original pore-space image. Each ganglion is assigned a unique color. (a) Per-pore mapping using PMM subimages: ganglia spanning multiple pores appear as disconnected fragments (e.g., the red ganglion occupying three adjacent pores). (b) Connectivity-corrected mapping: the medial axis (skeleton) of the pore space is used to bridge fragments across pores belonging to the same ganglion, restoring physical connectivity while preserving the local curvature and total ganglion volume.

Numerical verification

In this section, we verify that the proposed iPNM accurately reproduces key pore-scale phenomena associated with Ostwald ripening, before deploying it to complex domains in Section 5. The verification cases were originally presented in [16] to validate a quasi-static PNM against microfluidic experiments and direct numerical simulations. That PNM enforces instantaneous capillary equilibration within each ganglion at all times, whereas iPNM the finite timescale of this process through its two-phase flow solver. Reproducing the quasi-static predictions ensures that iPNM retains the validated behavior. In all cases, pores are semi-cubic: a pore with inscribed-sphere radius RpR_p and volume (2Rp)3(2R_p)^3. The pore shape is controlled by fitting parameters λ1\lambda_1 and λ2\lambda_2 in Eqs. (27)–(29), which we set to λ1=0.905\lambda_1 = 0.905, λ2=0.01\lambda_2 = 0.01, and n=2n = 2, matching those in [16]. Throats are prismatic with square cross-sections, with side half-length Rt=ξ×min⁡(Rp1,Rp2)R_t = \xi\times\min(R_{p1}, R_{p2}), where ξ\xi is the throat-to-pore aspect ratio, and length Lt=Ls−Rp1−Rp2L_t = L_s - R_{p1} - R_{p2}, where LsL_s is the lattice spacing. Networks have a 2D lattice topology but non-uniform out-of-plane thickness, since each pore’s depth equals 2Rp2R_p. Further details on the network geometry and boundary conditions can be found in [16]. Below, we compare iPNM against the quasi-static PNM [16] for three verification cases: dissolution-induced snap-off (§ 4.1), bubble dislocation (§ 4.2), and ganglion growth (§ 4.3).

Dissolution-induced snap-off

We first examine dissolution-induced snap-off, observed in a microfluidic experiment by Sahloul et al. [46], in which a ganglion spanning two pores dissolves into the surrounding wetting phase. The ganglion fragments into two singleton bubbles that then undergo simultaneous dissolution and ripening. This sequence was successfully captured by the quasi-static PNM of [16], which we now reproduce via iPNM. Following [16], we consider a four-pore network in which a ganglion initially occupies the two central pores and the connecting throat (Figure 5a, Sb=0.45S_b = 0.45). A small perturbation in pore size (∼1%\sim1\%) is introduced to break symmetry. Dissolution is driven by imposing a lower dissolved concentration at the two outer boundary pores, which act as concentration sinks similar to the experiment. A constant wetting-phase pressure is imposed at both boundaries to allow the wetting phase to enter or leave in response to volumetric changes of the bubbles. As shown in Figure 5, both models predict snap-off of the ganglion into two singleton bubbles at Sb≈0.39S_b \approx0.39, followed by continued dissolution and ripening of the fragments. In the quasi-static PNM, capillary equilibrium within the connected ganglion prior to snap-off is enforced instantaneously. In iPNM, this equilibration proceeds dynamically through two-phase flow. As the non-wetting phase redistributes between the two occupied pores, the wetting phase adjusts to accommodate the resulting volume changes. This finite equilibration timescale leads to minor differences in the spatial distribution at later stages (Sb=0.06S_b = 0.06 and Sb=0.04S_b = 0.04).

Figure 5. Comparison of dissolution-induced snap-off and ripening between (a) quasi-static PNM [16] and (b) iPNM. Snapshots show the non-wetting phase distribution (red) and dissolved gas concentration field (color map) at decreasing bubble saturations (annotated on the left). Snap-off occurs at Sb≈0.39S_b \approx0.39, followed by dissolution and ripening of the ganglion fragments. Throats carry no volume in either model.

Bubble dislocation

Mehmani and Xu [16] coined bubble dislocation to describe the pore-to-pore migration of a growing bubble from a small pore into an adjacent large pore. The mechanism arises because, upon invading the large pore, the bubble lacks sufficient volume to be halted by that pore’s walls. Hence, the bubble retracts from the small pore and moves entirely into the large pore, which is energetically favorable. This phenomenon was demonstrated by DNS and observed in microfluidic experiments. The quasi-static PNM of [16] captured this behavior, which we now reproduce with iPNM. Following [16], we consider a four-pore network with pores arranged in ascending size from left to right. A bubble is initially placed in the smallest pore (Figure 6a, Sb=0.01S_b = 0.01). Slightly elevated dissolved concentrations are imposed at boundary pores (not depicted) connected to each end of the network, driving growth via diffusive mass uptake. A constant wetting-phase pressure is imposed at both boundaries to allow the wetting phase to enter or leave. In the quasi-static PNM, the bubble jumps to the next pore at the moment of invasion. In iPNM, this transition has a finite timescale, because the non-wetting phase must flow into the larger pore while the wetting phase flows countercurrent through corner films. Figure 6b shows iPNM reproduces the same sequence of events, with the bubble migrating through progressively larger pores. We note dislocation is challenging to model because it involves two events (invasion and retraction) occurring simultaneously, which taxes the time stepping and Newton iterations of the flow solver.

Figure 6. Comparison of bubble dislocation between (a) quasi-static PNM [16] and (b) iPNM. Snapshots show the non-wetting phase distribution (red) and dissolved gas concentration field (color map) at increasing bubble saturations (annotated on the left). The bubble migrates sequentially from the smallest to the largest pore in a four-pore network. Throats carry no volume in either model.

Ganglion growth

We next simulate ganglion growth in 11×1111 \times11 heterogeneous pore networks for two aspect ratios, ξ=1/4\xi= 1/4 and ξ=3/4\xi= 3/4, following the supplementary material of [16]. Pore sizes are drawn from a uniform distribution with contrast ratio Rp,max⁡/Rp,min⁡=2R_{p,\max}/R_{p,\min} = 2, where Rp,max⁡R_{p,\max} and Rp,min⁡R_{p,\min} are the maximum inscribed-sphere radii of the largest and smallest pores. Boundary throats are made intentionally small to create a capillary barrier that prevents the non-wetting phase from leaving the network. A small bubble is initially placed at the center of the network, and elevated dissolved concentrations are imposed at all boundary pores to drive growth. A constant wetting-phase pressure is imposed at the boundaries, as in § 4.1–§ 4.2. Simulations continue until SbtotS_b^{\mathrm{tot}} reaches 95%. As the ganglion grows, variations in pore and throat geometry cause it to repeatedly fragment and reconnect, producing a complex topological evolution. We track two descriptors of the ganglion to compare iPNM against the quasi-static PNM: (1) Euler characteristic χb=N−L+O\chi_b = N - L + O, where NN is the number of disconnected ganglia, LL is the number of redundant loops, and OO is the number of internal voids (zero for the 2D lattice here). A high positive χb\chi_b entails a fragmented phase, while a low or

negative value entails increased connectivity; (2) Specific area γb=Γb/Vb\gamma_b = \Gamma_b/V_b, where Γb\Gamma_b is the total gas–liquid interfacial area and VbV_b is the total non-wetting phase volume. Both PNMs are run on the same networks with these metrics compared in Figure 7. The models agree on the overall oscillatory behavior of χb\chi_b and γb\gamma_b as functions of SbtotS_b^{\mathrm{tot}}. Agreement is closer for ξ=3/4\xi= 3/4, where growth is continuous and snap-off is suppressed. For ξ=1/4\xi= 1/4, the ganglion fragments and reconnects frequently, so mispredicting a single event can snowball into larger pointwise deviations between the two models. These deviations stem from iPNM resolving the finite timescale of capillary equilibration, which the quasi-static PNM does not. Nevertheless, the overall statistical behavior remains in agreement.

Comparison of ganglion growth between quasi-static PNM and iPNM for two aspect ratios

Figure 7. Comparison of ganglion growth between quasi-static PNM [16] and iPNM for two aspect ratios: ξ=1/4\xi= 1/4 (left column) and ξ=3/4\xi= 3/4 (right column). Top row shows the non-wetting phase distribution (red: quasi-static PNM; green: iPNM). Middle and bottom rows show the Euler characteristic χb\chi_b and specific interfacial area γb\gamma_b, respectively, as functions of total saturation SbtotS_b^{\mathrm{tot}} (black: quasi-static PNM; red: iPNM).

Experimental validation

We next benchmark iPNM against the microfluidic experiments of Salehpour et al. [19], which track the Ostwald ripening of residually trapped hydrogen ganglia in a silicon-glass bonded micromodel patterned from a 2D X-ray CT image of a Canadian sandstone. The micromodel measures 10.19 × 4.25 mm in the horizontal plane with a uniform gap thickness of g=7.8g = 7.8 µm, and has a median pore size of 42 µm, a median throat size of 10.6 µm, and a porosity of 0.3. These dimensions introduce sufficient pore-scale heterogeneity for the iPNM to demonstrate its capability in handling realistic pore geometries. For a detailed description of the experimental apparatus and imaging protocol, see [19]. A schematic of the experimental setup and a representative raw image of the micromodel are shown in Figure 8.

Experimental setup and microscope image

Figure 8. Experimental setup (top) showing the silicon-glass bonded micromodel, high-pressure syringe pump, and temperature controller. Raw microscope image (bottom) of the micromodel showing residually trapped hydrogen bubbles (white) within the sandstone pore space (gray) at t=1t = 1 hr in the 40HS case. The wider inlet and outlet channels on the left and right edges contain large bubbles that act as concentration sinks.

Experiments were conducted at two temperatures (40°C and 80°C). Hydrogen was first injected into the water-saturated micromodel, followed by water reinjection to residually trap the gas as disconnected bubbles within the pore space. Large bubbles were deliberately retained in the wider inlet and outlet channels (Rl=50R_l = 50 µm and Rr=85R_r = 85 µm, respectively) so that they served as concentration sinks, sustaining diffusive mass transfer from the porous matrix toward the channels. Four experimental cases are considered, combining two temperatures with two levels of initial gas saturation: 40°C low-saturation (40LS), 40°C high-saturation (40HS), 80°C low-saturation (80LS), and 80°C high-saturation (80HS). The initial gas saturations range from 0.47 to 0.67, and experiments lasted 15–24 days.2

We compare iPNM against both the experimental data and the continuum model of [19]. The continuum model predicts the spatiotemporal evolution of dissolved concentration and gas saturation along the longitudinal direction of the micromodel. It couples Henry’s law with a PcP_c–SbS_b relationship obtained by applying PMM to the entire micromodel image. Unlike iPNM, the continuum model assumes local capillary equilibrium within each representative elementary volume (REV) and therefore cannot resolve individual bubble curvatures or population statistics. The model equations and numerical implementation are given in Appendix § 10. In what follows, we first describe the iPNM simulation setup (§ 5.1), then compare macroscopic quantities (§ 5.2) and population statistics (§ 5.3).

iPNM simulation setup

The pore network is extracted from the binary image of the micromodel via marker-based watershed segmentation (§ 2). For each pore ii, the volume ViV_i and maximum inscribed radius Rp,iR_{p,i} are computed, and PMM is applied pore-wise to construct the κb\kappa_b–SbS_b curve as described in § 3.4. Throats are modeled as rectangular cross-sections with in-plane half-width aa equal to the inscribed radius at the constriction, out-of-plane half-width b=g/2b = g/2, and length LtL_t equal to the sum of Euclidean distances from the constriction to the centroids of its two neighboring pores.

The initial saturation field is obtained by mapping bubble volumes segmented from the experimental image at t=0t = 0 onto the extracted pore network. Dirichlet BCs on CC are imposed at the left and right boundary throats by evaluating Henry’s law (Eq. (31)) with the capillary pressure of an interface spanning each channel’s half-width, Pc=σ(1/Rl+2/g)P_c = \sigma(1/R_l + 2/g), consistent with Appendix § 10. The remaining boundaries are treated as no-flux. For the flow problem, a constant wetting-phase pressure Pw=1P_w = 1 atm is imposed at the left and right boundaries, with no-flow on the remaining sides, so that gas redistribution is driven entirely by diffusive mass transfer. The non-wetting phase molar density is computed from the ideal gas law as ρb=Pw/(RgT)\rho_b = P_w/(R_gT) and held constant throughout. This is because capillary pressure variations are small relative to the ambient pressure: Pc/Pw≈3%P_c/P_w \approx3\% for the median pore size here. The water viscosity μw\mu_w is computed from the correlation of Reid et al. [47], and the hydrogen viscosity μb\mu_b is linearly interpolated between reference values at 300 K and 400 K from Mehl et al. [48]. All fluid properties are listed in Table 1.

PropertySymbol40°C80°CUnit
Molecular diffusivityDmD_m7.34×10−57.34 \times10^{-5}1.53×10−41.53 \times10^{-4}cm²/s
Surface tensionσ\sigma68.962.6dyn/cm
Henry’s constantHH7.51×10107.51 \times10^{10}7.55×10107.55 \times10^{10}dyn/cm²
Vapor pressurePvP_v7.36×1047.36 \times10^{4}4.73×1054.73 \times10^{5}dyn/cm²
Water viscosityμw\mu_w6.70×10−36.70 \times10^{-3}3.60×10−33.60 \times10^{-3}poise
H₂ viscosityμb\mu_b9.16×10−59.16 \times10^{-5}9.96×10−59.96 \times10^{-5}poise
Water molar densityρw\rho_w0.055mol/cm³
H₂ molar densityρb\rho_b3.89×10−53.89 \times10^{-5}3.45×10−53.45 \times10^{-5}mol/cm³

Table 1. Fluid properties for each experimental temperature.

Macroscopic behavior

Total saturation. We compare the temporal evolution of the total non-wetting phase saturation SbtotS_b^{tot} in the micromodel across all four experimental cases (Figure 9). In all cases, SbtotS_b^{tot} declines monotonically, driven by diffusive transport toward the boundary channels. Both iPNM and the continuum model capture this trend. iPNM shows strong agreement with experiments, except for some deviation at later times in the 80HS case. The deviation is attributed primarily to non-ideal BCs at the channels. In iPNM, the concentration at each channel is set uniformly based on the capillary pressure of an interface spanning the channel’s half-width. In the experiments, however, the bubble distribution in the channels is highly irregular (Figure 8), and in some cases low-curvature interfaces form along (rather than across) the channel, creating locally stronger sinks than assumed (see supplementary videos). A likely secondary contributor is the presence of condensation droplets within trapped ganglia, which are most prevalent in the 80HS case (yellow in Figure 13). These droplets often condense and re-evaporate cyclically, with non-trivial effects on concentration and capillary pressure fields that are not captured by iPNM. Both sources of error are discussed further in § 6.3.

Temporal evolution of total non-wetting phase saturation for all four experimental cases

Figure 9. Temporal evolution of total non-wetting phase saturation SbtotS_b^{tot} for all four experimental cases. Red squares denote experiments, the black line is the iPNM prediction, and the green line is the continuum model. Cases correspond to 40°C and 80°C with low and high initial SbtotS_b^{tot}.

Spatial distributions for the 80HS case

Figure 13. Spatial distributions for the 80HS case. Left: phase distribution (red: non-wetting, blue: wetting, black: solid, yellow: condensation). Right: ganglia colored by capillary pressure (dyn/cm2^2) and wetting phase by dissolved concentration (mole fraction). The top row shows the experimental initial condition used to initialize the simulation. Subsequent rows compare experiments and iPNM at T=24T = 24 and 288 hrs.

Mean curvature. Next, we compare the temporal evolution of the mean in-plane radius of curvature rˉ\bar{r} in Figure 10. In experiments, rˉ\bar{r} is calculated by fitting circles to individual bubble interfaces extracted from acquired images using least-squares minimization, with quality control to exclude nearly straight arcs and interfaces with insufficient pixel coverage. These per-pore radii are averaged across each ganglion, then across all ganglia. In iPNM, the per-pore radii are obtained by evaluating the κb\kappa_b–SbS_b curves, subtracting the out-of-plane contribution κgap=2/g\kappa_{\mathrm{gap}} = 2/g, then computing the reciprocal of the resulting in-plane curvature. The per-pore radii are averaged across ganglia similar to the experiments. In the continuum model, rˉ\bar{r} is derived by mapping SbtotS_b^{\mathrm{tot}} to a capillary pressure via the global PcP_c–SbS_b curve, then subtracting the out-of-plane contribution, and taking the reciprocal of the resulting in-plane curvature.

Temporal evolution of the mean in-plane radius of curvature for four experimental cases

Figure 10. Temporal evolution of the mean in-plane radius of curvature rˉ\bar{r} for all four experimental cases. Red squares denote experiments, the black line is the iPNM prediction, and the green line is the continuum model. Cases correspond to 40°C and 80°C with low and high initial SbtotS_b^{\mathrm{tot}}.

In all cases, Figure 10 shows that rˉ\bar{r} increases monotonically over time. iPNM yields close quantitative agreement with experiments across all cases, though it tends to slightly underestimate rˉ\bar{r} (particularly at 80°C and at late times for 40°C). This is partly attributable to the PMM-based κb\kappa_b–SbS_b curves, which overestimate curvature for multi-pore ganglia, as examined further in Section 5.3. By contrast, the continuum model overestimates rˉ\bar{r} at early and intermediate times, but approaches experimental data at late times. This arises because individual ganglion curvatures are not resolved, and SbtotS_b^{\mathrm{tot}} is mapped to a single curvature value using the global PcP_c–SbS_b curve. This misses the small, high-curvature, and out-of-equilibrium bubbles that can pull the true population mean rˉ\bar{r} downward. As these bubbles vanish over time, the curvature distribution narrows and the continuum model approaches the experiment.

Pore-scale behavior

The macroscopic quantities of the previous section mask pore-scale information about ganglia. Here, we validate iPNM against experiments by comparing probability distributions of ganglion curvatures and spatial configurations of the non-wetting phase, capillary pressure, and dissolved concentration. The continuum model is inherently incapable of providing ganglion-level information, so its mean-property predictions are juxtaposed where appropriate.

Curvature distribution. Figure 11 compares the PDF of radius of curvature of ganglia at three times (T=4T = 4, 192, and 360 hrs). In all cases, we see that the experimental PDFs shift toward larger rr while simultaneously broadening. This reflects two concurrent processes governing ripening in the presence of boundary sinks [19]: (1) Local ripening, which is driven by the interaction between adjacent ganglia, forcing the PDFs to sharpen into a Dirac delta. Salehpour et al. [19] showed local ripening lasts ∼5\sim5 hrs in the central region of the micromodel at 80°C, consistent with the ∼24\sim24 hr plateau of rˉ\bar{r} in Figure 10 for the full domain; (2) Global ripening, which is driven by diffusive mass loss to the channels. As most ganglia are supercritical (deformed), mass loss lowers curvature and shifts the PDF to larger rr.

PDFs of the mean radius of curvature of ganglia at three times for all four experimental cases

Figure 11. PDF of the mean radius of curvature of ganglia at three times (T=4T = 4, 192, and 360 hrs) for all four experimental cases. Red lines denote experimental data, black lines the iPNM predictions, and vertical green lines the continuum model’s prediction of rˉ\bar{r}.

iPNM captures the peak positions and shifts of the PDFs across all cases. The agreement is strongest for 40°C, but deteriorates for 80°C at intermediate and late times. At 80°C, the experimental PDFs are broader than iPNM and have more pronounced tails toward small rr. We attribute this primarily to the irregular bubbles in the inlet and outlet channels (Figure 8), whose effective curvatures—thus the boundary concentrations they impose—differ from the uniform values assumed in iPNM. A possible secondary factor is the higher prevalence of water condensation droplets at 80°C, whose cyclic condensation and re-evaporation alters concentration and capillary pressure fields in ways not captured by iPNM. In Figure 11, the continuum model is shown as vertical lines denoting rˉ\bar{r}, which fall consistently to the right of experimental and iPNM peaks, except at late times. This explains the overestimation of rˉ\bar{r} in Figure 10 and is consistent with the assertion that the continuum model does not capture pre-equilibrium dynamics of ripening in an REV.

Spatial distribution. Figure 12–Figure 13 compare the spatial distribution of trapped ganglia between experiments and iPNM at T=24T = 24 and 288 hrs for the 40HS and 80HS cases. The experimental initial condition used to initialize the simulations is shown in the top row of each figure. These two cases are selected because they illustrate the range of agreement, from strong in 40HS to weak in 80HS. The remaining two cases are included in the supplementary material. The iPNM results are rendered using the connectivity-corrected visualization described in § 3.8 (Figure 4), which maps simulated pore saturations onto the micromodel’s pore-scale image. The continuum model does not resolve individual ganglia and is therefore excluded from this comparison.

Spatial distributions for the 40HS case

Figure 12. Spatial distributions for the 40HS case. Left: phase distribution (red: non-wetting, blue: wetting, black: solid, yellow: condensation). Right: ganglia colored by capillary pressure (dyn/cm2^2) and wetting phase by dissolved concentration (mole fraction). The top row shows the experimental initial condition used to initialize the simulation. Subsequent rows compare experiments and iPNM at T=24T = 24 and 288 hrs.

In Figure 12–Figure 13, the ganglia near the channels dissolve more rapidly due to their proximity to concentration sinks, creating a depletion zone that expands inward over time [18]. After 24 hrs, the depletion zone is already visible in the experiments, and by 288 hrs, it has expanded significantly. iPNM captures this dissolution front, though the agreement is stronger in 40HS than in 80HS. In the 80HS case, the left portion of the micromodel depletes faster in the experiment than in iPNM, likely due to a low-curvature interface in the bottom part of the left channel that acts as a stronger sink than the uniform-concentration BC assumed in iPNM (see supplementary videos). The right columns of Figure 12–Figure 13 compare capillary pressure and dissolved concentration fields. The latter is obtained by solving the Laplace equation over the wetting phase, with concentrations at ganglia set by Henry’s law and used as Dirichlet BCs. In 40HS, these fields from iPNM are in good agreement with experiments. In 80HS, the discrepancy is more pronounced, consistent with idealized BCs at channels and higher prevalence of condensation droplets discussed above. Another contributing factor to the more rapid depletion observed in iPNM is the PMM curvature bias, which we discuss next.

PMM curvature bias. The PMM-derived κb\kappa_b–SbS_b curves used by iPNM overestimate curvature for pores belonging to multi-pore ganglia, contributing to the underestimation of rˉ\bar{r} seen in Figure 10–Figure 11. Figure 14 demonstrates this visually by comparing experimental phase distributions against PMM reconstructions at the same pore locations. Red circles drawn tangent to various interfaces reveal smaller radii in PMM than in experiments. This arises because PMM constructs the κb\kappa_b–SbS_b curve for each pore by morphologically eroding its subimage independently. During erosion, connected throats are progressively emptied of the non-wetting phase, so the resulting bubble configuration is confined entirely to the pore body (Figure 3). For multi-pore ganglia, however, some of these throats are occupied and hold part of the ganglion’s volume. By excluding this throat volume, PMM effectively compresses the same amount of non-wetting phase into a smaller space, producing more deformed interfaces with higher curvature than observed.

Comparison between experimental phase distributions and PMM reconstructions at select pore locations

Figure 14. Comparison between experimental phase distributions and PMM reconstructions at select pore locations. Red circles are drawn tangent to the gas–water interface. PMM tends to yield smaller radii, indicating a higher curvature than observed experimentally.

Discussion

Continuum vs. pore-scale modeling

We showed in Section 5.2 that the continuum model of [19] captures SbtotS_b^{tot} well across all four experiments. However, three fundamental limitations distinguish it from iPNM. First, the continuum model cannot provide pore-scale information about the spatial position, size, curvature, or population statistics of individual ganglia, whereas iPNM resolves all of these by tracking each ganglion explicitly through pore-scale balance equations. Second, the continuum model assumes capillary equilibrium within an REV at all times, as dictated by the PcP_c–SbS_b curve. iPNM imposes no such assumption and naturally resolves the pre-equilibration dynamics of ripening. Depending on the size and specific microstructure of the REV, the pre-equilibration timescale can be on the order of macroscopic forcings imposed on the system (e.g., cyclic injections in H2_2 storage), which would prevent equilibrium from ever being reached. Even more challenging is when an REV itself is absent (e.g., fractal media) [49], for which the continuum framework breaks down entirely. Third, the PcP_c–SbS_b curve from PMM, used by the continuum model, corresponds to the maximal saturation achievable at any given PcP_c. If the actual saturation is non-maximal, as is often the case due to incomplete sweep during drainage-imbibition cycles, PcP_c is underestimated [18]. Mitigating this requires extended parameterizations that express PcP_c as functions of SbS_b, specific interfacial area [50], and Euler characteristic [51, 52], but these demand evolution equations for these extra variables, which themselves require closure. Pore-scale models like iPNM circumvent this entirely because each ganglion’s capillary pressure is computed from its own saturation and pore geometry, with no ambiguity from unoccupied pores. Population balance methods [22, 23] similarly avoid this limitation.

iPNM’s limitations

A key achievement of iPNM is that no geometric approximation in pore shape, typical of all existing PNMs, is introduced. Instead, different pore shapes are encoded by κb\kappa_b–SbS_b curves obtained from pore-wise applications of PMM. Despite the advance, several limitations remain. One is the PMM curvature bias for multi-pore ganglia identified in § 5.3. Because PMM erodes each pore’s subimage independently, connected throats are progressively emptied. This confines the non-wetting phase to the pore body and can overestimate curvature when throats are occupied. One antidote is a connectivity-aware PMM, analogous to the classical invasion-percolation formulation by [34], but the downside is that the number of occupied-throat permutations grows combinatorially with coordination number. A more tractable approximation is to apply a saturation correction to κb\kappa_b–SbS_b curves depending on the number of occupied throats. However, the details require further research. Another limitation of iPNM concerns the non-uniqueness of the subcritical branch in the κb\kappa_b–SbS_b curves for bridge pores (Eq. (29)). As shown in Figure 3, multiple interface configurations can exist in a bridge pore depending on which throats are occupied, but all are approximated with the same function. While the correct physical limits are satisfied (§ 3.4), the actual shape between these limits is uncertain.

Other limitations are less consequential for the experiments considered herein but may matter in other settings. First, a zero contact angle was assumed, which is appropriate for the H2H_2–water–silicon system in § 5. Extensions of PMM to non-zero contact angles exist [53, 54] and iPNM can accommodate them without modification. Note the effect enters solely through the κb\kappa_b–SbS_b curves. Second, while all simulations were on topologically 2D networks, iPNM and its associated PMM operations remain unchanged in 3D. Third, vapor pressure and temperature were assumed spatially uniform, which clearly is insufficient to capture the localized condensation and reevaporation observed at 80∘C80^\circ\mathrm{C}. Capturing this phenomenon requires an extra energy balance and the equation of state for water. Lastly, we treated the non-wetting phase density as constant, which was justified by Pc/Pw≈3%P_c/P_w \approx3\% in § 5.1, also typical of the subsurface. When this approximation breaks down, the flow and transport problems become nonlinear.

Sources of experimental discrepancy

The largest source of discrepancy between iPNM and the experiments is the idealization of the BCs at the channels. In iPNM, the concentration at each channel is set to be uniform and based on the capillary pressure of an interface spanning the channel’s half-width (§ 5.1). In experiments, however, the bubble distributions in the channels are highly non-uniform (Figure 8). For example, in the 40HS case, both channels are nearly filled with gas, with the left one containing small condensation droplets that raise the effective capillary pressure above the one assumed in iPNM. In the 80HS case, a low-curvature interface forms in the bottom of the left channel, which acts as a stronger sink than assumed. A similar low-curvature interface sits also in the right channel. These non-uniformities are likely why the micromodel depletes faster in the 80HS experiment than in iPNM (see Figure 13 and supplementary videos). The other two cases exhibit similar non-uniformities. In 40LS, the right channel initially contains a small bubble at the bottom that gradually grows upward, lengthening the diffusion path for the topmost ganglia. In 80LS, a low-curvature interface temporarily forms in the right channel but collapses quickly, consistent with the better agreement in SbtotS_b^{\mathrm{tot}}.

Condensation droplets are a secondary source of error, most prominent at $80^\circ\mathrm{C}$. In 80HS (Figure 13), droplets repeatedly condense within ganglia and re-evaporate over multiple cycles. A plausible contributor is small temperature fluctuations across the micromodel. Even on the order of tenths of a degree, such fluctuations can produce sizable swings in vapor pressure via the Clausius–Clapeyron relation (∼\sim4% per 1∘C1^\circ\mathrm{C}). The elevated gas pressure inside ganglia, by contrast, accounts for only a ∼\sim0.05% shift in vapor pressure due to the Poynting effect. Moreover, cyclic condensation and re-evaporation perturbs the wetting phase, which can mix dissolved H2H_2 across larger distances. Condensed droplets also increase local ganglion curvature, promoting dissolution by Henry’s law. Lastly, the experiments were initialized with deionized water [19], so the wetting phase was undersaturated at T=0T=0, whereas iPNM initializes $C from Henry’s law assuming local equilibrium. Despite this difference in initialization, no signature is visible in Figure 10, where a temporary drop in mean curvature (rˉ\bar{r}) would be expected. This effect is therefore of minor importance.

Conclusions

We presented an image-based pore network model (iPNM) that couples two-phase flow with Ostwald ripening to simulate the evolution of trapped ganglia in heterogeneous porous media. Unlike previous PNMs, iPNM eliminates geometric approximations of pore shape by parameterizing curvature–saturation curves directly from pore-scale images via the pore morphology method (PMM). It handles multi-pore ganglia of arbitrary size and complexity through a combination of pore-type classification, operator-split flow and transport solvers, and an iterative capillary stability checking algorithm that captures invasion, snap-off, retraction, fragmentation, coalescence, and dislocation. The sub-critical branch of the curvature–saturation curve, essential for ripening, is captured for all pore types.

iPNM was verified against a quasi-static PNM to ensure capillary-driven phenomena such as dissolution-induced snap-off, bubble dislocation, and ganglion growth in heterogeneous networks are captured accurately. iPNM was then validated against microfluidic experiments tracking hydrogen ripening in a sandstone-patterned micromodel over 15–24 days. The model predicts total saturation, mean curvature, curvature distributions, and spatial phase configurations in good agreement with experiments, particularly at 40°C. At 80°C, discrepancies arise primarily due to non-ideal boundary conditions at the micromodel’s inlet and outlet channels and, to a lesser extent, from condensation droplets not modeled by iPNM. A curvature bias by PMM was also identified for multi-pore ganglia, arising from the independent treatment of each pore during morphological erosion. Comparison with a continuum model confirmed that while macroscopic saturation is captured well by the continuum approach, it cannot resolve population statistics and spatial information about individual ganglia or pre-equilibrium dynamics of ripening within an REV.

Several directions for future work emerge from this work. Extending iPNM to non-zero contact angles, using existing PMM variants, would enable investigating whether ripening dynamics are qualitatively altered under partial wetting. In the context of underground hydrogen storage, generalizing iPNM along the multi-component framework of [20] would allow simulating ripening of gas mixtures. Moreover, iPNM opens the possibility of studying how cyclic injection and withdrawal, interacting with ripening, reshape ganglion populations over multiple seasons. Application to 3D rock images extracted from X-ray CT is straightforward, as the PMM operations remain unchanged.

Supplementary materials

Included as supplementary materials

Acknowledgments

This research is jointly supported by the National Science Foundation, United States under Grant No. CBET-2348723 (for ZL, YM) and the Natural Sciences and Engineering Research Council of Canada (NSERC) Alliance Grants under ALLRP 592525-23 and ALLRP 567631-24 (for MS, TL, BZ).

Entry pressure for rectangular throat

We derive the capillary entry pressure for a throat with rectangular cross-section using the Mayer-Stowe-Princen (MS-P) theory [42, 43, 44]. The derivation is based on an energy balance at the moment of invasion. Specifically, for a small displacement dx\mathrm{d}x of the meniscus, the balance of work done versus surface energy created/destroyed reads:

PcAe dx=(Lbwσbw+Lbsσbs−Lwsσws) dx,(39)P_c A_e \,\mathrm{d}x = \left(L_{bw}\sigma_{bw} + L_{bs}\sigma_{bs} - L_{ws}\sigma_{ws}\right)\,\mathrm{d}x, \tag*{(39)}

where AeA_e is the cross-sectional area occupied by the non-wetting phase (effective area), LbwL_{bw} is the length of the bubble-water (non-wetting/wetting/wetting) interface, LbsL_{bs} is the length of bubble-solid contact, and σbw\sigma_{bw}, σbs\sigma_{bs}, σws\sigma_{ws} are the respective interfacial tensions. Applying Young’s equation, σbs−σws=σbwcos⁡θ\sigma_{bs}-\sigma_{ws}=\sigma_{bw}\cos\theta, and denoting σ=σbw\sigma=\sigma_{bw}, Eq. A.1 simplifies to:

Pcσ=Lbw+Lbscos⁡θAe=PeAe,(40)\frac{P_c}{\sigma} = \frac{L_{bw}+L_{bs}\cos\theta}{A_e} = \frac{P_e}{A_e}, \tag*{(40)}

where Pe=Lbw+Lbscos⁡θP_e=L_{bw}+L_{bs}\cos\theta is the effective perimeter. The next step in the MS-P method is to equate the displacement curvature Pc/σP_c/\sigma to the curvature 1/r1/r of the arc menisci (AMs) residing in the corners [44]:

Pcσ=1r=Pe(r)Ae(r).(41)\frac{P_c}{\sigma}=\frac{1}{r}=\frac{P_e(r)}{A_e(r)}. \tag*{(41)}

Eq.A.3 is an implicit equation that must be solved for rr. We next derive explicit expressions for Pe(r)P_e(r) and Ae(r)A_e(r).

Figure 2 shows the cross-section of a rectangular throat occupied by the non-wetting phase in the center and wetting phase in the corners, forming four AMs. The throat has an in-plane side length of 2a2a and an out-of-plane side length of 2b2b. The distance from the contact line to the corner apex is denoted by ℓ\ell and it is related to rr via:

ℓ=2 rsin⁡(π4−θ).(42)\ell=\sqrt{2}\,r\sin\left(\frac{\pi}{4}-\theta\right). \tag*{(42)}

The bubble-solid contact length LbsL_{bs} is the perimeter of the throat minus the portions occupied by the AMs:

Lbs=4(a+b−2ℓ).(43)L_{bs}=4(a+b-2\ell). \tag*{(43)}

The bubble-water arc length LbwL_{bw} counting all four AMs is:

Lbw=8(π4−θ)r.(44)L_{bw}=8\left(\frac{\pi}{4}-\theta\right)r. \tag*{(44)}

Hence, the effective perimeter PeP_e can be computed via:

Pe=Lbw+Lbscos⁡θ=4[2(π4−θ)r+(a+b−2ℓ)cos⁡θ].(45)P_e=L_{bw}+L_{bs}\cos\theta=4\left[2\left(\frac{\pi}{4}-\theta\right)r+(a+b-2\ell)\cos\theta\right]. \tag*{(45)}

The effective cross-sectional area AeA_e of the non-wetting phase can be computed by subtracting the four corner-film areas from the throat’s cross-sectional area. Straightforward geometric analysis yields the area of a single corner film:

AAM=22ℓ(22ℓ+rcos⁡(π4−θ))−(π4−θ)r2.(46)A_{AM}=\frac{\sqrt{2}}{2}\ell\left(\frac{\sqrt{2}}{2}\ell+r\cos\left(\frac{\pi}{4}-\theta\right)\right)-\left(\frac{\pi}{4}-\theta\right)r^2. \tag*{(46)}

which in turns allows computing:

Ae=4ab−4AAM=4[ab−22ℓ(22ℓ+rcos⁡(π4−θ))+(π4−θ)r2].(47)A_e=4ab-4A_{AM}=4\left[ab-\frac{\sqrt{2}}{2}\ell\left(\frac{\sqrt{2}}{2}\ell+r\cos\left(\frac{\pi}{4}-\theta\right)\right)+\left(\frac{\pi}{4}-\theta\right)r^2\right]. \tag*{(47)}

Eqs.A.7 and A.9 are substituted into Eq.A.3 to allow solving for rr numerically. The entry pressure is Pce=σ/rP_{ce}=\sigma/r.

Special case of square throat. For a square throat with a=b=Ra=b=R, we show our formulation for rectangular throats reduces to the analytical solution of Ma et al. [44]:

PceRσ=cos⁡θ+sin⁡θcos⁡θ+π4−θ.(48)\frac{P_{ce}R}{\sigma}=\cos\theta+\sqrt{\sin\theta\cos\theta+\frac{\pi}{4}-\theta}. \tag*{(48)}

Let γ=π/4−θ\gamma=\pi/4-\theta. Substituting a=b=Ra=b=R and ℓ=2rsin⁡γ\ell=\sqrt{2}r\sin\gamma into Eqs.A.7 and A.9 yields:

Pe=4[2γr+(2R−22rsin⁡γ)cos⁡θ],(49)P_e=4\left[2\gamma r+(2R-2\sqrt{2}r\sin\gamma)\cos\theta\right], \tag*{(49)}
Ae=4[R2+r2γ−r2sin⁡2γ−r2sin⁡γcos⁡γ].(50)A_e=4\left[R^2+r^2\gamma-r^2\sin^2\gamma-r^2\sin\gamma\cos\gamma\right]. \tag*{(50)}

Inserting into Eq.A.3 and using cos⁡θ=12(cos⁡γ+sin⁡γ)\cos\theta=\frac{1}{\sqrt{2}}(\cos\gamma+\sin\gamma), we obtain after simplification:

R2=r2[γ−sin⁡γ(cos⁡γ+sin⁡γ)]+2Rr(cos⁡γ+sin⁡γ).(51)R^2=r^2[\gamma-\sin\gamma(\cos\gamma+\sin\gamma)]+\sqrt{2}Rr(\cos\gamma+\sin\gamma). \tag*{(51)}

Defining u=r/Ru=r/R, Eq. A.13 is a quadratic Au2+Bu−1=0Au^2+Bu-1=0, with A=γ−sin⁡γ(cos⁡γ+sin⁡γ)A=\gamma-\sin\gamma(\cos\gamma+\sin\gamma) and B=2(cos⁡γ+sin⁡γ)B=\sqrt{2}(\cos\gamma+\sin\gamma). The physical root gives:

PceRσ=1u=B+B2+4A2.(52)\frac{P_{ce}R}{\sigma}=\frac{1}{u}=\frac{B+\sqrt{B^2+4A}}{2}. \tag*{(52)}

Noting that B/2=cos⁡θB/2=\cos\theta and B2+4A=4sin⁡θcos⁡θ+4(π/4−θ)B^2+4A=4\sin\theta\cos\theta+4(\pi/4-\theta), this simplifies to Eq. A.10.

Derivation of non-wetting phase conductivity

We derive the hydraulic conductance of the non-wetting (or bubble) phase flowing through the central region of an invaded throat with rectangular cross-section (see Figure 2). The derivation employs the hydraulic diameter concept [38] and ensures that the correct asymptotic behavior is preserved as the bubble saturation approaches unity.

To establish some key relations, consider first the scenario where a throat is occupied by one fluid phase undergoing steady-state, fully-developed laminar flow. The corresponding axial momentum balance on a fluid element reads:

−dPdx=τwPtAt.(53)-\frac{\mathrm{d}P}{\mathrm{d}x}=\frac{\tau_w P_t}{A_t}. \tag*{(53)}

where τw\tau_w is the mean shear stress exerted by the fluid on the throat’s walls, AtA_t is the cross-sectional area of the throat, and PtP_t is its wetted perimeter. Next, introduce the Fanning friction factor defined as:

Cf=τw12ρu2.(54)C_f=\frac{\tau_w}{\frac{1}{2}\rho u^2}. \tag*{(54)}

where ρ\rho is the fluid density and uu the mean velocity. For laminar flow, CfC_f scales inversely with Reynolds number:

Cf=PoRe,whereRe=ρuDhμ.(55)C_f=\frac{Po}{Re},\qquad\text{where}\quad Re=\frac{\rho uD_h}{\mu}. \tag*{(55)}

where Dh=4At/PtD_h=4A_t/P_t is the hydraulic diameter. The Poiseuille number PoPo is a geometry-dependent constant that characterizes flow resistance for a given throat shape. For example, Po=16Po=16 for a circular duct and Po≈14.2Po\approx14.2 for a square duct [38]. Substituting Eqs. B.2–B.3 into Eq. B.1 yields:

u=Dh22PoμΔPLt.(56)u=\frac{D_h^2}{2Po\mu}\frac{\Delta P}{L_t}. \tag*{(56)}

The volumetric flow rate can be computed as follows:

qt=uAt=Dh2At2PoμΔPLt.(57)q_t=uA_t=\frac{D_h^2A_t}{2Po\mu}\frac{\Delta P}{L_t}. \tag*{(57)}

Defining the hydraulic conductance via qt=(Kt/μ)ΔPq_t=(K_t/\mu)\Delta P, we obtain:

Kt=Dh2At2PoLt.(58)K_t=\frac{D_h^2A_t}{2PoL_t}. \tag*{(58)}

For a rectangular throat with half-widths aa and bb (Figure 2), At=4abA_t=4ab, Pt=4(a+b)P_t=4(a+b), and Dh=4At/PtD_h=4A_t/P_t.

Now for the non-wetting phase in Figure 2, we define the hydraulic conductance in a similar fashion to Eq. B.6:

Kb=Dh,b2Ab2PobLt.(59)K_b=\frac{D_{h,b}^2A_b}{2Po_bL_t}. \tag*{(59)}

where AbA_b is the cross-sectional area of the bubble-occupied region in the throat, Dh,bD_{h,b} is its hydraulic diameter, and PobPo_b is its Poiseuille number. This region is bounded by solid walls and the four arc menisci occupying the corners, as shown in Figure 2. Under the assumption of perfect slip at the interface between the bubble and wetting phase, the wetted perimeter CbC_b consists only of the solid walls. Expressions for AbA_b, CbC_b, and Dh,bD_{h,b} are given in § 3.3.

The only quantity on the right-hand side of Eq. B.7 that is not directly available is the Poiseuille number PobPo_b. In general, PobPo_b depends on the shape of the bubble-occupied region (Figure 2), which varies with the throat’s saturation as the corner films grow or shrink. Since no analytical or numerical solution for this dependence exists, we approximate PobPo_b by the single-phase Poiseuille number PoPo. The latter can be back-calculated using:

Pob≈Po=Dh2At2KtLt(60)Po_b \approx Po = \frac{D_{h}^{2} A_t}{2 K_t L_t} \tag*{(60)}

where KtK_t is the known single-phase conductivity from Eq. (13). This approximation ensures the correct asymptotic behavior: as Sb→1S_b \to1, the wetting films shrink, causing Ab→AtA_b \to A_t, Cb→PtC_b \to P_t, and Dh,b→DhD_{h,b} \to D_h, so that Kb→KtK_b \to K_t.

Continuum model

The continuum model of Salehpour et al. [19] describes the macroscopic evolution of the non-wetting phase due to Ostwald ripening. It is derived by taking the continuum limit of the pore-scale mass balance equations of Bueno et al. [20], assuming ideal gas behavior for the non-wetting phase. The full derivation is in the supplementary material of [19]. The model is formulated in terms of the dissolved mole fraction CC in the aqueous phase. For the micromodel considered here, the domain is treated as 1D in the left-to-right direction xx (Figure 8), and the equation reads:

{[Hρw−1][HdSbdPc+Sb]+1}∂C∂t=Dmτ∂2C∂x2(61)\left\{\left[\frac{H}{\rho_w}-1\right]\left[H\frac{dS_b}{dP_c}+S_b\right]+1\right\}\frac{\partial C}{\partial t}=\frac{D_m}{\tau}\frac{\partial^2 C}{\partial x^2} \tag*{(61)}

where DmD_m is the molecular diffusion coefficient, HH is Henry’s constant, ρw\rho_w is the molar density of the wetting phase, and τ=ϕ−1\tau=\phi^{-1} is Bruggeman’s tortuosity. The dSb/dPcdS_b/dP_c term is obtained from a macroscopic PcP_c–SbS_b relationship computed by applying PMM to the binary image of the entire micromodel [18]. Unlike classical PMM [33, 34], the variant used here retains disconnected ganglia, and unlike the pore-wise application in iPNM (§ 3.4), here PMM is applied globally to produce a single PcP_c–SbS_b curve. The reader is referred to [19] for an illustration of this curve.

The non-wetting phase saturation SbS_b is recovered from CC via Henry’s law:

C=Pw+Pc(Sb)−PvH(62)C=\frac{P_w+P_c(S_b)-P_v}{H} \tag*{(62)}

where PwP_w is the wetting-phase pressure and PvP_v is the spatially uniform vapor pressure. To invert Eq. C.2 for SbS_b, the PcP_c–SbS_b curve is used. Dirichlet BCs on CC are imposed at the left and right ends of the domain by evaluating Eq. C.2 at Pc=σ(1/Rl,r+2/g)P_c=\sigma(1/R_l,r+2/g) from the Young–Laplace equation, where Rl=50 μmR_l=50\ \mu\mathrm{m} and Rr=85 μmR_r=85\ \mu\mathrm{m} are half-widths of the left and right channels, respectively (Figure 8). The initial condition for CC is prescribed by converting the initial non-wetting phase saturation profile to concentration via Eq. C.2 and the PcP_c–SbS_b curve. The initial saturation profile itself is constructed from the first experimental image by averaging the phase occupancy in the yy-direction (up-down in Figure 8) at each xx-coordinate using a sliding window of width equal to 5% of the micromodel length. The total non-wetting phase saturation in the micromodel at each time step is computed by spatially averaging the saturation profile.

Eq. C.1 is discretized along the xx direction with a uniform mesh of 100 cells. Backward Euler time integration with adaptive time stepping and a Newton solver are used at each time step. Fluid properties are assigned depending on the temperature corresponding to the particular experiment being modeled. These are listed in Table 1.

References

  1. [1]D. Zivar, S. Kumar, J. Foroozesh, Underground hydrogen storage: A comprehensive review, International journal of hydrogen energy 46 (45) (2021) 23436–23462.DOI
  2. [2]S. Bachu, Co2 storage in geological media: Role, means, status and barriers to deployment, Progress in energy and combustion science 34 (2) (2008) 254–273.
  3. [3]S. Iglauer, A. Paluszny, C. H. Pentland, M. J. Blunt, Residual co2 imaged with x-ray micro-tomography, Geophysical Research Letters 38 (21) (2011).DOI
  4. [4]M. Andersson, S. Beale, M. Espinoza, Z. Wu, W. Lehnert, A review of cell-scale multiphase flow modeling, including water management, in polymer electrolyte fuel cells, Applied Energy 180 (2016) 757–778.DOI
  5. [5]W. Ostwald, Studien über die bildung und umwandlung fester körper, Zeitschrift für physikalische Chemie 22 (1) (1897) 289–330.DOI
  6. [6]I. M. Lifshitz, V. V. Slyozov, The kinetics of precipitation from supersaturated solid solutions, Journal of physics and chemistry of solids 19 (1-2) (1961) 35–50.DOI
  7. [7]C. Wanger, Theorie der alterung von niederschlagen durch umlosen, Z. Elektrochem. 65 (1961) 581–591.
  8. [8]P. W. Voorhees, The theory of ostwald ripening, Journal of Statistical Physics 38 (1985) 231–252.DOI
  9. [9]K. Xu, R. Bonnecaze, M. Balhoff, Egalitarianism among bubbles in porous media: an ostwald ripening derived anticoarsening phenomenon, Physical review letters 119 (26) (2017) 264502.DOI
  10. [10]J. A. de Chalendar, C. Garing, S. M. Benson, Pore-scale modelling of ostwald ripening, Journal of Fluid Mechanics 835 (2018) 363–392.
  11. [11]Y. Mehmani, K. Xu, Pore-network modeling of ostwald ripening in porous media: How do trapped bubbles equilibrate?, Journal of Computational Physics 457 (5 2022). doi:10.1016/j.jcp.2022.111041.DOI
  12. [12]S. Goodarzi, Y. Zhang, S. Foroughi, B. Bijeljic, M. J. Blunt, Trapping, hysteresis and ostwald ripening in hydrogen storage: A pore-scale imaging study, International Journal of Hydrogen Energy 56 (2024) 1139–1151.
  13. [13]Z. Jangda, H. Menke, A. Busch, S. Geiger, T. Bultreys, H. Lewis, K. Singh, Pore-scale visualization of hydrogen storage in a sandstone at subsurface pressure and temperature conditions: Trapping, dissolution and wettability, Journal of Colloid and Interface Science 629 (2023) 316–325.DOI
  14. [14]M. Boon, T. Rademaker, C. W. Winardhi, H. Hajibeygi, Multiscale experimental study of h/brine multiphase flow in porous rock characterizing relative permeability hysteresis, hydrogen dissolution, and ostwald ripening, Scientific Reports 14 (1) (2024) 30170.DOI
  15. [15]C. Wang, Y. Mehmani, K. Xu, Capillary equilibrium of bubbles in porous media, Proceedings of the National Academy of Sciences 118 (17) (2021) e2024069118.DOI
  16. [16]Y. Mehmani, K. Xu, Capillary equilibration of trapped ganglia in porous media: A pore-network modeling approach, Advances in Water Resources 166 (8 2022). doi:10.1016/j.advwatres.2022.104223.DOI
  17. [17]Y. Li, C. Garing, S. M. Benson, A continuum-scale representation of ostwald ripening in heterogeneous porous media, Journal of Fluid Mechanics 889 (2020) A14.DOI
  18. [18]Y. Mehmani, K. Xu, A continuum theory of diffusive bubble depletion in porous media, Advances in Water Resources 185 (2024) 104625.DOI
  19. [19]M. Salehpour, T. Lan, N. Bueno, M. Z. I. Laku, Y. Mehmani, B. Zhao, Ostwald ripening in underground gas storage, arXiv preprint arXiv:2509.00044 (2025).arxiv.org/abs/2509.00044
  20. [20]N. Bueno, L. Ayala, Y. Mehmani, Ostwald ripening of multi-component bubbles in porous media: A theory and a pore-scale model of how bubble populations equilibrate, Advances in Water Resources 182 (2023). doi:10.1016/j.advwatres.2023.104581.DOI
  21. [21]Y. Yu, C. Wang, J. Liu, S. Mao, Y. Mehmani, K. Xu, Bubble coarsening kinetics in porous media, Geophysical Research Letters 50 (1) (2023) e2022GL100757.DOI
  22. [22]N. Bueno, L. Ayala, Y. Mehmani, A generalized kinetic theory of ostwald ripening in porous media, Advances in Water Resources 193 (2024). doi:10.1016/j.advwatres.2023.104826.DOI
  23. [23]N. Bueno, L. F. Ayala, Y. Mehmani, Kinetic theory of multicomponent ostwald ripening in porous media, arXiv preprint arXiv:2512.20665 (2025).arxiv.org/abs/2512.20665
  24. [24]D. Singh, H. A. Friis, E. Jettstuen, J. O. Helland, A level set approach to ostwald ripening of trapped gas bubbles in porous media, Transport in Porous Media 145 (2) (2022) 441–474.DOI
  25. [25]D. Singh, H. A. Friis, E. Jettstuen, J. O. Helland, Ostwald ripening of gas bubbles in porous media: Impact of pore geometry and spatial bubble distribution, Advances in Water Resources 187 (2024) 104688.
  26. [26]N. Joewondo, V. Garbin, R. Pini, Experimental evidence of the effect of solute concentration on the collective evolution of bubbles in a regular pore-network, Chemical Engineering Research and Design 192 (2023) 82–90.DOI
  27. [27]K. E. Thompson, Pore-scale modeling of fluid transport in disordered fibrous materials, AIChE journal 48 (7) (2002) 1369–1389.DOI
  28. [28]V. Joekar-Niasar, S. M. Hassanizadeh, H. Dahle, Non-equilibrium effects in capillarity and interfacial area in two-phase flow: dynamic pore-network modelling, Journal of fluid mechanics 655 (2010) 38–71.
  29. [29]S. Chen, C. Qin, B. Guo, Fully implicit dynamic pore-network modeling of two-phase flow and phase change in porous media, Water Resources Research 56 (11) (2020) e2020WR028510.DOI
  30. [30]X. Li, Y. Yortsos, Visualization and simulation of bubble growth in pore networks, AIChE Journal 41 (2) (1995) 214–222.
  31. [31]L. A. Dillard, M. J. Blunt, Development of a pore network simulation model to study nonaqueous phase liquid dissolution, Water Resources Research 36 (2) (2000) 439–454.DOI
  32. [32]A. Dominguez, S. Bories, M. Prat, Gas cluster growth by solute diffusion in porous media. experiments and automaton simulation on pore network, International Journal of Multiphase Flow 26 (12) (2000) 1951–1979.DOI
  33. [33]R. Hazlett, Simulation of capillary-dominated displacements in microtomographic images of reservoir rocks, Transport in porous media 20 (1) (1995) 21–35.
  34. [34]M. Hilpert, C. T. Miller, Pore-morphology-based simulation of drainage in totally wetting porous media, Advances in water resources 24 (3-4) (2001) 243–255.DOI
  35. [35]S. Beucher, C. Lantuéjoul, Use of watersheds in contour detection, in: International Workshop on Image Processing: Real-time Edge and Motion Detection/Estimation, Rennes, France, 1979.
  36. [36]J. T. Gostick, Versatile and efficient pore network extraction method using marker-based watershed segmentation, Physical Review E 96 (2) (2017) 023307.DOI
  37. [37]Y. Mehmani, H. A. Tchelepi, Minimum requirements for predictive pore-network modeling of solute transport in micromodels, Advances in water resources 108 (2017) 83–98.DOI
  38. [38]F. M. White, J. Majdalani, Viscous fluid flow, Vol. 3, McGraw-Hill New York, 2006.
  39. [39]T. Patzek, D. B. Silin, Shape factor and hydraulic conductance in noncircular capillaries: I. one-phase creeping flow, Journal of colloid and interface science 236 (2) (2001) 295–304.
  40. [40]Y. Muzychka, M. Yovanovich, Modeling friction factors in non-circular ducts for developing laminar flow, in: 2nd AIAA, Theoretical Fluid Mechanics Meeting, 1998, p. 2492.DOI
  41. [41]T. Patzek, J. Kristensen, Shape factor correlations of hydraulic conductance in noncircular capillaries: Ii. two-phase creeping flow, Journal of colloid and interface science 236 (2) (2001) 305–317.
  42. [42]R. P. Mayer, R. A. Stowe, Mercury porosimetry–breakthrough pressure for penetration between packed spheres, Journal of colloid Science 20 (8) (1965) 893–911.DOI
  43. [43]H. Princen, Capillary phenomena in assemblies of parallel cylinders: I. capillary rise between two cylinders, Journal of Colloid and Interface Science 30 (1) (1969) 69–75.
  44. [44]S. Ma, G. Mason, N. R. Morrow, Effect of contact angle on drainage and imbibition in regular polygonal tubes, Colloids and Surfaces A: Physicochemical and Engineering Aspects 117 (3) (1996) 273–291.DOI
  45. [45]R. Lenormand, C. Zarcone, A. Sarr, Mechanisms of the displacement of one fluid by another in a network of capillary ducts, Journal of Fluid Mechanics 135 (1983) 337–353.
  46. [46]N. Sahloul, M. Ioannidis, I. Chatzis, Dissolution of residual non-aqueous phase liquids in porous media: pore-scale mechanisms and mass transfer rates, Advances in water resources 25 (1) (2002) 33–49.DOI
  47. [47]R. Reid, J. Prausnitz, B. Poling, The Properties of Gases and Liquids, 4th Edition, McGraw-Hill, New York, 1987.
  48. [48]J. Mehl, M. Huber, A. Harvey, Ab initio transport coefficients of gaseous hydrogen, Int. J. Thermophys. 31 (2010) 740–755. doi :10.1007/s10765-009-0697-9.DOI
  49. [49]Y. Mehmani, T. Anderson, Y. Wang, S. A. Aryana, I. Battiato, H. A. Tchelepi, A. R. Kovscek, Striving to translate shale physics across ten orders of magnitude: What have we learned?, Earth-Science Reviews 223 (2021) 103848.DOI
  50. [50]S. M. Hassanizadeh, W. G. Gray, Thermodynamic basis of capillary pressure in porous media, Water resources research 29 (10) (1993) 3389–3405.DOI
  51. [51]J. E. McClure, R. T. Armstrong, M. A. Berrill, S. Schlüter, S. Berg, W. G. Gray, C. T. Miller, Geometric state function for two-fluid flow in porous media, Physical Review Fluids 3 (8) (2018) 084306.DOI
  52. [52]J. E. McClure, T. Ramstad, Z. Li, R. T. Armstrong, S. Berg, Modeling geometric state for fluids in porous media: Evolution of the euler characteristic, Transport in Porous Media 133 (2) (2020) 229–250.DOI
  53. [53]V. P. Schulz, E. A. Wargo, E. C. Kumbur, Pore-morphology-based simulation of drainage in porous media featuring a locally variable contact angle, Transport in Porous Media 107 (1) (2015) 13–25.DOI
  54. [54]X. Liu, A. Zhou, S.-I. Shen, J. Li, Modeling drainage in porous media considering locally variable contact angle based on pore morphology method, Journal of Hydrology 612 (2022) 128157.DOI

Paper details

vlnc:W_uv7upbr3
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents