Experiments and geochemical modeling of hydrogen generation during water–rock reactions: effects of mineralogy and physicochemical conditions

1 Introduction

Ultramafic rocks are igneous rocks rich in iron (Fe) and Magnesium (Mg) with a silica content less than 45% (Rezaee, 2025). Different ultramafic rock minerals react differently during serpentinization, one of the water-rock reactions in which iron-bearing rocks react with water at elevated temperatures to form serpentine and hydrogen as byproducts. In fact, Isah and Okoroafor (2025) determined that not all geologic H₂-generating reactions involving ultramafic and rocks result in the formation of serpentine, brucite, or magnetite. Olivine (Mg, Fe)₂SiO₄ can be found in high concentrations in peridotites and is considered the major reactant in the transformation to serpentine. Orthopyroxene (Mg, Fe)₂Si₂O_6, which is also common in peridotites, websterite, and harzburgite, can still contribute to serpentinization but is less reactive. Similarly, Clinopyroxene (Ca, Mg, Fe)₂Si₂O₆, found in wehrlite, websterite, and gabbro (Rezaee, 2025), is the most stable and least reactive of the three, typically surviving as a relic mineral in strongly serpentinized peridotites (Pens et al., 2016). Olivine’s alteration generates molecular hydrogen; a simplified reaction is presented by (Rezaee, 2025):

6 F e 2 S i O 4 O l i v i n e + 7 H 2 O → 3 F e 3 S i 2 O 5 O H 4 S e r p e n t i n e + F e 3 O 4 M a g n e t i t e + H 2

Serpentinization is accompanied by oxidation of ferrous iron to ferric iron due to reactions with water in minerals such as olivine and orthopyroxene, which generate hydrogen (Klein et al., 2013).

2 F e O + H 2 O → F e 2 O 3 + H 2 a q

The initial mineral composition of the rock and the temperature control the reaction path and the product composition. For example, an olivine-dominant peridotite undergoing serpentinization forms Fe-bearing brucite at temperatures less than 320 °C and magnetite at temperatures greater than 200 °C, while serpentinization of pyroxenite produces iron-rich serpentine and iron-poor talc. Fayalite-rich olivine is stable in the presence of water at low temperatures till 180 °C, so it is predicted that serpentinization of fayalite-rich olivine is improved at temperatures lower than 180 °C (Klein et al., 2013). The reaction path depends upon the percentage of fayalite in the olivine to the forsterite. Oze and Sharma (2007) studied different percentages of Earth’s surface conditions and those of the planet Mars. The thermodynamic calculations indicated that, at 1 bar and between 5 °C and 90 °C, serpentinization is favorable from Fo₃₅ to Fo₁₀₀, where X in Fo_X identifies the forsterite (Fo100: Mg₂SiO₄) component or the mol% of Mg in the olivine. The different chemical reaction equations are presented in Table 1. Later, thermodynamic studies by McCollom et al. (2022) showed that Fe-rich olivine generates greater amounts of H₂ per mole than serpentinization of Mg-rich olivine, by a factor of 2–10, depending on temperature and olivine composition.

TABLE 1. Serpentinization reactions based on compositional variations of olivine reproduced from (Oze and Sharma, 2007).

Changing the pH affects the serpentinization of the olivine-rich rocks. McCollom et al. (2020) reported that at a constant temperature of 230 °C, increasing the pH from 9 to 12 increased the generated hydrogen concentration from 25 to 47 m m o l / k g − 1, almost doubling the concentration. Studying the concentration of the products, it has been shown that increasing the pH increases Fe partitioning into brucite and chrysotile rather than into magnetite (McCollom et al., 2020).

Thermodynamics constrain the mineral distribution and stability of iron as a function of temperature, and, accordingly, the amount of hydrogen production from olivine. The rate of serpentinization at the same pressure (35 MPa) increases at high temperatures above 315 °C, and at temperatures below 150 °C, the kinetic reaction is slower (McCollom and Bach, 2009).

The serpentinization rate is a function of the water supply rate to the reaction surfaces (Martin and Fyfe, 1970). Rocks like peridotite, which are olivine-dominant, have inherently low porosity that decreases further as serpentinization products precipitate and occlude the available pore space. The reaction decreases with time as fluid pathways are reduced. Røyne et al. (2008) suggested that inducing fractures creates new fluid pathways and, accordingly, accelerates the reaction (Zhang et al., 2019).

Many experiments have been conducted, and some models have been built to study hydrogen generation rates during serpentinization under different conditions of pressure, temperature, and mineral composition, as reported by Rezaee (2025) in Table 2.

TABLE 2. Hydrogen generation during the serpentinization of ultramafic minerals and rocks, highlighting the influence of experimental conditions on the outcomes (Rezaee, 2025).

Data from Table 2 is plotted with hydrogen generated normalized by the number of experimental/modeling days and is presented in Figure 1.

Across the compiled laboratory dataset, the time-normalized H₂ yield rate, expressed as mmol kg^-1 day^-1, spans nearly six orders of magnitude, from approximately 10^–4 mmol kg^-1 day^-1 in low-temperature dunite experiments to approximately 30 mmol kg^-1 day^-1 in high-temperature, high-pressure olivine systems. This range immediately signals that yield alone is an insufficient basis for comparing across studies: duration normalization reveals the underlying kinetic differences that cumulative totals conceal.

Temperature emerges as the dominant control. Experiments conducted above 200 °C consistently produce rates of 0.2–30 mmol kg^-1 day^-1, whereas those conducted between 25 °C and 100 °C yield rates two to four orders of magnitude lower, in the range 10^–4 to 5 × 10⁻³ mmol kg^-1 day^-1. This steep dependence is consistent with Arrhenius kinetics, in which the rate-limiting olivine dissolution step carries an activation energy of 60–80 kJ mol^-1, sufficient to produce a tenfold increase in rate for every approximately 40 °C rise in temperature at conditions relevant to serpentinizing systems. The implication is consequential: laboratory experiments conducted at ambient or near-surface temperatures, however prolonged, are unlikely to reproduce the reaction rates operative in deeper crustal or hydrothermal settings, and comparisons across temperature regimes must be made with this kinetic offset explicitly in view.

Pressure exerts a secondary but measurable positive influence, most clearly at high temperature, where the dataset is sufficient to permit comparison. At 300 °C, olivine-bearing experiments show systematically higher rates as pressure increases from approximately 300 bar to 3,000 bar, a trend attributable to enhanced water activity, suppressed degassing of product H₂, and possible pressure-induced modification of the olivine dissolution mechanism. At low temperature, the near-ambient pressure in most experiments makes it difficult to isolate an independent pressure contribution, and the available data do not permit a rigorous separation of the two variables. A controlled experimental series varying pressure at a fixed temperature would be required to resolve this.

Mineralogy introduces a further axis of variability. Olivine-rich compositions tend to produce the highest rates under equivalent temperature and pressure conditions, consistent with the high Fe(II) content and relatively rapid dissolution kinetics of Mg-Fe olivine. This is similar to the findings by Isah and Okoroafor (2025). Mixed compositions incorporating orthopyroxene show lower rates in several studies, plausibly reflecting the dilution of reactive Fe(II) by the pyroxene component and differences in the serpentinization pathway for pyroxene-bearing assemblages. Whole-rock peridotite experiments exhibit the widest intra-group spread, likely because bulk mineralogy varies across studies, and the relative proportions of olivine, pyroxene, and accessory phases are not always reported in sufficient detail for quantitative comparison.

Modeling serpentinization reaction has had different approaches. Rezaee (2025) proposed a regression model predicting hydrogen generation as a function of pressure and temperature, with two separate equations: one for temperatures below 315 °C and another for temperatures between 315 °C and 400 °C. McCollom and Bach (2009) developed a numerical model based on chemical thermodynamics to examine the effects of temperature and water-to-rock ratio on changes in mineral composition during serpentinization. Many have used thermodynamic equilibrium modeling in PHREEQC to study the reaction paths for serpentinization reactions (Geymond et al., 2025; Osselin et al., 2022; Parkhurst and Appelo, 2013). Nwanwe et al. (2026) applied thermodynamic modeling to investigate the effects of temperature, pressure, and forsterite percentage on hydrogen production. The model validation was based on earlier modeling work by McCollom et al. (2022). Recent modeling studies have advanced the simulation of geological hydrogen generation through increasingly detailed reactive transport frameworks. Eslami et al. (2026) developed a reservoir-scale dual-permeability fracture–matrix model to evaluate continuous and cyclic injection strategies during serpentinization, with emphasis on operational controls and porosity–permeability evolution. Moradi D, et al. (2026) simulated reaction and transport during core-scale serpentinization of Fe-bearing olivine and examined the associated evolution of mineral phases and transport properties. The present study is complementary to these modeling-focused contributions.

A recent review by Moradi R. et al. (2026) summarized the current understanding of geological hydrogen generation by serpentinization, including reaction mechanisms, thermodynamic controls, kinetic limitations, and reactive transport processes. This broader synthesis highlights the need for experimentally constrained studies that link mineralogical controls, time-dependent reaction kinetics, and transport effects during H₂ generation.

The literature summarized above shows that H₂ generation during serpentinization is controlled by both thermodynamic reaction pathways and kinetic limitations. Thermodynamics determines the stability of secondary phases and the potential for Fe(II) oxidation to Fe(III)-bearing products such as magnetite, whereas kinetics controls the rate at which mineral dissolution, Fe release, oxidation, and secondary mineral precipitation occur. Temperature accelerates mineral dissolution and reaction progress through Arrhenius-type rate dependence, while pH and pressure influence aqueous speciation, mineral stability, and the partitioning of generated H₂ between gas and aqueous phases. In addition, the rate of serpentinization depends on the supply of water to reactive surfaces. Therefore, powdered samples, which provide high and nearly uniform mineral–fluid contact area, may not represent intact or fractured rocks, where flow pathways, residence time, and progressive coating of reactive surfaces can strongly alter the temporal evolution of H₂ generation.

1.1 Research gap

Despite extensive experimental and modeling studies on serpentinization, a key gap remains in coupling controlled laboratory measurements with kinetic geochemical modeling to quantify H₂ generation as a function of olivine composition, particularly the forsterite-to-fayalite ratio. Previous studies have commonly used equilibrium thermodynamic models to evaluate mineralogical controls on H₂ generation; however, fewer studies have validated time-dependent kinetic predictions against experimental measurements across systematically varied mineral compositions.

Another limitation is that most experimental studies rely on powdered samples. Although powdered systems are useful for isolating chemical reaction controls because of their high reactive surface area, they do not reproduce the flow geometry, spatially heterogeneous mineral–fluid contact, residence-time effects, or progressive surface occlusion that occur in intact or fractured subsurface rocks. As a result, batch experiments and equilibrium models may reproduce cumulative H₂ yields but do not fully capture the temporal production behavior or its sensitivity to fracture-scale transport processes.

Therefore, bridging the gap between laboratory-scale serpentinization experiments and field-scale geological hydrogen systems requires experimental configurations that span powdered and intact-core systems, together with modeling frameworks that incorporate kinetic rate laws and reactive transport. This study addresses this gap by combining controlled mineral-composition experiments, Arrhenius-based kinetic modeling, and one-dimensional reactive transport simulation of a core-scale flow-through experiment.

1.2 Objectives

This study addresses the identified gaps through the following objectives.

• Quantify the effect of olivine composition on H₂ generation. Systematic powdered-sample experiments were used to evaluate H₂ generation as a function of the forsterite-to-fayalite ratio and to identify the dominant mineralogical control on hydrogen yield.
• Evaluate the effects of temperature, pressure, and initial pH. The individual and coupled effects of these parameters were examined across different mineral compositions, with particular attention to pressure–pH interactions that cannot be resolved from single-variable analysis alone.
• Develop and calibrate kinetic batch models. Arrhenius-based kinetic models were developed to reproduce the temporal evolution of H₂ generation in powdered-sample experiments across the tested mineral compositions and operating conditions.
• Compare powdered and core-scale H₂ generation behavior. Powdered-sample batch experiments were compared with core-scale flow-through experiments to evaluate how flow geometry, residence time, and mineral–fluid contact conditions alter H₂ production trajectories.
• Develop a core-scale reactive transport model. A one-dimensional reactive transport model was constructed for the core-scale experiment to reproduce early-time H₂ generation behavior and identify processes responsible for later-time divergence between modeled and measured results.

1.3 Study novelty

This study makes three main contributions to the understanding of geological hydrogen generation.

• First, it provides a controlled experimental dataset for isolating the effect of olivine composition on H₂ generation. Pure forsterite, pure fayalite, and intermediate mixtures were tested under consistent experimental conditions. This design isolates the effect of the forsterite-to-fayalite ratio without the confounding influence of varying temperature, pressure, reaction duration, or experimental setup. The resulting dataset provides a direct experimental basis for evaluating mineralogical controls and calibrating geochemical models.
• Second, it integrates experimental measurements with time-dependent kinetic–thermodynamic modeling. The model combines thermodynamic equilibrium constraints with Arrhenius kinetic rate laws, allowing the temporal evolution of H₂ generation to be simulated rather than predicting equilibrium end states alone. After calibration against the powdered-sample experiments, the model is used to evaluate sensitivity to temperature, pressure, and initial pH beyond the directly tested experimental conditions.
• Third, it extends the analysis from powdered batch systems to fractured intact rock under flow-through conditions. Coreflooding results are used to demonstrate that transport processes in fractured media can fundamentally alter H₂ generation behavior relative to powdered-sample systems. Unlike powdered samples, core-scale systems introduce the coupled effects of flow geometry, residence time distribution, possible fluid channeling, and progressive surface occlusion along fracture pathways. This comparison highlights a transport–reaction coupling that must be considered before laboratory serpentinization results can be reliably translated to subsurface field conditions.

2 Materials and methods

This study consists of experimental work followed by geochemical modeling. The experimental work is conducted at two scales: powdered samples and a core-scale sample. The powdered-sample experiments include three controlled mineral-system experiments using pure fayalite, pure forsterite, and a 50% fayalite–50% forsterite mixture, as well as six peridotite experiments conducted under varying pressure and pH conditions. The core-scale experiment was performed using a Twin Sisters dunite core to investigate hydrogen generation under flow-through conditions. Table 3 shows the composition of all the samples used for the experiment.

TABLE 3. Initial mineral composition of samples in weight %.

* The summation of Fayalite and Forsterite represents the Olivine percentage

The modeling work is conducted in two stages. First, four batch geochemical reaction models are developed using the React module for the pure fayalite, pure forsterite, 50% fayalite–50% forsterite, and peridotite systems. The peridotite model is treated as a complex rock system. Second, a one-dimensional reactive transport model is developed using X1t to simulate reaction and transport behavior in the Twin Sisters dunite core. The validated models are then used to conduct sensitivity analyses on the effects of temperature and pressure on hydrogen generation. The research workflow methodology is presented in Figure 2.

2.1 Experimental work

2.1.1 Powdered-sample experiments

This study used powdered samples: a pure fayalite powder, a pure forsterite powder, a 50% fayalite/50% forsterite mixture, and a peridotite powder sourced from the Trinity Ophiolite at Eunice Bluff, California, to examine the effects of varying mineral compositions. The peridotite sample was prepared by initially crushing it in a porcelain mortar and pestle to prevent further contamination. The crushed sample was sieved to obtain a uniform fraction in the 150–300 µm size range. The 50% fayalite/50% forsterite mixture was prepared by weighing exactly 2.5 gm of each fayalite and forsterite, dry mixing them, and finally adding the deionized water.

The powdered sample experiments were conducted in a Hastelloy autoclave reactor with a 150 mL volume and a water-to-rock ratio of 2. The reactor (Figure 3) was pressurized in two steps: first, using high-purity nitrogen to 320 psi; then, by raising the reactor temperature to 150 °C, the pressure increased to 500 psi. Six experiments were conducted by varying the initial temperature, pressure, or pH to examine the effects of altering the reaction environment, as shown in Table 4 for experiments E1-E6. The generated hydrogen was recorded after 3, 7, and 14 days, and for some experiments it was extended to 21 and 28 days. At each sampling point, the system was allowed to cool to ambient temperature to prevent vaporized water from entering the sampled gas. The hydrogen gas was sampled in a transparent, sealed plastic bag, then measured using an electrochemical gas analyzer with a measurement accuracy of ±2 ppm. A similar design was conducted for the pure fayalite, pure forsterite, and 50% fayalite/50% forsterite experiments, using a reactor with a 450 mL volume, a temperature of 100 °C, and a pressure between 477 and 530 psi.

TABLE 4. Experimental design conditions for powdered samples.

2.1.2 Core-scale experiment

The iron-rich ultramafic rock samples were obtained from the Twin Sisters ultramafic complex in Washington, USA, with 92.3% Olivine (Table 3). Textural evidence of cataclasis and olivine recrystallization indicates deformation under moderate temperatures. Representing upper mantle material emplaced via NW-trending faults, Twin Sisters dunite serves as a well-preserved reference for geochemical, structural, and hydrogen generation studies.

The flooded rock sample was a cylindrical core (H1) measuring 1.5 inches in diameter and 3 inches in length. The core was cut into two equal halves (H1A and H1B) and then reassembled, with proppant placed between the halves to maintain a fracture aperture of about 1 mm (Figure 4). (Galvis-Silva et al., 2025; Shabdirova et al., 2027) provide further details of the experiment. This is realistic and consistent with natural and engineered systems, as indicated by fracture-aperture studies on geothermal outcrops (Forstner et al., 2026), hydraulic-fracturing modeling in oil and gas systems (Frash, 2025), and true-scale laboratory experiments (Ponziani et al., 2015). Moreover, field-scale core-through observations from the Hydraulic Fracturing Test Site (HFTS) project (Gale et al., 2018) and laboratory visualization studies (Chun et al., 2020) indicate that proppant grains in narrow fractures typically form monolayer-to-partial multilayer structures, bridging and packing along fracture walls under closure stress. The experimentally assembled configuration in this work reproduces this stable, laterally continuous proppant bed, representing a realistic analog of post-placement proppant distributions observed in subsurface hydraulic fractures.

The hydrogen generation experiment was conducted using a high-pressure, high-temperature (HPHT) core flooding system, and proceeded through the following steps: initially, the Thermal Stabilization Phase; to ensure thermal equilibrium throughout the core and associated fluid pathways, the system was heated and maintained at a constant temperature of 100 °C for a continuous period of 24 h. Followed by Synthetic Brine Injection: synthetic brine (0.1% NaCl) was injected into the core at a controlled flow rate of 0.01 cc/min, with inlet and outlet pressures of 124 and 75 bar, respectively. The gas-liquid separator remained closed to prevent early gas escape. Daily Sampling and Analysis: Gas and liquid effluents were collected daily for compositional analysis. The effluent gases were routed directly from the separator to a Micro Gas Chromatograph (MicroGC). And the effluent brine was collected and analyzed for pH, conductivity, and Total Dissolved Solids using handheld probes; and for major elements (Fe, Mg, Si, Al, Ca, K) via Inductively Coupled Plasma Optical Emission Spectrometry (ICP-OES) (Figure 5).

2.2 Modeling work

The models were built to represent each mineral composition, based on the Arrhenius equation to calculate the reaction rate constant as a function of activation energy, temperature, and specific area (Daval et al., 2010) as shown in Equation 1:

K = k 0 ⁡ exp − E a / R T (1)

Where K is the dissolution constant in m o l e . c m − 2 s − 1, E a is the activation energy in J . m o l e − 1 , R is the universal gas constant in J . m o l e − 1 . K − 1, and T is the absolute temperature in Kelvin, k 0 is the pre-exponent coefficient in m o l e . c m − 2 s − 1. The activation energy and the pre-exponential coefficient were treated as calibration parameters to match the experimental data within the range reported in the literature (U.S. Geological survey, 2004).

The thermal database used is the one defined as thermo.tdat, which represents 624 minerals, including the most common involved in the serpentinization reaction, such as fayalite, forsterite, brucite, chrysotile, and magnetite. The gas fugacity is calculated using the Tsonopoulos correlation (Tsonopoulos, 1974) Equations 2, 3:

B P C R T c = f 0 T r + ω f 1 T r + f 2 T r (2)

ln ⁡ φ = B P R T (3)

Where, T r is a reduced temperature, P C is the critical pressure, T c is the critical temperature, and ω is the acentric factor. B is the second virial coefficient account for the gas non-ideality.

Although H₂ concentration is the main response variable analyzed in this study, the modeled system is multicomponent. Mineral dissolution releases Fe, Mg, and Si-bearing components into solution, thermodynamic calculations evaluate aqueous speciation and secondary mineral stability, and H₂ generation occurs through oxidation of Fe(II) to Fe(III)-bearing phases. Thus, the predicted H₂ yield reflects the coupled evolution of mineral dissolution, aqueous chemistry, secondary mineral precipitation, and gas generation rather than a single-component reaction.

2.2.1 Powdered samples’ models

The powdered samples model was built using the “React” module in Geochemist’s Workbench (Bethke, C. M., and Farrell, B.). The model was built based on the conceptual model of having two main components: the initial system and the process that alters it. The model calculates the system’s initial equilibrium state, then the system changes by adding or removing reactants. The system-defined parameters are shown in Table 5.

TABLE 5. Modeling parameters for the 4 powdered samples models.

Only Fayalite and Forsterite are modeled in a form of kinetic reaction; the rest of the minerals are defined only as an initial weight percentage.

Surface was calculated knowing the average grain size using this formula: a r e a c m 2 g m = 6 g r a i n s i z e c m × d e n s i t y g m c c .

2.2.2 Reactive transport model “core sample model”

The core sample model is built using the “X1t” module in the Geochemist Workbench. The module couples the chemical reaction to fluid flow and transport in 1D. A one-dimensional flow model was adopted because fluid flow is expected to be dominated by the fracture pathway. The matrix permeability is (0.0001) mD, whereas the fracture permeability is (662) mD, corresponding to a permeability contrast of approximately (6.62 \times 10^6). This large contrast indicates that matrix flow is negligible relative to fracture flow, supporting the use of a 1D model to represent transport and reaction along the primary flow direction.

Transport calculations account for the movement of dissolved chemical species via advection, hydrodynamic dispersion, and molecular diffusion. Fick’s law calculates the dispersive flux q_D (mol/cm²/sec) of a chemical component in solution equation 4:

q D = − ∅ D ∂ C ∂ x (4)

Where ∅ is the sediment porosity, D is the coefficient of hydrodynamic dispersion (cm²/s), and C is the volumetric concentration (mol/cm³) of the component in question.

The coefficient of hydrodynamic dispersion is calculated as shown in equations 5–7 where D* is the diffusion coefficient (cm²/s), α is the dispersivity (cm), and υ is the average linear fluid velocity (cm/s), q is the flow rate (cc/sec), A is the cross-sectional area, ϕ is the porosity, and L is the length of the core (cm). The system-defined parameters are shown in Table 6.

D = D * + ∝ υ (5)

α = L 100 (6)

υ = q A × ∅ (7)

TABLE 6. Modeling parameters for the reactive transport model of the core (Twin-sister Dunite).

3 Results and discussion

3.1 Comparing experimental to model results

The lab results for the serpentinization of the powdered samples experiments, pure fayalite, 50% fayalite, pure forsterite, and the peridotite “complex rock” (21% fayalite, 38.5% fosterite, with other minerals) showed a continuous increase of hydrogen till reaching a plateau. The model results matched the lab results, with a maximum absolute error of 0.14 (Figure 6). The results show that pure fayalite has the highest hydrogen generation, which is almost five times that of the complex rock and more than twice that of the fayalite-forsterite mixture. In the pure forsterite experiment, no hydrogen gas was detected, so the results were not included in the comparison in Table 7. The lack of hydrogen from pure fosterite is likely due to the experimental temperature. The modeling framework was further extended to evaluate additional olivine compositions that were not experimentally tested. The model with 21% fayalite and 79% forsterite produced H₂ concentrations close to those of the complex rock sample, which contains a similar fayalite fraction. However, the results were not identical because the complex rock contains additional mineral components that may influence reaction pathways, Fe availability, and secondary mineral formation. These findings suggest that the presence of additional mineral components reduced hydrogen generation compared to a system containing 21% fayalite and 79% fosterite. Increasing the fayalite content to 61.5% (assuming all the other minerals were fayalite and the forsterite remained as 38.5%) substantially increased H₂ generation, producing nearly three times more hydrogen than the 21% fayalite case. These results confirm that fayalite content is a dominant control on hydrogen generation, because higher Fe(II) availability promotes greater H₂ production during serpentinization.

TABLE 7. Comparison results between experimental lab results and model results for H₂ generated “Powdered samples”.

Absolute error = M o d e l R e s u l t s − L a b R e s u l t s L a b R e s u l t s

The model-predicted end-member curves shown in Figure 6 carry important implications for stimulated geological hydrogen production in natural systems. The Fa₁₀₀ (pure fayalite) curve establishes the theoretical maximum H₂ generation achievable from a purely fayalitic olivine system under the experimental conditions (100 °C, 33 bar, 7.5 pH, water to rock ratio 4:1), providing an upper bound against which natural rock compositions can be benchmarked. The 21% fayalite model, which closely reproduces the complex peridotite experimental results, demonstrates that the geochemical model is well-calibrated for the actual mineral assemblage of the Trinity Ophiolite peridotite sample; the small residual offset between this model and the measured data reflects contributions from non-olivine phases (lizardite, magnetite, hematite, apatite) that influence Fe partitioning but are represented only as thermodynamic end-states rather than kinetic reactants. The 61.5% fayalite model represents a hypothetical iron-enriched ultramafic rock, such as a fayalite-rich dunite or an iron-metasomatized peridotite, and shows that a roughly threefold increase in H₂ yield over the natural peridotite is achievable simply by targeting Fe-richer lithologies. This has direct practical implications: geological hydrogen exploration and stimulation programs should prioritize formation screening for fayalite (Fe-endmember olivine) content, as it is the dominant predictor of H₂ yield. The close convergence of the 21% fayalite model and the complex rock experimental data also validates the use of simplified binary olivine models as useful screening tools for predicting H₂ potential across different ultramafic lithologies, even when the full mineral assemblage is complex, as long as the Fe-bearing olivine fraction is well-characterized. Importantly, neither end-member model accounts for surface passivation, flow channelization, or transport limitations that would apply at the core or reservoir scale; the end-member predictions therefore represent kinetically accessible reaction-controlled maxima that would require open-system or stimulated-flow conditions to approach in practice.

Figure 7 shows the model results for the core-scale experiment compared with the laboratory H₂ measurements. Both the experimental and modeling results show an initial increase in H₂ generation, reaching a peak around day 7, followed by a decline around day 11. This agreement suggests that the model reasonably captures the early-stage hydrogen-generation behavior Figure 7. The results of the model vs. the raw data are shown in Table 8.

TABLE 8. Comparison of results between experimental lab results and model results for H₂ generated “Core Sample”.

Absolute error = M o d e l R e s u l t s − L a b R e s u l t s L a b R e s u l t s

The main difference occurs from the plateau phase onward and extends into the declining stage. From day 7 onward, the model appears to be predicting much more than the experiment, suggesting that hydrogen reduction might have already begun. The model attributes the reduction of hydrogen solely to the depletion of the reactive mineral. In the experiment, however, the decline is likely controlled by a combination of reactive-mineral depletion at the core surface and passivation due to alteration products, flow channeling, or secondary mineral precipitation. This passivation and flow channeling may reduce fluid flow through the core, limiting access to reactive surfaces and decreasing the reaction rate. Since this passivation effect and the changes in the reactive available surface area of the core are not exactly quantified in the current geochemical model, the model deviates more strongly from the experimental results at later times. However, during the early reaction period, the model is able to predict the experimental behavior with relatively low error, as shown in Table 8.

Three coupled mechanisms are proposed to explain the late-time model-data divergence, which reaches the relative absolute error increased to 7.18, equivalent to 718%, indicating a larger deviation between the model prediction and laboratory measurement under this condition at day 19. First, the reactive transport model assumes a fixed reactive surface area throughout the simulation. In reality, serpentinization products, including chrysotile, lizardite, and magnetite, precipitate progressively on the surfaces of primary olivine and fayalite grains, forming passivating coatings that reduce the effective mineral–fluid contact area over time. This surface passivation is the same process observed experimentally as the post-peak H₂ decline: as secondary minerals deposit on reactive grain surfaces, the rate of Fe(II) oxidation and H₂ generation reduces. Because the model does not evolve surface area as a function of secondary mineral accumulation, it systematically overestimates the reaction rate at late times, which is the dominant source of divergence after the H₂ peak. Second, serpentinization is accompanied by a solid volume increase of approximately 40%–44% as low-density serpentine replaces denser olivine (Klein & Le Roux, 2020; Beinlich et al., 2020). In a confined core, this volume expansion can progressively occlude fracture apertures and pore throats, reducing permeability and limiting fluid access to fresh reactive surfaces (Malvoisin et al., 2021; Ross et al., 2025). For this experiment, we assume that the volume expansion was small due to the low temperature and confined pressure; however, the changes in permeability may not be insignificant. Malvoisin et al. (2021) demonstrated, using a 1-D coupled reaction–deformation model, that the competition between volume-expanding precipitation and deformation-induced fracturing controls whether serpentinization proceeds to completion or self-arrests; models that ignore deformation predict premature reaction termination. The current X1t model does not couple solid-volume changes to porosity–permeability evolution, so it cannot capture the progressive flow restriction that the experiment, if any, may have experienced after day 8–10. Third, the core used in this study is a single sample, which limits the ability to quantify experimental uncertainty and to distinguish between systematic geochemical effects and sample-specific heterogeneities such as pre-existing microfractures, mineral banding, or local compositional variation. These three mechanisms operate simultaneously and non-linearly: passivation reduces reactive area, volume expansion reduces porosity, and both effects compound over time, making the late-stage behavior inherently difficult to reproduce without a fully coupled reactive–transport–geomechanical framework. Incorporating dynamic surface-area evolution, porosity–permeability feedback, and rock deformation into future reactive transport models is therefore identified as the highest-priority modeling advance needed before reservoir-scale extrapolation can be made with confidence. The current model should be considered valid for the early-stage reaction period (days 0–8), during which surface area has not yet been substantially occluded and flow paths remain open.

3.2 Controlling factors: parameters’ sensitivity

For the complex rock sample, we conducted six experiments to study the effects of varying pressure and pH on hydrogen production, as shown in Table 9. On the modeling side, after the four powdered models and the core reactive transport model were verified, they were used as seeds for sensitivity analyses of temperature and pressure. An initial pH of 7.5 was selected as the baseline for the model simulations because it corresponds to the experimental condition used for the main powdered-sample calibration cases, including pure fayalite, pure forsterite, and the 50% fayalite–50% forsterite mixture, as well as the core-scale experiment. This baseline was therefore used to maintain consistency between the laboratory and modeling conditions and to isolate the effects of mineral composition, temperature, and pressure in the sensitivity analysis. We kept all the parameters constant and varied one parameter. The models ran for 30 days, and the results were compared at the final step. The models and results are shown in Table 10.

TABLE 9. Experimental lab results of the six experiments of the complex rock peridotite (powdered samples) at Temperature 150 °C, and different pressures and pH.

TABLE 10. Hydrogen generated in ppm, modeling sensitivity results, the base model is of 7.5 pH, pressure 35 bar, 100 °C, and the same water and rock weights as mentioned earlier in Table 5.

Core Sample results are taken at 7.9 days reading to be near the peak point; other results are taken at the end of 30 days.

TABLE 11. Data source used to plot the tornado chart.

3.2.1 Effect of changing pressure

The laboratory results show that increasing pressure led to a decrease in the measured H₂ concentration in the gas phase Figure 8. This trend is largely attributable to the experimental procedure used to increase pressure: additional N₂ was injected into the reaction cell, increasing the total gas moles and diluting H₂ in the headspace. Because the detector reports H₂ in ppm, a lower ppmv value can occur even if the absolute amount of H₂ produced increases. In addition, higher pressure can increase H₂ partitioning into the aqueous phase (i.e., higher dissolved H₂ at equilibrium), further reducing the headspace H₂ signal. Therefore, the observed decrease in gas-phase H₂ ppm with pressure does not necessarily imply reduced H₂ generation, but rather reflects dilution and phase partitioning effects under the experimental configuration.

To quantify the N₂ dilution effect and recover a physically meaningful estimate of total H₂ production, a back-calculation was performed for experiments E1 (35 bar), E2 (70 bar), and E3 (135 bar) at 14 days. In this setup, the 150 mL Hastelloy autoclave was pressurized entirely with high-purity N₂, so the headspace gas at reaction temperature was essentially a binary N₂–H₂ mixture. This is fundamentally different from the core-flooding setup, where H₂ is detected directly in the continuous effluent stream by Micro Gas Chromatograph with no inert gas diluent. Using the ideal gas law at 150 °C and the known headspace volumes (120, 101, and 80 mL for E1, E2, and E3, respectively), the N₂ inventories were estimated at 119, 201, and 307 mmol. From the measured H₂ ppm values and these N₂ inventories, gas-phase H₂ moles were back-calculated as approximately 43, 59, and 9 μmol for E1, E2, and E3. Dissolved H₂ contributions estimated via Henry’s law at 25 °C (K^H = 7.8 × 10⁻⁴ mol/L/atm) were negligible (<1 μmol) for all three experiments. Rock-mass-normalized corrected H₂ yields are approximately 2.9, 2.4, and 0.2 mmol/g for E1, E2, and E3, respectively. The N₂ dilution factor, the ratio of true to measured H₂ ppm, increases from approximately 618× at 35 bar to 2,826× at 135 bar, because each higher-pressure experiment required a larger N₂ charge, compounding the dilution. Despite this dominant artifact, the corrected mole-based yields still decline monotonically with pressure (E1 > E2 > E3), suggesting a genuine geochemical pressure suppression effect consistent with the core-scale modeling results. The batch autoclave ppmv data should therefore be treated as qualitative indicators only; the corrected mole estimates provide a more rigorous basis for comparison. Critically, because the core-flooding setup employs continuous brine injection and effluent-stream GC detection, it is entirely free of this N₂ artifact, making the coreflood results the more reliable dataset for evaluating pressure sensitivity at the intact-rock scale. Future batch experiments should consider pressurization via piston-accumulator or water-injection methods to decouple confining pressure from headspace gas composition.

Another explanation was presented by Portella et al. (2024), who reported that serpentinization in a peridotite sample occurred twice as fast at 13.4 MPa as at 20.7 MPa. The complexity of the serpentinization process in peridotite involves the hydration reactions of olivine and orthopyroxene, and lower pressure enhances spinel dissolution of aluminum (Al), accelerating olivine serpentinization and delaying orthopyroxene dissolution.

The modeling results for the powdered samples under varying pressure are more consistent with the expected behavior in the literature: as pressure increases, more hydrogen is generated, as shown in Figure 9.

The observed decrease in hydrogen generation with increasing pressure in the core-scale experiment contrasts with the trend observed in powdered-sample models Figure 10. This difference is likely due to a change in the controlling mechanism from reaction-controlled behavior in powdered systems to transport-limited behavior in the core sample. In powdered samples, the high reactive surface area allows more direct water–mineral contact, whereas in the core sample, reaction progress depends on connected porosity, fracture permeability, and fluid accessibility to reactive surfaces. Increasing pressure may reduce permeability by closing microfractures or pore throats, while serpentinization products may further passivate mineral surfaces and clog flow pathways. As a result, higher pressure can limit fluid access to fresh reactive surfaces and reduce the effective reaction rate (Malvoisin et al., 2021). built a 1-D model that showed that the models that take into consideration the solid volume increase, clog porosity, and stop the reaction, while models that account for the deformation that accompanies the solid volume increase allow the fluid pathways to still be open and the reaction to proceed.

However, it is noted that in both types of models, pressure has a secondary effect compared to changing other parameters, such as temperature, as it will be discussed in later sections.

3.2.2 Effect of changing pH

The effect of pH on hydrogen generation depends on the reaction path the minerals follow. Wang et al. (2019) reported results from experiments on iron-rich rocks, showing that adding sodium bicarbonate (NaHCO₃) decreased the pH from 10.92 to 8.25 and accelerated the dissolution of both olivine and brucite, thereby improving hydrogen generation. However, maintaining a high pH still improves olivine dissolution but suppresses brucite dissolution, which may have led to a decrease in the total available iron ions and, accordingly, in hydrogen generation. McCollom et al. (2020) had similar observations: increasing the pH improved olivine dissolution, but in that case, the improvement was sufficient to enhance hydrogen generation.

In this study, the pH effect was evaluated at 150 °C under two pressure conditions. At 35 bar, H₂ generation was consistently higher at pH 8.5 than at pH 9.5 throughout the experiment (Figure 11). The pH 8.5 case increased from 51 ppm at 3 days to 431 ppm at 28 days, whereas the pH 9.5 case remained much lower, increasing only from 12 ppm to 59 ppm over the same period. This indicates that, at low pressure, the less alkaline condition favored sustained H₂ generation, likely by enhancing mineral dissolution and Fe(II) release.

At 70 bar, the pH response was more time-dependent (Figure 12). During the first 7 days, pH 9.5 produced more H₂ than pH 8.5; however, by 14 days, pH 8.5 generated the highest H₂ concentration, reaching 294 ppm compared with 240 ppm at pH 9.5. The pH 10.5 case produced the lowest H₂ throughout the experiment, reaching only 91 ppm at 14 days. These results suggest that higher alkalinity suppresses H₂ generation under the tested conditions, while moderately alkaline pH supports greater reaction progress over longer reaction time.

The comparison between 35 and 70 bar indicates that the pH effect is pressure-dependent. Although pH 8.5 produced the highest final H₂ concentration in both cases, the early-time response at 70 bar differed from the low-pressure trend. This suggests that pressure may modify the influence of pH through changes in mineral dissolution, aqueous speciation, gas solubility, or secondary phase formation. Therefore, pH should be interpreted as a coupled geochemical control rather than an independent variable. Overall, the results show that moderately alkaline conditions enhance H₂ generation, whereas higher pH values reduce H₂ production, likely due to suppressed dissolution and/or passivation of reactive mineral surfaces.

3.2.3 Effect of changing temperature

Increasing temperature generally accelerates serpentinization kinetics, thereby promoting hydrogen generation. This temperature dependence is reflected in the simulation results for the powdered pure fayalite, the fayalite–forsterite mixture, and the complex ultramafic rock. However, the magnitude of the response varies with mineral composition. Increasing temperature from 50 °C to 100 °C increases the simulated hydrogen generation by approximately fourfold for the pure fayalite case, compared with a smaller increase (less than threefold) for the 50% fayalite mixture and less than a twofold increase for the complex rock. This reduced sensitivity at lower fayalite content is consistent with a lower availability of reactive Fe(II) phases and a shift in reaction pathways. Finally, the temperature-driven increase becomes progressively limited as reactive mineral mass is depleted; once the limiting mineral is largely consumed, further temperature increases produce only marginal additional hydrogen, as shown by comparing 100 °C and 200 °C in Figure 13. Temperature had a strong positive effect on H₂ generation. However, the temperature sensitivity is limited by the thermal dataset available, which covers only 0 °C–300 °C.

In the core-scale experiment, increasing temperature accelerates serpentinization and shifts the H₂ production peak to earlier times, as shown in Figure 14. The 200 °C case reaches the highest peak, approximately 6,700 ppm around day 3, while the 100 °C case peaks at about 3,300 ppm around days 7–9. Therefore, increasing the temperature from 100 °C to 200 °C nearly doubles the peak H₂ concentration and significantly increases the reaction rate. However, the higher-temperature cases also decline more rapidly after the peak, likely because reactive mineral surfaces are consumed earlier. As a result, high temperature enhances early H₂ generation, but it may lead to lower H₂ production later due to faster depletion of the reactive material.

It is important to note that the core-scale experimental study comprised a single core specimen (H1). This represents a significant experimental constraint: the absence of replicate cores means that sample-specific heterogeneities, including pre-existing microfractures, local mineral banding, olivine grain size distribution, and initial porosity variation, cannot be distinguished from systematic geochemical responses. Quantitative uncertainty bounds on the core-scale H₂ generation rate and peak concentration cannot be established from a single experiment. The core-scale results should therefore be interpreted as a proof-of-concept demonstration of the reaction–transport coupling behavior rather than as a statistically representative dataset. Future work should attempt to replicate core experiments under identical conditions to quantify inter-sample variability, and cores from different depth intervals or lithological units to assess the sensitivity of the results to natural rock heterogeneity. These limitations are explicitly acknowledged in the interpretation of transport-limited behavior and model–data divergence in Section 4.

Figure 15 is a tornado plot summarizing the sensitivity analyses of the parameters that impact hydrogen generation. It shows that fayalite content is the most influential parameter controlling H₂ generation, followed by temperature, pH, and finally pressure. Increasing fayalite content from 21% to 100% produced the largest increase in H₂ relative to the 50% fayalite baseline, confirming that Fe(II)-bearing silicates are the main source of reducing capacity during serpentinization. Temperature also had a strong positive effect, with increasing temperature from 60 °C to 80 °C substantially increasing H₂ generation, indicating enhanced reaction progress at higher temperatures.

In contrast, pH showed a moderate inverse effect: increasing pH from 9.5 to 10.5 reduced H₂ generation, while lowering pH to 8.5 increased it. This suggests that higher alkalinity may limit reaction progress, possibly by reducing mineral dissolution or promoting surface passivation by secondary phases. Pressure had the smallest effect in the simulations, with only minor changes in H₂ over the tested pressure range, indicating that pressure is not a primary control on modeled H₂ generation under these conditions.

Overall, the sensitivity ranking indicates that mineral composition and temperature are the dominant controls on H₂ generation, while pH acts as a secondary geochemical control, and pressure has a limited direct effect in the model.

4 Research gaps, limitations, and future directions

Although this study provides an integrated experimental and modeling framework for evaluating geochemical H₂ generation, several limitations remain. First, the laboratory measurements primarily quantify gas-phase H₂. Under elevated pressure, a portion of the generated H₂ may dissolve into the aqueous phase, while pressurizing gases can also dilute the headspace concentration. Therefore, gas-phase H₂ measurements alone may underestimate total hydrogen generation, particularly under high-pressure conditions. Future experiments should include a full hydrogen mass balance that accounts for both gas-phase and dissolved H₂ to improve comparison across pressure conditions.

Second, the current kinetic batch models successfully reproduce powdered-sample experiments because these systems have high and relatively uniform mineral–fluid contact area. However, this assumption becomes less valid in intact or fractured rocks, where fluid flow is controlled by connected pathways and reactive surfaces may not remain uniformly accessible. The core-scale reactive transport model reproduced the early-time H₂ generation trend but deviated from the laboratory results after the peak concentration was reached. This late-time divergence suggests that additional processes, such as depletion of accessible reactive material, permeability reduction, secondary-mineral coating, and surface passivation, are not fully represented in the current model.

Although the present work quantifies H₂ generation as a function of mineral composition and reaction conditions, it does not directly characterize post-reaction mineralogical alteration or rock deformation. These processes are important for field-scale geological hydrogen systems because serpentinization may produce secondary minerals that coat reactive surfaces, modify pore and fracture structure, and alter permeability.

A key research gap is the treatment of reactive surface area evolution. Most standard geochemical and reactive transport models assume that reactive surface area is fixed, homogeneous, or directly proportional to mineral abundance. In fractured systems, however, serpentinization products may precipitate preferentially along fracture walls or flow constrictions, reducing access to reactive surfaces in a spatially heterogeneous manner. Future modeling work should incorporate time-dependent and spatially variable reactive surface area to better represent surface coating, fracture-pathway occlusion, and the progressive reduction of mineral–fluid contact.

Another important limitation is the absence of geomechanical feedback in the present modeling framework. Serpentinization is associated with solid-volume expansion, which can either create new fractures and expose fresh reactive surfaces or seal existing pathways and reduce permeability. These competing processes may strongly influence H₂ generation in natural and stimulated systems, but were not explicitly represented in the current experiments or simulations. Future studies should therefore couple geochemical reactions with geomechanical deformation, fracture evolution, and permeability changes.

Porosity evolution and permeability changes were not directly measured or dynamically updated in the current model. However, they are expected to influence the late-time behavior of the core-scale experiment. During serpentinization, dissolution of primary minerals and precipitation of secondary phases can modify the pore and fracture network. Secondary mineral precipitation along fracture walls or pore throats may reduce effective porosity and permeability, restrict flow pathways, and decrease access to fresh reactive surfaces. Although in the core model, the porosity and permeability of the fracture are much higher than those of the matrix. The system is confined so that pore-volume expansion is not allowed. Flow is continuous within the fracture, thereby removing reaction products. As such, changes in matrix porosity and permeability will be negligible, and changes in fracture porosity and permeability were not measured, but should not have a significant impact at the beginning of the experiment, though perhaps at the end. This provides a plausible explanation for the late-time deviation between the core-scale experiment and the reactive transport model, particularly after the peak H₂ concentration was reached. Future work should track the time-dependent permeability evolution to better constrain the model.

Finally, additional work is needed to upscale laboratory observations to reservoir-scale geological hydrogen systems. Future research should focus on: (i) quantifying dissolved and gas-phase H₂ simultaneously under pressure; (ii) improving representation of evolving reactive surface area and secondary-mineral passivation; (iii) combining H₂ measurements with post-reaction mineralogical characterization; (iv) developing coupled geochemical–geomechanical models that capture volume expansion, fracture propagation, and permeability evolution; and (v) extending reactive transport simulations from core-scale fracture systems to heterogeneous reservoir-scale domains. These advances are necessary to translate laboratory serpentinization results into predictive tools for evaluating geological hydrogen generation at the field scale. In addition, Microbial processes were not included in this study because the experiments and models were designed to isolate abiotic hydrothermal water–rock reactions. However, microbial hydrogen production or consumption may be important in lower-temperature subsurface systems and should be considered in future studies that evaluate field-scale geological hydrogen generation and preservation.

5 Conclusion

This study investigated geochemical H₂ generation from ultramafic minerals using laboratory experiments, kinetic–thermodynamic batch modeling, and core-scale reactive transport modeling. The results show that the experimental scale strongly influences the interpretation of H₂ generation. Powdered-sample systems are primarily controlled by reaction kinetics because of their high exposed reactive surface area, whereas core-scale flow-through systems are controlled by coupled reaction and transport processes.

Mineral composition was the dominant control on H₂ generation in powdered samples. H₂ yield increased with increasing fayalite content, confirming the importance of Fe(II)-bearing silicates as the main source of reducing capacity during serpentinization. Temperature also showed a positive effect, with higher temperatures accelerating H₂ generation and increasing production, particularly in fayalite-rich systems.

The effect of initial pH was pressure-dependent. Lowering pH from 9.5 to 8.5 increased H₂ generation in the laboratory experiments, but the magnitude of this increase differed between 35 and 70 bar, highlighting the need to evaluate coupled parameter effects rather than interpreting pH independently. Pressure had a comparatively smaller direct effect in the powdered-sample simulations, producing only minor increases in total H₂ generation. However, laboratory gas-phase measurements showed lower detected H₂ at higher pressure, likely due to increased H₂ dissolution in the aqueous phase and dilution effects from pressurizing gas.

The kinetic batch models successfully reproduced the powdered-sample experiments, indicating that the thermodynamic and kinetic framework captures the main geochemical controls on H₂ generation. In contrast, the core-scale reactive transport model reproduced the early-time H₂ generation trend but diverged from the laboratory results after the peak concentration was reached. This late-time deviation suggests that transport limitations, evolving reactive surface accessibility, permeability reduction, and/or surface passivation become increasingly important in fractured rock systems.

Overall, the findings show that mineralogy and temperature are the primary controls on geochemical H₂ generation, while pH and pressure act as secondary controls whose effects depend on experimental conditions and measurement approach. The comparison between powdered and core-scale systems demonstrates that transport–reaction coupling must be considered before laboratory serpentinization results can be translated to field-scale geological hydrogen systems.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Eleni Gianni, Ionian university, Greece

Reviewed by: Muhammad Sahimi, University of Southern California, United States

Sarath Poda, Texas Tech University, United States