Nickel Nanoparticles Boost CO2 Dissolution in Seawater, With an AI Model Pinpointing the Sweet Spot

Dissolved ions such as sodium, chloride, magnesium, and sulfate hoard the water molecules that CO2 needs to dissolve, a phenomenon known as the salting-out effect, which makes gas uptake in the ocean measurably slower than in fresh water. Now a team at Florida Atlantic University reports a way to push past that limit. The catch, and the most interesting part of the finding, is that more nanoparticles are not always better.

At the flow-focusing junction, hydrodynamic shear sliced the gas stream into uniform microbubbles, which then traveled down a 20-millimeter observation channel. As CO2 molecules dissolved into the surrounding liquid, each bubble shrank, and a high-speed camera recording at 3,500 frames per second captured the decline. The relative reduction in projected bubble area between inlet and outlet served as a direct optical readout of dissolution efficiency, with reproducibility verified across 72 videos showing less than 2.3 percent variation in initial bubble size.

The catalytic logic behind the enhancement rests on the chemistry of nickel nanoparticle surfaces. In water, the roughly 100-nanometer nickel particles become hydrated, forming hydroxylated sites that act as adsorption platforms for CO2 right at the gas–liquid interface, where the nanoparticles accumulate to minimize interfacial energy. Once CO2 adsorbs, it is converted toward bicarbonate species; the nanoparticle–bicarbonate complex then diffuses into the bulk liquid through Brownian motion and microconvective flow, releases its bicarbonate cargo, and regenerates the hydroxylated surface for the next cycle.

This adsorption–transport–desorption loop keeps depleting the CO2 concentration near the bubble surface, sustaining the gradient that drives further dissolution. Smaller, well-dispersed particles offer more active surface area, which is precisely what salinity threatens to destroy. That threat is why the polymer matters.

Seawater's high ionic strength compresses the electrical double layer surrounding each nanoparticle, weakening electrostatic repulsion and letting van der Waals attraction pull particles into clusters. Aggregated particles present far less surface area and drift more slowly toward the bubble interface, gutting the catalytic machinery. CMC counters this with a double defense: its ionized carboxylate groups generate electrostatic repulsion while the polymer chains themselves provide steric hindrance, keeping particles apart.

The baseline controls told a clean story — plain seawater dissolved 21.8 percent of the CO2, deionized water 38.4 percent, and seawater with 100 milligrams per liter of CMC alone reached 48.3 percent. Adding nickel nanoparticles on top of that pushed dissolution to 57.2 percent, an increment the authors attribute specifically to catalytic activity rather than the polymer or the medium. But the headline result is the shape of the response curve.

At a fixed CMC concentration of 100 milligrams per liter, dissolution peaked at 57.2 percent with 10 milligrams per liter of nanoparticles, then fell steadily as loading increased — dropping to 26 percent at 70 and just 18.4 percent at 150 milligrams per liter. Below the peak, performance was also modest, at 43.7 percent for 6 and 50.5 percent for 8 milligrams per liter, because too few particles were available to work the interface. The explanation is a stabilization budget: each polymer chain can only coat so much particle surface.

Once nanoparticle loading outstrips the available CMC, repulsion weakens, clusters form, and the effective catalytic area collapses. The result is a bell-shaped curve with a single optimum at each polymer concentration — a nonlinear behavior that simple one-factor-at-a-time experiments would miss. Raising the polymer dose moved the sweet spot.

At 200 milligrams per liter of CMC, the optimum shifted to 70 milligrams per liter of nanoparticles, with dissolution peaking at 58.9 percent. At 300 milligrams per liter of CMC, the optimum landed at 90 milligrams per liter of nanoparticles, where dissolution reached 67.2 percent — the highest value recorded across all 24 tested conditions. Zeta potential measurements tracked the underlying colloidal stability.

At the lowest CMC level, zeta potentials hovered between 8 and 19 millivolts, consistently below the roughly 20 to 25 millivolt threshold associated with stable dispersions in high-salt media. At 300 milligrams per liter of CMC, values stayed above 29 millivolts from 30 milligrams per liter of nanoparticles onward, peaking at 36 millivolts at the optimum. The dissolution peak and the stability peak shifted together, exactly as Derjaguin–Landau–Verwey–Overbeek (DLVO) theory — the standard framework balancing van der Waals attraction against electrostatic double-layer repulsion — predicts.

The reverse failure mode proved just as instructive. When nanoparticles were scarce and polymer abundant, dissolution actually worsened with more CMC: at 6 milligrams per liter of nanoparticles, dissolution fell from 43.7 percent at 100 milligrams per liter of CMC to a dismal 11.6 percent at 300 milligrams per liter. In that regime, excess polymer chains crowd onto the limited particle surface and partially block the very hydroxylated sites where CO2 adsorption happens.

The particles stay perfectly dispersed, but they are effectively muzzled. So the optimum at each CMC level marks the narrow window where neither aggregation nor site-blocking dominates — where particle number, colloidal stability, and accessible surface area are simultaneously sufficient. The authors are careful to note that their microfluidic bubble-shrinkage measurements characterize time-dependent dissolution and mass transfer, not equilibrium solubility, so the reported optima apply to the investigated flow conditions rather than to thermodynamic solubility limits.

Because the response surface is nonlinear in two coupled variables, the team turned to machine learning to squeeze more out of their data. The model's performance was striking. Across all 24 experimental conditions, predicted values hugged the 1:1 parity line over the full dissolution range from 0.11 to 0.67, with a coefficient of determination of 0.9999 and a maximum point-by-point deviation of 1.03 percent.

Residuals scattered around zero with no systematic trend, and the only visible deviations — near dissolution values of 0.44 and 0.47 — sat in the steepest transition zones near each peak and stayed within 2 percent. To test generalization on genuinely fresh data, the team ran six new measurements at untrained concentration combinations; the model achieved a mean relative error of 2.65 percent, with individual deviations from 1.10 to 7.36 percent. The largest error occurred at 9 milligrams per liter of nanoparticles and 250 milligrams per liter of CMC, squarely inside the steepest part of the response curve, where dissolution jumps from about 0.15 to 0.46 over just 2 milligrams per liter of nanoparticle loading and no training point exists.

That error falls comfortably within the 2 to 5 percent range reported for comparable GPR models applied to CO2 capture systems. The trained model then painted a continuous prediction map across the entire concentration domain, and its maximum landed near 290 milligrams per liter of CMC and 85 milligrams per liter of nanoparticles — closely matching the experimental optimum and offering a concrete target that the coarse 24-point grid alone could never have resolved. The heatmap also revealed that below 200 milligrams per liter of CMC, dissolution is far more sensitive to nanoparticle loading than to polymer dose, while above that threshold the surface flattens as robust colloidal stability buffers moderate changes in particle concentration.

The authors suggest the system could eventually be integrated into a closed, recirculating seawater-based capture process in a confined reactor, with the magnetic properties of nickel enabling catalyst recovery by magnetic separation — though they caution that recovery efficiency and continuous large-scale performance remain untested.

Subject of Research: Enhancement of CO2 dissolution in seawater using CMC-stabilized nickel nanoparticles studied with microfluidics, DLVO colloidal stability analysis, and Gaussian Process Regression modeling

Article Title: CO2 dissolution enhancement in seawater by CMC-stabilized nickel nanoparticles

Article References: Abhishek, Donjuan, J., Berthelot Frot, A., Nasrabadi, A. M., Ratanpara, A., Venkat, V., Gumbinger, C., & Kim, M. (2026). CO2 dissolution enhancement in seawater by CMC-stabilized nickel nanoparticles. Frontiers in Materials, 13, Article 1961882. https://doi.org/10.3389/fmats.2026.1961882

Image Credits: AI Generated

DOI: 10.3389/fmats.2026.1961882

Keywords: carbon capture, seawater, nickel nanoparticles, carboxymethyl cellulose, CO2 dissolution, microfluidics, Gaussian Process Regression, machine learning, colloidal stability, DLVO theory, zeta potential, ocean alkalinity