The partial pressure of CO₂ (P _CO2) in Earth’s atmosphere has varied considerably over Phanerozoic time. Oceanic bicarbonate ion (HCO₃ ⁻) concentration and alkalinity of the ocean would have followed the changing P _CO2. It is currently assumed that carbon in the oceans is in equilibrium with Earth’s atmosphere and that it comprises a single phase. By knowing any two of the variables P _CO2, pH, dissolved inorganic carbon (DIC) and alkalinity, for a given temperature, salinity and pressure, all the other parameters can be calculated. Recent studies of calcium carbonate (CaCO₃) solubility and formation in ocean water with changing P _CO2, provide evidence for the effects of prenucleation clusters and phase separations of calcium (Ca²⁺) and carbonate ions (CO₃ ²⁻) (Gebauer and Coelfen, 2011) that affect carbonate chemistry.

We report here the existence of droplets of a liquid condensed phase (LCP) that are bicarbonate-rich, or hyperalkaline, relative to the bulk solution. The LCP droplets have been identified in ancient and modern seawater compositions (Figures 1A, B, respectively), industrial CO₂ capture and mineralization solutions for CO₂ sequestration (Figure 1C) and blood serum (Figure 1D). A LCP is present in simple sodium carbonate (Na₂CO₃) solutions over a range of pHs for HCO₃ ⁻ as well as in more complex solutions from which carbonate minerals form, such as electrolyte solutions forming CaCO₃ polymorphs similar to those seen in CaCO₃ skeletal biomineralization or solutions containing sodium bicarbonate (NaHCO₃) with varying ratios of Ca²⁺ and magnesium (Supplementary Figure S1). The presence of LCP requires that seawater, capture fluids in CO₂ sequestration, blood and biomineralizing solutions be viewed as two-phase rather than one-phase systems. As a result, calculated estimates of the above-mentioned variables of these systems from two measured values do not fully constrain the system. When Ca²⁺ is present in these systems, even at near-neutral pH where HCO₃ ⁻ is dominant, the HCO₃ ⁻ is particularly concentrated in the hyperalkaline LCP phase, which leads to CaCO₃ precipitation and evolution of gaseous CO₂.

Nuclear magnetic resonance (NMR) spectroscopy has been used in the past to characterize bicarbonate-bearing LCPs (Bewernitz et al., 2012) and early HCO₃ ⁻/CO₃ ²⁻ nucleation behavior (Gebauer et al., 2010) in those systems. ¹³C NMR data confirm that the LCPs are bicarbonate-rich relative to the bulk solution (Figures 2A, B). A NaHCO₃/sodium chloride solution was characterized using one- (1D) and two-dimensional (2D) transverse relaxation measurements in order to identify and analyze the LCP, (see Figures 2A, B, respectively). The 1D relaxation measurement in Figure 2A shows a doublet in the spectrum that is consistent with the presence of two similar, yet non-identical solution states coexisting in equilibrium; LCP and mother solution. Deconvolution of the two peaks suggests that, over the course of the four-second time average used to acquire the data, approximately 30% of the inorganic carbon in solution is present as LCP. The CPMG T₂ relaxation measurement in Figure 2B demonstrates that the two phases have different T₂ relaxation times, re-enforcing the notion that the two peaks are due to HCO₃ ⁻ residing in two distinct solution environments. Bewernitz et al. had previously used NMR in a similar fashion to support conclusions about LCP in supersaturated HCO₃ ⁻ solutions containing Ca²⁺ (Bewernitz et al., 2012). The results presented in this report are significant in that they demonstrate the presence of an LCP in the absence of divalent cations and at concentrations that are undersaturated with respect to any solid phases, such as NaHCO₃, suggesting that LCPs are not a specific step in the nucleation process, but rather an ubiquitous and fundamental electrolyte behavior occurring in solutions containing HCO₃ ⁻.

Our explanation for the “shoulder” observed in the NMR spectra of HCO₃ ⁻ solutions is centered on the assumption that there are two distinct chemical environments that are similar but non-identical in which HCO₃ ⁻ resides. However, an alternative interpretation of the data suggests that the shoulder is due to chemical shift anisotropy (Buckingham, 1960; Carper et al., 2004). This phenomenon occurs due to the asymmetric susceptibility of the ¹³C nucleus in a static magnetic field (B₀), and depends on the ¹³C nucleus orientation with respect to B₀. In solution-state NMR, the differences are usually averaged out due to rapid tumbling. In solid-state NMR, the differences are strongly manifested as broad peaks that encompass all the static orientations of the NMR-sensitive nuclei. For the NaHCO₃ solution presented in Figures 2A, B, an intermediate situation is present. According to a recent definition of the HCO₃ ⁻ tensors (Stueber et al., 2002), there appears to be rapid averaging of the HCO₃ ⁻ chemical shift tensors in the δ_P and δ_⊥ tensors, but much less rapid averaging in the δ_a tensor. This situation is consistent with some ordering that occurs between HCO₃ ⁻ aligning at an interface similar to amphiphilic species that make up the interface of micelles, emulsion droplets, or vesicles with the δ_P and δ_⊥ tensors aligned and rotating at the interface, and the δ_a tensor in relative stasis (Lindstrom et al., 2006). In order to verify this possibility, however, high-resolution magic angle spinning experiments, similar to those conducted on micelles, vesicles, and liposomes, need to be carried out with the bicarbonate-rich LCP.

Light scattering measurements by nanoparticle tracking analysis (NTA) were used to calculate a first-order approximation of the refractive index (RI) of bicarbonate-rich LCP droplets (see Supplementary Figure S2). NTA has proven successful in providing estimated RIs in earlier studies (Filipe et al., 2010; Bewernitz et al., 2012). In NaHCO₃-NaCl solution, the bicarbonate-rich LCP droplets show a low intensity, polydisperse size distribution. This suggests that the droplets have a lower RI than the mother solution and that their RI is a range rather than a single value. Figure 2C shows a best-fit to statistically relevant data points, giving a calculated RI of 1.355 for the LCP droplets. To encompass the range of the diverse RI, a lower-limit fit estimation, inclusive of all the LCP droplets, gives a RI of 1.347. Salinity of the bicarbonate-rich LCP was estimated by plotting saltwater RI vs. salinity (Quan and Fry, 1995), and is shown in Figure 2D. The values are consistent with a bicarbonate-rich LCP, having salinity between 20 and 60 g/L, that is phase-separated from a mother solution with salinity of 15 g/L. In other words, the bicarbonate-rich LCP has a salinity of four times greater (or more) compared to the mother solution. This demonstrates just how condensed the LCP can be and, to the best of our knowledge, represents the first attempt to estimate the salinity of bicarbonate-rich LCP droplets using this novel technique.

As the NMR characterization showed, the large fraction of carbon present in the LCP is consistent with the amount of “missing” HCO₃ ⁻, as would be defined by the various empirically measured fitting functions and activity coefficients used to model electrolyte solutions. This suggests that the formation of LCP is part of a fundamental mechanism that has been detected in previous studies but, to this point, has not been explicitly included in electrolyte solution models. The large fraction of carbon in the two-phase LCP system (see Figure 2B) plays a significant role in the chemistry and mineral nucleation of various HCO₃ ⁻-containing systems. Explicit accounting for LCP may in fact simplify our thermodynamic and kinetic models of electrolyte solutions and provide new insights into previously unexplained behaviors. Electrolyte solutions deviate from ideal behaviors by lowering the activity of individual ions. Many such mechanisms are suspected to contribute, at least in part, to the loss of activity. These include Debye-Hückel screening, ion-pairing, changes in ion hydration and solvent structure, and the recently discovered prenucleation clustering (PNC) (Gebauer et al., 2008). LCP (Bewernitz et al., 2012) is an additional phenomenon that can account for a portion, or even the majority of, the non-ideality of concentrated salt solutions, simply by lowering the ion activity due to the incorporation of ions into an LCP phase. The ions present in a separate immiscible phase do not contribute to thermodynamic measurements of the “mother solution” so would tend to lower the activities of those species (see Figure 3). We propose a new approach to modeling salt solution properties by considering the formation of a two-phase bicarbonate-rich LCP system. For additional data and experiments that correlate the LCP to non-ideal behavior, see Supplementary Figures S3, S4.

Perhaps the most significant implication of the discovery of bicarbonate-rich LCP lies in the application of carbon capture and sequestration. The ability to manipulate and isolate LCP is unlike any other approach to CO₂ capture and utilization. Based on the mineralization process used by calcifying aquatic invertebrates that form mineralized skeletons, CO₂ is converted to solid carbonate material via the reaction of Ca²⁺ with two equivalents of HCO₃ ⁻, Eq. 1 (ΔG = −27.3 kJ mol⁻¹). By carrying out the process at near-neutral pH where DIC speciation is predominantly HCO₃ ⁻, relative to dissolved CO₂ (H₂CO₃) and CO₃ ²⁻, organisms are able to control the crystal size, shape, orientation, phase, texture and location during the deposition of CaCO₃ biominerals. CaCl 2 aq + 2 NaHCO 3 aq = CaCO 3 s + CO 2 aq + H 2 O l + 2 NaCl aq (1)

That the mineralization process happens at physiological pH has huge implications for large-scale CO₂ sequestration and production of synthetic carbonate solids. In fact, we have verified, at laboratory scale, reaction conditions for Eq. 1 in the range of pH 6–8.5, producing yields of CaCO₃ and CO₂ of 80% and 90%, respectively (see Supplementary Figure S5). The biomimetic approach has been verified in a continuous process as well, whereby a solution of NaHCO₃ is introduced into a solution of CaCl₂ in a continuous flow set-up with flow-rates on the order of several liters per minute.

Typical approaches to commercial synthetic CaCO₃ minerals involve reacting Ca²⁺ with CO₃ ^2-, Eq. 2 (ΔG = −49.87 kJ mol⁻¹). This reaction occurs at roughly pH 10 due to the predominance of CO₃ ²⁻, relative to H₂CO₃ and HCO₃ ⁻. The high pH of the reaction in Eq. 2 is therefore a major limitation to prepare synthetic CaCO₃ because of the need to increase the alkalinity in the system. CaCl 2 aq + Na 2 CO 3 aq = CaCO 3 s + 2 NaCl aq (2)

It is commonly assumed that CO₂ capture and geological sequestration will provide the primary solution for managing carbon emissions on a world-wide sustainable basis. This requires that CO₂ be in a substantially pure form as it must be compressed and liquefied for transport and injection into subsurface geological reservoirs, which will require subsequent monitoring. The most significant quantities of carbon emissions originate from Portland cement plants, coal- and natural gas-fired power plants, all of which emit dilute streams of CO₂, but contain mainly nitrogen. Current state-of-the-art technologies to purify CO₂ from industrial flue gas, such as amine scrubbing (Rochelle, 2009), are energy intensive primarily due to the stripping of pure CO₂ out of the capture solution. At a coal-fired power plant, for example, purifying the CO₂ from a flue stream can require more than 35% of the electricity generated by the plant. In the context of the reaction in Eq. 1, however, where two equivalents of CO₂ (as HCO₃ ⁻) produce one equivalent of pure CO₂ and one equivalent of sequestered CO₂ (as CaCO₃ for building materials), such energy intensive loads might be circumvented.

At scale, we envision a solution to carbon capture and sequestration as a four-stage process (see process flow diagram in Supplementary Figure S6). CO₂ capture solutions are created using an alkali recovery process driven by ion concentration gradients. Cation-selective dialysis membranes remove protons (H⁺), i.e., acid, from NaCl feed solution into a draw solution, taking with it a chloride ion (Cl⁻). What is left behind is a feed solution that is higher in alkalinity. The capture solution is combined with a source of gaseous CO₂ in a gas-liquid contactor to form HCO₃ ⁻ solutions that have the propensity to separate into LCP droplets. The bicarbonate-rich LCP is separated and recovered from their bulk solution by established membrane desalination technology. Nanofiltration (NF) has proven to be especially effective. Using NF membranes we have found that the droplets do not behave like ions, but instead act like intact larger moieties that are rejected by the membrane, and therefore concentrating the LCP droplets (see Supplementary Figures S7, S8). Combination of the now concentrated bicarbonate-rich LCP with hard Ca²⁺ brine solution results in the formation of synthetic CaCO₃ and the concomitant evolution of CO₂, by way of the reaction in Eq. 1. If, for example, the process were fitted to a 500 MW coal-fired power plant, 9,410 tons per day (TPD) of solid CaCO₃ mineral and 4,138 TPD pure CO₂ would be produced, assuming 52% recovery of CO₂ (see Supplementary Figure S9 for detailed metrics at each stage). The energy consumption lies mainly in pumping the water required for the process.

Ideally, the synthetic CaCO₃ produced in the process will be used in the built environment in the form of concrete. A formulation of water, cement and aggregate constituents, concrete is the most used building material in the world and represents the largest potential sustainable reservoir for CO₂ sequestration. The carbon footprint of a cubic yard (yd³) of concrete, however, is tremendous, almost entirely due to the energy intensive process of manufacturing ordinary Portland cement (OPC) (National Ready-Mix Concrete Association, 2012). It is therefore of significant interest to replace traditional components of concrete with novel, carbon-reducing components that originate from LCP solutions. For example, Blue Planet’s CarbonMix liquid (Figure 1C) is a concentrated LCP solution that is used as a complete water replacement in concrete formulations. When used in formulations where OPC is replaced by, natural limestone, CarbonMix liquid has a significant impact on the carbon footprint per yd³ concrete. This includes the offset of CO₂ that would have otherwise come from the manufacturing of OPC, as well as storage of CO₂ sequestered in the concrete. Figure 4A shows the time-dependent compressive strength data for a series of mortar specimens that use CarbonMix liquid with a reduced OPC formulation, which has a pronounced effect on the curing properties. Other examples of carbon-reducing components for concrete include cement and aggregate replacement by synthetic CaCO₃. These carbon-reducing components have an even bigger impact on the carbon footprint per yd³ concrete (Figure 4B). This is illustrated further by the abovementioned mortar specimens, which have a lower carbon footprint the ordinary mortar specimen formulation (Figure 4C).

The carbonate minerals produced from LCP solutions are not only valuable to concrete, but we’ve also found that minerals precipitated via the reaction in Eq. 1 exhibit unusually high solar reflectance (SR) relative to known commercial and natural materials (Figure 4D). High SR is desirable for commercial cool roofing technologies as a means to mitigate heat island effects in urban areas (Akbari et al., 2009; Santamouris et al., 2011), reducing the convective and radiative thermal gain of a surface.

In the field of biomineralization, it has long been proposed that inorganic carbonates can be induced, through biological additives, to condense into a concentrated liquid phase just prior to mineral formation, specifically CaCO₃ (Gower and Odom, 2000; Faatz et al., 2004; Gower, 2008). The presence of charged organic moieties, such as those in seawater and in serum, appear to have a profound affect on the coalescence and abundance of the LCP droplets, and are known to be associated with biomineralization processes. For example, the putative role of the acidic glycoproteins that are ubiquitous in all carbonate and phosphate mineralized tissues has been proposed to be in mineral nucleation, crystal growth modulation and calcium transport. Given the strong influence that organic polymers in seawater and serum appear to have on the size and abundance of LCP, LCP-modulation may represent a functional role for these organic moieties, and explain their intracrystalline presence in these skeletal materials. Many CaCO₃ biomineralization processes have been difficult to explain with single-phase models. It has been assumed that pre-nucleation phases possess “CaCO₃” stoichiometry and not HCO₃ ⁻ stoichiometry. Here, we find that the initial stoichiometry may exist as a calcium bicarbonate association with LCP, especially in the presence of biological agents. We provide evidence that the bicarbonate-rich LCP does not require supersaturated conditions to form and is neither mineral specific nor transient, but rather, is a fundamental electrolyte behavior exhibited by HCO₃ ⁻, which exists at undersaturated and apparent equilibrium conditions. The resulting potential for additional control over the bicarbonate-rich LCP droplets has far-reaching ramifications in such fields as oceanography, environmental sciences (Lee et al., 2010), material sciences and carbon sequestration. Long-lived, stable LCP droplets at undersaturated conditions allow for a previously unknown means to manipulate solvated inorganic carbon chemistry, such as mechanical separation, concentration and storage of inorganic carbon in the solution state, and the engineering of synthetic minerals with superior properties.