MUMUCO: Multiscale and Multiphase Phenomena in Complex Fluids

MUMUCO

MUMUCO: Multiscale and Multiphase Phenomena in Complex Fluids

Research project selected under the 2024 call for proposals

Principal Investigator : Jérôme CUNY

Involved Teams :

• MAD (LCPQ)

Research project selected under the 2024 call for proposals

Project summary 

 

Despite their paramount importance in the global energy economy, interface phenomena in multi-phase fluids remain very poorly understood. Their understanding requires significant development and sharing of experimental and theoretical approaches. Various areas related to the energy industries are indeed expected to benefit from significant improvements thanks to a deeper understanding of these phenomena. In this context, the MUMUCO project aims to elucidate the phenomena that govern the formation, dissociation, aggregation and stability of interfacial structures in gas hydrates. These processes are of strong interest in the field of energy with applications in the areas of heat storage and environmentally friendly refrigeration. To achieve this goal, the MUMUCO project aims for the development of original modeling methods, based on the development of Deep Neural Network Potentials, allowing for the description of complex molecular systems with unprecedented accuracy. In addition to their development, application of those potentials to model gas hydrates is considered for validation of the proposed methodology.

 

1. Scientific context:

 

Gas hydrates have been proposed as materials for various technological applications. For example, hydrate sludges, composed of hydrate crystals dispersed in a carrier phase (water or liquid oil, emulsion, etc.) can be used as a two-phase secondary fluid to store and transport large quantities of cold thermal energy [1]. However, the performance of cold transfer and storage processes depends largely on the thermophysical properties of the fluid used such as its viscosity [2] or its heat transfer coefficient [3] which depend on the quantity of hydrates (fraction) and its particle size distribution. The kinetics of hydrate formation/dissociation is therefore an important aspect to take into account during the design and operation phases of these processes [4]. The use of kinetic additives (traditionally inhibitors, more recently promoters) makes it possible to improve the control of the nucleation of these species, but their development requires a better understanding of the interfacial interactions in these systems. However, the methodologies used so far remain insufficient to obtain an accurate representation of interfacial phenomena, such as the influence of additives on the rate and statistics of nucleation, on the growth and agglomeration of crystals, on the packing of particles, etc. It therefore clearly appears that a more in-depth study at the molecular scale is mandatory to make further progress in this area.

 

Experimentally, most published works at the molecular scale concern crystal growth and properties of single hydrate crystals or formation of polycrystalline crusts. Important topics such as the formation of gas hydrates in an emulsion or the influence of an interface (in the presence or not of impurities-crystallization inhibitors/promoters) on the formation of gas hydrates are barely studied. It is also surprising that for these systems, which are important for industry, no data on the nucleation ability, even for very simple hydrates (such as cyclopentane), are available in the literature. The main reason is that these systems display a large metastability zone, i.e. that significant subcooling, above 20°C, is necessary to trigger nucleation. To address this issue, it has been shown that hydrate nucleation is facilitated if ice is first crystallized. For instance, for high subcooling, Dieker et al. showed that using this kinetic pathway, cyclopentane nucleation was repeatable and instantaneous [5]. This reduction in the nucleation induction time is attributed, in the literature, to a “memory effect” [6]. This phenomenon has been reported with hydrates of CO2, hydrocarbons, and THF [7]. However, no experimental study has explained this memory effect, which is the key to understanding hydrate crystallization. To fill this gap, LGC (Laboratoire de Génie Chimique, LGC, UMR 5503) researchers, among which Dr. Sébastien Teychené, were able to adapt droplet microfluidic installations already developed at the LGC for the study of the nucleation of organic crystals, poorly soluble salts and proteins [8]. These facilities could make it possible to measure the nucleation probability, and therefore the nucleation kinetics, of a model gas hydrate (cyclopentane). These experiments will thus be able to provide quantitative data on the influence of the composition of the solution on the nucleation of gas hydrates. Furthermore, by coupling these microfluidic installations with diffusion techniques (SAXS, WAXS and XDR) and Raman spectroscopy, the structure of the pre-nucleation clusters can be followed during the crystallization of the hydrates. The LGC team has solid expertise in the development of such microfluidic chips compatible with X-ray scattering experiments [9]. As part of Raj Kumar Ramamoorthy’s post-doc (ANR funding 2019 – 2023), the LGC notably developed microfluidic systems particularly suited to the study of gas hydrates (transparent systems that can be cooled to temperatures of around -50°C, resistant to the use of organic solvents and to a few tens of bars of pressure). These systems made it possible to carry out preliminary measurements of gas hydrate dissociation. However, the relevant interpretation of SAXS, WAXS and XDR measurements requires theoretical support, currently not available.

 

The theoretical study of gas hydrates using modeling tools has been implemented for several years now. In addition to classical structural information, the use of molecular dynamics (MD) simulations provides access to important properties that can be linked to measurable macroscopic data. This is the case, for example, for viscosity, thermal conductivity and diffusion of species [10]. However, the question arises here of the level of theory used since most of the simulations carried out are based on force field potentials [11]. The latter make it possible to treat large systems, several thousand molecules, but at the cost of a drastic simplification of the nature of the interactions and limited transferability. This greatly limits their application to the description of interfaces and the study of the impact of chemical impurities on these interfaces. Higher theoretical level simulations have also been carried out at the DFT level [12], but the high cost of this approach also limits its scope of application. In recent years, the LCPQ has developed intermediate potentials through the self-consistent-charge density-functional based tight binding (SCC-DFTB) approach. This approach, which makes it possible to treat systems of intermediate size between DFT and force field, is also more transferable than the latter. We were recently able to develop an optimized SCC-DFTB potential for the description of water [13] which, as part of N. Cinq’s thesis (2019-2023, ANR funding), was applied to CO2 and CO2 hydrates. N2. In particular, we were able to observe the destructuring of the CO2 hydrate around 280-300K for a simulation time of 50 picoseconds. However, despite these important advances, the compromise between the size of the accessible systems and the accuracy of the potential is not yet satisfactory to address the question of interfaces and their behavior, which requires being able to considerably increase the size of the systems and the simulation times.

 

2. Scientific objectives:

 

MUMUCO aims to realistically model the behavior of gas hydrates at the molecular scale. To do this, it will be necessary to develop original modeling tools to elucidate the phenomena which (1) govern the formation, dissociation and stability of gas hydrates, (2) control their development, (3) have an impact on process efficiency in energy technologies.

 

At the molecular level, modeling gas hydrates is complex. It requires the development of accurate potentials, of ab-initio quality, making it possible to describe buildings of several thousand molecules to correctly model interfaces, phase change phenomena, crystallization and dissociation. To achieve this goal, we will develop potentials based on Deep Neural Network Potential (DNNP) which we will construct from data produced from high-quality quantum calculations. Such potentials can reach ab initio quality level and are very computationally efficient on CPUs, and even more so on GPUs [14]. To develop and validate DNNPs, we will take advantage of the large quantity of data generated at the SCC-DFTB level for liquid water and gas hydrates obtained as part of N. Cinq’s thesis defended at the Unviersité de Toulouse on June 16, 2023. Thousands of structures, under different temperature and pressure conditions, as well as liquid water/ice and liquid water/gas hydrate interfaces can thus be generated in a relatively short time thanks to previous work carried out at the LCPQ. Once re-calculated at the DFT level with a high quality functional (revPBE0-D3, SCAN, SCAN0 or other), the structures, energies, forces and stress tensors thus obtained will constitute the initial training set of DNNP potentials. The first challenge of the MUMUCO project will be to implement a systematic methodology to generate accurate DNNPs, be able to evaluate their accuracy and improve them. Indeed, the exact definition of the data set, the structural descriptors, the definition of the DNNP are all parameters which will have to be defined and tested. To do this, we will take advantage of the DeePMD kit [15], specifically developed to construct DNNP potentials for MD applications and which we have already used at LCPQ as part of two Master 2 internships.

 

Once the DNNPs are developed, we will implement MD/DNNP simulations using the i-PI code [16], interfaced with the DeePMD potentials. Simulations will be carried out in the NVT and NPT ensembles, on systems of several thousand molecules, with a simulation time of several hundreds of picoseconds, or even a few nanoseconds if porting to GPU systems is possible. The first step will be to conduct simulations on liquid water and on crystalline ice using a single DNNP. Those systems are well described in the literature and will serve as a test case. If successful, this methodology will be extended to the description of gas hydrates of CO2 and cyclopentane by including into the DNNP the interaction of water with gas molecules. This step is not straightforward as it can necessitate the automatic generation of structures not visited during a rough MD simulation. This is why a systematic methodology, as mentioned above, needs to be developed and be as efficient as possible to avoid unnecessary expensive DFT calculations to be done. Furthermore, the refinement of SCC-DFTB data at the DFT level can lead to discontinuities in the DNNP. Again, a systematic methodology to complete and improve the data set will allow us to correct such effects. Once the DNNP is properly implemented, the stability of water/cyclopentane hydrate/CO2 interfaces under different pressure and temperature conditions will be studied using MD/DNNP. Post-processing of these MD simulations will allow us to access radial distribution functions, angle distributions, crystallinity descriptors, diffusion coefficients, dissociation statistics, energy data and to provide phase diagrams never before obtained at this level of theory. It will also be possible for us to take into account the nuclear quantum effects in such simulations. These effects can be important on aqueous systems but are extremely expensive to take into account from a theoretical point of view [17]. Due to the low computational cost of DNNP, we will include them and therefore achieve a simulation quality never achieved so far for systems that aim to be as realistic as possible.

 

Finally, the addition of gas hydrate inhibitors/promoters, initially ethylene glycol, tetramethylammonium, tetrahydrofuran and acetone (chosen both for their chemical simplicity and their significant impact) will be considered. This asks the key question of the chemical complexity that can be reached with DNNPs. If an accurate enough DNNP can be developed, MD/DNNP simulations in combination with accelerated sampling tools will be implemented to study the influence of temperature, pressure and the presence of inhibitors or promoters on the structure of gas hydrates during their formation/dissociation.

 

In the longer term, our work will lead to structure/dynamic/property relationships for gas hydrates that are unique to date. The extrapolation of these results to the macroscopic scale will be considered following the MUMUCO project, most probably as part of an ANR funding request, to study the influence of the overall composition of the systems and the presence of selected additives on flow behaviors, dynamics of phase changes and heat and mass transfers in gas hydrates. 

[1] P. Clain, F. T. Ndoye, A. Delahaye, L. Fournaison, W. Lin, D. Dalmazzone, Int. J. Refrig. 2015, 50, 19-31. P. Clain, A. Delahaye, L. Fournaison, N. Mayoufi, D. Dalmazzone, W. Fürst. Chem. Eng. J. 2012, 193-194, 112-122.

 

[2] S. Jerbi, A. Delahaye, J. Oignet, L. Fournaison, P. Haberschill. Int. J. Refrig. 2013, 36 (4), 1294-1301. J. Oignet, A. Delahaye, J.-P Torré, C. Dicharry, H. M Hoang, P. Clain, V. Osswald, et al. Chem. Eng. Sci. 2017, 158, 294-303.

 

[3] J. Oignet, H. M.Hoang, V. Osswald, A. Delahaye, L. Fournaison, P. Haberschill, Appl. Therm. Eng. 2017, 118, 630-637.

 

[4] S. Jerbi, A. Delahaye, L. Fournaison, P. Haberschill, Int. J. Refrig. 2010, 33 (8), 1625-1631. H. Hashemi, S. Babaee, A. H. Mohammadi, P. Naidoo, D. Ramjugernath, J. Chem. Thermodyn. 2015, 82. 47-52. T. Dufour, H. M.Hoang, J. Oignet, V. Osswald, P. Clain, L. Fournaison, A. Delahaye, Appl. Energy 2017, 204, 641-652.

 

[5] L. E. Dieker, ZM. Aman, NC. George, AK. Sum, ED. Sloan, CA. Koh, Energy Fuels 2009, 23 (12), 5966–5971.

 

[6] J. S. Parent, P. R. Bishnoi, Chem. Eng. Comm. 1996, 144 (1) 51-64.

 

[7] S. Takeya, A. Hori, T. Hondoh, T. Uchida, J. Phys. Chem. B 2000, 104, 4164. H. Sefidroodi, EA. Malcolm, A. Kelland, Chem. Eng. Sci. 2013, 87, 133– 140. S. Gao, W. Chapman, W. House, Proc. of the 5th Int. Conf. on Gas Hydrates 2005, Tapir Ac. Press, Trondheim, Norway.

 

[8] S. Teychené, B. Biscans, Chem. Eng. Sci. 2012, 77, 242–248. S. Teychené, B. Biscans, Cryst. Growth Des. 2011, 11 (11), 4810–4818. Y. Vitry, S. Teychené, S. Charton, F. Lamadie, B. Biscans. Chem. Eng. Sci. 2015, 133, 54–61. S. Teychené, D. Radajewski, NV. Pham, I. Rodriguez-Ruiz, P. Pernot, M. Brennich, et al. Acta Cryst. 2017, A73, C611.

 

[9] N. Pham, D. Radajewski, A. Round, M. Brennich, P. Pernot, B. Biscans, et al. Anal. Chem. 2017, 89 (4), 2282–2287.

 

[10] A. Guerra, S. Mathews, M. Maric, P. Servio,A. D. Rey, Molecules 2022, 27, 5019.

 

[11] H. Docherty, A. Galindo, C. vega, E Sanz, J. Chem. Phys. 2006, 125, 074510.

 

[12] C. Pétuya, L. Martin-Gondre, P. Aurel, F. Damais, A. Desmedt, J. Chem. Phys. 2019, 150, 184705.

 

[13] N. Cinq, A. Simon, F. Louisnard, J.Cuny, J. Phys. Chem. B 2023, 127, 35, 7590–7601.

 

[14] L. Zhang, J. Han, H. Wang, R. Car, E. Weinan, Phys. Rev. Lett. 2018, 120, 143001.

 

[15] H. Wang, L. Zhang, J. Han and Weinan E. Weinan, Comput. Phys. Comm. 2018, 228, 178-184.

 

[16] V. Kapoil et al., Comput. Phys. Comm. 2019, 236, 214-223.

 

[17] M. Ceriotti et al.,Chem. Rev. 2016, 116, 13, 7529–7550.