Hostname: page-component-cd9895bd7-8ctnn Total loading time: 0 Render date: 2024-12-27T09:33:21.114Z Has data issue: false hasContentIssue false

Transport scaling in porous media convection

Published online by Cambridge University Press:  20 August 2024

Xiaojue Zhu*
Affiliation:
Max Planck Institute for Solar System Research, 37077 Göttingen, Germany
Yifeng Fu
Affiliation:
Max Planck Institute for Solar System Research, 37077 Göttingen, Germany
Marco De Paoli*
Affiliation:
Physics of Fluids Group and Max Planck Center for Complex Fluid Dynamics and J. M. Burgers Centre for Fluid Dynamics, University of Twente, P.O. Box 217, 7500AE Enschede, The Netherlands Institute of Fluid Mechanics and Heat Transfer, TU Wien, 1060 Vienna, Austria
*
Email addresses for correspondence: [email protected], [email protected]
Email addresses for correspondence: [email protected], [email protected]

Abstract

We present a theory to describe the Nusselt number, $\operatorname {\mathit {Nu}}$, corresponding to the heat or mass flux, as a function of the Rayleigh–Darcy number, $\operatorname {\mathit {Ra}}$, the ratio of buoyant driving force over diffusive dissipation, in convective porous media flows. First, we derive exact relationships within the system for the kinetic energy and the thermal dissipation rate. Second, by segregating the thermal dissipation rate into contributions from the boundary layer and the bulk, which is inspired by the ideas of the Grossmann and Lohse theory (J. Fluid Mech., vol. 407, 2000; Phys. Rev. Lett., vol. 86, 2001), we derive the scaling relation for $\operatorname {\mathit {Nu}}$ as a function of $\operatorname {\mathit {Ra}}$ and provide a robust theoretical explanation for the empirical relations proposed in previous studies. Specifically, by incorporating the length scale of the flow structure into the theory, we demonstrate why heat or mass transport differs between two-dimensional and three-dimensional porous media convection. Our model is in excellent agreement with the data obtained from numerical simulations, affirming its validity and predictive capabilities.

Type
JFM Papers
Creative Commons
Creative Common License - CCCreative Common License - BY
This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (http://creativecommons.org/licenses/by/4.0), which permits unrestricted re-use, distribution and reproduction, provided the original article is properly cited.
Copyright
© The Author(s), 2024. Published by Cambridge University Press.

1. Introduction

Carbon dioxide (${\rm CO}_2$) sequestration is a process aimed at long-term storage of large volumes of ${\rm CO}_2$ (Schrag Reference Schrag2007), primarily to mitigate climate change and support energy transition. One of the most promising sequestration strategies involves natural underground reservoirs. In this case, liquid ${\rm CO}_2$ is injected in saline aquifers, geological porous formations located hundreds of metres beneath the Earth's surface and confined by horizontal impermeable layers (Huppert & Neufeld Reference Huppert and Neufeld2014; De Paoli Reference De Paoli2021). Saline aquifers are filled with brine, highly salted water denser than ${\rm CO}_2$. Because of this density difference, the injected volume of ${\rm CO}_2$ will sit on top of the brine, and a critical configuration takes place: in case of fractures in the top confining layer of the formation, ${\rm CO}_2$ would migrate upwards and eventually reach the upper strata up to the atmosphere (Hidalgo et al. Reference Hidalgo, Dentz, Cabeza and Carrera2015). However, ${\rm CO}_2$ is partially soluble in brine and the resulting mixture, which is heavier than both starting fluids, sinks downwards through the porous rocks and safely traps the ${\rm CO}_2$ (Emami-Meybodi et al. Reference Emami-Meybodi, Hassanzadeh, Green and Ennis-King2015; Letelier et al. Reference Letelier, Ulloa, Leyrer and Ortega2023). To determine the optimal ${\rm CO}_2$ injection rate and predict the long-time behaviour of the injected ${\rm CO}_2$, it is therefore imperative to conduct a meticulous assessment of the flow dynamics and the associated mixing laws (MacMinn et al. Reference MacMinn, Neufeld, Hesse and Huppert2012; Guo et al. Reference Guo, Sun, Zhao, Li, Liu and Chen2021). An idealized representation of this complex system consists of a porous domain fully saturated with fluid and confined between a heated bottom plate and a cooled top plate (Hewitt, Neufeld & Lister Reference Hewitt, Neufeld and Lister2012; Wen, Chang & Hesse Reference Wen, Chang and Hesse2018). The top-to-bottom temperature difference induces a density gradient that drives the flow. This configuration, which we label here as Rayleigh–Darcy (RD) convection, serves as a fundamental model of the aforementioned process, where the ${\rm CO}_2$ concentration field, responsible for the density increase in the case of geological carbon sequestration, is replaced by a temperature field.

Indeed, thermal and solutal convective porous media flows can be considered equivalent and controlled by the same governing equations provided that: (i) in the temperature-driven flow, the solid phase is locally in thermal equilibrium with the liquid phase, and (ii) in the corresponding concentration-driven system, the solid is impermeable to the solute. Additional factors to be accounted for a proper comparison between these systems are the dependency of the viscosity and the fluid density with respect to the value of the scalar. While viscosity is generally weakly sensitive to temperature variations, it may be considerably affected by the local value of solute concentration. However, it has been previously shown by Hidalgo et al. (Reference Hidalgo, Fe, Cueto-Felgueroso and Juanes2012) that, in convective porous media flows, concentration-induced viscosity variations do not significantly affect the global transport properties of the system. In contrast, the shape of the density-concentration (or density-temperature) curves is shown to be crucial. For a general introduction to the RD convection, we refer to the reviews by Hewitt (Reference Hewitt2020) and De Paoli (Reference De Paoli2023). In this work, we will refer to fluid characterized by a constant viscosity and a linear dependency of density with the transported scalar (temperature).

The single control parameter in RD convection is the Rayleigh–Darcy number $\operatorname {\mathit {Ra}}$, which indicates the relative strength between driving forces (convection) and dissipative effects (diffusion and viscosity), while the major response parameter of the system is the Nusselt number $\operatorname {\mathit {Nu}}$, a dimensionless measure of the amount of heat (or solute) exchanged. Similar to the Rayleigh–Bénard (RB) convection (i.e. a fluid heated from below and cooled from the top, in the absence of any porous medium), in recent years, major efforts have been put into understanding the scaling relation between $\operatorname {\mathit {Nu}}$ and $\operatorname {\mathit {Ra}}$, where $\operatorname {\mathit {Ra}}$ is intended as the thermal Rayleigh number (Ahlers, Grossmann & Lohse Reference Ahlers, Grossmann and Lohse2009). The classical theory (Malkus Reference Malkus1954; Priestley Reference Priestley1954; Howard Reference Howard1966) posits that at significantly high $\operatorname {\mathit {Ra}}$, the buoyancy flux should be independent of the layer's height ($L$). In the high-$\operatorname {\mathit {Ra}}$ regime within a porous medium, this argument predicts a linear scaling of $\operatorname {\mathit {Nu}}\sim \operatorname {\mathit {Ra}}$. It has also been rigorously demonstrated that the linear scaling serves as an upper bound (Doering & Constantin Reference Doering and Constantin1998; Otero et al. Reference Otero, Dontcheva, Johnston, Worthing, Kurganov, Petrova and Doering2004; Wen et al. Reference Wen, Dianati, Lunasin, Chini and Doering2012; Hassanzadeh, Chini & Doering Reference Hassanzadeh, Chini and Doering2014). Interestingly, such scaling also means that the dimensional flux is independent of thermal diffusivity and, as a result, a realization of the scaling indicates the system reaches the so-called ultimate regime (Hewitt et al. Reference Hewitt, Neufeld and Lister2012; Pirozzoli et al. Reference Pirozzoli, De Paoli, Zonta and Soldati2021). In comparison, in RB convection, a similar argument leads to $\operatorname {\mathit {Nu}}\sim \operatorname {\mathit {Ra}}^{1/3}$ (Malkus Reference Malkus1954; Priestley Reference Priestley1954; Howard Reference Howard1966), different from the diffusion-free ultimate scaling $\operatorname {\mathit {Nu}}\sim \operatorname {\mathit {Ra}}^{1/2}$ proposed by Kraichnan (Reference Kraichnan1962) and Spiegel (Reference Spiegel1963). A detailed introduction to the scalings in RB is provided by Ahlers et al. (Reference Ahlers, Grossmann and Lohse2009), Chillà & Schumacher (Reference Chillà and Schumacher2012) and Xia et al. (Reference Xia, Huang, Xie and Zhang2023).

Direct numerical simulations (DNS) have been conducted in both two and three dimensions for RD convection to investigate heat transfer scaling at finite $\operatorname {\mathit {Ra}}$. The two-dimensional (2-D) DNS at high $\operatorname {\mathit {Ra}}$ ($10^3 \leqslant \operatorname {\mathit {Ra}} \leqslant 10^4$) by Otero et al. (Reference Otero, Dontcheva, Johnston, Worthing, Kurganov, Petrova and Doering2004) suggested a slightly sublinear $\operatorname {\mathit {Nu}}(\operatorname {\mathit {Ra}})$ scaling. Subsequent DNS, as reported by Hewitt et al. (Reference Hewitt, Neufeld and Lister2012), extended up to $\operatorname {\mathit {Ra}}=4\times 10^4$, and indicated that the scaling $\operatorname {\mathit {Nu}}\sim \operatorname {\mathit {Ra}}$ is asymptotically attained, albeit with a correction to the linear scaling. A simple fit, $\operatorname {\mathit {Nu}}=0.0069\operatorname {\mathit {Ra}}+2.75$, was proposed to accommodate the data within this range. For three-dimensional (3-D) RD convection, DNS conducted by Pirozzoli et al. (Reference Pirozzoli, De Paoli, Zonta and Soldati2021) and De Paoli et al. (Reference De Paoli, Pirozzoli, Zonta and Soldati2022) reached up to $\operatorname {\mathit {Ra}}=8\times 10^4$ and suggested that the appropriate scaling for $\operatorname {\mathit {Nu}}$ is given by $\operatorname {\mathit {Nu}}=0.0081\operatorname {\mathit {Ra}}+ 0.067\operatorname {\mathit {Ra}}^{0.61}$. This contrasts with an alternative fit proposed by Hewitt, Neufeld & Lister (Reference Hewitt, Neufeld and Lister2014), where $\operatorname {\mathit {Nu}}=0.0096\operatorname {\mathit {Ra}}+ 4.6$ was considered for the data within the range up to $\operatorname {\mathit {Ra}}=2\times 10^4$.

It is crucial to underline that the previously mentioned corrections to the linear scaling are purely empirical in nature. This leads to the fundamental question: can we provide an explanation for these corrections and quantify them? To address this, we turn our attention to the Grossmann–Lohse (GL) theory (Grossmann & Lohse Reference Grossmann and Lohse2000, Reference Grossmann and Lohse2001), a key tool for comprehending the effective scaling of Nusselt and Reynolds numbers in relation to $\operatorname {\mathit {Ra}}$ in turbulent RB convection. The central premise of the GL theory can be summarized as follows. First, it establishes a connection between $\operatorname {\mathit {Nu}}$ and $\operatorname {\mathit {Ra}}$ by considering their relationship with the kinetic energy dissipation rate and thermal dissipation rate through exact relations. Second, the theory dissects these dissipation rates into contributions from the boundary layer (BL) and the bulk flow. In this work, we will derive corresponding exact relations for RD convection. By applying the principles of the GL theory, we can deduce the BL and bulk contributions to the thermal dissipation rate, shedding light on the origins and expressions of the corrections to the linear scaling.

2. Governing equations

We consider a fluid-saturated porous domain heated from below and cooled from above, as sketched in figure 1(a i). Although we discuss here the problem of a thermally driven flow in a porous medium that is locally in thermal equilibrium with the fluid, the same conclusions apply when the scalar is a solute, provided that the governing parameter ($\operatorname {\mathit {Ra}}$) is matched. The size of the domain considered in $W$ is the wall-parallel directions $x,y$ and $L$ in wall-normal direction $z$, along which gravity ($\boldsymbol {g}$) acts. The flow is visualized in terms of dimensionless temperature $T$. For sufficiently high $\operatorname {\mathit {Ra}}$, a columnar flow structure develops both in 2-D and in 3-D flows, as one can observe from the cross-section relative to the domain midheight in figure 1(a ii), while the near-wall region (figure 1a iii) is populated by thin filamentary plumes. This structure differs significantly from the classical RB turbulence, reported in figure 1(b), which is controlled in the bulk by large-scale rolls. In RB convection, large-scale structures span the entire domain, with typical length scales comparable to the height of the system. In comparison, in RD convection, columnar-like structures prevail, and are characterized by length scales distinct from the system's height. This difference will be taken into account when we build up a theory for RD convection in the following sections.

Figure 1. Instantaneous dimensionless temperature field $T$ for convection in (a) porous media for $\operatorname {\mathit {Ra}}=10^4$, $W/L=1$ (Pirozzoli et al. Reference Pirozzoli, De Paoli, Zonta and Soldati2021), labelled as RD convection, and (b) in classical RB turbulence (thermal Rayleigh number $10^9$, Prandtl number $1$, $W/L=1$). The temperature distribution is shown over the entire volume (a i, b i), at the centreline $z=1/2$ (a ii, b ii) and near the upper wall (a iii, b iii).

Before presenting the dimensionless equations, we introduce the scaling quantities employed. The equations are made dimensionless with respect to convective flow scales (Pirozzoli et al. Reference Pirozzoli, De Paoli, Zonta and Soldati2021), namely, velocities are scaled with $\mathcal {U}=\alpha g\Delta K/\nu$, where $\alpha$ is the thermal expansion rate, $\varDelta$ the temperature difference between the bottom and the top plate, $g$ the acceleration due to gravity, $\nu$ the kinematic viscosity and $K$ the permeability of the porous medium, which we assume to be homogeneous and isotropic. Lengths are scaled with $L$ and time with $\phi L/\mathcal {U}$. The dimensionless temperature is obtained as $T=(T^*-T^*_{top})/\varDelta$, with $T^*$ and $T^*_{top}$ the dimensional temperature field and the temperature value at the top boundary, respectively. Finally, pressure is scaled by $g L (\rho ^*_{top}-\rho ^*_{bot})$, where the top-to-bottom fluid density difference ($\rho ^*_{top}-\rho ^*_{bot}$) is used. In an incompressible RD system, the heat transport is controlled by the dimensionless advection–diffusion equation (Pirozzoli et al. Reference Pirozzoli, De Paoli, Zonta and Soldati2021),

(2.1)\begin{equation} \frac{\partial T}{\partial t}+\boldsymbol{u}\boldsymbol{\cdot}\boldsymbol{\nabla} T=\frac{1}{\operatorname{\mathit{Ra}}}\nabla^2T { ,} \end{equation}

where $\boldsymbol {u}$ and $T$ are the velocity and temperature fields, respectively, $t$ is time and $\operatorname {\mathit {Ra}}$ is the Rayleigh–Darcy number defined as

(2.2)\begin{equation} \operatorname{\mathit{Ra}}=\frac{\alpha g\Delta KL}{\kappa\nu}. \end{equation}

In this parameter, the medium ($K$), domain ($g,L$) and fluid ($\alpha,\varDelta,\nu,\kappa$) properties are included, where $\kappa$ is the thermal diffusivity. The momentum transport and the flow incompressibility are accounted by the Darcy law and continuity equations,

(2.3)$$\begin{gather} \boldsymbol{u}={-}(\boldsymbol{\nabla} p-T\boldsymbol{k}), \end{gather}$$
(2.4)$$\begin{gather}\boldsymbol{\nabla}\boldsymbol{\cdot}\boldsymbol{u}=0, \end{gather}$$

respectively, where $p$ is the reduced pressure field and $\boldsymbol {k}$ is the unit vector aligned with $z$.

At the horizontal boundaries, we consider the temperature constant and equal to $T=1$ at the bottom plate and $T=0$ at the top, so that an unstable configuration is achieved and the flow is driven by convection. No-penetration boundary conditions are assumed at both plates for the velocity, while the sides are considered periodic. Equations (2.1), (2.3) and (2.4) together with these boundary conditions determine the flow dynamics, which are controlled by two dimensionless parameters, namely $\operatorname {\mathit {Ra}}$ and the horizontal domain width $W/L$. The latter does not appear explicitly in the equations, but may play a significant role in determining the flow structure, especially at low $\operatorname {\mathit {Ra}}$.

3. Nusselt number and exact conservation equations

First, we will correlate the thermal (Nusselt number) and the kinetic (Péclet number) response parameters to the control parameter (Rayleigh–Darcy number), and then the thermal dissipation will be linked to the Nusselt number.

The temporal and horizontal average of (2.1) can be written as

(3.1)\begin{equation} \frac{\partial}{\partial z} \left ( \operatorname{\mathit{Ra}} \overline{\left \langle u_{z} T \right \rangle}_{A} - \overline{\left \langle \frac{\partial T}{\partial z} \right \rangle}_{A} \right ) = 0. \end{equation}

The Nusselt number then reads (Letelier, Mujica & Ortega Reference Letelier, Mujica and Ortega2019; Ulloa & Letelier Reference Ulloa and Letelier2022)

(3.2)\begin{equation} \operatorname{\mathit{Nu}}=\operatorname{\mathit{Ra}} \overline{\left\langle u_zT\right\rangle}_A-\overline{\left\langle \frac{\partial T}{\partial z} \right\rangle}_A. \end{equation}

Here the following notations are used for different averaging procedures. Overbars $\overline {\cdots }$ correspond to the time average of a dimensionless value $f$, while an average over the horizontal surface and an average over the whole volume domain are denoted by $\langle \cdots \rangle _A$ and $\langle \cdots \rangle$,

(3.3)$$\begin{gather} \bar{f} = \frac{1}{\tau} \int_{t_0}^{t_0+\tau} f \,\mathrm{d} t, \end{gather}$$
(3.4)$$\begin{gather}\left \langle\, f \right \rangle_{A} = \frac{1}{A} \int_{0}^{W/L} \int_{0}^{W/L} f \,\mathrm{d}\kern 0.06em x \,\mathrm{d}y, \end{gather}$$
(3.5)$$\begin{gather}\left \langle\, f \right \rangle = \frac{1}{V} \int_{0}^{W/L} \int_{0}^{W/L} \int_{0}^{1} f \,\mathrm{d}\kern 0.06em x \,\mathrm{d}y \,\mathrm{d}z, \end{gather}$$

respectively, where $A=(W/L)^2$ is the dimensionless horizontal surface area and $V=(W/L)^2$ is the dimensionless volume of the whole domain based on our characteristic length scale $L$. Two exact relations exist in our system and can be derived from the governing equations. Using the dimensionless velocity $\boldsymbol {u}$ to dot product both sides of (2.3) and combining with the incompressible continuity equation (2.4), we get

(3.6)\begin{equation} \left | \boldsymbol{u} \right |^2 ={-}\boldsymbol{\nabla}\boldsymbol{\cdot} (p\boldsymbol{u}) + Tu_z. \end{equation}

The volume and time average of (3.6) reads

(3.7)\begin{equation} \overline{\left\langle \left | \boldsymbol{u} \right |^2 \right\rangle} ={-}\frac{1}{V} \iint_{\varSigma} (p\boldsymbol{u}) \boldsymbol{\cdot} \hat{\boldsymbol{n}} \,\mathrm{d}S + \overline{\left\langle T u_z \right\rangle}, \end{equation}

where $\varSigma$ is the boundary surface of the domain, $\mathrm {d}S$ denotes the surface element on the boundary and $\hat {\boldsymbol {n}}$ is the normal unit vector for the surface elements. The mean power given by the pressure gradient vanishes due to the non-penetration boundary condition,

(3.8)\begin{equation} \frac{1}{V} \iint_{\varSigma} (p\boldsymbol{u}) \boldsymbol{\cdot} \hat{\boldsymbol{n}} \,\mathrm{d}S ={-}\frac{1}{V} \iint_{\varSigma(z=0)} p u_z \,\mathrm{d}S + \frac{1}{V} \iint_{\varSigma(z=1)} p u_z \,\mathrm{d}S = 0. \end{equation}

The mean buoyancy power in (3.7) can be written as

(3.9)\begin{equation} \overline{\left\langle T u_z \right\rangle} = \frac{1}{V} \int_{0}^{1} \overline{\left \langle Tu_z \right \rangle}_A A \,\mathrm{d}z = \frac{1}{\operatorname{\mathit{Ra}}} \int_{0}^{1} \left (\operatorname{\mathit{Nu}} + \overline{\left\langle \frac{\partial T}{\partial z} \right\rangle}_A \right ) \mathrm{d}z. \end{equation}

Here, the last equivalence comes from the $\operatorname {\mathit {Nu}}$ definition (3.2). Since we use $L$ as our length scale, $z\in [0,1]$. The last term in the above equation reads

(3.10)\begin{equation} \frac{1}{\operatorname{\mathit{Ra}}}\int_{0}^{1} \overline{\left\langle \frac{\partial T}{\partial z} \right\rangle}_A \,\mathrm{d}z = \frac{1}{\operatorname{\mathit{Ra}}}\int_{0}^{1} \frac{\partial \overline{\left\langle T \right\rangle}_A}{\partial z} \,\mathrm{d}z = \frac{1}{\operatorname{\mathit{Ra}}}\left ( \overline{\left\langle T \right\rangle}_A |_{z=1} - \overline{\left\langle T \right\rangle}_A |_{z=0} \right) ={-}\frac{1}{\operatorname{\mathit{Ra}}}. \end{equation}

Reintroducing (3.10) back into (3.9), after some algebraic manipulations we get

(3.11)\begin{equation} \overline{\left\langle T u_z \right\rangle} = \frac{1}{\operatorname{\mathit{Ra}}} \int_{0}^{1} \left (\operatorname{\mathit{Nu}} + \overline{\left\langle \frac{\partial T}{\partial z} \right\rangle}_A \right ) \mathrm{d}z = \frac{1}{\operatorname{\mathit{Ra}}} (\operatorname{\mathit{Nu}} - 1). \end{equation}

Combining (3.7), (3.8) and (3.11), we obtain an expression for the mean dimensionless velocity square,

(3.12)\begin{equation} \overline{\left\langle \left | \boldsymbol{u} \right |^2 \right\rangle} = \frac{1}{\operatorname{\mathit{Ra}}} (\operatorname{\mathit{Nu}} - 1). \end{equation}

We introduce the velocity scale

(3.13)\begin{equation} \mathcal{V} = \mathcal{U}\sqrt{\overline{\left \langle\left | \boldsymbol{u} \right |^2 \right\rangle}}, \end{equation}

with $\mathcal {U}=\alpha g\Delta K/\nu$, and one finally obtains an exact relation

(3.14)\begin{equation} \operatorname{\mathit{Pe}}^2=(\operatorname{\mathit{Nu}}-1)\operatorname{\mathit{Ra}} \end{equation}

with

(3.15)\begin{equation} \operatorname{\mathit{Pe}} = \frac{\mathcal{V} L}{ \kappa} = \operatorname{\mathit{Ra}}\frac{\mathcal{V}}{\mathcal{U}}, \end{equation}

where $\operatorname {\mathit {Pe}}$ is the Péclet number. The relation (3.14) aligns with the findings reported by Hassanzadeh et al. (Reference Hassanzadeh, Chini and Doering2014), albeit derived from a slightly different set of equations for porous media convection. Note that for RB convection, the analogous exact relation is $\epsilon _u=\nu ^3/L^4(\operatorname {\mathit {Nu}}-1)\operatorname {\mathit {Ra}}\operatorname {\mathit {Pr}}^{-2}$, where $\epsilon _u$ is the kinetic energy dissipation rate and $Pr$ is the Prandtl number (Ahlers et al. Reference Ahlers, Grossmann and Lohse2009). To assess the validity of (3.14), we consider the numerical measurements available in the literature. For 2-D flows, $\operatorname {\mathit {Pe}}$ is measured by De Paoli et al. (Reference De Paoli, Yerragolam, Lohse and Verzicco2024) using (3.13) and (3.15). The velocity root mean square (r.m.s.) at the centreline (2-D and 3-D) is computed by Hewitt et al. (Reference Hewitt, Neufeld and Lister2012, Reference Hewitt, Neufeld and Lister2014) and reported in figure 2(a).

Figure 2. (a) The r.m.s. of velocity used to estimate the Péclet number, $\operatorname {\mathit {Pe}}$. Horizontal component (${\rm r.m.s.}(u)$), vertical component (${\rm r.m.s.}(w)$) and total (defined as $\sqrt {[{\rm r.m.s.}(u)]^2+[{\rm r.m.s.}(w)]^2}$ for 2-D cases, and $\sqrt {2[{\rm r.m.s.}(u)]^2+[{\rm r.m.s.}(w)]^2}$ for 3-D cases) are reported; data from Hewitt et al. (Reference Hewitt, Neufeld and Lister2012, Reference Hewitt, Neufeld and Lister2014). (b) Comparison of $\operatorname {\mathit {Pe}}$ compensated with $(\operatorname {\mathit {Nu}}-1)\operatorname {\mathit {Ra}}$ as from (3.14). Here $\operatorname {\mathit {Pe}}$ is computed with $\mathcal {V}$ defined as in (3.17) for the data from Hewitt et al. (Reference Hewitt, Neufeld and Lister2012, Reference Hewitt, Neufeld and Lister2014), and as in (3.13) for data from De Paoli et al. (Reference De Paoli, Yerragolam, Lohse and Verzicco2024). (c) Ratio of the Nusselt number to the dimensionless thermal dissipation, derived in (3.25); data from Pirozzoli et al. (Reference Pirozzoli, De Paoli, Zonta and Soldati2021) and De Paoli et al. (Reference De Paoli, Yerragolam, Lohse and Verzicco2024).

Since in all directions no mean flow exists, we have that $\mathcal {V}$ is obtained from the r.m.s. of the velocity components ($u_i$), namely

(3.16)\begin{equation} \mathcal{V} = \mathcal{U}\left(\overline{\left \langle[\text{r.m.s.}(u_x)]^2+[\text{r.m.s.}(u_y)]^2+[\text{r.m.s.}(u_z)]^2\right\rangle}\right)^{1/2}.\end{equation}

Assuming that the centreline flow is representative of the kinetic energy of the system, we have that

(3.17)\begin{equation} \mathcal{V} \approx\mathcal{U}\left(\overline{\left \langle[\text{r.m.s.}(u_x)]^2+[\text{r.m.s.}(u_y)]^2+[\text{r.m.s.}(u_z)]^2\right \rangle_{\varSigma(z=1/2)}}\right)^{1/2}. \end{equation}

We use this approximation to compute $\mathcal {V}$ and verify the validity of (3.14) for the data of Hewitt et al. (Reference Hewitt, Neufeld and Lister2012, Reference Hewitt, Neufeld and Lister2014). We finally observe in figure 2(b) that (3.14) (dashed line) is in excellent agreement with the measurements obtained from the exact definition of $\operatorname {\mathit {Pe}}$ (2-D and $\mathcal {V}$ computed with (3.13) from De Paoli et al. (Reference De Paoli, Yerragolam, Lohse and Verzicco2024)) and also with measurements obtained from the approximated definition of $\operatorname {\mathit {Pe}}$ (2-D and 3-D, $\mathcal {V}$ computed with (3.17) from Hewitt et al. (Reference Hewitt, Neufeld and Lister2012, Reference Hewitt, Neufeld and Lister2014)).

We will now derive an equation to correlate the mean thermal dissipation to the Nusselt number. Multiplying the dimensionless thermal advection–diffusion equation (2.1) by $T$, integrating over the whole domain and time-averaging, we have

(3.18)\begin{equation} \overline{\left \langle \frac{\partial}{\partial t} \left ( \frac{T^2}{2} \right ) \right \rangle} ={-} \overline{\left \langle T \boldsymbol{u} \boldsymbol{\cdot} \boldsymbol{\nabla} T \right\rangle} + \frac{1}{\operatorname{\mathit{Ra}}} \overline{\left \langle T \nabla^2 T \right\rangle}. \end{equation}

Under the assumption of a statistically steady state,

(3.19)\begin{equation} \overline{\left \langle \frac{\partial}{\partial t} \left ( \frac{T^2}{2} \right ) \right \rangle} = \frac{\partial}{\partial t} \overline{\left \langle \frac{T^2}{2} \right \rangle} = 0. \end{equation}

The first term on the right-hand side of (3.18) can be further written as

(3.20)\begin{equation} -\overline{\left \langle T \boldsymbol{u} \boldsymbol{\cdot}\boldsymbol{\nabla} T \right \rangle} = \overline{\left \langle \frac{T^2}{2} \boldsymbol{\nabla}\boldsymbol{\cdot}\boldsymbol{u} \right \rangle} - \overline{\left \langle \boldsymbol{\nabla}\boldsymbol{\cdot}\left ( \boldsymbol{u}\frac{T^2}{2} \right) \right \rangle} ={-}\frac{1}{V} \iint_{\varSigma} \left( \boldsymbol{u}\frac{T^2}{2} \right) \boldsymbol{\cdot} \hat{\boldsymbol{n}} \,\mathrm{d}S = 0. \end{equation}

Here we employed again continuity (2.4) and the no-penetration boundary condition. The second term on the right-hand side of (3.18) reads

(3.21)\begin{equation} \frac{1}{\operatorname{\mathit{Ra}}} \overline{\left \langle T \nabla^2 T \right\rangle} = \frac{1}{\operatorname{\mathit{Ra}}} \left[ \overline{\left \langle \nabla^2 \left( \frac{T^2}{2} \right) \right \rangle} - \overline{\left \langle \left\lvert \boldsymbol{\nabla} T \right\lvert^2 \right \rangle} \right]. \end{equation}

Combining results from (3.18)–(3.21), one obtains

(3.22)\begin{equation} \overline{\left \langle \left\lvert \boldsymbol{\nabla} T \right\lvert^2 \right \rangle} = \overline{\left \langle \nabla^2 \left( \frac{T^2}{2} \right) \right \rangle}. \end{equation}

We can use the following procedure to further simplify the right-hand side of (3.22):

(3.23)\begin{align} \overline{\left \langle \nabla^2 \left( \frac{T^2}{2} \right) \right \rangle} &= \frac{1}{V} \iint_{\varSigma} \left( \overline{T \boldsymbol{\nabla} T} \right) \boldsymbol{\cdot} \hat{\boldsymbol{n}} \,\mathrm{d}S \end{align}
(3.24)\begin{align} &= \frac{A}{V} \left ( \overline{\left\langle T\frac{\partial T}{\partial z} \right\rangle}_{\varSigma(z=1)} - \overline{\left\langle T\frac{\partial T}{\partial z} \right\rangle}_{\varSigma(z=0)} \right ) = \operatorname{\mathit{Nu}}. \end{align}

Here we considered that $V=A=(W/L)^2$, applied the boundary conditions for $T$ and $u_z$, as well as the $\operatorname {\mathit {Nu}}$ definition (3.2). Combining (3.22) and (3.24), we get

(3.25)\begin{equation} \overline{\left \langle \left\lvert \boldsymbol{\nabla} T \right\lvert^2 \right \rangle} = \operatorname{\mathit{Nu}}, \end{equation}

where $\overline { \langle \lvert \boldsymbol {\nabla } T \lvert ^2 \rangle }$ represents the dimensionless mean thermal dissipation. This relation, which holds also for RB convection (Ahlers et al. Reference Ahlers, Grossmann and Lohse2009), has been also presented before for RD flows (Otero et al. Reference Otero, Dontcheva, Johnston, Worthing, Kurganov, Petrova and Doering2004; Hidalgo et al. Reference Hidalgo, Fe, Cueto-Felgueroso and Juanes2012; De Paoli Reference De Paoli2023) and for Hele-Shaw convection the limit of infinitely thin domains (Letelier et al. Reference Letelier, Mujica and Ortega2019; Ulloa & Letelier Reference Ulloa and Letelier2022). The ratio of the Nusselt number to the dimensionless thermal dissipation is compared in figure 2(c) for the numerical results of Pirozzoli et al. (Reference Pirozzoli, De Paoli, Zonta and Soldati2021) and De Paoli et al. (Reference De Paoli, Yerragolam, Lohse and Verzicco2024). We observe that, also in this case, the agreement between theory and simulations is good, confirming the validity of (3.25). Finally, one obtains the dimensional thermal dissipation rate,

(3.26)\begin{equation} \epsilon=\kappa\frac{\varDelta^2}{L^2} \overline{\left \langle \left\lvert \boldsymbol{\nabla} T \right\lvert^2 \right \rangle} = \kappa\frac{\varDelta^2}{L^2} \operatorname{\mathit{Nu}}. \end{equation}

Equations (3.14) and (3.26) represent the two exact relations we derived in our system.

4. Application of GL theory and scaling relation for the Nusselt number

With the two exact relations derived in the previous section, we can now apply the main ideas of the GL theory to RD convection. The key idea of the GL theory (Grossmann & Lohse Reference Grossmann and Lohse2000, Reference Grossmann and Lohse2001) is to split the kinetic and thermal dissipation rates into contributions from the BLs and bulk. In RD convection, the procedure becomes even simpler than in RB, as only the thermal dissipation rate appears in the exact relations. We separate the mean thermal dissipation as

(4.1)\begin{equation} \epsilon=\epsilon_{BL}+\epsilon_{bulk}, \end{equation}

and apply the respective scaling relations for $\epsilon _{BL}$ and $\epsilon _{bulk}$, based on the BL theory and fully developed flow in the bulk. The horizontal- and time-averaged profiles of temperature, shown in figure 3(a), confirm that the flow can be split into two distinct regions: a well-mixed bulk with nearly uniform properties, and a thin BL characterized by a linear temperature profile, the slope of which is unitary when $z$ is rescaled with $\operatorname {\mathit {Nu}}$ (figure 3b). The thickness of this BL, $\lambda /L$, can be determined as the distance from the wall at which the linear function fitting the temperature profile in the bulk ($0.4\leqslant z \leqslant 0.6$) intersects the near-wall temperature fit. The measurement procedure is illustrated in figure 3(c), where the intersection is marked by the bullet. The Nusselt number sets the thickness of the BL $\lambda /L =1/(2\operatorname {\mathit {Nu}})$ (De Paoli et al. Reference De Paoli, Pirozzoli, Zonta and Soldati2022). In RD convection, it has been proposed by Otero et al. (Reference Otero, Dontcheva, Johnston, Worthing, Kurganov, Petrova and Doering2004) that the thermal BL thickness scales as $\lambda /L \sim \operatorname {\mathit {Ra}}^{-1}$ (consistent with $Nu \sim Ra$, from the classical theory (Malkus Reference Malkus1954; Priestley Reference Priestley1954; Howard Reference Howard1966) and the upper bound scaling derived by Doering & Constantin (Reference Doering and Constantin1998)). As illustrated in figure 3(d), this approximation is verified. Although both in 2-D and 3-D cases the thermal BL thickness follows $\lambda /L \sim \operatorname {\mathit {Ra}}^{-1}$, the prefactor differs (see figure 3d). This discrepancy arises from the distinct flow structures in 2-D and 3-D cases. In 3-D cases, owing to the additional degree of freedom compared with the 2-D case, plumes can freely move and reorganize towards the most efficient configuration, resulting in a different value of $\operatorname {\mathit {Nu}}$. Consequently, the BL thickness also varies. This phenomenon is analogous to RB convection, where 2-D and 3-D flows exhibit different BL thicknesses due to variations in $\operatorname {\mathit {Nu}}$ (Van Der Poel, Stevens & Lohse Reference Van Der Poel, Stevens and Lohse2013).

Figure 3. (a) Horizontal- and time-averaged profiles of temperature are shown for different $\operatorname {\mathit {Ra}}$ (De Paoli et al. Reference De Paoli, Pirozzoli, Zonta and Soldati2022). The near-wall region is magnified in (b), where the wall-normal coordinate is rescaled with $\operatorname {\mathit {Nu}}$. The profiles are self-similar, and in the BL follow a linear behaviour (solid line) with unitary slope. (c) The thickness of the BL is determined as the distance from the wall of the intersection between the linear profile fitting the bulk ($0.4\leqslant z \leqslant 0.6$) and the near-wall regions ($z\le 0.1\operatorname {\mathit {Nu}}/\operatorname {\mathit {Ra}}$). The case relative to $\operatorname {\mathit {Ra}}=10^4$ is reported. A close-up view of the near-wall region is shown in the inset. (d) Thickness of the thermal BL $\lambda /L$ as a function of the Rayleigh number. The 3-D measurements computed as discussed (bullets) are very well fitted by the correlation $50/\operatorname {\mathit {Ra}}$. The correlation proposed for 2-D flows by Otero et al. (Reference Otero, Dontcheva, Johnston, Worthing, Kurganov, Petrova and Doering2004) ($\lambda /L\sim 15/\operatorname {\mathit {Ra}}$) is also reported, as well as the value obtained from the Nusselt number $\lambda /L=1/(2\operatorname {\mathit {Nu}})$; data from De Paoli et al. (Reference De Paoli, Pirozzoli, Zonta and Soldati2022).

The profiles of dimensionless thermal dissipation $\overline {\langle |\boldsymbol {\nabla } T|^2\rangle _A}$ obtained from De Paoli et al. (Reference De Paoli, Pirozzoli, Zonta and Soldati2022) are shown in figure 4(a) for different values of the Rayleigh number. In the inset, the dissipation is rescaled by the respective Nusselt number, and shown up to 1. We observe that the BL contribution to the dissipation is more pronounced as $\operatorname {\mathit {Ra}}$ is increased. A more quantitative description is provided in the following. The thermal dissipation rate in the BL scales as $\sim \kappa (\varDelta /\lambda )^2$. Therefore, taking into account the layer extension ($\lambda /L\sim \operatorname {\mathit {Ra}}^{-1}$), the BL contribution towards the total thermal dissipation rate reads

(4.2)\begin{equation} \epsilon_{BL}\sim\kappa\frac{\varDelta^2}{\lambda^2}\frac{\lambda}{L}\sim\kappa\frac{\varDelta^2}{L^2}\operatorname{\mathit{Ra}}. \end{equation}

Assuming the flow in the bulk is well mixed (Grossmann & Lohse Reference Grossmann and Lohse2000; Bader & Zhu Reference Bader and Zhu2023; Song, Shishkina & Zhu Reference Song, Shishkina and Zhu2024), we get

(4.3)\begin{equation} \epsilon_{bulk}\sim\frac{\mathcal{V}\theta^2}{\ell}. \end{equation}

Here, $\mathcal {V}$ and $\theta$ are the typical velocity and temperature scales, respectively. The characteristic length scale is $\ell$, defined as the wavelength associated with the power-averaged mean wavenumber at the midheight $(\bar {k})$, i.e. $\ell /L=2{\rm \pi} /\bar {k}$. In numerical simulations, $\bar {k}$ is obtained from the time-averaged spectrum of the temperature field at $z=1/2$ (Hewitt et al. Reference Hewitt, Neufeld and Lister2014). The importance of $\epsilon _{BL}$ and $\epsilon _{bulk}$ relative to the total dissipation $\epsilon$, is reported in figure 4(b) for 3-D RD convection (De Paoli et al. Reference De Paoli, Pirozzoli, Zonta and Soldati2022). Here, $\epsilon _{BL}$ and $\epsilon _{bulk}$ are obtained from the profiles, and represent the mean value within the respective regions. In RB, to derive the scalings, the length scale is assumed to be the height of the domain $L$, which makes sense as there exist large-scale rolls. However, here in RD, the typical flow structures are columnar-like, making the length scale $\ell$ different from $L$. This difference is clearly visible in figures 1(a i) and 1(b i). From the definition of $\operatorname {\mathit {Nu}}$ (3.2) and assuming that in the bulk, $\theta \sim \operatorname {\mathit {Nu}}(\kappa \varDelta )/(\mathcal {V}L)$, and when $\operatorname {\mathit {Nu}}$ only comes from the fluctuation in the bulk, $\theta \sim (\epsilon _{bulk}/\mathcal {V})( L/\varDelta )$, we get

(4.4)\begin{equation} \epsilon_{bulk}\sim\kappa\frac{\varDelta^2}{L^2}{\operatorname{\mathit{Pe}}}\frac{\ell}{L}. \end{equation}

The same bulk scaling has also been reported for rapidly rotating convection (Song et al. Reference Song, Shishkina and Zhu2024) and magnetoconvection with strong vertical magnetic fields (Bader & Zhu Reference Bader and Zhu2023). In all these three systems, in the bulk there exists a new horizontal dominant length scale that is different from the height of the domain. In each of these three systems, a new dominant length scale emerges, distinct from the domain's height. This disparity constitutes a significant deviation from the original GL theory.

Figure 4. (a) Horizontal- and time-averaged profiles of dimensionless thermal dissipation, $\overline {\langle |\boldsymbol {\nabla } T|^2\rangle _A}$, are shown for different $\operatorname {\mathit {Ra}}$ (De Paoli et al. Reference De Paoli, Pirozzoli, Zonta and Soldati2022). In the inset, the dissipation is rescaled by the respective Nusselt number, and shown up to $\overline {\langle |\boldsymbol {\nabla } T|^2\rangle _A}/\operatorname {\mathit {Nu}}=1$. The BL contribution to the dissipation is more pronounced as $\operatorname {\mathit {Ra}}$ is increased. (b) Contributions to the mean scalar dissipation $\epsilon$ within the BL ($\epsilon _{BL}$) and in the bulk region ($\epsilon _{bulk}$); data are from De Paoli et al. (Reference De Paoli, Pirozzoli, Zonta and Soldati2022).

Determining the dominant wavelength in RD convection is a challenging task. The reason is linked to the complex way in which the dynamic near-wall flow structures interact with the stationary, large-scale columnar plumes controlling the bulk. A detailed review is provided by Hewitt (Reference Hewitt2020), which we summarize here with additional details including later results (De Paoli et al. Reference De Paoli, Pirozzoli, Zonta and Soldati2022). In 3-D cases, using asymptotic stability theory, Hewitt & Lister (Reference Hewitt and Lister2017) derived that

(4.5)\begin{equation} \ell/L\sim \operatorname{\mathit{Ra}}^{{-}1/2} \quad {\rm (3}\hbox{-}{\rm D)}. \end{equation}

Numerical simulations by Hewitt et al. (Reference Hewitt, Neufeld and Lister2014) and De Paoli et al. (Reference De Paoli, Pirozzoli, Zonta and Soldati2022), best fitted by scaling exponents of $-0.52$ and $-0.49$, respectively, agree well with this prediction (see also figure 5a). Therefore, we employ this scaling relation ($\ell /L\sim \operatorname {\mathit {Ra}}^{-1/2}$) for the centreline in 3-D cases. The situation is more complex in 2-D cases. Hewitt, Neufeld & Lister (Reference Hewitt, Neufeld and Lister2013b) derived analytically the scaling relation

(4.6)\begin{equation} \ell/L\sim \operatorname{\mathit{Ra}}^{{-}5/14} \quad {\rm (2}\hbox{-}{\rm D)}. \end{equation}

Wen, Corson & Chini (Reference Wen, Corson and Chini2015) have shown that for $\operatorname {\mathit {Ra}}\le 19\,976$ the centreline dominant length scale is well approximated by $\ell /L\sim \operatorname {\mathit {Ra}}^{-0.40}$. However, one can observe in figure 5(b) that when $\operatorname {\mathit {Ra}}\ge 39\,716$, the interplume spacing measured by Wen et al. (Reference Wen, Corson and Chini2015) is not unique. The conclusion is that in 2-D RD convection, at $\operatorname {\mathit {Ra}}\ge 39\,716$, a precise scaling remains to be established by running simulations in very wide domains and for very long times. In view of this, we consider the scaling proposed by Hewitt et al. (Reference Hewitt, Neufeld and Lister2013b), which represents the best theoretical prediction available, to be valid. Therefore, combining (4.4) with (3.14), (4.5) and (4.6), we get

(4.7)\begin{equation} \epsilon_{bulk}\sim\kappa\frac{\varDelta^2}{L^2}\operatorname{\mathit{Nu}}^{1/2} \quad {\rm (3}\hbox{-}{\rm D)}, \end{equation}

and

(4.8)\begin{equation} \epsilon_{bulk}\sim\kappa\frac{\varDelta^2}{L^2}{\operatorname{\mathit{Nu}}^{1/2}\operatorname{\mathit{Ra}}^{1/7}} \quad {\rm (2}\hbox{-}{\rm D)}. \end{equation}

Figure 5. Dominant length scales $\ell /L$ in the bulk ($z=1/2$) for (a) 3-D (Hewitt et al. Reference Hewitt, Neufeld and Lister2014; De Paoli et al. Reference De Paoli, Pirozzoli, Zonta and Soldati2022) and (b) 2-D (Hewitt et al. Reference Hewitt, Neufeld and Lister2012; Wen et al. Reference Wen, Corson and Chini2015) simulations. Theoretical scaling relations defined in (4.5) and (4.6), corresponding to 3-D and 2-D flows, respectively, are also indicated (dashed lines).

Finally, combining (3.26), (4.1), (4.2), (4.7) and (4.8), we reach an expression for $\operatorname {\mathit {Nu}}$ as a function of $\operatorname {\mathit {Ra}}$ for the 3-D and the 2-D cases, as follows:

(4.9)$$\begin{gather} \operatorname{\mathit{Nu}}=A_3\operatorname{\mathit{Ra}}+B_3\operatorname{\mathit{Nu}}^{1/2} \quad {\rm (3}\hbox{-}{\rm D)}, \end{gather}$$
(4.10)$$\begin{gather}\operatorname{\mathit{Nu}}=A_2\operatorname{\mathit{Ra}}+B_2\operatorname{\mathit{Nu}}^{1/2}\operatorname{\mathit{Ra}}^{1/7} \quad {\rm (2}\hbox{-}{\rm D)}. \end{gather}$$

As reported in figure 6, these scaling relations fit very well the data obtained from numerical simulations, both in the 2-D and in the 3-D cases. The values of the coefficients $A_2$, $A_3$, $B_2$, $B_3$, indicated in figure 6, are obtained as best fitting from the data shown, representing the numerical results available and with $\operatorname {\mathit {Ra}}>2\times 10^3$. The choice of considering values larger than this threshold is motivated by the flow topology: at low $\operatorname {\mathit {Ra}}$ the bulk flow structure is not columnar, as it is dominated by large-scale convective rolls (Graham & Steen Reference Graham and Steen1994), and therefore our theory does not apply. The expressions of $\operatorname {\mathit {Nu}}$ derived in (4.9) and (4.10) take the form of a linear scaling with a sublinear correction. The linear scaling was previously proposed for 2-D (Hewitt et al. Reference Hewitt, Neufeld and Lister2012) and 3-D (Hewitt et al. Reference Hewitt, Neufeld and Lister2014) flows. The scaling proposed here provides similar results to the linear scaling with sublinear corrections proposed for the 3-D case by Pirozzoli et al. (Reference Pirozzoli, De Paoli, Zonta and Soldati2021). However, in this case one fitting parameters fewer is used, i.e. all scaling exponents are known. The good agreement of the present scaling relations with the numerical measurements available in the literature suggests that the theory proposed is indeed valid and promising for higher $\operatorname {\mathit {Ra}}$. The RD system is completely defined by two parameters, namely the Rayleigh number $\operatorname {\mathit {Ra}}$ and the aspect ratio $W/L$. In 3-D flows at high-$\operatorname {\mathit {Ra}}$, it has been observed that all major flow statistics converge for an aspect ratio of $W/L=1$ (De Paoli et al. Reference De Paoli, Pirozzoli, Zonta and Soldati2022). This differs in 2-D systems: also at high $\operatorname {\mathit {Ra}}$ and due to the additional lateral confinement, the aspect ratio may have an influence on $\ell$ (Wen et al. Reference Wen, Corson and Chini2015). Therefore, it may be required to take $W/L$ into account in the present theory to better describe the transport properties in 2-D RD flows. To this aim and also to assess the physics of the scaling prefactors, additional simulations in large domains and at larger $\operatorname {\mathit {Ra}}$ are needed.

Figure 6. Compensated Nusselt number as a function of Rayleigh number. Results are for 2-D (Hewitt et al. Reference Hewitt, Neufeld and Lister2012; Wen et al. Reference Wen, Corson and Chini2015; De Paoli, Zonta & Soldati Reference De Paoli, Zonta and Soldati2016; Pirozzoli et al. Reference Pirozzoli, De Paoli, Zonta and Soldati2021; De Paoli et al. Reference De Paoli, Yerragolam, Lohse and Verzicco2024) and 3-D (Hewitt et al. Reference Hewitt, Neufeld and Lister2014; Pirozzoli et al. Reference Pirozzoli, De Paoli, Zonta and Soldati2021) simulations. New fitting curves are obtained considering all data with $\operatorname {\mathit {Ra}}>2\times 10^3$, with numerical values of the coefficients being $A_2 = (6.386 \pm 0.007)\times 10^{-3}$, $B_2=0.2533\pm 0.0050$ and $A_3 = (8.592 \pm 0.033)\times 10^{-3}$, $B_3=1.376\pm 0.059$. The coefficients of determination ($R^2$) for the best fitting curves provided are 0.991 and 0.999 in the 2-D and in the 3-D cases, respectively.

5. Conclusions

In summary, we have established two exact relationships, one pertaining to the Péclet number and the other to the thermal dissipation rate, in the context of Rayleigh–Darcy (RD) convection – a fundamental system for heat and mass transport in porous media. Inspired by previous models developed for RB convection (Grossmann & Lohse Reference Grossmann and Lohse2000, Reference Grossmann and Lohse2001), we have formulated a scaling theory for heat transfer in 2-D and 3-D RD flows, where the Nusselt number is expressed as a function of the Rayleigh number as described by (4.9) and (4.10). This theory enables us to provide a theoretical explanation to the sublinear empirical corrections proposed in prior studies (Pirozzoli et al. Reference Pirozzoli, De Paoli, Zonta and Soldati2021). Our investigations, supported by both the 2-D and 3-D literature results, confirm the validity of the proposed theory. Moreover, by taking the length scale of the flow structures into account, we also shed new light on the physical origins of the disparities in scaling relations between 2-D and 3-D RD convection.

Our findings are relevant to convective flows in homogeneous and isotropic porous media where the top-to-bottom density difference is defined. However, these hypotheses represent idealized conditions not taking into account additional flow features that occur in realistic processes, such as hydrodynamic dispersion (Liang et al. Reference Liang, Wen, Hesse and DiCarlo2018; Tsinober et al. Reference Tsinober, Rosenzweig, Class, Helmig and Shavit2022; Tsinober, Shavit & Rosenzweig Reference Tsinober, Shavit and Rosenzweig2023), medium heterogeneity (Simmons, Fenstemaker & Sharp Reference Simmons, Fenstemaker and Sharp2001; Dentz, Hidalgo & Lester Reference Dentz, Hidalgo and Lester2023), anisotropy (Ennis-King, Preston & Paterson Reference Ennis-King, Preston and Paterson2005; De Paoli et al. Reference De Paoli, Zonta and Soldati2016) and alternative flow configurations (Hidalgo et al. Reference Hidalgo, Fe, Cueto-Felgueroso and Juanes2012; De Paoli, Zonta & Soldati Reference De Paoli, Zonta and Soldati2017; Letelier et al. Reference Letelier, Ulloa, Leyrer and Ortega2023). Nonetheless, present findings represent a crucial step required to develop a robust and physically sound theory for convection in porous media flows, which can be further expanded to include the presence of the different variations mentioned above. We consider for instance the sequestration of carbon dioxide in saline aquifers. Such a system is usually modelled as a rectangular domain initially filled with brine and confined by two horizontal low-permeability layers, and therefore it is assumed to be impermeable at the bottom boundary (no-flux) (Huppert & Neufeld Reference Huppert and Neufeld2014). Here, the solute enters from the top, in correspondence of which the concentration of ${\rm CO}_2$ is constant. This flow configuration, defined as ‘one-sided’ (Hewitt, Neufeld & Lister Reference Hewitt, Neufeld and Lister2013a) or ‘semi-infinite’ (Ennis-King et al. Reference Ennis-King, Preston and Paterson2005), is subject to a transient behaviour: the average ${\rm CO}_2$ concentration within the system will increase over time, and it will be progressively harder to keep dissolving solute. In quantitative terms this means that, after a short initial phase in which dissolution increases due to the formation and growth of the fingers, the flux of solute through the top boundary will later reduce as a result of the saturation of the domain. The dynamics of such a system has been thoroughly characterized (Slim Reference Slim2014), and it is shown to be quantitatively related to the dynamics observed in RD convection (Hewitt et al. Reference Hewitt, Neufeld and Lister2013a; De Paoli et al. Reference De Paoli, Zonta and Soldati2017). In order to describe the evolution of the one-sided system with a simple box-model, accurate predictions of the transport scaling in RD convection are essential. In the presence of high-permeability formations such as the Utsira Sand reservoir at Sleipner (Bickle et al. Reference Bickle, Chadwick, Huppert, Hallworth and Lyle2007), the Rayleigh–Darcy number may be as high as $6\times 10^5$ (Hewitt et al. Reference Hewitt, Neufeld and Lister2013a), which is beyond current numerical capabilities. As a result, the theoretical results provided in our work will play a crucial role as a tool to determine the long-term evolution of flows in semi-infinite domains.

Funding

We gratefully acknowledge the financial support from the Max Planck Society, the German Research Foundation through grants 521319293 and 540422505, the Alexander von Humboldt Foundation and the Daimler Benz Foundation. This project has also received funding from the European Union's Horizon Europe research and innovation programme under the Marie Sklodowska–Curie grant agreement no. 101062123.

Declaration of interests

The authors report no conflict of interest.

References

Ahlers, G., Grossmann, S. & Lohse, D. 2009 Heat transfer and large scale dynamics in turbulent Rayleigh–Bénard convection. Rev. Mod. Phys. 81 (2), 503.CrossRefGoogle Scholar
Bader, S.H. & Zhu, X. 2023 Scaling relations in quasi-static magnetoconvection with a strong vertical magnetic field. J. Fluid Mech. 976, A4.CrossRefGoogle Scholar
Bickle, M., Chadwick, A., Huppert, H.E., Hallworth, M. & Lyle, S. 2007 Modelling carbon dioxide accumulation at Sleipner: implications for underground carbon storage. Earth Planet. Sci. Lett. 255 (1), 164176.CrossRefGoogle Scholar
Chillà, F. & Schumacher, J. 2012 New perspectives in turbulent Rayleigh–Bénard convection. Eur. Phys. J. E 35, 125.CrossRefGoogle ScholarPubMed
De Paoli, M. 2021 Influence of reservoir properties on the dynamics of a migrating current of carbon dioxide. Phys. Fluids 33 (1), 016602.CrossRefGoogle Scholar
De Paoli, M. 2023 Convective mixing in porous media: a review of Darcy, pore-scale and Hele-Shaw studies. Eur. Phys. J. E 46 (12), 129.CrossRefGoogle ScholarPubMed
De Paoli, M., Pirozzoli, S., Zonta, F. & Soldati, A. 2022 Strong Rayleigh–Darcy convection regime in three-dimensional porous media. J. Fluid Mech. 943, A51.CrossRefGoogle Scholar
De Paoli, M., Yerragolam, G.S., Lohse, D. & Verzicco, R. 2024 A finite difference solver for numerical simulations of convective porous media flows (in preparation).CrossRefGoogle Scholar
De Paoli, M., Zonta, F. & Soldati, A. 2016 Influence of anisotropic permeability on convection in porous media: implications for geological ${\rm CO}_2$ sequestration. Phys. Fluids 28 (5), 056601.CrossRefGoogle Scholar
De Paoli, M., Zonta, F. & Soldati, A. 2017 Dissolution in anisotropic porous media: modelling convection regimes from onset to shutdown. Phys Fluids 29 (2), 026601.CrossRefGoogle Scholar
Dentz, M., Hidalgo, J.J. & Lester, D. 2023 Mixing in porous media: concepts and approaches across scales. Transp. Porous Med. 146 (1–2), 553.CrossRefGoogle Scholar
Doering, C.R. & Constantin, P. 1998 Bounds for heat transport in a porous layer. J. Fluid Mech. 376, 263296.CrossRefGoogle Scholar
Emami-Meybodi, H., Hassanzadeh, H., Green, C.P. & Ennis-King, J. 2015 Convective dissolution of ${\rm CO}_2$ in saline aquifers: progress in modeling and experiments. Intl J. Greenh. Gas Control 40, 238266.CrossRefGoogle Scholar
Ennis-King, J., Preston, I. & Paterson, L. 2005 Onset of convection in anisotropic porous media subject to a rapid change in boundary conditions. Phys. Fluids 17 (8), 084107.CrossRefGoogle Scholar
Graham, M.D. & Steen, P.H. 1994 Plume formation and resonant bifurcations in porous-media convection. J. Fluid Mech. 272, 6790.CrossRefGoogle Scholar
Grossmann, S. & Lohse, D. 2000 Scaling in thermal convection: a unifying theory. J. Fluid Mech. 407, 2756.CrossRefGoogle Scholar
Grossmann, S. & Lohse, D. 2001 Thermal convection for large Prandtl numbers. Phys. Rev. Lett. 86 (15), 3316.CrossRefGoogle ScholarPubMed
Guo, R., Sun, H., Zhao, Q., Li, Z., Liu, Y. & Chen, C. 2021 A novel experimental study on density-driven instability and convective dissolution in porous media. Geophys. Res. Lett. 48 (23), e2021GL095619.CrossRefGoogle Scholar
Hassanzadeh, P., Chini, G.P. & Doering, C.R. 2014 Wall to wall optimal transport. J. Fluid Mech. 751, 627662.CrossRefGoogle Scholar
Hewitt, D.R. 2020 Vigorous convection in porous media. Proc. Math. Phys. Engng Sci. 476 (2239), 20200111.Google ScholarPubMed
Hewitt, D.R. & Lister, J.R. 2017 Stability of three-dimensional columnar convection in a porous medium. J. Fluid Mech. 829, 89111.CrossRefGoogle Scholar
Hewitt, D.R., Neufeld, J.A. & Lister, J.R. 2012 Ultimate regime of high Rayleigh number convection in a porous medium. Phys. Rev. Lett. 108 (22), 224503.CrossRefGoogle Scholar
Hewitt, D.R., Neufeld, J.A. & Lister, J.R. 2013 a Convective shutdown in a porous medium at high Rayleigh number. J. Fluid Mech. 719, 551586.CrossRefGoogle Scholar
Hewitt, D.R., Neufeld, J.A. & Lister, J.R. 2013 b Stability of columnar convection in a porous medium. J. Fluid Mech. 737, 205231.CrossRefGoogle Scholar
Hewitt, D.R., Neufeld, J.A. & Lister, J.R. 2014 High Rayleigh number convection in a three-dimensional porous medium. J. Fluid Mech. 748, 879895.CrossRefGoogle Scholar
Hidalgo, J.J., Dentz, M., Cabeza, Y. & Carrera, J. 2015 Dissolution patterns and mixing dynamics in unstable reactive flow. Geophys. Res. Lett. 42 (15), 63576364.CrossRefGoogle Scholar
Hidalgo, J.J., Fe, J., Cueto-Felgueroso, L. & Juanes, R. 2012 Scaling of convective mixing in porous media. Phys. Rev. Lett. 109 (26), 264503.CrossRefGoogle ScholarPubMed
Howard, L.N. 1966 Convection at high Rayleigh number. In Proceedings of the 11th International Congress of Applied Mechanics, pp. 1109–1115. Springer.CrossRefGoogle Scholar
Huppert, H.E. & Neufeld, J.A. 2014 The fluid mechanics of carbon dioxide sequestration. Annu. Rev. Fluid Mech. 46, 255272.CrossRefGoogle Scholar
Kraichnan, R.H. 1962 Turbulent thermal convection at arbitrary Prandtl number. Phys. Fluids 5 (11), 13741389.CrossRefGoogle Scholar
Letelier, J.A., Mujica, N. & Ortega, J.H. 2019 Perturbative corrections for the scaling of heat transport in a Hele-Shaw geometry and its application to geological vertical fractures. J. Fluid Mech. 864, 746767.CrossRefGoogle Scholar
Letelier, J.A., Ulloa, H.N., Leyrer, J. & Ortega, J.H. 2023 Scaling ${\rm CO}_2$–brine mixing in permeable media via analogue models. J. Fluid Mech. 962, A8.CrossRefGoogle Scholar
Liang, Y., Wen, B., Hesse, M.A. & DiCarlo, D. 2018 Effect of dispersion on solutal convection in porous media. Geophys. Res. Lett. 45 (18), 96909698.CrossRefGoogle Scholar
MacMinn, C.W., Neufeld, J.A., Hesse, M.A. & Huppert, H.E. 2012 Spreading and convective dissolution of carbon dioxide in vertically confined, horizontal aquifers. Water Resour. Res. 48 (11).CrossRefGoogle Scholar
Malkus, W.V.R. 1954 Discrete transitions in turbulent convection. Proc. R. Soc. Lond. A 225 (1161), 185195.Google Scholar
Otero, J., Dontcheva, L.A., Johnston, H., Worthing, R.A., Kurganov, A., Petrova, G. & Doering, C.R. 2004 High-Rayleigh number convection in a fluid-saturated porous layer. J. Fluid Mech. 500, 263281.CrossRefGoogle Scholar
Pirozzoli, S., De Paoli, M., Zonta, F. & Soldati, A. 2021 Towards the ultimate regime in Rayleigh–Darcy convection. J. Fluid Mech. 911, R4.CrossRefGoogle Scholar
Priestley, C.H.B. 1954 Convection from a large horizontal surface. Austral. J. Phys. 7 (1), 176201.CrossRefGoogle Scholar
Schrag, D.P. 2007 Preparing to capture carbon. Science 315 (5813), 812813.CrossRefGoogle ScholarPubMed
Simmons, C.T., Fenstemaker, T.R. & Sharp, J.M. Jr. 2001 Variable-density groundwater flow and solute transport in heterogeneous porous media: approaches, resolutions and future challenges. J. Contam. Hydrol. 52 (1–4), 245275.CrossRefGoogle ScholarPubMed
Slim, A.C. 2014 Solutal-convection regimes in a two-dimensional porous medium. J. Fluid Mech. 741, 461491.CrossRefGoogle Scholar
Song, J., Shishkina, O. & Zhu, X. 2024 Scaling regimes in rapidly rotating thermal convection at extreme Rayleigh numbers. J. Fluid Mech. 984, A45.CrossRefGoogle Scholar
Spiegel, E.A. 1963 A generalization of the mixing-length theory of turbulent convection. Astrophys. J. 138, 216.CrossRefGoogle Scholar
Tsinober, A., Rosenzweig, R., Class, H., Helmig, R. & Shavit, U. 2022 The role of mixed convection and hydrodynamic dispersion during ${\rm CO}_2$ dissolution in saline aquifers: a numerical study. Water Resour. Res. 58 (3), e2021WR030494.CrossRefGoogle Scholar
Tsinober, A., Shavit, U. & Rosenzweig, R. 2023 Numerical investigation of the influence of hydrodynamic dispersion on solutal natural convection. Water Resour. Res. 59, e2023WR034475.CrossRefGoogle Scholar
Ulloa, H.N. & Letelier, J.A. 2022 Energetics and mixing of thermally driven flows in Hele-Shaw cells. J. Fluid Mech. 930, A16.CrossRefGoogle Scholar
Van Der Poel, E.P., Stevens, R.J.A.M. & Lohse, D. 2013 Comparison between two-and three-dimensional Rayleigh–Bénard convection. J. Fluid mech. 736, 177194.CrossRefGoogle Scholar
Wen, B., Chang, K.W. & Hesse, M.A. 2018 Rayleigh–Darcy convection with hydrodynamic dispersion. Phys. Rev. Fluids 3 (12), 123801.CrossRefGoogle Scholar
Wen, B., Corson, L.T. & Chini, G.P. 2015 Structure and stability of steady porous medium convection at large Rayleigh number. J. Fluid Mech. 772, 197224.CrossRefGoogle Scholar
Wen, B., Dianati, N., Lunasin, E., Chini, G.P. & Doering, C.R. 2012 New upper bounds and reduced dynamical modeling for Rayleigh–Bénard convection in a fluid saturated porous layer. Commun. Nonlinear Sci. Numer. Simul. 17 (5), 21912199.CrossRefGoogle Scholar
Xia, K.-Q., Huang, S.-D., Xie, Y.-C. & Zhang, L. 2023 Tuning heat transport via coherent structure manipulation: recent advances in thermal turbulence. Natl Sci. Rev. 10 (6), nwad012.CrossRefGoogle ScholarPubMed
Figure 0

Figure 1. Instantaneous dimensionless temperature field $T$ for convection in (a) porous media for $\operatorname {\mathit {Ra}}=10^4$, $W/L=1$ (Pirozzoli et al.2021), labelled as RD convection, and (b) in classical RB turbulence (thermal Rayleigh number $10^9$, Prandtl number $1$, $W/L=1$). The temperature distribution is shown over the entire volume (a i, b i), at the centreline $z=1/2$ (a ii, b ii) and near the upper wall (a iii, b iii).

Figure 1

Figure 2. (a) The r.m.s. of velocity used to estimate the Péclet number, $\operatorname {\mathit {Pe}}$. Horizontal component (${\rm r.m.s.}(u)$), vertical component (${\rm r.m.s.}(w)$) and total (defined as $\sqrt {[{\rm r.m.s.}(u)]^2+[{\rm r.m.s.}(w)]^2}$ for 2-D cases, and $\sqrt {2[{\rm r.m.s.}(u)]^2+[{\rm r.m.s.}(w)]^2}$ for 3-D cases) are reported; data from Hewitt et al. (2012, 2014). (b) Comparison of $\operatorname {\mathit {Pe}}$ compensated with $(\operatorname {\mathit {Nu}}-1)\operatorname {\mathit {Ra}}$ as from (3.14). Here $\operatorname {\mathit {Pe}}$ is computed with $\mathcal {V}$ defined as in (3.17) for the data from Hewitt et al. (2012, 2014), and as in (3.13) for data from De Paoli et al. (2024). (c) Ratio of the Nusselt number to the dimensionless thermal dissipation, derived in (3.25); data from Pirozzoli et al. (2021) and De Paoli et al. (2024).

Figure 2

Figure 3. (a) Horizontal- and time-averaged profiles of temperature are shown for different $\operatorname {\mathit {Ra}}$ (De Paoli et al.2022). The near-wall region is magnified in (b), where the wall-normal coordinate is rescaled with $\operatorname {\mathit {Nu}}$. The profiles are self-similar, and in the BL follow a linear behaviour (solid line) with unitary slope. (c) The thickness of the BL is determined as the distance from the wall of the intersection between the linear profile fitting the bulk ($0.4\leqslant z \leqslant 0.6$) and the near-wall regions ($z\le 0.1\operatorname {\mathit {Nu}}/\operatorname {\mathit {Ra}}$). The case relative to $\operatorname {\mathit {Ra}}=10^4$ is reported. A close-up view of the near-wall region is shown in the inset. (d) Thickness of the thermal BL $\lambda /L$ as a function of the Rayleigh number. The 3-D measurements computed as discussed (bullets) are very well fitted by the correlation $50/\operatorname {\mathit {Ra}}$. The correlation proposed for 2-D flows by Otero et al. (2004) ($\lambda /L\sim 15/\operatorname {\mathit {Ra}}$) is also reported, as well as the value obtained from the Nusselt number $\lambda /L=1/(2\operatorname {\mathit {Nu}})$; data from De Paoli et al. (2022).

Figure 3

Figure 4. (a) Horizontal- and time-averaged profiles of dimensionless thermal dissipation, $\overline {\langle |\boldsymbol {\nabla } T|^2\rangle _A}$, are shown for different $\operatorname {\mathit {Ra}}$ (De Paoli et al.2022). In the inset, the dissipation is rescaled by the respective Nusselt number, and shown up to $\overline {\langle |\boldsymbol {\nabla } T|^2\rangle _A}/\operatorname {\mathit {Nu}}=1$. The BL contribution to the dissipation is more pronounced as $\operatorname {\mathit {Ra}}$ is increased. (b) Contributions to the mean scalar dissipation $\epsilon$ within the BL ($\epsilon _{BL}$) and in the bulk region ($\epsilon _{bulk}$); data are from De Paoli et al. (2022).

Figure 4

Figure 5. Dominant length scales $\ell /L$ in the bulk ($z=1/2$) for (a) 3-D (Hewitt et al.2014; De Paoli et al.2022) and (b) 2-D (Hewitt et al.2012; Wen et al.2015) simulations. Theoretical scaling relations defined in (4.5) and (4.6), corresponding to 3-D and 2-D flows, respectively, are also indicated (dashed lines).

Figure 5

Figure 6. Compensated Nusselt number as a function of Rayleigh number. Results are for 2-D (Hewitt et al.2012; Wen et al.2015; De Paoli, Zonta & Soldati 2016; Pirozzoli et al.2021; De Paoli et al.2024) and 3-D (Hewitt et al.2014; Pirozzoli et al.2021) simulations. New fitting curves are obtained considering all data with $\operatorname {\mathit {Ra}}>2\times 10^3$, with numerical values of the coefficients being $A_2 = (6.386 \pm 0.007)\times 10^{-3}$, $B_2=0.2533\pm 0.0050$ and $A_3 = (8.592 \pm 0.033)\times 10^{-3}$, $B_3=1.376\pm 0.059$. The coefficients of determination ($R^2$) for the best fitting curves provided are 0.991 and 0.999 in the 2-D and in the 3-D cases, respectively.