Research

From Newton’s mechanics, many scientists and mathematicians have devoted to studying differential equations to explain several phenomena in the nature. It is natural to ask whether we can solve boundary value problems of given partial differential equations. If we know an existence of solution of the problem, one can also ask some properties of solutions that have.

I have worked on various research projects to find a good perspective of studying those problems. Some of topics are

Mathematical Fluid Dynamics
- Free boundary problem from petroleum engineering
- Local regularity theory for Stokes equations
- Equations from magnetohydrodynamics
Long time behavior of solutions to Kinetic equations
Elliptic equations with singular drift terms

Mathematical Fluid Dynamics: free boundary problems from petroleum engineering

In 1934, Muskat introduced the free-boundary problem for two immiscible fluids in porous media to study the mechanism of oil recovery. This problem has received considerable attention due to the widespread use of waterflooding, which led to petroleum overproduction before the Great Depression (see Pennsylvania oil rush and references therein).

Since we consider a flow in porous media, the flow is governed by an experimental Darcy’s law

$$\frac{\mu}{\kappa}u +\nabla_{x,y} p =-\rho g e_{d+1},\quad \mathrm{div}_{x,y} u=0.

Here $u$ denotes the velocity of the fluid, $p$ denotes the pressure $\mu>0$ stands for the dynamic viscosity, $\kappa>0$ the permeability of the porous media, $\rho>0$ the density of the fluid, and $g>0$ the gravity constant. The fluid is occupied in

$$\Omega_f=\{(x,y) : x\in \mathbb{R}^d, y<f(x)\}$$

and we impose that the normal velocity is continuous

$$u\cdot n=0\quad \text{on } \Sigma_f$$

satisfies the kinematic boundary condition:

$$\partial_t f=\sqrt{1+|\nabla f|^2}(u\cdot n)\quad \text{on } \Sigma_f.$$

and by Young-Laplace eqation, the pressure jump is proportional to the mean curvature of the interface

$$p=\mathfrak{s}H(f)\quad \text{on } \Sigma_f(t).$$

This problem is called the one-phase Muskat problem. While the horizontal Hele-Shaw problem driven by injection or suction has been widely studied mathematically, the one-phase Muskat problem exhibits a different physical nature and mathematical structure due to gravity.

Pioneered by the work by Cordoba and Gancedo, there are plenty of works on local well-posedness, global well-posedness, finite-time singularity formation, and regularity properties of solutions to the Muskat problem such as desingularization of corners, existence of self-similar solutions which immediately smooth out from an exact corner. However, there are few results on global solvability of the one-phase Muskat problem. For the global solvability result, Gancedo, Garcia-Juarez, Patel, Strain (2019) global well-posedness of the one-phase Muskat problem when initial data is in Wiener spaces with medium size. Later, Nguyen (2020) proved small data global well-posedness of the one-phase Muskat problem whose initial data belongs to critical Besov spaces. These result allows viscosity jumps. However, both results do not include the effect of surface tensions. For peridioc Lipschitz initial data with arbitrary size, Dong, Gancedo, Nguyen (2023, 2026) proved global well-posedness of the one-phase Muskat problem by introducing layer potential techinques. Moreover, the solution is unique in viscosity solution sense.

If we include surface tension effects, only available global solvabilitiy results are due to Chen, Nguyen, Xu (2023, 2026) and Lazar (2026) for two-fluid cases. However, these result assumed equal viscosity and hence those results cannot cover the one-phase problem.

In my joint work with Hongjie Dong DK26, we proved small-data global well-posedness of the one-phase Muskat problem with surface tensions. Moreover, we proved that the Lipschitz norm of interface converges to zero as $t\rightarrow\infty$.

We reformulate the one-phase Muskat problem in terms of interface. Then the problem becomes an evolution equation in terms of interface

$$\partial_t f + G(f)(f+H(f))=0.$$

Then the operator $G(\eta)(\eta+H(\eta))$ is a third order parabolic equations and hence it is delicate to apply viscosity solution approach due to lack of comparision pricniple which is not available in higher order parabolic equations.

To overcome this difficulty, we prove that $L^2$-norm of interface is Lyapunov functional. This was first observed by Alazard and Bresch (2024) in periodic domain case. We extend this to whole space case by a suitable approximation argument and regularity esetimates for harmonic functions. Moreover, using a rigidity estimate $$|D^2u|^2 |Du|^2-|D^2 u Du|^2 \geq \frac{1}{d+1}|D^2 u|^2 |Du|,$$ we can further show that $f$ satisfies $$ \frac{1}{2}\frac{d}{dt}\int_{\mathbb{R}^d} |f|^2 dx+\frac{1}{\mathcal{F}(\Vert f \Vert_{H^s})} \Vert f\Vert_{\dot{H}^{3/2}}^2 \leq 0.$$ This and a refined expansion of the Dirichlet-Neumann operator, we can show that the solution exists globally and the Lipschitz norm of the interface converges to zero as $t\rightarrow\infty$.

Mathematical Fluid Dynamics: local regularity theory for Stokes equations

Recently, I worked on local regularity theory for Stokes equations. To explain this result, let us consider nonstationary Stokes equations with variable viscosity coefficients: $$ \partial_t u - a^{ij}(t,x)D_{ij} u + \nabla p = f,\quad \mathrm{div}\, u=g\quad \text{in } U. $$ Here $a^{ij}(t,x)$ is uniformly elliptic as in the first section. Also, $U$ is a cylindrical domain in $\mathbb{R}^{d+1}$. Besides mathematical interests, such an equation can be used to model non-Newtonian fluids that have thixotropy, i.e., time-dependent shear thinning properties. Also, the equation is also naturally introduced when we introduce Stokes equations on a manifold. In the case of the heat equation, it is easy to show that if $u \in L^2(Q_1)$ is a very weak solution to heat equations $\partial_t u-\Delta u =0$ in $Q_1$, then $u$ is smooth in $Q_1$. Moreover, one can also show that $$ \Vert D^2 u \Vert_{L_{2}(Q_{1/2})}\leq N \Vert u \Vert_{L_{2}(Q_1)}.$$ However, unlike heat equation, Stokes equation may not have good regularity properties. This is captured by a well-known example given by Serrin (1962): consider $$ u(t,x)=c(t)\nabla \phi(x),\quad p(t,x)=-c(t)\phi(x),$$ where $\phi$ is harmonic. Then $(u,p)$ satisfies $$ \partial_t u -\Delta u + \nabla p=0,\quad \mathrm{div} u= 0.$$ Note that we can choose $c$ as an absolutely continuous function.

When $g=0$ and $a^{ij}=\delta^{ij}$, such estimate was first obtained by Chen-Strain-Yau-Tsai (2008) proved that if $1<s,q<\infty$ and $g=0$, then $$ \Vert D^2 u \Vert_{L_{s,q}(Q_{1/2})} \leq N \Vert u \Vert_{L_{s,q}(Q_1)} + N \Vert f \Vert_{L_{s,q}(Q_1)} $$ holds. Independently, Jin (2013) and Wolf (2015) also proved the validity of such an estimate when $s=q=2$. Later, Dong-Phan (2021) obtained that if $a^{ij}$ is in $VMO_x$, then such regularity estimate is possible even if $g$ is nonzero. However, they need to assume that $u\in \tilde{W}^{1,2}_{s,q}(Q_1)$ to get the desired regularity estimate, where $\tilde{W}^{1,2}_{s,q}(Q_1)=\{ u, Du, D^2 u \in L_{s,q}(Q_1), \partial_t u\in L_1(Q_1)\}$. In the same paper, they also obtained the corresponding results for Stokes equations in divergence form. In this case, the leading coefficient need not be bounded. Observe that these results are a priori results to obtain such regularity results. Recently, Dong-Li (2023) obtained an interior regularity estimate when $s=q$ and $a^{ij}$ is merely measurable in $t$ and Holder continuous in $x$ via level set argument following Caffarelli-Peral, Wang, Byun, etc.

Another natural question is the validity of such regularity estimates when we impose boundary conditions. Chang-Kang (2020) proved that it is not possible to obtain a gradient estimate for Stokes equations if we impose a no-slip boundary condition due to the non-local effect near the boundary. Very recently, Dong-Kim-Phan (2022) proved that Hessian estimates for Stokes equations in nondivergence form are possible if we impose the Lions boundary conditions, which are special cases of Navier-slip boundary conditions. These results are also a priori similar to Dong-Phan (2021). Very recently, Chen-Liang-Tsai (2023) proved that gradient estimate for very weak solutions is possible if we impose the Navier boundary condition, $d=3$, $a^{ij}=\delta^{ij}$, and $g=0$.

In the joint work with Hongjie Dong, we proved in [(DK23)](https://arxiv.org/abs/2308.09220) that when $a^{ij}$ is merely measurable in $t$ and has small mean oscillation in small cylinders, we proved that interior local $L_{s,q}$-Hessian estimates are possible for $u\in\tilde{W}^{1,2}_{q_0}(Q_1)$ in the case of nondivergence form equation and $u\in W^{0,1}_{q_0}(Q_1)$ in the case of equations in divergence form. For the boundary estimate, we proved the similar result for the equation in nondivergence form if we impose Lions boundary conditions. This result was further extended to curved boundary case in [[DK24]](http://arxiv.org/abs/2408.17321).

As a natural counterpart to Sobolev estimates, one can also ask the validity of $C^1$ or $C^2$ estimates for Stokes equations with variable viscosity coefficients. In a joint work with Hongjie Dong, we also obtained $C^1_x$-estimates for equations in divergence form when the viscosity coefficients satisfy $L_2$-$\mathrm{DMO}_x$ condition. As a corollary, we also obtained spatial Schauder estimates as well. Unlike usual parabolic equations, we cannot expect temporal smoothing of the equation as we can observe this phenomenon via Serrin's example.

Mathematical Fluid Dynamics: equations from Magnetohydrodynamics

I worked on the global well-posedness of fluid equations that have fractional dissipations. To explain my result, let us first consider the generalized magnetohydrodynamics equations:

\begin{equation} \left\{\begin{aligned} \partial_t u -\nu\Lambda^{2\alpha}u+(u\cdot\nabla)u+\nabla p &=(b\cdot \nabla)b,\\\ \partial_t b -\eta\Lambda^{2\beta}b + (u\cdot \nabla)b &=(b\cdot\nabla)u,\\ \mathrm{div}\, u =\mathrm{div}\, b&=0,\\ u=u_0,\quad b=b_0& \end{aligned} \right. \end{equation}

Here $u:\mathbb{R}^d\times (0,\infty)\rightarrow \mathbb{R}^d$ denotes the velocity field and the magnetic field, respectively. For $s>0$, $\Lambda^s$ denotes the fractional Laplacian which is defined by the Fourier transform:

\begin{equation} \widehat{\Lambda^s f}(\xi)=|\xi|^s \hat{f}(\xi). \end{equation}

When $\alpha=\beta=1$, the system is reduced to the classical MHD equation which was first introduced by Hannes Alfvén to describe the fluid of motions that were induced by magnetic fields.

Many researchers have devoted themselves to establishing well-posedness theory and regularity properties of solutions to the equation. Duvaut-Lions [DL72] proved the existence of weak solutions to the equation. Later, the result was extended by Wu [W03] for general $\alpha$ and $\beta$.

Starting from Arnol’d pioneering work, many researchers studied differential equations via dynamical points of view. It is natural to ask a equilibrium point of the corresponding dynamical systems. Arnol’d considered incompressible Euler equations that are regarded as a geodesic equations to find a fixed point of the equations, which is a steady Euler equation.

Later, Moffatt considered similar situations by working on magnetic fields. He constructed magnetohydrodyanmics equilibria from ideal MHD equations by letting $t\rightarrow \infty$. This procedure is called a magnetic relaxation. However, his construction has a limitation because of the problem on discontinuity in time variables. Later, Moffatt suggested a model that could lead to a desirable magnetic relaxation:

Moffatt introduced this model when $\alpha=1$ and $\eta=0$. When $\nu,\eta>0$, $\alpha=\beta=1$, and $d=2,3$, McCormick-Robinson-Rodrigo [MRR14] proved global existence of weak solutions of (4). For the two-dimensional case, they also proved uniqueness and regularity of weak solutions. Recently, Ji-Tan [JT21] proved global existence of strong solutions of (4) when $d=3$, $\alpha=1$, and $\beta\geq 3/2$. On the other hand, when $\nu>0$, $\eta=0$, and $\alpha=0$, Brenier [B14] proved global existence of dissipative weak solutions on the two-dimensional torus $\mathbb{T}^2$. When $\nu>0$, $\eta=0$, and $\alpha=1$, Fefferman et.al. [FMRR14] proved local existence and uniqueness of strong solutions of (4) on $\mathbb{R}^d$, $d=2,3$. Recently, when $\nu>0$, $\eta=0$, and $\alpha>d/2+1$, Beekie-Friedlander-Vicol [BFV21] established global existence of strong solutions of (4) on the torus $\mathbb{T}^d$, $d=2,3$. Moreover, they also investigated 2D stability and 3D instability as well as the long-time behavior of solutions. One natural question is whether we can relax these regularity assumptions on $\alpha$ and $s$ to guarantee the global solution to this Stokes-Magneto system.

In a joint work with Hyunseok Kim [HK22], we proved the existence of global weak solutions of the problem for general $\alpha,\beta$ on whole space $\mathbb{R}^d$, $d\geq 2$. More precisely, if $\alpha$, $\beta$ satisfy $1/2<\alpha<(d+1)/2$, $\beta >0$, and $\min(\alpha+\beta,2\alpha+\beta-1)>d/2$, then we proved the existence of global weak solutions of (4). Moreover, we also proved that weak solutions are unique if $\beta \geq 1$ and $\min (\alpha+\beta,2\alpha+\beta-1)\geq d/2+1$, in addition.

In the previous research work, we only proved the uniqueness of weak solutions when $\beta \geq 1$. Even though we can guarantee the existence of weak solutions for $\beta$ below 1, due to the structure that we used, we cannot proceed to guarantee the uniqueness of weak solutions. In the recent joint work with Hantaek Bae and Jaeyong Shin, we proved that when $\alpha$, $\beta$ satisfy $1/2<\alpha<(d+1)/2$, $\beta>1/2$, $\alpha+\beta<d+1$ (and some additional technical conditions if $\alpha>(d+2)/4$), local-in-time mild solutions exist and global-in-time mild solutions exist for sufficiently small initial data which belongs to the natural scaling critical $L^p$-space corresponding to our Stokes-Magneto system. We also proved that $\lim_{t\rightarrow\infty}\Vert {b}(t) \Vert_{L^p}=0$. In the same paper, when $\eta=0$ and on the torus, we proved that a local-in-time strong solution exists for more irregular data and it becomes global when $\alpha\geq d/2+1$ and $s\geq 1$. Our result gives a partial answer on the global existence part given in Question 1 of Beekie-Friedlander-Vicol when $\alpha \geq d/2+1$.

Kinetic equations

Also, I have worked kinetic equations in partiuclar, long-time dynamics of kinetic equations induced by Riesz interactions. To explain this problem, consider the following systems of ODEs:

\begin{equation} \left\{ \begin{aligned} \dot{x}_i &= v_i\\ \dot{v}_i &=-\frac{1}{N} \nabla_{x_i} H_N(x_1,\dots,x_N),\quad i=1,\dots,N,\\ x_i(0)=x_i^0,\quad v_i(0)=v_i^0, \end{aligned} \right. \end{equation} where \begin{equation} H_N^\alpha(x_1,\dots,x_N)=\lambda\sum_{i\neq j} V_\alpha(x_i-x_j),\quad V_\alpha(x)=\frac{\Gamma((3-2\alpha)/2)}{\pi^{3/2}2^{2\alpha}\Gamma(\alpha)}\frac{1}{|x|^{3-2\alpha}} \end{equation} for $\lambda\in\{-1,1\}$ and $0<\alpha<3/2$. We call $V_\alpha(x)$ as a Riesz kernel. When $\alpha=1$, the kernel becomes the classical Coulomb potential. As $N\rightarrow \infty$, one can formally derive a system of partial differential equations, we call this equation as Vlasov-Riesz systems of order $\alpha$: \begin{equation} \left\{ \begin{aligned} \partial_t f + v\cdot \nabla_x f +\lambda E[f]\cdot\nabla_v f=0\quad \text{in } (0,\infty)\times\mathbb{R}^3_x\times \mathbb{R}^3_v,\\ f=f_0\quad \text{on } \{t=0\}\times \mathbb{R}^3_x\times \mathbb{R}^3_v, \end{aligned} \right. \end{equation} where the vector field $E[f]$ is given by \begin{equation} E[f](t,x)=c_\alpha \nabla_x \int_{\mathbb{R}^3_{y,v}} \frac{1}{|x-y|^{3-2\alpha}} f(t,y,v)dydv \end{equation} and $\lambda \in \{-1,1\}$. The case $\lambda =1$ describes an attracdtive case that can be observed in galactic settings. When $\lambda=-1$, the system describes a repulsive case that can be seen in a plasma or ion gas. However, we are interested in small data long-time dynamics, so the sign of $\lambda$ does not play an important role. When $\alpha=1$, the system is called Vlasov-Poisson system which was studied by many reseachers.

Let me first describe the Vlasov-Poisson system, $\alpha=1$. Global existence with small initial data was first obtained by Bardos-Degond. Later, the global existence of classical solution with large initial data was independently obtained by Lions-Perthame and Pfaffelmoser. Hence, it is natural to ask its asymptotic behavior. Choi-Ha proved that any nontrivial solution does not have linear scattering, i.e., the nonlinear solution does not behave like linear kinetic transport equation. Instead, Choi-Kwon proved the existence of the modified scattering of the Vlasov-Poisson system, that is, the solution to a nonlinear equation converges to a solution of a linear equation with a suitable correction. In this situation, the solution converges to a function following natural linear trajectory with logarithmic correction. However, their trajector was not explicit. Recenetly, there are several ongoing works on kinetic equations by applying techniques from dispersive equations. In this philosophy, Ionescu-Pausader-Wang-Widmayer gave another proof with a concrete scattering trajectory by using so called $Z$-norm method. Later, Flynn-Ouyang-Pausader-Widmayer gave another proof by using pseudo-conformal symmetry of the equation as well as the Hamiltonian structure of the equation.

However, there are few works when $\alpha \neq 1$. For local well-posedness, Choi-Jeong proved the local existence and uniqueness of solution in a weighted Sobolev space on $\mathbb{R}^3$ when $\alpha>3/8$. Moreover, they proved a finite-time singularity formation of solutions for the gravitational case.

In a joint work with Wenrui Huang [(HK26)](https://link.springer.com/article/10.1007/s00023-026-01735-7), we have obtained global well-posedness with small initial data. Moreover, the solution exhibits scattering when $\alpha \in (1/2,1)$ and modified scattering when $\alpha \in (1,1+\delta)$. Unlike the case $\alpha=1$, the nonlinear solution converges to a solution along the natural linear trajectory with polynomial correction. Moreover, we constructed modified wave operator as well.

In a joint work with Wenrui Huang (HK26b), we included repulsive harmonic potentials into the problem and show linear scattering in $d\geq 3$ and modified scattering in $d=2$. We also constructed modified wave operator as well.

In both papers, we adapted a technique from dispersive equations: using a pseudoconformal symmetry and a bootstrap argument, we can show the desired result. In particular, the recent paper introduced kinetic lens transform which relaxes decay condition appears in the previous papers due to Bigorgne, Velozo Ruiz, and Velozo Ruiz.

Elliptic equations with singular drift terms

For several years, I have devoted to studying unique solvability of linear elliptic equations with singular drift terms in Sobolev spaces. To explain the results, let $\Omega$ be a bounded Lipschitz domain in $\mathbb{R}^n$, $n\geq 2$. We consider the following Dirichlet problems of elliptic equations of second-order: \begin{equation}\tag{$D$} -\mathrm{div}(A \nabla u)+\mathrm{div}(u\mathbf{b})=f\quad \text{in } \Omega\qquad u=0\quad \text{on } \partial\Omega \end{equation} and \begin{equation}\tag{$D’$} -\mathrm{div}(A \nabla v)-\mathbf{b}\cdot \nabla v =g\quad \text{in } \Omega\qquad v=0\quad \text{on } \partial\Omega \end{equation} Here $A=(a^{ij}):\mathbb{R}^n\rightarrow \mathbb{R}^{n\times n}$ denotes an $n\times n$ real matrix-valued function which is uniformly elliptic, that is, there exists $0<\delta<1$ such that \begin{equation} \delta|\xi|^2 \leq a^{ij}(x)\xi_i \xi_j\quad \text{and}\quad |a^{ij}(x)|\leq \delta^{-1}\quad \text{for all } x,\xi \in \mathbb{R}^n, \end{equation} $\mathbf{b}=(b^1,b^2,\dots,b^n) : \Omega\rightarrow \mathbb{R}^n$ is a given vector field. Here we follow the usual summation convention for repeated indices.

When $\mathbf{b}$ is zero or a bounded vector field, $L^p$-theory for these equations is well-known by several authors. One might ask $L^p$-theory for unbounded vector field, $\mathbf{b}\in L^q(\Omega;\mathbb{R}^n)$.

During the master course, I wrote a research article [KK18] with my academic advisor Prof. Hyunseok Kim. At the Republic of Korea Air Force Academy, I wrote two research articles [K20] and [K21].

In a joint work with Hyunseok Kim [KK18], when $A=I$ and $\mathbf{b} \in L^{n}(\Omega;\mathbb{R}^n)$, we proved $L^{\alpha,p}$–solvability for the problems $(D)$ on arbitrary bounded Lipschitz domains $\Omega$ in $\mathbb{R}^n$, $n\geq 3$ for appropriate $(\alpha,p)\in \mathscr{A}\cap \mathscr{B}$. Here $L^{\alpha,p}(\Omega)$ denotes the Bessel potential spaces over $\Omega$, which generalizes the classical Sobolev spaces $W^{k,p}(\Omega)$. Here $\mathscr{A}$ is the set of admissible pairs $(\alpha,p)$ such that the Dirichlet problem for the Poisson equation has a unique solution in $L^{\alpha,p}_0(\Omega) ={ u\in L^{\alpha,p}(\Omega) : u=0\text{ on } \partial\Omega}$. Also, $\mathscr{B}$ denotes the set of some pairs $(\alpha,p)$ for which $\mathbf{b} \cdot \nabla u \in L^{\alpha-2,p}(\Omega)$ for all $u\in L_0^{\alpha,p}(\Omega)$. Similar result also proved for the problem $(D’)$. This result extends the classical result of Jerison-Kenig [JK95]. In the same paper, Neumann problems were also considered.
When $A$ satisfies (1) and $\mathbf{b} \in L^{n,\infty}(\Omega;\mathbb{R}^n)$, $n\geq 2$, $\mathrm{div} \mathbf{b} \geq 0$ in $\Omega$, I proved that if $q>2$, then there exists $\varepsilon>0$ such that if $g\in W^{-1,q}(\Omega)$, then there exists a unique $v\in W_0^{1,2+\varepsilon}(\Omega)$ of the problem $(D’)$. As an application, I proved that if $f\in \bigcap_{q<2} W^{-1,q}(\Omega)$, there exists a unique $u\in \bigcap_{q<2} W_0^{1,q}(\Omega)$ satisfying the problem $(D)$. The result can be found in [K20]. When $n=2$, this result extends Kim-Tsai [KT20] and Chernobai-Shilkin [CS19].
In [K21], I proved that if $2<p<\infty$, $A$ satisfies mean oscillation in small balls and $\mathrm{div} A \in L^{2}(\Omega;\mathbb{R}^2)$, $\Omega$ is a bounded Lipschitz domain in $\mathbb{R}^2$ which has a small Lipschitz constant, and $\mathbf{b} \in L^{2}(\Omega;\mathbb{R}^2)$, then the problem $(D’)$ has a unique solution in $W_0^{1,p}(\Omega)$. A similar result was also proved for the problem $(D)$. Results are new even if $A=I$. The results complement Kim-Kim [KK15] and Kang-Kim [KK17] when $n=2$. Further research should be done to remove an additional assumption on $\mathrm{div} A$.