Free boundary regularity and well-posedness of physical solutions to the supercooled Stefan problem

1. Problem Statement

The paper investigates the supercooled Stefan problem, a mathematical model describing the freezing of supercooled water into ice. In the one-phase setting, the temperature in the solid phase is held constant at 0, while the liquid phase temperature $u(x,t)$ evolves according to the heat equation. The free boundary $\Lambda_t$ represents the interface between the liquid and solid.

The system is given by:
$$\begin{cases} u_t - \frac{1}{2}u_{xx} = 0 & t > 0, \, x > \Lambda_t \\ \dot{\Lambda}_t = \frac{\alpha}{2} u_x(\Lambda_t, t) & t > 0 \\ u(x, 0) = u_0(x), \quad u(\Lambda_t, t) = 0 & x \ge 0, \, t \ge 0 \end{cases}$$
where $\alpha > 0$ and $u_0$ is a probability density.

Key Challenges:

• Ill-posedness: Unlike the classical Stefan problem, the supercooled version is generally ill-posed. Classical solutions can blow up in finite time, leading to discontinuities in the free boundary $\Lambda_t$ and instantaneous temperature spikes.
• Jump Discontinuities: The free boundary $\Lambda_t$ is known to exhibit jump discontinuities. The size of these jumps is not uniquely determined by the classical formulation.
• Physical Solutions: To resolve non-uniqueness, the concept of a physical solution is used, defined via a probabilistic formulation (McKean-Vlasov equation) where the jump size is the minimal possible value satisfying the conservation of mass.
• Open Questions: Previous works established existence and local uniqueness under restrictive assumptions (e.g., bounded initial data, finite monotonicity changes). Open questions remained regarding:
  1. Whether jump times can accumulate (i.e., can there be infinitely many jumps in a finite time interval?).
  2. The global regularity of the free boundary (specifically, is it $C^1$ or smoother?).
  3. Global uniqueness for general initial data, particularly those with high oscillations near the origin.

2. Methodology

The author employs a combination of probabilistic analysis, PDE estimates, and free boundary regularity theory.

A. Probabilistic Reformulation and Obstacle Problem:

• The problem is reformulated using a system of Brownian particles interacting via a moving boundary.
• A key transformation is introduced: defining the potential $w(x,t) = \int_t^\infty u(x,s) ds$.
• This potential $w$ satisfies a weighted parabolic obstacle problem:
  $$w_t - \frac{1}{2}w_{xx} = -\nu(x)\chi_{\{w>0\}}$$
  where $\nu(x)$ is the density of stopped particles. This transformation allows the application of established regularity theory for obstacle problems.

B. A Priori Estimates:

• Second-order estimates: The author derives bounds on $u_t$ and $u_{xx}$ (specifically $(u_t)^+$ and $u_{xx}\chi_{u>0}$) for any probabilistic solution. These are inspired by recent results for the classical Stefan problem but adapted to handle the discontinuous source term.
• Nondegeneracy: A linear nondegeneracy estimate is proven for physical solutions: $u(x,t) \ge c(x - \Lambda_t)_+$. This implies the free boundary moves with a strictly positive velocity where continuous and provides a lower bound on the freezing time $s(x) = \inf\{t : \Lambda_t > x\}$.

C. Backward Propagation of Oscillation:

• A novel technique is developed to relate future jumps to past behavior.
• Lemma 4.6: The author constructs continuous curves along which the spatial derivative $u_x$ maintains a constant sign, propagating backward from a future time $t_1$ to an earlier time $t_0$.
• Lemma 4.7: It is shown that if a jump occurs at time $t^*$, the physical jump condition forces an oscillation (a change in sign of $u_x$) near the boundary at $t^*$. This oscillation can be propagated backward to $t_0$.
• Since the number of oscillations at any fixed time $t_0 > 0$ is finite (due to local monotonicity properties), this mechanism limits the number of future jumps, preventing accumulation.

D. Blow-up Analysis:

• Near free boundary points (excluding jump intervals), the author performs a blow-up analysis on the potential $w$.
• This classifies boundary points into regular (where $u$ vanishes continuously) and singular (where $u$ jumps from the critical value $\alpha^{-1}$ to 0, or endpoints of jumps).

3. Key Contributions and Results

1. Non-Accumulation of Jumps (Theorem 1.3)

• Result: The set of discontinuities (jump times) of the physical solution $\Lambda_t$ is locally finite in $(0, \infty)$.
• Significance: This resolves an open question from [8]. It proves that while jumps can occur, they cannot accumulate at any positive time. Oscillations in the initial data are resolved instantly, and jumps can only accumulate at $t=0$.

2. Regularity of the Free Boundary (Theorems 1.4 & 1.6)

• Result: The freezing time function $s(x)$ (the inverse of $\Lambda_t$) is of class $C^1$ on the entire domain $(\Lambda_0, \alpha)$.
• Smoothness: $s(x)$ is $C^\infty$ on an open set $\mathcal{R}$ whose complement is countable.
• Classification: The free boundary points are fully classified:
  • Regular Points: Where $u$ vanishes continuously or points strictly inside a jump interval. Here, the boundary is smooth.
  • Singular Points: Endpoints of jumps or points where $u$ drops from $\alpha^{-1}$ to 0. These form a countable set.
• Significance: This provides a complete description of the free boundary's structure, showing that despite the potential for jumps, the solution is remarkably smooth almost everywhere. It improves upon previous results by removing assumptions on the initial data (boundedness, monotonicity) and providing two-sided regularity (not just one-sided).

3. Global Uniqueness (Theorem 1.7 & Corollaries)

• Result: Local uniqueness implies global uniqueness. If two physical solutions agree on $[0, t_0]$ for any $t_0 > 0$, they are identical globally.
• Mechanism: The paper proves that for any $t > 0$, the solution $u(\cdot, t-)$ automatically satisfies the "finite monotonicity change" condition required by previous uniqueness theorems (e.g., [8]), even if the initial data $u_0$ is highly oscillatory.
• Applications:
  • Corollary 1.8: Establishes global uniqueness for initial data that are not too flat near $x=0$ (specifically $u_0(x) \le \alpha^{-1} - cx^n$).
  • Corollary 1.9: Establishes global uniqueness for a specific family of highly oscillatory initial densities (a critical case of [29]), answering an open question regarding their global uniqueness.

4. Significance and Impact

• Resolution of Open Problems: The paper definitively answers two major open questions regarding the supercooled Stefan problem: the non-accumulation of jumps and the global uniqueness of solutions for general initial data.
• Methodological Advance: The introduction of the "backward propagation of oscillation" property is a significant technical innovation. It provides a powerful tool to control the dynamics of the free boundary by linking future singular events to past regularity.
• Generality: The results hold for very general initial data (any probability density with finite moments), removing the restrictive assumptions (boundedness, finite monotonicity) required in previous literature.
• Regularity Theory: The proof that the free boundary is $C^1$ globally and $C^\infty$ outside a countable set represents a major step forward in the regularity theory of free boundary problems with discontinuous sources.
• Interdisciplinary Relevance: Since the supercooled Stefan problem models phenomena in financial mathematics (contagion in markets) and neuroscience (integrate-and-fire models), these well-posedness and regularity results provide a rigorous foundation for analyzing these systems in broader contexts.

In summary, Sebastian Munoz's work establishes a robust well-posedness theory for the supercooled Stefan problem, proving that physical solutions are unique and possess a highly regular free boundary structure, provided one looks at times $t > 0$.