Dimension of the singular set in the parabolic obstacle problem

This paper establishes that the singular set in the parabolic obstacle problem for general $C^{2,1}$ obstacles has a parabolic Hausdorff dimension of at most $n-1$, extending a previous result limited to constant Laplacian obstacles by combining a truncated parabolic frequency formula with monotonicity estimates and an iterative argument.

The Gist

Here is an explanation of the paper "Dimension of the Singular Set in the Parabolic Obstacle Problem" using simple language and creative analogies.

The Big Picture: The "Muddy Floor" Problem

Imagine you are trying to smooth out a very sticky, muddy floor (this is the obstacle). You have a heavy, flexible rubber sheet (the solution) that you want to lay over the mud.

• The Rules: The sheet must never go under the mud (it can touch it, but not penetrate it). Also, the sheet wants to be as flat and relaxed as possible (minimizing energy), but it's being pulled by gravity or pressure (the heat equation).
• The Result: The sheet will lift off the mud in some areas and stick to it in others. The line where the sheet lifts off is called the Free Boundary.

Usually, this line is smooth and well-behaved, like a clean river bank. However, sometimes the line gets messy, jagged, or forms weird "knots." These messy spots are called Singular Points.

The Question: How "big" or "complex" can these messy knots get? Can they form a whole wall? A whole room? Or are they just thin lines or tiny dots?

The Main Discovery

The authors of this paper, Alejandro Martínez and Xavier Ros-Oton, proved a very specific limit on how messy these knots can get.

They showed that even in the most complicated, time-changing scenarios (where the mud might be shifting or the pressure changing), these messy knots can never form a full 3D object or a thick sheet. They are essentially flat sheets or lines.

In mathematical terms, if you are in a world with $n$ dimensions (like our 3D space plus time), the "messy" part of the boundary has a dimension of at most $n-1$.

• If you are in 3D space + time, the messy part is at most 2D (like a sheet).
• If you are in 2D space + time, the messy part is at most 1D (like a line).

The Challenge: The "Perfect" vs. "Real" World

Before this paper, mathematicians knew this rule was true only in a "perfect" world where the mud was perfectly uniform (mathematically, where the obstacle was a simple parabola with a constant slope).

In the real world, the mud is uneven. The obstacle $\phi$ is "rough" (only $C^{2,1}$ smooth, meaning it has bumps and wiggles).

• The Problem: When the obstacle is rough, the math tools used to measure the "messiness" of the boundary start to break. The usual rulers (monotonicity formulas) give you a slightly crooked reading, making it hard to prove the dimension limit.

The Solution: The "Iterative Ladder"

The authors didn't just fix the ruler; they built a ladder to climb out of the mess. Here is how they did it, step-by-step:

1. The "Truncated Frequency" (The Rough Ruler)

Imagine you are trying to measure how "wiggly" the rubber sheet is near a knot. You use a tool called a Frequency Formula.

• In the perfect world, this tool gives a perfect number.
• In the rough world, the tool gives a number that is almost right, but has a small error.
• The authors created a "Truncated" version of this tool. Think of it like a speedometer that caps out at a certain speed. If the sheet is getting too wild, the tool just says "It's maxed out" rather than giving a wrong number. This helps them get a foothold.

2. The "Iterative Argument" (Climbing the Ladder)

This is the cleverest part of the paper.

• Step 1: They start with a very rough estimate. They prove that the "wiggles" are small enough to get a slightly better measurement.
• Step 2: Because the measurement is better, they can use a slightly more sensitive version of their tool (a higher "truncation" parameter).
• Step 3: This new tool gives an even better measurement, which allows them to use an even more sensitive tool.

It's like climbing a ladder where every rung you stand on gives you the leverage to reach the next, higher rung.

• They start at a low rung (proving the error is small for a specific range).
• They use that success to prove the error is even smaller for a wider range.
• They repeat this process until they have climbed all the way to the top, proving the rule holds for all possible roughness levels.

3. The "Saturated" Frequency

As they climb the ladder, they discover that the "frequency" of the wiggles hits a "saturation point." It's like a balloon that expands until it hits a ceiling. Once it hits that ceiling, it can't get any bigger.

• They proved that for the most complex knots (the "top stratum"), the frequency is always "saturated" at a specific value.
• This saturation is the key that locks the dimension of the messy set to be exactly $n-1$ (or less).

The Final Result: Cleaning Up the Mess

By combining this "ladder" technique with some geometric logic (showing that if the sheet is too messy, it would have to violate the laws of physics), they proved:

No matter how bumpy the obstacle is, the "messy" part of the boundary where the sheet sticks to the mud cannot be a thick, 3D blob. It is always a thin, flat structure (or a line).

Why Does This Matter?

This isn't just about rubber sheets and mud. This math describes:

• Finance: How American stock options are priced (when to exercise the option).
• Physics: How ice melts into water (the Stefan problem).
• Engineering: How materials deform under pressure.

The authors showed that even in the most chaotic, real-world scenarios, nature has a limit on how "jagged" these transitions can be. The universe prefers order, and even in the messiest places, the chaos is confined to a lower dimension.

In short: They took a math problem that was only solved for "perfect" conditions and used a clever, step-by-step "ladder" method to prove it holds true for the messy, real world too.

Technical Summary

Here is a detailed technical summary of the paper "Dimension of the Singular Set in the Parabolic Obstacle Problem" by Alejandro Martínez and Xavier Ros-Oton.

1. Problem Statement

The paper investigates the parabolic obstacle problem, a free boundary problem arising in contexts such as optimal stopping for Brownian motion and the pricing of American options (Black-Scholes model). The problem is formulated as finding a function $v$ satisfying:
$$\begin{cases} \partial_t v - \Delta v = 0 & \text{in } \{v > \phi\} \cap \Omega \times (0, T) \\ v \ge \phi & \text{in } \Omega \times (0, T) \\ \partial_t v - \Delta v \ge 0 & \text{in } \Omega \times (0, T) \end{cases}$$
where $\phi \in C^{2,1}$ is the obstacle. By setting $u = v - \phi$, the problem is localized to:
$$\partial_t u - \Delta u = -f(x) \chi_{\{u>0\}}, \quad u \ge 0, \quad \partial_t u \ge 0,$$
where $f = -\Delta \phi$ is a non-negative smooth function.

The central challenge is understanding the regularity of the free boundary $\partial\{u > 0\}$. The free boundary consists of:

• Regular points: Where the free boundary is locally $C^\infty$.
• Singular points ($\Sigma$): Where the contact set $\{u=0\}$ has zero density. At these points, the solution admits a quadratic expansion $u(x,t) \approx p_2(x)$, where $p_2$ is a homogeneous quadratic polynomial.

The Open Question: While it was known that for a fixed time $t$, the singular set $\Sigma_t$ is contained in an $(n-1)$-dimensional manifold, and recently (in the case $f \equiv 1$) that the space-time singular set has parabolic Hausdorff dimension at most $n-1$, it remained unproven whether this dimension bound holds for general obstacles (i.e., when $f$ is not constant).

2. Main Result

The authors prove the following theorem:

Theorem 1.1: Let $u$ be a bounded solution to the parabolic obstacle problem with $f \in C^{0,1}$ (Lipschitz continuous) and $f > 0$. Let $\Sigma$ be the set of singular points. Then, the parabolic Hausdorff dimension of the singular set satisfies:
$$\dim_{\text{par}}(\Sigma) \le n - 1.$$

This generalizes the recent result by Figalli, Ros-Oton, and Serra (FRS24), which was restricted to the case $f \equiv 1$ (constant obstacle Laplacian).

3. Methodology and Key Techniques

The proof relies on a sophisticated combination of monotonicity formulas, blow-up analysis, and an iterative argument to handle the non-constant obstacle $f$.

A. Classification of Singular Points

Singular points are classified based on the dimension of the zero set of the quadratic polynomial $p_2$ in their expansion:
$$\Sigma = \bigcup_{m=0}^{n-1} \Sigma_m, \quad \text{where } \Sigma_m = \{ (x_0, t_0) \in \Sigma : \dim\{p_2 = 0\} = m \}.$$
The goal is to show $\dim_{\text{par}}(\Sigma_m) \le n-1$ for all $m$.

B. Truncated Frequency Formula

The core tool is a truncated parabolic frequency function $\phi_\gamma(r, w)$, defined for a rescaled solution $w = (u-p)\zeta$ (where $\zeta$ is a cutoff):
$$\phi_\gamma(r, w) = \frac{D(r, w) + \gamma r^{2\gamma}}{H(r, w) + r^{2\gamma}}$$
where $D$ and $H$ are Dirichlet and $L^2$ energy functionals, and $\gamma$ is a parameter.

• The Challenge: In the constant case ($f \equiv 1$), the frequency function is almost-monotone for all $\gamma > 2$. In the general case ($f \in C^{0,1}$), the error terms in the monotonicity formula are larger, and almost-monotonicity is initially only guaranteed for $\gamma \in (2, 5/2)$.
• The Strategy: The authors use an iterative argument. They show that if the frequency is "saturated" (i.e., the limit $\lambda = \gamma$) for a certain range of $\gamma$, the error terms in the expansion of $u$ improve. This improved expansion allows them to extend the range of $\gamma$ for which almost-monotonicity holds, eventually reaching any $\gamma < 3$.

C. Iterative Improvement of Error Terms

1. Initial Step: Using the almost-monotonicity for $\gamma \in (2, 5/2)$, they improve the error term in the expansion of $u$ from $o(|x|^2 + |t|)$ to $o(|x|^{5/2} + |t|^{5/4})$.
2. Iteration: This sharper error term allows the derivation of almost-monotonicity for a larger range of $\gamma$ (up to $11/4$).
3. Limit: By repeating this process, they establish that for any $\gamma \in (2, 3)$, the frequency is saturated ($\lambda = \gamma$). This implies the expansion can be refined to:
   $$u(x_0+x, t_0+t) = p_2(x) + o(|x|^{3-\varepsilon} + |t|^{(3-\varepsilon)/2}).$$

D. Handling the Lipschitz Obstacle

A crucial technical contribution is handling $f \in C^{0,1}$ rather than $f \in C^\infty$. The authors prove a specific regularity result (Proposition 2.7): if a subharmonic function is Lipschitz in $n-1$ directions, it is $C^{0,\alpha}$ in the remaining direction. This allows them to control the error terms generated by the non-constant $f$ without requiring higher regularity.

4. Key Contributions

1. Generalization to Non-Constant Obstacles: The paper removes the restrictive assumption $f \equiv 1$, proving the dimension bound for arbitrary Lipschitz obstacles. This is non-trivial because the lack of constant $f$ introduces significant error terms in the monotonicity formulas.
2. Iterative Saturation Argument: The authors develop a novel iterative scheme to extend the range of the frequency parameter $\gamma$ from $(2, 5/2)$ to $(2, 3)$. This overcomes the limitation that standard monotonicity formulas fail for general $f$ at higher frequencies.
3. Refined Blow-up Analysis: They perform a detailed analysis of the "second blow-up" in the top stratum ($\Sigma_{n-1}$), proving that the frequency limit $\lambda^*$ equals $\gamma$ for all $\gamma \in (2, 3)$. This leads to the optimal expansion rate $o(r^{3-\varepsilon})$.
4. Geometric Measure Theory Application: Using the refined expansion, they prove that the singular set is "cleaned" by a parabolic cone of dimension $n-1$, allowing the application of a GMT lemma to bound the parabolic Hausdorff dimension.

5. Significance

• Completeness of the Theory: This result completes the picture for the parabolic obstacle problem, establishing that the singular set behaves similarly to the elliptic case and the constant-coefficient parabolic case, regardless of the obstacle's smoothness (provided it is Lipschitz).
• Robustness of Methods: The techniques developed, particularly the iterative frequency saturation and the handling of Lipschitz coefficients in monotonicity formulas, are likely to be applicable to other free boundary problems with non-constant coefficients or lower regularity data.
• Optimality: The bound $n-1$ is sharp, as examples exist where the singular set is an $(n-1)$-dimensional manifold (e.g., a hyperplane).

In summary, Martínez and Ros-Oton successfully extend the dimension bound of the singular set to the general parabolic obstacle problem by overcoming the analytical difficulties posed by non-constant obstacles through a refined, iterative monotonicity argument.