Space-time geometry of Hele-Shaw flow

Abstract: We study the dynamic behavior of the free boundary of the Hele-Shaw flow with a nonnegative source, in which the pressure satisfies $-Δp=c$ in ${p&gt;0}$ and the free boundary moves with normal velocity $V=|\nabla p|$. For initial domains satisfying an interior ball condition, we prove that the patch ${p&gt;0}$ expands at a locally positive rate, so that its hitting time is locally Lipschitz; this is optimal, since distinct portions of the patch may collide. We also show that the free boundary is locally either regular, or of collision type, or undergoing parabolic-scale hole closings of various dimensions: such a space-time characterization of singularities is new in Hele-Shaw flow. We further show that, near any time at which no collision occurs, the space-time free boundary is a $C<sup>1$ hypersurface. This implies that at such times every free boundary point has a space-time normal: the space-time unit normal is parallel to the time direction precisely where the spatial free boundary has singular geometry. The results apply in particular to the tumor growth model with nutrients introduced by Perthame, Quirós, and Vázquez, but the analysis of singularities is new even for the classical injection problem.

• The paper proves that the hitting time is locally Lipschitz under quantitative positivity, boundedness, spatial regularity, and time-integrability conditions on the source, improving earlier Hölder estimates.
• It establishes sharp or nearly sharp closing rates for lower-stratum singularities, with parabolic scaling in most dimensions and a logarithmic correction for point singularities in two dimensions, while rescaled zero sets converge to ellipsoidal cylinders.
• It shows that, away from collision-type singularities, the space-time free boundary is a C¹ hypersurface, and applies these results to nutrient-driven tumor growth in dimensions d≤3, including quantitative long-time volume expansion.

Setting and problem

The paper studies the free boundary of the Hele-Shaw flow with a nonnegative source $c(x,t)$, in which the pressure satisfies $-\Delta p = c$ in $\{p>0\}$ and the free boundary moves with normal velocity $V = |\nabla p|$ (Darcy's law). This pressure formulation arises as the patch limit of the constrained transport equation $\partial_t\rho - \nabla\cdot(\rho\nabla p) = c\rho$ with $0\leq\rho\leq1$, and its primary motivating instance is the tumor growth model of Perthame–Quirós–Vázquez, in which the source is a nutrient satisfying a diffusion-consumption equation. The authors' objective is a dynamic characterization of the singularities of the free boundary, complementing the extensive literature on short-time regularity and on stationary profiles via the elliptic obstacle problem.

The analysis proceeds through the Baiocchi transform $w(x,t) = \int_0^t p(x,s)\,ds$, whose positive phase $\Omega_t = \{w(\cdot,t)>0\} = \{T<t\}$ is described by the hitting time $T$. Away from the initial patch, each slice $w(\cdot,t)$ solves an elliptic obstacle problem whose source $-\Delta p = c$0 depends on the hitting time itself — a structural feature distinguishing this problem from the injection and Stefan problems, where the Baiocchi transform yields a homogeneous obstacle problem. Quadratic blowups of $-\Delta p = c$1 are either half-space solutions (regular points) or nonnegative 2-homogeneous polynomials $-\Delta p = c$2 with $-\Delta p = c$3 (singular points); the latter stratify by the dimension $-\Delta p = c$4 of the spine $-\Delta p = c$5, with the top stratum $-\Delta p = c$6 comprising collision-type singularities.

Nondegeneracy and Lipschitz hitting time

Under hypotheses (A1) on the source — uniform positivity, boundedness, and spatial Lipschitz continuity on compact time intervals — (A2), an interior ball condition (IB) on the initial patch, and time regularity $-\Delta p = c$7 (the endpoint $-\Delta p = c$8 of hypothesis S$-\Delta p = c$9), the paper proves that the hitting time is locally Lipschitz in $\{p>0\}$0, with constants depending only on $\{p>0\}$1, $\{p>0\}$2, $\{p>0\}$3, and quantitative norms of $\{p>0\}$4. Since the normal velocity equals $\{p>0\}$5, this is a quantitative lower bound on the boundary speed valid at singular as well as regular points; it resolves in the affirmative the Lipschitz regularity conjectured in prior work of Collins–Jacobs–Kim, where only Hölder continuity was obtained. The Lipschitz bound is optimal: at a collision, $\{p>0\}$6 behaves like the minimum of two smooth functions.

The proof compares the solution with a time-rescaled copy of itself via a reparametrization $\{p>0\}$7 solving an ODE adapted to the temporal decrease of $\{p>0\}$8; the interior ball condition enters solely to establish that the patch covers an $\{p>0\}$9-neighborhood of $V = |\nabla p|$0 within time proportional to $V = |\nabla p|$1, using explicit radial solutions planted in interior balls. The same section establishes linear pressure nondegeneracy at the interface, $V = |\nabla p|$2 for $V = |\nabla p|$3.

Hopf–Lax inequality for the pressure

A tool of independent interest is a Hopf–Lax type inequality: for any two times $V = |\nabla p|$4,

$V = |\nabla p|$5

where $V = |\nabla p|$6 encodes the relative spatial oscillation $V = |\nabla p|$7 and the accumulated time decrease $V = |\nabla p|$8 of the source. Formally this integrates the Hamilton–Jacobi inequality $V = |\nabla p|$9. The inequality improves on the version previously derived through porous medium approximation and Aronson–Bénilan estimates by carrying no additive error term and requiring less time regularity; it holds under S$\partial_t\rho - \nabla\cdot(\rho\nabla p) = c\rho$0 for any $\partial_t\rho - \nabla\cdot(\rho\nabla p) = c\rho$1, without interior ball assumptions. The proof is a comparison argument against a barrier built from a single time slice of the pressure, translated at constant velocity while subtracting a growing constant; completing the square shows the free boundary condition holds identically for the barrier. The main technical difficulty is the low regularity of this barrier, handled by mollification and by restricting initially to "admissible" time slices, which form a set of full measure.

As a corollary, combined with annular barrier arguments, the inequality yields Hölder continuity of the hitting time under S$\partial_t\rho - \nabla\cdot(\rho\nabla p) = c\rho$2 alone, with no geometric assumption on the initial patch, at an explicit exponent $\partial_t\rho - \nabla\cdot(\rho\nabla p) = c\rho$3 that is attained when the time integrability is strengthened to some $\partial_t\rho - \nabla\cdot(\rho\nabla p) = c\rho$4.

Closing rates and asymptotic shape at lower-stratum singularities

The core contribution concerns a singular point $\partial_t\rho - \nabla\cdot(\rho\nabla p) = c\rho$5 with $\partial_t\rho - \nabla\cdot(\rho\nabla p) = c\rho$6. Measuring the transverse radius $\partial_t\rho - \nabla\cdot(\rho\nabla p) = c\rho$7 of the zero set outside a thin cone around the spine $\partial_t\rho - \nabla\cdot(\rho\nabla p) = c\rho$8, the paper proves:

• Sharp rates for point singularities ($\partial_t\rho - \nabla\cdot(\rho\nabla p) = c\rho$9): $0\leq\rho\leq1$0 for $0\leq\rho\leq1$1, and $0\leq\rho\leq1$2 in $0\leq\rho\leq1$3, the logarithm reflecting the Green's function behavior.
• Intermediate strata ($0\leq\rho\leq1$4): $0\leq\rho\leq1$5, determining the exact exponent though not the full asymptotics.
• Asymptotic shape: rescaled zero sets converge locally in Hausdorff distance to a cylinder $0\leq\rho\leq1$6 over a $0\leq\rho\leq1$7-dimensional ellipsoid $0\leq\rho\leq1$8 determined solely by the blowup polynomial $0\leq\rho\leq1$9.

Thus every singularity below the top stratum is a hole closing at essentially the parabolic rate of the explicit cylindrical solution. The upper bounds follow from Taylor expansion of $w(x,t) = \int_0^t p(x,s)\,ds$0 against the blowup profile; the lower bounds combine static pressure barriers on conical annuli — constructed by separating radial and angular variables via a Sturm–Liouville problem whose principal eigenvalue $w(x,t) = \int_0^t p(x,s)\,ds$1 degenerates as $w(x,t) = \int_0^t p(x,s)\,ds$2 when $w(x,t) = \int_0^t p(x,s)\,ds$3, a degeneration available precisely because $w(x,t) = \int_0^t p(x,s)\,ds$4 — with the Hopf–Lax inequality preventing overly fast closing. The $w(x,t) = \int_0^t p(x,s)\,ds$5 upper bound requires a separate comparison against an explicit radial annular solution tracking the extinction of an inner hole.

Two consequences quantify the divergence of the velocity. First, $w(x,t) = \int_0^t p(x,s)\,ds$6 near the singularity for every $w(x,t) = \int_0^t p(x,s)\,ds$7, improving to $w(x,t) = \int_0^t p(x,s)\,ds$8 when $w(x,t) = \int_0^t p(x,s)\,ds$9 has positive lower $\Omega_t = \{w(\cdot,t)>0\} = \{T<t\}$0-density; taking $\Omega_t = \{w(\cdot,t)>0\} = \{T<t\}$1 shows the known global $\Omega_t = \{w(\cdot,t)>0\} = \{T<t\}$2 estimate is sharp. Second, the hitting time satisfies $\Omega_t = \{w(\cdot,t)>0\} = \{T<t\}$3 at scale $\Omega_t = \{w(\cdot,t)>0\} = \{T<t\}$4 for every $\Omega_t = \{w(\cdot,t)>0\} = \{T<t\}$5: since $\Omega_t = \{w(\cdot,t)>0\} = \{T<t\}$6, the free boundary sweeps a ball of radius $\Omega_t = \{w(\cdot,t)>0\} = \{T<t\}$7 around a lower-stratum singularity in time $\Omega_t = \{w(\cdot,t)>0\} = \{T<t\}$8, i.e., with diverging speed.

$\Omega_t = \{w(\cdot,t)>0\} = \{T<t\}$9 regularity of the space-time free boundary away from collisions

Combining the vanishing gradient at singular points with the classical formula $T$0 at regular points, the paper obtains its structural result: if $T$1, then $T$2 in a neighborhood of $T$3, and the space-time free boundary is a $T$4 hypersurface whose singular points are exactly those where the space-time normal is parallel to the time direction — equivalently, where the patch expands with infinite speed.

The local ingredient is an intrinsic-window gradient estimate $T$5 on $T$6, proved by comparing $T$7 with directional derivatives of $T$8 in the spirit of Caffarelli's barrier method for the Stefan problem. Two difficulties specific to this setting arise: the lack of time regularity of $T$9 forces separate arguments before $w(\cdot,t)$0 (via the closing rate) and after (via slice comparison), and the dependence of the obstacle-problem source $w(\cdot,t)$1 on $w(\cdot,t)$2 prevents the use of fixed quadratic correctors; the authors instead introduce Green's function correctors normalized so that they vanish at the base point, with a bootstrap simultaneously widening the intrinsic time window and improving the modulus.

The global ingredient addresses the fact that collisions act nonlocally: a merger elsewhere can instantaneously change $w(\cdot,t)$3 at far-away regular boundary points. The paper shows that lower-stratum singular sets have zero Sobolev 2-capacity, being contained in countably many $w(\cdot,t)$4 manifolds of dimension at most $w(\cdot,t)$5, and hence cannot disturb the Dirichlet problem discontinuously. Concretely, absence of the top stratum implies Mosco convergence of the spaces $w(\cdot,t)$6 as $w(\cdot,t)$7; since $w(\cdot,t)$8 minimizes the Dirichlet energy over $w(\cdot,t)$9, Mosco stability gives strong $-\Delta p = c$00 continuity in time, and elliptic estimates up to the regular boundary upgrade this to local uniform convergence of $-\Delta p = c$01. A notable contrast emerges with the Stefan problems: in the supercooled problem every singularity, collisions included, forms with infinite speed and $-\Delta p = c$02 is globally $-\Delta p = c$03; in the classical problem the failure of $-\Delta p = c$04 at the top stratum is local; in Hele-Shaw flow a single collision can affect $-\Delta p = c$05 globally in space, and the theorem identifies collisions as the only mechanism for such failure.

Application to tumor growth

For the nutrient model in dimensions $-\Delta p = c$06, with patch initial data, bounded initial tumor satisfying the interior ball condition, and $-\Delta p = c$07 strictly positive, the paper verifies all standing hypotheses. The key step is deriving $-\Delta p = c$08 from the Duhamel representation of the nutrient equation, using the sharp $-\Delta p = c$09 estimate for $-\Delta p = c$10 at full strength inside heat semigroup smoothing; the resulting time singularity $-\Delta p = c$11 is integrable exactly when $-\Delta p = c$12, which explains the dimensional restriction. All preceding conclusions therefore apply to the tumor patch, including the Lipschitz hitting time, which in turn supplies a hypothesis needed to conclude $-\Delta p = c$13 for almost every $-\Delta p = c$14.

On large scales, the paper proves the volume lower bound $-\Delta p = c$15 for large $-\Delta p = c$16, via a replenishment argument showing that nutrient diffusing into the tumor region sustains geometric growth of the cumulative mass. The exponent $-\Delta p = c$17 matches the radial scaling of the nutrient-limited regime, and the estimate shows the tumor region cannot remain bounded — contrasting with the zero-nutrient-diffusion variant, where boundedness or exponential growth may occur.

Limitations and open questions

Several restrictions are acknowledged explicitly. The dimensional restriction $-\Delta p = c$18 for the tumor application is tied to the integrability of the heat-kernel singularity in the $-\Delta p = c$19 estimate; whether the conclusions hold in higher dimensions is open. Sign-changing sources break the correspondence between the transport equation and the pressure formulation altogether, and free-boundary regularity in that regime remains largely unresolved. For the injection problem, which does not satisfy (A1), the authors argue heuristically that their results extend — indicating the necessary modifications via comparison estimates, Harnack chains, and boundary Harnack arguments — but the full verification is left open. Most substantively, the finite-time analysis does not determine the geometry of the eventual tumor region $-\Delta p = c$20: whether thin-finger growth persists and what the asymptotic shape of $-\Delta p = c$21 is as $-\Delta p = c$22 remain open, as does the question of whether $-\Delta p = c$23 exhausts the complement of the initial patch in the tumor model (known to hold for the injection problem). The intermediate-stratum rates leave the precise second-order asymptotics of $-\Delta p = c$24 undetermined.

Conclusion

This paper delivers a complete space-time classification of free-boundary singularities for Hele-Shaw flow with a nonnegative source: the hitting time is locally Lipschitz (optimally), every non-collisional singularity is a parabolic-scale hole closing toward a cylinder over an ellipsoid determined by the blowup profile, and away from collision times the space-time free boundary is a $-\Delta p = c$25 hypersurface whose sole singularities are points of infinite normal speed. These results apply to the tumor growth model with nutrients in physical dimensions and constitute new contributions even for the classical injection problem, anchored by a Hopf–Lax inequality for the pressure proved by a direct comparison argument.