---
title: Two-Phase Elliptic Measure
url: https://www.emergentmind.com/topics/two-phase-elliptic-measure
type: topic
---

# Two-Phase Elliptic Measure

Two-phase elliptic measure concerns a pair of complementary domains, typically denoted $\Omega^+$ and $\Omega^-$, together with the corresponding interior and exterior harmonic or elliptic measures $\omega^+$ and $\omega^-$. The central datum is the Radon–Nikodym derivative
$$
h(Q)=\frac{d\omega^-}{d\omega^+}(Q),\qquad Q\in \partial\Omega,
$$
or, more generally, the mutual absolute continuity of $\omega^+$ and $\omega^-$. In the harmonic case this leads to a two-phase free-boundary problem with transmission condition on the common boundary $\Gamma=\partial\Omega$; in the variable-coefficient setting it leads to structural questions for the boundary via blow-ups, tangent measures, and geometric measure theory [1409.4460] [1904.00751] [2509.04189].

## 1. Core setting and principal objects

In the basic harmonic-measure formulation, $\Omega^+\subset \mathbb R^n$ is an open 2-sided non-tangentially accessible (NTA) domain, and
$$
\Omega^-:=\operatorname{Int}(\mathbb R^n\setminus \overline{\Omega^+}),
$$
with $\partial\Omega$ as their common boundary. One writes $\omega^+$ and $\omega^-$ for the harmonic measures on $\partial\Omega$ seen from $\Omega^+$ and $\Omega^-$, assumes mutual absolute continuity $\omega^-\ll \omega^+\ll \omega^-$, and defines the Poisson-kernel ratio $h=d\omega^-/d\omega^+$ [1409.4460]. In the lecture-note formulation of Engelstein’s work, the Green’s functions $u^+$ and $u^-$ are taken with poles at $\infty$, equivalently on unbounded NTA domains, and the boundary condition is expressed in transmission form as
$$
\partial_\nu u^+(Q)=h(Q)\,\partial_\nu u^-(Q),\qquad Q\in \Gamma,
$$
where $\partial_\nu$ denotes the normal derivative pointing into $\Omega^+$ [1409.4460].

The geometric hypotheses are formulated in terms of NTA connectivity and flatness. For $\Omega^+\subset \mathbb R^{n+1}$, the NTA condition is given by the corkscrew condition and the Harnack-chain condition in the sense of Jerison–Kenig; if $\Omega^-=\mathbb R^{n+1}\setminus \overline{\Omega^+}$ is also NTA, one has a two-sided NTA pair $(\Omega^+,\Omega^-)$ [1904.00751]. Reifenberg flatness is quantified by
$$
D_E(x,r,L):=r^{-1}\max\Big\{\sup_{y\in E\cap B(x,r)}\operatorname{dist}(y,L),\sup_{y\in L\cap B(x,r)}\operatorname{dist}(y,E)\Big\},
$$
with $D_E(x,r)=\inf_L D_E(x,r,L)$, and vanishing Reifenberg flatness means
$$
\lim_{r\to 0}\sup_{x\in E}D_E(x,r)=0
$$
[1904.00751].

In the elliptic setting, one replaces the Laplacian by divergence-form operators
$$
L_{A^+}u=-\operatorname{div}(A^+\nabla u)\quad\text{in }\Omega^+,\qquad
L_{A^-}u=-\operatorname{div}(A^-\nabla u)\quad\text{in }\Omega^-,
$$
where $A^\pm:\mathbb R^n\to \mathbb R^{n\times n}$ are uniformly elliptic, symmetric, matrix-valued coefficients, each 2-quasicontinuous, satisfying
$$
\Lambda_0^{-1}|\xi|^2\le \langle A^\pm(x)\xi,\xi\rangle \le \Lambda_0|\xi|^2
$$
for almost every $x$ and all $\xi\in\mathbb R^n$ [2509.04189]. For fixed poles $x^\pm\in \Omega^\pm$, the corresponding elliptic measures $\omega^\pm$ are supported on $\partial\Omega$ and solve the Dirichlet problem via the Riesz representation [2509.04189].

## 2. Two-phase free-boundary formulation for harmonic measure

The sharp regularity theory in the harmonic case is organized around the assumption
$$
\log h\in C^{k,\alpha}(\partial\Omega),
$$
with $k\ge 0$ and $\alpha\in (0,1)$ [1409.4460]. The main theorem stated in Engelstein’s exposition is that if $\Omega^+\subset \mathbb R^n$ is a 2-sided NTA domain, or a Lipschitz domain, and $\log h\in C^{k,\alpha}(\partial\Omega)$, then:

1. if $n=2$, $\Gamma$ is locally given by the graph of a $C^{k+1,\alpha}$ function;
2. if $n\ge 3$, then under an a priori small Reifenberg-flatness or Lipschitz assumption, the same conclusion holds.

Moreover, if $\log h\in C^\infty$ then $\Gamma$ is $C^\infty$, and if $\log h$ is real-analytic then $\Gamma$ is analytic. In the basic case $\log h\in C^{0,\alpha}$ one obtains $\Gamma\in C^{1,\alpha}$ [1409.4460].

A central feature of this theory is that there is no a priori non-degeneracy in the free boundary condition. The regularity argument therefore does not begin from a standard quantitative separation of phases; instead, non-degeneracy must be established from monotonicity formulae [1409.4460]. This distinguishes the two-phase harmonic-measure problem from one-phase free-boundary results of Alt–Caffarelli, Jerison, and De Silva–Ferrari–Salsa, which are described in the source exposition as requiring an a priori non-degeneracy in the free-boundary condition [1409.4460].

This formulation places the Poisson-kernel ratio at the center of the analysis. The boundary regularity of $\Gamma$ is not prescribed directly; rather, it is recovered from the regularity of $\log h$ and the structure of the harmonic measures themselves. A plausible implication is that two-phase elliptic measure serves simultaneously as boundary data and as a geometric probe of the interface.

## 3. Monotonicity, blow-ups, and quantitative regularity

The basic harmonic-measure argument uses three nearly-monotone functionals for
$$
v(x):=h(Q)u^+(x)-u^-(x).
$$
The first is Almgren’s frequency,
$$
N(r;v):=\frac{r\int_{B_r}|\nabla v|^2}{\int_{\partial B_r} v^2},
$$
when $v(0)=0$. For a truly harmonic $v$, one has $N'(r)\ge 0$; in the two-phase setting $v$ is “almost harmonic,” and the source states that
$$
N'(r)\ge -Cr^{\alpha/2-1},
$$
which is integrable at $0$, so that $N(r)\to 1$ as $r\to 0$ [1409.4460].

The second is the Alt–Caffarelli–Friedman functional
$$
J(r;v):=\frac{1}{r^4}\Big(\int_{B_r}|\nabla v^+|^2\Big)\Big(\int_{B_r}|\nabla v^-|^2\Big).
$$
In the truly harmonic two-phase setting $J(r)$ is monotone; in the present setting it is “almost” monotone, and this yields the pointwise lower bound
$$
\lim_{r\to 0} r^{1-n}\omega^+(B(Q,r))>0,\qquad \forall Q\in \Gamma
$$
[1409.4460]. The third is Monneau’s functional
$$
M(r;v,p):=\frac{1}{r^{n+1}}\int_{\partial B_r}|v-p|^2,
$$
with $p$ a 1-homogeneous polynomial. By relating $M'$ to $N-1$, one deduces that $M(0;p)$ exists; together with $J(0)>0$, this forces uniqueness of the tangent half-plane $p$ and rules out degenerate free boundary behaviour [1409.4460].

These monotonicity arguments yield the radial density
$$
\Theta(Q):=\lim_{r\to 0} r^{1-n}\omega^+(B(Q,r)),
$$
which exists and satisfies $0<\Theta(Q)<\infty$ for each $Q\in \Gamma$. Consequently, $\Gamma$ can be decomposed into a regular set where the tangent is a plane and a negligible remainder [1409.4460].

The proof strategy then proceeds through blow-ups and compactness. After obtaining an initial Lipschitz bound for $u:=u^+-u^-$ from the ACF functional, one fixes $Q\in \Gamma$ and rescales
$$
v_r(x):=\frac{v(Q+rx)}{r^{n-2}},
$$
extracting a subsequence with $v_{r_j}\to p$ in $C^0_{\mathrm{loc}}$, where $p$ is a 1-homogeneous harmonic polynomial with $\{p=0\}$ a hyperplane. The almost-monotonicity of $N$ and Monneau’s functional shows that $p$ is unique, so $\Gamma$ has a unique tangent plane at every $Q$ [1409.4460].

The passage from $C^1$ to $C^{1,\alpha}$ uses an improvement-of-flatness scheme. Once $\Theta(Q)>0$ and the unique tangent plane $P(Q)$ are known, the boundary lies within $\varepsilon r$ of $P(Q)$ in every sufficiently small ball $B_r(Q)$, with $\varepsilon\to 0$ uniformly in $Q$; this implies that $\Gamma$ is locally a $C^1$ graph [1409.4460]. The one-sided and two-sided Harnack lemmata then allow one to push a gap from the boundary of a ball into its center when the solution is trapped between rotated two-plane solutions. In the key iterative lemma, if in $B_1(0)$ one has
$$
U_\varepsilon^-(x\cdot \nu-\varepsilon)\le u(x)\le U_\varepsilon^+(x\cdot \nu+\varepsilon),
$$
and $\log h$ varies by at most $\varepsilon$, then for any $\rho\ll 1$ one finds a new direction $\nu'$, a new height $y'$ with $y'\sim y$, and a smaller error $\varepsilon'\ll \varepsilon$ so that
$$
U_{\varepsilon'}^-(x\cdot \nu'-\varepsilon')\le u(x)\le U_{\varepsilon'}^+(x\cdot \nu'+\varepsilon')
\quad\text{in }B_\rho.
$$
Iterating yields $C^{1,s}$ regularity for some $s>0$ depending only on dimension and $\alpha$, and a second iteration with the sharp Hölder control on $\log h$ yields exactly $C^{1,\alpha}$ [1409.4460].

Higher regularity is obtained by a partial hodograph transform in each phase $\Omega^\pm$, which converts the free-boundary PDE into a fully elliptic system for two height-functions $\psi^\pm$ over the common boundary. The system is checked to be elliptic with coercive boundary conditions, and a nonlinear Schauder theory of Agmon–Douglis–Nirenberg / Kinderlehrer–Stampacchia promotes $\psi^\pm$ from $C^{m,\beta}$ to $C^{m+1,\beta}$, yielding $C^{k+1,\alpha}$ regularity of $\Gamma$ by iteration [1409.4460].

## 4. VMO regularity and geometric equivalences

The VMO regime replaces pointwise Hölder control of $\log h$ by vanishing mean oscillation relative to harmonic measure. For a doubling measure $\mu$, the source defines $\mathrm{BMO}(\mu)$ by
$$
\|f\|_{\mathrm{BMO}(\mu)}:=\sup_B \mu(B)^{-1}\int_B |f-f_B|\,d\mu,
$$
and $f\in \mathrm{VMO}(\mu)$ precisely when
$$
\lim_{r\to 0}\sup_x \mu(B(x,r))^{-1}\int_{B(x,r)}|f-f_{B(x,r)}|\,d\mu =0
$$
[1904.00751]. In the two-phase setting one asks that
$$
\log \frac{d\omega^-}{d\omega^+}\in \mathrm{VMO}(\omega^+).
$$

Prats and Tolsa prove a two-phase analogue of the Kenig–Toro one-phase characterization. If $\Omega^+\subset \mathbb R^{n+1}$ is a bounded NTA domain, $\Omega^-=\mathbb R^{n+1}\setminus \overline{\Omega^+}$ is also NTA, and $\Omega^+$ is $\delta$-Reifenberg flat for $\delta$ sufficiently small, then the following are equivalent:

- $\omega^+$ and $\omega^-$ are mutually absolutely continuous and
  $$
  \log\frac{d\omega^-}{d\omega^+}\in \mathrm{VMO}(\omega^+);
  $$
- $\Omega^+$ is vanishing Reifenberg flat, the inner unit normal $N$ to $\partial\Omega^+$ exists $\omega^+$-a.e. and belongs to $\mathrm{VMO}(\omega^+)$, and both Radon–Nikodym derivatives satisfy a reverse Hölder inequality of exponent $3/2$;
- $\Omega^+$ is vanishing Reifenberg flat, $N\in \mathrm{VMO}(\omega^+)$, and either $d\omega^-/d\omega^+\in RH_{3/2}(\omega^+)$ or $d\omega^+/d\omega^-\in RH_{3/2}(\omega^-)$;
- $\Omega^+$ is vanishing Reifenberg flat, $\Omega^+$ and $\Omega^-$ have joint big pieces of chord-arc subdomains, and
  $$
  \lim_{r\to0}\sup_{x\in\partial\Omega^+}\frac{1}{r^n}\int_{B(x,r)\cap \partial\Omega^+}\bigl|N(y)-N_{B(x,r)}\bigr|^2\,d\sigma(y)=0
  $$
  where $N_{B(x,r)}$ is the unit normal to the best-approximating plane over $B(x,r)$ [1904.00751].

The proof uses several distinct ingredients. In the implication $(a)\Rightarrow (b)$, the mutual absolute continuity of $\omega^\pm$ and the capacity-density condition yield rectifiability and tangent $n$-planes $\omega^+$-a.e.; jump formulas for the Riesz transform relate boundary integrals of the normal $N$ to non-tangential limits of singular integrals against $\omega^\pm$; a Christ–David stopping-time cube decomposition isolates good regions where the blow-up of $\log(d\omega^-/d\omega^+)$ is small; and Carleson-measure or square-function estimates, combined with the John–Nirenberg/VMO condition, produce the vanishing oscillation estimate for $N$ [1904.00751].

The implication $(c)\Rightarrow (a)$ proceeds through approximating chord-arc subdomains obtained by bumping $\partial\Omega^+$ up and down on a Whitney-type covering of the bad set; on each approximator, the one-phase Kenig–Toro theorem gives that the logarithm of its Poisson kernel belongs to VMO of surface measure, and the estimate is transferred back by the maximum principle for harmonic measures [1904.00751]. This equivalence makes explicit that VMO regularity in the two-phase logarithm encodes both vanishing flatness and quantitative geometric approximation by chord-arc pieces.

## 5. Multi-operator two-phase elliptic measure

The multi-operator theory replaces a single operator by a pair of divergence-form operators with distinct coefficients on the two sides of the boundary. Let
$$
F_1=\{p\in \partial\Omega: 0< d\omega^-/d\omega^+(p)<\infty\},
$$
so that on $F_1$ the two measures are mutually absolutely continuous [2509.04189]. The main structural theorem states that if $\Omega^\pm$ are complementary NTA domains and $A^\pm$ are 2-quasicontinuous uniformly elliptic symmetric matrices, then there exists $F^*\subset F_1$ of full $\omega^\pm$-measure and a decomposition
$$
\partial\Omega = F^*\sqcup S\sqcup N
$$
such that:

1. for $p\in F^*$,
   $$
   \operatorname{Tan}(\omega^\pm,p)\subset \{c\,\mathcal H^{n-1}\!\llcorner \pi:\ \pi\text{ is an }(n-1)\text{-plane}\},
   $$
   one has $\omega^+\ll \omega^-\ll \omega^+$ on $F^*$, and $\dim_H F^*\le n-1$;
2. on $S=F_2\cup F_3$, one has $\omega^+\perp \omega^-$;
3. $\omega^\pm(N)=0$ [2509.04189].

Here $F_2$ consists of the points where $d\omega^-/d\omega^+=\infty$, and $F_3$ those where the Radon–Nikodym derivative vanishes [2509.04189]. The theorem therefore separates the boundary into a full-measure set with flat tangent measures, a singular part where the two elliptic measures are mutually singular, and a null remainder.

The blow-up construction is explicit. For $p\in F^*$ and radii $r_i\downarrow 0$, one rescales
$$
\Omega_i^\pm = (\Omega^\pm-p)/r_i,\qquad
u_i^\pm(x)=\frac{r_i^{n-2}u^\pm(p+r_i x)}{\omega^\pm(B(p,r_i))},\qquad
\omega_i^\pm(E)=\frac{\omega^\pm(p+r_iE)}{\omega^\pm(B(p,r_i))},
$$
where $u^\pm=G(\cdot,x^\pm)$ are the Green’s functions with fixed poles $x^\pm$ [2509.04189]. After passing to a subsequence, one obtains locally Hausdorff convergence of the rescaled domains, local uniform convergence of $u_i^\pm$, and weak-* convergence of $\omega_i^\pm$ to a common limit measure $\omega_\infty$ [2509.04189].

In the limit, $\Omega_\infty^\pm$ are unbounded complementary NTA domains with common boundary $\Sigma=\partial\Omega_\infty$, $\omega_\infty$ is their harmonic measure with pole at $\infty$, and
$$
u:=u_\infty^+-u_\infty^-
$$
is a global $W^{1,2}_{\mathrm{loc}}$ solution of
$$
L^A_p u := -\operatorname{div}(A(u)\nabla u)=0 \quad\text{in }\mathbb R^n,
$$
with
$$
A(u)(x)=
\begin{cases}
A^+(p), & x\in \{u>0\},\\
A^-(p), & x\in \{u\le 0\}.
\end{cases}
$$
Equivalently, after the linear change of variable $\Lambda(p)=\sqrt{A^+(p)}$, one reduces to
$$
\Delta u^+=0 \text{ in }\{u>0\},\qquad
\operatorname{div}(M_p\nabla u^-)=0 \text{ in }\{u<0\},
$$
where
$$
M_p=\Lambda(p)^{-1}A^-(p)\Lambda(p)^{-1},
$$
with $u=0$ on $\partial\{u>0\}$ and a matching normal-flux condition that becomes a two-plane-solution jump inequality in the viscosity formulation [2509.04189].

This reduction shows that multi-operator two-phase elliptic measure is controlled by a corresponding multi-operator free-boundary problem. The geometric content of the structural theorem is therefore mediated by the rigidity of the free-boundary blow-ups.

## 6. Tangent measures, rigidity, and unresolved questions

The tangent-measure framework follows Preiss. A nonzero Radon measure $\nu$ is a tangent measure $\nu\in \operatorname{Tan}(\mu,p)$ if for some $r_i\downarrow 0$ and $c_i>0$,
$$
c_i\,T_{p,r_i}[\mu]\rightharpoonup \nu
$$
weak-*, where $T_{p,r}(x)=(x-p)/r$ [2509.04189]. A $d$-cone is a family of measures closed under multiplication by positive constants and dilations, and Preiss’s connectedness lemma gives a dichotomy once the ambient cone has the requisite property $(P)$ [2509.04189].

In the multi-operator application, the relevant cone is the cone of blow-up elliptic measures
$$
M=D(A^+(p),A^-(p)),
$$
and the flat cone $F$ consists of $(n-1)$-plane measures [2509.04189]. The source verifies property $(P)$ by a Liouville-type rigidity statement: if a blow-up measure is sufficiently close to flat at all large scales, then a global Liouville theorem for the two-phase free boundary, obtained through Caffarelli’s “flat$\to$Lipschitz$\to C^{1,\alpha}\to$two-plane” program, forces the measure itself to be flat [2509.04189]. Since touching corkscrew balls in $\Omega^\pm$ produce a two-plane tangent measure, the alternative $\operatorname{Tan}(\omega,p)\cap F=\varnothing$ is excluded, and one concludes that all tangent measures are flat for $\omega^\pm$-almost every $p\in F_1$ [2509.04189].

Several limitations are explicit. In the single-operator harmonic case, mutual absolute continuity $\omega^+\ll \omega^-\ll \omega^+$ forces $\dim_H F^*=n-1$ and $H^{n-1}$-rectifiability. In the multi-operator setting, one still has
$$
\omega^+\ll \omega^-\ll \omega^+ \Longrightarrow \dim_H F^*\le n-1,
$$
but the reverse inequality and full rectifiability remain open because an Alt–Caffarelli–Friedman monotonicity formula “in the case of coefficient jumps fails to characterize flatness” [2509.04189]. Accordingly, the paper gives only a partial answer to Bishop’s question and leaves the multi-operator Oksendal conjecture open [2509.04189].

Open directions also appear in the harmonic and VMO theories. The lecture-note exposition of Engelstein’s work lists possible extensions to second-order elliptic operators with variable coefficients and corresponding elliptic measure, to two-phase problems for $p$-harmonic measure or nonlinear PDEs, and to finer geometric analysis of the singular set where the tangent polynomial has degree at least $2$ [1409.4460]. Prats and Tolsa state that the same equivalences should hold for unbounded NTA domains with poles at $\infty$, and they identify singular sets and stratification, as well as higher-order analogues such as biharmonic or polyharmonic measures and fully nonlinear elliptic operators, as active directions [1904.00751].

A recurrent misconception is that mutual absolute continuity alone automatically yields the full smooth or rectifiable picture. The cited results are more specific. In the Hölder regime, one obtains sharp $C^{1,\alpha}$ and then $C^{k+1,\alpha}$ regularity under the stated assumptions on $\log h$ and the ambient geometry [1409.4460]. In the VMO regime, one obtains an equivalence with vanishing Reifenberg flatness, VMO normal, and chord-arc approximation under the stated small-flatness hypotheses [1904.00751]. In the multi-operator setting, mutual absolute continuity yields flat tangent measures almost everywhere and the upper bound $\dim_H F^*\le n-1$, but not the missing lower bound on dimension [2509.04189]. Together these results describe two-phase elliptic measure as a boundary invariant whose analytic regularity, free-boundary structure, and geometric consequences depend sensitively on the operator, the oscillation class of the logarithmic density, and the quantitative flatness available a priori.

Source: https://www.emergentmind.com/topics/two-phase-elliptic-measure