Metric Space Theory of BV Functions

• Metric BV theory is the study of functions of bounded variation in metric measure spaces, defined via both relaxation (upper gradients) and approximate modulus methods.
• It establishes equivalence between these definitions under doubling measures and 1–Poincaré inequalities, ensuring a robust analytic framework in non-smooth settings.
• The theory supports fine geometric analysis and extends classical BV results to PI-spaces, Carnot groups, and other complex, non-Euclidean environments.

A function of bounded variation (BV) is a function whose generalized gradient is a finite measure, with the BV theory in metric measure spaces extending the classical Euclidean framework to highly non-smooth settings using analytic tools adapted to metric geometry. The metric space BV theory rigorously characterizes, compares, and relates several relaxation-based and curve-modulus-based definitions, yielding robust seminorm equivalences under doubling and $1$-Poincaré inequalities and leading to the emergence of deep underlying geometric structures such as the Semmes family of curves.

1. Competing Definitions of BV in Metric Spaces

Two main approaches to defining BV-functions in complete metric measure spaces $(X,d,\mu)$, $\mu$ Borel and locally finite, have emerged:

• Miranda Jr.’s Relaxation Definition: A function $u\in L^1_{\mathrm{loc}}(X)$ is in $BV(X)$ if

$$|Du|(X) = \inf\Bigl\{\liminf_{i\to\infty}\int_X g_{u_i}\,d\mu \Bigr\} < \infty,$$

where the infimum is over sequences $u_i\in \mathrm{Lip}_{\mathrm{loc}}(X)$ converging to $u$ in $L^1_{\mathrm{loc}}$, and $g_{u_i}$ is the minimal $1$--weak upper gradient of $u_i$. This extends the distributional approach of the Euclidean setting to metric-measure geometries (Durand-Cartagena et al., 2018).

• Martio’s Approximate Modulus Definition: For a sequence $(\rho_i)$ of Borel functions, $(\rho_i)$ is AM-admissible for a curve family $\Gamma$ if for all rectifiable $\gamma\in\Gamma$,

$\liminf_{i\to\infty} \int_\gamma \rho_i\,ds \ge 1.$

The AM-modulus of $\Gamma$ is

$\AM(\Gamma) = \inf_{(\rho_i)\,\mathrm{AM-\!adm.}} \liminf_{i\to\infty} \int_X \rho_i\,d\mu.$

A function $u$ is in $BV_{\AM}(X)$ if, modulo a curve family of null AM-modulus, there exists a sequence $(\rho_i)$ with finiteness properties so that for every rectifiable $\gamma$ not in the exceptional family and all $s<t$,

$$|u(\gamma(t)) - u(\gamma(s))| \le \liminf_{i\to\infty}\int_{\gamma|[s,t]} \rho_i\,ds.$$

The AM-variation seminorm is

$|D_{\AM}u|(X) = \inf\Bigl\{\liminf_i \|\rho_i\|_{L^1(X)}\Bigr\},$

the infimum again over all such sequences (Durand-Cartagena et al., 2018).

These frameworks respectively reflect variational relaxation and rectifiable-path-based approaches to bounded variation in the absence of a linear structure.

2. Fundamental Equivalence under Doubling and $1$-Poincaré

For a complete metric space $(X,d,\mu)$ with $\mu$ doubling (i.e., $\mu(B(x,2r)) \le C_d \mu(B(x,r))$) and supporting a $1$-Poincaré inequality,

$\fint_B |u-u_B|\,d\mu \le C_P\,r\,\fint_{\lambda B} g\,d\mu,$

for all balls $B=B(x,r)$, $u\in L^1_{\mathrm{loc}}$ and any $1$-weak upper gradient $g$, the two notions above coincide:

$BV(X) = BV_{\AM}(X),$

with the seminorms comparable:

$c_1 |Du|(X) \le |D_{\AM}u|(X) \le |Du|(X), \quad c_1>0.$

Similarly, for the $N^{1,1}$ (Newton–Sobolev) and $N_{\AM}^{1,1}$ modules, the function spaces coincide (Durand-Cartagena et al., 2018).

The proof proceeds by showing:

• Any relaxation limit for $|Du|$ yields an AM-upper bound sequence, thus $BV(X)\subset BV_{\AM}(X)$.
• For $BV_{\AM}(X)\subset BV(X)$, one first demonstrates inclusion under the AM–Poincaré inequality, then compares the two Poincaré inequalities using the Semmes family (see next section), ultimately providing quadratic partition-of-unity arguments and Lipschitz (or Newtonian) approximations.

This equivalence validates the analytic robustness of the "metric BV class" in non-smooth settings subject to the doubling and $1$-Poincaré framework.

3. Semmes Family of Curves and its Structural Role

A central geometric consequence of the doubling and $1$-Poincaré hypotheses is the existence of a Semmes family of curves:

• For each pair $x,y\in X$, there exists a compact family $\Gamma_{x,y}$ of $C$-quasiconvex curves joining $x$ to $y$, equipped with a Borel probability measure $\sigma_{x,y}$.
• For every Borel set $A\subset X$,

$$\int_{\Gamma_{x,y}} |\gamma \cap A|\,d\sigma_{x,y}(\gamma) \leq C \int_{CB_{x,y} \cap A} R_{x,y}(z)\,d\mu(z),$$

with $CB_{x,y}$ a controlled neighborhood of $x$ and $y$, and $R_{x,y}(z)$ scalar densities involving $\mu$-measured balls at $z$ relative to $x$ and $y$.

The construction relies on modulus bounds for path families, minimax arguments, and measure selection strategies. The existence of a Semmes family is both a consequence and, via modulus duality, a characterization of the $1$-Poincaré property in metric measure spaces (Durand-Cartagena et al., 2018).

The Semmes pencil underpins both the equivalence of BV notions and the fine structure theory (e.g., coarea, trace, and fine continuity properties) of BV functions in metric geometry.

4. Key Structural Formulas: Poincaré and Coarea

Important analytic ingredients of the theory include:

• $1$-Poincaré Inequality:

$\fint_B |u - u_B|\,d\mu \leq C_P\, r\, \fint_{\lambda B} g\,d\mu,$

for all $u\in L^1_{\mathrm{loc}}$ and suitable upper gradients $g$.

• Coarea Formula (Miranda Jr.):

$$\int_X f\,d|Du| = \int_{-\infty}^{\infty}\int_X f\,d|D\chi_{\{u>t\}}|\,dt,$$

and, in particular,

$|Du|(X) = \int_{-\infty}^\infty P(\{u>t\}, X)\,dt,$

where $P(E,X)=|D\chi_E|(X)$ is the perimeter of $E$.

These allow BV analysis to recover the full machinery of fine properties, such as lower semicontinuity, fine traces, Gauss–Green formulas, and De Giorgi–type structure theorems (Durand-Cartagena et al., 2018).

5. Examples, Generalizations, and Applications

• PI-spaces: Any complete, doubling metric space supporting a $1$-Poincaré inequality, called a PI-space, is encompassed in this theory. This includes Riemannian manifolds with lower Ricci curvature bounds, Carnot groups (e.g., the Heisenberg group), Ahlfors–regular Loewner spaces, and fractal-type spaces.
• Carnot Groups: On Carnot groups, the equivalence of "horizontal BV" and the AM–BV class holds.
• Finite Perimeter Sets: Federer's characterization in $\mathbb{R}^n$—that a set $E$ has finite perimeter if and only if its measure-theoretic boundary has finite Hausdorff $(n-1)$-measure—is extended to metric spaces via the AM-modulus notion: the perimeter measure coincides with the AM-modulus of curves crossing the measure-theoretic boundary.
• Fine Properties: The outcome is a unified BV-theory in very general non-smooth spaces, with all fine properties—lower semicontinuity, full coarea formula, fine traces, the De Giorgi structure theorem, and their applications in geometric and variational problems—carrying over under the doubling and $1$-Poincaré hypothesis (Durand-Cartagena et al., 2018).

Notion	Definition type	Relation under (doubling + $1$-Poincaré)
Miranda Jr. BV	Relaxation via upper gradients/Lipschitz approximation	$BV(X) = BV_{\AM}(X)$ (equivalent seminorms)
Martio AM–BV	Approximate modulus via rectifiable curves	$BV(X) = BV_{\AM}(X)$ (equivalent seminorms)
Semmes family	Family of quasicontinuous curves with measure control	Characterizes $1$-Poincaré and ensures BV equivalence

6. Connections to Broader Metric BV Theory

The equivalence of the relaxation (energy-based) and AM–modulus (modulus-based) definitions anchors the entire analytic framework of BV in metric spaces. The existence of a Semmes family of curves is both necessary and sufficient for the underlying modulus structure and thus for the duality arguments crucial in geometric analysis and $\Gamma$-convergence.

This comprehensive metric space theory of functions of bounded variation is now the foundation for further developments in geometric measure theory, analysis on non-smooth spaces, and metric variational problems (Durand-Cartagena et al., 2018).