---
title: Defect Measure of Strong Compactness
url: https://www.emergentmind.com/topics/defect-measure-of-strong-compactness
type: topic
---

# Defect Measure of Strong Compactness

Defect measure of strong compactness denotes the quantitative object that records the gap between weak convergence and strong convergence for a non-compact embedding. For a continuous embedding \(E\hookrightarrow F\) of Banach spaces and a bounded sequence \(u_k\rightharpoonup u\) in \(E\), the defect of compactness is the part of \(u_k-u\) that fails to go to zero in \(F\). In Sobolev theory this defect is expressed by a profile decomposition, while in microlocal and PDE settings it is encoded by a Radon or phase-space measure whose vanishing is equivalent to strong convergence [1804.00531] [1212.1430].

## 1. Abstract notion and profile-decomposition viewpoint

The basic definition is relative to an embedding \(E\hookrightarrow F\). If \((u_k)\) is bounded in \(E\) and
\[
u_k \rightharpoonup u \quad \text{weakly in } E,
\]
then the defect of compactness is the part of \(u_k-u\) which fails to go to zero in \(F\). In the presence of a profile decomposition one writes, for each finite \(J\),
\[
u_k = u + \sum_{j=1}^J g_k^j\Phi^j + r_k^J,
\]
where \(u\) is the weak limit in \(E\), the terms \(g_k^j\Phi^j\) are elementary concentrations built from concentration profiles \(\Phi^j\), and the remainder satisfies \(\|r_k^J\|_F\to 0\) as \(J\to\infty\), uniformly in \(k\). When the embedding is compact, one can choose the decomposition so that there are no nontrivial profiles and hence the entire defect vanishes [1804.07950].

This viewpoint is closely related to cocompactness. In the \(BV(\mathbb R^n)\) setting one fixes a group \(D\) of linear isometries and calls a bounded sequence \(D\)-vanishing if for every sequence of dislocations \(g_k\in D\) one has \(\|g_k u_k\|_Y\to 0\). The embedding is cocompact with respect to \(D\) if \(D\)-vanishing implies strong convergence in \(Y\). In that framework, the defect of compactness is precisely the part that remains after all possible dislocations have been tested [1408.4583].

A complementary formulation is available in \(L^p\)-spaces through microlocal compactness forms. There, oscillations and concentrations precisely discriminate between weak and strong compactness, and the corresponding microlocal object vanishes if and only if the sequence converges strongly in \(L^p\) [1212.1430].

## 2. Sobolev embeddings on manifolds with bounded geometry

A central instance is the subcritical Sobolev embedding on a smooth, complete Riemannian manifold \(M\) of bounded geometry. In one formulation, bounded geometry means that the injectivity radius \(r(M)>0\) and all covariant derivatives of the curvature tensor are uniformly bounded. For \(1<p<N\) and
\[
2^*=\frac{Np}{N-p},
\]
the continuous embedding
\[
H^{1,p}(M)\hookrightarrow L^q(M), \qquad p<q<2^*,
\]
need not be compact on non-compact \(M\) [1804.00531].

The profile-decomposition theorem gives a complete description of the failure of compactness. Let \(\{u_k\}\subset H^{1,p}(M)\) be bounded with \(u_k\rightharpoonup w^{(0)}\). After extraction of a subsequence, there exist a countable family of points \(\{y_k^{(n)}\}_{k\in\mathbb N}\subset Y\), with \(Y\subset M\) an \(r\)-discretization, such that for any \(m\neq n\),
\[
d\!\bigl(y_k^{(n)},y_k^{(m)}\bigr)\to\infty \quad \text{as } k\to\infty,
\]
together with manifolds at infinity \(M^{(n)}\) of bounded geometry, nontrivial global profiles \(w^{(n)}\in H^{1,p}(M^{(n)})\), and elementary concentrations \(W_k^{(n)}\in H^{1,p}(M)\) such that for every \(N\in\mathbb N\),
\[
u_k = w^{(0)} + \sum_{n=1}^N W_k^{(n)} + r_k^N,
\]
where
\[
\lim_{k\to\infty}\|r_k^N\|_{L^q(M)}=0 \quad \text{uniformly in } N,
\]
and the series \(\sum_{n=1}^\infty W_k^{(n)}\) converges unconditionally in \(H^{1,p}(M)\), uniformly in \(k\) [1804.00531].

In the Hilbertian case treated for \(H^{1,2}(M)\hookrightarrow L^p(M)\), \(2<p<2^*\), the same structure appears with weak limit \(w^{(0)}\), countably many discrete sequences \(y_k^{(n)}\), manifolds at infinity \(M_\infty^{(n)}\), global profiles \(w^{(n)}\in H^{1,2}(M_\infty^{(n)})\), and elementary concentrations \(W_k^{(n)}\in H^{1,2}(M)\). For every finite \(J\),
\[
u_k = w^{(0)} + \sum_{n=1}^J W_k^{(n)} + r_k^J,
\]
with
\[
\|r_k^J\|_{L^p(M)}\to 0 \quad \text{as } J\to\infty \text{ uniformly in } k,
\]
and
\[
u_k-w^{(0)}-\sum_{n=1}^J W_k^{(n)} \to 0 \quad \text{in } L^p(M)
\]
[1804.07950].

The significance of this theorem is that the defect of strong compactness is not left as an unspecified failure of convergence. It is resolved into countably many asymptotically disjoint concentration modes, each attached to its own asymptotic geometry.

## 3. Manifolds at infinity and elementary concentrations

A distinctive feature of the manifold setting is that concentration profiles need not live on the original manifold \(M\). Each discrete sequence \((y_k)\) escaping to infinity carries its own asymptotic chart-by-chart geometry, and this geometry is assembled into a new complete Riemannian manifold \(M_\infty\) of bounded geometry [1804.07950].

The construction proceeds by fixing a small radius \(r<\operatorname{inj}(M)\) and a uniformly discrete lattice \(Y\subset M\). For each \(k\), one lists the neighbours of \(y_k\) in increasing distance:
\[
y_{k;0}=y_k,\quad y_{k;1},y_{k;2},\dots \in Y.
\]
On overlapping balls \(B(y_{k;i},r)\cap B(y_{k;j},r)\) one has transition maps
\[
v_{ij,k}=e_{y_{k;i}}^{-1}\circ e_{y_{k;j}}:Q_{2r}\to\mathbb R^N.
\]
By bounded geometry one extracts a \(C^\infty\)-convergent subsequence so that \(v_{ij,k}\to v_{ij}\) in \(C^\infty(Q_{2r})\). The collection of domains \(Q_{2r}\) with gluing maps \(v_{ij}\) satisfies the hypotheses of a standard manifold-gluing theorem, and their union is the new manifold \(M_\infty\), with coordinate charts \(\phi_i:Q_r\to M_\infty\) and transition functions \(\phi_i\circ \phi_j^{-1}=v_{ij}\). The metric on \(M_\infty\) is defined chartwise as the \(C^\infty\)-limit of the pulled-back metrics on \(B(y_{k;i},r)\), so \(M_\infty\) again has bounded curvature and injectivity radius \(\ge r\) [1804.07950].

The corresponding elementary concentration is obtained by “spotlighting” the local copies of the profile back into \(M\). With a trailing system of nearest-neighbour points \(y_{k;i}^{(n)}\), normal-coordinate charts
\[
e_{y_{k;i}^{(n)}}:B(0,r)\subset\mathbb R^N \to B(y_{k;i}^{(n)},r)\subset M,
\]
and a partition of unity \(\chi_{y_{k;i}^{(n)}}\), one defines
\[
W_k^{(n)}(x)=\sum_{i=0}^\infty \chi_{y_{k;i}^{(n)}}(x)\cdot
w^{(n)}\circ \phi_i^{(n)}\circ [e_{y_{k;i}^{(n)}}]^{-1}(x).
\]
The profile \(w^{(n)}\) is therefore an honest Sobolev function on its own limit manifold, while \(W_k^{(n)}\) is the corresponding concentration mode on \(M\) [1804.07950].

This construction shows that the “location” of the defect is not only a sequence of points running off to infinity. It also includes an induced limit geometry, and different escaping sequences may generate different manifolds at infinity.

## 4. Quantitative measurement by norm splitting

The profile decomposition becomes a defect measure in a quantitative sense through energy-decoupling and norm-splitting identities. In the Hilbertian case \(H^{1,2}(M)\), the decomposition yields a Plancherel-type inequality
\[
\liminf_{k\to\infty}\|u_k\|_{H^{1,2}(M)}^2
\ge
\|w^{(0)}\|_{H^{1,2}(M)}^2
+
\sum_{n=1}^\infty \|w^{(n)}\|_{H^{1,2}(M_\infty^{(n)})}^2
+
\limsup_{k\to\infty}\|r_k^k\|_{H^{1,2}(M)}^2.
\]
If one lets the remainder index \(J\to\infty\), the tail \(\|r_k^J\|\) may be made arbitrarily small, recovering the orthogonality of energies of the profiles. At the same time, the Lebesgue norms satisfy a Brezis–Lieb-type relation:
\[
\|u_k\|_{L^p(M)}^p
=
\|w^{(0)}\|_{L^p(M)}^p
+
\sum_{n=1}^J \|w^{(n)}\|_{L^p(M_\infty^{(n)})}^p
+
\|r_k^J\|_{L^p(M)}^p
+o(1),
\qquad k\to\infty.
\]
In the special case \(w^{(0)}\equiv 0\), the first term drops out [1804.07950].

For the general \(H^{1,p}(M)\) theory, the corresponding energy-decoupling estimate is
\[
\|w^{(0)}\|_{H^{1,p}(M)}^p
+
\sum_{n=1}^\infty \|w^{(n)}\|_{H^{1,p}(M^{(n)})}^p
\le
\liminf_{k\to\infty}\|u_k\|_{H^{1,p}(M)}^p,
\]
while for each fixed \(N\),
\[
\|u_k\|_{L^q(M)}^q
=
\|w^{(0)}\|_{L^q(M)}^q
+
\sum_{n=1}^N \|W_k^{(n)}\|_{L^q(M)}^q
+
o(1),
\qquad k\to\infty
\]
[1804.00531].

These identities provide the exact sense in which the defect is “measured.” The locations of defect are the sequences \(y_k^{(n)}\) running off to infinity; the geometry of each concentration region is captured by \(M_\infty^{(n)}\); and the strength of each defect is encoded by the Sobolev and Lebesgue norms of the corresponding profile. In this formulation, the embedding is compact exactly when all profiles vanish [1804.07950].

## 5. Measure-theoretic variants: \(BV\) and tensor-rectifiable defects

The defect-measure viewpoint is not confined to Sobolev embeddings on manifolds. In \(BV(\mathbb R^n)\), the non-compact embedding
\[
BV(\mathbb R^n)\hookrightarrow L^{1^*}(\mathbb R^n),\qquad 1^*=\frac{n}{n-1},
\]
admits a profile decomposition relative to the group of dyadic dilations and integer translations
\[
g[j,y]u(x)=2^{(n-1)j}u(2^j x-y), \qquad (j\in\mathbb Z,\ y\in\mathbb R^n).
\]
If \((u_k)\) is bounded in \(BV(\mathbb R^n)\), then after passing to a subsequence there exist profiles \(w^{(n)}\in BV(\mathbb R^n)\) and dislocation parameters \((j_k^{(n)},y_k^{(n)})\in\mathbb Z\times\mathbb R^n\) satisfying
\[
|j_k^{(n)}-j_k^{(m)}|+|y_k^{(n)}-y_k^{(m)}|\to\infty \qquad \text{whenever } n\neq m,
\]
such that
\[
u_k(x)=\sum_{n=1}^\infty g[j_k^{(n)},y_k^{(n)}]\,w^{(n)}(x)+r_k(x),
\]
with \(\|r_k\|_{L^{1^*}}\to 0\), and
\[
\sum_{n=1}^\infty |D w^{(n)}|(\mathbb R^n)+|D r_k|(\mathbb R^n)
\le |D u_k|(\mathbb R^n)+o(1).
\]
In this setting the total variation \(|Du_k|\) is the seminorm that carries the concentrations and defects [1408.4583].

A different but explicitly measure-theoretic example arises for energies penalizing simultaneous oscillations in two independent directions. With \(\mathbb R^n=X_1\oplus X_2\), \(\dim X_\ell=n_\ell\), and \(u\in L^1(\Omega)\), the failure of the decomposition \(Du=D_1u+D_2u\) with \(D_1u\otimes D_2u=0\) is measured by the mixed derivative
\[
\mu[u]:=\nabla_1\nabla_2 u,
\]
viewed as a distribution with values in \(X_1\otimes X_2\). Under the sole assumption \(E_\theta(u)<\infty\), \(\mu[u]\) extends to a finite Radon measure and satisfies
\[
|\mu[u]|(\Omega)\le C(\Omega,\theta_1,\theta_2)\,\|u\|_{L^\infty}^{1-\theta}\,E_\theta(u)<+\infty.
\]
For \(\theta<1\), \(\mu[u]\) is \((n_1-1,n_2-1)\)-tensor-rectifiable, meaning that it is concentrated on a countable union of tensor-products of \((n_1-1)\)- and \((n_2-1)\)-planes. In the case \(n_1=n_2=1\), \(\theta=1\), and \(u\in \operatorname{Lip}(\Omega)\), one recovers a 1-rectifiable defect:
\[
\mu = m(x)\,\mathcal H^1\lfloor \Sigma
\]
for a 1-rectifiable set \(\Sigma\subset\Omega\) and a Borel \(m:\Sigma\to\mathbb R\) [2309.17067].

These examples show that the phrase “defect measure” may designate either a structured family of elementary concentrations or an actual Radon measure. What remains invariant is the role: it captures all residual non-compactness once the weak limit has been removed.

## 6. Microlocal defect measures and phase-space formulations

Microlocal compactness forms provide a phase-space refinement of defect measurement for \(L^p\)-bounded sequences. For \(1<p<\infty\), \(\Omega\subset\mathbb R^d\) open and bounded, and a norm-bounded sequence \((u_j)\subset L^p(\Omega;\mathbb C^N)\), one obtains, after passing to a subsequence,
\[
\omega=(\{\omega_x\}_{x\in\Omega},\lambda_\omega,\{\omega_x^\infty\}_{x\in\overline\Omega})\in MCF^p(\Omega;\mathbb C^N),
\]
characterized by the double-limit representation
\[
\langle\!\langle f\otimes\overline\Psi,\omega\rangle\!\rangle
=
\lim_{R\to\infty}\lim_{j\to\infty}
\int_\Omega
h\bigl(x,u_j(x)\bigr)\cdot
\overline{T_{(1-\eta_R)\Psi}[u_j](x)}\,dx.
\]
Here \(\omega_x\) captures oscillations at point \(x\), while \(\omega_x^\infty\), together with \(\lambda_\omega\), captures concentrations. The concentration measure may be chosen canonically as
\[
\lambda_\omega=\mathrm{w}^*-\lim_{j\to\infty}|u_j(x)|^p\,dx
\quad\text{in } M^+(\overline\Omega).
\]
The decisive criterion is:
\[
u_j\to u \text{ strongly in } L^p(\Omega;\mathbb C^N)
\quad\Longleftrightarrow\quad
\omega=0.
\]
Thus \(\omega\) is exactly the defect measure for strong compactness in \(L^p\), and it unifies the information carried by generalized Young measures and \(H\)-measures [1212.1430].

A related but more classical phase-space object is the microlocal defect measure used in high-frequency limits of PDEs. For a sequence \(v_n=x(h_n-h_0)\in L^2_{\mathrm{loc}}\), one obtains a non-negative Radon measure
\[
\mu=\mu(x,\hat\xi)\in \mathcal M^+(S^*M)
\]
such that for every matrix-valued, order-zero pseudodifferential operator \(A=\operatorname{Op}(a)\),
\[
\lim_{n\to\infty}\int_{\mathbb R^{2+1}}\langle A v_n,v_n\rangle\,dx
=
\int_{S^*M}\operatorname{tr}[a(x,\hat\xi)]\,d\mu(x,\hat\xi).
\]
In the Einstein-vacuum setting with \(\mathbb U(1)\) symmetry and elliptic gauge, this measure is supported on the null-cone
\[
\{(x,\hat\xi)\in S^*M:\,g_0^{\alpha\beta}(x)\xi_\alpha\xi_\beta=0\},
\]
and it obeys the transport equation
\[
\nabla_\alpha(\xi^\alpha\mu)=0.
\]
The effective stress-energy in the limit Einstein equations is
\[
T_{\alpha\beta}^{\mathrm{eff}}(x)=\int \xi_\alpha\xi_\beta\,d\mu(x,\hat\xi).
\]
In this setting \(\mu\) records the defect in
\[
\bigl\langle A(h_n-h_0),\,h_n-h_0\bigr\rangle_{L^2},
\]
and its transport law identifies the limiting effective matter as massless Vlasov [1907.10743].

Across these formulations, a common principle emerges. Strong compactness fails through oscillation, concentration, escape to infinity, or high-frequency propagation; the defect measure is the object that isolates these mechanisms, quantifies their size, and shows that no further hidden defect remains.

Source: https://www.emergentmind.com/topics/defect-measure-of-strong-compactness