Thresholded Weak Convergence in L1 Spaces

Updated 25 February 2026

Thresholded weak convergence is defined by dual parameters that control the greedy approximation process, ensuring both convergence and uniform boundedness in L1.
The method employs a dyadic chain selection and a two-stage weak thresholding approach that filters Haar coefficients based on predefined ratios.
This framework distinguishes itself from classical greedy methods by enabling effective approximation even when the multivariate Haar basis is not quasi-greedy.

Thresholded weak convergence is a concept arising from the study of greedy approximation algorithms for the multivariate Haar basis in $L_{1}([0,1]^d)$ , specifically in the context where the underlying basis is not quasi-greedy. The thresholded weak greedy algorithm introduces two real parameters, $0 < t < s < 1$, governing the weakness (threshold) and a secondary chain-length threshold. The central result is that, for this algorithm, the sequence of greedy approximants converges uniformly and is bounded for all $f \in L_{1}([0,1]^d)$ , in contrast to the failure of classical thresholding greedy algorithms in this setting (Dilworth et al., 2012).

1. Multivariate Haar Basis and Haar Coefficients

Let $d \ge 1$ . The setting is the Banach space $X = L_{1}([0,1]^d)$ , with the normalized multivariate Haar system $\{h_{I}^{(i)} : I \in \mathcal D^d,\, 1 \le i < 2^d\}$ , augmented by the constant $h_{[0,1]^d}^{(0)} = 1$ . The set $\mathcal D^d = \bigcup_{n \ge 0} \{I_1 \times \cdots \times I_d: I_j \subset [0,1)$ dyadic of length $2^{-n}$ } $indexes dyadic cubes of all scales. For$ f \in L_{1}([0,1]^{d) $, the Haar coefficient on cube$ I $and direction$ i $ is defined by $ c_{I}^{(i)}(f) = \int_{[0,1]^d} f(x) h_{I}^{(i)}(x)\,dx. $ A fixed total order$ < $is imposed on the collection of multi-indexed Haar functions. <h2 class='paper-heading' id='weak-thresholding-greedy-algorithm-construction'>2. Weak Thresholding Greedy Algorithm: Construction</h2> The weak thresholding greedy algorithm (WTGA) is parameterized by two real numbers$ 0 < t < s < 1 $. The approximation to$ f $is constructed inductively via $ G_m = G_{s, t}^{(m)}(f), \qquad R_m = f - G_m, \qquad (m \ge 0) $ with$ G_0 = 0 $and$ R_0 = f $. The iteration at$ m \mapsto m+1 $proceeds via three substeps: <ul> <li>(A) Branch Selection: Find the minimal$ (I_m, j_m) $for which$ |c_{I_m}^{(j_m)}(R_{m-1})| = \max_{(I, i)} |c_{I}^{(i)}(R_{m-1})| $.</li> <li>(B) Weak Thresholding on Dyadic Chain: Define$ A_m $as the maximal dyadic ancestor of$ I_m $such that all cubes$ I $in the dyadic chain$ \mathcal C(I_m, J) $from$ I_m $to$ J $admit some Haar direction$ i $with$ |c_{I}^{(i)}(R_{m-1})| \ge s\,|c_{I_m}^{(j_m)}(R_{m-1})| $.</li> <li>(C) Threshold-$ t $Selection in$ A_m $: Among directions$ 1 \le i < 2^d $in$ A_m $, select the smallest index$ i_m $such that</li> </ul> $ |c_{A_m}^{(i_m)}(R_{m-1})| \ge t \max_{1 \le j < 2^d} |c_{A_m}^{(j)}(R_{m-1})|. $ Set$ G_m = G_{m-1} + c_{A_m}^{(i_m)}(f) h_{A_m}^{(i_m)} $,$ R_m = f - G_m $. This selection rule is branch-greedy, focused on$ t $-admissible large coefficients, and is independent of minor variants of the selection order (<a href="/papers/1209.1378" title="" rel="nofollow" data-turbo="false" class="assistant-link" x-data x-tooltip.raw="">Dilworth et al., 2012</a>). <h2 class='paper-heading' id='main-results-convergence-and-uniform-boundedness'>3. Main Results: Convergence and Uniform Boundedness</h2> The primary theorem establishes: For$ 0 < t < s < 1 $, for every$ f \in L_{1}([0,1]^d) $: <ul> <li>Convergence:$ \lim_{m \to \infty} G_{s, t}^{(m)}(f) = f $in$ L_{1} $.</li> <li>Uniform Boundedness: There exists$ C(d, s, t) < \infty $such that for all$ m $,</li> </ul> $ \|G_{s, t}^{(m)}(f)\|_{1} \le C(d, s, t)\|f\|_{1}. $ An explicit bound, $ C(d, s, t) \le \left(\frac{5}{t} + 12\right) \left(1 + (2^d - 1)\,\min\{s(1-s), s-t\}/24\right), $ is provided. If either$ s \to t $or$ s \to 1 $, the algorithm fails to remain bounded and diverges on suitable examples. Thus, the multivariate Haar basis is not quasi-greedy in$ L_1 $(<a href="/papers/1209.1378" title="" rel="nofollow" data-turbo="false" class="assistant-link" x-data x-tooltip.raw="">Dilworth et al., 2012</a>). <h2 class='paper-heading' id='proof-structure-and-technical-lemmas'>4. Proof Structure and Technical Lemmas</h2> The proof of boundedness and convergence follows three stages: I. Norm-vs-Coefficient Estimates: <ul> <li>Lemma 3.1: For$ J \subset I $,$ |c_I(f)| \le \frac{|I|}{|J|}\|f\|_{1, J} $.</li> <li>Lemma 3.2: If$ |c_I(f)| \le 1 $for all$ I $, then$ \|P_I f\|_1 \le 1 $, with$ P_I $as the Haar-projection.</li> <li>Lemma 3.3: For$ J \subset I $,$ |J| = \frac{1}{2}|I| $,</li> </ul> $ \|P_I f\|_1 - \|P_J f\|_1 \ge \frac{1}{2}|c_I(f)| - |c_J(f)|. $ II. Combinatorial Decomposition of Active Cubes: <ul> <li>The minimal generalized chain representation (MGCR) provides a partition into dyadic chains whose tips correspond to cubes where ancestors drop below threshold.</li> <li>Lemma 4.6: For$ p, q $with disjoint active coefficient sets and opposite$ s $-weakness/$ t $-smallness,$ \|p+q\|_1 \ge C(s, t)|\text{union of minimal tips}| $.</li> </ul> III. Symmetrization and Final Patching: <ul> <li>Symmetrization via$ L_{i} $ensures all active cubes of a function lie in fixed dyadic siblings, facilitating lower bounds on norms.</li> <li>Iteration leads to$ \|G_m(f)\|_1 \le C(t)\,\|G_m(f) + \text{next increment}\|_1 \le C(t)\,\|f\|_1 $, with$ C(t) = 5/t + 12 $.</li> <li>Once uniform boundedness is established, convergence is achieved by the standard “basis-projection” (gliding hump) argument (Wojtaszczyk [11]).</li> </ul> <h2 class='paper-heading' id='algorithmic-quantities-and-threshold-parameter-effects'>5. Algorithmic Quantities and Threshold Parameter Effects</h2> No nontrivial asymptotic rate for$ \|f - G_m(f)\|_1 \to 0 $as$ m \to \infty $is provided, but explicit uniform bounds hold at each step:$ \|G_{s, t}^{(m)}(f)\|_1 \le C(d, s, t)\|f\|_1, $and consequently,$ \|R_m\|_1 \le (1 + C(d, s, t))\|f\|_1 $. The algorithm thus avoids blow-up. If parameter$ s $is chosen too close to$ t $or to$ 1$, divergence results, demonstrating the criticality of a gap between the two thresholds for the algorithm's stability in $L_1([0,1]^d) $(<a href="/papers/1209.1378" title="" rel="nofollow" data-turbo="false" class="assistant-link" x-data x-tooltip.raw="">Dilworth et al., 2012</a>). <h2 class='paper-heading' id='comparison-with-classical-greedy-approximations'>6. Comparison with Classical Greedy Approximations</h2> The thresholded weak greedy algorithm contrasts with two classical paradigms in Banach space approximation: <ul> <li>Thresholding Greedy Algorithm (<a href="https://www.emergentmind.com/topics/transition-aware-graph-attention-network-tga" title="" rel="nofollow" data-turbo="false" class="assistant-link" x-data x-tooltip.raw="">TGA</a>): Selects at each step the largest coefficient. The TGA fails to converge in$ L_1([0,1]^d) $for the Haar basis due to lack of quasi-greediness.</li> <li>Weak Greedy Algorithm: At each step, selects coefficients within a factor$ t < 1 $of the current maximum; this converges in$ L_p $for$ p > 1 $when the basis is quasi-greedy but fails in$ L_1 $without additional control.</li> <li>Thresholded Weak Algorithm (as in this context): Introduces two parameters$ t < s < 1 $. The branch-greedy step, governed by a secondary threshold$ s $, limits the climb in the dyadic tree and ensures selection of coefficients of sufficient size. Convergence in$ L_1 $is attained without the Haar basis being quasi-greedy (<a href="/papers/1209.1378" title="" rel="nofollow" data-turbo="false" class="assistant-link" x-data x-tooltip.raw="">Dilworth et al., 2012</a>).</li> </ul> In Banach space terminology, this demarcates a new regime: even for bases lacking quasi-greediness, convergence of a greedy algorithm can be restored by suitably “throttling” the selection mechanism via dual thresholds. <h2 class='paper-heading' id='significance-and-extensions'>7. Significance and Extensions</h2> Thresholded weak convergence demonstrates that controlled weakening of the greedy step, combined with dyadic chain-length moderation via auxiliary threshold$ s $, suffices for uniform approximation in$ L_1 $by members of the Haar basis. This is structurally distinct from classical greedy and weakly greedy algorithms, underpinning further studies in approximation theory for bases failing standard (quasi-)greedy conditions. The framework suggests avenues for developing analogous strategies for other non-quasi-greedy systems in$ L_p$ spaces and more general Banach spaces (Dilworth et al., 2012).}

Markdown Report Issue Upgrade to Chat

References (1)

On the Convergence of a Weak Greedy Algorithm for the Multivariate Haar Basis (2012)

Topic to Video (Beta)

No one has generated a video about this topic yet.

Whiteboard

No one has generated a whiteboard explanation for this topic yet.

Follow Topic

Get notified by email when new papers are published related to Thresholded Weak Convergence.