--- title: Contour Bootstrapping in Matrices & Shapes url: https://www.emergentmind.com/topics/contour-bootstrapping-method type: topic --- # Contour Bootstrapping in Matrices & Shapes “Contour bootstrapping” designates two distinct technical constructions in recent arXiv literature. In matrix analysis, it is a resolvent-based method that converts contour-integral representations of spectral quantities into perturbation bounds for projectors, low-rank approximants, and analytic matrix functionals under a spectral gap assumption [2407.05230]. In statistical shape analysis, it denotes a bootstrap-based inferential procedure for 2D contours extracted from images, implemented within Elastic Shape Analysis (ESA) through percentile contours, square-root velocity functions (SRVFs), elastic shape distance (ESD), and an \(m\)-out-of-\(N\) bootstrap designed for non-smooth functionals [2606.04879]. The shared term is therefore terminologically ambiguous: in one setting the relevant contours lie in the complex plane, whereas in the other they are planar object boundaries. ## 1. Distinct meanings and mathematical settings The matrix-analytic usage arises in the study of perturbations of real symmetric \(n\times n\) matrices. Given a symmetric matrix \(A\) with eigen-decomposition \[ A=\sum_{i=1}^n \lambda_i u_i u_i^{\!*}, \] and a perturbation \(E\), one studies \(\widetilde A=A+E\) and seeks bounds on quantities such as the largest eigenvalue, spectral norm, leading eigenvector, eigenspace projectors, leading singular subspaces, best rank-\(p\) approximants, and more general matrix functionals of the form \[ f_S(A)=\sum_{i\in S} f(\lambda_i)u_i u_i^{\!*}. \] The methodological core is spectral calculus via complex contours and resolvents [2407.05230]. The ESA usage arises in image-based shape inference. A 2D contour is modeled as an absolutely continuous parameterized closed planar curve \[ c:\mathcal D\to \mathbb R^2, \] with \(\mathcal D=[0,1]\) or \(\mathbb S^1\), and in the closed case \(c(0)=c(1)\). Contours are extracted from images as percentile contours. If image intensities are \(\{S_{ij}\}\) with empirical distribution \(G\), then for a percentile \(p\in(0,1)\), \[ g_p=G^{-1}(p),\qquad c_p=f_p(S)=\{X_{ij}\in\mathbb R^2:S_{ij}=g_p,\ (ij)\in\mathcal G\}. \] That point set is then ordered and interpolated into a smooth closed curve. The inferential target is not Euclidean curve proximity but shape equivalence modulo translation, rotation, global scaling, and reparameterization [2606.04879]. This terminological duality is central. The matrix method is an analytic perturbation technique; bootESA is a nonparametric hypothesis-testing framework. Any unified treatment must therefore begin by separating the two senses rather than conflating them. ## 2. Contour bootstrapping in matrix perturbation theory For symmetric matrices, contour bootstrapping rests on the contour representation of spectral quantities. If \(\Gamma\) is a closed contour enclosing exactly the eigenvalues \(\{\lambda_i:i\in S\}\), then for analytic \(f\), \[ f_S(A)=\frac{1}{2\pi i}\oint_\Gamma f(z)(zI-A)^{-1}\,dz. \] The same formula holds for \(\widetilde A=A+E\) provided \(\Gamma\) encloses the corresponding perturbed eigenvalues and no others. Consequently, \[ f_S(\widetilde A)-f_S(A)=\frac{1}{2\pi i}\oint_\Gamma f(z)\big[(zI-\widetilde A)^{-1}-(zI-A)^{-1}\big]\,dz. \] The key algebraic input is the resolvent identity, \[ (zI-\widetilde A)^{-1}-(zI-A)^{-1}=(zI-A)^{-1}E(zI-\widetilde A)^{-1}. \] Tran–Vu then define \[ F(f,S)=\frac{1}{2\pi}\oint_\Gamma \big\|f(z)\big[(zI-\widetilde A)^{-1}-(zI-A)^{-1}\big]\big\|\,|dz| \] and obtain \[ \|f_S(\widetilde A)-f_S(A)\|\le F(f,S). \] A second expansion yields the self-referential inequality \[ F(f,S)\le F_1(f,S)+\alpha F(f,S), \] where \[ F_1(f,S)=\frac{1}{2\pi}\oint_\Gamma \|f(z)(zI-A)^{-1}E(zI-A)^{-1}\|\,|dz| \] and \[ \alpha=\max_{z\in\Gamma}\|(zI-A)^{-1}E\|. \] The “bootstrapping” step is the absorption of the self-term. Under the gap condition \[ \min_{i\in S,\;j\notin S}|\lambda_i-\lambda_j|\ge 4\|E\|, \] one can choose \(\Gamma\) so that its distance to the spectrum is at least \(2\|E\|\), implying \(\alpha\le 1/2\). Therefore \[ F(f,S)\le 2F_1(f,S), \] and the exact perturbation problem is reduced to estimating an integral involving only the resolvent of \(A\), the perturbation \(E\), and the contour geometry [2407.05230]. The method is carried out with rectangular contours whose vertical sides roughly bisect spectral gaps and whose horizontal sides are placed at imaginary height proportional to the relevant eigenvalue magnitude, for example \(T=2|\lambda_p|\). This contour design is not auxiliary: it encodes the separation needed both for spectral stability and for explicit logarithmic integral estimates. ## 3. Principal matrix bounds, refinements, and applications The first major application is eigenspace perturbation. For the projector onto the top \(p\) eigenspace, \[ \Pi_p(A)=\sum_{i=1}^p u_i u_i^{\!*}, \] let \(\delta_p=\lambda_p-\lambda_{p+1}\), let \(r>p\) be the smallest index such that \(\lambda_p-\lambda_{r+1}\ge |\lambda_p|\), and define \[ x=\max_{1\le i,j\le r}|u_i^{\!*}Eu_j|. \] Under \[ 4\|E\|\le \delta_p\le |\lambda_p|, \] Tran–Vu prove \[ \|\Pi_p(\widetilde A)-\Pi_p(A)\| \le 24\,\frac{\|E\|}{|\lambda_p|}\Big(\log\frac{60|\lambda_p|}{\delta_p}+2\Big) + 24\,\frac{r^2x}{\delta_p}. \] This refines Davis–Kahan by separating a term controlled by \(|\lambda_p|\) from a term controlled by the finer directional quantity \(x\), rather than only \(\|E\|/\delta_p\) [2407.05230]. For best rank-\(p\) approximation, the paper derives several bounds. In the positive semi-definite case, \[ \|\widetilde A_p-A_p\|\le 6\|E\|\Big(\log\frac{60|\lambda_p|}{\delta_p}+2\Big), \] under the corresponding gap assumption. For general analytic functionals \[ f_p(A)=\sum_{i=1}^p f(\lambda_i)u_i u_i^{\!*}, \] the contour argument yields \[ \|f_p(\widetilde A)-f_p(A)\| \le 24\,\max_{z\in\Gamma}|f(z)|\, \frac{\|E\|}{|\lambda_p|} \Big(\log\frac{60|\lambda_p|}{\delta_p}+2\Big). \] Projectors and rank-\(p\) approximants arise as the special cases \(f(z)=1\) and \(f(z)=z\). The significance of these bounds lies in the structural split between global perturbation size and directional incoherence. The paper explicitly notes that for random noise models such as Wigner perturbations, \(\|E\|\) may be of order \(\sqrt n\) while each bilinear form \(u_i^{\!*}Eu_j\) may be only \(O(\log n)\) with high probability. This opens regimes in which contour bootstrapping is substantially sharper than generic Davis–Kahan or Eckart–Young–Mirsky-based estimates. Two applications are emphasized. In matrix sparsification, \(E=\widetilde A-A\) is often random and structured, and the resulting bounds quantify when spectral subspaces or low-rank surrogates of a sparse approximation remain close to those of the dense matrix. In differential privacy, one adds noise \(E\) to a sensitive matrix \(A\), and the contour-bootstrapping bounds convert the required noise scale into explicit spectral utility guarantees. The paper highlights that this analysis is not tied to Gaussian noise and can therefore support privacy mechanisms with more general noise distributions [2407.05230]. ## 4. bootESA: contour bootstrapping in elastic shape analysis The second usage, bootESA, belongs to statistical shape analysis of image-derived contours. Its geometric foundation is ESA, which compares shapes in a manner invariant to translation, rotation, global scaling, and reparameterization. The basic representation is the SRVF \[ q(t)=\frac{\dot c(t)}{\sqrt{\|\dot c(t)\|}}, \] defined for a parameterized curve \(c\). Translation invariance is immediate because adding a constant vector to \(c\) does not change \(\dot c\), and scale invariance is enforced by normalization [2606.04879]. Shape comparison takes place in the quotient space \[ \mathcal S=\mathcal C/(SO(2)\times \Gamma), \] where \(\Gamma\) is the reparameterization group and \(SO(2)\) the rotation group. The elastic shape distance between two shape classes is defined by optimizing over rotations and reparameterizations and then applying a \(\cos^{-1}\) geodesic map to the resulting \(\mathbb L^2\) inner product. In practice, computation alternates between optimizing the rotation and optimizing the reparameterization, then evaluates the ESD after registration. The inferential target is a one-sample shape test. From reconstructed images \(\hat S^1,\dots,\hat S^N\), one forms the learned mean image \[ \bar S_{ij}=\frac{1}{N}\sum_{n=1}^N \hat S^n_{ij}, \] extracts the percentile contour \(\bar c_p\), converts it to its SRVF \(\bar q_p\), and compares it with a hypothesized true shape \(c^0\) having SRVF \(q^0\). The question is whether the underlying true shape has equivalence class \([q^0]\) in elastic shape space. A central obstacle is non-smoothness. The ESD functional is not Fréchet differentiable everywhere because the alignment optimizer can be non-unique and because the derivative of \(\cos^{-1}(x)\) diverges at \(x=1\), exactly the regime corresponding to zero shape distance. The paper therefore does not use a standard delta-method bootstrap directly on ESD. Instead, it adopts an \(m\)-out-of-\(N\) bootstrap tailored to non-smooth functionals [2606.04879]. ## 5. Bootstrap construction, confidence regions, and hypothesis testing Given reconstructed images \(\hat S^1,\dots,\hat S^N\), the bootstrap resamples in image space. For each bootstrap replicate \(b=1,\dots,B\), sample indices \(n_1,\dots,n_m\) with replacement from \(\{1,\dots,N\}\) and define \[ \bar S_b^*=\frac{1}{m}\sum_{k=1}^m \hat S^{n_k}. \] From each \(\bar S_b^*\), extract the percentile contour \(\bar c_{p,b}^*\), compute the SRVF \(\bar q_{p,b}^*\), and evaluate the elastic shape distance to the reference mean shape \([\bar q_p]\). The bootstrap statistic is \[ (R_m^*)_b=\sqrt m\, d_{\mathcal S}([\bar q_p],[\bar q_{p,b}^*]). \] The resulting empirical distribution approximates the sampling law of \(\sqrt m\, d_{\mathcal S}([q_p],[\bar q_p])\) [2606.04879]. The paper defines the unknown confidence region \[ C_{\alpha,m}=\{[q_p]:\sqrt m\, d_{\mathcal S}([q_p],[\bar q_p])z\big)<\alpha\right\}. \] For one-sided intervals, the empirical confidence region for ESD is \([0,\hat Z_\alpha]\). The one-sample test against a hypothesized shape \(q^0\) uses the observed statistic \[ R_N^{\mathrm{obs}}=\sqrt N\, d_{\mathcal S}([q^0],[\bar q_p]), \] and the p-value \[ p\text{-value}=\frac{\sum_{b=1}^B \mathbb I\big((R_m^*)_b>R_N^{\mathrm{obs}}\big)}{B}. \] One rejects the null at level \(\alpha\) when the p-value is at most \(\alpha\), equivalently when \(R_N^{\mathrm{obs}}>\hat Z_\alpha\). The \(m\)-out-of-\(N\) device is the essential modification. The paper states that with the standard choice \(m=N\), bootstrap ESDs are often conservative because the non-smooth ESD overestimates variability near zero. It therefore recommends \(m=N^\epsilon\) with \(\epsilon\in(0,1)\) and \(m/N\to 0\), while keeping the observed-data statistic scaled by \(\sqrt N\). In simulations for ICF-like images, \(\epsilon\approx 0.8\) provided reasonable type I error behavior for typical percentile levels \(p\in\{0.65,0.75,0.85\}\). In real data examples, \(B=1000\) bootstrap replicates were used [2606.04879]. ## 6. Empirical behavior, comparison with classical methods, and limitations For the matrix method, the comparison class is classical perturbation theory. Davis–Kahan bounds eigenspace error in terms of \(\|E\|/\delta_S\), and Eckart–Young–Mirsky-based bounds control low-rank perturbation by \(2(\sigma_{p+1}(A)+\|E\|)\). Contour bootstrapping differs by using a self-consistent resolvent inequality, by making the denominator depend on eigenvalue magnitude as well as spectral gap, and by admitting fine structure through quantities such as \(x=\max |u_i^{\!*}Eu_j|\). Its limitations are also explicit: a nontrivial spectral gap is required, typically \(4\|E\|\le \delta_p\), and the treatment in the paper is restricted to symmetric or Hermitian matrices rather than non-normal ones [2407.05230]. For bootESA, the comparison class includes Legendre polynomial or spherical harmonic summaries, Euclidean or Procrustes approaches, and existing ESA tests. The paper argues that basis-coefficient methods are well suited to smooth near-spherical shapes but can miss localized asymmetries; Euclidean and Procrustes approaches generally require landmark correspondences and are sensitive to parameterization; and prior ESA-based tests were chiefly multi-sample procedures rather than one-sample tests against a specified shape. bootESA fills that one-sample inferential gap by working directly with contour-derived shapes and ESD [2606.04879]. The simulation evidence reported for bootESA is consistent with its design. As sample size \(N\) increases, both empirical and bootstrap ESD distributions concentrate near zero. With \(m=N\), type I error is often below the nominal level and sometimes effectively zero; with \(\epsilon=0.8\), type I error is closer to the target \(\alpha=0.1\) for many percentiles. Power is high for alternatives such as indented circles, ellipses with varying aspect ratio, and polygons with few sides, and it increases with both sample size and effect size. In two ICF experiments with \(N=55\) pinholes, some percentile contours were significantly different from circles while others were not; the second shot showed more contours, especially in middle and high percentile ranges, that were not significantly different from circles [2606.04879]. A common misconception would be to treat “contour bootstrapping” as a single transferable recipe. Current usage indicates otherwise. In matrix perturbation theory, the contour is a complex-analytic integration path and the bootstrap is the absorption of a self-term in a resolvent inequality. In ESA, the contour is an image-derived planar boundary and the bootstrap is literal resampling for inference on a non-smooth distance functional. This suggests a family resemblance only at the level of indirect stabilization: both methods replace a difficult quantity by a more tractable surrogate under structural assumptions, but they do so in mathematically unrelated ways. The principal open directions likewise diverge. The matrix paper points toward fuller treatments of sparsification, privacy mechanisms, structured random perturbations, and possible extensions to non-normal matrices. The ESA paper points toward 3D surfaces, joint inference across multiple percentile contours, symmetry measures beyond circularity, alternative shape metrics, and other bootstrap schemes such as wild bootstrap, residual bootstrap in reconstruction space, or subsampling approaches. Taken together, the two literatures establish “contour bootstrapping” as a technically rich but context-dependent term rather than a single canonical method [2407.05230; 2606.04879]. Source: https://www.emergentmind.com/topics/contour-bootstrapping-method