Abstract
We analyze the approximation of the solutions of second-order elliptic problems, which have point singularities and belong to a countably normed space of analytic functions, with a first-order, h-version finite element (FE) method based on uniform tensor-product meshes. The FE solutions are well known to converge with algebraic rate at most 1 / 2 in terms of the number of degrees of freedom, and even slower in the presence of singularities. We analyze the compression of the FE coefficient vectors represented in the so-called quantized-tensor-train format. We prove, in a reference square, that the resulting FE approximations converge exponentially in terms of the effective number N of degrees of freedom involved in the representation: \(N={\mathcal {O}} ( \log ^{5} \varepsilon ^{-1} ) \), where \(\varepsilon \in (0,1)\) is the accuracy measured in the energy norm. Numerically we show for solutions from the same class that the entire process of solving the tensor-structured Galerkin first-order FE discretization can achieve accuracy \(\varepsilon \) in the energy norm with \(N={\mathcal {O}} ( \log ^{\kappa } \varepsilon ^{-1} ) \) parameters, where \(\kappa <3\).
Similar content being viewed by others
Notes
I. V. Oseledets. QTT decomposition of the characteristic function of a simplex. September 2010, private communication.
We use the master branch of the GitHub version 2.2+ of July 24, 2014 (git tag http://github.com/oseledets/TT-Toolbox/tree/v2.3-4-ge1a3f2c).
References
Andreev, R., Tobler, C.: Multilevel preconditioning and low-rank tensor iteration for space time simultaneous discretizations of parabolic pdes. Numer. Linear Algebra Appl. 22(2), 317–337 (2015). doi:10.1002/nla.1951
Babuska, I., Szabo, B., Katz, I.: The p-version of the finite element method. SIAM J. Numer. Anal. 18(3), 515–545 (1981). doi:10.1137/0718033
Babuška, I.: Error-bounds for finite element method. Numer. Math. 16(4), 322–333 (1971). doi:10.1007/BF02165003
Babuška, I., Dorr, M.R.: Error estimates for the combined \(h\) and \(p\) versions of the finite element method. Numer. Math. 37(2), 257–277 (1981). doi:10.1007/BF01398256
Babuška, I., Guo, B.: The \(h\)-\(p\) version of the finite element method for domains with curved boundaries. SIAM J. Numer. Anal. 25(4), 837–861 (1988). doi:10.1137/0725048
Babuška, I., Guo, B.: Regularity of the solution of elliptic problems with piecewise analytic data. Part I. Boundary value problems for linear elliptic equation of second order. SIAM J. Math. Anal. 19(1), 172–203 (1988). doi:10.1137/0519014
Babuška, I., Guo, B.: Regularity of the solution of elliptic problems with piecewise analytic data. II: The trace spaces and application to the boundary value problems with nonhomogeneous boundary conditions. SIAM J. Math. Anal. 20(4), 763–781 (1989). doi:10.1137/0520054
Babuška, I., Guo, B.Q.: The h, p and h-p version of the finite element method; basis theory and applications. Advances in Engineering Software 15(3–4), 159–174 (1992). doi:10.1016/0965-9978(92)90097-Y. http://www.sciencedirect.com/science/article/pii/096599789290097Y
Babuška, I., Kellogg, R.B., Pitkäranta, J.: Direct and inverse error estimates for finite elements with mesh refinements. Numer. Math. 33(4), 447–471 (1979). doi:10.1007/BF01399326
Bachmayr, M., Dahmen, W.: Adaptive near-optimal rank tensor approximation for high-dimensional operator equations. Found. Comput. Math. 15(4), 839–898 (2015). doi:10.1007/s10208-013-9187-3
Bachmayr, M., Dahmen, W.: Adaptive low-rank methods: problems on Sobolev spaces. SIAM J. Numer. Anal. 54(2), 744–796 (2016). doi:10.1137/140978223
Ballani, J., Grasedyck, L.: A projection method to solve linear systems in tensor format. Numer. Linear Algebra Appl. (2012). doi:10.1002/nla.1818
Ballani, J., Grasedyck, L.: Tree adaptive approximation in the hierarchical tensor format. SIAM J. Sci. Comput. 36(4), A1415–A1431 (2014). doi:10.1137/130926328
Brenner, S.C., Scott, L.R.: The Mathematical Theory of Finite Element Methods, Texts in Applied Mathematics, vol. 15. Springer (2008). http://link.springer.com/book/10.1007/978-0-387-75934-0
Canuto, C., Hussaini, M.Y., Quarteroni, A., Zang, T.A.: Spectral Methods in Fluid Dynamics. Springer Series in Computational Physics. Springer-Verlag (1988). http://link.springer.com/book/10.1007/978-3-642-84108-8
Canuto, C., Hussaini, M.Y., Quarteroni, A., Zang, T.A.: Spectral Methods. Scientific Computation. Springer-Verlag (2006). http://www.springer.com/book/9783540307259
Canuto, C., Hussaini, M.Y., Quarteroni, A., Zang, T.A.: Spectral Methods. Scientific Computation. Springer-Verlag (2007). http://www.springer.com/book/9783540307273
Canuto, C., Nochetto, R.H., Stevenson, R., Verani, M.: Adaptive spectral Galerkin methods with dynamic marking. SIAM J. Numer. Anal. 54(6), 3193–3213 (2016). doi:10.1137/15M104579X
Canuto, C., Tabacco, A., Urban, K.: The wavelet element method: Part i. construction and analysis. Applied and Computational Harmonic Analysis 6(1), 1–52 (1999). doi:10.1006/acha.1997.0242. http://www.sciencedirect.com/science/article/pii/S1063520397902426
Canuto, C., Tabacco, A., Urban, K.: The wavelet element method part ii. realization and additional features in 2d and 3d. Appl. Comput. Harmon. Anal. 8(2), 123–165 (2000). doi:10.1006/acha.2000.0282. http://www.sciencedirect.com/science/article/pii/S1063520300902823
Dahmen, W., Schneider, R.: Composite wavelet bases for operator equations. Math. Comput. 68(228), 1533–1567 (1999). doi:10.1090/S0025-5718-99-01092-3. http://www.ams.org/journals/mcom/1999-68-228/S0025-5718-99-01092-3
Dolgov, S., Khoromskij, B., Oseledets, I.: Fast solution of parabolic problems in the tensor train/quantized tensor train format with initial application to the fokker-planck equation. SIAM J. Sci. Comput. 34(6), A3016–A3038 (2012). doi:10.1137/120864210
Dolgov, S.V., Khoromskij, B.N.: Tensor-product approach to global time-space-parametric discretization of chemical master equation. Preprint 68, Max-Planck-Institut für Mathematik in den Naturwissenschaften (2012). http://www.mis.mpg.de/publications/preprints/2012/prepr2012-68.html
Dolgov, S.V., Savostyanov, D.V.: Alternating minimal energy methods for linear systems in higher dimensions. Part I: SPD systems. arXiv preprint 1301.6068 (2013). http://arxiv.org/abs/1301.6068
Grasedyck, L.: Hierarchical singular value decomposition of tensors. SIAM J. Matrix Anal. Appl. 31(4), 2029–2054 (2010). doi:10.1137/090764189
Grasedyck, L.: Polynomial approximation in Hierarchical Tucker Format by vector-tensorization. Preprint 308, Institut für Geometrie und Praktische Mathematik, RWTH Aachen (2010). http://www.igpm.rwth-aachen.de/Download/reports/pdf/IGPM308_k.pdf
Grasedyck, L., Kressner, D., Tobler, C.: A literature survey of low-rank tensor approximation techniques. GAMM-Mitteilungen 36(1), 53–78 (2013). doi:10.1002/gamm.201310004
Grisvard, P.: Elliptic problems in nonsmooth domains. Soc. Ind. Appl. Math. (2011). doi:10.1137/1.9781611972030
Guo, B., Babuška, I.: The h-p version of the finite element method. part 1: The basic approximation results. Comput. Mech. 1(1), 21–41 (1986). doi:10.1007/BF00298636
Guo, B., Babuška, I.: The h-p version of the finite element method. Part 2: general results and applications. Comput. Mech. 1(3), 203–220 (1986). doi:10.1007/BF00272624
Guo, B.Q., Babuska, I.: On the regularity of elasticity problems with piecewise analytic data. Adv. Appl. Math. 14(3), 307–347 (1993). doi:10.1006/aama.1993.1016. http://www.sciencedirect.com/science/article/pii/S019688588371016X
Hackbusch, W.: Tensor Spaces and Numerical Tensor Calculus, Springer Series in Computational Mathematics, vol. 42. Springer (2012). doi:10.1007/978-3-642-28027-6. http://www.springerlink.com/content/l62t86
Hackbusch, W., Kühn, S.: A new scheme for the tensor representation. J. Fourier Anal. Appl. 15(5), 706–722 (2009). doi:10.1007/s00041-009-9094-9. http://www.springerlink.com/content/t3747nk47m368g44
Harbrecht, H., Schneider, R.: Wavelet Galerkin schemes for boundary integral equations–implementation and quadrature. SIAM J. Sci. Comput. 27(4), 1347–1370 (2006). doi:10.1137/S1064827503429387
Hiptmair, R., Moiola, A., Perugia, I., Schwab, C.: Approximation by harmonic polynomials in star-shaped domains and exponential convergence of Trefftz hp-dGFEM. ESAIM. Math. Model. Numer. Anal. 48, 727–752 (2014). doi:10.1051/m2an/2013137. http://www.esaim-m2an.org/articles/m2an/abs/2014/03/m2an130137/m2an130137.html
Holtz, S., Rohwedder, T., Schneider, R.: The alternating linear scheme for tensor optimization in the Tensor Train format. SIAM J. Sci. Comput. 34(2), A683–A713 (2012). doi:10.1137/100818893
Kazeev, V.: Quantized tensor-structured finite elements for second-order elliptic PDEs in two dimensions. Ph.D. thesis, SAM, ETH Zurich, ETH Dissertation No. 23002 (2015). doi:10.3929/ethz-a-010554062. http://e-collection.library.ethz.ch/view/eth:48314
Kazeev, V.: Tensor-structured multilevel approximation of polynomial and piecewise-analytic functions (in preparation)
Kazeev, V., Khammash, M., Nip, M., Schwab, C.: Direct solution of the Chemical Master Equation using Quantized Tensor Trains. PLOS Comput. Biol. 10(3) (2014). doi:10.1371/journal.pcbi.1003359. http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1003359
Kazeev, V., Oseledets, I., Rakhuba, M., Schwab, C.: QTT-finite-element approximation for multiscale problems I: model problems in one dimension. Adv. Comput. Math. 1–32 (2016). doi:10.1007/s10444-016-9491-y
Kazeev, V., Schwab, C.: Tensor approximation of stationary distributions of chemical reaction networks. SIAM J. Matrix Anal. Appl. 36(3), 1221–1247 (2015). doi:10.1137/130927218
Khoromskij, B.N.: \(\cal{O}(d \log n)\)-quantics approximation of \(n\)-\(d\) tensors in high-dimensional numerical modeling. Constr. Approx. 34(2), 257–280 (2011). doi:10.1007/s00365-011-9131-1. http://www.springerlink.com/content/06n7q85q14528454/
Khoromskij, B.N., Schwab, C.: Tensor-structured Galerkin approximation of parametric and stochastic elliptic PDEs. SIAM J. Sci. Comput. 33(1), 364–385 (2011). http://epubs.siam.org/sisc/resource/1/sjoce3/v33/i1/p364_s1
Kondrat’ev, V.A.: Boundary value problems for elliptic equations in conical regions. Sov. Math. 4, 1600–1602 (1963)
Kondrat’ev, V.A.: Boundary problems for elliptic equations with conical or angular points. Transactions of the Moscow Mathematical Society 16, 209–292 (1967). http://mi.mathnet.ru/eng/mmo186
Kondrat’ev, V.A., Oleinik, O.A.: Boundary-value problems for partial differential equations in non-smooth domains. Russ. Math. Surv. 38, 1–86 (1983). http://stacks.iop.org/0036-0279/38/i=2/a=A01
Kressner, D., Steinlechner, M., Uschmajew, A.: Low-rank tensor methods with subspace correction for symmetric eigenvalue problems. Technical report 40, MATHICSE EPFL (2013). http://sma.epfl.ch/~anchpcommon/publications/EVAMEN.pdf
Kressner, D., Tobler, C.: Preconditioned low-rank methods for high-dimensional elliptic PDE eigenvalue problems. Comput. Methods Appl. Math. 11(3), 363–381 (2011). http://cmam.info/index.php?do=issues/art&vol=11&num=3&art=323
Lions, J.L., Magenes, E.: Non-Homogeneous Boundary Value Problems and Applications, Die Grundlehren der mathematischen Wissenschaften, vol. 181. Springer (1972). doi:10.1007/978-3-642-65161-8
Lubich, C.: From Quantum to Classical Molecular Dynamics: Reduced Models and Numerical Analysis. Zurich Lectures in Advanced Mathematics. European Mathematical Society (2008). doi:10.4171/067. http://www.ems-ph.org/books/book.php?proj_nr=87
Maz’ya, V.A., Plamenevskiĭ, B.A.: \(L_p\)-estimates of solutions of elliptic boundary value problems in domains with edges. Transactions of the Moscow Mathematical Society 37, 49–93 (1978). http://mi.mathnet.ru/eng/mmo357
Melenk, J.M.: hp-Finite Element Methods for Singular Perturbations. Lecture Notes in Mathematics. Springer-Verlag (2002). http://www.springer.com/en/book/9783540442011
Oseledets, I.: Approximation of matrices with logarithmic number of parameters. Dokl. Math. 80, 653–654 (2009). doi:10.1134/S1064562409050056
Oseledets, I., Dolgov, S.: Solution of linear systems and matrix inversion in the TT-format. SIAM J. Sci. Comput. 34(5), A2718–A2739 (2012). doi:10.1137/110833142
Oseledets, I.V.: Approximation of \(2^{d} \times 2^{d}\) matrices using tensor decomposition. SIAM J. Matrix Anal. Appl. 31(4), 2130–2145 (2010). doi:10.1137/090757861. http://link.aip.org/link/?SML/31/2130/1
Oseledets, I.V.: Tensor Train decomposition. SIAM J. Sci. Comput. 33(5), 2295–2317 (2011). doi:10.1137/090752286
Oseledets, I.V.: Constructive representation of functions in tensor formats. Constr. Approx. 37(1), 1–18 (2013). http://link.springer.com/article/10.1007/s00365-012-9175-x
Oseledets, I.V. (ed.): TT Toolbox, 2.2 edn. (2015). http://spring.inm.ras.ru/osel/download/quick_start.pdf. http://github.com/oseledets/TT-Toolbox
Oseledets, I.V., Tyrtyshnikov, E.E.: Breaking the curse of dimensionality, or how to use SVD in many dimensions. SIAM J. Sci. Comput. 31(5), 3744–3759 (2009). doi:10.1137/090748330. http://epubs.siam.org/sisc/resource/1/sjoce3/v31/i5/p3744_s1
Rohwedder, T., Uschmajew, A.: Local convergence of alternating schemes for optimization of convex problems in the TT format. Preprint 112, DFG-Schwerpunktprogramm 1324 (2012). http://www.dfg-spp1324.de/download/preprints/preprint112.pdf
Schollwöck, U.: The density-matrix renormalization group in the age of matrix product states. Ann. Phys. 326(1), 96–192 (2011). doi:10.1016/j.aop.2010.09.012. http://www.sciencedirect.com/science/article/pii/S0003491610001752. January 2011 Special Issue
Schwab, C.: \(p\)- and \(hp\)-FEM: Theory and Application to Solid and Fluid Mechanics. Oxford University Press, Oxford (1998)
Stevenson, R.: Stable three-point wavelet bases on general meshes. Numer. Math. 80(1), 131–158 (1998). doi:10.1007/s002110050363
Stevenson, R.: Composite wavelet bases with extended stability and cancellation properties. SIAM J. Numer. Anal. 45(1), 133–162 (2007). doi:10.1137/060651021
Trefethen, L.N.: Spectral Methods in MATLAB. Society for Industrial and Applied Mathematics (2000). doi:10.1137/1.9780898719598
Tyrtyshnikov, E.E.: Tensor approximations of matrices generated by asymptotically smooth functions. Sbornik: Mathematics 194(5), 941–954 (2003). doi:10.1070/SM2003v194n06ABEH000747. http://iopscience.iop.org/1064-5616/194/6/A09
Beirão da Veiga, L., Buffa, A., Rivas, J., Sangalli, G.: Some estimates for \(h\)-\(p\)-\(k\)-refinement in isogeometric analysis. Numer. Math. 118(2), 271–305 (2011). doi:10.1007/s00211-010-0338-z
Wang, H., Thoss, M.: Multilayer formulation of the multiconfiguration time-dependent hartree theory. J. Chem. Phys. 119(3), 1289–1299 (2003). doi:10.1063/1.1580111
Rockett, A.M.: Sum of the inverses of binomial coefficients. Fibonacci Quart. 19(5), 433–437 (1981). http://www.fq.math.ca/Scanned/19-5/rockett.pdf
Author information
Authors and Affiliations
Corresponding author
Additional information
Dedicated to Prof. Reinhold Schneider on the occasion of his 60th birthday.
The research of VK was performed mainly at the Seminar for Applied Mathematics, ETH Zurich. During the preparation of this work, CS was supported in part by the European Research Council through the FP7 Advanced Grant AdG247277.
Appendices
A Auxiliary results
Lemma A.1
Let \(n\in {\mathbb {N}}\). Then, for the sum of the inverses of the corresponding \(n+1\) binomial coefficients, one has
Proof
For \(n\in {\mathbb {N}}\), let us denote the left-hand side of (A.1) by \(I_n\). For every \(n\in {\mathbb {N}}\), we have the recurrence relation \(I_{n+1} = ( n+2 ) / ( 2n+2 ) \, I_n + 1\), see [69, Theorem 1]. By induction, starting from \(n=3\), it follows that \(I_n \ge 2 ( 1+n^{-1} ) \) for all \(n \ge 3\). Then \(I_{n+1} \le I_n + ( 2n+2 ) ^{-1} I_n - 1\) for all \(n \ge 3\). Again by induction, we obtain \(I_n \le I_3 = 8/3\) for all \(n > 3\). Since \(I_1,I_2 \le I_3\), this proves the claim. \(\square \)
As above, for all \(p\in {\mathbb {N}}\) and \(s=0,1,\ldots ,p\), we use the notation
Lemma A.2
For all \(\varrho > 1\) and \(p\in {\mathbb {N}}\) such that \(p \ge \varrho \), let us set \(s= \bigl \lfloor p / \varrho \bigr \rfloor \), so that \(1 \le s \le p\). Then the bound
holds with \(c^{2} = e^{5}/\sqrt{2\pi }\).
Proof
Using Stirling’s bound for Euler’s Gamma function, we obtain the bounds
which yield together
For \(s= \lfloor p/\varrho \rfloor \), we have \(p/\varrho - 1 < s \le p/\varrho \), so that \( ( \varrho - 1 ) s \le p-s\). We denote \(t=2s / ( p+s ) \) and, using that \( ( 1-t ) ^{1/t} < e^{-1} \sqrt{1-t}\) holds for \(0 < t \le 1\), obtain
\(\square \)
B Bounds for univariate quasi-interpolation
We shall use the following bound, for which one may refer to either of [62, Corollary 3.15] or [67, Corollary 2].
Proposition B.1
Assume that \(p\in {\mathbb {N}}\) and \(s\in {{\mathbb {N}}}_{0}\) are such that \(s \le p\). Then, for any function \({\hat{u}} \in \mathbb {H}^{s+1}_{} ( \hat{\mathrm {J}} ) \), the interpolant \(\hat{\pi }_{p} {\hat{u}}\) satisfies the error bounds
We shall also use the following stability bound on the second derivative of the interpolant.
Lemma B.2
For every \(p\in {\mathbb {N}}\) and for every \(u\in \mathbb {H}^{2}_{} ( \hat{\mathrm {J}} ) \), the following bounds hold:
Proof
Consider \(\mathbb {L}^{2}_{w} ( \hat{\mathrm {J}} ) \), a weighted space of square-integrable functions defined on \(\hat{\mathrm {J}}\), with the weight w given by \(w ( x ) =1-x^{2}\) for all \(x\in \hat{\mathrm {J}}\). For this weight, we have \({\hat{u}}''\in \mathbb {L}^{2}_{w} ( \hat{\mathrm {J}} ) \) with \( || {\hat{u}}'' || _{\mathbb {L}^{2}_{w} ( \hat{\mathrm {J}} ) } \le || {\hat{u}}'' || _{\mathbb {L}^{2}_{} ( \hat{\mathrm {J}} ) }\). Note that \(L_{i}'\), \(i\in {\mathbb {N}}\), are orthogonal in \(\mathbb {L}^{2}_{w} ( \hat{\mathrm {J}} ) \), namely
Below, we shall also use that \( || L_{i}' || _{\mathbb {L}^{2}_{} ( \hat{\mathrm {J}} ) }^{2}=i ( i+1 ) \) for every \(i\in {{\mathbb {N}}}_{0}\). This follows readily from integration by parts and the orthogonality of the Legendre polynomials, as in [62, theorem 3.91]:
Since \({\hat{u}}'\in \mathbb {L}^{2}_{} ( \hat{\mathrm {J}} ) \), we have a Legendre representation \({\hat{u}}'=\sum _{i=0}^{\infty } c_{i} L_{i}\) in \(\mathbb {L}^{2}_{} ( \hat{\mathrm {J}} ) \) with coefficients \(c_{i}\), \(i\in {{\mathbb {N}}}_{0}\). Then \({\hat{u}}''=\sum _{i=1}^{\infty } c_{i} L_{i}'\) holds in \(\mathbb {L}^{2}_{} ( \hat{\mathrm {J}} ) \) and, thus, also in \(\mathbb {L}^{2}_{w} ( \hat{\mathrm {J}} ) \). This results in the bound
By Definition 5.3, we have \( ( \hat{\pi }_{p} {\hat{u}} ) '=\sum _{i=0}^{p-1} c_{i} L_{i}\), and by the triangle inequality we obtain
Finally, we apply the Cauchy–Bunyakovsky–Schwarz inequality to (B.4) and (B.5) to arrive at the bounds
\(\square \)
By rescaling from an interval \(\mathrm {I}\) to \(\hat{\mathrm {J}}=(-1,1)\), from Proposition B.1 and Lemma B.2 we obtain the following statement.
Corollary B.3
Consider an interval \(\mathrm {I}= (2a,2a+2 h)\) with \(0< h < \infty \). Let \(\varphi \) be an affine map from \(\hat{\mathrm {J}}\) onto \(\mathrm {I}\). Then, for all \(p,s\in {\mathbb {N}}\) such that \(s \le p\), for every \(u \in \mathbb {H}^{s+1}_{} ( \mathrm {G} ) \) and for \(v\in \mathbb {Q}_{p ,{\,}\!p}\) given by \(v {{\mathrm{\circ }}}\varphi = \hat{\pi }_{k ,{\,}\!p} \, ( u {{\mathrm{\circ }}}\varphi ) \), the following inequalities hold:
C Bounds for tensor-product bivariate quasi-interpolation
We shall use the following error bound for the projection \(\hat{\Pi }_{p_1,p_2}\). For similar bounds for tensor-product interpolation operators, we also refer to [62, lemma 4.67] and [67, theorem 5].
Lemma C.1
Let \(p= ( p_1,p_2 ) \in {\mathbb {N}}^2\) and \(q_1,q_2,r_1,r_2,s_1,s_2\in {{\mathbb {N}}}_{0}\) be such that \(q_1,r_1,s_1 \le p_1\) and \(q_2,r_2,s_2 \le p_2\). Then the following bounds hold true for every \({\hat{u}} \in \mathbb {H}^{q_1+1,2}_{{\mathrm{mix}}} ( \hat{\mathrm {Q}} ) \cap \mathbb {H}^{2,q_2+1}_{{\mathrm{mix}}} ( \hat{\mathrm {Q}} ) \cap \mathbb {H}^{r_1+1,r_2+1}_{{\mathrm{mix}}} ( \hat{\mathrm {Q}} ) \cap \mathbb {H}^{s_1+1 ,{\,}\!1}_{{\mathrm{mix}}} ( \hat{\mathrm {Q}} ) \cap \mathbb {H}^{1,s_2+1}_{{\mathrm{mix}}} ( \hat{\mathrm {Q}} ) \):
Proof
Note that the inequality
follows from Definition 5.3.
Let \({\hat{u}} \in \mathbb {H}^{s_1+1 ,{\,}\!1}_{{\mathrm{mix}}} ( \hat{\mathrm {Q}} ) \cap \mathbb {H}^{1,s_2+1}_{{\mathrm{mix}}} ( \hat{\mathrm {Q}} ) \). Since \(\hat{\Pi }_{p_1 {{\,}\!}p_2} = \hat{\pi }_{1 ,{\,}\!p_1} {{\mathrm{\circ }}}\, \hat{\pi }_{2 ,{\,}\!p_2}\), we have the decomposition
\(\underline{\mathbb {L}^{2}_{}\text { error bound.}}\) By the triangle inequality, from (C.2) we obtain
Using Proposition B.1, we bound the first two terms as follows: for every \(k\in \{ 1,2 \} \), we have
for every \(s_k\in {{\mathbb {N}}}_{0}\) such that \(s_k \le p_k\). Similarly, for the third term we obtain
for all \(r_1,r_2\in {{\mathbb {N}}}_{0}\) such that \(r_1\le p_1\) and \(r_2\le p_2\). By combining the bounds for the three terms, we obtain the \(\mathbb {L}^{2}_{}\)-norm estimate claimed.
\(\underline{\mathbb {H}^{1}_{}\text { error bound.}}\) We use (C.2), the triangle inequality, (C.1) and Proposition B.1 to arrive at
An analogous bound holds for \( || \partial _{2}^{} ( {\hat{u}} - \hat{\Pi }_{p_1 {{\,}\!}p_2} {\hat{u}} ) || _{\mathbb {L}^{2}_{} ( \hat{\mathrm {Q}} ) }^{2}\). Together, the two bounds prove the \(\mathbb {H}^{1}_{}\)-norm error estimate claimed.
\(\underline{\mathbb {H}^{2}_{}\text {-stability estimate.}}\) Let us decompose the interpolant as follows:
Using the triangle inequality and the properties of the interpolation operators, we obtain
For these two terms, let us use the corresponding bounds of Lemma B.2. For the second, we obtain \( || \partial _{1}^{2} \hat{\pi }_{1 ,{\,}\!p_1} {\hat{u}} || _{\mathbb {L}^{2}_{} ( \hat{\mathrm {J}} ) }^{2} \le \frac{1}{2} ( p_1^2 - 1 ) || \partial _{1}^{2} {\hat{u}} || _{\mathbb {L}^{2}_{} ( \hat{\mathrm {J}} ) }^{2} \). For the first term, we use the error bound of Proposition B.1: \( || ( \mathsf {id}_{}- \hat{\pi }_{2 ,{\,}\!p_2} ) \partial _{1}^{2} \hat{\pi }_{1 ,{\,}\!p_1} {\hat{u}} || _{\mathbb {L}^{2}_{} ( \hat{\mathrm {J}} ) }^{2} \le p_2^{-1} ( p_2 + 1 ) ^{-1} || \partial _{2}^{}\partial _{1}^{2} \hat{\pi }_{1 ,{\,}\!p_1} {\hat{u}} || _{\mathbb {L}^{2}_{} ( \hat{\mathrm {J}} ) }^{2} = p_2^{-1} ( p_2 + 1 ) ^{-1} || \partial _{1}^{2} \hat{\pi }_{1 ,{\,}\!p_1} \partial _{2}^{}{\hat{u}} || _{\mathbb {L}^{2}_{} ( \hat{\mathrm {J}} ) }^{2} \). Then, by the first bound of Lemma B.2, the inequality \( || ( \mathsf {id}_{}- \hat{\pi }_{2 ,{\,}\!p_2} ) \partial _{1}^{2} \hat{\pi }_{1 ,{\,}\!p_1} {\hat{u}} || _{\mathbb {L}^{2}_{} ( \hat{\mathrm {J}} ) }^{2} \le p_2^{-1} ( p_2 + 1 ) ^{-1} p_1^2 / 4 \, ( p_1^2 - 1 ) || \partial _{1}^{}\partial _{2}^{}{\hat{u}} || _{\mathbb {L}^{2}_{} ( \hat{\mathrm {J}} ) }^{2} \) holds true. By combining the bounds for the two terms, we arrive at
An analogous estimate follows for \( || \partial _{2}^{2} \hat{\Pi }_{p_1 {{\,}\!}p_2} {\hat{u}} || _{\mathbb {L}^{2}_{} ( \hat{\mathrm {J}} ) }^{2}\). For the mixed derivative, (C.1) allows to obtain
Together, the bounds on the second-order derivatives yield the stability estimate. \(\square \)
Corollary C.2
Consider a rectangle \(\mathrm {G}= (2a_1,2a_1+2 h_1) {{\mathrm{\times }}}(2a_2,2a_2+2 h_2)\) with \(\lambda h \le h_1,h_2 \le \Lambda h\), where \(h>0\), \(0 < \lambda \le 1\) and \(\Lambda \ge 1\). Let \(\varphi \) be an affine map from \(\hat{\mathrm {Q}}\) onto \(\mathrm {G}\). Then, for all \(p,s\in {\mathbb {N}}\) such that \(s \le p\), for every \(u \in \mathbb {H}^{s+2}_{} ( \mathrm {G} ) \) and for \(v\in \mathbb {Q}_{p ,{\,}\!p}\) given by \(v {{\mathrm{\circ }}}\varphi = \hat{\Pi }_{p ,{\,}\!p} \, ( u {{\mathrm{\circ }}}\varphi ) \), the following inequalities hold:
Proof
The statement follows by a rescaling argument from Lemma C.1 with \(s_1=s_2=s\) and \(r_1=s-1\), \(r_2=0\). \(\square \)
Lemma C.3
Consider \(p_1,p_2\in {\mathbb {N}}\) and \(s_2\in {{\mathbb {N}}}_{0}\) such that \(s_2\le p_2\). Then the following bounds on \(\hat{\Gamma }_{}\), the left edge of \(\hat{\mathrm {Q}}\), hold for every \({\hat{u}} \in \mathbb {H}^{1,s_2+1}_{{\mathrm{mix}}} ( \hat{\mathrm {Q}} ) \):
Proof
First, we note that
holds with every \(\varkappa >0\). This explicit form of the trace theorem follows immediately from the formula \(\hat{v} ( -1,y ) = \hat{v} ( x,y ) - \int _{-1}^{x} \partial _{1}^{}\hat{v} ( t,y ) \,{\mathrm{d}}t\), valid for all \( ( x,y ) \in \hat{\mathrm {Q}}\), and from the Cauchy–Bunyakovsky–Schwarz inequality for \({\mathbb {R}}^{2}\) and \(\mathbb {L}^{2}_{} ( \hat{\mathrm {J}} ) \).
The interpolation property of \(\hat{\pi }_{1 ,{\,}\!p_1}\) and \(\hat{\pi }_{2 ,{\,}\!p_2}\) implies the relations
for all \(y\in \hat{\mathrm {J}}\) and \(\alpha _2\in \{ 0,1,2 \} \). Therefore, by Proposition B.1 and Lemma B.2, the following inequalities hold:
Then, applying inequality (C.5) with \(\varkappa =\frac{1}{2}\), we obtain the claim. \(\square \)
Corollary C.4
Consider a rectangle \(\mathrm {G}= (2a_1,2a_1+2 h_1) {{\mathrm{\times }}}(2a_2,2a_2+2 h_2) \subset \mathrm {Q}\). Let \(\gamma \) denote the left edge of \(\mathrm {G}\) and \(\varphi \) be an affine map from \(\hat{\mathrm {Q}}\) onto \(\mathrm {G}\).
Then, for all \(p\in {\mathbb {N}}\) and \(s\in {{\mathbb {N}}}_{0}\) such that \(s\le p\), for every \(u \in \mathbb {H}^{s+3}_{} ( \mathrm {G} ) \) and for \(v\in \mathbb {Q}_{p ,{\,}\!p}\) given by \(v {{\mathrm{\circ }}}\varphi = \hat{\Pi }_{p ,{\,}\!p} \, ( u {{\mathrm{\circ }}}\varphi ) \), the following inequalities hold:
Proof
Follows from Lemma C.3 by a rescaling argument. \(\square \)
Proposition C.5
Consider a rectangle \(\mathrm {G}= (2a_1,2a_1+2 h_1) {{\mathrm{\times }}}(2a_2,2a_2+2 h_2) \subset \mathrm {Q}\). Let \(w\in \mathbb {P}_{1}\) be the polynomial satisfying \(w ( 2a_1 ) =1\) and \(w ( 2a_1+2h_1 ) =0\).
Then, for every \(v\in \mathbb {P}_{p}\) with \(p\in {\mathbb {N}}\), the polynomial \(\xi =w {{\mathrm{\otimes }}}v \in \mathbb {Q}_{1 ,{\,}\!p}\) satisfies
with \(m=0,1\).
Proposition C.6
Consider a rectangle \(\mathrm {Q}=(0,1)^{2}\) and \(\beta \in [0,1)\). Then there exist positive constants \(D_{0}\) and \(D_{1}\) such that, for every \(u\in \mathbb {H}^{2 ,{\,}\!2}_{\beta } ( \mathrm {Q} ) \), the polynomial \(v \in \mathbb {Q}_{1 ,{\,}\!1}\) interpolating u at the vertices of \(\mathrm {Q}\) is well-defined and satisfies the error bounds
Proof
By Proposition 3.1, the function u admits continuous extension to \({{\mathrm{\mathbf{cl}\,}}}\mathrm {Q}\). This ensures that the interpolant is well defined. The error estimates follow from [29, Lemma 3.6], see also [62, Lemma 4.25]. \(\square \)
D Proof of Proposition 5.2
Proof
Let \(u\in \mathbb {H}^{2}_{\beta } ( \mathrm {Q} ) \) and \(l\ge 2\). Applying Corollary C.2 in every except with \(i_{0}= ( \mu ^{}_{1 {{\,}\!}1}-1 ,{\,}\!\mu ^{}_{2 {{\,}\!}1}-1 ) \) and Proposition C.6 with a rescaling argument in , we obtain
where \(D_{0}\) and \(D_{1}\) are positive constants depending only on \(\beta \). The claim follows immediately. \(\square \)
E Proofs of theorems for hp quasi-interpolation
1.1 E.1 Proof of Lemma 5.8
Proof
For an arbitrary function \(u\in \mathbb {H}^{1 ,{\,}\!1}_{{\mathrm{mix}}} ( \mathrm {Q} ) \), we shall define a lifting term so that and the mapping \(u \mapsto w^{l}\) is linear.
For \(1 \le j \le l\), let \(\gamma ^{{{\,}\!}j}_{1}\) and \(\tilde{\gamma }^{{{\,}\!}j}_{1}\) denote the right edges of \(\mathrm {G}^{{{\,}\!}l ,{\,}\!j}_{1 {{\,}\!}1}\) and \(\mathrm {G}^{{{\,}\!}l ,{\,}\!j}_{1 {{\,}\!}0}\) respectively, \(\gamma ^{{{\,}\!}j}_{2}\) and \(\tilde{\gamma }^{{{\,}\!}j}_{2}\), the top edges of \(\mathrm {G}^{{{\,}\!}l ,{\,}\!j}_{1 {{\,}\!}1}\) and \(\mathrm {G}^{{{\,}\!}l ,{\,}\!j}_{0 {{\,}\!}1}\) respectively, \(\tilde{\Gamma }^{{{\,}\!}j}_{1}\) and \(\tilde{\Gamma }^{{{\,}\!}j}_{2}\), the left edge of \(\mathrm {G}^{{{\,}\!}l ,{\,}\!j}_{1 {{\,}\!}0}\) and the bottom edge of \(\mathrm {G}^{{{\,}\!}l ,{\,}\!j}_{0 ,{\,}\!1}\) respectively.
Let us set consider the polynomial interpolants
given in Definition 5.7. For \(1 \le j < l\) and \(k\in \{ 1,2 \} \), we define linear univariate polynomials \(\psi ^{{{\,}\!}j}_{k}\in \mathbb {P}_{1}\) by requiring
Using these as factors, we introduce bivariate lifting polynomials \(\eta ^{{{\,}\!}j}_{1},\tilde{\eta }^{{{\,}\!}j}_{1}\in \mathbb {P}_{1 ,{\,}\!p}\) and \(\eta ^{{{\,}\!}j}_{2},\tilde{\eta }^{{{\,}\!}j}_{2}\in \mathbb {P}_{p ,{\,}\!1}\) with \(1 \le j < l\) by setting
Additionally, we introduce lifting polynomials \(\zeta _{1}\in \mathbb {P}_{1 ,{\,}\!p}\) and \(\zeta _{2}\in \mathbb {P}_{p ,{\,}\!1}\):
From the lifting polynomials defined in (E.1a)–(E.1b), we construct the lifting term :
By Definition 5.5, the interpolants of u in any two elements sharing an entire edge coincide on that edge. Then, due to the nodal exactness of interpolation in each element, the construction of (E.1a)–(E.1b) ensures that, first, \(w^{l}\) vanishes in \(\mathrm {G}^{{{\,}\!}l ,{\,}\!0}_{}\) and on \({{\mathrm{\varvec{\partial }}}}\mathrm {Q}\) and, second, the lifted interpolant extends to a function continuous across the edges of the elements of : .
Since the operator is linear, so is the mapping \(u \mapsto w^{l}\). This allows to define a linear operator on u by setting . \(\square \)
1.2 E.2 Preliminary bounds
In this section, using the auxiliary results presented in Appendices B and C below, we prove preliminary approximation and stability results of hp approximation, which are specified for the analyticity class in Theorem 5.10.
For all \(r\in {\mathbb {N}}\), \(\sigma \in {\mathbb {R}}\), we introduce \( [ \,\cdot \, ] _{r+1 ,{\,}\!\sigma }\), a broken Sobolev seminorm on \(\mathbb {H}^{r+1}_{} ( \mathrm {Q}\setminus \mathrm {G}^{{{\,}\!}l ,{\,}\!0}_{} ) \):
We note that \( [ \cdot ] _{r+1 ,{\,}\!\sigma }\) depends on \(\mu ^{}\) and l, which define . For the sake of brevity, we do not indicate this dependence explicitly in the notation.
Lemma E.1
(Estimates for hp quasi-interpolation in terms of the broken Sobolev seminorms) Let \(\beta \in [0,1)\). Then there exist positive constants \(D_{0}\) and \(D_{1}\) such that, for every \(\mu ^{}\in \{ 0,1 \} ^{2\times 2}\), for all \(l,p,s\in {\mathbb {N}}\) such that \(l \ge 2\) and \(s \le p\) and for every \(u\in \mathbb {H}^{2 ,{\,}\!2}_{\beta } ( \mathrm {Q} ) \cap \mathbb {H}^{s+2 ,{\,}\!2}_{\beta } ( \mathrm {G}^{l} ) \), the hp approximation satisfies the following error and stability bounds:
Furthermore, on the boundary we have the inequalities
where with \(j=1,\ldots ,l\) and \(\nu \in \mathcal {N}\) are given by (5.8a).
Proof
By Proposition C.6 and a rescaling argument, the interpolant \(v^{l}\) satisfies the error bounds
with positive constants \(D_{0}\) and \(D_{1}\) depending only on \(\beta \).
For all \(j=1,\ldots ,l\), \(\nu \in \mathcal {N}\) and \(s\in {\mathbb {N}}\) such that \(s \le p\), Corollary C.2 yields the following:
By combining the inequalities (E.6a)–(E.6b) and (E.7), we obtain the bounds of (E.4a)–(E.4c).
Finally, for all \(j=1,\ldots ,l\) and \(\nu \in \mathcal {N}\) such that of (5.8a) is nonempty, we apply Corollary C.4 to \(\mathrm {G}^{{{\,}\!}l ,{\,}\!j}_{\nu }\) and :
By summing over all \(j=1,\ldots ,l\) and \(\nu \in \mathcal {N}\), we obtain (E.5a)–(E.5c). \(\square \)
Let us denote the right and upper halves of \(\mathrm {G}^{{{\,}\!}l ,{\,}\!0}_{}\) as follows:
Lemma E.2
(Estimates for trace lifting in terms of the broken Sobolev seminorms) Let \(\beta \in [0,1)\). Then, for every \(\mu ^{}\in \{ 0,1 \} ^{2\times 2}\), for all \(l,p,s\in {\mathbb {N}}\) such that \(l \ge 2\) and \(s \le p\) and for every \(u\in \mathbb {H}^{3 ,{\,}\!2}_{} ( \mathrm {Q} ) \cap \mathbb {H}^{s+2 ,{\,}\!2}_{\beta } ( \mathrm {G}^{l} ) \), the trace-lifting term of Lemma 5.8 satisfies the bounds
where
with \(\tilde{\mathrm {G}}^{{{\,}\!}l ,{\,}\!0}_{1}\) and \(\tilde{\mathrm {G}}^{{{\,}\!}l ,{\,}\!0}_{2}\) given by (E.8).
Proof
Using the Cauchy–Bunyakovsky–Schwarz inequality in each \(\mathrm {G}^{{{\,}\!}l ,{\,}\!j}_{\nu }\), we may bound the lifting term \(w^{l}\) of Lemma 5.8 as follows:
for \(m=0,1,2\).
Applying Proposition C.5 and the Cauchy–Bunyakovsky–Schwarz inequality on edges of the elements, we obtain from (E.2) and (E.11) that
Rearranging the terms and using Corollary C.4, we arrive at
where \(\tilde{\mathrm {G}}^{{{\,}\!}l ,{\,}\!0}_{1}\) and \(\tilde{\mathrm {G}}^{{{\,}\!}l ,{\,}\!0}_{2}\) are given by (E.8).
Analogously to (E.12a), we obtain the bounds
and
The final inequalities of (E.12a)–(E.12c) prove the bounds of (E.9a)–(E.9b). \(\square \)
Let \(\mathrm {G}\subset \mathrm {Q}\) be a rectangle. Assume that in the sense of Definition 3.3 with positive constants \(C_{u}\) and \(\delta _{u}\). Then, using Lemma A.1, we obtain for every \(r\in {\mathbb {N}}\) that
Let \(l\in {\mathbb {N}}\), \(j=1,\ldots ,l\) and \(\nu \in \mathcal {N}\). Using (5.4), we obtain the bound for all \(x\in \mathrm {G}^{{{\,}\!}l ,{\,}\!j}_{\nu }\) and \(r\in {\mathbb {N}}\). The corresponding Sobolev seminorm is thus bounded as follows:
for all \(r\in {\mathbb {N}}\), \(l\in {\mathbb {N}}\), \(j=1,\ldots ,l\) and \(\nu \in \mathcal {N}\).
For \(\tilde{\mathrm {G}}^{{{\,}\!}l ,{\,}\!0}_{k}\) with \(k=1,2\) given by (E.8), we obtain a similar bound: for all \(x\in \mathrm {G}^{{{\,}\!}l ,{\,}\!j}_{\nu }\) and \(r\in {\mathbb {N}}\), and, therefore,
holds for all \(r\in {\mathbb {N}}\), \(l\in {\mathbb {N}}\), and \(k=1,2\).
Lemma E.3
(estimates for the broken Sobolev seminorms) Assume \(\beta \in [0,1)\), \(\alpha \ge 1\), \(\mu ^{}\in \{ 0,1 \} ^{2\times 2}\) and in the sense of Definition 3.3 with positive constants \(C_{u}\) and \(\delta _{u}\). Then, with p and s given by (5.10), the following bound is satisfied for all \(l,r\in {\mathbb {N}}\) and \(\sigma \in {\mathbb {R}}\):
where
Proof
Applying (E.13b) for all \(j=1,\ldots ,l\) and \(\nu \in \mathcal {N}\), we deduce the following bound for \( [ u ] ^{2}_{r+1 ,{\,}\!\sigma }\) of (E.3) with arbitrary \(r\in {\mathbb {N}}\) and \(\sigma \in {\mathbb {R}}\):
For p and s given by (5.10), using Lemma A.2, we arrive at
\(\square \)
1.3 E.3 Proof of Theorem 5.10
Proof
For p and s given by (5.10), Lemma E.3 yields the following bounds:
where \(\tilde{\beta }=\min \bigl \{ \frac{1}{2} ,{\,}\!\beta \bigr \} \). Also, using (E.3) and (E.13b), we obtain the estimates
Combining the bounds of Lemma E.1 with (E.14a)–(E.15b) and applying Definition 3.3 again, we obtain
where \(\tilde{\beta }=\min \bigl \{ \frac{1}{2} ,{\,}\!\beta \bigr \} \),
with any positive \(C_{1},C_{2},c_{0},c_{1},c_{2}\) such that
where the expressions on the right-hand side are monotonically decreasing with respect to l. \(\square \)
1.4 E.4 Proof of Theorem 5.11
Proof
Let \(l\in {\mathbb {N}}\) be greater than one. Consider p and s given by (5.10). According to Lemma E.2, there exists vanishing in \(\mathrm {G}^{{{\,}\!}l ,{\,}\!0}_{}\) and on \({{\mathrm{\varvec{\partial }}}}\mathrm {Q}\) and such that and the bounds (E.9a)–(E.9b) are satisfied. From those bounds, we obtain
where \(Z_0^2\) is given by (E.10) and can be estimated using (E.13c) as follows:
By Lemma E.3, there holds a bound
From (E.3), (E.13c) and Definition 3.2, we obtain also that
Combining the bounds (E.17)–(E.18b) with (E.16a)–(E.16b), we obtain the inequalities (5.13) with any positive \(\tilde{C}_{1},\tilde{C}_{2}\) such that
where the expressions on the right-hand side are monotonically decreasing with respect to l. \(\square \)
Rights and permissions
About this article
Cite this article
Kazeev, V., Schwab, C. Quantized tensor-structured finite elements for second-order elliptic PDEs in two dimensions. Numer. Math. 138, 133–190 (2018). https://doi.org/10.1007/s00211-017-0899-1
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00211-017-0899-1