Tensor FEM for Spectral Fractional Diffusion


We design and analyze several finite element methods (FEMs) applied to the Caffarelli–Silvestre extension that localizes the fractional powers of symmetric, coercive, linear elliptic operators in bounded domains with Dirichlet boundary conditions. We consider open, bounded, polytopal but not necessarily convex domains \(\varOmega \subset {\mathbb {R}}^d\) with \(d=1,2\). For the solution to the Caffarelli–Silvestre extension, we establish analytic regularity with respect to the extended variable \(y\in (0,\infty )\). Specifically, the solution belongs to countably normed, power-exponentially weighted Bochner spaces of analytic functions with respect to y, taking values in corner-weighted Kondrat’ev-type Sobolev spaces in \(\varOmega \). In \(\varOmega \subset {\mathbb {R}}^2\), we discretize with continuous, piecewise linear, Lagrangian FEM (\(P_1\)-FEM) with mesh refinement near corners and prove that the first-order convergence rate is attained for compatible data \(f\in \mathbb {H}^{1-s}(\varOmega )\) with \(0<s<1\) denoting the fractional power. We also prove that tensorization of a \(P_1\)-FEM in \(\varOmega \) with a suitable hp-FEM in the extended variable achieves log-linear complexity with respect to \({\mathscr {N}}_\varOmega \), the number of degrees of freedom in the domain \(\varOmega \). In addition, we propose a novel, sparse tensor product FEM based on a multilevel \(P_1\)-FEM in \(\varOmega \) and on a \(P_1\)-FEM on radical-geometric meshes in the extended variable. We prove that this approach also achieves log-linear complexity with respect to \({\mathscr {N}}_\varOmega \). Finally, under the stronger assumption that the data be analytic in \(\overline{\varOmega }\), and without compatibility at\(\partial \varOmega \), we establish exponential rates of convergence ofhp-FEM for spectral fractional diffusion operators in energy norm. This is achieved by a combined tensor product hp-FEM for the Caffarelli–Silvestre extension in the truncated cylinder with anisotropic geometric meshes that are refined toward \(\partial \varOmega \). We also report numerical experiments for model problems which confirm the theoretical results. We indicate several extensions and generalizations of the proposed methods to other problem classes and to other boundary conditions on \(\partial \varOmega \).

This is a preview of subscription content, log in to check access.

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6
Fig. 7
Fig. 8
Fig. 9
Fig. 10


  1. 1.

    Abramowitz, M., Stegun, I.: Handbook of mathematical functions with formulas, graphs, and mathematical tables, National Bureau of Standards Applied Mathematics Series, vol. 55. For sale by the Superintendent of Documents, U.S. Government Printing Office, Washington, D.C. (1964)

  2. 2.

    Acosta, G., Borthagaray, J.: A fractional Laplace equation: regularity of solutions and finite element approximations. SIAM J. Numer. Anal. 55(2), 472–495 (2017). https://doi.org/10.1137/15M1033952.

    MathSciNet  MATH  Article  Google Scholar 

  3. 3.

    Adams, R.: Sobolev spaces. Academic Press [A subsidiary of Harcourt Brace Jovanovich, Publishers], New York-London (1975). Pure and Applied Mathematics, Vol. 65

  4. 4.

    Ahlfors, L.: Complex analysis, third edn. McGraw-Hill Book Co., New York (1978). An introduction to the theory of analytic functions of one complex variable, International Series in Pure and Applied Mathematics

  5. 5.

    Apel, T.: Interpolation of non-smooth functions on anisotropic finite element meshes. M2AN Math. Model. Numer. Anal. 33(6), 1149–1185 (1999). https://doi.org/10.1051/m2an:1999139.

    MathSciNet  MATH  Article  Google Scholar 

  6. 6.

    Apel, T., Melenk, J.: Interpolation and quasi-interpolation in \(h\)- and \(hp\)-version finite element spaces. In: E. Stein, R. de Borst, T. Hughes (eds.) Encyclopedia of Computational Mechanics, second edn., pp. 1–33. John Wiley & Sons, Chichester, UK (2018). Extended preprint at http://www.asc.tuwien.ac.at/preprint/2015/asc39x2015.pdf.

  7. 7.

    Aurada, M., Feischl, M., Führer, T., Karkulik, M., Praetorius, D.: Energy norm based error estimators for adaptive BEM for hypersingular integral equations. Appl. Numer. Math. 95, 15–35 (2015). https://doi.org/10.1016/j.apnum.2013.12.004.

    MathSciNet  MATH  Article  Google Scholar 

  8. 8.

    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). https://doi.org/10.1137/0725048.

    MathSciNet  MATH  Article  Google Scholar 

  9. 9.

    Băcuţă, C., Li, H., Nistor, V.: Differential operators on domains with conical points: precise uniform regularity estimates. Rev. Roumaine Math. Pures Appl. 62(3), 383–411 (2017)

    MathSciNet  MATH  Google Scholar 

  10. 10.

    Bernardi, C., Dauge, M., Maday, Y.: Polynomials in the Sobolev world (version 2). Tech. Rep. 14, IRMAR (2007). https://hal.archives-ouvertes.fr/hal-00153795.

  11. 11.

    Birman, M., Solomjak, M.: Spektralnaya teoriya samosopryazhennykh operatorov v gilbertovom prostranstve. Leningrad. Univ., Leningrad (1980)

    Google Scholar 

  12. 12.

    Bonito, A., Borthagaray, J.P., Nochetto, R.H., Otárola, E., Salgado, A.J.: Numerical methods for fractional diffusion. Computing and Visualization in Science (2018). https://doi.org/10.1007/s00791-018-0289-y.

  13. 13.

    Bonito, A., Pasciak, J.: Numerical approximation of fractional powers of elliptic operators. Math. Comp. 84(295), 2083–2110 (2015). https://doi.org/10.1090/S0025-5718-2015-02937-8.

    MathSciNet  MATH  Article  Google Scholar 

  14. 14.

    Brändle, C., Colorado, E., de Pablo, A., Sánchez, U.: A concave-convex elliptic problem involving the fractional Laplacian. Proc. Roy. Soc. Edinburgh Sect. A 143(1), 39–71 (2013). https://doi.org/10.1017/S0308210511000175.

    MathSciNet  MATH  Article  Google Scholar 

  15. 15.

    Cabré, X., Sire, Y.: Nonlinear equations for fractional Laplacians II: Existence, uniqueness, and qualitative properties of solutions. Trans. Amer. Math. Soc. 367(2), 911–941 (2015). https://doi.org/10.1090/S0002-9947-2014-05906-0.

    MathSciNet  MATH  Article  Google Scholar 

  16. 16.

    Cabré, X., Tan, J.: Positive solutions of nonlinear problems involving the square root of the Laplacian. Adv. Math. 224(5), 2052–2093 (2010). https://doi.org/10.1016/j.aim.2010.01.025.

    MathSciNet  MATH  Article  Google Scholar 

  17. 17.

    Caffarelli, L., Silvestre, L.: An extension problem related to the fractional Laplacian. Comm. Part. Diff. Eqs. 32(7-9), 1245–1260 (2007). https://doi.org/10.1080/03605300600987306.

    MathSciNet  MATH  Article  Google Scholar 

  18. 18.

    Caffarelli, L., Stinga, P.: Fractional elliptic equations, Caccioppoli estimates and regularity. Ann. Inst. H. Poincaré Anal. Non Linéaire 33(3), 767–807 (2016). https://doi.org/10.1016/j.anihpc.2015.01.004.

    MathSciNet  MATH  Article  Google Scholar 

  19. 19.

    Capella, A., Dávila, J., Dupaigne, L., Sire, Y.: Regularity of radial extremal solutions for some non-local semilinear equations. Comm. Partial Differential Equations 36(8), 1353–1384 (2011). https://doi.org/10.1080/03605302.2011.562954.

    MathSciNet  MATH  Article  Google Scholar 

  20. 20.

    Cartan, H.: Théorie élémentaire des fonctions analytiques d’une ou plusieurs variables complexes. Avec le concours de Reiji Takahashi. Enseignement des Sciences. Hermann, Paris (1961)

    Google Scholar 

  21. 21.

    Chen, X., Zeng, F., Karniadakis, G.: A tunable finite difference method for fractional differential equations with non-smooth solutions. Comput. Methods Appl. Mech. Engrg. 318, 193–214 (2017). https://doi.org/10.1016/j.cma.2017.01.020.

    MathSciNet  Article  Google Scholar 

  22. 22.

    Costabel, M., Dauge, M.: General edge asymptotics of solutions of second-order elliptic boundary value problems. I, II. Proc. Roy. Soc. Edinburgh Sect. A 123(1), 109–155, 157–184 (1993). https://doi.org/10.1017/S0308210500021272.

  23. 23.

    Costabel, M., Dauge, M., Nicaise, S.: Analytic regularity for linear elliptic systems in polygons and polyhedra. Math. Meths. Appl. Sci. 22(8) (2012)

  24. 24.

    D’Elia, M., Gunzburger, M.: The fractional Laplacian operator on bounded domains as a special case of the nonlocal diffusion operator. Comput. Math. Appl. 66(7), 1245 – 1260 (2013). https://doi.org/10.1016/j.camwa.2013.07.022. http://www.sciencedirect.com/science/article/pii/S0898122113004707.

  25. 25.

    DeVore, R., Lorentz, G.: Constructive Approximation. Springer Verlag. Berlin (1993)

    Google Scholar 

  26. 26.

    NIST Digital Library of Mathematical Functions. http://dlmf.nist.gov/, Release 1.0.19 of 2018-06-22. http://dlmf.nist.gov/. F. W. J. Olver, A. B. Olde Daalhuis, D. W. Lozier, B. I. Schneider, R. F. Boisvert, C. W. Clark, B. R. Miller and B. V. Saunders, eds.

  27. 27.

    Duoandikoetxea, J.: Fourier analysis, Graduate Studies in Mathematics, vol. 29. American Mathematical Society, Providence, RI (2001). Translated and revised from the 1995 Spanish original by David Cruz-Uribe

  28. 28.

    Fabes, E., Kenig, C., Serapioni, R.: The local regularity of solutions of degenerate elliptic equations. Comm. Part. Diff. Eqs. 7(1), 77–116 (1982). https://doi.org/10.1080/03605308208820218.

    MathSciNet  MATH  Article  Google Scholar 

  29. 29.

    Gaspoz, F., Heine, C.J., Siebert, K.: Optimal grading of the newest vertex bisection and \(H^1\)-stability of the \(L_2\)-projection. IMA J. Numer. Anal. 36(3), 1217–1241 (2016). https://doi.org/10.1093/imanum/drv044.

    MathSciNet  MATH  Article  Google Scholar 

  30. 30.

    Gaspoz, F., Morin, P.: Convergence rates for adaptive finite elements. IMA J. Numer. Anal. 29(4), 917–936 (2009). https://doi.org/10.1093/imanum/drn039.

    MathSciNet  MATH  Article  Google Scholar 

  31. 31.

    Gold́shtein, V., Ukhlov, A.: Weighted Sobolev spaces and embedding theorems. Trans. Amer. Math. Soc. 361(7), 3829–3850 (2009). https://doi.org/10.1090/S0002-9947-09-04615-7.

  32. 32.

    Gui, W., Babuška, I.: The \(h,\;p\) and \(h\)-\(p\) versions of the finite element method in \(1\) dimension. II. The error analysis of the \(h\)- and \(h\)-\(p\) versions. Numer. Math. 49(6), 613–657 (1986). https://doi.org/10.1007/BF01389734.

    MathSciNet  MATH  Article  Google Scholar 

  33. 33.

    Harbrecht, H., Peters, M., Siebenmorgen, M.: Combination technique based \(k\)-th moment analysis of elliptic problems with random diffusion. J. Comput. Phys. 252, 128–141 (2013). https://doi.org/10.1016/j.jcp.2013.06.013.

    MathSciNet  MATH  Article  Google Scholar 

  34. 34.

    Khoromskij, B., Melenk, J.: Boundary concentrated finite element methods. SIAM J. Numer. Anal. 41(1), 1–36 (2003)

    MathSciNet  MATH  Article  Google Scholar 

  35. 35.

    Kufner, A., Opic, B.: How to define reasonably weighted Sobolev spaces. Comment. Math. Univ. Carolin. 25(3), 537–554 (1984)

    MathSciNet  MATH  Google Scholar 

  36. 36.

    Landkof, N.: Foundations of modern potential theory. Springer-Verlag, New York-Heidelberg (1972). Translated from the Russian by A. P. Doohovskoy, Die Grundlehren der mathematischen Wissenschaften, Band 180

  37. 37.

    Lions, J.L., Magenes, E.: Non-homogeneous boundary value problems and applications. Vol. I. Springer-Verlag, New York (1972)

    Google Scholar 

  38. 38.

    Lynch, R.E., Rice, J.R., Thomas, D.H.: Direct solution of partial difference equations by tensor product methods. Numer. Math. 6, 185–199 (1964). https://doi.org/10.1007/BF01386067.

    MathSciNet  MATH  Article  Google Scholar 

  39. 39.

    McLean, W.: Strongly elliptic systems and boundary integral equations. Cambridge University Press, Cambridge (2000)

    Google Scholar 

  40. 40.

    Meidner, D., Pfefferer, J., Schürholz, K., Vexler, B.: \(hp\)-finite elements for fractional diffusion. SIAM Journal on Numerical Analysis 56(4), 2345–2374 (2018). https://doi.org/10.1137/17M1135517.

    MathSciNet  MATH  Article  Google Scholar 

  41. 41.

    Melenk, J.: On the robust exponential convergence of \(hp\) finite element method for problems with boundary layers. IMA J. Numer. Anal. 17(4), 577–601 (1997). https://doi.org/10.1093/imanum/17.4.577.

    MathSciNet  MATH  Article  Google Scholar 

  42. 42.

    Melenk, J.: \(hp\)-finite element methods for singular perturbations, Lecture Notes in Mathematics, vol. 1796. Springer-Verlag, Berlin (2002). https://doi.org/10.1007/b84212.

  43. 43.

    Melenk, J., Schwab, C.: \(hp\) FEM for reaction-diffusion equations. I. Robust exponential convergence. SIAM J. Numer. Anal. 35(4), 1520–1557 (1998). https://doi.org/10.1137/S0036142997317602.

    MathSciNet  MATH  Article  Google Scholar 

  44. 44.

    Melenk, J., Schwab, C.: Analytic regularity for a singularly perturbed problem. SIAM J. Math. Anal. 30(2), 379–400 (1999). https://doi.org/10.1137/S0036141097317542.

    MathSciNet  MATH  Article  Google Scholar 

  45. 45.

    Miller, K., Samko, S.: Completely monotonic functions. Integral Transform. Spec. Funct. 12(4), 389–402 (2001). https://doi.org/10.1080/10652460108819360.

    MathSciNet  MATH  Article  Google Scholar 

  46. 46.

    Muckenhoupt, B.: Weighted norm inequalities for the Hardy maximal function. Trans. Amer. Math. Soc. 165, 207–226 (1972)

    MathSciNet  MATH  Article  Google Scholar 

  47. 47.

    Müller, F., Schötzau, D., Schwab, C.: Symmetric interior penalty discontinuous Galerkin methods for elliptic problems in polygons. SIAM J. Numer. Anal. 55(5), 2490–2521 (2017). https://doi.org/10.1137/17M1120634.

    MathSciNet  MATH  Article  Google Scholar 

  48. 48.

    Nochetto, R., Otárola, E., Salgado, A.: A PDE approach to fractional diffusion in general domains: a priori error analysis. Found. Comput. Math. 15(3), 733–791 (2015). https://doi.org/10.1007/s10208-014-9208-x.

    MathSciNet  MATH  Article  Google Scholar 

  49. 49.

    Nochetto, R., Otárola, E., Salgado, A.: Piecewise polynomial interpolation in Muckenhoupt weighted Sobolev spaces and applications. Numer. Math. 132(1), 85–130 (2016). https://doi.org/10.1007/s00211-015-0709-6.

    MathSciNet  MATH  Article  Google Scholar 

  50. 50.

    Nochetto, R., Veeser, A.: Primer of adaptive finite element methods. In: Multiscale and adaptivity: modeling, numerics and applications, Lecture Notes in Math., vol. 2040, pp. 125–225. Springer, Heidelberg (2012). https://doi.org/10.1007/978-3-642-24079-9.

  51. 51.

    Olshanskii, M., Reusken, A.: On the convergence of a multigrid method for linear reaction-diffusion problems. Computing 65(3), 193–202 (2000). https://doi.org/10.1007/s006070070006.

    MathSciNet  MATH  Article  Google Scholar 

  52. 52.

    Otárola, E.: A PDE approach to numerical fractional diffusion. Ph.D. thesis, University of Maryland, College Park (2014)

  53. 53.

    Roos, H.G., Stynes, M., Tobiska, L.: Numerical methods for singularly perturbed differential equations, Springer Series in Computational Mathematics, vol. 24. Springer-Verlag, Berlin (1996). https://doi.org/10.1007/978-3-662-03206-0. Convection-diffusion and flow problems

  54. 54.

    Sauter, S., Schwab, C.: Boundary element methods, Springer Series in Computational Mathematics, vol. 39. Springer-Verlag, Berlin (2011). https://doi.org/10.1007/978-3-540-68093-2. Translated and expanded from the 2004 German original

  55. 55.

    Schneider, R.: Multiskalen- und Wavelet-Matrixkompression. Advances in Numerical Mathematics. B. G. Teubner, Stuttgart (1998). https://doi.org/10.1007/978-3-663-10851-1. Analysisbasierte Methoden zur effizienten Lösung großer vollbesetzter Gleichungssysteme. [Analysis-based methods for the efficient solution of large nonsparse systems of equations]

  56. 56.

    Schneider, R., Reichmann, O., Schwab, C.: Wavelet solution of variable order pseudodifferential equations. Calcolo 47(2), 65–101 (2010). https://doi.org/10.1007/s10092-009-0012-y.

    MathSciNet  MATH  Article  Google Scholar 

  57. 57.

    Schötzau, D., Schwab, C.: Exponential convergence for \(hp\)-version and spectral finite element methods for elliptic problems in polyhedra. M3AS 25(9), 1617–1661 (2015)

  58. 58.

    Schötzau, D., Schwab, C.: Exponential convergence of hp-fem for elliptic problems in polyhedra: Mixed boundary conditions and anisotropic polynomial degrees. Journ. Found. Comput. Math. 18(3), 595–660 (2018). https://doi.org/10.1007/s10208-017-9349-9.

    MathSciNet  MATH  Article  Google Scholar 

  59. 59.

    Schwab, C.: \(p\)- and \(hp\)-finite element methods. Numerical Mathematics and Scientific Computation. The Clarendon Press, Oxford University Press, New York (1998). Theory and applications in solid and fluid mechanics

  60. 60.

    Schwab, C., Suri, M.: The \(p\) and \(hp\) versions of the finite element method for problems with boundary layers. Math. Comp. 65(216), 1403–1429 (1996). https://doi.org/10.1090/S0025-5718-96-00781-8.

    MathSciNet  MATH  Article  Google Scholar 

  61. 61.

    Schwab, C., Suri, M., Xenophontos, C.: The \(hp\) finite element method for problems in mechanics with boundary layers. Comput. Methods Appl. Mech. Engrg. 157(3-4), 311–333 (1998). https://doi.org/10.1016/S0045-7825(97)00243-0. Seventh Conference on Numerical Methods and Computational Mechanics in Science and Engineering (NMCM 96) (Miskolc)

  62. 62.

    Scott, L., Zhang, S.: Finite element interpolation of nonsmooth functions satisfying boundary conditions. Math. Comp. 54(190), 483–493 (1990). https://doi.org/10.2307/2008497.

    MathSciNet  MATH  Article  Google Scholar 

  63. 63.

    Stinga, P., Torrea, J.: Extension problem and Harnack’s inequality for some fractional operators. Comm. Partial Differential Equations 35(11), 2092–2122 (2010). https://doi.org/10.1080/03605301003735680.

    MathSciNet  MATH  Article  Google Scholar 

  64. 64.

    Tartar, L.: An introduction to Sobolev spaces and interpolation spaces, Lecture Notes of the Unione Matematica Italiana, vol. 3. Springer, Berlin (2007)

  65. 65.

    Turesson, B.: Nonlinear potential theory and weighted Sobolev spaces, Lecture Notes in Mathematics, vol. 1736. Springer-Verlag, Berlin (2000). https://doi.org/10.1007/BFb0103908.

Download references

Author information



Corresponding author

Correspondence to Enrique Otárola.

Additional information

This paper is dedicated to Professor Ivo Babuška on the occasion of his 90th birthday.

The results in this paper were obtained when the authors met at the MFO Oberwolfach during the WS 1711 in March 2017. The research of RHN was supported in part by NSF Grants DMS-1109325 and DMS-1411808. The research of EO was supported in part by CONICYT through Project FONDECYT 3160201. The research of AJS was supported in part by NSF Grant DMS-1418784. The research of JMM was supported by the Austrian Science Fund (FWF) Project F 65. The research of CS was supported by the Swiss National Science Foundation (SNSF) under Grant No. 159940.

Communicated by Philippe G. CIARLET.


A Proof of Lemma 11

We will only show (5.53), (5.54) as the estimates (5.55), (5.56) are proved using similar arguments; see, for instance, the proof of [6, Theorem 8]. We distinguish between the first element \(I_1\), the terminal element \(I_M\), and the remaining ones. We write \(h_i = |I_i|\). We simplify the exposition by assuming \(X = {{\mathbb {R}}}\). It is convenient to define, for each interval \(I_i\), \(i=2,\ldots ,M\), the quantity \(C_i\) by

$$\begin{aligned} C_i^2:= \sum _{\ell =1}^\infty (2 K_u)^{-\ell } \frac{1}{\ell !^2} \Vert u^{(\ell )}\Vert ^2_{L^2(\omega _{\alpha + 2\ell -2\beta ,\gamma }, I_i)}. \end{aligned}$$

We observe that, since \(u \in {{\mathscr {B}}}_{\beta ,\gamma }^1(C_u,K_u)\),

$$\begin{aligned} \sum _{i=2}^M C_i^2 \le 2 C_u^2, \end{aligned}$$

where we recall that the space \({{\mathscr {B}}}_{\beta ,\gamma }^1(C_u,K_u)\) corresponds to a class of analytic functions and is defined as in (5.51). We begin the proof with an auxiliary result about linear interpolation on the reference element.

Lemma 15

(linear interpolant) Let X be a Hilbert space, \({\widehat{K}} = (0,1)\), and let \(\widetilde{\pi }_1\) be the linear interpolant in the points 1 / 2, 1. Let \(\alpha > -1\) and \(\delta \le 1\). Then, for \(u \in C((0,1];X)\) and provided the terms on the right-hand side are finite, we have

$$\begin{aligned} \int _{{\widehat{K}}} y^\alpha \Vert u - \widetilde{\pi }_1 u\Vert _X^2\, \, {\mathrm{d}}y&\lesssim \int _{{\widehat{K}}} y^{\alpha +2\delta } \Vert u^\prime \Vert _X^2\, \, {\mathrm{d}}y , \end{aligned}$$
$$\begin{aligned} \int _{{\widehat{K}}} y^\alpha \Vert (u - \widetilde{\pi }_1 u)^\prime \Vert _X^2\, \, {\mathrm{d}}y&\lesssim \int _{{\widehat{K}}} y^{\alpha +2\delta } \Vert u^{\prime \prime }\Vert _X^2\, \, {\mathrm{d}}y, \end{aligned}$$

where the hidden constant is independent of u.


For notational simplicity, we will prove the lemma only for the case \(X = {{\mathbb {R}}}\).

We begin with the proof of (A.3). Since \((u - \widetilde{\pi }_1 u)(1) = 0\), we have, for \(y \in {\widehat{K}}\),

$$\begin{aligned} (u - \widetilde{\pi }_1 u)(y) = \int _1^y (u - \widetilde{\pi }_1 u)^\prime (t)\,\, {\mathrm{d}}t, \end{aligned}$$

so that

$$\begin{aligned} \int _0^1 y^\alpha |u - \widetilde{\pi }_1 u|^2\, \, {\mathrm{d}}y\le & {} 2 \int _0^1 y^\alpha \left| \int _{y}^1 |u^\prime (t)|\,\, {\mathrm{d}}t\right| ^2 \, {\mathrm{d}}y\\&\quad +\, 2 \int _0^1 y^\alpha \left| \int _{y}^1 |(\widetilde{\pi }_1 u)^\prime (t)|\,\, {\mathrm{d}}t\right| ^2 \, {\mathrm{d}}y. \end{aligned}$$

From Hardy’s inequality (e.g., [25, Chapter 2, Theorem 3.1]), we infer

$$\begin{aligned} \int _0^1 y^\alpha \left| \int _{y}^1 |u^\prime (t)|\,\, {\mathrm{d}}t\right| ^2\, \, {\mathrm{d}}y \le 4(\alpha +1)^{-2} \int _0^1 y^{\alpha +2} |u^\prime (y)|^2\, \, {\mathrm{d}}y. \end{aligned}$$

From \( (\widetilde{\pi }_1 u)^\prime = 2\int _{1/2}^1 u^\prime (t)\,\, {\mathrm{d}}t \), we obtain \( | (\widetilde{\pi }_1 u)^\prime |^2 \le C \int _{1/2}^1 t^{\alpha +2\delta } |u^\prime (t)|^2\,\, {\mathrm{d}}t \) and therefore, in view of \(\alpha > -1\), the estimate

$$\begin{aligned} \int _0^1 y^\alpha | (\widetilde{\pi }_1 u)'|^2\, \, {\mathrm{d}}y \lesssim \int _0^1 y^{\alpha +2} |u^\prime (y)|^2\, \, {\mathrm{d}}y. \end{aligned}$$

This concludes the proof of (A.3) for the case \(\delta = 1\). Since the integration range is \(y \in \widehat{K} = (0,1)\), we may replace \(y^{\alpha +2}\) by \(y^{\alpha +2 \delta }\).

Next, we prove (A.4) and assume that the right-hand side of (A.4) is finite. Again, it suffices to consider the limiting case \(\delta = 1\) and use Hardy’s inequality. We mention in passing that (A.5) actually shows that \(u \in C^1((0,1]; X)\). We write

$$\begin{aligned} (u - \widetilde{\pi }_1 u)^\prime (y)&= u^\prime (y) - 2 \int _{1/2}^1 u^\prime (t)\, \, {\mathrm{d}}t = 2\int _{1/2}^1 \left( u^\prime (y) - u^\prime (t) \right) \,\, {\mathrm{d}}t \nonumber \\&= 2\int _{1/2}^1 \int _{t}^y u^{\prime \prime }(\tau )\,\, {\mathrm{d}}\tau \,\, {\mathrm{d}}t . \end{aligned}$$


$$\begin{aligned} \int _{0}^1 y^\alpha |(u -&\widetilde{\pi }_1 u)^\prime (y)|^2\,\, {\mathrm{d}}y = 4 \int _{0}^1 y^\alpha \left| \int _{1/2}^1 \int _{t}^y u^{\prime \prime }(\tau )\,\, {\mathrm{d}}\tau \, \, {\mathrm{d}}t \right| ^2\, \, {\mathrm{d}}y \\&\le 2 \int _{1/2}^1 \int _{0}^1 y^\alpha \left| \int _{t}^y |u^{\prime \prime }(\tau )|\,\, {\mathrm{d}}\tau \right| ^2 \, {\mathrm{d}}y \, {\mathrm{d}}t\\&\lesssim \int _{1/2}^1 \int _{0}^1 y^\alpha \left[ \left| \int _{y}^1 |u^{\prime \prime }(\tau )|\,\, {\mathrm{d}}\tau \right| ^2 + \left| \int _{t}^1 |u^{\prime \prime }(\tau )|\,\, {\mathrm{d}}\tau \right| ^2 \right] \, {\mathrm{d}}y \, {\mathrm{d}}t\\&\lesssim \int _{0}^1 y^{\alpha +2} |u^{\prime \prime }(y)|^2\,\, {\mathrm{d}}y + \int _{1/2}^1 y^{\alpha +2} |u^{\prime \prime }(y)|^2\, \, {\mathrm{d}}y, \end{aligned}$$

where in the last step we applied Hardy’s inequality. Lemma 15 is thus proved. \(\square \)

With Lemma 15, we can estimate the error on the first element \(I_1\) as follows: Scaling the estimate (A.3) gives

$$\begin{aligned} \Vert u - \pi ^{\mathbf {r}}_{y} u\Vert _{L^2(\omega _{\alpha ,0},I_1)} \le C h_1^{\beta } \Vert u^\prime \Vert _{L^2(\omega _{\alpha +2-2\beta ,0},I_1)}, \end{aligned}$$

and the assumption allows us to insert the weight \(e^{\gamma y}\) on both sides of (A.6).

We now bound the error contributions from elements away from the origin, i.e., on the elements \(I_i\), \(i=2,\ldots ,M\). These elements satisfy \(h_i \sigma /(1-\sigma )= {{\mathrm{dist}}}(I_i,0) \). For \(I_i = (y_{i-1}, y_i)\), the pull-back \({\widehat{u}}_i: = u|_{I_i} \circ F_{I_i}\) satisfies

$$\begin{aligned}&\Vert {\widehat{u}}_i^{(\ell +1)}\Vert ^2_{L^2(-1,1)} =(h_i/2)^{-1+2(\ell +1)} \Vert u^{(\ell +1)}\Vert ^2_{L^2(I_i)} \\&\quad \le (h_i/2)^{-1+2(\ell +1)} e^{-\gamma y_{i-1}} \max _{y \in I_i} y^{-\alpha -2(\ell +1)+2\beta } \Vert u^{(\ell +1)}\Vert ^2_{L^2(\omega _{\alpha +2(\ell +1)-2\beta ,\gamma },I_i)} \\&\quad \lesssim e^{-\gamma y_{i-1}} h_i^{-1 + 2 (\ell +1)} h_i^{-\alpha - 2 (\ell +1)+2\beta } (2 \sigma /(1-\sigma ))^{-2(\ell +1)} C_i^2 (2 K_u)^{2(\ell +1)} (\ell +1)!^2, \end{aligned}$$

where in the last step we have used (A.1). The operator \({\widehat{\varPi }}_r\) given by Lemma 10 then yields the existence of a \(b>0\) that depends solely on \(K_u\) and \(\sigma \) for which

$$\begin{aligned} \Vert {\widehat{u}} - {\widehat{\varPi }}_{{\mathbf {r}}_i}{\widehat{u}}\Vert _{L^2(-1,1)} \lesssim C_i e^{-\gamma y_{i-1}} e^{-b {\mathbf {r}}_i} h_i^{-(1+\alpha )/2 + \beta }. \end{aligned}$$

Scaling back to \(I_i\) and using again \(h_i \sim {\text {*}}{dist}(I_i,0)\) yields

$$\begin{aligned} \Vert u - \pi ^{{\mathbf {r}}}_{y} u\Vert ^2_{L^2(\omega _{\alpha ,\gamma },I_i)} \le C h_i^{2\beta } C_i^2 e^{-2 b {\mathbf {r}}_i}. \end{aligned}$$

Summation over i and taking the slope of the linear degree vector sufficiently large (see, for instance, the proof of [6, Theorem 8] for details) give

for suitable \(b' > 0\). Combining this with (A.6) gives the desired (5.53).

It remains to prove (5.54). We begin with a preparatory result.

Lemma 16

(exponential decay) Let X be a Hilbert space, and let \(\delta \in {{\mathbb {R}}}\), \(\gamma > 0\), . Then, the following holds for in items (i), (ii) and for in items (iii), (iv) with implied constants depending solely on \(\delta \), \(\gamma \), and :

  1. (i)

    If \(\lim _{y \rightarrow \infty } u(y) = 0\) and , then

  2. (ii)

    If then \(\lim _{y \rightarrow \infty } u(y) = 0.\)

  3. (iii)

    If \(\lim _{y \rightarrow \infty } u^{(j)}(y)=0\) for \(j=0\), 1 and , then

  4. (iv)

    If , then \(\lim _{y \rightarrow \infty } u(y) = \lim _{y\rightarrow \infty } u^\prime (y) = 0\).


We will only prove items (i) and (ii) as the remaining two are proved by similar arguments.

We begin the proof with the following observation: There is a constant \(c>0\) (that depends only on \(\delta \), , and \(\gamma \)) such that for


For \(\delta \ge 0\), this is immediate. For \(\delta < 0\), one integrates by parts once to discover that the leading-order asymptotics (as ) of the integral is .

We now proceed with the proof of (A.7): Since \(\gamma > 0\), we can write

and (A.7) follows from (A.9). The assertion of item (ii) follows by a similar argument, starting from \(u(y) = u(\eta ) + \int _{\eta }^y u^\prime (t)\,\, {\mathrm{d}}t\), squaring, multiplying by \(\exp (-\gamma \eta )\), and integrating in \(\eta \) from y to \(\infty \). \(\square \)

To prove (5.54) we have to estimate . Lemma 16, (i) shows


Since is obtained from \(\pi ^{\mathbf {r}}_{y}\) by a correction on the terminal element \(I_M\), the desired (5.54) follows easily from (A.10), if we recall that .

B Analysis of the Decoupling Eigenvalue Problem

Lemma 17

(weighted Poincaré) Let and \(\alpha \in (-1,1)\). Then, for with there holds



From we get . Hence, for ,

which finishes the proof. \(\square \)

Lemma 18

(eigenvalue upper bound) Let and \(\alpha \in (-1,1)\). Assume that \((v,\mu )\) satisfy


Then, .


We compute, using Lemma 17

which finishes the proof. \(\square \)

We also need lower bounds for eigenvalues.

Lemma 19

(eigenvalue lower bound) Let \(\alpha > -1\). Let \({\mathscr {G}}^M\) be an arbitrary mesh on with the property that for all elements \(I_i\), \(i=2,\ldots ,M\), not abutting \(y=0\) there holds \(|I_i| \le C_{geo} {{\mathrm{dist}}}(I_i,0)\). Let be a subspace of the space of piecewise polynomials of degree q on \({\mathscr {G}}^M\). Then, with \(h_{min}\) denoting the smallest element size,


where the hidden constant depends solely on \(C_{geo}\) and \(\alpha \).


We emphasize that the condition \(h_i \le C_{geo} {\text {*}}{dist}(I_i,0)\) is satisfied for all meshes where neighboring elements have comparable size. We also remark that (slightly) sharper estimates (in the dependence on the polynomial degree q) are possible on geometric meshes with linear degree vector. We write \(h_i = |I_i|\). We will use the polynomial inverse estimate (B.6). For the first element \(I_1 = (0,y_1)\), we calculate for \(v \in V_h\) and its pull-back \(\widehat{v}:= v|_{I_1} \circ F_{I_1}\)

$$\begin{aligned} \Vert v^\prime \Vert ^2_{L^2(y^\alpha ,{\widehat{K}})}&= (h_1/2)^{\alpha +1-2} \int _{-1}^1 (1+y)^\alpha |\widehat{v}^\prime (y)|^2\,\, {\mathrm{d}}y \nonumber \\&{\mathop {\lesssim }\limits ^{(B.6)}} h_1^{\alpha +1-2} q^4 \int _{-1}^1 (1+y)^\alpha |\widehat{v}(y)|^2\,\, {\mathrm{d}}y \sim h_1^{-2} q^4 \Vert v\Vert ^2_{L^2(y^\alpha ,I_1)}. \end{aligned}$$

For the remaining elements \(I_i\), we exploit that the assumption \(h_i \le C_{geo} {{\mathrm{dist}}}(I_i,0)\) ensures that the weight is slowly varying within them, i.e.,

$$\begin{aligned} \max _{y \in I_i} y^\alpha \le (1+C_{geo})^{|\alpha |} \min _{y \in I_i} y^\alpha , \qquad i=2,\ldots ,M. \end{aligned}$$

Hence, the polynomial inverse estimate (B.6) (with \(\alpha =\beta = 0\) there) yields by scaling arguments

$$\begin{aligned} \Vert v^\prime \Vert _{L^2(y^\alpha ,I_i)} \le C h_i^{-1} q^2 \Vert v\Vert _{L^2(y^\alpha ,I_i)}. \end{aligned}$$

Combining (B.4), (B.5) yields the result. \(\square \)

Lemma 20

(polynomial inverse estimate) Let \({\widehat{K}} = (-1,1)\). For \(\alpha \), \(\beta > -1\), there is \(C_{\alpha ,\beta }>0\) such that for all \(q \in {{\mathbb {N}}}_0\) and all \(w \in {{\mathbb {P}}}_q({\widehat{K}})\):

$$\begin{aligned} \int _{-1}^1 (1+y)^\alpha (1-y)^\beta |w^\prime (y)|^2\, {\mathrm{d}}y&\le C_{\alpha ,\beta } q^4 \int _{-1}^1 (1+y)^\alpha (1-y)^\beta |w(y)|^2\, {\mathrm{d}}y. \end{aligned}$$


Step 1: We assert (B.6) for \(\alpha = \beta \). From [10, Chap. III, Props. 6.1, 6.3], we get for \(w \in {{\mathbb {P}}}_q({\widehat{K}})\)

$$\begin{aligned} \int _{-1}^1 (1-y^2)^\alpha |w^\prime (y)|^2\, {\mathrm{d}}y&{\mathop {\lesssim }\limits ^{[{10, Chap.~{III}, Prop.~{6.1}}]}} q^2 \int _{-1}^1 (1-y^2)^{\alpha +1} |w^\prime (y)|^2\, {\mathrm{d}}y \\&{\mathop {\lesssim }\limits ^{[10, Chap.~{III}, Prop.~{6.3}]}} q^4 \int _{-1}^1 (1-y^2)^{\alpha } |w(y)|^2\, {\mathrm{d}}y. \end{aligned}$$

Simple scaling arguments imply for arbitrary (fixed) finite intervals (ab) and \(\alpha ' > -1\) for all \(w \in {{\mathbb {P}}}_q(a,b)\):

$$\begin{aligned} \int _{a}^b (y - a)^{\alpha '} (b-y)^{\alpha '} |w^\prime (y)|^2\, {\mathrm{d}}y \lesssim q^4 \int _{a}^b (y - a)^{\alpha '} (b-y)^{\alpha '} |w(y)|^2\, {\mathrm{d}}y . \end{aligned}$$

Step 2: We show (B.6) for \((\alpha ,\beta ) = (\alpha ,0)\). (By symmetry, this also shows the case \((\alpha ,\beta ) = (0,\beta )\)). Since (B.7) implies for \(w \in {{\mathbb {P}}}_q({\widehat{K}})\)

$$\begin{aligned} \int _{0}^1 (1+y)^\alpha |w^\prime (y)|^2\, {\mathrm{d}}y \lesssim \int _{0}^1 |w^\prime (y)|^2\, {\mathrm{d}}y \lesssim q^4 \int _{0}^1 |w(y)|^2\, {\mathrm{d}}y, \end{aligned}$$

it suffices to prove the bound \(\int _{-1}^0 (1+y)^\alpha |w^\prime (y)|^2\, {\mathrm{d}}y \lesssim q^4 \int _{-1}^1 (1+y)^\alpha |w(y)|^2\, {\mathrm{d}}y\). To that end, define \(\widetilde{w}(y):= (1-y) w(y) \in {{\mathbb {P}}}_{q+1}({\widehat{K}})\) and note \(\widetilde{w}^\prime (y) = (1-y) w^\prime (y) - w(y)\). Then,

$$\begin{aligned}&\int _{-1}^0(1+y)^\alpha |w^\prime (y)|^2\, {\mathrm{d}}y \le \int _{-1}^0(1+y)^\alpha |w^\prime (y) (1-y)|^2\, {\mathrm{d}}y \\&\quad \le 2 \int _{-1}^0(1+y)^\alpha \left( |\widetilde{w}^\prime (y)|^2 + |w(y)|^2\right) \, {\mathrm{d}}y \\&\quad {\mathop {\lesssim }\limits ^{(B.7)}} (q+1)^4 \int _{-1}^1 (1+y)^\alpha \left( (1-y)^\alpha |\widetilde{w}(y)|^2 + |w(y)|^2\right) \, {\mathrm{d}}y \\&\quad \lesssim (q+1)^4 \int _{-1}^1 (1+y)^\alpha \left( (1-y)^\alpha (1-y)^2 |w(y)|^2 + |w(y)|^2\right) \, {\mathrm{d}}y. \end{aligned}$$

Since \(\alpha > -1\), we have \((1- y)^{2+\alpha } \lesssim 1\), which allows us to conclude the proof of the case \(\beta = 0\) in (B.6).

Step 3: For arbitrary \(\alpha \), \(\beta > -1\), we use (scaled versions of) Step 2:

$$\begin{aligned}&\int _{-1}^1 (1+y)^\alpha (1-y)^\beta |w^\prime (y)|^2\, {\mathrm{d}}y \lesssim \int _{-1}^0 (1+y)^\alpha |w^\prime (y)|^2\, {\mathrm{d}}y + \int _{0}^1 (1-y)^\beta |w^\prime (y)|^2\, {\mathrm{d}}y\\&\quad {\mathop {\lesssim }\limits ^{\text {Step 2}}} q^4 \int _{-1}^0 (1+y)^\alpha |w(y)|^2\, {\mathrm{d}}y + q^4 \int _{0}^1 (1-y)^\beta |w(y)|^2\, {\mathrm{d}}y\\&\quad \lesssim q^4 \int _{-1}^1 (1+y)^\alpha (1-y)^\beta |w(y)|^2\, {\mathrm{d}}y. \end{aligned}$$

This concludes the proof. \(\square \)

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Banjai, L., Melenk, J.M., Nochetto, R.H. et al. Tensor FEM for Spectral Fractional Diffusion. Found Comput Math 19, 901–962 (2019). https://doi.org/10.1007/s10208-018-9402-3

Download citation


  • Fractional diffusion
  • Nonlocal operators
  • Weighted Sobolev spaces
  • Regularity estimates
  • Finite elements
  • Anisotropic hp-refinement
  • Corner refinement
  • Sparse grids
  • Exponential convergence

Mathematics Subject Classification

  • 26A33
  • 65N12
  • 65N30