Abstract
We consider the model problem of mixing of passive tracers by an incompressible viscous fluid. Addressing questions of optimal control in realistic geometric settings or alternatively the design of fluid-confining geometries that successfully effect mixing requires a meaningful norm in which to quantify mixing that is also suitable for easy and efficient computation (as is needed, e.g., for use in gradient-based optimization methods). We use the physically inspired reasonable surrogate of a negative index Sobolev norm over the complex fluid mixing domain \(\Omega\), a task which could be seen as computationally expensive since it requires the computation of an eigenbasis for \(L^{2}(\Omega )\) by definition. Instead, we compute a representant of the scalar concentration field in an appropriate Sobolev space in order to obtain an equivalent definition of the Sobolev surrogate norm. The representant, in turn, can be computed to high-order accuracy by a Padé approximation to certain fractional pseudo-differential operators, which naturally leads to a sequence of elliptic problems with an inhomogeneity related to snapshots of the time-varying concentration field. Fast and accurate potential theoretic methods are used to efficiently solve these problems, with rapid per-snapshot mix-norm computation made possible by recent advances in numerical methods for volume potentials. We couple the methodology to existing solvers for Stokes and advection equations to obtain a unified framework for simulating and quantifying mixing in arbitrary fluid domains. We provide numerical results demonstrating the convergence of the new approach as the approximation order is increased.
Similar content being viewed by others
Notes
Note that even in the presence of diffusion (with diffusion constant, \(\kappa > 0\)) it is worthwhile to consider alternative metrics of mixing, since the limit \(\kappa \rightarrow 0\) is singular in a way that impacts mixing studies; see, e.g., [3] for a detailed discussion.
References
Optimal stirring strategies for passive scalar mixing: Lin, Z. Thiffeault, J.-L. Doering. C.R. J. Fluid Mech. 675, 465–476 (2011)
Lunasin, E., Lin, Z., Novikov, A., Mazzucato, A., Doering, C.R.: Optimal mixing and optimal stirring for fixed energy, fixed power, or fixed palenstrophy flows. J. Math. Phys. 53(11),(2012)
Thiffeault, J.-L.: Using multiscale norms to quantify mixing and transport. Nonlinearity 25(2), R1 (2012)
Mathew, G., Mezić, I., Petzold, L.: A multiscale measure for mixing. Physica D 211(1–2), 23–46 (2005)
Eggl, M.F., Schmid, P.J.: Mixing enhancement in binary fluids using optimised stirring strategies. J. Fluid Mech. 889, A24 (2020)
Darbas, M., Darrigrand, E., Lafranche, Y.: Combining analytic preconditioner and Fast Multipole Method for the 3D Helmholtz equation. J. Comput. Phys. 236, 289–316 (2013)
Bonito, A., Pasciak, J.E.: Numerical approximation of fractional powers of elliptic operators. Math. Comput. 84, 2083–2110 (2015)
Duan, B., Lazarov, R.D., Pasciak, J.E.: Numerical approximation of fractional powers of elliptic operators. IMA J. Numer. Anal. 40, 1746–1771 (2020)
Barnett, A., Hassell, A.: Fast computation of high-frequency Dirichlet eigenmodes via spectral flow of the interior Neumann-to-Dirichlet map. Commun. Pure Appl. Math. 67.3, 351–407 (2014)
Gholami, A., Malhotra, D., Sundar, H., Biros, G.: FFT, FMM, or multigrid? A comparative study of state-of-the-art Poisson solvers for uniform and nonuniform grids in the unit cube. SIAM J. Sci. Comput. 38(3), C280–C306 (2016)
McLean, W.: Strongly elliptic systems and boundary integral equations. Cambridge, (2000)
Lions, J.L., Magenes, E.: Non-homogeneous boundary value problems and applications vol. 1. Springer (1972)
Brezinski, C.: Padé-type approximation and general orthogonal polynomials. vol. 50. Springer (1980)
Baker, G.A., Graves-Morris, P.: Padé approximants (second edition). Cambridge University Press, (1996)
Clawpack Development Team. Clawpack software. Version 5.7.1. (2020). https://doi.org/10.5281/zenodo.4025432.http://www.clawpack.org
Mandli, K.T., Ahmadia, A.J., Berger, M., Calhoun, D., George, D.L., Hadjimichael, Y., Ketcheson, D.I., Lemoine, G.I., LeVeque, R.J.: Clawpack: building an open source ecosystem for solving hyperbolic PDEs. PeerJ Computer Science 2, e68 (2016). https://doi.org/10.7717/peerj-cs.68
LeVeque, R.J.: Finite Volume Methods for Hyperbolic Problems. Cambridge University Press, (2002). http://www.clawpack.org/book.html
Schlottke-Lakemper, M., Gassner, G.J., Ranocha, H., Winters, A.R., Chan, J.: Trixi.jl: Adaptive high-order numerical simulations of hyperbolic PDEs in Julia. (2021) https://github.com/trixiframework/Trixi.jl. https://doi.org/10.5281/zenodo.3996439
Kress, R.: Linear Integral Equations. vol. 82. Springer, (1989)
Barnett, A., Wu, B., Veerapaneni, S.: Spectrally accurate quadratures for evaluation of layer potentials close to the boundary for the 2D Stokes and Laplace equations. SIAM J. Sci. Comput. 37(4), B519–B542 (2015)
Anderson, T.G., Zhu, H., Veerapaneni, S.: A fast, high-order scheme for evaluating volume potentials on complex 2D geometries via area-to-line integral conversion and domain mappings. J. Comput. Phys. (2022). https://doi.org/10.1016/j.jcp.2022.111688
Hsiao, G.C., Wendland, W.L.: Boundary Integral Equations. Springer, (2008)
Driscoll, T.A., Hale, N., Trefethen, L.N.: Chebfun Guide. Pafnuty Publications, (2014). http://www.chebfun.org/docs/guide/
Acknowledgements
This research was supported in part through computational resources and services provided by the Advanced Research Computing (ARC) at the University of Michigan. The authors are grateful to the referees for their helpful comments that strengthened the paper.
Funding
We acknowledge support from NSF under grant DMS-2012424.
Author information
Authors and Affiliations
Contributions
All authors designed research and wrote the manuscript. M.B. developed the formulation and T.A. implemented all the algorithms and compiled the results.
Corresponding author
Appendices
Appendix 1. Eigenfunction expansion in circular or annular domains
Circular domain
Let \(\Omega=B(0,a)=\{x(\rho,\theta),\;0\leq\rho<a,\,0\leq\theta<2\pi\}\) be the disk of radius a. The Dirichlet Laplace eigenfunctions for \(\Omega\) are
where \(J_{n}\) is the Bessel function of first kind and integer order n and \(z_{nm}\) (\(m= 1,2,\ldots\)) are the (real, positive) zeros of \(J_{n}\) (the excluded zero \(z=0\) of \(J_{n}\) (\(n\ge 1\)) not producing nonzero eigenfunctions). Setting the normalization constants to \(\gamma _{0m} = \sqrt{\pi }\big [ aJ_{1}(z_{0m}) \big ]^{-1}\) and \(\gamma _{nm} = \sqrt{\pi /2}\big [ aJ_{n+1}(z_{nm}) \big ]^{-1}\) \((n\ge 1)\), the eigenfunctions (28) are \(L^{2}(\Omega )\)-orthonormal and satisfy
As (Dirichlet) Laplace eigenfunctions, the functions (28) define a Hilbert basis of \(L^{2}(\Omega )\), so that any \(c\in L^{2}(\Omega )\) admits the expansion
with \(c_{0m} = \big ( \varphi _{0m},c \big )_{L^{2}(\Omega )}\) and \(c^{(1,2)}_{nm} = \big ( \varphi ^{(1,2)}_{nm}, c \big )_{L^{2}(\Omega )}\). The \(H^{-r}(\Omega )\) norm of c is therefore given by
Annular domain
Let now \(\Omega = B(0, \rho _{1})\setminus \overline{B(0, \rho _{2})}\) be the annulus of internal radius \(\rho _{2}\) and external radius \(\rho _{1}\). The (unnormalized) radial Dirichlet eigenfunctions for \(\Omega\) are given by the expression
so that eigenfunctions are given (analogously to the disc case, and before normalization) via \(\varphi _{nm}^{(1,2)}(\rho , \theta ) = f_{nm}(\rho ) \left\{ \begin{array}{c} \cos (n\theta )\\ \sin (n\theta )\end{array}\right\}\). While the functional form (32) of the eigenfunctions is clearly known, the corresponding annular eigenvalues \(\lambda _{nm} = \zeta _{nm}^{2}\) are required for the basis to be fully determined. We solve for the eigenvalues using a Newton iteration on the eigenvalue equation, with the method bootstrapped using an approximate eigenvalue obtained using the chebfun system [23] (whose values, at least for larger eigenvalues fail to provide adequate accuracy but are still highly useful to start a Newton iteration).
Appendix 2. Derivation of Padé approximations
Let f have the (\(2n-1\))-th order Taylor expansion
about \(X=0\). In particular, we have \(a_{0}=1\) and \(a_{k+1}=a_{k}(2k+ r)/(2k+2)\) for \(f(X)=(1- X)^{-r/2}\). The coefficients of the polynomials
such that \(\Pi _{n}[f]=P_{n-1}/Q_{n}\) (with the adopted normalization \(q_{0}=1\) ensuring uniqueness of \(P_{n-1},\,Q_{n}\)) are found from the linear relations
where \(q_{1},\ldots q_{n}\) solve the n linear equations (a) and \(p_{0},\ldots ,p_{n-1}\) are then given explicitly by relations (b). Numerical experiments for \(f(X)=(1- X)^{-1/2}\) indicate however that the linear system (35a) becomes ill-conditioned for n larger than about 10. We therefore solved (35) using symbolic computation.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Anderson, T.G., Bonnet, M. & Veerapaneni, S. Quantifying mixing in arbitrary fluid domains: a Padé approximation approach. Numer Algor 93, 441–458 (2023). https://doi.org/10.1007/s11075-022-01423-7
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11075-022-01423-7