Despite being one of the most dependable methods used by applied mathematicians and engineers in handling complex systems, the finite element method commits variational crimes. This paper contextualizes the concept of variational crime within a broader account of mathematical practice by explaining the tradeoff between complexity and accuracy involved in the construction of numerical methods. We articulate two standards of accuracy used to determine whether inexact solutions are good enough and show that, despite violating the justificatory principles of one, the finite element method nevertheless succeeds in obtaining its legitimacy from the other.
This is a preview of subscription content, log in to check access.
Buy single article
Instant access to the full article PDF.
Price includes VAT for USA
Subscribe to journal
Immediate online access to all issues from 2019. Subscription will auto renew annually.
This is the net price. Taxes to be calculated in checkout.
There are passing mentions of the finite element method in Peressini (1997), Winsberg (2006), Humphreys (2009), and Morrison (2009) and Bursten (2015), but no sustained discussion of the specific features of this method, or their philosophical implications. Without mentioning finite element methods explicitly, Wilson (e.g., 2013, 2017) has often discussed closely related themes.
Some would say that z is not a solution at all. This is incorrect if, as we do in this paper, one wishes to say that inexact solutions are solutions in some sense.
We added the labels (A)–(C) in the quote for future reference.
We remark that this proof relies on the discrete sampling necessary for computational work, and fails if the computer program is allowed to use interval arithmetic, or otherwise validate its results by faithfully interpolating between samples. This highlights the importance of interpolation that enables the computation of continuous numerical solutions. This also highlights the fact that, even though interval analysis is not a panacea to conceptually analyze the concept of approximation, approaches based on it (e.g., Laymon 1990) can capture this situation quite well.
If we use a continuously differentiable numerical method, such as the continuous explicit Runge-Kutta methods implemented in Matlab, we will then be in a position to compute the residual as well.
There are interesting additional issues. In a recent talk at UBC (where he presented results published in Chen 2017), D. M. Kaufman (Adobe Research and Columbia) pointed out that, in the context of computer graphics and animation, “realistic isn’t the same as accurate.” In some circumstances, such as solving systems with many millions of moving components, detailed accuracy may not be possible—but one can still get a realistic movie by presenting motions that “could have happened.” A good structured backward error can ensure this. There are many kinds of backward error, however, and they do need the proper physical context in order to make sense.
One often sees the maximum mesh diameter used; but with equidistribution of error, one recovers a mean mesh diameter. This does not seem widely known, and we will be pursuing this in future work. For the purposes of this paper, either the maximum mesh width or the mean mesh width, conditioned on equidistribution, can be considered to be going to zero in the asymptotic limit.
Ardourel, V., & Jebeile, J. (2017). On the presumed superiority of analytical solutions over numerical methods. European Journal for Philosophy of Science, 7(2), 201–220.
Bellen, A., & Zennaro, M. (2003). Continuous Runge–Kutta methods for ODEs. In Blah (Ed.) Numerical methods for delay differential equations: Oxford University Press.
Bursten, J. (2015). Surfaces, scales and synthesis: scientific reasoning at the nanoscale. PhD thesis: University of Pittsburgh.
Butcher, J.C. (2016). Numerical methods for ordinary differential equations. Wiley.
Chen, D., Levin, D.I., Matusik, W., Kaufman, D.M. (2017). Dynamics-aware numerical coarsening for fabrication design. ACM Transactions on Graphics, 36(4), 84.
Corless, R.M. (1993). Six, lies, and calculators. The American Mathematical Monthly, 100(4), 344–350.
Corless, R.M., & Fillion, N. (2013). A graduate introduction to numerical methods, from the viewpoint of backward error analysis, (p. 868). New York: Springer.
Corless, R.M., & Fillion, N. (2019). Backward error analysis for perturbation methods. In Algorithms and complexity in mathematics, epistemology, and science (pp. 35–79): Springer.
Courant, R. (1943). Variational methods for the solution of problems of equilibrium and vibrations. Bulletin of the American Mathematical Society, 43, 1–23.
Fillion, N., & Bangu, S. (2015). Numerical methods, complexity, and epistemic hierarchies. Philosophy of Science, 82, 941–955.
Fillion, N., & Corless, R.M. (2014). On the epistemological analysis of modeling and computational error in the mathematical sciences. Synthese, 191, 1451–1467.
Fillion, N., & Moir, R.H.C. (2018). Explanation and abstraction from a backward-error analytic perspective. European Journal for Philosophy of Science, 8 (3), 735–759.
Heyting, A. (1966). After thirty years. In Studies in logic and the foundations of mathematics, (Vol. 44 pp. 194–197): Elsevier.
Higham, N.J. (2002). Accuracy and stability of numerical algorithms, 2nd edn. Philadelphia: SIAM.
Humphreys, P. (2004). Extending ourselves: computational science, empiricism and scientific method. USA: Oxford University Press.
Humphreys, P. (2009). The philosophical novelty of computer simulation methods. Synthese, 169(3), 615–626.
Ilie, S., Söderlind, G., Corless, R.M. (2008). Adaptivity and computational complexity in the numerical solution of ODEs. Journal of Complexity, 24(3), 341–361.
Kahan, W. (1980). Handheld calculator evaluates integrals. Hewlett-Packard Journal, 31(8), 23–32.
Laymon, R. (1990). Computer simulations, idealizations and approximations. In PSA: Proceedings of the biennial meeting of the philosophy of science association, (Vol. 1990 pp. 519–534): Philosophy of Science Association.
Moir, R.H. (2010). Reconsidering backward error analysis for ordinary differential equations. Master’s thesis: The University of Western Ontario.
Morrison, M. (2009). Models, measurement and computer simulation: the changing face of experimentation. Philosophical Studies, 143, 33–57.
Peressini, A. (1997). Troubles with indispensability: applying pure mathematics in physical theory. Philosophia Mathematica, 5(3), 210–227.
Ramsey, J.L. (1992). Towards an expanded epistemology for approximations. In PSA: Proceedings of the biennial meeting of the philosophy of science association, (Vol. 1992 pp. 154–164): Philosophy of Science Association.
Strang, G. (1972). Variational crimes in the finite element method. In Aziz, A. (Ed.) The mathematical foundations of the finite element method with applications to partial differential equations (pp. 689–710): Elsevier.
Strang, G. (1973). Piecewise polynomials and the finite element method. Bulletin of the American Mathematical Society, 79(6), 1128–1137.
Strang, G., & Fix, G. (1973). Analysis of the finite element method. Prentice-Hall.
Turing, A.M. (1952). The chemical basis of morphogenesis. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences, 237(641), 37–72.
Werschulz, A.G. (1991). The computational complexity of differential and integral equations: an information-based approach. Oxford University Press.
Wilkinson, J.H. (1971). Modern error analysis. SIAM Review, 13(4), 548–568.
Wilson, M. (1998). Mechanics, classical. In Craig, E. (Ed.) Routledge encyclopedia of philosophy. Routledge.
Wilson, M. (2013). Enlarging one’s stall or how did all of these sets get in here? Philosophia Mathematica, 21, 2.
Wilson, M. (2017). Physics avoidance: and other essays in conceptual strategy. Oxford University Press.
Winsberg, E. (2006). Handshaking your way to the top: simulation at the nanoscale. In Lenhard, J., Küppers, G., Shinn, T. (Eds.) Simulation: pragmatic constructions of reality (pp. 139–151): Springer.
We would like to thank [removed for review]. We would also like to thank the two reviewers who made valuable suggestions.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
About this article
Cite this article
Fillion, N., Corless, R.M. Concepts of Solution and the Finite Element Method: a Philosophical Take on Variational Crimes. Philos. Technol. (2019). https://doi.org/10.1007/s13347-019-00371-w
- Mathematical solution
- Finite element method
- Variational crimes