SYSTEMATIC MATHEMATICS

Discrete Exterior Calculus & Geometric Numerical Integration

This 107-chapter Stage 5 specialization starts from a practical question: when a smooth field is replaced by finitely many numbers on a mesh, which identities should remain exact and which claims require approximation evidence? Thirteen content-sized units build oriented complexes, chains and cochains, incidence, discrete Stokes, primal-dual meshes, Hodge stars, compatible differential operators, FEEC interfaces, Hodge and field solvers, variational and Lie-group integrators, convergence and implementation audits, and eight original reproducible dossiers. Every chapter reads its arrays by type, derives one bounded result, completes a hand-checkable model, activates a failure case, and pairs practice with a full bilingual solution.

Before this course: Completed Linear Algebra, Multivariable Calculus, Vector Calculus, Differential Equations, Partial Differential Equations, Numerical Analysis, Topology, Differential Geometry & Manifolds, Differential Forms/de Rham/Hodge Theory, Sobolev Spaces/Distributions/Weak PDE, Calculus on Manifolds & Geometric Integration, and basic programming with arrays. Cell complexes and sparse assembly are rebuilt from first principles.

COURSE FACTSLevel, chapters, units, prerequisite, and outcome
Chapter 1

What should a discretization preserve?

Objective: Which part of “What should a discretization preserve?” is an exact consequence of topology or variation, and which part depends on mesh geometry, approximation, or solver evidence?

Unit 1, chapter 1 begins with the computational question behind “What should a discretization preserve?.” A finite model cannot preserve every continuous feature, so the invariant, approximation target, and acceptable evidence must be named before an algorithm is chosen. Before writing a matrix, identify the continuous object, the mesh object that stores it, its orientation, and the invariant that the discretization is expected to preserve.

The exact definition is: A structure-preserving discretization is a finite system equipped with typed maps for which specified algebraic identities hold exactly, while a consistency map relates finite data to the declared continuous objects. Every symbol therefore has a type: a chain is not a cochain, an incidence matrix is not a metric matrix, a primal cell is not its dual, and an exact algebraic identity is not the same as an observed convergence rate.

The chapter’s usable statement is: Exact preservation of d_h²=0 or a discrete variational identity is logically separate from convergence of a numerical solution to a smooth one. The derivation follows this visible route: Write the continuous complex, choose discrete spaces and maps, mark commuting squares that are exact, then list every norm and limiting statement that still needs analysis or experiment. Topological operations are audited separately from metric-dependent approximations so an exact cancellation cannot hide a poor Hodge star or an invalid mesh.

The complete worked problem is: Compare two updates for x′=0: x_{n+1}=x_n and x_{n+1}=−x_n; identify which preserves |x| and which is consistent. We declare ordering, signs, units, boundary treatment, solver tolerance, and arithmetic precision before computing. The result is Both updates preserve |x| exactly, but only x_{n+1}=x_n has zero local defect for arbitrary initial data; invariant preservation alone does not select the correct dynamics.

The failure case is part of the concept: A method may conserve one quantity exactly while being inconsistent, unstable, or inaccurate for the requested solution. A defensible numerical claim therefore reports the exact identity, the approximation hypothesis, the residual or error norm, refinement evidence, and the first conclusion that stops being justified.

Exact preservation of d_h²=0 or a discrete variational identity is logically separate from convergence of a numerical solution to a smooth one.

Start from the typed definition for “What should a discretization preserve?.” Write dimensions, cell orientations, primal or dual location, coefficient space, and boundary conditions before multiplying any arrays.

Write the continuous complex, choose discrete spaces and maps, mark commuting squares that are exact, then list every norm and limiting statement that still needs analysis or experiment. Display each sign, transpose, inverse, quadrature weight, and projection. Verify exact identities symbolically or with integer incidence arithmetic before introducing floating-point metric data.

Evaluate the declared model and independently check its invariant, residual, and refinement behavior. This yields Both updates preserve |x| exactly, but only x_{n+1}=x_n has zero local defect for arbitrary initial data; invariant preservation alone does not select the correct dynamics. The argument stops at the boundary: A method may conserve one quantity exactly while being inconsistent, unstable, or inaccurate for the requested solution.

Compare two updates for x′=0: x_{n+1}=x_n and x_{n+1}=−x_n; identify which preserves |x| and which is consistent.

  1. List every cell with its orientation and index, declare the stored cochain values and metric weights, and translate A structure-preserving discretization is a finite system equipped with typed maps for which specified algebraic identities hold exactly, while a consistency map relates finite data to the declared continuous objects. into a dimensioned algebraic expression.
  2. Write the continuous complex, choose discrete spaces and maps, mark commuting squares that are exact, then list every norm and limiting statement that still needs analysis or experiment. Keep exact incidence arithmetic separate from quadrature, linear solves, and rounding.
  3. Compute the requested quantity, test the structural identity and numerical error independently, and report Both updates preserve |x| exactly, but only x_{n+1}=x_n has zero local defect for arbitrary initial data; invariant preservation alone does not select the correct dynamics.

Result: Both updates preserve |x| exactly, but only x_{n+1}=x_n has zero local defect for arbitrary initial data; invariant preservation alone does not select the correct dynamics.