10.21 Tensor Change of Basis Verification Procedure

Tensor Change of Basis Verification Procedure is the complete, ordered sequence of checks applied to a finished change-of-basis computation, beginning with purely syntactic inspection of the formula and ending with numerical confirmation of coordinate-independent invariants, organized so that cheaper, faster checks are performed first and more demanding checks are reserved for confirming a result that has already passed the earlier stages.

Stage One: Syntactic Inspection

Index Balance Before Any Arithmetic

The procedure begins, before any numerical values are computed, by confirming that every free index in the proposed transformation formula appears with matching letter and placement on both sides of the equation, and that every dummy index appears exactly twice, once upper and once lower, catching formula-writing errors at essentially no computational cost.

Direction and Variance Assignment

Immediately following the index balance check, the procedure confirms that each upper index has been assigned a forward Jacobian factor and each lower index an inverse Jacobian factor, and that the overall direction of the transformation, from the intended source chart to the intended target chart, matches what was specified at the outset.

Stage Two: Structural Numerical Checks

Basis Pairing Verification

With the formula confirmed syntactically sound, the procedure computes the primary and dual basis vectors in the target chart and confirms their pairing still reduces to the Kronecker delta, verifying that the underlying basis machinery, not merely the component formula, has been applied consistently.

Jacobian Product Identity Spot Check

The procedure multiplies the computed forward Jacobian matrix by the computed inverse Jacobian matrix at the point of interest and confirms the product equals the identity matrix, catching any error in computing the matrix inverse before that error can propagate into the component-level checks that follow.

Stage Three: Component-Level Verification

Round-Trip Component Recovery

The procedure applies the reverse transformation to the newly computed components and compares the recovered values against the original components individually, index by index, isolating any single component where a discrepancy appears rather than only detecting an aggregate failure.

Stage Four: Invariance Confirmation

Contracted Scalars and Structural Invariants

Finally, the procedure computes every available fully contracted scalar, along with, for a rank-two tensor, its determinant, trace, and eigenvalues, in both the source and target charts, confirming agreement across every one of these coordinate-independent quantities as the concluding, most demanding layer of verification.

Route Agreement When an Intermediate Chart Exists

Where an intermediate coordinate chart is naturally available, the procedure additionally compares the direct transformation result against the result obtained by transforming through the intermediate chart in two steps, confirming that both routes agree as required by the coordinate change compatibility of the tensorial rule.

Diagram of the Full Procedure

Ordered Stages From Cheap to Demanding

Handling a Failure at Any Stage

Localizing the Error by Stage

Because the procedure is ordered from cheapest to most demanding, a failure detected at an earlier stage should be resolved before proceeding to later stages, since an unresolved syntactic error at stage one, for instance, will almost certainly cause every later stage to fail as well, wasting effort on more expensive checks that cannot succeed until the earlier problem is fixed.

Passing All Stages as Confidence, Not Certainty

Successfully passing every stage of the procedure provides strong practical confidence that a change-of-basis computation has been carried out correctly, though, as with any finite set of checks, it remains logically possible, though highly unlikely in practice, for a sufficiently unusual combination of errors to survive every stage simultaneously; repeating the procedure with an independently chosen coordinate transformation offers an additional layer of confidence when the stakes of an undetected error are high.

Adapting the Procedure to the Situation

Omitting Stages for Simple Cases

For simple, well-understood transformations, such as a pure coordinate relabeling with a diagonal Jacobian, later stages of the procedure may be safely abbreviated or skipped, since the earlier stages alone are typically sufficient to catch any plausible error in such a simplified setting.

Emphasizing Later Stages for Complex or High-Stakes Cases

For a complex curvilinear transformation involving a higher-rank tensor, or for a computation whose result will be relied upon in further downstream work, carrying out every stage of the procedure in full, including checking multiple independent invariants and, where possible, an alternative route through an intermediate chart, is warranted to minimize the risk of an undetected error propagating further.