06-reference / concepts

coordinate change as core move

Sun Apr 19 2026 20:00:00 GMT-0400 (Eastern Daylight Time) ·concept ·status: draft
coordinate-changerepresentationproblem-solvinglaplacetopologylinear-algebra

Coordinate Change as the Core Problem-Solving Move

The one-sentence claim

A problem is hard in one coordinate system and trivial in another; the core move in mathematical problem-solving is finding the system in which the answer falls out, and the “solution” is often just the restatement of the transformed problem.

The pattern

Every hard problem arrives wearing clothes. Axes, variables, units, a domain. The novice tries to solve the problem in the clothes it showed up in. The move is to recognize the clothes are a choice — someone picked them — and a different choice collapses the work. You are not solving the problem. You are picking the representation in which the problem has already been solved.

Three things travel together in the pattern. First, there is a symmetry in the problem that the original coordinates hide. Second, there is a target representation in which that symmetry becomes manifest — a circle instead of an ellipse, an addition instead of a multiplication, a polynomial instead of a derivative. Third, the transformation between the two is the content of the solution. The computation afterward is bookkeeping.

Six instances, three flavors

Space-lift coordinate changes. Take a 2D problem and view it as the shadow of a 3D object. The hexagon-tiling problem from 2026-04-20-3blue1brown-five-puzzles-thinking-outside-the-box is transparent once you notice a rhombus tiling is a projection of a stack of cubes; max moves between any two tilings is exactly n³, because the cube is three-dimensional and the number is its volume. Same video: Tarski planks on a disk are hemispherical caps viewed from above, and Archimedes’ cylinder projection gives the area directly, so the minimum sum of plank widths is 2. Same video: Monge’s three-circle theorem reads off cones with a common tangent plane. And 2026-04-20-3blue1brown-topology-open-problem pushes the same move further — the inscribed-rectangle problem becomes a question about a Möbius strip in 4D whose self-intersections are the rectangles. The shared grammar: when the current dimension is not enough, lift.

Rescaling coordinate changes. Change the metric, and a hostile shape becomes a friendly one. In 2026-04-20-3blue1brown-why-colliding-blocks-compute-pi, energy conservation traces an ellipse in (v1, v2) space; rescaling by sqrt(mass) on each axis converts the ellipse into a circle, and the inscribed-angle theorem finishes the job. In 2026-04-20-3blue1brown-how-and-why-to-take-a-logarithm-of-an-image, the complex logarithm takes Escher’s recursive self-referential distortion and unwraps it into a flat strip with a fixed point at the blank spot. The log takes a multiplicative structure and makes it additive. That is the same move a data engineer makes log-transforming a skewed price series, and it is the same move SVD makes axis-aligning correlated features.

Basis and transform coordinate changes. The Laplace-series pair is the cleanest paired example the vault holds. 2026-04-20-3blue1brown-but-what-is-a-laplace-transform is the inner move: the Laplace transform is a change of basis from the time-domain basis into the exponential-function basis, and the poles in the new basis are the coefficients. 2026-04-20-3blue1brown-why-laplace-transforms-are-so-useful is the outer move: once you change domains from t to s, a differential equation becomes polynomial algebra. Grant’s closing, “differential expressions turn into polynomials,” is the most operationally explicit statement of the pattern in the corpus. Both videos are coordinate changes. Different flavors — one picks a basis, the other picks a domain — same move.

The meta-move

The skill is not solving the problem. The skill is noticing which coordinate system makes it trivial. In Sanderson’s framing across the Five Puzzles video: step up a dimension, find the symmetry, move the symmetry to where it is convenient. The content of mathematical taste is the library of coordinate systems you have internalized, cross-indexed by the kind of problem each one dissolves.

This also tells you when you are stuck for the right reason. If you are three hours into a calculation and it is still ugly, the question is not “am I computing this correctly” — it is “am I in the wrong coordinates.” The good mathematician does less arithmetic than the bad one.

Why RDCO cares (load-bearing)

Every skill-design decision is a coordinate-change decision. “Should this be a skill or a bash alias?” is really “which coordinate system makes this problem trivial — LLM-space or shell-space?” If the problem’s natural structure is pattern-matching under ambiguity, LLM-space. If it’s deterministic text manipulation, shell-space. Picking the wrong one means doing the work in the wrong clothes.

The RDCO graph database is a coordinate change. Raw markdown is prose in the time-domain of reading. Typed edges are the s-plane — questions impossible to ask in text (“what docs cite author X who validates position Y”) become trivial lookups in the graph. The graph is not a better representation of the vault; it is the vault in a different basis.

Sanity Check is a coordinate change applied weekly. Take a technical problem phrased in the operator’s domain (a production pipeline failure, a dashboard anomaly, a retry storm), re-express it in the reader’s mental framework (a physical metaphor, a state-space picture, a pole diagram), and the solution becomes visible. The re-expression is the article. Readers who come away saying “I never thought about it that way” are reporting a successful coordinate change.

verifier-as-epistemology is itself a coordinate-change thesis. “Is this output correct?” is hard in the natural coordinates of the output. “Does the mechanical verifier pass it?” is computable in the coordinates of an audit script. The epistemic move is the change of coordinates, not the verification step. layered-defense-architecture generalizes it — each layer is a coordinate change on failure detection. Do not check “is the model wrong” (hard in model-coordinates); check “does the audit script fail” (trivial in verifier-coordinates). binary-decision-around-continuous-probability points the same way from the other side: collapsing a probability distribution to argmax is a coordinate change that destroys information, and the fix is staying in the original coordinate system a step longer.

When coordinate change fails

The honest counter. Coordinate changes do not always exist. Some problems are hard because they are hard in every representation — P vs NP is the canonical example, and the whole field of complexity theory is the study of problems whose difficulty is coordinate-invariant. The move is not universal. It is a heuristic, and a strong one, but it is not a theorem.

There is also the analysis-vs-intuition cap Sanderson names explicitly in the Five Puzzles closing. We can lift 2D problems into 3D because we are 3D creatures and intuition comes along for the ride. Once the lift goes to 4D or 10,000D — a transformer’s embedding space, for example — the coordinate change still works mechanically, but the intuitive guardrails are gone. The answer becomes analysis rather than recognition. This is the honest frame for LM writing: these models do coordinate-change reasoning in dimensions where no human has intuition, and whether that counts as “understanding” is a category question, not a capability one. See high-dim-surface-concentration for the companion concept — high-dimensional geometry is counter-intuitive because the coordinate change costs us the intuition we used to check our work.

Confidence

Six sources, all Grant Sanderson / 3Blue1Brown. This is the same caveat that applies to high-dim-surface-concentration and verifier-as-epistemology at their current source counts — the vault evidence is single-author-cluster even though the pattern itself is 80+ years old in mathematics and computer science. The right next sources are Polya’s How to Solve It (the “find a related problem you can solve” heuristic is the same move), the change-of-variables theorem in multivariable calculus (the mechanical heart of the pattern), and any standard treatment of the Fourier or Laplace transform outside the 3B1B corpus. The pattern is not under-supported in the literature; it is under-supported in this vault, which is honest. A fourth-author source would move confidence from “well-illustrated single-source cluster” to “well-triangulated canon.”