Metrics separate and compare points
Objective: Teach Metrics separate and compare points to someone who knows ordinary calculus but has never seen a metric-measure space. Begin with “a metric assigns a nonnegative symmetric distance, vanishes only on identical points, and obeys the triangle inequality,” read (X,d,m) is a metric space with a declared Borel measure; B_r(x) is an open ball; γ is a rectifiable or absolutely continuous curve; f is a real-valued function; g is a candidate upper gradient; p is the declared exponent; N01.01 is the audit label., reconstruct N01.01, state the theorem bridge, and diagnose the mutation.
Unit 1, chapter 1 starts from a concrete learner question: what can “Metrics separate and compare points” mean when there may be no coordinates, tangent vectors, or classical derivative? In ordinary language, a metric assigns a nonnegative symmetric distance, vanishes only on identical points, and obeys the triangle inequality. The course names the object before displaying a formula, because a familiar derivative symbol can otherwise hide assumptions that the space does not possess.
Read every symbol as a typed sentence. (X,d,m) is a metric space with a declared Borel measure; B_r(x) is an open ball; γ is a rectifiable or absolutely continuous curve; f is a real-valued function; g is a candidate upper gradient; p is the declared exponent; N01.01 is the audit label. A point, a curve, a function, an almost-everywhere class, a measure, a module element, and a number are different kinds of objects. The letter m measures sets; it is not a point mass unless that is explicitly declared.
The local meaning is: a metric assigns a nonnegative symmetric distance, vanishes only on identical points, and obeys the triangle inequality. It is interpreted inside the unit focus “a complete or explicitly incomplete metric space, a Borel measure finite on bounded sets, and curves whose length is computed from partitions.” This definition says which data are compared. It does not silently supply doubling, completeness, a Poincaré inequality, geodesics, differentiability charts, quadratic energy, curvature bounds, or uniqueness.
The bounded theorem route is: metric length is invariant under increasing reparameterization; absolutely continuous curves admit an almost-everywhere metric speed whose integral controls endpoint distance and subcurve length. Every noun in that sentence carries a hypothesis. The learner highlights where completeness, separability, bounded finiteness of m, p>1, doubling, a weak (1,p)-Poincaré inequality, infinitesimal Hilbertianity, or a curvature-dimension condition first enters.
Use four separate ledgers. The geometry ledger contains distances, balls, curve lengths, and parameterizations. The measure ledger contains measurable sets, null sets, densities, and integrals. The energy ledger contains slopes, upper gradients, relaxations, and exponents. The conclusion ledger records whether equality is pointwise, m-almost everywhere, along p-almost every curve, in Lp, weakly, or only as an inequality.
The finite ledger is deliberately small: Use X={0,…,7} with d(i,j)=|i−j| and counting measure m. The declared open ball B_1.5(4) is {3,4,5} and has mass 3; the ledger distinguishes radius, diameter, cardinality, and measure. It is not a miniature proof of the infinite theorem. Its job is to expose a wrong radius convention, missing absolute value, unnormalized mass, inadmissible curve density, hidden endpoint, or mistaken exponent before notation becomes dense.
Rebuild N01.01 by two independent routes. Route A follows the definition directly through distances, curve integrals, relaxation, or transport cost. Route B uses symmetry, subdivision, a comparison function, a dual inequality, or a smooth calibration case. Agreement certifies only the finite row; disagreement stops the argument at its first divergent operation.
A curve inequality replaces a missing coordinate derivative only after the curve class and exceptional-family convention are fixed. “For every curve,” “for p-almost every curve,” and “for one sampled curve” are three different statements. Changing a function on an m-null set can still affect values along an exceptional curve, which is why representatives and capacity matter.
A slope, a weak upper gradient, a minimal weak upper gradient, a Cheeger differential, and a module norm are related constructions, not interchangeable spellings. The chapter says which identification theorem is available in the declared setting and retains inequality signs when an isometric identification has not been proved.
The required failure mutation is: treat an arbitrary measure as doubling, identify support with all of X, confuse open and closed balls, or assume every pair of points is joined by a geodesic. Apply it to “Metrics separate and compare points,” locate the first unavailable definition or theorem hypothesis, and state the strongest smaller conclusion that survives. A boundary is part of the lesson, not an apology after the formula.
Pictures of balls, branching curves, level sets, or transported masses are explanations of typed data. They do not prove doubling, a Poincaré inequality, rectifiability, curvature bounds, or convergence. Every visual therefore labels the metric, measure, radius convention, scale, and whether an edge is a real curve, a schematic relation, or a numerical approximation.
When an integral appears, name its measure and units. ∫_γg ds integrates along arc length, ∫_Xg^p dm integrates over the ambient measure, and ∫P({u>t},A)dt integrates perimeters over level values. The same integral sign does not make these quantities the same kind of object.
When curvature language appears, distinguish a metric-measure curvature-dimension condition from pointwise sectional or Ricci tensors on a smooth manifold. The course uses optimal transport, entropy, quadratic Cheeger energy, and Bochner inequalities only in the stated direction. It does not infer a smooth manifold or a classical tensor from the letters RCD.
Close by teaching the chapter without the page. Explain “Metrics separate and compare points” in ordinary language, type every symbol, reproduce N01.01 twice, state the bounded theorem with all hypotheses, activate the mutation, and name one tempting but invalid stronger claim. Vocabulary recognition is not yet the ability to reconstruct a proof.
Keep an evidence table with columns object, type, domain, measure, exponent, representative, curve quantifier, equality meaning, hypothesis source, and first failure. Translate the main displayed relation into a complete sentence. If the sentence cannot say what is integrated over what, the formula has not yet been understood.
Finish with a neighboring-case comparison. Keep the same function but change exactly one of the metric, measure, exponent, curve family, representative, or structural hypotheses. Recompute the finite ledger, mark which quantities stay identical and which change, and decide whether the theorem remains available. This controlled comparison prevents a learner from memorizing N01.01 as a free-floating recipe.
Bounded theorem N01.01: metric length is invariant under increasing reparameterization; absolutely continuous curves admit an almost-everywhere metric speed whose integral controls endpoint distance and subcurve length. This chapter applies that route only to Metrics separate and compare points after its listed hypotheses are checked.
Declare the metric-measure data for N01.01: X, d, m, support, finiteness on balls, completeness or separability if used, the curve class, function representatives, exponent p, and the exact claim represented by “Metrics separate and compare points.”
Verify that every ball, curve length, composition, integral, and norm is defined. If a curve is reparameterized, check that the arc-length integral is unchanged; if a function is an almost-everywhere class, choose only the representative allowed by the theorem.
Construct the exact finite record: Use X={0,…,7} with d(i,j)=|i−j| and counting measure m. The declared open ball B_1.5(4) is {3,4,5} and has mass 3; the ledger distinguishes radius, diameter, cardinality, and measure. Preserve fractions, endpoint conventions, exponent powers, mass normalization, and inequality directions. A rounded diagram never replaces this row.
Prove the local estimate by its declared mechanism: triangle inequality, absolute continuity along curves, admissible density, relaxation lower bound, truncation, locality, convexity, or integration by parts. Name which operation uses a complete or explicitly incomplete metric space, a Borel measure finite on bounded sets, and curves whose length is computed from partitions.
Cross from the local record to the theorem only through the stated bridge “metric length is invariant under increasing reparameterization; absolutely continuous curves admit an almost-everywhere metric speed whose integral controls endpoint distance and subcurve length.” Record the compactness, maximal estimate, modulus argument, lower semicontinuity, measurable chart, module representation, entropy convexity, or Bochner step that supplies the passage.
Audit quantifiers and equivalence classes. Separate every point from m-almost every point, every rectifiable curve from p-almost every curve, one upper gradient from the minimal class, and equality of representatives from equality in Lp or capacity.
Recompute N01.01 by the independent route and compare exact outputs. If approximation is unavoidable, report the mesh or sample, precision, stopping rule, interval enclosure, and a refinement or perturbation check.
Activate the failure mutation “treat an arbitrary measure as doubling, identify support with all of X, confuse open and closed balls, or assume every pair of points is joined by a geodesic.” Stop at the first lost hypothesis and state the surviving conclusion. This proof does not establish arbitrary metric differentiability, every endpoint p=1 result, general Alexandrov or sub-Riemannian structure, full RCD regularity, or scientific validity of a fitted metric.
Reconstruct the exact N01.01 ledger twice: Use X={0,…,7} with d(i,j)=|i−j| and counting measure m. The declared open ball B_1.5(4) is {3,4,5} and has mass 3; the ledger distinguishes radius, diameter, cardinality, and measure. Explain what is geometric, what is measured, what is integrated, and which general theorem the row does not prove.
- Write the complete typed input for N01.01: metric, measure, balls, curve or level set, function, candidate gradient, exponent, and representative.
- Calculate distances, masses, path lengths, slopes, level sets, or transport pairs one row at a time without decimal rounding.
- Check admissibility and normalization directly, including endpoints, absolute values, p powers, and the measure used by each integral.
- Recompute by subdivision, symmetry, a comparison function, a dual bound, or the smooth calibration appropriate to this unit.
- Compare the two exact ledgers and attach N01.01; if they disagree, repair the first divergent line instead of averaging the answers.
- Apply the unit mutation, name the first missing hypothesis, weaken the conclusion, and only then interpret the picture or application.
Result: Audited result N01.01: both exact routes reproduce the declared finite record and preserve its inequality direction. This certifies the record only; the bounded theorem still requires every stated hypothesis.