When a Description Becomes a Verdict

1. The extra conclusion

A proof is described as technical. An exposition begins with elementary material. A theory reformulates something already known. These can be useful descriptions. Sometimes even “merely” is warranted: a claimed innovation really is a standard construction once its notation has been unpacked.

The difficulty begins when the description is asked to settle a further question. Familiar ingredients are treated as settling what their arrangement contributes. An elementary beginning becomes a measure of an inquiry’s reach. A known conclusion becomes a reason to overlook what another proof explains.

No intellectual life can dispense with abbreviated judgments. Attention is finite, expertise is unevenly distributed, and much of what we learn comes from trusting someone who has already done the work. A useful judgment compresses an examination that others cannot always repeat. The question is what must survive that compression.

2. Enough for which question?

Ask whether an eigenvalue list determines what happens when a transformation is applied repeatedly. Start with the identity and look for a change that preserves the eigenvalues but accumulates under iteration. Adding the second coordinate to the first does this: the transformation sends to , leaving the second coordinate available to add again. Its triangular matrix has the same eigenvalues, 1 and 1, as the identity. Those eigenvalues also correctly determine that both transformations have determinant one and are invertible.

The list does not determine their behavior under iteration. Starting from , the first stays at ; the second reaches after n applications. The eigenvalue list has forgotten a distinction that matters to this question.

There is no fault in the list for omitting it. Useful descriptions routinely omit information. Their adequacy depends on the inference we want to draw. A richer account can be necessary for one purpose and wasteful for another.

This gives a test: if a description determines an answer, cases it cannot distinguish must agree on that answer. A disagreement identifies a limit of the description. Agreement in a few cases does not establish its general adequacy; that requires an argument that the answer depends only on what the description retains.

A description can also provide evidence for an answer without determining it. Its force then depends on how reliably the retained features bear on the question. The strength of the judgment should reflect what remains uncertain.

“No new theorem” can answer a legitimate question about a paper’s results. It leaves open whether the paper offers a better explanation, a usable procedure, or a connection between existing arguments. Those gains need demonstration. Their presence may justify a place in a textbook while remaining insufficient for a research journal.

3. What the arrangement does

A description can preserve individual answers while losing the relationships among them. In topology, recovering the groups assigned to each space leaves a further question: how does a map between spaces act on those groups? The universal coefficient theorem makes this distinction exact. It recovers integral cohomology groups from integral homology through a short exact sequence. The sequence respects maps of spaces—it is natural—and splits as abelian groups. Its splittings cannot in general be chosen to respect those maps.

To test that limitation, look for a map whose homology action vanishes while its cohomology action does not. Torsion offers a small place to try: multiplication by an integer greater than one has zero kernel over the integers but a nonzero cokernel. Realize that calculation by attaching one cell. Form by attaching an -cell to by a degree- map, with and . Collapsing gives . The reduced cellular chain complex of has boundary in degrees and . Its only nonzero reduced integral homology is therefore in degree . The sphere has reduced homology only in degree , so induces zero on reduced homology, just as a constant map does.

On top cellular chains, however, is the identity. Dualizing gives the identity on top cochains; taking cohomology reduces integers modulo . In degree , the two pullbacks from to satisfy

The homology maps cannot distinguish these two maps; cohomology does.¹

Homology thus determines the cohomology groups up to isomorphism, while its induced maps do not determine the induced cohomology maps. A list of abstract groups can be a complete answer to one question and an inadequate input to the next. Eilenberg and Mac Lane made compatibility with induced maps an explicit object of study.² Recovering each answer and recovering how the answers vary together are different achievements.

This example distinguishes three comparisons: agreement of the resulting groups, agreement of induced maps, and agreement of the constructions producing them. A claim of equivalence needs to specify which of these is intended.

A comparison of methods also needs consistent assumptions. Prerequisites, available operations, and the examples being tested must be stated on both sides. A favorable example does not establish general superiority, while one counterexample can refute a universal claim.

Enough information can still be difficult to use. Two equivalent accounts may differ in how readily a reader can locate a dependence or apply a construction. That explanatory gain must be considered alongside the definitions, translations, and machinery required to use the account.

4. What recognition knows

Expertise often makes this comparison almost immediate. A specialist may recognize a theorem under different notation, notice that an alleged extension follows from a standard argument, or locate a fatal counterexample. A brief judgment can contain years of relevant experience. Requiring every assessment to begin from first principles would squander that competence.

Nor is explicit explanation a perfect measure of knowledge. People can recognize a pattern reliably before articulating every cue. Research on intuitive expertise locates its reliability in learnable regularities and opportunities for practice and feedback; felt certainty alone is a poor test.³

Recognizing a false step, recognizing an old theorem, and assessing what a new organization explains are different tasks. Familiarity can guide all three, but success at one does not automatically settle another. “These moves are standard” becomes a strong objection to a claimed conceptual gain when the comparison shows that the gain is standard too.

The author therefore needs a usable entrance: a question, a relevant antecedent, and a worked difference. “You have missed the architecture” is no substitute. A critic who supplies the missing reference may have identified exactly what the author needed to learn. The comparison must be capable of changing what either party claims.

5. Reasons for stopping

A reader may reasonably stop because a contribution looks unlikely to repay further attention. A journal may decline a correct paper because other submissions better serve its aims. Such decisions need not wait until every question about the work is settled.

The distinction is between having enough reason to act and having enough reason for a particular assertion. Declining a full review does not establish that a paper has no explanatory value. A relevant antecedent can support a specific claim of redundancy; its force comes from the comparison.

Mathematical publishing guidelines differ on whether editors owe reasons for nonacceptance.⁴ Whenever a substantive reason is given, its scope can remain precise: “The claimed theorem follows from this result under these hypotheses”; or “The comparison is informative, but the journal requires new research results.”

Correcting a claim of redundancy may leave the decision unchanged while withdrawing an unsupported description of the work. The contribution has not been shown absent merely because it was insufficient for that venue.

6. Keeping the boundary visible

A useful assessment locates the claim, identifies the comparison, and makes clear how it bears on the purpose. An exact counterexample may suffice; a judgment about explanatory gain may require reconstructing more of the argument.

An assessment can state what was checked: the hypotheses of a cited theorem, the equivalence between two constructions, or the behavior of a proposed procedure on a counterexample. That makes the scope of the conclusion available to another reader.

A decision to decline further review can then remain distinct from a finding that the claimed result fails or follows from known work.