Constructive reals and choice principles
Carry quantitative approximation data explicitly and state the extra principles used to compare real-number constructions.
On this page
The question for today
What you’ll understand
- Build a rational approximation with a verified convergence modulus.
- Distinguish supplied approximation data from merely existing approximations.
- State choice-dependent comparison results for the specified Cauchy and Dedekind constructions without universalizing them.
A constructive real number must support reliable mathematical use, but there are several ways to package the information. A rational sequence with a modulus gives explicit approximation instructions; a located Dedekind cut describes order relations with rationals. Classically these constructions fit together easily, while constructively the required simultaneous selections deserve attention. We will compute a real with a transparent rate, identify the choice hidden in some quotient arguments, and use interval refinement as an independent example of quantitative evidence.
A modulus is part of usable approximation data
A rational Cauchy sequence has arbitrarily small tails. If this is supplied as untruncated dependent data, it includes a function selecting a tail index for every requested precision and a proof that the selection works. If the definition instead uses merely existing indices, extracting a simultaneous modulus is an additional step whose justification depends on the ambient theory. These packages should not be conflated simply because both are described verbally as a Cauchy sequence.
A chosen modulus makes approximation quantitative: ask for a precision, compute the indicated index, and use the accompanying estimate. Different rational sequences can represent the same real, so a quotient or other identification construction is still needed to make representation-independent equality. Equality of real numbers is not automatically a decidable comparison just because rational arithmetic is decidable. Positive separation can often be witnessed by a rational gap, whereas absence of a known separation is not an equality certificate.
Dedekind location need not provide every binary decision at once
A constructive Dedekind real can be given by lower and upper rational cuts satisfying inhabitation, roundedness, disjointness, and a locatedness condition. Locatedness says that whenever one rational is strictly below another, the real lies above the first or below the second, with logical disjunction interpreted according to the chosen foundation. It does not assert an exact comparison with every rational, since equality with a boundary rational may prevent a strict alternative at that same boundary.
In the HoTT book's specified constructions, the canonical map from Cauchy reals into Dedekind reals is an embedding. The book proves that countable choice is sufficient to make it an equivalence, and gives excluded middle as another sufficient assumption. This is a scoped comparison result, not a claim that every constructive foundation must adopt choice or that every presentation of Cauchy reals has identical completeness properties. The distinction between truncated existence and selected rational approximations is central to the proof.
Completeness proofs must disclose their selections
Suppose a sequence consists of equivalence classes of rational Cauchy sequences. A diagonal construction may begin by selecting one representative from every class. Mere membership in a quotient does not in general supply that countable list of representatives. A countable choice assumption can justify such a step in suitable formulations, but an argument should identify where it was used. Assuming an arbitrary sequence already comes with chosen representatives is a stronger input specification than starting with bare quotient elements.
Other constructions organize completion differently. The HoTT book develops a higher inductive Cauchy completion that builds in the needed limit operation without simply repeating a representative-selection proof. Its comparison with Dedekind reals remains a separate theorem. For concrete numerical work, one can often avoid these foundational choices by supplying explicit rational approximations and bounds from the outset. Such a calculation proves the stated approximation guarantee; it does not settle every possible equivalence between competing foundational definitions of the real continuum.
Work it through
Produce a strict precision certificate
For qₙ=1−2⁻ⁿ, compute q₅ and give an index guaranteeing error strictly below 1/64 from the limit one. Supply a single modulus that also verifies the Cauchy-tail condition at every binary precision.
At index five the rational value is thirty-one over thirty-two, and its error from one is one over thirty-two. This is larger than the requested one over sixty-four, so that index does not meet the target.
At index seven the value is one hundred twenty-seven over one hundred twenty-eight. Its error is one over one hundred twenty-eight, strictly below one over sixty-four. Index six would give equality with the requested tolerance, so it would not meet the strict inequality.
For general precision k, choose M(k)=k+1. If n and m are at least that index, the difference of the two approximants is no larger than the larger of their two errors from one, which is at most 2 to the power minus k minus one. This is strictly below the requested binary precision.
The value q₅ is 31/32, index seven meets the strict 1/64 error target, and M(k)=k+1 is a valid Cauchy modulus.
The living notebook
Explore the model
Interactive lab · root-finding
Inspect interval evidence for a quadratic root
Pause and Reset, then set target c to 2 and initial upper bound U to 2. After five Steps predict bracket [1.375,1.4375], midpoint 1.40625 and error bound 0.03125. For a fresh comparison Reset and set c to 0.25 and U to 2: two Steps find the exact root 0.5 and collapse the bracket. These are certified refinements for one monotone quadratic, not a general equality decision for constructive reals. Edits retain elapsed time; Reset and set both controls each run.
- Midpoint after 0 refinements
- 1.5
- Bracket error bound
- 1.5
- Computed residual · x² − c
- 0.25
Bisection of x² − c on [0,U], capped at 50 iterations for floating-point resolution. The error bound is the maximum distance from the displayed midpoint to either bracket endpoint (half the width in exact arithmetic). Compensated square evaluation preserves the bracket when ordinary multiplication rounds a residual to zero. A truly exact endpoint or midpoint collapses the bracket. This is one monotone quadratic, not a general root solver. Parameter changes pause playback and recompute the model at the current elapsed time. Reset restores both parameters and returns time to zero. Playback starts paused. Adjust either parameter, then choose Play or Step.
Look forAn interval plus a verified bound supplies usable approximation evidence, including an explicitly detected exact special case.
Pause & reason
Check your understanding
Your turn
Try it yourself
Starting from [1,2], perform four rational bisections for the positive root of x²−2. Give the final bracket, its midpoint, and a rigorous error bound for that midpoint without decimal root estimates.
Take a moment to make a prediction or work it out on paper. Open a hint when you need one.
Give me a hint
Compare each rational midpoint's square with two, preserving a lower endpoint below the root and an upper endpoint above it.
Compare with the solution
The tested midpoints are 3/2, 5/4, 11/8, and 23/16. Their square comparisons respectively move the upper, lower, lower, and upper endpoint. The final bracket is [11/8,23/16], since 121/64<2<529/256. Its width is 1/16 and its midpoint is 45/32. The midpoint error is therefore at most half the width, namely 1/32. This rational bracket certificate requires no decision about equality of arbitrary real numbers.
Keep the idea
Take it with you
- A supplied modulus turns approximation into a quantitative operation.
- Truncated locatedness and simultaneously chosen approximation data are different inputs.
- Choice-sensitive comparison and completeness claims must name the real construction and the selection step.
Read further
Original sources and deeper explanations to follow at your own pace.
Optional progress, saved in this browser when available. Revisit any lesson at any time.
Lesson not marked complete.