Density, Discreteness, Midpoints, and Approximation

The Big Idea

The real number line appears smooth and unbroken, but many sets and practical systems select only certain points from it. The integers form separated steps. The rational numbers reach into every real interval. Measuring instruments and computers usually work with finite or discrete selections even when the quantity being modeled is continuous.

Four ideas help us move between these viewpoints:

• Density tells us whether a set reaches inside every interval.
• Discreteness tells us whether each selected point stands apart from the others.
• Midpoints locate exact centers and divide intervals systematically.
• Approximation replaces a value with a nearby usable value while tracking the resulting error.

Together, these ideas prepare us for measurement, numerical computation, limits, sequences, continuity, optimization, statistics, and computer mathematics.

Four Views of the Number Line

A picture can suggest density or closeness, but the definitions use intervals and inequalities. Those definitions keep the ideas reliable at every scale, including scales too small to draw.

Density Means Every Interval Is Reached

A subset \(D\) of the real numbers is dense in \(\mathbb R\) if every open interval contains at least one member of \(D\). In symbols,

\[ a<b \quad\Longrightarrow\quad \text{there is some }d\in D\text{ with }a<d<b. \]

The rational numbers \(\mathbb Q\) are dense in \(\mathbb R\). No matter how close two distinct real numbers are, some fraction lies strictly between them. The irrational numbers are also dense in \(\mathbb R\).

Density is not about how many dots can be printed on a diagram. It is the guarantee that every positive-width interval—however small—contains a selected point.

There Is Always Another Real Number Between

If \(a<b\), their midpoint

\[ \frac{a+b}{2} \]

lies strictly between them. Once one intermediate point is found, the same midpoint process can be repeated on either smaller interval. Therefore, there are infinitely many real numbers between any two distinct real numbers.

This is one reason the real line has no “next” real number after \(x\). Given any proposed next number \(y>x\), the midpoint \((x+y)/2\) is closer and still greater than \(x\).

Fractions Can Approximate Every Real Location

The density of \(\mathbb Q\) says that rational numbers can come arbitrarily close to any real number. For example,

\[ 1.4<\sqrt2<1.5, \] \[ 1.41<\sqrt2<1.42, \] \[ 1.4142<\sqrt2<1.4143. \]

The decimal bounds are rational numbers. They can be refined indefinitely, even though no rational number equals \(\sqrt2\).

There is no closest rational number to \(\sqrt2\). If a rational approximation is not exact, density allows a still closer rational approximation.

Irrational Numbers Also Appear in Every Interval

Between any two distinct real numbers \(a<b\), there is an irrational number as well as a rational number. One way to produce an irrational value is to begin with a rational \(q\) strictly between \(a\) and \(b\), then add a sufficiently small irrational displacement such as \(\sqrt2/n\).

Thus no interval belongs exclusively to rational or irrational numbers. These two sets are thoroughly interwoven along the real line, even though a number cannot be both rational and irrational.

Density and Number of Elements Are Different Questions

The rational numbers are countable, while the real numbers are uncountable. Nevertheless, both reach every real interval. Density describes placement, not cardinality.

A set can therefore be smaller in the sense of cardinality yet still be distributed throughout the entire line. This distinction becomes important in set theory, analysis, probability, and numerical computation.

Discrete Points Stand Apart

A subset \(D\subseteq\mathbb R\) is discrete if every point of \(D\) is isolated from the other points of \(D\). For each \(x\in D\), there is some radius \(\varepsilon>0\) such that

\[ (x-\varepsilon,x+\varepsilon)\cap D=\{x\}. \]

The integers \(\mathbb Z\) are discrete. Around the integer \(n\), the interval \((n-\tfrac12,n+\tfrac12)\) contains no other integer.

Unlike the real line, the integers do have immediate neighbors: the next integer after \(n\) is \(n+1\), and the preceding integer is \(n-1\).

A Uniform Grid Has a Common Spacing

A set is uniformly discrete when one positive minimum separation works throughout the entire set. The grid

\[ G_{h,c}=\{c+nh:n\in\mathbb Z\}, \qquad h>0, \]

has spacing \(h\). The integers use \(c=0\) and \(h=1\); currency measured only to cents uses a conceptual spacing of \(0.01\) dollar.

Not every discrete set has one global spacing. In \({1,\tfrac12,\tfrac13,\ldots}\), each listed point is isolated, but the gaps shrink toward \(0\). The set becomes non-discrete if \(0\) is included, because every neighborhood of \(0\) then contains other members of the set.

Dense and Discrete Are Not Simple Opposites

“Dense in \(\mathbb R\)” asks whether every real interval contains a member of the set. “Discrete” asks whether each member can be isolated from the set’s other members. These are different tests.

A set can be neither dense in the whole line nor discrete. The two terms are not exhaustive labels.

Choose a Continuous or Discrete Model Deliberately

The quantity, the model, and the recording system may have different structures. A continuous quantity can be sampled, rounded, or stored on a discrete grid.

The Midpoint Is the Exact Center of Two Positions

For real numbers \(a\) and \(b\), the midpoint is

\[ m=\frac{a+b}{2}. \]

It is equally distant from both endpoints:

\[ |m-a|=|b-m|=\frac{|b-a|}{2}. \]

For \(a=-5\) and \(b=9\),

\[ m=\frac{-5+9}{2}=2. \]

Do not confuse the midpoint with half the distance. The midpoint is a location; \(|b-a|/2\) is a distance.

Why the Midpoint Formula Works

Start at \(a\). The signed displacement from \(a\) to \(b\) is \(b-a\). Travel half of that displacement:

\[ a+\frac{b-a}{2} =\frac{2a+b-a}{2} =\frac{a+b}{2}. \]

This form is useful because it shows the action: start, measure the change, and take the chosen fraction of it. The average form \((a+b)/2\) is shorter; both describe the same point.

Interpolation Locates Any Fraction of the Way

The midpoint uses one-half of the displacement. More generally, the point a fraction \(t\) of the way from \(a\) to \(b\) is

\[ P(t)=a+t(b-a)=(1-t)a+tb. \]

• \(t=0\) gives \(a\).
• \(t=\tfrac12\) gives the midpoint.
• \(t=1\) gives \(b\).
• \(0<t<1\) gives a point strictly between \(a\) and \(b\).

This is linear interpolation. Values \(t<0\) or \(t>1\) extend beyond the endpoints and are called extrapolation.

Interpolation Is a Weighted Average

In

\[ (1-t)a+tb, \]

the coefficients \(1-t\) and \(t\) add to \(1\). When both are nonnegative, the result lies between \(a\) and \(b\). This is a two-value weighted average.

The same structure later appears in averages, mixtures, probability, line segments in coordinate geometry, convex combinations, computer animation, and numerical methods.

Repeated Midpoints Create Finer Grids

Begin with the interval \([0,1]\). Its midpoint is \(1/2\). Bisect the resulting intervals to obtain \(1/4\) and \(3/4\). Continue, and the grid points have the form

\[ \frac{k}{2^n}, \qquad k\in\mathbb Z,\quad n\in\mathbb N. \]

These are dyadic rational numbers. They do not include every rational number—for example, \(1/3\) is not dyadic—but they form a dense subset of \(\mathbb R\). Repeated bisection can therefore produce grid points as close as desired to any real location.

Approximation Replaces Exactness with Controlled Nearness

An exact value identifies a quantity without approximation. An approximate value is a nearby replacement chosen for measurement, communication, or computation.

\[ \sqrt2\approx1.4142 \]

The symbol \(\approx\) means “is approximately equal to.” It does not assert equality. By contrast,

\[ \sqrt2=1.41421356237\ldots \]

is an exact equality because the nonterminating decimal notation—with the ellipsis understood as the continuing expansion—names the same real number.

Error Measures the Cost of Approximation

If \(x\) is the exact value and \(a\) is an approximation, the signed error and absolute error may be written

\[ e=a-x, \qquad E=|a-x|. \]

The sign of \(e\) tells whether the approximation is above or below the exact value. The magnitude \(E\) tells how far apart they are.

For \(x=\sqrt2\) and \(a=1.41\),

\[ E=|1.41-\sqrt2|\approx0.00421. \]

Relative Error Compares Error with Scale

The same absolute error can be minor for a large quantity and serious for a small one. When \(x\ne0\), the relative error is

\[ \frac{|a-x|}{|x|}. \]

Multiply by \(100\%\) for percent error. If an exact length is \(80\) cm and a reported value is \(79\) cm, then

\[ \text{relative error}=\frac1{80}=0.0125, \qquad \text{percent error}=1.25\%. \]

Relative error is undefined when the exact value is zero, so use absolute error or another context-appropriate scale there.

An Error Bound Is Often More Useful Than an Error

The exact value may be unknown, so the exact error may also be unknown. We can still guarantee that the error does not exceed a chosen amount \(E\):

\[ |x-a|\le E. \]

This absolute-value statement is equivalent to the interval statement

\[ a-E\le x\le a+E. \]

Thus an approximation and an error bound create a guaranteed enclosure for the exact value.

Bracketing Turns Uncertainty into an Interval

Suppose we know only that

\[ L\le x\le U. \]

The midpoint of the bracket,

\[ m=\frac{L+U}{2}, \]

is the estimate that minimizes the largest possible absolute error. The guarantee is

\[ |x-m|\le\frac{U-L}{2}. \]

The bracket width \(U-L\) measures total uncertainty; the midpoint estimate has a maximum error equal to half that width.

Rounding Maps a Continuous Range to a Discrete Grid

Rounding to the nearest multiple of a positive step \(h\) replaces \(x\) by a nearest grid point \(r\). Any standard tie convention that selects one of the two nearest grid points preserves the bound

\[ |x-r|\le\frac h2. \]

Rounding is therefore a transformation from many possible inputs to fewer reportable outputs. It loses information and is not reversible without additional knowledge.

Truncation is different: it discards digits beyond a selected place rather than choosing the nearest grid point. Its absolute error can approach one full grid step \(h\), so do not automatically assign the half-step rounding bound to a truncated value.

Floor and Ceiling Give One-Sided Integer Approximations

The floor \(\lfloor x\rfloor\) is the greatest integer not exceeding \(x\). The ceiling \(\lceil x\rceil\) is the least integer not less than \(x\):

\[ \lfloor x\rfloor\le x\le\lceil x\rceil. \]

For \(x=3.7\), the floor is \(3\) and the ceiling is \(4\). For \(x=-3.7\), the floor is \(-4\) and the ceiling is \(-3\).

Floor and ceiling are useful when a practical answer must be a whole number. Whether to round down or up depends on the decision: complete boxes that can be filled use floor, while vehicles required to carry everyone use ceiling.

Resolution Determines Which Differences Can Be Reported

Resolution is the smallest increment a system can distinguish or report under its operating convention. A scale displaying mass to the nearest \(0.1\) kg cannot communicate every real-valued difference in mass.

Resolution is not automatically the same as accuracy. A display may show many decimal places and still be systematically wrong. Likewise, a coarse but well-calibrated instrument may be accurate within its stated tolerance.

Decimal Places Form Successively Finer Grids

Numbers with at most \(n\) digits after the decimal point lie on the grid

\[ \left\{\frac{k}{10^n}:k\in\mathbb Z\right\}. \]

Increasing \(n\) changes the spacing from \(1\) to \(0.1\), \(0.01\), \(0.001\), and so on. Each fixed grid is discrete, but the union of all terminating-decimal grids is dense in \(\mathbb R\).

This is a productive pattern: a sequence of discrete models can approximate a continuous space with increasingly fine resolution.

A Rational Grid Can Reach Any Requested Accuracy

Given a real number \(x\) and any tolerance \(\varepsilon>0\), choose a positive integer \(n\) large enough that \(1/n<\varepsilon\). Let

\[ q=\frac{\lfloor nx\rfloor}{n}. \]

Then \(q\) is rational and

\[ q\le x<q+\frac1n, \qquad |x-q|<\varepsilon. \]

This argument turns “as close as desired” into a procedure: select a fine enough grid, then choose the grid point immediately below \(x\).

Bisection Refines a Bracket Predictably

Suppose a target is known to lie in \([L,U]\). Test the midpoint \(m=(L+U)/2\), decide which half still contains the target, and discard the other half. Each step divides the width and maximum midpoint error by \(2\).

After \(n\) successful bisections, the interval width is

\[ \frac{U-L}{2^n}, \]

and the midpoint of the remaining bracket has maximum error

\[ \frac{U-L}{2^{n+1}}. \]

Bisection is simple, dependable, and foundational in numerical root-finding, search algorithms, measurement, and optimization.

Computers Use Finite Representations

A real interval contains infinitely many values, but a computer has finite memory. It therefore stores a finite collection of representable numbers. Many familiar decimals and fractions must be approximated in binary floating-point arithmetic.

The representable points are discrete, and their spacing is not generally uniform across the entire number range. Very large magnitudes usually have wider spacing between adjacent representable values than small magnitudes.

This does not make numerical computation unreliable. It means that algorithms must account for rounding, accumulated error, scale, and tolerance rather than assuming every written real number is stored exactly.

Approximation Error Can Travel Through Operations

Suppose \(a\) approximates \(x\) with \(|a-x|\le E_a\), and \(b\) approximates \(y\) with \(|b-y|\le E_b\). For addition,

\[ |(a+b)-(x+y)| \le E_a+E_b. \]

The bound comes from the triangle inequality. Subtraction has the same worst-case bound. Multiplication and division require bounds that also account for the sizes of the quantities, so premature rounding can create avoidable downstream error.

A practical rule is to retain extra working digits and round once at the end, unless the problem’s reporting procedure requires otherwise.

A Reliable Approximation Workflow

1. Identify the exact target. State what value or quantity is being approximated.
2. Choose the model. Decide whether the permitted values are continuous, discrete, or grid-based.
3. State the method. Rounding, truncation, interpolation, and midpoint estimation have different effects.
4. Keep equality honest. Use \(=\) for equality and \(\approx\) for approximation.
5. Record the scale or resolution. Include units and the grid step when relevant.
6. Find an error or bound. Use absolute, relative, or interval error as the context requires.
7. Carry sufficient working precision. Avoid repeated unnecessary rounding.
8. Check plausibility. Confirm that the estimate lies in a known bracket and has the correct magnitude.
9. Report responsibly. Do not display more certainty than the data or method supports.

Common Confusions

Working Summary

A dense set reaches into every interval; a discrete set has isolated members. The rationals and irrationals are each dense in the real line, while the integers form a uniformly discrete grid. Density concerns placement, not how many elements a set has.

The midpoint \(m=(a+b)/2\) is equally distant from two endpoints. The interpolation rule \(a+t(b-a)\) locates any fractional position between or beyond them. Repeated midpoint construction creates increasingly fine dyadic grids.

Approximation should be controlled by error, bounds, resolution, and purpose. Rounding maps continuous possibilities to discrete outputs; bracketing and bisection refine uncertainty; and numerical systems use finite representations to work effectively with continuous mathematical models.

Check Your Understanding