Geometric proof and coordinate methods

A geometric proof explains what follows from what

Join A = (0, 0), B = (4, 0), C = (5, 2) and D = (1, 2) in that order. Averaging the coordinates of A and C gives (5/2, 1); averaging B and D gives the same point. The diagonals share a midpoint. This is an exact check for one figure. To prove the same result for every parallelogram, we will explain why the equality follows from its opposite parallel sides.

A theorem is a claim for every figure satisfying specified hypotheses. In “If ABCD is a parallelogram, its diagonals bisect each other”, being a parallelogram is the hypothesis and diagonal bisection is the conclusion. The letters must identify vertices in boundary order, and a quadrilateral must have four distinct vertices and a simple closed boundary. An argument cannot quietly assume the conclusion it is meant to prove.

A proof connects the hypotheses to the conclusion through definitions, established results and valid calculations. Checking several exact examples can expose an error or suggest a theorem, but does not establish a claim about every admissible figure. A rounded measurement from a drawing has an additional limitation: the apparent equality may be a drawing or measurement approximation.

The converse reverses hypothesis and conclusion and requires its own argument. “Every rectangle has equal diagonals” does not by itself prove “Every quadrilateral with equal diagonals is a rectangle”. One counterexample satisfying the proposed hypothesis and failing the conclusion disproves that universal converse. A figure that violates the stated hypothesis is not a counterexample to the original theorem.

A convenient coordinate frame must still represent every allowed figure

Use the Euclidean plane with perpendicular axes and the same length unit on both axes. A coordinate describes a directed displacement, so negative and fractional values are permitted. Changing the origin or turning the perpendicular frame does not stretch the figure: the preceding transformation lesson explains why rigid movements preserve distances and angles. A reflected frame can reverse the sign of an oriented area while preserving its ordinary area.

For an arbitrary nondegenerate triangle, place A at (0, 0), direct the positive x-axis along AB, and write B = (b, 0), C = (c, d), with b > 0 and d ≠ 0. These conditions represent every such triangle after a choice of frame. The coordinate c need not lie between 0 and b. Setting c = 0 as well would impose a right angle at A, so that extra choice is unavailable unless perpendicularity is already known.

For a parallelogram, the useful representation is A = (0, 0), B = u, D = v and C = u + v, where u and v are nonparallel directed pairs. The definition of a parallelogram justifies this representation: the line through B parallel to AD and the line through D parallel to AB meet at u + v. Choosing B = (b, 0) is optional; choosing D = (0, d) would additionally assume the parallelogram is a rectangle.

A diagram may use different screen scales to fit its panel. Read the exact coordinates rather than measuring that panel. If physical coordinates themselves use unequal units, such as metres horizontally and centimetres vertically, convert to a common unit before applying a distance or right-angle formula. The unadjusted numerical pair would describe a different scale in the two directions.

Exact coordinate differences describe lengths and whole segments

From P = (1, 1) to Q = (5, 4), the horizontal change is 4 and the vertical change is 3. Pythagoras gives PQ² = 4² + 3² = 25, hence PQ = 5 units. Half of each change is (2, 3/2), so the midpoint is M = (3, 5/2). Coordinate differences describe the movement; averaging endpoints describes the point halfway along it.

For P = (pₓ, pᵧ) and Q = (qₓ, qᵧ), set u = Q − P = (qₓ − pₓ, qᵧ − pᵧ). Define q(u) = uₓ² + uᵧ². Perpendicular coordinate offsets and Pythagoras give PQ² = q(u), so PQ = √q(u). If an offset is zero, the formula reduces to the horizontal or vertical distance; if both are zero, the points coincide and the distance is zero. Squared distances avoid unnecessary radicals without changing equality of nonnegative lengths.

The midpoint M is (P + Q)/2, meaning that each coordinate is averaged separately. Indeed, M − P = (Q − P)/2 and Q − M = (Q − P)/2. These equal directed halves place M on PQ and give equal distances to its endpoints. For coincident endpoints this formula returns their common point, but it does not create a positive segment or a direction.

The complete segment PQ consists of P + t(Q − P) for every real t with 0 ≤ t ≤ 1. The endpoints correspond to t = 0 and t = 1, the midpoint to t = 1/2, and the remaining parameters cover all intermediate points. With unrestricted real t and P ≠ Q, the same expression describes the supporting line. This connects vertex coordinates to entire straight sides rather than treating a polygon as a list of isolated markers.

Parallel and perpendicular conditions must also handle vertical lines

Take A = (0, 0), B = (3, 0) and C = (0, 4). The changes from A are (3, 0) and (0, 4), and their coordinate products add to 3 × 0 + 0 × 4 = 0. The squared lengths are AB² = 9, AC² = 16 and BC² = 25. This valid triangle satisfies 9 + 16 = 25, so the converse of Pythagoras proves a right angle at A. We will now name the product sum and show why the same test works for arbitrary nonzero directions.

For two directed pairs u and v, define the coordinate sums u · v = uₓvₓ + uᵧvᵧ and cross(u, v) = uₓvᵧ − uᵧvₓ. These are algebraic abbreviations for the displayed products, not assumptions from a later linear-algebra course. For line directions, first require u ≠ (0, 0) and v ≠ (0, 0). A zero pair identifies no line or angle even though both product expressions then vanish.

The equation cross(u, v) = 0 is equivalent to v being a real multiple of nonzero u. If uₓ ≠ 0, use t = vₓ/uₓ: the equation gives vᵧ = t uᵧ. If uₓ = 0, then uᵧ ≠ 0; the equation forces vₓ = 0 and t = vᵧ/uᵧ works. Thus the two directions lie along the same line direction, including the vertical case. Their supporting lines may be distinct parallel lines or the same line, so coincidence must be checked separately.

For perpendicularity, expand q(v − u) = q(u) + q(v) − 2(u · v). Take the common starting point as O, with endpoints U = O + u and V = O + v. If the nonzero directions are perpendicular, Pythagoras on OUV gives UV² = OU² + OV², so u · v = 0. Conversely, if u · v = 0, the directions cannot be collinear: v = t u with v ≠ 0 would give u · v = t q(u) ≠ 0. The triangle is therefore nondegenerate, and its squared lengths satisfy the Pythagorean equality. The converse of Pythagoras proves the right angle at O.

For nonvertical directions, dividing the coordinate-sum condition by uₓvₓ gives the familiar slope condition m₁m₂ = −1. That division is valid only when both horizontal components are nonzero. The coordinate-sum test itself also covers a vertical direction and a horizontal direction. A negative multiple reverses a ray's direction, so parallel supporting lines do not mean the rays have the same directed angle.

A signed area records orientation; its magnitude is an ordinary area

For A, B and C, let u = B − A and v = C − A. Define the signed double area Δ = cross(u, v). The sign distinguishes the two boundary orientations in the chosen frame; reversing B and C changes Δ to −Δ. The ordinary triangular area is |Δ|/2 square units. In particular, Δ = 0 means the three points are collinear, including repeated-point cases, and gives no nondegenerate triangle.

Here is a derivation rather than a formula guessed from a picture. If u ≠ 0, put t = (u · v)/q(u), H = A + t u, and w = v − t u. Then u · w = 0, so a nonzero w is perpendicular to AB, while w = 0 means C lies on that line. Expanding gives w = (Δ/q(u))(−uᵧ, uₓ): positive Δ places C to the left of directed AB, and negative Δ to the right. H lies on the complete supporting line; it need not lie between A and B. The distance CH is √q(w), the perpendicular height to the base of length √q(u).

Expansion gives q(u)q(w) = q(u)q(v) − (u · v)² = (uₓvᵧ − uᵧvₓ)² = Δ². The base-times-height formula therefore gives area = √(q(u)q(w))/2 = |Δ|/2. If u = 0, A and B coincide, Δ = 0, and the area is zero directly; there is no division by q(u). For a nondegenerate parallelogram with adjacent directions u and v, its diagonal separates it into two triangles with area |Δ|/2, giving parallelogram area |Δ|.

For a valid simple quadrilateral, choose a diagonal lying inside it and divide it into two triangles. Adding their signed double areas cancels the shared diagonal terms and gives Δ₄ = cross(A, B) + cross(B, C) + cross(C, D) + cross(D, A), where cross(A, B) means AₓBᵧ − AᵧBₓ. Both triangles have the same boundary orientation, so the quadrilateral's ordinary area is |Δ₄|/2. This consecutive-coordinate sum is often called the shoelace expression. A nonzero value by itself does not certify a simple boundary: opposite sides must not cross or overlap and every side must have positive length. For repeated, collinear or crossed data, the expression remains an algebraic calculation without automatically representing the requested geometric area.

A shared diagonal midpoint characterises a nondegenerate parallelogram

For a parallelogram in boundary order, write A, B = A + u, D = A + v and C = A + u + v, with cross(u, v) ≠ 0. The midpoint of AC is A + (u + v)/2. The midpoint of BD is ((A + u) + (A + v))/2, the same point. It lies halfway along each whole diagonal, so the diagonals bisect each other.

Prove the converse separately. If the diagonal midpoints agree, (A + C)/2 = (B + D)/2, hence C = B + D − A. Set u = B − A and v = D − A. Then the four positions are A, A + u, A + u + v, A + v. Their consecutive side directions are u, v, −u, −v. When cross(u, v) ≠ 0, all sides are positive and opposite supporting lines are distinct and parallel, which proves a nondegenerate parallelogram.

The same condition also controls the boundary. In this representation AB lies on A + t u and DC on A + v + t u; these lines cannot meet because v is not a multiple of u. AD and BC are similarly disjoint. Adjacent sides meet only at their intended common endpoint. Thus the ordered boundary is simple and has the expected four corners; the conclusion concerns those entire segments.

The hypotheses matter. A = (0, 0), B = (1, 0), C = (3, 0), D = (2, 0) have the same diagonal midpoint (3/2, 0), but all four points are collinear and the boundary overlaps. The algebraic equality holds, while a nondegenerate parallelogram has not been obtained. With a valid simple quadrilateral, diagonal bisection and being a parallelogram are equivalent; do not omit the geometric validity conditions.

Equal and perpendicular diagonals have different consequences

In a nondegenerate parallelogram, use the same adjacent directions u and v. The diagonal directions are u + v and v − u. Expanding their squared lengths gives q(u + v) − q(v − u) = 4(u · v). Equal diagonal lengths are therefore equivalent to u · v = 0. The adjacent sides are perpendicular, and the other corners use the same directions with signs reversed, so all four interior angles are right angles: the parallelogram is a rectangle.

For perpendicular diagonals, their coordinate sum is (u + v) · (v − u) = q(v) − q(u). Both diagonal directions are nonzero because u and v are nonparallel. The diagonals are therefore perpendicular exactly when the adjacent side lengths are equal. Opposite sides of the parallelogram have the same lengths already, so all four sides are equal and it is a rhombus. A square satisfies both tests, but neither condition alone forces both conclusions.

Removing the parallelogram hypothesis changes the claim. The simple trapezoid A = (−2, 0), B = (2, 0), C = (1, 2), D = (−1, 2) has AC² = BD² = 13, but its diagonal midpoints (−1/2, 1) and (1/2, 1) differ. It is not a rectangle. The simple kite A = (0, 3), B = (2, 0), C = (0, −1), D = (−2, 0) has perpendicular vertical and horizontal diagonals, but side-length squares 13, 5, 5, 13, so it is not a rhombus. Each refutes an unrestricted converse, not the guarded parallelogram theorem.

Equal squared distances turn a circle problem into two linear equations

For A = (0, 0), B = (4, 0) and C = (0, 3), try O = (2, 3/2). The three squared distances are all 25/4, so this point gives a circle with positive radius 5/2. Try O = (4/3, 1) instead: the squared distances to A, B and C are 25/9, 73/9 and 52/9. That candidate fails, even though the correct circle still exists. The centre equations will find the successful candidate without guessing.

A circle has a centre O = (h, k) and a positive radius r. Its circumference is the complete set of points X = (x, y) satisfying (x − h)² + (y − k)² = r². Equal nonnegative distances can be compared by squaring. The condition r > 0 is still necessary: r = 0 gives only the centre point, while a negative number is not a circle radius.

For given A, B and C, a candidate circumcentre must satisfy OA² = OB² = OC². Subtracting the expanded squared-distance expressions cancels h² + k². If E = Bₓ² + Bᵧ² − Aₓ² − Aᵧ² and F = Cₓ² + Cᵧ² − Aₓ² − Aᵧ², the two equations are 2(Bₓ − Aₓ)h + 2(Bᵧ − Aᵧ)k = E and 2(Cₓ − Aₓ)h + 2(Cᵧ − Aᵧ)k = F. Every equal-distance centre satisfies them, and expanding backwards proves that every solution has the three equal squared distances.

For distinct A and B, the first equation is the complete perpendicular-bisector line of AB. To see why, let M = (A + B)/2 and u = B − A. The equation is exactly u · (O − M) = 0. Its solutions include M and all points in the perpendicular direction (−uᵧ, uₓ); the component argument from the direction tests shows these exhaust the solutions. At O = M the offset is zero, and M belongs to the line without a claimed angle for that zero offset. This derivation also handles vertical or horizontal AB without slopes.

Meeting just one bisector is insufficient for three vertices. After finding a candidate, substitute it into both linear equations and compare all three squared distances. Use r = √OA² only after those checks and the positive-radius guard. The centre may lie outside the triangle; that location does not invalidate a circumcircle.

Three noncollinear points give one circle; other data need a different conclusion

Let u = B − A, v = C − A and Δ = cross(u, v). The coefficient combination of the two centre equations is 4Δ, their two-by-two determinant. Here that name abbreviates a product difference, not an assumed matrix theory. If Δ ≠ 0, eliminate k by multiplying the first equation by vᵧ and the second by uᵧ and subtracting. This gives 2Δ h = E vᵧ − F uᵧ. Eliminate h similarly to get 2Δ k = uₓF − vₓE. Thus h = (E vᵧ − F uᵧ)/(2Δ) and k = (uₓF − vₓE)/(2Δ), with valid nonzero denominators.

Substitution verifies that these values solve both equations, proving existence. Any solution must have the same h and k by the elimination identities, proving uniqueness. The common squared radius is positive: if OA² = 0, O = A, and equality with OB² would force B = A, contradicting Δ ≠ 0. Hence every three noncollinear points lie on a unique positive-radius circle, and the centre's equations supply the exact certificate.

If A, B and C are three distinct collinear points, choose a perpendicular common-unit frame with A = (0, 0), B = (b, 0), C = (c, 0), where b > 0, c ≠ 0 and c ≠ b. Equal distances to A and B require h = b/2; equality to A and C requires h = c/2. These requirements conflict, so no circle passes through all three. Thus three distinct points admit exactly one circle if and only if they are noncollinear.

If exactly two distinct points are listed, the repeated vertex contributes no new condition. Every centre on their perpendicular bisector supplies a positive-radius circle, so there are infinitely many possibilities. If all three listed points coincide at P, any centre O ≠ P with radius OP supplies a circle; choosing O = P gives radius zero and fails the circle definition. Δ = 0 is therefore a signal to classify the data, not permission to divide by zero or a universal verdict of “no circle”.

An exact certificate must establish the requested geometric object

Write an argument in three linked stages: verify the hypotheses, calculate an exact relation, then name the result that turns that relation into the conclusion. For a right-angle claim, distinct noncollinear points come before the squared-length equality. For a rectangle claim, a valid parallelogram comes before equal diagonals. For a circumcircle claim, the full centre equations and a positive common radius come before the circle conclusion.

Coordinates of vertices can determine a polygon when its complete consecutive side segments are specified, but a list of matching numbers is not automatically a proof about those sides or interior angles. The segment formula proves the supporting directions along every side. In a parallelogram the directions u, v, −u, −v establish all corner angles; a single measured or computed angle alone would not certify an arbitrary quadrilateral as a rectangle. Reordering the same vertex set can produce a crossed boundary and invalidate the quadrilateral hypotheses.

A circle certificate through A, B and C proves membership of those vertices on one complete distance-defined circumference. It does not say the straight sides of triangle ABC lie on that circumference: they are chords. Likewise, selected displayed circle markers do not exhaust a circle. Distinguish a constructed exact centre, a proposed centre being tested, and an approximate drawing of either.

Successful explorer cases illustrate the proofs and check calculations. A failed valid case can be a counterexample to an overbroad proposed claim. A failed hypothesis requires a statement such as “no nondegenerate triangle” or “parallelogram theorem unavailable”, rather than an angle or shape classification the data do not support. Keep that distinction even when the algebra gives a tidy zero.

Explore proofs, counterexamples and the cases where a theorem cannot be applied

The explorer has four scenes with eight exact choices each: right-angle checks, diagonal-midpoint parallelogram checks, circumcentre equations, and diagonal converses. Each calculation uses the displayed rational coordinates in one perpendicular common-unit frame. The drawing helps locate the points; exact squared lengths, product differences and equations decide the certificate. The panel writes its consecutive-coordinate area expression as 2S: for a triangle it equals Δ, and for a quadrilateral it equals Δ₄. For a valid parallelogram this is twice cross(B − A, D − A); for invalid boundaries it remains an algebraic expression. In the centre scene, O labels the proposed centre, K the solved centre, and D the coefficient determinant 4Δ.

In the right-angle scene, inspect distinctness and noncollinearity before interpreting the coordinate-sum or Pythagorean equality. Vertical sides and fractional coordinates remain valid. A repeated point or collinear triple cannot receive a nondegenerate triangle angle merely because an expression vanishes.

In the parallelogram scene, compare both diagonal midpoints and inspect the complete ordered boundary. A common midpoint with the nondegeneracy conditions proves the entire parallelogram by the general argument above. An unequal midpoint refutes diagonal bisection for that valid figure; repeated, collinear or self-crossing data need their own failed-hypothesis explanation.

In the circumcentre scene, follow both linear equations, the product difference and every radius-distance check. Noncollinear data have one exact centre even when a proposed candidate is wrong. Three distinct collinear points have none, while repeated positions can have infinitely many centres. Keep a zero-radius candidate separate from the existence of other valid positive-radius circles.

In the converse scene, establish whether the polygon is a valid parallelogram before applying the equal-diagonal or perpendicular-diagonal test. A trapezoid with equal diagonals and a kite with perpendicular diagonals explain why the missing hypothesis matters. A general identity proves the guarded theorem; the eight illustrated choices do not replace that proof.

Worked examples

Try it yourself

1. For an arbitrary nondegenerate triangle ABC, a proposed proof starts A = (0, 0), B = (b, 0), C = (0, d), with b and d positive. Is this a valid general coordinate choice? Give a valid representation, and explain what fails when its vertical coordinate is zero.

   Show the answer and reasoning

   The proposed coordinates assume a right angle at A. A general representation is A = (0, 0), B = (b, 0), C = (c, d), with b > 0 and d ≠ 0. If d = 0, the three positions are collinear and no nondegenerate triangle is represented.

   Directing the x-axis along AB is a choice of perpendicular frame. Placing C on the y-axis is an additional geometric restriction, not a frame choice available for every triangle. Allowing arbitrary c and a nonzero signed d covers every nondegenerate triangle without assuming perpendicular sides.

2. For A = (1, −2), B = (1, 1), C = (5, −2), prove the angle at A is right without using slopes. Find the signed double area in the order A, B, C and the ordinary area.

   Show the answer and reasoning

   The angle at A is 90°. AB² = 9, AC² = 16 and BC² = 25. The signed double area is −12 square units, and the ordinary triangular area is 6 square units.

   The adjacent pairs are u = (0, 3) and v = (4, 0), both nonzero, with cross(u, v) = −12 ≠ 0. Their coordinate sum is 0 × 4 + 3 × 0 = 0. Equivalently 9 + 16 = 25, so the converse of Pythagoras proves the right angle. A vertical slope quotient is unnecessary, and the negative area sign records boundary orientation rather than negative physical area.

3. A = (−2, 1), B = (2, 2) and D = (−1, 4) are adjacent vertices of a proposed parallelogram in the order A, B, C, D. Find C using diagonal bisection, then verify nondegeneracy, the common midpoint and the area.

   Show the answer and reasoning

   C = (3, 5). Both diagonal midpoints are (1/2, 3), the adjacent product difference is 11, and the parallelogram area is 11 square units.

   The midpoint condition gives A + C = B + D, hence C = B + D − A = (3, 5). With u = (4, 1) and v = (1, 3), cross(u, v) = 12 − 1 = 11 ≠ 0. The complete side directions u, v, −u, −v give a simple nondegenerate parallelogram. Averaging either diagonal gives (1/2, 3), and the absolute product difference gives its area.

4. An argument claims that equal diagonals always imply a rectangle, and perpendicular diagonals always imply a rhombus. Test the first claim with A = (−2, 0), B = (2, 0), C = (1, 2), D = (−1, 2), and the second with A = (0, 3), B = (2, 0), C = (0, −1), D = (−2, 0). What hypothesis repairs both claims?

   Show the answer and reasoning

   Both unrestricted claims are false. The first simple quadrilateral has equal diagonal squares 13 but is not a rectangle. The second has perpendicular diagonals but side-length squares 13, 5, 5, 13, so it is not a rhombus. Requiring a nondegenerate parallelogram repairs both implications.

   The first diagonal midpoints are (−1/2, 1) and (1/2, 1), so the figure is not a parallelogram. The second midpoints are (0, 1) and (0, 0); its unequal adjacent side lengths disprove the rhombus conclusion. Both boundaries are simple, so these are admissible counterexamples to claims about arbitrary simple quadrilaterals. Neither contradicts a theorem whose premise already requires a parallelogram.

5. A = (−1, 1), B = (3, 1), C = (−1, 4). A proposed centre is O₀ = (1, 2). Test it, then find the true circumcentre and positive radius and explain uniqueness.

   Show the answer and reasoning

   O₀ fails: its squared distances to A, B, C are 5, 5, 8. The true centre is O = (1, 5/2), with radius 5/2 units, and it is unique.

   The first equal-distance equation is 8h = 8, giving h = 1. The second is 6k = 15, giving k = 5/2. The coordinate product difference is 12 ≠ 0, so these equations have one solution. Substitution gives all three squared distances 25/4, a positive common radius square. O₀ belongs to the first bisector but not the second; its failure does not mean the noncollinear points lack a circumcircle.

6. Classify positive-radius circles through each listed triple: (0, 0), (2, 0), (5, 0); (0, 0), (2, 0), (2, 0); and (1, 1), (1, 1), (1, 1). For the last triple, is centre (1, 1) allowed?

   Show the answer and reasoning

   Respectively: no circles; infinitely many with centres (1, k) for any real k; infinitely many with any centre other than (1, 1). Centre (1, 1) in the last case gives radius zero and is excluded.

   Three distinct collinear positions force conflicting centre coordinates h = 1 and h = 5/2. With only two distinct positions, the sole nontrivial equation is h = 1, and r² = 1 + k² is always positive. With one repeated position P, any O ≠ P gives r = OP > 0. A zero determinant distinguishes none of these cases by itself; repeated constraints and inconsistent constraints have different meanings.

7. A calculation reports u · v = 0 for u = (0, 0), v = (2, 1). Another reports a shared diagonal midpoint for A = (0, 0), B = (1, 0), C = (3, 0), D = (2, 0). A third orders the four square positions as (0, 0), (2, 2), (0, 2), (2, 0). May these certify a right angle, nondegenerate parallelogram or simple square boundary?

   Show the answer and reasoning

   None supplies the proposed certificate. The zero direction has no angle; the midpoint example is collinear and overlapping; the reordered square positions form a self-crossing boundary rather than a simple square.

   A zero product is meaningful only after the directions and triangle satisfy the angle hypotheses. The collinear midpoint equality is an algebraic fact, but its adjacent product difference is zero. In the final ordered boundary, the first and third sides cross at (1, 1), so matching a square's vertex set does not establish its required edge order. Check the whole object and hypotheses before naming the geometric conclusion.

What to carry forward

• A theorem proves a conclusion for every figure satisfying its hypotheses. Selected examples support checking; a converse needs its own proof or counterexample.
• Choose perpendicular common-unit coordinates without imposing an extra shape: moving the frame is allowed, assuming an unknown right angle is not.
• Squared coordinate differences prove distance relations, averaged coordinates prove midpoints, and the segment parameter describes complete sides.
• For nonzero directions, the product difference tests a shared line direction and the coordinate sum tests perpendicularity; neither divides by a horizontal component.
• A triangle's ordinary area is half the absolute signed double area. Zero area, repeated vertices and crossed boundaries require a failed-hypothesis explanation.
• Shared diagonal midpoints characterise a nondegenerate parallelogram. Within that class, equal diagonals characterise rectangles and perpendicular diagonals characterise rhombuses.
• Three noncollinear points have a unique positive-radius circumcircle, proved by both linear centre equations. Distinct collinear or repeated-point data need separate classifications, and every proposed centre must be checked.