Mistake museum

Mistake Museum

The exhibits here are not silly slips — they are mistakes that clever people make too. For each one, find the flaw yourself first, then open the diagnosis. 208 exhibits so far.

Vol. 1 Sets hall

Exhibit

Solving for the parameter and forgetting to check distinctness

The wrong step: , so or , giving or .
What went wrong? Think first, then open

Diagnosis When , both elements and equal , and the set reads . The elements of a set must be distinct, so this is not a two-element set at all.

The right step: After solving, substitute each candidate back. makes the two elements equal, so reject it; gives the set , which works. So .

“When a set contains a parameter, always substitute back and check that the elements are distinct”

From: Elements and Sets

Exhibit

Treating ∅ and {∅} as the same thing

The wrong step: , because both sides are “empty”.
What went wrong? Think first, then open

Diagnosis has no elements at all; has one element, which happens to be the empty set. Think of an empty box versus a box containing one empty box.

The right step: and , but : one has elements, the other has .

“Count the elements first, then ask whether the sets are equal”

From: Elements and Sets

Exhibit

Moving brackets around in a set difference

The wrong step:
What went wrong? Think first, then open

Diagnosis Take . On the left, , so the left side is ; on the right, , and removing more still leaves . The two sides disagree on the region .

The right step: Check element by element, or rewrite the difference as an intersection: .

“Only union and intersection let you move brackets freely; difference does not”

From: Operations on Sets

Exhibit

Taking the complement inside the brackets but not flipping the operation

The wrong step:
What went wrong? Think first, then open

Diagnosis An element outside is only missing from at least one of them: it may be in but not in . For example, with , , , the left side is and the right side is .

The right step: De Morgan's laws do two things at once: complement each set, and swap intersection for union: .

“When the complement goes inside the brackets, intersection and union swap”

From: Operations on Sets

Exhibit

Non-empty proper subsets, but only one thing subtracted

The wrong step: has non-empty proper subsets.
What went wrong? Think first, then open

Diagnosis Out of the subsets, two have to go: the empty set (not non-empty) and the whole set itself (not proper).

The right step: There are non-empty proper subsets. Before counting, name each object you are excluding, one by one.

“Exclude as many as you name, and name every one you exclude”

From: Counting the Elements of a Finite Set

Exhibit

Counting through a correspondence that is not one-to-one

The wrong step: Match each subset of with the sum of its elements. The sum can only be , so there are subsets.
What went wrong? Think first, then open

Diagnosis Different subsets can have the same sum: and both sum to . This correspondence is “many-to-one”, so the two sides do not have the same size (there are actually subsets).

The right step: Before counting through a correspondence, check two things: different objects go to different images (injective), and every image is hit by some object (surjective). Subsets ↔ 0-1 strings passes both checks.

“Only a one-to-one correspondence lets you count the other side instead”

From: Counting the Elements of a Finite Set

Exhibit

Counting an empty piece when counting partitions

The wrong step: Split into two classes: each element picks a class, ways; the classes are unordered, so divide by to get .
What went wrong? Think first, then open

Diagnosis Two of the colorings use a single color for everything, and then the other class is the empty set. Every piece of a partition must be non-empty, so these two do not count.

The right step: ways. In general, an -element set can be split into two pieces in ways.

“No piece of a partition may be empty”

From: Partitions of a Set

Exhibit

Adding pieces that overlap

The wrong step: students in the class are in the math club and in the physics club, so are in at least one club.
What went wrong? Think first, then open

Diagnosis Some students may be in both clubs. “Math club” and “physics club” are not a partition: the two pieces intersect, and anyone in both is counted twice.

The right step: The addition principle only works for pairwise disjoint pieces. With overlap you must subtract what was counted twice: . That is the inclusion–exclusion principle of station 9.

“Check that the pieces are pairwise disjoint before you use the addition principle”

From: Partitions of a Set

Exhibit

Thinking a largest intersecting family must all share one element

The wrong step: For an intersecting family of subsets of to reach members, it must be “all subsets containing some element ”.
What went wrong? Think first, then open

Diagnosis Counterexample: together with all three-element subsets and . That is members, every two of which intersect (a three-element subset misses only one element, so it meets every two-element subset), yet no element is shared by all of them.

The right step: The proof of the bound only says “at most one subset from each complementary pair”. Any family that takes exactly one from each pair and is pairwise intersecting reaches the bound, and there is more than one such family.

“A construction that reaches the bound is often not unique”

From: Families of Subsets

Exhibit

Thinking any layer is a largest antichain

The wrong step: Subsets in the same layer never contain one another, so all -element subsets of form the largest family in which no member contains another.
What went wrong? Think first, then open

Diagnosis Subsets in one layer really do not contain one another, but the layers have different sizes: there are only one-element subsets, but two-element subsets.

The right step: The layer sizes are largest in the middle. Sperner's theorem says the maximum is exactly , reached by the middle layer.

“Take the widest layer”

From: Families of Subsets

Exhibit

Saying “equals” when you have only shown “must contain”

The wrong step: is closed under subtraction, and , so .
What went wrong? Think first, then open

Diagnosis What follows is , , and from there every integer is in , that is, . But the problem never said contains only integers: is also closed under subtraction and also contains and .

The right step: The conclusion should be . To claim equality you would also have to prove the other direction: contains nothing else.

“What you derive is a lower bound, not the whole set”

From: Properties of Sets

Exhibit

Thinking S could have just one element

The wrong step: satisfies . Take ; all we need is .
What went wrong? Think first, then open

Diagnosis becomes , whose discriminant is : no real solution. So in a set of real numbers, and are always two different elements.

The right step: First check whether the fixed-point equation has a solution. Here it has none, so has at least two elements; pushing further, it actually has at least three.

“Before you make the rule “stand still”, solve the equation”

From: Properties of Sets

Exhibit

Leaving out the boundary points in a case split

The wrong step: Solve . When : …; when : …; when : …
What went wrong? Think first, then open

Diagnosis and belong to no case: something is left out. And this equation holds at every point of , including both endpoints, so leaving them out gives a wrong answer.

The right step: The cases must cover everything: , , . A boundary point may go on either side, but it must go on one side, and only one.

“Nothing left out (the union is everything); nothing counted twice (the classes are pairwise disjoint)”

From: The Principle of Classification

Exhibit

Mixing two criteria, so the classes overlap

The wrong step: Split into the two classes “even numbers” and “multiples of ”, count each, and add.
What went wrong? Think first, then open

Diagnosis are both even and multiples of , so they are counted twice; and are in neither class. This is not a partition.

The right step: Use one criterion at a time (for example, “the remainder on division by ”), or handle the overlap with inclusion–exclusion.

“One classification, one criterion”

From: The Principle of Classification

Exhibit

Taking “the smallest” in an infinite set

The wrong step: Let be the set of all positive real numbers, and let be the smallest element of …
What went wrong? Think first, then open

Diagnosis has no smallest element: for any , is smaller. The well-ordering principle only promises a smallest element for a finite non-empty set of reals, or a non-empty set of natural numbers.

The right step: Before using the extremal principle, check that there are finitely many objects, or that the values are non-negative integers. Otherwise, first find a way to make the problem finite.

“Make sure the extreme object exists before you pick it”

From: The Extremal Principle

Exhibit

Assuming there is only one “largest”

The wrong step: Let have the most wins. If beat , then has more wins than .
What went wrong? Think first, then open

Diagnosis beating only means won one game that did not; may also have won many games that did not. And several players may be tied for the most wins.

The right step: The extremal principle gives you just one sentence: “nobody is larger”. Your argument has to produce a concrete reason why someone has strictly more, for example beat and everyone beat.

“Use “nobody is larger”; do not swap it for “it is larger than everyone else””

From: The Extremal Principle

Exhibit

Stopping after subtracting the pairwise intersections

The wrong step:
What went wrong? Think first, then open

Diagnosis Look at the middle region : the first three terms count it times, the last three subtract it times, leaving times. It has been left out.

The right step: Add back. On the workbench, switch off the last term and the middle region immediately turns amber and is marked “”.

“If you subtract too much, add it back, until every region is counted exactly once”

From: Inclusion–Exclusion

Exhibit

Counting “neither” and stopping after subtracting

The wrong step: The numbers in that are neither multiples of nor multiples of number .
What went wrong? Think first, then open

Diagnosis The numbers are multiples of both and , so they were subtracted twice. In fact there are : .

The right step: Find the union first: , then take the complement: .

“To count a complement, first get the union right”

From: Inclusion–Exclusion

Vol. 2 Functions and Functional Equations hall

Exhibit

Base and exponent swapped when counting mappings

The wrong step: and , so there are mappings from to .
What went wrong? Think first, then open

Diagnosis The elements of are the ones making choices: each picks an image from the elements of , and choices are made. That is , not .

The right step: The number of mappings is . To remember it: the exponent is “how many choices are made” (the size of ), the base is “how many options each choice has” (the size of ).

“Whoever makes the choices is the exponent”

From: Mappings

Exhibit

Checking “is there an arrow?” but not “how many arrows?”

The wrong step: , , with , , . Every element of has an image, so this is a mapping.
What went wrong? Think first, then open

Diagnosis A mapping needs every element of to have a unique image. Here goes to both and , so its image is not unique.

The right step: Check both: every element of has at least one arrow (none left out) and at most one arrow (none doubled). How often an element of is hit, or whether it is hit at all, has nothing to do with being a mapping.

“On the A side, none missing and none doubled; on the B side, anything goes”

From: Mappings

Exhibit

Seeing $+1$ and shifting right

The wrong step: The graph of is the graph of shifted unit to the right.
What went wrong? Think first, then open

Diagnosis The point of the new graph at has the same height as the point of the old graph at . So what the old graph looked like at now appears at : everything has moved to the left.

The right step: is shifted right by (left if ). , so : shift left by .

“Write it as f(x − h), then read off h”

From: Functions

Exhibit

Treating $|f(x)|$ and $f(|x|)$ as the same thing

The wrong step: ; the graph of is the graph of with the part below the -axis folded up.
What went wrong? Think first, then open

Diagnosis That is : the absolute value sits on the function value, so it acts up and down. In the absolute value sits on the variable: and give the same value, so it is an even function. Keep the part to the right of the -axis and replace the left part by the mirror image of the right.

The right step: : fold the bottom up, nothing moves sideways. : copy the right onto the left, nothing moves up or down. Toggle between the two buttons on the workbench to compare.

“The absolute value acts on whatever it wraps”

From: Functions

Exhibit

The graph of an odd function always passes through the origin

The wrong step: is odd, so and the graph passes through the origin.
What went wrong? Think first, then open

Diagnosis is odd, but is not in its domain, so the graph does not pass through the origin at all. “” assumes that is in the domain.

The right step: The correct statement: if the domain of an odd function contains , then gives .

“Before you use f(0), ask whether 0 is in the domain”

From: Odd and Even Functions

Exhibit

Simplifying first, then deciding odd or even

The wrong step: , so is even.
What went wrong? Think first, then open

Diagnosis Simplifying threw away a restriction on the domain: the original function needs . So is in the domain but is not. The domain is not symmetric about the origin, so the function can be neither odd nor even.

The right step: The first step in deciding parity is always to check whether the domain is symmetric about the origin. Here it is not, so is neither odd nor even.

“No symmetric domain, no parity”

From: Odd and Even Functions

Exhibit

Decreasing on two pieces, so decreasing on their union

The wrong step: is decreasing on .
What went wrong? Think first, then open

Diagnosis Take : , so the right-hand point is higher. Monotonicity must hold for any two points of the interval, and a pair straddling breaks it.

The right step: Write “ is decreasing on and on ”, separating the intervals with “and” or a comma. Do not join them with .

“Do not glue monotone intervals together carelessly”

From: Monotonicity

Exhibit

“Proving” monotonicity with a few special values

The wrong step: , so is increasing on .
What went wrong? Think first, then open

Diagnosis Three points say nothing about what happens in between: has , yet it is decreasing on .

The right step: A proof must work for any : form the difference , turn it into a product, and decide the sign of each factor. Special values can only disprove monotonicity.

“Examples can only disprove; a proof needs “any””

From: Monotonicity

Exhibit

Seeing $f(x+2)=-f(x)$ and calling $2$ the period

The wrong step: , so is periodic with period .
What went wrong? Think first, then open

Diagnosis After a shift by the graph does not land on itself: it lands upside down. A period needs , and here there is an extra minus sign.

The right step: Shift once more: . The period is (and is not a period unless is identically ).

“Flip twice to get back to where you started”

From: Periodicity

Exhibit

Every periodic function has a least positive period

The wrong step: Every periodic function has a least positive period; just work it out.
What went wrong? Think first, then open

Diagnosis The constant function has every positive number as a period, so there is no smallest one. The Dirichlet function ( when is rational, when is irrational) has every positive rational as a period, and also has no least positive period.

The right step: Before you talk about “the least positive period”, make sure it exists. When we say “ has period ”, we mean its least positive period.

“Having a period does not mean having a least positive period”

From: Periodicity

Exhibit

Writing only $f(1)>0$ for “both roots greater than 1”

The wrong step: Both roots of are greater than , because .
What went wrong? Think first, then open

Diagnosis only says that does not lie between the roots: both roots could be to the right of , or both to the left. Here the roots are and , both on the left.

The right step: You also need the axis: both roots greater than , , . Here , so it fails.

“The sign at an end point only tells “same side” or “opposite sides”; which side comes from the axis”

From: Quadratic Functions

Exhibit

Checking the axis and the end points, forgetting the discriminant

The wrong step: : the axis lies in , and , so both roots lie in .
What went wrong? Think first, then open

Diagnosis : there are no real roots at all. The whole parabola floats above the -axis; the end-point and axis conditions hold, but there are no roots to place.

The right step: “Both roots in ” needs all four conditions: , , , .

“First ask whether there are roots, then ask where they are”

From: Quadratic Functions

Exhibit

Finding where a composite log function increases, forgetting the domain

The wrong step: : the inner is decreasing on and the outer function is decreasing, so is increasing on .
What went wrong? Think first, then open

Diagnosis The argument of a logarithm must be positive: , that is, or . Points of are not in the domain at all; at , for instance, the argument is .

The right step: Find the domain first, then study the inner function's monotonicity inside it: is increasing on and decreasing on .

“With a logarithm, settle the argument first, then talk about monotonicity”

From: Power, Exponential and Logarithmic Functions

Exhibit

Equality cannot be reached, yet the minimum is still given as $2a$

The wrong step: The minimum of on is .
What went wrong? Think first, then open

Diagnosis Equality in holds at , and . On this interval the value is always greater than , so is not the minimum.

The right step: lies entirely to the right of the turning point , where the function is increasing, so the minimum is .

“When you use AM–GM to find an extreme value, first ask whether equality can actually be reached”

From: The Function f(x) = x + a²/x

Exhibit

Substituting without checking the range of the new variable

The wrong step: , so the minimum is .
What went wrong? Think first, then open

Diagnosis Equality needs , which is impossible. Let ; then , which is on the increasing part of .

The right step: The minimum is reached at (that is, ): .

“After a substitution, write down the range of the new variable first”

From: The Function f(x) = x + a²/x

Exhibit

Equality in AM–GM cannot hold, yet you still write “minimum”

The wrong step: For , , so the minimum is .
What went wrong? Think first, then open

Diagnosis only when , and is not in . So is only a lower bound. It is never reached, so it is not the minimum.

The right step: is decreasing on , so on its minimum is the value at the right end: .

“Before you write “minimum”, find the $x$ where equality holds”

From: Maxima and Minima of Functions

Exhibit

The discriminant method, forgetting the leading coefficient can be 0

The wrong step: : from for we get , so the minimum is .
What went wrong? Think first, then open

Diagnosis When the equation becomes , which has no solution at all. only means something for a quadratic equation; here tells you nothing.

The right step: Check separately by substituting it back: it is not reached. So the range is : there is no minimum, and the maximum is (reached at ).

“Last step of the discriminant method: test separately the value that makes the leading coefficient 0”

From: Maxima and Minima of Functions

Exhibit

Checking only the endpoints for a quadratic as well

The wrong step: , and , so on .
What went wrong? Think first, then open

Diagnosis “The two ends are enough” is true only for a linear function, whose segment lies between its two ends. is a parabola opening upward, and it dips in the middle: .

The right step: First make sure the function you built is linear (or monotonic) in the chosen variable. For a quadratic, look at the vertex and the discriminant.

“The endpoint method needs “linear” or “monotonic””

From: Solving Problems by Constructing a Function

Exhibit

Concluding $u=v$ from $f(u)=f(v)$ when $f$ is not monotonic

The wrong step: is odd and , so , , and the only solution is .
What went wrong? Think first, then open

Diagnosis is not monotonic on : , so different inputs can give the same value. For example gives , so it is a solution too.

The right step: To get from you must first prove that is strictly monotonic ( is, is not). Otherwise, expand and solve the equation the honest way.

“Oddness makes the two sides match; monotonicity makes the inputs equal”

From: Solving Problems by Constructing a Function

Exhibit

Reading $f^{(n)}(x)$ as $[f(x)]^n$

The wrong step: , so .
What went wrong? Think first, then open

Diagnosis means composing three times: . is a power of the value. They are completely different operations.

The right step: Substitute one layer at a time: , , .

“Iterating means “do it again”, not “multiply again””

From: Iteration of a Function

Exhibit

Announcing a formula after computing three terms

The wrong step: , , , so . QED.
What went wrong? Think first, then open

Diagnosis The first three terms fitting a formula does not mean every term fits it. “Spotting the pattern” is only a guess; the proof is still missing.

The right step: Add the induction: if , then . Or use the fixed point directly: .

“A guessed formula needs a proof by induction, or a method that derives it directly”

From: Finding the n-th Iterate

Exhibit

Solving the fixed-point equation without proving the limit exists

The wrong step: , . Let the limit be ; then , so .
What went wrong? Think first, then open

Diagnosis is , which grows without bound and has no limit at all. Solving only says: if there is a limit, it must be a fixed point.

The right step: First prove that the limit exists (for example, monotonic and bounded, or, as on this page, the distance to the fixed point shrinks by a fixed ratio), then solve . Here : the distance doubles at every step, and the orbit moves away from .

“A fixed point is only a candidate; whether the orbit goes there needs its own proof”

From: Applications of Iteration

Exhibit

Forgetting the range of the new variable

The wrong step: for every real . Let , so and .
What went wrong? Think first, then open

Diagnosis can be positive or negative, but can only be non-negative, and both must satisfy the equation. gives , while gives : a contradiction.

The right step: This equation has no solution: no such function exists. When you substitute, state the range of the new variable clearly, and check with every value of that can occur.

“The substitution must be reversible, and the range must be stated”

From: Functional Equations: Substitution

Exhibit

Stopping once you have a formula, without substituting back

The wrong step: Elimination in the system gives a formula for , so that is the function we want.
What went wrong? Think first, then open

Diagnosis Elimination only proves: “if satisfies the original equation, then can only be this formula.” That is a necessary condition. Sometimes the formula is undefined at some points, or does not satisfy the original equation at all.

The right step: Substitute the you found back into the original equation (the “Substituted back” meter on the workbench does exactly that), and state the domain, for example .

“Elimination gives a candidate; substituting back makes it an answer”

From: Functional Equations: Substitution

Exhibit

Taking what plugging in gives you as the answer

The wrong step: . gives , shows is even, so .
What went wrong? Think first, then open

Diagnosis and “even” are only necessary conditions. Plenty of functions satisfy them (for example ), and does not satisfy the original equation; on the other hand does satisfy it too. Plugging in values by itself cannot give “”.

The right step: Plugging in gives properties and special values. Either keep going until you derive a formula for (with suitable extra conditions), or guess a solution, substitute it back into the original equation to check, and explain why there are no others.

“Plugging in gives clues, not the answer”

From: Functional Equations: Plugging in Values

Exhibit

Proving it for the rationals, then claiming it for all reals

The wrong step: From we get for every rational , so for every real .
What went wrong? Think first, then open

Diagnosis The equation only “strings together” the rational points. An irrational number like can never be reached by finitely many additions, subtractions and integer multiples. Without an extra condition, there really are solutions that are not (it is just impossible to write one down).

The right step: The last rung must use the condition given in the problem (monotonic, continuous or bounded) to squeeze each irrational number between rationals (see the last two steps of the proof at the start of this page).

“The rung from ℚ to ℝ needs an extra condition”

From: Functional Equations: Cauchy's Method

Exhibit

Still writing $c_1x_1^n+c_2x_2^n$ for a double root

The wrong step: has characteristic roots , so .
What went wrong? Think first, then open

Diagnosis really has only one unknown constant, which cannot match two initial values and . For example, with and , contradicts .

The right step: For a double root the general term is . The example above gives ; check it term by term:

“Two initial values need two independent constants”

From: Functional Equations: Recursion

Vol. 3 Trigonometric Functions hall

Exhibit

Taking a square root without checking the quadrant

The wrong step: , so .
What went wrong? Think first, then open

Diagnosis only tells you that , and that has two square roots, one positive and one negative. An angle with can be in quadrant I or in quadrant II: the line meets the unit circle at two points.

The right step: If no quadrant is given, split into cases: in quadrant I, ; in quadrant II, . If the quadrant is given, take the sign from the direction of the cosine line .

“The Pythagorean identity gives the size; ask the unit circle for the sign”

From: The Unit Circle and the Trigonometric Functions

Exhibit

Treating tan 90° as a very large number

The wrong step: , so .
What went wrong? Think first, then open

Diagnosis At the terminal side points straight up and is parallel to the tangent , so it never meets it: is undefined. It is not “the number infinity”. And if you approach from the side, the tangent line points down and heads off in the negative direction.

The right step: Write down the domain: needs . Only compare values inside one interval where the function is monotonic.

“No intersection, no function value”

From: The Unit Circle and the Trigonometric Functions

Exhibit

Reading the shift straight off the constant in the brackets

The wrong step: The graph of is the graph of shifted left by .
What went wrong? Think first, then open

Diagnosis A shift adds to or subtracts from itself. Shifting left by gives . For we need .

The right step: Factor out the coefficient of first: . What is added inside the brackets now is the shift: left by .

“The shift is what gets added to x, so factor out ω first”

From: Graphs and Their Transformations

Exhibit

Thinking a larger ω makes the graph wider

The wrong step: The graph of is the graph of with every -coordinate stretched to times its value.
What went wrong? Think first, then open

Diagnosis already reaches its maximum at , while only gets there at : the same height comes earlier, so the graph has been squeezed. The period is shorter than .

The right step: Every -coordinate is shrunk to of its value (-coordinates unchanged). Think of as “how far it turns per unit of ”: the faster it turns, the denser the waves.

“The larger ω, the denser the waves”

From: Graphs and Their Transformations

Exhibit

Thinking sin|x| also has period 2π

The wrong step: is periodic with least positive period .
What went wrong? Think first, then open

Diagnosis For it is just , but for the graph is the mirror image of the right half. Take : , while . One counterexample is enough.

The right step: has just one axis of symmetry, , and it is not periodic. To show that “ is not a period”, it is enough to find one with .

“A period must work for every x; one x is enough to break it”

From: Period, Monotonicity and Symmetry

Exhibit

Copying the increasing-interval recipe when ω is negative

The wrong step: The increasing intervals of come from solving .
What went wrong? Think first, then open

Diagnosis The inner function is decreasing in . While increases, actually decreases as grows, so this gives exactly the decreasing intervals.

The right step: First rewrite it as . The increasing intervals of are the decreasing intervals of : .

“Make ω positive first, then substitute the whole block”

From: Period, Monotonicity and Symmetry

Exhibit

Distributing sin as if it were multiplication

The wrong step: , so .
What went wrong? Think first, then open

Diagnosis The right-hand side is about , which is bigger than : it cannot be a sine. The picture shows it clearly: the two pieces that make up the height of are not and ; each one is multiplied by a cosine, giving and .

The right step: .

“sin is not multiplication, so you cannot just open the brackets”

From: Sum, Double-Angle and Half-Angle Formulas

Exhibit

Taking only the positive root in a half-angle formula

The wrong step: and , so .
What went wrong? Think first, then open

Diagnosis If is in , then is in , an angle in quadrant II, whose cosine is negative. The formula only gives the size; the sign depends on where lands.

The right step: . (In fact and .)

“Before taking the root in a half-angle formula, work out the range of the half angle”

From: Sum, Double-Angle and Half-Angle Formulas

Exhibit

Dropping the minus sign in cos α − cos β

The wrong step: .
What went wrong? Think first, then open

Diagnosis Cosine is decreasing on , so and the difference must be positive. The formula is , with a minus sign in front. In the picture: is perpendicular to the angle bisector, and turning the bisector direction by turns its into .

The right step: . When you are done, use monotonicity to sanity-check the sign.

“Of the four formulas, only “cos minus cos” carries a minus sign”

From: Sum-to-Product and Product-to-Sum

Exhibit

Turning a sum into a product when the functions differ

The wrong step: .
What went wrong? Think first, then open

Diagnosis Sum-to-product needs the same function in both terms: both vectors must be projected onto the same axis. is a projection onto the -direction and onto the -direction, so they do not combine into one diagonal.

The right step: Make the names match first: , so .

“Same function first, then the product”

From: Sum-to-Product and Product-to-Sum

Exhibit

Finding the auxiliary angle from tan φ = b/a alone

The wrong step: : , take , so the expression is .
What went wrong? Think first, then open

Diagnosis Expand: , off by exactly a minus sign. only fixes the direction of a line, so could still be off by . The auxiliary angle must satisfy two conditions, and : the point is in quadrant II.

The right step: and , so and the expression is . Try on the workbench: the reading is , not .

“φ is the direction of the point (a, b), not just a slope”

From: The Auxiliary Angle and Evaluating Expressions

Exhibit

Using ±R as the extreme values on an interval

The wrong step: For , the minimum of is .
What went wrong? Think first, then open

Diagnosis would need , that is , which is not in the interval. The L may only turn through , so the hypotenuse never gets to point straight down.

The right step: , where the smallest value of sine is . So the minimum is (at ) and the maximum is (at ).

“After combining, find the range of x + φ first”

From: The Auxiliary Angle and Evaluating Expressions

Exhibit

Finding an angle from its sine and keeping only the acute one

The wrong step: In , , , . By the law of sines , so .
What went wrong? Think first, then open

Diagnosis In there are two angles whose sine is : and . The law of sines only tells you , not whether is acute or obtuse. Here , so , and neither angle clashes with .

The right step: or , and both triangles exist: the first has , the second . To get everything in one go, use the law of cosines instead and write a quadratic in : its two positive roots are exactly the two triangles.

“When you get an angle from its sine, ask: does the obtuse one work too?”

From: The Laws of Sines and Cosines

Exhibit

Concluding A = B straight from sin 2A = sin 2B

The wrong step: In , , so and the triangle is isosceles.
What went wrong? Think first, then open

Diagnosis Two angles in with the same sine are either equal or supplementary. Both and lie in , and is perfectly possible, for example , .

The right step: , so or : the triangle is isosceles or right-angled.

“Equal sines? Think “equal or supplementary” first”

From: Identities and Area in a Triangle

Exhibit

The tangent is −1, so the angle must be −π/4

The wrong step: , so .
What went wrong? Think first, then open

Diagnosis Angles with the same tangent differ by an integer multiple of , so the value of the tangent alone does not pin down the angle. Both and lie in , so their sum lies in : it cannot possibly be negative.

The right step: Estimate the range first: the sum lies in , and the only angle in that interval with tangent is . So .

“Estimate the range of the angle first, then use the trig value to pin it down”

From: Inverse Trigonometric Functions

Exhibit

Dividing both sides by sin x and losing roots

The wrong step: Solve : divide both sides by to get , so .
What went wrong? Think first, then open

Diagnosis Dividing by silently assumes . But when both sides are , so are solutions too, and they have been quietly thrown away.

The right step: Bring everything to one side and factor: , so or . The solutions are or ().

“A factor containing the unknown may be factored out, never cancelled”

From: Simple Trigonometric Equations

Exhibit

Solving sin x > ½ as if it were a linear inequality

The wrong step: and , so .
What went wrong? Think first, then open

Diagnosis Sine is not monotonic. satisfies , yet ; does not satisfy , yet … Nothing matches. The solution set comes in separate pieces, and it repeats every .

The right step: Look at the graph over one period: the arc above the horizontal line is . Then add the period: ().

“Draw one period first, then add the period”

From: Trigonometric Inequalities

Exhibit

Forgetting to restrict the angle after substituting

The wrong step: Range of : let ; then and .
What went wrong? Think first, then open

Diagnosis is non-negative, but can be negative. Only with is true; without that restriction the equation is wrong whenever is on the lower half of the circle, and the minimum is never actually reached.

The right step: Let with (then takes every value in exactly). So and .

“A substitution must match both ways: x covers its whole domain, and t ranges over exactly enough”

From: Trigonometric Substitution and Applications

Vol. 4 Complex Numbers and Vectors hall

Exhibit

Is “real part 0” enough for a purely imaginary number?

The wrong step: is purely imaginary, so and .
What went wrong? Think first, then open

Diagnosis For we get , which is a real number, not a purely imaginary one. A purely imaginary number needs real part and imaginary part not ; the second condition was dropped.

The right step: Solve to get , and then .

“Purely imaginary = real part zero + imaginary part nonzero: you need both conditions”

From: Complex Numbers and Their Algebra

Exhibit

Using √a·√b = √(ab) for negative numbers

The wrong step: .
What went wrong? Think first, then open

Diagnosis only holds for . Write the square root of a negative number as a multiple of first: , .

The right step: .

“When you see the square root of a negative number, bring out the i before you calculate”

From: Complex Numbers and Their Algebra

Exhibit

A negative modulus

The wrong step: is in trigonometric form, with modulus and argument .
What went wrong? Think first, then open

Diagnosis Trigonometric form requires . The modulus is a length and can never be negative. The minus sign in front amounts to a further turn of .

The right step: : modulus , principal argument .

“Trigonometric form has three requirements: r positive, cos first, a plus sign in the middle”

From: Modulus and Argument (I)

Exhibit

Reading the argument of cos θ − i sin θ as θ

The wrong step: , so .
What went wrong? Think first, then open

Diagnosis There is a minus sign in the middle, so this is not trigonometric form. The point is in the fourth quadrant; its argument is plus a full turn.

The right step: , so .

“Rewrite in standard trigonometric form first, then read off the argument”

From: Modulus and Argument (I)

Exhibit

Is a constant sum always an ellipse?

The wrong step: is the ellipse with foci .
What went wrong? Think first, then open

Diagnosis The distance between the two fixed points is , and the sum of the distances is also . Two sides of a triangle add up to more than the third side, with equality only when lies on the segment, so the locus is just a segment.

The right step: Compare with : if it is larger, an ellipse; if equal, the segment (here the part of the real axis); if smaller, the empty set.

“Compare the constant with the distance between the two fixed points before you conclude anything”

From: Complex Numbers and Equations

Exhibit

Using the discriminant to test for real roots when the coefficients are not real

The wrong step: The equation has discriminant , which is not a non-negative real number, so it has no real root.
What went wrong? Think first, then open

Diagnosis “If there are real roots” was derived for equations with real coefficients. When the coefficients contain , the discriminant cannot even be “greater than zero”. In fact, substitute : , so there is a real root.

The right step: Substitute a real root and separate real and imaginary parts: , so and . The real root is (the other root is ).

“The discriminant is only for real coefficients; with complex coefficients, turn the complex into the real first”

From: Complex Numbers and Equations

Exhibit

Subtraction pointing the wrong way

The wrong step: .
What went wrong? Think first, then open

Diagnosis For two vectors with the same starting point, the difference points from the head of the vector being subtracted to the head of the vector it is subtracted from: goes from to . Check: .

The right step: . Rule of thumb: same tail, join the heads, point at the first vector.

“After a subtraction, add back to check it”

From: Adding and Subtracting Vectors

Exhibit

Can lengths simply be added?

The wrong step: , , so .
What went wrong? Think first, then open

Diagnosis Adding vectors combines directions. You get only when point the same way; when they point opposite ways, and when they are perpendicular. In general there is only the triangle inequality: .

The right step: There is not enough information: all we know is that lies in . We also need the angle between them.

“When vectors add, their lengths usually do not”

From: Adding and Subtracting Vectors

Exhibit

Can you “cancel” in a dot product?

The wrong step: and , so .
What went wrong? Think first, then open

Diagnosis only sees the shadow of in the direction of , and infinitely many vectors cast the same shadow: with , both and have dot product .

The right step: The correct conclusion is , that is, (or ).

“A dot product is a number: the direction information is lost, so you cannot cancel”

From: The Inner Product

Exhibit

A negative dot product means an obtuse angle

The wrong step: , so the angle between and is obtuse.
What went wrong? Think first, then open

Diagnosis only says , that is, . When (the vectors point in opposite directions) the dot product is negative too, but that is not an obtuse angle.

The right step: “The angle is obtuse” and do not point in opposite directions. When you find a range for a parameter, remove the value that makes them opposite.

“Once you know the sign of the dot product, check the collinear case as well”

From: The Inner Product

Exhibit

Taking the angle between two lines to be obtuse

The wrong step: and , so the angle between the skew lines is .
What went wrong? Think first, then open

Diagnosis The angle between two lines lies in . The angle between two vectors can be obtuse, but a line has no direction, so an obtuse angle has to be replaced by its supplement.

The right step: Use : the angle between the lines is .

“For the angle between lines, put an absolute value in the cosine formula”

From: Vectors in Space

Exhibit

Using the cosine for the angle between a line and a plane

The wrong step: The line has direction , the plane has normal vector , and , so the angle between the line and the plane is .
What went wrong? Think first, then open

Diagnosis is the angle between the line and the normal; the angle between the line and the plane is its complement.

The right step: , so .

“A normal vector measures the angle with the normal: for the line–plane angle, take the complement”

From: Vectors in Space

Exhibit

Do all roots of x³ = 1 satisfy x² + x + 1 = 0?

The wrong step: , so , and therefore .
What went wrong? Think first, then open

Diagnosis A product of two factors being only tells you that one of them is . is also a root of , and .

The right step: Only when (that is, or ) do we get . Before using , make sure is one of the non-real cube roots.

“Before dividing by x − 1, make sure x ≠ 1”

From: Roots of Unity and Their Uses

Exhibit

One term too many

The wrong step: is a th root of unity (), so .
What went wrong? Think first, then open

Diagnosis What adds up to is exactly one full turn, terms: . The extra term makes the sum .

The right step: .

“Count the terms first, then cross out the full turns”

From: Roots of Unity and Their Uses

Exhibit

Treating the modulus of a sum as the sum of the moduli

The wrong step: , so .
What went wrong? Think first, then open

Diagnosis is the distance from the point to the point . is fixed, so runs around the circle of radius , and this distance ranges from to .

The right step: The triangle inequality gives , that is, ; the value is at and at .

“The triangle inequality gives a range, not an equation”

From: Modulus and Argument (II)

Exhibit

Remembering the equality condition as “equal”

The wrong step: if and only if .
What went wrong? Think first, then open

Diagnosis Equality needs the two vectors to point in the same direction; their lengths may differ: , but .

The right step: one of them is , or is a positive real number (same argument).

“For equality, look at the directions, not the sizes”

From: Modulus and Argument (II)

Exhibit

Rotating the wrong way

The wrong step: Construct the equilateral triangle outward on , so .
What went wrong? Think first, then open

Diagnosis Multiplying by turns counterclockwise by . Whether lands to the left or the right of depends on whether the vertices of the triangle go counterclockwise or clockwise; get it wrong and you construct the equilateral triangle pointing inward.

The right step: First agree that go counterclockwise. Then the triangle lies to the left of , and outward means turning clockwise: .

“Before you express a rotation as a product, fix the orientation of the vertices”

From: Applications of Complex Numbers and Vectors

Exhibit

Calling the points concyclic because a ratio is real

The wrong step: is real, so the four points are concyclic.
What went wrong? Think first, then open

Diagnosis When the four points lie on one line, both ratios are real and so is their quotient, yet the four points are not concyclic.

The right step: The conclusion should be “the four points are concyclic or collinear”. When a problem asks you to prove they are concyclic, also show that three of them are not collinear.

“A line is a “circle of infinite radius”; do not forget to rule it out”

From: Applications of Complex Numbers and Vectors

Vol. 5 Sequences and Mathematical Induction hall

Exhibit

Proving the step, forgetting the base case

The wrong step: Claim: is even. Proof: suppose is even. Then is even too. By induction, the claim holds for every .
What went wrong? Think first, then open

Diagnosis The inductive step really is fine, but for we get , which is odd. The first domino never falls, so it does not matter that each one “would knock over the next”. In fact is always even, so is always odd: the claim is false for every .

The right step: Induction needs both parts: first check , then prove . Choose “ is even” in the domino chain above: every gap shows ✓, yet not a single domino falls.

“Without the first push, even the strongest chain stays still”

From: The First Principle of Induction

Exhibit

Every step must work, including the smallest one

The wrong step: “Any horses all have the same color.” For it is obvious. Suppose any horses have the same color. Take horses. Remove the first: the remaining have the same color. Remove the last: those have the same color too. The two groups share a horse, so all have the same color.
What went wrong? Think first, then open

Diagnosis “The two groups share a horse” needs . Going from to , removing the first horse leaves and removing the last leaves : the two groups have no element in common, so the argument breaks at the step . One broken gap is enough to stop everything after it.

The right step: When you write the inductive step, check that the argument also works for the smallest . If it only works for , you must check separately (and here is exactly the one that is false).

“Break the chain in one place and nothing after it falls”

From: The First Principle of Induction

Exhibit

The step uses two earlier terms, but only one base case is checked

The wrong step: The sequence . “Proof” that : for , . Suppose and . Then .
What went wrong? Think first, then open

Diagnosis The inductive step uses two earlier terms. The first time it can be used is for , and then it needs and . But was never checked. In fact , then , and the conclusion is wrong all the way.

The right step: If the step uses several earlier terms, check that many base cases: here you must check both and . As soon as you do, you find that is false.

“Look back as many cells as each step looks back”

From: The Second Principle of Induction

Exhibit

Descending where there is no smallest element

The wrong step: Proof that has no solution in positive rationals: if is a solution, then so is , and it is smaller. This can go on getting smaller forever, a contradiction.
What went wrong? Think first, then open

Diagnosis is obviously a solution. The mistake: “getting smaller forever” is no contradiction at all among the positive rationals, since is an infinite decreasing sequence of positive rationals. The well-ordering principle only holds for the positive integers (or any set of integers that is bounded below).

The right step: The quantity that descends must be a positive integer (or a non-negative integer): a denominator , the value of one variable, , and so on. First make sure that the descending quantity takes positive integer values; only then say “it cannot keep getting smaller forever”.

“Descent needs positive integers, so that there is a floor”

From: Well-Ordering and Infinite Descent

Exhibit

Using Sₙ − Sₙ₋₁ for the general term and forgetting n = 1

The wrong step: Given , we have , so .
What went wrong? Think first, then open

Diagnosis only makes sense for ( is not defined). must be found separately: , while the formula gives at . Check: if held for every , then would be , not .

The right step: , and for , . So . The two pieces merge into one formula only when the constant term of is .

“Always check a₁ = S₁ separately”

From: General Terms and Sums

Exhibit

Treating aₘ + aₙ as aₘ₊ₙ

The wrong step: In the arithmetic sequence we have , because .
What went wrong? Think first, then open

Diagnosis , but . The property says that “two pairs of terms with the same index sum have the same sum”: when . Both sides must have two terms, and is only one.

The right step: . If you insist on comparing with , the right statement is , and : exactly the difference of .

“Pair the indices, two terms on each side”

From: Arithmetic and Geometric Sequences

Exhibit

Seeing 1, 2, 4, 8, 16 and deciding the next term is 32

The wrong step: Take points on a circle and join every pair by a chord (no three chords through one point). The numbers of regions are . So for there are regions.
What went wrong? Think first, then open

Diagnosis There are actually . The number of regions is , a polynomial of degree , which happens to agree with for the first five terms. Finitely many terms never determine the next one: any number at all can come next, if you allow polynomials of high enough degree.

The right step: The difference method comes with a condition: you must know that the sequence is arithmetic of order (for example, because you can prove from the problem that its general term is a polynomial of degree ), and then use at least terms to pin it down. Type into the difference machine: only the fourth differences are constant.

“A pattern must be proved, not just seen”

From: Higher-Order Arithmetic Sequences and Differences

Exhibit

The initial values are a₁, a₂, but the constants are solved as if they were a₀, a₁

The wrong step: . The characteristic roots are . Put ; from we get .
What went wrong? Think first, then open

Diagnosis The system of equations uses and , but the problem gives and . Check: the formula gives , while the problem says . The whole sequence has been shifted by one place.

The right step: Set up the equations with : and , which give , that is, . Check that .

“Once you have the constants, substitute the initial values back to check”

From: Recursive Sequences

Exhibit

Combining two periods and thinking only of the least common multiple

The wrong step: A sequence has period and also period , so its least positive period is .
What went wrong? Think first, then open

Diagnosis is certainly a period, but it is not the least one. From and we get (go forward , then back ). So is a period too. In general, the greatest common divisor of two periods is also a period (by Bézout's identity, ).

The right step: Periods can be “subtracted”: is a period. Whether the least positive period is or depends on the sequence itself. Also, every period is a multiple of the least positive period.

“Periods can be subtracted, so look at the greatest common divisor”

From: Periodic Sequences

Exhibit

The number of tilings is off from the Fibonacci index by one

The wrong step: A 2×n strip has tilings, so a 2×5 strip has tilings.
What went wrong? Think first, then open

Diagnosis In this book's notation . A 2×1 strip has tiling and a 2×2 strip has (two upright or two lying), so the tiling counts start — one place ahead of , which starts .

The right step: A 2×n strip has tilings; a 2×5 strip has . Line up the indices with the two or three smallest before using a formula.

“Check the indices against n = 1, 2 first”

From: The Fibonacci Sequence

Exhibit

Treating a necessary condition as a sufficient one

The wrong step: We know: if is prime (), then is prime. is prime, so is prime.
What went wrong? Think first, then open

Diagnosis The first sentence says: if is composite (), then has a proper factor ( a proper divisor of ), so it is not prime. It does not say that is prime whenever is. In fact .

The right step: “ is prime” is only a necessary condition for to be prime. To decide about you have to factor it separately.

“The converse needs its own proof”

From: The Fibonacci Sequence

Exhibit

Proving it for powers of 2 only, then claiming it for every n

The wrong step: For , ; if it holds for , it holds for . By mathematical induction the AM–GM inequality holds for every .
What went wrong? Think first, then open

Diagnosis Starting from , “” only reaches . It never touches : only a few rungs of the ladder are lit, and the ones in between stay dark.

The right step: Add one more step, “” (or in general for ): above every there is a proved , and you walk down from there one rung at a time.

“If you jumped up, you must also be able to climb down”

From: Several Proofs of the AM–GM Inequality

Exhibit

Padding with the wrong number

The wrong step: To prove the case of numbers, append a to and use the result for numbers: .
What went wrong? Think first, then open

Diagnosis That inequality is true, but it does not give : the added changes the mean, and both sides now contain something you do not want.

The right step: Pad with their own mean . The mean of the numbers is still , so , that is , so .

“The number you add must leave the mean unchanged”

From: Several Proofs of the AM–GM Inequality

Exhibit

Step size 2, but only one base case checked

The wrong step: We want for every . We checked and proved , so the statement holds.
What went wrong? Think first, then open

Diagnosis knocks over , knocks over , … only the odd numbers fall. Nobody pushes , so not one domino in the even chain falls.

The right step: With step size you must check consecutive base cases . Each base starts one chain, and together they cover every remainder on division by .

“Step size k means k base cases”

From: Choosing the Right Step Size

Exhibit

A “convenient” step size whose step does not actually work

The wrong step: If cents can be paid, so can cents (): just pay cent more.
What went wrong? Think first, then open

Diagnosis There are only -cent and -cent stamps, no -cent ones. The inductive step must really be doable: from a way of paying , build a way of paying .

The right step: Either change the step size — add one -cent stamp, — or do step size properly: if the payment uses a , swap one for two s; if it has no (so at least three s), swap three s for two s.

“Choose the step size so that the step really works”

From: Choosing the Right Step Size

Exhibit

Inducting on m, but “freezing” n as well

The wrong step: Induct on to prove . Induction hypothesis: . Then
What went wrong? Think first, then open

Diagnosis The step needs : the second variable has changed from to . If the hypothesis only says “true for this ”, it is of no use.

The right step: Put the other variable into the hypothesis, and make it arbitrary: “for every , ”. Then use it at : .

“Induct on one variable, and let the hypothesis cover every value of the other”

From: Choosing What to Induct On

Exhibit

Inducting on m+n when the step does not make the sum smaller

The wrong step: The inductive step for uses . Induct on : is a smaller case, so it holds by the induction hypothesis.
What went wrong? Think first, then open

Diagnosis : the sum has not decreased, so this case is not in the hypothesis “”. Inducting on only works if every case the step uses has a smaller sum.

The right step: Look for the quantity that really decreases in this step: here it is . Induct on , assuming the statement for and every .

“Whatever you induct on must get smaller at every step”

From: Choosing What to Induct On

Exhibit

“The added term is tiny”, so it cannot go over

The wrong step: Suppose . Then , and is tiny, so .
What went wrong? Think first, then open

Diagnosis The induction hypothesis only tells you ; it says nothing about how far is from . If were exactly (the hypothesis allows that), adding any positive number would go over . “Tiny” is not a proof.

The right step: Write “how far from ” into the statement: prove . The inductive step then needs exactly , which is clearly true.

“Room that is not in the hypothesis cannot be used in the conclusion”

From: Changing the Statement

Exhibit

Strengthening too much

The wrong step: To be safe, prove the stronger . The inductive step holds, so the original statement is proved.
What went wrong? Think first, then open

Diagnosis The inductive step does hold, but the base case fails: for , while . The strengthened statement is simply false, so of course it cannot be proved.

The right step: A strengthening must satisfy both ends: the step must go through (enough room), and the base case must stand (not too much room). For , the step needs and the base needs : only works.

“The strengthened statement must, first of all, be true”

From: Changing the Statement

Exhibit

Checking five cases and calling it a proof

The wrong step: Join points on a circle by all possible chords: the disc is cut into regions (). So points give regions.
What went wrong? Think first, then open

Diagnosis For there are only regions, not . The first five agreeing is a coincidence: the true answer is , a polynomial of degree four, while is exponential; the two agree only for the first few terms.

The right step: Small cases are only for guessing. A guessed formula must be confirmed by induction (or some other proof). Try to prove and you will find you cannot give any reason for the step “one more point doubles the number of regions”.

“Checking cases gives only a conjecture”

From: Guess, Then Prove

Exhibit

Secretly using the formula you are proving

The wrong step: Guess . Proof: ; substitute into the recurrence , check that both sides are equal, so the guess is right.
What went wrong? Think first, then open

Diagnosis This uses the very conclusion to be proved for . The right direction is: assume only , compute from the recurrence, and see whether it comes out as .

The right step: . Work from the hypothesis to the conclusion, not backward from the conclusion.

“Assume only the n-th term; compute the (n+1)-th”

From: Guess, Then Prove

Exhibit

I could not arrange it, so it does not exist

The wrong step: Arrange to by “always place the smallest number allowed”: , and halfway through there is nothing left to place. So no such arrangement exists.
What went wrong? Think first, then open

Diagnosis One failed attempt only shows that this way of arranging does not work. “Does not exist” is a claim about every arrangement: either check them all, or give a reason that applies to every arrangement. In fact works.

The right step: To prove existence, look for a construction; to prove non-existence, find a reason that “no arrangement can escape” (counting, pigeonhole, parity, an invariant).

“One failure does not prove non-existence”

From: Existence Problems for Sequences

Exhibit

The pigeonhole principle says “there is one”, so you think you know which

The wrong step: Any sequence of distinct numbers has a monotone subsequence of length , so the first terms (or the last ) must be monotone.
What went wrong? Think first, then open

Diagnosis The pigeonhole principle only guarantees that such a subsequence exists; it does not say where. In neither the first nor the last terms are monotone; a monotone subsequence is , taken with gaps.

The right step: An indirect existence proof only tells you “there is one”; to point to it you must search separately (for example by following the labels used in the proof).

“An indirect proof guarantees there is one, but does not point the way”

From: Existence Problems for Sequences

Vol. 6 The Mean Inequalities and the Cauchy Inequality hall

Exhibit

Forgetting the “non-negative” condition

The wrong step: For every , .
What went wrong? Think first, then open

Diagnosis The mean inequality only holds for non-negative numbers. At the left side is , far below . The arithmetic mean of two negative numbers is negative, while the geometric mean is positive.

The right step: Split into cases: for , ; for , apply the inequality to and get .

“Before using AM–GM, make sure every term is non-negative”

From: The Mean Inequalities

Exhibit

Writing “≥” without saying when equality holds

The wrong step: and , so and the maximum of is . (No word on when it is reached.)
What went wrong? Think first, then open

Diagnosis Here it happens to be right: it is reached at . But “” only says that is an upper bound. Whether it is the maximum depends on whether equality can actually hold. The next stop and Chapter 2 show many bounds that cannot be reached.

The right step: Every time you use the mean inequality, add a sentence “equality holds if and only if …”, and check that this condition is compatible with the other conditions of the problem.

“An inequality is only half the story; the equality condition is the other half”

From: The Mean Inequalities

Exhibit

Averaging pairs and assuming everything becomes equal in finitely many steps

The wrong step: For , keep replacing two numbers by their mean. After finitely many steps all three become , and the product keeps growing, so .
What went wrong? Think first, then open

Diagnosis The conclusion is right, the argument is not. After averaging, the new number equals if and only if the two averaged numbers add up to , which means the third number is already . There is no at the start, so a can never appear: the process does not stop after finitely many steps.

The right step: Use the adjustment method: at each step make one number exactly equal to the mean , and let the other one absorb the difference. Each step adds one more , so it ends after at most steps. (Or take a limit of the infinite process, but that needs more argument.)

“For an argument by “finitely many adjustments”, every step must make definite progress on something”

From: Proofs of the Mean Inequalities

Exhibit

Taking “half and half” as the weights without thinking

The wrong step: ; adding the cyclic versions gives .
What went wrong? Think first, then open

Diagnosis The conclusion happens to be true, but the first step is false: with , the left side is and the right side is . On the right, the average exponent of is , not : the two sides do not match up.

The right step: Choose the weights so that the exponents match: the exponent of is , so , that is, .

“Match the exponents first, then write the inequality”

From: Proving Inequalities with AM–GM

Exhibit

Using AM–GM twice with clashing equality conditions

The wrong step: and . Since and , the minimum of is .
What went wrong? Think first, then open

Diagnosis The two inequalities need and for equality, but then . Equality cannot be reached, so is only a lower bound, not the minimum. In fact at the product is , and that is the minimum.

The right step: Expand: . Since and is decreasing on , we get ; also . Both equalities hold at , so they are compatible, and the minimum is .

“Every “≥” must reach equality at the same point”

From: Extrema with AM–GM

Exhibit

A chain of inequalities, each with its own equality case

The wrong step: For a triangle with perimeter : , so the maximum area is .
What went wrong? Think first, then open

Diagnosis Every step is true, but the three equalities need , and (that is, ) respectively. They cannot all hold at once, so is never reached. The true maximum is , less than half of that.

The right step: Switch to the tangent lengths : and , with equality at , the equilateral triangle with side . This one is reached.

“In a chain of bounds, all the equality cases must hold at once, in the same figure”

From: AM–GM in Geometric Inequalities

Exhibit

Minimizing 1/a + 4/b under a + b = 1 with two AM–GMs in a row

The wrong step: , so the minimum is .
What went wrong? Think first, then open

Diagnosis The first equality needs , that is, ; the second needs . Both hold together only at , which is impossible. is never reached; it is only a lower bound.

The right step: Multiply by : , with equality at , that is, , , which is allowed. The minimum is .

“Get there with one application; do not let two equality cases fight”

From: Variants of AM–GM

Exhibit

Bounding pair by pair without a parameter

The wrong step: and ; adding, , so the maximum is .
What went wrong? Think first, then open

Diagnosis Every step is true, but equality cannot be reached. The first two steps need , and the last step enlarges and to and , so equality there needs , and then is as well. The true maximum is .

The right step: Split in an unknown ratio , and choose so that the three square terms get equal coefficients: .

“Wherever the coefficients are unbalanced, that is where the slack hides”

From: AM–GM with Parameters

Exhibit

Writing only one sum of squares

The wrong step: By the Cauchy inequality, .
What went wrong? Think first, then open

Diagnosis The right-hand side of the Cauchy inequality is the product of two sums of squares. Here the second list is ; its sum of squares is , and you cannot drop it. Take : the left side is , the right side , so the inequality as written is simply false.

The right step: , that is, .

“Write down both lists first, then apply the formula”

From: The Cauchy Inequality and Its Proofs

Exhibit

Mistaking “proportional” for “equal”

The wrong step: , with equality when .
What went wrong? Think first, then open

Diagnosis Take : the left side is and the right side , not equal. The two lists are and ; equality needs them to be proportional, not equal to .

The right step: Equality holds if and only if , i.e. . Check: at both sides are .

“Equality is about the ratio between the two lists, not about the numbers inside one list”

From: The Cauchy Inequality and Its Proofs

Exhibit

Merging even with a negative denominator

The wrong step: .
What went wrong? Think first, then open

Diagnosis The left side is , far below . The fraction form comes from the Cauchy inequality applied to and , and needs .

The right step: Before using , make sure every denominator is positive.

“Only positive denominators can be merged”

From: Variants and Generalizations of Cauchy

Exhibit

Merging when the numerator is not a square

The wrong step: .
What went wrong? Think first, then open

Diagnosis The left side is only . The merging formula needs each numerator to be a square: . When the numerator is , the matching is , not .

The right step: Write ; that is correct. Or first multiply top and bottom by the same number to make the numerator a square.

“Read each numerator as a square first, then add up the things being squared”

From: Variants and Generalizations of Cauchy

Exhibit

Carrying on while the products still contain variables

The wrong step: Given , prove : by Cauchy, the left side is .
What went wrong? Think first, then open

Diagnosis The first step is fine, but the right side still contains variables. And it is in fact at most (Cauchy again), so the second step goes the wrong way.

The right step: Here bound each factor on its own: and ; multiplying gives .

“A good construction makes the products constant; if variables are left, check the direction first”

From: Proving Inequalities with Cauchy

Exhibit

A lower bound that is never reached

The wrong step: , , , so , hence the minimum is .
What went wrong? Think first, then open

Diagnosis Equality needs , i.e. , but the problem requires . The inequality itself is true; the value simply can never be reached.

The right step: On the function is increasing, so its minimum is at : .

“Before you write “the minimum is”, put the equality condition back into the constraints”

From: Proving Inequalities with Cauchy

Exhibit

Checking proportionality for equality but forgetting the sign

The wrong step: , . From the equality condition , take .
What went wrong? Think first, then open

Diagnosis Substituting into the second equation gives , not . In the proportionality the factor can be positive or negative; needs .

The right step: Set ; substituting into gives , so the solution is . Then check .

“Write the equality condition as $(x,y)=k(a,b)$, then let the equations decide $k$”

From: Cauchy for Systems and Extrema

Exhibit

The merged lower bound is smaller than the target

The wrong step: To prove : .
What went wrong? Think first, then open

Diagnosis The first step is right, the second is wrong: by Cauchy, , so this lower bound is at most and only reaches it when . The merge threw away too much.

The right step: Use instead. Merging gives , and always holds.

“Before merging, ask whether the step that remains will still be true”

From: Cauchy and Fractional Inequalities

Exhibit

Treating “the average row” as “every row”

The wrong step: A board without rectangles has filled cells. On average a row has cells and uses column pairs, so ; and from this we conclude that every row has exactly cells.
What went wrong? Think first, then open

Diagnosis Different rows can have different numbers of cells. What lets you replace by “the case of the average” is the Cauchy inequality , and that only gives an inequality; it does not say the rows really are equal.

The right step: Write it out: (Cauchy), so . Equal rows are only the equality condition of Cauchy.

“When Cauchy replaces “every row” by “the average”, that is an inequality, not an assumption”

From: Cauchy in Combinatorial Estimates

Exhibit

Using AM–GM twice with clashing equality conditions

The wrong step: , . , so the minimum is .
What went wrong? Think first, then open

Diagnosis Equality in the first step needs , i.e. ; equality in the second step, , needs . The two conditions cannot hold together, so is never reached. The true minimum is about .

The right step: Use Cauchy (fraction form), which gets there in one step: , with equality when .

“When you chain several inequalities, their equality conditions must hold together”

From: AM–GM and Cauchy Together

Vol. 7 Analytic Geometry hall

Exhibit

Taking the signed value for the length

The wrong step: On the number line has coordinate and has coordinate , so .
What went wrong? Think first, then open

Diagnosis is the length . The signed value also carries a direction: going from to means moving in the negative direction, so . Length only asks “how far”; the signed value also asks “which way”.

The right step: Always take “end coordinate minus start coordinate”: , and .

“Signed values have a sign, lengths do not; write the end first, then subtract the start”

From: Directed Segments

Exhibit

Thinking λ can be any real number

The wrong step: On the line there is always a point with .
What went wrong? Think first, then open

Diagnosis means , that is , so : the two endpoints coincide, which contradicts having a segment. The farther runs, the closer gets to , but it never gets there.

The right step: The denominator in the division formula already warns you: .

“A denominator in a formula marks the value the quantity can never take”

From: Directed Segments

Exhibit

Putting the whole segment in the denominator of λ

The wrong step: is the trisection point of nearer to , so .
What went wrong? Think first, then open

Diagnosis For a dividing point, is : the piece before the point over the piece after it, not over the whole segment. The trisection point nearer to splits the segment as , so .

The right step: Substitute into to get , which is exactly “one third of the way from ”.

“From the start to the dividing point, then from the dividing point to the end”

From: Basic Problems of the Coordinate Method

Exhibit

Getting a negative area and assuming it is wrong

The wrong step: , , : . An area cannot be negative, so something must have gone wrong.
What went wrong? Think first, then open

Diagnosis The determinant gives a signed area: positive when the vertices go counterclockwise, negative when they go clockwise. is clockwise (up first, then down to the right), so is correct.

The right step: Take the absolute value: . To add the areas of several triangles, list all their vertices in the same direction, or take each absolute value before adding.

“The minus sign tells you the direction, not that you made a mistake”

From: Basic Problems of the Coordinate Method

Exhibit

Choosing coordinates “too well” and losing the general case

The wrong step: To prove a property of a triangle, take , , .
What went wrong? Think first, then open

Diagnosis With this choice, has quietly been restricted to an isosceles right triangle (). You may choose the coordinate system freely, but the coordinate system must not decide the shape of the figure “on the side”.

The right step: Choosing the axes only uses up the freedom of the figure’s position (translation and rotation). Once the triangle is placed, it still has independent quantities: , , , with , , unrelated.

“You may pick the axes, not the figure”

From: Choosing a Good Coordinate System

Exhibit

Writing down only half of the locus

The wrong step: The locus of points at distance from the -axis has equation .
What went wrong? Think first, then open

Diagnosis The distance from a point to the -axis is , not . The point is also at distance from the -axis but does not satisfy : condition ① (every point of the curve satisfies the equation) is broken.

The right step: , i.e. or : two lines. It can also be written as .

“Distances need absolute values, and removing an absolute value means splitting into cases”

From: Curves and Equations

Exhibit

Squaring, and gaining points that do not belong

The wrong step: Square both sides of the curve to get its equation .
What went wrong? Think first, then open

Diagnosis requires : it is the upper half of the circle, while is the whole circle. The point satisfies the second but is not on the original curve: squaring was not an equivalent transformation.

The right step: Write . At every step of simplifying an equation, ask: could this step add solutions, or lose solutions?

“Simplifying only works if every step is an equivalence”

From: Curves and Equations

Exhibit

“Equal intercepts”, but only the intercept form was tried

The wrong step: The line through with equal intercepts on the two axes: let ; substituting gives , so the only answer is .
What went wrong? Think first, then open

Diagnosis The intercept form requires both intercepts to be nonzero. The line through the origin has both intercepts equal to , so they are “equal” too, yet the intercept form rules it out.

The right step: Split into two cases. Nonzero intercepts give ; zero intercepts mean the line passes through the origin, giving . There are two answers.

“Whatever form you use, first ask which lines it leaves out”

From: Equations of a Line

Exhibit

Saying a vertical line has slope 0

The wrong step: The line is perpendicular to the -axis, its angle of inclination is , so its slope is .
What went wrong? Think first, then open

Diagnosis Slope belongs to a horizontal line (angle of inclination ). is undefined, so a vertical line has no slope; in the formula the denominator is .

The right step: has angle of inclination and no slope. It cannot be written in point–slope or slope–intercept form, but it can be written in general form as .

“Slope 0 is flat; no slope is vertical”

From: Equations of a Line

Exhibit

Getting the sign of the normalizing factor wrong

The wrong step: To put in normal form, divide both sides by : , so .
What went wrong? Think first, then open

Diagnosis The constant term of the normal form is , which must be . After multiplying by the constant term is , so this is not the normal form. The sign of must be opposite to that of : , so take .

The right step: : , , .

“Give μ the sign opposite to C, and p will never be negative”

From: The Normal Form of a Line

Exhibit

Taking θ for the angle of inclination of the line

The wrong step: The line in normal form has angle of inclination .
What went wrong? Think first, then open

Diagnosis is the direction of the normal , and the line is perpendicular to it. The angle of inclination of the line is or (whichever lies in ).

The right step: Here the angle of inclination is . Check: the slope is .

“θ points at the line; α runs along it”

From: The Normal Form of a Line

Exhibit

Distance between parallel lines without matching the coefficients

The wrong step: The distance between and is .
What went wrong? Think first, then open

Diagnosis The formula needs the two equations to have exactly the same and . The coefficients of and in the second equation are twice those in the first, so its constant term must be divided by as well.

The right step: Write the second line as ; the distance is . Or take a point on the first line and find its distance to the second: .

“Match A and B first, then subtract the C’s”

From: Relations between Points and Lines

Exhibit

Setting up l₁ + λl₂ = 0 and losing l₂

The wrong step: Find the line through the intersection of and that also passes through . Let ; substituting gives , which has no solution, so no such line exists.
What went wrong? Think first, then open

Diagnosis happens to lie on (). The family contains every line through the intersection point except itself, so “no solution” is telling you precisely that the answer is .

The right step: The answer is . To miss nothing, you can use (, not both ) instead.

“When λ has no solution, first check whether l₂ is the answer”

From: Families of Lines

Exhibit

Seeing $x^2+y^2$ and calling it a circle

The wrong step: is a circle with center .
What went wrong? Think first, then open

Diagnosis Completing the square gives . The left side is a sum of two squares and can never be negative, so no point at all satisfies the equation. already told you so.

The right step: Compute first: greater than gives a circle, equal to a single point, less than no points at all.

“After completing the square, check the sign of the right side first”

From: The Circle

Exhibit

Reading off the center when the square terms do not have coefficient 1

The wrong step: The center of is .
What went wrong? Think first, then open

Diagnosis The formula is for an equation in which and both have coefficient . Here the coefficient is , so plugging in directly doubles the center.

The right step: Divide both sides by first: , so the center is and .

“Make the square-term coefficients 1 first”

From: The Circle

Exhibit

Always putting the foci on the $x$-axis

The wrong step: The foci of are .
What went wrong? Think first, then open

Diagnosis The major axis follows the larger denominator. Here is under , so the major axis is the -axis, with and . The points are not on the major axis at all; they are even outside the ellipse ().

The right step: , so the foci are .

“Compare the denominators first; the larger one tells you which axis is the major axis”

From: The Ellipse

Exhibit

Computing the ellipse's $c^2$ as $a^2+b^2$

The wrong step: For the half focal distance is .
What went wrong? Think first, then open

Diagnosis In an ellipse, is the hypotenuse: the distance from an end of the minor axis to a focus is exactly (the string folded in half), so . With larger than , the foci would lie outside the ellipse.

The right step: , so . ( is the relation for the hyperbola.)

“In an ellipse, a is the largest”

From: The Ellipse

Exhibit

Getting the asymptote slopes upside down

The wrong step: The asymptotes of are .
What went wrong? Think first, then open

Diagnosis The transverse axis of this hyperbola is the -axis, and and got mixed up when applying . Check: for large , , so .

The right step: Replace the on the right side by : , that is, . This way you never have to remember which is and which is .

“For the asymptotes, replace 1 by 0”

From: The Hyperbola

Exhibit

Finding a hyperbola's foci with the ellipse relation

The wrong step: The foci of the hyperbola are .
What went wrong? Think first, then open

Diagnosis is the relation for the ellipse. The vertices of the hyperbola are at , yet these foci at would lie between the two branches, while each focus of a hyperbola should be “wrapped” inside its own branch.

The right step: For a hyperbola , so the foci are . How to remember: is half the diagonal of the box.

“In an ellipse a is the largest; in a hyperbola c is the largest”

From: The Hyperbola

Exhibit

Reading off the focus without the standard form

The wrong step: The focus of is .
What went wrong? Think first, then open

Diagnosis In the squared term is , so the axis of symmetry is the -axis and the focus must lie on the -axis. Also, is not : first write the equation in the form .

The right step: , so , ; the focus is and the directrix is .

“First write it as x²=2py or y²=2px, then read off p”

From: The Parabola

Exhibit

Taking $p$ as the $x$-coordinate of the focus

The wrong step: In , , so the focus is .
What went wrong? Think first, then open

Diagnosis is the distance from the focus to the directrix, and the vertex lies halfway between them, so the focus is only from the vertex.

The right step: , the focus is and the directrix is .

“The focus is at p/2”

From: The Parabola

Exhibit

Using the replacement rule at a point off the curve

The wrong step: Tangents from the point to the ellipse : the replacement rule gives , that is, .
What went wrong? Think first, then open

Diagnosis The replacement rule gives the tangent only at a point on the curve. is not on the ellipse, and the line does not even pass through ; it is actually the line joining the two points of tangency (the chord of contact).

The right step: Let the tangent be , substitute into the ellipse and set the discriminant to : , , . Both points of tangency have -coordinate : they lie exactly on that “wrong answer”.

“First check whether the point is on the curve”

From: Tangents and Normals of Conics

Exhibit

Getting the signs of the translation formulas backwards

The wrong step: : put , and get .
What went wrong? Think first, then open

Diagnosis You need to write as . The new origin is in the old coordinates, and the formulas are , , that is, , .

The right step: , : , . The equation becomes , with center .

“The old coordinates of the new origin are (h, k), and the brackets hold a minus sign”

From: Simplifying by Translating the Axes

Exhibit

Remembering the angle formula with the wrong sign

The wrong step: To remove the -term of : , .
What went wrong? Think first, then open

Diagnosis Substitute back into : , , so . With the sign flipped, the -term actually got bigger.

The right step: gives , . If in doubt, check with the formula for .

“cot 2θ = (A − C)/B, with A first”

From: Simplifying by Rotating the Axes

Exhibit

Taking half the $xy$-coefficient as $B$

The wrong step: : , , so it is of elliptic type.
What went wrong? Think first, then open

Diagnosis Here is the whole coefficient of the -term, . (Some books write the equation as ; the discriminant is then . Do not mix the two conventions.)

The right step: : parabolic type. In fact , which is a parabola.

“First make sure which convention the equation is written in”

From: The General Quadratic Equation

Exhibit

Forgetting the one member the family leaves out

The wrong step: Find a circle or line through the intersection points of the circle and the line that also passes through : let ; substituting gives , so there is none.
What went wrong? Think first, then open

Diagnosis can never be itself (that would be ). The point lies on the line , and that very line is the answer.

The right step: Write ( not both ). Substituting gives : the answer is the line . The three points are collinear, so a circle through them really does not exist.

“When you get a contradiction, check λ = ∞ first”

From: Families of Conics

Exhibit

Using $\theta=\arctan\dfrac yx$ without looking at the quadrant

The wrong step: The point : , , so its polar coordinates are .
What went wrong? Think first, then open

Diagnosis only takes values in , so it only ever points into the first or fourth quadrant. is in the second quadrant, but is the fourth-quadrant point : exactly half a turn away.

The right step: only tells you the line the terminal side lies on. Now look at and take the angle in the second quadrant: , so the polar coordinates are .

“The tangent only sees the line; the quadrant is your job”

From: Polar Coordinates

Exhibit

Seeing $\theta$ in the first quadrant and assuming the point is there too

The wrong step: The point has polar angle , so it lies in the first quadrant.
What went wrong? Think first, then open

Diagnosis When the point is not on the terminal side of but on its extension backward through the pole. and are the same point, in the third quadrant.

The right step: The formulas are the safest way: , . Both are negative.

“Negative ρ means walk backward”

From: Polar Coordinates

Exhibit

One substitution fails, so you conclude there is no symmetry

The wrong step: : replacing by gives . The equation changed, so the curve is not symmetric about the polar axis.
What went wrong? Think first, then open

Diagnosis The mirror image of in the polar axis can be written as , but also as , and a point of the curve may satisfy the equation under only one of its names. Substitute the second form: . The equation is unchanged, so the curve is in fact symmetric.

The right step: Equation unchanged after a substitution ⇒ definitely symmetric. Equation changed ⇒ no conclusion yet: try the other form, or look at the graph and test points.

“A substitution test is sufficient, not necessary”

From: Equations of Curves in Polar Coordinates

Exhibit

Letting θ run only from 0 to 2π and thinking the curve is complete

The wrong step: has period , but one full turn of the polar angle is , so letting go from to draws the whole curve.
What went wrong? Think first, then open

Diagnosis and point the same way, but : changes sign and the point lands on the other side. Drawing only loses half the curve.

The right step: To graph , let run through a full period of (here ), or first prove that the second half is a mirror image of the first.

“One turn of the ray is not one period of the equation”

From: Equations of Curves in Polar Coordinates

Exhibit

Taking $p$ to be the distance from the directrix to the center

The wrong step: For the ellipse , and the directrix is , so and the polar equation is .
What went wrong? Think first, then open

Diagnosis In , is the distance from the focus to its directrix, not from the center to the directrix. is the distance from the directrix to the center ; you still have to subtract .

The right step: , and (half the latus rectum). Check: at , , which is exactly half the latus rectum.

“ep is half the latus rectum; check it with θ = π/2”

From: Simplifying Calculations and Proofs with Polar Coordinates

Exhibit

Copying the ellipse formula for a focal chord of a hyperbola, with absolute values

The wrong step: A focal chord of a hyperbola also satisfies .
What went wrong? Think first, then open

Diagnosis When the chord meets both branches of the hyperbola, and are on the same side of the focus and one of the radii is negative. The identity always holds for the signed ; with absolute values it becomes .

The right step: In the widget, set to and turn : watch when the readout turns yellow with “”. First decide whether the chord meets one branch or two, then decide between sum and difference.

“The formula holds for signed ρ; absolute values need cases”

From: Simplifying Calculations and Proofs with Polar Coordinates

Exhibit

Taking the $\varphi$ in the ellipse’s parametric equations to be the inclination of $OP$

The wrong step: On the ellipse , the point where makes a angle with the -axis is the point with , namely .
What went wrong? Think first, then open

Diagnosis is the eccentric angle: it is the angle of the point on the auxiliary large circle, not the angle of . The line from to has inclination with , about , not .

The right step: You need , that is , so . The two angles agree only for .

“The eccentric angle lives on the auxiliary circle, not on OP”

From: Parametric Equations of Curves

Exhibit

Treating $|t|$ as a distance when the direction vector is not a unit vector

The wrong step: On the line , the point with is at distance from .
What went wrong? Think first, then open

Diagnosis The distance from to is . Only when the direction is a unit vector such as is the distance.

The right step: The distance is . Or first rewrite in standard form, ; then is the signed distance.

“t is a distance only if the coefficients’ squares add up to 1”

From: Parametric Equations of Curves

Exhibit

Handing in as soon as the parameter is gone, forgetting the ranges of x and y

The wrong step: : from we get , so the curve is a parabola.
What went wrong? Think first, then open

Diagnosis can never be negative. The relation used in the elimination also holds for , so the Cartesian equation “invites in” the left half as well, although no ever gets there.

The right step: The answer must be written : the right half of the parabola (including the vertex ).

“Was the elimination reversible? Take a look at the ranges of x and y”

From: Eliminating and Choosing Parameters

Exhibit

Changing the parameter and quietly losing a point

The wrong step: The circle can be written , and the two describe the same curve.
What went wrong? Think first, then open

Diagnosis These come from setting , and is undefined at . The corresponding point is never reached, whatever is: would need , that is .

The right step: Either add “together with the point ”, or use instead.

“The half-angle substitution loses the point θ = π”

From: Eliminating and Choosing Parameters

Exhibit

Finding the “chord with a given midpoint” from $t_1+t_2=0$ without checking that the chord exists

The wrong step: The chord of the ellipse with midpoint : from the slope is , so the chord lies on the line .
What went wrong? Think first, then open

Diagnosis is outside the ellipse (). Substituting into the ellipse gives with discriminant : this line does not meet the ellipse at all, so where is the chord? only means something when there are two intersection points.

The right step: First check that is inside the curve (substituting gives a value less than ), or, after finding the line, check that the discriminant is positive. When is outside the ellipse, no chord has it as midpoint.

“Before Vieta, ask the discriminant”

From: Simplifying Calculations and Proofs with Parametric Equations

Exhibit

Taking $t_1t_2$ to be $|P_0A|\cdot|P_0B|$ directly

The wrong step: is inside the circle , and a chord through it is cut into two pieces whose product is .
What went wrong? Think first, then open

Diagnosis is a signed distance: when lies between and , and have opposite signs and their product is negative. A product of lengths needs an absolute value.

The right step: . Conversely, the sign of itself tells you whether lies between and .

“t carries a sign; lengths need absolute values”

From: Simplifying Calculations and Proofs with Parametric Equations

Exhibit

“Above the line means $>0$”

The wrong step: is the region above the line .
What went wrong? Think first, then open

Diagnosis Test a point: the origin gives , but the origin is below the line . The coefficient of is , so the larger is, the smaller is: above the line, . “Above means ” is true only when the coefficient of is positive.

The right step: Rewrite the inequality as before reading it, or simply test a point not on the line: is the region below the line (not including the line).

“When in doubt, test a point”

From: How Curves Divide the Plane into Regions

Exhibit

Testing with the origin when the origin is on the curve

The wrong step: Which part is ? Substituting the origin gives , so… no idea. Maybe the whole plane?
What went wrong? Think first, then open

Diagnosis The origin lies on the parabola , and cannot stand for either side. The test point must not be on the curve.

The right step: Use another point, for example : , and it is “inside” the parabola (on the side it opens toward). So is the piece outside the parabola.

“Keep the test point off the curve”

From: How Curves Divide the Plane into Regions

Exhibit

Comparing only the vertices when the objective is not linear

The wrong step: Minimum of on the triangle with vertices , , : the values at the three vertices are , , , so the minimum is .
What went wrong? Think first, then open

Diagnosis “The extremum is at a vertex” holds only for a linear objective: its level curves are lines, and a line sliding off a convex polygon touches a vertex last. The level curves of are circles centered at the origin, and a growing circle may first touch the middle of an edge.

The right step: The foot of the perpendicular from the origin to the edge is , which lies between and , so the minimum is .

“Look at the shape of the level curves before deciding where to search”

From: On Extremum Problems

Exhibit

“The bigger the intercept, the bigger z”

The wrong step: Maximum of : slide the line upward; the bigger the intercept, the bigger , so take the highest point of the region.
What went wrong? Think first, then open

Diagnosis becomes , whose -intercept is : the bigger the intercept, the smaller . When the coefficient of is negative, every direction is reversed.

The right step: Write down how the intercept is related to before deciding the direction; or, easiest of all, substitute the candidate vertices one by one and compare.

“Whether intercept and z move together depends on the sign of y’s coefficient”

From: On Extremum Problems

Exhibit

Removing an absolute value and keeping only one case

The wrong step: The locus of points equidistant from the -axis and the -axis: , so .
What went wrong? Think first, then open

Diagnosis is equivalent to or . Writing only leaves out the points in the second and fourth quadrants (such as ), which satisfy the condition but are not on the curve: completeness fails.

The right step: The locus is the two lines and , which can be written together as .

“At every step ask, “Is this an equivalence?””

From: How Curves and Equations Correspond

Exhibit

Counting A and B in the locus of a right-angle vertex

The wrong step: With and , the locus of points with has equation .
What went wrong? Think first, then open

Diagnosis When coincides with , the “line ” does not even exist, let alone being perpendicular to anything. and lie on the circle but do not satisfy the condition: purity fails.

The right step: The locus is the circle without the two points , written .

“Where the figure in the condition degenerates, that point does not count”

From: How Curves and Equations Correspond

Exhibit

Substituting the wrong way round in the substitution method

The wrong step: is on the circle and is the midpoint of and . Let ; then , and substituting into the circle’s equation gives .
What went wrong? Think first, then open

Diagnosis We want the locus of , so should be the coordinates of ; and what gets substituted into the circle’s equation must be the point , which is the one satisfying it. The version above pushes ’s coordinates into ’s equation, and the resulting curve has nothing to do with .

The right step: Let and express through it: . is on the circle, so , that is .

“Name the point you want, express the known point, substitute into the known equation”

From: Methods for Locus Problems

Exhibit

Forgetting the excluded points after eliminating the parameter in the intersecting-loci method

The wrong step: The line through with slope meets the line through with slope at . Multiplying the two equations gives , so the locus is the whole hyperbola .
What went wrong? Think first, then open

Diagnosis (otherwise makes no sense), and the vertices of the hyperbola could only come from or an “infinite” slope. At the two lines are parallel and there is no intersection: those are the directions of the asymptotes. The vertices are extra points.

The right step: The locus is without the two points . In the widget, pull to and to and see.

“After eliminating the parameter, go back to 9.1 and check purity”

From: Methods for Locus Problems

Vol. 8 Method lines hall

Exhibit

Substituting, then forgetting the range of the new variable

The wrong step: Find the minimum of . Let . Then , so the minimum is .
What went wrong? Think first, then open

Diagnosis , which equals only for . In that range , and only gets down to .

The right step: Whenever you substitute, write down the range of the new variable as well: . Then , and the minimum is (reached at ).

“New variable, new range”

From: Reduction line

Exhibit

Substituting only once and hoping to find f(x)

The wrong step: From we get , and that is the answer.
What went wrong? Think first, then open

Diagnosis The right-hand side still contains the unknown , so nothing has been solved. There are two unknowns in one equation.

The right step: Replace by to get a second equation. Two equations in two unknowns and : the problem has been reduced to a system of two linear equations.

“As many unknowns, as many equations”

From: Reduction line

Exhibit

The negation is wrong

The wrong step: To prove “at least one of is greater than ”, assume: “none of is greater than ” … that is fine. But to prove “ are all positive”, the assumption is written as “none of is positive”.
What went wrong? Think first, then open

Diagnosis The opposite of “all are” is “not all are”, that is, “at least one is not”. Writing “none is” leaves out the case where some are and some are not, so the contradiction only rules out part of the possibilities.

The right step: Learn the common negations by heart: all are ↔ not all are; at least one ↔ none at all; at most one ↔ at least two; there exists ↔ every one fails. After writing the assumption, ask yourself: is exactly one of the original conclusion and the assumption true?

“The assumption must be the whole opposite of the conclusion”

From: Proof by Contradiction line

Exhibit

A “contradiction” that never uses the assumption

The wrong step: Assume is rational. Since , we have , so is not an integer: contradiction.
What went wrong? Think first, then open

Diagnosis “Not an integer” does not contradict “rational”: is not an integer either. The assumption never appeared in the argument, so it has not ruled anything out.

The right step: Write the assumption as something you can compute with: (in lowest terms), , so … The contradiction has to come out of this equation.

“Actually use the assumption”

From: Proof by Contradiction line

Exhibit

Proving the step but never checking the base

The wrong step: Prove . Suppose . Adding to both sides gives , so holds. Hence it holds for all .
What went wrong? Think first, then open

Diagnosis The inductive step really is fine, but says , which is false. Every domino would knock over the next one, but nobody pushes the first, so not a single one falls.

The right step: Induction needs both halves: the base case ( holds) and the inductive step ().

“Push the first domino first”

From: Mathematical Induction line

Exhibit

Step size 3, but only one base case

The wrong step: Prove that every postage of cents can be made with -cent and -cent stamps: works; if can be made, add one -cent stamp to get . Done.
What went wrong? Think first, then open

Diagnosis Starting from and stepping by only proves . Nothing ever reaches , or the numbers that follow them.

The right step: With step size you need three base cases: , , . Only the three chains together cover every .

“Step by k, check k starting points”

From: Mathematical Induction line

Exhibit

The boxes are built, but sharing a box proves nothing

The wrong step: Take numbers from and prove that one is a multiple of another: build boxes from “consecutive pairs” . Two of the numbers share a box, so one is a multiple of the other.
What went wrong? Think first, then open

Diagnosis and share a box, but neither is a multiple of the other. The number of boxes is right, but the link “same box conclusion” is broken. Consecutive pairs guarantee coprime numbers.

The right step: Build the boxes by odd part: , of them. Two numbers in the same box differ only by some factors of , so one is a multiple of the other.

“First check that two in the same box are the pair you want”

From: The Pigeonhole Principle line

Exhibit

Showing “enough” but not “one fewer fails”

The wrong step: What is the smallest number of numbers that guarantees two coprime ones? numbers go into the boxes , so the answer is .
What went wrong? Think first, then open

Diagnosis The pigeonhole principle only shows that numbers are enough. To call the smallest, you must also show that are not enough.

The right step: Add an example: the even numbers are pairwise not coprime. So are not enough, are, and the answer is .

“Minimum = enough + one fewer is not enough”

From: The Pigeonhole Principle line

Exhibit

Stopping after subtracting the pairwise intersections

The wrong step: Among there are multiples of , or .
What went wrong? Think first, then open

Diagnosis When you add the three sets, the numbers that are multiples of all of () are counted times; when you subtract the three pairwise intersections they are subtracted times. In the end they are not counted at all.

The right step: Add back the intersection of all three sets: . The formula for three sets has seven terms, with signs alternating .

“Added three times and subtracted three times means never counted”

From: Inclusion–Exclusion line

Exhibit

Dividing by the wrong number for an intersection

The wrong step: Among there are numbers that are multiples of both and .
What went wrong? Think first, then open

Diagnosis Being a multiple of both and means being a multiple of their least common multiple , not of their product . The number itself was missed.

The right step: There are . The lcm equals the product only when the two numbers are coprime.

“For an intersection, use the lcm”

From: Inclusion–Exclusion line

Exhibit

Taking the “smallest solution” among positive reals

The wrong step: Suppose has a solution in positive reals. Take the one with the smallest , and then …
What went wrong? Think first, then open

Diagnosis There is no smallest positive real: if is a solution, so is . Besides, this equation really does have solutions in positive reals (for example , ).

The right step: The well-ordering principle holds only for positive integers (or for finite sets). Infinite descent has to run among the positive integer solutions: take the positive integer solution with the smallest .

“First make sure the “smallest” exists”

From: The Extremal Principle line

Exhibit

Most wins, so they must have beaten everyone

The wrong step: In a round robin, take the player with the most wins. Since has the most wins, beat everyone.
What went wrong? Think first, then open

Diagnosis Most wins is not the same as winning every game. In a five-player round robin everyone can have exactly wins and losses, and then nobody is unbeaten. The extreme element is only “the most”; what it guarantees has to be proved.

The right step: What you can prove is this: for every other player, either beat them directly or beat someone who beat them. The proof is by contradiction: if some beat and also beat everyone beat, then has more wins than .

“The extreme element has an extreme property, but do not exaggerate it”

From: The Extremal Principle line

Exhibit

The sum is even, so both must be even

The wrong step: is even, so and are both even. Let , …
What went wrong? Think first, then open

Diagnosis An even only says that and have the same parity. is a sum of two odd numbers. Miss the “both odd” case and you have done only half of the argument.

The right step: Write: is even . Then treat the two cases “both even” and “both odd”.

“Even = even + even = odd + odd”

From: Parity line

Exhibit

The parity matches, so it can be done

The wrong step: The board shows ; each move erases two numbers and writes their difference. The sum is (odd), so the last number can be any odd number, for example .
What went wrong? Think first, then open

Diagnosis A parity invariant only tells you that an even number cannot be reached. It does not promise that every odd number can. In fact each difference written is no bigger than the largest number on the board, so the last number is at most : is out of reach.

The right step: Parity can only rule things out. To say “it can be done”, you must give an actual sequence of moves.

“Parity can only say “impossible””

From: Parity line

Exhibit

Different bases, yet the area ratio is taken as the ratio of heights

The wrong step: The areas of and are in the ratio , so their heights are in the ratio .
What went wrong? Think first, then open

Diagnosis Area base height. The area ratio equals the ratio of heights only when the bases are the same (or equal in length), and equals the ratio of bases only when the heights are the same.

The right step: Before using an area ratio, find “what is shared”: a common side (ratio of heights), a common height (ratio of bases), or a common angle (ratio of the products of the two sides around it).

“First find a common side, a common height or a common angle”

From: The Area Method line

Exhibit

Forgetting the absolute value when finding an area from coordinates

The wrong step: , , , .
What went wrong? Think first, then open

Diagnosis is a signed area: positive when the vertices go counterclockwise, negative when they go clockwise. Here is clockwise.

The right step: Take the absolute value for the area: . The sign of the signed area is useful in itself: it tells you which way the three points turn.

“An area is never negative; a signed area can be”

From: The Area Method line

Exhibit

Counting the middle term twice when pairing

The wrong step: , find : the terms make pairs, each adding up to … but is not a whole number, so pairing fails.
What went wrong? Think first, then open

Diagnosis is odd, so the middle term is paired with itself: . Four pairs plus half a pair is exactly . Pairing has not failed.

The right step: A safer way to write it is “adding in reverse order”: add to the same written backwards. Every position gives , there are positions, so and .

“Add in reverse order, and odd or even no longer matters”

From: Looking at the Whole line

Exhibit

Keeping the wrong ends after telescoping

The wrong step: .
What went wrong? Think first, then open

Diagnosis The two terms in each difference are two apart, so after cancelling, two terms survive at the front, , and two at the back, . Keeping only one at each end loses some.

The right step: . Write out the first three and the last three terms and see with your own eyes what cancels and what is left.

“Terms that cancel k apart leave k terms at each end”

From: Looking at the Whole line

Exhibit

Thinking that x² + y² = (x + y)²

The wrong step: and , so .
What went wrong? Think first, then open

Diagnosis , which has an extra . After switching to the notation you still have to expand honestly.

The right step: . For higher powers use the recurrence .

“New notation, but the expansion must still be complete”

From: Choosing Good Notation line

Exhibit

Reading f⁽ⁿ⁾(x) as (f(x))ⁿ

The wrong step: , so .
What went wrong? Think first, then open

Diagnosis means composed with itself times: , not the square of . Misread the notation and everything after it is wrong.

The right step: .

“A superscript in brackets means iteration, not a power”

From: Choosing Good Notation line

Exhibit

Joining two points on the same side with a straight line

The wrong step: Minimum of : it is the sum of the distances from to and , so the minimum is .
What went wrong? Think first, then open

Diagnosis and are both above the -axis. The segment never meets the -axis, so can never lie on it, and is never reached. “A straight segment is the shortest path” only helps when the broken line can actually be pulled straight.

The right step: First reflect in the -axis to , so that . Then join : it crosses the -axis, and the minimum is .

“Same side? Reflect first. Only points on opposite sides can be joined straight”

From: Combining Numbers and Figures line

Exhibit

Treating a semicircle as a full circle

The wrong step: Squaring both sides of gives : a circle and a line. Whenever they meet there are two roots, so there are two roots for .
What went wrong? Think first, then open

Diagnosis requires , so it is only the upper half of the circle. Squaring invites the lower half in as well, and the extra intersection points are not roots of the original equation.

The right step: Draw the semicircle and the family of lines of slope , and count only the intersections with the semicircle: there are two roots for .

“See exactly which part of the curve you have before counting crossings”

From: Combining Numbers and Figures line

Exhibit

The correspondence is not one-to-one

The wrong step: The non-negative integer solutions of correspond to “ balls in a row, with dividers placed in the gaps”, so there are of them.
What went wrong? Think first, then open

Diagnosis Dividers go only into gaps, so every part gets at least one ball: this matches the positive integer solutions. Solutions with (two dividers side by side, or one at the far left) are never counted. This “correspondence” is not a surjection.

The right step: First set and so on, which turns it into positive integer solutions of : . Or line up the balls and dividers together and choose where the dividers go: .

“Check that a correspondence has no repeats and also no gaps”

From: Correspondence and Pairing line

Exhibit

Reflecting in the wrong line

The wrong step: Count the paths from to that never go above : reflect each “bad” path in , and its end point becomes… still . Reflection gets nowhere.
What went wrong? Think first, then open

Diagnosis Paths may touch , so “bad” means touching . The reflection must be in the line the path touches at its “first foul”, which is .

The right step: Reflect the part of a bad path after it first touches in the line . The end point becomes , so there are bad paths.

“Reflect in the line where the foul happens”

From: Correspondence and Pairing line

Exhibit

The right recurrence with the wrong initial values

The wrong step: The number of subsets of with no two consecutive numbers satisfies , and , so .
What went wrong? Think first, then open

Diagnosis The empty set also qualifies. should be ( and ), and (). Once an initial value is wrong, every later term is wrong too: the correct value is .

The right step: Always count the initial values by hand from the definition, and say clearly whether the empty set counts. A recurrence that uses the two previous terms needs two initial values.

“Initial values are the foundation of a recursion; count them by hand”

From: Recursion line

Exhibit

A case left out of the split

The wrong step: Derangements: goes to some position ( choices); then has to go to position , and the remaining numbers are deranged, so .
What went wrong? Think first, then open

Diagnosis does not have to go to position . “ and swap” is only one case. In the other case does not go to position ; treat position as 's “forbidden” place and you get a derangement of numbers, contributing .

The right step: When splitting by the last element, check that the cases do not overlap and miss nothing, so together they make up everything. .

“The cases must add up to the whole”

From: Recursion line

Exhibit

The color counts are equal, so it must be possible

The wrong step: Color a 6×6 board black and white by : cells of each. Each 1×4 strip covers two black and two white cells, and strips cover exactly black and white, so it can certainly be tiled.
What went wrong? Think first, then open

Diagnosis A coloring can only prove “impossible”. If the color counts do not match, it certainly cannot be done; if they match, it only means that this one coloring raises no objection. Switch to the four-color diagonals : on a 6×6 board the four colors have cells, so it cannot be tiled.

The right step: To say “possible”, show the tiling. To say “impossible”, find a coloring that really gets stuck.

“Coloring only ever says no; “possible” needs a construction”

From: Coloring line

Exhibit

A coloring that does not suit the tile

The wrong step: Analyze 1×3 strips with a checkerboard coloring: there is one more black cell than white left, which shows no contradiction, so it can probably be tiled.
What went wrong? Think first, then open

Diagnosis A 1×3 strip may cover two black and one white, or one black and two white. The numbers of black and white cells it covers are not fixed, so this coloring puts no constraint on it.

The right step: The first rule of coloring: every tile (every move) must cover the same number of cells of each color. For 1× strips, use the -color diagonals .

“Look at the shape of the tile first, then decide how to color”

From: Coloring line

Exhibit

Substituting only x = 1

The wrong step: . Setting gives , so the sum of the even-numbered terms, , is about half of that.
What went wrong? Think first, then open

Diagnosis only gives the sum of all the coefficients; it cannot separate the odd and even terms. To “cancel” the odd terms you need a value that changes the sign of odd powers and leaves even powers alone: .

The right step: Set : . Set : . Add and divide by : .

“To separate some terms, choose a value that changes their sign”

From: Assigning Values line

Exhibit

Plugging in values gives only a necessary condition

The wrong step: : setting gives , so , and the answer is this family of functions.
What went wrong? Think first, then open

Diagnosis Plugging in values only shows “if there is a solution, it must look like this”. Here has not been found yet, and you must substitute back into the original equation before you can say it really is a solution.

The right step: First set : . If we get , a contradiction, so and . Substitute back: . It holds.

“Plugging in finds candidates; substituting back makes them answers”

From: Assigning Values line

Exhibit

Each object is counted twice, but you forget to divide

The wrong step: people each shook hands with exactly others, so there were handshakes in all.
What went wrong? Think first, then open

Diagnosis Counting person by person, each handshake is counted once by each of its two people. Here the “incidence table” is people × handshakes, and each column (one handshake) has two marks.

The right step: , so there were handshakes. And by the way: if people each shook hands times, would be odd, which is impossible. That is the handshake lemma.

“First ask how many marks each column has”

From: Counting in Two Ways line

Exhibit

Taking the bound from double counting as the answer

The wrong step: A family of -element subsets of , any two sharing at most one element. Count “(pair of elements, set)”: , so is at most .
What went wrong? Think first, then open

Diagnosis is only an upper bound. For , count once more by elements: the sets containing a given element meet pairwise only in that element, so there are at most of them. Then , so , which is sharper.

The right step: Once you have an upper bound, either give a construction that reaches it, or count in another way to improve the bound.

“A bound becomes an answer only with a construction”

From: Counting in Two Ways line

Exhibit

An adjustment with no end

The wrong step: To prove that positive numbers with a fixed sum have the largest product when they are all equal: each time, replace the largest and the smallest by their average. The product grows, and if you keep going they all become equal.
What went wrong? Think first, then open

Diagnosis With real numbers, “replace by the average” may never finish. For example becomes in one step, but becomes , then again… and every step leaves new unequal numbers. Whether they are equal after infinitely many steps needs a limit argument; you cannot just claim it.

The right step: Instead, make one number exactly equal to the average at each step: replace the largest and the smallest by and . Each step adds at least one more number equal to , so it ends after at most steps.

“An adjustment must be sure to end after finitely many steps”

From: Step-by-Step Adjustment line

Exhibit

Pushing integers toward “all equal” too

The wrong step: Write as a sum of positive integers. The sum of squares is smallest when all three equal , so the minimum is .
What went wrong? Think first, then open

Diagnosis The variables must be integers, so is not allowed. The end point of the adjustment is “any two numbers differ by at most 1”, not “all equal”.

The right step: Adjust to : any two differ by at most , and moving them closer only swaps them. The minimum is .

“The constraints decide the end point; for integers it is “differ by at most 1””

From: Step-by-Step Adjustment line

Exhibit

A constructed example that was never checked

The wrong step: Choose numbers from , none dividing another: take all the odd numbers , since odd numbers do not divide each other.
What went wrong? Think first, then open

Diagnosis divides everything, divides … Odd numbers can certainly divide each other. A constructed example must be checked against every condition; “looks right” does not count.

The right step: Take : the ratio of any two of them is greater than and less than , so neither can divide the other.

“Check every condition of your example”

From: Construction line

Exhibit

The factorial construction is off by one

The wrong step: To get consecutive composite numbers, take .
What went wrong? Think first, then open

Diagnosis divides only for . Since , there is no guarantee that is composite.

The right step: Take : and , so all ten numbers are composite.

“Every term of a construction needs a reason”

From: Construction line

Exhibit

The invariant does not object, so it must be possible

The wrong step: The board shows ; each time, erase and write . The parity of the sum does not change (the sum is , even), so the last number can be any even number, such as or .
What went wrong? Think first, then open

Diagnosis An invariant gives only a necessary condition. There are other restrictions: the largest number never increases, so the last number is at most , and is out of the question. Whether can be reached has to be shown by actually doing it (, , ).

The right step: Use invariants (and monovariants) to rule out impossible end points, then give a sequence of moves for each remaining candidate.

“The invariant says “no”; only a construction says “yes””

From: Invariants and Monovariants line

Exhibit

Taking a quantity that changes for an invariant

The wrong step: Erase and write : the total does not change, so starting from , the number left at the end is .
What went wrong? Think first, then open

Diagnosis becomes , so the total drops by : the total is a decreasing monovariant, not an invariant. What stays the same is its parity, because the drop is even.

The right step: Check first: by how much does the quantity change in one move? Only if the change is always is it an invariant; if the change is always even, the parity is invariant.

“First compute the “change” caused by one move”

From: Invariants and Monovariants line

Exhibit

Counting each handshake only once

The wrong step: Can each of people shake hands with exactly of the others? The number of handshakes is , a positive integer, so yes.
What went wrong? Think first, then open

Diagnosis Each handshake involves two people and is counted once by each of them: is the sum of the degrees, which must equal twice the number of handshakes. is odd, so it cannot be twice an integer.

The right step: Draw it as a graph: the sum of the degrees the number of edges, which is always even. is odd, so it cannot be done.

“The sum of the degrees is twice the number of edges”

From: Graph-Theoretic Methods line

Exhibit

Thinking that 5 people cannot avoid it either

The wrong step: Among any people there must be who all know each other or all do not know each other, by the same argument as for people.
What went wrong? Think first, then open

Diagnosis The proof for people uses “one person has edges, so of them have the same color”. With people, each has only edges, which can be red and blue. And there really is a counterexample: color the sides of a pentagon red and the edges of the pentagram blue; there is no monochromatic triangle.

The right step: is the smallest: has a coloring that avoids it, does not. See it for yourself on the on this page.

“Every number in a proof is there for a reason”

From: Graph-Theoretic Methods line