Pólya's Four Steps
Understand: restate the problem in your own words; identify the unknown, the data, the condition; draw a figure; try small cases. Plan: connect to a known problem; specialize, generalize, or work backwards; introduce notation. Execute: carry it out, checking each step. Look back: check the result, find another route, ask what the method can also solve.
The first step is where beginners spend too little time and experts spend the most. Computing the first five cases of a problem about all is not a waste — it is how the pattern, and usually the proof, is discovered. Working backwards from the goal ("what would I need to be true one line before the end?") is the second most useful habit; it turns a blank page into a specific sub-goal.
Fence posts are placed every m along a straight m fence, including both ends. How many posts? The reflex answer counts the gaps. A drawing shows posts at : six posts for five gaps. Understanding the problem — here, literally drawing it — is the step that catches the error.
How many regions do lines in general position (no two parallel, no three through a point) cut the plane into? : . Differences : the th line crosses the earlier ones at points, which cut it into pieces, and each piece splits one region in two. So , and , proved by induction with exactly that step.
Ask: "what would make this easy?" If the numbers were smaller, if the shape were symmetric, if one variable were zero — then solve THAT and see whether the solution survives being un-wished.
Jumping from the small cases to a formula without the reason. The data for lines also fit , which is wrong from on. The pattern becomes trustworthy only when you can say why each new line adds what it adds.
You must reach exactly starting from using only "add " and "double". What is the fewest moves? (Hint: go backwards from .)
Show the answerHide the answer
Backwards, halve when even, subtract when odd: : moves. Halving whenever possible is optimal because, run forwards, doubling is the only move that gains more than , and delaying a doubling never helps.
Invariants and Parity
When a process or game is described by moves, look for an invariant: a quantity (often a parity, a sum mod , or a coloring count) that every move preserves. If the start and the goal have different values of the invariant, the goal is unreachable — a proof of impossibility with no case analysis at all.
The power of an invariant is that it replaces "try every sequence of moves" — usually infinitely many — by one check per kind of move. You never follow the process; you only verify that a single move cannot change the quantity. Finding the invariant is the creative part: compute a few candidate quantities (sum, parity, sum of squares, a count mod ) before and after one move, and keep the one that does not change.
Seven cups, all face up; a move flips exactly two. Can all end face down? Each move changes the number of face-up cups by , , or : the parity of face-up cups never changes. It starts odd () and the goal is , even. Impossible.
Remove two opposite corners of an board. Can the rest be tiled by dominoes? Each domino covers one dark and one light square. Opposite corners have the same color, so of that color and of the other remain, and dominoes would need of each. No.
Three numbers are on a board. A move replaces two of them, and , by and . Can the board ever read ? The sum of squares is invariant: , since the terms cancel. It starts at , while . Impossible.
Finding a quantity that changes "usually" but not always, or proving a state is unreachable when the problem asked whether it is reachable (an invariant proves impossibility only; possibility needs a construction).
The numbers are on a board. A move erases two numbers and writes . What single number remains at the end?
Show the answerHide the answer
The SUM is invariant: erasing and and writing leaves it unchanged. It starts at , so the last number is .
Coloring Arguments
To show a region cannot be tiled by certain pieces, color it so that every piece, wherever it is placed, covers a fixed pattern of colors. Then compare what the pieces must cover with what the region has.
A coloring is an invariant in disguise: the "move" is placing one more piece, and the quantity preserved is a relation between the numbers of cells of each color covered. The checkerboard handles pieces that cover equally many dark and light cells; pieces that do not balance, or longer pieces, need a finer coloring, such as coloring cell by for bars.
Color a board like a checkerboard with dark corners: dark and light cells. Remove a light cell next to a corner: dark and light remain, and dominoes would need of each, so no tiling exists. Remove a corner instead: and remain, and the coloring proves nothing — in fact that board can be tiled. A balanced count is not a proof of possibility.
Can a board be tiled by T-shaped pieces of four cells? Checkerboard it: dark and light. Wherever a T is placed, its row of three alternates (two cells of one color around a middle of the other), and the fourth cell, touching the middle, has the color opposite to the middle. So each T covers cells of one color and of the other: an odd number of dark cells. Nine odd numbers add to an odd number, but the board has dark cells. No tiling.
Can a board be tiled by bars? The area, , is a multiple of , and the checkerboard balances ( and ) — neither rules it out. Color cell by . Any bar, horizontal or vertical, covers four consecutive values of , one of each class. So bars cover cells of each class. But the classes have cells. No tiling.
Stopping at the first coloring. The checkerboard balances on the board, and it is tempting to conclude the tiling "should work". A coloring that balances only means that coloring is too weak; try one matched to the shape of the piece.
Show that an board cannot be tiled by T-tetrominoes and one square.
Show the answerHide the answer
Checkerboard: dark cells. The square covers exactly dark cells. Each T covers or dark cells — an odd number — and odd numbers add to an odd number. So the pieces cover (odd) = an odd number of dark cells, never .
Monovariants and Games
A monovariant is a quantity that changes in only one direction with every move. If it is a non-negative integer and strictly decreases at every move, the process must stop. In games, the analogue is a property of positions that the winning player can always restore.
Termination proofs are well-ordering in action: a strictly decreasing sequence of non-negative integers cannot go on forever. The skill is finding the quantity. Count something the moves are designed to fix — pairs out of order, friendships inside a room, distance from the goal — and check that every allowed move reduces it by at least .
A row of numbers is sorted by repeatedly swapping two neighbors that are out of order. Does it always finish? Count the inversions: pairs (not just neighbors) in the wrong order. Swapping an out-of-order neighboring pair fixes that pair and changes no other pair's order, so the count drops by exactly . It cannot drop below , so the process stops — and it takes exactly as many swaps as there were inversions. For the inversions are : exactly swaps.
Prove that the members of any club can be split into two rooms so that everyone has at most half of their friends in their own room. Start with any split. If someone has more than half of their friends in their room, move them. Their friendships inside a room change from (friends in old room) to (friends in new room), a strict decrease; nobody else's friendships inside rooms change. So the total number of friendships inside rooms strictly decreases, and since it is a non-negative integer, the moves stop. When they stop, nobody has more than half of their friends in their own room.
There are counters. Two players alternately take , or ; whoever takes the last counter wins. The first player wins: take , leaving . Whenever the opponent takes , take . After each of your turns the number left is a multiple of — the property you restore — so the opponent never faces on their turn, and you take the last counter.
A decreasing quantity that is not a non-negative integer. "The distance to the target halves every move, so we arrive" is false: a positive real can halve forever without reaching . Termination needs integer values and a floor.
There are counters; players alternately take to , and whoever takes the last one wins. Which player can force a win, and how?
Show the answerHide the answer
The second player. Multiples of are losing for the player about to move: whatever they take ( to ), the other replies with and leaves a multiple of again. is a multiple of , so the second player answers every with and takes the last counter.
Why It Works
Let be a quantity with for every allowed move. By induction on the number of moves, every position reachable from the start has : true after moves, and each move preserves it. So a position with is not reachable. The converse is not part of the argument: positions with may still be unreachable, which is why a matching invariant never proves that a goal can be reached.
The numbers are on a board; a move erases and writes . What can the last number be? Flawed: "the sum is invariant, so the last number is ". It is not invariant — erasing and and writing lowers the sum by . What is invariant is the PARITY of the sum: and differ by , an even number. The sum starts at , odd, so the last number is odd. The invariant does not say which odd number; that would need a further argument.