Is elimination or substitution better?

Neither in general; it depends on the coefficients. If some unknown has a coefficient of 1 or −1, substitution rearranges without fractions and is quicker. If two coefficients of the same unknown are equal or opposite, elimination cancels them with no multiplying and is quicker. With neither, elimination usually wins because multiplying keeps whole numbers where dividing does not.

What does it mean when both variables disappear?

Look at what is left. If it is 0 = 0, the two equations said the same thing and there are infinitely many solutions. If it is 0 = some non-zero number, they contradict each other and there is no solution. Those are the only two possibilities, and they are opposite answers.

When should I use matrices instead?

At three unknowns matrices become competitive and at four they become the only sensible option. They are also the only route that tells you the shape of the solution set when it is not a single point — the rank comparison says whether a failure is inconsistency or a family, and the reduced form hands you the family.

Can a system have exactly two solutions?

No. A linear system has exactly one solution, none, or infinitely many — never two, never seven. Two lines either cross once, never cross, or are the same line. Systems with more than one but finitely many solutions do exist, but they are not linear: a line meeting a circle is two points, and the circle makes the system quadratic.

Does the order of the equations matter?

Not to the answer. Swapping two equations, multiplying one by a non-zero number, or adding a multiple of one to another all leave the solution set exactly as it was — those three operations are the whole of row reduction, and they are chosen precisely because they change nothing that matters.

Elimination vs Substitution vs Matrices

The five-second rule

Look at the coefficients before you commit. If any unknown has a coefficient of 1 or −1 anywhere, substitution is easiest: that equation rearranges to "x = …" with no division and therefore no fractions. If instead two coefficients of the same unknown are equal or opposite — a 3y in one equation and a −3y in the other — elimination is easiest, because they cancel with no multiplying at all. If neither is true, elimination with small whole-number multipliers is usually still quicker than substitution, because multiplying keeps integers where dividing does not.

Matrices earn their place at three unknowns and become the only sensible option at four. They also answer a question the other two cannot: whether a system that fails has no solution or infinitely many, and if the latter, what the family of solutions actually is.

±1 coefficient → substitute · equal or opposite coefficients → eliminate · otherwise → eliminate with whole-number multipliers

The same system, three ways

Take 3x + 2y = 16 and 5x − 4y = 12. By elimination: the y coefficients are 2 and −4, so multiply the first by 2 to get 6x + 4y = 32, then add to get 11x = 44 and x = 4. Substituting back gives y = 2.

By substitution: no coefficient is ±1, so some fraction is unavoidable. Isolating y in the first gives y = 8 − (3/2)x; putting that into the second gives 5x − 4(8 − (3/2)x) = 12, which expands to 11x = 44 and the same x = 4. Same answer, one more fractional line.

By row reduction: the augmented matrix reduces to the identity with 4 and 2 in the constants column. It takes more writing on a 2×2 and less on a 4×4, which is the whole trade.

Worked example

3x + 2y = 16, 5x − 4y = 12 → x = 4, y = 2. Elimination needs one multiplier; substitution needs a fraction; row reduction needs neither but writes more lines.

Where each method actually goes wrong

Elimination fails on signs. You multiply one equation by a negative and forget to apply it to the constant on the right, or you add when the coefficients were equal rather than opposite. Both produce a clean-looking equation that is not equivalent to the one you started with.

Substitution fails on brackets. The isolated expression replaces a single term, so it needs brackets, and a minus in front of those brackets applies to every term inside: 5x − 4(8 − 1.5x) is 5x − 32 + 6x, not 5x − 32 − 6x. That one sign is the single most common lost mark in the method.

Row reduction fails on arithmetic volume. Nothing about it is conceptually hard; there are simply more numbers, and one slip early propagates through every later row. Checking the answer in the original equations catches all three failures at once and takes ten seconds.

No solution, or infinitely many?

Both look the same at first: the unknowns all vanish. What is left decides which you have. If elimination leaves 0 = 0, the equations carried the same information and there are infinitely many solutions. If it leaves 0 = 7, or any other non-zero number, the equations contradict each other and there is no solution at all.

Geometrically these are the only two ways two lines can fail to cross once: they can be the same line, or they can be parallel. There is no third failure. With three unknowns the same logic applies to planes, and with more unknowns the geometry stops being drawable but the algebra does not change.

The general statement is about rank. Row-reduce and count the non-zero rows of the coefficient part and of the whole augmented matrix. Equal, and matching the number of unknowns, gives one solution. Equal but smaller gives a family, with as many free parameters as the shortfall. Different means no solution — the extra pivot sits in the constants column, which is exactly the 0 = 7 row written another way.

"Infinitely many" is not the answer — the family is

When a system has infinitely many solutions, saying so is only half the work. The answer is the family: for x + y = 5, it is x = 5 − t and y = t, where t is anything you like. Every solution to the system is on that list and nothing else is.

The number of free parameters is the number of unknowns minus the rank. One unknown free means a line of solutions; two means a plane. A system with four unknowns and three independent equations has exactly one free parameter, and its solution set is a line in four-dimensional space — which sounds exotic and is written down the same way as any other.

When there are more equations than unknowns

An overdetermined system — three equations, two unknowns — usually has no solution, because each equation is another constraint and there is nothing left to satisfy it with. But not always: the extra equation may repeat information the first two already carried, in which case the solution survives.

This is where rank earns its keep. Three equations in two unknowns with rank 2 in both matrices have exactly one solution, and all three lines pass through the same point. The same three equations with the augmented rank at 3 have none, and the three lines form a triangle instead. Nothing about counting equations tells you which; only reducing them does.

Common mistakes

Multiplying an equation by a number and forgetting the constant on the right. Every term gets multiplied, including the one after the equals sign.
Adding when the coefficients were equal rather than opposite. Equal coefficients cancel on subtraction; opposite ones cancel on addition. Getting it backwards doubles the term instead of removing it.
Dropping the brackets when substituting. The isolated expression replaces one term, so a minus in front of it distributes over everything inside.
Eliminating a different variable in the second pair of a 3×3. Use the same one both times, or you end up with two equations still containing three unknowns.
Checking the answer against a transformed equation rather than the original. An error in the transformation survives that check untouched.
Reporting "no solution" when the system reduces to 0 = 0. That is infinitely many solutions, which is the opposite outcome.

When not to rely only on the calculator

These three methods apply to LINEAR systems — every unknown to the first power, with no products of unknowns. An equation containing xy, x² or 1/x is not linear, and none of this applies to it. The methods also assume exact coefficients: where the numbers come from measurements rather than a textbook, a system can be technically solvable and practically meaningless, because a small change in a coefficient moves the answer a long way. That sensitivity is a real subject in its own right and nothing here addresses it.

Elimination, Substitution, or Matrices — Choosing a Route Through a System