Beyond: The binomial theorem and proofs by counting¶
Counting and combinations · step 5 of 5
A question to start¶
Work out the powers of 11:
Look at the digits: 1 1, then 1 2 1, then 1 3 3 1, then 1 4 6 4 1. They are the rows of Pascal's triangle! Why should multiplying by 11 have anything to do with choosing groups? (And what goes wrong at \(11^5 = 161051\)?) The answer is one of the most useful results in algebra.
The idea¶
Multiplying out brackets by choosing¶
A binomial is an expression with two terms, like \(a + b\). Let us multiply out \((a + b)^3\) in a careful way:
To multiply out, you pick one term from each bracket, multiply your three picks, and do this for every possible way of picking; then add everything up. Each way of picking is a string of three letters, like \(a\,b\,a\), giving \(a^2 b\).
- To get \(a^3\) you must pick \(a\) from all three brackets: 1 way.
- To get \(a^2 b\) you pick \(b\) from one bracket and \(a\) from the other two. Which bracket gives the \(b\)? There are \(\dbinom{3}{1} = 3\) choices: \(baa\), \(aba\), \(aab\).
- To get \(a b^2\), choose which two brackets give \(b\): \(\dbinom{3}{2} = 3\) ways.
- To get \(b^3\): \(\dbinom{3}{3} = 1\) way.
So
The numbers in front, called coefficients, are 1, 3, 3, 1: row 3 of Pascal's triangle. That is no coincidence. The coefficient of each term counts how many ways to choose which brackets supply the \(b\)s, and that is a "choose" number.
The binomial theorem¶
The same argument works for any whole number power \(n\). The term with \(k\) copies of \(b\) (and so \(n - k\) copies of \(a\)) appears once for every way of choosing \(k\) brackets out of \(n\):
The \(\Sigma\) means "add up these terms as \(k\) runs from 0 to \(n\)" (the Beyond step of Patterns introduces this notation).
The powers of 11. \(11^n = (10 + 1)^n\). Put \(a = 10\) and \(b = 1\): each coefficient of Pascal's row gets multiplied by a power of 10, so it lands in its own column, like digits:
In row 5 the numbers 10 and 10 are too big to be single digits, so they carry: \(1, 5, 10, 10, 5, 1\) becomes 161051.
Things that fall out of it¶
Choose values for \(a\) and \(b\) and the theorem gives facts about Pascal's triangle for free.
Row sums. Put \(a = b = 1\):
Every row adds up to a power of 2. There is a counting reason too: a set of \(n\) things has \(2^n\) subsets (groups you can make from it, including the empty group and the whole set), because each thing is either in or out; and sorting those subsets by their size gives the row of choose numbers.
Alternating sums. Put \(a = 1\), \(b = -1\). The left side is \(0^n = 0\) (for \(n \ge 1\)):
For row 4: \(1 - 4 + 6 - 4 + 1 = 0\). ✓
Quick approximations. When \(x\) is small, the later terms of \((1 + x)^n\) are tiny, so
For example, \(1.01^{10} \approx 1 + 10(0.01) + 45(0.0001) = 1.1045\). The true value is \(1.10462\ldots\) This is how people worked out compound interest before calculators.
Proof by counting¶
A combinatorial proof shows two expressions are equal by explaining that both count the same thing. Pascal's rule from the Master step was one: both sides counted the groups of \(r\) chosen from \(n\). Here is another classic.
Stars and bars. In how many ways can 10 identical sweets be shared among 3 children, if some children may get none? Lay the 10 sweets in a row and add 2 dividers, which split the row into 3 parts (child 1, child 2, child 3):
This one means 3, 5, 2. Every sharing gives exactly one row of 10 stars and 2 bars, and every such row gives one sharing. A row has 12 positions, and you only have to choose which 2 of them hold the bars:
In general, \(n\) identical things shared among \(k\) people can be done in \(\dbinom{n + k - 1}{k - 1}\) ways.
Worked examples¶
1. Expand \((x + 2)^4\).
Row 4 of Pascal's triangle is 1, 4, 6, 4, 1. With \(a = x\) and \(b = 2\):
Check with \(x = 1\): the left side is \(3^4 = 81\), and the right side is \(1 + 8 + 24 + 32 + 16 = 81\). ✓
2. One coefficient only. Find the coefficient of \(x^3\) in \((2x - 1)^7\).
- The term with \(x^3\) uses \(2x\) from 3 brackets and \(-1\) from the other 4.
- It is \(\dbinom{7}{3} (2x)^3 (-1)^4 = 35 \times 8x^3 \times 1 = 280x^3\).
The coefficient is 280.
3. A term with no \(x\). Find the term that does not contain \(x\) in \(\left(x + \dfrac{2}{x}\right)^6\).
- A general term is \(\dbinom{6}{k} x^{6-k} \left(\dfrac{2}{x}\right)^{k} = \dbinom{6}{k} 2^k x^{6 - 2k}\).
- No \(x\) means \(6 - 2k = 0\), so \(k = 3\).
- The term is \(\dbinom{6}{3} \times 2^3 = 20 \times 8 = 160\).
Try it¶
- Expand \((x + 1)^4\).
- Expand \((2a - b)^3\).
- Find the coefficient of \(x^2\) in \((1 + 3x)^6\).
- Use the approximation to estimate \(0.99^{10}\), writing it as \((1 - 0.01)^{10}\).
- How many subsets does a set of 10 things have? How many of them have exactly 4 things?
- In how many ways can 7 identical sweets be shared among 3 children, if some may get none?
Answers
- Row 4 is 1, 4, 6, 4, 1, so \((x + 1)^4 = x^4 + 4x^3 + 6x^2 + 4x + 1\).
- Row 3 is 1, 3, 3, 1, with \(a \to 2a\) and \(b \to -b\): \((2a)^3 + 3(2a)^2(-b) + 3(2a)(-b)^2 + (-b)^3\), which is \(8a^3 - 12a^2 b + 6a b^2 - b^3\).
- \(\dbinom{6}{2} (3x)^2 = 15 \times 9x^2\), so the coefficient is 135.
- \(1 + 10(-0.01) + 45(0.0001) = 1 - 0.1 + 0.0045 = 0.9045\). (The true value is \(0.90438\ldots\))
- \(2^{10} = 1024\) subsets, of which \(\dbinom{10}{4} = 210\) have exactly 4 things.
- 7 stars and 2 bars: 9 positions, choose 2 for the bars: \(\dbinom{9}{2} = 36\).
Common mistake¶
Forgetting to raise the whole term to the power. In \((2x - 1)^7\) the \(b\) is \(-1\) and the \(a\) is \(2x\), so a term contains \((2x)^3 = 8x^3\), not \(2x^3\), and \((-1)^4 = 1\), with the sign decided by the power. Put brackets round each term before you raise it to a power, and the slip disappears.
Where it leads¶
- Back a step: Master: combinations and Pascal's triangle
- The binomial theorem is the engine behind the binomial distribution in Probability: the chance of exactly \(k\) heads in \(n\) tosses is \(\dbinom{n}{k} \big/ 2^n\).
- Proving identities for every \(n\), by counting or by induction, is the theme of MAT137 Unit 1 (logic, sets and proofs). When the power is not a whole number, \((1 + x)^n\) becomes an endless sum, a power series: see MAT137 Unit 14.
- The other topics: all topics.