Hints on How to Do Math

1. Be tidy and systematic. Whenever you have data in tables, especially when place-value is important, be sure to keep things lined up in neat columns and neat rows. See section 4 for an example of how this works in practice.

2. If your columns are wobbly, get a pad of graph paper and see if that helps. If you don’t have a supply of store-bought graph paper, you can make graph paper on your computer printer. There are freeware programs that do a very nice job of this.

If that’s too much bother, start with a plain piece of paper and sketch in faint guidelines when necessary.

3. Paper is cheap. If you find that you are running out of room on this sheet of paper, get another sheet of paper, rather than squeezing the calculation into a smaller space. See section 4 for an example of using extra paper to achieve a better result.

4. The whole calculation should be structured as a succession of true statements. The first statement is true, the next statement is true, and the statement after that, et cetera. Each statement is a consequence of the previous statements (in conjunction with known theorems, and the “givens” of the problem). Finally, we get to the bottom line, and we know it is true. An example of this can be seen in reference 1.

5. Sometimes, especially for long and/or complex calculations, it helps to organize your calculation in two columns. An example of this can be seen in reference 1. In the left column, you write an equation. In the right column, make a note as to how you derived that equation. (Numbering all your equations makes this easier.) If there isn’t enough width to do this easily, turn the paper 90 degrees, so it becomes wider and less tall. (Writing paper isn’t suitable for this, so use graph paper ... or lacking that, plain white paper.)

6. Avoid writing down an un-named number like 17, or an un-named expression like a+17 ... because you might forget the meaning thereof. Instead, whenever possible, write down an equation such as b=a+17. That way you can point to every item on the page and say that’s true, that’s true, that’s true... in accordance with the strategy described in item 4.

7. If the meaning of a variable-name such as a and b is not obvious, construct a dictionary, glossary, or legend (like the legend of a map) that explains in a sentence or two the meaning of each variable. That is, a name is not the same as an explanation. Do not expect the structure of a name or symbol to tell you everything you need to know. Most of what you need to know belongs in the legend. The name or symbol should allow you to look up the explanation in the legend. For more on this, see the discussion of names in reference 2.

Suppose the name is constructed according to a pattern, such as F_rT where F stands for force, the subscript r stands for rolling resistance, and T stands for truck. In the glossary, explain what each element means, so the reader doesn’t have to guess. Does T stand for truck? Or does it stand for trailer? Or?????

8. It is important to be able to go back and check the correctness of the calculation you have done. See section 4 and/or reference 1 for examples of what this means in practice.

9. As a corollary of item 8: Don’t perform surgery on your equations. That is, once you write down a correct equation, don’t start crossing out terms (or, worse, erasing terms) and replacing them with other expressions. Such a substitution may be “mathematically” correct, if the replacement is equal to the thing being replaced ... but it is a bad strategy, because it makes it hard for you to check your work. Instead, write a new equation. Leave the old equation as it is. Paper is cheap. An example of this can be seen in reference 1.

10. Keep track of the units for each expression. For example, the statement “x=2.5 inches” means something rather different from “x=2.5 meters” ... and if you shorthand it as “x=2.5” you’re just asking for trouble.

Sometimes the penalty for getting this wrong is 328 million dollars, as in the case of the Mars Climate Orbiter (reference 3).

Figure 1: Mars Climate Orbiter mission logo

Most computer languages do not automatically keep track of the units, so you will have to do it by hand, in the comments. If the calculation is nicely structured, it may suffice to have a legend somewhere, spelling out the units for each of the variables. If variables are re-used and/or converted from one set of units to another, you need more than just a legend; you will need comments (possibly quite a lot of comments) to indicate what units are being used at each point in the code.

One policy that is sometimes helpful (but sometimes risky) is to convert everything to SI units as soon as it is read in, even in fields where SI units are not customary. Then you can do the calculation in SI units and convert back to conventional units (if necessary) immediately before writing out the results. (This is problematic when writing an “intermediate” file. Should it be SI or customary? How do you know the difference between an “intermediate” result and a final result?)

It is certainly possible for computer programs to keep track of the units automatically. A nice example is reference 4. It is a shame that such features are not more widely available

11. The factor-label method is a convenient and powerful way of converting units when necessary. The correctness of this method is a direct consequence of the axioms of algebra, since it starts by multiplying by unity, which is allowed by the axioms.

12. Haste makes waste, especially with multi-step processes. If you work methodically, you’ll get the right answer the first time, and that’s all there is to the story. If you try to do it twice as fast, you’ll get the wrong answer, and then you’ll have to do it over again ... and again ... and again.

2 Multiple Methods of Solution

Here’s a classic example: The task is to add 198 plus 215. The easiest way to solve this problem in your head is to rearrange it as (215 + (200 − 2)) which is 415 − 2 which is 413. The small point is that by rearranging it, a lot of carrying can be avoided.

One of the larger points is that it is important to have multiple methods of solution. This and about ten other important points are discussed in reference 5.

3 Drawing Inequalities

The classic "textbook" diagram of an inequality uses shading to distinguish one half-plane from the other. This is nice and attractive, and is particularly powerful when diagramming the relationship between two or more inequalities, as shown in figure 2.

Figure 2: Inequalities Represented using Shading

However, when you are working with pencil and paper, shading a region is somewhere between horribly laborious and impossible.

It is much more practical to use hatching instead, as shown in figure 3.

Figure 3: Inequalities Represented using Hatching

Obviously the hatched depiction is not as beautiful as the shaded depiction, but it is good enough. It is vastly preferable on cost/benefit grounds, for most purposes.

Some refinements:

I recommend hatching the excluded half-plane, so that the solution-set remains unhatched rather than hatched. This is particularly helpful when constructing the conjunction (logical AND) of multiple inequalities.
I recommend using solid lines versus dashed lines to distinguish “≥” inequalities from “>” inequalities. If you’re going to hatch the excluded half-plane, use solid lines for the “>” inequalities, to show that the boundary itself is excluded.
As suggested in figure 4, for linear inequalities you can do the hatching faster and more accurately with the help of a ruler or straightedge, which makes it easy to ensure that none of the hatches stray into the wrong region.

Figure 4: How To Draw Inequalities

4 Long Multiplication

See reference 6.

5 Long Division

See reference 7.

6 Hindu Numeral System

The modern numeral system is based on place value. As we understand it today, each numeral can be considered a polynomial in powers of b, where b is the base of the numeral system. For decimal numerals, b=10. As an example:

2468		=		2 b³	+	4 b²	+	6 b¹	+	8 b⁰
		=		2000	+	400	+	60	+	8

(1)

As discussed in reference 6, this allows us to understand long multiplication in terms of the more general rule for multiplying polynomials.

7 Multiplying Multi-Term Expressions

Given two expressions such as (a+b+c) and (x+y), each of which has one or more terms, the systematic way to multiply the expressions is to make a table, where the rows correspond to terms in the first expression, and the rows correspond to terms in the second expression:

a |

b |

c |

(a+b+c) · (x+y)

ax + ay + bx + by + cx + cy

(2)

In the special case of multiplying a two-term expression by another two-term expression, the mnemonic FOIL applies. That stands for First, Outer, Inner, Last. As shown in figure 5, we start with the First contribution, i.e. we multiply the first term from in each of the factors. Then we add in the Outer contribution, i.e. the first term from the first factor times the last term from the last factor. Then we add in the Inner contribution, i.e. the last term from the first factor times the first term from the last factor. Finally we add in the Last contribution, i.e. we multiply the last terms from each of the factors.

Figure 5: FOIL = First + Outer + Inner + Last

8 Square Roots

8.1 Newton’s Method

If you can do long division, you can do square roots.

Most square roots are irrational, so they cannot be represented exactly in the decimal system. (Decimal numerals are, after all, rational numbers.) So the name of the game is to find a decimal representation that is a sufficiently-accurate approximation.

We start with the following idea: For any nonzero x we know that x÷√x is equal to √x. Furthermore, if s₁ is greater than √x it means x/s₁ is less than √x. If we take the average of these two things, s₁ and x/s₁, the average is very much closer to √x. So we set

s₂

s₁ + x/s₁

(3)

and then iterate. The method is very powerful; the number of digits of precision doubles each time. It suffices to use a rough estimate for the starting point, s₁. In particular, if you are seeking the square root of an 8-digit number, choose some 4-digit number as the s₁-value.

This is a special case of a more general technique called Newton’s method, but if that doesn’t mean anything to you, don’t worry about it.

8.2 First-Order Expansion

Note that the square of 1.01 is very nearly 1.02. Similarly, the square of 1.02 is very nearly 1.04. Turning this around, we find the general rule that if x gets bigger by two percent, then √x gets bigger by one percent ... to a good approximation.

We can illustrate this idea by finding the square root of 50. Since 50 is 2% bigger than 49, the square root of 50 is 1% bigger than 7 ... namely 7.07. This is a reasonably good result, accurate to better than 0.02%.

If we double this result, we get 14.14, which is the square root of 200. That is hardly surprising, since we remember that the square root of 2 is 1.414, accurate to within roundoff error.

9 Simple Trig Functions

Sine and cosine are transcendental functions. Evaluating them will never be super-easy, but it can be done, with reasonably decent accuracy, with relatively little effort, without a calculator.

In particular:

You can always start with a zeroth-order approximation: For angles near zero, the sine will be near zero. For angles near 90^∘, the sine will be near 1. For angles near 30^∘, the sine will be near 0.5, et cetera.
You can draw a graph, using the anchor points in equation 4 as a guide. You can then use graphical interpolation to obtain values for any angle.
A simple Taylor series gives a result accurate to 2.1% or better using only a couple of multiplications. Remember to express the angle in radians before using these formulas.

The following facts serve to “anchor” our knowledge of the sine and cosine:

sin(0^∘)

√

0/4

cos(90^∘)

(4a)

sin(30^∘)

√

1/4

0.5

cos(60^∘)

(4b)

sin(45^∘)

√

2/4

0.70711

cos(45^∘)

(4c)

sin(60^∘)

√

3/4

0.86603

cos(30^∘)

(4d)

sin(90^∘)

√

4/4

cos(0^∘)

(4e)

Actually, that hardly counts as “remembering” because if you ever forget any part of equation 4 you should be able to regenerate it from scratch. The 0^∘ and 90^∘ values are trivial. The 30^∘ is a simple geometric construction. Then the 60^∘ and 45^∘ values are obtained via the Pythagorean theorem. The value for 45^∘ should be particularly easy to remember, since √2 = 1.414 and √½ = ½√2.

The rest of this section is devoted to the Taylor series. A low-order expansion works well if the point of interest is not too far from the nearest anchor.

For angles between −10^∘ and +10^∘, the approximation sin(x)≈x is accurate to better than 0.51%. This is a one-term Taylor series. Let’s call it the Taylor[1] approximation. It is super-easy to evaluate, since it involves no additions and no multiplications, or at most one multiplication if we need to convert to radians from degrees or some other unit of measure. See equation 5c and the blue line near x=0 in figure 6.
For angles from 25^∘ to 65^∘, all the points are with a few degrees of one of the anchor points in equation 4. This means the first-order Taylor series is accurate to better than 1 percent in this region. We can call this the Taylor[0,1] approximation. It requires knowing the sine and the cosine at the nearest anchor point ... which we do in fact know from equation 4. See equation 9 and the blue line in figure 6.

Figure 6: Piecewise 1st-Order Taylor Approximation to Sine
For a rather broad range of angles near the top of the sine, from 65^∘ to 115^∘, the approximation sin(x)≈1−x²/2 is accurate to better than half a percent. This is a second-order Taylor series with only two terms, because the linear term is zero. Let’s call this the Taylor[0,2] approximation. See equation 5e and the green line near x=90^∘ in figure 6.
This leaves us with a region from 10^∘ to 25^∘ that requires some special attention. Options include the following:
For most purposes, the best option is to use the Taylor[1,3] approximation anchored at zero. This requires a couple more multiplications, but the result is accurate to better than 0.07%.
If you really want to minimize the number of multiplications, we can start by noting that the Taylor[1] extrapolation coming up from zero is better than the Taylor[0,1] extrapolation coming down from 30^∘, so rather than using the closest anchor we use the 0^∘ anchor all the way up to 20^∘ and use the 30^∘ anchor above that. This has the advantage of minimizing the number of multiplications. Disadvantages include having to remember an obscure fact, namely the need to put the crossover at 20^∘ rather than halfway between the two anchors. The accuracy is better than 2.1%, which is not great, but good enough for some applications. The error is shown in figure 7.

Figure 7: Taylor Approximation Error
There are many other options, but all the options I know of involve either more work or less accuracy.
The spreadsheet that produces these figures is given by reference 8.

Here are some additional facts that are needed in order to carry out the calculations discussed here.


1^∘	=	17.4532925 milliradian		(5a)

1 radian	=	57.2957795^∘		(5b)

sin(x)	≈	x	for small x, measured in radians	(5c)

cos(x)	=	sin(π/2 − x)	for all x	(5d)

	≈	1 − x²/2	for small x, measured in radians	(5e)

Last but not least, we have the Pythagorean trig identity:

sin² + cos²

(6)

and the sum-of-angles formula:

sin(a + b)

sin(a)cos(b) + cos(a)sin(b)

(7)

If you can maintain even a vague memory of the form of equation 7, you can easily reconstruct the exact details. Use the fact that it has to be symmetric under exchange of a and b (since addition is commutative on the LHS). Also it has to behave correctly when b=0 and when b=π/2.

If we assume b is small and use the small-angle approximations from equation 5, then equation 7 reduces to the second-order Taylor series approximation to sin(a+b).

sin(a + b)

≈

sin(a) + cos(a) b − sin(a) b²/2

for small b, measured in radians

(8)

If we drop the second-order term, we are left with the first-order series, suitable for even smaller values of b:

sin(a + b)

≈

sin(a) + cos(a) b

for smaller b, measured in radians

(9)

You can use the Taylor series to interpolate between the values given in equation 4. Since every angle in the first quadrant is at least somewhat near one of these values, you can find the sine of any angle, to a good approximation, as shown in figure 6.

10 References

: 1.
John Denker,
“Balancing Reaction Equations using Gaussian Elimination”
www.av8n.com/math/gauss-elim.htm
: 2.
John Denker,
“Suggestions for Writing Good Software”
www.av8n.com/computer/htm/good-software.htm
: 3.
Douglas Isbell, Mary Hardin, Joan Underwood
“MARS CLIMATE ORBITER TEAM FINDS LIKELY CAUSE OF LOSS”
http://mars.jpl.nasa.gov/msp98/news/mco990930.html
: 4.
Google Calculator. Example: furlongs per fortnight, converted to SI units:
http://www.google.com/search?q=1+furlong+per+fortnight+in+m/s
: 5.
John Denker,
“Learning, Remembering, and Thinking”
www.av8n.com/math/thinking.htm
: 6.
John Denker,
“Long Multiplication
www.av8n.com/math/long-multiplication.htm
: 7.
John Denker,
“Long Division
www.av8n.com/math/long-division.htm
: 8.
John Denker,
“spreadsheet for checking trig approximations
./trig-approx.xls

[Contents]