top of page

Rational approximations are always useful when approximating real numbers.

While 22/7 is an okay sort-of-but-not-so-good approximation of π, 355/113 is significantly better. The relative error of the first is 0.040%, while the relative error of the second is 0.0000085%, comparing the approximation at nine significant digits with 3.14159292 and π rounded to nine digits 3.14159265. Ton contrast, 22/7 = 3.14285714.

The best approximation for the length of a year in terms of days is to:

  1. have each year be 365 days,

  2. unless the year is a multiple of four, in which case it is a leap year with 366 days,

  3. unless it is also a multiple of 128, in which case it is a 365-day year.

This gives the average number of days in each year to be 46751/128 = 365.2421875 days. A tropical year is 365.24219 days, so the error is less than one-quarter of one second per year.

A lunar month is approximation 29.530588 days, and a reasonable approximation is 29 + 26/49, so if 26 out of every 49 months is 30 days, and the remaining 23 are 29 days, this will be a good enough approximation that this will lose only slightly more than two seconds per lunar month. After a century, the synchronization would be off by less than three-quarters of an hour.

This could be implemented by having one 15-month period alternating 30-29-30-...-30-29-30,

followed by two 17-month periods alternating in the same pattern,

totaling 49 months, with 8 + 9 + 9 = 26 30-day months.

Of course, this is only an approximation because there is significant variation throughout the year as to how long a lunar month is, so it may happen that there are four months in a row that have 30 days, and it is also possible to have a lunar month being only 28 days. Over the long term, however, this should be a reasonable approximation.

A really good approximation to an equilateral triangle is to have the base 8 and the height 7.

The hypotenuse is the square root of 65, or 8.0623, so a relative error of less than 1% and the angles are 60.255°, 60.255° and 59.49°.

If you want greater accuracy, use a base of 22 and a height of 19

The hypotenuse is the square root of  482, or 21.9545, so a relative error of 0.2% and the angles are 59.93°, 59.93° and 60.14°.

If you require even greater accuracy, use a base of 52 and a height of 45.

The hypotenuse is the square root of 2701, or 91.97, so a relative error of 0.0555% and the angles are 59.98°, 59.98° and 60.04°

The first approximation is an excellent means of turning square grid paper into a hexagonal grid. Depending on the size of the grid, either of these solutions is quite acceptable as an approximation to a hexagonal grid:

There is no reasonable approximation of e. Really, 2.7182818285 is easy to remember and reasonably accurate. There is one interesting event that occurred in 1828, something that should be of interest to all ECE students: it was the year that Ányos Jedlik created the first electric motor.

Personal pet-peeve. I was a TA for a calculus course, and a student wrote 17¼ or something like that. The instructor required us to take marks off, because obviously this meant 17/4 = 4.25 and not the correct answer of 17.25. Honestly, you don't have to give a student 50% on a question simply because the format is potentially questionable: the student who got the right answer, but formatted the answer in a way the instructor did not even specify is questionable at best. My approach might be to give the student -ε , where we all know that ε is a small value, but never-the-less highlighting that there is an issue here. 17+¼ is a format I've never seen anyone use...

Again, another pet-peeve: a teacher I liked failed students because they came from French immersion and used a comma instead of a period for the decimal point. What a welcome to an English-language science class: you get zero because you wrote 3,14 and not 3.14. Just mark it as correct and ask the student to use a period in the future...

If anyone takes any of these rational approximations seriously, you need a life. They're here more for fun and interest, and not any serious suggestion of any real application...

296/167 is a nice approximation of √π
1632/271 is a nice approximation of the multiplier of the Avogadro constant (6.022140221402214 versus 6.02214076)

53/8 is a nice approximation of the multiplier of the Planck constant (6.625 versus 6.62607015)

58/55 is a nice approximation of the Planck constant divided by 2π (1.054545454545454 versus 1.054571817)

These are most interesting when the rational number uses fewer digits than the number of digits of accuracy.

bottom of page