Appendix X

Understanding Logarithms Intuitively

Adam A. Smith

Logarithms make a lot of people anxious. A lot of this has to do with the way
they’re often taught in high school and secondary school: by memorizing all
the proper steps, without imparting much deeper meaning. For example,
maybe you were once taught to solve problems like this:

log7 49 = ?

If you are like most people, you knew how to solve this years ago, when
it was important to pass a math test. You were probably taught how to
rearrange this into an exponential equation using the following equivalence:

logb x = y ⇐⇒ by = x,
and from there you solved for the unknown. But without any solid intuition
for what this all really means, you were probably just going through a series
of memorized steps in order to get the right answer. This isn’t helpful: if
you just memorize without understanding, you’ll forget all about it once it’s
no longer important.∗
The problem is that logarithms are important. They are especially cru-
cial in computer science when one wants to analyze two different algorithms,
figuring out which one is the more efficient. Thus, the purpose of this work is
to give you a more intuitive feel for what a logarithm represents, making you
more comfortable with what they mean. You’ll be able to estimate them in
your head, and hopefully this knowledge will stay with you for a long time.

Scales of Ten
The most important thing to know about logarithms is this:

A logarithm represents the scale of a number.

Think of all the one-digit numbers, 1 through 9. (For now we’re skipping
over 0.) Of course these numbers are all different, but they’re close enough
to each other to be easily comparable. However the two-digit numbers, 10
through 99, are on a totally different scale. They’re easily comparable to each
∗
Maybe it even convinced you that you’re “not a math person”, even though you had
done perfectly fine in math class before. This is a problem, because when people become
convinced that they “can’t do math”, it can limit their career options in the future.
Appendix X: Understanding Logarithms Intuitively 2

other, but most of them are so much bigger than the one-digit numbers, that
it’s hard to think of them in quite the same way. If you were drawing them
next to each other on a graph, the two-digit numbers would dominate, and
you could barely read the one-digit values at all. Next come the three-digit
numbers, 100 through 999. And these are on an even higher scale: it’s easy
to compare them with each other, but harder to compare them with the two-
digit numbers and even harder to compare them with the one-digit numbers.
They’re all on different orders of magnitude.
These scales of size 10 are the base-10 logarithms, which are represented
by “log10 ” in equations. Base-10 logarithms are sometimes called common
logarithms. That small “10” is the base, and it can be any positive number
except 1 (though we’ll soon see that only a few bases are commonly used).
If you have a round number like 1, 10, 100, and so on, the base-10 logarithm
of that number is just the number of zeros it has:

log10 1 = 0 log10 1000 = 3 log10 1,000,000 = 6

log10 10 = 1 log10 10,000 = 4 log10 10,000,000 = 7
log10 100 = 2 log10 100,000 = 5 log10 100,000,000 = 8

For a number that lies between two of these round numbers, the loga-
rithm will be some intermediate decimal value (probably one that goes on
forever without repeating). So all the one-digit numbers greater than or
equal to 1 have a logarithm between 0 and 1. Likewise, the two-digit num-
bers have a log between 1 and 2, and so on. So if you want to estimate the
logarithm of a number, all you have to do is count its digits. For example the
number 83,176,000 has eight digits, and therefore its log must be between 7
and 8. And since it’s a large eight-digit number, the log is closer to 8 than
7. (In fact, the log of this number is approximately 7.92.)
Here’s the graph of positive base-10 logarithms:

Look at how flat it is—it increases very slowly. Nevertheless, it always goes
up: bigger numbers always have bigger logarithms. You can also see that
it gets flatter and flatter as it goes on; it takes bigger and bigger inputs for
each subsequent increase of 1. But it never gets totally flat: if you increase
the input, the output is always at least a little bigger. And unlike other
functions, it doesn’t have an asymptote that it approaches forever but never
reaches. It will reach any value you can think of, eventually.
In the base-10 logarithm, each scale is 10 times bigger than the previous
one. Another way to think of it is “Starting from 1, how many times do I
3 Appendix X: Understanding Logarithms Intuitively

need to multiply by 10 to get to this number?” For example, if we write

83,176,000 in “scientific notation”, we get:

83,176,000 = 8.3176 × 107 .

Writing out all the 10s separately, this becomes:

83,176,000 = 8.3176 × 10 × 10 × 10 × 10 × 10 × 10 × 10.

So how many factors of ×10 are there in this number? It’s easy to see that
there are 7, and then most of an 8th (the 8.3176 is “most of” an 8th ×10).
So the log of about 7.92 makes sense. If the one-digit numbers are “in the
zeroth scale”, and the two-digit numbers are “in the first scale”, then we can
think of 83,176,000 as being “in the seventh scale, but almost to the eighth
scale”.
Remember that log10 100 is 2, and log10 1000 is 3. So which number has
a log of 2.5, halfway between them?

log10 x = 2.5.

To solve this, we need to multiply 1 by a factor of ×10, two and a half times.
But how do you multiply half a time? The answer is to use a square root:
√
x = 10 × 10 × 10
x =∼ 316.23.

(The ∼ means “approximately”.) This might seem surprising, since ∼ 316.23

isn’t halfway between 100 and 1000, in the way you usually think of “half-
way”. But remember, we’re talking about scales and multiplicative factors,
√
so ∼ 316.23 really is halfway between them. When you multiply
√ 100 × 10,
you get ∼ 316.23. And when you multiply that by × 10 again, you get
1000. Multiplicitavely, ∼ 316.23 is halfway between 100 and 1000. That is
why:
log10 ∼ 316.23 = 2.5.
We can tell a lot about numbers just from their logarithms. For example,
let us say that there are two numbers, a and b. And let us say that we don’t
know the numbers themselves, but we do know their base-10 logs:

log10 a = 6.96
log10 b = 4.26

What can we deduce about the numbers themselves, before we calculate

them?
• a must be bigger, because it has the bigger log.
• In fact, a must be a lot bigger, because it has two more digits. (The
difference between the logarithms tells us the difference in the number
of digits a and b have.)
Appendix X: Understanding Logarithms Intuitively 4

• We can see that a is a rather high seven-digit number, and b is a low

five-digit number.

So we really can tell a lot about numbers by just looking at their loga-
rithms, without calculating them. But we can calculate a and b if we choose,
by raising 10 to the two logs:

a = 106.96 = ∼ 9,120,108
b = 104.26 = ∼ 18,197

Everything we deduced above was true, and we were able to do so without

any difficult calculation.

An Interplanetary Example
So why are logarithms useful? Sometimes you might want to compare items
that are so different from each other, that it becomes difficult. For example,
let us say that we wanted to compare the sizes of the major bodies of the
solar system. We want to communicate clearly, so we decide to make a graph
of the volumes in cubic megameters (Mm). (A megameter is 1000 km: about
the distance from Portland, Oregon to Salt Lake City, Utah, or from Paris,
France to Madrid, Spain. It’s a good unit to use when talking about sizes
of planets.)

The bar graph above shows what happens when we display these values
in the traditional way. We can see the problem right away: the Sun is so
much bigger than everything else, that it’s almost impossible to read any
of the other values! Is Saturn bigger or smaller than Neptune? How does
the size of Venus compare to the size of the Moon? The only thing that’s
obvious here is that the Sun is gigantic, but most people already know that.
This graph just isn’t that useful.
But what if instead make a log-scaled graph, like this:
5 Appendix X: Understanding Logarithms Intuitively

Here, the height of each bar is the logarithm of the value. See how every
notch on the left is 10× the value of the one before it? This makes the
overall graph much clearer, because it means that we can graph values that
differ by quite a lot. Effectively, we’re graphing the scales of the volumes.
The Sun is still clearly the biggest thing, but now we can see better how the
planets compare. Jupiter and Saturn are about three scales smaller than
the Sun, so they’re about 1000× smaller. Uranus and Neptune are one scale
smaller still: each one is about one tenth the size of Jupiter or Saturn. And
Earth and Venus are two scales smaller still. So therefore Earth is six scales
smaller than the Sun—or about a million times smaller! This graph is much
more useful and more readable than the one above, thanks to logarithms.
Be careful with log-scaled graphs—they can be misleading if you don’t
read them carefully. For example, it might look to some people that Earth
is about one third the size of the Sun, because its bar is about a third the
height. This just isn’t true—it’s six whole scales smaller. Log-scaled graphs
should only be used when there’s a clear benefit like there is here.

Natural and Binary Logarithms

But wait—aren’t there other bases of logarithms? Everything we’ve done so

far deals with factors of 10: each scale is 10 times bigger than the one before
it. When you memorized (and probably forgot) how to do logarithms when
you were younger, you probably had to do calculations with lots of wacky
bases, like base 7 and base 6 and base 11.
Remember, the base of the logarithm determines the size of the scale.
But what your old math teacher probably didn’t tell you is that most bases
just aren’t used that much in real scientific and engineering work. You
usually don’t need to worry about them. In fact, there are only three bases
that you will commonly see: common logs (base 10), natural logs (base e),
and binary logs (base 2). Here’s how they compare to each other:
Appendix X: Understanding Logarithms Intuitively 6

Notice that the logarithm of 1 is always 0, regardless of base. This is because

no doubling, ×10s, or any other multiplying is needed to attain it. Also, two
different logarithms are always proportional to each other in the same way.
For example, the binary log of some number x will always be ∼ 3.32 times
as big as the base-10 logarithm of x. We’ll talk more about this later.
Let’s talk about these important bases one at a time:
• You’ve already seen base-10 logarithms like this:

log10 31 ≈ 1.49.

Humans like using 10 as a base because it’s intutive. A lot of human-

scaled measurements use base 10. For example, the prefixes of the met-
ric system are a kind of base-10 logarithm, with milli-, centi-, deci-, and
no prefix all referring to different scales that differ by a factor of ×10.
Also, the Richter scale that used to be used to measure earthquakes is
base-10 logarithmic, with a 5.0 earthquake being ten times as strong
as a 4.0 earthquake. And the decibel system used for acoustics makes
use of the base-10 log.
• People doing calculations in calculus frequently use the number e as
the base. This number e (sometimes called “Euler’s number”∗ ) is an
infinitely-long non-repeating decimal number, approximately equal to
2.71828. Like π, it is a mathematical constant of the universe, that
appears in many very important equations and theorems. The base-e
logarithm is often called the natural logarithm because it is funda-
mental to many calculus equations and to natural phenomena. It is
commonly written in equations as “ln” (or rarely as “loge ”).
If you wish to estimate a natural logarithm, it is the same as asking
how many factors of ×e there are in a number. For example,

e × e × e ≈ 20.
∗
Since “Euler” is a German name, it is pronounced much like the English word “oiler”.
7 Appendix X: Understanding Logarithms Intuitively

That is, there are about 3 factors of ×e in 20. Therefore:

ln 20 ≈ 3.

Calculating natural logs in your head is more difficult than the other
common logs, since e is a non-repeating decimal number. Fortunately,
they are rare when studying computer science. You will encounter
them much more when you study calculus, physics, or chemistry.

• People who work with information theory (like computer scientists)

often use a logarithm of base 2, also known as the binary log.∗ Here
the scale is 2 rather than 10. Therefore, the binary log asks “Starting
from 1, how many times was this number doubled in order to reach
its current value?” (Equivalently, you may also ask “How many times
does this number need to be halved in order to get back down to 1?”)
The binary log can be expressed as “log2 ” to distinguish it from other
logarithms. (It is also sometimes written as “lg”.) For example:

log2 64 = 6.

This is because when 1 is doubled six times, it will be exactly 64.

When working with base-2 logarithms, the powers of two are very
important.∗∗ The first few powers are:

20 = 1 23 = 8 26 = 64 29 = 512
21 = 2 24 = 16 27 = 128 210 = 1024
22 = 4 25 = 32 28 = 256

This means that:

log2 1 = 0 log2 8 = 3 log2 64 = 6 log2 512 = 9

log2 2 = 1 log2 16 = 4 log2 128 = 7 log2 1024 = 10
log2 4 = 2 log2 32 = 5 log2 256 = 8

Remember, the binary log of a value is the number of doublings needed

to attain that value. So if you were to double the number 1 ten times,
you would get 1024.
Notice that 210 (1024) is very close to 1000, so:

log2 1000 ≈ 10.

∗
Musicians also work with binary logs, though most of them don’t know it. Each
octave in a musical score represents a doubling of frequency. Therefore a score is basically
a binary-log-scaled graph of notes’ frequencies!
∗∗
You do not need to memorize these powers of 2 now. When you work with computers
long enough, you will memorize them without even realizing that you’ve done so.
Appendix X: Understanding Logarithms Intuitively 8

This useful fact can be used to approximate the binary logarithm of

very large numbers. For example, what is the binary logarithm of
16,000,000,000 (sixteen billion)? Separating out the factors of ×1000,
we see that:
16,000,000,000 = 16 × 1000 × 1000 × 1000.
Since we need 4 doublings to get the 16, and we need about 10 dou-
blings to get each of the three 1000s, the total number of doublings
(the log of 16,000,000,000) must be:
log2 16,000,000,000 = 4+ ∼ 10+ ∼ 10+ ∼ 10
= ∼ 34.
In fact, the real answer is ∼ 33.90, so our approximation is quite good!
• Sometimes, you’ll also see trinary logarithms (base 3) when working
with information theory. These are a lot like base-2 logarithms, except
of course the scale is 3 and the log is a count of triplings, not doublings.
There is no special way of expressing the trinary log, so they are almost
always just “log3 ”. These logs are quite rare, though.

Analyzing Algorithms
It turns out that logarithms are very important for analyzing and comparing
different algorithms to each other. Sometimes, there are two different ways
to solve a problem, and both of them will get the right answer. But if
one method is much quicker and more efficient than the other, than that
method should be used—even if the slower method would eventually solve
the problem.
For example, let us say that you are programming some kind of word
game in which the players take turns playing cards or tiles with letters on
them, in order to spell out words. The computer needs to judge whether or
not a certain play is a legitimate word, rejecting nonwords like “AQUR” and
“NUMPY”. Say you have a big list containing all the allowed words, and
you need to check it every time a play is made, to make sure it’s allowed.
We will call the word we’re checking a “query”. If a query is “valid”, that
means that it’s in the list. An “invalid” query is not.
One approach is to do an exhaustive search, like is shown on the next
page. You just check all the words in the big list one at a time, seeing if
you can find the queried word. But this can take a lot of time. If the query
is valid, on average you’ll have to search through half of them before you
find it. But if the query is invalid, you’ll have to look through every word
in the list and rule them all out one at a time. This search is an example of
a linear algorithm, because the number of checks it makes is proportional
to the number of words in the list. If there were twice as many words, the
9 Appendix X: Understanding Logarithms Intuitively

search would take twice as long. If there were three

times as many words, it would take three times as
long. And so on. It’s not very efficient.
One better way to do this task is to use a binary
search, which can rule out many words simultane-
ously. All it needs is for the list to be presorted al-
phabetically. It starts by checking the middle word
in the list (instead of the first one) and uses that
to rule out half the list. If the middle word comes
before the query, the query cannot be in the first
half. Likewise if the middle word comes after the
query, it cannot be in the second half. (If the mid-
dle word happens to be the right word, the search
is very short indeed.) Keep doing this with all the
words that haven’t been ruled out, until you either
find the query or prove that it can’t be in the list.
Thus each query can rule out half of the remaining
list!
So let’s say that a player has just played
“QUAIL”, and we want to make sure that it’s a
valid play. We’ll start by comparing that to the
word in the middle of the list, which happens to be
“LESSENING”. “QUAIL” comes after “LESSEN-
ING”, so every word before “LESSENING” can be
ruled out. So either “QUAIL” is an invalid word, or
it’s in the second half. (We don’t know yet.) Next
we compare it with the word that is in the mid-
dle of the remaining list: “ROOMED”. “QUAIL”
comes before “ROOMED”, so if QUAIL is valid, it
must be in the first half of what remains. The en-
tire second half (that is, the fourth quarter of the
whole list) can be ruled out. We’ve already elim-
inated three quarters of the whole list, after only
comparing “QUAIL” to two words! This process is
kept up, until “QUAIL” is found.
Every time we check a word, we can rule out
half the remaining words in the list. So how many
checks do you need to make for a list of n words?
That’s the same as asking how many times do you
need to halve n to get down to 1. The answer is
(of course) log2 n. The amount of time it’ll take
is proportional to the log of n, which means that if
you were to double the size of the list, the process of
checking a query would only get slightly longer. We
say that binary search is a logarithmic algorithm. It
Appendix X: Understanding Logarithms Intuitively 10

is much more efficient than a linear algorithm like exhaustive search. Many
other computer algorithms depend on a doubling or halving factor like this,
and so their analysis will involve binary logs. Thus, understanding logs can
help you pick out the fastest method to solve a problem.

Negative Logarithms

So far, we’ve been concentrating on the positive numbers. But not every
logarithm is positive—so it’s natural to ask what the relation is between
negative numbers and logarithms.
First, we usually don’t allow the base or the input of a logarithm to
be negative. We start running into the same problem as when taking the
square root of a negative number: the result is often an imaginary or complex
number. So we’re not going to worry about those cases here.
But if you take the logarithm of a positive number less than 1, the result
will be negative. This makes sense: 1 has a logarithm of 0, so smaller
numbers must have logs less than 0. Negative logarithms represent smaller
and smaller scales. For base-10 logs:

log10 1 = 0 log10 0.001 = −3 log10 0.000001 = −6

log10 0.1 = −1 log10 0.0001 = −4 log10 0.0000001 = −7
log10 0.01 = −2 log10 0.00001 = −5 log10 0.00000001 = −8

So to go a higher scale, we have to multiply by 10. But to go a scale lower,

we have to divide by 10.
Binary logs are similar. Each step down means that we have to halve
one more time:

1 1 1
log2 1 = 0 log2 8 = −3 log2 64 = −6 log2 512 = −9
log2 12 = −1 1
log2 16 = −4 1
log2 128 = −7 1
log2 1024 = −10
log2 14 = −2 1
log2 32 = −5 1
log2 256 = −8

Here’s the graph of base-10, natural, and binary logarithms including

some negative logs:
11 Appendix X: Understanding Logarithms Intuitively

No matter which base we’re using, the logarithms get steeper and steeper as
the input approaches 0. But you can’t really take the logarithm of 0 itself—0
just isn’t on any scale, so the answer isn’t defined. However, you can say
that the log approaches negative infinity as x approaches 0. The log of any
infinitesimally small number will be a very negative number.
Let’s try to estimate the log of a number less than 1. What is the binary
log of 31 ? We know that 13 is between 14 and 12 , whose logs we already know.
If we take the logs of all the numbers in this relation, we can find upper and
lower bounds on the log of 13 :
1 1 1
4 < 3 < 2
1 1 1
log2 4 < log2 3 < log2 2
−2 < log2 13 < −1

Therefore, we know that the binary log of 13 must be between −2 and −1.
(Indeed, it is ∼ −1.58—roughly halfway between them.)

Logarithm Identities
Finally, let’s work through some common logarithm identities. These are the
equations that you might see when studying logarithms in a more formal way.
After each identity, we’ll explain why it is true on an intuitive level.
The first identity can be used to relate different logarithms with bases b
and d to each other:
1
logb d = .
logd b
What does this mean? For now, imagine that b = 2 and d = 10. Remember
that the binary log is counting the number of doublings, and the base-10
Appendix X: Understanding Logarithms Intuitively 12

log is counting the number of ×10 factors, to make up some number x. But
how many doublings are there in one ×10 factor? That’s just the binary log
of 10:
log2 10 ≈ 3.32.
In other words, multiplying something by 10 is the same thing as doubling
it about 3.32 times. Multiplying a number by 100 is the same as doubling it
about 6.64 times. The two logarithms are very closely related. We can also
go the other way:
log10 2 ≈ 0.301.
This means that doubling something is the same thing as multiplying it by 10
about 0.301 times. And those numbers log2 10 (∼ 3.32) and log10 2 (∼ 0.301)
must be reciprocals of each other, since they represent the “exchange rate”
between the two logarithms.
The above boxed equation simply makes it clear that this is true no
matter what b and d are. So long as they’re valid logarithmic bases, you
can find this “exchange rate”, and the two rates must be reciprocals of one
another.
Furthermore, we can take advantage of this to calculate binary logarithms
on calculators that can only do base-10 logarithms. This is expressed in our
next important equation:

logb x = logd x ÷ logd b .

That is, a base-b logarithm of some number is equal to the base-d logarithm
of that same number, divided by the base-d logarithm of b. Let us say
you wanted to calculate the binary log of 256, but your calculator only has
base-10 logs. Not a problem! Here are the steps you can do:

1. Start by calculating the base-10 log of 256, which is ∼ 2.41. So there

are ∼ 2.41 factors of ×10 in 256.

2. Now divide that number by log10 2, which is ∼ 0.301.

Why do we divide by log10 2? Well, really we’d like to multiply by log2 10

(∼ 3.32), because we know that each ×10 is “worth” about 3.32 doublings.
But our calculator doesn’t do binary logs, so we can’t easily calculate that
number. Fortunately, we can calculate its reciprocal: log10 2. So instead
of multiplying by log2 10, we divide by log10 2. It’ll get us the exact same
answer, like so:

log2 64 = log10 64 ÷ log10 2

=∼ 2.41 ÷ ∼ 0.301
= 8.

This could easily be done on a calculator that does base-10 logs. Of course,
the above boxed formula works no matter what b and d are. But you will
13 Appendix X: Understanding Logarithms Intuitively

find this most useful when you’re trying to calculate binary logarithms, when
your equipment won’t let you do it directly.
Next:
logb xy = logb x + logb y .

All this is saying is that the number of ×b factors in the product xy is just
equal to the number of ×b factors in x plus the number of ×b factors in y.
For example, what is log2 4096? Remember, that means how many doublings
you must do to attain 4096. First, notice that:

4096 = 1024 × 4.

And we already know that you need 10 doublings to get to 1024, and 2
doublings to get to 4. Therefore, the total number of doublings to get to
4096 must be 12:

log2 4096 = log2 (1024 × 4)

= log2 1024 + log2 4
= 10 + 2
= 12.

And of course, this is true no matter what base you are using.
A related formula is:
x
logb = logb x − logb y
y

This one just goes in the opposite direction. What if you’re trying to cal-
culate log2 512, but you don’t have it memorized? But you remember that
log2 1024 is 10, and 512 is half of 1024. So you can think of 512 this way:
start at 1 and double it ten times to get up to 1024, but then halve it once
to get back down to 512. Mathematically, this is:

log2 512 = log2 ( 1024

2 )
= log2 1024 − log2 2
= 10 − 1
= 9.

This identity is also related:

logb xp = p × logb x .

If you have some number x raised to a power p, that means that you’re
multiplying together p different instances of x. For example:

x5 = x × x × x × x × x.
Appendix X: Understanding Logarithms Intuitively 14

Each of those instances of x must have the exact same factors. So if you
know the logarithm of x, the logarithm of xp must be p times as great. For
example, what is the binary log of 85 , which is 32,768?

log2 32,768 = log2 85

= 5 × log2 8
=5×3
= 15.

All this really means is that 85 is five different factors of ×8, and each
individual ×8 is 3 doublings. Therefore, 85 must be 5 × 3 = 15 doublings.

Conclusion
Logarithms can be hard when you’re first getting used to them. They’re
a different kind of math, working with scalings of numbers rather than di-
rectly with the numbers themselves. However, you can be at ease with them
once you develop an intuitive sense for the way they work. Hopefully this
appendix has helped with that, and will let you go forward and use them
with more confidence.