1
00:00:15,259 --> 00:00:18,960
The goal here is simple, explain what a derivative is.

2
00:00:19,160 --> 00:00:21,658
The thing is though, there's some subtlety to this topic,

3
00:00:21,658 --> 00:00:24,200
and a lot of potential for paradoxes if you're not careful.

4
00:00:24,780 --> 00:00:27,551
So a secondary goal is that you have an appreciation

5
00:00:27,551 --> 00:00:30,219
for what those paradoxes are and how to avoid them.

6
00:00:31,219 --> 00:00:35,591
You see, it's common for people to say that the derivative measures an instantaneous

7
00:00:35,591 --> 00:00:39,759
rate of change, but when you think about it, that phrase is actually an oxymoron.

8
00:00:40,240 --> 00:00:43,542
Change is something that happens between separate points in time,

9
00:00:43,542 --> 00:00:46,646
and when you blind yourself to all but just a single instant,

10
00:00:46,646 --> 00:00:48,600
there's not really any room for change.

11
00:00:49,500 --> 00:00:51,840
You'll see what I mean more as we get into it,

12
00:00:51,840 --> 00:00:56,022
but when you appreciate that a phrase like instantaneous rate of change is actually

13
00:00:56,021 --> 00:01:00,103
nonsense, I think it makes you appreciate just how clever the fathers of calculus

14
00:01:00,103 --> 00:01:02,991
were in capturing the idea that phrase is meant to evoke,

15
00:01:02,991 --> 00:01:05,980
but with a perfectly sensible piece of math, the derivative.

16
00:01:07,540 --> 00:01:11,877
As our central example, I want you to imagine a car that starts at some point A,

17
00:01:11,876 --> 00:01:15,839
speeds up, and then slows down to a stop at some point B 100 meters away,

18
00:01:15,840 --> 00:01:19,000
and let's say it all happens over the course of 10 seconds.

19
00:01:20,519 --> 00:01:23,899
That's the setup to have in mind as we lay out what the derivative is.

20
00:01:23,900 --> 00:01:29,099
Well, we could graph this motion, letting the vertical axis represent the

21
00:01:29,099 --> 00:01:34,579
distance traveled, and the horizontal axis represent time, so at each time t,

22
00:01:34,578 --> 00:01:38,723
represented with a point somewhere on the horizontal axis,

23
00:01:38,724 --> 00:01:44,134
the height of the graph tells us how far the car has traveled in total after

24
00:01:44,134 --> 00:01:45,540
that amount of time.

25
00:01:46,760 --> 00:01:50,160
It's pretty common to name a distance function like this s of t.

26
00:01:50,159 --> 00:01:52,706
I would use the letter d for distance, but that

27
00:01:52,706 --> 00:01:55,359
guy already has another full time job in calculus.

28
00:01:56,500 --> 00:01:59,760
Initially, the curve is quite shallow, since the car is slow to start.

29
00:02:00,280 --> 00:02:04,340
During that first second, the distance it travels doesn't change that much.

30
00:02:04,980 --> 00:02:07,582
For the next few seconds, as the car speeds up,

31
00:02:07,581 --> 00:02:10,454
the distance traveled in a given second gets larger,

32
00:02:10,455 --> 00:02:13,219
which corresponds to a steeper slope in this graph.

33
00:02:13,800 --> 00:02:17,520
Then towards the end, when it slows down, that curve shallows out again.

34
00:02:20,759 --> 00:02:25,516
If we were to plot the car's velocity in meters per second as a function of time,

35
00:02:25,516 --> 00:02:27,199
it might look like this bump.

36
00:02:27,860 --> 00:02:30,000
At early times, the velocity is very small.

37
00:02:30,460 --> 00:02:34,224
Up to the middle of the journey, the car builds up to some maximum velocity,

38
00:02:34,223 --> 00:02:36,619
covering a relatively large distance each second.

39
00:02:37,659 --> 00:02:39,919
Then it slows back down towards a speed of zero.

40
00:02:41,379 --> 00:02:44,180
These two curves are definitely related to each other.

41
00:02:44,840 --> 00:02:47,159
If you change the specific distance vs.

42
00:02:47,259 --> 00:02:50,299
time function, you'll have some different velocity vs.

43
00:02:50,419 --> 00:02:51,079
time function.

44
00:02:51,759 --> 00:02:55,039
What we want to understand is the specifics of that relationship.

45
00:02:55,680 --> 00:02:59,099
Exactly how does velocity depend on a distance vs.

46
00:02:59,400 --> 00:02:59,819
time function?

47
00:03:01,939 --> 00:03:04,681
To do that, it's worth taking a moment to think

48
00:03:04,681 --> 00:03:07,539
critically about what exactly velocity means here.

49
00:03:08,379 --> 00:03:11,879
Intuitively, we all might know what velocity at a given moment means,

50
00:03:11,879 --> 00:03:14,979
it's just whatever the car's speedometer shows in that moment.

51
00:03:17,180 --> 00:03:21,457
Intuitively, it might make sense that the car's velocity should be higher at times when

52
00:03:21,457 --> 00:03:25,639
this distance function is steeper, when the car traverses more distance per unit time.

53
00:03:26,699 --> 00:03:30,719
But the funny thing is, velocity at a single moment makes no sense.

54
00:03:31,360 --> 00:03:34,895
If I show you a picture of a car, just a snapshot in an instant,

55
00:03:34,895 --> 00:03:38,540
and I ask you how fast it's going, you'd have no way of telling me.

56
00:03:39,620 --> 00:03:42,379
What you'd need are two separate points in time to compare.

57
00:03:43,180 --> 00:03:47,310
That way you can compute whatever the change in distance across those times is,

58
00:03:47,310 --> 00:03:48,860
divided by the change in time.

59
00:03:49,560 --> 00:03:49,740
Right?

60
00:03:49,819 --> 00:03:54,159
I mean, that's what velocity is, it's the distance traveled per unit time.

61
00:03:55,620 --> 00:03:59,069
So how is it that we're looking at a function for velocity that

62
00:03:59,069 --> 00:04:02,359
only takes in a single value of t, a single snapshot in time?

63
00:04:02,900 --> 00:04:04,280
It's weird, isn't it?

64
00:04:04,280 --> 00:04:07,868
We want to associate individual points in time with a velocity,

65
00:04:07,868 --> 00:04:12,300
but actually computing velocity requires comparing two separate points in time.

66
00:04:14,639 --> 00:04:17,399
If that feels strange and paradoxical, good!

67
00:04:17,920 --> 00:04:20,959
You're grappling with the same conflicts that the fathers of calculus did.

68
00:04:21,379 --> 00:04:25,339
And if you want a deep understanding for rates of change, not just for a moving car,

69
00:04:25,339 --> 00:04:29,346
but for all sorts of things in science, you're going to need to resolve this apparent

70
00:04:29,346 --> 00:04:29,719
paradox.

71
00:04:32,199 --> 00:04:34,705
First, I think it's best to talk about the real world,

72
00:04:34,706 --> 00:04:36,939
and then we'll go into a purely mathematical one.

73
00:04:37,540 --> 00:04:40,460
Let's think about what the car's speedometer is probably doing.

74
00:04:41,199 --> 00:04:43,932
At some point, say 3 seconds into the journey,

75
00:04:43,932 --> 00:04:48,757
the speedometer might measure how far the car goes in a very small amount of time,

76
00:04:48,757 --> 00:04:52,420
maybe the distance traveled between 3 seconds and 3.01 seconds.

77
00:04:53,360 --> 00:04:57,514
Then it could compute the speed in meters per second as that tiny

78
00:04:57,514 --> 00:05:01,860
distance traversed in meters divided by that tiny time, 0.01 seconds.

79
00:05:02,899 --> 00:05:05,555
That is, a physical car just side-steps the paradox and

80
00:05:05,555 --> 00:05:08,259
doesn't actually compute speed at a single point in time.

81
00:05:08,779 --> 00:05:11,679
It computes speed during a very small amount of time.

82
00:05:13,180 --> 00:05:18,687
So let's call that difference in time dt, which you might think of as 0.01 seconds,

83
00:05:18,687 --> 00:05:22,360
and let's call that resulting difference in distance ds.

84
00:05:22,959 --> 00:05:26,743
So the velocity at some point in time is ds divided by dt,

85
00:05:26,744 --> 00:05:30,400
the tiny change in distance over the tiny change in time.

86
00:05:31,579 --> 00:05:35,339
Graphically, you can imagine zooming in on some point of this distance vs.

87
00:05:35,500 --> 00:05:37,680
time graph above t equals 3.

88
00:05:38,560 --> 00:05:43,143
That dt is a small step to the right, since time is on the horizontal axis,

89
00:05:43,142 --> 00:05:47,001
and that ds is the resulting change in the height of the graph,

90
00:05:47,002 --> 00:05:50,439
since the vertical axis represents the distance traveled.

91
00:05:51,220 --> 00:05:55,472
So ds divided by dt is something you can think of as the rise

92
00:05:55,471 --> 00:05:59,519
over run slope between two very close points on this graph.

93
00:06:00,699 --> 00:06:03,439
Of course, there's nothing special about the value t equals 3.

94
00:06:03,939 --> 00:06:06,797
We could apply this to any other point in time,

95
00:06:06,797 --> 00:06:10,665
so we consider this expression ds over dt to be a function of t,

96
00:06:10,665 --> 00:06:15,605
something where I can give you a time t and you can give me back the value of this

97
00:06:15,605 --> 00:06:18,879
ratio at that time, the velocity as a function of time.

98
00:06:19,600 --> 00:06:22,838
For example, when I had the computer draw this bump curve here,

99
00:06:22,838 --> 00:06:27,240
the one representing the velocity function, here's what I had the computer actually do.

100
00:06:27,939 --> 00:06:32,620
First, I chose a small value for dt, I think in this case it was 0.01.

101
00:06:33,439 --> 00:06:38,207
Then I had the computer look at a whole bunch of times t between 0 and 10,

102
00:06:38,208 --> 00:06:41,386
and compute the distance function s at t plus dt,

103
00:06:41,386 --> 00:06:44,820
and then subtract off the value of that function at t.

104
00:06:45,420 --> 00:06:51,064
In other words, that's the difference in the distance traveled between the given time,

105
00:06:51,064 --> 00:06:53,660
t, and the time 0.01 seconds after that.

106
00:06:54,519 --> 00:06:58,305
Then you can just divide that difference by the change in time, dt,

107
00:06:58,305 --> 00:07:02,480
and that gives you velocity in meters per second around each point in time.

108
00:07:04,420 --> 00:07:08,721
So with a formula like this, you could give the computer any curve representing any

109
00:07:08,721 --> 00:07:12,920
distance function s of t, and it could figure out the curve representing velocity.

110
00:07:13,540 --> 00:07:17,437
Now would be a good time to pause, reflect, and make sure this idea

111
00:07:17,437 --> 00:07:21,622
of relating distance to velocity by looking at tiny changes makes sense,

112
00:07:21,622 --> 00:07:25,520
because we're going to tackle the paradox of the derivative head on.

113
00:07:27,480 --> 00:07:32,771
This idea of ds over dt, a tiny change in the value of the function s divided by

114
00:07:32,771 --> 00:07:38,000
the tiny change in the input that caused it, that's almost what a derivative is.

115
00:07:38,699 --> 00:07:43,908
And even though a car's speedometer will actually look at a concrete change in time,

116
00:07:43,908 --> 00:07:49,055
like 0.01 seconds, and even though the drawing program here is looking at an actual

117
00:07:49,055 --> 00:07:54,447
concrete change in time, in pure math the derivative is not this ratio ds over dt for a

118
00:07:54,447 --> 00:07:59,962
specific choice of dt. Instead, it's whatever that ratio approaches as your choice for dt

119
00:07:59,963 --> 00:08:00,760
approaches 0.

120
00:08:02,540 --> 00:08:07,333
Luckily there is a really nice visual understanding for what it means to ask what

121
00:08:07,333 --> 00:08:11,075
this ratio approaches, Remember, for any specific choice of dt,

122
00:08:11,074 --> 00:08:15,810
this ratio ds over dt is the slope of a line passing through two separate points

123
00:08:15,810 --> 00:08:16,980
on the graph, right?

124
00:08:17,740 --> 00:08:22,357
Well as dt approaches 0, and as those two points approach each other,

125
00:08:22,357 --> 00:08:26,314
the slope of the line approaches the slope of a line that's

126
00:08:26,314 --> 00:08:30,140
tangent to the graph at whatever point t we're looking at.

127
00:08:30,579 --> 00:08:33,928
So the true honest-to-goodness pure math derivative is not the

128
00:08:33,928 --> 00:08:37,119
rise over run slope between two nearby points on the graph,

129
00:08:37,119 --> 00:08:41,000
it's equal to the slope of a line tangent to the graph at a single point.

130
00:08:42,360 --> 00:08:45,864
Now notice what I'm not saying, I'm not saying that the derivative is

131
00:08:45,864 --> 00:08:49,420
whatever happens when dt is infinitely small, whatever that would mean.

132
00:08:50,000 --> 00:08:52,340
Nor am I saying that you plug in 0 for dt.

133
00:08:53,039 --> 00:08:58,899
This dt is always a finitely small non-zero value, it's just that it approaches 0 is all.

134
00:09:03,620 --> 00:09:04,960
I think that's really clever.

135
00:09:05,379 --> 00:09:08,277
Even though change in an instant makes no sense,

136
00:09:08,277 --> 00:09:12,003
this idea of letting dt approach 0 is a really sneaky backdoor

137
00:09:12,003 --> 00:09:16,379
way to talk reasonably about the rate of change at a single point in time.

138
00:09:17,019 --> 00:09:17,519
Isn't that neat?

139
00:09:18,059 --> 00:09:20,495
It's kind of flirting with the paradox of change in

140
00:09:20,495 --> 00:09:22,980
an instant without ever needing to actually touch it.

141
00:09:23,299 --> 00:09:25,743
And it comes with such a nice visual intuition too,

142
00:09:25,744 --> 00:09:28,660
as the slope of a tangent line to a single point on the graph.

143
00:09:30,159 --> 00:09:33,098
And because change in an instant still makes no sense,

144
00:09:33,099 --> 00:09:37,427
I think it's healthiest for you to think of this slope not as some instantaneous

145
00:09:37,427 --> 00:09:41,543
rate of change, but instead as the best constant approximation for a rate of

146
00:09:41,543 --> 00:09:42,720
change around a point.

147
00:09:44,340 --> 00:09:46,940
By the way, it's worth saying a couple words on notation here.

148
00:09:47,340 --> 00:09:51,743
Throughout this video I've been using dt to refer to a tiny change in t with

149
00:09:51,743 --> 00:09:55,403
some actual size, and ds to refer to the resulting change in s,

150
00:09:55,403 --> 00:09:59,807
which again has an actual size, and this is because that's how I want you to

151
00:09:59,807 --> 00:10:00,779
think about them.

152
00:10:01,659 --> 00:10:05,795
But the convention in calculus is that whenever you're using the letter d like this,

153
00:10:05,796 --> 00:10:08,910
you're kind of announcing your intention that eventually you're

154
00:10:08,909 --> 00:10:11,100
going to see what happens as dt approaches 0.

155
00:10:11,919 --> 00:10:16,828
For example, the honest-to-goodness pure math derivative is written as ds divided by dt,

156
00:10:16,828 --> 00:10:19,697
even though it's technically not a fraction per se,

157
00:10:19,697 --> 00:10:23,779
but whatever that fraction approaches for smaller and smaller nudges in t.

158
00:10:25,779 --> 00:10:27,679
I think a specific example should help here.

159
00:10:28,259 --> 00:10:31,374
You might think that asking about what this ratio approaches

160
00:10:31,374 --> 00:10:35,303
for smaller and smaller values would make it much more difficult to compute,

161
00:10:35,303 --> 00:10:37,500
but weirdly it kind of makes things easier.

162
00:10:38,200 --> 00:10:43,160
Let's say you have a given distance vs time function that happens to be exactly t cubed.

163
00:10:43,159 --> 00:10:47,771
So after 1 second the car has traveled 1 cubed equals 1 meters,

164
00:10:47,772 --> 00:10:52,240
after 2 seconds it's traveled 2 cubed, or 8 meters, and so on.

165
00:10:53,019 --> 00:10:55,635
Now what I'm about to do might seem somewhat complicated,

166
00:10:55,635 --> 00:10:57,800
but once the dust settles it really is simpler,

167
00:10:57,801 --> 00:11:01,680
and more importantly it's the kind of thing you only ever have to do once in calculus.

168
00:11:03,100 --> 00:11:06,951
Let's say you wanted to compute the velocity, ds divided by dt,

169
00:11:06,951 --> 00:11:09,299
at some specific time, like t equals 2.

170
00:11:09,940 --> 00:11:13,257
For right now let's think of dt as having an actual size,

171
00:11:13,256 --> 00:11:16,459
some concrete nudge, we'll let it go to 0 in just a bit.

172
00:11:17,139 --> 00:11:22,323
The tiny change in distance between 2 seconds and 2 plus dt

173
00:11:22,323 --> 00:11:27,939
seconds is s of 2 plus dt minus s of 2, and we divide that by dt.

174
00:11:28,620 --> 00:11:34,659
Since our function is t cubed, that numerator looks like 2 plus dt cubed minus 2 cubed.

175
00:11:35,259 --> 00:11:38,100
And this is something we can work out algebraically.

176
00:11:38,100 --> 00:11:42,320
Again, bear with me, there's a reason I'm showing you the details here.

177
00:11:42,799 --> 00:11:49,843
When you expand that top, what you get is 2 cubed plus 3 times 2 squared dt

178
00:11:49,844 --> 00:11:57,260
plus 3 times 2 times dt squared plus dt cubed, and all of that is minus 2 cubed.

179
00:11:58,379 --> 00:12:01,961
Now there's a lot of terms, and I want you to remember that it looks like a mess,

180
00:12:01,961 --> 00:12:02,879
but it does simplify.

181
00:12:03,779 --> 00:12:05,899
Those 2 cubed terms cancel out.

182
00:12:06,519 --> 00:12:11,606
Everything remaining here has a dt in it, and since there's a dt on the bottom there,

183
00:12:11,606 --> 00:12:13,559
many of those cancel out as well.

184
00:12:14,279 --> 00:12:19,682
What this means is that the ratio ds divided by dt has boiled down into

185
00:12:19,682 --> 00:12:24,860
3 times 2 squared plus 2 different terms that each have a dt in them.

186
00:12:25,580 --> 00:12:30,129
So if we ask what happens as dt approaches 0, representing the idea of looking at a

187
00:12:30,129 --> 00:12:34,679
smaller and smaller change in time, we can just completely ignore those other terms.

188
00:12:36,100 --> 00:12:39,250
By eliminating the need to think about a specific dt,

189
00:12:39,250 --> 00:12:43,100
we've eliminated a lot of the complication in the full expression.

190
00:12:43,879 --> 00:12:47,360
So what we're left with is this nice clean 3 times 2 squared.

191
00:12:48,360 --> 00:12:52,490
You can think of that as meaning that the slope of a line tangent to

192
00:12:52,490 --> 00:12:56,919
the point at t equals 2 of this graph is exactly 3 times 2 squared, or 12.

193
00:12:57,820 --> 00:13:01,060
And of course, there's nothing special about the time t equals 2.

194
00:13:01,559 --> 00:13:04,720
We could more generally say that the derivative

195
00:13:04,721 --> 00:13:08,080
of t cubed as a function of t is 3 times t squared.

196
00:13:10,740 --> 00:13:13,220
Now take a step back, because that's beautiful.

197
00:13:13,820 --> 00:13:16,280
The derivative is this crazy complicated idea.

198
00:13:16,600 --> 00:13:19,623
We've got tiny changes in distance over tiny changes in time,

199
00:13:19,623 --> 00:13:22,208
but instead of looking at any specific one of those,

200
00:13:22,207 --> 00:13:24,500
we're talking about what that thing approaches.

201
00:13:24,500 --> 00:13:26,980
I mean, that's a lot to think about.

202
00:13:27,639 --> 00:13:31,559
And yet what we've come out with is such a simple expression, 3 times t squared.

203
00:13:32,960 --> 00:13:36,060
And in practice, you wouldn't go through all this algebra each time.

204
00:13:36,419 --> 00:13:40,413
Knowing that the derivative of t cubed is 3t squared is one of those things that all

205
00:13:40,413 --> 00:13:44,500
calculus students learn how to do immediately without having to re-derive it each time.

206
00:13:45,059 --> 00:13:48,339
And in the next video, I'm going to show you a nice way to think about

207
00:13:48,340 --> 00:13:51,759
this and a couple other derivative formulas in really nice geometric ways.

208
00:13:52,500 --> 00:13:56,384
But the point I want to make by showing you all of the algebraic guts

209
00:13:56,384 --> 00:14:00,326
here is that when you consider the tiny change in distance caused by a

210
00:14:00,326 --> 00:14:04,600
tiny change in time for some specific value of dt, you'd have kind of a mess.

211
00:14:05,259 --> 00:14:08,902
But when you consider what that ratio approaches as dt approaches 0,

212
00:14:08,902 --> 00:14:13,020
it lets you ignore much of that mess, and it really does simplify the problem.

213
00:14:13,779 --> 00:14:16,720
That right there is kind of the heart of why calculus becomes useful.

214
00:14:18,019 --> 00:14:21,528
Another reason to show you a concrete derivative like this is that it

215
00:14:21,528 --> 00:14:25,088
sets the stage for an example of the kind of paradoxes that come about

216
00:14:25,089 --> 00:14:28,700
if you believe too much in the illusion of instantaneous rate of change.

217
00:14:30,000 --> 00:14:34,812
So think about the actual car traveling according to this t cubed distance function,

218
00:14:34,812 --> 00:14:38,720
and consider its motion at the moment t equals 0, right at the start.

219
00:14:39,700 --> 00:14:43,379
Now ask yourself whether or not the car is moving at that time.

220
00:14:45,559 --> 00:14:50,298
On the one hand, we can compute its speed at that point using the derivative,

221
00:14:50,298 --> 00:14:53,700
3t squared, which for time t equals 0 works out to be 0.

222
00:14:54,779 --> 00:14:59,990
Visually, this means that the tangent line to the graph at that point is perfectly flat,

223
00:14:59,990 --> 00:15:03,269
so the car's quote-unquote instantaneous velocity is 0,

224
00:15:03,269 --> 00:15:06,139
and that suggests that obviously it's not moving.

225
00:15:07,159 --> 00:15:11,860
But on the other hand, if it doesn't start moving at time 0, when does it start moving?

226
00:15:12,580 --> 00:15:14,540
Really, pause and ponder that for a moment.

227
00:15:15,100 --> 00:15:17,779
Is the car moving at time t equals 0?

228
00:15:22,600 --> 00:15:23,379
Do you see the paradox?

229
00:15:24,259 --> 00:15:26,000
The issue is that the question makes no sense.

230
00:15:26,539 --> 00:15:30,439
It references the idea of change in a moment, but that doesn't actually exist.

231
00:15:30,860 --> 00:15:32,600
That's just not what the derivative measures.

232
00:15:33,480 --> 00:15:38,369
What it means for the derivative of a distance function to be 0 is that the best

233
00:15:38,369 --> 00:15:43,320
constant approximation for the car's velocity around that point is 0 m per second.

234
00:15:44,080 --> 00:15:47,580
For example, if you look at an actual change in time,

235
00:15:47,580 --> 00:15:51,080
say between time 0 and 0.1 seconds, the car does move.

236
00:15:51,500 --> 00:15:53,700
It moves 0.001 m.

237
00:15:54,600 --> 00:15:59,853
That's very small, and importantly, it's very small compared to the change in time,

238
00:15:59,852 --> 00:16:02,979
giving an average speed of only 0.01 m per second.

239
00:16:03,679 --> 00:16:08,704
And remember, what it means for the derivative of this motion to be 0 is that

240
00:16:08,705 --> 00:16:13,860
for smaller and smaller nudges in time, this ratio of m per second approaches 0.

241
00:16:14,840 --> 00:16:16,720
But that's not to say that the car is static.

242
00:16:17,539 --> 00:16:20,966
Approximating its movement with a constant velocity of 0 is,

243
00:16:20,966 --> 00:16:22,820
after all, just an approximation.

244
00:16:24,340 --> 00:16:29,211
So whenever you hear people refer to the derivative as an instantaneous rate of change,

245
00:16:29,211 --> 00:16:33,472
a phrase which is intrinsically oxymoronic, I want you to think of that as a

246
00:16:33,472 --> 00:16:37,679
conceptual shorthand for the best constant approximation for rate of change.

247
00:16:39,179 --> 00:16:42,126
In the next couple videos, I'll be talking more about the derivative,

248
00:16:42,126 --> 00:16:45,241
what it looks like in different contexts, how do you actually compute it,

249
00:16:45,241 --> 00:16:48,399
why is it useful, things like that, focusing on visual intuition as always.