1
00:00:00,500 --> 00:00:03,840
GILBERT STRANG: OK, I'm going
to explain Fourier series,

2
00:00:03,840 --> 00:00:06,300
and that I can't
do in 10 minutes.

3
00:00:06,300 --> 00:00:09,460
It'll take two,
maybe three, sessions

4
00:00:09,460 --> 00:00:15,690
to see enough examples
to really use the idea.

5
00:00:15,690 --> 00:00:20,640
Let me start with what
we're looking for.

6
00:00:20,640 --> 00:00:23,190
We have a function.

7
00:00:23,190 --> 00:00:29,100
And we want to write it as
a combination of cosines

8
00:00:29,100 --> 00:00:30,920
and sines.

9
00:00:30,920 --> 00:00:35,880
So those our basis functions--
the cosines and the sine.

10
00:00:35,880 --> 00:00:39,600
And a n's and the b n's
are the coefficients

11
00:00:39,600 --> 00:00:41,480
that we have to look for.

12
00:00:41,480 --> 00:00:47,790
That tells us how much of
cosine nx is in the big function

13
00:00:47,790 --> 00:00:49,270
f of x.

14
00:00:49,270 --> 00:00:55,000
Notice that the cosines start at
n equals 0, because cosine of 0

15
00:00:55,000 --> 00:00:55,950
is 1.

16
00:00:55,950 --> 00:01:00,020
So there's an a0 in our sum.

17
00:01:00,020 --> 00:01:05,090
But there isn't a b0, because
n equals zero of the sine

18
00:01:05,090 --> 00:01:08,680
would be zero, and we
don't get anything there.

19
00:01:08,680 --> 00:01:11,820
So we're looking for
the a n's and b n's.

20
00:01:11,820 --> 00:01:16,000
And, really, I want to
show you, at the same time,

21
00:01:16,000 --> 00:01:21,580
the complex form
with coefficient cn.

22
00:01:21,580 --> 00:01:25,830
And now n goes from minus
infinity to infinity.

23
00:01:25,830 --> 00:01:28,820
That's really the
more beautiful form

24
00:01:28,820 --> 00:01:36,490
because that one formula
for cn does the job,

25
00:01:36,490 --> 00:01:41,390
whereas here I will need
a separate formula for a n

26
00:01:41,390 --> 00:01:43,641
and for bn.

27
00:01:43,641 --> 00:01:44,140
OK.

28
00:01:46,740 --> 00:01:52,440
So this is natural when
the function is real,

29
00:01:52,440 --> 00:01:56,220
but in the end, and for the
discrete Fourier transform,

30
00:01:56,220 --> 00:01:58,560
and for the fast
Fourier transform,

31
00:01:58,560 --> 00:02:00,490
the complex case will win.

32
00:02:00,490 --> 00:02:05,170
And, of course, everybody
sees that e to the inx,

33
00:02:05,170 --> 00:02:10,250
by Euler's great formula, is
a combination of cosine nx

34
00:02:10,250 --> 00:02:11,370
and sine nx.

35
00:02:11,370 --> 00:02:16,470
So, I can use those, or I
can use cosine and sine.

36
00:02:16,470 --> 00:02:17,860
OK.

37
00:02:17,860 --> 00:02:21,360
So, how do you
find these numbers?

38
00:02:21,360 --> 00:02:23,600
The key is orthogonality.

39
00:02:23,600 --> 00:02:28,030
So that's the first central
idea here in Fourier series,

40
00:02:28,030 --> 00:02:30,460
is the idea of orthogonality.

41
00:02:30,460 --> 00:02:31,540
Now what does that mean?

42
00:02:31,540 --> 00:02:33,390
That means perpendicular.

43
00:02:33,390 --> 00:02:36,930
And for a vector,
and a second vector,

44
00:02:36,930 --> 00:02:40,100
we have an idea of what
perpendicular means.

45
00:02:40,100 --> 00:02:42,670
The 90 degree
angle between them.

46
00:02:42,670 --> 00:02:47,730
And we check that by the dot
product-- or inner product,

47
00:02:47,730 --> 00:02:52,250
whichever name you like--
between the two vectors

48
00:02:52,250 --> 00:02:54,090
should be 0.

49
00:02:54,090 --> 00:02:54,590
OK.

50
00:02:54,590 --> 00:03:01,560
But here we have functions--
like cosine functions.

51
00:03:01,560 --> 00:03:06,780
So here's one cosine, and
here's a different cosine.

52
00:03:06,780 --> 00:03:13,840
So those are two different basis
functions-- say, cosine of 7x

53
00:03:13,840 --> 00:03:16,155
and cosine of 12 x.

54
00:03:18,920 --> 00:03:23,310
The coefficients a7
and a12 would tell us

55
00:03:23,310 --> 00:03:27,840
how much of cosine 7x
is in the function.

56
00:03:27,840 --> 00:03:32,180
You see, we're separating the
function into frequencies.

57
00:03:32,180 --> 00:03:37,800
We're looking into pure
oscillations, pure harmonics.

58
00:03:37,800 --> 00:03:42,140
And we expect, probably,
that's the lower harmonics

59
00:03:42,140 --> 00:03:46,930
the smoother ones cos
x, cos 2x, cos 3x,

60
00:03:46,930 --> 00:03:49,270
have most of the energy.

61
00:03:49,270 --> 00:03:54,660
And the high harmonics,
cosine 12x, cosine 100x,

62
00:03:54,660 --> 00:03:57,340
probably those are
quickly alternating,

63
00:03:57,340 --> 00:04:02,990
those contain noise,
and high frequency.

64
00:04:02,990 --> 00:04:05,360
Quick changes in the
function will show up

65
00:04:05,360 --> 00:04:07,210
in the high frequencies.

66
00:04:07,210 --> 00:04:08,470
OK.

67
00:04:08,470 --> 00:04:13,480
So what's the answer to
this integral-- cosine

68
00:04:13,480 --> 00:04:21,589
of 7x times cosine of 12x dx,
over the range minus pi to pi?

69
00:04:21,589 --> 00:04:25,220
Orthogonality comes
in, the answer is 0.

70
00:04:25,220 --> 00:04:27,140
That's the crucial fact.

71
00:04:27,140 --> 00:04:32,460
That's what makes it possible
to separate out a7 and a12

72
00:04:32,460 --> 00:04:34,040
and get hold of them.

73
00:04:34,040 --> 00:04:36,480
So let me show you
how to do that.

74
00:04:36,480 --> 00:04:42,780
So I'm going to use this
fact, which is the function

75
00:04:42,780 --> 00:04:46,300
version of 90 degree angle.

76
00:04:46,300 --> 00:04:49,775
So, you see, it's a
little like a dot product.

77
00:04:53,190 --> 00:04:56,730
Well, let me remember, a dot
product would be something

78
00:04:56,730 --> 00:05:09,650
like c1 d1 plus c2 d2 equals
0, if I had a vector c1 c2

79
00:05:09,650 --> 00:05:12,610
and a vector d1 d2.

80
00:05:12,610 --> 00:05:15,730
That would be the dot
product, and it would be 0

81
00:05:15,730 --> 00:05:17,600
if the vectors are orthogonal.

82
00:05:17,600 --> 00:05:21,020
Here, instead of
adding, I'm integrating

83
00:05:21,020 --> 00:05:23,480
because I have functions.

84
00:05:23,480 --> 00:05:26,710
So just that's the
meaning of dot product--

85
00:05:26,710 --> 00:05:29,120
the integral of
one function times

86
00:05:29,120 --> 00:05:31,090
the other function gives 0.

87
00:05:31,090 --> 00:05:31,820
OK.

88
00:05:31,820 --> 00:05:33,320
I'll use that now.

89
00:05:33,320 --> 00:05:35,450
OK, how will I use this?

90
00:05:35,450 --> 00:05:40,850
I will look what I want.

91
00:05:40,850 --> 00:05:43,310
This is my goal.

92
00:05:43,310 --> 00:05:52,850
I'll multiply both sides of
this equation by cosine kx.

93
00:05:52,850 --> 00:05:55,050
And then I'll integrate.

94
00:05:55,050 --> 00:05:59,820
And the beauty is, that when
I multiply by cosine kx,

95
00:05:59,820 --> 00:06:06,520
and I integrate, everything
goes to zero except what I want.

96
00:06:06,520 --> 00:06:13,880
By the way, all the sines
times cosine kx integrate to 0.

97
00:06:13,880 --> 00:06:17,980
All the sines are orthogonal
to all the cosines.

98
00:06:17,980 --> 00:06:21,120
And all the cosines will
be orthogonal to all

99
00:06:21,120 --> 00:06:22,260
the other cosines.

100
00:06:22,260 --> 00:06:27,150
So let me show you what I get.

101
00:06:27,150 --> 00:06:32,930
So I multiply my f
of x by cosine kx,

102
00:06:32,930 --> 00:06:36,420
and I integrate
from minus pi to pi.

103
00:06:36,420 --> 00:06:37,240
OK?

104
00:06:37,240 --> 00:06:41,610
Now, on the right-hand
side, this is my integral

105
00:06:41,610 --> 00:06:46,530
from minus pi to pi, of my
big sum of all these terms,

106
00:06:46,530 --> 00:06:56,300
0 to infinity, a n cos nx,
etcetera-- including the sines

107
00:06:56,300 --> 00:06:58,150
but I'm not even put
them in because they're

108
00:06:58,150 --> 00:07:04,152
going to get killed by this
integration-- times cosine kx

109
00:07:04,152 --> 00:07:05,940
dx.

110
00:07:05,940 --> 00:07:12,370
All I did was take the f
of x equal that formula,

111
00:07:12,370 --> 00:07:20,180
multiplied both sides by
cosine kx, and integrated.

112
00:07:20,180 --> 00:07:23,550
And, now the
orthogonality pays off,

113
00:07:23,550 --> 00:07:26,930
because this times
this, when I integrate

114
00:07:26,930 --> 00:07:29,690
gives 0, with one exception.

115
00:07:29,690 --> 00:07:36,230
When n equals k, then
I do get the integral.

116
00:07:36,230 --> 00:07:46,550
The only term I get is
ak, cosine kx, twice dx.

117
00:07:46,550 --> 00:07:50,840
Only k equal n
survives this process.

118
00:07:50,840 --> 00:07:53,200
And then that integral
of cosine squared

119
00:07:53,200 --> 00:07:59,730
happens to be pi, so
this is just ak times pi.

120
00:07:59,730 --> 00:08:02,880
Look, I've discovered
what ak is.

121
00:08:02,880 --> 00:08:08,590
I've discovered the k
Fourier cosine coefficient.

122
00:08:08,590 --> 00:08:11,290
I just divide by pi.

123
00:08:11,290 --> 00:08:16,250
So can I just divide by pi
to get this formula for ak?

124
00:08:16,250 --> 00:08:18,180
Ak is 1 over pi.

125
00:08:18,180 --> 00:08:24,120
The integral from minus pi
to pi of my function, times

126
00:08:24,120 --> 00:08:28,320
cosine kx dx.

127
00:08:28,320 --> 00:08:30,330
That's the formula.

128
00:08:30,330 --> 00:08:34,880
That tells me the coefficient.

129
00:08:34,880 --> 00:08:38,549
And I could only do
that with orthogonality

130
00:08:38,549 --> 00:08:42,059
to knock out all but one term.

131
00:08:42,059 --> 00:08:46,170
And now, if I wanted
the sine coefficients,

132
00:08:46,170 --> 00:08:52,220
bk, it would be the same formula
except that would be a sine.

133
00:08:52,220 --> 00:08:55,740
And if I wanted the
complex coefficient,

134
00:08:55,740 --> 00:09:00,980
ck, it turns out it'd be
the same formula expect--

135
00:09:00,980 --> 00:09:04,390
well maybe it's 2 pi
there, 1 over 2 pi--

136
00:09:04,390 --> 00:09:08,810
and this becomes an
e to the minus ikx.

137
00:09:08,810 --> 00:09:14,720
In a complex case, the complex
conjugate e to the minus ikx

138
00:09:14,720 --> 00:09:15,910
shows up.

139
00:09:15,910 --> 00:09:19,880
So this is really the dot
product, the inner product,

140
00:09:19,880 --> 00:09:22,180
of the function with the cosine.

141
00:09:22,180 --> 00:09:23,050
OK.

142
00:09:23,050 --> 00:09:25,050
So let me do some examples.

143
00:09:29,700 --> 00:09:34,930
Maybe I should write up the sine
formula that I just mentioned.

144
00:09:34,930 --> 00:09:43,250
So bk is the integral 1 over
pi, the integral of my function,

145
00:09:43,250 --> 00:09:47,360
times sine kx dx.

146
00:09:47,360 --> 00:09:50,600
And there's one exception.

147
00:09:50,600 --> 00:09:55,500
A0 has a little bit
different formula,

148
00:09:55,500 --> 00:09:58,380
the pi changes to 2 pi.

149
00:09:58,380 --> 00:10:00,030
I'm sorry about that.

150
00:10:00,030 --> 00:10:04,990
When k is 0 or it's the integral
of 1, from minus pi to pi,

151
00:10:04,990 --> 00:10:06,140
and I get 2 pi.

152
00:10:06,140 --> 00:10:15,610
So, a0 is 1 over 2 pi-- the
integral of f of x times

153
00:10:15,610 --> 00:10:20,980
when k is zero
cosine-- this is 1 dx.

154
00:10:20,980 --> 00:10:22,970
That has a simple meaning.

155
00:10:22,970 --> 00:10:30,766
That's the average of f of x.

156
00:10:34,870 --> 00:10:36,050
OK.

157
00:10:36,050 --> 00:10:42,360
So the basis function was
just 1 when k was zero.

158
00:10:42,360 --> 00:10:46,610
When k is 0, the function
of my cosine is just one,

159
00:10:46,610 --> 00:10:51,320
and I get the integral
of the function

160
00:10:51,320 --> 00:10:55,250
times 1 divided by 2 pi.

161
00:10:55,250 --> 00:10:57,980
Could we just do an example?

162
00:10:57,980 --> 00:11:00,020
So I want to take a function.

163
00:11:00,020 --> 00:11:02,610
And in this video
why don't I take

164
00:11:02,610 --> 00:11:07,250
an easy, but very important,
function-- the delta function.

165
00:11:07,250 --> 00:11:11,010
So I plan to use these
formulas on the delta function.

166
00:11:11,010 --> 00:11:14,090
Let me draw a little picture
of the delta function.

167
00:11:14,090 --> 00:11:17,850
I'm only going between
minus pi and pi,

168
00:11:17,850 --> 00:11:20,820
and the delta function,
as we know, is 0,

169
00:11:20,820 --> 00:11:26,690
it's infinite, at the
spike, and 0 again.

170
00:11:26,690 --> 00:11:31,270
The reason I wanted to draw it
is, that's an even function.

171
00:11:31,270 --> 00:11:36,560
That's a function which is
symmetric between x and minus

172
00:11:36,560 --> 00:11:37,570
x.

173
00:11:37,570 --> 00:11:42,520
And in that case,
there will be no sines.

174
00:11:42,520 --> 00:11:45,680
Sine functions are odd.

175
00:11:45,680 --> 00:11:52,870
The integral from minus pi to
pi of an odd function gives 0.

176
00:11:52,870 --> 00:12:00,170
The odd means that when
you cross x equals 0

177
00:12:00,170 --> 00:12:05,860
you get minus the result
for x greater than 0.

178
00:12:05,860 --> 00:12:10,110
So my point is, this
is an even function--

179
00:12:10,110 --> 00:12:17,390
delta of x is the same as delta
of minus x, and only cosines.

180
00:12:21,070 --> 00:12:23,920
Good.

181
00:12:23,920 --> 00:12:28,690
The sine coefficients
automatically dropped our 0 so,

182
00:12:28,690 --> 00:12:30,490
of course, the
integral would show it.

183
00:12:30,490 --> 00:12:33,880
But we see it even
before we integrate.

184
00:12:33,880 --> 00:12:37,820
OK I'm ready for
the delta function.

185
00:12:37,820 --> 00:12:43,010
So I'm going to
write delta of x,

186
00:12:43,010 --> 00:12:45,040
and we remember what
the delta function

187
00:12:45,040 --> 00:12:50,930
is-- a combination of cosines.

188
00:12:54,710 --> 00:12:55,330
OK.

189
00:12:55,330 --> 00:12:59,100
That's the delta function
between minus pi and pi.

190
00:12:59,100 --> 00:13:00,210
OK.

191
00:13:00,210 --> 00:13:02,690
And what's our
formula for the a n?

192
00:13:02,690 --> 00:13:07,770
Well, you remember we had a
special formula for a0, which

193
00:13:07,770 --> 00:13:14,930
was 1/2 pi times the
integral, from minus pi to pi,

194
00:13:14,930 --> 00:13:22,110
of our function, which is delta,
times the basis function, which

195
00:13:22,110 --> 00:13:26,140
n equals 0, the basis
function is 1 dx.

196
00:13:26,140 --> 00:13:28,210
OK, we know the answer to that.

197
00:13:28,210 --> 00:13:31,400
We can integrate
the delta function.

198
00:13:31,400 --> 00:13:34,910
The one key thing about the
integral of the delta function

199
00:13:34,910 --> 00:13:40,990
is, it's always 1-- if we cross
x equals 0, which we will.

200
00:13:40,990 --> 00:13:43,510
So that integral is 1
so I'm getting 1/2 pi.

201
00:13:46,180 --> 00:13:50,260
What about the other
for a coefficient?

202
00:13:50,260 --> 00:13:52,960
So that's 1/pi, now.

203
00:13:52,960 --> 00:14:00,520
The integral from minus pi to
pi of all of my function times

204
00:14:00,520 --> 00:14:02,387
cosine kxdx.

205
00:14:05,380 --> 00:14:06,490
You know what I'm doing.

206
00:14:06,490 --> 00:14:09,520
I'm using my formula to
find the coefficients.

207
00:14:09,520 --> 00:14:13,180
My formula says take the
function, whatever it is--

208
00:14:13,180 --> 00:14:16,290
and in this example,
it's the delta function--

209
00:14:16,290 --> 00:14:19,010
multiply by the
cosine, integrate,

210
00:14:19,010 --> 00:14:22,200
and divide by the factor pi.

211
00:14:22,200 --> 00:14:23,000
OK.

212
00:14:23,000 --> 00:14:25,780
Well, of course, we
can do that integral.

213
00:14:25,780 --> 00:14:28,610
Because when you integrate
a delta function,

214
00:14:28,610 --> 00:14:36,410
times some other function, all
the action is at x equals 0.

215
00:14:36,410 --> 00:14:40,660
At x equals 0,
this function is 1.

216
00:14:40,660 --> 00:14:44,190
And I don't care what it
is elsewhere, it's just 1.

217
00:14:44,190 --> 00:14:48,820
So this is the same as
integrating delta of x times 1,

218
00:14:48,820 --> 00:14:52,670
which gives us-- well, the
interval the delta function 1.

219
00:14:52,670 --> 00:14:56,520
So that integral is one,
so I'm getting 1/pi.

220
00:14:59,100 --> 00:15:01,110
Good.

221
00:15:01,110 --> 00:15:02,020
OK.

222
00:15:02,020 --> 00:15:05,090
So now, do you want
me to write out

223
00:15:05,090 --> 00:15:06,720
the series for the
delta function?

224
00:15:06,720 --> 00:15:09,280
It looks kind of unusual.

225
00:15:09,280 --> 00:15:13,370
This is telling us
something quite remarkable.

226
00:15:13,370 --> 00:15:18,740
It's telling us that all these
coefficients are the same.

227
00:15:18,740 --> 00:15:22,870
All the frequencies,
all the harmonics,

228
00:15:22,870 --> 00:15:27,840
are in the delta function
in equal amounts.

229
00:15:27,840 --> 00:15:33,710
Usually, we would see a big
drop off of the coefficients ak,

230
00:15:33,710 --> 00:15:37,000
but for the delta function,
which is so singular,

231
00:15:37,000 --> 00:15:39,940
all a big spike at
one point, there's

232
00:15:39,940 --> 00:15:43,690
no drop off and no decay
in the coefficients,

233
00:15:43,690 --> 00:15:45,170
they just constant.

234
00:15:45,170 --> 00:15:45,740
OK.

235
00:15:45,740 --> 00:15:52,010
So I'm saying that the delta
function is the constant term,

236
00:15:52,010 --> 00:16:04,200
1/2pi, and then 1/pi times
cosine of x, and cosine of 2x,

237
00:16:04,200 --> 00:16:05,390
and so on.

238
00:16:10,760 --> 00:16:12,480
OK.

239
00:16:12,480 --> 00:16:15,540
All frequencies
there are the same.

240
00:16:15,540 --> 00:16:21,530
And I'll stop with
that one example here.

241
00:16:21,530 --> 00:16:24,520
So the key points
were orthogonality,

242
00:16:24,520 --> 00:16:29,390
the formulas for the the
coefficients, and this example.

243
00:16:29,390 --> 00:16:31,200
Thank you.