Proofs of identities in FM

On The Floor

9 May 2016

1 Binomial expansion (P3)

 
n−1
Y
∞ 
 (k − i) 

i=0
X
(1 + x)k = 1 + xn  for k ∈ Q.
 
n!

 
n=1  
 
k k! k(k − 1) · · · (k − n + 1)
Proof: Notice, at first, that for nonnegative integers k and n, = = .
n n!(k − n)! n!
 n−1 
Y
 (k − i) 
 i=0 
For n ≥ 1 this is also equivalent to   n!
. We temporarily extend this notation

 
 
k
of to all real numbers k and all nonnegative integers n, so what we need to prove is
n
∞  
k
X k n
(1 + x) = x .
n
n=0
We split the problem into few steps, each expanding the set of the numbers for which the
above statement is true.

1. k ∈ N0 .
Assume k ≥ 1 since k = 0 is trivial. Write the expression as the product of k identical
factors 1 + x, and we know that for i ∈ [0, k] we have ki ways to choose x for i times and

k
1 for k − i times. This yields the coefficient of xi as . For xi where i > k simply note
i
i−1
Y
(k − j)
j=0
that the term contains factor 0 when j = k, so we are done for this case.
i!
2. k ∈ Z.
1
Proof: Let’s do induction, each time reducing k. For base case k = −1 notice that 1+x is
X∞ ∞
X (−1)(−2) · · · · · (−n)
equal to the sum to infinity 1 − x + x2 − x3 + · · · = (−1)n xn = xn
(1)(2) · · · · · (n)
n=1 n=1
∞  
X −1 n
= x .
n
n=0

1
 ∞  
X k
Inductive step: Suppose that (1 + x)k = xn for some k < 0. Differentiat-
n
n=0
∞   ∞  
k−1
X k n−1 k−1
X n k n−1
ing both sides yields k(1 + x) = nx , or (1 + x) = x =
n k n
n=0 n=0
∞ ∞ ∞  
X n k(k − 1) · · · (n − k + 1) n−1 X (k − 1) · · · (n − k + 1) n−1 X k − 1 n−1
· x = x = x
k 1 · 2 · ··· · n 1 · 2 · · · · · (n − 1) n−1
n=1 n=1 n=1
∞  
X k−1 n
= x , as desired.
n
n=0

3. k ∈ Q.
∞  
! ∞   ! ∞   !
X k n X l
n
X k+l n
Proof: First, we prove that x x = x . Notice
n n n
n=0 n=0 n=0
that this identity already holds for k, l ∈ Z, since from the frst two subproblems we have
∞   ∞   ∞ 
! !  !
X k n X l n X k+l n
x x = (1 + x)k · (1 + x)l , while x = (1 + x)k+l .
n n n
n=0 n=0 n=0
To prove this, we need to verify that for each n ≥ 0, coefficient of xn is the same for LHS
  n   
k+l X k l
and RHS. For RHS it is . For LHS this is . Observe also that
n i n−i
i=0
  X n   
k+l k l
− is a polynomial in variables k and l if n is fixed, so name it
n i n−i
i=0
P (k, l). Our aim is to prove that P (k, l) is identically zero.
The fact P (k, l) ≡ 0 for k, l ∈ Z already follows from above. Now fix k as arbitrary integer
  X n   
k+l k l
and vary l, we know that − is a polynomial in variable l and
n i n−i
i=0
has degree at most n. However, every integer is the root of this monovariable polynomial,
it has infinitely many roots and hance must be identically zero. Also that for an k, l ∈ R,
P (k, l) = P (l, k) (P is a symmetric polynomial) so we can assert that P (k, l) ≡ 0 whenever
one of k, l is an integer. Now turn back to P (k, l) again, and fix k as any real number and
vary l (so P (k, l) is again a polynomial with variable l), and this time P (k, l) = 0 for any
integer l (again!) Therefore this monovariable polynomial must be identically zero, again,
and we have P (k, l) = 0 for any pair of k, l ∈ R.
∞  
! ∞   ! ∞   !
p
X k n X l
n
X k+l n
Finally, let k = q , and x x = x means that
n n n
!q n=0 n=0 n=0
∞   ∞   ∞  
! !
X k n X kq n X p n
x = x = x = (1 + x)p (since p is an integer and
n n n
n=0 n=0 n=0
∞  
!
X k n
we already prove this grand identity true for all integers). This means x =
n
n=0
p
(1 + x) q = (1 + x)k , or possibly −(1 + x)k if q is even. Nevertheless, comparing the
constant term (LHS=RHS=1) yields that −(1 + x)k is impossible, so we are done.
Comment 1: It is possibe to establish P (k, l) ≡ 0 by algebraic expansion, but too messy.
Comment 2: We can jump from problem 1 directly to problem 3, but the proof of problem
2 shows the beauty of differentiation.

2
2 Calculus.
1. If a polynomial an xn + an−1 xn−1 + · · · + a0 has roots z1 , z2 , · · · , zk , each with multiplic-
ity b1 , b2 , · · · , bk respectively (so b1 + b2 + · · · + bk = n and bi ≥ 1), then the equation
dn y dn−1 y dy
Aij xj−1 ezi x for
P
an n + an−1 n−1 + · · · + a1 + a0 y = 0 has general solution y =
dx dx dx
i = 1, 2, · · · , k and j = 1, 2, · · · bi and for Aij arbitrary constant.
Proof: Name the polynomial P . First, notice that this general solution has exactly n
parameters (or n degree of freedom) because we can decide arbitrarily on what should
the values f (0), f 0 (0), f 00 (0), · · · f (n−1) (0) and let the rest follow according to our differ-
ential equation (we can’t go further than this since the differential equation also implies
dn+k y dn−1+k y dk+1 y dk y
an n+k + an−1 n−1+k + · · · + a1 k+1 + a0 k = 0 for any k ≥ 0) and these n degrees
dx dx dx dx
of freedom gives n parameters, according to Maclaurin’s theorem. Hence it suffices to
show that xj−1 ezi x works (we can omit any constant since that will only multiply LHS by
Aij .
Now let’s investigate what happens as we differentiate xk eax in general. We claim, by
n
dn k ax
 
n−i n
X
induction, that n
(x e )=a x e + k ax a k(k − 1) · · · (k − i + 1)xk−i eax (notice
dxn i
i=1
that when i > k such terms contain factor k − (k + 1) + 1 = 0 so it can be disregarded (in
other words, no term has factor xj eax for j < 0).
n
dn k ax
 
n−i n
X
Base case n = 0 is trivial. Now assume that n
(x e )=a x e + k ax a k(k −
dxn i
i=1
n
dn+1 k ax
 
k−i ax d n k ax
X
n−i n
1) · · · (k − i + 1)x e , then (x e )= (a x e + a k(k − 1) · · · (k −
dxn+1 dx i
n   i=1
k−i ax
X
n−i n
i + 1)x e )=a(an xk eax ) + kan xk−1 eax + a(a k(k − 1) · · · (k − i + 1)xk−i eax ) +
i
i=1
n  
X n
(k − i)(an−i k(k − 1) · · · (k − i + 1)xk−i−1 eax ). Now the term xk eax has coeffi-
i
i=1  
n+1 k−i ax n−i+1 n
cient a and x e has coefficient a k(k − 1) · · · (k − i + 1)+ an−i+1 k(k −
i
dn+1
   
n n + 1 n−i+1
1) · · · (k−i+1) = a k(k−1) · · · (k−i+1) so n+1 (xk eax )=an+1 xk eax +
i−1 i dx
n+1  
X n+1
an+1−i k(k − 1) · · · (k − i + 1)xk−i eax , consistent with our hypothesis.
i
i=1

To complete our proof, we need to verify that the coefficient of xk−i eax is 0 is the differential
dn k ax
 
equation (for i ≤ k). Now the coefficient of xk−i eax in (x e ) is a n−i n k(k −
dxn i
1) · · · (k − i + 1)x k−i ax with k ≤ bj − 1
  e , so for a = zj (a root of the original  polynomial)

n n − 1
we need an zjn−i k(k − 1) · · · (k − i + 1) + an−1 zjn−1−i k(k − 1) · · · (k − i + 1) +
  i   i
1 0
· · · + a1 zj1−i k(k − 1) · · · (k − i + 1) + a0 zj−i k(k − 1) · · · (k − i + 1) = 0. (Of course
  i i
k
= 0 for k < l). Observe that we actually assumed that zj 6= 0; this limit case
l
can be established easily wherby the general solution contains term x for some , (so

3
 di 
the polynomial is divisible by x+1 and ai = 0 for i ≤ . But x = 0 for i >  so
dxi
di
ai i x = 0 for any i. Now with this assumption the above is same as proving that
dx
n  
p−i p
X
ap zj =0.
i
p=0
To prove this, observe that the root xj has multiplicity of more than k times, and i ≤
k. This means that (x − zj )k+1 divides the polynomial P , and it is well-known that
dα
P (x) is divisible by (x − zj )k+1−α for α ≤ k so substituting i into alpha yields that
dxα
dα i
k+1−i (and k + 1 − i ≥ 1). So d P (x) = 0 when
P (x) is divisible by (x − z j )
dxα dxi
x = zj . Also observe that the t-th derivative of x is u(u − 1) · · · (u − t + 1)xu−t , so
u
n n n
di di X p
X
p−i
X
0= P (x) = ap x = p(p − 1) · · · (p − i + 1)a p x = i!p(p − 1) · · · (p − i +
dxi dxi
p=0 p=0 p=0
n  
p−i
X p
1)ap x =i! ap xp−i (for x = zj ), as claimed.
i
p=0

3 Matrices
1. Row space and column space of any matrix have the same dimension.   
a11 a12
 a21   a22 
Proof: Let dimension of column space be n. W.L.O.G. let   . . . ,  . . . ,. . . ,
  

am1 am2
 
a1n
 a2n  Pn
 . . .  be pairwise linearly independent. Then for j > n, aij = k=1 λkj aik for some
 

amn
real numbers λkj aik . Multiplying each element in row i by a1i1 we have:

n
 
a12 a1n X a1k
 1 ... { λkj }
a11 a11 a11

 
k=1
n
 
 1 a22 a2n a2k
 X 
... { λkj }
a21 a21 a21
 
 
 k=1 
 .. 
 . 
n
 
 1 am2 amn amk 
 X
... { λkj } 
am1 am1 am1
k=1

where {} means that j runs from n + 1, n + 2, . . . , g where g is the number of columns of

this matrix. For convinience we name j as anything between n + 1 and g, inclusive. (We
assume that ai1 6= 0. If it is, then just swap the whole row with some ’lower’ rows with
nonzero first entry.)

4
 Now r.e.f. for the first element in each row yields
n
 
a12 a1n X a1k
 1 ... { λkj }
a11 a11 a11

 
k=1
n
 
 0 a22 − a12 a2n a1n a2k a1k
 X 
... − { λkj ( − )} 
a21 a11 a21 a11 a21 a11
 
 
 k=1 
 .. 
 . 
n
 
 0 am2 − a12 amn a1n amk a1k
 X 
... − { λkj ( − )} 
am1 a11 am1 a11 am1 a11
k=1

axy a1y
If a0xy = ax1 − a11 then the original matrix becomes
n
 
a12 a1n X a1k
 1 ... { λkj }
a11 a11 a11

 
k=1
n
 
 X 
 0 a0
 22 ... a02n { λkj a02k } 

 
 k=1 
 .. 
 . 
n
 
 X 
 0 a0
m2 ... a0mn { λkj a0mk } 
k=1

Replacing each a0xy with axy yields we can assume that the first element of eaach (except
the first) row is zero.
Repeating the r.e.f. process yields:
n
 
a12 a1(n−1) a1n X a1k
 1 ... { λkj }
a11 a11 a11 a11

 
k=1
n
 
 X 
 0 a22 . . . a2(n−1) a2n { λkj a2k }
 

 
 k=1 
 .. 
 . 
n
 
 X 
 0 0 ...
 0 ann { λkj ank } 

 k=1

..
 
.

But notice that the first n elements of j-th row (j > n) are all zero. It follows that
Xn
λkj aik (∀1 ≤ i ≤ i) must be zero. So all other elements must be zero and we have
k=1
columns onsisting entirely of zeroes after the n-th row.
This yields that, the dimension of row space cannot exceed n (some of the rows 1 to n
might be entirely zero, in which case we know that dimension of row space < n), i.e.
cannot exceed the dimension of column space. Similarly the dimension of column space
cannot exceed the dimension of row space. Hence they are equal.

2. Multplicative associativity of matrices: (AB)C = A(BC).

Proof: Let’s check the dimension of each matrix for the equation to be valid. Let the

5
 dimensions of A, B, C be a1 × a2 , b1 × b2 and c1 × c2 , respectively. For AB to be valid we
must have a2 = b1 and the resulting matrix has dimension a1 × b2 . For (AB)C to be valid
we need b2 = c1 and the matrix at LHS has dimension a1 × c2 . A similar check at RHS
yields a2 = b1 , b2 = c1 and resulting matrix with dimension a1 × c2 , hence the dimension
check is done.
Now let’s investigate every single entry x at i-th row and j-th column. Denote [ab]xy
as the element of i-th row and j-th column of matrix AB; define [bc]xy similarly. Now
bX
2 =c1 bX
2 =c1 aX
2 =b1 X
for (AB)C this entry is [ab]ik ckj = ( aid bdk )ckj = aid bdk ckj .
k=1 k=1 d=1 1≤d≤a2 ,1≤k≤b2
aX
2 =b1 aX
2 =b1 bX2 =c1 X
For A(BC) this entry is aid [bc]dj = aid ( bdk ckj ) = aid bdk ckj .
d=1 d=1 k=1 1≤d≤a2 ,1≤k≤b2
Hence the every corresponding entries of (AB)C and A(BC) are equal. Q.E.D.

3. Given
 T amatrix P with non-zero determinant, and r1 , r2 , r3 are 3×3 vectors such that P =
r1
 r2T  (where T stands for transpose). Then P −1 = 1

r2 × r3 r3 × r1 r1 × r2 .
det(P )
r3T
Proof: Let P −1 = s1 s2 s3 . Observe that ri · sj = 0 for i 6= j and 1 otherwise. This

means that si is a scalar multiple of ri+1 × ri+2 (indices taken modulo 3). (Why? si is
perpendicular to both ri+1 and ri+2 .) Now let constants ki be that si = ki (ri+1 × ri+2 ),
so 1 = ki · ri · (ri+1 × ri+2 ). However, it is well-known that r1 · (r2 × r3 ) = r2 · (r3 × r1 ) =
1
r3 · (r1 × r2 ) = det(P ) so k1 = k2 = k3 = det(P and we are done. To verify this for k1 , if
     )
a11 a21 a31
r1 =  a12  , r2 =  a22 , r3 =
  a32  then r1 ·(r2 ×r3 )=a11 a22 a33 −a11 a32 a23 +
a13 a23 a33
 
a11 a12 a13
a12 a23 a31 − a12 a21 a33 + a13 a21 a32 − a13 a32 a21 =det  a21 a22 a23  = det(P ).
a31 a32 a33
4. If an n×n matrix Q has n linearly independent eigenvectors r1 , r2 , · · · rn and correspond-
ing eigenvalues a1 , a2 , · · · , an , then Q = P DP −1 , where P = (r1 r2 . . . rn ) and D is the
diagonal matrix with ai at i-th row and i-th column.  T 
s1
 sT 
 2 
Proof: Denote, again, T as the transpose of vector. If P −1 =  .  then ri · sj =1 if
 .. 
sTn
i = j and 0 otherwise. Now let n × 1 matrix x be (ri ), thenP  −1 x has 0 for all elements
0
 0 
 
 .. 
 . 
−1
 
except the i-th entry, which is 1. This means DP x = D   1  has 0 for all elements

 0 
 
 .. 
 . 
0

6
  
0

 0 

 .. 

 . 

except the i-th entry, which is ai . Therefore (r1 r2 . . . rn ) 
 ai  = ai (ri ) = ai x. This


 0 

 .. 
 . 
0
means that P DP −1
has exactly the same eigenvectors (with corresponding eigenvalues)
as Q, and we show that such matrix is unique.
To this end, let’s show that P DP −1 −Q ≡ 0. Now we have Q(ri ) = P DP −1 (ri ) = ai (ri ), so
(P DP −1 −Q)(ri ) = 0 for i = 1, 2, · · · , n. But since r1 , r2 , · · · , rn are linearly independent,
we know that the dimension of null space of P DP −1 − Q is n, or the rank of this matrix
is 0. Hence P DP −1 − Q is a zero matrix. Q.E.D.

5. Let A be an m × n matrix. If there exists a matrix B such that AB and BA are identity
matrices, then m = n.
Proof: If AB and BA are both defined, then the dimension of B is n×m, and AB and BA
have dimensions m × m and n × n, respectively. If m 6= n, without the loss of generality
we can assume that m < n. We prove that the matrix BA has rank of at most m, which
forces its determinant to be zero (and therefore cannot be identity matrix).
 
c1i
 c2i 
Now let C = BA, and denote its entry in vector manner (v1 v2 · · · vn ). Then vi = . 
 
 .. 
cni
 
b11 a1i + b12 a2i + · · · + b1m ami
 b21 a1i + b22 a2i + · · · + b2m ami 
=  = (ai1 w1 + ai1 w2 + · · · + aim wm ) where wj rep-
 
..
 . 
bn1 a1i + bn2 a2i + · · · + bnm ami
 
b1j
 b2j 
resents the vector  . , or the j-th column of matrix B. Notice that every vector vi
 
 .  .
bnj
can be represented as x1 w1 + x2 w2 + · · · + xm wm , where x1 , x2 , · · · , xn ∈ R. This means
the set {v1 , v2 , · · · , vn } can only span in at most m dimensions, i.e. the dimension of
column space of this matrix BA is at most m.

6. det(A) · det(B) = det(AB), where A, B are n × n matrices.

Proof: Let C = AB, and denote aij as the entry of i-th row and j-th column of A.
Define bij and cij similarly. The term apq brs will appear as a term in C iff q = r, in
which case it will be contained in the entry cps . Moreover, to be consideredinto the
Yn n
Y X n
determinant of C, each term must be in the form of ciσ(i) =  aij bjσ(i) , where
i=1 i=1 j=1
{σ(i)|1 ≤ i ≤ n} is the permutation of {1, 2, · · · , n}. Individually speaking, the product
Y n
of sums above comprises terms in the form aiα(i) bα(i)σ(i) , where α(1), α(2), · · · α(n) is
i=1
a sequence of n numbers satisfying 1 ≤ α(j) ≤ n, ∀j ∈ [1, n]. Notice also that det(C) =

7
 n
!
X Y
N (γ)
(−1) ciγ(i) , where he sum is taken across all permutations γ and N (γ) is the
i=1
n
Y
number of inversions in γ. This means that the coefficient of aiα(i) bα(i)σ(i) in det(C)
i=1
is well-defined, i.e. the sums of coefficient of this term in all the terms (−1)N (γ) ciγ(i) . We
show that if {α(i)|1 ≤ i ≤ n} is not a permutation of {1, 2, · · · , n} (i.e. some of them are
n
Y
equal), then the product aiα(i) bα(i)σ(i) has total coefficient zero in det(C). And if it is a
i=1
permutation (meaning each term is different), then the coefficient is +1 if {σ(i)|1 ≤ i ≤ n}
is an even permutation, and -1 otherwise.
Now for each particular permutation σ, the terms in the (multi-)set {aij bjσ(i) |1 ≤ i, j ≤ n}
Yn
is pairwise distinct, which means the term aiα(i) bα(i)σ(i) has coefficient of exactly
 i=1 
Yn n
Y Xn
1 in the expansion of ciσ(i) =  aij bjσ(i) . (I.e. every term in the form
i=1 i=1 j=1
Y
n
)x = 1 axy byσ(x) (σ= permutation of x) can only appear at most once in this ex-
n
Y
pansion). Define the occurences of the term aiα(i) bα(i)σ(i) as the total number of the
i=1
n
Y
permutation σ0 such that this term is a term in the expansion of ciσ0 (i) . Then for each
i=1
k we name f (k) as the number of i ∈ [1, n] s.t. α(i) = k. We show that the occurences of
Yn n
Y
aiα(i) bα(i)σ(i) is ni=1 f (i)!. Indeed, this is equivalent to saying that
Q
aiα0 (i) bα0 (i)σ0 (i) =
i=1 i=1
n
Y n
Y n
Y
aiα(i) bα(i)σ(i) , or simply bα(i)σ0 (i) = bα(i)σ(i) as we can take α0 ≡ α for the pur-
i=1 i=1 i=1
pose of establishing this proof of our claim. Now define Sk as {i|1 ≤ i ≤ n, α(i) = k} (for
k = 1, 2, · · · , n) and what we need is {σ 0 (i)|i ∈ S(k)} = {σ(i)|i ∈ S(k)}, ∀k ∈ [1, n]. Now
σ 0 ≡ σ is definitely a solution, and for each such k we know that {σ(i)|i ∈ S(k)} has f (k)!
solutions. For each k, there are f (k)! possibilities to very σ, and multiplying this for all
such k yields the answer f (1)!f (2)! · · · f (n)!.

Qn finish this lemma, we 0need to prove that if f (k) ≥ 2 for some k, then among all
To
i=1 f (i)! permutations
Qn σ , exactly half of them is odd and half even. This means that
i=1 f (i)!
there must be occurences of aiα(i) bα(i)σ(i) that contibutes positively to the
2 Qn
f (i)!
determinant of C (i.e. have coefficient +1 in this determinant calculation) and i=1
2
that contributes negatively, which gives an overall coefficient of zero. We need the following
well-known lemma (not hard to prove, though) :
If integer x ≥ 2, swapping any two elements in a permutation of x distinct objects will
change the permutation from odd to even, and vice versa. Q
n
0 f (i)!
This allows us to pair up all feasible permutations σ into i=1 pairs such that ech
2
pair has an odd permutation and even permutation. Let j be an index in [1, n] such that
f (j) ≥ 2, and choose j1 and j2 such that α(j1 ) = α(j2 ) = j. Then for each permutation
σ 0 , pair it up with another permutation σ10 such that:

8
 • σ10 (i1 ) = σ 0 (i2 )
• σ10 (i2 ) = σ 0 (i1 )
• σ10 (i) = σ 0 (i), ∀ i ∈ [1, n]\{i1 , i2 }
Clearly, if σ 0 fulfills the claim that {σ 0 (i)|i ∈ S(k)} = {σ(i)|i ∈ S(k)}, ∀k ∈ [1, n] then
σ10 also fulfills the claim that {σ10 (i)|i ∈ S(k)} = {σ(i)|i ∈ S(k)}, ∀k ∈ [1, n]. Moreover,
each pair is disjoint (meaning no element can exist in two groups) since the other element
(i.e. permutation) in the same pair with a permutation is defined uniquely before. This
completes our bijection (and hence our claim).
Yn
If f (k) = 1 for all k, then there is only one such occurence of aiα(i) bα(i)σ(i) , or σ 0 ≡ σ
i=1
is the only possible permutation. The two previous terms have coefficient 1 in det(C) iff
Yn
ciσ(i) has a coefficient 1 in det(C), i.e. σ is an even permutation or N (σ) is even.
i=1
n
Y
Finally, let’s prove that for if α is a permutation, then the coefficient of aiα(i) bα(i)σ(i)
i=1
n
Y
in det(C) is the product of the coefficient of aiα(i) in det(A) and the coefficient of
i=1
n
Y
bα(i)σ(i) n det(B). Let I be the identity permutation, i.e. I(x) = x, ∀x ∈ [1, n]. De-
i=1
note also the permutation that brings α to σ as ω (means if α(i) = j and σ(i) = k,
then ω(j) = k. Try to think ω as σα−1 .) Now we know that if a permutation σ
is even, then I can be mapped to σ by swapping two neighbouring elements only by
an even number of times (and vice versa). In other words, if N denotes the mini-
mum number of swappings needed, then we can map I to σ by N (α) + N (ω) of neigh-
bouring swappings (first N (α) swapping from I to α, then N (ω) swappings from α to
σ). This entails N (σ) ≡ N (α) + N (ω) (mod 2), or (−1)N (σ) = (−1)N (α) · (−1)N (ω) .
Yn n
Y n
Y
Since bα(i)σ(i) = biω(i) , we have: coefficient of aiα(i) bα(i)σ(i) is (−1)N (σ) , coeffi-
i=1 i=1 i=1
n
Y n
Y
cient of aiα(i) is (−1)N (α) , and bα(i)σ(i) is (−1)N (ω) . This proves our claim, and
i=1 i=1
n
Y n
XY
det(C)= (−1)N (σ) ciσ(i) = (−1)N (σ) aiα(i) bα(i)σ(i) (σ any permutation, α any n-
i=1 i=1
n
XY
N (σ)
tuple) = (−1)
aiα(i) bα(i)σ(i) (same definition for σ but restrict α to permuta-
i=1 X
tions, since the other case has coefficient zero and can be disregarded)= (−1)N (α) aiα(i)
X X X
· (−1)N (ω) biω(i) = (−1)N (α) aiα(i) · (−1)N (ω) bα(i)σ(i) = det(A) det(B), as desired.

4 Centroid of bodies.
1. Solid hemisphere with radius r.
Answer: 83 r from the centre, and lying on the line through centre and perpendicular to
the plane of the great circle.
Proof:
√ Let the radius of the hemisphere be r, then the equation of this hemisphere is just
y = r2 − x2 from 0 to r having rotated 360◦ . We know that ȳ = 0. To find x̄ simply

9
 Rr Rr 2 2 4 4
π 0 xy 2 dx 0 x(r2 − x2 ) dx [ x 2r − x4 ]r0 r4
take the fraction Rr = Rr 2 = x3 r
= 2r3 = 38 r.
π 0 y 2 dx 0 r − x 2 dx 2
[xr − 3 ]0 3

2. Hemispherical shell with radius r.

Answer: 12 r from the centre, and lying on the line through centre and perpendicular to
the plane of the great circle. s  2
Rr dy
2π 0 xy 1 + dx
dx
Proof: same equation as above for x and y, but now we need x̄= s  2 .
Rr dy
2π 0 y 1 + dx
dx
s  2
d 2 2 0.5 −x dy p
Observe that ((r −x ) )= 2 2 0.5
so y 1 + = (r2 −x2 )0.5 · r2 (r2 − x2 )−1 =
dx (r − x ) dx
x2 r r3
Rr
xr dx [r( 2 )]0 1
r. The original integral then becomes R0 r = r = 2
2
= r.
0 r dx
[r(x)]0 r 2
3. Circular sector with radius r and subtended angle 2α.
2r sin α
Answer: from the centre of the circle (i.e. sector) and lying on the angle bisector
3α
of the sector.
Proof: Same equation as R rabove, for x from r cos 2α to Rr, and y√= tan 2α for x from
R r cos 2α 2 r
xy dx 0 x tan 2α dx + r cos 2α x r2 − x2 dx
0 to r cos 2α. Now x̄= R0 r = R r cos 2α Rr √ . Now to
0 y dx 0 x tan 2α dx + r cos 2α r2 − x2 dx
deal with the second integral R r in numerator
√ we needR 0 the substitution
√ x = r cos θ, and
we have dx = −r sin θdθ. r cos 2α x r − x dx = 2α −r cos θ r − r cos2 θr sin θdθ =
2 2 2 2
2α
sin3 θ r3 sin3 2α

R 2α 3 2 3
0 r cos θ sin θdθ = r = . (We know from the area of sector that
3 0 3
 3 r cos 2α
2
R r cos 2α 2 x tan 2α (r cos 2α)3 sin 2α
the denominator is r α.) 0 x tan 2α dx = = =
3 0 3 cos 2α
r3 sin 2α cos2 2α r3 sin 2α r3 sin 2α
. Adding the two terms up yield (cos2 2α + sin2 2α) = .
3 3 3
r3 sin 2α r sin 2α
This yields x̄ = = .
3r2 α 3α
It is obvious that the centre of mass of this sector lies on the angle bisector of the sector
due to symmetry, hence have gradient cos α rom centre O (the origin). The length from
r sin 2α 2r sin α
the origin to the centre of mass has therefore length = .
3α cos α 3α
4. Circular arc with radius r and subtended angle 2α.
r sin α
Answer: from the centre of the circle (i.e. sector) and lying on the angle bisector
α
of the sector. √
Proof: The equation is given by yR = r2 − x2 , x from rcos 2α to r. As shown above,
r
√ xr
r cos 2α r2 −x2 dx
r  2
dy
1 + dx = √r2r−x2 . Here, x̄ = R r . The denominator is simply the arc
√ r
r cos 2α r2 −x2 dx
√ √
length 2rα, while the numerator is equal to [−r r2 − x2 ]rr cos 2α = r r2 − r2 cos2 2α =
r2 sin 2α
r2 sin 2α. Therefore x̄ = . Again, divide it by cos α to find he distance from the
2rα
2
r sin 2α r sin α
centre and we have = .
2rα cos α α

10
 5. Triangular lamina ABC.
Answer: Let the midpoint of BC to be M , then the centre of mass is the point G on
segment AM , satisfying AG : GM = 2 : 1.
Proof: We prove that this centroid must lie on all medians of a triangle by establishing
that the magnitude of moment with respect to AM at both sides of AM must be the
same. To do this, we cut the triangle into very tiny strips, each one parallel to AM , and
prove that for each strip, we can find another strip on the other side of the median that
has the same length (or same mass) and with the same distance from the line AM .
We need to prove first that the number of strips on both sides must the same, given the
each has width d (infinisimal). Now, the product of d and the total number of strips
on side BM is the perpendicular distance from B to AM , which is BM sin ∠AM B; the
product of d and the total number of strips on side CM is the perpendicular distance from
C to AM , which is CM sin ∠AM C. However, BM = CM (M is the midpoint of BC)
and sin ∠AM B = sin ∠AM C (these angles are supplementary) so this claim is proven.
Next, consider a point P on side BM , and Q a point on side AM such that P Q is parallel
to AM . Next, reflect P across M to get P 0 , and let Q0 be a point on AC such that
PQ BP CP 0 P 0 Q0
P 0 Q0 k AM . Now = = = so P Q=P 0 Q0 . (The first equivalence
AM BM CM AM
is because triangles BQP and BAM are similar; the second equivalence follows when
BP = CP 0 by the definition of reflection and BM = CM ; the third equivalence is the
consequence of the fact that triangles CAM and CQ0 P 0 are similar). Finally, the distance
between strip P Q and line AM is P M sin ∠BAM and distance between strip P 0 Q0 and
line AM is P M sin ∠CAM . It is easy to verify that the are the same.
The similar proof above can be used to verify that the other medians contain the centre
of mass of triangle ABC as well.

5 Moment of inertia.
1. Thin rod of mass m length 2r about the perpendicular axis through the midpont.
mr2
Answer: .
3 Rr 2 3 2r3
−r x dx [ x3 ]r−r 3 mr2
Proof: Moment of inertia=m · R r =m· = m · = .
−r 1 dx
[x]r−r 2r 3

2. Rectangular lamina of mass m and dimensions 2a × 2b about the perpendicular axis

through the centre of mass.
1
Answer: m(a2 + b2 ).
3
Proof: Consider al the particles
Xon the rectangular lamina (with mass dm), so
Xthe mass of
the particles is actually dm and the moment of inertia is (x2 +
−a≤x≤a,−b≤y≤b −a≤x≤a,−b≤y≤b
2
y )dm. Turning infinitely many particles into double integration (because there are two
Ra Rb Ra Rb
dimensions!) we have −a ( −b 1 dy) dx for mass, and −a ( −b x2 + y 2 dy) dx for moment
Rb Ra Rb Ra
of inertia. Now ( −b 1 dy)=[y]b−b = 2b so −a ( −b 1 dy) dx = −a 2b dx = [2bx]a−a = 4ab.
Rb 3 3 Ra 3 2x3 b + 2xb3 a
( −b x2 + y 2 dy)=[x2 y + y3 ]b−b = 2x2 b + 2b3 so −a 2x2 b + 2b3 dx = [ ]−a =
3
4a3 b + 4ab3 a2 + b2
 
=4ab . Dividing it by 4ab and multiplying by m yields the moment
3 3

11
 a2 + b2
 
of inertia as m .
3
3. Disc/solid cylinder with mass m and radius r about the axis pependicular to the circular
plane and pasing through its centre.
mr2
Answer: .
2
Proof: Let the height of solid cylinder be h (in the event of a circular lamina we can
treat this h as infinisimal (which doesn’t affect our answer anyway). Now consider the
sub-ring (or sub-cylinderic shell) with centre the axis
R r and3 radius Rx,r and the surface area is
3 dx
2πx h dx x [x4 /4]r r2
definitely 2πxh. The desired ratio now becomes R0r = R0r = 2 0r = ,
0 2πxh dx 0 x dx
[x /2]0 2
which leads to our answer.

4. Solid sphere of mass m and radius r about an axis passing through its centre.
2mr2
Answer: .
5
Proof: Treat this sphere as x2 + y 2 + z 2 ≤ r2 about the z-axis, and consider the lo-
2 2 2 2
cus of points of √distance d from √ this z-axis. Now √ d = x + y , and d + z ≤ r .
2 2 2 2 2
Therefore |z| ≤ r − d or − r − d ≤ z ≤ r − d√ . The locus is therefore the 2
2 2
curved√face of cylindrical √ shell with radius d and height 2 r − d , and the surace area is
2 2
2πd(2 r − d ) = 4πd r − d . 2 2
Rr 2 √ Rr √
0 (x )4πx r2 − x2 dx 0 x3 r2 − x2 dx
The desired fraction is now R r √ = Rr √ . Now for both de-
4πx r 2 − x2 dx x r 2 − x2 dx
0 0
nominator and numerator, we use the substitution x = r sin θ to get dx = r cos dθ, so that
π
Rr √ − cos3 θ 2 r3

R π2 R π2 3 2 3
2 2
0 x r − x dx= 0 (r sin θ)(r cos θ)(r cos θ dθ)= 0 r sin θ cos θ dθ=r 3
= .
3
0
R r 3√ R π2 2
R π2 5 3 2
Similarly, 0 x r2 − x2 dx= 0 (r sin θ) (r sin θ)(r cos θ)(r cos θ dθ) = 0 r sin θ cos θ dθ
π
− cos3 θ 2

R π2 5 2 2
R π2 5 2
R π2 5 4 5
= 0 r sin θ(1−cos θ) cos θ dθ = 0 r sin θ cos θ dθ - 0 r sin θ cos θ dθ = r
3 0
 5
 π2 5 5 5
− cos θ r r 2r
- r5 = − = . Summarizing above yields the desired fraction as
5 0 5 3 15
2r5 r3 2r2
÷ = , whch yields the answer.
15 3 5
5. Spherical shell of mass m and radius r about an axis passing through its centre.
2mr2
Answer: .
3 √
Proof: Treat this shell as the equation y = r2 − x2 , rotated through 360◦ the x-axis.
(x rnging from −r to r). Assume, too, that the x-axis is our axis of reference.√Now the
distance of each point from this axis is its y- coordinate,sso for each y, x = ± r2 − x2 .
 2
dy
For each of these two x, a circle with circumference 2πy 1 + is generated upon
dx
reolution (recall how we found surface
s area of revolution of a graph!) As always, 1 +
 2  2
dy r 2 dy √ r
= 2 2
. This means 2πy 1 + =2π r2 − x2 √ =2πr. This means
dx r −x dx r 2 − x2

12
 s
dy 2


Rr
−r(y 2 )2πy1+ dx Rr 2 2
dx −r 2πr(r − x ) dx
that our desired fraction is now s = R r =
−r 2πr dx
 2
Rr dy
−r 2πy 1 + dx dx
3
[2πr3 x − 2rx r
3 ]−r 4πr4 − 34 r4 2r2 2mr2
= = so the answer is .
[2πrx]r−r 4πr2 3 3
6. Perpendicular axis theorem. (applies only to a lamina)
Proof: We assume that the lamina is on the x − y plane, and LetR fR(x, y) be the density
at point x, y. Now the mass of this lamina is (ignoring limits) ( f (x, y) dx) dy. For
each particle at point (x, y), thepdistance from x-axis is y and the distance from y-axis is
x. Its distance from z-axis is 2 2
R R 2 2
R R x 2 + y (i.e. distance
R 2 from the origin).
R R 2Therefore Iz =
R (R (x2 +y )f (x, y) dx) dy = ( x f (x, y) dx+ y f (x, y) dx) dy = ( x f (x, y) dx)dy +
( y f (x, y) dx) dy=Iy + Ix , as desired.
7. Parallel axis theorem. (applies to a solid and a lamina) R R R
Proof: Let the density at point (x, y, z) be f (x, y, z), and its mass is m = ( ( f (x, y, z) dx) dy) dz
(again, dropping R R limits).
R Assume, W.L.O.G., R that
R R the centre of mass lies at R Rpoint
R (0,0,0),
meaning that ( ( xf (x, y, z) dx) dy) dz = ( ( yf (x, y, z) dx) dy) dz = ( ( zf (x, y, z) dx) dy) dz
= 0. Now let the axis through the centre of mass to be the z-axis, p and the other axis
to be x = a, y = b. The distance p of solid from the first axis is x2 + y 2 and dis-
2 2
tance from the second R R Raxis is (x − a) + 2(y − b) . Now the moment R R R of 2inertia about
the second axis is ( ( ((x − a)2 + (y − b) )f (x, y, z) dx) dy) dz= ( ( (x + y 2 + a2 +

b2 − 2ax − 2by)f (x, y, z) Rdx)R dy)

R R R 2 2 )f (x, y, z) dx) dy) dz+ ( ( (a2 +
R R R
dz = ( ( (x + y
b2 )f (x, y, z) dx) dy)RdzR -R2 ( ( axf (x, y, z) dx) dy)
R R R R
dz - 2 ( ( byf (x, y, z) dx) dy) dz =
2 2
R R R
Iz−axis
R R+ R(a + b ) ( ( f (x, y, z) dx) dy) dz - 2a ( ( xf (x, y, z) dx) dy) dz
- 2b ( ( yf (x, y, z) dx) dy) dz = Iz−axis + (a + b2 )m - 0, i.e. the moment of inertia
2

about the first axis + mr2 where r is the distance of the second axis from the origin.

6 Regression lines
1. Given n points (xi , yi ), i = 1, 2, · · · , n. Then the equation of line of regression of y on x
Xn
(xi − x̄)(yi − ȳ)
i=1
is given by y − ȳ = n (x − x̄).
X
(xi − x̄)2
i=1
Proof: First, we prove that for a set of lines with common gradient, the sum of square of
vertical deviation of each point from the line is minimum when the line passes through
the centre of mass of the points.
Suppose that the gradient is k, and let θ = tan−1 k be the angle (anticlockwise) of the line
with the x-axis. Now perform a rotation of everything (the n points and the line) of angle
θ clockwise about any point, say, the origin. Then the line now becomes horizontal, the
initial vertical deviation of each point to the line becomes the ”θ -degree to the vertical”
deviation to the line (which is, | cos1 θ | times the new vertical deviation (or perpendicular
distance) to the line: multipliation by a constant across all points). Moreover, the centre of
mass of the points also follows the same
 rotation. (It makes
  prefect
 sense intuitively, but to
cos α − sin α x
rigorize the argument, notice that rotates the point (x, y) an-
sin α cos α y

13
gle α anticlockwise. If α = −θ then the rotation of points (xi , yi ), i = 1, 2, · · · , n have cen-
n      n  
1 X cos α − sin α xi 1 cos α − sin α X xi
tre of mass = , which
n sin α cos α yi n sin α cos α yi
i=1 i=1
is the original centre of mass having rotated angle α).
Now let (x0i , yi0 ) be our new points, and since the line is now horizontal we can assume
that its equation is y = w, w constant. Now the vertical distance of each point to the line
Xn Xn n
X
0 0
0
is |yi − w| so the sum of square of distance is 2 2
(yi − w) =nw − 2w yi + (yi0 )2 =
i=1 i=1 i
 n 2 n
!2
X X
 yi0  yi0 n
i=1 i=1
X
(yi0 )2 . Now the ”θ degree to the normal” distance
 
w −
n  − +
n  n
  i

can be obtained by dividing each term by | cos θ|, which is a costant across each term!
Xn
yi0
i=1
Therefore, the minimum distance can be achieved when w = n , i.e. passing through
the centre of mass.
To find the gradient of this line, notice that the equation of the line must be in the form
y − ȳ = b(x− x̄). The vertical deviation from each point to the line is thus yi −b(xi − x̄)− ȳ.
Xn Xn
2
Therefore the sum of square of vertical deviation is [(yi − ȳ) − b(xi − x̄)] = b2 (xi −
i=1 i=1
n
X n
X
x̄)2 − 2b (xi − x̄)(yi − ȳ) + (yi − ȳ)2 . Again, by completing the square we know that
i=1 i=1
n
X
(xi − x̄)(yi − ȳ)
i=1
the least square sum is achieved when b = n . Notice, also, that the
X
(xi − x̄)2
i=1
n
X n
X
n
xi yi
i=1 i=1
X
numerator is equivalent to xi yi − and that the denominator is equivalent
n
i=1
n
!2
X
n
xi
i=1
X
to (x2i ) − .
n
i=1

14