ISI MMath Solutions and Discussion 2017 : PMB

ISI MMath PMB 2017 Subjective Questions along with hints and solutions and discussions

Note: You must try the problems on your own before looking at the solutions. I have made the solutions so that you don’t have to join any paid program to get the solutions. After you have failed several times in your attempt to solve a question even with the hints provided, then only look at the solution to that problem. If you cannot solve a single question on your own then this exam is not for you. And if you like my work then please do share this among your friends and do comment below.  

To view ISI MMATH 2020 solutions, hints, and discussions: Click Here

To view ISI MMATH 2019 solutions, hints, and discussions: Click Here

To view ISI MMATH 2018 solutions, hints, and discussions: Click Here

To view ISI MMATH 2016 solutions, hints, and discussions: Click Here

To view previous year’s question papers: Click Here

Group A

Problem 1.

Let a_n\in\mathbb{R}, such that \sum^\infty_{n=1}|a_n|=\infty and \sum^m_{n=1}a_n\to a\in\mathbb{R} as m\to\infty. Let a^+_n=max \{a_n,0\}. Show that \sum^{\infty}_{n=1}a^+_n=\infty.

Topic: Real Analysis, Difficulty level: Medium

Indian Statistical Institute, ISI MMath 2017 Problem 1 Hints along with Solution & Discussion

Hint 1

Use the representation

    \[a^+_n=\frac{a_n+|a_n|}{2}.\]

Fix any C>0. There exists m_0 such that for all m\geq m_0

    \[\left|\sum^m_{n=1}a_n-a\right|<1\]

Hint 2

    \[\sum^m_{n=1}|a_n|>2C+|a|+1.\]

Hint 3

Then for all m\geq m_0 we have

    \[\sum^m_{n=1}a^+_n=\frac{1}{2}\sum^m_{n=1}a_n+\frac{1}{2}\sum^m_{n=1}|a_n|\]

Full Solution

We use the representation

    \[a^+_n=\frac{a_n+|a_n|}{2}.\]

Fix any C>0. There exists m_0 such that for all m\geq m_0

    \[\left|\sum^m_{n=1}a_n-a\right|<1, \mbox{ and }\]

    \[\sum^m_{n=1}|a_n|>2C+|a|+1.\]

Then for all m\geq m_0 we have

    \[\sum^m_{n=1}a^+_n=\frac{1}{2}\sum^m_{n=1}a_n+\frac{1}{2}\sum^m_{n=1}|a_n|=\]

    \[=\frac{1}{2}a+\frac{1}{2}\left(\sum^m_{n=1}a_n-a\right)+\frac{1}{2}\sum^m_{n=1}|a_n|\geq\]

    \[\geq \frac{1}{2}\sum^m_{n=1}|a_n|-\frac{1}{2}|a|-\frac{1}{2}\left|\sum^m_{n=1}a_n-a\right|>\]

    \[>\frac{1}{2}(2C+|a|+1)-\frac{1}{2}|a|-\frac{1}{2}=C.\]

For any C>0 there exists m_0 such that for all m\geq m_0

    \[\sum^m_{n=1}a^+_n>C.\]

In other words, \sum^\infty_{n=1}a^+_n=\infty.

 

 

Problem 2.

Let E= \{(x,y,z)\in\mathbb{R}^3|x,y,z>0,xy+yz+zx=1\}. Prove that there exists (a,b,c)\in E such that abc\geq xyz, for all (x,y,z)\in E.

Topic: Real Analysis, Difficulty Level: Medium

Indian Statistical Institute, ISI MMath 2017 Problem 2 Hints along with Solution & Discussion

Hint 1

Apply the arithmetic mean-geometric mean inequality:

 

for all n\geq 1 and a_1,\ldots,a_n\geq 0

    \[(a_1\ldots a_n)^{1/n}\leq \frac{a_1+\ldots+a_n}{n}\]

Hint 2

Let n=3, a_1=xy, a_2=yz, a_3=xz, where (x,y,z)\in E. Then

    \[(a_1a_2a_3)^{1/3}=(xyz)^{2/3}\leq \frac{a_1+a_2+a_3}{3}=\frac{xy+yz+xz}{3}=\frac{1}{3}.\]

Hint 3

Take

    \[(a,b,c)=\left(\frac{1}{\sqrt{3}},\frac{1}{\sqrt{3}},\frac{1}{\sqrt{3}}\right)\in E.\]

Full Solution

We apply the arithmetic mean-geometric mean inequality:

 

for all n\geq 1 and a_1,\ldots,a_n\geq 0

    \[(a_1\ldots a_n)^{1/n}\leq \frac{a_1+\ldots+a_n}{n}\]

Let n=3, a_1=xy, a_2=yz, a_3=xz, where (x,y,z)\in E. Then

    \[(a_1a_2a_3)^{1/3}=(xyz)^{2/3}\leq \frac{a_1+a_2+a_3}{3}=\frac{xy+yz+xz}{3}=\frac{1}{3}.\]

So, for any (x,y,z)\in E we have

    \[xyz\leq \frac{1}{3^{3/2}}.\]

Take

    \[(a,b,c)=\left(\frac{1}{\sqrt{3}},\frac{1}{\sqrt{3}},\frac{1}{\sqrt{3}}\right)\in E.\]

Then

    \[abc=\frac{1}{3^{3/2}}\geq xyz\]

for any (x,y,z)\in E. The answer is

    \[(a,b,c)=\left(\frac{1}{\sqrt{3}},\frac{1}{\sqrt{3}},\frac{1}{\sqrt{3}}\right).\]

 

Problem 3.

Let f:\mathbb{R\to\mathbb{R}} be an increasing function. Suppose there are sequences (x_n) and (y_n) such that x_n<0<y_n for all n\geq 1 and f(y_n)-f(x_n)\to 0 as n\to\infty. Prove that f is continuous at 0.

Topic: Real Analysis, Difficulty Level: Medium

Indian Statistical Institute, ISI MMath 2017 Problem 3 Hints along with Solution & Discussion

Hint 1

Try to form the mathematical expression with \epsilon >0 and \epsilon \in \mathbb{R} from this given statement:
“Suppose there are sequences (x_n) and (y_n) such that x_n<0<y_n for all n\geq 1 and f(y_n)-f(x_n)\to 0 as n\to\infty“.

Hint 2

Let \epsilon>0. There exists n_0 such that for all n\geq n_0

    \[f(y_n)-f(x_n)<\epsilon.\]

Hint 3

Fix such n\geq n_0 and take \delta=\min(y_n,-x_n)>0. Consider arbitrary x\in (-\delta,\delta), x\ne 0. If x>0, then x_n<0<x<\delta\leq y_n,

    \[f(x)\leq f(y_n), f(0)\geq f(x_n)\]

and

    \[|f(x)-f(0)|=f(x)-f(0)\leq f(y_n)-f(x_n)<\epsilon.\]

If x<0, then x_n\leq -\delta<x<0<y_n,

    \[f(x)\geq f(x_n), f(0)\leq f(y_n)\]

Full Solution

Let \epsilon>0. There exists n_0 such that for all n\geq n_0

    \[f(y_n)-f(x_n)<\epsilon.\]

Fix such n\geq n_0 and take \delta=\min(y_n,-x_n)>0. Consider arbitrary x\in (-\delta,\delta), x\ne 0. If x>0, then x_n<0<x<\delta\leq y_n,

    \[f(x)\leq f(y_n), f(0)\geq f(x_n)\]

and

    \[|f(x)-f(0)|=f(x)-f(0)\leq f(y_n)-f(x_n)<\epsilon.\]

If x<0, then x_n\leq -\delta<x<0<y_n,

    \[f(x)\geq f(x_n), f(0)\leq f(y_n)\]

and

    \[|f(x)-f(0)|=f(0)-f(x)\leq f(y_n)-f(x_n)<\epsilon.\]

In any case,

    \[|f(x)-f(0)|<\epsilon\]

as soon as |x|<\delta. f is continuous at 0.

Problem 4.

Do there exist continuous functions P and Q on [0,1] such that y(t)=\sin(t^2) is a solution to y''+Py'+Qy=0 on [\frac{1}{n},1] for all n\geq 1 ? justify your answer.

Topic: Elements of Ordinary Differential Equations, Difficulty Level: Medium

Indian Statistical Institute, ISI MMath 2017 Problem 4 Hints along with Solution & Discussion

Hint 1

Assume that such functions exist and then proceed.

Hint 2

Derivatives of y(t)=\sin(t^2) are equal to

    \[y'(t)=2t\cos(t^2), \ y''(t)=2\cos(t^2)-4t^2\sin(t^2).\]

Hint 3

In particular,

    \[\lim_{t\to 0}y(t)=0, \lim_{t\to 0}y'(t)=0, \lim_{t\to 0}y''(t)=2.\]

Full Solution

Assume that such functions exist. Derivatives of y(t)=\sin(t^2) are equal to

    \[y'(t)=2t\cos(t^2), \ y''(t)=2\cos(t^2)-4t^2\sin(t^2).\]

In particular,

    \[\lim_{t\to 0}y(t)=0, \lim_{t\to 0}y'(t)=0, \lim_{t\to 0}y''(t)=2.\]

If for all t\in (0,1] the equality

    \[y''(t)+P(t)y'(t)+Q(t)y(t)=0,\]

then taking the limit as t\to 0 we get

    \[2+P(0)\cdot 0+Q(0)\cdot 0=0,\]

which is impossible. So, such functions P and Q do not exist.

Problem 5.

Let f:\mathbb{R}\to\mathbb{R} be defined by

    \[f(x)=\int^{1+e^{x^3+x}}_{e^{x^3+x}} e^{r^2} dr\]

for all x\in\mathbb{R}. Prove that f is monotone.

Topic: Real Analysis, Difficulty Level: Easy

Indian Statistical Institute, ISI MMath 2017 Problem 5 Hints along with Solution & Discussion

Hint 1

Observe the relation

    \[f(x)=h(g(x)),\]

where g(x)=e^{x^3+x}, h(y)=\int^{1+y}_ye^{r^2}dr.

Hint 2

We have

    \[g'(x)=e^{x^3+x}(3x^2+1)>0\]

Hint 3

    \[h'(y)=e^{(1+y)^2}-e^{y^2}=e^{y^2}(e^{1+2y}-1).\]

Full Solution

Observe the relation

    \[f(x)=h(g(x)),\]

where g(x)=e^{x^3+x}, h(y)=\int^{1+y}_ye^{r^2}dr. We have

    \[g'(x)=e^{x^3+x}(3x^2+1)>0\]

    \[h'(y)=e^{(1+y)^2}-e^{y^2}=e^{y^2}(e^{1+2y}-1).\]

In particular, h'(y)>0 for y>-1/2. Since g(x)>0, we have h'(g(x))>0, g'(x)>0 and

    \[f'(x)=h'(g(x))g'(x)>0.\]

The function f(x) is strictly increasing.

 

Problem 6.

Let w=\{w(i,j)\}_{1\leq i,j\leq m} be an m\times m symmetric matrix with non-negative real entries such that w(i,j)=0 if and only if i=j. Show that d(i,j)=\min \{\sum^{k-1}_{j=0}w(i_j,i_{j+1})|k\geq 1,i_0=i,i_k=j,i_j\in \{1,\dots,m\}\} is a metric on \{1,\dots,m\}.

Topic: Metric Space, Difficulty Level: High

Indian Statistical Institute, ISI MMath 2017 Problem 6 Hints along with Solution & Discussion

Hint 1

At first we verify that the minimum is achieved. Consider a path i_0,i_1,i_2,\ldots,i_k between i and j of length k+1\geq m+1. Among elements

    \[i_0,i_1,i_2,\ldots,i_k\]

there are at least two identical elements: i_r=i_s for some r<s. Then

    \[i_0,i_1,\ldots,i_{r-1},i_s,\ldots,i_k\]

is a path of length <k+1. Also, w(i_{r-1},i_r)=w(i_{r-1},i_s) and

    \[\sum^{k-1}_{l=0}w(i_l,i_{l+1})\geq \sum^{r-2}_{l=0}w(i_l,i_{l+1})+w(i_{r-1},i_s)+\sum^{k-1}_{l=s}w(i_l,i_{l+1}).\]

It means that it is enough to consider only paths of length \leq m. There are only finitely many such paths, hence the minimum in the definition of d(i,j) is achieved. We check axioms of a metric.

 

Hint 2

For any path i_0=i,i_1,\ldots,i_{k-1},i_k=j between i and j we have

    \[\sum^{k-1}_{l=0}w(i_l,i_{l+1})\geq 0.\]

Hence, d(i,j)\geq 0. Assume d(i,j)=0. Since the minimum in the definition of d(i,j) is achieved, we have

    \[\sum^{k-1}_{l=0}w(i_l,i_{l+1})= 0 \eqno(*)\]

for some path i_0=i,i_1,\ldots,i_{k-1},i_k=j between i and j. Since all terms in the left-hand side of (*) are non-negative, we deduce that

    \[w(i_l,i_{l+1})=0, \ l=0,\ldots,k-1.\]

By definition of w, it follows that

    \[i_0=i_1=\ldots=i_{k-1}=i_k.\]

In particular, i=i_0=i_k=j. We’ve proved that d(i,j)\geq 0 and d(i,j)=0\Rightarrow i=j.

Hint 3

Consider a path i_0=i,i_1=i. Then d(i,i)\leq w(i,i)=0. But d(i,i)\geq 0, hence

    \[d(i,i)=0.\]

This proves that d(i,j)=0\Leftrightarrow i=j.

Full Solution

At first we verify that the minimum is achieved. Consider a path i_0,i_1,i_2,\ldots,i_k between i and j of length k+1\geq m+1. Among elements

    \[i_0,i_1,i_2,\ldots,i_k\]

there are at least two identical elements: i_r=i_s for some r<s. Then

    \[i_0,i_1,\ldots,i_{r-1},i_s,\ldots,i_k\]

is a path of length <k+1. Also, w(i_{r-1},i_r)=w(i_{r-1},i_s) and

    \[\sum^{k-1}_{l=0}w(i_l,i_{l+1})\geq \sum^{r-2}_{l=0}w(i_l,i_{l+1})+w(i_{r-1},i_s)+\sum^{k-1}_{l=s}w(i_l,i_{l+1}).\]

It means that it is enough to consider only paths of length \leq m. There are only finitely many such paths, hence the minimum in the definition of d(i,j) is achieved. We check axioms of a metric.

 

  • For any path i_0=i,i_1,\ldots,i_{k-1},i_k=j between i and j we have

    \[\sum^{k-1}_{l=0}w(i_l,i_{l+1})\geq 0.\]

Hence, d(i,j)\geq 0. Assume d(i,j)=0. Since the minimum in the definition of d(i,j) is achieved, we have

    \[\sum^{k-1}_{l=0}w(i_l,i_{l+1})= 0 \eqno(*)\]

for some path i_0=i,i_1,\ldots,i_{k-1},i_k=j between i and j. Since all terms in the left-hand side of (*) are non-negative, we deduce that

    \[w(i_l,i_{l+1})=0, \ l=0,\ldots,k-1.\]

By definition of w, it follows that

    \[i_0=i_1=\ldots=i_{k-1}=i_k.\]

In particular, i=i_0=i_k=j. We’ve proved that d(i,j)\geq 0 and d(i,j)=0\Rightarrow i=j.

  • Consider a path i_0=i,i_1=i. Then d(i,i)\leq w(i,i)=0. But d(i,i)\geq 0, hence

    \[d(i,i)=0.\]

This proves that d(i,j)=0\Leftrightarrow i=j.

  • Consider a path i_0=i,i_1,\ldots,i_{k-1},i_k=j between i and j, such that

    \[d(i,j)=\sum^{k-1}_{l=0}w(i_l,i_{l+1}).\]

Observe that i_k=j,i_{k-1},\ldots,i_1,i_0=i is a path between j and i. So,

    \[d(j,i)\leq \sum^{k-1}_{l=0}w(i_{k-l},i_{k-l+1})=\sum^{k-1}_{l=0}w(i_{l+1},i_{l})=\sum^{k-1}_{l=0}w(i_{l},i_{l+1})=d(i,j).\]

We’ve proved that

    \[d(j,i)\leq d(i,j).\]

Since this is true for all i,j, we also get

    \[d(i,j)\leq d(j,i).\]

Finally, d(i,j)=d(j,i).

  • Consider two paths i_0=i,i_1,\ldots,i_k=j, r_0=j,r_1,\ldots,r_s=r, such that

    \[d(i,j)=\sum^{k-1}_{l=0}w(i_l,i_{l+1}), d(j,r)=\sum^{s-1}_{p=0}w(r_p,r_{p+1}).\]

Observe that i_0,i_1,\ldots,i_k,r_1,\ldots,r_s is a path between i and r. Hence

    \[d(i,r)\leq \sum^{k-1}_{l=0}w(i_j,i_{j+1})+w(i_k,r_1)+\sum^{s-1}_{p=1}w(r_p,r_{p+1})=\]

use that w(i_k,r_1)=w(j,r_1)=w(r_0,r_1)

    \[=\sum^{k-1}_{l=0}w(i_j,i_{j+1})+\sum^{s-1}_{p=0}w(r_p,r_{p+1})=d(i,j)+d(j,r).\]

The triangle inequality is proved.

d is a metric on \{1,\ldots,m\}.

Group B

Problem 7.

Factory A produces 1 bad watch in 100 and factory B produces 1 bad watch in 200. You are given two watches from one of the factories and you don’t know which one.

  1. What is the probability that the second watch works?
  2. Given that the first watch works, what is the probability that the second watch works ?

Topic: Probability Theory, Difficulty Level: Medium

Indian Statistical Institute, ISI MMath 2017 Problem 7 Hints along with Solution & Discussion

Hint 1

Introduce two hypothesis

    \[H_1=\{\mbox{ the watches are from the Factory } A\},\]

    \[H_2=\{\mbox{ the watches are from the Factory } B\}.\]

Hint 2

Then P(H_1)=P(H_2)=\frac{1}{2}. Consider events

    \[E_1=\{\mbox{ the first watch works }\}\]

    \[E_2=\{\mbox{ the second watch works }\}.\]

It is given that P(E_j|H_1)=0.99, P(E_j|H_2)=0.995.

Hint 3

By the law of total probability

    \[P(E_2)=P(E_2|H_1)P(H_1)+P(E_2|H_2)P(H_2)=\frac{0.99+0.995}{2}=0.9925\]

Full Solution

Introduce two hypothesis

    \[H_1=\{\mbox{ the watches are from the Factory } A\},\]

    \[H_2=\{\mbox{ the watches are from the Factory } B\}.\]

Then P(H_1)=P(H_2)=\frac{1}{2}. Consider events

    \[E_1=\{\mbox{ the first watch works }\}\]

    \[E_2=\{\mbox{ the second watch works }\}.\]

It is given that P(E_j|H_1)=0.99, P(E_j|H_2)=0.995.

  • By the law of total probability

    \[P(E_2)=P(E_2|H_1)P(H_1)+P(E_2|H_2)P(H_2)=\frac{0.99+0.995}{2}=0.9925\]

  • We compute the probability P(E_2|E_1) by the definition:

    \[P(E_2|E_1)=\frac{P(E_1\cap E_2}{P(E_1)}\]

From the previous part, we know P(E_1)=0.9925. By the law of total probability

    \[P(E_1\cap E_2)=P(E_1\cap E_2|H_1)P(H_1)+P(E_1\cap E_2|H_2)P(H_2)=\frac{0.99^2+0.995^2}{2}=0.9850625\]

So,

    \[P(E_2|E_1)=\frac{0.9850625}{0.9925}=0.9925063\]

Problem 8.

Let R be a commutative ring containing a field k as a sub-ring. Assume that R is a finite dimensional k-vector space. Show that every prime ideal of R is maximal.

Topic: Abstract Algebra- Ring and Field Theory, Difficulty Level: Medium

Indian Statistical Institute, ISI MMath 2017 Problem 8 Hints along with Solution & Discussion

Hint 1

Let I be a prime ideal in R. It is enough to check that R/I is a field

Hint 2

The infinite sequence 1,a,a^2,\ldots is linearly dependent and there exists samllest m such that

    \[c_0+c_1a+c_2a^2+\ldots+c_ma^m=0\]

for some non-zero coefficients c_0,c_1,\ldots,c_m\in k. If c_0=0,

Hint 3

Then

    \[a(c_1+c_2a+\ldots+c_ma^{m-1})=0.\]

Full Solution

Let I be a prime ideal in R. It is enough to check that R/I is a field, i.e. every nonzero element of a\in R/I has a multiplicative inverse. R/I is a finite-dimensional k-vector space. Hence the infinite sequence 1,a,a^2,\ldots is linearly dependent and there exists smallest m such that

    \[c_0+c_1a+c_2a^2+\ldots+c_ma^m=0\]

for some non-zero coefficients c_0,c_1,\ldots,c_m\in k. If c_0=0, then

    \[a(c_1+c_2a+\ldots+c_ma^{m-1})=0.\]

Let x\in R be a representative of a. Then

    \[x(c_1+c_2x+\ldots+c_mx^{m-1})\in I.\]

a\in R/I\setminus \{0\}, hence x\not\in I. The ideal I is prime, so

    \[c_1+c_2x+\ldots+c_mx^{m-1}\in I\]

and

    \[c_1+c_2a+\ldots+c_ma^{m-1}=0.\]

The latter contradicts minimality of m. So, c_0\ne 0, and

    \[1=a\left(-\frac{c_1}{c_0}-\frac{c_2}{c_0}a-\ldots-\frac{c_m}{c_0}a^{m-1}\right)\]

a\in R/I has a multiplicative inverse. R/I is a field and I is a maximal ideal.

Problem 9.

Let p,q be prime numbers and n\in\mathbb{N} such that p \nmid n-1. \mbox{ If }p\mid n^q-1 then show that q\mid p-1.

Topic: Number Theory, Difficulty Level: Easy

Indian Statistical Institute, ISI MMath 2017 Problem 9 Hints along with Solution & Discussion

Hint 1

By assumptions of the problem (n,p)=1.

Hint 2

Let k be the order of n modulo p, i.e. k\geq 2 is the smallest positive integer such that p\mid n^k-1.

Hint 3

From p\mid n^q-1 we deduce that k\mid q. q is prime, so k=q.
Now use “Fermat’s Little Theorem”.

Full Solution

By assumptions of the problem (n,p)=1. Let k be the order of n modulo p, i.e. k\geq 2 is the smallest positive integer such that p\mid n^k-1. From p\mid n^q-1 we deduce that k\mid q. q is prime, so k=q. By the Fermat Little Theorem,

    \[p\mid n^{p-1}-1.\]

Since q is the order of n modulo p, it follows that q\mid p-1.

Problem 10.

Determine all finite groups which have exactly 3 conjugacy classes.

Topic: Group Theory, Difficulty Level: Hard

Indian Statistical Institute, ISI MMath 2017 Problem 10 Hints along with Solution & Discussion

Hint 1

Let G be a finite group with exactly 3 conjugacy classes. One conjugacy class is always the identity class \{e\}.

Hint 2

Let m and n be sizes of other two conjugacy classes. Conjugacy classes form a partition of a group, so

    \[|G|=1+m+n.\]

Hint 3

The size of a conjugacy class of g\in G equals to the index of a centralizer of g,

Full Solution

Let G be a finite group with exactly 3 conjugacy classes. One conjugacy class is always the identity class \{e\}. Let m and n be sizes of other two conjugacy classes. Conjugacy classes form a partition of a group, so

    \[|G|=1+m+n.\]

The size of a conjugacy class of g\in G equals to the index of a centralizer of g, i.e. it divides |G|. So,

    \[m\mid |G|, \ n\mid |G|.\]

Further,

    \[m\mid n+1=|G|-m, \ n\mid m+1=|G|-n.\]

Assume that m=n. Then m\mid m+1, i.e. m=n=1, |G|=3. The only group of order 3 is the cyclic group Z_3. It is an abelian group, hence it has exactly 3 conjugacy classes.

 

Assume that m<n. Then m+1\leq n, hence n=m+1, m\mid 2=n+1-m. If m=1, then n=2 and |G|=4. The only groups of order 4 are the Klein 4 group or the cyclic group Z_4. Both are abelian and thus have 4 conjugacy classes.

 

If m=2, then n=3 and |G|=6. There are two groups of order 6: an abelian group Z_6, and a nonabelian group S_3 (permutations on three elements). We check that S_3 has three conjugacy classes:

 

\{()\}, \{(12),(13),(23)\}, \{(123),(132)\}

 

The only finite groups with three conjugacy classes are Z_3 and S_3.

 

Problem 11.

Let F be a field, a\in F, p a prime integer. Suppose the polynomial x^p-a is reducible in F[x]. Prove that this polynomial has a root in F.

Topic: Field Theory, Difficulty Level: Medium

Indian Statistical Institute, ISI MMath 2017 Problem 11 Hints along with Solution & Discussion

Hint 1

In the case a=0 obviously x=0 is the needed root.

Hint 2

We assume firther that a\ne 0. Let K be the splitting field of x^p-a over F.

Hint 3

Then

    \[x^p-a=(x-b_1)\ldots(x-b_p)\]

for some b_1,\ldots,b_p\in K.

Full Solution

In the case a=0 obviously x=0 is the needed root. We assume firther that a\ne 0. Let K be the splitting field of x^p-a over F. Then

    \[x^p-a=(x-b_1)\ldots(x-b_p)\]

for some b_1,\ldots,b_p\in K. Since the polynomial x^p-a is reducible, we may relabel its roots in such a way that for some n<p a polynomial

    \[f(x)=(x-b_1)\ldots(x-b_n)\]

is a factor of x^p-a, i.e. f(x) has coefficients in F. Observe that for each i, b^p_i=a. Consider the free term of f(x). It is of the form \pm b_1b_2\ldots b_n. Hence, b=b_1b_2\ldots b_n\in F. Further,

    \[b^p=(b_1b_2\ldots b_n)^p=b^p_1\ldots b^p_n=a^n.\]

Since (n,p)=1 we can find integers r,s such that

    \[1=rn+sp.\]

We have a^sb^r\in F, and

    \[(a^sb^r)^p=a^{sp}b^{rp}=a^{sp+rn}=a\]

a^sb^r is a root of x^p-a in F.

Problem 12.

Let V be a finite-dimensional vector space over a field F and let T:V\to V be a linear transformation. Let W\subseteq V be a subspace such that T(W)\subseteq W. Suppose T is diagonalizable. Is T restricted to W also diagonalizable?

Topic: Linear Algebra, Difficulty Level: Hard

Indian Statistical Institute, ISI MMath 2017 Problem 12 Hints along with Solution & Discussion

Hint 1

By assumption, V is a direct sum of subspaces

    \[V=V_1\oplus \ldots\oplus V_k,\]

and on each subspace V_k T acts as a multiplication by \lambda_k, all \lambda_1,\ldots\lambda_k are distinct.

Hint 2

Consider w\in W. Then

    \[w=v_1+\ldots+v_k\]

for some v_i\in V_i, 1\leq i\leq k. Let m be the number of nonzero summands v_1,\ldots,v_k.

Hint 3

By induction on m we will prove that all v_1,\ldots,v_k\in W. If m=0, then v_1=\ldots=v_k=0 and there is nothing to prove.

Full Solution

By assumption, V is a direct sum of subspaces

    \[V=V_1\oplus \ldots\oplus V_k,\]

and on each subspace V_k T acts as a multiplication by \lambda_k, all \lambda_1,\ldots\lambda_k are distinct.

 

Consider w\in W. Then

    \[w=v_1+\ldots+v_k\]

for some v_i\in V_i, 1\leq i\leq k. Let m be the number of nonzero summands v_1,\ldots,v_k. By induction on m we will prove that all v_1,\ldots,v_k\in W. If m=0, then v_1=\ldots=v_k=0 and there is nothing to prove. Assume the result is proved for number of nonzero summands <m, and let v_i\ne 0. Then

    \[Tw=\lambda_1 v_1+\ldots +\lambda_k v_k,\]

    \[Tw-\lambda_i w=\sum_{j\ne i}(\lambda_j-\lambda_i)v_i\in W.\]

By inductive assumption, for all j\ne i

    \[(\lambda_j-\lambda_i)v_i\in W.\]

Eigenvalues \lambda_1,\ldots,\lambda_k are distinct, so v_j\in W for j\ne i. But then

    \[v_i=w-\sum_{j\ne i}v_j\in W.\]

This statement shows that in the decomposition

    \[w=v_1+\ldots+v_k\]

all summands are in W, i.e.

    \[W=(W\cap V_1)\oplus\ldots \oplus (W\cap V_k).\]

The restriction of T to W is diagonalizable.

To view ISI MMATH 2020 solutions, hints, and discussions: Click Here

To view ISI MMATH 2019 solutions, hints, and discussions: Click Here

To view ISI MMATH 2018 solutions, hints, and discussions: Click Here

To view ISI MMATH 2016 solutions, hints, and discussions: Click Here

To view previous year’s question papers: Click Here

6 Comments

Leave a Reply to Mausami Cancel reply