STEP II/III 2014 solutions

Badges: 18

Rep:

#1

Report Thread starter 8 years ago

#1

Papers found here. (Credit to Tarquin Digby for those [and Charles98 for his photos])

STEP II:
1: Solution by mikelbird
2: Solution by LightBlueSoldier
3: Solution by Stray
4: Solution by Khallil
5: Solution by Blazy
6: Solution by DFranklin
7: Solution by cliverlong
8: Solution by LightBlueSoldier, Solution by Stray
9: Solution by Brammer
10: Solution by JosephML
11: Solution by Brammer
12: Solution by shamika
13: Solution by DJMayes

STEP III:
1: Solution by jtSketchy
2: Solution by Brammer
3: Solution by edrraa, Solution by mikelbird
4: Solution by Farhan.Hanif93
5: Solution by metaltron
6: Solution by ctrls
7: Solution by Farhan.Hanif93
8: Solution by DFranklin
9: Solution by JosephML
10: Solution by JosephML
11: Solution by Brammer
12: Solution by DJMayes
13: Solution by DJMayes

0

reply

JosephML

Badges: 2

Rep:

#2

Report 8 years ago

#2

II/10:

Spoiler:

Show

For the first part, get the two equations for

and

and combine them:

$y = u \sin \theta - \frac{1}{2}gt^2$ and $x=u \cos \theta t$ , giving:

$y(x, \lambda) =\lambda x - x^2 \frac{\lambda^2 + 1}{2u^2}g$

Now we wish to maximize

for a given value of

so we take what is known as the partial derivative of

with respect to $\lambda$ i.e. ignore

and differentiate wrt to $\lambda$ :

$\displaystyle \frac{\partial y}{\partial \lambda} = - \frac{gx^2}{u^2} \lambda + x$ .

To get the max, we solve $\frac{\partial y}{\partial \lambda} = 0$ which gives $\lambda = \frac{u^2}{gx}$ . Putting this back into $y(x, \lambda)$ will give us

:

$\displaystyle Y = \frac{u^2}{2g} - \frac{g}{2u^2}x^2$ .

The sketch of this is fairly straight forward with the set of points which can be reached being those with

and $0\leq y \leq Y$ .

Finally the maximum possible distance reached by the particle is when

which gives us $x = \frac{u^2}{g}$ .

0

reply

shamika

Badges: 16

Rep:

#3

Report 8 years ago

#3

II/12:

(i)

Spoiler:

Show

$P(t < T < t + \delta t | T > t) = \frac{P(t < T < t + \delta t)}{P(T>t)}$ . The denominator is just 1 - F(t). The numerator is $\frac{F(t + \delta t) - F(t)}{\delta t} \cdot \delta t = f(t) \cdot \delta t$ , since the first term is tends to the derivative of F(t) (i.e. f(t)), as t tends to 0.

(ii)

Spoiler:

Show

In this case, $f(t)=\frac{1}{a}$ . giving immediately $h(t) = \frac{1}{a-t}$ .

(iii)

Spoiler:

Show

Note that $f(t) =\frac{dF(t)}{dt}$ . Therefore we want to solve $\frac{1}{t}= \frac{\frac{dF(t)}{dt}}{1-F(t)}$ . Separating the variables gives

$\int {\frac{dt}{t}} = \int{\frac{dF(t)}{1-F(t)}}$ , i.e.

$\log t = -\log(1 - F(t)) + c$ . When t = a, F(t) = 0, so c = log a, and so rearranging we get $F(t) = 1 - \frac{a}{t}$ . Differentiating, we have $\frac{a}{t^2}$ .

As a check, we can insert F(t) and f(t) into the definition of the hazard function; we get $h(t) = \frac{a}{t^2} \cdot \frac{t}{a} = \frac{1}{t}$ , as required.

(iv)

Spoiler:

Show

We prove each implication separately. If $f(t) = k\exp(-k(t-b))$ , then $F(t)=k\int_b^t exp(-k(x-b)dx$ , where x is just a dummy variable. This gives $F(t) = 1 - \exp(-k(t-b))$ . Subbing this into the formula of h(t), we get h(t) = k, a constant.

On the other hand, if h(t) = k, then if we solve $k(t - F(t)) = \frac{dF(t)}{dt}$ , which (after separating the variables and integrating to find F(t)), gives $F(t) = 1 - \exp(-k(t-b))$ , which differentiates to the required form of f(t).

(v)

Spoiler:

Show

Similar to (iii) and (iv), solve $\frac{\lambda}{\theta^{\lamda}} \cdot t^{\lamda -1} \cdot (1 - F(t)) = \frac{dF(t)}{dt}$ , to give $F(t) = 1 - \exp(-(\frac{t}{\theta})^{\lambda})$ . Finally, differentiating, we get $f(t) = \frac{\lambda}{\theta}\cdot (\frac{t}{\theta})^{\lambda - 1} \codt \exp(-(\frac{t}{\theta})^{\lambda})$

Note: this is the Weibull distribution, which is an important distribution in actuarial science to model very extreme losses. Of course, there was no reason you needed to know that to complere the question. The concept of the hazard function is also a fundamental theoretical actuarial tool, but I've never used it in practice.

0

reply

LightBlueSoldier

Badges: 14

Rep:

#4

Report 8 years ago

#4

Some photos: II 2:

Name: ImageUploadedByStudent Room1403971645.464489.jpg
Views: 2501
Size: 157.7 KB

Name: ImageUploadedByStudent Room1403971658.601121.jpg
Views: 1997
Size: 125.1 KB

Name: ImageUploadedByStudent Room1403971668.513346.jpg
Views: 1958
Size: 140.5 KB

And II 8

Name: ImageUploadedByStudent Room1403971688.538401.jpg
Views: 2256
Size: 138.5 KB

Name: ImageUploadedByStudent Room1403971696.480341.jpg
Views: 2098
Size: 136.9 KB

Posted from TSR Mobile

1

reply

JosephML

Badges: 2

Rep:

#5

Report 8 years ago

#5

III/9

Spoiler:

Show

We have
$\displaystyle \bold{r} = \frac{kt - 1 + e^{-kt}}{k^2}\bold{g} + \frac{1- e^{-kt}}{k}\bold{u}$ ,

so differentiating gives us:

$\displaystyle \bold{v} = \frac{1-e^{-kt}}{k}\bold{g} + e^{-kt}\bold{u}$

and again:

$\displaystyle \bold{a} = e^{-kt} \bold{g} - k e^{-kt} \bold{u}$ .

For the initial conditions just put in

.

Now we must show that this fits with $\bold{F} = m \bold{a}$ :

$\bold{F} &= m \bold{g} - mk\bold{v} \\ &= m\bold{g} - m((1-e^{-kt})\bold{g} - ke^{-kt}\bold{u}) \\ &= m(e^{-kt}\bold{g} - ke^{-kt}\bold{u})$

And so we indeed have $\bold{F} = m\bold{a}$ .

Now letting $\bold{u} = u\cos \alpha \bold{i} + u \sin \alpha \bold{j}$ and $\bold{g} = -g \bold{j}$ and subbing into $\bold{r}$ we get:

$\displaystyle \bold{r} = (- \frac{kt - 1 + e^{-kt}}{k^2} + \frac{1-e^{-kt}}{k} u \sin \alpha) \bold{j} + \frac{1- e^{-kt}}{k} u \cos \alpha \bold{i}$

For $\bold{r . j} = \bold{0}$ we must have the $\bold{j}$ component of $\bold{r}$ to be zero. Hence giving:

$\displaystyle ku \sin \alpha = (\frac{kT}{1- e^{-kT}} -1)g$ .

Similarly writing $\bold{v}$ in terms of $\bold{i}$ and $\bold{j}$ we have:

$\displaystyle \bold{v} = (e^{-kt} u \sin \alpha - \frac{1- e^{-kt}}{k}g) \bold{j} + e^{-kt} u \cos \alpha \bold{i}$

which then gives

$\displaystyle \tan \beta = - \frac{e^{-kT} u \sin \alpha - k^{-1}(1-e^{-kT}g}{e^{-kT}u \cos \alpha}$

which simplifies to the required result. Combining the two results gives us:

$\displaystyle \tan \beta = \tan \alpha \left(\frac{e^{kT} -1 - kT}{kT + e^{-kT} - 1}\right)$
We are given that $\sinh \left(kT\right) > kT$ . So from the definition of $\sinh$ we have:

$\displaystyle \frac{e^{kT} - e^{-kT}}{2} > kT \\ \Rightarrow e^{kT} - kT - 1< e^{-kT} + kT -1 \Rightarrow \frac{e^{kT} - kT - 1}{e^{-kT} + kT -1} > 1$

And so we get the required $\tan \beta > \tan \alpha$ . Finally since $\alpha$ and $\beta$ are both acute we must have $\beta > \alpha$

0

reply

user2020user

Badges: 16

#6

Report 8 years ago

#6

II/4:

Spoiler:

Show

$\begin{aligned} \text{I} = \int_{1/b}^{b} \dfrac{x\ln x}{\left( a^2 + x^2 \right) \left( a^2 x^2 + 1 \right)} \text{ d}x & \ \overset{u=1/x}= \ \int_{b}^{1/b} \dfrac{\frac{1}{u} \ln \left( \frac{1}{u} \right)}{\left( a^2 + \frac{1}{u^2} \right) \left( \frac{a^2}{u^2} + 1 \right)} \cdot \dfrac{\text{ d}u}{-u^2} \\ & = \int_{b}^{1/b} \dfrac{ - \frac{1}{u} \ln u}{- \left( a^2 u^2 +1 \right) \frac{1}{u^2} \left(a^2 + u^2 \right)} \text{ d}u \\ & \int_{b}^{1/b} \dfrac{u\ln u}{\left( a^2 u^2 +1 \right) \left(a^2 + u^2 \right)} \text{ d}u \\ & = - \int_{1/b}^{b} \dfrac{u\ln u}{\left( a^2 u^2 +1 \right) \left(a^2 + u^2 \right)} \text{ d}u \end{aligned}$

$\begin{aligned} 2\text{I} & = \int_{1/b}^{b} \dfrac{x\ln x}{\left( a^2 + x^2 \right) \left( a^2 x^2 + 1 \right)} \text{ d}x + \underbrace{\left( - \int_{1/b}^{b} \dfrac{u\ln u}{\left( a^2 u^2 +1 \right) \left(a^2 + u^2 \right)} \text{ d}u \right)}_{\text{Rewriting } u \text{ as } x} \\ & = \int_{1/b}^{b} \dfrac{x\ln x}{\left( a^2 + x^2 \right) \left( a^2 x^2 + 1 \right)} \text{ d}x - \int_{1/b}^{b} \dfrac{x\ln x}{\left( a^2 x^2 +1 \right) \left(a^2 + x^2 \right)} \text{ d}x = 0 \end{aligned} \\ \\ \text{ I} \ = 0 \end{aligned}$

$\displaystyle \begin{aligned} \text{J} = \int_{1/b}^{b} \dfrac{\arctan x}{x} \text{ d}x & \ \overset{u=1/x}= \ \int_{b}^{1/b} \dfrac{\arctan \left( \frac{1}{u} \right)}{\left( \frac{1}{u} \right)} \cdot \dfrac{\text{ d}u}{-u^2} \\ & = \int_{b}^{1/b} \dfrac{ -\arctan \left( \frac{1}{u} \right)}{u} \text{ d}u \\ & = \int_{1/b}^{b} \dfrac{ \arctan \left( \frac{1}{u} \right)}{u} \text{ d}u \end{aligned}$

$\begin{aligned} 2\text{J} & = \int_{1/b}^{b} \dfrac{ \arctan x}{x} \text{ d}x + \underbrace{\int_{1/b}^{b} \dfrac{ \arctan \left( \frac{1}{u} \right)}{u} \text{ d}u}_{\text{ Rewriting } u \text{ as } x} \\ & = \int_{1/b}^{b} \dfrac{ \arctan x + \arctan \left( \frac{1}{x} \right)}{x} \text{ d}x \ = \ \dfrac{\pi}{2} \int_{1/b}^{b} \dfrac{\text{d}x}{x} \ = \ \dfrac{\pi}{2} \ln (b^2) \end{aligned}$

$\text{ J} \ = \dfrac{\pi \ln b}{2}$

$\begin{aligned} I & = \int_{0}^{\infty} \dfrac{1}{\left( a^2 + x^2 \right)^2} \text{ d}x \\ & \overset{u = a{^2}/x}= \dfrac{1}{a^{2}} \int_{0}^{\infty} \dfrac{u^{2}}{(u^2 + a^2)^{2}} \text{ d}u \\ & \overset{+a^{2}-a^{2}}= \dfrac{1}{a^{2}} \left( \int_{0}^{\infty} \dfrac{1}{u^{2}+a^{2}} \text{ d}u - a^{2} I \right) \implies 2I = \dfrac{1}{a^{2}} \underbrace{ \left( \int_{0}^{\infty} \dfrac{1}{u^{2}+a^{2}} \text{ d}u \right)}_{= \pi/2a} \end{aligned}$

I'll edit the last part into this post in a short while.

Last edited by user2020user; 2 years ago

2

reply

31541

Badges: 2

Rep:

#7

Report 8 years ago

#7

(Original post by JosephML)
III/9
.

I only got up as far as the result involving uksin - how many marks do you think I'd have got?

0

reply

Kasel

Badges: 0

Rep:

#8

Report 8 years ago

#8

(Original post by JosephML)
III/9

I know it's trivial but didn't you also have to show that when t=0, v=u (initial conditions)?

0

reply

ctrls

Badges: 9

Rep:

#9

Report 8 years ago

#9

III/6

Spoiler:

Show

By letting

with the given inequality and noting that the inequality

holds for

where

, you get,

$\int_0^{x_0} f''(t) dt = f'(x_0) - f(0) = f'(x_0)>0.$

Since

Repeating the process by setting

gives the required inequality.

(i) Choosing $f(x)=1 - \cos{x} \cosh{x}$ , if we can prove that

for all $0<x<\frac{\pi}{2}$ , then the result follows. To do this, we differentiate it twice, so

$f'(x) = \sin{x}\cosh{x} - \cos{x}\sinh{x},$
$f'(x) = \cos{x}\cosh{x} + \sin{x}\sinh{x} + \sin{x}\sinh{x} - \cos{x}\cosh{x} = 2\sin{x}\sinh{x}.$

Noting that for $0<x<\frac{\pi}{2},$ $\sin{x}>0$ and $\sinh{x}>0$ so it follows that

for this range of

Since

the conditions for the inequality in the original question is met and we can deduce

for the given range of

and the result follows.

(ii) Here you can prove each bound separately. Since for $0<x<\frac{\pi}{2}$ you can show that

, $\sin{x}>0$ and $\sinh{x}>0$ , we can multiply over the denominators without worrying about flipping the inequality. This gives,

$\frac{1}{\cosh{x}} < \frac{\sin{x}}{x} \Leftrightarrow \sin{x}\cosh{x}-x>0.$

So we set $f(x)=\sin{x}\cosh{x}-x$ and prove it is positive in the given domain using the inequality again. Differentiating twice gives,

$f'(x)=\cos{x}\cosh{x} + \sin{x}\sinh{x} - 1,$
$f''(x)=-\sin{x}\cosh{x} + \cos{x}\sinh{x} + \cos{x}\sinh{x} + \cos{x}\cosh{x} = 2\cos{x}\sinh{x}.$

Noting again the the two functions are positive in the domain and that

the result follows.

Finally for the upper bound, rearranging the inequality as,

$\frac{\sin{x}}{x} < \frac{x}{\sinh{x}} \Leftrightarrow x^2 - \sin{x}\sinh{x}>0,$

And setting $f(x)=x^2 - \sin{x}\sinh{x}>0$ , again it is sufficient to show that

for the given domain to establish the inequality. Differentiating twice,

$f'(x)=2x - \cos{x}\sinh{x} - \sin{x}\cosh{x},$
$f''(x)=2+\sin{x}\sinh{x} - \cos{x}\cosh{x} - \cos{x}\cosh{x} - \sin{x}\sinh{x} = 2\left(1-\cos{x}\cosh{x}\right).$

Note that this gives the same inequality from part (i), so we can quickly deduce that

for the given domain. Since

we can deduce that

and the result follows.

Note: Most of this is just tedious manipulation to show the initial inequalities are met, so I've omitted some of the details of showing certain functions are positive and that f(0) and f'(0) are equal to zero in the various cases. If there's an faster way to do this question that I've missed, I would be interested to hear about it.

1

reply

JosephML

Badges: 2

Rep:

#10

Report 8 years ago

#10

(Original post by Kasel)
I know it's trivial but didn't you also have to show that when t=0, v=u (initial conditions)?

Ah, you're right. I missed that bit. I'll correct the solution, thanks.

(Original post by CD315)
I only got up as far as the result involving uksin - how many marks do you think I'd have got?

I should suspect somewhere between 8 and 12 marks.

0

reply

Kasel

Badges: 0

Rep:

#11

Report 8 years ago

#11

(Original post by Khallil)
II/4

Do you think we would have had to prove arctan(x) + arctan(1/x)=pi/2?
There are a few short proofs but I forgot to include any in my solution.

0

reply

Farhan.Hanif93

Badges: 18

Rep:

#12

Report Thread starter 8 years ago

#12

STEP III, Q4

(i)

Note, expanding the square in the integrand of

:

$I - I_1 = -\displaystyle\int_0^1 y^2(1+\tan^2 x) + 2yy'\tan x dx$

$= - \displaystyle\int_0^1 y^2\sec^2x +\left(\dfrac{d}{dx}(y^2)\right) \tan x dx$

$= - \displaystyle\int_0^1 \dfrac{d}{dx}[y^2\tan x] dx$

$= y(0)^2\underbrace{\tan (0)}_{= \ 0} - \underbrace{y(1)^2}_{= \ 0} \tan (1)$

$\Rightarrow \boxed{I-I_1 = 0}$ , as required.

Hence $I = \displaystyle\int_0^1 \underbrace{(y' + y\tan x)^2}_{\geq 0} dx$ , which can interpreted as the area under a curve that is always above the

-axis. It follows that $\boxed{I\geq 0}$ with $I=0 \iff y'+y\tan x = 0$ over the interval of integration i.e. for every

s.t. $0\leq x\leq 1$ . Solving:

$\iff \displaystyle\int\dfrac{y'}{y} dx= \displaystyle\int -\tan x dx$

$\iff y= A\cos x$ .

Note $y(1) = 0 \Rightarrow A=0$ so that $\boxed{I=0 \iff y\equiv 0}$ on $0\leq x\leq 1$ , as desired.

(ii)

Let $J_1 = \displaystyle\int_0^1 (y' + ay\tan b x)^2 dx$ and observe that $J_1 \geq 0$ by the argument in

. Moreover, we require $b<\dfrac{\pi}{2}$ in order to avoid integrating over the discontinuity of $\tan bx$ at $x= \dfrac{\pi}{2b}$ i.e. for our integral to make sense.

Furthermore:

$J-J_1 = - a\displaystyle\int_0^1 a y^2 \sec^2 bx + 2yy'\tan bx dx$

Setting

allows us to write the integrand as the derivative of a product, provided that $a<\pi /2$ for the above reasons.

So, in the case $a<\pi /2$ :

$J - J_1 = -a \displaystyle\int_0^1 \dfrac{d}{dx}[y^2 \tan ax] dx$

$=a[y(0)^2\underbrace{\tan 0}_{= \ 0} - \underbrace{y(1)^2}_{= \ 0}\tan a]$

$\Rightarrow \boxed{J = J_1 \geq 0}$ , as desired, with the argument being valid for $\boxed{a< \dfrac{\pi}{2}}$ .

In the case $a=\pi / 2$ :

Integrating the first term in the integrand of $J = \displaystyle\int_0^1 \left(y'^2 - \left(\dfrac{\pi}{2}\right)^2 y^2\right) dx$ by parts with

, we obtain:

$J = y(0)y'(0) - \displaystyle\int_0^1 y(y''+(\pi /2 )^2 y) dx$

So it is sufficient to search for a function $y\not\equiv 0$ s.t. $y''+(\pi /2)^2y =0$ with either

or

.

Seeing that the general solution of the second order differential equation is $y = A\cos \dfrac{\pi x}{2} + B\sin \dfrac{\pi x}{2}$ ,

forces us to take

and observing that $\dfrac{d}{dx}[A\cos \dfrac{\pi x}{2} ]|_{x=0} = 0$ , it follows that $y= \cos \dfrac{\pi x}{2}$ is s.t.

,

and $y''+(\pi /2)^2 y = 0$ on

.

Hence $y=\cos \dfrac{\pi x}{2} \Rightarrow J = 0 - \displaystyle\int_0^1 y\cdot 0 dx = 0$ so $\boxed{y=\cos \dfrac{\pi x}{2} }$ will do the trick.

It's worth pointing out that a very similar question appeared in a IB tripos exam for the variational principles course.

1

reply

user2020user

Badges: 16

#13

Report 8 years ago

#13

(Original post by Kasel)
...

I did but I left it out of the solution above for brevity. The one I wrote in the paper was very similar to this, but it looks really forced with regards to the 1-1 in the denominator.

$\begin{aligned} \phi = \arctan b + \arctan \frac{1}{b} \implies \tan \phi & = \dfrac{\tan \left( \arctan b \right) + \tan \left( \arctan \frac{1}{b} \right)}{1 - \tan \left( \arctan b \right)\tan \left( \arctan \frac{1}{b} \right)} \\ & = \dfrac{b + \frac{1}{b}}{1 - 1} \end{aligned}$

$\therefore \ \tan \phi = \displaystyle \lim_{n \to 0} \left( \dfrac{b + \frac{1}{b}}{n} \right) \implies \phi = \arctan \left( \lim_{n \to 0} \left( \dfrac{b + \frac{1}{b}}{n} \right) \right) = \dfrac{\pi}{2}$

0

reply

Kasel

Badges: 0

Rep:

#14

Report 8 years ago

#14

(Original post by Khallil)
...

To avoid limits you can say let y=arctan(x), then tan(y)=x
and
cot(y)=(1/x)=tan(pi/2 - y)
which gives arctan(x)+arctan(1/x)= y+pi/2-y=pi/2

0

reply

user2020user

Badges: 16

#15

Report 8 years ago

#15

(Original post by Kasel)
...

Ah, of course! :facepalm2:

0

reply

DJMayes

Badges: 16

Rep:

#16

Report 8 years ago

#16

STEP III - Q12

Spoiler:

Show

i)

Spoiler:

Show

Note that

is increasing. Thus:

$P(x \leq x_m ) = \frac{1}{2} \iff P(y \leq e^{x_m} ) = \frac{1}{2}$

So $y_m = e^{x_m}$

ii)

Spoiler:

Show

$P(Y \leq y ) = P(e^X \leq y) = P(X \leq lny)$

Differentiating, we deduce a p.d.f of $\dfrac{f(lny)}{y}$ , as required.

The mode occurs at a stationary point of the p.d.f (not at an endpoint; this is not possible as the range is infinite) so occurs when the derivative of the p.d.f is 0:

$\Rightarrow \dfrac{f'(lny) - f(lny)}{y^2} = 0$

From which $f'(ln\lambda) = f(ln\lambda)$ follows.

iii)

Spoiler:

Show

Our integral is merely the integral of the p.d.f of the random variable $N(\mu + \sigma^2, \sigma)$ , so equals 1. Then:

$E(Y) = E(e^X) = \dfrac{1}{\sigma\sqrt{2\pi}} \displaystyle\int_{-\infty}^{\infty} exp(-\dfrac{1}{2\sigma^2}((x-\mu)^2-2\sigma^2x)) \ dx$

$\dfrac{1}{\sigma\sqrt{2\pi}} \displaystyle\int_{-\infty}^{\infty} exp(-\dfrac{1}{2\sigma^2}((x-\mu-\sigma^2)^2-2\mu \sigma^2 - \sigma^4)) \ dx$

$= exp(\mu + \frac{1}{2}\sigma^2) I$

Where I is the integral mentioned above equalling one, from which our result follows.

iv):

Spoiler:

Show

Using our p.d.f. X has median $\mu$ so Y has median $e^{\mu}$

Applying ii) with f the p.d.f. given above, we find that $\lambda = e^{\mu - \sigma^2}$ :

$f(x) = \dfrac{1}{\sqrt{2 \pi \sigma^2}} exp(-\dfrac{(x-\mu)^2}{2\sigma^2})$

$f'(x) = -\dfrac{(x-\mu)}{\sigma^2 \sqrt{2 \pi \sigma^2}} exp(-\dfrac{(x-\mu)^2}{2\sigma^2})$

$f(x) = f'(x) \Rightarrow -\dfrac{(x-\mu + \sigma^2)}{\sigma^2 \sqrt{2 \pi \sigma^2}} exp(-\dfrac{(x-\mu)^2}{2\sigma^2}) = 0$

$\Rightarrow x = \mu - \sigma^2$

Noting that $x = log\lambda$ , the result follows.

Then the inequality follows easily from the fact the exponential function is increasing.

0

reply

Stray

Badges: 6

Rep:

#17

Report 8 years ago

#17

STEP II - Q8 (LaTeX)

Spoiler:

Show

$\{n, a, b, c_r\} \mathbb{Z}+$
$c_r, (0 \leq r \leq n) = \texttt{coefficient of } x^r \texttt{ in } (a + bx)^n$

$c_r = ^n\mathrm{C}_r \cdot a^{n-r} b^r= \frac{n!}{r!(n-r)!} \cdot a^{n-r} b^r$

$c_m \geq c_r \texttt{ for } 0 \leq r \leq n$

If

is a maximum, it must be $\geq c_{m+1}$ and $\geq c_{m-1}$

Taking first, $c_{m+1}$ :

$\frac{n!a^{n-(m+1)} b^{(m+1)}}{(m+1)!(n-(m+1))!} \leq \frac{n!a^{n-m} b^m}{m!(n-m)!}$

Divide both sides by the common factor

$\frac{n!a^{n-(m+1)}b^m}{m!(n-(m+1))!}$

Giving

$\frac{b}{m+1} \leq \frac{a}{n-m}$

$bn-bm \leq am+a$

$bn-a \leq am+bm$

$\frac{bn-a}{a+b} \leq m$

This is almost the LHS of what we are required to show, just needs a little fiddling

$\frac{bn+b-(a+b)}{a+b} \leq m$

$\frac{b(n+1)}{a+b} -1 \leq m (**)$

Now repeat the process for $c_{m-1}$ :

$\frac{n!a^{n-(m-1)} b^{(m-1)}}{(m-1)!(n-(m-1))!} \leq \frac{n!a^{n-m} b^m}{m!(n-m)!}$

Divide both sides by the common factor

$\frac{n!a^{n-m}b^{m-1}}{(m-1)!(n-m)!}$

Giving

$\frac{a}{n-(m-1)} \leq \frac{b}{m}$

$am \leq bn + b - bm$

$m(a+b) \leq b(n+1)$

$m \leq \frac{b(n+1)}{a+b}$

Which combines with (**) to give the required solution

$\frac{b(n+1)}{a+b} -1 \leq m \leq \frac{b(n+1)}{a+b}$

Considering the possible values of

$\texttt{Let } y=\frac{b(n+1)}{a+b}$

$y-1 \leq m \leq y$

Either

and

are both integers, or

and

are non-integer values either side of an integer, so if

is an integer, it is either

and

(2 values) or it lies between them (1 value).

For the next parts, we are asked to find either the value of

where there is 1 value, or the larger value if there are 2. This means we can consider the function

to produce the Floor of the upper limit - so if the upper limit evaluates to 8.9 then the value is 8, but if it is 9.0 it is 9, etc. The Floor symbol is $\lfloor$ .

$G(n,a,b) = \lfloor{\frac{b(n+1)}{a+b}}$

i)
$G(9,1,3) = \lfloor{\frac{3(9+1)}{1+3}} = \lfloor\frac{30}{4} = \floor{7.5} = 7$

$G(9,2,3) = \lfloor{\frac{3(9+1)}{2+3}} = \lfloor\frac{30}{5} = 6$

ii)
$G(2k,a,a) = \lfloor{\frac{a(2k+1)}{a+a}} = \lfloor\frac{2k+1}{2} = \lfloor{(k + 0.5)} = k$

$G(2k-1,a,a) = \lfloor{\frac{a(2k-1+1)}{a+a}} = \lfloor\frac{2k}{2} = k$

iii)
If we fix

and

, the only part of

that can alter is the denominator containing

, so for large

we want small

, so $G_{max}$ is when

. (Zero is not a positive integer).

iv)
For fixed

and (unknown) fixed

, the part of the function varying is:

$\frac{b}{b+1}$

For positive integer values of $b, \frac{1}{2} \leq \frac{b}{b+1} < 1.$

So $\frac{1}{2} \leq G(n,1,b) < n+1$
giving $G(n,1,b)_{max}=n$

LightBlueSoldier points out below that my original $b = \infty$ is wrong... of course the constraint isn't nearly as harsh as that, regardless of whether infinity is acceptable...

$\frac{b(n+1)}{b+1} \geq n$

$bn+b \geq nb+n$

$b \geq n$

This was a nice question once you get into it, but I can see that the layout of the question might make it look more complex than it actually was.

0

reply

LightBlueSoldier

Badges: 14

Rep:

#18

Report 8 years ago

#18

(Original post by Stray)
STEP II - Q8 (LaTeX)

$\{n, a, b, c_r\} \mathbb{Z}+$
$c_r, (0 \leq r \leq n) = \texttt{coefficient of } x^r \texttt{ in } (a + bx)^n$

$c_r = ^n\mathrm{C}_r \cdot a^{n-r} b^r= \frac{n!}{r!(n-r)!} \cdot a^{n-r} b^r$

$c_m \geq c_r \texttt{ for } 0 \leq r \leq n$

If

is a maximum, it must be $\geq c_{m+1}$ and $\geq c_{m-1}$

Taking first, $c_{m+1}$ :

$\frac{n!a^{n-(m+1)} b^{(m+1)}}{(m+1)!(n-(m+1))!} \leq \frac{n!a^{n-m} b^m}{m!(n-m)!}$

Divide both sides by the common factor

$\frac{n!a^{n-(m+1)}b^m}{m!(n-(m+1))!}$

Giving

$\frac{b}{m+1} \leq \frac{a}{n-m}$

$bn-bm \leq am+a$

$bn-a \leq am+bm$

$\frac{bn-a}{a+b} \leq m$

This is almost the LHS of what we are required to show, just needs a little fiddling

$\frac{bn+b-(a+b)}{a+b} \leq m$

$\frac{b(n+1)}{a+b} -1 \leq m (**)$

Now repeat the process for $c_{m-1}$ :

$\frac{n!a^{n-(m-1)} b^{(m-1)}}{(m-1)!(n-(m-1))!} \leq \frac{n!a^{n-m} b^m}{m!(n-m)!}$

Divide both sides by the common factor

$\frac{n!a^{n-m}b^{m-1}}{(m-1)!(n-m)!}$

Giving

$\frac{a}{n-(m-1)} \leq \frac{b}{m}$

$am \leq bn + b - bm$

$m(a+b) \leq b(n+1)$

$m \leq \frac{b(n+1)}{a+b}$

Which combines with (**) to give the required solution

$\frac{b(n+1)}{a+b} -1 \leq m \leq \frac{b(n+1)}{a+b}$

Considering the possible values of

$\texttt{Let } y=\frac{b(n+1)}{a+b}$

$y-1 \leq m \leq y$

Either

and

are both integers, or

and

are non-integer values either side of an integer, so if

is an integer, it is either

and

(2 values) or it lies between them (1 value).

For the next parts, we are asked to find either the value of

where there is 1 value, or the larger value if there are 2. This means we can consider the function

to produce the Floor of the upper limit - so if the upper limit evaluates to 8.9 then the value is 8, but if it is 9.0 it is 9, etc. The Floor symbol is $\lfloor$ .

$G(n,a,b) = \lfloor{\frac{b(n+1)}{a+b}}$

i)
$G(9,1,3) = \lfloor{\frac{3(9+1)}{1+3}} = \lfloor\frac{30}{4} = \floor{7.5} = 7$

$G(9,2,3) = \lfloor{\frac{3(9+1)}{2+3}} = \lfloor\frac{30}{5} = 6$

ii)
$G(2k,a,a) = \lfloor{\frac{a(2k+1)}{a+a}} = \lfloor\frac{2k+1}{2} = \lfloor{(k + 0.5)} = k$

$G(2k-1,a,a) = \lfloor{\frac{a(2k-1+1)}{a+a}} = \lfloor\frac{2k}{2} = k$

iii)
If we fix

and

, the only part of

that can alter is the denominator containing

, so for large

we want small

, so $G_{max}$ is when

. (Zero is not a positive integer).

iv)
For fixed

and (unknown) fixed

, the part of the function varying is:

$\frac{b}{b+1} (***)$

This tends to 1 as b tends to infinity, but will never actually reach 1, so

tends towards but never reaches

so for fixed

$G(n,1,b)_{max} =n \texttt{ when } b = \infty$

---

This was a nice question once you get into it, but I can see that the layout of the question might make it look more complex than it actually was.

The last part is incorrect. In particular the value of infinity does not exist.

Posted from TSR Mobile

1

reply

Blazy

Badges: 8

Rep:

#19

Report 8 years ago

#19

STEP II, Q5: (somebody quickly check through this please though I think it should be fine)

(i)

Let $\displaystyle y = ux$ , so $\displaystyle \frac{dy}{dx} = u + x\frac{du}{dx}$ by Product Rule.

Using this substitution, the question becomes:

Solve
$\displaystyle u + x\frac{du}{dx} = \frac{2ux+x}{ux-2x} = \frac{2u+1}{u-2}$ .

Rearranging: $\displaystyle x\frac{du}{dx} = \frac{2u+1}{u-2} - u = \frac{u^2 - 4u - 1 }{2-u}$

Separating the variables and integrating:
$\displaystyle \int \frac{2-u}{u^2 - 4u - 1 } du = \int \frac{1}{x} dx$

$\displaystyle \Rightarrow -\frac{\log(u^2 - 4u - 1)}{2} = \log x + A$

$\displaystyle \Rightarrow \log(u^2 - 4u - 1)= -2\log x + B$

$\displaystyle \Rightarrow u^2 - 4u - 1= Cx^{-2}$

$\displaystyle \Rightarrow \left ( \frac{y}{x} \right )^2 - 4\left ( \frac{y}{x} \right ) - 1= Cx^{-2}$

$\displaystyle \Rightarrow y^2 - 4xy - x^2= C$

The solution curve passes through (1,1) so C = -4

Hence: $\boxed{x^2 + 4xy - y^2 = 4}$

(ii)

We look for a and b such that:

and

in order to make the DE homogenous.

and

works.

Then substituting

and

:

$\displaystyle \frac{dY}{dX} = \frac{X-2Y}{2X+Y}$

Making the same substitution

and rearranging:

$\Rightarrow \displaystyle X\frac{dU}{dX} = \frac{-U^2 - 4U + 1}{2+U}$

$\Rightarrow \displaystyle \int \frac{2+U}{-U^2 - 4U + 1} dU = \int \frac{1}{X}dX$

$\Rightarrow \displaystyle -\frac{\log(-U^2 - 4U + 1)}{2} = \log X + A$

$\Rightarrow \displaystyle \log(-U^2 - 4U + 1) = -2\log X + B$

$\Rightarrow \displaystyle -U^2 - 4U + 1 = CX^{-2}$

$\Rightarrow \displaystyle -Y^2 - 4XY + X^2 = C$

$\Rightarrow \displaystyle -(y+1)^2 - 4(x-2)(y+1) + (x-2)^2 = C$

The solution curve goes through (1,1), so C = 5. Hence:

$\boxed {-(y+1)^2 - 4(x-2)(y+1) + (x-2)^2 = 5}$

The only real difficulty here was spotting how to get rid of the -4 and -3.

3

reply

metaltron

Badges: 11

Rep:

#20

Report 8 years ago

#20

STEP III Q5

Beginning

Spoiler:

Show

If ABCD is a parallelogram, then $\displaystyle \vec{AB} =b - a = \vec {DC} = c - d$ . So $\displaystyle a + c = b + d$ .
If $\displaystyle a + c = b+ d$ then $\displaystyle b-a = c-d$ so $\displaystyle \vec{AB} = \vec {DC}$ so ABCD is a parallelogram.

If ABCD is a parallelogram then it is a square if and only if its diagonals are perpendicular. That is $\displaystyle i(a-c) = b-d$ .

Part i)

Spoiler:

Show

Half the diagonal of a square is $\displaystyle \frac{\sqrt{2}}{2}$ times the side length of a square, and the diagonal makes an angle of 45 degrees with the side. So by a spiral enlargement:

$\displaystyle \frac{\sqrt{2}}{2}e^{i\frac{\pi}{4}} (p-q) = x - q$

So $\displaystyle x = q + \frac{\sqrt{2}}{2} \times \frac{\sqrt{2}}{2} (1+i)(p-q)$

So $\displaystyle x = q + \frac{1}{2}(p(1+i) - q(1+i)) = \frac{1}{2}(p(1+i) + q(1-i))$

Part ii)

Spoiler:

Show

By symmetry:

$\displaystyle x= \frac{1}{2}(p(1+i) + q(1-i))$

$\displaystyle y= \frac{1}{2}(q(1+i) + r(1-i))$

$\displaystyle z= \frac{1}{2}(r(1+i) + s(1-i))$

$\displaystyle t= \frac{1}{2}(s(1+i) + p(1-i))$

So $\displaystyle 2(y - t) = (q-s)(1+i) + (r-p)(1-i) = (q+r-s-p) + (q-s+p-r)i$

$\displaystyle 2i(x-z) = i((p-r)(1+i) + (q-s)(1-i)) = i((p-r+q-s) + (p-r+s-q)i)$

So

so diagonals of quadrilateral are perpendicular.

So XYZT is a square if and only if $\displaystyle \vec{XY} = \vec{TZ}$ .

That is $\displaystyle (q-p)(1+i) + (r-q)(1-i) = (r-s)(1+i) + (s-p)(1-i)$ .

Looking at the co-efficient of i (which is complex itself) $\displaystyle r - p = r - p$ which is always true.

Equating the other parts $\displaystyle 2q - p - r = r + p - 2s$ .

So XYZT is a square if and only if $\displaystyle q +s = p + r$ which is the condition for PQRS to be a parallelogram.

0

reply

Quick Reply

Oops, nobody has posted
in the last few hours.

Have you seen...?

Poll

What is missing from the school curriculum?

Watched Threads

Oops, nobody has posted
in the last few hours.

TSR Support Team

Get Started

Using TSR

Info

TSR Group

Personal statements

Connect with TSR

STEP II/III 2014 solutions

Quick Reply

Related discussions

Oops, nobody has postedin the last few hours.

Oops, nobody is replying to posts.

Have you seen...?

Poll

What is missing from the school curriculum?

Watched Threads

Spotlight

Oops, nobody has postedin the last few hours.

Oops, nobody is replying to posts.

TSR Support Team

Get Started

Using TSR

Info

TSR Group

Personal statements

Connect with TSR

Oops, nobody has posted
in the last few hours.

Oops, nobody has posted
in the last few hours.