- (30%) Suppose a random value y follows the g distribution, i.e., y can be expressed as a function of a normally distributed variable as follows.
y=δ+λexp(gz)−1g,
where δ and g are real numbers, λ is a positive real number, and z follows the normal distribution with mean to be μ and standard deviation to be σ. Given the constraint [g(y−δ)/λ+1]>0, solve the following problems.- (4%) Express z in terms of y, i.e., find the function h(y)=z.
- (4%) With the function h(y), derive the density function of y, f(y), according to the definition f(y)=1σ√2π|h′(y)|e−12σ2(h(y)−μ)2, where μ and σ are the mean and the standard deviation of z.
- (8%) Derive the expectation of y, E(y).
- (8%) Derive the variance of y, var(y).
- (6%) Show that when g is zero, y follows a normal distribution. [Hint: Consider the result of limg→0exp(gz)−1g.]
- 移項便有
g(y−δ)λ+1=exp(gz)
由於 [g(y−δ)λ+1]>0,故可同取自然對數便有 gz=ln(1+g(y−δ)λ),故得z=1gln(1+g(y−δ)λ):=h(y)
- 先求 h 的導函數有
h′(y)=1g⋅gλ+g(y−δ)=1λ+g(y−δ)
由於 g(y−δ)/λ+1>0,故 λ+g(y−δ)>0,故 |h′(y)|=h′(y)。從而 y 的機率密度函數為f(y)=1σ√2π⋅1λ+g(y−δ)exp[−12σ2(ln(λ+g(y−δ))−lnλg−μ)2]
其中 y∈(y∗,∞),y∗=−λ/g+δ。 - 那麼由期望值的定義可知
E(y)=∫∞y∗yf(y)dy
令 t=h(y)−μσ√2,那- 當 y→y∗ 時有 t→−∞;
- 當 y→∞ 時有 t→∞;
- 求導有 dt=h′(y)σ√2dy。再者整理有 y=h−1(μ+tσ√2)。
E(y)=1√π∫∞−∞h−1(μ+tσ√2)e−t2dt=1√π∫∞−∞[δ−λg+λgexp(gμ+tgσ√2)]e−t2dt=1√π∫∞−∞(δ−λg)e−t2dt+λegμg√π∫∞−∞exp(−t2+tgσ√2)dt=δ−λg+λegμ+g2σ2/2g
- 按照變異數的定義有 var(y)=E(y2)−(E(y))2,故我們僅需計算 E(y2),即計算
E(y2)=∫∞y∗y2f(y)dy
沿用前一小題的代換可知E(y2)=1√π∫∞−∞[δ−λg+λgexp(gμ+tgσ√2)]2e−t2dt=1√π∫∞−∞(δ−λg)2e−t2dt+2λegμg√π(δ−λg)∫∞−∞exp(−t2+tgσ√2)dt+λ2e2gμg2√π∫∞−∞exp(−t2+2tgσ√2)dt=(δ−λg)2+2λegμ+g2σ2/2g(δ−λg)+λ2e2gμ+2g2σ2g2
因此所求為var(y)=[(δ−λg)2+2λegμ+g2σ2/2g(δ−λg)+λ2e2gμ+2g2σ2g2]−(δ−λg+λegμ+g2σ2/2g)2=λ2g2e2gμ+g2σ2(eg2σ2−1)
- 由於當 g 趨於零時可以發現 limg→0exp(gz)−1g=limg→0zexp(gz)1=z。故當 g=0 時有 y=δ+λz,此表明經由線性轉換能使 y 成為平均為 δ+λμ,而標準差則為 λσ。由 (c) 與 (d) 能知 limg→0E(y)=δ+λμ 且 limg→0var(y)=λ2σ2。
- (8%) Find the maximum and minimum values of f(x,y,z)=x+3y−z subject to 4x2+2y2+z2=4.
- (12%) A 10,000-cubic-foot-room has 500 smoke particles per cubic foot. A ventilation system is turned on that each minute brings in 500 cubic feet of smoke-free air, while an equal volume of air leaves the room. Also, during each minute, smoker in the room add a total of 10,000 particles of smoke to the room. Assume that the air in the room mixes thoroughly.
- (4%) Find a differential equation and initial condition that govern the total number y(t) of smoke particles in the room after t minutes.
- (4%) Solve this differential equation.
- (4%) Find how soon the smoke level will fall to 100 smoke particles per cubic foot. (If the answers are with decimal numbers, please round to the nearest hundredth.)
- 由於 y(t) 表在時刻 t 分鐘後的煙霧粒子的個數,那麼 y′(t) 便表示在 t 分鐘後增加煙霧粒子的瞬間增加量,按照題意可知
y′(t)=10000−y(t)10000⋅500
再者一開始房間內有 500⋅104 個煙霧粒子,因此 y(0)=5⋅106。 - 經由移項可知
dy200000−y=dt20
同取積分有−ln|200000−y|=t20+C
透過初始條件可知 C=−ln4800000,從而所求為y(t)=200000+4800000e−t/20
- 要達到濃度為每立方英尺有 100 顆煙霧粒子,則該房間要有總數 106 顆煙霧粒子。因此解方程式 y(t)=106,即 2+48e−t/20=10,故有 t=20ln6。
- Evaluate:
- (5%) ϕ(α)=∫∞0αe−αxdx for α>0.
- (10%) Evaluate ∫10ex2dx, approximately by using Taylor's theorem of the mean and estimate the maximum error.
- 直接計算瑕積分可知
ϕ(α)=−e−αx|∞0=1
- 回憶自然指數的泰勒展開式為
ex=n∑k=0xkk!+eξ(n+1)!xn+1
此處 ξ 介於 0 與 x 之間。現用 x2 取代 x 有ex2=n∑k=0x2kk!+eη2(n+1)!x2n+2
而此時的 η 介於 0 與 x 之間。由此計算定積分可知其近似值為∫10ex2dx≈∫10n∑k=0x2kk!dx=n∑k=01k!(2k+1)
而其最大的誤差估計為|∫10ex2dx−n∑k=01k!(2k+1)|≤∫10eη2x2n+2(n+1)!dx=eη2(n+1)!(2n+3)≤e(n+1)!(2n+3)
這是因為 η∈[0,x]⊆[0,1],因此 eη2≤e。 - Let φ be a continuous real function on [0,α]. Assume that φ(x)>0 if 0<x≤α and that φ(x)∼Axr as x→0 (A>0, r≥0). For all t>0 set
F(t)=∫α0dxt+φ(x).
- (10%) Show that if r>1, then as t→0
F(t)∼πrA1/rsin(π/r)⋅1t1−1/r.
- (10%) Show that if r=1, then as t→0
F(t)∼1Alog1t.
- (10%) Show that if r>1, then as t→0
- 給定正數 ε<1,那麼存在 δ>0使得 0<x<δ 蘊含 Axr(1−ε)<φ(x)<Axr(1+ε)。據此我們可寫 F 如下
F(t)=∫δ0dxt+φ(x)+∫αδdxt+φ(x)
對於第二項可得如下的估算0<∫αδdxt+φ(x)≤∫αδdxmin[δ,α]φ=α−δmin[δ,α]φ<∞
故有 t∗1 使得 0<t<t∗1 蘊含∫αδ1t+φ(x)<ε2rA1/r(1+ε)1/rt1−1/r
至於第一項則容易知道1t∫δ0dx1+Axr(1+ε)/t=∫δ0dxt+Axr(1+ε)<∫δ0dxt+φ(x)<∫δ0dxt+Axr(1−ε)=1t∫δ0dx1+Axr(1−ε)/t
首先我們處理積分 ∫δ0dx1+Axr(1+ε)/t,令 y=1+Axr(1+ε)/t,那麼有- 當 x=0 時有 y=1;
- 當 x=δ 時有 y=1+Aδr(1+ε)/t;
- 整理有 x=(t(y−1)A(1+ε))1/r,求導有 dx=1r(tA(1+ε))1/r(y−1)(1−r)/rdy。
∫δ0dx1+Axr(1+ε)/t=t1/rrA1/r(1+ε)1/r∫1+Aδr(1+ε)/t11y(y−1)1−1/rdy=t1/rrA1/r(1+ε)1/r[∫∞11y(y−1)1−1/rdy−∫∞1+Aδr(1+ε)/t1y(y−1)1−1/rdy]
由於 limy→∞1y(y−1)1−1/r1y=0,故瑕積分 ∫∞1+Aδr(1+ε)/t1y(y−1)1−1/rdy 有限,從而 limt→0+∫∞1+Aδr(1+ε)/t1y(y−1)1−1/rdy=0,即當 t<t∗2 時有∫∞1+Aδr(1+ε)/t1y(y−1)1−1/rdy<ε2
另一方面,透過變換 u=y−1 可知∫∞11y(y−1)1−1/rdy=∫∞0u1/r−11+udu
運用複變函數論中的留數定理與路徑積分(可參見「複變數-原理及題解」,中央圖書出版社,吳新生編著,第 403 頁)可知∫∞0u1/r−11+udu=πsin(π/r)
綜合上述的推導可得1rA1/r(1+ε)1/r(πsin(π/r)−ε)1t1−1/r<F(t)
類似地,我們也有F(t)<1rA1/r(1−ε)1/r(πsin(π/r)+ε)1t1−1/r
將兩者結合可知1rA1/r(1+ε)1/r(πsin(π/r)−ε)<t1−1/rF(t)<1rA1/r(1−ε)1/r(πsin(π/r)+ε)
故取由 t→0+ 與 ε 的任意性可知limt→0+t1−1/rF(t)=πrA1/rsin(π/r)
這就給出了F(t)∼πrA1/rsin(π/r)1t1−1/r
- 承前述步驟可知
∫δ0dxt+Ax(1+ε)<∫δ0dxt+φ(x)<∫δ0dxt+Ax(1−ε)
直接計算可給出1A(1+ε)log(1+Aδ(1+ε)t)<∫δ0dxt+φ(x)<1A(1−ε)log(1+Aδ(1−ε)t)
同乘以 Alog(1/t) 有11+εlog(1+Aδ(1+ε)/t)log(1/t)<Alog(1/t)∫δ0dxt+φ(x)<11−εlog(1+Aδ(1−ε)/t)log(1/t)
取 t 趨於零且由 ε 的任意性可知 limt→0+Alog(1/t)∫δ0dxt+φ(x)=1,從而有limt→0+Alog(1/t)F(t)=1
這就給出了F(t)∼1Alog1t
- Suppose that an→c as n→∞ and that {αi}∞i=1 is a sequence of positive terms for which n∑i=1ai→∞ as n→∞.
- (10%) Show that
n∑i=1αiain∑i=1αi→c as n→∞.
In particular, if αi=1 for all i, then1nn∑i=1ai→c as n→∞.
- (5%) Show that the converse of the special case in (a) does not always hold by giving a counterexample of a sequence {an}∞n=1 that does not converges, yet 1nn∑i=1ai converges as n→∞.
- (10%) Show that
- 設 bn=n∑i=1αi、cn=n∑i=1αiai,按題設可知 limn→∞bn=∞。再者,有
limn→∞cn+1−cnbn+1−bn=limn→∞αn+1an+1αn+1=limn→∞an+1=c
那麼根據Stolz-Cesàro定理可知 limn→∞cnbn=c,此即所欲證。而當 αi=1,那麼容易知道 bn=n,cn=n∑i=1ai。 - 考慮 an=(−1)n,容易直接計算
n∑i=1ai={−1,if n is odd,0,if n is even.
由此可知 limn→∞1nn∑i=1ai=0,但數列 {ai}∞i=1 發散。
訣竅
前兩小題根據題目給定的資訊整理即可,而後三小題則分別按照期望值、變異數與常態分配的定義處理。解法
訣竅
運用初等不等式求極值;亦可使用拉格朗日乘子法求解。解法一
由柯西不等式可知23=[(2x)2+(√2y)2+z2][(12)2+(3√2)2+(−1)2]≥(x+3y−z)2
因此函數 f 在給定的限制條件下的最大最小值分別為 √23 與 −√23,此時的等號成立條件為 4x=2y3=−z,即 (1√23,6√23,−4√23) 與 (−1√23,−6√23,4√23)。解法二
設定拉格朗日乘子函數如下F(x,y,z,λ)=x+3y−z+λ(4x2+2y2+z2−4)
據此解下列的聯立方程組{Fx(x,y,z,λ)=1+8λx=0Fy(x,y,z,λ)=3+4λy=0Fz(x,y,z,λ)=−1+2λz=0Fλ(x,y,z,λ)=4x2+2y2+z2−4=0
由於明顯 λ≠0,故 x=−18λ、y=−34λ、z=12λ,由此代入第四式為464λ2+2⋅916λ2+14λ2=4
即得 λ2=2364,故有 λ=±√238。如此得兩座標(x,y,z)=(1,6,−4)√23, (x,y,z)=(−1,−6,4)√23
而最大最小值分別為 √23 與 −√23。
沒有留言:
張貼留言