由买买提看人间百态

topics

全部话题 - 话题: integrand
1 (共1页)
k****g
发帖数: 67
1
最关键的是第三个变量Integrand不知道怎么typedef
CALL CUBATR &
(DIMENS,NumFun,Integrand,NumRgn,Vertices,RgType,Value,AbsErr, &
! and optional parameters
IFAIL,Neval,EpsAbs,EpsRel,Restart,MinPts,MaxPts,Key,Job,Tune)
!-----------------------------------------------------------------------
! Input parameters
! ----------------
!
! DIMENS Integer.
! The dimension of the region of integration.
!
! NumFun Integer.
! Number of components of the integrand.
!
! Integrand
!
T*******x
发帖数: 8565
2
来自主题: Military版 - 向黎曼猜想发起总攻
我现在开始研究Alpha(z)了。这是黎曼Zeta函数的分子,分母是Gamma函数。研究黎曼
Zeta函数的零点,等价于研究Alpha函数和Gamma函数的零点和pole。
Alpha(z)=integration 0 to infinity of
1/(e^x-1) x^(z-1) dx
这个函数的直接定义域是实部大于1的复数,这和Gamma函数不一样了,Gamma函数的直
接定义域是实部大于0的复数。
直接看一下z=0.5+ib,这是黎曼猜想零点所在的直线。代入积分定义,先看一下实部,
integrand变成,
1/(e^x-1) 1/sqrt(x) cos(b*ln(x)),
先看x=0附近,integrand第一部分以1/x的方式趋于无穷,第二部分是根号倒数,合起
来以1/x^1.5的方式趋于无穷,不可积。第三部分cos提供了正负变号的部分,合在一起
还是不可积。
所以积分在这里无定义,不管是实部还是虚部。Alpha函数在z=0.5+ib的直线上必须通
过解析延拓来定义。
我在想,怎么才能算出比如z=0.5+i的Alpha函数值。
c****e
发帖数: 2097
3
来自主题: Physics版 - 广义泛函积分怎么处理?
not familiar with any numerical methods, but generally speaking, you have to
know the asymptotic behavior of your integrand to decide what to do with
your -\infty.
if it dies away over there, for example, i guess you can take some large
negative value.
however, if the case is not so, then you should manipulate your integrand
and integration domain a bit doing some mapping so as not to lose the
important stuff.

实函数,而是比较复杂的含时算符(矩阵)的函数, 不能直接积分得到解析表达式,
而且积分表达式随着里头算符的具体表达式不同而不同。
数来代替负无穷?
x***u
发帖数: 6421
4
来自主题: Physics版 - mathematica 积分问题
In[1]:= f[x_, y_, z_] := x^2 + y^2 + z^2;
g[x_, y_] := NIntegrate[f[x, y, z], {z, 0, 1}];
NIntegrate[g[x, y]^2 + g[x, y]^0.5, {x, 0, 1}, {y, 0, 1}]
During evaluation of In[1]:= NIntegrate::inumr: The integrand \
x^2+y^2+z^2 has evaluated to non-numerical values for all sampling \
points in the region with boundaries {{0,1}}. >>
During evaluation of In[1]:= NIntegrate::inumr: The integrand \
x^2+y^2+z^2 has evaluated to non-numerical values for all sampling \
points in the region with boundaries
l****o
发帖数: 2909
5
来自主题: Quant版 - today's interview
First of all, |W(t)| is F(t)-measurable, where F(t) is generated by W(t).
Secondly, E[\int_0^T|W_t|^2 dt] =\int_0^T E[|W_t|^2] dt=\int_0^T t dt=T^2/2
<\infty. This proves that it sufficiently satisfy ITO itegration
definition. Therefore it is a martingale. Here you can use standard machine and
L^P space definition for more rigorous proof on how to construct a general integrand from simple integrand.
Thirdly, E[[\int_0^T|W_t| dw(t)]^2]=E[\int_0^T|W_t|^2 dt] =\int_0^T E[|W_t|^
2] dt=\int_0^T t ... 阅读全帖
x********o
发帖数: 519
6
来自主题: Quant版 - today's interview
didn't you say it is not a martingale?
thought you would give us a surprise.

2
machine and
integrand from simple integrand.
|^
T
T*******x
发帖数: 8565
7
来自主题: Military版 - 向黎曼猜想发起总攻
记录一个不完全成功的尝试吧。
Gamma函数定义为,从0到无穷对x积分,
x^z/(xe^x),结果为z的函数,z为复数,可以通过解析延拓把该函数定义到全复平面。
Gamma函数的reflection formula是
Gamma(z)Gamma(1-z)=pi/sin(z*pi)。
我试证了一下,把两个积分integrand分别写为x和y的函数,然后积分乘积组合为二重
积分,再用换元法,
x=u^2,y=v^2,r^2=u^2+v^2,u/v=tan t,的方法,得到
Gamma(z)Gamma(1-z)=
2 integration 0 to pi/2 of (tan t)^(2z-1) dt。
C**o
发帖数: 10373
8
来自主题: Military版 - 向黎曼猜想发起总攻
抽死你丫的
盹盹盹
[在 TheMatrix (TheMatrix) 的大作中提到:]
:记录一个不完全成功的尝试吧。
:Gamma函数定义为,从0到无穷对x积分,
:x^z/(xe^x),结果为z的函数,z为复数,可以通过解析延拓把该函数定义到全复平面。
:Gamma函数的reflection formula是
:Gamma(z)Gamma(1-z)=pi/sin(z*pi)。
:我试证了一下,把两个积分integrand分别写为x和y的函数,然后积分乘积组合为二重
:积分,再用换元法,
:x=u^2,y=v^2,r^2=u^2+v^2,u/v=tan t,的方法,得到
:Gamma(z)Gamma(1-z)=
:☆ 发自 iPhone 买买提 1.24.07
T*******x
发帖数: 8565
9
来自主题: Military版 - 向黎曼猜想发起总攻
研究了一下Gamma函数的integrand,x^n/e^x,
如图橙黄色曲线是n=6。能看出一些性质。
首先最高点在x=6。这个时候被积函数是n^n/e^n,这是Stirling formula,结果~n!。
其次,积分的主要部分就是曲线下面积,尾部到无穷远基本可以忽略。
如果把曲线近似看作一个三角形的话,高和底都和Stirling formula给出的值有关,是
三角形的面积近似等于n!。
再考虑z=n+bi也就是z有虚部的情况,
x^(bi)=e^(ib ln(x)),也就是只提供相位,对modulus无贡献。b ln(x)随x不均匀变化
,所以考虑一个x=e^y的变换。蓝色曲线就是变换后的函数。横坐标方向压缩了。
考虑b越来越大的情况,也就是相位变化,一圈一圈的越来越快。形象的看就是x方向拧
螺旋,拧的圈数越来越多,螺旋越来越密。对于任意固定的n,b越大积分越小,因为相
位的变化,都相互抵消了。所以不管实部是多少,沿着虚部向无穷远处走,Gamma函数
都趋于0。
Alpha函数,也就是把被积函数分母替换为e^x-1。这个被积函数画个图可以看出和
Gamma函数的被积函数基本重合... 阅读全帖

发帖数: 1
10
来自主题: Military版 - 向黎曼猜想发起总攻
高点在x=7处严格发散
盹盹盹
[在 TheMatrix (TheMatrix) 的大作中提到:]
:研究了一下Gamma函数的integrand,x^n/e^x,
:如图橙黄色曲线是n=6。能看出一些性质。
:首先最高点在x=6。这个时候被积函数是n^n/e^n,这是Stirling formula,结果~n!。
:其次,积分的主要部分就是曲线下面积,尾部到无穷远基本可以忽略。
:如果把曲线近似看作一个三角形的话,高和底都和Stirling formula给出的值有关,
是三角形的面积近似等于n!。
:再考虑z=n+bi也就是z有虚部的情况,
:x^(bi)=e^(ib ln(x)),也就是只提供相位,对modulus无贡献。b ln(x)随x不均匀变
化,所以考虑一个x=e^y的变换。蓝色曲线就是变换后的函数。横坐标方向压缩了。
:考虑b越来越大的情况,也就是相位变化,一圈一圈的越来越快。形象的看就是x方向
拧螺旋,拧的圈数越来越多,螺旋越来越密。对于任意固定的n,b越大积分越小,因为
相位的变化,都相互抵消了。所以不管实部是多少,沿着虚部向无穷远处走,Gamma函数
:都趋于0... 阅读全帖
T*******x
发帖数: 8565
11
来自主题: Military版 - 向黎曼猜想发起总攻
继续。Zeta函数或者Gamma函数Alpha函数在z虚部不为零的情况下如何计算,这个还没
有进展。回头再补一些已知的gap。其中一个是还没有证明Zeta函数的级数定义与积分
定义等价。
Zeta函数的级数定义是
Zeta(z)=1/1^z+1/2^z+1/3^z+...
Zeta函数的积分定义是Zeta(z)=Alpha(z)/Gamma(z),
Alpha(z)=integration from 0 to infinity of x^(z-1)/(e^x-1) dx,
Gamma(z)=integration from 0 to infinity of x^(z-1)/e^x dx,
这两个的等价性还没有证明。
考虑一个具体的特例:z=2。
Gamma(2)=1,这个容易证明。
因此要证明integration from 0 to infinity of x/(e^x-1) dx
=1/1^2+1/2^2+1/3^2+...
这个也不容易啊。
思路是把integrand用泰勒级数展开,分别积分就会出现级数。
但是困难是积分的上下限是0到无穷,无法分项做定积分。
解决思路是做变量换元... 阅读全帖
T*******x
发帖数: 8565
12
来自主题: Military版 - 做题了,定积分
介绍一下此题的背景吧。
此题是证明Gamma函数的reflection formula中遇到的一步。
Gamma函数定义为,从0到无穷对x积分,
x^z/(xe^x),结果为z的函数,z为复数,可以通过解析延拓把该函数定义到全复平面。
Gamma函数的reflection formula是
Gamma(z)Gamma(1-z)=pi/sin(z*pi)。
我试证了一下,把两个积分integrand分别写为x和y的函数,然后积分乘积组合为二重
积分,再用换元法,
x=u^2,y=v^2,r^2=u^2+v^2,u/v=tan t,的方法,得到
Gamma(z)Gamma(1-z)=
2 integration 0 to pi/2 of (tan t)^(2z-1) dt。
当z=0.5+ib时,积分应该给出reflection formula所给出的值。这就是此题的来源。
n*****b
发帖数: 2235
13
x当然可以小于零
但这道题的integrand在在-5/2是没有定义的
所以x不能等于或小于-5/2
x******a
发帖数: 6336
14
来自主题: Programming版 - 请教一个monte carlo问题
求$\int_0^1e^{x^2} \,dx$, 为什么下面这段code总是返回1?谢谢
#include
#include
using namespace std;
double uniRand()
{
return double(rand()/RAND_MAX);
}
int main()
{
srand((unsigned)time(0));
//number of draws;

int N;
cout<<"enter N: "< cin>>N;
double result=0;
//integrand

for(int i=0; i< N; i++)
{
double temp=pow(uniRand(),2);
re... 阅读全帖
q***t
发帖数: 3
15
来自主题: Computation版 - mathematica 提问 "integrand not numerical"
列位大虾有用mathematica的吧?

请教一个问题。希望绘出一个曲线 f(t);其定义为一个积分:

\int g(x,y,t) dx
f(t) = \int ------------------ dy (1)
\int h(x,y,t) dx

g(x,y,t)和h(x,y,t)都是给定的函数。里外的三个积分限都是
(-Infinity,Infinity),一般是没有解析表达的。如果用

Integrate[ g(x,y,t), x ]
f(t) = NIntegrate[ --------------------------, y ] (2)
Integrate[ h(x,y,t), x ]

计算就太慢了。原因是 mathematica 试图简化 Integrate[ g(x,y,t), x ] 未果。
可是改成数值积分:

a***u
发帖数: 72
16
来自主题: Computation版 - Re: mathematica 提问 "integrand not numeri
第一个例子,
定义吧:
functmp0[y_,t_]:=Block[{x},
NIntegrate[ g(x,y,t), {x,-Infinity,Infinity} ]/
NIntegrate[ h(x,y,t), {x,-Infinity,Infinity} ]
]
f[t_]:=Block[{y},
NIntegrate[functmp0[y,t],{y,-Infinity,Infinity}]
]
Plot[f[t],{t,tmin,tmax}];
NIntegrate[ NIntegrate[ x+y, {x,0,1} ], {y,0,1} ] 简单
NIntegrate[x+y,{x,0,1},{y,0,1}];
Solve[ x + NIntegrate[x+y,{y,0,1}] == 1, x ] 不好
定义:
functmp1[x_]:=Block[{y},
-1+x + NIntegrate[ x+y, {
h***o
发帖数: 539
17
function pointer
void cubatr(int* dimens, int* numfun, void* (*integrand)(double*, int*), ....)
m****n
发帖数: 51
18
来自主题: Mathematics版 - 关于这个积分
Integrate[-Exp[-x*x]*Sin[x]^2/x^2,{x,-Infinity,
Infinity}]=Sqrt[Pi]*(-1+1/e+Erf[1])=1.527
这个积分的答案我早已经知道,但是我不知道Mathematica是怎么作出来的, 所以贴
在这里, 看谁能作出就象水请教。
了解计算这个积分的目的是想解析计算一个更一般的积分
Integratep[-Exp[-C^2*x^2]*Sin[x-x1]*Sin[x-x2]/((x-x1)*(x-x2)),{x,-Infinity,
Infinity}]
关于上面的积分,我曾经尝试过围道积分,但是得到的结果错误。因此写信给Mathematic
al Methods for Physicists 的作者写信请教。得到的答案如下
The integrand has no singularities except at infinity. Therefore, the
residue theorem cannot be applied. I have been unable to integrate it
analyti
w****o
发帖数: 20
19
来自主题: Mathematics版 - Re: 一个积分问题
My Mathematica shows that:
Integrate[Exp[-x^2]*Sin[x^2]/x^2,{x,-Infinity,
Infinity}]=2^(5/4)*Pi^(1/2)*Sin[Pi/8]=1.613;
Integrate[Exp[-x^2]*Sin[x]^2/x^2,{x,-Infinity,
Infinity}]=Sqrt[Pi]*(-1+1/e+Sqrt[Pi]*Erf[1])=1.527;
I think you can do your integral analytically (integration by parts):
Exp[-x^2]*Sin[x]^2/x^2 dx
=-Exp[-x^2]*Sin[x]^2 d(1/x)
=...
Finally you can reduce the integrand into a linear combination of
Integrate[Exp[-x^2]*Sin[x]/x] and Integrate[Exp[-x^2]*Sin[x]], which you all
know how t
w**a
发帖数: 1024
20
来自主题: Mathematics版 - 问个stationary phase integration method
考虑这个积分,对x 从-inf. 到 +inf.
integrand :
exp[i*a*f(x)], when a is large,i = sqrt(-1)
f(x) is a real function
假如只有一个x0使得f'(x0)=0,那么人们经常就用x0附近的那段路径[x0-e,x0+e]代替整个
积分,然后用taylor expansion去代替原来的函数f(x),现在的问题是,为什么要用这个
点?难道仅仅是因为这一点的f(x)的值变化量最小吗?
w********s
发帖数: 59
21
I think GCQ (Gauss Chebshev quadrature) works well in your case. Do a
search. It is because the integrand tail vanishes very quickly, so the
partition of x should be chosen as non-uniform.

)/
w**a
发帖数: 1024
22
来自主题: Mathematics版 - delta function question
in physics, we know:
the integration of \exp{i*k*x}dx, i=\sqrt{-1}, on the real line from -\infty
to +\infty
is a delta function I=2*\pi*\delta(k).
The usual way to show this identity is to use
Fourier transform of a delta function then study its inverse transform.
Here is my question: integrate the same integrand, but from x=0 to +\infty.
Shall we get one-half of I (above value)? How to show this? I really want to
use distribution technique to show this. Thank you.
R******y
发帖数: 651
23
来自主题: Mathematics版 - mathematica 积分问题 (转载)
【 以下文字转载自 Physics 讨论区 】
发信人: RICIIkey (昵称太短!), 信区: Physics
标 题: mathematica 积分问题
发信站: BBS 未名空间站 (Sat Nov 8 18:53:40 2008)
f[x_, y_, z_] := x^2 + y^2 + z^2;
g[x_, y_] := NIntegrate[f[x, y, z], {z, 0, 1}]
NIntegrate[g[x, y]^2 + g[x, y]^(1/2), {x, 0, 1}, {y, 0, 1}]
出错信息是
NIntegrate::inum: Integrand f[x, y, z] is not numerical at {z} = {0.5`}
我只要NIntegrate 的结果,Integrate 是没问题的。
n**h
发帖数: 22
24
来自主题: Mathematics版 - stochastic integration
多谢楼上。现在integrand是一个gaussian process,所以不一定有continuous sample
path。但是知道它是tight的。能否从tight推出continuous sample path呢?
R******y
发帖数: 651
25
来自主题: Physics版 - mathematica 积分问题
f[x_, y_, z_] := x^2 + y^2 + z^2;
g[x_, y_] := NIntegrate[f[x, y, z], {z, 0, 1}]
NIntegrate[g[x, y]^2 + g[x, y]^(1/2), {x, 0, 1}, {y, 0, 1}]
出错信息是
NIntegrate::inum: Integrand f[x, y, z] is not numerical at {z} = {0.5`}
我只要NIntegrate 的结果,Integrate 是没问题的。
R******y
发帖数: 651
26
来自主题: Physics版 - 一个百思不得其解的积分问题
1) integral diverges
2) integrand is even!
3) Contour integral diverges at pole.
e**********n
发帖数: 359
27
来自主题: Physics版 - 问个弱问题
Integrand, including e^{ipr}, goes to zero in the upper half-plane. It does
not need Jordan lemma.
t**g
发帖数: 522
28
of course this integral diverges. Check your integrand to see if it is
correct.
c****e
发帖数: 2097
29
来自主题: Physics版 - 考考大家的数学
had to google what iid is. and I have no interest in this subject.
is there some jargon used? F(a) means probability for x<=a?
has to be integrated distribution, otherwise, the whole 'solution' makes no
sense.
unless you were talking about discrete distributions, of which normal
distribution doesn't belong according to wiki (so many jargons).
given the above interpretation, the formula:
F(a)*P(X2 > a) + (1-F(a))*P(X2 is stupid.
how can the person not consider the integrand mo
c******n
发帖数: 49
30
来自主题: Quant版 - 问两道probability的题
(

is
应该是pdf吧。
sigma2
这个跟上面一样, identify two integrands: a-x (xa+b-x); y-b (x and b
t**o
发帖数: 64
31
来自主题: Quant版 - an interview question
Write out integrals explicitly, change of variable to make the integrand a
total derivative.
j*****4
发帖数: 292
32
来自主题: Quant版 - 请教一个随机积分的题目
no,the integrand is an adapted process rather than a simple process.
but ito integral process has zero expectation,so Var(Y)=E(Y^2)
then apply Ito Isometry property.
k*******d
发帖数: 1340
33
来自主题: Quant版 - 菜鸟问一个积分
Why it is normal distributed? The integrand is not deterministic, I don't
see it immediately why it should be normal distributed.
Recall that \int_0^T B(t) dB(t) = 1/2 * (B(T)^2 - T), which is not normal
distributed.
u******s
发帖数: 157
34
来自主题: Quant版 - 菜鸟问一个积分
Is it the integrand has to be finite?
Can you give some intuition on counter examples? Thanks.
y********l
发帖数: 11
35
来自主题: Quant版 - one question urgent
Agree with idrunk. Shouldn't the integrand be:
E[e^{i a (w(T) - w(s)}]*E[e^{i a w(s)} w(s)]
= e^{a^2(T-s)/2}*E[e^{i a w(s)} w(s)]?
Thanks~
a*********r
发帖数: 139
36
来自主题: Quant版 - 问一道stochastic
First, SDE is another way to state the Ito integral. Since the integrand is
not stochastic, the integral is actually a one-dimensional Wiener integral.
The variance is readily computed from the well-known formula.
a*********r
发帖数: 139
37
来自主题: Quant版 - 问一道stochastic
Since the integrand is deterministic, it's a one-dimensional Wiener integral
which is normally distributed with mean 0 and variance \int_0^t\exp{-2rs}ds.
a*********r
发帖数: 139
38
来自主题: Quant版 - 问一道stochastic
Sure. One can find many different approach for this problem. For example,
write in the integral form then apply Ito isometry (which is actually an
overkill for a deterministric integrand), but the answer will be the same.
Advantage of Wiener integral for this problem: We get more than we need: we
also know X_t is normally distributed with mean 0 and variance given above.
However, using Ito isometry, we only know X_t is a random variable with mean
0 and given above.
x******a
发帖数: 6336
39
if we write $z=Rexp(i\theta)$, after changing variable,
then the absolute value of integrand is bounded by
$|exp(iz)|=exp(-R\sin\theta)$.
and the integral is from $0$ to $\pi$. now using dominated convergence
theorem to get 0.
h*****u
发帖数: 204
40
We know that T*W_T is normal
and int_0^T t dW_t is normal,(integrand is non-random)
If T*W_T and int_0^T t dW_t are independent.
then it is normal.
But T*W_T and int_0^T t dW_t are NOT independent.
Then I don't know.
and for this one I think T*W_T and int_0^T t dW_t are NOT independent.
So could you show me how do you get that? Thanks
L**********u
发帖数: 194
41
来自主题: Quant版 - [Ito formula] 一道题求解
1/dW_t^2-t is a martingale and the integrand should be interpreted as
2/d \sum_{s=1}^d W_s dW_s.
In Riemann Geometry, we always omit the sum notation by Einstein summation
convention
n******m
发帖数: 169
42
来自主题: Quant版 - 问道题目
先回答 ghlian:
概率是不一样,但是payoff的分布也不一样。当然,说“对称”是不妥的。
不过这两种情况确实是抵消的。
假设在某个时间 t1, barrier hit了,此时 S(t1)=H, 考虑后续发展,可能最后 S(T)
高于 H,(senario2) 也可能低于 H (senario3)
现在你对所有后续路径积分,integrand = payoff = (S(T)-H) (in both case), 积分
的结果就是
forward(S(t1))-H = S(t1)-H = 0
正解见 xuzhikai19 和 rrua,的 replication.
不过我现在被这种有决策参与的replication搞晕了,
你能用一个exercise price=100的forward replicate出这个barier option的payoff,
好像只能说明这个forward比这个option值钱阿??就像你可以选择永远不early
exercise, 那么american option的payoff也完全跟european的一样,但这不说明两者
同价。
是我哪个地方没弄... 阅读全帖
L*******t
发帖数: 2385
43
一时半会没想出来好办法,不负责任的说说一个不是很靠谱的吧,希望能抛砖引玉:
你的积分上下限是-a到a,可以转化成积分上下限是-无穷到+无穷,但是integrand要乘
上一个indicator function,区间是[-a,a],如果能验证NIG的Levy measure乘上这个
indicator function还是一个Levy Measure,那么你构造了一个新的Levy Process,(
不再是NIG了),然后用PDE方法求出这个积分式满足的PDE,然后看你的灵感了,能否
guess and verify一个closed form solution。
c******0
发帖数: 59
44
integrand 到底是什么?
我感觉你前后打了两个不同的式子..
m********1
发帖数: 368
45
来自主题: Statistics版 - 请教SAS IML调用DATA step数据的问题
向各位SAS IML达人请教一个SAS IML调用DATA step数据的问题。程序见下方。print
alpha beta可以正常输出,但是计算积分却报错,不知道是什么缘故?
注:我在写一个很复杂的宏,这是我简化的一个程序。我必须调用DATA eval里面的数
据,而不能直接在IML里面定义这些数据。
/* Define the Integrand in AUC Quaduature */
%MACRO TPR(FPR, alpha, beta);
1-1/(1+exp((log(&FPR./(1-&FPR.))+&alpha.)*exp(&beta./2)));
%MEND TPR;
DATA eval;
input alpha beta;
cards;
2.6728 -0.1220
;
RUN;
PROC IML;
use eval;
read all;
print alpha beta;
start fun(FPR);
TPR=%TPR(FPR, alpha, beta);
return(TPR);
finish;
/* Call QUAD */
bound = { 0 1
i**f
发帖数: 1195
46
来自主题: Statistics版 - 请教SAS IML调用DATA step数据的问题
add global option when you define a module..
try this:
/* Define the Integrand in AUC Quaduature */
%MACRO TPR(FPR, alpha, beta);
1-1/(1+exp((log(&FPR./(1-&FPR.))+&alpha.)*exp(&beta./2)));
%MEND TPR;
DATA eval;
input alpha beta;
cards;
2.6728 -0.1220
;
RUN;
PROC IML;
use eval;
read all;
print alpha beta;
start fun(FPR) global(alpha,beta);
TPR=%TPR(FPR, alpha, beta);
return(TPR);
finish;
/* Call QUAD */
bound = { 0 1 };
call quad(AUC, "fun", bound) eps=1E-10;
print AUC;
QUIT;
k******2
发帖数: 111
47
integrand 等于 p(\tilt{x},\vartheta|\mathcal{X})
积分是叫marginalized 这个joint 概率得到了你的公式了
l*********s
发帖数: 5409
48
let y=x+c, the integrand becomes (y-c)^n * exp(-y^2)
expand the polynomial (y-c)^n, to reduce the problem to linear combination
of integrations of the form integral_c_inf y^m * exp(-y^2), which again can
be solved using integration by parts and induction
1 (共1页)