Distribusi


7

Apakah ada ekspresi kanonik atau analitik untuk distribusi probabilitas untuk variabel acak kompleks sirkular-simetris Z:

Z=ejθ,
dimana θU(0,2π)?

Catatan samping:

Diketahui bahwa bagian nyata dan imajiner, yaitu:

(Z)=cosθ(Z)=sinθ
memiliki kepadatan marginal yang diberikan oleh :
f(Z)(z)=f(Z)(z)=1π1z2,1<z<1,
tetapi karena mereka tidak independen, menghitung PDF bersama mereka tidak trivial.

EDIT: Z berbeda dari normal kompleks di sini, amplitudo |Z|adalah deterministik dan identik 1, sedangkan jika Z normal kompleks, |Z| akan didistribusikan Rayleigh.


Bagaimana hal ini berbeda dari kompleks sirkular-simetris normal ?
Maxtron

1
@ Maxtron amplitudo di sini adalah 1, sedangkan kompleks normal memiliki amplitudo yang itu sendiri adalah variabel acak dengan distribusi Rayleigh.
Robert L.

@OlliNiemitalo punya jawaban untuk hal yang sama seperti yang saya ingat ...
Fat32

@ Fat32 Ya ini berdering. Pertanyaan lain adalah: Apa distribusi itu?
Olli Niemitalo

@OlliNiemitalo Ya itu pertanyaannya!
Fat32

Jawaban:


6

Karena bagian nyata dan imajiner sangat tergantung satu sama lain (jika Anda memiliki nilai satu, Anda tahu nilai yang lain persis), sepertinya Anda bisa menerapkan pdf marginal dari bagian nyata r, diberi nilai bagian imajiner i:

fri(r,i)=fr|i(r | i)fi(i)

Anda mencatat pdf dari bagian nyata dan imajiner secara individual:

fr(z)=fi(z)=1π1z2

Itu meninggalkan pdf marginal fr|i(r | i). Ingatlah bahwa untuk realisasi variabel acak yang diberikanZ, dua komponen terkait secara deterministik:

r2+i2=cos2(θ)+sin2(θ)=1

Dengan adanya hubungan ini, kita dapat menyelesaikannya r dengan kondisi i:

r2=1i2
r=±1i2

Oleh karena itu, pdf marginal dari r diberi nilai i adalah sepasang impuls:

fr|i(r | i)=12δ(r1i2)+12δ(r+1i2)

Menyatukan ini akan menghasilkan:

fri(r,i)=δ(r1i2)+δ(r+1i2)2π1i2

Memikirkan hal ini secara geometris, untuk setiap garis horizontal i=i0 (untuk i0[1,1]) dalam ri pesawat, hanya ada dua titik r0=±1i02itu bukan nol, dan pdf memiliki ketinggian tak terbatas pada titik-titik tersebut. Seperti yang kita harapkan, titik-titik persimpangan (yaitu titik di mana pdf adalah nol) adalah di mana garis horizontal bersilangan dengan lingkaran satuan!

Ini berarti bahwa pdf bersama bernilai nol, kecuali di sepanjang lingkaran unit, di mana ia mengambil ketinggian tak terbatas. Itu sejalan dengan intuisi, sebagai definisi dari variabel acakZ memastikan bahwa hanya dapat mengambil nilai yang ada di lingkaran unit.

Tidak ada yang istimewa tentang cara spesifik saya menguraikan ini; Anda juga dapat memindahkan masalah dan melihat garis vertikal diri bidang bentuk r=r0 dan Anda akan menemukan hubungan yang sama karena kopling dekat dari dua variabel acak.

Saya percaya formulasi ini setara dengan yang ada di jawaban AlexTP , tetapi turunannya mungkin lebih intuitif.


1
apakah ada salah ketik di r=1i2; mungkin maksudmur2=1i2? Juga dapat kami tunjukkan bahwa pdf gabungan terintegrasi ke 1 dalam bidang xy (atau eqv ri). Lagi-lagi juga pdf bersamafri(r,i) is not circularly symmetric? (the denominator?). Cannot we simply use the logic that given z=ejθ=cos(θ)+jsin(θ)=x+jy with θ uniform in [0,2π] and the constraint x2+y2=1 simply offers a circularly symmetric joint pdf of the type fxy(x,y)=Kδ2(x2+y21) or fR,θ=Kδ2(R1) ; a ring impulse on x-y plane? where K is possibly 1/2π ?
Fat32

1
I corrected the typos you referred to, thanks.
Jason R

@Fat32: I made some fixes that led me to what sounds like what you were getting at in your comment above. AlexTP's answer still is probably more intuitively pleasing though.
Jason R

1
There is a typo in copying fr|i(r|i) to fr,i(r,i). Dan saya perhatikan itufr,i(0,1)= berbeda dengan fr,i(1,0)=12π. Saya pikir mereka harus memiliki nilai yang sama, bukan?
AlexTP

1
semuanya, mungkin simulasi cepat akan bermanfaat untuk menunjukkan poin Anda :)
AlexTP

9

Hindari perhitungan yang rumit, biarkan X dan Ymenjadi variabel acak normal standar , variabel acak AndaZ memiliki distribusi yang sama antara V

V(XX2+Y2,YX2+Y2)
(easy to see V=1 and the angle of V is equivalent to the angle of a circularly symmetric Normal hence uniform).

This kind of V is one of the constructions of a point uniformly distributed on circle (which can be generalized to (n1)-sphere, see Sphere Point Picking and for example this answer).

Thus the PDF of Z is simply the reciprocal of the unit circle's circumference. For Zρ=ρejΘ with fixed ρ and uniform Θ,

in polar coordinates (where infinitesimal area is rdrdθ),

fR,Θ(r,θ)=12πδ(rρ)


2
i think this is correct.
robert bristow-johnson

The intuition seems to be so simple behind the proof. All you need to do is pick points from the surface of a unit sphere :)
Maxtron

Can't we just remove the ||z|| in the denominator? Otherwise, how would you evaluate the expression for ||z||0? Of course, we would like fZ(0)=0 to hold.
Matt L.

@MattL. you are right, it is wrong having z in the denominator, not due to the evaluation at limits but because it is simply wrong. I have fixed the final result.
AlexTP

1
@AlexTP: I still think that there shouldn't be any variable in the denominator. Have a look at my answer and please tell me if/where I go wrong.
Matt L.

4

Based on the existing answers, which opened my eyes for what's going on here, I would like to present yet another very simple expression for the solution, which is just slightly different from the one in AlexTP's answer (and which turned out to be equivalent to the one given in Jason R's answer, as shown below in the EDIT-part).

[EDIT: now that AlexTP has edited his answer, our expressions for the PDF are identical; so all three answers finally agree with each other].

Let the complex random variable Z=X+jY be defined as

(1)Z=ρejθ

where the radius ρ is deterministic and given, whereas the angle θ is random and uniformly distributed on [0,2π). I state without further proof that Z is circularly symmetrical, from which it follows that its probability density function (PDF) must satisfy

(2)fZ(z)=fZ(x+jy)=fZ(r),withr=x2+y2

i.e., it can be written as a function of the radius (magnitude) r.

Since the PDF must be zero everywhere except for r=ρ, and since it must integrate to unity (when integrated over the 2-dimensional plane), the only possible PDF is

(3)fZ(r)=12πδ(rρ)

It can be shown that (3) leads to the correct marginal densities for the random variables X and Y.


EDIT:

After some very useful discussion in the comments it appears that we've managed to agree on one solution to the problem. I will show in the following that the unassuming formula (3) is actually equivalent to the more involved looking formula in Jason R's answer. Note that I use r for the magnitude (radius) of the complex RV Z, whereas in Jason's answer r denotes the real part of Z. I will use x and y for the real and imaginary parts, respectively. Here we go:

(4)fZ(r)=12πδ(rρ)=12πδ(x2+y2ρ)

We know that δ(g(x)) is given by

(5)δ(g(x))=iδ(xxi)|g(xi)|

where xi are the (simple) roots of g(x). We have

(6)g(x)=x2+y2ρandg(x)=xx2+y2=xr

The two roots xi are

(7)x1,2=±ρ2y2

Consequently,

(8)|g(x1)|=|g(x2)|=ρ2y2ρ=1(yρ)2

With (5)-(8), Eq. (4) can be written as

(9)fX,Y(x,y)=12π1(yρ)2[δ(xρ2y2)+δ(x+ρ2y2)]

For ρ=1, Eq. (9) is identical to the expression given in Jason R's answer.

I think we can now agree that Eq. (3) is a correct (and very simple) expression for the PDF of the complex RV Z=ρejθ with deterministic ρ and uniformly distributed θ.


I think the determinant of Jacobian transform should be r then
θ=02πr=0+fZ(r)rdrdθ1
AlexTP

@AlexTP: Take as an extreme example ρ=0, which makes the RV Z deterministic, it's always zero. It's PDF should then be a Dirac impulse at r=z=y=0, which is what I get with the formula I suggested. With the formula in your answer we get an undefined term in that case.
Matt L.

it's true, but I think the value of PDF can be infinity and only the CDF should be bounded. In other words, the Dirac impulse matters only inside integrals.
AlexTP

Yes, but if you integrate your pdf (for ρ=0), what do you get (and how would you do it)?
Matt L.

I would do like this
limρ0+θr=0+12πρδ(rρ)rdrdθ=limρ0+θ12πρρdθ=θ12πdθ=1
The integral is finite while the PDF is not thanks to the Jacobian determinant. My point is that you need that Jacobian determinant r whenever you integrate the PDF in polar coordinate. Tell me if you find it unconvinced.
AlexTP
Dengan menggunakan situs kami, Anda mengakui telah membaca dan memahami Kebijakan Cookie dan Kebijakan Privasi kami.
Licensed under cc by-sa 3.0 with attribution required.