You are on page 1of 126

TFY-44.

130 Kvanttimekaniikka II

S. Virtanen 2007

Chapter 1

Tila-avaruusformalismin kertausta ja kehittely


Kvanttimekaniikan formalismissa kutakin fysikaalista systeemi vastaa tila-avaruus H = {| }, joka on Hilbert-avaruus:1 1. H on lineaariavaruus:2 a| + b| H a, b C, | , | H,

2. H:ss on mritelty kompleksinen sistulo eli pistetulo (| , | ) C. Yleens kytetn Diracin bracket-notaatiota | (| , | ). Ptee: (a) | = | , (b) |a1 + b2 = a |1 + b |2 , (c) | 0; | = 0 vain jos | = 0 (nolla-alkio)

Kohta (b) tarkoittaa, ett sistulo on toisen argumentin suhteen lineaarinen; kohdista (a) ja (b) seuraa toisaalta antilineaarisuus ensimmisen argumentin suhteen. Sistulo mritt lisksi avaruuden normin = | 0, ja sille ptee | | | (Cauchyn-Schwarzin epyhtl) + + (Kolmioepyhtl) 3. Avaruus on suljettu norminsa suhteen, eli kaikki Cauchy-jonot (joille ptee limn,m n m = 0) suppenevat.

1 Siis tila-avaruus on erilainen erilaisille systeemeille, mutta se on kuitenkin aina Hilbert-avaruus tai ns. laajennettu Hilbert-avaruustst tarkemmin myhemmin. 2 Eli vektoriavaruus; tst syyst sen alkioita sanotaan usein vektoreiksi.

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

Fysikaalisen systeemin tilaa tietyll ajanhetkell vastaa jokin tila-avaruuden normitettu ( | = 1) vektori | . Kysymyst Miten konstruoidaan annettua fysikaalista systeemi vastaava tila-avaruus eksplisiittisesti ksitelln myhemmin kappaleessa 1.7.

1.1 Bra- ja ket-vektorit Tila-avaruuden alkioita | H sanotaan ket-vektoreiksi. Olkoon | H. Nyt f (| ) = | on rajoitettu lineaarinen funktionaali H:ssa (lin. kuvaus H C s.e. |f (| )| < C | H jollekin C R). H:n rajoitettujen lin. funktionaalien joukkoa kutsutaan duaaliavaruudeksi H . Duaaliavaruus on helposti osoitettavissa lineaariavaruudeksi, ja sen alkioita (funktionaaleja) sanotaan kvanttimekaniikassa bra-vektoreiksi. Erityisesti f H , ja merkitsemme | f . Jokaista ket-vektoria | vastaa tmn mukaisesti bra-vektori |. Edellesitetyn mukaisesti ptee |(| ) f (| ) = | . (1.1)

Eli kun ketti | vastaavalla bra-vektorilla | operoidaan ket-vektoriin | , saadaan Diracin bracket (pistetulo) | C. Bra-vektoreiden notaatio on siis luonnollisella tavalla sopusoinnussa pistetulonotaation kanssa. Seuraava tulos osoittaa, ett Hilbert-avaruuksille knteisesti mys jokaista bra-vektoria vastaa ket-vektori: Frechet-Riesz: Jokaista H :n rajoitettua (=jatkuvaa) funktionaalia f vastaten on olemassa yksiksitteinen ket | ja tt vastaava bra-vektori | s.e. f (| ) = | kaikille | H. H:n ja H :n vektoreilla on siis 1-1 -vastaavuus. Erityisesti kaikki duaaliavaruuden alkiot voidaan esitt muodossa |, miss | H.

TFY-44.130 Kvanttimekaniikka II 1.2 Lineaariset operaattorit

S. Virtanen 2007

Lineaariset operaattorit ovat kuvauksia A : H H, usein merkitn lyhyesti A(| ) A| |A H, s.e. A(a| + b| ) = aA| + bA| . Tulo: (AB )| A(B | ) Kommutaattori: [A, B ] AB BA (| |)| = | | (operaattori, tyypillisesti = 0!) Ulkotulo: Ulkotulo | | tarkoittaa lin. operaattoria, jolle ptee34
( | a = a | )

(1.2)

| | .

(1.3)

avulla. Suomeksi: Selvsti |(A| ) on lineaarinen funktionaali H:ssa, joten Frechet-Rieszin mukaan on olemassa | s.e. |(A| ) = | .

Adjungaatti: Operaattoreiden operointi bra-vektoreihin (eli H:ss) voidaan mritell yhtln ( |A)| = |(A| ) C (1.4)

(1.5)

Nyt siis yhtliden (1.4) ja (1.5) mukaisesti asetetaan |A = |. Lisksi, koska bra- ja ket-vektoreilla on 1-1 -vastaavuus, mrittelee bra-vektoreihin operointi operaattorilla A lineaarikuvauksen A vastaaville ket-vektoreille s.e. A| = | . (1.6) Operaattori A on operaattorin A adjungaatti eli hermiittiskonjugaatti.5 Edellesitetyn perusteella saadaan A | = | = |(A| ) = |A .
(1.6) (1.5)

(1.7)

Vrt. sistulo eli bracket | . Huom. Operaattori voidaan mritell yksiksitteisesti kertomalla miten se operoi mielivaltaiseen H:n alkioon. 5 Periaatteessa Frechet-Riesz edellytt funktionaalin olevan mys rajoitettu, ja nin ei ole tss tapauksessa jos A ei ole rajoitettu. Tyypillisesti kuitenkin A on rajoitettu (eli A C ) H:n tihess osajoukossa, ja tmn avulla A voidaan konstruoida.
4

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

Siis oleellisesti kaikilla lin. operaattoreilla on Rieszin teoreeman mukaisesti olemassa adjungaatti, jolle ptee yhtl (1.7) kaikilla | , | H. Adjungaatille ptee: ( A ) = A (aA) = a A ( A + B ) = A + B (AB ) = B A (| |) = | | Erityisesti yhtlst (1.4) huomataan, ett Diracin merkint on assosiatiivinen, ja sulut voidaan poistaa kokonaan: Matriisielementit: |A| = ( |A)| = |(A| ). Hermiittiset operaattorit: Operaattori A on hermiittinen, jos A = A. Unitaariset operaattorit: Operaattori U on unitaarinen, jos U U = U U = I .6

Tss I on identiteettioperaattori; I | = | .

TFY-44.130 Kvanttimekaniikka II 1.3 Tila-avaruuden kannat

S. Virtanen 2007

Vektorijoukko {|i } H on lineaarisesti riippumaton, jos yhtl 7 i ci |i = 0 toteutuu vain jos ci = 0 kaikille i. Avaruuden dimensio on maksimaalinen lukumr lineaarisesti riippumattomia vektoreita. Vektorijoukko on tydellinen, jos mielivaltainen | voidaan lausua lineaarikombinaationa | = ci |i . (1.8)
i

Lineaarisesti riippumaton tydellinen vektorijoukko on avaruuden kanta. Kantavektorijoukolle esitys (1.8) on aina yksiksitteinen. Jos vektorijoukko on retn, her kysymys yllolevan summan konvergoinnista. Numeroituvasti retn kanta (jolle siis summat konvergoivat) on Schauderin kanta. Usein halutaan kytt ortonormitettuja kantoja, joiden vektoreille ptee i |j = ij . Tllaisille kannoille esityksen (1.8) kehityskertoimet saadaan suoraan projektioista kantavektoreille:8 ci = i | (todista!). Erityisesti ortonormaalille kannalle ptee siis | = i | |i
( a | = | a )

(1.9)

(1.10)

|i i | =

(1.3)

|i i | | . (1.11)

Koska tm ptee kaikille | H, saadaan ortonormaalin kannan vektoreille sulkeumarelaatio


i

|i i | = I.

(1.12)

7 8

Emme usein merkitse eksplisiittisesti indeksijoukkoase voi olla summalausekkeissa mik tahansa diskreetti joukko. Huomaa, ett tm siis ei pde, ellei kanta ole ortonormaali!

TFY-44.130 Kvanttimekaniikka II Jatkuvaindeksiset kannat:

S. Virtanen 2007

Koska tllin | = (0) = , on kantavektoreiden normi retn eivtk ne kuulu Hilbert-avaruuteen H. Tila-avaruutta on tmn vuoksi laajennettava sopivasti, jotta tllaiset kannat ovat matemaattisesti hyvksyttvitst laajennuksesta tarkemmin kappaleessa 1.5. Joka tapauksessa tllaisten jatkuvien kantojen tapauksessa voidaan mielivaltainen tilavektori esitt muodossa | = d | | = d | |

Usein kvanttimekaniikassa halutaan kytt mys jatkuvaindeksisi, numeroitumattomasti rettmi kantoja, esim. {| }, R. Tyypillisesti niden kantojen vektorit ovat ortonormaaleja Diracin delta -distribuution mieless, eli ptee (1.13) | = ( ).

(1.14)

ja tten erityisesti ptee sulkeumarelaatio d | | = I (Vrt. yhtl (1.12) diskreetille ortonormaalille kannalle). (1.15)

1.4 Esitykset Kaikki laskut tila-avaruuden vektoreilla ja operaattoreilla voidaan tehd kytten mielivaltaista kantaa vastaavaa esityst. Esityksess oleellisesti projisoidaan tilat ja operaattorit jollekin kannalle, ja suoritetaan laskut nin saatavia projektiokertoimia, jotka ovat yksiksitteisi, kytten. Olkoon {|i } ortonormaali kanta. Sulkeumarelaatiota (1.12) kytten saadaan | = I | = ja A = IAI =
ij i

ci |i ,

ci = i | Aij = i |A|j .

(1.16)

Aij |i j |,
6

(1.17)

TFY-44.130 Kvanttimekaniikka II Nyt esim. | = A| = =


ij

S. Virtanen 2007

ijk

Aij |i j |ck |k =
i

ijk

Aij ck jk |i (1.18)

Aij cj |i =

ci |i

eli operaattorilla operoiminen voidaan esitt projektiokertoimien matriisikertolaskuna: ci = Aij cj .


j

(1.19)

Jos esityskanta on jatkuva (esim. yhden hiukkasen systeemin paikkaesityksess kanta {|r }), saadaan tietysti mys projektiokertoimista jatkuva-argumenttisia funktioita: (r) = r| , A(r, r) = r|A|r , (1.20)

ja summaukset korvautuvat integraaleilla. Tss erityisesti (r) on hiukkasen aaltofunktio paikkaesityksess.

1.5 Operaattoreiden ominaisarvot ja -vektorit Jos ja | = 0, on | operaattorin A ominaisvektori ja vastaava ominaisarvo. Operaattorin ominaisarvojen joukkoa kutsutaan sen spektriksi. Ominaisarvoa vastaavaa ominaisvektoreiden joukkoa kutsutaan :n ominaisavaruudeksi. Ominaisavaruudet ovat suljettuja summan ja skalaarilla kertomisen suhteen, joten ne ovat H:n aliavaruuksia, jos nollavektori lasketaan niihin mukaan. Ominaisavaruuksien dimensioita sanotaan vastaavien ominaisarvojen degeneraatioiksi. Hermiittiset operaattorit Kvanttimekaniikka I:ss osoitettiin, ett hermiittisten operaattoreiden ominaisarvot ovat reaalisia ja eri ominaisarvoja vastaavat ominaisvektorit
7

A| = |

(1.21)

TFY-44.130 Kvanttimekaniikka II ortogonaalisia, eli9

S. Virtanen 2007

Koska avaruuteen H kuuluvat ominaisvektorit voidaan aina normittaa ykksnormiin, ja samaa ominaisarvoa vastaava ominaisvektorikanta voidaan ortonormittaa Gram-Schmidt -menetelmll, voidaan hermiittisen operaattorin ominaisvektoreista muodostaa ortonormitettu joukko. Kokonaan eri asia on, onko tm joukko tydellinen, eli voidaanko hermiittisen operaattorin ominaisvektoreista muodostaa H:n ortonormitettu kanta. Tm olisi erittin haluttavaa kvanttimekaniikan formalismin todennkisyystulkinnan kannalta. Tyypillisesti kvanttimekaniikan oppikirjat postuloivat tmn aina onnistuvan, mutta valitettavasti asia on kaikkea muuta kuin selvi. Esim. vapaan hiukkasen liikemroperaattorilla ei ole H:ssa ainutta ominaistilaa, koska paikkaesityksen ominaisfunktiot p (r) eipr/ ovat normiltaan rettmi eivtk vastaavat tilavektorit kuulu tila-avaruuteen H. Ongelma voidaan ratkaista laajentamalla tila-avaruutta H siten, ett mukaan tulevat kaikki sellaiset vektorit, joiden pistetulo H:n jonkin aidon osajoukon (tm testifunktio-osajoukko sislt paikkaesityksess vain rettmn sileit ja oleellisesti vhintn eksponentiaalisen nopeasti rettmyydess vaimenevia funktioita) alkioiden kanssa on rellinen. Tm sallii uusien vektoreiden olevan normiltaan rettmi, kunhan ne eivt ole sit liian patologisesti (esim. eksponentiaalisesti rettmyydess divergoivat aaltofunktiot eivt sislly laajennettuun avaruuteen, mutta liikemrn ja paikan ominaistilat eli tasoaallot ja delta-funktiot sisltyvt). Nin saadaan laajempi, ns. rikitetty Hilbert-avaruus H , joka ei ole kuitenkaan en tavallisessa mieless Hilbert-avaruus, koska pistetulo ei ole rellinen kaikkien alkioiden vlill. Nyt ns. yleistetty spektraaliteoreema kertoo, ett kaikilla hermiittisill operaattoreilla on olemassa tydellinen (ts. koko avaruuden virittv) ominaiskanta (ts. ominaisvektoreista koostuva kanta) rikitetyss Hilbertin avaruudessa.10
9 Usein normitettuja ominaisvektoreita merkitn lyhyesti ket-symbolilla, joka sislt vain vastaavan ominaisarvon, mikli esim. degeneraatio ei aiheuta sekaannusta. Siis esim. A| = | jne. 10 Tarkempia yksityiskohtia kurssilla Muutaman kappaleen systeemit. . .

| = 0,

jos = .

(1.22)

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

Esim.: Liikemrn ominaistilat |p , joille r|p = p (r) = (2 )3/2eipr/ kuuluvat rikitettyyn Hilbertin avaruuteen, koska p| dr eipr/ (r) < , (1.23)

kaikille testifunktioille (r) (sileit + vaimenevat oleellisesti eksponentiaalisesti rettmyydess). Lisksi nm ominaistilat muodostavat H:n jatkuvaindeksisen kannan, jolle ptee ortonormitusrelaatio p|p = (p p ). (1.24)

Summa summarum: Kaikilla hermiittisill operaattoreilla on olemassa tydellinen ominaiskanta, kun tila-avaruutta laajennetaan sopivasti. On kuitenkin huomattava, ett esim. skalaaritulo ei ole rajoitettu koko laajennetussa avaruudessa H , joka sislt nyt tiloja joille | = . Mutta erityisesti fysikaaliset tilat kuuluvat aina Hilbert-avaruuteen H H ja niiden normi on rellinen. Kommutoivat hermiittiset operaattorit Perustulos: Olkoon A, B , C ,. . . joukko hermiittisi operaattoreja, joilla on olemassa tydellinen ominaisvektorikanta ja jotka kommutoivat pareittain keskenn: [A, B ] = 0, [A, C ] = 0, [B, C ] = 0, jne. (1.25) Tllin ptee: On olemassa ortonormaali kanta {|i }, jonka vektorit ovat kaikkien niden operaattorien yhteisi ominaisvektoreita, eli A|i = ai |i , B |i = bi |i , C |i = ci |i , jne. (1.26)

Tllaisen kommutoivan operaattorijoukon sanotaan olevan tydellinen, jos yhteinen ominaisvektorikanta on degeneroitumaton. Tllin kantavektorit voidaan spesioida yksiksitteisesti pelkstn ominaisarvojen avulla. Usein tllin merkitn |i = |ai bi ci . Jos kommutoiva operaattorijoukko ei ole tydellinen, siit saadaan tydellinen lismll siihen sopivia operaattoreita (Mieti miksi!).
9

TFY-44.130 Kvanttimekaniikka II 1.6 Kvanttimekaniikan postulaatit

S. Virtanen 2007

Fysikaalinen systeemi Laajennettu Hilbert-avaruus H . Mitattava suure A Lin. op. A : H H , A = A

Systeemin tila | H, | = 1 (modulo ei )

Tss ei viel oteta kantaa siihen, miten annetun systeemin tila-avaruus muodostetaan ja mitk operaattorit vastaavat mitattavia suureita. Tt kytkent reaalimaailmaan ksitelln kappaleessa 1.7. Pelkstn kompleksivaihetekijll eroavat vektorit vastaavat samaa fysikaalista tilaa. Siis fysikaalinen tila vastaa itse asiassa tila-avaruuden sdett (engl. ray) eli vektorijoukkoa {ei | }, R. Yleisemmin systeemin tila mritelln tilaoperaattorin avulla. Tilaoperaattori redusoituu kyttmmme systeemin tilan mritelmn ns. puhtaille tiloille, mutta sen avulla voidaan esitt mys ns. sekatiloja, jotka koostuvat jonkin todennkisyysjakauman mukaisesti sekoituksesta puhtaita tiloja. Tilaoperaattoriksittely on tarpeen esim. tarkasteltaessa systeemej, joiden vuorovaikutukset ympristn kanssa sekoittavat puhtaan tilan. Aiheesta lis statistisen fysiikan ja muutaman kappaleen kvantti-ilmiiden kursseilla.

Fysikaalista suuretta A mitattaessa voidaan saada vain arvoja, jotka ovat vastaavan operaattorin A ominaisarvoja.

10

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

Mitattaessa operaattoria A vastaavaa suuretta systeemin ollessa tilassa |

gn 1. Jos A:n spektri on diskreetti, ja {|ui n }i=1 on gn -kertaisesti degeneroitunutta ominaisarvoa an vastaavan ominaisaliavaruuden ortonormaali kanta, on todennkisyys P (an ) saada mittaustulokseksi an gn

P (an ) =

i=1

2 | ui n | | .

(1.27)

2. Jos A:n spektri on jatkuva (ja ei-degeneroitunut, yksinkertaisuuden vuoksiyleistys helppo), ja |u ominaisarvoa vastaava ominaisvektori jolle ptee u |u = ( ), on todennkisyys dP () saada mittaustulokseksi arvo vlilt ja + d dP () = | u | |2 d. (1.28)

On huomattava, ett ominaistilojen |u (ns. ltteritilojen) ei tarvitse olla normittuvia, kunhan ne kuuluvat laajennettuun H:een. Sen sijaan fysikaalisen tilan | tytyy aina kuulua avaruuteen H. Jos operaattoria A vastaavan suureen mittaus tilassa | antaa tulokseksi arvon an , on systeemin tila vlittmsti mittauksen jlkeen | = Pn | , Pn (1.29)

n i i miss Pn = g i=1 |un un | on projektio-operaattori ominaisarvon an ominaisavaruudelle.

Mittaustapahtumien vlill systeemin tilan aikakehitys mrytyy Schrdingerin yhtlst i d | (t) = H (t)| (t) , dt (1.30)

miss operaattori H (t) on systeemin kokonaisenergiaa vastaava operaattori.

11

TFY-44.130 Kvanttimekaniikka II 1.7 Construction of the State Space

S. Virtanen 2007

The general formalism of quantum mechanics described up to now is valid for all systems. But how to construct the state space H and the operators corresponding to physical quantities for a given system?
1.7.1 A particle interacting with external elds

Let us rst consider a system consisting of only one particle (mass m) moving in given external elds. The particle has three (external) degrees of freedom corresponding to the spatial position of its center of mass. In addition the particle may have independent internal degrees of freedom, such as spin. The external and internal degrees of freedom are measurable, hence we require as a starting point that the corresponding Hermitian operators exist. Let r = ( x, y , z ) be the operators corresponding to the position of the center = (S x, S y , S z ) to the spin state of the particle (we know that of mass, and S is an angular spin is an internal angular momentum of the particle, hence S momentum operator). Especially, we assume the following: The three center-of-mass operators commute and they have a continuous spectrum consisting of all vectors in the three-dimensional space:11 r|r = r|r , r R3 . (1.31)

The spin operators are angular momentum operators: x, S y ] = i S z [S etc. (1.32)

The internal and external degrees of freedom are independent, i.e. the operators related to them commute: ] = 0. [ r, S
11

(1.33)

Essentially, this means that the space is a continuum with independent x, y and z degrees of freedom. In attempts to quantize gravitation this assumption has to be relaxed.

12

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

The common eigenstate of all these operators is uniquely dened. 2 and S z form a complete commuting set Now since for example operators S for the spin degrees of freedom, and a given particle type is always in the 2,12 the spin state of the particle is completely specied same eigenstate of S by the z -component of the spin. It is thus natural to assume that the set z } is a complete commuting set of operators for the one-particle {x , y , z , S system. Consequently, their common eigenstates |r, , such that r|r, = r|r, , z |r, = |r, , S (1.35) span the state space for the system composed of one particle. Especially, the set {|r, } is the basis vector set for the coordinate representation. The projection of a ket onto this basis (r, ) = r, | r, | r|
( r = r)

r R3 , = s, s + 1, . . . , s

(1.34)

(1.36)

is the wave function of the state in the coordinate representation. Since = r r, | , (1.37)

the position operator operates in the coordinate representation as13 r ( r, ) = r ( r, ) . (1.38)

Its eigenfunctions r0 (r) are proportional to the Dirac delta functions (r r0), and since dr (r r0) (r r0) = (r0 r0), the normalization choice r0 (r) = (r r0 ) (1.40) (1.39)

2 with eigenvalue 2 s(s + 1) = 3 2 . particle is always in an eigenstate of S For example, spin- 1 2 4 For convenience, we usually denote an operator in H and its representations with the same symbol, although mathematically they are different operators.
12 13

13

TFY-44.130 Kvanttimekaniikka II yields the convenient orthonormality relation r, |r , = (r r ) . Especially, this implies the closure relation dr |r, r, | = I.

S. Virtanen 2007

(1.41)

(1.42)

Okay, now we have constructed the state space (rigged, it turns out) for the one-particle system. How about the operators corresponding to physical quantities other than the position of the center of mass and the spin? It turns ), linear out that the basic operators for energy (the Hamiltonian H ) can be constructed by requiring ) and angular momentum (J momentum (p the formalism to be Galilean invariant, i.e. that the same formalism is valid in all inertial frames (and, with suitable modications, in rotating frames). For example, one nds out that = p i in the coordinate representation.14 Especially, the Hamiltonian operator for a one-particle moving in an external electromagnetic eld described by the vector potential A(r, t) and the scalar potential (r, t), and in an additional external potential V (r, t), can be shown to be of the form 2 1 q A( r, t) + V ( r, t); (1.44) H (t) = p r, t) + q( 2m c here q is the charge of the particle, and c the velocity of light; we use the Gaussian system of units. In the coordinate representation the Hamiltonian takes the form q (t) = 1 A(r, t) H 2m i c
14

(1.43)

+ q(r, t) + V (r, t).

(1.45)

See for example Ballentine: Quantum Mechanics for details.

14

TFY-44.130 Kvanttimekaniikka II
1.7.2 Composite systems and tensor product spaces

S. Virtanen 2007

Having constructed the state space and physically important operators for one-particle systems, we consider the generalization to many-particle systems. Mathematically, this can be done by glueing several one-particle state spaces together with the so-called tensor product. To see what kind of properties the many-particle system state space must satisfy, let us consider a system that has (at least) two independent degrees of freedom. These can be for example the position coordinate x1 of particle one in the x-direction, and the position coordinate x2 of particle two. In quantum mechanics, the term independent means that the corresponding operators x 1 and x 2 commute, i.e. [ x1 , x 2 ] = 0. (1.46) This is intuitive in the sense that for independent degrees of freedom one must be able to prepare quantum states in which for example particle 1 is located very precisely (with small variance) at some point and particle 2 at another point. If the two operators would not commute, this would be impossible because of the Heisenberg uncertainty relation. Now, let {|x1, 1 (1) } be a complete eigenbasis for the state space H(1) of particle 1, and correspondingly {|x2 , 2 (2) } a complete eigenbasis for the state space H(2) of particle 2. Here for example |x1, 1 (1) is an eigenstate of x 1 with eigenvalue x1, and simultaneous eigenstate of other operators needed to form some complete commuting set of operators for H(1) . Because of (1.46), the state space H of of the whole system must contain states that are simultaneous eigenstates of x 1 and x 2. I.e., it must contain states |x1, x2, such that x 1|x1, x2, = x1|x1, x2, , x 2|x1, x2, = x2 |x1, x2, . (1.47)

This fundamental property that systems with several independent degrees of freedom have can be realized by constructing their state spaces H as tensor products (or Kronecker products) of the state spaces H(1) and H(2) of the subsystems. The tensor product vectors are of the form |
(1)

(2)

(1)

H(1) , |

(2)

H(2) ,

(1.48)

15

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

and the tensor product space H consists of all linear combinations of tensor product vectors. We denote also H = H(1) H(2) . Properties of the tensor product: The tensor product is linear with respect to multiplication by complex numbers: (| ) | = | (| ) = (| | ), and distributive with respect to both arguments separately: | (|1 + |2 ) = | |1 + | |2 , but note that in general (|1 + |2 ) | = |1 | + |2 | , (1.50) (1.51) (1.52) (1.49)

|1 |1 + |2 |2 = (|1 + |2 ) (|1 + |2 ).

From these properties it follows that if {|i (1) } is a basis of H(1) and {|j (2) } is a basis of H(2) , then {|i (1) |j (2) } is a basis of H = H(1) H(2) (see exercises).

Implication: If the dimension of H(1) is N and the dimension of H(2) is M , the dimension of H is MN .

Note that not all vectors in H are of the factorized form (1.48)but they are all sums of factorized vectors. Scalar product in H may be dened as a linear mapping that yields for factorized kets | = |u (1) |v (2) and | = |u (1) |v (2) | = u|u
(1)

v |v

(2)

(1.53)

The scalar product is extended for non-factorized kets by linearity. Operators: If A(1) and B (2) are linear operators in H(1) and H(2) , respectively, then C = A(1) B (2)
(2)

(1.54) ) (B (2) |v
(2)

is a linear operator in H dened by the rule (A(1) B (2) )(|u


(1)

|v

) = (A(1) |u
16

(1)

).

(1.55)

The operation on non-factorized kets is determined by linearity.

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

Notation: The superscripts referring to subsystems are often dropped, as is also the tensor product symbol. Thus one often denotes | for example.
1.7.3 Many-particle systems
(1)

(2)

| | | | ,

(1.56)

Now, by using the principle presented above to construct state spaces for composite systems, it is straightforward to construct the state space for a system composed of N particles: such a system can be decomposed to N one-particle systems with independent degrees of freedom, i.e. the state space is spanned by vectors |r1 , 1 (1) |r2 , 2 (2) |rN , N |r1 , 1 |r2 , 2 |rN , N |r1 , 1, r2, 2, . . . , rN , N ,
(N )

(1.57)

which are common eigenvectors for the position operators r(1) , r(2) ,. . . , r(N ) (1) (2) (N ) z z 2)(1) etc.) of the individual and spin operators S , Sz ,. . . , S (and (S particles.15 If the particles are noninteracting, the Hamiltonian operator is of the form =H (1) I (2) I (N ) H (2) I (N ) + I (1) H . . . (N ) , + I (1) I (2) H

(1.58)

(i) are the one-body Hamiltonians for the particlesusually one just where H writes in shorthand =H (1) + H (2) + + H (N ) . H (1.59)

In fact, because the x, y and z coordinates and the spin of the individual particles are also independent degrees of freedom, we can construct even the one-particle state spaces as tensor products. For example we can write |r1 , 1 = |x1 |y1 |z1 |1 |x1 |y1 |z1 |1 etc. Here for example the set {|1 } spans the 2s + 1 -dimensional Hilbert space of the spin degree of freedom for particle 1, and these are the objects we have called spinors in Kvanttimekaniikka I.

15

17

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

Interactions between particles can be taken into account as external potentials for each individual particle: For example, for the Coulomb interaction the electric eld generated by N 1 other particles can be considered as an external eld for each particle, and taken into account as in (1.44). Especially, in the coordinate representation a state vector is represented by the wave function16 (r1 , 1, r2, 2, . . . , rN , N ) = r1 , 1, r2, 2, . . . , rN , N | , (1.60)

and in the absence of vector potentials the Schrdinger equation takes the form (neglecting also spins)
N n=1

2 2 + U (rn, t) + V (r1, r2, . . . , rN ) (r1 , r2, . . . , rN , t) 2mn n n=1 (r1, r2, . . . , rN , t). t (1.61)

=i

Here U (rn , t) denotes a common external potential for the particles, and V (r1, r2, . . . , rN ) an interaction potential between the particles (which is usually time-independent).17

Note that the wave function is not a wave in the ordinary three-dimensional space, but rather a wave in an abstract 3N -dimensional conguration space. 17 Note that a factorized vector | = |u |v has the coordinate representation (r1 , r2 ) = u(r1 )v (r2 ), and the wave function space is spanned by such factorized functions.

16

18

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

1.8 Equivalent Formulations of Quantum Mechanics


1.8.1 Unitary transformations of states and operators

From the postulates of quantum mechanics it follows that all physical predictions of the theory (possible results of measurements and their probabilities) are given by the eigenvalue spectra of operators and the absolute values of scalar products between ket vectors. All other information is physically irrelevant (the complex phase factors of the state vectors, for example). Thus, any new choice of the operators and the state vectors is equivalent to the old formalism we have developed if: 1. The new operators corresponding to physical quantities have the same eigenvalue spectra as the old operators, 2. The scalar products of new vectors representing a physical state with the eigenvectors of the new operators have the same absolute values as the corresponding old scalar products. These conditions are satised if we transform the old vectors and operators with a unitary transformation in the following way:18 | = U | , Here U is a unitary operator, i.e. U U = U U = I. (1.63) | = |U , A = U AU . (1.62)

It is easy to see that scalar products and the spectra of operators remain invariant in this transformation: For example, if | = U | , are any two vectors, it follows | = U |U = |U U | = |
Vectors and operators with primes are new ones, and those without primes old ones.

| = U |

(1.64)

(1.65)

18

19

TFY-44.130 Kvanttimekaniikka II and

S. Virtanen 2007

|A | = |U U AU U | = |A| . On the other hand, if it follows A| = | ,

(1.66)

(1.67)

A| = (U AU )(U | ) = U A| = U | = | , and it follows that the spectra of A and A are identical.


1.8.2 Heisenberg and Schrdinger pictures

(1.68)

The unitary transformation yielding an equivalent formulation of quantum theory may be chosen to be time-dependent in such a way, that the transformation exactly cancels the time-dependence of the state vector |S (t) of the systemthe subscript S here stands for Schrdinger picture which is the choice of kets and operators for physical states and quantities we have used so far. Naturally, as the time-dependent transformation removes the time-dependence of state vectors, we will see that it simultaneously renders most of the operators time-dependent. However, the resulting formalism is completely equivalent to the Schrdinger picture and is called the Heisenberg picture. In our familiar Schrdinger picture, the time-dependence of the state vector is determined by the Schrdinger equation 1 d |S (t) = HS (t)|S (t) . (1.69) dt i If the initial state at time t0 is |S (t0 ) , we may dene a time evolution operator U (t2 , t1) such that |S (t) = U (t, t0)|S (t0) . (1.70)

Inserting this formal solution into the Schrdinger equation, we nd that U (t, t0) satises the equation 1 U (t, t0) = HS (t)U (t, t0) t i
20

(1.71)

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

with the initial condition U (t0, t0 ) = I . Using this operator equation and its adjoint, it follows that 1 U U HS U ) . ( U U ) = ( U HS t i (1.72)

(t), the right hand side vanishes, and we nd that If and because HS (t) = HS

U (t, t0)U (t, t0) = U (t0 , t0)U (t0, t0) = I,

(1.73)

and similarly that U (t, t0)U (t, t0 ) = I , i.e. the time-evolution operator is unitary at all times. This unitary operator (or its adjoint, to be exact) is used to transform the formalism, as described in subsection 1.8.1, into the Heisenberg picture. From now on we x the arbitrary initial time t0 = 0, and write U (t) = U (t, 0). The state vector |S (t) of the system is transformed to |H = U (t)|S (t) = U (t)U (t)|S (0) = |S (0) ,
(1.62) (1.70)

(1.74)

(1.75)

i.e. in the new Heisenberg picture (denoted here by the subscript H) the state vector of the system is time-independent and equals the state vector at a xed time t = 0! On the other hand, the operators are transformed time-dependently as AH (t) = U (t)AS (t)U (t), (1.76) which typically renders even previously time-independent operators time-dependent. Essentially, one may conclude that typically in the Schrdinger picture the state vector is time-dependent and operators time-independent, and vice versa in the Heisenberg picture. The time evolution of the state vector is trivial in the Heisenberg picture, but

21

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

let us calculate the equation of motion for the operators. Eq. (1.76) implies d d d d AH (t) = U (t) AS (t)U (t) + U (t) AS (t) U (t) + U (t)AS(t) U (t) dt dt dt dt 1 d (1.71) = U (t)HS(t)AS (t)U (t) + U (t) AS (t) U (t) i dt 1 + U (t)AS (t)HS(t)U (t). i (1.77) By inserting the identity operators I = U (t)U (t) between HS (t)s and AS (t)s and by using the transformation relation (1.76), we nally get i d AH (t) = [AH (t), HH(t)] + i dt d AS (t) dt .
H

(1.78)

This is the fundamental equation essentially determining the dynamical evolution of the systemit is the counterpart of the Schrdinger equation in the Heisenberg picture. For conservative systems HS = HS (t), and (1.71) can be easily integrated to yield the solution U (t) = eitHS/ . (1.79) This implies that for any operator AS (t) commuting with the Hamiltonian, i.e. [AS (t), HS] = 0, we get AH (t) = U (t)AS(t)U (t) = eitHS / AS (t)eitHS/ = eitHS / eitHS / AS (t) = AS (t); (1.80) especially, for a conservative system HH = HS = H = const. (1.81)

22

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

Chapter 2

Quantum Theory of Scattering


Literature: Schwabl, chapter 18 Merzbacher, chapter 11 (detailed treatment) 2.1 Motivation and Problem Setting Many experiments in physics consist of colliding a beam of particles of type A with a target consisting of particles of type B, and measuring the properties of particles resulting from the collisions. Such experiments provide crucial information about elementary particles and interactions between them, but collision experiments play an important role also in atomic, molecular and nuclear physics, in which the colliding particles are often composite particles. Different types of collisions: 1. Rearrangement collisions: Resulting particles are of different type than the initial particles A and B. 2. Scattering collisions: Resulting particles are of the same type than the initial particles. However, if the A and B are composite particles

23

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

,AJA?J

5?=JJAHA@F=HJE? AI

?E@A

J>A=

6H= 6=HCAJ I EJJA@>A=

Figure 2.1: Geometry of a scattering experiment.

(atoms, for example), the nal particles can be in different internal states (different energy eigenstates) than the initial ones. (a) Elastic collisions: Internal states of the particles are not changed total kinetic energy of the particles conserved

(b) Inelastic collisions: Internal states changed total kinetic energy not conserved

The basic setup for collision experiments is shown in Figure 2.1. A beam of particles directed along the z -axis collides with the target particles. As a result of the interaction of the beam with the target, some particles are scattered at angles (, ) = (0, 0) from the origin point dened to be at the target, and the nal angular distribution of particles is measured by a detector. The most important quantity characterizing scattering is the differential cross section (, ), that can be measured directly and that we also want to calculate theoretically. Suppose that a ux of ji particles per unit area and unit time is incident on the target. The number of particles scattered in unit time into a cone with solid angle d about the direction (, ) is obviously proportional to ji d, and is written as ji (, )d, which denes the differential cross section. Let js (r, , ) be the ux of scattered particles far from the target. Since the area subtended by solid angle d at distance r is r2 d, we nd ji (, )d ji (, ) js (r, , ) = = , (2.1) r 2 d r2
24

TFY-44.130 Kvanttimekaniikka II which implies1

S. Virtanen 2007

r2js (r, , ) . (2.2) (, ) = ji Essentially, (, ) is proportional to the probability density for a particle to scatter into direction (, ). As seen from Eq. (2.2), (, ) has the dimension of an area ([m2 ]). When integrated over a nite solid angle, it can be interpreted as the cross-sectional area (with respect to the incoming particle ux) from which all particles scatter into this solid angle. The scattering probability to any directions = 0 is proportional to the total cross section = (, )d. (2.3)
4

For example, it will be seen in the exercises that the total cross section of a hard sphere with radius a is = a2 for classical Newtonian scatteringthis is natural, since the total cross-sectional area of the ball is just a2 . In accordance with the notation for the total cross section, the differential d cross section is also often denoted as d .
2.1.1 Idealization of a scattering experiment

In order to avoid complications arising from modelling the most general situation, we assume the following simplications and idealizations: We consider only elastic scattering and assume that the interaction between incoming and target particles can be modelled with spin-independent interaction potential V ( r1 , r2 ) = V ( r1 r2 ) . Thus we can neglect the spins of the particles.2 We assume that the density of the incident beam is low enough such that the wave packets of the incoming particles do not overlap (
Despite of the r appearing in Eq. (2.2), (, ) is independent of r , because js (r, , ) 1/r 2 . In reality, spins typically have to be taken into account. . .

(2.4)

25

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

incident particles do not interact). Hence we can consider only what happens to one wave packet (one particle) approaching the target. The average distance of particles in the target is assumed to be large compared to the dimensions of the approaching wave packet. This amounts to neglecting simultaneous coherent scattering from many target particles (cf. neutron scattering experiments from a crystal lattice). However, the approaching wave packet is assumed to be much larger than the size of one target particle.3 Furthermore, the target is assumed to be thin enough that multiple scattering, i.e. subsequent scattering from several target particles, is negligible. All in all, according to these assumptions we can restrict to consider only scattering of one incoming wave packet from one target particle. The detectors are assumed to be far from the target, such that they essentially see the target as one point. Consequently the nite size of the target can be neglected in determining the angular distribution of scattering, and we may restrict to consider scattering from one target particle initially located at the origin.
2.1.2 Direct and inverse scattering problems

The two kinds of scattering problems: A. Direct scattering problem: V (r1 r2) is known, calculate (, ), B. Inverse scattering problem: (, ) is known, calculate V (r1 r2 ). Typically, the inverse scattering problem is much more difcult and subtle than the direct problem, and in the following we analyse only the latter.
In the limit that the sizes of the wave packets of the colliding particles approach zero, we would be able to describe the problem in terms of classical mechanics. In that limit the impact parameter b, i.e. the distance of the incoming particle from the z -axis, is crucial in determining the result of scattering.
3

26

TFY-44.130 Kvanttimekaniikka II 2.2 How to Solve the Direct Problem?

S. Virtanen 2007

According to the above-mentioned simplications, the problem is reduced to a two-particle problem with the Hamiltonian H
(2) 2

2m1

2 1

2m2

2 2 + V ( r1 r2 )

(2.5)

determining the time-evolution; here 1 refers to the incident and 2 to the target particle.4 As the interaction depends only on the relative displacement of the particles, it is convenient to transform to the center of mass (CM) coordinates, in which the problem reduces to considering the relative Hamiltonian5
2

2 where r = r1 r2 and is the reduced mass of the particles. The CM equation can be forgot, but nally the results have to be converted to the laboratory frame.6 So how to calculate scattering of a wave packet from a target particle?

H = Hrel =

2 + V ( r) ,

(2.6)

1. Initial condition: At time t = 0 a wave packet centered at r0 , a point near the negative z -axis far left from the origin: (r, 0) = dk 0 (k)eikr = dk 0 (k)eikr . (2.7)

This is the expansion of the initial state in the plane wave basis. Here 0 (k) is smooth function of narrow width k, centered about k0 z (the momentum of the particle is essentially k0). Recall that the width of the wave packet in real space is approximately r 1 . k (2.8)

4 In fact, this Hamiltonian applies to the case in which the target really consists of only one particle. If the target consists for example of atoms bound tightly into a solid state lattice, the kinetic energy term of the target particle should be omitted (essentially, in that case the effective mass of the target atom is the mass of the whole target lattice of atoms). 5 See the lecture notes of Kvanttimekaniikka I. 6 That is a straightforward kinematic problem, see Liboff sec. 14.3.

27

TFY-44.130 Kvanttimekaniikka II 2. Calculate the time-development of (r, t) from

S. Virtanen 2007

(r, t) = H (r, t). t Noting that the Hamiltonian is time-independent, we recall that i (r, t) =
n

(2.9)

cn n (r)eint ,

(2.10)

where n (r) = r|n are the wavefunctions of some complete energy eigenbasis {|n } with eigenenergies En = n, and cn are coefcient such that (2.7) holds. Note that for continuous parts of the spectrum an integration instead of summing is needed. So, rst we should nd a representation of the form (2.10) holding for (r, 0). Now it turns out that: In the next section we show that there exists eigenfunctions k (r) for H corresponding to energies (k = |k|) E k = k = k , 2
2 2 (+)

(2.11)

the so-called stationary scattering states, that behave asymptotically as eikr i (r) + s (r) e + fk (, ) ; (2.12) r here (r, , ) are the spherical coordinates of r. These solutions thus look asymptotically as a superposition of an incoming plane wave i and an outgoing spherical wave s . It can be shown7 that the initial wave packet has the same expansion coefcients in the plane wave basis (Eq. (2.7)) than in terms of (+) functions {k }.
r (+) k (r) ikr

Accordingly, we nd the implicit but complete solution at any time (r, t) =


7

dk 0(k)k (r)eik t .

(+)

(2.13)

See Schwabl, sec. 18.1.3.

28

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

3. Infer how the solution (2.13) behaves long after the particles have been so close to each other that they interact. Thus calculate the differential cross section (, ). In order to simplify the analysis, we rst note that the solution (2.13) is a superposition of components of the form k (r, t) = eik t k (r).
(+)

(2.14)

Let us nd out, what kind of scattering do these components represent. In order to do this, let us calculate the probability ux related to them. Recall that the probability ux density is 2mi m Now Eqs. (2.12) and (2.14) imply that asymptotically the ux of k (r, t) is j(r)
r

j(r, t) =

[ ] =

Im{ }.

(2.15)

Im{(i + s )(i + s )}
Im{i i }

(2.16) + cross terms.

Im{s s }

The rst term represents the incoming ux ji , which is easily seen to be ji (r) = k z . (2.17)

The second term is the scattered ux js that propagates radially outwards: In the exercises it is shown that it behaves asymptotically as js (r) = k |fk (, )|2 r, r2 (2.18)

Thus we nd that the components with denite k represent scattering with an incoming ux ji and scattered ux js . As the probability uxes here are up to a normalization constant proportional to particle uxes in
How about the cross terms? They represent the mixing of the incoming ux and the scattered ux. The incoming beam is in reality not a plane wave, but it has a nite width that is accomplished by superposition of solutions of the form (2.14) for k approximately within k around k0 . Due to the nite width of the incoming beam, the superposition of all the cross terms vanishes in the detector regime outside it, i.e. for = 0 and for large r . Thus we can neglect them in our analysis.
8

where r = r/|r| is the radial unit vector.8

29

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

case we have a continuous beam of incoming particles, we nd Eqs. (2.2), (2.17) and (2.18) to imply r2 js (r, , ) = |fk (, )|2. (, ) = ji (2.19)

Because of this connection to the differential cross section, the fk (, ) is called the scattering amplitude. The one-particle wave packet is a superposition of components with denite k, so some kind of averaging over k is required for it in (2.19). However, typically fk varies very slowly within the range k for broad wave packets, and the result (2.19) is essentially correct for k = k0.9 The direct problem thus essentially reduces to nding the fk (, ), given the potential V (r1 r2 ). 2.3 Integral Scattering Equation Due to Eq. (2.19), we have solved the scattering problem if we can nd the scattering amplitude fk (, ). In addition, we would like to show that the (+) relative Hamiltonian H indeed has eigenstates k (r) of the form (2.12). In order to do this, we want to solve the eigenproblem
2

H (r) = i.e.

2 + V ( r) ( r) = E k ( r) ,

(2.20)

( 2 + k 2 ) ( r) = U ( r) ( r) , where (cf. Eq. (2.11)) Ek = and U ( r) =


9

(2.21)

k 2 2
2

2 2

(2.22)

V ( r) .

(2.23)

At certain values of k so-called scattering resonances (see Schwabl, section 18.8) may occur, in which fk varies rapidly as a function of k.

30

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

If and when a differential equation has complicated boundary conditions, it is often more convenient to transform it into an integral equation. Now the boundary conditions are of the complicated asymptotic form (r) eikr + fk (, ) so we use integral equation formalism. Let us formulate the problem as an integral equation by using Greens functions. Consider the inhomogeneous Helmholtz equation (Poisson equation) ( 2 + k 2 ) ( r) = A ( r) , (2.25) (2 + k 2 )G(r, r) = (r r); (2.26)
r

eikr , r

(2.24)

where A(r) is a xed source function. Let G(r, r) be a function that satises such a function is a Greens function for the Helmholtz operator 2 + k 2 . Furthermore, let (r) be a solution of the homogeneous Helmholtz equation (2 + k 2 )(r) = 0. Now it follows that (r) = (r) + is a solution of the Eq. (2.25):10 (2 + k 2 ) (r) = (2 + k 2 )(r) + =0+

(2.27)

dr G(r, r)A(r)

(2.28)

dr (2 + k 2)G(r, r)A(r)

(2.29)

dr ( r r ) A ( r ) = A ( r) .

The desired boundary conditions for (r) are obtained by choosing (r) and G(r, r) appropriatelythe equations (2.26) and (2.27) have innitely many different solutions. We would like to solve Eq. (2.21), which is not quite of the form (2.25), because (r) also appears in the source term. Anyway, setting A(r) = U (r) (r) above we see that if (r) satises the integral equation (r) = (r) +
10

dr G(r, r )U (r) (r),

(2.30)

Assuming the order of integration and differentiation may be changed.

31

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

it is also a solution for Eq. (2.21). Hence we can alternatively try to solve this integral equation, which is a Fredholm equation of the second kind.11
2.3.1 Building in the boundary conditions

First of all, the boundary conditions satised by the solution of (2.30) depend on the choice of (r) and G(r, r). Our rst task is to nd such a and G that satises (2.24). In principle we can use all tricks of the trade to nd suitable and Gfor example, we can make a sophisticated guess and check afterwards that the desired boundary conditions are fullled. Let us use heuristic reasoning to nd rst. If the interaction potential V (r r ) between the particles is very weak, so is also U (r) and the integral term in (2.30), which implies (r) (r). On the other hand, scattering should be weak, implying fk to be very small, and (2.24) would imply (r) to approach the plane wave eikr asymptotically. This plane wave satises the homogeneous equation (2.27), hence it seems natural to choose (r) = eikr at least in the weak scattering regime. How about G(r, r)? It will be shown in the exercises that the functions 1 eik|rr | G (r r ) = 4 |r r |

(2.31)

(2.32)

satisfy Eq. (2.26). Furthermore, we will show that G+ (r r) corresponds to the boundary conditions (2.24). Consequently, the original problem stated in Eqs. (2.21) and (2.24) is equivalent to solving the integral equation (r) = eikr + dr G+ (r r )U (r) (r ) (2.33)

which is the so-called Lippmann-Schwinger integral equation for scattering states in the position representation.
11

See for example Arfken: Mathematical Methods for Physicists, chapter 16.

32

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

But let us check that the solution of (2.33) really satises the required boundary conditions. For simplicity, we suppose that the interaction potential V (r) has a nite range, i.e. that there exists a radius a such that V ( r) = 0 , r > a. (2.34)

It follows that the integration region in (2.33) may be restricted to r a. Let us analyze the behaviour of the solution of (2.33) in the region r a. We may expand |r r | = r + r 2r r = r
2 2 2 2

r r 12 +O r

r r

(2.35)

where r = r/r. It follows that in the exponent of G+ we get k |r r | = kr

r r 12 +O r

r r r r ,

r r = kr 1 +O r

r r

= kr k r + kr O where

(2.36) k = k r. Similarly, 1 = |r r | r 1 1
rr r r 2 r

(2.37)

+O

1 r r = +O 1+ r r

r r

1 = 1+O r (2.38)

r r

Using these asymptotic forms in G+ in (2.33) we get ( r) = e hence (r) eikr


r ikr

dr

1 i e 4

krk r +kr O

r r

1 1+O r

r r

U ( r ) ( r ) , (2.39)

1 eikr 4 r
33

dr eik r U (r) (r),

(2.40)

TFY-44.130 Kvanttimekaniikka II which is indeed of the form (2.24) with fk (, ) = 1 4

S. Virtanen 2007

dr eik r U (r) (r ).

(2.41)

The solutions of (2.33) satisfy these boundary conditions also for non-nite range potentials that do not satisfy (2.34), but decay faster that r1 for r . We will show this for spherically symmetric potentials in section 2.4.2.
2.3.2 Neumann series solution and Born approximation

In order to nd the scattering amplitude, we should now solve (2.33) for (r) and plug the result into the integral of Eq. (2.41).12 Let us try to solve Eq. (2.33) recursively as a perturbation series in the interaction potential U (r). If the interaction is weak, scattering is weak and the integral term on the right hand side has only a small effect. Correspondingly, we use as a zeroth-order approximation (r) 0(r) = eikr . (2.42)

In order to nd a hopefully better rst order approximation, we take the integral term into account by using the zeroth-order approximation for (r) in it: 1 (r) = eikr + dr G+ (r r )U (r)eikr . (2.43) Continuing in this fashion, we nd the general nth-order approximation as
n

n (r) = e where u j ( r) = dr1

ikr

+
j =1

u j ( r) ,

(2.44)

drj G+ (r r1 )U (r1)G+(r1 r2 )U (r2) G+ (rj 1 rj )U (rj )eikrj . (2.45)

12 Recall that k = kinc z is the wave vector of the incident beam, and k = kscatt is a wave vector of equal length but having direction (, ), i.e. it points to the scattering direction.

34

TFY-44.130 Kvanttimekaniikka II If the series

S. Virtanen 2007

uj (r) converges, it is straightforward to show that (r) = lim n (r) = e


n ikr

j =1

u j ( r)

(2.46)

is a solution for (2.33). As uj is of j th power in the interaction potential, it is plausible that the series converges if the interaction and scattering is not too strong.13 The series solution (2.46) is called the Neumann series solution for the integral scattering equation. However, often only the rst order approximation in the interaction potential is used to estimate the scattering amplitude. This approximation is obtained from Eq. (2.41) by using only the zeroth-order approximation (r) 0 (r) in the integral. The result (recall the relation (2.23))
(Born) (, ) = fk

dr eiKr V (r),

(2.47)

where K = k k = kinc kscatt (2.48) is the difference between the wave vectors corresponding to incidence and scattering directions. This is the famous Born approximation for the scattering amplitude. Remarks on the Born approximation: Very simple to calculate, thus popular when expected to be reliable. Essentially just a Fourier transform of the potential. Expected to be reliable for weak interaction potentials and high incident energies (such that the Fourier transform of the potential is small). The corresponding approximation for the differential cross section is given by relation (2.19):
(Born)

2 (, ) = 2 4

2 4

dr e

iKr

V (r ) .

(2.49)

13 Typically the Neumann series converges, but in some physical situations it does not and other solving methods are needed.

35

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

2.4 Scattering by a Central Potential; Method of Partial Waves In the previous section we solved the eigenvalue problem for scattering states by using an integral equation formulationfor interaction potential of the most general form the problem is rather intractable in the differential equation formulation. However, in case of a spherically symmetric interaction potential, i.e. for V (r1 r2 ) = V (|r1 r2 |), (2.50)

the rotational symmetry can be utilized and the differential equation formulation becomes manageable. In this case the original equations (2.21) and (2.24) can be written as 2 (r) + [k 2 U (r)] (r) = 0, eikr , r where U (r) = 2 2 V (r ). Now because of the rotationally symmetric interaction, the Ansatz (r) eikr + fk (, )
r

(2.51) (2.52)

( r) =
lm

(2l + 1)il AlmRl (r)Plm (cos )eim

(2.53)

consisting of partial waves having a denite angular momentum14 reduces the 3D equation (2.51) into the 1D radial equations l(l + 1) 1 d 2d 2 R ( r ) + k U ( r ) r Rl (r) = 0 l r2 dr dr r2 (2.54)

for functions Rl (r). The problem reduces to nding out coefcients Alm and radial functions Rl (r) such that (r) satises the asymptotic boundary condition (2.52).
The Plm (cos ) are Legendre functions. Partial waves are angular momentum eigenstates, since they are proportional to the spherical harmonicsrecall that 1/2 2l + 1 (l m)! Ylm (, ) = Plm (cos )eim . 4 (l + m)!
14

The factors (2l + 1)il are included in the Ansatz only for convenience in the following calculations. Note that the Ansatz does not restrict generality in any way, because {Rl (r )Ylm (, )} forms a complete basis.

36

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

For simplicity, we rst assume that the potential U (r) has only a nite range, i.e. that there exists a nite radius a such that V ( r ) = 0, r > a. (2.55)

Recall that in regions where U (r) = 0, the solution of (2.54) is a linear combination Ajl (kr) + Bnl (kr) of the spherical Bessel functions. For the kind of solutions we are looking for, it turns out that both A and B can be required to be real, and furthermore we may require the normalization A2 + B 2 = 1note that we can still adjust normalization using the undetermined coefcients Alm . The solutions can then be written outside the interaction radius a as Rl (r) = cos(l )jl (kr) sin(l )nl (kr), r a. (2.56)

The partial wave phase shifts l are real because A2 + B 2 = 1. We want to t (r) to its asymptotic boundary conditions, and thus we want to nd out the asymptotic behaviour of Rl (r). The asymptotic behaviour of the Bessel functions is of the form sin(kr l 2) , jl (kr) kr
r

(2.57) (2.58)

cos(kr l ) 2 , nl (kr) kr
r

which imply
l 2 ) + sin(l ) cos(kr l 2 ) Rl (r) kr sin(kr l 2 + l ) = . kr r cos(l ) sin(kr

(2.59)

Note that if V (r) = 0 everywhere (no interaction, no scattering), the only possible solution is Rl (r) = jl (kr) (nl (kr)Ylm (, ) does not satisfy the original Schrdinger equation at r = 0, thus one must set B = 0) implying l = 0 for all l. Far away from the interaction region, the only effect of the interaction to the wavefunction is to give the partial waves possibly nonzero phase shifts l !

37

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

In order to t to the asymptotic form (2.52), it is useful to express also the plane wave eikr in terms of angular momentum eigenstates. In course Kvanttimekaniikka I we proved that15 eikz = eikr =
l

(2l + 1)il jl (kr)Pl (cos ).

(2.60)

Especially, we nd the asymptotic form e


ikr r

(2l + 1)i
l

l sin(kr

l 2) Pl (cos ). kr

(2.61)

Now we are ready to solve for the coefcients Alm and to nd the relation between the phase shifts l and the scattering amplitude fk (, ) by tting (r) to its asymptotic form. Equations (2.53), (2.59) and (2.61) imply ( r)
r lm
(2.52)

sin(kr l 2 + l ) m (2l + 1)i Alm Pl (cos )eim kr


l

(2l + 1)i
l

l sin(kr

) l eikr 2 Pl (cos ) + fk (, ) . kr r

(2.62)

By using sin(x) = (eix eix)/(2i), we get eil 2 eil eikr eil 2 eil eikr m (2l + 1)i Alm Pl (cos )eim 2ikr
l

lm

=
l

(2l + 1)i

il 2 eikr le

eil 2 eikr eikr Pl (cos ) + fk (, ) . (2.63) 2ikr r

This equation has to hold for all values of r (large enough), and . Since ikr ikr the functions e r and e r are linearly independent, their r-independent coefcients must be the same on both sides. Noting that eil 2 = il , the ikr coefcients of e r yield eil m Pl (cos )eim = (2l +1)i Alm 2ik
2l

(2l +1)i2l
l

lm
15

1 Pl (cos ). (2.64) 2ik

Pl (cos ) Pl0 (cos ); we choose the coordinate system so that k points in the positive z -direction.

38

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

Since the spherical harmonics are linearly independent, this implies Alm = eil m0. (2.65)

We note that the expansion of (r) contains only terms with m = 0, and consequently (r) is independent of ! This is a natural consequence of the fact that the interaction is rotationally symmetric. It follows that also fk (, ) = fk () and (, ) = (). By using (2.65) and il eil 2 = 1, the coefcients of fk ( ) = or 1 fk ( ) = k 1 2ik

eikr r

yield (2.66)

l l=0

(2l + 1) e2il 1 Pl (cos ),

(2l + 1) sin(l )eil Pl (cos ).

(2.67)

This implies the total cross section to be = d|fk ()|2 = d 1 k 1 k (2l + 1) sin(l )eil Pl (cos )
l

(2.68)

(2l + 1) sin(l )eil Pl (cos ) .


l

Using the orthogonality relation for the Legendre polynomials dPl (cos )Pl (cos ) = we get 4 = 2 k
l=0

4 ll , 2l + 1

(2.69)

(2l + 1) sin2 l .

(2.70)

Equations (2.67) and (2.70) express the desired quantities in terms of phase shifts l . Note that these results are exact. The problem is thus reduced to determining the phase shifts. Note that l = l (k ), i.e. that the phase shifts depend on the energy of the particles in the incident beam.

39

TFY-44.130 Kvanttimekaniikka II One interesting observation: Since Pl (1) = 1, we nd 1 Imfk (0) = k i.e.
l=0

S. Virtanen 2007

1 (2l + 1) sin(l )Im{e }Pl (1) = k


il

l=0

(2l + 1) sin2 l , (2.71)

k (2.72) . 4 This relation is known as the optical theorem. The optical theorem arises from total ux conservation: The scattered ux leads to diminished ux behind the scatterer. This diminishing of ux in the shadow region is due to destructive interference corresponding to scattering in the direction = 0; thus the relation between fk (0) and . The optical theorem is valid also more generally for nonsymmetric potentials and even for nonelastic scattering. Imf (0) =
2.4.1 How to determine the phase shifts?

What remains to be done is to nd the phase shifts l , that determine the denite asymptotic behaviour of the partial waves far from the scattering region. In order to determine the phase shifts, we need the solutions (for all l) of the radial Schrdinger equation (2.54) for all values of r. We have already obtained the general solutions in the region r > a, and it remains to solve this equation for different partial waves also in the interaction region r a, and nally to t the solutions together at r = a. This tting procedure yields the phase shifts in the same way as usually this kind of tting determines the unknown constants in the general solution for a differential equation. You may have noted that we have not yet utilized the detailed form of V (r) in any way, except for requiring it to have a nite range a, i.e. V ( r ) = 0, r > a. (2.73)

Now, the exact form of V (r) determines the solutions of the radial equation in the region r < a, and by the tting procedure it nally determines the phase shifts. In this way we can see the connection from V (r) to (, ), which is the backbone of the direct scattering problem.

40

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

In the regions where V (r) is nite, the radial wave function and its derivative have to be continuous, and usually these conditions can be used to t the solutions in the inner and outer regions together. Especially, if V (r) is nite in the vicinity of r = a, the radial function Rl and its derivative have to be continuous at r = a. In particular, the logarithmic derivative l = 1 dRl Rl dr = d log |Rl | dr (2.74)

has to be continuous at r = a.16 Eq. (2.56) thus implies k [jl (ka) cos l nl (ka) sin l ] , l (a) = jl (ka) cos l nl (ka) sin l which gives tan l = kjl (ka) l (a)jl (ka) . knl (ka) l (a)nl (ka) (2.76) (2.75)

We can thus nd the phase shifts in the following way: 1. Solve (numerically, if not otherwise) the radial equation Eq. (2.54) in the scattering region r < a. Calculate the logarithmic derivatives l (a) = limra l (r) for the scattering region solution. 2. Use Eq. (2.76). Note that if the potential is not bounded at r = a, one has to use other methods to t the solutions in the scattering region and outside it. As an example of such a situation we consider a bit later scattering from a hard sphere potential. Anyway, the point is that always one can determine the phase shifts by solving the radial equation in the whole space, no matter in which way one does it. Often it can be in fact simpler to solve the radial equation directly in the whole space instead of using the logarithmic derivative method, for example.
16 We use the logarithmic derivative, because it does not depend on the normalization of Rl . Thus we can determine the phase shifts by using only one condition instead of two (continuity of Rl and Rl ).

41

TFY-44.130 Kvanttimekaniikka II
2.4.2 Validity for potentials with nonnite range

S. Virtanen 2007

We have so far restricted our analysis to potentials that vanish outside a nite radius a. The results derived are, however, valid also for potentials that vanish only asymptotically as long as the scattering states satisfy the boundary conditions eikr . (2.77) r Let us derive a sufcient condition for this to be the case when the scattering potential is spherically symmetric. Using the partial wave analysis applicable in this case we see that the desired boundary conditions can be satised if the radial equation has solutions Rl (r) that behave asymptotically as (r) eikr + fk (, )
r

eikr Rl (r) (2.78) r in such a case solutions of the form (2.59) exists and tting to the boundary condition (2.77) succeeds as before. Dening as usual the new radial functions ul (r) = rRl (r) (2.79)
r

which satisfy the radial equation (implied by Eq. (2.54)) l(l + 1) d2 u l ( r ) 2 + k U ( r ) ul ( r ) = 0, dr2 r2 we look for solutions of the form ul (r) = eh(r) eikr ;
r

(2.80)

(2.81)

if h(r) constant, Rl (r) = ul (r)/r satisfy (2.78), hence (r) can be made to satisfy (2.77) and everything is OK. Inserting the general Ansatz (2.81) into (2.80), one nds the equation for h(r): h (r) + h (r)2 2ikh (r) = U (r) + l(l + 1) . r2 (2.82)

This can be thought as rst order differential equation for h (r). If it has r solutions for which h (r) 0 sufciently rapidly, h(r) will approach constant for large r and everything is OK.
42

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

If U (r) decays for large r as r1 with > 0, we see that (2.82) should have a solution that behaves asymptotically as h (r) r1 , where 17 = min{, 2}. Integrating this result, we nd h(r) = dr h (r) a + b r 1 , (2.83)

which approaches a constant for large r if > 1, i.e. if > 1. Hence we see that If the potential decays faster than
1 r

asymptotically, our approach works.

However, for Coulomb potential V (r) r1 one nds h(r) log r, which does not converge, and different methods are required to analyse pure Coulomb scattering.18 If the potential has a non-nite range but decays faster than r1, we can still determine the phase shifts by using Eq. (2.76), but by calculating the limit a instead of using a xed, nite tting radius a. If the potential would decay slower than r1, this limit would not exist any more, i.e. the phase shifts would not saturate asymptotically!
2.4.3 Relative contributions of different partial waves

Although for spherically symmetric potentials we have managed to reduce the problem to solving a 1D radial equation instead of the original 3D Schrdinger equation,19 one has to solve the radial equation for each l in order to nd the corresponding phase shifts l . Hence the partial wave method is practically useful only when good approximations are obtained by taking only a rather small number of partial waves into account. We will now show this to be the case in the low-energy limit for potentials that vanish for r > a. In order to see the relative importance of different partial waves, let us rst just neglect the interaction term U (r)Rl (r) in the radial equation (2.54).
17

Note that for this kind of solution one may neglect the terms h (r ) and h (r )2 in the equation for large r . See Merzbacher, sec. 11.8, for example. 19 For example solving numerically a 1D equation is MUCH faster than solving a 3D one!
18

43

TFY-44.130 Kvanttimekaniikka II
(0)

S. Virtanen 2007

Then we get the solution Rl (r) = jl (kr). Since the spherical Bessel functions have asymptotic forms20 (kr)l jl (kr) , (2l + 1)!! nl (kr)
(0)

(kr l), (kr l),

(2.84) (2.85)

(2l 1)!! , (kr)l+1

we see that Rl (r) is vanishingly small in the regime kr l. Let us now take the interaction term U (r)Rl (r) into account in rst-order perturbation theory. All the matrix elements relevant for perturbation theory are in the (0) position representation integrals that contain the term U (r)Rl (r) in the integrand. However, this term is negligible if l ka, since for r < a we have kr < ka l and the radial function is very small according to (2.84), and for r > a the potential vanishes. Thus, the effect of the scattering potential U (r) to partial waves with l ka (2.86)

should be negligible, and we should get a good approximation by calculating explicitly only the phase shifts for l ka, and setting l = 0 otherwise. Comments:21 The lower the energy, the smaller the ka and the number of nonnegligible phase shifts. Hence the partial wave analysis is most efcient and useful in describing scattering at low energies.22 In the low energy limit ka 1 we can neglect according to (2.86) all partial waves except the s-wave corresponding to l = 0. Eq. (2.67) then implies 1 (ka 1), (2.87) fk () ei0 sin 0 k since P0 (cos ) = 1. It follows that in the low energy limit scattering is always isotropic, i.e. independent of !
l!! = l(l 2)(l 4) 2 or 1 depending on whether l is odd or even. These hold for potentials that vanish outside a given radius. The results hold also for potentials that decay asymptotically sufciently rapidlysee for example Landau & Lifshitz: Quantum Mechanics, Non-Relativistic Theory for details. 22 Recall that for the Born approximation the situation is the opposite!
21 20

44

TFY-44.130 Kvanttimekaniikka II It can be shown generally that

S. Virtanen 2007

which also shows that s-wave scattering dominates in the low-energy limit. However, to these general results there is an exception: so-called scattering resonances. Such resonances may occur if the scattering potential supports bound states. If the energy of the incoming beam is near such a bound state energy, the wavefunction inside the scattering region becomes very sensitive with respect to small perturbations and our argument based on rst-order perturbation theory may not hold. Scattering resonances will be demonstrated in exercise problems.
2.4.4 Example: Scattering from a hard sphere

l k 2l+1,

k 0

(2.88)

Let us consider scattering from a hard sphere potential V (r ) = +, r a, 0, r > a. (2.89)

Note that in this case V (r) is not nite in the vicinity of r = a, so that we can not use the formula (2.76) to determine the phase shifts! Instead, the phase shifts are easily determined using the original expression (2.56) for Rl (r). Due to the innite potential inside the sphere, the radial function has to vanish at r = a, and (2.56) implies jl (ka) . (2.90) tan l = nl (ka)
Low energy limit (ka 1)

In this case we may use the asymptotic expressions (2.84) and (2.85) in (2.90) to get23 (ka)2l+1 l tan l . (2.91) (2l + 1)[(2l 1)!!]2
23

This result accidentally holds also for l = 0, although the asymptotic expressions do not!

45

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

Thus the phase shifts indeed fall off rapidly as l increases, since the factorial function wins the power function 6-0 for large arguments. In the limit k 0 all but the s-wave phase shift 0 = ka can be neglected, and Eq. (2.70) gives 4 2 4 (ka)2 = 2 sin (ka) = 4a2 . 2 k k (2.93) (2.92)

Note that this is four times the geometric cross section a2 for classical hard sphere scattering! In this limit the de Broglie wavelength = 2/k is very large, and the sphere strongly modies the wavefunction even much farther than the radius ahence the large cross section.
High energy limit (ka 1)

As expected, the high energy limit is much more difcult to analyse because all partial waves contribute to scattering. Substitution of (2.90) into (2.70) yields (2l + 1)jl2(ka) 4 = 2 . (2.94) k jl2(ka) + n2 l (ka)
l=0

This can be calculated in the limit ka analytically (but not easily. . . ), and the result is = 2a2 . (2.95) This is twice the geometric cross section, which is surprising since one would expect the high energy, short wavelength limit to correspond to classical scattering. The reason for this seemingly anomalous result is that in the relations (, ) = |fk (, )|2,
ikr

(2.96)

eikr (2.97) e + fk (, ) r on which we based our analysis of quantum mechanical scattering all deviations from the asymptotic plane wave form eikr are interpreted as
r (+) k (r)

46

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

scattering. Now the hard sphere modies the incoming ux in the classical limit in two ways: particles that hit the ball are scattered, and the missing particles form a shadow behind the sphere. Hence fk is nonzero not only in scattering directions = 0, but also in the forward direction = 0, where it produces the shadow via destructive interference with the plane wave. Consequently, in the quantum mechanical analysis the missing particles behind the scatterer are also considered as scattered, while in the classical calculation they are not. This causes the total cross section in the quantum case to be twice the classical value. It is to be noted that this failure of the QM scattering is an artefact of the limit k . For very large but nite k diffraction in fact takes place and the diffracted, i.e. scattered particles ll the shadow far away from the sphere. Hence the quantum mechanical result is in fact correct, and, strictly speaking, the classical hard sphere scattering is never realized!

47

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

Chapter 3

Interaction of Atoms with Electromagnetic Radiation


Literature: Schwabl 16.3 (Harmonic perturbation theory), 16.4 (Atom-radiation interaction)

3.1 Motivation and Problem Setting The subject of interaction of atoms with electromagnetic elds is naturally of fundamental importance for atomic physics and optics. In Kvanttimekaniikka I we considered the effect of certain time-independent and time-dependent external electric and magnetic elds on hydrogen-like atoms (Zeeman effect, Stark effect). Now we aim to investigate the effect of electromagnetic radiation on atoms. As before, the treatment is only semiclassical in the sense that we treat the atoms quantum mechanically, but the electromagnetic eld classically. This description is somewhat insufcient, and in a more precise treatment the electromagnetic eld is also quantized (giving rise to quantum electrodynamics (QED)). The semiclassical model is able to describe rather well absorption and stimulated emission processes, but a proper treatment of spontaneous emission would require the use of QED.

48

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

We use time-dependent perturbation theory in our analysis. First we briey review the main results of time-dependent perturbational analysis. In order to describe radiation, we specialize to consider harmonically time-dependent perturbations and derive the so-called Fermis golden rule.

3.2 Time-Dependent Perturbation Theory


3.2.1 The general rst-order results

We rst review the basic result of time-dependent perturbation theory; for details, see the lecture notes of Kvanttimekaniikka I. The system under consideration is assumed to be described by the Hamiltonian H (t) = H0 + H (t), where H0 is time-independent with orthonormalized eigenstates H0 |n = En |n , (3.2)

(3.1)

and H (t) a time-dependent perturbation. Assume that the system is at the initial time t = 0 in the eigenstate | (0) = |i , (3.3)

of H0 . It follows that at later times t the projection of the state vector | (t) onto an eigenstate |k = |i is within the rst-order perturbation theory k | (t) where ki =
(1) ck (t)

1 = i

dt k |H (t )|i eiki t ,

(3.4)

Ek Ei

(3.5)

is the Bohr frequency corresponding to the energy difference between the states |k and |i . The rst-order result is assumed to be accurate for t small enough such that the state vector has not changed markedly from its initial value, i.e. as long as (1) ck (t) ki. (3.6)
49

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

If the perturbation is very weak, the rst order approximation may be accurate up to rather large t, but for strong perturbations it loses accuracy very quickly.
3.2.2 Special case: Harmonic perturbation

and H (t) = 0 for t < 0; above, the operator A is to be time-independent, and we can choose 0 without restricting generality. Note that H (t) has to be Hermitian for H = H0 + H (t) to be Hermitian. Substituting this perturbation to the general result (3.4), we nd in the rst order in perturbation theory for t > 0

We apply now the above-mentioned results to the important special case of harmonically time-dependent perturbations. The perturbation is taken to be of the form H (t) = Aeit + Aeit (t 0), (3.7)

Now the transition probability from the initial state |i to a nal state |f is1 Pif (t) = | f | (t) | =
2

t t 1 1 i(ki )t k | (t) = dt e + dt ei(ki +)t k |A|i k |A |i i i 0 0 (3.8) i(ki )t i(ki + )t k |A |i 1 e k |A|i 1 e + . = ki ki +

f |A|i 1 ei(f i)t f |A |i 1 ei(f i+)t . + f i f i + (3.9)

Let us consider the state after a long-lasting perturbation, i.e. |t| 1, |f i t| 1. (3.10)

Essentially this means that we consider the system after several periods of the harmonic perturbation and that the energy difference between the initial and nal states is not much smaller than the energy corresponding to the
One often speaks about transition probability in this connection from state |i to state |f , but strictly speaking the state | (t) is not in general any of the states |k but just a superposition of them, and the nomenclature is a bit misleading. Pif (t) is rather only the probability for the energy measurement of the system to give the result Ef at time t (assuming Ef to be nondenerate).
1

50

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

perturbation. We assume the matrix elements f |A|i , f |A |i to be small enough such that the criterion (3.10) is satised while Pif (t) is still small so that the rst-order perturbation theory is reliable. Let us estimate the contribution of the two terms in (3.9). Since 1 ei(f i)t f i
2

sin

f i ) t 1 2 ( f i )

1 ( 2

(3.11)

is as a function of sharply peaked at = f i (height t2 , width 4t ; see Figure 13.9 in Liboff) for times such that (3.10) holds, the transition probability is essentially nonzero only for f i. These maxima in the transition probability correspond to resonance frequencies of the system between its eigenstates. The overlap of the two peaks is negligible, and we nd to an excellent approximation 2 sin[ 1 1 2 ( +f i )t] 2 , for f i 1 2 | f |A |i | 2 ( +f i ) 2 1 (3.12) Pif (t) = 1 2 sin[ 2 ( f i )t] , for 1 2 | f |A|i | f i ( ) fi 2 0, otherwise. Now because 0, only one of these peaks is physically relevant for given initial and nal stateswhich one, is determined by the sign of f i. If f i > 0, it is the peak at = f i (i.e. = Ef Ei > 0), and this corresponds to resonant absorption of energy by the system. If f i < 0, it is the peak at = f i (i.e. = Ei Ef > 0), and this corresponds to resonant emission of energy (emission stimulated by the perturbation).

Since the width of the peaks is 4/t, the transition probability is essentially nonzero only if 2 |Ef Ei | < . (3.13) t This can be interpreted in terms of energy-time uncertainty relation2 : After the perturbation has acted time t, the spread E in the energy of the
1 The energy-time uncertainly relation E t is not a true Heisenberg uncertainty relation such as xpx 2 . Instead, it only describes in an approximate manner the relation between transition times t between states that have energy spread E . 2

51

TFY-44.130 Kvanttimekaniikka II nal state is approximately E . t


Fermis golden rule

S. Virtanen 2007

(3.14)

In the previous analysis we have assumed that the spectrum of the system on which the perturbation acts (eigenvalues of H0 ) is discrete. If the initial state |i or the nal state |f belongs to a continuous part of the spectrum, we can develop the analysis further to obtain a simple and very useful result for the transition probability. To be specic, we assume |i to be a normalized state belonging to a discrete part of the spectrum, and we consider transitions to a continuous part of the spectrum. The general result (3.4) and the results presented above for harmonic perturbation remain valid also when |f belongs to spectral continuum. Let us calculate the total transition probability to the continuous part of the system. This means that we sum (integrate) over all possible nal state transition probabilities. Let n(E ) be the density of energy eigenstates per unit energy in the continuum, i.e. n(E )dE is the number of states in the range [E, E + dE ]. According to (3.12), the probability that after time t the system has made a transition to the spectral continuum due to absorption of energy is3 Pabs (t) = 1
2

| f |A|i |2

sin

f i ) t 1 2 ( f i )

1 2 (

n(Ef )dEf

(3.15)

Recall that we still assume |t| 1, |f i t| 1 to hold.4 Note that above the matrix element depends on the integration variable Ef via the state |f , and the bracketed expression via f i . Recall also that for t large, the bracketed function has a sharp peak of width 4t at f i . Typically the dependence of the matrix element and the level density on Ef is much slower, and to a good approximation they may be approximated to be
For simplicity, we assume that the spectrum is nondegenerate, such that Ef species the state |f uniquely. It is in principle straightforward to generalize the analysis to the case of a degenerate spectrum. 4 The integration may be taken over all energies, because for |t| 1 the transition probability decays fast enough away from the resonant energy such that the contribution from the extended regions becomes negligible.
3

52

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

constant in the small neigborhood of f i = in which the bracketed function is nonzero. Thus we nd to presumably good accuracy Pabs (t) 1
2

| f |A|i |2 n(Ei + )

sin

f i ) t 1 2 ( f i )

1 2 (

dEf .

(3.16)

The remaining integral is found to integrate to 2 t, and we nd Pabs (t) 2 | f |A|i |2 n(Ei + )t. (3.17)

We note that the transition probability is linearly proportional to time t. Thus the result implies a constant transition rate Rabs = Pabs (t) 2 | f |A|i |2 n(Ei + ). t (3.18)

This is the famous Fermis golden rule for transition rates, a simple and very useful result!

3.3 Interaction of Atoms with Electromagnetic Radiation In this section we aim to study the effect of electromagnetic radiation on atoms. As the time-dependence of monochromatic radiation is harmonic, we can utilize the results derived in the previous section in our analysis. We suppose that the wavelength of the radiation is much larger than atomic sizes. This excludes X rays and gamma radiation. Typical application would be a laser eld imposed on an atom. Also we treat the electromagnetic eld classically, and only the atom quantum mechanically. QED would be required to account for spontaneous emission processes, for example. Assuming that the nucleus of the atom is so massive compared to electrons that it can be thought xed, the Hamiltonian of an atom in an external

53

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

electromagnetic (EM) eld can be written in the position representation as


N

H= 1 + 2
i=1 N

q 1 i A(ri, t) p 2m c q2 , |ri rj |

Zq 2 si B( + q(ri , t) M r, t) ri

i=1 j =i

(3.19) where m is the electron mass, q = e its charge, ri the position operator of si its spin magnetic i = i i its momentum operator and M electron i, p moment operator, and Z the charge number of the nucleus. The classical elds A(r, t), (r, t) are the vector and scalar potentials for the electromagnetic eld:5 E = B = A. 1 A , c t (3.20) (3.21)

si B( The terms M r, t) represent the interaction of the electron spin with the magnetic eld: = gs q S q S s = gs B S (3.22) M 2mc mc is the spin magnetic moment operator of an electron, gs 2.002 the gyromagnetic ratio of the electron, B = 2e mc the Bohr magneton and S the spin operator of the electron. It is easy to believe that solving the exact time-development of the atom using this Hamiltonian is rather difcult. Thus we make the following simplications / approximations: We assume that the radiation is not too strong such that rst-order perturbation theory can be used. This is sensible because electromagnetic elds of a lasers are typically much weaker than the elds inside of atoms.
5

We use the Gaussian system of units.

54

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

We neglect the correlations and interactions between electrons, and use the fact that for non-X radiation it is the valence electron(s) that make transitions, and the effect on the inner electrons can usually be neglected. The problem thus reduces to analysing the physics of lets say one electron moving under the Coulomb attraction of a nucleus that has effective charge Z e (recall the variational calculation of the ground state energy of helium in Kvanttimekaniikka I). This model should be sensible at least for alkali atoms with one valence electron. These simplications / approximations amount to modeling the physics with the effective one-electron Hamiltonian q 1 A( p r, t) H= 2m c
2

Z e2 s B( + q( r, t) M r, t). r

(3.23)

To use perturbation theory, we write this Hamiltonian as H = H0 + H (t), where (3.24)

2 p Z e2 (3.25) 2m r is the time-independent Hamiltonian of the hydrogen-like free atom, and the time-dependent perturbation H0 = q2 q sB ) + ( pA+Ap A2 + q M H = 2 2mc 2mc

(3.26)

describes the interaction of the atom with the radiation eld. Note that the eigenstates of H0 are the the familiar one-electron |nlm states, but possibly corresponding to Z = 1 (for Z = 1 they are just the eigenstates of hydrogen). The electromagnetic elds and the Schrdinger equation i = H t (3.27)

55

TFY-44.130 Kvanttimekaniikka II are invariant under gauge transformations of the form A = A + , 1 , = c t = ei(q/ c) ,

S. Virtanen 2007

(3.28) (3.29) (3.30)

where = (r, t) is an arbitrary scalar function (Verify this!). The gauge degree of freedom may be xed at will, and calculations can be often simplied considerably by a judicious choice of gauge. In the following we use the Coulomb gauge, in which the gauge is xed by the gauge condition The Maxwell equations and this gauge condition imply where = (r, t) is the charge distribution. Since we neglect the effect of the atom to the EM eld, (r, t) 0 and we may choose (r, t) 0. 2 = 4, (3.32) A = 0. (3.31)

(3.33)

Let us rst take the perturbing external electromagnetic eld to be a monocromatic plane electromagnetic wavein section 3.3.2 we discuss the case of nonmonochromatic radiation. In the Coulomb gauge, this is represented by the vector potential
i(krt) A(r, t) = [A0ei(krt) + A ], 0e

(3.34)

where is a unit polarization vector and k the wave vector pointing to the propagation direction. The dispersion relation for EM waves is = ck . The transversality condition k=0 (3.35) always satised by freely propagating EM waves also implies the gauge condition (3.31) to be satised.

On the other hand, due to (3.31) and (3.33), our time-dependent perturbation H (t) reduces to the form6 q q2 sB + H = Ap A2 M 2 mc 2mc

(3.36)

A= Recall that p

A, and A = [ A] + A .

56

TFY-44.130 Kvanttimekaniikka II in the Coulomb gauge.

S. Virtanen 2007

In fact, the two last terms may typically be neglected to good approximation: The intensity of the radiation is proportional to |A0|2 , and the intensity q2 q2 2 2 of usual light sources is low enough such that 2mc 2 A 2mc2 |A0 | is negligible compared to the rst term.7 Also we may make the crude order of magnitude estimate s B|i f |M q )|i (A p f | mc
q SB mc q Ap mc 0

kA0 k = , A0 p p

(3.37)

since S and B kA0 due to B = A. Now the xp-uncertainty relation suggests that /p a0 , where a0 is the Bohr radius characterizing the radius of the atom. We thus nd s B|i f |M q )|i (A p f | mc a0 , (3.38)

where = 2/k is the wavelength of the radiation. In the optical region 500nm, and because a0 0.5= 0.05nm, it follows a0 0.0001. (3.39) The third term is thus in the optical region also negligible compared to the rst one!
q )|i may Note that for certain states the matrix element f | mc (A p vanish identically, and then of course our order of magnitude estimate fails. In fact this happens quite often due to symmetry effects, and will lead to selection rules for transitions (we will come back to this a bit later). In such a case the dominating (although small) contribution is q )|i is nonzero, (A p given by the other terms. However, when f | mc our order of magnitude estimate shows that it typically gives the dominating contribution.
The radiation eld is usually much weaker than the very strong internal EM elds within the atom. This is essential for our perturbational treatment in the rst place.
7

57

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

Consequently, in the optical region it should be a good approximation to take simply q q i(krt) = [A0ei(krt) + A ] p (3.40) H A p 0e mc mc as the perturbation Hamiltonian. Now, assuming the wave to have inuenced the atom over several frequency periods (i.e. t 1), we can apply the general result (3.12), nding for the electron state of the atom the transition probabilities 2 1 ( +f i )t] sin[ 2 q 2 |A0 |2 i k r 2 |i | , for f i < 0 p 1 2 2 2 | f |e 2 ( +f i ) m c 2 1 sin[ 2 ( f i )t] Pif (t) = q2 |A0 |2 ikr 2 |i | , for f i + > 0 | f |e p 1 m 2 c2 2 2 ( f i ) 0, otherwise. (3.41) The time-averaged energy density of the plane wave EM eld can be shown to be 2 1 (E2 + B2) = |A0|2 , (3.42) E = 2 8 2c so that we can express our result in the form 2 1 2 sin[ 2 ( +f i )t] 2 q i k r 2 E |i | , for f i < 0 p 1 2 2 2 | f |e 2 ( +f i ) m 2 1 Pif (t) = 2q2 E ikr 2 sin[ 2 ( f i )t] |i | , for f i + > 0 p 1 m2 2 2 | f |e 2 ( f i ) 0, otherwise. (3.43) In the rst case Ef Ei = f i < 0, i.e. the energy of the atom is reduced in the transition. Accordingly, it corresponds to a stimulated emission process, in which the electromagnetic eld stimulates the atom to emit energy. Actually, the atom emits energy by photon emission, but the emitted photon is not seen in our formalism which treats the EM eld classically.

Correspondingly, in the second case the Ef Ei = f i > 0, i.e. the energy of the atom is increased in the transition. The process corresponds to an
58

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

absorption of energy by the atom from the EM eld. Actually, the atom absorps a photon in such a process.
3.3.1 Electric dipole transitions; Selection rules

In order to nd out the transition rates, we still have to calculate the matrix elements |i Mf i = f |eikr p (3.44) that appear in (3.43). The initial and nal states are energy eigenstates of an unperturbed hydrogenic atom.

We note that the wavefunctions of the atomic states essentially vanish outside few Bohr radii a0 . Since a0 |k r| 0.0001 (3.45) in the region in which the wavefunctions are nonzero we may approximate eikr = 1 ik r + 1 with good accuracy in the matrix elements: |i = f |p |i . Mf i f | p (3.47) This is the so-called electric dipole approximation; note that within it the matrix elements do not depend at all on the wave vector k of the radiation. The approximation may be interpreted as follows: The atom is much smaller than the optical wavelength of the EM eld, and thus the external eld is essentially constant within the atomic volume. By taking into account higher order terms in the expansion of eikr , one gets magnetic dipole, electric quadrupole etc. matrix elements, which are corrections to the electric dipole approximation. These corrections are important if the electric dipole matrix element vanishes. Note, however, that in higher order calculations also the second and third terms in the interaction Hamiltonian (3.36) have to be taken into account to be consistent. Now since 2 1 1 p = = + V ( r) , x ] = 2[ p, x ] p 2(i ex) p p x, [ H0 , x ] = [ 2m 2m 2m im (3.48)
59

(3.46)

TFY-44.130 Kvanttimekaniikka II and similarly for the y - and z -components, we nd |i = f |p im f |[H0 , r]|i Ef Ei f | r|i

S. Virtanen 2007

= im

(3.49)

= imf i f | r|i . Thus the problem reduces to calculating the matrix elements f | r|i , hence the term electric dipole approximation. In order to show that in fact most of these dipole matrix elements vanish, let us resort to an explicit calculation: The wave functions of hydrogenic atoms are angular momentum eigenstates of the form8 nlm (r) = Rnl (r)Ylm (, ). (3.50)

By denoting the quantum numbers of the initial and nal states as ni , li, mi and nf , lf , mf , correspondingly, we get f | r|i = =
0 0

r2dr

d Rn ( r ) Yl f f lf

mf

(, ) rRnili (r)Ylimi (, )
mf

(r)Rnili (r) dr r3 Rn f lf

d Yl f

(, ) r0 Ylimi (, ), (3.51)

where r0 = r/|r|. Let us concentrate on the angular integral. In spherical coordinates we have r0 = x sin cos + y sin sin + z cos . 3 cos , 4 4 3 3 sin ei, 8 (3.52)

Let us try to express also this in terms of spherical harmonics. Recalling that Y10(, ) = we nd r0 =
8

Y11(, ) =

(3.53)

z Y10 +

x + iy 1 x + iy 1 Y1 + Y1 . 2 2

(3.54)

Recall from Kvantti I that the radial functions do not depend on the magnetic quantum number m.

60

TFY-44.130 Kvanttimekaniikka II The angular integral thus contains factors of the form d Yl f Since
0 2 mf

S. Virtanen 2007

(, )Y1m (, )Ylimi (, ).

(3.55)

d eimf eim eimi = 2m,mf mi ,

(3.56)

the angular integral can be nonzero only if m = mf mi = m = 1, 0, 1. (3.57)

This is the rst selection rule for electric dipole transitionssuch selection rules have to be obeyed in order to have a nonvanishing transition probability. In order to derive a second selection rule based on orthogonality of the spherical harmonics, we use the so-called addition theorem for spherical harmonics: One can show that
l1 + l2

Yl 1

m1

2 (, ) (, )Ylm 2

=
l=|l1 l2 |

C (l1l2m1 m2 ; l, m1 + m2 )Ylm1 +m2 (, ),

(3.58) where C (l1l2 m1 m2 ; lm) are the Clebsch-Gordan coefcients between the coupled and uncoupled angular momentum bases (recall Kvantti I. . . ). By using this result, we nd d Yl f =
mf
i (, ) (, )Y1m (, )Ylm i

li +1

d Yl f

mf

(, )
l=|li1|

C (li, 1, mi, m; l, mi + m)Ylmi+m (, ), (3.59)

which vanishes due to orthogonality of spherical harmonics unless lf = |li 1|, li, li + 1. (3.60)

This is the second selection rule for electric dipole transitions. In fact, a further constraint comes from parity conservation: Recall from Kvantti I that nlm(r) = (1)l nlm (r),
61

(3.61)

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

i.e. the unperturbed eigenstates have parity (1)l . Now because r is odd under parity reections r r, the parity of nf lf mf (r) r nili mi (r) is (1)lf +li1 . Unless this function has even parity, i.e. l = lf li is an odd integer, (3.62) the matrix element f |r|i vanishes. Altogether we nd the selection rules m = 0, 1, for electric dipole transitions. Note that if a given transition is forbidden on the basis of these selection rules, it may be allowed as a higher-order transitionmagnetic dipole, electric quadrupole etc. transitions have their own, somewhat different selection rules. Anyway, since the higher-order transitions are much weaker, transitions that break the electric dipole selection rules are in general much slower than those that obey them.
3.3.2 Nonmonochromatic radiation

l = 1

(3.63)

In practice, the radiation which strikes the atom is at least slightly non-monochromatic, and we should integrate the perturbation Hamiltonian (3.40) over all wave vectors k and sum over polarizations , with A0 = A0(k, ), and use the resulting perturbation to calculate the transition probabilities. Due to the square of the matrix element, the resulting expression would contain cross terms between different wave vector components of the radiation. However, if the radiation is incoherent, the cross terms will average to zero and the correct result is obtained just by integrating the result (3.43) over the different wave vectors. In the dipole approximation the dependence on the direction of k vanishes, and it is sufcient to integrate over the frequency ; recall that = ck . If we denote E , the energy density of the radiation eld per unit frequency interval by d d we nd by using Eq. (3.43) for the absorption probability from an

62

TFY-44.130 Kvanttimekaniikka II inhomogeneous radiation eld the result Pabs (t) =


(3.49)

S. Virtanen 2007

1 2q 2 2 sin 2 ( f i )t |i | | f | p 1 m2 2 2 2 ( f i ) 2 2q 2 f i 2

dE d d
2

| f | r|i |2

1 2

sin

f i ) t 1 2 ( f i )

1 2 (

(3.64) dE d d

within the electric dipole approximation. If the radiation energy density is a smooth function of , for t 1 we nd to good approximation Pabs (t) = =
2 2q 2 f i 2

1 dE | f | r|i |2 2 f i d dE d ,
f i

sin
f i

4 q t
2

2 2

f i ) t 1 2 ( f i )

1 2 (

| f | r|i |2

(3.65) implying the transition rate Rabs = 4 2 q 2


2

| f | r|i |2

dE d

.
f i

(3.66)

If in addition the radiation is unpolarized, one should also take the average of the result over all polarization directions: Since9 1 1 | f |x |i |2 + | f |y |i |2 + | f |z |i |2 = | f | r|i |2 , 3 3 (3.67) we nd for unpolarized radiation the absorption rate | f | r|i |2 = Rabs 4 2 q 2 2 dE = | f | r | i | 3 2 d .
f i

(3.68)

Note that this result and its derivation is very much analogous to Fermis golden rulein fact, it can be derived by quantizing the EM eld and by noting that the nal state consisting of atomic state + photon states forms a spectral continuum.
Note that in the last step f | r|i is an valued) 3D vector and | f | r|i |2 its length squared. In R ordinary (but complex 2 1 2 addition, by denition | f | r|i | = 4 d| f | r|i | , where the integral is taken over all directions of vector . One can verify that this gives just the average over orthogonal directions.
9

63

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

Chapter 4

Non-Relativistic Many-Particle Systems


Literature: Schwabl: Advanced Quantum Mechanics, Part I Fetter & Walecka: Quantum Theory of Many-Particle Systems, Chapter 1 4.1 Second Quantization1 Naturally the most interesting and important physical systems consist of large numbers of particles. Most of the particles are identicaltwo electrons, for example, have identical physical properties. Experimentally it is found that all elementary particles are either bosons or fermions. In theoretical modelling the quantum mechanical state vector has to be symmetric with respect to interchange of identical bosons, and antisymmetric with respect to interchange of identical fermions. This division of particles to bosons and fermions is supported by all experiments up to date. Furthermore, the so-called spin-statistics theorem states that particles with integer spins are bosons, and particles with half an odd integer spins are fermions. The spin-statistics theorem can be derived from very fundamental principles of relativistic quantum eld theories, and it is also supported by all experiments up to date.
1

We follow the structure of Part I in Schwabls book.

64

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

In principle we can model a system consisting of N identical particles by just forming its Hamiltonian operator H and by solving the Schrdinger equation i (r1 , r2, . . . , rN ; t) = H (r1 , r2, . . . , rN ; t); t (4.1)

here ri is the coordinate of the ith particle.2 One also has to take care that the wave function has proper symmetry properties depending on whether the particles are bosons or fermions. For large systems, this wave function formulation is, however, very inconvenient: If for example N = 1023, it is impractical to try to directly solve Eq. (4.1), taking care of the proper symmetry of the wave function. Thus we want to develop a more convenient but equivalent formulation of the quantum mechanics of many-body systems. Such is the so-called second quantization formalism we are going to present.
4.1.1 Identical Particles, Many-Particle States, and Permutation Symmetry
Permutations

An N -permutation is a map P : {1, 2, . . . , N } {1, 2, . . . , N } that rearranges the N elements into a different order; we denote often P = (1 P (1), 2 P (2), . . . , N P (N )). There are altogether N ! different N -permutations. They form the permutation group SN . SN is called a group because 1. SN is closed under multiplication, i.e. two subsequent permutations form a permutation: If P, Q SN then also P Q = P Q SN . 2. SN contains an identity permutation I SN , a permutation that does not change the order at all. 3. For each permutation P SN , there exists an inverse permutation P 1 SN such that P P 1 = P 1 P = I .
2

(4.2)

For simplicity, we neglect spin degrees of freedom here.

65

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

Transpositions Pij SN are permutations which only change the order of elements i and j : Pij = (i j, j i). All permutations can be written as products of transpositions. The signature of a permutation P is (1)|P |, (4.3)

where |P | is the minimum number of transpositions P contains. If |P | is an even (odd) integer, P is said to be an even (odd) permutation. Permutations can be written as products of cycles. The simplest cycles are the two-cycles (ij ), which are just transpositions Pij . Three-cycles (ijk ) = Pijk = (i j, j k, k i) etc.
Permutation operators

Consider a system consisting of N identical particles. Let {|i } be a complete orthonormal basis for 1-particle state space H. The direct product states |i1 |i2 |iN |i1 |i2 |iN |i1 , . . . , iN (4.4)

form a basis {|i }N for the state space HN = H(1) H(2) H(N ) of the N -particle system (see sections 1.7.2 and 1.7.3). Let us dene the linear corresponding to permutation P SN by permutation operator P specifying its operation on the basis states: |i1 |i2 |iN = |iP 1 (1) |iP 1 (2) |iP 1 (N ) . P (4.5)

Often we drop the hat and denote the permutation and the corresponding operator with the same symbol P . For example let us choose N = 3, and the permutation P123 = (1 2, 2 3, 3 1). We see that 1 P123 = (2 1, 3 2, 1 3) and P123 | | | = | | | (4.6) etc. Note the inverse order in applying the permutation to state vectors! In the operation above the state of particle 1 is changed to be the state of
66

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

particle 2, the state of particle 2 is changed to be the state of particle 3, and the state of particle 3 is changed to be the state of particle 1. Let us next nd the representation of the permutation operators in the {|i }N basis of HN . Let | =
i1 ,...,iN

c(i1, . . . , iN )|i1 |iN

(4.7)

be a vector in HN . It follows that P | = =


i1 ,...,iN i1 ,...,iN

c(i1 , . . . , iN )P |i1 |iN c(i1 , . . . , iN )|iP 1(1) |iP 1 (N ) . (4.8)

Let us now rename the dummy summation indices as i iP () . It follows P | = =


iP (1) ,...,iP (N ) iP (1) ,...,iP (N )

c(iP (1) , . . . , iP (N ))|iP (P 1 (1)) |iP (P 1 (N )) c(iP (1) , . . . , iP (N ))|i1 |iN (4.9)

=
i1 ,...,iN

c(iP (1) , . . . , iP (N ))|i1 |iN .

Hence the expansion coefcients of the state P | are c(iP (1) , . . . , iP (N )), i.e. the permutation operator has in this basis the representation P c(i1 , . . . , iN ) = c(iP (1) , . . . , iP (N ) ). Note that as usual, we denote the permutation operator and its various representations with the same symbol, and thus we have simply P | = c(i1, . . . , iN )P |i1 |iN = [P c(i1 , . . . , iN )]|i1 |iN . (4.10)

i1 ,...,iN

i1 ,...,iN

(4.11) where in the second step P is the permutation operator operating on vectors in HN , and in the last step P is the representation of the permutation operator in basis {|i }N , operating on expansion coefcients c(i1, . . . , iN )!
67

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

Especially, consider the coordinate representation corresponding to basis vectors |x1 , 1 |x2 , 2 |xN , N |1 |2 |N , (4.12) where the s denote the spin degrees of freedom (see section 1.7), and we denote for simplicity often just 1 x1 , 1 etc. The expansion coefcients in this basis are the wave functions (1, 2, . . . , N ), and the permutation operators act on them as3 P (1, 2, . . . , N ) = (P (1), P (2), . . . , P (N )). Properties: Permutation operators are unitary: | = =
i1 ,...,iN

(4.13)

c(i1, . . . , iN )d(i1, . . . , iN )
i1 ,...,iN

c(iP (1), . . . , iP (N ) )d(iP (1), . . . , iP (N ) ) = P |P , (4.14)

hence |P = P 1 |P 1 P = P 1 | | , | P = P 1 . (4.15)

All permutation operators do not commute for N 3. Hence there does not exist a complete basis for HN such that the basis vectors would be eigenvectors of all P SN for N 3.4 However, one can nd proper subspaces that are invariant under all permutations: A subspace S HN is invariant under permutations if | S P | S P SN . (4.16) See Schwabl for examples of different invariant subspaces in case N = 3.
One can show this just by replacing the sums by integrals in the above derivation. For N = 2 this is possible; an arbitrary wave function (1, 2) can be expressed as a sum of symmetric and antisymmetric states s (1, 2) = (1, 2) + (2, 1) and a (1, 2) = (1, 2) (2, 1).
4 3

68

TFY-44.130 Kvanttimekaniikka II
Symmetric operators

S. Virtanen 2007

An operator O in HN is symmetric (with respect to permutations of particles) if it commutes with all P SN : [P, O] = 0. Assertion: An operator O is symmetric P |O|P = |O| | . | (4.18) (4.17)

Proof: is clear due unitarity of P , and follows from

|O| = P |O|P = |P OP | = |P 1 OP | P 1 OP = O,

(4.19)

where the last implication follows from the general result derived in QM I.
Identical particles

What does one mean by saying that two particles are identical? Essentially the following: States that differ only by permutations of identical particles cannot be distinguished by any observations. Since expectation values of observables are certainly measurable, this principle implies |O| = P |O|P | (4.20)

for any operator O corresponding to an observable. However, according to the result (4.18) this implies All operators corresponding to observables are symmetric under permutations of identical particles: [P, O] = 0 P SN .

Especially, the Hamiltonian operator of an N -particle system consisting of identical particles is symmetric, i.e. it commutes with all permutation operators: [P, H ] = 0. (4.21)
69

TFY-44.130 Kvanttimekaniikka II
4.1.2 Completely Symmetric and Antisymmetric States

S. Virtanen 2007

The state | HN is totally symmetric, if Pij | = +| Pij | = | (4.22) for all transposition operators Pij SN . The state is totally antisymmetric, if (4.23) for all transpositions. Since all permutations can be written as products of transpositions, totally symmetric states are in fact simultaneous eigenstates of all permutation operators P SN .5 We nd in general P | = (1)|P || , where the upper (lower) sign corresponds to totally symmetric (antisymmetric) state. Let us dene the symmetrization operators 1 S = N! (1)|P |P, (4.25) (4.24)

where the sum goes over all permutation operators. Properties: We note rst that follows that 1 P S = N! 1 = N!
Q f ( Q)

Q f ( P Q)

for xed P SN .6 It (1)|P


1

1 (1)|Q|P Q = N! (1)|P
1

P Q|

PQ (4.26)

Q|

1 ( ) = (1)|P | N!
5 6

(1)|Q|Q

Note that the noncommutativity of P s does not prevent the existence of some common eigenstates. This follows from the fact that {P Q|Q SN } = SN : Clearly {P Q|Q SN } SN . Moreover, P Q1 = P Q2 implies Q1 = Q2 , hence {P Q|Q SN } contains all the N ! permutations.

70

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

where in step () we have noted that (1)|P Q| = (1)|P |(1)|Q| and 1 (1)|P | = (1)|P |. It follows P S = (1)|P |S . (4.27) If | H is an arbitrary state vector, we see that S+ | is totally symmetric and S | totally antisymmetric! If | HN is totally (anti)symmetric, then 1 S| = N! 1 (4.24) = N!
P

(1)|P |P | N! (1) (1) | = | = N !| , N!


|P | |P |

(4.28)

where S+ is used if | is totally symmetric, and S if it is totally antisymmetric. Let then | =


i1 ,...,iN

c(i1, . . . , iN )|i1 |iN

(4.29)

be a totally symmetric / antisymmetric vector in HN . According to (4.28) we get 1 1 | = S| = c(i1, . . . , iN )S |i1 |iN N! N ! i1 ,...,iN 1 [Sc(i1 , . . . , iN )]|i1 |iN , = N ! i1 ,...,iN

(4.30)

where in the last step S is the representation of S in the {|i }N basiscompare with (4.11). But due to orthonormality of the basis vectors |i1 |iN , the expansion coefcients are unique and Eqs. (4.29) and (4.30) imply 1 (4.31) c(i1, . . . , iN ) = Sc(i1 , . . . , iN ). N! This means that the expansion coefcients of totally (anti)symmetric states in orthonormal bases are totally (anti)symmetric.7 Furthermore, Eq. (4.30)
7 Total (anti)symmetricity of expansion coefcients is dened analogous to the case of state vectors, i.e. P c(i1 , . . . , iN ) = (1)|P | c(i1 , . . . , iN ).

71

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

implies that totally (anti)symmetric vectors can be written as linear combinations of the totally (anti)symmetric basis vectors S |i1 |iN . (4.32)

Note that these vectors span only the subspace of totally (anti)symmetric states in HN , not the whole HN (a vast majority of vectors in HN are not totally (anti)symmetric)!
4.1.3 Bosons
States, Fock Space, Creation and Annihilation Operators

Let us now assume that the N identical particles we are considering are bosons. Based on experiments, we know that the state vector of such a system has to be (totally) symmetric with respect to interchange of the particle indices, and thus the physical states of the system belong to the + subset HN HN consisting of totally symmetric N -particle states. Since symmetric operators map totally symmetric states to totally symmetric states (prove this!), and all physical operators have to be symmetric, we may just + instead of HN . restrict the quantum mechanical state space to be HN
+ In the previous section we showed that HN is spanned by vectors S+ |i1 |iN . Let us calculate the norm of these vectors: Let ni be the number of times that the state |i occurs in the vector |i1 |iN (so-called occupation number of state |i ). When expanded, S+ |i1 |iN is a sum of ! N ! basis vectors. Of these terms n1 !N n2 ! are different, and each of different terms occurs n1 !n2! times. Since the different basis states are orthonormal, we nd

N! = n1 ! n2 ! . n1 ! n2 ! (4.33) On the other hand, order of |i s in S+ |i1 |iN is irrelevant (all information contained in occupation numbers ni ), hence we can specify the states by using just the occupation numbers. The normalized occupation
i1 , . . . , iN |S+ S+ |i1 , . . . , iN

1 N!

( n1 ! n2 ! ) 2

72

TFY-44.130 Kvanttimekaniikka II number states are dened as |n1 , n2, . . . =

S. Virtanen 2007

1 S+|i1 |iN . n1 ! n2 !

(4.34)

Moreover, if the occupation numbers of two such states are different, all the terms in their expansions in the |i1 , . . . , iN basis are orthogonal, and we nd altogether the orthonormalization relation n1 , n2, . . . |n1 , n2, . . . = n1 n1 n2 n2 (4.35)

+ , the closure and, because these states also form a complete basis in HN relation |n1, n2, . . . n1 , n2, . . . | = (4.36) n1 ,n2 ,... + holds in HN . Note that above the occupation numbers always sum up to the total particle number:

ni = N ;

(4.37)

i=1 + this is equivalent to the fact that |n1 , n2, . . . HN .

Now comes a crucial point in transferring to second quantized formalismwe enlarge the state space to contain state spaces corresponding to all different total particle numbers:
+ + + H+ = H0 H1 HN .

(4.38)

Here, the direct sum symbol means that all | H+ are of the form | = |
0

(4.39)

+ + . Especially, the vectors of HN now correspond to the where | i Hi subspace {| = 0 0 | N 0 } (4.40)

of H+ .

The direct sum space is a Hilbert space: If we have | , | H+ , i.e. | = |


0

| = | 0 |
73

(4.41)

TFY-44.130 Kvanttimekaniikka II then we dene their sum to be | + | = (| and if a C, we have a| = (a| 0 ) (a| 1 ) .
0

S. Virtanen 2007

+ | 0 ) (|

+ | 1 ) ,

(4.42)

(4.43)

Furthermore, the scalar product is dened in the direct sum space as | =


0

+ 1 |

+ .

(4.44)

Note that for example R3 = R R R; this gives a good intuitive idea what the direct sum is all about. Usually physicists are not this rigorous in notation, and just write + instead of . But now back to the space H+ . It is called the (bosonic) Fock space. The + peculiar zero-particle space H0 is one-dimensional, and contains only the scalar multiples of the vacuum state |0 . Note that the occupation number states |n1, n2, . . . span the whole H+, if the restriction (4.37) is relaxed! Thus {|n1 , n2, . . . } is an orthonormal basis of H+ .8 Ok, up to now we have done nothing but just enlarged the state space to contain all different particle number states. This is necessary because we want to express all physical operators in term of particle creation and annihilation operators, and these operators indeed change the total particle numbers of states. We dene the (bosonic) creation operator a such that ni + 1| . . . , ni + 1, . . . . a i | . . . , ni , . . . =
8

(4.45)

Lets see how the adjoint of this, the annihilation operator a operates on the
Note that the relations (4.35) and (4.36) automatically hold also in the Fock space, and fulll the orthogonality requirement of states having different total particle numbers.

74

TFY-44.130 Kvanttimekaniikka II basis states: ai | . . . , ni, . . . =


(4.36)

S. Virtanen 2007

n 1 ,n2 ,...

| . . . , ni, . . . . . . , ni, . . . |ai | . . . , ni, . . . | . . . , ni, . . . | . . . , ni, . . . ni + 1 . . . , ni + 1, . . . | . . . , ni, . . . ni + 1 n1 n1 ni+1,ni (4.46)

(4.45)

n 1 ,n2 ,...
(4.35)

n 1 ,n2 ,...

ni | . . . , ni 1 , . . . .

We thus see that a i adds one particle into state i, and ai removes one particle 9 from state i. Let us calculate the commutators of these operators. If i < j , we get [ai , aj ]| . . . , ni, . . . , nj , . . . = (aiaj aj ai )| . . . , ni, . . . , nj , . . . = ( ni nj nj ni ) | . . . , ni 1 , . . . , nj 1 , . . . = 0; (4.47) similarly if i > j . If i = j , naturally [ai , aj ] = 0. Altogether, we thus see that [ai , aj ] = 0 for any i, j , which also implies [a i , aj ] = [ai , aj ] = 0. On the other hand, if i < j , we get
[ai , a j ]| . . . , ni , . . . , nj , . . . = (ai aj aj ai )| . . . , ni , . . . , nj , . . . = ( ni nj + 1 nj + 1 ni ) | . . . , ni 1 , . . . , nj + 1 , . . . = 0, (4.48)

and similarly for i > j . If i = j , we have


[ai, a i ]| . . . , ni , . . . = (ai ai ai ai )| . . . , ni , . . . = ( ni + 1 ni + 1 ni ni ) | . . . , ni , . . . = | . . . , ni , . . . .
9

(4.49)
ni = 0.

Note that for ni = 0 the state | . . . , ni 1, . . . is ill-dened, but this does not matter since in that case

75

TFY-44.130 Kvanttimekaniikka II Altogether, we thus nd the commutation relations [ai, aj ] = 0,


[a i , aj ] = 0,

S. Virtanen 2007

[ai, a j ] = ij .

(4.50)

These are called the canonical Bose commutation relations. Note that we can create all occupation number basis states by operating with the creation operators on the vacuum state |0 |0, 0, . . . : 1 | . . . , ni , . . . = a | . . . , ni 1 , . . . , ni i hence recursively we nd 1 n1 n2 (a |n1 , n2, . . . = 1 ) (a2 ) |0 . n1 ! n2 !
The Particle-Number Operator

(4.51)

(4.52)

The particle number operator n i = a i ai counts the number of particles in state i, because n i | . . . , ni , . . . = ni | . . . , ni , . . . (4.54) (4.53)

the occupation number state basis is thus a common eigenbasis of these operators with eigenvalues ni. The total particle number operator is = N
i

n i,

(4.55)

because |n1 , n2, . . . = N


i

ni |n1 , n2, . . . .

(4.56)

Note that not all states in H+ are particle number eigenstates! For example 1 1 . | = |0, 2, 0, . . . is not an eigenstate of n 1, n 2, nor N |1, 0, 0, . . . + 2 2
76

TFY-44.130 Kvanttimekaniikka II
General Single- and Many-Particle Operators

S. Virtanen 2007

It turns out that not only the particle number operator, but all symmetric operators can be expressed in terms of the creation and annihilation operators a i , ai . In order to see what kind of operators we encounter for many-particle systems, we recall that the Hamiltonian for a system consisting of N identical, charge- and spinless particles can be usually written in the form H =T +U +F =

t +

u +

1 2

f ,
=

(4.57)

where , = 1, 2, . . . , N , and 2 p t = , u = U ( r ) (4.58) 2m are the kinetic and external potential energy operators of particle , and f = V ( r , r ) (4.59)

is the interaction energy operator between particles and . All the operators T , U and F are symmetric (verify!), as is required for physical operators for identical particles. The operators T and U are single-particle operators, since they are sums of operators of the form s = I1 I1 s I+1 IN (4.60)

that operate nontrivially only on the state of particle . Similarly, F is a two-particle operator, since it is a sum of operators of the form p = I1 I1 p I+1 I 1 p I +1 IN . (4.61) Similarly one can dene a general n-particle operator, but it turns out that we usually need to consider only symmetric single- and two-particle operators. Using the basis {|i }N , it is straightforward to show that we can represent an operator acting only on particle as t =
ij

tij |i
77

j | ,

(4.62)

TFY-44.130 Kvanttimekaniikka II where the outer product operator |i and tij = i|t |j

S. Virtanen 2007

j | = I1 |i j | IN ,
particle

(4.63)

= i|t|j

(4.64)

are the matrix elements of the operator in the single-particle basis (note that we can drop the particle indices in the single particle matrix element, because the matrix element gets the same value for all ). From this one nds the representation of the corresponding symmetric single-particle operator |i j | . (4.65) t = tij T =
ij

Similarly, one nds for a symmetric two-particle operator the representation F = where |i |j and are the matrix elements of operator f between two-particle states. For example, typical 2-particle interaction operators are in the coordinate representation of the diagonal form10 r | r |V ( r , r )|r |r = (r r ) (r r )V (r r ),
10

1 2

i,j,k,m

i, j |f |k, m

|i |j

k | m| ,

(4.66)

k | m| = I1 |i k | |j m| IN ,
particle particle

(4.67)

i, j |f |k, m = i| j | f |k |m

(4.68)

(4.69)

and by calculating the matrix elements in the coordinate representation, we


This kind of interaction is termed local.

78

TFY-44.130 Kvanttimekaniikka II get i, j |f |k, m = = dr dr dr dr

S. Virtanen 2007

i| j | |r |r r | r |V ( r , r )|r |r r | r ||k |m dr dr dr dr

i (r )j (r ) (r r ) (r r )V (r r )k (r )m (r )

dr dr i (r )j (r )V (r r )k (r )m (r ),

(4.70) where i (r) = r|i is the one-particle wavefunction of state |i . Okay, now we are going to show that we can express the symmetric + operators that operate in HN H+ in terms of the creation and annihilation operators in a very simple way. Consider rst the symmetric single-particle operator O = |i j | . Assume rst that i = j . We nd O | . . . , ni , . . . , nj , . . . = 1 |i j | S+ |i1 , i2, . . . , iN n1 ! n2 ! (4.71) 1 = |i j | |i1 , i2, . . . , iN , S+ n1 ! n2 !
(4.34)

since [O, S+] = 0 due to symmetricity of O and the fact that S+ consists of permutations. Now since i|j = ij , the operator |i j | acting on vector |i1 , i2, . . . , iN yields nj terms in each of which one i = j is replaced by i = i. Subsequent operation by S+ thus yields O | . . . , ni , . . . , nj , . . . 1 = nj n1 !n2! (ni + 1)! (nj 1)! | . . . , ni + 1, . . . , nj 1, . . . n1 ! n2 ! nj ni + 1 | . . . , ni + 1 , . . . , nj 1 , . . . = nj = nj ni + 1 | . . . , ni + 1 , . . . , nj 1 , . . . = a i aj | . . . , ni , . . . , nj , . . . . (4.72)
79

TFY-44.130 Kvanttimekaniikka II Hence we nd

S. Virtanen 2007

|i

j | = a i aj

(4.73)

if i = j . If on the other hand i = j , above just index i is replaced ni times by itself, and one nds O| . . . , ni, . . . = ni| . . . , ni, . . . = a i ai | . . . , ni , . . . , (4.74)

so that Eq. (4.73) holds also in this case. Altogether this result implies that a general symmetric single-particle operator can be written in the form T =
ij

tij

|i

j | =

tij a i aj .
ij

(4.75)

This is a very simple and beautiful result that contains a whole lot of combinatorics and proper normalization conditions! Note that if the operator t is diagonal in the {|i } basis, i.e. tij = i|t|j = tii ij , we get T =
i

(4.76) (4.77)

tii a i ai =
i

tiin i,

i.e. the operator just counts the particles in each state i and multiplies the result by the diagonal matrix elements. This is very intuitive! Similarly one nds the following representation for symmetric 2-particle operators: F = 1 2 i, j |f |k, m |i |j

i,j,k,m

k | m| =

1 2

i,j,k,m

i, j |f |k, m a i aj am ak ,

(4.78) and correspondingly for a general symmetric n-particle operator! Heuristic understanding of this: In diagonal basis k = i and m = j , and
a in j ij n i (4.79) i aj ai aj = ai ai aj aj ai [ai , aj ]aj = ai ai aj aj ij ai aj = n

counts the number of pairs that can be formed from particles in states i and j . Note that this works correctly also in the case i = j !
80

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

In conclusion, in our quantum mechanical description of N identical bosons we have rst neglected all but totally symmetric states of the state space by + restricting the state space to HN HN that is spanned by the occupation number states |n1, n2, . . . that obey Eq. (4.37) (this is possible since all the + physical states belong to HN and all physical operators are symmetric, i.e. + + ). Then we have enlarged the state they map vectors in HN to vectors in HN + space to H by allowing states with any particle numbers; H+ is spanned by occupation number states without the restriction (4.37). Then we have expressed all physical (symmetric) operators in terms of bosonic creation and annihilation operators. In this new formulation one can represent state vectors as linear combinations of vectors of the form 1 n1 n2 (a |n1 , n2, . . . = 1 ) (a2 ) |0 , n1 ! n2 ! i.e. all states can be written in the form
| = f (a 1 , a2 , . . .)|0 ,

(4.80)

and since also operators can be written in terms of a i and ai , all calculations reduce to the algebra of creation and annihilation operators! The other main advantage we have gained is that states of the form (4.80) are always properly symmetrized (totally symmetric), so that we do not have to explicitly symmetrize all the physical states! Furthermore, because of the reduction to creation and annihilation operator calculations it is natural to use the Heisenberg picture in which all the time-dependence is transferred to the operators, instead of the Schrdinger picture we have used so far. However, nothing prevents in principle using the Schrdinger picture.

81

TFY-44.130 Kvanttimekaniikka II
4.1.4 Fermions
States, Fock Space, Creation and Annihilation Operators

S. Virtanen 2007

Up to now we have developed the creation/annihilation operator formalism only for bosons. It turns out that this works also for fermions, if we only replace the canonical bosonic commutation relations by canonical fermionic anticommutation relations. So let us consider a system consisting of N identical fermions. The physical states of identical fermions are totally antisymmetric, and the state space can be restricted to subset HN HN consisting of all totally antisymmetric is spanned by the N -particle states. As shown earlier in Section 4.1.2, HN vectors S |i1 |iN that can be also written in the form of Slater determinants S |i1 , . . . , iN 1 = N! |i1 . . . |iN
1

|i1 . . . |iN

|i1 N . . . . |iN N

...

(4.81)

Due to total antisymmetricity of these states, they have the property S | . . . , i , . . . , i , . . . = S | . . . , i , . . . , i , . . . , (4.82)

which implies that occupation numbers ni cannot take values larger than one! This is the Pauli exclusion principle, stating that no two identical fermions can occupy the same quantum state. Analogously to the bosonic case we dene the (fermionic) occupation number states by setting |n1 , n2, . . . = S |i1, . . . , iN . Note two things: We do not specify yet which sign should be used above. The sign depends on the order of i s on the rhs, and we will x it a bit later. The additional normalization factor 1/ n1 !n2! (cf. Eq. (4.34)) is not required here, because ni = 0, 1.
82

(4.83)

TFY-44.130 Kvanttimekaniikka II Again analogously the occupation number states with ni = N


i

S. Virtanen 2007

(4.84)

form an orthonormal basis for HN . And again we enlarge the state space to contain all totally antisymmetric states having any total particle number, and get the (fermionic) Fock space H = H0 H1 HN ,

(4.85)

which is spanned by all occupation number states without the restriction (4.84) (but always ni = 0, 1, of course). And consequently we have n1 , n2, . . . |n1 , n2, . . . = n1 n1 n2 n2 and
1 1

(4.86)

n1 =0 n2 =0

|n1, n2, . . . n1 , n2, . . . | =

(4.87)

holds in H . Let us now dene the fermionic creation and annihilation operators. The creation operator acts on occupation number states as
Si a i | . . . , ni , . . . = (1) ni ,0 | . . . , ni + 1, . . . ,

(4.88)

where the phase factor Si for the given occupation number state is Si =
j<i

nj .

(4.89)

Note the two differences compared to the bosonic case: The phase factor (1)Si . The factor ni,0 ensures that that the occupation number of the resulting state does not exceed the maximum possible value 1.

83

TFY-44.130 Kvanttimekaniikka II The denition of a i implies its adjoint to act as ai | . . . , ni, . . . =


(4.88)

S. Virtanen 2007

n 1 ,n2 ,...

| . . . , ni, . . . . . . , ni, . . . |ai | . . . , ni, . . . | . . . , ni, . . . (1)Si ni,0 . . . , ni + 1, . . . | . . . , ni, . . .


n 1 ,n2 ,...

= (1) | . . . , 0, . . . ni ,1

n 1 ,n2 ,... Si

(1)Si ni,0 | . . . , ni, . . . n1 n1 ni+1,ni

= (1)Si ni,1 | . . . , ni 1, . . . ,

(4.90)

where Si = j<i nj . We see that indeed a i creates a particle into state i, and ai removes such an occupation. These operators are in many respects similar to their bosonic counterparts. For example
Si Si a i ai | . . . , ni , . . . = (1) ni ,1 (1) | . . . , ni , . . . = ni | . . . , ni , . . . , (4.91) hence n i = ai ai is again the occupation number operator of state i, and the total particle number operator is given by = N n i. (4.92) i

Lets see how one can create the occupation number states by operating with creation operators on the vacuum state. When particles with largest state indices are created rst, the phase factors Si appearing in (4.88) vanish and we obtain n1 n2 |n1 , n2, . . . = (a (4.93) 1 ) (a2 ) . . . |0 , to be compared to (4.52). However, in one important respect the fermionic operators differ from the bosonic ones. Lets see how the fermionic creation and annihilation operators commute: For i < j we get ai aj | . . . , ni, . . . , nj , . . . = (1)Sj (1)Si nj ,1 ni ,1| . . . , 0, . . . , 0, . . . , (4.94)
84

TFY-44.130 Kvanttimekaniikka II but

S. Virtanen 2007

aj ai | . . . , ni, . . . , nj , . . . = (1)Si (1)Sj 1nj ,1ni ,1| . . . , 0, . . . , 0, . . . , (4.95) hence we nd ai aj = aj ai . Note that the sign changes when the order of two annihilation operators is changed! Such operators are said to anticommute. We dene the notation {A, B } [A, B ]+ AB + BA, [A, B ] [A, B ] AB BA, (4.96) (4.97)

for arbitrary operators A and B ; the rst line denes the anticommutator of these operators. Especially we get {ai , aj } = ai aj + aj ai = 0 for i = j . But in fact this holds also for i = j , since (4.90) implies (ai)2 = 0! Furthermore, by taking the Hermitian conjugate we nd {a i , aj } = 0. How about commutation of ai and a j ? If i < j , we get
(ai a j + aj ai )| . . . , ni , . . . , nj , . . .

= [(1)Sj (1)Si + (1)Si (1)Sj 1]nj ,0 ni,1 | . . . , 0, . . . , 1, . . . = 0, (4.98)

hence {ai , a j } = 0. But if i = j , we have


Si Si Si Si (ai a i + ai ai )| . . . , ni , . . . = [ni ,0 (1) (1) + ni ,1 (1) (1) ]| . . . , ni , . . . = | . . . , ni , . . . , (4.99)

hence {ai , a i } = 1. Altogether we get the relations {ai , aj } = 0,


{a i , aj } = 0,

{ai , a j } = ij ,

(4.100)

which are called canonical fermionic anticommutation relations. These relations reect the fact that fermionic states are totally antisymmetric. Now we nally also specify the sign in (4.83) by stating that
S|i1 , i2, . . . , iN = a i1 ai2 aiN |0 ;

(4.101)

85

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

compared to Eqs. (4.83) and (4.93), we see that by using the anticommutation relations for creation operators, this relation xes the sign that was previously left undetermined. This is just a matter of convention, any other consistent choise would in principle work as well.
Single- and Many-Particle Operators

Analogously to the bosonic case also operators that are symmetric with respect to permutations of identical fermions can be expressed in terms of fermionic creation and annihilation operators. Furthermore, the operators are exactly of the same form as in the bosonic case! Especially, one gets T =
ij

tij

|i

j | =

tij a i aj . ,
ij

(4.102)

and F = 1 2 i, j |f |k, m |i |j

i,j,k,m

k | m| =

1 2

i,j,k,m

i, j |f |k, m a i aj am ak ,

(4.103) for symmetric one- and two-particle operators; see Schwabl, subsection 1.4.2 and Eq. (1.3.26) on page 16. Note that since fermionic operators do not commute, the order of the operators is important. Especially, note the different order of indices k and m in the matrix element and in the annihilation operator part of the 2-particle operator!

86

TFY-44.130 Kvanttimekaniikka II
4.1.5 Field Operators

S. Virtanen 2007

This section applies both for the bosonic and the fermionic case.
Transformations Between Different Basis Systems

Above we have developed the theory using a single-particle basis set {|i }. Naturally we can choose any orthonormal basis for this purpose, and the results are invariant. In fact the formalism expressed in terms of creation and annihilation operators a i , ai corresponding to basis {|i } can be rather simply expressed in terms of creation and annihilation operators a , a corresponding to another basis {| }. This can be shown rigorously, but we content ourselves to the following a bit heuristic argument: The single-particle state | can be expanded in the basis |i as | =
i

|i i| .

(4.104)

Now the operator a i creates a particle in the state |i . It follows that the operator a i| a (4.105) i =
i

creates one particle in the state | , i.e. it is the creation operator corresponding to this state. For the corresponding annihilation operator it follows a = |i ai . (4.106)
i

Field Operators

Let us transform to use the creation and annihilation operators corresponding to the position basis |x , instead of the {|i } basis. We have x|i = i (x),
87

(4.107)

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

where i (x) is the wave function of the single-particle state |i in the position representation. According to (4.105) and (4.106) the creation operator corresponding to position basis is (x) =
i i (x)ai ,

(4.108)

and the annihilation operator is (x) =


i

i (x)ai .

(4.109)

These operators are called eld operators, and they create and annihilate, correspondingly, a particle at the position x. From the canonical commutation/anticommutation relations (4.50) and (4.100) we nd [ (x), (x)] = [ (x), (x)] = [ (x), (x)] =
ij

i (x)j (x )[ai, aj ] = 0,
i (x)j (x )[ai , aj ] = 0, i (x) j (x )[ai , aj ] = i (x) i (x ) = (x x ),

ij

ij

(4.110) where the last step follows from the representation of the closure relation i |i i| = I in the position basis, and the upper sign applies to fermions and the lower one to bosons. Note the similarity to the canonical (anti)commutation relations, the only difference is just that the |x basis is continuous, while {|i } is discrete. Now we can express all symmetric operators in terms of the eld operators;

88

TFY-44.130 Kvanttimekaniikka II for example the kinetic energy operator takes the form T =
ij

S. Virtanen 2007

a i i|t|j aj a i aj dx i (x) a i aj
ij 2 2

=
ij 2

2m

2 j (x) (4.111)

2m

dx i (x)j (x)

(4.108,4.109)

2m where in the second step we have used partial integration and the i (x)s vanish at innity such that no boundary terms appear.11 Similarly the two-particle interaction operator gets the form 1 F = i, j |f |k, m a i aj am ak 2
ijkm

dx (x) (x),

= =

1 2 1 2

a i aj am ak ijkm

dx

dx i (x)j (x )V (x, x )k (x)m (x )

dx

dx (x) (x )V (x, x) (x) (x), (4.112)

and the Hamiltonian


2

H=

dx

2m

(x) (x) + U (x) (x) (x) 1 2 dx dx (x) (x )V (x, x) (x) (x). (4.113)

+ Particle density:

The particle density operator is a symmetric single-particle operator, whose position representation is of the form (see exercises) n(x) =

(x x )

n (x),

(4.114)

11 Note that (x) (x) = (x) (x), i.e. we have a dot product of vectors hereofter this dot is not denoted explicitly.

89

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

where x denotes the coordinate of particle . For single-particle states its matrix elements are just

i|n (x)|j

dx i (x ) (x x )j (x ) = i (x)j (x),

(4.115)

hence it can be written as n(x) =


ij

a i aj i|n (x)|j

= (x) (x).

(4.116)

Total-particle-number operator: Total number of particles is the integral of the particle density, hence = N dx n(x) = dx (x) (x).
i ai ai .

(4.117)

= This can be veried also by using the fact that N Current density operator:

In exercises it is shown that the current density operator gets the form j(x) = 2im (x) (x) [ (x)] (x) . (4.118)

Field equations

So far we have presented the operators in the Schrdinger picture. However, as mentioned earlier, in the formalism we have developed it is more convenient to use the Heisenberg picture, in which the states are time-independent, but operators develop in time according to12 OH (t) = eiHt/ OS (t)eiHt/ . (4.119)

This implies the operators to satisfy the Heisenberg equation of motion d i OH = [H, OH ] + dt OS t (4.120)
H

12 This form applies only for conservative systems, while the following Heisenberg equation of motion is valid generally. We derive the eld equations assuming the system to be conservative, but the nal result is valid also more generally.

90

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

As we have expressed the relevant operators above in terms of the (Schrdinger picture) eld operators (x) and (x), it is sufcient to nd the time-development of the corresponding Heisenberg picture eld operators (x, t) H (x, t). The Heisenberg equation of motion implies the following equation of motion for the eld operator:13 i (x, t) = [H, (x, t)] = [H, eiHt/ (x)eiHt/ ] t = eiHt/ [H, (x)]eiHt/ , (x, 0) = (x). Now we would like to nd out the commutator [H, (x)] =
(4.113)

(4.121)

where we have used the initial condition (4.122)

dx + 1 2

2m dx

[ (x) (x), (x)] + U (x)[ (x ) (x), (x)] dx [ (x ) (x)V (x, x ) (x) (x), (x)] (4.123)

for the bosonic and fermionic casenote that the eld operators naturally commute with scalar functions, as is seen from their denitions. Since [AB, C ] = ABC CAB = A(BC CB )+(AC CA)B = A[B, C ]+[A, C ]B, (4.124) and on the other hand [AB, C ] = ABC CAB = A(BC +CB )(AC +CA)B = A{B, C }{A, C }B, (4.125) we get altogether [AB, C ] = A[B, C ] [A, C ]B. (4.126)

This is useful in evaluating the commutators of eld operators: for bosons it is convenient to use the lower signs, and for fermions the upper. We get for
13

The eld operator does not have time dependence in the Schrdinger picture, hence the last term in (4.120) is zero.

91

TFY-44.130 Kvanttimekaniikka II the kinetic energy term dx [ (x) (x ), (x)] = Now

S. Virtanen 2007

dx (x )[ (x ), (x)] [ (x), (x)] (x ) . (4.127)

1 (4.110) (x ), (x)] = lim [ (x + hex ) (x ), (x)] = 0 h0 h x etc., and similarly [ [ (x ), (x)] = (x x). It follows dx [ (x) (x ), (x)] = =

(4.128)

(4.129)

dx { (x x) (x )}

dx (x x)2 (x ) = 2 (x), (4.130)

where we have used partial integration. Similarly we get for the potential energy term dx U (x )[ (x ) (x), (x)] = and for the interaction energy term 1 2 dx = since [ (x) (x), (x)] = 0 (4.133) for both bosons and fermions. Since [ (x ) (x), (x)] = (x )[ (x), (x)] [ (x ), (x)] (x) = (x x) (x ) (x x) (x ),
92

dx U (x)[ (xx) (x)] = U (x) (x), (4.131)

dx [ (x) (x)V (x , x) (x) (x), (x)] dx dx [ (x ) (x ), (x)]V (x, x ) (x) (x), (4.132)

1 2

(4.134)

TFY-44.130 Kvanttimekaniikka II we get 1 2 = 1 2 1 = 2 1 = 2 = dx dx

S. Virtanen 2007

dx [ (x ) (x)V (x , x) (x) (x), (x)] dx [ (x x) (x ) (x x) (x )]V (x, x ) (x) (x) dx (x )V (x, x) (x) (x) dx (x )V (x, x) (x) (x) dx (x )V (x, x) (x ) (x) dx (x)V (x, x) (x) (x)

dx (x)V (x, x) (x ) (x), (4.135)

if and when the interaction between particles is symmetric: V (x, x) = V (x , x). All in all we get the equation of motion i (x, t) = t
2

(4.136)

2m

2 + U (x) (x, t) +

dx (x, t)V (x, x) (x, t) (x, t). (4.137)

Note that if we neglect the interaction term, this looks very similar to the Schrdinger equation for the quantum mechanics of a single particle! Also, for example for a single particle the probability density is |(x)|2 = (x)(x), which can be compared with expression (4.116) for the particle density of our formalism. The only difference seems to be that the single-particle wave function (x, t) has been replaced by the eld operator (x, t). Due to this formal correspondence the many-particle formalism we have developed is termed second quantization, as if the complex wave function had been quantized to become a eld operator. However, it has to be kept in mind that we have not done anything else than presented the same theory in a different language!

93

TFY-44.130 Kvanttimekaniikka II
4.1.6 Momentum Representation

S. Virtanen 2007

If the Hamiltonian of the system is translationally invariant (commutes with , no spatially dependent external potentials), and the state of the system is p translationally invariant, it is by far most convenient to use momentum representation (because the energy eigenstates can be chosen to be also momentum eigenstates). Translationally invariant system is always innite (no boundaries), but to easily derive the state density etc. it is useful to rst consider a system in a nite rectangular box with volume V = LxLy Lz and use periodic boundary conditions14 (x) = (x + Lxex ) etc. (4.138)

These boundary conditions are satised by the momentum eigenstate wavefunctions nx ny nz 1 , , , nx , ny , nz Z , x|k = k (x) = eikx , k = 2 Lx Ly Lz V (4.139) which form an orthonormal basis for the coordinate representation of the one-particle state space H. The Hamiltonian operator in terms of momentum state creation and annihilation operators is 1 + ( k|t|k + k|u|k ) a a H= k, k |f |p, p a k k k ak ap ap . 2 k,k k,k ,p,p (4.140) The matrix elements are straightforward to calculate, with the result (see Schwabl) H=
k

k 1 ak ak + 2m V

2 2

k,k

Uk k a k ak +

1 2V

q,p,k

Vq a p+q akq ak ap ,

(4.141) where Uq =
14

dx eiqx U (x),

Vq =

dx eiqx V (x)

(4.142)

One could also use Dirichlet boundary conditions (wave function vanishes at the boundaries), they yield the same result in the thermodynamic limit V, N so that density N/V = const. However, Dirichlet conditions break the translational symmetry, and we could not use the momentum representation.

94

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

are the Fourier transforms of the external and interaction potentials.


Fourier Transformation of the Density

See Schwabl, p. 27.


The Inclusion of Spin

Up to now we have neglected the spin index in the arguments (x, ) of the wave functions and operators. When taking the spin degree of freedom into account, in the second quantized formalism we just denote (x) and ap , for example. The (anti)commutation relations for the eld operators become [ (x), (x)] = 0,
[ (x), (x )] = 0,

[ (x), (x )] = (x x ).

(4.143)

In calculating for example the total density, one naturally sums over densities of particles in different spin states, resulting in n(x) =
(x) (x).

(4.144)

For more details, see Schwabl, p. 27-28.

4.2 Spin-1/2 Fermions In this section we investigate the ground state of a homogeneous system consisting of spin-1/2 fermions using second quantization. In subsection 4.2.1 we consider the rather simple case of noninteracting fermion gas, then in subsection 4.2.2 we calculate the rst-order corrections for the ground state energy due to Coulomb interaction between the fermions. Finally, in section 4.2.3 we discuss the Hartree-Fock approximation for atoms.

95

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

In principle the methods presented can be straightforwardly generalized for inhomogeneous systems at nite temperatures and even for time-dependent systems, but in practice this leads to very intricate calculations. . . 15
4.2.1 Noninteracting Fermions
The Fermi Sphere, Excitations

For noninteracting (V (x) 0) free (U (x) 0) fermions the Hamiltonian can be written in the form (compare with (4.141)) H=
k

k a ak = 2m k

2 2

k n k , 2m

2 2

(4.145)

and we note that the Hamiltonian is diagonal in the occupation number basis corresponding to single-particle momentum eigenstatesa manifestation of the conveniency of the momentum basis for homogeneous systems. What is the ground state (lowest energy state) |0 H of this Hamiltonian for an N -particle system? Since all the occupation number operators commute with the Hamiltonian, the ground state may be required to be also the common eigenstate of all n k . But this means that it has to be an occupation number eigenstate! Thus E |0 = H |0 = k n k |0 = 2m
2 2

k nk |0 . 2m

2 2

(4.146)

Since nk = 0 |n k |0 = 0, 1, we see that the lowest energy state corresponds to lling all the N lowest-energy single-particle states, i.e.16 |0 =

|k|<kF

a k |0 ,

(4.147)

where the Fermi wave number kF is chosen such that N=


k
15 16

nk = 2
|k|<kF

1.

(4.148)

See, for example Fetter & Walecka: Quantum Theory of Many-Particle Systems. 1 is the spin index, see Schwabl section 1.6.3. Here = 2

96

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

The sum is the number of states within the Fermi sphere |k| < kF . From Eq. (4.139) we see that in the k-space there is one state per volume 2 2 2 (2 )3 k = = . Lx Ly Lz V (4.149)

In the limit V the states become more and more dense and we may approximate the sums over states of smooth functions f (k) by integrals: f (k) =
k k

k
(2 )3 V

f (k)

V (2 )3

dk f (k).

(4.150)

Especially, Eq. (4.148) can be approximated as V N =2 (2 )3 implying kF = (3 2n)1/3, where n = N/V


2 2 kF 3 V 1 3 V kF dk = 2 4 k = , (2 )3 3 F 3 2

(4.151)

k<kF

(4.152) (4.153)

is the average density of the system. The energy corresponding to kF , F = is called the Fermi energy. In fact it is straightforward to show (see Schwabl) that the (local) density is n(x) = n = N/V , i.e. the system is indeed homogeneous as expected. The total energy of the ground state is E = 0 |H |0 = 0 | k n k |0 = 2m
2 2 2 k nk = 2m 2m 2 2

2m

(4.154)

(4.155) where (x) is the Heavyside step function. In the thermodynamic limit one may again replace the sum by the corresponding integral, yielding E= V 2m (2 )3 2
2

k 2 (kF k ),

dk k 2(kF k ) =

2 V 1 5 (4.152) 3 2kF 3 4 k N = F N. = F m (2 )3 5 10m 5 (4.156) 2

97

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

The simplest excitation for the system is seen to correspond to transferring a particle from the Fermi sphere (creating a hole) to occupy a state outside the Fermi sphere. Such an excited state can be written as | = a k2 2 ak1 1 |0 , (4.157)

where |k1 | < kF and |k2 | > kF . The energy of such an elementary excitation with respect to the ground state energy is
2

2m

2 (k2 2 k1 ).

(4.158)

Single-Particle Correlation Function, Pair Distribution Function and Density Correlation Functions

The pair distribution function for a homogeneous system is dened in the coordinate representation as g (x) = V N (N 1) (x x + x ) . (4.159)

Compare this with the expression for density, Eq. (4.114). Essentially, this quantity characterizes the probability density that particle pairs have separation x. By calculating the two-particle matrix elements of g (x) between position eigenstates (see Schwabl (1.5.13)), one nds that in terms of eld operators g (x) = V dx dx (x ) (x) (x x + x ) (x ) (x) N (N 1) V = dx (x ) (x x) (x x) (x) N (N 1) V V (x ) (x x) (x x) (x) = N (N 1) V2 = (x) (x x) (x x) (x ) , N (N 1) (4.160)

where translational invariance has been used in the penultimate step. For V2 1 V2 large N the prefactor can be approximated as N (N 1) N 2 = n2 .
98

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

1 (x) (x) (x ) (x) . (4.161) g (x x ) n Up to now we have neglected spinsa straightforward generalization for spin- 1 2 case is

Translational invariance allows us to write nally (set rst x x x and then x x above)
2

g (x x )

2 n

2 (x) (x ) (x ) (x) .

(4.162)

The density-density correlation function is closely related to g (x x ): G(x x ) = n (x) n(x ) = (x) (x) (x ) (x) (4.163)

G(x x ) = (x) (x) (x) (x )

= (x) (x ) (x) (x) + (x)[ (x), (x) (x)] = (x) (x ) (x) (x) + (x) (x) (x x ) N (N 1) g (x x ) + (x x )n. = V

(4.164)

Let us calculate the pair distribution function for the ground state of a homogeneous spin- 1 2 system: n 2 1 ei(kk )xi(qq )x 0 |a g (x x ) 2 k aq aq ak |0 . 2 V kk qq (4.165) If = the matrix element vanishes unless17 k = k and q = q , hence one gets n 2 1 0 |n k n q |0 g (x x ) 2 2 V
kq

1 = 2 V
17

nk n
kq

1 n 1 NN = 2 N N = 2 = V V 2 2 2

(4.166)
Note that both |0 and a k aq aq ak |0 are proportional to occupation number basis states, and two occupation number states corresponding to different occupation number sets are orthogonal.

99

TFY-44.130 Kvanttimekaniikka II i.e. g (x x ) = 1 = const.

S. Virtanen 2007

( = ).

(4.167)

One concludes that particle densities corresponding to different spin species do not correlate in any way. On the other hand, for = the matrix element has two possibilities to yield a nonzero contribution: Either k = k and q = q , or k = q and q = k . Altogether, one gets
0 |a k aq aq ak |0 = kk qq 0 |ak aq aq ak |0 + kq qk 0 |ak aq ak aq |0 = (kq qk kk qq ) 0 |a k aq ak aq |0

= (kq qk kk qq )

( 0 |a k {aq , ak }aq |0 0 |ak ak aq aq |0 ) (4.168)

Since the anticommutator vanishes unless k = q, but the Kronecker delta term is then zero, we nd nally
0 |a k aq aq ak |0 = (kq qk kk qq ) 0 |ak ak aq aq |0

= (kk qq kq qk )nk nq ,

(4.169)

yielding n 2
2

g (x x ) =

1 V2

(4.166)

kq n 2

1 ei(kq)(xx ) nk nq |G (x x )|2,

(4.170)

where
G (x x ) = 0 | (x) (x)|0 = 0 |

kk

1 ikx+ik x e ak ak |0 V

1 V

eik(xx ) nk
k

(4.171)

100

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

is the so-called single-particle correlation function. By replacing the sum with the corresponding integral, one gets G (x x ) = 1 ik(xx ) d k e (kF k ) (2 )3 1 kF 2 2 ik |xx | cos dk k d (cos ) e = (2 )3 0 1 kF 1 dk k sin k |x x | = 2 2 |x x | 0 3n sin x x cos x , = 2 x3 x = kF |x x |. Combined with (4.170) one nally gets g (x x ) 1 9 (sin x x cos x)2. 6 x (4.174)

(4.172)

where (4.173)

This function is plotted in Schwabl, Fig. 2.4. One observes essentially that 2 fermions with same spin do not want to be closer than distance F = k , the F so-called Fermi wavelength, to each other. This is a manifestation of the Pauli exclusion principle in a homogeneous fermion gas!

101

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

4.2.2 Ground State Energy and Elementary Theory of the Electron Gas

In the previous subsection we treated mainly ideal (noninteracting) fermion gas/plasma, although the general treatment (denitions etc.) for the correlation functions is valid for all homogeneous systems. The ideal gas case was rather straightforward, although it already yielded some interesting physics, for example the fermionic pair distribution function. However, fermionic ideal gas is a rather academic example and has no direct practical applications. In this subsection the analysis is extended to a homogeneous electron gas, interacting via Coulomb repulsion. In order to have a charge balance in the system, we assume the electrons to move in a positive background charge distribution. Typically this positive charge consists of pointlike ions, but for simplicity we assume it to be smoothed to a homogeneous positive charge distribution.18 The Hamiltonian for such a system can be written as
N

H=
=1

1 p2 + e2 2m 2

e|x x | e2 n |x x | =1 1 + e2 n2 2 dx

e|x x| dx |x x| e|xx | dx , (4.175) |x x |

where x are the electron position vectors, n = N V is the average density of electrons and the density of the positive background charge distribution, and e2 is the square of the elementary charge (we use Gaussian units). The screened Coulomb potential for the interaction is used in order to render all the integrals convergent. In the end of the calculation one lets 0+.

In order to nd the second quantized form of this Hamiltonian, lets calculate the matrix elements for the electron-electron interaction for momentum eigenstates (compare with Eq. (4.70)): k1 1k2 2|V |k3 3k44 = e 1 3 2 4
2

dx1

1 ik1 x1 ik2 x2 e|x1 x2 | ik3 x1 ik4 x2 dx2 2 e e . e e V |x1 x2 | (4.176)

18

This is typically an oversimplication, but the more realistic case is much too difcult for us to analyse. . .

102

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

Changing to new integration variables x = x2 and y = x1 x2 ( x1 = y + x, x2 = x) (verify that the Jacobian determinant for this transformation is 1), we get k1 1k2 2|V |k33 k44 =
y e2 i(k1 +k2 k3 k4 )x i(k3 k1 )y e d x e d y e V2 1 3 2 4 y 2 e 4 = 1 3 2 4 k1 +k2 ,k3 +k4 . V (k1 k3 )2 + 2 (4.177)

The electron interaction term in the Hamiltonian thus takes the form Hint 4 a k1 1 ak2 2 ak4 4 ak3 3 . 2 2 (k1 k3 ) + k1 1 k2 2 k3 3 k4 4 (4.178) Now it is useful to make the following change of variables: 1 3 2 4 k1 +k2 ,k3 +k4 k1 = k + q and 1 = 3 = 2 = 4 = ; (4.180) note that (4.179) guarantees that k1 + k2 = k3 + k4 , and reduces the number of wave vector variables from 4 to 3 according to the factor k1 +k2 ,k3 +k4 . In terms of the new variables, we can write Hint e2 = 2V 4 a a k + q , k q, ak ak . 2 + 2 q (4.181) k2 = k q k3 = k k4 = k (4.179) e2 = 2V

kk q

Now the q = 0 terms can be shown to cancel with the ion-ion and ion-electron interactions (see Schwabl, p. 42; in fact here a term that vanishes in the thermodynamic limit is neglected.)19, and nally letting 0+ we can write the Hamiltonian as H=
k

k e2 a ak + 2m k 2V

2 2

kk q

4 ak+q, a k q, ak ak , 2 q

(4.182)

where the prime in the summation symbol means that the term q = 0 is to be omitted.
19

For a more detailed treatment, see Fetter & Waleckas book.

103

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

We are going to approximately calculate the energy of the ground state of the system, but in order to nd the sensible approximation method it is useful to estimate the relative importance of the kinetic energy and interaction terms by introducing dimensionless variables. Let r0 be the radius of the sphere having the average volume per particle: 4 3 N. V = r0 3 (4.183)

Let the dimensionless length rs be the ratio of r0 and the Bohr radius 2 a0 = me2 : r0 (4.184) rs = ; a0 rs characterizes the density of the system. Dening furthermore the dimensionless volume and wave vectors = r 3 V V 0 = r0 k k = r0 k k = r0q, q (4.185)

we can write the Hamiltonian in the form 4 rs 1 2 e2 + k a a a ak H= ak a (4.186) k + q , k k 2 2 q , k a0 rs 2 2V q


k kk q

, k and q , k stay Now varying V and letting N be xed, we see that V constant and only rs changes, and we see that the relative importance of the interaction term with respect to the kinetic energy terms should become negligible in the limit rs 0, corresponding to the high-density limit!20 Hence, if we restrict to consider the limit rs 1, we may treat the interaction term as a small perturbation, and use rst-order perturbation theory to estimate its effect. Neglecting rst the small perturbation, the zeroth-order Hamiltonian is just the ideal (noninteracting) gas Hamiltonian analysed in the previous section, and we know the unperturbed ground state to be a (4.187) |0 = k |0 ,
|k|<kF

20 This may come a bit as a surprise, since one would expect interactions to be dominant in the high-density limit. However, it is to be remembered that also the Fermi energy (kinetic energy) grows in that limit, and it grows in fact even faster!

104

TFY-44.130 Kvanttimekaniikka II where kF = (3 2n)1/3.

S. Virtanen 2007

(4.188)

In addition we have already calculated the ground state energy in the zeroth order: E
(0)
(4.156)

2 3 3 2 kF e2 N 3 = F N = N= 25 5 5 2m 2a0 rs

9 4

2/3

e2 2.21 N 2 2a0 rs

(4.189)

The rst-order correction from perturbation theory is given by the expectation value of the interaction term in the noninteracting ground state: E
(1)

e2 = 0 | 2V e2 = 2V

kk q

4 ak+q, a k q, ak ak |0 2 q (4.190)

kk q

4 0 |a k+q, ak q, ak ak |0 . 2 q

Since q = 0, k + q = k in all terms of the sum, and the only possibility for the matrix element to give a nonvanishing contribution is k + q = k , k q = k, and = . Hence
0 |a k+q, ak q, ak ak |0 = k+q,k 0 |ak+q, ak ak+q, ak |0

= k+q,k 0 |n k+q, n k |0 = k+q,k nk+q, nk = k+q,k (kF |k + q|)(kF k ), (4.191)

and in the thermodynamic limit we get E


(1)

e2 = 2V = 2

kq

4 (kF |k + q|)(kF k ) q2 dk dq 1 (kF |k + q|)(kF k ). q2 (4.192)

V2 e2 4 2V (2 )6

The integral is left as an exercise problem, with the result E


(1)

e2 0.916 N . = 2a0 rs
105

(4.193)

TFY-44.130 Kvanttimekaniikka II Collecting the results, we get the expression E e2 2.21 0.916 = + O(ln rs) 2 N 2a0 rs rs

S. Virtanen 2007

(4.194)

for the energy per particle in the limit high-density limit rs 1; note that e2 = 1Rydberg 13.6 eV. Above, O(ln rs ) represents the higher-order 2 a0 corrections, which are termed simply the correlation energy. They are much harder to calculate, for details you may consult Fetter & Waleckas book. Our result is correct in the limit rs 1, but because we have just calculated the expectation value of energy for the state |0 , the variational principle implies that we have anyway got an upper bound for the energy of the true ground state. In the low-density limit rs 1 one can show that the so-called Wigner solid state gives a lower energy than |0 . For the Wigner solid one can derive the expansion 1.79 2.66 e2 E + 3/2 + = N 2a0 rs rs for rs 1.
4.2.3 Hartree-Fock Equations for Atoms

(4.195)

Hartree-Fock (HF) approximation essentially means that one tries to approximate the state of the system within a restricted set of states consisting of occupation number eigenstates {|n1, n2, . . . } corresponding to some orthonormal one-particle basis set {|i }. For the ground state of an ideal (noninteracting) fermion gas the HF approximation was seen to be exact, because the true ground state |0 is an occupation number basis state. For the interacting gas we in fact used the HF approximation, and in that case it really is just an approximation because the true ground state is not an occupation number eigenstate (except in the limit rs 0). In this section we derive the so-called Hartree-Fock equations for an atom consisting of N electrons and a nucleus (whose motion is neglected because

106

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

of the much larger mass) having charge number Z . The Hamiltonian is H=


ij

( i|t|j + i|u|j )a i aj + 2 p t= , 2m

1 2

ijkm

i, j |V |k, m a i aj am ak ,

(4.196)

with

Ze2 u= , r

where r = |x|. The single-particle eigenstates |i may be chosen at willin fact we determine them in the following way: In the spirit of the HF approximation, we try to nd the best possible approximation for the ground state of the system within occupation number eigenstates by using the HF Ansatz
| = a 1 aN |0 .

e2 V = , |x x |

(4.197)

(4.198)

The idea is to nd single-particle basis states |i such that the Ansatz state yields the lowest possible expectation value for the energy of the system.21 This is just the principle for the variational approximation of the ground state. First we have to nd the expectation value of energy for the Ansatz state: |H | = ( i|t|j + i|u|j ) |a i aj | + 1 2
i, j |V |k, m |a i aj am ak | .

ij

ijkm

(4.199) Since | is an occupation number eigenstate with N lowest states lled, we nd 1, if i = j N |a (4.200) i aj | = 0, otherwise, and
|a i aj am ak | = im jk |am ak am ak | + ik jm |ak am am ak | . (4.201)
Note that the Ansatz state depends on the choice of the single-particle states, because the creation operators a i depend on the single-particle states! Note also that in the Ansatz the lowest N single-particle states are lled, other are left empty.
21

107

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

Since (ai)2 = 0 always for fermions, only terms with m = k give nonzero contributions, and canonic anticommutation relations imply
|a i aj am ak | = im jk |am ak am ak | + ik jm |am am ak ak |

= (ik jm im jk ) |n mn k | =

ik jm im jk , if m, k N 0, otherwise. (4.202)

Hence 1 |H | = ( i|t|i + i|u|i ) + ( i, j |V |i, j i, j |V |j, i ) 2 i=1 i,j =1


2 N i=1 N N N N

2m +

dx |i | + dx

i=1

dx U (x)|i(x)|2

1 2 i,j =1

dx V (x x ) |i (x)|2|j (x)|2
msi msj i (x)j (x )i (x )j (x) , (4.203)

where i (x) = x|i is the wave function and msi the spin index of state |i . Minimization of this expression is left as an exercise, but the result is the conditions i i (x) = Ze2 2m r
2 N 2 N

i (x) +
j =1

e2 dx |j (x )|2i (x) |x x |

msi msj
j =1

dx

e2 j (x )i (x )j (x), (4.204) |x x |

which are the famous Hartree-Fock equations. These can be also derived directly by minimizing the energy of a many-particle wave function consisting of a Slater determinant of single-particle wave functions (presented in almost all condensed matter textbooks), but the derivation based on second quantization formalism is shorter and more elegant.
108

TFY-44.130 Kvanttimekaniikka II 4.3 Bosons


4.3.1 Free Bosons

S. Virtanen 2007

See Schwabl, section 3.1.1.


4.3.2 Weakly Interacting, Dilute Bose Gas

Only the particles mediating the four fundamental interactions (gravitation, electromagnetic, weak and strong interactions) are known to be bosonic elementary particlesquarks and leptons are fermions. Thus is seems at rst sight that there does not exist (mutually) interacting bosonic systems in the nature. However, composite particles having an even number of fermions can in fact be considered as bosons, according to the rules for calculating the total spin of a composite particle and the spin-statistics theorem.22 Thus for example atoms which consist of an even number of fermions (total number of electrons, protons and neutrons) are effectively bosons, which are furthermore identical if all the atoms have the same internal electronic state (all in the ground state, for example). The quantum statistical nature of atoms is usually too weak to be observable: typically it is totally masked by the high thermal energy and/or strong interactions between the atoms. However, in certain cases it is manifested in quite spectacular ways: For example the quantum uids 3 He and 4He become superuids at low enough temperatures, which means for example that their viscocity becomes exactly zero. The superuid transition of He superuids can be understood as Bose-Einstein condensation, in which a macroscopic fraction of particles condenses to the lowest energetic state of the system. However, because of the strong interactions present in these liquids, their physics has escaped a thorough microscopic understanding, and the relations of superuidity and Bose-Einstein condensation are still under investigation. One way to try to understand liquid helium is to study the physics of a
22

An even number of half odd-integer spins gives an integer total spin, which corresponds to a boson.

109

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

weakly interacting Bose gas, which is theoretically much easier to analyse. Recently, the theory developed for the weakly interacting Bose gas has found a concrete physical application in modelling the dilute alkali-atom Bose-Einstein condensates (found in 1995 by Eric Cornells group). These systems are nowadays a very hot topic in physics, and intensely investigated by both theorists and experimentalists.
Bogoliubov Theory of the Weakly Interacting Bose Gas

We consider only a homogeneous gas consisting of N particles per volume V , and thus use momentum representation with periodic boundary conditions. The Hamiltonian reads (see Section Momentum Representation) H = H0 + Hint =
k

1 k2 ak ak + 2m 2V

kpq

Vq a k+q apq ap ak ,

(4.205)

where we have used units in which Vq =

= 1 and dx eiqx V (x) (4.206)

is the Fourier transform of the interaction potential. Due to the symmetry V (x) = V (x) of the interaction, one can readily verify that Vq = Vq (4.207) (Prove this!). At low enough temperatures, a condensate is formed into the the lowest-energy single particle mode corresponding to k = 0. This mode becomes macroscopically occupied, i.e. for the state | of the system23 n0 = |a 0 a0 | N0 N (4.208) for k = 0, nk N . Since a 0 |N0 , . . . =
23

a0 |N0 , . . . =

N0|N0 1, . . . ,

(4.209) (4.210)

N0 + 1|N0 + 1, . . . ,

At nite temperatures, the system is not in a pure state but instead in a mixed state described by a state operator, but we neglect this aspect heresee the statistical physics courses for more details. Our treatment is essentially valid at zero-temperature.

110

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

and [a0, a 0 ] = 1, the noncommutatory operator nature of a0 and a0 is negligible when operating on this kind of states (for which N0 1; then N0 + 1 N0 to a very good accuracy, and adding or subtracting one particle to/from N0 has only a small effect). Following the idea of Bogoliubov, we thus approximate these operators just by complex numbers:

a0

N0 ,

a 0

N0 .

(4.211)

The motivation for this approximation is of course rather heuristic, but the analysis can be carried out rigorously with essentially the same results as long as N0 1. Furthermore, as the matrix elements of ak , a k with k = 0 for states under consideration are supposed to be much smaller than those for a0 , a 0 , we approximate the Hamiltonian by neglecting all terms in Hint that do not contain at least two a0 , a 0 operators. There is one interaction term which contains four of these operators (k = p = q = 0), no terms which contain exactly three (because of momentum conservation), and four different groups of terms which contain two: Hint 1 V0 a 0 a0 a0 a0 2V +2
k k k k V0 a 0 a0 ak ak

(q = 0, k = 0 or p = 0) (q = 0 , k + q = 0 , p = 0 ) (q = 0 , p q = 0 , k = 0 ) (q = 0 , k = p = 0 ) (q = 0, k = q, p = q), (4.212)

+ + + +
k

Vk a 0 ak a0 ak Vk a k a0 ak a0 Vk a k ak a0 a0 Vk a 0 a0 ak ak

where the prime in the summations means excluding the term k = 0. By using (4.207) and also the approximation (4.211), we get Hint
2 N0 V0 N0 + 2V V k

1 [(V0 + Vk )a k ak + Vk (ak ak + ak ak )]. (4.213) 2


111

TFY-44.130 Kvanttimekaniikka II Furthermore, the approximation (4.211) implies = N


k

S. Virtanen 2007

a k ak N0 +

a k ak ,

(4.214)

and at the same level of approximation we may write N N0 + This implies V0 2 V0 2 N V0 N0 N 2V 2V V V0 2 N V0 N 2V V


k k k

a k ak .

(4.215)

a k ak + a k ak ,

V0 2V

kk

a k ak ak ak

(4.216)

where in the last step we have neglected the double summation term because for the physical states under consideration |
k

a k ak | N N0 N.

(4.217)

Inserting this result into (4.213) gives Hint N 2 V0 (N N0 )V0 2V V


k

a k ak +

N0 V

1 [Vk a k ak + Vk (ak ak +ak ak )]. 2 (4.218)


a k ak ak ak

Now the term (N N0)V0 V


k

a k ak

V0 V

(4.219)

kk

may again be neglected, and we get nally24 H = H0 + Hint k2 N 2 V0 N0 ak ak + + 2m 2V V


k
24

Vk a k ak +

N0 2V

Vk (a k ak + ak ak ).

(4.220)
Note that this differs from the result given by Schwabl if N0 < N our result is more accurate in the general case.

112

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

This approximate form may at rst sight look even more complicated than the exact Hamiltonian, but there is a crucial difference: the approximate Hamiltonian is quadratic in the creation/annihilation operators, while the original one contains the quartic interaction term. Now, quadratic forms can be diagonalized by a rather simple coordinate transformation, but no such simple means apply to the original, nonlinear Hamiltonian. The coordinate transformation here is in fact a Bogoliubov transformation to a new set of creation/annihilation operators:
a k = uk k + vk k , ak = uk k + vk k

(4.221)

where uk , vk are complex numbers (we demand them in fact to be real) and the new operators k , k are demanded to be also creation/annihilation operators, satisfying
[k , k ] = [k , k ] = 0, [k , k ] = kk .

(4.222)

Such a transformation is called canonical. One can easily show that this requirement implies (see Schwabl)
2 u2 k vk = 1

(4.223)

for all k. Inserting the transformation (4.221) into the approximate Hamiltonian (4.220), one nds (see exercises) H N 2V0 + 2V
k k

k2 + n0 Vk 2m

2 u2 k k k + vk k k + uk vk (k k + k k )

N0 + 2V where

2 Vk (u2 k + vk )(k k + k k ) + 2uk vk (k k + k k )

(4.224) N0 (4.225) V is the condensate density (not to be confused with the corresponding occupation number!). This contains terms k k , k k that are diagonal in the occupation number basis corresponding to the new occupation number operators n k = k k , (4.226) n0 =
113

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

and nondiagonal terms containing k k , k k . Now the idea is to choose the transformation such that the Hamiltonian becomes diagonal in the new occupation number basisthis is the case if we choose the coefcients uk , vk such that

k2 n0 2 + n0Vk uk vk + Vk (u2 (4.227) k + v k ) = 0. 2m 2 With such coefcients, the Hamiltonian gets the diagonal form (see exercises) N 2 V0 1 H 2V 2 where k = k2 2m
2 k

k2 + n0Vk k + 2m n0 k 2Vk + m

k 1/2

k k k ,

(4.228)

(4.229)

In this form, the spectrum of the system is clearly visible: The ground state of the system is the new vacuum state |0 that satises k |0 = 0 (4.230)
k nk (k ) |0 ,

for all k. Excited states are of the form 1 |n1 , n2, . . . = with excitation energies E E0 =

(4.231)

nk !

k nk .

(4.232)

It is to be noted that the new particles created by k are not the original bosons, but instead a combination of a particle and a holesee the Bogoliubov transformation for this interpretation. The new particles are termed quasiparticles.

The energy-momentum dispersion relation is given by k . For high momenta and energy we nd k2 k = 2m k2 2mn0Vk 4mn0Vk 1 + 1+ k2 2m k2
114

k2 + n0Vk , (4.233) = 2m

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

which is just the dispersion relation for a single particle moving in the mean-eld potential created by the average condensate density. For low momenta, we nd instead k ck, (4.234) where n0 V0 . (4.235) m This is the characteristic dispersion relation for sound waves, with sound velocity c! For low momenta, the excitations of the system are density waves, which consist of collective motion of all the atoms, instead of the single-particle-like high-energy excitations. c=
Superuidity

The weakly interacting Bose-Einstein condensate has superuid properties, i.e. it can ow past objects without exerting a drag force! Let us demonstrate this by considering a condensate owing in a channel. What is dissipation, that for normal uids degrades the ow by converting kinetic energy to heat via frictional viscocity? Essentially, it is an excitation process of higher-energy states of the uid, caused by objects that move with respect to it. If it would be energetically unfavourable to create those excitations in the uid, the viscocity of the uid would vanish and the ow would not degrade, i.e. the uid would be superuid! Let us thus calculate the excitation energy of elementary excitations, i.e. quasiparticles, for the weakly interacting condensate owing in a channel with a constant velocity v. The thermodynamic equilibrium state for all systems is to be determined in the coordinate frame in which the external potentials are time-independent. The external potentials are now the walls of the channel, thus we have to calculate the excitation energy of quasiparticles in the laboratory frame in which the walls are static. However, thus far we have determined the quasiparticle energies only in the frame in which the condensate is at rest, i.e. in the frame moving with velocity v with respect to the laboratory frame. The momentum Plab and energy Elab in the laboratory
115

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

frame are related to momentum P and energy E in the frame moving with velocity v by the Galilei transformation: Plab = P + M v, M v2 , Elab = E + P v + 2 (4.236) (4.237)

where M is the total mass of the system. Now, if an elementary excitation with wave vector k would be created, the total momentum of the system in the condensate rest frame would be changed by the amount P = k p, and the energy by the amount E = k (p). Equation (4.237) now implies that the energy of the system in the laboratory frame would be correspondingly changed by the amount Elab = E + P v = (p) + p v. (4.238)

If Elab < 0, it is favourable for the excitation to be excited, leading to dissipation. We see that if v < vc min (p) , p (4.239)

no excitations are energetically favourable to be created. For example, for an ideal gas (p) = p2 /2m, hence vc = 0, and excitations are always energetically favourablean ideal gas is not a superuid, even while it is noninteracting! However, for a weakly interacting Bose condensate the low energy dispersion relation is (p) = cp, where c is the sound velocity. This dispersion gives vc = c, hence the condensate ow should be dissipationless for velocities that do not exceed the sound velocity c!25

25 If the dispersion curve has a local minimum in the transition region between the phonon region and the single-particle region, the critical superuid velocity is somewhat smaller, but still nonzerosee Schwabl Fig. 3.7.

116

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

Chapter 5

Brief Introduction to Relativistic Wave Equations


Literature: Schwabl: Advanced Quantum Mechanics, Part II 5.1 Introduction Up to now we have neglected special relativistic effects in our quantum mechanical calculations. However, small particles can easily travel fast, and in situations where quantum mechanical effects are important (comparing to classical mechanics), also relativistic effects are often observable. One example is our familiar hydrogen atom: The speed of the electron is approximately v Zc, where Z is the atomic number, 1/137 the ne structure constant and c the speed of light. Thus for hydrogen the electron velocity is of order 1% of the speed of light, and relativistic effects are already observable in the ne structure of its spectrumfor heavier atoms, the inner electrons are strongly relativistic. Thus, even to be able to calculate accurately the spectrum of hydrogen, one should develop a quantum mechanical formalism that takes into account special relativistic effects.1
It turns out that special relativity and quantum mechanics can be combined to a unied theory, but combining general relativity and quantum mechanics to a consistent theory is extremely difcult. The latter is still to large degree an open problem in physics, although it has been investigated intensively (naturally already by Einstein). However, if one contents oneself with not quantizing the gravitational eld, one can extend the following analysis to include general relativity.
1

117

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

5.2 Special relativity and postulates of quantum mechanics How did we end up having non-relativistic quantum mechanics in the rst place? In our postulates of quantum mechanics (see the rst Chapter) there is nothing that precludes special relativity. It was only in the practical realization of the last postulate Between measurement the state vector of the system evolves according to the Schrdinger equation i d | (t) = H (t)| (t) , dt (5.1)

where H (t) is the operator corresponding to the total energy of the system. that we have neglected relativistic effects: For example for a free particle we have chosen 2 p H= , (5.2) 2m according to the non-relativistic Hamiltonian function for such a system. Thus in order to obtain a relativistic theory, one just has to choose the Hamiltonian operator such that it takes into account relativistic effects. Otherwise the quantum theory remains the same! However, it turns out that this systematic approach to marry quantum mechanics and special relativity necessarily leads to creation and destruction of particles, properly described only by relativistic many-particle quantum eld theories. We are not going to follow this long and complicated path towards the greatest intellectual achievements of mankind, but rather follow the rst historical steps in the 1920s and 1930s when physicists tried to develop one-particle wave equations to describe a relativistic electron. These one-particle equations are always only approximate, but at low energy scales they are very accurate approximations to the proper quantum eld theories. Derivation of relativistic one-particle wave-equations is necessarily a bit intuitive and contains handwaving arguments, because as said these theories are only approximate and contain paradoxes that can be only solved within relativistic quantum eld theories. An important criterion in deriving such
118

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

equations follows from the principle of relativity, i.e. that the physical laws must be the same in all inertial coordinate systems. It follows that also the form of the wave-equations must be the same in all inertial coordinate systems, i.e. they have to be relativistically invariant. In order to see what kind of restrictions this imposes to the theory, we rst analyze Lorentz transformations between different inertial frames.

5.3 Four-vectors and Lorentz transformations In order to nd out what kind of one-particle wave equations are relativistically invariant, we rst analyze the properties of Lorentz transformations. Let x (x0, x1, x2, x3) = (ct, x, y, z ) (5.3)

denote the coordinates of the same event in an inertial frame O . We restrict to the case that the frames O and O coincide at time t = t = 0. These sets of coordinates are contravariant space-time four-vectors. Corresponding covariant space-time four-vectors are denoted and dened to be x (x0, x1, x2, x3) = (ct, x, y, z ) and x (x0, x1, x2, x3) = (ct , x, y , z ).

denote the time and spatial coordinates of an event in an inertial frame O, and let x (x0, x1, x2, x3) = (ct , x, y , z ) (5.4)

(5.5) (5.6)

One uses the convention that components of four-vectors are assigned Greek indices, and the components of spatial three-vectors are denoted by Latin indices or by the usual x, y , and z . In addition, one uses the Einstein summation convention: indices that appear twice, one contravariant (superscript index) and one covariant (subscript index), are summed over. For example
3

x x

=0

x x = c2 t2 x2 y 2 z 2 .
119

(5.7)

TFY-44.130 Kvanttimekaniikka II Creatures like a can be regarded usual matrices, i.e. 0 a 0 a0 1 a0 2 a0 3 a1 0 a1 1 a1 2 a1 3 a = a2 0 a2 1 a2 2 a2 3 , a3 0 a3 1 a3 2 a3 3

S. Virtanen 2007

(5.8)

and thus we can denote multiplication of a contravariant vector w by the matrix a compactly as v = a w (remember the summation convention!). Now Lorentz transformations tell how all four-vectors (for example space-time coordinates of an event) are related between two different inertial frames.2 From the requirement that uniform motion must be uniform in all inertial frames, one can deduce that Lorentz transformations must be linear transformations, i.e. they can always be represented as multiplication of four-vectors by a matrix. Let be the Lorentz transformation between the frames O and O . This means that x = x (5.9)

is the relation between the coordinates of all events in these two frames. For example, if the frame O moves with respect to O with velocity v = c (0 < 1) in the x-direction, one has 1 2 2 0 0 ct ct 1 1 x 1 2 0 0 = x 2 (5.10) 1 1 y y 0 0 1 0 z z 0 0 0 1 The point in dening separately contravariant and covariant four-vectors is that one can show that combinations like x x in which one contracts (sums over) co- and contravariant indices are Lorentz invariant, i.e. they are the same in all inertial frames (relativistically invariant).
2 Vice versa, four-vectors are dened to be such four component vectors that transform between inertial frames according to Lorentz transformations.

120

TFY-44.130 Kvanttimekaniikka II 5.4 Klein-Gordon equation

S. Virtanen 2007

The energy operator (5.2) corresponds to the classical energy-momentum relation p2 , (5.11) E= 2m from which one can formally derive the Schrdinger equation (in position representation) by using the operator substitutions Ei , t p . i (5.12)

Analogously, one could try to nd out a relativistic Schrdinger equation by quantizing the relativistic energy-momentum relation E= c2 p2 + m2 c4 (5.13)

by the same operator substitution. By this way one arrives at i (x, t) = t 2c2 2 + m2 c4 (x, t). 2 c2 4 + (x, t). 2m 8m3 (5.14)

Expanding the square root we can write this in the form (x, t) = i t mc2 (5.15)

Now suppose we transform this equation to a different inertial frame using the Lorentz transformation (5.9), i.e. we express the equation in terms of t and . Since a general Lorentz transformation mixes time and spatial variables, it is obvious that the transformed equation is generally not of the same form (with just unprimed derivatives replaced with primed ones), and thus the above equation is not Lorenzt invariant (relativistically invariant). But if, instead of (5.13) we start from the squared relation E 2 = c2 p2 + m2 c4 , after operator substitution we end up having the wave equation
2

(5.16)

2 (x, t) = ( 2c2 2 + m2 c4 ) (x, t). 2 t


121

(5.17)

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

Is this equation Lorentz invariant? It can be written in the form + where mc


2

(x, t) = 0,

(5.18)

, = (5.19) x x are the four-dimensional gradient vectors, and as implied by the notation, they are four-vectors (exercise!). Thus is Lorentz invariant, and because the rest mass m of the particle and the natural constants c and are the same in all inertial frames, the time evolution equation (5.18) is really relativistically invariant, i.e. a good candidate for describing some kind of relativistic particle. This equation is the Klein-Gordon equation. It has free particle plane wave solutions of the form = p (x, t) = ei(Etpx)/ with energy E= c2 p2 + c4 m2 , (5.21) (5.20)

and we see that the spectrum is not bounded from below (arbitrarily large negative energies). This is somewhat of a problem, and it can be remedied only by upgrading the theory into a relativistic quantum eld theory. It turns out that the negative-energy particles should be interpreted as antiparticles in order to solve all the problems! Hence this way we necessarily end up having a many-particle theory, although we intented to develop only a one-particle equation. The wave function (x, t) in the KG equation is a one-component (scalar) quantity, and thus the equation could describe only spinless particles. There are not many spinless but massive elementary particles in the nature, but neutral mesons are such and indeed it turns out that the KG equation can be used to describe them in not-too-large energy scales (in which many-particle effects are not important)!

122

TFY-44.130 Kvanttimekaniikka II 5.5 Dirac equation

S. Virtanen 2007

At the time of its invention, the Klein-Gordon equation was not considered very important because the main goal was to nd a relativistic wave equation particles. for electrons, that are spin- 1 2 One of the problems with the KG equation is that the probability density = (i /2mc2) ( /t /t) related to it (see Schwabl 5.2.2) is not positive denite. This originates from the fact that the KG equation is second order in time (and space) derivatives. In order to avoid this problem, Paul Dirac tried to develop a relativistic equation for electrons that is of rst order in time derivatives. Lorentz invariance requires it to be of rst order also in the spatial derivatives. Dirac wanted the equation to have the original form (5.1) of the Schrdinger equation, i.e. i d (x, t) = H (x, t), dt (5.22)

and searched for a free electron Hamiltonian that is rst order in spatial derivatives: H= c 1 1 + 2 2 + 3 3 i x x x + mc2 = c k k + mc2 (5.23) i

(recall the Eintein summation convention and that Latin indices run only over spatial components). The coefcients i and have to be constants in time and space in order to get a translationally invariant theory (free particle!). In order to nd suitable s and , we require that energy eigenstate solutions E (x) = H (x) (5.24) satisfy the relativistic dispersion relation E 2 = c2 p2 + c4 m2 . (5.25)

123

TFY-44.130 Kvanttimekaniikka II Using (5.23) and (5.24), we get

S. Virtanen 2007

E 2 (x) = H 2 (x) 2 2 mc3 i c i j j i ( + ) i j + ( + i )i + 2m2 c4 (x) = 2 i c2 i j ( + j i ) pi p j + mc3 (i + i ) pi + 2 m2 c4 (x). = 2 (5.26) Note that we have not assumed that the constant coefcients commute with each otherfor example they can be matrices. In order for this equation to be consistent with (5.25), the constant coefcients should obviously satisfy the conditions i j + j i = 2ij I, (5.27) i + i = 0, and (i)2 = 2 = I, (5.29) where I denotes 1 or possibly a unit matrix. These relations mean that s and s must form a representation of a Clifford algebra. One easily sees that they cannot be usual complex numbers, but on the other hand one knows that Clifford algebras always have matrix representations. It turns out that the smallest possible matrices that satisfy these relations are the 4 4 matrices =
i

(5.28)

0 i , i 0

where i are the 2 2 Pauli matrices 1 = and I= For example 0 1 , 1 0 2 =

I 0 , 0 I

(5.30)

0 i , i 0 1 0 . 0 1

3 =

1 0 , 0 1

(5.31)

(5.32)

0 0 0 0 2 = 0 i i 0
124

0 i i 0 . 0 0 0 0

(5.33)

TFY-44.130 Kvanttimekaniikka II So Dirac arrived at the wave equation i d (x, t) = dt

S. Virtanen 2007

c k k + mc2 (x, t) i

(5.34)

called the Dirac spinor. This wave equation is the famous Dirac equation.

with i s and the 4 4 matrices given above. Naturally (x, t) has to be a 4-component column vector 1(x, t) 2(x, t) (x, t) = (5.35) 3(x, t) 4(x, t)

But how to interpret the four components of the Dirac spinor? How could it describe electrons, because electrons are described with a two-component spinor in nonrelativistic quantum mechanics? By investigating the nonrelativistic limit of the Dirac equation (exercises!), one nds that essentially two of the four components of the Dirac spinor can be interpreted to be just the electron spinor components. The other two components describe a particle having an electron mass, but positive charge!3 This kind of analysis made Dirac convinced that there should be an antiparticle for electron, the positron, essentially because of the requirement of relativistic invariance. Indeed his prediction turned out to be correct as the positron was later found experimentally!

5.6 Dirac theory of hydrogen atom Above, we have presented only the free particle Dirac equation. In analogy with the nonrelativistic treatment, one can take into account the coupling of electron (and positron) to external electromagnetic elds by doing the so-called minimal substitution e p A( r, t) (5.36) p c
Roughly speaking, the upper two components represent electrons and the lower two positrons, but to be precise, the lower (upper) components are small but nonvanishing for electrons (positrons).
3

125

TFY-44.130 Kvanttimekaniikka II

S. Virtanen 2007

in the free particle Dirac equation and adding the scalar electrical potential term e( r, t) to it. This results in the position representation in the equation e d (x, t) = ck k Ak (x, t) + mc2 + e(x, t) (x, t). i dt i c (5.37) Now one can test the Dirac electron theory by applying it to the hydrogen atom. The static Coulomb potential due to proton is described by setting Ze2 A(x, t) 0, e(x, t) = , (5.38) r and one searches for energy eigenstates corresponding to Dirac equation (5.37). It turns out to be a rather formidable calculation4 , but the energy eigenstates can be solved exactly analytically! The eigenenergies as function of the principal quantum number n and the total angular momentum (orbital + spin) quantum number j are 1 2 2 Z En,j = mc2 1 + 1 2 2 n (j + 1 ( j + ) + ) ( Z ) (5.39) 2 2 = mc
2

Z 22 (Z)4 1 2 n2 2 n3

1 j+

1 2

3 4n

+ O((Z)6) ,

where 1/137 is the ne structure constant. The rst term in the last line is the electron rest mass energy, and the second the usual Bohr energy obtained in nonrelativistic treatment neglecting the electron spin. The third term can be also derived from nonrelativistic hydrogen theory, by taking the spin-orbit, lowest order relativistic corrections and the so-called Darwin term into account perturbatively.5 The main point is that the spectrum given by the Dirac theory agrees very well with experiments, much better than the spectrum given by the nonrelativistic calculation! It is to be noted that the degeneracy with respect to the angular momentum quantum number l is broken when the relativistic effects are taken into account.6
An exercise problem a few years ago in this course. . . See Schwabl, Section 8.2 for details. In these calculations the gyromagnetic ratio of the electron has to be known independently, but the Dirac theory gives it automatically the very precise approximate value gs = 2. 6 Dirac equation can not explain the hyperne splitting of hydrogen spectral lines. This effect arises from the interaction of proton and electron magnetic moments, and the magnetic moment of proton is not taken into account in the Dirac theory, which is a one-particle theory. In addition, the nite mass of the nucleus should be taken into account by using the reduced mass instead of the electron mass m.
5 4

126

You might also like