Sgribhinni Institiuid Ard-Leighinn Bhaile Atha Cliath. Sraith A Uimh. 1 Communications of the Dublin Institute for Advanced Studies. Series A No. 1 QUANTUM ELECTRODYNAMICS by PAUL A. M. DIRAC DUBLIN THE DUBLIN INSTITUTE FOR ADVANCED STUDIES 64-3 MERRION SQUARE 1943 (Reprinted 160)QUANTUM ELECTRODYNAMICS. By P. A. M. DIRAC. ‘Tue present leotares, like those of Eddington, are concerned with unifying relativity and quantum theory, but they approach the question from a different point of view, Eddingon's method is first to get the physical ideas clear and then gradually to build up a mathematical scheme, Tho present method is just the opposite—first to set up a mathematical scheme aud then to try to get its physical interpretation, ‘This method ought to be the easier one, becanse it takes the easier task first. It should be easier to get the mathematical scheme, because for this purpose the number of things one has to choose between is small. ‘The scheme, to be acceptable, must be neat and beautiful, and the number of such schemes which pure mathematies can provide is very limited. When the mathematical scheme is decided on, the subsequent task of finding-its physical intorpretation is rendered easier by the existence of many points of contact between the mathematies and the physics. ‘We shall be concerned with the problem of the interaction of a number of charged particles with the electromagnetic field, and our object will be to get a method for calculating experimental results, such as the probability of certain particles, which are incident in a certain way, being scattered in a certain way, with perhaps the emission and absorption of certain photons, ‘The method has to be in conformity with the general principles of physics, such as the conservation laws and Lorentz invariance, and with the laws of the quanbuun theory so far as they are understood. The line of attack will be to start from the mathematical scheme of the quantum theory as it is known for the non-relativistic domain and-to try to extend it to make it relativistic. We shall find that there is one natural way of doing this. But when we come to consider the physical interpre- tation of the mathematical scheme, we shall find that the natural extension of the non-relativistic theory leads to the appearance of negative energies and probabilities. Tt hecomes necessary to make some new assumptions, which result in the interpretation being somewhat less ditect than one could wish for, but which is all the same in agresment with all essential requirements. G42 Communications of the Dublin tnstitule—Ser, (A). §L. The Bracket Notation ‘The most convenient notation to uso in quantum mechanies for goneral thoovetieal purposes in which one wants to be able to pass freely om ono representation to another is the author’s bracket notation.’ ‘The vectors corresponding to the quantum states are denoted by symbols |), consisting of 4 vertical line and an incompleted.bracket. If one wants to give a vector a label, a say, ono inserts it inthe middle, thus |a). Exch of those vectors has a conjugate imaginary vector, corresponding to the same quantum state, which is written ¢{ with the same label. ‘Thus, in our example, {a} has the conjugate imaginary (a. A veetor (2) and a vector ja) havea scalar produet whieh is a numbor, and which is written as a symbolic product { {a ). (For brevity one doos not put two vertical lines). ‘The following axioms hold :— (alb) = (dja), (v1) tho bar denoting the conjugate complex number, and (aja) > 0. (2) ‘One can operate on a |) with a linear operator, the result being another {). If the linear operator & operates on the vector | a), the result is written asa symbolic product £|a). Similarly, one ean operate on a <| with a linear operator, the result being another (|, and if the linear operator £ operates on the vector ¢ bj, the result is written as a symbolic product ¢ b |, A veotor |a) anda vector (}| have a symbolic product |a ) ¢|, which is a linear operator, If this linear operator operates on jc), the result is |a) (dle), @ numerical multiple of |a), and if it operates on o|, the result is ¢o|a) (d|, a namorical multiple of ¢ 5}. In all our symbolic multiplication processes the associative axiom of multiplication holds, but not, in general, the commutative one. Of course when a numerical factor appears in a product, such as the (|e) and the = fz). and (1°6) follows. The conditions (1-5) and (1-6) are sufficient as well as necessary, since (13) and (1-4) can be deduced from them, To deduce (1-3) from them, take any vector (a | and multiply it into (1°6) on the left, ‘The result is Brdelr) Cr] = <2], which gives (| expanded in the form (1°38), with ¢ = (2|r). To deduce (1-4) from them, multiply (16) by (| on the left. The result is Beal rrr] = 6| = Bd dr] or 3+ (Ks |r) = dr) Cr] = 0. Applying the independence condition (1°5), (1-4) follows. ‘The general formula (1°6) leads immediately to the law of matrix multiplication of quantum mechanics in all its various cases, e.g. (sl En] ty = SCs Elrdcrinley and Kr] Ele = Bakr PE lsyis| zy. Instead of the discrete set of basic vectors (r{, one can have a con- tinuous range of them, labelled by a parameter, g' say, that can take on all values in a certain continuous range. ‘The sums in (1°3), (1-0), and (1°6) are then to be replaced by integrals over q’, and (1-4) is to be replaced by (C11) = 8-9") a7) [42]4 Communications of the Dublin fustitule—Ser, (A). There now exists an operator, galled the opera ating on any |g’) or (Kle=cia'l. (rs) ‘There exisis another important operator, which ay call the operator ae which is such that when it operates on any | ), the g/ representative of that | > gots differentiated. In symbols ap oo flag) = ag S71) ea : Hence 1 Vig = * x wave function is norma presontatives (qi'qx...4) 6 ridinger’s theory. If such « has the physical interpretation has 1€ai'qx'...|)|? is the probability of the observables qi, 92 - having values close to gy’, qx... per unit range of variation of these values. are the wave functions of $2. Fock's Treatment of the Harmonie Oscillator. A. comer-stono of electrodynamie theory is the simple harmonic oscii- lator, Ic may .be treated in quantum mechanies by ‘Heisenberg’s matrix method or by Schrédinger's wave functions, but a simpler method has been discovered by Fock,? which is much more suitable for our presont purposes. Neglecting unessential numerical coefficients, we have to deal with a coordinate q and momentum p satisfying ww - p= 4. (2:1) Fock’s method consists in introducing the variable z - 5 E-7,@+@) (22) and setting up a representation referring to & Such a representation is of a more general kind than the representation referring to g of the pro- coding section, on account of & not being a real dynamical variable. We may take a L p - 7 #7 we - 3) sinee this leads to the commutation relation a@ a sa HR ; i GE cae WE ee IP) EF DA re es like (1-11). ‘The energy operator, excluding the zero point energy, is a 3p? + 9%) - 2 =4(p + Mp - tg) = b 5 sentnnetions of thi |) vectors) aro * w tor (which ar bh n=O, Lopreseutatives of the giving the enor rock, Zeits f. Phys. 49, p. $89 (1028). There is @ oorrection in Phys, Zeits. de Soviet. U. footnote (1934).6 Communications of the Dublin Institule—Ser. (A), levels, It follows that the representative of any |a ) ean be expressed as a power series (Ela) mag tarG tart sass... (25) the various terms in the expansion corresponding to the various energy levels. Let us determine the form of the tepresentatives of the eigen ¢ | vectors. An eigen ( | satisfies pee legend or CERT Eye nce): Assuming that relations like (18), (1-10) hold for the E-representation, we get ~ lL ED) eal) ae’ = showing that (|) is of the form &-"-". It follows that the repre- sentative of any (| can be expressed as a power series COLE) = bet +B, Eh + bs nH, (26) the various terms in the expansion again corresponding to the various energy levels. One may look upon the representative ¢€|a) for any value of E as the scalar product of |) with a basic (&|, and similarly (| £) may be looked upon as the scalar product of (&| with a basie | £). However, <£| and |£) are not conjugate imaginary. vectors. ‘The rule that the same label is used for conjugate imaginary vectors does not apply when the label is a value of a complex, dynamical variable, and the conjugate imaginary of | § ) must be written ¢ & |. a Suppose that the | a) represented by the right-hand side of (2°5} and the (| represented by the right-hand side of (2:6) are conjugate imaginary vectors, ‘Then the coefficients 5, and a,, which refer to the same energy level, must be connected by a relation of the type ba = An ane (27) To determine X, we note that, if ¢b] and | a) aré conjugate imagi- nary vectors, then so are (3| (pig) and (p+ ig)|a) or ls and £| a). These vectors are represented by ST (n+1)b,€-""? and Sa, &"*, which must therefore be connected in the same way in which the right-hand sides of (26) and (25) are connected by (2:7), 80 that (n 4 1) dn = AnsiGaeDirac—Quantum Electrodynamics. 7 Trion 2 OGPetleD ack and we may tale e Aw aal. ‘The condition (2°7) now becomes 5 We eae 28) As a check, let us examine the requirement that /6| (p+ %) and (p - ig) |) shall also be conjugate imaginary veetors, or ¢¢|& and ae |a@). The representatives of these vectors are %5d,E-" and 32 nay E*-!. We see that these series are connected in the same way in which the right-hand sides of (2:6) and (2%) are connected by (2'8), with the exception that there is a constant term }, in the former which is not connected with any term in the latter, and which has no analogue in the right-hand side of (26). To get over this difficulty, we must assume that an arbitrary constant may be added to: the representative ¢ |) of any (| and it still represents the same |. Generalizing the argument, one finds the need for a more eee) assumption, that an arbitrary ascend- ing power series co, + ¢,£+£'+... may be added to the repre- sentative ¢|€) ofany ¢ | and it ceil represents the same < | - ‘The scalar product ¢b|a@) of the vectors (5| and |a) whose representatives are given by (2°6) and (25) must be of the form (b)a) = pede + wibies Hardee tone, where the y's are real positive numbers, By suitably redefining the coofficients a, and b,, one ean arrange to have the p’s unity, 6o that . Kd ad) = bya, 4 Bay + bea, + (29) This may be expressed as a contour integral round the origin, if DEFOE 4 BE GEG tabs Ee) ae (210) - Fa P MID AECE LO. ‘This result may be put into operator form 1 } -_— (2°1. = Pl eraecel, qe ‘and is then ‘the analogue of (1, 6) for the present kind of representation. We can now see how it comes about that an arbitrary ascending power series E may be added to (| £), since sucha series. would not affect the of the integral (2:10).8 Communications of the Dublin Tnstitute—Ser, (A), Taking the vector <6 | te be the conjugate imaginary of Ja), so that the labels 6 and @ are the same and the coefficients 6, and a, aro connected by (2°8), we get from (2:9) Cala) =X nllag|* = x (al) |on]?. (212) Theso sums must be convergent. for vectors that correspond to quantum states: It follows: that thie series “Xa, €" representing |a) must be convergent for all § but the series 3, £-*-! represontinlg (> | need not converge for any £ For this reason one usually works only with ¢ £ | ) representatives, and (| &') representatives are nob suitable for general theoretical inyestigations.- ‘The form of (2:12) shows that if |) is normalized, it corresponds to 4 state for which the probability of the energy having the value n is Pa = atlas |". (213) : ‘ Let us work out the representatives of some simple operators, The unit operator is represented by (PITIED = PLE oy 1 yi (214) since this makes 1a Pelle dece tay = Cee). ‘The operator § is represented by . ” CETELE) = pty, pip. (215) " These two expressions count as oqual in the representative of an operator, since their difference is unity, which does not contribute anything when ‘wultiplied into ¢£”|) and integrated round the origin with respect to €”, and contributes only a constant when multiplied into ¢ | &)> and integrated round the origin with respect to .£’. The operator fo ie represented by : (2-16) since this makes eo doy Tei P CEL gel En aR CENIY = Ce) 1 +) fen @ ” ri P SLED aE CR Ay ery <|". (34) ‘There arises a correspondence between the assembly of bosons and a set of harmonic oscillators, each of the boson states specified by a value of g corresponding to one oscillator. ‘The power series in the £’s representing a state of the assembly of bosons may be identified with that describing a state of the oscillators, the probability (3-4) of there being various numbers of bosons in the various states for a boson being the same as the probability (2°17) of the oscillators being in various quantum states. The assembly of bosons and the set of oscillators are mathematically equivalent systems. ‘The transformation from the set of components (3'2)' to the function (|) is a linear transformation and is therefore expressible in terms of a transformation function ¢ —| 9 9”...g*),. thus CEL) = Bae CELT". gg." |>- (85) Comparing (3:5) with (3°83), one finds 3 : CELA gg") = why Eqn... Ege. (3°6) The operators E,, /dE,, when operating to the right, correspond to the emission of a boson into and the absorption of a boson from the state g respectively. From the theory of § 2, one sees that these operators are adjoint (ie. (a|&, and d/d&,|«) are conjugate imaginary vectors, and so are (a|0/a&, and &,|@))-Dmac—Quantum Blectrodynamics. uw §4. Relativistic Theory of a Particle. Let us take for simplicity a particle of zero rest-mass, as we shall afterwards apply the theory to photons, We want our work to bo of relativistic form throughout, so we must first generalize the conditions of * $1 for basic vectors to get them into relativistic form. ‘A state of a particle without spin is represented by @ wave function (2,2, 2,%|), ot |? (he sy by” sy... Keo! Sux)", where nq of kk”...k* are equal to k*, , of them are equal to k*, and son, This expression equals coofficient of EX, EM... in (Ey |) |? he 840)" (ky? 9)". 2. (5-10) For &, + 0, Ej may be considered to describe a harmonic oseiliator. Tt must be taken as (k, s,)4 times the & of §2 or 8, to get the same, weight factor in (5-10) as in (3-4), and thus satisfies with its adjoint &f the commutation relation - . EEE, - Exte =k, sy. nal mol. 11)16 Communications of the Dublin Institute—Ser. (A), For &, <0, one no longer has the analogy with the harmonic oseillator but (5°11) atill holds, as may be verified thus :— Consider the degree of freedom corresponding to Ey by itself, and take two states, a and 2, represented by (Eula) = 3 a Ef (Ex) = 35, ER. Corresponding to the physical interpretation (6+10), we must have the multiplication law Ca] b> = Ent ay by (ky ay)". Using this law Cal FELD) = Sint dn das (hy oy". Taking the conjugate complex equation, DEE |) = Stil au bus (he sy)" = S (m= 1)! By (mary hy 84) (hey sy)", showing that Ef|a) is represented by Shaeky ox Et'= kyonge CExle)- Thus Em kenge and (5°11) follows, whether i is positive or negative. (511) leads to Eke Eue - ExsE fa = hey ‘Passing back from discrete to continuous kk Sas Ede E ue Exo Efe = hot (bit — k,») 3 (kee — bY) (Rt - ke) when 2,4, &,' have the same sign + (5°12) = 0 whon &,*, k,' have opposite sigus.. Returning to the Ey’s, according to (5-9) BE Exe — Exe E8= Bx)? Baga, [fof 10029 (ER, Eien ~ Eten EE.) ae Ok” = (2x)-? By, fea oe Vey =- 2(2n)°* [sin (ke, x - x”) oF, where in the last expression I’, is restricted to be positive: Put x'-x”=-xDmac—Quantum Electrodynamics, n and let |2|= (;*+2,"+ 2)4, and @ be the angle between the three- dimensional vectors 1, #., #3 and ki, ky, ks, Then ELE ge — Exe EL =~ tet {fin [6 (a, ~ | 2 | o0sf0)] By dd 008 0 cos #1 = = int |2[-4 confia (as ~ | [008 1.) = - ie] (J teosttstes [2101 ~ comfy (e+ |= D]letke lal (B@.- 121) - 8+ 121)) = -40(@) 6 -ta@ +2’). (613) ‘The operators £,, E,, when operating to the right, correspond to the emission of a boson into and the absorption of a boson from the momentum- energy state kh, and likewise the operators £,, £7 correspond to the emission of a boson into and the absorption of a boson from the position state x, The commutation relations (5-12), (5-13) which these operators satisfy should be compared with (4-14), (46) of the one-particle theory, ‘The above relativistio theory for spinless bosons of zero rest-mass may easily be extended to apply to photons. The wave function for a single boson must be made into a 4-vector with components corresponding to the four components of the electromagnetic potential. The 4-vector wave function may be odnsidered as a function of the four coordinates x and of another variable taking on the four values 0, 1, 2,3, and may be written (xu). The w here is not a suffix attached to x, but is an independent variable, which is written in the lower position to express that the whole function (x, |) is subject to the contravariant law under Lorentz trans- formations, and which may be raised by the usual rule (xe|) = gur(xrl). (614) ‘The multiplication rule (4°5) must be extended to : pres = ge (diy xp) (T - 2) 0) Fi = ~pu +0; A, (i) + [mit Se[ pri ~ 6s A-()]*{b= 0, (6:31) where r= 1, 2, or 3. In terms of the /;, the equations of motion are easily seen to be f a ; eye FA. Now put all the times s. equal to ¢. Thon, for £ independent of the Pa and sy, Ee avO)- (1 (632) where H=3:F/@, (6:33) or, aa will do equally well HE = 3; (0; As + (me + B-(Pa~ 0: An) NHL (634) § 7. Elimination of the Longitudinal Waves. ‘The field 4 (x) may be split up into longitudinal aud transverse waves, as follows : : Split up the theee-dimensional veetor field 4, (x), into a divergence-free pert M,(x) and a rotation-free part L,(x), ie. 4, = M+ L, (7) where (72) (73) and do this eo that OM,=-0, of, (Tay Then M, gives the transverse waves, and Z, and A, tho longitudinal waves. This splitting up is not relativi:‘e, but it is of interest in spite of this, - because the longitudinal waves can be eliminated from the Hamiltonian formiilation by a contact transformation, thereby effecting a simplification of Ube equations.4 Communications of the Dublin Institwte—Ser, (A), ‘The contact transformation ean be given tost conveniently by building it up from infnitesmal contact (ransformations, each dynamical variable £ being transforined according to o He” 2 fe, avy O 0, causing possibly a—> 0. It is necessary that both positive and negative values of &, should occur in (8:8) and (89), otherwise one would have, as well as (811), integrals of the form [bo costa dky, fone sink,a dk, , (8°12) and no procedure of taking the mean of an oscillating part would enable one to assign a finite value to such integrals in the limit a=0. One would get instead terms of the form a-™-*, a-™-!, The earlier quantum electro- dynamics of Heisenberg and Pauli, which worked with only positive-energy photons, was faced with just this difficulty.Dirac—Quantum Llectrodynamics. 31 §9. Lhe Redundant Variables, Corresponding to (8°5), define Buz) = Gb) Exsayn - ER w)s (91) which provides us with a now field. We have, from (5-18), [Ba@), 4-(@)] = 0, and thus the B,(x) commute with all the dynamical variables occurring in the Hamiltonians. Such quantities are called redundant variables. ‘They are, of course constants of the motion. If the longitudinal part of -A,,(x) has been eliminated, then the transverse part of By (x) provides the only interesting redundant variables. Let us examine the significance of redundant variables in an elementary case. A system of one degree of freedom, with the canonieal variables 9, p, may be considered as a system of two degrees of freedom by putting = Feta), p= dy(ort vo. 2) ‘Then qi-gs and p,~pz are redundant variables, as they commute with g and p, Make a canonical transformation which separates the redun- dant and non-redundant variables, by putting a= FyQs~ 99 Dem Galt Pa. (93) so that (9-2) and (9'3) give the required contact transformation. Referred to the new variables, the wave function ¢9q,|) is afunction of g, g, satis- fying a wave equation which involves only g and 2. It must therefore satisfy this wave equation for each value of g,, which thus appears in it only as a parameter. Since g, has no physical meaning, to get a physical interpretation we must integrate |qg,|)|? forall g,, and so get the probability of ¢ having a specified value or lying in a specified range. One can now see that ¢9q,|)} does not represent a pure quantum state. Such a state is represented by a function of the single variable g satisfying the wave equation. (It is the state ensuing from a maximum observation, in this case an observation of g.) (qq,|) represents a mixture of several pure states, namely, all those pure states represented by functions of g, which one can get by giving numorical values to ¢, in (gq,|)- Such a mixture is the analogue of a Gibbs ensemble in, classical mechanics. One can conclude that:—Zhe existence of redundant variables means that a wave function represents a Gibbs ensemble and not a pure state, This conclusion must still hold if one is using a representation which does not separate the redundant and non-redundant variables.32 Communications of the Dublin. Institule—Ser, (A), One could tiy to make a canonicalsttansformation to separate the redun- dant and non-redundant variables injoitr prosent quantum electrodynamics. Let us work with the limiting ease 4 = 0, Put Elke = Siw for ky>0. (94) Then, from (5°9), (8°5), and (9'1), An(e) = (AA)? (ny fff (Gee ~ Sha) F084 © Zee + Eb) et} ab (95) By(z) = Gn} ew {ff (CE in + Liu) 4008) + (~ Zu ~ Eke) e° 10em)} ah (96) with y= +(kit+ kat ket)h. ‘Thus putting Vy Ey CE dak, = EE, - Li o7 VB ee = Su + EE, Vis th + Ee’ a and a* are the only field variables occurring in the Hamiltonians, and B and B* commute with « aud a* and are the redundant variables. Let us confine our attention to one particular’ #-value (with &, > 0), which we may suppose to be one of a discreet set, and one particular j-valuo, corresponding to transverse waves. Then, using (519) and dropping the suffixes, Beer e ree rep gre wee |? 8) where ¢ is some positive number. From (97) and (9:8) a*a-aat=o ) i 9°9) B*B-pp* = -e.$ (99) One can now set up a contact transformation from the, variables § E*, % 2%, to the variables «, «*, 8, 6% The transformation func- tion has been worked out by the author.’ It enables one to transform the wave function from a power series in EZ to a power series in «, 3. [From the commutation relations (9'8), (9:9), two forms of transformation are possible, one of which trarisforms an ascending power series in £, £ to an ascending power series in a, 8, the other transforms it to a descending power series in a, 8. We choose the former, which is the one worked out in the reference." ; ‘The Hamiltonians get transformed to functions of a, 2/da independent of B, 2/98. Thus the redundant variable 8 may be eliminated from the ‘Dirne, Pros. Roy. Soo. A. 180, p. 1 (1942). Appendix III.Dirac—Quantum Electrodynamics, 33 wave equations. However, if one carries out this elimination procedure for all k-values simultaneously, one gets the wave equations of Heisenberg and Pauli’s quantum electrodynamics, which do not have any solution, Decause of the divergent integrals discussed at the end of §8. Thus one cannot apply the transformation which eliminates all the redundant variables to solutions of the wave equations. All the same this transformation has a meaning because one can apply it to suitable functions of the &'s, Z's nob satisfying the © equations. $10. The Negative Bnergics and Probabilities. We may give the redundant variables any values we like, subject to no inconsistency arising from their Poisson Bracket relationships. It leads to a reasonable interpretation for the negative energies and probabilities to take the transverse part of By (x) equal to the transverse part of the: initial value of A, (x), after elimination of the longitudinal waves. With this assumption, for the initial state | init ), ([4.n) transverse — Bia (2) transverse] | init) = 0. (10-1) Making the Fourier resolution of this equation we get, from (9°5) and (9°6), , ER, | init) = 0 Ck, | emit) = . (102) where ¢ denotes a transverse component. These conditions are evidently all consistent, since the £*s and Z*s all commute. They show that a i a he : aE, (ES | init) = 0 Bey, (ES limit = 0, (108) so that the representative (££ | init) of the initial state is independent of the &'s and Z's, and is a function only of the electron variables 2,;. It thus corresponds to no photons being present, of either positive or negative energies ; The following natural interpretation for the probabilities (5°10) given by the wave function at some later time now appears. ‘That part of the wave function corresponding to no photons present may be supposed to give the probability of no change having taken place in the field of photons; that part corresponding to one positive-energy plioton present may be supposed to give. the probability of a photon having been emitted; that corresponding to one negative-energy photon present may be supposed to give the probability of a photon having been absorbed; and so on for the parts corresponding to two or more photons present. The rarious parts of the wave function whieh referred fe]a Communications of the Dublin Institut (AD. to the existence of positive-and negative-energy photons in the old interpretation now refer to the emissions and absorptions of photons. ‘This disposes of the negative-energy difficulty in a satisfactory way, conforming to the laws of conservation of energy and momentum. Lt is possible only because of the redundant variables, enabling one to arrange that tho initial wave function shall correspond in its entirety to no emissions or absorptions having taken place, { ‘The assumptions (10-1) or (10-2) ate not compatible with the equations of motion and can therefore hold only for a particular value for the particle times zz. ‘These assumptions are thus suitable only for problems in which the initial conditions apply to one time for each particle, for example if the initial positions of the particles in, space-time are given. In practical problems one is usually given, not the initial positions of the particles, but their initial momenta for some collision process) The momenta p,q do not commute with the Hamiltonians (6-1) or (8'1) and the requirement that they should have certain initial values means that one must solve the wave equations by a perturbation method, taking as the zero-order wave function one for which they have these values and arranging that the higher-order waye-funetions should refer only to outgoing particles or else latent particles (Je. particles in transient states for which the momenta have values not satisfying pi p= m,*). The zero-order wave function in this perturbation method is then the initial wave function for which the couditions (10-1) or (10-2) hold, and any emission or absorption of photons that takes place according to the new interpretation given above refers to changes from the state represonted by this zero-order wave funetion. ‘The physical interpretation is not yet complete, because at present it would give a negative probability for a process involving the absorption of a photon, or the absorption of any odd number of photons, ‘To track down these negative probabilities, let us study the initial state by transforming the initial wave function to the «, 8 variables introduced in §9. From (10:2) and (9°7) (a + B)| init) = 0, (B" - a) | init) = 0, (10-4) which give, from the commutation relations (9-9) (Z +B) as 16 »=0 (cg +4) (0B laait = 0. (108) Hence Caps | init) = constant e~ °F", (10°6) ‘Thus, applying (610) to the «, 8 variables, the probability of there being for the initial state m photons of the A field and » of the B field in theDmac—Quantum Electrodynamics. 35 Photon stato considered is zoro if m+n and is proportional to (- 1) if m =n. Tho probability of there being » photons in the A field and in the B fiold must be taken equal to Pa = 2(e-1)", qo-7) where ¢ is a small positive quantity tonding to zero, in order that we may have 33.P, = 1. é The probabilities (107), equal to 2 and —2, are not physically under- standable, but one can use them mathematically in accordance with the rules for working with a Gibbs ensemble. One can suppose a hypothetical mathe- matical world with the initial probability distribution (10-7) for the photons, and one can work out the probabilities of radiative transition processes occurring in this world. Ono can deduce the carresponding probability Coefficients, i.e. the probabilities per unit intensity of each beam of incident radiation concerned, by using Kinstein’s laws. of radiation. For example, for a process involving the absorption of a photon, if the probability coetlicient is B, the probability of the process is SpnotPnB = -4B, (108) and for a process involving the emission of a photon, if the probability coefficient is A, the probability of the process is Zi (M+1) Pd = 4A. (109) Now the probability of an absorption process, as calculated from the theory, is negative, and that for an emission process is positive, so that, equating these calculated probabilities to (10-8) and (10'9) respectively, one obtains positive values for both B and A, Generally, it is easily verified that any radiative transition probability coefficient obtained by this method is positive. : It now becomes reasonable to assume that these probability coeficients obtained for a hypothetical world are the same as those of the actual world. One gets in this way a general physical interpretation for the quantum theory of photons, When applied to elementary examples, it gives the same results as Heisenberg and Pauli’s quantum electrodynamics with neglect of the divergent integrals, since the extra factor 2°} occurring in the matrix elements of the present theory owing to the \/2 in (97) compensates the factor } in the right-hand side of (10°8) or (10-9), * The above interpretation enables one to caleulate the probability coefficients for all transition processes, which is essentially all that one needs for comparing the theory with experiment. It involves the logical36 Communications of the Dublin Institute—Ser. (A), defect that it requires one to assume Einstein's laws of radiation, instead of enabling one to deduce them from the theory. ‘This defect can be reduced, however, by the circumstance that one can deduce Einstein’s Jaws of radiation for the hypothetical world by working with initial wave functions which are not completély independent of the £’s and 2's, as (10°3) requires, but contain small powers of some of the &’s and 2's, and then seeing how the resulting change in the initial probability distribution of the photons (10°7) affects the probability for transition processes. It then remains only to assume that the same laws of radiation hold for the actual world as for the hypothetical one, an assumption of the same nature as the main assumption above, ; A further point in connection with the interpretation should be noted: The theory involves processes in which a certain photon is emitted and the same photon is absorbed—two actions which cancel each: other and leave nothing observable. However, according to Einstein's laws, such processes would be stimulated by incident radiation in a different way from what they would be if these actions did not occur, and thus there is a possibility of getting experimental evidence for such actions, This point was brought up by correspondenee with W. Paull.