History
The Smatrix was first introduced by John Archibald Wheeler in the 1937 paper "On the Mathematical Description of Light Nuclei by the Method of Resonating Group Structure".^{[1]} In this paper Wheeler introduced a scattering matrix – a unitary matrix of coefficients connecting "the asymptotic behaviour of an arbitrary particular solution [of the integral equations] with that of solutions of a standard form",^{[2]} but did not develop it fully.
In the 1940s, Werner Heisenberg independently developed and substantiated the idea of the Smatrix. Because of the problematic divergences present in quantum field theory at that time, Heisenberg was motivated to isolate the essential features of the theory that would not be affected by future changes as the theory developed. In doing so, he was led to introduce a unitary "characteristic" Smatrix.^{[2]}
Today, however, exact Smatrix results are a crowning achievement of conformal field theory, integrable systems, and several further areas of quantum field theory and string theory. Smatrices are not substitutes for a fieldtheoretic treatment, but rather, complement the end results of such.
In onedimensional quantum mechanics
A simple prototype in which the Smatrix is 2dimensional is considered first, for the purposes of illustration. In it, particles with sharp energy E scatter from a localized potential V according to the rules of 1dimensional quantum mechanics. Already this simple model displays some features of more general cases, but is easier to handle.
Each energy E yields a matrix S = S(E) that depends on V. Thus, the total Smatrix could, figuratively speaking, be visualized, in a suitable basis, as a "continuous matrix" with every element zero except for 2 × 2blocks along the diagonal for a given V.
Definition
Consider a localized one dimensional potential barrier V(x), subjected to a beam of quantum particles with energy E. These particles are incident on the potential barrier from left to right.
The solutions of Schrödinger's equation outside the potential barrier are plane waves given by
$\psi _{\rm {L))(x)=Ae^{ikx}+Be^{ikx))$
for the region to the left of the potential barrier, and
$\psi _{\rm {R))(x)=Ce^{ikx}+De^{ikx))$
for the region to the right to the potential barrier, where
$k={\sqrt {2mE/\hbar ^{2))))$
is the wave vector. The time dependence is not needed in our overview and is hence omitted. The term with coefficient A represents the incoming wave, whereas term with coefficient C represents the outgoing wave. B stands for the reflecting wave. Since we set the incoming wave moving in the positive direction (coming from the left), D is zero and can be omitted.
The "scattering amplitude", i.e., the transition overlap of the outgoing waves with the incoming waves is a linear relation defining the Smatrix,
${\begin{pmatrix}B\\C\end{pmatrix))={\begin{pmatrix}S_{11}&S_{12}\\S_{21}&S_{22}\end{pmatrix)){\begin{pmatrix}A\\D\end{pmatrix)).$
The above relation can be written as
$\Psi _{\rm {out))=S\Psi _{\rm {in))$
where
$\Psi _{\rm {out))={\begin{pmatrix}B\\C\end{pmatrix)),\quad \Psi _{\rm {in))={\begin{pmatrix}A\\D\end{pmatrix)),\qquad S={\begin{pmatrix}S_{11}&S_{12}\\S_{21}&S_{22}\end{pmatrix)).$
The elements of S completely characterize the scattering properties of the potential barrier V(x).
Unitary property
The unitary property of Smatrix is directly related to the conservation of the probability current in quantum mechanics.
The probability current J of the wave function ψ(x) is defined as
$J={\frac {\hbar }{2mi))\left(\psi ^{*}{\frac {\partial \psi }{\partial x))\psi {\frac {\partial \psi ^{*)){\partial x))\right).$
The current density to the left of the barrier is
$J_{\rm {L))={\frac {\hbar k}{m))\left(A^{2}B^{2}\right),$
while the current density to the right of the barrier is
$J_{\rm {R))={\frac {\hbar k}{m))\left(C^{2}D^{2}\right).$
For conservation of the probability current density, J_{L} = J_{R}. This implies the Smatrix is a unitary matrix.
Proof
${\begin{aligned}&J_{\rm {L))=J_{\rm {R))\\&A^{2}B^{2}=C^{2}D^{2}\\&B^{2}+C^{2}=A^{2}+D^{2}\\&\Psi _{\text{out))^{\dagger }\Psi _{\text{out))=\Psi _{\text{in))^{\dagger }\Psi _{\text{in))\\&\Psi _{\text{in))^{\dagger }S^{\dagger }S\Psi _{\text{in))=\Psi _{\text{in))^{\dagger }\Psi _{\text{in))\\&S^{\dagger }S=I\\\end{aligned))$
Timereversal symmetry
If the potential V(x) is real, then the system possesses timereversal symmetry. Under this condition, if ψ(x) is a solution of Schrödinger's equation, then ψ*(x) is also a solution.
The timereversed solution is given by
$\psi _{\rm {L))^{*}(x)=A^{*}e^{ikx}+B^{*}e^{ikx))$
for the region to the left to the potential barrier, and
$\psi _{\rm {R))^{*}(x)=C^{*}e^{ikx}+D^{*}e^{ikx))$
for the region to the right to the potential barrier,
where the terms with coefficient B*, C* represent incoming wave, and terms with coefficient A*, D* represent outgoing wave.
They are again related by the Smatrix,
${\begin{pmatrix}A^{*}\\D^{*}\end{pmatrix))={\begin{pmatrix}S_{11}&S_{12}\\S_{21}&S_{22}\end{pmatrix)){\begin{pmatrix}B^{*}\\C^{*}\end{pmatrix))\,$
that is,
$\Psi _{\rm {in))^{*}=S\Psi _{\rm {out))^{*}.$
Now, the relations
$\Psi _{\rm {in))^{*}=S\Psi _{\rm {out))^{*},\quad \Psi _{\rm {out))=S\Psi _{\rm {in))$
together yield a condition
$S^{*}S=I$
This condition, in conjunction with the unitarity relation, implies that the Smatrix is symmetric, as a result of time reversal symmetry,
$S^{T}=S.$
Transmission coefficient and reflection coefficient
The transmission coefficient from the left of the potential barrier is, when D = 0,
$T_{\rm {L))={\frac {C^{2)){A^{2))}=S_{21}^{2}.$
The reflection coefficient from the left of the potential barrier is, when D = 0,
$R_{\rm {L))={\frac {B^{2)){A^{2))}=S_{11}^{2}.$
Similarly, the transmission coefficient from the right of the potential barrier is, when A = 0,
$T_{\rm {R))={\frac {B^{2)){D^{2))}=S_{12}^{2}.$
The reflection coefficient from the right of the potential barrier is, when A = 0,
$R_{\rm {R))={\frac {C^{2)){D^{2))}=S_{22}^{2}.$
The relations between the transmission and reflection coefficients are
$T_{\rm {L))+R_{\rm {L))=1$
and
$T_{\rm {R))+R_{\rm {R))=1.$
This identity is a consequence of the unitarity property of the Smatrix.
Optical theorem in one dimension
In the case of free particles V(x) = 0, the Smatrix is^{[3]}
$S={\begin{pmatrix}0&1\\1&0\end{pmatrix)).$
Whenever V(x) is different from zero, however, there is a departure of the Smatrix from the above form, to
$S={\begin{pmatrix}2ir&1+2it\\1+2it&2ir^{*}{\frac {1+2it}{12it^{*))}\end{pmatrix)).$
This departure is parameterized by two complex functions of energy, r and t.
From unitarity there also follows a relationship between these two functions,
$r^{2}+t^{2}=\operatorname {Im} (t).$
The analogue of this identity in three dimensions is known as the optical theorem.
Definition in quantum field theory
Interaction picture
A straightforward way to define the Smatrix begins with considering the interaction picture.^{[4]} Let the Hamiltonian H be split into the free part H_{0} and the interaction V, H = H_{0} + V. In this picture, the operators behave as free field operators and the state vectors have dynamics according to the interaction V. Let
$\left\Psi (t)\right\rangle$
denote a state that has evolved from a free initial state
$\left\Phi _{\rm {i))\right\rangle .$
The Smatrix element is then defined as the projection of this state on the final state
$\left\langle \Phi _{\rm {f))\right.$
Thus
$S_{\rm {fi))\equiv \lim _{t\rightarrow +\infty }\left\langle \Phi _{\rm {f))\Psi (t)\right\rangle \equiv \left\langle \Phi _{\rm {f))\rightS\left\Phi _{\rm {i))\right\rangle ,$
where S is the Soperator. The great advantage of this definition is that the timeevolution operator U evolving a state in the interaction picture is formally known,^{[5]}
$U(t,t_{0})=Te^{i\int _{t_{0))^{t}d\tau V(\tau )},$
where T denotes the timeordered product. Expressed in this operator,
$S_{\rm {fi))=\lim _{t_{2}\rightarrow +\infty }\lim _{t_{1}\rightarrow \infty }\left\langle \Phi _{\rm {f))\rightU(t_{2},t_{1})\left\Phi _{\rm {i))\right\rangle ,$
from which
$S=U(\infty ,\infty ).$
Expanding using the knowledge about U gives a Dyson series,
$S=\sum _{n=0}^{\infty }{\frac {(i)^{n)){n!))\int _{\infty }^{\infty }dt_{1}\cdots \int _{\infty }^{\infty }dt_{n}T\left[V(t_{1})\cdots V(t_{n})\right],$
or, if V comes as a Hamiltonian density,
$S=\sum _{n=0}^{\infty }{\frac {(i)^{n)){n!))\int _{\infty }^{\infty }dx_{1}^{4}\cdots \int _{\infty }^{\infty }dx_{n}^{4}T\left[{\mathcal {H))(t_{1})\cdots {\mathcal {H))(t_{n})\right].$
Being a special type of timeevolution operator, S is unitary. For any initial state and any final state one finds
$S_{\rm {fi))=\left\langle \Phi _{\rm {f))S\Phi _{\rm {i))\right\rangle =\left\langle \Phi _{\rm {f))\left\sum _{n=0}^{\infty }{\frac {(i)^{n)){n!))\int _{\infty }^{\infty }dx_{1}^{4}\cdots \int _{\infty }^{\infty }dx_{n}^{4}T\left[{\mathcal {H))(t_{1})\cdots {\mathcal {H))(t_{n})\right]\right\Phi _{\rm {i))\right\rangle .$
This approach is somewhat naïve in that potential problems are swept under the carpet.^{[6]} This is intentional. The approach works in practice and some of the technical issues are addressed in the other sections.
In and out states
Here a slightly more rigorous approach is taken in order to address potential problems that were disregarded in the interaction picture approach of above. The final outcome is, of course, the same as when taking the quicker route. For this, the notions of in and out states are needed. These will be developed in two ways, from vacua, and from free particle states. Needless to say, the two approaches are equivalent, but they illuminate matters from different angles.
From vacua
If a^{†}(k) is a creation operator, its hermitian adjoint is an annihilation operator and destroys the vacuum,
$a(k)\left*,0\right\rangle =0.$
In Dirac notation, define
$*,0\rangle$
as a vacuum quantum state, i.e. a state without real particles. The asterisk signifies that not all vacua are necessarily equal, and certainly not equal to the Hilbert space zero state 0. All vacuum states are assumed Poincaré invariant, invariance under translations, rotations and boosts,^{[6]} formally,
$P^{\mu }*,0\rangle =*,0\rangle ,\quad M^{\mu \nu }*,0\rangle =*,0\rangle ,$
where P^{μ} is the generator of translation in space and time, and M^{μν} is the generator of Lorentz transformations. Thus the description of the vacuum is independent of the frame of reference. Associated to the in and out states to be defined are the in and out field operators (aka fields) Φ_{i} and Φ_{o}. Attention is here focused to the simplest case, that of a scalar theory in order to exemplify with the least possible cluttering of the notation. The in and out fields satisfy
$(\Box ^{2}+m^{2})\phi _{\rm {i,o))(x)=0,$
the free Klein–Gordon equation. These fields are postulated to have the same equal time commutation relations (ETCR) as the free fields,
${\begin{aligned}{[\phi _{\rm {i,o))(x),\pi _{\rm {i,o))(y)]}_{x_{0}=y_{0))&=i\delta (\mathbf {x} \mathbf {y} ),\\{[\phi _{\rm {i,o))(x),\phi _{\rm {i,o))(y)]}_{x_{0}=y_{0))&={[\pi _{\rm {i,o))(x),\pi _{\rm {i,o))(y)]}_{x_{0}=y_{0))=0\end{aligned)),$
where π_{i,j} is the field canonically conjugate to Φ_{i,j}. Associated to the in and out fields are two sets of creation and annihilation operators, a^{†}_{i}(k) and a^{†}_{f} (k), acting in the same Hilbert space,^{[7]} on two distinct complete sets (Fock spaces; initial space i, final space f). These operators satisfy the usual commutation rules,
${\begin{aligned}{[a_{\rm {i,o))(\mathbf {p} ),a_{\rm {i,o))^{\dagger }(\mathbf {p} ')]}&=i\delta (\mathbf {p} \mathbf {p'} ),\\{[a_{\rm {i,o))(\mathbf {p} ),a_{\rm {i,o))(\mathbf {p'} )]}&={[a_{\rm {i,o))^{\dagger }(\mathbf {p} ),a_{\rm {i,o))^{\dagger }(\mathbf {p'} )]}=0.\end{aligned))$
The action of the creation operators on their respective vacua and states with a finite number of particles in the in and out states is given by
${\begin{aligned}\left\mathrm {i} ,k_{1}\ldots k_{n}\right\rangle &=a_{i}^{\dagger }(k_{1})\cdots a_{\rm {i))^{\dagger }(k_{n})\lefti,0\right\rangle ,\\\left\mathrm {f} ,p_{1}\ldots p_{n}\right\rangle &=a_{\rm {f))^{\dagger }(p_{1})\cdots a_{f}^{\dagger }(p_{n})\leftf,0\right\rangle ,\end{aligned))$
where issues of normalization have been ignored. See the next section for a detailed account on how a general nparticle state is normalized. The initial and final spaces are defined by
${\mathcal {H))_{\rm {i))=\operatorname {span} \{\left\mathrm {i} ,k_{1}\ldots k_{n}\right\rangle =a_{\rm {i))^{\dagger }(k_{1})\cdots a_{\rm {i))^{\dagger }(k_{n})\left\mathrm {i} ,0\right\rangle \},$
${\mathcal {H))_{\rm {f))=\operatorname {span} \{\leftf,p_{1}\ldots p_{n}\right\rangle =a_{\rm {f))^{\dagger }(p_{1})\cdots a_{\rm {f))^{\dagger }(p_{n})\left\mathrm {f} ,0\right\rangle \}.$
The asymptotic states are assumed to have well defined Poincaré transformation properties, i.e. they are assumed to transform as a direct product of oneparticle states.^{[8]} This is a characteristic of a noninteracting field. From this follows that the asymptotic states are all eigenstates of the momentum operator P^{μ},^{[6]}
$P^{\mu }\left\mathrm {i} ,k_{1}\ldots k_{m}\right\rangle =k_{1}^{\mu }+\cdots +k_{m}^{\mu }\left\mathrm {i} ,k_{1}\ldots k_{m}\right\rangle ,\quad P^{\mu }\left\mathrm {f} ,p_{1}\ldots p_{n}\right\rangle =p_{1}^{\mu }+\cdots +p_{n}^{\mu }\left\mathrm {f} ,p_{1}\ldots p_{n}\right\rangle .$
In particular, they are eigenstates of the full Hamiltonian,
$H=P^{0}.$
The vacuum is usually postulated to be stable and unique,^{[6]}^{[nb 1]}
$\mathrm {i} ,0\rangle =\mathrm {f} ,0\rangle =*,0\rangle \equiv 0\rangle .$
The interaction is assumed adiabatically turned on and off.
Heisenberg picture
The Heisenberg picture is employed henceforth. In this picture, the states are timeindependent. A Heisenberg state vector thus represents the complete spacetime history of a system of particles.^{[8]} The labeling of the in and out states refers to the asymptotic appearance. A state Ψ_{α, in} is characterized by that as t → −∞ the particle content is that represented collectively by α. Likewise, a state Ψ_{β, out} will have the particle content represented by β for t → +∞. Using the assumption that the in and out states, as well as the interacting states, inhabit the same Hilbert space and assuming completeness of the normalized in and out states (postulate of asymptotic completeness^{[6]}), the initial states can be expanded in a basis of final states (or vice versa). The explicit expression is given later after more notation and terminology has been introduced. The expansion coefficients are precisely the Smatrix elements to be defined below.
While the state vectors are constant in time in the Heisenberg picture, the physical states they represent are not. If a system is found to be in a state Ψ at time t = 0, then it will be found in the state U(τ)Ψ = e^{−iHτ}Ψ at time t = τ. This is not (necessarily) the same Heisenberg state vector, but it is an equivalent state vector, meaning that it will, upon measurement, be found to be one of the final states from the expansion with nonzero coefficient. Letting τ vary one sees that the observed Ψ (not measured) is indeed the Schrödinger picture state vector. By repeating the measurement sufficiently many times and averaging, one may say that the same state vector is indeed found at time t = τ as at time t = 0. This reflects the expansion above of an in state into out states.
From free particle states
For this viewpoint, one should consider how the archetypical scattering experiment is performed. The initial particles are prepared in well defined states where they are so far apart that they don't interact. They are somehow made to interact, and the final particles are registered when they are so far apart that they have ceased to interact. The idea is to look for states in the Heisenberg picture that in the distant past had the appearance of free particle states. This will be the in states. Likewise, an out state will be a state that in the distant future has the appearance of a free particle state.^{[8]}
The notation from the general reference for this section, Weinberg (2002) will be used. A general noninteracting multiparticle state is given by
$\Psi _{p_{1}\sigma _{1}n_{1};p_{2}\sigma _{2}n_{2};\cdots },$
where
 p is momentum,
 σ is spin zcomponent or, in the massless case, helicity,
 n is particle species.
These states are normalized as
$\left(\Psi _{p_{1}'\sigma _{1}'n_{1}';p_{2}'\sigma _{2}'n_{2}';\cdots },\Psi _{p_{1}\sigma _{1}n_{1};p_{2}\sigma _{2}n_{2};\cdots }\right)=\delta ^{3}(\mathbf {p} _{1}'\mathbf {p} _{1})\delta _{\sigma _{1}'\sigma _{1))\delta _{n_{1}'n_{1))\delta ^{3}(\mathbf {p} _{2}'\mathbf {p} _{2})\delta _{\sigma _{2}'\sigma _{2))\delta _{n_{2}'n_{2))\cdots \quad \pm {\text{ permutations)).$
Permutations work as such; if s ∈ S_{k} is a permutation of k objects (for a kparticle state) such that
$n_{s(i)}'=n_{i},\quad 1\leq i\leq k,$
then a nonzero term results. The sign is plus unless s involves an odd number of fermion transpositions, in which case it is minus. The notation is usually abbreviated letting one Greek letter stand for the whole collection describing the state. In abbreviated form the normalization becomes
$\left(\Psi _{\alpha '},\Psi _{\alpha }\right)=\delta (\alpha '\alpha ).$
When integrating over freeparticle states one writes in this notation
$d\alpha \cdots \equiv \sum _{n_{1}\sigma _{1}n_{2}\sigma _{2}\cdots }\int d^{3}p_{1}d^{3}p_{2}\cdots ,$
where the sum includes only terms such that no two terms are equal modulo a permutation of the particle type indices. The sets of states sought for are supposed to be complete. This is expressed as
$\Psi =\int d\alpha \ \Psi _{\alpha }\left(\Psi _{\alpha },\Psi \right),$
which could be paraphrased as
$\int d\alpha \ \left\Psi _{\alpha }\right\rangle \left\langle \Psi _{\alpha }\right=1,$
where for each fixed α, the right hand side is a projection operator onto the state α. Under an inhomogeneous Lorentz transformation (Λ, a), the field transforms according to the rule
$U(\Lambda ,a)\Psi _{p_{1}\sigma _{1}n_{1};p_{2}\sigma _{2}n_{2}\cdots }=e^{ia_{\mu }((\Lambda p_{1})^{\mu }+(\Lambda p_{2})^{\mu }+\cdots )}{\sqrt {\frac {(\Lambda p_{1})^{0}(\Lambda p_{2})^{0}\cdots }{p_{1}^{0}p_{2}^{0}\cdots ))}\sum _{\sigma _{1}'\sigma _{2}'\cdots }D_{\sigma _{1}'\sigma _{1))^{(j_{1})}(W(\Lambda ,p_{1}))D_{\sigma _{2}'\sigma _{2))^{(j_{2})}(W(\Lambda ,p_{2}))\cdots \Psi _{\Lambda p_{1}\sigma _{1}'n_{1};\Lambda p_{2}\sigma _{2}'n_{2}\cdots },$


(1)

where W(Λ, p) is the Wigner rotation and D^{(j)} is the (2j + 1)dimensional representation of SO(3). By putting Λ = 1, a = (τ, 0, 0, 0), for which U is exp(iHτ), in (1), it immediately follows that
$H\Psi =E_{\alpha }\Psi ,\quad E_{\alpha }=p_{1}^{0}+p_{2}^{0}+\cdots ,$
so the in and out states sough after are eigenstates of the full Hamiltonian that are necessarily noninteracting due to the absence of mixed particle energy terms. The discussion in the section above suggests that the in states Ψ^{+} and the out states Ψ^{−} should be such that
$e^{iH\tau }\int d\alpha \ g(\alpha )\Psi _{\alpha }^{\pm }=\int d\alpha \ e^{iE_{\alpha }\tau }g(\alpha )\Psi _{\alpha }^{\pm ))$
for large positive and negative τ has the appearance of the corresponding package, represented by g, of freeparticle states, g assumed smooth and suitably localized in momentum. Wave packages are necessary, else the time evolution will yield only a phase factor indicating free particles, which cannot be the case. The right hand side follows from that the in and out states are eigenstates of the Hamiltonian per above. To formalize this requirement, assume that the full Hamiltonian H can be divided into two terms, a freeparticle Hamiltonian H_{0} and an interaction V, H = H_{0} + V such that the eigenstates Φ_{γ} of H_{0} have the same appearance as the in and outstates with respect to normalization and Lorentz transformation properties,
$H_{0}\Phi _{\alpha }=E_{\alpha }\Phi _{\alpha },$
$(\Phi _{\alpha }',\Phi _{\alpha })=\delta (\alpha '\alpha ).$
The in and out states are defined as eigenstates of the full Hamiltonian,
$H\Psi _{\alpha }^{\pm }=E_{\alpha }\Psi _{\alpha }^{\pm },$
satisfying
$e^{iH\tau }\int d\alpha \ g(\alpha )\Psi _{\alpha }^{\pm }\rightarrow e^{iH_{0}\tau }\int d\alpha \ g(\alpha )\Phi _{\alpha }.$
for τ → −∞ or τ → +∞ respectively. Define
$\Omega (\tau )\equiv e^{+iH\tau }e^{iH_{0}\tau },$
then
$\Psi _{\alpha }^{\pm }=\Omega (\mp \infty )\Phi _{\alpha }.$
This last expression will work only using wave packages.From these definitions follow that the in and out states are normalized in the same way as the freeparticle states,
$(\Psi _{\beta }^{+},\Psi _{\alpha }^{+})=(\Phi _{\beta },\Phi _{\alpha })=(\Psi _{\beta }^{},\Psi _{\alpha }^{})=\delta (\beta \alpha ),$
and the three sets are unitarily equivalent. Now rewrite the eigenvalue equation,
$(E_{\alpha }H_{0}\pm i\epsilon )\Psi _{\alpha }^{\pm }=\pm i\epsilon \Psi _{\alpha }^{\pm }+V\Psi _{\alpha }^{\pm },$
where the ±iε terms has been added to make the operator on the LHS invertible. Since the in and out states reduce to the freeparticle states for V → 0, put
$i\epsilon \Psi _{\alpha }^{\pm }=i\epsilon \Phi _{\alpha ))$
on the RHS to obtain
$\Psi _{\alpha }^{\pm }=\Phi _{\alpha }+(E_{\alpha }H_{0}\pm i\epsilon )^{1}V\Psi _{\alpha }^{\pm }.$
Then use the completeness of the freeparticle states,
$V\Psi _{\alpha }^{\pm }=\int d\beta \ (\Phi _{\beta },V\Psi _{\alpha }^{\pm })\Phi _{\beta }\equiv \int d\beta \ T_{\beta \alpha }^{\pm }\Phi _{\beta },$
to finally obtain
$\Psi _{\alpha }^{\pm }=\Phi _{\alpha }+\int d\beta \ {\frac {T_{\beta \alpha }^{\pm }\Phi _{\beta )){E_{\alpha }E_{\beta }\pm i\epsilon )).$
Here H_{0} has been replaced by its eigenvalue on the freeparticle states. This is the Lippmann–Schwinger equation.
In states expressed as out states
The initial states can be expanded in a basis of final states (or vice versa). Using the completeness relation,
$\Psi _{\alpha }^{}=\int d\beta (\Psi _{\beta }^{+},\Psi _{\alpha }^{})\Psi _{\beta }^{+}=\int d\beta \Psi _{\beta }^{+}\rangle \langle \Psi _{\beta }^{+}\Psi _{\alpha }^{}\rangle =\sum _{n_{1}\sigma _{1}n_{2}\sigma _{2}\cdots }\int d^{3}p_{1}d^{3}p_{2}\cdots (\Psi _{\beta }^{+},\Psi _{\alpha }^{})\Psi _{\beta }^{+},$
$\Psi _{\alpha }^{}=\left\mathrm {i} ,k_{1}\ldots k_{n}\right\rangle =C_{0}\left\mathrm {f} ,0\right\rangle \ +\sum _{m=1}^{\infty }\int {d^{4}p_{1}\ldots d^{4}p_{m}C_{m}(p_{1}\ldots p_{m})\left\mathrm {f} ,p_{1}\ldots p_{m}\right\rangle }~,$
where C_{m}^{2} is the probability that the interaction transforms
$\left\mathrm {i} ,k_{1}\ldots k_{n}\right\rangle =\Psi _{\alpha }^{))$
into
$\left\mathrm {f} ,p_{1}\ldots p_{m}\right\rangle =\Psi _{\beta }^{+))$
.
By the ordinary rules of quantum mechanics,
$C_{m}(p_{1}\ldots p_{m})=\left\langle \mathrm {f} ,p_{1}\ldots p_{m}\right\mathrm {i} ,k_{1}\ldots k_{n}\rangle =(\Psi _{\beta }^{+},\Psi _{\alpha }^{})$
and one may write
$\left\mathrm {i} ,k_{1}\ldots k_{n}\right\rangle =C_{0}\left\mathrm {f} ,0\right\rangle \ +\sum _{m=1}^{\infty }\int {d^{4}p_{1}\ldots d^{4}p_{m}\left\mathrm {f} ,p_{1}\ldots p_{m}\right\rangle }\left\langle \mathrm {f} ,p_{1}\ldots p_{m}\right\mathrm {i} ,k_{1}\ldots k_{n}\rangle ~.$
The expansion coefficients are precisely the Smatrix elements to be defined below.
The Smatrix
The Smatrix is now defined by^{[8]}
$S_{\beta \alpha }=\langle \Psi _{\beta }^{}\Psi _{\alpha }^{+}\rangle =\langle \mathrm {f} ,\beta \mathrm {i} ,\alpha \rangle ,\qquad \mathrm {f} ,\beta \rangle \in {\mathcal {H))_{\rm {f)),\quad \mathrm {i} ,\alpha \rangle \in {\mathcal {H))_{\rm {i)).$
Here α and β are shorthands that represent the particle content but suppresses the individual labels. Associated to the Smatrix there is the Soperator S defined by^{[8]}
$\langle \Phi _{\beta }S\Phi _{\alpha }\rangle \equiv S_{\beta \alpha },$
where the Φ_{γ} are free particle states.^{[8]}^{[nb 2]} This definition conforms with the direct approach used in the interaction picture. Also, due to unitary equivalence,
$\langle \Psi _{\beta }^{+}S\Psi _{\alpha }^{+}\rangle =S_{\beta \alpha }=\langle \Psi _{\beta }^{}S\Psi _{\alpha }^{}\rangle .$
As a physical requirement, S must be a unitary operator. This is a statement of conservation of probability in quantum field theory. But
$\langle \Psi _{\beta }^{}S\Psi _{\alpha }^{}\rangle =S_{\beta \alpha }=\langle \Psi _{\beta }^{}\Psi _{\alpha }^{+}\rangle .$
By completeness then,
$S\Psi _{\alpha }^{}\rangle =\Psi _{\alpha }^{+}\rangle ,$
so S is the unitary transformation from instates to out states.
Lorentz invariance is another crucial requirement on the Smatrix.^{[8]}^{[nb 3]} The Soperator represents the quantum canonical transformation of the initial in states to the final out states. Moreover, S leaves the vacuum state invariant and transforms inspace fields to outspace fields,^{[nb 4]}
$S\left0\right\rangle =\left0\right\rangle$
$\phi _{\mathrm {f} }=S\phi _{\mathrm {i} }S^{1}~.$
In terms of creation and annihilation operators, this becomes
$a_{\rm {f))(p)=Sa_{\rm {i))(p)S^{1},a_{\rm {f))^{\dagger }(p)=Sa_{\rm {i))^{\dagger }(p)S^{1},$
hence
${\begin{aligned}S\mathrm {i} ,k_{1},k_{2},\ldots ,k_{n}\rangle &=Sa_{\rm {i))^{\dagger }(k_{1})a_{\rm {i))^{\dagger }(k_{2})\cdots a_{\rm {i))^{\dagger }(k_{n})0\rangle =Sa_{\rm {i))^{\dagger }(k_{1})S^{1}Sa_{\rm {i))^{\dagger }(k_{2})S^{1}\cdots Sa_{\rm {i))^{\dagger }(k_{n})S^{1}S0\rangle \\&=a_{\rm {o))^{\dagger }(k_{1})a_{\rm {o))^{\dagger }(k_{2})\cdots a_{\rm {o))^{\dagger }(k_{n})S0\rangle =a_{\rm {o))^{\dagger }(k_{1})a_{\rm {o))^{\dagger }(k_{2})\cdots a_{\rm {o))^{\dagger }(k_{n})0\rangle =\mathrm {o} ,k_{1},k_{2},\ldots ,k_{n}\rangle .\end{aligned))$
A similar expression holds when S operates to the left on an out state. This means that the Smatrix can be expressed as
$S_{\beta \alpha }=\langle \mathrm {o} ,\beta \mathrm {i} ,\alpha \rangle =\langle \mathrm {i} ,\beta S\mathrm {i} ,\alpha \rangle =\langle \mathrm {o} ,\beta S\mathrm {o} ,\alpha \rangle .$
If S describes an interaction correctly, these properties must be also true:
 If the system is made up with a single particle in momentum eigenstate k⟩, then Sk⟩ = k⟩. This follows from the calculation above as a special case.
 The Smatrix element may be nonzero only where the output state has the same total momentum as the input state. This follows from the required Lorentz invariance of the Smatrix.