5.2 Low-energy theory
5.2.2 Action expansion
According to this, nA↑ −nB↓ = 0 except for D>U = 1. Although the equation suggests the inexistence of a broken-symmetry phase, the singular behaviour for a fixed value ofUsuggests otherwise. This is an indicator that the extremum ofSdoes not correspond to a minimum in the broken-symmetry phase and that we must approach the problem from a different standpoint.
We will, however, make use of the solution arrived at here, which amounts to the saddle-point solution of the action, valid for the normal phase. But before proceeding, we derive an additional result, which relates the Fermi levelµforU6= 0and the Fermi level of the non-interactive system µ0, in virtue of charge conservation. This reads
n = nA↑ +nB↑ +nA↓ +nB↓
= 2D>(µ0−ε>Λ) + 2D<(−∆<−ε<Λ)
(5.9)
= 2D> µ−U 2
X
τ,σ
nτσ−ε>Λ
!
+ 2D<(−∆<−ε<Λ), (5.13)
where, in the second equality, the electron densities in the non-interactive system were used, ε>Λ, ε<Λ < 0are cone-dependent cutoffs and nis the total electron density. Noting that, in the normal phase, the following relations hold
X
σ
nAσ =X
σ
nBσ =X
τ
nτ↑ =X
τ
nτ↓ = n
2 , (5.14)
we arrive at
µ=µ0+U n
2 . (5.15)
This relation holds for the normal phase as well as for the broken symmetry phase in the low doping regime, but not for any other, as will be shown further on. Note that due to the constancy of the DOS and, therefore, simplicity of the integrals forT →0, from now on most computations involving electron densities will be written directly, as in Eq. (5.13).
¯
nτσ holds. Absorbing the saddle-point contribution into the inverse of the propagator Gˆ∗, we arrive at the bare propagator in the non-interactive systemGˆ0,
~
Gˆ0−1τ
k,σ =i~ωn−ετk,σ+µ−UX
τ0
nτσ¯0 =i~ωn−ετk,σ+µ0,
where Eqs. (5.14) and (5.15) were used in the second equality. Integrating out the fermionic fields,
S[{δφτσ}]/~=N βUX
q
X
τ,τ0
i¯nτ↑δq,0+δφτq,↑
i¯nτ↓0δq,0+δφτ−q,↓0
−Tr ln
"
β~ Gˆ0−1+iU
~ X
τ
δφˆτ
!#
, (5.17) factoring out the bare propagator and simplifying the quadratic term in virtue of Eq. (5.14), we obtain the grand-partition function
Z=Z0
ˆ
D{δφτσ}e−S0[{δφτσ}]/~, (5.18) whereZ0is the grand-partition function of the non-interactive system and the newφ-field action S0reads
S0[{δφτσ}]/~=−N βUn 2
2
+N βUin 2
X
q
X
τ
δφτq,↑+δφτ−q,↓
+N βUX
q
X
τ,τ0
δφτq,↑δφτ−q,↓0 −Trln
"
1 + iU
~ Gˆ0
X
τ
δφˆτ
# .
Now, due to the magnitude of the perturbations{δφτq,σ}relatively to the electron densities{nτσ}, hypothesized previously, it is possible to obtain the action up to quadratic order inδφby Taylor expanding the fermionic contribution of the action which, in virtue of the trace, reads
F[{δφτσ}] = Tr ln
"
1 +iU
~ Gˆ0
X
τ0
δφˆτ0
#
=
=iU
~ X
σ,τ
X
k,k0
Gˆ0
τ,σ
k,k0
X
τ0
δφˆτ0
!σ¯
k0,k
+1 2
U
~ 2
X
σ,τ
X
k,k0,q,q0
Gˆ0
τ,σ
k,k0
X
τ0
δφˆτ0
!σ¯
k0,q0
Gˆ0
τ,σ
q0,q
X
τ0
δφˆτ0
!¯σ
q,k
.
(5.19)
Note that Gˆ0τ,σ
k0,k
= Gˆ0τ
k,σ
δk0,k and P
τ0δφˆτ0σ¯ k0,k
= P
τ0δφτk−k0 0,¯σ . Note also that, within the validity of the parabolic approximation, this expansion is exact, as shown in Sec. D.2 of Appendix D. We now make an ansatz for the decoupling fieldsφ, imposing space- and time- homogeneity
δφτq,σ ≡iηστδq,0,∀τ, σ . (5.20) This ansatz is variational, in the sense that the solution will be determined in virtue of minimiza- tion of the action: equivalently to the motivation of Eq. (5.8), the structure of the path integral Eq. (5.18) is such that the solutions which minimize the action constitute the dominant contribu- tion to the grand-partition function.
Why is this approach necessary, considering it is founded essentially on the same principle as the previous one? Eq. (5.12) suggests that the saddle-point solution fails to account for a minimum in the broken symmetry phase, so that we must take a slightly wider look at the action in the vicinity of the transition. Indeed, this approach will allow for a more intricate manipulation of the action, essentially by including constraints due to conservation laws unaccounted for so far.
Eq. (5.19) then becomes
F({ητσ})/~=−U
~ X
σ,τ
X
k
P
τ0ηστ¯0 iωn−ετk,σ+µ0
−1 2
U
~ 2
X
σ,τ
X
k
P
τ0ητ¯σ0 iωn−ετk,σ+µ0
!2
. (5.21)
We can perform the Matsubara frequency integrations, yielding
F({ητσ})/~=−βUX
σ
X
τ0
ηστ¯0X
k,τ
nF(ετk,σ−µ0)−βU2 2
X
σ
X
τ0
ητ¯σ0
!2 X
k,τ
∂nF(ετk,σ−µ0)
∂ετk,σ , (5.22) where the derivative of the Fermi distribution in the second term appears in virtue of taking the residue of a second-order pole [118]. Replacing summations we have, for the 1st order and 2nd order terms, respectively,
X
σ
X
τ0
ηστ¯0X
k,τ
nF(ετk,σ−µ0) =
T→0
= N
D> µ0−ε>Λ
−D<(∆<+ε<Λ) X
τ
ηστ(5.13)= Nn 2
X
τ,σ
ητσ,
(5.23)
X
σ
X
τ0
ητσ¯0
!2 X
k,τ
∂nF(ετk,σ−µ0)
∂ετk,σ =
T→0
= −N
"
D>
ˆ −∆>
−ε>Λ
dε δ(µ0−ε) +D<
ˆ −∆<
−ε<Λ
dε δ(µ0−ε)
# X
σ
X
τ0
ηστ0
!2
=−N D>X
σ
X
τ0
ητσ0
!2
,
(5.24)
We can now use charge conservation to impose a constraint on the variational parameters{ητσ}.
Retrieving again the fermionic propagator from the effective action of Eq. (5.17), applying the
ansatz of Eq. (5.20) yields
~
Gˆ−1τ k,σ
=i~ωn−ετk,σ+µ0−UX
τ0
ηστ¯0.
This is, in fact, the propagator in the broken symmetry phase, which implies that the non- interactive electronic bands become modified by a spin-dependent shift, reading
Ek,στ =ετk,σ+UX
τ0
ηστ¯0=ετk,σ−σU 2
X
τ0,σ0
σ0ηστ00+U 2
X
τ0,σ0
ητσ00, (5.25)
which shall be referred to as magnetic bands (or cones). The second equality is obtained in virtue of
X
τ0
ητσ¯0 = 1 2
X
τ0,σ0
ηστ00−σ 2
X
τ0,σ0
σ0ηστ00.
Using this result to compute electron densities in the interactive system, we can now compare the cone fillings between the non-interactive and the interactive system, setting a constraint to the parameters{ηστ}in virtue of charge conservation. This reads
n= 2D>(µ0−ε>Λ) + 2D<(−∆<−ε<Λ)
=D> µ0−UX
τ
η↓τ−ε>Λ
!
+D> µ0−UX
τ
ητ↑−ε>Λ
!
+ 2D<(−∆<−ε<Λ),
⇒ X
τ,σ
ητσ= 0 , (5.26)
rendering the 1st order term of the fermionic contribution of the expanded action, Eq. (5.22), vanishing. Moreover, note that
X
σ
X
τ0
ηστ0
!2
= X
τ,σ
ητσ
!2
−2X
τ,τ0
η↑τητ↓0, (5.27)
holds for the 2nd order term and, in turn
X
τ,τ0
η↑τητ↓0 = 1 4
X
τ,σ
ητσ
!2
−1 4
X
τ,σ
σηστ
!2
(5.28)
⇒X
σ
X
τ0
ητσ0
!2
= 1 2
X
τ,σ
ητσ
!2
+1 2
X
τ,σ
σητσ
!2
,
so that the fermionic contribution reads
F({ητσ})/~=N βU D>U X
τ,σ
σητσ 2
!2 ,
yielding, for the the total action,
S0({ητσ})/~=−N βUn 2
2
−N βUn 2
X
τ,σ
ηστ−N βUX
τ,τ0
η↑τητ↓0−N βD>U2 X
τ,σ
σητσ 2
!2
=N βU(1−D>U) X
τ,σ
σηστ 2
!2
−N βUn 2
2 ,
where, in the second equality, Eqs. (5.26) and (5.28) were used. Finally, recovering thatη↓A = ηB↑ = 0holds - accounting for the fact that, in the low doping regime, the↓-Aand↑-B are not affected by charge imbalance - yields
S0(ξ)
N β~U =S0({ηστ}) N β~U +n
2 2
= 1−D>U ξ
2 2
, (5.29)
where ξ = P
τ,σσηστ = ηA↑ −ηB↓. Note that, for D>U > 1, the action is minimized by maxi- mizing the variational parameterξand, thus, charge imbalance becomes non-vanishing. This implies the existence of a broken symmetry phase, wherein the hole-doped system becomes spontaneously spin and valley polarized, provided that the on-site Coulomb interaction energy is greater than the critical value
Uc= 1 D> ,
with the DOSD>given by, according to the fourth equality of Eq. (5.11), D>= 2π
ABZ
m>
~2
=
√3 4π
a
~ 2
m>, (5.30)
whereABZ = (2/√
3)(2π/a)2 is the area of the Brillouin Zone andm> is the effective mass of charge carriers in the high-lying cones of the spin-split valence band, as given by Eq. (3.26). As expected, the onset of the phase transition is defined by a Stoner criterion. Values ofD> and Uc for the different TMDCs are shown in Table 5.1.
Table 5.1: DOS of the high-lying conesD> and critical valueUc for the various TMDCs - low doping. (Units: eV)
M X2 MoS2 WS2 MoSe2 WSe2
D> −1
0.0745 0.0503 0.0876 0.0565
Uc 13.4 19.9 11.4 17.7
These appear to be higher than current estimates for the interaction strength, ranging in 2÷10 eV[119], especially forWcompounds. There are reasons to believe, however, that the strength of the on-site Coulomb interaction (or, put simply, of the Hubbard coupling) might cover these critical values, based on studies carried out for graphene [114], where the energy of the partially screened on-site Coulomb interaction is found to be9.3 eV. TMDCs, in turn, due to their richer electronic structure, are expected to have significantly larger values for the Coulomb interaction, so that these values do not stand to compromise the realization of this phase.
Note, also, that the values used for the effective mass are an estimate, since they are ob- tained from the non-interacting, undoped electronic structure and, as such, are not accurate for the system studied here.
Failure of the saddle-point solution: charge conservation
This result accounts for the findings from the saddle-point solution: indeed, forU = Uc the action is flat and, trivially, the order parameter is arbitrary, and forU > Ucthe action inverts its concavity, so that its extremum is actually a maximum in the broken symmetry phase.
In this sense, the action derived so far is not physical, since it implies that charge imbalance would increase unboundedly to minimize the action. The minimum of the action must then be fixed by charge conservation. In turn, charge conservation imposes that the cone↑-Acan only be filled up to its capacity, which reads, for the spin densitynA↑ −nB↓,
nA↑ −nB↓ =D>
ˆ −∆>
ε>Λ
dε−
ˆ µ0−U2ξ ε>Λ
!
=D>
ˆ −∆>
µ0−U2ξ
dε
=D>U
2 ξ−D>(∆>+µ0), On the other hand, charge imbalance in the magnetic cones is given by
nA↑ −nB↓ =D>
ˆ dε
nF
ε−µ0−U ξ 2
−nF
ε−µ0+U ξ 2
=
ˆ µ0+U2ξ µ0−U2ξ
dε
=D>U ξ .
Equating, we arrive at a constraint for the parameterξ, reading D>U
2 ξ+D>(∆>+µ0) = 0. (5.31) Note that the filling constraint of Eq. (5.31) is equivalent to imposing directly that the the top of
the↑-Acone coincides with the Fermi level, that is,
−∆>=µ0+U ξ 2 ,
and, as will be shown, directly yields the mean-field solution in the broken symmetry phase, which we callξU. Now, we introduce this constraint in the unphysical action of Eq. (5.29) using a Lagrange multiplierλ, which is promptly eliminated by variatingξ,
δ
S0(ξ)−λ D>U
2 ξ+D>(∆>+µ0)
= 0⇒λ=N βU1−D>U D>U ξ . Substituting back, we arrive at the physical action,
S(ξ) N β~U =−
U Uc
−1 ξU −ξ 2
ξ
2, U > Uc, ξ >0 . (5.32) This is the most general form of the action, up to second order of the variational order parameter ξ, valid for all doping regimes. The extremum ofSis then given by
∂S
∂ξ(ξU) = 0⇔ ξU = 2|µ0| −∆>
U , U > Uc , (5.33)
which, clearly, is a minimum in the broken symmetry phaseU > Uc:
∂2S
∂ξ2 = N βU 2
U Uc
−1
>0.
Figure 5.1 depicts the physical picture for this transition, in terms of the cones in the magnetic bands. Additionally, it is possible to determineµA, the Fermi level down to which the↓-Acone remains completely filled, according to the conditionnB↑ =nA↓ forξ=ξU. This condition reads
nB↑ −nA↓ =D<
ˆ −∆<
µA−U ξU/2
dε= 0 ⇔ |µA|= ∆>+ ∆<
2 = ∆. Furthermore, the magnetization density (i.e. spin density) yields
m=nA↑ −nB↓ =D>U ξU = 2D>(|µ0| −∆>), (5.34) which, as expected, is equal to the hole densityp= 2D>(|µ0| −∆>), meaning that all charge transferred to the↑-Acone originates from the↓-B cone, as required by charge conservation.
Indeed, considering the conservation law Eq. (5.26) and thatη↓A=ηB↑ = 0holds, in this regime,
Figure 5.1: Magnetic bands, according to Eq. 5.25, for (a) the normal phase,D>U <1, and (b) the magnetic phase,D>U >1. In (b) we have assumedξ >0.
we can determine the solutionsηA↑,UandηB↓,U in the broken symmetry phase, yielding
η↑,UA =−ηB↓,U =|µ0| −∆>
U . (5.35)
These solutions confirm our ansatz: they are proportional (by a factor ofD>U) to the variation of charge density in each cone and, indeed, such a variation is smaller in magnitude that the total charge density of a cone.
Finally, note that the same results obtained with the formalism used can be obtained us- ing a variational mean-field approach based on the Gibbs-Bogoliubov-Feynman inequality, as discussed in Sec. D.1 of Appendix D.