The Staggered Fermion for the Gross-Neveu Model at Non-Zero Temperature and Density ()
1. Introduction
The chiral phase transition in quantum chromodynamics (QCD) from the hadronic phase at low temperature T (low density
) to the quark-gluon plasma phase at high temperature (high density) has been studied intensively in the last decade. Although the relative firm statements for the phase structure can be made in two limit cases: finite T with small baryon density
and asymptotically high density
, the phase structures at the intermediate baryon density are not clear. For a recent and review and related work of QCD with finite density, see Ref. [1] - [9].
Since the chiral symmetry breaking and restoration are intrinsically non-perturbative, the number of techniques is limited and most results come from the lattice QCD. Unfortunately, the lattice QCD at finite density suffers from the notorious sign problem, especially for the intermediate or large baryon density. For some simpler quantum field models, e.g., the dense two-color QCD [10], the sign problem can be avoided. The recent progress of the sign problem in lattice field models can refer to [11] and references therein. In the last decades, the tensor network becomes very popular in condensed matter physics and high energy physics, especial for lower dimension models, since probability is not used and thus it is free of sign problem [12] [13] [14] [15].
This paper addresses a simplest four-fermion model with
symmetry: Gross-Neveu model at non-zero temperature and density [16] [17] [18] [19] [20]. The 2 + 1d Gross-Neveu model has an interesting continuum limit and there is a critical coupling indicating the threshold for the symmetry breaking at zero temperature and density. Although the 2 + 1d Gross-Neveu model is not renormalisable in the weak coupling expansion, it is renormalisable in
expansion [16], where
is the number of flavors of fermions.
The symmetry breaking of Gross-Neveu model for the 1 + 1d case has been studied extensively [21] - [29]. Recently, 2 + 1d Gross-Neveu model is used to study the inhomogeneous phases [30] and the symmetry breaking [31].
Compared with the Wilson fermion, the staggered fermion is more adequate for studying spontaneous chiral symmetry breaking. Another advantage of the staggered fermion is due to the reduced computational cost since the Dirac matrices have been replaced by the staggered phase factor. The reconstruction of the Wilson-like fermion from the staggered fermion is rather technique, thus needing a careful explanation of the physical fermions for lattice QCD [32] and for Gross-Neveu model [18].
In this paper, we revisit the staggered fermion for the 1 + 1d, 2 + 1d and 3 + 1d Gross-Neveu model at non-zero temperature and finite density. The gap equation, which is based on the large
limit, is solved in the momentum space. Moreover, we derive an explicit formula for the inverse matrix of the staggered fermion matrix, which is easy to be implemented by parallelization and thus make the large scale calculation of the gap equation feasible.
The arrangement of the paper is as follows. The continuum 2 + 1d Gross-Neveu model at finite density and non-zero temperature is introduced in Section 2. In Section 3, the 2 + 1d staggered fermion is shown and non-dimensional quantities are introduced. The kinetic part of staggered fermion in the momentum space is given in Section 4, where the trace of the inverse matrix and elements of inverse matrix are given explicitly in momentum space. In Section 5, the results in Section 4 are generalized to the 1 + 1d and 3 + 1d staggered fermion. The gap equation is given in Section 6, where the chiral condensate and fermion density are calculated. The simulation results in the large
limit are obtained in Section 7. Finally, the conclusion is given in Section 8.
2. The Gross-Neveu Model
The Gross-Neveu model for interacting fermions in 2 + 1d is defined by the continuum Euclidiean Lagrangian density at finite density
(1)
where
,
is the chemical potential,
the bare mass,
and
are an
-flavor 4 component spinor fields. Here we choose the Gamma matrices
(2)
(3)
where
are the Pauli matrices. The Gamma matrices satisfies
There is a discrete
symmetry
,
, which is broken by the mass term but not the interaction. Introducing the bosonic field
, the interaction between fermions is decoupled with the Lagrangian density,
(4)
The dimension of quantities for the 2 + 1d Gross-Neveu model is as follows
(5)
The partition function for this model is
(6)
where
with the inverse temperature
and the space size L.
and
are antiperiodic in
direction, and are periodic in
and
directions. We want to calculate the chiral condensate for one flavor
(7)
where
is the volume of 2 + 1d system. In the second equality we used
Since the Lagrangian density is translation invariant,
and
does not depend on x. This model in the large
limit can be solved exactly [18] in the chiral limit
, which is based on the saddle approximation (gap equation) in (6)
(8)
where in the third equality we write the trace of operator in momentum space and the summation over
3. The Staggered Fermion
The staggered fermion discretization of the action
is
(9)
with staggered phase factor
,
,
.
,
. The boundary condition for
and
are accounted for by the sign
and
(10)
Here
is defined on lattice
by
(11)
where
denotes 8 dual lattices
which is neighbour to
. The auxiliary field on dual lattice for two dimensional Gross-Neveu model was first studied in Ref. [33].
According to (5), the non-dimensional quantities are introduced by
(12)
(13)
and thus the action in (9) can be rewritten as
The partition function for the Gross-Neveu model with
flavors is:
(14)
where
and
denote the Grassmann fields of flavors
at the sites
,
is the real field defined at the dual lattice sites
. The action is
(15)
where
(16)
The derivative of this matrix D with respect to the chemical potential and bare mass are rather simple
The real matrix
satisfies the following symmetry
where
is the parity of site
.
By integrating the Grassmann fields, the partition function in (14) can be rewritten as
(17)
with the effective action
(18)
and
(19)
The computational results, e.g., non-dimensional chiral condensate and fermion density, depend on the non-dimensional quantities
The physical dimensional quantities can be recovered from the non-dimensional ones by introducing lattice size a according to (12), (13). For notation simplicity, we set
and thus
in the following discussion.
4. Staggered Fermion in Momentum Space
The kinetic part in (15) in one flavor is
where
(20)
and
are the Grassmann fields defined on lattices. A Wilson-like fermion can be obtained from the stagger fermion
[18].
Assume that
and
are even integers. Let
denotes a site on a lattice of twice the spacing of the original, and
is a lattice vector, which ranges over the corners of the elementary cube associated with Y, so that each site on the original lattice
uniquely corresponds to A and Y:
. Introducing notation
A shift along
direction can be represented by
(21)
Similarly,
(22)
is defined on the fine lattice sites
with lattice size
(23)
while
on the coarse lattice sites Y with lattice size
(24)
A unitary transformation of
is defined by [34]
(25)
(26)
where
matrices
and
is given by
(27)
and
satisfies the following properties (The indices
and b always run from 1 to 2)
(28)
(29)
(30)
(31)
Equation (31) is also valid if
is replaced by B.
(32)
(33)
See Appendix A for these properties.
Using (29), the inverse transformation of (25) and (26) are
(34)
(35)
Let us introduce the two Dirac fields with 4 components (
)
From the properties (30), it is easy to show that
where in the last equality the inner produce between
and q is given in momentum space corresponding to the coarse lattice with lattice size 2
(36)
For any fixed
,
where in the second equality (21) and (22) are used. According to the properties of
and
in (30) (31) (32) and (33)
(37)
where we used the notations
and the summation over
is taken for all modes in (36). Similarly, we have (see Appendix B)
(38)
Using
and (37) (38), the kinetic part
can be rewritten as in the momentum space
(39)
where the summation over
is taken for all momentum mode of coarse lattice according to (36), and the staggered matrix in the momentum space is diagonal
(40)
where
and
depends on k. The inverse matrix of
is
(41)
where
(42)
We can calculate the trace of inverse matrix D in (20) from (39)
(43)
where the summation over
is given by (36). Note that the right hand side of (43) is real since
for any
and
modes in (36). Similarly,
(44)
and
(45)
The inverse matrix of D in (20) is
(46)
See Appendix C for the derivation of (44)-(46).
Since D is diagonal in momentum space, the inverse matrix in the
basis is
where the notation with tilde denotes the inverse Fourier transformation, e.g.,
for
,
. We first use the fast Fourier transformation to calculate
and thus
for
,
. Then
for
,
can be obtained since it is anti-periodic in
direction and periodic in
and
direction.
Each term in
has a tensor product
between
matrix
with
matrix
and
matrix B. The indices of
of the inverse matrix
in (46) is related to
. The analytic formula for the inverse matrix of the staggered fermion is the main contribution of this paper. Compared to the computational complexity
of the usual inverse matrix, the computational cost is
since each element of the inverse matrix needs the summation over
. Moreover a parallel implementation can be realized easily for the formula (46).
The trace of the inverse matrix in (43) can be derived from (46)
5. The 1 + 1d and 3 + 1d Staggered Fermion
The staggered fermion matrix in (20) can be generalized to the 1 + 1d and 3 + 1d case, where
is 1 for the 1 + 1d case and
run from 1 to 3 for the 3 + 1d case.
For the 1 + 1d case, the
matrices
are defined to be
The unitary transformation in (25) and (26) are modified to be
The kinetic part
can be written as
(47)
where the summation is taken over all modes
(48)
The fermion matrix in momentum space is diagonal
(49)
with its inverse
(50)
where
(51)
The trace of the inverse matrix is
(52)
The inverse matrix of D can be calculated
(53)
where
(54)
For the 3 + 1d case, the
matrices
are defined to be
The unitary transformation in (25) and (26) are modified to be
The kinetic part can also be written as (47) where the summation is taken for all modes
Equations (49) - (51) are still valid except that
runs from 1 to 3. Equations (52) - (54) are modified to be
(55)
(56)
(57)
respectively. We have checked the formula (46), (53), (56) for the inverse matrices by Matlab.
6. The Gap Equation
The main contribution of the effective action (18) to the partition function can be obtained by the gap equation if
,
(58)
Here D is defined in (20) where
is replaced by
. The right hand side of (58) can be calculated from (42), (43) where
is replaced by
. The first derivative of
with respect to
can be computed from the gap equation (For simplicity, we assume that
)
(59)
If the average
of
has been calculated from the gap equation, the free energy density in the large
limit is
where
up to a constant. The other thermodynamic quantities can be calculated. For example, the fermion density can be analytically calculated
(60)
where
, and two sums over
in (60) are given in (59), (44) and (45), respectively. The
for each mode
in (44), (45), (59) is given by (42) with the replacement of
by
(Here for simplicity we assume that
) and
is solved from the gap equation (58).
7. Simulation Results
7.1. Large Volume Limit
Let us consider the large volume limit for the non-interacting 2 + 1d Gross-Neveu model. The partition function
, where the stagger fermion matrix D is given by (20). The ratio of the non-dimensional chiral condensate
and non-dimensional mass
is
(61)
where in the last equality we used Equation (43) where
, depending on
and
, is given by (61). Note that there are
modes
in (61). The ratio of the non-dimensional fermion density
and
(62)
where in the last equality we used (44) and (45).
We consider the case
,
and thus
. We fix
and
and then calculate
and
in the large
limit for fixed lattice size
. In fact
and
does not depend on the lattice size
since the non-dimensional mass
and non-dimensional chemical potential
does not depends on lattice size
. Figure 1 shows the dependence of
on
with fixed
. The linear fitting with respect to 1/N shows that the large
limit of
is close to 1.008 for all four cases, this is because
and
both vanish for large
limit. Figure 2 shows the dependence of
on
, where
and
. The large
limit is close to 1.9271 for
and 1.9234 for
, respectively.
7.2. Phase Diagram
The phase diagram of the 2 + 1d Gross-Neveu model in the large
limit is well known [16] [17] [18]. In this limit the phase diagram of
is based on the calculation of
. Basically for
and
, there is a critical coupling
such that the chiral symmetry is broken
if the coupling is
Figure 1. The dependence of
on
,
. (1)
with fitting
, (2)
with fitting
, (3)
with fitting
, (4)
with fitting
.
Figure 2. The dependence of
on
,
. (1)
with fitting
, (2)
with fitting
.
strong enough
. This critical coupling depends on the regularization of the continuum model. For the lattice regularization in this paper,
where
is the lattice size. For fixed coupling
which is not far away from the critical coupling (Otherwise, the continuum limit
cannot be taken), denote
be the value of
at this coupling
with vanishing temperature T and chemical potential
. The gap Equation (8), which is solved exactly in the chiral limit in Ref. [18], shows that there exists a critical temperature
such that the chiral symmetry is broken if
at this coupling
and
. Moreover, there is another critical chemical potential
such that this symmetry is broken only if
at this coupling
and
. The mean field results predict that the first order transition only exists at
and
for this coupling
.
For the 2 + 1d Gross-Neveu model, we first study the dependence of
on the coupling g and temperature
with vanishing chemical potential
. Figure 3 is the phase diagram of
for
and
. We always choose
to ensure the thermodynamic limit is achieved: the simulation results change very small for larger
. The marks + separate the symmetry phase
(above marks) and the chiral symmetry broken phase
(below marks). For fixed temperature T there is a critical coupling
such that
decreases to zero if
is increasing to
. Figure 3 shows that
is a increasing function of
and it will close to 1 at very low temperature. On the other hand, if
is fixed, there is a critical temperature
such that
is increasing from zero if T is decreasing from
.
Figure 4 shows the dependence of
on
for the different coupling
. For small
, e.g.,
,
changes small with the temperature.
Figure 3. Phase diagram of
for
,
,
. Below the marks + is the broken phase
.
Figure 4.
versus
,
,
,
.
from top to bottom.
For these range of parameters, it is in the deep chiral symmetry broken phase and we cannot obtain the chiral symmetry phase
even at very high temperature. For a slightly larger
, for example,
(black dots in
Figure 4), we can find a transition point
, which is between
and
in lattice unit. The symmetry phase and broken phase are realized for
and
, respectively.
Figure 5 shows the dependence of
on
at different temperature.
drops continuously to 0 if
is increasing to
from below, which show that the transition at the critical coupling constant
is second order. At very low temperature
,
is close to 1, which is consistent with those obtained in [19]. This is because in the limit of
, the gap equation at
is reduced to
The critical temperature
at the coupling
and
can be
verified numerically. Here we choose
and
which is not too far away from the critical coupling
. We also choose
such that it is very close to zero temperature, the value of
at the zero temperature and vanishing chemical potential is
. To calculate the critical temperature at this coupling, we calculate
at
and found that
is zero if
is between 14 and 16. Thus the critial temperature is between
and
which is very close to
.
Now let us study the effect of chemical potential on the chiral condensate
. Figure 6 shows the dependence of
on the chemical potential at the different
Figure 5.
versus
for different
.
,
,
.
Figure 6.
versus
,
,
(
),
.
temperature
.
drops sharply near
in the limit of zero temperature
, i.e.,
, which suggest a first order transition at the zero temperature. This first order transition at the zero temperature is verified by the analytical calculation,
where
is the
with
[18]. For the temperature
,
is slightly larger than
. If the temperature is raised, e.g.,
, it is more difficult to find a critical chemical potential such that the chiral symmetry is restored. This is not caused by the smallness of
, since the our results is always obtained for
, which is very close to the thermodynamics limit, i.e., the result changes very small if
is larger than 36. We also note that the transition at finite temperature is the second order, as explained in [18]. Figure 7 shows the dependence of
on
for a larger
. Compared with Figure 6,
at
and the critical chemical potential in Figure 7 become smaller, and thus the figures in Figure 7 is obtained by moving those figures of Figure 6 in the left-down direction. For the same temperature, for example,
, it is more difficult to find the critical chemical potential in Figure 7 than those in Figure 6. Both Figure 6 and Figure 7 show that the critical chemical potential
is decreased if the temperature is increased. At zero temperature, the mean field exact result show the critical chemical potential
is just the value of
at the vanishing chemical potential. This is exactly recovered in Figure 7 where
for
with
.
Figure 8 shows the dependence of
and fermion density on the chemical potential at
. At low temperature
,
drops rapidly near the critical chemical potential
, and the fermion density increase very fast, which suggest
and fermion density are not continuous at
at zero temperature and thus they can be regarded as the order parameters.
For the 3 + 1d Gross-Neveu model, we also calculate the dependence of
on the coupling and chemical potential at different temperature. Figure 9 shows the value of
depending on the coupling for the vanishing chemical potential.
Figure 7.
versus
,
,
(
),
.
Figure 8.
and fermion density vs
,
,
(
),
.
Figure 9.
versus
for different
.
,
,
.
Compared to Figure 5 for the 2 + 1d model, the critical coupling becomes smaller. Moreover, the dependence of
on the temperature is less sensitive. Figure 10 shows the dependence of
on the chemical potential at the coupling
for the 2 + 1d and 3 + 1d Gross-Neveu model, the critical chemical potential is larger for the 2 + 1d model than those for the 3 + 1d model.
8. Conclusions
The staggered fermion for the Gross-Neveu model at finite density and temperature is revisited. In the large
limit, this model in 1 + 1d, 2 + 1d and 3 + 1d dimension can be easily solved in momentum space. Moreover, an explicit formula for the inverse matrix for the 1 + 1d, 2 + 1d and 3 + 1d staggered fermion matrix is found, which can be implemented by parallelization. This formula can also be generalized to the other space dimensions. For the odd space dimension,
Figure 10.
versus
,
, (
),
. Left (3 + 1d), Right (2 + 1d).
the orthogonal transformation was found [33]. The key point to find the explicit formula for the inverse matrix is to use the properties of
and
as shown in Section 4. These properties for the even number of space dimension are simpler, as shown in the supplement material.
The dependence of chiral condensate and fermion density on the coupling, temperature and chemical potential are obtained by solving the gap equation. Our results for the 2 + 1d case reproduce the analytical results. We also compare the chiral condensate for the 2 + 1d and 3 + 1d case in the same range of parameters, showing that the reason for symmetry breaking and restoration can be explained by the suitable choice of the coupling, temperature and chemical potential.
Acknowledgements
Daming Li was supported by the National Science Foundation of China (No. 11271258, 11971309).
Appendix A. Proof of Properties of
and
The notations for
in (27) is a little awkward. I replace
,
and
in (27) by
,
and
, respectively. Thus
(A1)
The three Pauli matrices
satisfies the completeness relation
(A2)
We first have
by (A2) (A3)
which is also valid if
is replaced by
or
. Secondly,
Inserting
in the above equality, we have
.
where in the last equality we used
To prove that
(A4)
we want to prove that
i.e.,
This is obvious since the left hand side is
, by (A3) if
(A5)
Similarly, (A4) is also valid if
and -2 are replaced by
and +2, respectively. This is because the sign
in (A5) is replaced by
. Obviously,
For example,
,
Finally, we have
since the left hand side is
where we used
Here the we define
.
Appendix B: The Derivation of (38)
The derivation of (38) is similar to the calculation of
.
where
In the fourth equality, we used the formula like
Appendix C. The Derivation of (44)-(46)
First,
by (38) (39)
by (41)
Similarly,
The inverse matrix of D in (20) can be calculated as follows