Quantum-Criticality in Dissipative Quantum Two-Dimensional XYand Ashkin-Teller Models: Application to the Cuprates

# Quantum-Criticality in Dissipative Quantum Two-Dimensional XY and Ashkin-Teller Models: Application to the Cuprates

## Abstract

In a recent paper(1) we have shown that the dissipation driven quantum phase transition of the 2D xy model represents a universality class where the correlations at criticality is local in space and power law in time. Here we provide a detailed analysis of the model. The local criticality is brought about by the decoupling of infrared singularities in space and time. The former leads to a Kosterlitz Thouless transition whereby the excitations of the transverse component of the velocity field (vortices) unbind in space. The latter on the other hand leads to a transition among excitations (warps) in the longitudinal component of the velocity field, which unbind in time. The quantum Ashkin-Teller model, with which the observed loop order in the Cuprates is described maps in the critical regime to the quantum xy model. We also discuss other models which are expected to have similar properties.

The dissipative quantum 2D xy model was introduced (2); (3) to describe experiments on ultrathin granular Superconducting films, where it was observed that above a normal-state sheet resistance of order the resistivity does not decrease towards even at the lowest temperatures studied. (4). The granular superconductor is represented by an array of islands with finite superfluid density with Josephson coupling among the nearest neighbors. Dissipation of the Caldeira-Leggett (5) form is included, whereby the phase difference between the nearest neighbors is coupled to an external bosonic bath. the external bath effectively act as a resistor, , shunting nearest neighbor islands and energy is dissipated at the rate , where is the voltage drop which is proportional to the rate of change of the phase difference between the grains. A quantum phase transition occurs as the value of the resistance is tuned. For small values of the resistance, phase slips, whereby a voltage develops across the junction between two grains, cost too much energy and so lead to an ordered phase. On the other hand, for large value of the resistance, phase slip events are energetically favored and global superconductivity is destroyed.

In the absence of dissipation, the quantum model at maps to a 3d xy model. It was suggested that the physics in the presence of dissipation may not be continuously connected to the 3d xy universality class (6). It was also suggested that there is a possibility that the singularity introduced by dissipation may be linked to a transition driven by the unbinding in time of phase slips (7). The calculations are done in the weak coupling regime where one considers the Josephson coupling perturbatively. The same approach was previously used to study the phase transition and correlation functions (2); (7); (8). Arguments were given also that the critical fluctuations due to the unbinding of the phase slips in time which have long-time singularities may be local in space (2). However later work by some of the same authors presents a rather different picture (9). Moreover none of these investigations into the dissipative transition are applicable near the critical point.

In a previous paper we have shown that the the quantum phase transition is obtained due to the proliferation of a new kind of excitation, which we termed warps, whose physics is indeed related to the phase slips in time; more specifically they represent events in time where a change in the longitudinal component of the velocity field(1) occur. The possibility that the longitudinal component may lead to the introduction of sources and sinks which control the critical behavior was also recently emphasized in the context of one dimensional systems (10); (11). The correct identification of the relevant degrees of freedom in higher dimensions turns out to be interesting. The introduction of warps simplifies the analysis of the problem so that the correlation function of the order parameter can be obtained near the critical point. We find spatially local correlations with an infrared divergence () in temporal correlations at the critical point. We also the cross-over from the quantum critical to the quantum-regime.

The motivation for this study is that the existence of such a critical point naturally explains some of the most important features of the phase diagram of the Cuprate superconductors. Polarized neutron scattering (12); (13); (14); (15) in three distinct family of Cuprates have confirmed previous finding in a third family of Cuprates through dichroism in Angle Resolved Photoemission spectra (16), which reported the observation of time reversal violation in the pseudogapped phase of underdoped cuprates. In particular, the experiments lend support to the idea that the ground state in the pseudogapped regime is one with ordered current loops without loss of translation symmetry. (17); (18); (19). The low energy sector of this theory contains four states that are related by time reversal and reflection along the or axis. Alternatively, the four states can be represented by two Ising variable in each unit cell, the values representing currents in the and directions. A classical statistical mechanical model for such degrees of freedom that supports a phase transition without any divergence in the specific heat (20); (21) is the Ashkin-Teller model (22). Such a model is known to undergo a Gaussian phase transition (23). A quantum generalization of the classical model in terms of operators that induce fluctuations within the low energy phase space of the four states, leads to the dissipative 2D xy model studied here. The local criticality with a singularity is the fluctuation spectrum necessary to explain the anomalous properties in a number of properties such a resistivity, nuclear relaxation rate and optical conductivity in the normal state near optimal doping (24) abutting the pseudogapped phase. At long wavelengths, such correlations are directly observed by Raman scattering experiments. Recently, we have also shown that such fluctuations couple to fermions to promote pairing in the -symmetry channel. (25)

It is worth pointing out that a number of classical statistical mechanical models such as the 6-vertex and 8-vertex models also have Gaussian criticality in part of their phase diagram, as well as phase transitions which do not manifest any anomaly in their specific heat. Such models when augmented with ohmic dissipation do belong to the same universality class considered here. Local quantum criticality happens also to be a hallmark of the quantum phase transitions in heavy fermions (26); (27), although it is not clear to us how the class of models solved here may be connected to the microscopic models for heavy fermions.

This paper is organized as follows: The dissipative quantum XY model and the Villain transformation are discussed in sec II. In section III, we show how dissipation changes the critical properties of the model. We show that the action can be written in a quadratic form in terms of two orthogonal topological excitations, vortices which interact only in space and warps which interact only in time. We also discuss the difficulties in analyzing the model based on other choice of variables. In sec IV we discuss the properties of the warps. Sec V deals with the scaling equations for the vortices and warps and the phase transition which occurs through the proliferation of warps. In sec VI we present a detailed calculation of the fluctuation spectra near the quantum critical point. The spectra has the same form as that proposed to explain the normal state marginal fermi-liquid properties of the high cuprates. In sec VII we present the zero temperature phase diagram. The connection to Ashkin Teller model and cuprates is presented in sec VIII. The effect of fourfold anisotropy on the critical properties is anayzed in sec IX. In sec X we show how dissipation arises in cuprates. In sec XI we discuss other forms of dissipation and how they may lead to different forms of singularities in the fluctuation spectra.

## I Dissipative Quantum 2DXY Model

The classical 2d xy model consists of degrees of freedom, represented by an angle , living on the sites of a regular lattice, assumed to be a square lattice here, with a nearest neighbor interaction of the form

 H=J∑⟨ij⟩[1−cos(θi−θj)], (1)

where is Josephson coupling. Since a continuous symmetry cannot be spontaneously broken in two dimensions (28); (29), this model does not support a long range ordered phase. Nevertheless a phase transition does occur at finite temperature where the correlation function of the order parameter changes from exponential to power law. This is the Kosterlitz-Thouless-Berezinskii transition (30); (31). The physics of this phase transition is better understood in terms of the topological defects of the system. To do so we follow the standard procedure of using the Villain transform and integrating out the phase degrees of freedom. We include the algebra here so as to make the later discussion of the quantum version easier to follow.

The Villain transformation expresses the periodic function in terms of a periodic Gaussian:

 exp⎡⎣−βJ∑⟨ij⟩[1−cos(θi−θj)]⎤⎦≈∑mijexp⎡⎣−βJ∑⟨ij⟩(θi−θj−2πmij)2/2⎤⎦, (2)

where are integers that live on the links of the original square lattice. On a square lattice we choose to identify each site as where and are integers. We can combine the two link variables and into one two component vector that lives on the site of the lattice. The choice is purely for convenience and none of the results depend on how one chooses to organize the degrees of freedom of the model. Instead of introducing a new field to linearize the quadratic term via the Hubbard-Stratonovich transformation, we follow an alternative procedure. We expand the quadratic term and transform to Fourier space. Keeping the leading quadratic term , where a is the lattice constant we get

 (θxy−θx+1,y−2πmxx,y)2≈a2∇2xθxy + 4πa∇xθxymxx,y (3) + 4π2mx2x,y,

where is the component of the vector field given by the integer . Fourier transforming and integrating out the field in terms of which the action is quadratic we get

 Z≈∑mexp⎡⎣−4βJπ2∫dk⎡⎣k2m⋅mk−(k⋅mk)2k2⎤⎦⎤⎦. (4)

Combining the two terms and Fourier transforming to real space we get

 Z≈∑mexp[−βJ∑r∑r′(∇×mr)⋅(∇×mr′)log∣∣r−r′∣∣]. (5)

Since the vector field is two dimensional the curl is a scalar and is the vorticity of the vector field. Defining integer charges we get the standard action of the two dimensional coloumb gas. On the lattice this definition is equivalent to

 ρx,y = mx,y+1;x+1,y+1−mx,y;x+1,y (6) + mx+1,y;x+1,y+1−mx,y;x,y+1.

In this model, a phase transition occurs from a state at high temperatures where the vortices are free to a low temperature state where they are bound in pairs.

We now turn to the description of the Quantum version of the 2d xy model. First we consider the model in the absence of dissipation. The quantum 2d xy model is written in terms of operators and its conjugate, the number operator . The Hamiltonian is of the form

 H=∑in2i2C−J∑⟨ij⟩cos(θi−θj), (7)

where is the capacitance and . Since , the first term is the kinetic energy of the phase with a mass . There is competition between the kinetic energy and the potential energy terms, the former minimized by a state where is disordered stabilizing an insulating phase while the latter minimized by a fixed value of stabilizing a superconductor. The partition function can be recast in the form

 Z=∫Dθi(τ)exp⎡⎣−∫β0dτ⎛⎝∑iC2(∂τθ2i)−J∑⟨ij⟩cos(θi−θj)⎞⎠⎤⎦ (8)

.

At , the (imaginary) time direction becomes infinite in extent, and the model is in the 3d xy universality class. The dualization procedure used for the classical 2d xy model can now be implemented on the 3d xy action as well. First we discretize the imaginary time direction in units of and work on a three dimensional lattice. Introducing the variables which only live on the spatial links, the action is

 Z ≈ ∑mexp⎡⎣∑k,ω−4π2JΔτm⋅m (9) + 4π2J2ca2Δτ2(k⋅m)2(Ca2/cΔτ)ω2+Jca2Δτk2⎤⎦,

where . Rearranging terms as before we get

 Z=∑mexp⎡⎣∑k,ω−4π2J⎛⎝Jc∣∣k×m∣∣2+(C/c)ω2∣∣m∣∣2Jck2+(C/c)ω2⎞⎠⎤⎦, (10)

where , and . The isotropic model is recovered when the discretization in time is such that . Combining (which is a scalar since m is two dimensional) and into one three dimensional field we see that eqn.10 is the partition function for the three dimensional loop gas model which is in the same universality class as the 3d xy model(32). A point to note here is that a vector field coupling via a three dimensional kernel does undergo a phase transition while a scalar field does not (33). Alternatively, the loop gas model undergoes a 3d xy transition while the three dimensional Coloumb gas is always in a disordered phase.

Alternate derivations (34) map the model directly to the 3d xy model instead of the loop gas model. We find that the procedure described above is preferable in the presence of dissipation.

In the dissipative 2d xy model, one adds to eqn.8 a term which generates in the equations of motion of the phase variable a term proportional to the velocity, . The dissipative part of the action is of the form (35)

 Missing or unrecognized delimiter for \left (11)

where where . For a single Josephson junction the voltage drop is given by . The power dissipated by a resistor connecting the two superconducting grains is . Integrating over time one obtains total energy dissipated through the resistor. For an array of junctions the corresponding term in the action is of the form in eqn.11. Such a dissipative term can also be derived by assuming that the phase difference across every junction is coupled to an external bath of harmonic oscillators. Integrating out the bath the dissipative term is obtained in Fourier space as

 Sdiss=α∑ωn∑⟨ij⟩|ωn|η∣∣θi,ωn−θj,ωn∣∣2. (12)

The exponent reflects the form of the spectral density of the external oscillator bath at low frequencies. For a linear spectral density and a Ohmic dissipation is obtained (5). It is indeed interesting to ask what other forms of dissipation is allowed. We will postpone the discussion of the effect of such dissipative terms to section X.

## Ii Excitations of the Quantum 2D XY Model

We can now carry through the program of introducing the vector field m for the quantum 2d xy model with Ohmic dissipation (). The integration of the fields can be performed since the dissipative term is also quadratic. The effective model is now written as

 Z = ∑mexp⎡⎣∑k,ω−4π2Jm⋅m+4π2J2c(k⋅m)2(C/c)ω2n+Jck2+α|ωn|k2⎤⎦ (13) = ∑mexp⎡⎣∑k,ω−4π2JJc∣∣k×m∣∣2(C/c)ω2n+Jck2+α|ωn|k2−4π2Jω2n%m⋅m(C/c+αk2/|ωn|)(C/c)ω2n+Jck2+α|ωn|k2⎤⎦,

where we have redefined . The action in Eqn.13 has two possible phase transitions, depending on whether the capacitance or the dissipation term in the numerator of the second term in Eq. (13) dominates in long wavelength and low frequency limit. The former corresponds a critical point with the dynamic critical , i.e. we recover the loop gas model which belongs to the 3d xy universality class. The latter corresponds, as we will show to , i.e. a fixed point with local criticality. The two terms have the same scaling form for . In the rest of the paper we analyze the dissipation driven transition

The vector field m can be written as a sum of a longitudinal, , and a transverse component, which by definition satisfy and . The vorticity field is related to the transverse component alone and is given by

 →∇×mt(r,τ)=ρv(r,τ)=ρvc∑iδ(r−%ri)δ(τ−τi), (14)

where are the location of the core of the vortices and is the quantized charge of the vortex.

Unlike the theory where is the two component current that acts as the additional topological defect, we find that the topological excitation in terms of which the problem with dissipation is most conveniently discussed involves only the longitudinal component and is given by a ”warp” field defined through a non-local relation in space,

 dml(r,τ)dτ=∫d2Rρw(R,τ)r−R∣∣% r−R∣∣3. (15)

are events where the longitudinal component changes (see Fig.1). The warp charge is defined by

 ρw(R,τ)=c2πρwc∑iδ(R−Ri)δ(τ−τi), (16)

in a way analogous to the definition of the vortex charge above. We will show in Sec. (III) that the warp-charge is also quantized.

Given Eq.(15), the relation between the Fourier-transformed warp field and the longitudinal component of the velocity field is

 ıωnckk⋅ml(ωn,τ)=ρw(k,ωn) (17)

In terms of , the action, in the continuum limit, neatly splits into three parts:

 Sv = 1L2β∑Jk2∣∣ρv(kωn)∣∣2 (18) Sw = 1L2β∑α4π|ω|∣∣ρw(kωn)∣∣2 S′w = Missing or unrecognized delimiter for \left ×∣∣ρw(kωn)∣∣2.

where

 G(k,ωn)=1Jck2+(C/c)ω2n+α|ωn|k2 (19)

The factorization of the action in eqn.18 is only possible due to the choice of variables made in eqn.17. With this choice, it follows from Eq.(18) there are logarithmically interacting vortex charges in space with local interactions in time, and a logarithmically interacting warps in time with local interactions in space. The former is the standard Kosterlitz-Thouless-Berezinskii theory (31); (30) while the latter is the charge representation of the Kondo problem (36) near the Ferromagnet/Anti-Ferromagnet quantum critical point. Quite unlike the theory where Lorentz invariance is manifest, the space and time degrees of freedom are completely decoupled in the theory describing the dissipation driven transition. Being orthogonal the charges and are uncoupled; the action is a product of the action over configurations of and of . Any physical correlations are determined by correlations of both charges. The third piece of the action is nonsingular because it corresponds to the three dimensional Coloumb gas problem which is known to be disordered at all temperatures (33).

The singularity induced by the presence of Ohmic dissipation is transparent in our formulation. Suppose, one works instead with the degrees of freedom appropriate to the 2+1 d xy model which are the vortex density and the vortex current, . The latter is defined as

 Jv=^z×dmdτ (20)

The action in eqn.13 is

 Z=∑mexp⎡⎣∑k,ω − 4π2JJc∣∣ρv(kω)∣∣2(C/c)ω2n+Jck2+α|ωn|k2−4π2JJv⋅Jv(C/c)(C/c)ω2n+Jck2+α|ωn|k2 (21) − 4π2JJv⋅Jv(αk2/|ωn|)(C/c)ω2n+Jck2+α|ωn|k2],

The first two terms can be combined to give the dual representation of the model. In the absence of dissipation, these terms lead to the superconductor to insulator phase transition. The third term introduces a new singularity in the vortex currents. Due to the continuity equation between and , see below, the two singularities cannot be decoupled as before and the analysis of the model is not straightforward.

In terms of warps and vortices the vector field can be written as,

 m=ρv^z×ık/k2+ρwc^k/ıω (22)

and

 Jv=−ρvωk/k2+ρw^z×c^k (23)

Separating the longitudinal and transverse components of the vector field m and the vector current allows us to factorize the action. Notice that the new term in the vector current is transverse and the continuity equation. is satisfied.

Since warps are excitation that introduce longitudinal components in the vector field, alternate formulations could have been considered. In particular, we could have chosen to define monopoles (sources and sinks in two dimensions) which are of the form

 →∇⋅m=ρm (24)

Such excitations determine the longitudinal component of the vector field and their relation to warps in momentum space is,

 ρm(ω,k)=ckωρw(ω,k). (25)

In terms of monopoles one cannot separate the singularities and the simple description of the dissipation driven transition is no longer possible. The physics remains same as the scaling dimension of monopoles and warps is the same. We choose to work with warps as the analysis is more transparent and we do not have to consider the added complication of the renormalization of velocity.

## Iii Warps as topological defects in space time

In this section we discuss the topological properties and the physical content of the warps. The definitions in Eqs. (17) and (15) say that the warps are source for the longitudinal component of the vector field m. Since m is an integer vector field, it is important also to show that this property is preserved by Eq. (15) as well as that the warp-charge is quantized.

Let us first review the familiar case of the vortex in Eqn.26. In real space the vector field due to a vortex at is

 mt(r,τ)=ρvcˆz×(r−Ri)∣∣r−Ri∣∣2 (26)

Given the integer vector field m the vortex is defined by its quantized charge and the singular azimuthal vector field . This vector field is undefined inside a core radius where the source of the velocity field with discrete charge sits. The discreteness of is respected once such a singularity is identified. In particular, the corresponding field has a discontinuity along a line going from the core out to infinity. One can then proceed to the continuum to analyze the critical properties of the model .

We see from the solution of Eq. (15) that the field due to a warp created at site and time has the form:

 ml(r,τ)=cρwcΘ(τ−τi)r−Ri|r−Ri|3 (27)

Analogous to the vortex, this solution is a quantized charge in a core which serves as the source of a radial velocity field falling off as . To see this consider the solution to eq.17 with ,

 θ(r,τ) = Θ(τ−τi)(2πna∣∣% r−Ri∣∣+θc),∣∣r−Ri∣∣≥a (28) = Θ(τ−τi)θc,∣∣r−Ri∣∣

The boundary conditions chosen are and . The first guarantees the quantization of the warp charge, imposed by the phase jump of that occurs at . In other words a warp is an event which produces a discrete phase jump across the core. The corresponding discontinuity in the phase is discrete, corresponding to a discrete field localized along a loop encircling the core. This localized discrete serves as the source of the radial field outside the core. As one moves along any radial direction crossing this loop, which defines the boundary of the core, the phase jumps by . Thus the warp charge is discrete and satisfies the constraint that the vector field m is discrete. The second boundary condition where is chosen because we are only describing the changes with respect to the fields because of the warp event at .

## Iv Dissipation driven phase transition

The action for the warps, Eq. (18), has been studied before in the context of quantum coherence of two state system coupled to dissipative baths (37); (38). Let us introduce a core-energy for the ’s just as is done to control the fugacity of ’s, the vortices. Next consider how the renormalization of and proceeds. (39). Including the core-energy, the action is

 Sw=∑i[T∑nα4π1|ωn||ρwi(ωn)|2+∫dτΔ|ρwi(τ)|2]. (29)

The RG equations for eqn.29 are well known (36); (37); (38); (40)

 d˜Δdl = (1−α)˜Δ (30) dαdl = −α˜Δ2,

where and is the short time cutoff. The critical point of interest is at , where scales to 0; for the charges freely proliferate as ”screening” due to becomes effective. represents the ordered or confined region. We are interested here only in the region . Well in the quantum critical region, the (singular part of the) propagator for is

 ⟨ρw(ωn)ρw(−ωn)⟩=(1/4π|ωn|τc)−1. (31)

Note that the form of the propagator in eqn.31 is valid only at the critical point. The crossover to the quantum-disordered or screened state is given for when which is of the order of the inverse of the characteristic screening time, which may be estimated similarly to Kosterlitz’s estimate (41) of the screening length in the -problem:

 ωx≈τ−1cexp(−b/√1−α), (32)

where b is a numerical constant of O(1). At finite temperatures and low frequencies, the crossover temperature is of O(). On the ordered side , we have to consider the effect of the charging energy and the flow to the critical point. In fig.2 we plot the phase diagram in the plane where only the the region is correctly described. The crossover line is also shown where well above we expect to observe critical behavior controlled by the quantum critical point while below the system is phase disordered and hence an insulator.

## V Correlation Function

We can compute the correlation function for the order parameter field following a procedure which is a generalization of the method developed for the classical 2DXY model (23). We are interested in the expectation value of given by

 Ci,j,μ,ν=⟨eıθiμe−ıθjν⟩≡∫d[θ]e−¯S∫d[θ]e−S, (33)
 C=⟨eıθiμ−ıθjν−2πı∑pathm⟩ (34)

Notice that the exponential of the sum along the path is always one as m is an integer fields. We need to include it in our calculations to reproduce the correlation functions in the absence of dissipation. Since m is two dimensional vector, the contribution is only from the space like segment of the path chosen. We can now proceed as before. Going over to imaginary time implies that the factor of is absorbed in the definition of the path integral.

To compute the correlation function we introduce a field that lives on the sites of the space time lattice. Its and component are or depending on whether or not the path includes the link to or respectively. Similarly we introduce a field which is or depending on whether or not the path includes the link to . The path we choose to compute the correlation function is shown in fig.3. We can now integrate out the phase degrees of freedom as before to obtain the functional in terms of the vector fields alone.

 ¯S=Sv + Sw (35) + 1L2β∑ln(2π)JG(kl,ωn)[(−ıωnη0⋆/2)(ıkl⋅mln) +(−ıkl×→η⋆)⋅(ıkl×mln)+c.c. +((C/c)ω2n+α|ωn|k2)(m⋅→η) +∣∣ωnη0/2+kl⋅→η⋆/2∣∣2].

Given the definition of vortex and warps, we can write the vector field as

 m=ρvık×ˆzk2+ρwcıωˆk (36)

Replacing in the expression above we see that the vortex part of the correlation simplifies just as it did in the case of the action itself.

 ¯S=Sv + Sw (37) + Missing or unrecognized delimiter for \right +((C)+cαk2/|ω|)klρw(k⋅→η)+c.c.} +∣∣ωnη0/2+kl⋅→η⋆/2∣∣2] +{2πJk2(−ıkl×→η⋆)⋅ρv+cc.}

The net correlation is a product of three distinct contributions. The term independent of vortices and warps is the spin wave contribution. Notice that the associated propagator is three dimensional and as such is non singular. This reproduces the well known result that spin waves do not disorder the long range order in three dimensions. The other two contributions come from vortices (the last line in eqn.37) and warps. At the dissipative critical point, the singular fluctuations are associated with the warps. Thus the vortices and spin wave contributions are completely regular. To compute the contribution from the warps we follow a procedure similar to Jose et al. ((23)). Consider the contribution from the warps,

 ¯Sw = Sw+1L2β∑ln(2π)JG(kl,ωn)[{(−ıωnη0⋆/2)ckωρw (38) +(C+αck2/|ω|)klρw(k⋅→η)+c.c.}]

In real space the terms of the action can be rewritten as

 ¯Sw = Sw+∫dRdR′dτdτ′[∂r→η(R,τ) (39) G(Rτ;R′τ′)[(C+αck2/|ω|)ıklρw](R′τ′) + ∂τη0(R,τ)G(Rτ;R′τ′)[ckωρw](R′τ′)],

where the derivatives are taken along the path and the term is meant to represent the fourier transform of the terms within the square brackets. Consider the derivatives along the path of the fields and . They represent the change in value of the fields computed from one point to the next along the chosen path. Since the first term in eqn.39 couples only to the spatial derivative it has finite values only at and (ends of the space like component of the path) while the second term couples to time derivatives which are finite at and .

 ¯Sw = Sw+∫dRdτ1[ckωρw](Rτ1) (40) {G(r′τ,Rτ1)−G(r′τ′,Rτ1)} + ∫dRdτ1[(C+αck2/|ωn|)klρw](Rτ1) {G(rτ,Rτ1)−G(r′τ,Rτ1)}

Fourier transforming back we get

 ¯S=Sv + Sw (41) + 1L2β∑ln(2π)JG2(kl,ωn)[(eıkl⋅(rj)+ıωn(τμ) − eıkl⋅(rj)+ıωn(τν))(ckωρw)+(eıkl⋅(ri)+ıωn(τμ) − eıkl⋅(rj)+ıωn(τμ))((C+αck2/|ωn|)klρw)]

Since the action is quadratic, we can compute the correlation to give

 C(r−r′,τ−τ′)∝exp(−F), (42) F=−J2c24T∑n∫dk⟨ρw(ωn)ρw(−ωn)⟩× G2(k,ωn)[k2ω2n(1−cos(ωn(τ−τ′))) +(((C/c)+αk2/|ωn|)kl)2(1−cos(kl⋅(ri−rj))) +(((C/c)+αk2/|ωn|)ωn)(eıkl⋅(ri−rj)+eıωn(τμ−τν) −eıkl⋅(ri−rj)+ıωn(τμ−τν)−1)]

The order parameter correlation function has three contributions from the warps. The first term in the square brackets in eqn.42 is from the time like segment of the path; the second term corresponds to the space like segment of the path and the last term is an interference term between the two. At the critical point the correlation of the warps is given by eqn.31. Consider the space like segment. It is singular (of the form ) as goes to zero. This implies and the correlation function is zero unless . Thus the correlation function is local. For the last two terms are zero and only the time like path contributes to the correlation function. Performing the integral over momentum k, the correlation function is given by , where

 F(0,τ−τ′)=−2πT∑n1−cos(ωn(τ−τ′))|ωn|log(|ωn|τc). (43)

To perform the matsubara sum we consider the analytic properties of the logarithm in eqn.43. The branch cut introduced on the real axis implies that on integrating over a contour infinitesimally above the real axis from to and infinitesimally below the real axis from to the real part cancel while the imaginary part enforces a sum over positive frequencies giving

 F(0,τ−τ′)=−2πT∞∑n=11−cos(ωn(τ−τ′))|ωn|. (44)

Such correlation functions have been calculated in other contexts. In particular using the calculation by Ghaemi et al. (42), the spectral function of the correlation is

 Imχ(q,ω)=¯τctanh(ω/2T),  ω≲τ−1c. (45)

is . The order parameter correlation at is local in space and power law in time. Thus the dissipative 2DXY model describes a new type of quantum criticality where the correlations are local in space and are functions of , with a cut-off provided above. Precisely such criticality had been postulated to describe the ”Strange metal” or marginal fermi-liquid region of the phase diagram of high temperature superconductors (24).

## Vi Zero Temperature Phase Diagram

So far we have concentrated on the dissipation driven transition. We now consider the phase diagram in the plane. Our analysis clearly shows that the physics at finite dissipation and at zero dissipation cannot be continuously connected. This is in accord with the suggestion in Ref.( (6)), that the zero dissipation limit, described by the model at zero temperature, is in a different universality class compared to the model at infinitesimal dissipation. The other limit where the physics is easy to analyze is at infinite dissipation. In this case, the fluctuations in time are energetically unfavored and the model maps to a model. We now make a guess on the structure of the rest of the phase diagram based on the well-defined theory in the limits discussed.

In reference [6] it is argued that a weak coupling analysis shows that the entire Josephson coupling renormalizes to zero at small dissipation. Furthermore a self consistent calculation allows one to map the phase boundary within this scheme. Such an analysis does not capture the physics of topological defects that are necessary to determine the entire phase diagram. Given our identification of the relevant degrees of freedom and the fugacities associated with the new excitations, we propose an alternative understanding of the physics of the dissipative model.

We have explicitly shown above that the critical theory has two independent singularities. Thus the true phase diagram can be determined by studying the flow of stiffness, fugacity of vortices and fugacity of warps. The only algebraically ordered phase is one in which both the vortices and warps are bound. This occurs for and . Furthermore this phase is continuously connected to the 2d XY phase at infinite dissipation. Thus one expects a phase boundary to exist that starts at and and approaches and asymptotically as shown in fig.4. For the fugacity of warps grows and the stiffness vanishes. Similarly, for the vortices proliferate driving the stiffness to zero. We find no evidence for the ordered phase for small and as in reference [6] where the large fugacity of the vortices was ignored. Similarly, for large a minimum value of is required to overcome the fugacity of the warps. In other words for any given value of as one increases the fugacity of warps, which is infinite at infinitesimal , remains large until we cross the phase boundary, upon which it renormalizes to zero. Whether the resulting phase is algebraically ordered or not depends on the fugacity of vortices.

Consider the regime where warps proliferate. The vanishing stiffness implies that such excitations should also lead to screening in the interaction of the vortices. In analogy with the disordered phase in the KT theory, we should introduce a mass for both vortices and warps. This in turn will lead to unbinding of vortices as well. We believe that for small , this proliferation of warps and resultant screening of interaction among vortices will push the phase boundary away from to larger values of . At large we expect our theory to be valid and a dissipation driven transition will occur at . A schematic phase diagram in shown in fig.4.

At first sight it might appear that a long range interaction in time of the form is inconsistent with a disordered ground state, where the correlation function decays faster than the interaction. The reason such a state is allowed is the fact that the existence of excitations which correspond to phase slip events in time introduces a second time scale in the problem. The competition between the short time scale physics associated with the fugacity of the warps and the long range interaction between them is precisely the reason for the observed phase transition.

## Vii Ashkin Teller model and the Cuprates

In reference [1] we have argued that the long wavelength theory of a model for the Pseudogap state for the Cuprate superconductors is the Ashkin-Teller model, The quantum version of this model with dissipation is in the universality class of dissipative quantum model. The Marginal Fermi Liquid (24) properties of the optimally doped system originate from the fluctuation spectrum of the local quantum critical point. In this model the phase variable represents the possible current loop order within a given unit cell and the four fold anisotropy restricts it to four values representing states that break time reversal and one of two possible reflection symmetries. The origin of the dissipation in this system is due to the coupling of the order parameter, which corresponds to the coherent part of the current operator, to the incoherent part of the current operator which correspond to fermions near the Fermi surface. For long-wavelength fluctuations such a coupling leads to Ohmic dissipation of the form in eqn.11, while term of the form in eqn.56(see below) is not generated. Detailed derivation of the long wavelength theory will be provided in a later publication.

The classical Ashkin Teller model is defined in terms of two Ising degrees of freedom per unit cell. If we define and as Ising variables taking values to parameterize the currents along the the and direction, the four resulting states in each unit-cell (see fig.5) form the basis of the low energy sector of the long wavelength theory.

The Hamiltonian of the classical Ashkin Teller model is (22)

 HAT=−∑[J2(σiσj+τiτj)+J4σiσjτiτj] (46)

For the model reduces to two decoupled Ising models which is known to undergo a Gaussian phase transition (43). The analysis has been extended to map out the entire phase diagram by observing that is a marginal operator in this theory (44). For the system undergoes a Gaussian phase transition form a ferromagnetically ordered state to a paramagnetic state with continuously varying exponents. Quite remarkable, for , the specific heat exponent is negative. The Ashkin Teller model explains why no singularity is observed in specific heat measurements at the pseudogap temperature accompanying the breaking of time reversal symmetry. Defining and we can rewrite the Hamiltonian as

 HAT=−∑[2J2cos(θi−θj)+J4cos(2θi−2θj)] (47)

where ’s take values and . The constraint can be implemented as a four fold anisotropy term,

 HAT= − ∑[2J2cos(θi−θj)+J4cos(2θi−2θj)] (48) + ∑ih4cos(4θi)

Given this mapping of the Ashkin Teller model to a xy model with four fold anisotropy, we now analyze the effect of on the local quantum critical point of the dissipative 2dxy model.

## Viii Effect of four fold anisotropy

The discrete nature of the underlying degrees of freedom of the Ashkin Teller model enforce an Ising type order at low temperatures. On increasing temperature, it undergoes a Gaussian transition into a state which can be described by proliferation of vortices. It is a unique property of four fold anisotropy that the disordering, which occurs via proliferation of vortices, and ordering, which is enforced by a diverging anisotropy field, happen at the same temperature (23). We now analyze the effect of such an anisotropy field on the local quantum critical point. To do so we introduce the term in the Hamiltonian. To handle such a term in the action we employ the following approximation scheme

 eh4cos(4θi)≈∑pieln(y4)p2i+ı4piθi, (49)

where is an integer field that lives at each site and . For large values of the approximation is reliable as the sum will be dominated by the terms with . Eq.49 introduces two new terms in the action, one linear in the phase and the other independent of it. We can proceed as before in integrating out the phase degrees of freedom to get a new action which is of the form

 S=Sv+Sw + 1L2β∑Gc2(4π2Jkωρw(k,ω)p(−k,−ω) (50) + ∣∣p(k,ω)∣∣2)+ln(y4)∣∣p(k,ω)∣∣2.

Given that the coupling between the integer field and the phase variable is linear we can follow a self consistent procedure to calculate its effect on the dissipative fixed point. In this context it is useful to recall that in the absence of dissipation, the disordered side is characterized by the anisotropy field going to zero, i.e. being marginally irrelevant. Furthermore, in the absence of anisotropy the dissipative transition leads to a state which is characterized by the proliferation of warps. In other words, the most singular part of the action is the interaction between warps in time. Since the action for the integer field, , is quadratic, the cumulant expansion implies that the a new term in the action for warps appears of the form

 ˜S4 = 12L2β∑(4π2JGc2kω)2ρw(k,ω)ρw(−k,−ω)× (51) ⟨p(k,ω)p(−k,−ω)⟩.

The correlation is obtained from a self consistent solution of the coupled equations shown diagrammatically represented in Fig.6.

A quadratic equation is obtained for each of the propagators. Since we are looking at the effect of the anisotropy on the effective action for the warps, we can solve these equations to obtain the most singular contribution to the propagator in Eqn.51. Defining and , symbolically the equation satisfied by can be written as

 λ2G0wG2p−Gp+G0p=0. (52)

where and