Individual complex Dirac eigenvalue distributions from random matrix theory and comparison to quenched lattice QCD with a quark chemical potential

Individual complex Dirac eigenvalue distributions from random matrix theory and comparison to quenched lattice QCD with a quark chemical potential

G. Akemann, J. Bloch, L. Shifrin, and T. Wettig Department of Mathematical Sciences & BURSt Research Centre, Brunel University West London, Uxbridge UB8 3PH, United Kingdom
Institute for Theoretical Physics, University of Regensburg, 93040 Regensburg, Germany
November 21, 2007

We analyze how individual eigenvalues of the QCD Dirac operator at nonzero quark chemical potential are distributed in the complex plane. Exact and approximate analytical results for both quenched and unquenched distributions are derived from non-Hermitian random matrix theory. When comparing these to quenched lattice QCD spectra close to the origin, excellent agreement is found for zero and nonzero topology at several values of the quark chemical potential. Our analytical results are also applicable to other physical systems in the same symmetry class.

12.38.Gc, 02.10.Yn

Introduction. Hermitian random matrix theory (RMT), which describes systems with real spectra, enjoys many applications in physics and beyond. Dropping the Hermiticity constraint results in matrices whose eigenvalues are, in general, complex. Examples are the Ginibre ensembles Ginibre (1965) or their chiral counterparts Halasz et al. (1997). Although these ensembles describe non-Hermitian operators, they have found many applications (see Fyodorov and Sommers (2003) for a recent review), ranging from dissipation in quantum maps Grobe et al. (1988) over quantum chromodynamics (QCD) at nonzero quark chemical potential Stephanov (1996) to the brain auditory response described by nonsymmetric correlation matrices Kwapien et al. (2000).

Observables that are typically computed in RMT are spectral correlation functions. Alternatively, one can study the distributions of individual eigenvalues, provided that the latter can be ordered. For RMT with real eigenvalues, all such distributions are known and have found a variety of important applications. For example, the largest eigenvalue follows the Tracy-Widom distribution Tracy and Widom (1994) and appears in the longest increasing sub-sequence of partitions Baik et al. (1999) or growth processes Praehofer and Spohn (2000). The smallest eigenvalue distribution in chiral RMT has become a standard tool in lattice QCD to extract the low-energy constant (LEC) that appears in chiral perturbation theory (chPT) and is related to the chiral condensate Fukaya et al. (2007). This distribution is also sensitive to the gauge field topology and can be used to distinguish different patterns of chiral symmetry breaking Edwards et al. (1999).

In this paper, we generalize some of these results to the case of non-Hermitian chiral RMT in the unitary symmetry class. We study the distributions of individual eigenvalues in the complex plane and derive analytical results for the chiral RMT introduced in Ref. Osborn (2004). Our main focus will be on QCD, but our findings are also relevant for other systems with complex eigenvalues in the same symmetry class.

In QCD, a nonzero quark chemical potential leads to a complex spectrum of the Dirac operator. In the large-volume limit, chiral RMT is equivalent Basile and Akemann (2007) to the chiral effective theory for the epsilon-regime of QCD Gasser and Leutwyler (1987), which is a particular low-energy limit of the full theory. Here, a virtue of is that couples to the second LEC in leading order of chPT, , which is related to the pion decay constant Toublan and Verbaarschot (2001). A comparison of lattice QCD data to individual complex Dirac eigenvalue distributions from RMT thus allows us to determine both and (for related methods, see Refs. Damgaard et al. (2005); Osborn and Wettig (2006)).

Unfortunately, lattice QCD with dynamical fermions at faces a serious difficulty due to the loss of reality of the action. It is very hard to obtain significant statistics in unquenched simulations, and therefore we will only compare to quenched simulations below. However, for or (where is the pion mass and is the volume) the sign problem is not severe Spl (), and our method can be used to determine from such unquenched lattice data. Therefore we also derive RMT results for unquenched QCD, thus adding to the predictions for spectral densities Akemann et al. (2005) and the average phase factor Spl ().

What is known from RMT for individual eigenvalue correlations in the complex plane? For the non-chiral, unitary Ginibre ensemble the repulsion (or spacing distribution) of complex levels was computed in Grobe et al. (1988) and successfully compared to lattice QCD data in the bulk of the spectrum Markum et al. (1999). For maximal non-Hermiticity, the distribution of the largest eigenvalue with respect to radial ordering is also known Kanzieper (2005). However, in QCD it is the eigenvalues closest to the origin that carry information about topology and LECs, and therefore we concentrate on these in the following.

The complex spectral correlation functions of the QCD Dirac operator at were computed from different (but equivalent) chiral RMTs in Refs. Splittorff and Verbaarschot (2004); Osborn (2004); Akemann et al. (2005) and compared to quenched lattice QCD in Refs. Ake (a); Osborn and Wettig (2006). Later, a Dirac operator with exact chiral symmetry at was constructed Bloch and Wettig (2006, 2007) and tested against chiral RMT for topological charge . Here, we compare the data of Ref. Bloch and Wettig (2006) to our newly derived individual complex eigenvalue distributions, resulting in a much improved signal. For a recent review we refer to Ref. Akemann (2007).

Complex eigenvalue distributions. We start by defining the gap probability and the distribution of an individual eigenvalue in the complex plane. Suppose the partition function can be written in terms of complex eigenvalues of some operator, with a joint probability distribution function (jpdf) , symmetric in all its arguments, to be specified. (For simplicity, we consider only jpdf’s with additional symmetry , restricting ourselves to the upper half-plane .) The complex eigenvalue density correlation functions are defined as


The simplest example is just the spectral density. The gap probability is defined as the probability that there are exactly eigenvalues inside the set and eigenvalues in its complement ,


If all are known, the gap probabilities follow as in the real case Akemann and Damgaard (2004),


Parameterizing the boundary of in as , we can define the probability for eigenvalues to be inside , for the eigenvalue to be on the contour at , and for eigenvalues to be in the complement ,


(Because eigenvalues repel each other in RMT, the probability of finding two eigenvalues at is zero.) An ordering on is induced by a family of sets of increasing area with mutually nonintersecting contours. Via the Riemann mapping theorem, this can always be reduced to radial ordering. Definitions (2) and (4) are related through a variational derivative,


Employing the expansion (3), we can express the through densities. For example, for the first eigenvalue,


Results from RMT. The above considerations hold for any jpdf, including the jpdf appearing in the lattice QCD partition function in terms of complex Dirac operator eigenvalues and the jpdf of chiral RMT. We now consider the latter. The RMT for unquenched QCD with Osborn (2004) we use here is given by


The Vandermonde, , coming from the diagonalization of complex matrices of dimension (we take for convenience), leads to a repulsion of eigenvalues. (For the chiral RMTs corresponding to adjoint or two-color QCD, the Jacobians will be different, leading to different patterns of eigenvalue repulsion, see, e.g., Ref. Akemann (2007).) The weight depends on dynamical quark flavors with masses () and on the number of exactly zero eigenvalues (corresponding to the topological charge),


where is a modified Bessel function and is the chemical potential in the random matrix model. The first factor in Eq. (8) originates from the Dirac determinants. The non-Gaussian weight function results from an integration over angular and auxiliary variables Osborn (2004). For the are back on the imaginary axis. Complex RMT yields the following result for the densities Ake (b),


given in terms of the kernel of (bi-)orthogonal polynomials with respect to the weight of Eq. (8). In the quenched case (), these are given by Laguerre polynomials in the complex plane Osborn (2004). All unquenched density correlations are given explicitly in Ref. Akemann et al. (2005). A determinental expression follows for the in terms of the kernel operator times the characteristic function of . Eq. (3) is called its Fredholm determinant expansion.

As mentioned above, in the limit of large volume , RMT is equivalent to QCD in the epsilon-regime Basile and Akemann (2007). In this regime, the chemical potential, the quark masses, and the Dirac eigenvalues are rescaled such that the parameters , , and stay finite in the large- (large-) limit. In parentheses, we have given the scaling of these parameters in terms of the LECs of chPT.

Quenched case. In the quenched case, the RMT result for the microscopic spectral density is given by Splittorff and Verbaarschot (2004); Osborn (2004)


where is a modified Bessel function. The rescaled kernel giving all correlation functions according to Eq. (9) was derived in Refs. Osborn (2004); Akemann et al. (2005). In Fig. 1 we show as an example the density and the distribution of the first eigenvalue from Eq. (6) (in which is chosen to be semi-circular and only the first three terms are included). As in the case of real eigenvalues Akemann and Damgaard (2004), we see that the expansion converges rapidly. Higher-order terms merely assure that remains zero for large .

Figure 1: Quenched density of Eq. (10) (left), and quenched from Eq. (6) (including the first three terms) for circular (right), both for and .

For increasing , the quenched density Eq. (10) rapidly becomes rotationally invariant close to the origin. In terms of the new variable , it becomes


In this limit, we can derive a closed expression for the gap probability ASI (). Because of the rotational symmetry we choose to be a semi-circle of radius and obtain


where we have introduced the incomplete Bessel function for , and zero otherwise. Our quenched expression Eq. (12) generalizes the corresponding result of Ref. Grobe et al. (1988) for the non-chiral Ginibre ensemble, which is given in terms of incomplete exponentials .

Denoting each factor in Eq. (12) by , expressions for the easily follow in terms of the Mehta (2004). The radially ordered eigenvalue distributions are then obtained from the via Eq. (5), leading to


Figure 2 shows that the individual eigenvalue distributions nicely add up to the density Eq. (11).

Figure 2: Quenched spectral density Eq. (11) and distributions of the first eight eigenvalues Eq. (13), as well as their sum, all in the large- limit, for (left) and (right).

Comparison with lattice data. We now come to the comparison of our analytical results to quenched lattice QCD data. For details of the simulation we refer to Ref. Bloch and Wettig (2006). The gauge fields were generated in the quenched approximation on a lattice at (see Bloch and Wettig (2006) for an explanation of these choices). The Dirac operator introduced in Ref. Bloch and Wettig (2006) is a generalization of the overlap Dirac operator ove () to . It satisfies a Ginsparg-Wilson relation Ginsparg and Wilson (1982) and has exact zero modes at finite lattice spacing. We can therefore test our predictions in different sectors of topological charge . In Ref. Bloch and Wettig (2006) complete spectra of the generalized overlap operator were computed for several values of and large numbers of configurations, and these data are used in the comparisons to the RMT results below. We also used the fit parameters and from Ref. Bloch and Wettig (2006) to determine and , i.e., no additional fits were performed.

For the contours we again choose semi-circles, for all values of . Since we prefer to show 2D plots we have integrated over the phase of the complex number and display only the radial dependence. Results for are shown in Fig. 3 for and , corresponding to and , and in Fig. 4 for and , corresponding to and . (The lattice spacing has been set to unity.)

Figure 3: Integrated distribution of the first eigenvalue for as a function of the radius for (left) and (right). The solid lines are the RMT results from Eq. (6), the histograms are the quenched lattice data of Ref. Bloch and Wettig (2006). The bending-up of the RMT curves for large is an artifact of using only the first three terms in the expansion (6), see text.

For all values of we compare the data to the expansion Eq. (6), in which only the first three terms were used.

For the data were found to be approximately rotationally invariant, and we also compare them to the exact result in the large- limit from Eqs. (12) and (13). (Because of the rotational invariance, only the ratio could be determined for in Ref. Bloch and Wettig (2006), see Eq. (11). In this case the value of used in Eq. (6) is an extrapolation, assuming that is independent of .) The agreement between data and analytical curves is excellent except for at (see Fig. 4). In these two cases we have left the range of validity of RMT.

We emphasize that while the rise of the distributions from zero was in principle already tested in Ref. Bloch and Wettig (2006) through the density (see Fig. 1 or 2), their decrease represents a new, parameter-free test. Note also that because of the integration over the phase, the signal is much better than in Ref. Bloch and Wettig (2006). This allows us, for the first time, to successfully test the RMT predictions for .

Figures 3 and 4 also show the effect of truncating the Fredholm expansion (6): The analytical curves bend up for large after (almost) touching zero. Higher-order terms in the expansion (6) only affect the tail of the distributions. They will “repair” the bending-up and ensure that the tails remain zero, just as the data. The same effect was observed earlier for real eigenvalue distributions Akemann and Damgaard (2004). This feature of our approximation can be seen most clearly when comparing to the exact result in the large- limit, see Fig. 4 (right), in which we can observe how the expansion converges in the case of large .

Figure 4: Same as Fig. 3, but for (left) and (right). For we also show the exact RMT result in the large- limit from Eq. (13).

Conclusions. We have shown that the distributions of individual complex eigenvalues from non-Hermitian RMT agree very well with the corresponding distributions of the complex eigenvalues of the quenched QCD Dirac operator closest to the origin in three different topological sectors. As in the Hermitian case, these distributions are much easier to compare with than the density, in which a plateau may not be observable due to appreciable finite-volume corrections. Our analytical results are also relevant for other non-Hermitian systems in the chiral unitary symmetry class. In the future, it would be interesting to compute (and apply) similar results for the orthogonal and symplectic symmetry classes.

Acknowledgments. We thank K. Splittorff and J.J.M. Verbaarschot for useful discussions. This work was supported by EPSRC grant EP/D031613/1 (GA & LS), by EU network ENRAGE MRTN-CT-2004-005616 (GA), and by DFG grant FOR 465 (JB & TW).


  • Ginibre (1965) J. Ginibre, J. Math. Phys. 6, 440 (1965).
  • Halasz et al. (1997) M. A. Halasz, J. C. Osborn, and J. J. M. Verbaarschot, Phys. Rev. D56, 7059 (1997).
  • Fyodorov and Sommers (2003) Y. V. Fyodorov and H.-J. Sommers, J. Phys. A36, 3303 (2003).
  • Grobe et al. (1988) R. Grobe, F. Haake, and H.-J. Sommers, Phys. Rev. Lett. 61, 1899 (1988).
  • Stephanov (1996) M. A. Stephanov, Phys. Rev. Lett. 76, 4472 (1996).
  • Kwapien et al. (2000) J. Kwapien, S. Drozdz, and A. A. Ioannides, Phys. Rev. E62, 5557 (2000).
  • Tracy and Widom (1994) C. A. Tracy and H. Widom, Commun. Math. Phys. 159, 151 (1994).
  • Baik et al. (1999) J. Baik, P. Deift, and K. Johansson, J. Amer. Math. Soc. 12, 1119 (1999).
  • Praehofer and Spohn (2000) M. Praehofer and H. Spohn, Phys. Rev. Lett. 84, 4882 (2000).
  • Fukaya et al. (2007) H. Fukaya et al. (JLQCD collaboration), Phys. Rev. Lett. 98, 172001 (2007).
  • Edwards et al. (1999) R. G. Edwards, U. M. Heller, J. E. Kiskis, and R. Narayanan, Phys. Rev. Lett. 82, 4188 (1999).
  • Osborn (2004) J. C. Osborn, Phys. Rev. Lett. 93, 222001 (2004).
  • Basile and Akemann (2007) F. Basile and G. Akemann, JHEP 12, 043 (2007).
  • Gasser and Leutwyler (1987) J. Gasser and H. Leutwyler, Phys. Lett. B188, 477 (1987).
  • Toublan and Verbaarschot (2001) D. Toublan and J. J. M. Verbaarschot, Int. J. Mod. Phys. B15, 1404 (2001).
  • Damgaard et al. (2005) P. H. Damgaard, U. M. Heller, K. Splittorff, and B. Svetitsky, Phys. Rev. D72, 091501 (2005).
  • Osborn and Wettig (2006) J. C. Osborn and T. Wettig, PoS LAT2005, 200 (2006).
  • (18) K. Splittorff and J. J. M. Verbaarschot, Phys. Rev. Lett. 98, 031601 (2007); Phys. Rev. D75, 116003 (2007).
  • Akemann et al. (2005) G. Akemann, J. C. Osborn, K. Splittorff, and J. J. M. Verbaarschot, Nucl. Phys. B712, 287 (2005).
  • Markum et al. (1999) H. Markum, R. Pullirsch, and T. Wettig, Phys. Rev. Lett. 83, 484 (1999).
  • Kanzieper (2005) E. Kanzieper, in Frontiers in Field Theory, edited by O. Kovras (Nova Science, NY, 2005), p. 33.
  • Splittorff and Verbaarschot (2004) K. Splittorff and J. J. M. Verbaarschot, Nucl. Phys. B683, 467 (2004).
  • Ake (a) G. Akemann and T. Wettig, Phys. Rev. Lett. 92, 102002 (2004) [Erratum ibid. 96, 029902(E) (2006)].
  • Bloch and Wettig (2006) J. Bloch and T. Wettig, Phys. Rev. Lett. 97, 012003 (2006).
  • Bloch and Wettig (2007) J. Bloch and T. Wettig, Phys. Rev. D76, 114511 (2007).
  • Akemann (2007) G. Akemann, Int. J. Mod. Phys. A22, 1077 (2007).
  • Akemann and Damgaard (2004) G. Akemann and P. H. Damgaard, Phys. Lett. B583, 199 (2004).
  • Ake (b) G. Akemann, Phys. Rev. Lett. 89, 072002 (2002); J. Phys. A36, 3363 (2003).
  • (29) G. Akemann and L. Shifrin, to be published.
  • Mehta (2004) M. L. Mehta, Random matrices (Elsevier, Amsterdam, 2004), 3rd ed., Eq. (6.4.30).
  • (31) R. Narayanan and H. Neuberger, Nucl. Phys. B443, 305 (1995); H. Neuberger, Phys. Lett. B417, 141 (1998).
  • Ginsparg and Wilson (1982) P. H. Ginsparg and K. G. Wilson, Phys. Rev. D25, 2649 (1982).
Comments 0
Request Comment
You are adding the first comment!
How to quickly get a good reply:
  • Give credit where it’s due by listing out the positive aspects of a paper before getting into which changes should be made.
  • Be specific in your critique, and provide supporting evidence with appropriate references to substantiate general statements.
  • Your comment should inspire ideas to flow and help the author improves the paper.

The better we are at sharing our knowledge with each other, the faster we move forward.
The feedback must be of minimum 40 characters and the title a minimum of 5 characters
Add comment
Loading ...
This is a comment super asjknd jkasnjk adsnkj
The feedback must be of minumum 40 characters
The feedback must be of minumum 40 characters

You are asking your first question!
How to quickly get a good answer:
  • Keep your question short and to the point
  • Check for grammar or spelling errors.
  • Phrase it like a question
Test description