A Simple Introduction to Grobner Basis Methods in String Phenomenology

In this talk I give an elementary introduction to the key algorithm used in recent applications of computational algebraic geometry to the subject of string phenomenology. I begin with a simple description of the algorithm itself and then give 3 examples of its use in physics. I describe how it can be used to obtain constraints on flux parameters, how it can simplify the equations describing vacua in 4d string models and lastly how it can be used to compute the vacuum space of the electroweak sector of the MSSM.

Despite the wide range of physical problems which have been addressed within this context, the computational tools which are being used are all based, finally, on the same algorithm. The Buchberger algorithm [15] is at once what lends these methods their power and also the rate limiting step -placing bounds on the size of problem that can be dealt with. The recent burst of activity in this field has been fueled, in part, by the advent of freely available, efficient implementations of this algorithm [16,17]. There are also interfaces available between the commutative algebra program [17] and Mathematica [18,8], with [8] being particularly geared towards physicist's needs. The aim of this talk is to give an elementary introduction to the Buchberger algorithm and some of its recent applications.
In order to give an idea of how one simple algorithm can make so much possible, I will present the Buchberger algorithm and then show how it may be applied to physics in 3 elementary examples. Firstly, I will describe how it can be used to obtain constraints on the flux parameters in four dimensional descriptions of string phenomenological models which are necessary and sufficient for the existence of certain types of vacuum [8]. Secondly, I will describe how the Buchberger algorithm can be used to simplify the equations describing the vacua of such systems -making problems of finding minima much more tractable [8]. Finally, I shall describe how the same simple algorithm can be used to calculate the supersymmetric vacuum space geometry of the electroweak sector of the MSSM [1].
The remainder of this talk is structured as follows. In the next section, I take a few pages to explain the algorithm and the few mathematical concepts that we will require. In the three sections following that, I then describe the three examples mentioned above. I shall conclude by making a few final comments about the versatility and scaling of the Buchberger algorithm.

A tiny bit of commutative algebra
Two pages of simple mathematics will suffice to achieve all of the physical goals mentioned in the introduction. First of all we define the notion of a polynomial ring. In this paper we will call the fields of the physical systems we study φ i and any parameters present, such as flux parameters, a α . The polynomial rings C φ i , a α and C [a α ] are then simply the infinite set of all polynomials in the fields and parameters, and the infinite set of all polynomials in the parameters respectively.
Another mathematical concept we will require is that of a monomial ordering. This is simply an unambiguous way of stating whether any given monomial is formally bigger than any other given monomial. We may denote this in a particular case by saying m 1 > m 2 where m 1 , m 2 ∈ C φ i , a α are monomials in the fields and parameters. It is important to say what is not meant by this. We are not saying that we are taking values of the variables such that the monomial m 1 is numerically larger than the monomial m 2 . Rather we are saying that, in our formal ordering, m 1 is considered to come before m 2 .
For our purposes we will require a special type of monomial ordering called an elimination ordering. This means that our formal ordering of monomials has the following property.
In words this just says that if the largest monomial in P according to our ordering, LM(P ), does not depend on φ i then P does not depend on the fields at all. The monomial ordering classes all monomials with fields in them as being bigger than all of those without such constituents.
Given this notion of monomial orderings we can now present the one algorithm we will need to use -the Buchberger algorithm [15]. The Buchberger algorithm takes as its input a set of polynomials. These may be thought of as a system of polynomial equations by the simple expedient of setting all of the polynomials to zero. The algorithm returns a new set of polynomials which, when thought of as a system of equations in the same way, has the same solution set as the input. The output system, however, has several additional useful properties as we will see.

The Buchberger Algorithm
1. Start with a set of polynomials, call this set G.
2. Choose a monomial ordering with the elimination property described above.
3. For any pair of polynomials P i , P j ∈ G multiply by monomials and form a difference so as to cancel the leading monomials with respect to the monomial ordering: 4. Perform polynomial long division of S with respect to G. That is, formh = S − m 3 P k where m 3 is a monomial and P k ∈ G such that m 3 LM(P k ) cancels a monomial in S. Repeat until no further reduction is possible. Call the result h.

5.
If h = 0 consider the next pair. If h = 0 add h to G and return to step 3.
The algorithm terminates when all S-polynomials which may be formed reduce to 0. The final set of polynomials is called a Gröbner basis.
As mentioned above, the resulting set of polynomials has several nice properties. The feature which is often taken as defining is that polynomial long division with respect to this new set of polynomials always gives the same answer -it does not matter in which order we divide the polynomials out by.
For us, however, the important point about our Gröbner basis G is that it has what is called the elimination property. G ∩ C [a α ], the set of all polynomials in G which depend only upon the parameters, gives a complete set of equations on the a α which are necessary and sufficient for the existence of a solution to the set of equations we started with. The reason why this is so is actually very straightforward. Our elimination ordering says that any monomial with a field in it is greater than any monomial only made up of parameters. Looking back at step 3 of the Buchberger algorithm we see that we are repeatedly canceling off the leading terms of our polynomials -those containing the fields -as much as we can. Thus if it is possible to rearrange our initial equations to get expressions which do not depend upon the fields φ i then the Buchberger algorithm will do this for us. Clearly, while we have interpreted the a α as parameters and the φ i as fields in the above, as this is what we will require for the next section, this was not necessary. The Buchberger algorithm can be used to eliminate any unwanted set of variables from a problem, in the manner we have described.
This completes all of the mathematics we will need for our entire discussion and we may now move on to apply what we have learnt.

Constraints
The first physical question we wish to answer is the following. Given a four dimensional N = 1 supergravity describing a flux compactification, what are the constraints on the flux parameters which are necessary and sufficient for the existence of a particular kind of vacuum? This question can be asked, and answered [8], for any kind of vacuum, but in the interests of concreteness and brevity let us restrict ourselves to the simple case of supersymmetric Minkowski vacua.
Here is the superpotential of a typical system, taken from [19]. It describes a non-geometric compactification of type IIB string theory.
This system has some known constraints on its parameters which are necessary for the existence of a permissible vacuum. These come from, for example, tadpole cancellation conditions.
We also have the same constraints with the hats and checks switched around. In this example the fields, which we have been calling φ i , are S, τ and U and everything else is a "flux" parameter, or an a α in our notation.
In total, the equations which must be satisfied if a supersymmetric Minkowski vacuum is to exist are W = 0, ∂ S W = 0, ∂ τ W = 0, ∂ U W = 0 and the constraints on the flux parameters given above.
To extract a set of constraints solely involving the parameters which are necessary and sufficient for the existence of a solution to these equations we simply follow the procedure outlined in the previous section.
We can carry out this calculation trivially in Stringvacua [8] and, in fact, this example is provided for the user in the help system. The result is The reader will note that the result is a somewhat lengthy set of equations. In principle one has to find quantized solutions to these expressions, an obviously intractable Diophantine problem, and therefore it might be asked why this result is of any use. In fact, knowledge of such constraints on the flux parameters is hugely useful for a number of reasons.
• Firstly, we note that, while the full result of this process is often complex, some of the constraints can give us simple information about the system. In the current case, for example it can be seen thatc 2 = 0 is required for the existence of a supersymmetric Minkowski vacuum.
• Secondly, if one is scanning over a range of flux parameters and trying to numerically solve the equations to find vacua one can speed up ones analysis by first substituting any given set of flux parameters into the constraints we have obtained. If the constraints are not satisfied then vacua do not exist and there is no point in searching numerically for a solution. This turns what would be a time consuming numerical process giving inconclusive results (no solution was found) into a quick analytic conclusion (no solution exists).
• Lastly, knowledge of such constraints can greatly speed up algebraic approaches to finding vacua such as those outlined in [8].

Simplifying equations for vacua
Another use for the mathematics we learnt in section 2 are the so called "splitting tools" used in work such as [8]. The physical idea here is simple. Consider trying to solve the equations ∂V /∂φ i = 0 to find the vacua, including those which spontaneously break supersymmetry, of some supergravity theory. These equations are often extremely complicated. One way of viewing why this is so is that the equations for the turning points of the potential contain a lot of information. They describe not only the isolated minima of the potential which are of interest, but also lines of maxima, saddle points of various sorts and so forth. A useful tool to have, therefore, would be an algorithm that takes such a system as an input and returns a whole series of separate sets of equations, each individually describing fewer turning points. Since each separate equation system would then contain less information one might expect them to be easier to solve. It would be beneficial to choose a division of these equations which has physical interest. The choice we will make here, and which programs like Stringvacua implement [8], is to split up the equations for the turning points according to how they break supersymmetry -that is according to which F-terms vanish when evaluated on those loci.
The ability that packages such as Stringvacua have to split up equations in this manner is based upon the following splitting tool (see [21] for a nice set of more detailed notes on this kind of mathematical technique). Say that one of the F-terms of our theory is called F . Then we can split the equations describing turning points of the potential into two pieces.
The first of these expressions is a set of equations which is easier to solve, in general, than ∂V /∂φ i = 0 alone. We can use the F-term to simplify the equations for the turning points of the potential. On the other hand, expression (7) is not even a set of equations -it contains an inequality. We can convert (7) into a system purely involving equalities by making use of the mathematics we learned in section 2.
Consider the following set of equations, including a dummy variable t.
The second equation in (8) (7) do. Unfortunately they also depend upon one extra, and unwanted, variable -t. This is not a problem as we already know how to remove unwanted variables from our equations. We can simply eliminate them, as we did the fields in section 2. This will leave us with a necessary and sufficient set of equations in φ i and a α for a solution to (8) and thus to (7). So we can split the equations for the turning points of our potential into two simpler systems. One describes the turning points of V for which F = 0 and the other those for which F = 0. We can of course perform such a splitting many times -once for each F-term! In fact, using additional techniques from algorithmic algebraic geometry [20,8], which are essentially based upon the same trick, one can go much further. One can split the equations for the turning points up into component parts gaining one set of equations for every separate locus. Because we know which F-terms are non-zero on each of them these are classified according to how they break supersymmetry. The researcher interested in a certain type of breaking can therefore select the equations describing the vacua of interest and throw everything else away.
The above process of splitting up the equations for the vacua of a system can be very simply carried out in Stringvacua. Numerous examples can be found in the Mathematica help files which come with the package [8]. Here, let us consider the example of M-theory compactified on the coset SU (3)×U (1) U (1)×U (1) . The Kähler and superpotential for this coset, which has SU (3) structure, has been presented in [22].
Even this, relatively simple, model results in a potential of considerable size. Defining T i = −it i + τ i and U = −ix + y we find, To find the turning points of this potential we naively need to take eight different derivatives of (10) and solve the resulting set of inter-coupled equations in eight variables. This is clearly prohibitively difficult. Using the techniques described in this section, however, Stringvacua, can separate off parts of the vacuum space for us with ease. Consider, for example, the vacua which are isolated in field space and for which the real parts of all of the F-terms are non-zero, with the imaginary parts vanishing. To find these, the package tells us, we need only solve the equations, Because they only describe a small subset of all of the turning points of the full potential these equations are extremely simple in form and may be trivially solved. For this particular example the physically acceptable turning point that results is a saddle -something which can be readily ascertained once its location has been discovered.

Geometry of vacuum spaces
As a final example of what we can do with the simple techniques introduced in section 2 we will show how to calculate the vacuum space of a globally supersymmetric gauge theory. It is a well known result (see [23] and references therein) that the supersymmetric vacuum space of a such a theory, with gauge group G, can be described as the space of holomorphic gauge invariant operators (GIO's) built out of F-flat field configurations. What does this space look like? Consider a space, the coordinates of which Type Explicit Sum Index Number LHe L i αHβ ǫ αβ e j i, j = 1, 2, 3 9 ν ν i i = 1, 2, 3 1 If we take these GIO's to be built out of the F-flat field configurations then there will be still further relations among them. The vacuum space of the theory is the subspace defined by the solutions of these equations describing relations amongst the gauge invariant operators, once F-flatness has been taken into account. How can we calculate such a thing? The holomorphic gauge invariant operators of a globally supersymmetric gauge theory are given in terms of the fields.
Here S I are our GIO's and the f I are the functions of the fields that define them. Let us write the F-terms of the theory as F i . Consider the following set of equations.
These equations have solutions whenever the S I are given by functions of the fields in the correct way and when those field configurations which are being used are F-flat. However, according to the proceeding discussion, we wish to simply have equations in terms of the GIO's to describe our vacuum space. As in previous sections, we can eliminate the unwanted variables in our problem, in this case the fields φ i , using the algorithm of section 2 to obtain the equations describing the vacuum space. As a simple example, let us take the electroweak sector of the MSSM [1] (with right handed neutrinos). Given the field content of the left handed leptons, L i α , the right handed leptons, e i and ν i , and the two Higgs, H andH, one can build the elementary GIO's given in table 1. The indices i, j run over the 3 flavours and the indices α, β label the fundamental of SU (2).
To compute the F-terms we require the superpotential. Let us take the most general renormalizable form which is compatible with the symmetries of the theory and R-parity.
Here ǫ is the invariant tensor of SU (2) and C 0 , C 3 ij , C 4 ij , and, C 5 ij are constant coefficients. We now just follow the procedure outlined at the start of this section. We calculate the F-terms by taking derivatives of the superpotential, we label the gauge invariant operators S 1 to S 23 , we form the equations (13) and then simply run the elimination algorithm given in section 2.
The result is, upon simplification, given by six quadratic equations in 6 variables. It is a simple description of an affine version of a famous algebraic variety -the Veronese surface [1]. What can be done with such a result? The first observation we can make is that this vacuum space is not a Calabi-Yau. This means, for example, that one can say definitively that it is not possible to engineer this theory by placing a single D3 brane on a singularity in a Calabi-Yau manifold, without having to get into any details of model building.
Secondly one can study such vacuum spaces in the hope of finding hints at the structure of the theory's higher energy origins. In the case we have studied in this section, for example, we can "projectivize" (pretend the GIO's are homogeneous coordinates on projective space rather than flat space coordinates) and study the Hodge diamond of the result. The structure of supersymmetric field theory tells us that this Hodge diamond should depend on 4 arbitrary integers, but there is nothing at low energies which prevents us from building theories with any such integers we like. Interestingly, in the case of electroweak theory, these integers are all zero or one.
Whether this structure is indeed a hint of some high energy antecedent or just a reflection of the simplicity of the theory is debatable. This example does, however, demonstrate the idea of searching for such evidence of new physics in vacuum space structure. We should also add here that similar techniques can be used to show that the vacuum space of SQCD is a Calabi-Yau [5].

Final Comments
To conclude we shall make several points -one of which is a note of caution, with the rest being more optimistic. The first point which we shall make is that we should be careful lest the above discussion makes the algorithm we have been describing sound like an all-powerful tool. There is, as ever, a catch. In this case it is the way the algorithm scales with the complexity of the problem. A "worst case" upper bound for the degree of the polynomials in a reduced Gröbner basis can be found in [24]. If d is the largest degree found in your original set of equations then this bound is, where n is the number of variables. This worst case bound is therefore scaling doubly exponentially in the number of degrees of freedom. These very high degree polynomials are an indication that the problem is becoming very complex and thus computationally intensive. Despite this, physically useful cases can be analysed using this algorithm quickly, as demonstrated in this talk and in the references. This scaling does mean that one is not likely to gain much by putting one's problem on a much faster computer. One good point about (16) is that if you can find a way, using physical insight, to simplify the problem under study, then what you can achieve may improve doubly exponentially. Such a piece of physical insight was one of the keystones of the application of these methods to finding flux vacua [8].
We finish by commenting that the methods of computational commutative algebra which we have discussed here are extremely versatile. We have been able to perform three very different tasks simply utilizing one algorithm in a very simple manner. These methods are of great utility in problems taken from the literature and their implementation in a user friendly way in Stringvacua means that they may be tried out on any given problem with very little expenditure of time and effort by the researcher. Many more techniques from the field of algorithmic commutative algebra could be applied to physical systems than those described here, or indeed in the physics literature. We can therefore expect that this subject will only increase in importance in the future.