Diophantine Polynomial Systems

Introduction

A Diophantine polynomial system is an expression constructed with polynomial equations and inequalities

combined using logical connectives and quantifiers

where the variables represent integer quantities.

An occurrence of a variable inside or is called a bound occurrence; any other occurrence of is called a free occurrence. A variable is called a free variable of a polynomial system if the system contains a free occurrence of . A Diophantine polynomial system is quantifier-free if it contains no quantifiers. A decision problem is a system with all variables existentially quantified, that is, a system of the form

where are all variables in . The decision problem (1) is equivalent to True or False, depending on whether the quantifier-free system of polynomial equations and inequalities has integer solutions.

An example of a Diophantine polynomial system is

Goldbach's conjecture [1], formulated in 1742 and still unproven, asserts that system (2) is equivalent to True. This suggests that the Wolfram Language may not be able to solve arbitrary Diophantine polynomial systems. In fact, Matiyasevich's solution of Hilbert's tenth problem [2] shows that no algorithm can be constructed that would solve arbitrary Diophantine polynomial systems, not even quantifier-free systems or decision problems. Nevertheless, the Wolfram Language functions Reduce, Resolve, and FindInstance are able to solve several reasonably large classes of Diophantine systems. This tutorial describes these classes of systems and methods used by the Wolfram Language to solve them. The methods are presented in the order in which they are used. The tutorial also covers options affecting the methods used and how they operate.

Linear Systems

Systems of Linear Equations

Conjunctions of linear Diophantine equations are solvable for an arbitrary number of variables. The Wolfram Language uses a method based on the computation Hermite normal form of matrices, available in the Wolfram Language directly as HermiteDecomposition. The result may contain new unrestricted integer parameters. If the equations are independent, the number of parameters is equal to the difference between the number of variables and the number of equations.

This system has four variables and two independent equations, hence the result is expressed in terms of two integer parameters:

Frobenius Equations

A Frobenius equation is an equation of the form

where are positive integers, is an integer, and the coordinates of solutions are required to be nonnegative integers.

For finding solution instances of Frobenius equations the Wolfram Language uses a fast algorithm based on the computation of the critical tree in the Frobenius graph [11]. The algorithm applies when the smallest of does not exceed the value of the MaxFrobeniusGraph system option (the default is 1,000,000). Otherwise the more general methods for solving bounded linear systems are used. Functions FrobeniusSolve and FrobeniusNumber provide specialized functionality for solving Frobenius equations and computing Frobenius numbers.

This finds a solution of a Frobenius equation:

Bounded Systems of Linear Equations and Inequalities

If a real solution set of a system of linear equations and inequalities is a bounded polyhedron, the system has finitely many integer solutions. To find the solutions, the Wolfram Language uses the following procedure.

You may assume the system has the form , where is a integer matrix, is a length integer vector, is an integer matrix, and is a length integer vector. First, the method for solving systems of linear equations is used to find an integer vector such that and a integer matrix N whose rows generate the null space of . The integer solution set of is equal to . Put and . The integer solution set of is equal to , where is the integer solution set of . To improve efficiency of finding the set , the Wolfram Language simplifies using LatticeReduce, obtaining . Note that if the columns of are linearly dependent, has no solutions (otherwise it would have infinitely many solutions, which contradicts the assumptions). Hence you may assume that has linearly independent columns and so has columns. Put . Lattice reduction techniques are also used to find a small vector in the lattice . Let be such that . The set can be computed from the set of all such that using the formula .

To find the set a simple recursive algorithm can be used. The algorithm finds the bounds on the first variable using LinearProgramming and, for each integer value between the bounds, calls itself recursively with the first variable set to . This algorithm is used when the system option BranchLinearDiophantine is set to False. With the default setting True a hybrid algorithm combining the recursive algorithm and a branch-and-bound-type algorithm is used. At each level of the recursion, the recursion is continued for the "middle" values of the first variable (defined as a projection of the set of points contained in the real solution set together with a unit cube) while the remaining parts of the real solution set are searched for integer solutions using the branch-and-bound type algorithm. FindInstance finds the single element of it needs using a branch-and-bound type algorithm.

There are two system options, BranchLinearDiophantine and LatticeReduceDiophantine, that allow you to control the exact algorithm used. In some cases changing the values of these options may greatly improve the performance of Reduce.

This finds all integer points in a triangle:

The Wolfram Language enumerates the solutions explicitly only if the number of integer solutions of the system does not exceed the maximum of the th power of the value of the system option DiscreteSolutionBound, where is the dimension of the solution lattice of the equations, and the second element of the value of the system option ExhaustiveSearchMaxPoints.

Here Reduce does not give explicit solutions because their number would exceed the default limit of 10000:
This increases the value of the system option DiscreteSolutionBound to 1000:
Since there are two variables and no equations, the limit on the number of solutions is now , and Reduce can enumerate the solutions explicitly:
This resets DiscreteSolutionBound to the default value:

Arbitrary Systems of Linear Equations and Inequalities

Quantifier-free systems of linear Diophantine equations and inequalities are solvable for an arbitrary number of variables. The system is written in the disjunctive normal form, that is, as a disjunction of conjunctions, and each conjunction is solved separately. If a conjunction contains only equations, the method for solving systems of linear equations is used. If the difference between the number of variables and the number of equations is at most one, the Wolfram Language solves the equations using the method for solving systems of linear equations, and if the solution contains at most one free parameter (which is true in the generic case), back substitutes the solution into the inequalities to determine inequality restrictions for the parameter. For all other quantifier-free systems of linear Diophantine equations and inequalities the Wolfram Language uses the algorithm described in [3], with some linear-programming-based improvements for handling bounded variables. The result may contain new nonnegative integer parameters, and the number of new parameters may be larger than the number of variables.

This system has three variables; however, to express the solution set, you need eight nonnegative integer parameters:

Univariate Systems

Univariate Equations

To find integer solutions of univariate equations the Wolfram Language uses a variant of the algorithm given in [4] with improvements described in [5]. The algorithm can find integer solutions of polynomials of much higher degrees than can be handled by real root isolation algorithms and with much larger free terms than can be handled by integer factorization algorithms.

Here Reduce finds integer solutions of a sparse polynomial of degree 100,000:
The free term of this polynomial has 609,152 digits and cannot be easily factored:

Systems of Univariate Equations and Inequalities

Systems of univariate Diophantine equations and inequalities are written in the disjunctive normal form, and each conjunction is solved separately. If a conjunction contains an equation, the method for solving univariate equations is used, and the solutions satisfying the remaining equations and inequalities are selected.

Here Reduce finds integer solutions of and selects the ones that satisfy the inequality :

Conjunctions containing only inequalities are solved over the reals. Integer solutions in the resulting real intervals are given explicitly if their number in the given interval does not exceed the value of the system option DiscreteSolutionBound. The default value of the option is 10. For intervals containing more integer solutions, the solutions are represented implicitly.

Bivariate Systems

Quadratic Equations

The Wolfram Language can solve arbitrary quadratic Diophantine equations in two variables. The general form of such an equation is

If , where and are linear polynomials, the equation (1) is equivalent to , and methods for solving linear Diophantine equations are used. For irreducible polynomials , the algorithms used and the form of the result depend on the determinant of the quadratic form. The algorithms may use integer factorization and hence the correctness of the results depends on the correctness of the probabilistic primality test used by PrimeQ.

Hyperbolic-Type Equations with Square Determinants

If and is an integer, then , where and are linear polynomials and . In this case, the equation (1) is equivalent to the disjunction of linear systems , for all divisors of . Each of the linear systems has one solution over the rationals, hence the equation (1) has a finite number of integer solutions.

Here is a binary quadratic equation with :
Hyperbolic-Type Equations with Nonsquare Determinants

If and is not an integer, then the equation (1) is a Pell-type equation. Methods for solving such equations have been developed since the 18 th century and can be constructed based on [6] and [7] (though these books do not contain a complete description of the algorithm). The solution set is empty or infinite, parametrized by an integer parameter appearing in the exponent.

A Pell equation is an equation of the form , where is not a square. Solutions to Pell equations with small coefficients can be quite complicated:
Here is the solution of a Pell-type equation with :
Even though the solutions are expressed using nonrational numbers, they are in fact integers, as they should be:
Reduce can solve systems consisting of a Pell-type equation and inequalities giving simple bounds on variables:
Parabolic-Type Equations

If , set , , and . Since , and are nonzero integers, and . Then

Set and . Then the equation (1) is equivalent to

Suppose . If the equation (1) had integer solutions, would have integer solutions in , and so would be a product of two linear polynomials. Since here is irreducible, the equation (1) has no integer solutions.

If , then the equation (2) implies

If the modular equation (3) has no solutions in , the equation (1) has no integer solutions. (If , the modular equation (3) has one solution, .) Otherwise , for some solution of the modular equation (3). Replacing in the equation (2) and solving the resulting linear equation for gives

Note that since satisfies the modular equation (3), the division in the last term of (4) gives an integer result. Since and , . Taking the equation (4) and the fact that into account gives

Therefore, the full solution of the equation (1) of parabolic type consists of one-parameter solution families given by equations (4) and (5) for each solution of the modular equation (3), for which is an integer.

Here Reduce finds integer solutions of a quadratic equation of the parabolic type:
Elliptic-Type Equations

If , the solutions of equation (1) are integer points on an ellipse. Since an ellipse is a bounded set, the number of solutions must be finite. An obvious algorithm for finding integer points would be to compute the solutions for for each of the finite number of possible integer values of . This, however, would be prohibitively slow for larger ellipses. The Wolfram Language uses a faster algorithm described in [8].

Here Reduce finds positive integer solutions of a quadratic equation of the elliptic type. There are more than possible positive integer values of , so the obvious algorithm would not be practical for this ellipse:

Thue Equations

A Thue equation is a Diophantine equation of the form

where is an irreducible homogeneous form of degree

The number of solutions of Thue equations is always finite. The Wolfram Language can in principle solve arbitrary Thue equations, though the time necessary to find the solutions lengthens very fast with degree and coefficient size. The hardest part of the algorithm is computing a bound on the size of solutions. The Wolfram Language uses an algorithm based on the BakerWustholz theorem to find the bound [9]. If the input contains inequalities that provide a reasonable size bound on solutions, the Wolfram Language can compute the solutions much faster.

This finds integer solutions of a cubic Thue equation:
If Reduce is given a bound on the size of solutions, it can solve the equation much faster:
Here Reduce finds the only solution of a degree 15 Thue equation with at most a 100-digit coordinate. Without the bound on the solution size, Reduce did not finish in 1000 seconds:

Multivariate Nonlinear Systems

Systems Solvable with the Modular Sieve Method

The Wolfram Language uses a variant of the modular sieve method (see e.g. [9]). The method may prove that a system has no solutions in integers modulo an integer , and therefore, it has no integer solutions. Otherwise, it may find a solution with small integer coordinates or prove that the system has no integer solutions with all coordinates between and . The number of candidate solution points that the sieve method is allowed to test is controlled by the system option SieveMaxPoints.

This equation has no solutions modulo 2:
Here FindInstance finds a small solution using the modular sieve:

Systems with More than One Equation

If a Diophantine polynomial system contains more than one equation, the Wolfram Language uses GroebnerBasis in an attempt to reduce the problem to a sequence of simpler problems.

Systems Solvable by Recursion over Finitely Many Partial Solutions

The Wolfram Language attempts to solve an element of the Gröbner basis that depends on the minimal number of the initial variables. If the number of solutions is finite, then for each solution the Wolfram Language substitutes the computed values and attempts to solve the obtained system for the remaining variables.

Here the first equation has four integer solutions for and . For each of the solutions, the second equation becomes a univariate equation in . The four univariate equations have a total of two integer solutions:
Here the first equation is a Thue equation with one solution. After replacing and with the values computed from the first equation, the second equation becomes a Pell equation:
Systems with Linear-Triangular Gröbner Bases

This heuristic applies to systems with Gröbner bases of the form

In this case, the Wolfram Language solves the system of congruences

The solutions are represented using new integer parameters. These are substituted into the equation and the inequalities present in the original system, and the Wolfram Language attempts to solve the so-obtained systems for the new parameters.

This system reduces to solving a congruence and a Pell equation:
This system reduces to solving a system of two congruences and a quadratic Diophantine equation of the parabolic type for each family of congruence solutions:

Sums of Squares

The Wolfram Language can solve equations of the form

using the algorithm described in [10]. For multivariate quadratic equations, the Wolfram Language attempts to find an affine transformation that puts the equation in the form (2). A heuristic method based on CholeskyDecomposition is used for this purpose. However, the method may fail for some equations that can be represented in the form (2).

This solves a sum-of-squares equation in three variables:

To find a single solution of (2) FindInstance uses an algorithm based on [12].

This finds a decomposition of a 10000-digit integer into a sum of seven squares. N is applied to make the printed result smaller:
This proves that the decomposition found is correct:

Pythagorean Equation

The Wolfram Language knows the solution to the Pythagorean equation

This gives the general solution of the Pythagorean equation:

For quadratic equations in three variables, the Wolfram Language attempts to find a transformation of the form

transforming the equation to the Pythagorean equation.

This equation can be transformed to the Pythagorean equation:

Equations with Reducible Nonconstant Parts

If the sum of nonconstant terms in an equation factors, the Wolfram Language uses the formula

to reduce the equation to a disjunction of pairs of equations with lower degrees. Note that this reduction depends on the ability to find all divisors of , hence the correctness of the results depends on the correctness of the probabilistic primality test used by PrimeQ.

This cubic equation reduces to 12 pairs of quadratic and linear equations:

Equations with a Linear Variable

The Wolfram Language attempts to solve Diophantine systems of the form

where is a conjunction of inequalities or True, by reducing them to

The first part of the system (3) is solved using the method for solving systems with more than one equation. The Wolfram Language recognizes three cases when the second part of the system (3) is solvable. If , the solution is given by and by the restrictions on obtained by solving the inequalities . Nonlinear systems of inequalities are solved using CylindricalDecomposition. If for an integer constant , the solution of the second part of the system (3) is given by and by the restrictions on obtained by solving the congruence and then solving the inequalities for each solution of the congruence. If is nonconstant, the Wolfram Language can solve the second part of the system (3) if . Since the Wolfram Language factors all equations at the preprocessing stage, and can be assumed to be relatively prime. Then

for an integer and polynomials and with integer coefficients and . If is an integer, then is an integer, and so or . Since , the last condition is satisfied only by a finite number of integers . Hence the solutions of the second part of the system (3) can be selected from a finite number of solution candidates.

Additionally, the Wolfram Language uses the following heuristic to detect cases when the system (3) has no solutions. If there is an integer , such that is always divisible by , and is never divisible by , then the system (3) has no solutions. Candidates for are found by computing the GCD of the values of at several points.

The last two methods use exhaustive search over finite sets of points. The allowed number of search points is controlled by the system option SieveMaxPoints.

This reduces to (3) with :
This reduces to (3) with :
This reduces to the case of system (3):
Here Reduce detects that the equation has no solutions, because is always divisible by 5, and is never divisible by 5:

Systems with Empty or Bounded Real Solution Sets

If a Diophantine polynomial system is not solved by any other methods, the Wolfram Language solves the system over the reals using the Cylindrical Algebraic Decomposition (CAD) algorithm. If the system has no real solutions, then clearly it has no integer solutions. If the real solution set is bounded, then the number of integer solutions is finite. In principle, all the integer solutions can be found in this case from a cylindrical decomposition. Namely, for each cylinder, you enumerate all possible integer values of the first coordinate, then for each value of the first coordinate, you enumerate all possible integer values of the second coordinate, and so on. However, for large bounded solution sets this method could lead to a huge number of points to try. Therefore, the Wolfram Language has a bound on the number of explicitly enumerated integer solutions in a real interval. For systems for which the real solution set is unbounded or bounded but large, the solution is represented implicitly by returning the CAD and a condition that all variables are integers. Note that for multivariate systems such an implicit representation may not even be enough to tell whether integer solutions exist. This should be expected, given Matiyasevich's solution of Hilbert's tenth problem [2].

Here Reduce enumerates all integer solutions in a bounded real region:
Here Reduce returns an implicit solution:
Setting DiscreteSolutionBound->Infinity makes Reduce explicitly enumerate integer solutions in arbitrary bounded sets:
This resets DiscreteSolutionBound to the default value:
Here the modular sieve method shows that there are no solutions in . After adding inequalities to eliminate this cube, Reduce then recognizes that this equation has no solutions anywhere:

Equations of the Form

The Wolfram Language attempts to solve Diophantine systems of the form

where is a conjunction of inequalities or True, by transforming them to

The resulting system (4) may, or may not, be easier to solve. Systems exist for which this transformation could be applied recursively arbitrarily many times; therefore, the Wolfram Language uses a recursion bound to ensure the heuristic terminates.

This transforms to a system (4) with no real solutions:
Here the system (4) obtained after three recursive transformations has a reducible nonconstant part:

Systems Solvable by Exhaustive Search

For systems containing explicit lower and upper bounds on all variables, the Wolfram Language uses exhaustive search to find solutions. The bounds of the search are specified by the value of the system option ExhaustiveSearchMaxPoints. The option value should be a pair of integers (the default is ). If the number of integer points within the bounds does not exceed the first integer, the exhaustive search is used instead of any solution methods other than univariate polynomial solving. Otherwise, if the number of integer points within the bounds does not exceed the second integer, the exhaustive search is performed after all other methods fail.

This transcendental Diophantine equation with bounded variable values is solved by exhaustive search:

Options

The Wolfram Language functions for solving Diophantine polynomial systems have a number of options that control the way they operate. This tutorial gives a summary of these options.

option name
default value
GeneratedParametersCspecifies how the new parameters generated to represent solutions should be named

Reduce options affecting the behavior for Diophantine polynomial systems.

GeneratedParameters

To represent infinite solutions of some Diophantine systems, Reduce needs to introduce new integer parameters. The names of the new parameters are specified by the option GeneratedParameters. With GeneratedParameters->f, the new parameters are named f[1], f[2], .

By default, the new parameters generated by Reduce are named C[1], C[2], :
The option GeneratedParameters allows users to customize the parameter names:

ReduceOptions Group of System Options

Here are the system options from the ReduceOptions group that may affect the behavior of Reduce, Resolve, and FindInstance for Diophantine polynomial systems. The options can be set with

SetSystemOptions["ReduceOptions"->{"option name"->value}].
option name
default value
"BranchLinearDiophantine"Truewhether Reduce should use a branch-and-bound-type algorithm to compute solutions of bounded systems of linear Diophantine inequalities
"DiscreteSolutionBound"10the bound on the number of explicitly enumerated integer solutions in a real interval
"ExhaustiveSearchMaxPoints"{1000,10000}the maximal number of integer points within variable bounds for which the exhaustive search is used before and after all other solution methods
"ImplicitIntegerSolutions"Automaticwhether Reduce should be allowed to return real solutions with Element conditions when it cannot find explicit integer solutions
"LatticeReduceDiophantine"Truewhether LatticeReduce should be used to preprocess bounded systems of linear Diophantine inequalities
"MaxFrobeniusGraph"1000000the maximal size of the smallest coefficient in a Frobenius equation for which FindInstance computes the critical tree in the Frobenius graph
"SieveMaxPoints"{10000,1000000}the maximal number of points at which the modular sieve method evaluates the system before and after all other solution methods

ReduceOptions group options affecting the behavior of Reduce, Resolve, and FindInstance for Diophantine polynomial systems.

BranchLinearDiophantine

The value of the system option BranchLinearDiophantine specifies which variant of the algorithm should be used in the final stage of solving bounded linear systems. With the default setting "BranchLinearDiophantine"->False, a simple recursive enumeration is used. With "BranchLinearDiophantine"->True, Reduce uses a hybrid method combining a branch-and-bound-type algorithm and a simple recursive enumeration.

This finds integer points in a long, narrow, four-dimensional simplex. Here the simple recursive enumeration is significantly faster than the hybrid method:
This solves a system of two randomly generated equations and three randomly generated inequalities in seven variables inside a simplex bounded by . Here the speed difference between the algorithms is smaller:
For this system, the nondefault simple recursion method is faster:
DiscreteSolutionBound

The value of the system option DiscreteSolutionBound specifies whether integer solutions in a real interval should be enumerated explicitly or represented implicitly as . With DiscreteSolutionBound->n, the integer solutions in the given real interval are enumerated explicitly if their number does not exceed n. The default value of the option is 10.

There are 10 integers in the real interval . Reduce writes them out explicitly:
There are 11 integers in the real interval . Reduce represents them implicitly:
This increases the DiscreteSolutionBound to 11:
Now Reduce represents the solutions explicitly:

The value of DiscreteSolutionBound also affects the solving of bounded linear systems and other Diophantine systems with finitely many solutions. When you need an explicit enumeration of a large number of solutions in a bounded region, set DiscreteSolutionBound to Infinity.

With the default setting of DiscreteSolutionBound, Reduce finds a parametrized description of the solution set:
Setting DiscreteSolutionBound->Infinity makes Reduce explicitly enumerate the solutions. In this example finding an explicit enumeration of solutions is faster:
This resets DiscreteSolutionBound to the default value:
ExhaustiveSearchMaxPoints

The system option ExhaustiveSearchMaxPoints specifies the maximal number of search points used by the exhaustive search method. The option value should be a pair of integers (the default is ). If the number of integer points within the bounds does not exceed the first integer, the exhaustive search is used instead of any solution methods other than univariate polynomial solving. Otherwise, if the number of integer points within the bounds does not exceed the second integer, the exhaustive search is performed after all other methods fail.

With the default setting of ExhaustiveSearchMaxPoints, Reduce is unable to solve this equation:
This increases the value of the second element of ExhaustiveSearchMaxPoints to :
Now Reduce can solve the equation:
With the default setting of ExhaustiveSearchMaxPoints, Reduce solves this equation using the method for solving Pell equations:
Increasing the first element of ExhaustiveSearchMaxPoints to makes Reduce use the exhaustive search first. In this example the search is much slower than the Pell equation solver:
For this equation, the Pell equation solver is slower than the exhaustive search:
The exhaustive search is faster here:
This resets ExhaustiveSearchMaxPoints to the default value:
ImplicitIntegerSolutions

The value of the system option DiscreteSolutionBound specifies whether integer solutions in a real interval should be enumerated explicitly or represented implicitly as . With DiscreteSolutionBound->n, the integer solutions in the given real interval are enumerated explicitly if their number does not exceed n. The default value of the option is 10.

By default, when Reduce cannot find explicit integer solutions it returns real solutions with Element conditions:
With ImplictIntegerSolutions->False, Reduce allows only explicit solutions to Diophantine problems:
With the default value ImplictIntegerSolutions->Automatic, Reduce allows only explicit solutions to recursively generated Diophantine subproblems:
Setting ImplictIntegerSolutions->True makes Reduce accept implicit solutions to recursively generated Diophantine subproblems:
This resets ImplicitIntegerSolutions to the default value:
LatticeReduceDiophantine

The value of the system option LatticeReduceDiophantine specifies whether LatticeReduce should be used to preprocess systems of bounded linear inequalities. The use of LatticeReduce is important for systems of inequalities describing polyhedra whose projections on some nonaxial lines are much smaller than their projections on the axes. However, there are systems for which LatticeReduce, instead of simplifying the problem, makes it significantly harder.

This finds the only two integer points in a triangle whose projections on both axes have sizes greater than but whose projection on the line has size one:
This sets the value of the system option LatticeReduceDiophantine to False:
The nondefault method is much slower for this system, and the speed difference grows with :
Here is a system that contains a set of simple inequalities , which bound solutions to a reasonably small size polyhedron, combined with a set of relatively complicated inequalities . For such systems, using LatticeReduce tends to increase the timing:
The nondefault method is faster for this system:
This resets LatticeReduceDiophantine to the default value:
MaxFrobeniusGraph

The system option MaxFrobeniusGraph specifies the maximal size of the smallest coefficient in a Frobenius equation for which FindInstance uses an algorithm based on the computation of the critical tree in the Frobenius graph [11]. Otherwise, the more general methods for solving bounded linear systems are used. Unlike the general method for solving bounded linear systems, the method based on the computation of the Frobenius graph depends very little on the number of variables, hence it is the faster choice for equations with many variables. On the other hand, the method requires storing a graph of the size of the smallest coefficient, so for large coefficients it may run out of memory.

To find a solution of a Frobenius equation with the smallest coefficient larger than , FindInstance by default uses the general method for solving bounded linear systems. For this example the method is relatively slow but uses little memory. The kernel has been restarted to show the memory usage by the current example:
This increases the value of MaxFrobeniusGraph to :
Now FindInstance uses the method based on the computation of the Frobenius graph. It finds the solution faster, but it uses more memory:
This resets MaxFrobeniusGraph to the default value:
SieveMaxPoints

The system option SieveMaxPoints specifies the maximal number of search points used by the modular sieve method and by searches used in solving equations with a linear variable. The value of the option is a pair . Search involving up to points is tried before some of the slower solving methods. For decision problems, another search involving up to points is tried after all other methods fail. The default value of the option is .

With the default setting of SieveMaxPoints, FindInstance is unable to find a solution for the first equation or show that the second equation has no solutions:
Increasing the second element of SieveMaxPoints to enables FindInstance to solve the problems:
Reduce makes a second call to the modular sieve, with an increased limit on the number of points, only for decision problems:
Increasing the first element of SieveMaxPoints to enables Reduce to solve the problem completely:
This resets SieveMaxPoints to the default value:

References

[1] Goldbach, C. Letter to L. Euler, June 7, 1742. http://mathworld.wolfram.com/GoldbachConjecture.html. [online version]
[2] Matiyasevich, Yu. V. "The Diophantineness of Enumerable Sets (Russian)" Dokl. Akad. Nauk SSSR 191 (1970): 279282. English translation: Soviet Math. Dokl. 11 (1970): 354358.
[3] Contejean, E. and H. Devie. "An Efficient Incremental Algorithm for Solving Systems of Linear Diophantine Equations." Information and Computation 113 (1994): 143172.
[4] Cucker, F., P. Koiran, and S. Smale. "A Polynomial Time Algorithm for Diophantine Equations in One Variable." Journal of Symbolic Computation 27, no. 1 (1999): 2130.
[5] Strzebonski, A. "An Improved Algorithm for Diophantine Equations in One Variable." Paper presented at the ACA 2002 Session on Symbolic-Numerical Methods in Computational Science, Volos, Greece. Notebook with the conference talk available at members.wolfram.com/adams.
[6] Dickson, L. E. History of the Theory of Numbers. Chelsea, 1952.
[7] Nagell, T. Introduction to Number Theory. Wiley, 1951.
[8] Hardy, K., J. B. Muskat and K. S. Williams. "A Deterministic Algorithm for Solving in Coprime Integers and ." Mathematics of Computation 55 (1990): 327343.
[9] Smart, N. The Algorithmic Resolution of Diophantine Equations. Cambridge University Press, 1998.
[10] Bressoud, D. M. and S. Wagon. A Course in Computational Number Theory. Key College Publishing, 2000.
[11] Beihoffer, D. E., J. Hendry, A. Nijenhuis, and S. Wagon. "Faster Algorithms for Frobenius Numbers." to appear in The Electronic Journal of Combinatorics.
[12] Rabin, M. O. and J. O. Shallit. "Randomized Algorithms in Number Theory." Communications on Pure and Applied Mathematics 39 (1986): 239256.