15 Number Theory GAP provides a couple of elementary number theoretic functions. Most of these deal with the group of integers coprime to m, called the prime residue group. The order of this group is ϕ(m) (see Phi (15.2-2)), and λ(m) (see Lambda (15.2-3)) is its exponent. This group is cyclic if and only if m is 2, 4, an odd prime power p^n, or twice an odd prime power 2 p^n. In this case the generators of the group, i.e., elements of order ϕ(m), are called primitive roots (see PrimitiveRootMod (15.3-3)). Note that neither the arguments nor the return values of the functions listed below are groups or group elements in the sense of GAP. The arguments are simply integers. 15.1 InfoNumtheor (Info Class) 15.1-1 InfoNumtheor InfoNumtheor  info class InfoNumtheor is the info class (see 7.4) for the functions in the number theory chapter. 15.2 Prime Residues 15.2-1 PrimeResidues PrimeResidues( m )  function PrimeResidues returns the set of integers from the range [ 0 .. Abs( m )-1 ] that are coprime to the integer m. Abs(m) must be less than 2^28, otherwise the set would probably be too large anyhow.  Example  gap> PrimeResidues( 0 ); PrimeResidues( 1 ); PrimeResidues( 20 ); [ ] [ 0 ] [ 1, 3, 7, 9, 11, 13, 17, 19 ]  15.2-2 Phi Phi( m )  operation Phi returns the number ϕ(m) of positive integers less than the positive integer m that are coprime to m. Suppose that m = p_1^{e_1} p_2^{e_2} ⋯ p_k^{e_k}. Then ϕ(m) is p_1^{e_1-1} (p_1-1) p_2^{e_2-1} (p_2-1) ⋯ p_k^{e_k-1} (p_k-1).  Example  gap> Phi( 12 ); 4 gap> Phi( 2^13-1 ); # this proves that 2^(13)-1 is a prime 8190 gap> Phi( 2^15-1 ); 27000  15.2-3 Lambda Lambda( m )  operation Lambda returns the exponent λ(m) of the group of prime residues modulo the integer m. λ(m) is the smallest positive integer l such that for every a relatively prime to m we have a^l ≡ 1 mod m. Fermat's theorem asserts a^{ϕ(m)} ≡ 1 mod m; thus λ(m) divides ϕ(m) (see Phi (15.2-2)). Carmichael's theorem states that λ can be computed as follows: λ(2) = 1, λ(4) = 2 and λ(2^e) = 2^{e-2} if 3 ≤ e, λ(p^e) = (p-1) p^{e-1} (i.e. ϕ(m)) if p is an odd prime and λ(m*n) =Lcm( λ(m), λ(n) ) if m, n are coprime. Composites for which λ(m) divides m - 1 are called Carmichaels. If 6k+1, 12k+1 and 18k+1 are primes their product is such a number. There are only 1547 Carmichaels below 10^10 but 455052511 primes.  Example  gap> Lambda( 10 ); 4 gap> Lambda( 30 ); 4 gap> Lambda( 561 ); # 561 is the smallest Carmichael number 80  15.2-4 GeneratorsPrimeResidues GeneratorsPrimeResidues( n )  function Let n be a positive integer. GeneratorsPrimeResidues returns a description of generators of the group of prime residues modulo n. The return value is a record with components primes:  a list of the prime factors of n, exponents:  a list of the exponents of these primes in the factorization of n, and generators:  a list describing generators of the group of prime residues; for the prime factor 2, either a primitive root or a list of two generators is stored, for each other prime factor of n, a primitive root is stored.  Example  gap> GeneratorsPrimeResidues( 1 ); rec( exponents := [ ], generators := [ ], primes := [ ] ) gap> GeneratorsPrimeResidues( 4*3 ); rec( exponents := [ 2, 1 ], generators := [ 7, 5 ],   primes := [ 2, 3 ] ) gap> GeneratorsPrimeResidues( 8*9*5 ); rec( exponents := [ 3, 2, 1 ],   generators := [ [ 271, 181 ], 281, 217 ], primes := [ 2, 3, 5 ] )  15.3 Primitive Roots and Discrete Logarithms 15.3-1 OrderMod OrderMod( n, m[, bound] )  function OrderMod returns the multiplicative order of the integer n modulo the positive integer m. If n and m are not coprime the order of n is not defined and OrderMod will return 0. If n and m are relatively prime the multiplicative order of n modulo m is the smallest positive integer i such that n^i ≡ 1 mod m. If the group of prime residues modulo m is cyclic then each element of maximal order is called a primitive root modulo m (see IsPrimitiveRootMod (15.3-4)). If no a priori known multiple bound of the desired order is given, OrderMod usually spends most of its time factoring m for computing λ(m) (see Lambda (15.2-3)) as the default for bound, and then factoring bound (see FactorsInt (14.4-7)). If an incorrect bound is given then the result will be wrong.  Example  gap> OrderMod( 2, 7 ); 3 gap> OrderMod( 3, 7 ); # 3 is a primitive root modulo 7 6 gap> m:= (5^166-1) / 167;; # about 10^113 gap> OrderMod( 5, m, 166 ); # needs minutes without third argument 166  15.3-2 LogMod LogMod( n, r, m )  function LogModShanks( n, r, m )  function computes the discrete r-logarithm of the integer n modulo the integer m. It returns a number l such that r^l ≡ n mod m if such a number exists. Otherwise fail is returned. LogModShanks uses the Baby Step - Giant Step Method of Shanks (see for example [Coh93, section 5.4.1]) and in general requires more memory than a call to LogMod.  Example  gap> l:= LogMod( 2, 5, 7 ); 5^l mod 7 = 2; 4 true gap> LogMod( 1, 3, 3 ); LogMod( 2, 3, 3 ); 0 fail  15.3-3 PrimitiveRootMod PrimitiveRootMod( m[, start] )  function PrimitiveRootMod returns the smallest primitive root modulo the positive integer m and fail if no such primitive root exists. If the optional second integer argument start is given PrimitiveRootMod returns the smallest primitive root that is strictly larger than start.  Example  gap> # largest primitive root for a prime less than 2000: gap> PrimitiveRootMod( 409 );  21 gap> PrimitiveRootMod( 541, 2 ); 10 gap> # 327 is the largest primitive root mod 337: gap> PrimitiveRootMod( 337, 327 ); fail gap> # there exists no primitive root modulo 30: gap> PrimitiveRootMod( 30 ); fail  15.3-4 IsPrimitiveRootMod IsPrimitiveRootMod( r, m )  function IsPrimitiveRootMod returns true if the integer r is a primitive root modulo the positive integer m, and false otherwise. If r is less than 0 or larger than m it is replaced by its remainder.  Example  gap> IsPrimitiveRootMod( 2, 541 ); true gap> IsPrimitiveRootMod( -539, 541 ); # same computation as above; true gap> IsPrimitiveRootMod( 4, 541 ); false gap> ForAny( [1..29], r -> IsPrimitiveRootMod( r, 30 ) ); false gap> # there is no a primitive root modulo 30  15.4 Roots Modulo Integers 15.4-1 Jacobi Jacobi( n, m )  function Jacobi returns the value of the Kronecker-Jacobi symbol J(n,m) of the integer n modulo the integer m. It is defined as follows: If n and m are not coprime then J(n,m) = 0. Furthermore, J(n,1) = 1 and J(n,-1) = -1 if m < 0 and +1 otherwise. And for odd n it is J(n,2) = (-1)^k with k = (n^2-1)/8. For odd primes m which are coprime to n the Kronecker-Jacobi symbol has the same value as the Legendre symbol (see Legendre (15.4-2)). For the general case suppose that m = p_1 ⋅ p_2 ⋯ p_k is a product of -1 and of primes, not necessarily distinct, and that n is coprime to m. Then J(n,m) = J(n,p_1) ⋅ J(n,p_2) ⋯ J(n,p_k). Note that the Kronecker-Jacobi symbol coincides with the Jacobi symbol that is defined for odd m in many number theory books. For odd primes m and n coprime to m it coincides with the Legendre symbol. Jacobi is very efficient, even for large values of n and m, it is about as fast as the Euclidean algorithm (see Gcd (56.7-1)).  Example  gap> Jacobi( 11, 35 ); # 9^2 = 11 mod 35 1 gap> # this is -1, thus there is no r such that r^2 = 6 mod 35 gap> Jacobi( 6, 35 ); -1 gap> # this is 1 even though there is no r with r^2 = 3 mod 35 gap> Jacobi( 3, 35 ); 1  15.4-2 Legendre Legendre( n, m )  function Legendre returns the value of the Legendre symbol of the integer n modulo the positive integer m. The value of the Legendre symbol L(n/m) is 1 if n is a quadratic residue modulo m, i.e., if there exists an integer r such that r^2 ≡ n mod m and -1 otherwise. If a root of n exists it can be found by RootMod (15.4-3). While the value of the Legendre symbol usually is only defined for m a prime, we have extended the definition to include composite moduli too. The Jacobi symbol (see Jacobi (15.4-1)) is another generalization of the Legendre symbol for composite moduli that is much cheaper to compute, because it does not need the factorization of m (see FactorsInt (14.4-7)). A description of the Jacobi symbol, the Legendre symbol, and related topics can be found in [Bak84].  Example  gap> Legendre( 5, 11 ); # 4^2 = 5 mod 11 1 gap> # this is -1, thus there is no r such that r^2 = 6 mod 11 gap> Legendre( 6, 11 ); -1 gap> # this is -1, thus there is no r such that r^2 = 3 mod 35 gap> Legendre( 3, 35 ); -1  15.4-3 RootMod RootMod( n[, k], m )  function RootMod computes a kth root of the integer n modulo the positive integer m, i.e., a r such that r^k ≡ n mod m. If no such root exists RootMod returns fail. If only the arguments n and m are given, the default value for k is 2. A square root of n exists only if Legendre(n,m) = 1 (see Legendre (15.4-2)). If m has r different prime factors then there are 2^r different roots of n mod m. It is unspecified which one RootMod returns. You can, however, use RootsMod (15.4-4) to compute the full set of roots. RootMod is efficient even for large values of m, in fact the most time is usually spent factoring m (see FactorsInt (14.4-7)).  Example  gap> # note 'RootMod' does not return 8 in this case but -8: gap> RootMod( 64, 1009 ); 1001 gap> RootMod( 64, 3, 1009 ); 518 gap> RootMod( 64, 5, 1009 ); 656 gap> List( RootMod( 64, 1009 ) * RootsUnityMod( 1009 ), >  x -> x mod 1009 ); # set of all square roots of 64 mod 1009 [ 1001, 8 ]  15.4-4 RootsMod RootsMod( n[, k], m )  function RootsMod computes the set of kth roots of the integer n modulo the positive integer m, i.e., the list of all r such that r^k ≡ n mod m. If only the arguments n and m are given, the default value for k is 2.  Example  gap> RootsMod( 1, 7*31 ); # the same as `RootsUnityMod( 7*31 )' [ 1, 92, 125, 216 ] gap> RootsMod( 7, 7*31 ); [ 21, 196 ] gap> RootsMod( 5, 7*31 ); [ ] gap> RootsMod( 1, 5, 7*31 ); [ 1, 8, 64, 78, 190 ]  15.4-5 RootsUnityMod RootsUnityMod( [k, ]m )  function RootsUnityMod returns the set of k-th roots of unity modulo the positive integer m, i.e., the list of all solutions r of r^k ≡ n mod m. If only the argument m is given, the default value for k is 2. In general there are k^n such roots if the modulus m has n different prime factors p such that p ≡ 1 mod k. If k^2 divides m then there are k^{n+1} such roots; and especially if k = 2 and 8 divides m there are 2^{n+2} such roots. In the current implementation k must be a prime.  Example  gap> RootsUnityMod( 7*31 ); RootsUnityMod( 3, 7*31 ); [ 1, 92, 125, 216 ] [ 1, 25, 32, 36, 67, 149, 156, 191, 211 ] gap> RootsUnityMod( 5, 7*31 ); [ 1, 8, 64, 78, 190 ] gap> List( RootMod( 64, 1009 ) * RootsUnityMod( 1009 ), >  x -> x mod 1009 ); # set of all square roots of 64 mod 1009 [ 1001, 8 ]  15.5 Multiplicative Arithmetic Functions 15.5-1 Sigma Sigma( n )  operation Sigma returns the sum of the positive divisors of the nonzero integer n. Sigma is a multiplicative arithmetic function, i.e., if n and m are relatively prime we have that σ(n ⋅ m) = σ(n) σ(m). Together with the formula σ(p^k) = (p^{k+1}-1) / (p-1) this allows us to compute σ(n). Integers n for which σ(n) = 2 n are called perfect. Even perfect integers are exactly of the form 2^{n-1}(2^n-1) where 2^n-1 is prime. Primes of the form 2^n-1 are called Mersenne primes, and 42 among the known Mersenne primes are obtained for n = 2, 3, 5, 7, 13, 17, 19, 31, 61, 89, 107, 127, 521, 607, 1279, 2203, 2281, 3217, 4253, 4423, 9689, 9941, 11213, 19937, 21701, 23209, 44497, 86243, 110503, 132049, 216091, 756839, 859433, 1257787, 1398269, 2976221, 3021377, 6972593, 13466917, 20996011, 24036583 and 25964951. Please find more up to date information about Mersenne primes at https://www.mersenne.org. It is not known whether odd perfect integers exist, however [BC89] show that any such integer must have at least 300 decimal digits. Sigma usually spends most of its time factoring n (see FactorsInt (14.4-7)).  Example  gap> Sigma( 1 ); 1 gap> Sigma( 1009 ); # 1009 is a prime 1010 gap> Sigma( 8128 ) = 2*8128; # 8128 is a perfect number true  15.5-2 Tau Tau( n )  operation Tau returns the number of the positive divisors of the nonzero integer n. Tau is a multiplicative arithmetic function, i.e., if n and m are relative prime we have τ(n ⋅ m) = τ(n) τ(m). Together with the formula τ(p^k) = k+1 this allows us to compute τ(n). Tau usually spends most of its time factoring n (see FactorsInt (14.4-7)).  Example  gap> Tau( 1 ); 1 gap> Tau( 1013 ); # thus 1013 is a prime 2 gap> Tau( 8128 ); 14 gap> # result is odd if and only if argument is a perfect square: gap> Tau( 36 ); 9  15.5-3 MoebiusMu MoebiusMu( n )  function MoebiusMu computes the value of Moebius inversion function for the nonzero integer n. This is 0 for integers which are not squarefree, i.e., which are divided by a square r^2. Otherwise it is 1 if n has a even number and -1 if n has an odd number of prime factors. The importance of μ stems from the so called inversion formula. Suppose f is a multiplicative arithmetic function defined on the positive integers and let g(n) = ∑_{d ∣ n} f(d). Then f(n) = ∑_{d ∣ n} μ(d) g(n/d). As a special case we have ϕ(n) = ∑_{d ∣ n} μ(d) n/d since n = ∑_{d ∣ n} ϕ(d) (see Phi (15.2-2)). MoebiusMu usually spends all of its time factoring n (see FactorsInt (14.4-7)).  Example  gap> MoebiusMu( 60 ); MoebiusMu( 61 ); MoebiusMu( 62 ); 0 -1 1  15.6 Continued Fractions 15.6-1 ContinuedFractionExpansionOfRoot ContinuedFractionExpansionOfRoot( f, n )  function The first n terms of the continued fraction expansion of the only positive real root of the polynomial f with integer coefficients. The leading coefficient of f must be positive and the value of f at 0 must be negative. If the degree of f is 2 and n = 0, the function computes one period of the continued fraction expansion of the root in question. Anything may happen if f has three or more positive real roots.  Example  gap> x := Indeterminate(Integers);; gap> ContinuedFractionExpansionOfRoot(x^2-7,20); [ 2, 1, 1, 1, 4, 1, 1, 1, 4, 1, 1, 1, 4, 1, 1, 1, 4, 1, 1, 1 ] gap> ContinuedFractionExpansionOfRoot(x^2-7,0); [ 2, 1, 1, 1, 4 ] gap> ContinuedFractionExpansionOfRoot(x^3-2,20); [ 1, 3, 1, 5, 1, 1, 4, 1, 1, 8, 1, 14, 1, 10, 2, 1, 4, 12, 2, 3 ] gap> ContinuedFractionExpansionOfRoot(x^5-x-1,50); [ 1, 5, 1, 42, 1, 3, 24, 2, 2, 1, 16, 1, 11, 1, 1, 2, 31, 1, 12, 5,   1, 7, 11, 1, 4, 1, 4, 2, 2, 3, 4, 2, 1, 1, 11, 1, 41, 12, 1, 8, 1,   1, 1, 1, 1, 9, 2, 1, 5, 4 ]  15.6-2 ContinuedFractionApproximationOfRoot ContinuedFractionApproximationOfRoot( f, n )  function The nth continued fraction approximation of the only positive real root of the polynomial f with integer coefficients. The leading coefficient of f must be positive and the value of f at 0 must be negative. Anything may happen if f has three or more positive real roots.  Example  gap> ContinuedFractionApproximationOfRoot(x^2-2,10); 3363/2378 gap> 3363^2-2*2378^2; 1 gap> z := ContinuedFractionApproximationOfRoot(x^5-x-1,20); 499898783527/428250732317 gap> z^5-z-1; 486192462527432755459620441970617283/ 14404247382319842421697357558805709031116987826242631261357  15.7 Miscellaneous 15.7-1 PValuation PValuation( n, p )  function For an integer n and a prime p this function returns the p-valuation of n, that is the exponent e such that p^e is the largest power of p that divides n. The valuation of zero is infinity.  Example  gap> PValuation(100,2); 2 gap> PValuation(100,3); 0  15.7-2 TwoSquares TwoSquares( n )  function TwoSquares returns a list of two integers x ≤ y such that the sum of the squares of x and y is equal to the nonnegative integer n, i.e., n = x^2 + y^2. If no such representation exists TwoSquares will return fail. TwoSquares will return a representation for which the gcd of x and y is as small as possible. It is not specified which representation TwoSquares returns if there is more than one. Let a be the product of all maximal powers of primes of the form 4k+3 dividing n. A representation of n as a sum of two squares exists if and only if a is a perfect square. Let b be the maximal power of 2 dividing n or its half, whichever is a perfect square. Then the minimal possible gcd of x and y is the square root c of a ⋅ b. The number of different minimal representation with x ≤ y is 2^{l-1}, where l is the number of different prime factors of the form 4k+1 of n. The algorithm first finds a square root r of -1 modulo n / (a ⋅ b), which must exist, and applies the Euclidean algorithm to r and n. The first residues in the sequence that are smaller than sqrt{n/(a ⋅ b)} times c are a possible pair x and y. Better descriptions of the algorithm and related topics can be found in [Wag90] and [Zag90].  Example  gap> TwoSquares( 5 ); [ 1, 2 ] gap> TwoSquares( 11 ); # there is no representation fail gap> TwoSquares( 16 ); [ 0, 4 ] gap> # 3 is the minimal possible gcd because 9 divides 45: gap> TwoSquares( 45 ); [ 3, 6 ] gap> # it is not [5,10] because their gcd is not minimal: gap> TwoSquares( 125 ); [ 2, 11 ] gap> # [10,11] would be the other possible representation: gap> TwoSquares( 13*17 ); [ 5, 14 ] gap> TwoSquares( 848654483879497562821 ); # argument is prime [ 6305894639, 28440994650 ]