--- title: 'Multiplicative Primitivity: Theory & Applications' url: https://www.emergentmind.com/topics/multiplicative-primitivity type: topic --- # Multiplicative Primitivity: Theory & Applications Multiplicative primitivity is a context-dependent notion describing when finite multiplicative operations attain a maximal reachability property. In elementary number theory, it is realized by a primitive root \(g\) modulo a prime \(p\), namely an element of \((\mathbb Z/p\mathbb Z)^\times\) with order \(p-1\), so that every nonzero residue is a power of \(g\) [2205.11694]. In matrix theory, a tuple or set of nonnegative matrices is multiplicatively primitive when some finite product, or an appropriate Hurwitz product, is entrywise positive [2402.18586, 1306.0729]. In quantum information, the term appears both in the classical primitivity theory of channels and in a stronger notion for randomized quantum trajectories, where repeated left-multiplication by Kraus operators must generate a spanning set from every pure state [2109.01340, 2603.28664]. Across these settings, the common theme is that multiplication eventually eliminates local obstructions such as periodicity, reducibility, or nongenerating behavior. ## 1. Primitive roots and multiplicative order For a prime \(p\), the ring of integers modulo \(p\) is \[ \mathbb Z/p\mathbb Z=\{0,1,2,\dots,p-1\}, \] and its nonzero residues \[ (\mathbb Z/p\mathbb Z)^\times=\{1,2,\dots,p-1\} \] form a group under multiplication modulo \(p\) [2205.11694]. If \(a\in(\mathbb Z/p\mathbb Z)^\times\), its order is \[ \operatorname{ord}_p(a)=\min\{k\ge 1: a^k\equiv 1 \pmod p\}, \] and Fermat’s little theorem guarantees that this minimum exists and satisfies \(\operatorname{ord}_p(a)\le p-1\) [2205.11694]. A primitive root modulo \(p\) is an element \(g\) of order \(p-1\), equivalently an element for which \[ \{g^1,g^2,\dots,g^{p-1}\}\pmod p=(\mathbb Z/p\mathbb Z)^\times \] [2205.11694]. Several standard structural facts govern this notion. The order of any element divides \(p-1\); \(\operatorname{ord}_p(a^{-1})=\operatorname{ord}_p(a)\); and if \(\operatorname{ord}_p(a)=m\) and \(\operatorname{ord}_p(b)=n\) with \(\gcd(m,n)=1\), then \(\operatorname{ord}_p(ab)=mn\) [2205.11694]. In the broader setting of a finite cyclic group \(G\) with identity \(e\), the multiplicative order of \(a\in G\) is the least positive \(n\) such that \(a^n=e\), and an element is a generator precisely when its order is \(|G|\) [1408.4942]. For general moduli \(n>1\), the unit group is \(U_n=(\mathbb Z/n\mathbb Z)^\times\). An element \(a\in U_n\) is a primitive root modulo \(n\) when its index, or order, \[ \operatorname{ind}_n(a)=\min\{k\ge 1:a^k\equiv 1\pmod n\}, \] equals \(\varphi(n)\) [1911.08176]. This recovers the prime case when \(n=p\), but the data also emphasize order classes more generally: one studies \[ S_d(n)=\sum_{\substack{a\in U_n\\ \operatorname{ind}_n(a)=d}} a, \qquad P_d(n)=\prod_{\substack{a\in U_n\\ \operatorname{ind}_n(a)=d}} a, \] so that primitive roots correspond to the special case \(d=\varphi(n)\) [1911.08176]. ## 2. Constructive existence and algorithmic search A central constructive result is that every prime number has a primitive root [2205.11694]. The proof begins by factoring \[ p-1=\prod_{i=1}^t q_i^{e_i}, \] where the \(q_i\) are distinct primes. For each prime-power divisor \(q^e\), one considers the congruence \(x^{q^e}-1\equiv 0 \pmod p\). Using the fact that a nonzero polynomial of degree \(n\) over a field has at most \(n\) roots, together with the splitting behavior of \(x^{p-1}-1\), one shows that \(x^{q^e}-1\) has exactly \(q^e\) distinct roots in \((\mathbb Z/p\mathbb Z)^\times\), while \(x^{q^{e-1}}-1\) has exactly \(q^{e-1}\) roots. Hence there exists \(\alpha\) such that \(\alpha^{q^e}\equiv 1\) but \(\alpha^{q^{e-1}}\not\equiv 1\), forcing \(\operatorname{ord}_p(\alpha)=q^e\) [2205.11694]. Once elements \(c_i\) with \(\operatorname{ord}_p(c_i)=q_i^{e_i}\) are obtained, one forms \[ g=\prod_{i=1}^t c_i \pmod p. \] Because the orders \(q_i^{e_i}\) are pairwise relatively prime, the coprime-order lemma gives \[ \operatorname{ord}_p(g)=\prod_i q_i^{e_i}=p-1, \] so \(g\) is a primitive root [2205.11694]. The associated search routine is explicit: `findRootModP(q,n,p)` scans \(x=1,2,\dots,p-1\) until it finds an \(x\) satisfying \[ x^{q^n}\equiv 1 \pmod p, \qquad x^{q^{n-1}}\not\equiv 1 \pmod p, \] and `primitiveRootAux` recursively combines such witnesses [2205.11694]. The constructive character of the method is matched by explicit complexity remarks. The basic implementation factors \(p-1\) by trial division up to \(\sqrt{p-1}\), performs a linear scan of at most \(p\) candidates for each prime-power factor, computes two modular exponentiations at each step by repeated squaring, and multiplies the resulting witnesses. The stated overall cost is polynomial in \(p\), “actually about \(O(\sqrt p + tp\log p)\)” [2205.11694]. When the complete factorization \(p-1=\prod_{i=1}^k p_i^{e_i}\) is known, more refined order and primitive-root algorithms are available [1408.4942]. The classical multiplicative-order and randomized primitive-root procedures require \(O(k)\) modular exponentiations, for total bit complexity \[ O(k\cdot (\log p)^3). \] Using the paper’s K-Exponentiation subroutine, which computes the complement exponents \(a^{(p-1)/p_i^{e_i}}\) in a balanced binary-tree fashion, the dependence on \(k\) is reduced from linear to logarithmic. The modified multiplicative-order and primitive-root algorithms run in expected time \[ O(\log k \cdot (\log p)^3) \] [1408.4942]. The worked examples \(p=7\) and \(p=11\) in both papers illustrate the same underlying principle: factor \(p-1\), produce elements of the relevant prime-power orders, and combine them into a generator [2205.11694, 1408.4942]. ## 3. Arithmetic extensions: finite fields, fixed-order classes, and near-primitive roots The primitive-root paradigm extends from \((\mathbb Z/p\mathbb Z)^\times\) to finite fields. For a prime \(p\) and integer \(n>0\), the multiplicative group \[ \mathbb F_{p^n}^*=\mathbb F_{p^n}\setminus\{0\} \] is cyclic of order \(p^n-1\), and a primitive element is an \(\alpha\in\mathbb F_{p^n}^*\) whose order is exactly \(p^n-1\) [1304.1206]. Huang and Narayanan describe a deterministic algorithm which, in time polynomial in \(p\) and \(n\), either outputs an element that is provably a generator or declares failure. The method embeds \(\mathbb F_{p^n}\) into a specially chosen extension \(K=\mathbb F_{q^2}[X]/(g(X))\), generates multiplicative relations on a factor base using a variant of Joux’s relation-generation technique, computes a Smith normal form for the resulting relation lattice, and extracts an element of order divisible by almost all of \(q^{2m}-1\); exponentiation then yields a generator of \(\mathbb F_{p^n}^*\) [1304.1206]. Another extension studies not only primitive roots but all units of a fixed order \(d\mid \lambda(n)\). Zhong and Cai prove that \[ P_d(n)\equiv \begin{cases} -1 \pmod n, & d=2\text{ and }U_n\text{ is cyclic},\\ 1 \pmod n, & \text{otherwise}, \end{cases} \] and derive the general identity \[ S_d(n)=\sum_{e\mid d}\mu\!\left(\frac de\right)\sum_{\substack{a\in U_n\\ a^e\equiv 1\,(n)}} a \] for the corresponding sums [1911.08176]. In particular, \(S_1(n)=1\), \(S_2(n)\equiv -1 \pmod n\) for \(n>2\), and \(S_d(n)\equiv 0 \pmod n\) whenever \(4\mid d\) [1911.08176]. This places primitive roots inside a broader stratification of \(U_n\) by multiplicative order. A different arithmetic relaxation replaces exact generation by large order. For a prime \(p\), an integer \(a\) is termed a near-primitive root when \(\operatorname{ord}_p(a)\gg p^\delta\) for some \(\delta<1\) [2006.15200]. Agrawal and Pollack show that if \(a,b\) are multiplicatively independent, then for almost all primes \(p\), at least one of \[ S=\{a,\ b,\ ab,\ a^2b,\ ab^2\} \] satisfies \[ \operatorname{ord}_p(g)>p^{8/15}/\exp(2\sqrt{\log p}) \] [2006.15200]. More generally, for multiplicatively independent integers \(a_1,\dots,a_k\) and \[ A_N=\{a_1^{e_1}\cdots a_k^{e_k}:0\le e_ip^\delta, \qquad \delta=\left(1-\frac1{k+1}\right)\left(1-\frac1N\right), \] and hence for every \(\epsilon>0\) one can construct an explicit finite set \(A_\epsilon\) of size \[ |A_\epsilon|=\exp(O((1/\epsilon)\log(1/\epsilon))) \] such that for almost all \(p\) some element of \(A_\epsilon\) has order exceeding \(p^{1-\epsilon}\) [2006.15200]. This suggests that multiplicative primitivity in arithmetic is often studied both in exact and approximate forms. ## 4. Matrix products, \(k\)-primitivity, and complexity For a single nonnegative matrix \(A\), classical primitivity means that \(A^m>0\) entrywise for some positive integer \(m\), and the least such \(m\) is the exponent of \(A\) [2402.18586]. A direct multiplicative generalization considers a finite set \[ \mathcal M=\{A_1,A_2,\dots,A_m\}\subset \mathbb R_{\ge 0}^{n\times n} \] and declares it primitive when there exists a product \[ A_{i_1}A_{i_2}\cdots A_{i_k}>0. \] The length of the shortest positive product is \[ \ell^*(\mathcal M)=\min\{k>0:\exists\, i_1,\dots,i_k \text{ with }A_{i_1}\cdots A_{i_k}>0\} \] [1306.0729]. The survey on \(k\)-primitivity formulates the notion for tuples \(\mathcal A=(A_1,\dots,A_k)\) of nonzero nonnegative matrices using Hurwitz products [2402.18586]. For a multi-index \((m_1,\dots,m_k)\), the \((m_1,\dots,m_k)\)-Hurwitz product is the sum of all words containing exactly \(m_i\) copies of \(A_i\), and \(\mathcal A\) is \(k\)-primitive if there exists \((r_1,\dots,r_k)\), not all zero, such that \[ (A_1,\dots,A_k)^{(r_1,\dots,r_k)}>0. \] Equivalently, there is at least one word of length \(r_1+\cdots+r_k\) whose ordinary product is entrywise positive. The corresponding exponent is \[ \exp(\mathcal A)=\min\{r_1+\cdots+r_k:(A_1,\dots,A_k)^{(r_1,\dots,r_k)}>0\} \] [2402.18586]. Graph-theoretic formulations are central in both accounts. For a single matrix, primitivity is equivalent to irreducibility together with the condition that the greatest common divisor of all cycle lengths in its digraph equals \(1\) [2402.18586]. For a \(k\)-tuple, one forms a \(k\)-colored multidigraph \(D^k\) on \(\{1,\dots,n\}\), placing a color-\(i\) arc \(u\to v\) whenever \((A_i)_{u,v}>0\). Then \(\mathcal A\) is \(k\)-primitive if and only if for every ordered pair \((u,v)\) there exists a directed walk from \(u\) to \(v\) using exactly \(r_i\) arcs of color \(i\) for some fixed positive integers \(r_1,\dots,r_k\) [2402.18586]. In the language of sets of matrices, the path-product lemma asserts that \((A_1\cdots A_k)_{ij}>0\) exactly when there is a path from \(i\) to \(j\) in the sequence of digraphs \(G_1,\dots,G_k\) [1306.0729]. Known bounds show both classical continuity with Perron–Frobenius theory and genuinely new asymptotic behavior. For one matrix, Wielandt’s bound gives \[ \exp(A)\le (n-1)^2+1. \] For \(k=2\), Shader and Suwilo proved \[ \exp(A,B)\le (3n^3+2n^2-2n)/2. \] More generally, the maximal exponent of a primitive \(k\)-tuple in \(M_n(\mathbb R)\) grows as \(\Theta(n^{k+1})\) for fixed \(k\) [2402.18586]. Algorithmically, the situation bifurcates. Protasov’s dichotomy yields a polynomial-time test for \(k\)-primitivity under mild full-support assumptions, and consequently deciding \(k\)-primitivity is in \(P\) for fixed \(k\) in the usual bit model [2402.18586]. By contrast, for general finite sets of matrices, primitivity is decidable but NP-hard as soon as the set contains three matrices; unless \(P=NP\), there is no polynomial-time algorithm for deciding it [1306.0729]. The shortest positive product can be superpolynomial in the dimension, with lower bounds of the form \[ \ell^*(\mathcal M)\ge \bigl((1-\epsilon)e\bigr)^{\sqrt{n_k/2}} \] for infinitely many dimensions \(n_k\) when \(m\ge 4\), while a general upper bound is \(2^{n^2}\) [1306.0729]. Under the additional assumption that each matrix has no zero row and no zero column, there is a polynomial-time characterization via Protasov–Voynov block obstructions, and every primitive family admits a positive product of length \(O(n^3)\); any bound on synchronizing automata immediately transfers to this setting [1306.0729]. ## 5. Quantum channels and operator-theoretic variants In quantum information theory, primitivity first appears for completely positive trace-preserving maps. A quantum channel \(\Phi:M_n(\mathbb C)\to M_n(\mathbb C)\) is primitive if there exists \(m\ge 1\) such that for every nonzero positive semidefinite \(X\), \[ \Phi^m(X)>0 \] in the sense of strict positive definiteness. The least such \(m\) is the index of primitivity \[ q(\Phi)=\min\{m:\Phi^m(\rho)>0\ \forall\,\rho\ge 0,\ \operatorname{Tr}\rho=1\} \] [2109.01340]. For a column-stochastic matrix \(P\), primitivity means that \(P^k>0\) entrywise for some \(k\), and the least such \(k\) is \[ p(P)=\min\{k:P^k_{ij}>0\ \forall\, i,j\} \] [2109.01340]. For entanglement breaking channels, these two notions are tightly linked. If \[ \Phi(\rho)=\sum_{k=1}^r \operatorname{Tr}(F_k\rho)\,R_k \] is a Holevo form, the associated stochastic matrix is \[ S=(s_{ij}),\qquad s_{ij}=\operatorname{Tr}(F_iR_j). \] The nonzero part of the spectrum of \(\Phi\) coincides, with multiplicity, with the nonzero spectrum of any such \(S\) [2109.01340]. Moreover, \(S\) primitive together with \(\sum_k R_k>0\) implies that \(\Phi\) is primitive; conversely, primitivity of \(\Phi\) forces \(\sum_k R_k>0\) and primitivity of \(S\). The indices satisfy \[ |q(\Phi)-p(S)|\le 1, \] and if the Holevo rank is \(r\), then classical Wielandt theory yields \[ q(\Phi)\le r^2-2r+3 \] [2109.01340]. A newer and stronger notion, explicitly named multiplicative primitivity, arises in the study of randomized quantum trajectories [2603.28664]. Given a Kraus decomposition \[ \Phi^*(\rho)=\sum_{i=1}^k v_i\,\rho\,v_i^*, \qquad \sum_i v_i^*v_i=\mathbb I_d, \] define \[ \mathcal V_p=\operatorname{span}\{v_{i_1}\cdots v_{i_p}:1\le i_j\le k\}\subset M_d(\mathbb C). \] For a unit vector \(x\in\mathbb C^d\), one recursively defines \(\mathcal V_1^p x\), and \(\Phi^*\) is multiplicatively primitive if for every pure-state ray \(\hat x=[x]\) there exists \(p\in\mathbb N\) such that \[ \mathcal V_1^p x=\mathbb C^d \] [2603.28664]. The comparison theorem is one-sided in general: \[ \text{positivity improving}\ \Longrightarrow\ \text{multiplicative primitivity}\ \Longrightarrow\ \text{primitive}. \] In dimension \(d=2\), primitivity implies multiplicative primitivity; for \(d\ge 3\), the strictness question remains open [2603.28664]. This strengthened condition is used to analyze invariant measures for randomized quantum trajectories. If \(\Phi^*\) is multiplicatively primitive and the randomization measure dominates Haar measure on the Kraus-unitary group, then the induced Markov chain on projective space is \(\phi\)-irreducible with respect to Fubini–Study volume, has a unique invariant measure \(\nu_{\rm inv}\), and satisfies \[ \nu_{\rm inv}\gg \nu_{\rm unif}. \] If almost every Kraus operator in the randomization is invertible, then \(\nu_{\rm inv}\sim \nu_{\rm unif}\); under irreducibility and non-singular randomization, a Wasserstein convergence theorem gives exponential mixing [2603.28664]. In this operator-theoretic usage, multiplicative primitivity is therefore an ergodicity condition stronger than ordinary primitivity but weaker than positivity improving. ## 6. Related formulations and current directions The phrase also appears in adjacent multiplicative-generation problems. In the cyclic group \((\mathbb Z/q\mathbb Z)^\times\), Walker studies sets \[ P_\eta=\{p\text{ prime}:p<\eta q\} \] and their product sets \[ P_\eta^{(k)}=\{p_1p_2\cdots p_k:p_i\in P_\eta\}. \] For \(\eta\gg q^{-1/4+\epsilon}\), there exists a constant \(k\) depending only on \(\epsilon\) such that \[ P_\eta^{(k)}=(\mathbb Z/q\mathbb Z)^\times, \] while \(P_1^{(2)}\) has density at least \(\frac1{64}(1+o(1))\) [1505.03328]. This is not the primitive-root problem, but it is a multiplicative covering theorem in the same ambient group. The terminology of primitivity also appears in free groups. For the free group \(F_k\), a word \(w\) is primitive when it belongs to some basis of \(F_k\), equivalently when \(\langle w\rangle\) is a free factor [1104.3991]. A second criterion is measure preservation of the word map \(w:G^k\to G\) for every finite group \(G\). The paper proves that primitivity implies measure preservation, and for \(k=2\) the two properties are equivalent [1104.3991]. This suggests an abstract parallel: in both free-group and quantum settings, primitivity can be reformulated through an induced probabilistic uniformity property. Open problems are explicit in several of the cited works. In \(k\)-primitivity, open directions include the Beasley–Kirkland conjecture for 3-colored tournaments and classification problems for strongly connected tournaments that are \(k\)-primitive but fail to be \((k+1)\)-primitive [2402.18586]. For entanglement breaking channels, the continuity of the channel–matrix correspondence, the relation between \(q(\Phi)\) and Holevo rank in realistic settings, extensions beyond the entanglement breaking case, and the tightness of the bound \(q(\Phi)\le r^2-2r+3\) are listed as open problems [2109.01340]. In randomized quantum trajectories, the main unresolved structural question stated in the data is whether multiplicative primitivity is strictly stronger than classical primitivity in dimensions \(d\ge 3\) [2603.28664]. A persistent misconception is that multiplicative primitivity names a single invariant notion. The literature instead uses the term for several rigorously defined but nonidentical reachability properties: maximal order in cyclic groups, positivity of matrix products, positivity or spanning under iteration of quantum channels, and basis-membership phenomena in free groups [2205.11694, 2402.18586, 2109.01340, 2603.28664, 1104.3991]. What unifies these usages is not a common formal definition but a common structural pattern: finite multiplication overcomes the relevant obstruction and yields full generation, full positivity, or full support. Source: https://www.emergentmind.com/topics/multiplicative-primitivity