In number theory, a branch of mathematics, the Carmichael function λ(n) of a positive integer n is the smallest positive integer m such that

${\displaystyle a^{m}\equiv 1{\pmod {n))}$

holds for every integer a coprime to n. In algebraic terms, λ(n) is the exponent of the multiplicative group of integers modulo n. As this is a finite abelian group, there must exist an element whose order equals the exponent, λ(n). Such an element is called a primitive λ-root modulo n.

The Carmichael function is named after the American mathematician Robert Carmichael who defined it in 1910.^[1] It is also known as Carmichael's λ function, the reduced totient function, and the least universal exponent function.

The following table compares the first 36 values of λ(n) (sequence A002322 in the OEIS) with Euler's totient function φ (in bold if they are different; the ns such that they are different are listed in OEIS: A033949).

Numerical examples

1. Carmichael's function at 5 is 4, λ(5) = 4, because for any number ${\displaystyle 0<a<5}$ coprime to 5, i.e. ${\displaystyle a\in \{1,2,3,4\}~,}$ there is ${\displaystyle a^{m}\equiv 1\,({\text{mod ))5)}$ with ${\displaystyle m=4,}$ namely, 1^1⋅4 = 1⁴ ≡ 1 (mod 5), 2⁴ = 16 ≡ 1 (mod 5), 3⁴ = 81 ≡ 1 (mod 5) and 4^2⋅2 = 16² ≡ 1² (mod 5). And this m = 4 is the smallest exponent with this property, because ${\displaystyle 2^{2}=4\not \equiv 1\,({\text{mod ))5)}$ (and ${\displaystyle 3^{2}=9\not \equiv 1\,({\text{mod ))5)}$ as well.)
   Moreover, Euler's totient function at 5 is 4, φ(5) = 4, because there are exactly 4 numbers less than and coprime to 5 (1, 2, 3, and 4). Euler's theorem assures that a⁴ ≡ 1 (mod 5) for all a coprime to 5, and 4 is the smallest such exponent. Both 2 and 3 are primitive λ-roots modulo 5 and also primitive roots modulo 5.
2. Carmichael's function at 8 is 2, λ(8) = 2, because for any number a coprime to 8, i.e. ${\displaystyle a\in \{1,3,5,7\}~,}$ it holds that a² ≡ 1 (mod 8). Namely, 1^1⋅2 = 1² ≡ 1 (mod 8), 3² = 9 ≡ 1 (mod 8), 5² = 25 ≡ 1 (mod 8) and 7² = 49 ≡ 1 (mod 8).
   Euler's totient function at 8 is 4, φ(8) = 4, because there are exactly 4 numbers less than and coprime to 8 (1, 3, 5, and 7). Moreover, Euler's theorem assures that a⁴ ≡ 1 (mod 8) for all a coprime to 8, but 4 is not the smallest such exponent. The primitive λ-roots modulo 8 are 3, 5, and 7. There are no primitive roots modulo 8.

Recurrence for λ(n)

The Carmichael lambda function of a prime power can be expressed in terms of the Euler totient. Any number that is not 1 or a prime power can be written uniquely as the product of distinct prime powers, in which case λ of the product is the least common multiple of the λ of the prime power factors. Specifically, λ(n) is given by the recurrence

${\displaystyle \lambda (n)={\begin{cases}\varphi (n)&{\text{if ))n{\text{ is 1, 2, 4, or an odd prime power,))\\{\tfrac {1}{2))\varphi (n)&{\text{if ))n=2^{r},\ r\geq 3,\\\operatorname {lcm} {\Bigl (}\lambda (n_{1}),\lambda (n_{2}),\ldots ,\lambda (n_{k}){\Bigr )}&{\text{if ))n=n_{1}n_{2}\ldots n_{k}{\text{ where ))n_{1},n_{2},\ldots ,n_{k}{\text{ are powers of distinct primes.))\end{cases))}$

Euler's totient for a prime power, that is, a number p^r with p prime and r ≥ 1, is given by

${\displaystyle \varphi (p^{r}){=}p^{r-1}(p-1).}$

Carmichael's theorems

Carmichael proved two theorems that, together, establish that if λ(n) is considered as defined by the recurrence of the previous section, then it satisfies the property stated in the introduction, namely that it is the smallest positive integer m such that ${\displaystyle a^{m}\equiv 1{\pmod {n))}$ for all a relatively prime to n.

This implies that the order of every element of the multiplicative group of integers modulo n divides λ(n). Carmichael calls an element a for which ${\displaystyle a^{\lambda (n)))$ is the least power of a congruent to 1 (mod n) a primitive λ-root modulo n.^[3] (This is not to be confused with a primitive root modulo n, which Carmichael sometimes refers to as a primitive ${\displaystyle \varphi }$-root modulo n.)

If g is one of the primitive λ-roots guaranteed by the theorem, then ${\displaystyle g^{m}\equiv 1{\pmod {n))}$ has no positive integer solutions m less than λ(n), showing that there is no positive m < λ(n) such that ${\displaystyle a^{m}\equiv 1{\pmod {n))}$ for all a relatively prime to n.

The second statement of Theorem 2 does not imply that all primitive λ-roots modulo n are congruent to powers of a single root g.^[5] For example, if n = 15, then λ(n) = 4 while ${\displaystyle \varphi (n)=8}$ and ${\displaystyle \varphi (\lambda (n))=2}$. There are four primitive λ-roots modulo 15, namely 2, 7, 8, and 13 as ${\displaystyle 1\equiv 2^{4}\equiv 8^{4}\equiv 7^{4}\equiv 13^{4))$. The roots 2 and 8 are congruent to powers of each other and the roots 7 and 13 are congruent t to powers of each other, but neither 7 nor 13 is congruent to a power of 2 or 8 and vice versa. The other four elements of the multiplicative group modulo 15, namely 1, 4 (which satisfies ${\displaystyle 4\equiv 2^{2}\equiv 8^{2}\equiv 7^{2}\equiv 13^{2))$), 11, and 14, are not primitive λ-roots modulo 15.

For a contrasting example, if n = 9, then ${\displaystyle \lambda (n)=\varphi (n)=6}$ and ${\displaystyle \varphi (\lambda (n))=2}$. There are two primitive λ-roots modulo 9, namely 2 and 5, each of which is congruent to the fifth power of the other. They are also both primitive ${\displaystyle \varphi }$-roots modulo 9.

Properties of the Carmichael function

In this section, an integer ${\displaystyle n}$ is divisible by a nonzero integer ${\displaystyle m}$ if there exists an integer ${\displaystyle k}$ such that ${\displaystyle n=km}$. This is written as

${\displaystyle m\mid n.}$

A consequence of minimality of λ(n)

Suppose a^m ≡ 1 (mod n) for all numbers a coprime with n. Then λ(n) | m.

Proof: If m = kλ(n) + r with 0 ≤ r < λ(n), then

${\displaystyle a^{r}=1^{k}\cdot a^{r}\equiv \left(a^{\lambda (n)}\right)^{k}\cdot a^{r}=a^{k\lambda (n)+r}=a^{m}\equiv 1{\pmod {n))}$

for all numbers a coprime with n. It follows that r = 0 since r < λ(n) and λ(n) is the minimal positive exponent for which the congruence holds for all a coprime with n.

λ(n) divides φ(n)

This follows from elementary group theory, because the exponent of any finite group must divide the order of the group. λ(n) is the exponent of the multiplicative group of integers modulo n while φ(n) is the order of that group. In particular, the two must be equal in the cases where the multiplicative group is cyclic due to the existence of a primitive root, which is the case for odd prime powers.

We can thus view Carmichael's theorem as a sharpening of Euler's theorem.

Divisibility

${\displaystyle a\,|\,b\Rightarrow \lambda (a)\,|\,\lambda (b)}$

Proof.

By definition, for any integer ${\displaystyle k}$ with ${\displaystyle \gcd(k,b)=1}$ (and thus also ${\displaystyle \gcd(k,a)=1}$), we have that ${\displaystyle b\,|\,(k^{\lambda (b)}-1)}$ , and therefore ${\displaystyle a\,|\,(k^{\lambda (b)}-1)}$. This establishes that ${\displaystyle k^{\lambda (b)}\equiv 1{\pmod {a))}$ for all k relatively prime to a. By the consequence of minimality proved above, we have ${\displaystyle \lambda (a)\,|\,\lambda (b)}$.

Composition

For all positive integers a and b it holds that

${\displaystyle \lambda (\mathrm {lcm} (a,b))=\mathrm {lcm} (\lambda (a),\lambda (b))}$.

This is an immediate consequence of the recurrence for the Carmichael function.

Exponential cycle length

If ${\displaystyle r_{\mathrm {max} }=\max _{i}\{r_{i}\))$ is the biggest exponent in the prime factorization ${\displaystyle n=p_{1}^{r_{1))p_{2}^{r_{2))\cdots p_{k}^{r_{k))}$ of n, then for all a (including those not coprime to n) and all r ≥ r_max,

${\displaystyle a^{r}\equiv a^{\lambda (n)+r}{\pmod {n)).}$

In particular, for square-free n ( r_max = 1), for all a we have

${\displaystyle a\equiv a^{\lambda (n)+1}{\pmod {n)).}$

Average value

For any n ≥ 16:^[6][7]

${\displaystyle {\frac {1}{n))\sum _{i\leq n}\lambda (i)={\frac {n}{\ln n))e^{B(1+o(1))\ln \ln n/(\ln \ln \ln n)))$

(called Erdős approximation in the following) with the constant

${\displaystyle B:=e^{-\gamma }\prod _{p\in \mathbb {P} }\left({1-{\frac {1}{(p-1)^{2}(p+1)))}\right)\approx 0.34537}$

and γ ≈ 0.57721, the Euler–Mascheroni constant.

The following table gives some overview over the first 2²⁶ – 1 = 67108863 values of the λ function, for both, the exact average and its Erdős-approximation.

Additionally given is some overview over the more easily accessible “logarithm over logarithm” values LoL(n) := ln λ(n)/ln n with

• LoL(n) > 4/5 ⇔ λ(n) > n^4/5.

There, the table entry in row number 26 at column

•  % LoL > 4/5   → 60.49

indicates that 60.49% (≈ 40000000) of the integers 1 ≤ n ≤ 67108863 have λ(n) > n^4/5 meaning that the majority of the λ values is exponential in the length l := log₂(n) of the input n, namely

${\displaystyle \left(2^{\frac {4}{5))\right)^{l}=2^{\frac {4l}{5))=\left(2^{l}\right)^{\frac {4}{5))=n^{\frac {4}{5)).}$

ν	n = 2ν – 1	sum ${\displaystyle \sum _{i\leq n}\lambda (i)}$	average ${\displaystyle {\tfrac {1}{n))\sum _{i\leq n}\lambda (i)}$	Erdős average	Erdős / exact average	LoL average	% LoL > 4/5	% LoL > 7/8
5	31	270	8.709677	68.643	7.8813	0.678244	41.94	35.48
6	63	964	15.301587	61.414	4.0136	0.699891	38.10	30.16
7	127	3574	28.141732	86.605	3.0774	0.717291	38.58	27.56
8	255	12994	50.956863	138.190	2.7119	0.730331	38.82	23.53
9	511	48032	93.996086	233.149	2.4804	0.740498	40.90	25.05
10	1023	178816	174.795699	406.145	2.3235	0.748482	41.45	26.98
11	2047	662952	323.865169	722.526	2.2309	0.754886	42.84	27.70
12	4095	2490948	608.290110	1304.810	2.1450	0.761027	43.74	28.11
13	8191	9382764	1145.496765	2383.263	2.0806	0.766571	44.33	28.60
14	16383	35504586	2167.160227	4392.129	2.0267	0.771695	46.10	29.52
15	32767	134736824	4111.967040	8153.054	1.9828	0.776437	47.21	29.15
16	65535	513758796	7839.456718	15225.430	1.9422	0.781064	49.13	28.17
17	131071	1964413592	14987.400660	28576.970	1.9067	0.785401	50.43	29.55
18	262143	7529218208	28721.797680	53869.760	1.8756	0.789561	51.17	30.67
19	524287	28935644342	55190.466940	101930.900	1.8469	0.793536	52.62	31.45
20	1048575	111393101150	106232.840900	193507.100	1.8215	0.797351	53.74	31.83
21	2097151	429685077652	204889.909000	368427.600	1.7982	0.801018	54.97	32.18
22	4194303	1660388309120	395867.515800	703289.400	1.7766	0.804543	56.24	33.65
23	8388607	6425917227352	766029.118700	1345633.000	1.7566	0.807936	57.19	34.32
24	16777215	24906872655990	1484565.386000	2580070.000	1.7379	0.811204	58.49	34.43
25	33554431	96666595865430	2880889.140000	4956372.000	1.7204	0.814351	59.52	35.76
26	67108863	375619048086576	5597160.066000	9537863.000	1.7041	0.817384	60.49	36.73

Prevailing interval

For all numbers N and all but o(N)^[8] positive integers n ≤ N (a "prevailing" majority):

${\displaystyle \lambda (n)={\frac {n}{(\ln n)^{\ln \ln \ln n+A+o(1)))))$

with the constant^[7]

${\displaystyle A:=-1+\sum _{p\in \mathbb {P} }{\frac {\ln p}{(p-1)^{2))}\approx 0.2269688}$

Lower bounds

For any sufficiently large number N and for any Δ ≥ (ln ln N)³, there are at most

${\displaystyle N\exp \left(-0.69(\Delta \ln \Delta )^{\frac {1}{3))\right)}$

positive integers n ≤ N such that λ(n) ≤ ne^−Δ.^[9]

Minimal order

For any sequence n₁ < n₂ < n₃ < ⋯ of positive integers, any constant 0 < c < 1/ln 2, and any sufficiently large i:^[10][11]

${\displaystyle \lambda (n_{i})>\left(\ln n_{i}\right)^{c\ln \ln \ln n_{i)).}$

Small values

For a constant c and any sufficiently large positive A, there exists an integer n > A such that^[11]

${\displaystyle \lambda (n)<\left(\ln A\right)^{c\ln \ln \ln A}.}$

Moreover, n is of the form

${\displaystyle n=\mathop {\prod _{q\in \mathbb {P} )) _{(q-1)|m}q}$

for some square-free integer m < (ln A)^{c ln ln ln A}.^[10]

Image of the function

The set of values of the Carmichael function has counting function^[12]

${\displaystyle {\frac {x}{(\ln x)^{\eta +o(1)))},}$

where

${\displaystyle \eta =1-{\frac {1+\ln \ln 2}{\ln 2))\approx 0.08607}$

Use in cryptography

The Carmichael function is important in cryptography due to its use in the RSA encryption algorithm.

Proof of Theorem 1

For n = p, a prime, Theorem 1 is equivalent to Fermat's little theorem:

${\displaystyle a^{p-1}\equiv 1{\pmod {p))\qquad {\text{for all ))a{\text{ coprime to ))p.}$

For prime powers p^r, r > 1, if

${\displaystyle a^{p^{r-1}(p-1)}=1+hp^{r))$

holds for some integer h, then raising both sides to the power p gives

${\displaystyle a^{p^{r}(p-1)}=1+h'p^{r+1))$

for some other integer ${\displaystyle h'}$. By induction it follows that ${\displaystyle a^{\varphi (p^{r})}\equiv 1{\pmod {p^{r))))$ for all a relatively prime to p and hence to p^r. This establishes the theorem for n = 4 or any odd prime power.

Sharpening the result for higher powers of two

For a coprime to (powers of) 2 we have a = 1 + 2h₂ for some integer h₂. Then,

${\displaystyle a^{2}=1+4h_{2}(h_{2}+1)=1+8{\binom {h_{2}+1}{2))=:1+8h_{3))$,

where ${\displaystyle h_{3))$ is an integer. With r = 3, this is written

${\displaystyle a^{2^{r-2))=1+2^{r}h_{r}.}$

Squaring both sides gives

${\displaystyle a^{2^{r-1))=\left(1+2^{r}h_{r}\right)^{2}=1+2^{r+1}\left(h_{r}+2^{r-1}h_{r}^{2}\right)=:1+2^{r+1}h_{r+1},}$

where ${\displaystyle h_{r+1))$ is an integer. It follows by induction that

${\displaystyle a^{2^{r-2))=a^((\frac {1}{2))\varphi (2^{r})}\equiv 1{\pmod {2^{r))))$

for all ${\displaystyle r\geq 3}$ and all a coprime to ${\displaystyle 2^{r))$.^[13]

Integers with multiple prime factors

By the unique factorization theorem, any n > 1 can be written in a unique way as

${\displaystyle n=p_{1}^{r_{1))p_{2}^{r_{2))\cdots p_{k}^{r_{k))}$

where p₁ < p₂ < ... < p_k are primes and r₁, r₂, ..., r_k are positive integers. The results for prime powers establish that, for ${\displaystyle 1\leq j\leq k}$,

${\displaystyle a^{\lambda \left(p_{j}^{r_{j))\right)}\equiv 1{\pmod {p_{j}^{r_{j))))\qquad {\text{for all ))a{\text{ coprime to ))n{\text{ and hence to ))p_{i}^{r_{i)).}$

From this it follows that

${\displaystyle a^{\lambda (n)}\equiv 1{\pmod {p_{j}^{r_{j))))\qquad {\text{for all ))a{\text{ coprime to ))n,}$

where, as given by the recurrence,

${\displaystyle \lambda (n)=\operatorname {lcm} {\Bigl (}\lambda \left(p_{1}^{r_{1))\right),\lambda \left(p_{2}^{r_{2))\right),\ldots ,\lambda \left(p_{k}^{r_{k))\right){\Bigr )}.}$

From the Chinese remainder theorem one concludes that

${\displaystyle a^{\lambda (n)}\equiv 1{\pmod {n))\qquad {\text{for all ))a{\text{ coprime to ))n.}$