# How to find the inverse modulo $m$?

For example: $$7x \equiv 1 \pmod{31}$$ In this example, the modular inverse of $$7$$ with respect to $$31$$ is $$9$$. How can we find out that $$9$$? What are the steps that I need to do?

Update
If I have a general modulo equation:
$$5x + 1 \equiv 2 \pmod{6}$$

What is the fastest way to solve it? My initial thought was: $$5x + 1 \equiv 2 \pmod{6}$$ $$\Leftrightarrow 5x + 1 - 1\equiv 2 - 1 \pmod{6}$$ $$\Leftrightarrow 5x \equiv 1 \pmod{6}$$

Then solve for the inverse of $$5$$ modulo 6. Is it a right approach?

Thanks.

## 9 Answers

1. One method is simply the extended Euclidean algorithm: \begin{align*} 31 &= 4(7) + 3\\\ 7 &= 2(3) + 1. \end{align*} So $$1 = 7 - 2(3) = 7 - 2(31 - 4(7)) = 9(7) - 2(31)$$. Viewing the equation $$1 = 9(7) -2(31)$$ modulo $$31$$ gives $$1 \equiv 9(7)\pmod{31}$$, so the multiplicative inverse of $$7$$ modulo $$31$$ is $$9$$. This works in any situation where you want to find the multiplicative inverse of $$a$$ modulo $$m$$, provided of course that such a thing exists (i.e., $$\gcd(a,m) = 1$$). The Euclidean Algorithm gives you a constructive way of finding $$r$$ and $$s$$ such that $$ar+ms = \gcd(a,m)$$, but if you manage to find $$r$$ and $$s$$ some other way, that will do it too. As soon as you have $$ar+ms=1$$, that means that $$r$$ is the modular inverse of $$a$$ modulo $$m$$, since the equation immediately yields $$ar\equiv 1 \pmod{m}$$.

2. Another method is to play with fractions Gauss's method: $$\frac{1}{7} = \frac{1\times 5}{7\times 5} = \frac{5}{35} = \frac{5}{4} = \frac{5\times 8}{4\times 8} = \frac{40}{32} = \frac{9}{1}.$$ Here, you reduce modulo $$31$$ where appropriate, and the only thing to be careful of is that you should only multiply and divide by things relatively prime to the modulus. Here, since $$31$$ is prime, this is easy. At each step, I just multiplied by the smallest number that would yield a reduction; so first I multiplied by $$5$$ because that's the smallest multiple of $$7$$ that is larger than $$32$$, and later I multiplied by $$8$$ because it was the smallest multiple of $$4$$ that is larger than $$32$$. Added: As Bill notes, the method may fail for composite moduli.

Both of the above methods work for general modulus, not just for a prime modulus (though Method 2 may fail in that situation); of course, you can only find multiplicative inverses if the number is relatively prime to the modulus.

Update. Yes, your method for general linear congruences is the standard one. We have a very straightforward method for solving congruences of the form $$ax \equiv b\pmod{m},$$ namely, it has solutions if and only if $$\gcd(a,m)|b$$, in which case it has exactly $$\gcd(a,m)$$ solutions modulo $$m$$.

To solve such equations, you first consider the case with $$\gcd(a,m)=1$$, in which case $$ax\equiv b\pmod{m}$$ is solved either by finding the multiplicative inverse of $$a$$ modulo $$m$$, or as I did in method $$2$$ above looking at $$\frac{b}{a}$$.

Once you know how to solve them in the case where $$\gcd(a,m)=1$$, you can take the general case of $$\gcd(a,m) = d$$, and from $$ax\equiv b\pmod{m}$$ go to $$\frac{a}{d}x \equiv \frac{b}{d}\pmod{\frac{m}{d}},$$ to get the unique solution $$\mathbf{x}_0$$. Once you have that unique solution, you get all solutions to the original congruence by considering $$\mathbf{x}_0,\quad \mathbf{x}_0 + \frac{m}{d},\quad \mathbf{x}_0 + \frac{2m}{d},\quad\ldots, \mathbf{x}_0 + \frac{(d-1)m}{d}.$$

• @Arturo Magidin: Thanks, it's a lot of work :(. I thought it's just a simple calculation. Mar 6, 2011 at 23:38
• @Chan: Your definition of "a lot of work" needs revising, if you plan on studying more number theory and maths. The above is surprising little work, in my estimation... Mar 6, 2011 at 23:40
• @Arturo Magidin: Thanks for the advice. However, I have many other things that I want to study. I like Number Theory, but also Geometry, Computer Graphics, Algorithm, C++, Ruby, Compiler, AI..., plus my school classes. I wish a day could have 48 hours. Mar 6, 2011 at 23:43
• the euclidean algorithm is simple and easily programed...
– yoyo
Mar 6, 2011 at 23:48
• @Chan: Just for your information: the Euclidean Algorithm is considered a very fast algorithm; certainly faster than factoring and many other calculations that one often needs to do. In fact, many factoring algorithms work by making educated guesses and then computing gcds by using the Euclidean Algorithm in the hope of getting a nontrivial factor that way, precisely because the Euclidean Algorithm is so efficient. Mar 6, 2011 at 23:50

We know that $(7,31) = 1$. So you can use the extended Euclidean algorithm. In particular, you can find two integers $s,t$ such that $7s+31t = 1$. So $7s \equiv 1 \pmod {31}$.

A hint: Try to use Fermats Little theorem:

$a^{p-1}=1 \mod p$ for $p$ prime, and all $a\in \mathbb{Z}$.

• 7^29 mod 31 certainly works in this case, but for larger numbers it might be a bit of an effort compared with the Euclidean algorithm, and it would get more complicated by involving Euler's totient function for a non-prime modulus. Mar 6, 2011 at 23:57
• You can use binary exponentiation for high powers.
– qwr
Oct 8, 2023 at 12:09

We can visualize the solution to $7x \equiv 1 \pmod{31}$ as the intersection of the two "lines"

$\ \ \ \ \ y \equiv 7x \pmod{31}$ and $y \equiv 1 \pmod{31}$.

The first line is closed under addition and subtraction, since it passes through the origin. If $(x_1, y_1)$ and $(x_2, y_2)$ are points on the line $y \equiv 7x \pmod{31}$, then so are $(x_1+x_2, y_1+y_2)$ and $(x_1-x_2, y_1-y_2)$.

To find a point where the lines intersect, we start with the two points $(1, 7)$ and $(0,31)$ on the first line, and repeatedly subtract one point from another until the $y$-coordinate is 1.

$\ \ \ \ \ (0,31) - (1,7) = (-1, 24)$

$\ \ \ \ \ (-1, 24) - (1,7) = (-2, 17)$

$\ \ \ \ \ (-2, 17) - (1, 7) = (-3, 10)$

$\ \ \ \ \ (-3, 10) - (1, 7) = (-4, 3)$

$\ \ \ \ \ (1,7) - (-4, 3) = (5, 4)$

$\ \ \ \ \ (5,4) - (-4, 3) = (9,1)$

Therefore, $x \equiv 9 \pmod{31}$.

You can speed up this procedure by subtracting a multiple of one point from another point instead. This leads to the Euclidean algorithm.

There are many methods available, e.g. the extended Euclidean algorithm,  or a special case of Euclid's algorithm that computes inverses modulo primes that I call Gauss's algorithm.  The calculations are usually simpler using modular fraction arithmetic e.g. see here, and here and here for circa $$20$$ motley worked examples via a handful of methods (and see the sidebar "Linked" questions lists there for many more).

Update  March $$16, 2020\!:\,$$ for completeness we apply some of the linked methods below.

$$\!\!\bmod 31\!:\,\ \dfrac{1}{7}\equiv \dfrac{4}{28}\equiv\dfrac{-27}{-3}\equiv 9\$$ by Gauss's algorithm.

$$\!\!\bmod 31\!:\,\ \dfrac{1}{7}\equiv \dfrac{1}{-6}\,\dfrac{1}{4}\equiv \dfrac{-30}{-6}\,\dfrac{32}{4}\equiv 5\cdot 8\equiv 9$$

As here:  the freedom to choose $$\rm\color{#c00}{even}$$ residue reps $$\!\bmod\!$$ odds makes division by 2 easy:

$$\!\!\bmod 31\!:\,\ \dfrac{1}{7}\equiv\dfrac{\color{#c00}{32}}{\color{#c00}{-24}}\equiv\dfrac{4}{-3}\equiv\dfrac{-27}{-3}\equiv 9$$

By the fractional extended Euclidean algorithm, or associated equational form

\begin{align} \bmod 31\!:\ \ \dfrac{0}{31}\overset{\large\frown}\equiv\color{#c00}{\dfrac{1}7}\ \, &\!\!\!\overset{\large\frown}\equiv\color{#0a0}{\dfrac{-4}3}\overset{\large\frown}\equiv\color{#90f}{\dfrac{9}1}\\[.7em] \text{said equationally}\ \ \ \ [\![1]\!]\ \ \ \ 31\, x&\,\equiv \ 0\ \\ [\![2]\!] \ \ \ \ \ \ \color{#c00}{7\,x}&\ \color{#c00}{ \equiv\ 1}\!\!\!\\ [\![1]\!]-4\,[\![2]\!] \rightarrow [\![3]\!]\ \ \ \ \ \ \color{#0a0}{3\,x} &\ \color{#0a0}{\equiv\ {-}4}\ \\ [\![2]\!] - 2\,[\![3]\!] \rightarrow [\![4]\!]\, \ \ \ \ \ \ \ \ \color{#90f}{x}&\ \color{#90f}{ \equiv\ 9} \end{align}

Below we explain the basic idea behind the method of Inverse Reciprocity.

$$\!\!\bmod 31\!:\,\ n \equiv \dfrac{1}7\equiv \dfrac{1+31\color{#c00}k}7.\$$ For an exact quotient we seek $$\,k\,$$ with $$\,7\mid 1\!+\!31k,\,$$ i.e.

\!\!\bmod 7\!:\,\ \begin{align}0&\equiv 1\!+\!31k\\ &\equiv 1 + 3k\end{align}\! \iff\! \begin{align}3k&\equiv-1\\ &\equiv\,\ 6\end{align}\! $$\iff \color{#c00}{k\equiv 2},\$$ so $$\ n \equiv \dfrac{1\!+\!31(\color{#c00}2)}7\equiv 9\pmod{\!31}$$

Or $$\rm\ (a,m)\!=\!1 \overset{\rm Euler\ \phi}\Longrightarrow\ a^{-1} \equiv a^{\phi(m)-1}\!\pmod{\! m},\,$$ quickly computable by repeated squaring. It suffices to find any (small) period $$\rm\,e,\,$$ of $$\rm\,a,\,$$ i.e. $$\rm\,e\,$$ such that $$\rm\,a^{\large e}\equiv 1\,$$ (e.g. by Euler's totient or Fermat's little theorem or Carmichael's lambda theorem), since this implies $$\rm\,a^{-1}\equiv a^{e-1}$$.

Remark  The latter yields a simple closed form for CRT (Chinese Remainder Theorem)

$$\quad$$ if $$\rm\,\ (m,n)\!=\!1\,\$$ then $$\rm\quad \begin{eqnarray}\rm x\!&\equiv&\rm a\ \ (mod\ m)\\ \rm x\!&\equiv&\rm b\ \ (mod\ n)\end{eqnarray} \iff\ x\equiv a\,n^{\phi(m)}\!+b\,m^{\phi(n)}\ \ (mod\ mn)$$

More generally see the Peirce decomposition.

• @BillDubuque can we invert if totient is unknown? Sep 1, 2017 at 1:10
• @777 Yes, see the other two methods I mentioned above. Sep 1, 2017 at 2:45
• @BillDubuque which ones? all seems to use factorization of $m$ or $\phi(m)$ and nothing avoids these. Sep 1, 2017 at 2:51
• @777 The extended Euclidean algorithm and Gauss's algorithm do not use factorization. Follow the links to learn these methods. Sep 1, 2017 at 3:13
• @BillDubuque so both these methods do not need factorization of $m$ or knowledge of $\phi(m)$. Then doesn't that help to get decryption key $d$ in RSA from encryption key $e$? Sep 1, 2017 at 22:47

An alternative way to solve would be to use the fast powering algorithm.

$a^{-1}≡a^{p-2} \mod p$ for $p$, $a\in \mathbb{Z}$.

$7^{-1} \mod 31 = 7^{29} \mod 31 ≡ 9 \mod 31$

According to An Introduction to Mathematical Cryptography by Hoffstein et al, in practice this is about the same time complexity as the extended Euclidean algorithm given in other answers.

Before using the fast powering algorithm I would first check to see if a and p are relatively prime, so that you know an inverse exists before performing the calculation.

• Binary exponentiation
– qwr
Oct 8, 2023 at 12:10

Solve $$7x \equiv 1 \pmod{31}$$.

$$\quad 7x = 1 + 31y \implies 3y \equiv 6 \pmod{7} \implies y \equiv 2 \pmod{7}$$

and

$$\quad y : 2 \; \mid \; 1 + 31y = 63 \quad \quad \text{AND is divisible by } 7$$

So,

$$\tag{ANS} 7^{-1} \equiv 9 \pmod{31}$$

Here is another more interesting example:

Solve $$25x \equiv 1 \pmod{41}$$.

$$\quad 25x = 1 + 41y \implies y \equiv 4 \pmod{5}$$

and

$$\quad y : \;\,4 \; \mid \; 1 + 41y = 165 \quad \quad \text{NO GOOD!}$$
$$\quad y : \;\,9 \; \mid \; 1 + 41y = 370 \quad \quad \text{NO GOOD!}$$
$$\quad y : 14 \; \mid \; 1 + 41y = 575 \quad \quad \text{AND is divisible by } 25$$

So,

$$\tag{ANS} 25^{-1} \equiv 23 \pmod{41}$$

Here is another more interesting example:

Solve $$23x \equiv 1 \pmod{41}$$.

$$\quad 23x = 1 + 41y \implies 18y \equiv 22 \pmod{23} \implies 9y \equiv 11 \pmod{23}$$

Sub Solve: $$9y \equiv 11 \pmod{23}$$.

$$\quad 9y = 11 + 23z \implies 2z \equiv 1 \pmod{3} \implies z \equiv 2 \pmod{3}$$

and

$$\quad z : \;\,2 \; \mid \; 11 + 23z = 057 \quad \quad \text{NO GOOD!}$$
$$\quad z : \;\,5 \; \mid \; 11 + 23z = 126 \quad \quad \text{AND is divisible by } 9$$

So,

$$\quad 9y \equiv 11 \pmod{23} \text{ iff } y \equiv 14 \pmod{23}$$

Continuing where we left off with the original problem,

$$\quad y : \;\,14 \; \mid \; 1 + 41y = 575 \quad \quad \text{AND is divisible by } 23$$

So,

$$\tag{ANS} 23^{-1} \equiv 25 \pmod{41}$$

• Thank you. How you come up with $3y \equiv 6 \pmod{7}$ in $\quad 7x = 1 + 31y \implies 3y \equiv 6 \pmod{7} \implies y \equiv 2 \pmod{7}$?
– Avv
Mar 18, 2021 at 21:12
• If $7x=1+31y$ then $0 \equiv 1 + 31y \pmod7$ then $31y \equiv -1 \pmod7$ then $\big[(4\cdot7 +3)y \equiv 6 - (1\cdot7) \big]\pmod7$ then $3y \equiv 6 \pmod7$. Jun 13, 2021 at 15:58

The Arithmetical Progression f(n) = a + (n-1)d has only a finite number of terms for any modulus d and where a is the "initial" value, f(1), even when n is allowed to range over ℤ, so no holds barred trying negative values to quickly reach a coefficient of 1 or a -1.

In this example all moduli are to 31, starting from coefficient 7 one possibility is:

7x ≡ 1

⇒ -24x ≡ 32 (subtracting the modulus on the left and adding the modulus to the right)
⇒  -3x ≡ 4 (∵ gcd (12, 31) = 1)
⇒ -30x ≡ 40
⇒   1x ≡ 9


The above is an application of congruence article in the chapter on theory of numbers in Higher Algebra By H.S. Hall and S.R. Knight May 1887 (pdf probably available online).

$$7x\equiv 1\pmod{31}$$

We are seeking integer solutions of the equation,

$$7x+31y=1$$

Find a linear transformation $$\begin{pmatrix}\cdot & \cdot \\ \cdot &\cdot\end{pmatrix}$$ consisting of integer entries which maps the vector $$\begin{pmatrix}7\\ 31\end{pmatrix}$$ to some vector of the form $$\begin{pmatrix}1\\ \cdot\end{pmatrix}$$ or $$\begin{pmatrix}\cdot\\ 1\end{pmatrix}$$.

The identity transformation: $$\begin{pmatrix} 1 & 0\\ 0 & 1\end{pmatrix}\begin{pmatrix} 7\\ 31\end{pmatrix} =\begin{pmatrix}7\\31\end{pmatrix}$$

Now apply successive elementary row operations on the augmented matrix until you get the desired form on the RHS.

$$\Longrightarrow \begin{pmatrix} 9 & -2\\ -4 & 1 \end{pmatrix}\begin{pmatrix} 7\\ 31 \end{pmatrix} =\begin{pmatrix} 1\\ 3 \end{pmatrix}$$

$$\Longrightarrow 7( 9) +31( -2) =1\\ \Longrightarrow x\equiv 9\pmod{31}$$

This is just a different perspective to the extended Euclidean algorithm. I think this is useful, if you want to code the algorithm :)

Ideally, if you are up for finding $$x$$, you would not calculate the values on the second column because those are of no use.

• This forward form of the extended Euclidean algorithm was already linked in my answer long ago (I show a simpler fractional version in my answer). Here is a worked example. Nov 22, 2023 at 5:34
• @BillDubuque I did not see that. It is very detailed. Should I delete this post or let it be? Nov 22, 2023 at 5:45