This is a statement in linear algebra that I can't seem to understand the proof behind.

For a square matrix $A$, why is: $$\mathrm{adj}(A)A = \det(A) \cdot I$$

Any explanation would be greatly appreciated.


There is a nice short explanation. You probably know that the determinant of any matrix can be expanded row or column-wise using the minors:

$$\det A=\sum_{i=1}^n a_{ij}(-1)^{i+j}\det A(i\mid j)$$


$$\det A=\sum_{j=1}^n a_{ij}(-1)^{i+j}\det A(i\mid j)$$

for any $j$ (resp. $i$) or your liking where we obtain $A(i\mid j)$ by striking out the $i$-th row and the $j$-th column.

We define the coefficients of the adjoint $\hat A$ by $$(\hat A)_{ij}=(-1)^{{i+j}}\det A(j\mid i)$$

Now, upon matrix multiplication, we have

$$(A\cdot\hat A)_{k\ell}=\sum_{i=1}^n (-1)^{i+\ell}a_{ki} \det A(\ell\mid i)$$

If $k=\ell$, then $$(A\cdot \hat A)_{\ell\ell}=\sum_{i=1}^n(-1)^{i+\ell}a_{\ell i}\det A(\ell \mid i)=\det A$$ since we're expanding the determinant through the $\ell$-th row.

If $k\neq \ell$

$$(A\cdot\hat A)_{k\ell}=\sum_{i=1}^n (-1)^{i+\ell}a_{ki} \det A(\ell\mid i)=0$$

for it is the expansion of the determinant of the matrix $A^{k\ell}$ defined by $$(A^{k\ell})=\begin{cases} a_{ij} \text{ if }i\neq \ell\\a_{kj}\text{ if }i=\ell\end{cases}$$ which has two equal rows.

  • 1
    $\begingroup$ I can't seem to understand, why is this true? "since we're expanding the determinant through the ℓ-th column. If k≠ℓ (A⋅A^)kℓ = 0" $\endgroup$ – vondip Mar 31 '13 at 10:18
  • $\begingroup$ @vondip That sentence refers to the $\ell =k$ part. See the first two summations at the beginning. $\endgroup$ – Pedro Tamaroff Mar 31 '13 at 16:32
  • 1
    $\begingroup$ Sorry but i too, can't understand why if k≠ℓ (A⋅A^)kℓ = 0. is it possible to make it even simpler? $\endgroup$ – George Smyridis Feb 10 '16 at 21:05
  • $\begingroup$ @GeorgeSmyridis math.stackexchange.com/questions/1404189 $\endgroup$ – BCLC Sep 19 '16 at 14:45
  • $\begingroup$ @vondip math.stackexchange.com/questions/1404189 $\endgroup$ – BCLC Sep 19 '16 at 14:45

Another way of stating Cramer's Rule. See e.g. Wikipedia.

  • 3
    $\begingroup$ it seems to be as though Cramer's Rule relays on this. But what is the right way to prove this theorem? $\endgroup$ – vondip Mar 29 '13 at 9:43

Let $A=[\alpha_{ij}]$ be matrix of order $n$ on field $K$ and $s,r \in \{1,\dots,n\}$, $n>1,$ fixed elements. Let $\hat{A}_{rs}=(-1)^{r+s} \det A_{rs}$ . Then $$\det A = \alpha_{1s}\hat{A}_{1s}+\alpha_{2s}\hat{A}_{2s}+\dots+\alpha_{ns}\hat{A}_{ns}$$ and similary $$\det A = \alpha_{r1}\hat{A}_{r1}+\alpha_{r2}\hat{A}_{r2}+\dots+\alpha_{rn}\hat{A}_{rn}\tag{1}$$ Then we find that $$\alpha_{i1}\hat{A}_{r1}+\dots+\alpha_{in}\hat{A}_{rn}=\begin{cases} 0, &\text{ for } i \neq r,\\ \det A,& \text{ for } i=r.\end{cases}\tag{2}$$

And similary for $\alpha_{1i}\hat{A}_{1s}+\dots+\alpha_{ni}\hat{A}_{ns}=0 \quad (i \neq s).$

Now conclude that for every matrix of order $n$ on field $K$ it belongs one matrix

$$\operatorname{adj}A = \begin{bmatrix} \hat{A}_{11} & \cdots & \hat{A}_{n1} \\ \vdots & \ddots & \vdots \\ \hat{A}_{1n} & \cdots & \hat{A}_{nn} \end{bmatrix}$$ from $M_n(K)$, such that $$A \cdot \operatorname{adj}A = \det A \cdot I$$

Edit (explanation for $(2)$): From $(1)$, left side of $(2)$ is determinant of matrix $B$ which arise from matrix $A$ when we change $r$-th row its $i$-th row. If $i \neq r$ then that matrix have two same rows $(B_{r\to}=A_{i\to}=B_{i\to}$), so its determinant is zero.

  • 1
    $\begingroup$ Why is it true that for each i≠r, 0 aiAr is 0 ? $\endgroup$ – vondip Mar 29 '13 at 11:36
  • $\begingroup$ @vondip I edited the answer. $\endgroup$ – Cortizol Mar 29 '13 at 19:38

It is well known that the determinant, viewed as function of the columns of a square matrix, is linear in each of the columns. This means that if we fix any matrix $A\in M_n(K)$ and a column index $j$, and denote by $f_{A,j}$ the function $K^n\to K$ defined by $f_{A,j}(x_1,\ldots,x_n)=\det(A')$ where $A'$ is the matrix obtained by replacing column $j$ of $A$ by the entries $x_1,\ldots,x_n$, then $f_{A,j}$ is a linear function, so there exist constants $c_1,\ldots,c_n$ such that $f_{A,j}(x_1,\ldots,x_n)=c_1x_1+\cdots+c_nx_n$. These constants of course depend on $A$ and on $j$ (although the column $j$ of $A$ is irrelevant, as it gets systematically replaced), and indeed the Laplace expansion of $\det(A')$ by column $j$ gives explicit expressions for these coefficients. By definition, the $1\times n$ matrix $(c_1~c_2~\ldots~c_n)$ is precisely row $j$ of the adjugate matrix $\def\adj{\operatorname{adj}}\adj A$.

Now consider the matrix product $(c_1~c_2~\ldots~c_n)\cdot A=(p_1~p_2~\ldots~p_n)$. This can be computed separately by the columns of $A$: one has $p_i=f_{A,j}(A_{1,i},\ldots,A_{n,i})$ (the arguments form column $i$ of $A$), which is $\det(A')$ where $A'$ is the matrix obtained by replacing column $j$ of $A$ by (a copy of) column $i$ of $A$. Obviously one has $A'=A$ if $i=j$, so that $p_i=\det A$, and all other values $p_i$ are zero due to the alternating property of the determinant, since in these cases we are taking the determinant of a matrix $A$ whose columns $i$ and $j$ are identical, with $i\neq j$. In formula $$ p_i=\delta_{i,j}\det A \qquad\text{for $i\in\{1,\ldots,n\}$.} $$ But what we have computed is just row $j$ of the matrix product $\adj A\cdot A$, and what we have said is valid for all $j\in\{1,\ldots,n\}$. Combining all these computations one gets $$ \adj A\cdot A=I_n\det A $$ (where the right hand side is just a scalar multiple of the identity matrix $I_n$), which is the identity in the question.

All in all you can see there is not much more to this identity than the multi-linear and alternating properties (by columns) of the determinant, Laplace expansion and the definition of the adjugate matrix. Since it is a formal algebraic identity, one may take any commutative ring in the place of $K$ (but commutativity is essential; multi-linearity depends on it). To prove that $A\cdot\adj A$ also gives the same result, one needs the multi-linear and alternating properties by rows of the determinant.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.