As far as linear algebra is concerned, the two most important operations with vectors are vector addition [adding two (or more) vectors] and scalar multiplication (multiplying a vectro by a scalar). Analogous operations are defined for matrices.

**Matrix addition**. If *A* and *B* are matrices *of the same size*, then they can be added. (This is similar to the restriction on adding vectors, namely, only vectors from the same space **R** ^{n }can be added; you cannot add a 2‐vector to a 3‐vector, for example.) If *A* = [*a*_{ij}] and *B* = [*b*_{ij}] are both *m* x *n* matrices, then their sum, *C* = *A* + *B*, is also an *m* x *n* matrix, and its entries are given by the formula

Thus, to find the entries of *A* + *B*, simply add the corresponding entries of *A* and *B*.

**Example 1**: Consider the following matrices:

Which two can be added? What is their sum?

Since only matrices of the same size can be added, only the sum *F* + *H* is defined (*G* cannot be added to either *F* or *H*). The sum of *F* and *H* is

Since addition of real numbers is commutative, it follows that addition of matrices (when it is defined) is also commutative; that is, for any matrices *A* and *B* of the same size, *A* + *B* will always equal *B* + *A*.

**Example 2**: If any matrix *A* is added to the zero matrix of the same size, the result is clearly equal to *A*:

This is the matrix analog of the statement *a* + 0 = 0 + *a* = *a*, which expresses the fact that the number 0 is the additive identity in the set of real numbers.

**Example 3**: Find the matrix *B* such that *A* + *B* = *C*, where

If

then the matrix equation *A* + *B* = *C* becomes

Since two matrices are equal if and only if they are of the same size and their corresponding entries are equal, this last equation implies

Therefore,

This example motivates the definition of matrix *subtraction*: If *A* and *B* are matrices of the same size, then the entries of *A* − *B* are found by simply subracting the entries of *B* from the corresponding entries of *A*. Since the equation *A* + *B* = *C* is equivalent to *B* = *C* − *A*, employing matrix subtraction above would yield the same result:

**Scalar multiplication**. A matrix can be multiplied by a scalar as follows. If *A* = [*a*_{ij}] is a matrix and *k* is a scalar, then

That is, the matrix *kA* is obtained by multiplying each entry of *A* by *k*.

**Example 4**: If

then the scalar multiple 2 *A* is obtained by multiplying every entry of *A* by 2:

**Example 5**: If *A* and *B* are matrices of the same size, then *A* − *B* = *A* + (− *B*), where − *B* is the scalar multiple (−1) *B*. If

then

This definition of matrix subtraction is consistent with the definition illustrated in Example 8.

**Example 6**: If

then

**Matrix multiplication**. By far the most important operation involving matrices is *matrix multiplication*, the process of multiplying one matrix by another. The first step in defining matrix multiplication is to recall the definition of the dot product of two vectors. Let **r** and **c** be two *n*‐vectors. Writing **r** as a 1 x *n* row matrix and **c** as an *n* x 1 column matrix, the dot product of **r** and **c** is

Note that in order for the dot product of **r** and **c** to be defined, both must contain the same number of entries. Also, the order in which these matrices are written in this product is important here: The row vector comes first, the column vector second.

Now, for the final step: How are two general matrices multiplied? First, in order to form the product *AB, the number of columns of A must match the number of rows of B*; if this condition does not hold, then the product *AB* is not defined. This criterion follows from the restriction stated above for multiplying a row matrix **r** by a column matrix **c**, namely that the number of entries in **r** must match the number of entries in **c**. If *A* is *m* x *n* and *B* is *n* x *p*, then the product *AB* is defined, and the size of the product matrix *AB* will be *m* x *p*. The following diagram is helpful in determining if a matrix product is defined, and if so, the dimensions of the product:

Thinking of the *m* x *n* matrix *A* as composed of the row vectors **r**_{1}, **r**_{2},…, **r**_{m} from **R**^{n} and the *n* x *p* matrix *B* as composed of the column vectors **c**_{1}, **c**_{2},…, **c**_{p} from **R**^{n},

and

the rule for computing the entries of the matrix product *AB* is **r** _{i }· **c** _{j }= ( *AB*) _{ij }, that is,

**Example 7**: Given the two matrices

determine which matrix product, *AB* or *BA*, is defined and evaluate it.

Since *A* is 2 x 3 and *B* is 3 x 4, the product *AB*, in that order, is defined, and the size of the product matrix *AB* will be 2 x 4. The product *BA* is *not* defined, since the first factor ( *B*) has 4 columns but the second factor ( *A*) has only 2 rows. The number of columns of the first matrix must match the number of rows of the second matrix in order for their product to be defined.

Taking the dot product of row 1 in *A* and column 1 in *B* gives the (1, 1) entry in *AB*. Since

the (1, 1) entry in *AB* is 1:

The dot product of row 1 in *A* and column 2 in *B* gives the (1, 2) entry in *AB*,

and the dot product of row 1 in *A* and column 3 in *B* gives the (1, 3) entry in *AB*:

The first row of the product is completed by taking the dot product of row 1 in *A* and column 4 in *B*, which gives the (1, 4) entry in *AB*:

Now for the second row of *AB*: The dot product of row 2 in *A* and column 1 in *B* gives the (2, 1) entry in *AB*,

and the dot product of row 2 in *A* and column 2 in *B* gives the (2, 2) entry in *AB*:

Finally, taking the dot product of row 2 in *A* with columns 3 and 4 in *B* gives (respectively) the (2, 3) and (2, 4) entries in *AB*:

Therefore,

**Example 8**: If

and

compute the (3, 5) entry of the product *CD*.

First, note that since *C* is 4 x 5 and *D* is 5 x 6, the product *CD* is indeed defined, and its size is 4 x 6. However, there is no need to compute all twenty‐four entries of *CD* if only one particular entry is desired. The (3, 5) entry of *CD* is the dot product of row 3 in *C* and column 5 in *D*:

**Example 9**: If

verify that

but

In particular, note that even though both products *AB* and *BA* are defined, *AB* does not equal *BA*; indeed, they're not even the same size!

The previous example gives one illustration of what is perhaps the most important distinction between the multiplication of scalars and the multiplication of matrices. For real numbers *a* and *b*, the equation *ab = ba* always holds, that is, multiplication of real numbers is commutative; the order in which the factors are written is irrelevant. However, it is decidedly false that matrix multiplication is commutative. For the matrices *A* and *B* given in Example 9, both products *AB* and *BA* were defined, but they certainly were not identical. In fact, the matrix *AB* was 2 x 2, while the matrix *BA* was 3 x 3. Here is another illustration of the noncommutativity of matrix multiplication: Consider the matrices

Since *C* is 3 x 2 and *D* is 2 x 2, the product *CD* is defined, its size is 3 x 2, and

The product *DC*, however, is not defined, since the number of columns of *D* (which is 2) does not equal the number of rows of *C* (which is 3). Therefore, *CD ≠ DC*, since *DC* doesn't even exist.

Because of the sensitivity to the order in which the factors are written, one does not typically say simply, “Multiply the matrices *A* and *B*.” It is usually important to indicate which matrix comes first and which comes second in the product. For this reason, the statement “Multiply *A* on the right by *B*” means to form the product *AB*, while “Multiply *A* on the left by *B*” means to form the product *BA*.

**Example 10**: If

and **x** is the vector (−2, 3), show how *A* can be multiplied on the right by **x** and compute the product.

Since *A* is 2 x 2, in order to multiply *A* on the right by a matrix, that matrix must have 2 rows. Therefore, if **x** is written as the 2 x 1 *column* matrix

then the product *A* **x** can be computed, and the result is another 2 x 1 column matrix:

**Example 11**: Consider the matrices

If *A* is multiplied on the right by *B*, the result is

but if *A* is multiplied on the left by *B*, the result is

Note that both products are defined and of the same size, but they are not equal.

**Example 12**: If *A* and *B* are square matrices such that *AB = BA*, then *A* and *B* are said to *commute*. Show that any two square diagonal matrices of order 2 commute.

Let

be two arbitrary 2 x 2 diagonal matrices. Then

and

Since *a* _{11} *b* _{11} = *b* _{11} *a* _{11} and *a* _{22} *b* _{22} = *b* _{22} *a* _{22}, *AB* does indeed equal *BA*, as desired.

Although matrix multiplication is usually not commutative, it is *sometimes* commutative; for example, if

then

Despite examples such as these, it must be stated that *in general, matrix multiplication is not commutative*.

There is another difference between the multiplication of scalars and the multiplication of matrices. If *a* and *b* are real numbers, then the equation *ab* = 0 implies that *a* = 0 or *b* = 0. That is, the only way a product of real numbers can equal 0 is if at least one of the factors is itself 0. The analogous statement for matrices, however, is not true. For instance, if

then

Note that even though neither *G* nor *H* is a zero matrix, the product *GH* is.

Yet another difference between the multiplication of scalars and the multiplication of matrices is the lack of a general cancellation law for matrix multiplication. If *a, b*, and *c* are real numbers with *a* ≠ 0, then, by canceling out the factor *a*, the equation *ab = ac* implies *b = c*. No such law exists for matrix multiplication; that is, the statement *AB = AC* does *not* imply *B = C*, even if *A* is nonzero. For example, if

then both

and

Thus, even though *AB = AC* and *A* is not a zero matrix, *B* does not equal *C*.

**Example 13**: Although matrix multiplication is not always commutative, it *is* always *associative*. That is, if *A, B*, and *C* are any three matrices such that the product *(AB)C* is defined, then the product *A(BC)* is also defined, and

That is, as long as the order of the factors is unchanged, how they are *grouped* is irrelevant.

Verify the associative law for the matrices

First, since

the product *(AB)C* is

Now, since

the product *A(BC)* is

Therefore, *(AB)C = A(BC)*, as expected. Note that the associative law implies that the product of *A, B*, and *C* (in that order) can be written simply as *ABC*; parentheses are not needed to resolve any ambiguity, because there is no ambiguity.

**Example 14**: For the matrices

verify the equation ( *AB*) ^{T} = *B* ^{T} *A* ^{T}.

First,

implies

Now, since

*B* ^{T} *A* ^{T} does indeed equal ( *AB*) ^{T}. In fact, the equation

holds true for *any* two matrices for which the product *AB* is defined. This says that if the product *AB* is defined, then the transpose of the product is equal to the product of the transposes *in the reverse order*.

**Identity matrices**. The zero matrix *0 *_{m x n} plays the role of the additive identity in the set of *m x n* matrices in the same way that the number 0 does in the set of real numbers (recall Example 7). That is, if *A* is an *m x n* matrix and *0 = 0 *_{m x n} , then

This is the matrix analog of the statement that for any real number *a*,

With an additive identity in hand, you may ask, “What about a *multiplicative* identity?” In the set of real numbers, the multiplicative identity is the number 1, since

Is there a matrix that plays *this* role? Consider the matrices

and verify that

and

Thus, *AI = IA = A*. In fact, it can be easily shown that for this matrix *I*, both products *AI* and *IA* will equal *A* for *any* 2 x 2 matrix *A*. Therefore,

is the multiplicative identity in the set of 2 x 2 matrices. Similarly, the matrix

is the multiplicative identity in the set of 3 x 3 matrices, and so on. (Note that *I* _{3} is the matrix [δ _{ij }] _{3 x 3}.) In general, the matrix *I *_{n} —the *n x n* diagonal matrix with every diagonal entry equal to 1—is called the **identity matrix** of order *n* and serves as the multiplicative identity in the set of all *n x n* matrices.

Is there a multiplicative identity in the set of all *m x n* matrices if *m ≠ n*? For any matrix *A* in *M *_{m x n} ( **R**), the matrix *I *_{m} is the **left identity** ( *I *_{m}A = A ), and *I *_{n} is the **right identity** ( *AI *_{n} = A ). Thus, unlike the set of *n x n* matrices, the set of nonsquare *m x n* matrices does not possess a qunique *two‐sided* identity, because *I *_{m} ≠ I _{n} if *m ≠ n*.

**Example 15**: If *A* is a square matrix, then *A* ^{2} denotes the product *AA,A* ^{3} denotes the product *AAA*, and so forth. If *A* is the matrix

show that *A* ^{3} = − *A*.

The calculation

shows that *A* ^{2} = − *I*. Multiplying both sides of this equation by *A* yields *A* ^{3} = − *A*, as desired. [Technical note: It can be shown that in a certain precise sense, the collection of matrices of the form

where *a* and *b* are real numbers, is structurally identical to the collection of *complex numbers, a + bi*. Since the matrix *A* in this example is of this form (with *a* = 0 and *b* = 1), *A* corresponds to the complex number 0 + 1 *i = i*, and the analog of the matrix equation *A* ^{2} = − *I* derived above is *i* ^{2} = −1, an equation which defines the imaginary unit, *i*.]

**Example 16**: Find a nondiagonal matrix that commutes with

The problem is asking for a nondiagonal matrix *B* such that *AB = BA*. Like *A*, the matrix *B* must be 2 x 2. One way to produce such a matrix *B* is to form *A* ^{2}, for if *B = A* ^{2}, associativity implies

(This equation proves that *A* ^{2} will commute with *A* for *any* square matrix *A*; furthermore, it suggests how one can prove that *every* integral power of a square matrix *A* will commute with *A*.)

In this case,

which is nondiagonal. This matrix *B* does indeed commute with *A*, as verified by the calculations

and

**Example 17**: If

prove that

for every positive integer *n*.

A few preliminary calculations illustrate that the given formula does hold true:

However, to establish that the formula holds for *all* positive integers *n*, a general proof must be given. This will be done here using *the principle of mathematical induction*, which reads as follows. Let *P(n)* denote a proposition concerning a positive integer *n*. If it can be shown that

and

then the statement *P(n)* is valid for *all* positive integers *n*. In the present case, the statement *P(n)* is the assertion

Because *A* ^{1} = *A*, the statement *P*(1) is certainly true, since

Now, assuming that *P(n)* is true, that is, assuming

it is now necessary to establish the validity of the statement *P*( *n* + 1), which is

But this statement does indeed hold, because

By the principle of mathematical induction, the proof is complete.

**The inverse of a matrix**. Let *a* be a given real number. Since 1 is the multiplicative identity in the set of real numbers, if a number *b* exists such that

then *b* is called the *reciprocal* or *multiplicative inverse* of *a* and denoted *a* ^{−1} (or 1/ *a*). The analog of this statement for square matrices reads as follows. Let *A* be a given *n* x *n* matrix. Since *I* = *I *_{n} is the multiplicative identity in the set of *n* x *n* matrices, if a matrix *B* exists such that

then *B* is called the (multiplicative) **inverse** of *A* and denoted *A* ^{−1} (read “ *A* inverse”).

**Example 18**: If

then

since

and

Yet another distinction between the multiplication of scalars and the multiplication of matrices is provided by the existence of inverses. Although every nonzero real number has an inverse, *there exist nonzero matrices that have no inverse*.

**Example 19**: Show that the nonzero matrix

has no inverse.

If this matrix had an inverse, then

for some values of *a, b, c*, and *d*. However, since the second row of *A* is a zero row, you can see that the second row of the product must also be a zero row:

(When an asterisk, *, appears as an entry in a matrix, it implies that the actual value of this entry is irrelevant to the present discussion.) Since the (2, 2) entry of the product cannot equal 1, the product cannot equal the identity matrix. Therefore, it is impossible to construct a matrix that can serve as the inverse for *A*.

If a matrix has an inverse, it is said to be **invertible**. The matrix in Example 23 is invertible, but the one in Example 24 is not. Later, you will learn various criteria for determining whether a given square matrix is invertible.

**Example 20**: Example 18 showed that

Given that

verify the equation ( *AB*) ^{−1} = *B* ^{−1} *A* ^{−1}.

First, compute *AB*:

Next, compute *B* ^{−1} *A* ^{−1}:

Now, since the product of *AB* and *B* ^{−1} *A* ^{−1} is *I*,

*B* ^{−1} *A* ^{−1} is indeed the inverse of *AB*. In fact, the equation

holds true for *any* invertible square matrices of the same size. This says that if *A* and *B* are invertible matrices of the same size, then their product *AB* is also invertible, and the inverse of the product is equal to the product of the inverses *in the reverse order*. (Compare this equation with the one involving transposes in Example 14 above.) This result can be proved in general by applying the associative law for matrix multiplication. Since

and

it follows that ( *AB*) ^{−1} = *B* ^{−1} *A* ^{−1}, as desired.

**Example 21**: The inverse of the matrix

is

Show that the inverse of *B* ^{T} is ( *B* ^{−1}) ^{T}.

Form *B* ^{T} and ( *B* ^{−1}) ^{T} and multiply:

This calculation shows that ( *B* ^{−1}) ^{T} is the inverse of *B* ^{T}. [Strictly speaking, it shows only that ( *B* ^{−1}) ^{T} is the *right inverse* of *B* ^{T}, that is, when it multiplies *B* ^{T} on the right, the product is the identity. It is also true that ( *B* ^{−1}) ^{T} *B* ^{T} = *I*, which means ( *B* ^{−1}) ^{T} is the *left inverse* of *B* ^{T}. However, it is not necessary to explicitly check both equations: If a square matrix has an inverse, there is no distinction between a left inverse and a right inverse.] Thus,

an equation which actually holds for *any* invertible square matrix *B*. This equation says that if a matrix is invertible, then so is its transpose, and the inverse of the transpose is the transpose of the inverse.

**Example 22**: Use the distributive property for matrix multiplication, *A*( *B* ± *C*) = *AB* ± *AC*, to answer this question: If a 2 x 2 matrix *D* satisfies the equation *D* ^{2} − *D* − 6 *I* = *0*, what is an expression for *D* ^{−1}?

By the distributive property quoted above, *D* ^{2} − *D* = *D* ^{2} − *DI* = *D(D − I)*. Therefore, the equation *D* ^{2} − *D* − 6 *I* = *0* implies *D(D − I)* = 6 *I*. Multiplying both sides of this equation by 1/6 gives

which implies

As an illustration of this result, the matrix

satisfies the equation *D* ^{2} − *D* − 6 *I* = *0*, as you may verify. Since

and

the matrix 1/6 ( *D−I*) does indeed equal *D* ^{−1}, as claimed.

**Example 23**: The equation ( *a* + *b*) ^{2} = *a* ^{2} + 2 *ab* + *b* ^{2} is an identity if *a* and *b* are real numbers. Show, however, that ( *A* + *B*) ^{2} = *A* ^{2} + 2 *AB* + *B* ^{2} is *not* an identity if *A* and *B* are 2 x 2 matrices. [Note: The distributive laws for matrix multiplication are *A*( *B* ± *C*) = *AB* ± *AC*, given in Example 22, and the companion law, ( *A* ± *B*) *C* = *AC* ± *BC*.]

The distributive laws for matrix multiplication imply

Since matrix multiplication is not commutative, *BA* will usually not equal *AB*, so the sum *BA* + *AB* cannot be written as 2 *AB*. In general, then, ( *A* + *B*) ^{2} ≠ *A* ^{2} + 2 *AB* + *B* ^{2}. [Any matrices *A* and *B* that do not commute (for example, the matrices in Example 16 above) would provide a specific counterexample to the statement ( *A* + *B*) ^{2} = *A* ^{2} + 2 *AB* + *B* ^{2}, which would also establish that this is not an identity.]

**Example 24**: Assume that *B* is invertible. If *A* commutes with *B*, show that *A* will also commute with *B* ^{−1}.

*Proof*. To say “ *A* commutes with *B*” means *AB* = *BA*. Multiply this equation by *B* ^{−1} on the left and on the right and use associativity:

**Example 25**: The number 0 has just one square root: 0. Show, however, that the (2 by 2) zero matrix has infinitely many square roots by finding all 2 x 2 matrices *A* such that *A* ^{2} = *0*.

In the same way that a number *a* is called a square root of *b* if *a* ^{2} = *b*, a matrix *A* is said to be a square root of *B* if *A* ^{2} = *B*. Let

be an arbitrary 2 x 2 matrix. Squaring it and setting the result equal to *0* gives

The (1, 2) entries in the last equation imply *b*( *a* + *d*) = 0, which holds if (Case 1) *b* = 0 or (Case 2) *d* = − *a*.

Case 1. If *b* = 0, the diagonal entries then imply *a* = 0 and *d* = 0, and the (2, 1) entries imply that *c* is arbitrary. Thus, for any value of *c*, every matrix of the form

is a square root of *0* _{2x2}.

Case 2. If *d* = − *a*, then the off‐diagonal entries will both be 0, and the diagonal entries will both equal *a* ^{2} + *bc*. Thus, as long as *b* and *c* are chosen so that *bc* = − *a* ^{2}, *A* ^{2} will equal *0*.

A similar chain of reasoning beginning with the (2, 1) entries leads to either *a* = *c* = *d* = 0 (and *b* arbitrary) or the same conclusion as before: as long as *b* and *c* are chosen so that *bc* = − *a* ^{2}, the matrix *A* ^{2} will equal *0*.

All these cases can be summarized as follows. Any matrix of the following form will have the property that its square is the 2 by 2 zero matrix:

Since there are infinitely many values of *a, b*, and *c* such that *bc* = − *a* ^{2}, the zero matrix *0* _{2x2} has infinitely many square roots. For example, choosing *a* = 4, *b* = 2, and *c* = −8 gives the nonzero matrix

whose square is