What is the Hadamard product?

The Hadamard product, written A circle-dot B, is the element-wise product of two matrices of the same shape. Each entry of the result equals the product of the corresponding entries in A and B. It is also called the Schur product or the element-wise matrix product.

How is the Hadamard product different from standard matrix multiplication?

The Hadamard product multiplies matrices cell by cell, requires identical shapes, and produces a result of the same shape. Standard matrix multiplication pairs rows of A with columns of B, requires the inner dimensions to match, and produces a result whose shape depends on the outer dimensions. They are different operations with different shape rules and different results.

Can you take the Hadamard product of matrices of different shapes?

No. Both matrices must have exactly the same number of rows and the same number of columns. Without matched shapes there is no pairing of corresponding entries, so the element-wise rule is not defined.

Is the Hadamard product commutative?

Yes. Because every entry of the result is just the scalar product of the matching entries in A and B, the Hadamard product inherits the commutativity of scalar multiplication: A circle-dot B equals B circle-dot A. It is also associative and distributes over matrix addition.

Where is the Hadamard product used?

It appears in machine learning for gating and masking, in image processing for pixel-wise filters, in statistics for weighted scaling, and in numerical linear algebra for preconditioning. The common pattern is operations on matrix-shaped data where rows and columns should not be mixed.

Hadamard Product (element-wise)

Symbolic visualization of A ⊙ B = C, cell by cell. Not to be confused with standard matrix multiplication.

Dimensions (shared by A and B)?The Hadamard product (denoted A ⊙ B) multiplies element-by-element: each entry of the result is the product of the corresponding entries of the two operands. It requires them to have the same shape — the result keeps that same shape. The same idea applies to vectors and to matrices. Note: this is NOT the standard matrix product A × B, which pairs rows with columns and has different shape requirements; the Hadamard product just pairs entries.

A, B2×3

A2×3

a1,1

a1,2

a1,3

a2,1

a2,2

a2,3

⊙

B2×3

b1,1

b1,2

b1,3

b2,1

b2,2

b2,3

C2×3

Step 1 / 8

Step explanations

1Hadamard product (element-wise)

Both A and B are 2×3. To compute C = A ⊙ B, pair up each cell of A with its counterpart in B and multiply them. The result C has the same shape.

Identical shapes required - the opposite of what standard matrix multiplication asks for. Learn more about the opening scene · how it differs from AB

Key Terms

Getting Started with the Visualizer

Reading the Scene Player

The Opening Scene: Two Matrices, Same Shape

One Cell at a Time

The Completed Element-wise Product

Choosing Dimensions

What the Hadamard Product Is

Hadamard vs Standard Matrix Multiplication

Key Properties

Where the Hadamard Product Appears

Worked Example

Common Mistakes

Related Concepts

Key Terms

Hadamard product — the element-wise product of two matrices of the same shape, denoted

A \odot B

. Each entry of the result is the product of the corresponding entries:

c_{i,j} = a_{i,j} \cdot b_{i,j}

.

Schur product — alternative name for the Hadamard product.

Element-wise (pointwise) operation — an operation applied independently to each pair of corresponding entries; the result at

(i,j)

depends only on the inputs at

(i,j)

.

Same-shape requirement — both operands must have identical dimensions. A

2 \times 3

matrix cannot be Hadamard-multiplied with a

3 \times 2

.

Standard matrix product — the row-by-column product

A \times B

, a different operation with different shape rules and a different result.

$\odot$ symbol — the circle-dot operator, the standard notation distinguishing the Hadamard product from

A \times B

AB

Getting Started with the Visualizer

Set the shape, then watch

A \odot B = C

build one cell at a time.

• Use the Dimensions steppers to set the shared shape of

A

and

B

(1 to 5 in each direction)
• Hover the ? icon for a full explanation of the Hadamard product and how it differs from the standard matrix product
• Press play on the scene player or step manually with the back and next buttons
• Adjust speed and use the step log on the right to scroll through completed cells

There is no operation toggle — the Hadamard product is a single operation, fully determined by the shape of the operands.

Reading the Scene Player

Each scene focuses on one cell of

C

and combines highlights with arrows.

• The active cell in

A

is highlighted primary, the matching cell in

B

as secondary, and the destination cell in

C

as accent
• Two curved arrows flow from

a_{i,j}

and

b_{i,j}

into

c_{i,j}

, making the data flow explicit
• Each filled cell of

C

shows its symbolic content

a_{i,j} \cdot b_{i,j}

• The step log on the right keeps a record of every completed cell

By the final scene, every cell of

C

contains its symbolic product and the operation is complete.

The Opening Scene: Two Matrices, Same Shape

The player starts with

A

and

B

side by side and

C

empty on the right, all three

2 \times 3

at the default dimensions.

The setup looks exactly like the addition tool's, and that is not a coincidence: the Hadamard product has the same precondition and the same output shape. Only the operation applied to each pair differs.

Opening scene, frozen

A and B at 2×3 with C empty. The same setup as matrix addition - identical shapes in, identical shape out - which is exactly what standard multiplication does not require.

Getting the precondition right matters more here than anywhere else in this section, because the operation shares a name with something that requires the opposite. Standard matrix multiplication needs the inner dimensions to agree — an

m \times n

times an

n \times p

— and produces an

m \times p

result. The Hadamard product needs the shapes to be identical and returns that same shape.

So

A \odot B

is defined here where

AB

is not: two

2 \times 3

matrices cannot be multiplied in the standard sense at all, since

3 \neq 2

One Cell at a Time

Each step highlights a cell of

A

, the cell in the same position of

B

, and the destination in

C

, then writes the plain product

a_{i,j} \cdot b_{i,j}

into it.

The frozen picture below is a step partway through the

2 \times 3

run: some cells of

C

already hold their product, one pair is being multiplied, and the rest are placeholders.

Mid-sweep, frozen

One cell of A, the cell in the same position of B, and the destination in C. Two numbers multiplied, no row or column involved.

Compare this sweep against standard matrix multiplication and the difference is stark. There, a single output entry consumes an entire row of

A

and an entire column of

B

, and is a sum of products. Here, an output entry consumes exactly two numbers and is a single product.

That is why the Hadamard product is cheap —

mn

multiplications against the roughly

mnp

of the standard product — and why it parallelises trivially. Nothing in one cell depends on anything in another.

The Completed Element-wise Product

The final scene fills every cell, so

C

reads

c_{i,j} = a_{i,j} \, b_{i,j}

throughout, at the same

2 \times 3

shape it started with.

Written that way, the Hadamard product is to multiplication what matrix addition is to addition: the underlying arithmetic applied entry by entry, with no mixing across positions.

Completed product, frozen

All six cells filled with the plain product of the two entries above, and C still 2×3. No mixing across positions at any point.

The properties follow from the entries, as they did for addition. It is commutative,

A \odot B = B \odot A

— which standard matrix multiplication emphatically is not — and associative, and it distributes over matrix addition. Its identity is the all-ones matrix, not the identity matrix

I

: multiplying element-wise by

I

would zero out everything off the diagonal.

That last point is the cleanest illustration that the two products are genuinely different operations wearing similar notation. An element-wise inverse exists only when every entry of

A

is non-zero, and is found by reciprocating each entry — nothing like a matrix inverse, which needs a non-zero determinant and mixes every entry into every other.

Choosing Dimensions

The dimension steppers control the shape shared by all three matrices.

• Start with a small shape like

2 \times 2

2 \times 3

to see each cell pairing clearly
• Increase to

4 \times 4

5 \times 5

to see the operation scale — total scenes equal

m \times n

• Symbolic cell contents in

C

shrink automatically when the shape is large, so

a_{i,j} \cdot b_{i,j}

stays readable at

5 \times 5

• Square (

n \times n

) and rectangular (

m \times n

m \neq n

) shapes follow identical rules

The shape of

C

is forced to match the shared shape of

A

and

B

— no separate control needed.

What the Hadamard Product Is

The Hadamard product is the element-wise product of two matrices of the same shape. For

m \times n

matrices

A

and

B

C = A \odot B, \quad c_{i,j} = a_{i,j} \cdot b_{i,j}

Each entry of

C

depends only on the matching pair of entries in

A

and

B

— no row or column interaction. The result

C

has the same

m \times n

shape as the operands.

The Hadamard product is the multiplicative analogue of matrix addition: both are element-wise, both require matched shapes, both preserve dimensions. It contrasts sharply with the standard matrix product

A \times B

, which mixes rows with columns and changes shape.

For comprehensive theory, see matrix operations.

Hadamard vs Standard Matrix Multiplication

These two operations share the word "multiplication" but are fundamentally different.

• Notation:

A \odot B

(Hadamard) vs

A B

A \times B

(standard)
• Shape requirement: identical shapes (Hadamard) vs inner dimensions match (standard,

m \times k

times

k \times n

)
• Result shape: same as operands (Hadamard) vs

m \times n

(standard)
• Computation: pairwise product per cell (Hadamard) vs sum of row-column products (standard)
• Commutativity: commutative (Hadamard) vs generally non-commutative (standard)

For the standard product, see matrix multiplication. The two are confused often enough that the Hadamard product is sometimes spelled out as "element-wise product" to avoid ambiguity.

Key Properties

The Hadamard product satisfies the algebraic properties one would expect from an element-wise multiplication.

• Commutativity:

A \odot B = B \odot A

• Associativity:

(A \odot B) \odot C = A \odot (B \odot C)

• Distributivity over addition:

A \odot (B + C) = A \odot B + A \odot C

• Identity: the all-ones matrix

J

acts as identity:

A \odot J = A

• Scalar pull-out:

(kA) \odot B = k(A \odot B)

• Transpose:

(A \odot B)^T = A^T \odot B^T

These properties mirror ordinary scalar multiplication exactly — which is unsurprising, since the operation is just scalar multiplication applied entry by entry.

Where the Hadamard Product Appears

The Hadamard product shows up wherever data lives in matrix form but the operation needs to be local.

• Machine learning: masking, gating in LSTMs and GRUs, attention weights applied element-wise
• Image processing: pointwise filters and masks applied to pixel grids
• Statistics: covariance scaling, weighted moment computations
• Numerical linear algebra: preconditioners and diagonal scaling can be expressed as Hadamard products
• Signal processing: windowing and apodization

The common thread: the matrix shape carries spatial or indexing structure, but the operation itself should not mix rows or columns.

Worked Example

Take

A

and

B

2 \times 3

matrices:

A = \begin{pmatrix} 1 & 2 & 3 \\ 4 & 5 & 6 \end{pmatrix}, \quad B = \begin{pmatrix} 7 & 8 & 9 \\ 0 & 1 & 2 \end{pmatrix}

Then

C = A \odot B

is computed cell by cell:

C = \begin{pmatrix} 1 \cdot 7 & 2 \cdot 8 & 3 \cdot 9 \\ 4 \cdot 0 & 5 \cdot 1 & 6 \cdot 2 \end{pmatrix} = \begin{pmatrix} 7 & 16 & 27 \\ 0 & 5 & 12 \end{pmatrix}

Compare with the standard product

A \times B

: it is undefined here because

A

2 \times 3

and

B

is also

2 \times 3

— the inner dimensions (

3

and

2

) do not match. The Hadamard product has no such restriction beyond matching shapes.

Common Mistakes

The Hadamard product is simple, but a few recurring mistakes appear.

• Confusing it with the standard matrix product — by far the most common; check the notation (

\odot

vs juxtaposition) and the shape rules
• Trying to Hadamard-multiply matrices of different shapes — there is no broadcasting rule in the strict mathematical definition
• Assuming the result is a different shape —

A \odot B

has the same shape as both operands, unlike standard multiplication
• Using $\cdot$ or $\times$ for the Hadamard product — these notations strongly suggest the standard product and cause confusion
• Forgetting the operation is commutative — unlike standard matrix multiplication,

A \odot B = B \odot A

always holds