torch.js has not been released yet.

torch.reciprocal

function reciprocal(input: Tensor, options?: UnaryOptions): Tensor

Computes the element-wise reciprocal of the input tensor.

Calculates 1/x for every element in the tensor. Essential for:

Normalization: Dividing by scales or standard deviations
Ratio computation: Computing inverse rates or frequencies
Mathematical modeling: Implementing inverse relationships
Numerical methods: Scaling by inverse magnitudes

\text{reciprocal}(x) = \frac{1}{x}

Zero handling: Produces Infinity for input of zero
Identity: reciprocal(reciprocal(x)) = x
Gradient: d/dx[1/x] = -1/x² = -(1/x)² = -reciprocal(x)²

Division by zero: Returns Infinity or NaN for elements that are zero

Parameters

inputTensor: The input tensor (any shape, any dtype)
optionsUnaryOptionsoptional: Optional settings including out parameter

Returns

Tensor– A new tensor with 1/x values element-wise

Examples

// Basic computation
const x = torch.tensor([1, 2, 4]);
torch.reciprocal(x);  // [1.0, 0.5, 0.25]

// Standard deviation inverse
const std = torch.tensor([2.0, 5.0]);
const inv_std = torch.reciprocal(std);  // [0.5, 0.2]

See Also

PyTorch torch.reciprocal()
div - General division: div(1, x) is equivalent to reciprocal(x)
rsqrt - Reciprocal square root (1/√x)