jax-ml/jax · error · ValueError

Expected 'output' to be either 'real' or 'complex', got {out

Error message

Expected 'output' to be either 'real' or 'complex', got {output=}.

What it means

jax.scipy.linalg.schur requires the `output` argument to be exactly the string 'real' or 'complex'; anything else (including typos, wrong case, or None) raises ValueError before dispatching to the underlying Schur decomposition.

Source

Thrown at jax/_src/scipy/linalg.py:575

    upper-triangular in this case because the input matrix is symmetric:

    >>> T  # doctest: +SKIP
    Array([[-2.0000005 ,  0.5066295 , -0.43360388],
           [ 0.        ,  1.5505103 ,  0.74519426],
           [ 0.        ,  0.        ,  6.449491  ]], dtype=float32)

    The transformation matrix ``Z`` is unitary:

    >>> jnp.allclose(Z.T @ Z, jnp.eye(3), atol=1E-5)
    Array(True, dtype=bool)

    The input can be reconstructed from the outputs:

    >>> jnp.allclose(Z @ T @ Z.T, a)
    Array(True, dtype=bool)
  """
  if output not in ('real', 'complex'):
    raise ValueError(
      f"Expected 'output' to be either 'real' or 'complex', got {output=}.")
  return _schur(ensure_arraylike("scipy.schur", a), output)


def inv(a: ArrayLike, overwrite_a: bool = False, check_finite: bool = True) -> Array:
  """Return the inverse of a square matrix

  JAX implementation of :func:`scipy.linalg.inv`.

  Args:
    a: array of shape ``(..., N, N)`` specifying square array(s) to be inverted.
    overwrite_a: unused in JAX
    check_finite: unused in JAX

  Returns:
    Array of shape ``(..., N, N)`` containing the inverse of the input.

  Notes:

View on GitHub (pinned to 1e1c6a8fc0)

Solutions

1. Pass output='real' for real Schur form or output='complex' for complex Schur form explicitly
2. If building output dynamically, validate/normalize the value before calling schur

Example fix

// before
T, Z = jax.scipy.linalg.schur(A, output=np.iscomplexobj(A))
// after
T, Z = jax.scipy.linalg.schur(A, output='complex' if np.iscomplexobj(A) else 'real')

null

def is_valid_schur_output(o: str) -> bool: return o in ('real', 'complex')

try: T, Z = jax.scipy.linalg.schur(A, output=out) except ValueError as e: if 'output' in str(e): out = 'complex' if np.iscomplexobj(A) else 'real'; T, Z = jax.scipy.linalg.schur(A, output=out) else: raise

When it happens

Trigger: Calling jax.scipy.linalg.schur(A, output='r'), output='REAL', output=None, or any other string.

Common situations: Passing a variable that was never validated, or assuming schur defaults output to None like older SciPy versions; porting code where output was optional.

Understand the failure class

Background: "Must be a positive integer", "Invalid value", "Unsupported": the invalid-argument-value error family, when a library rejects the value you pass — this error's family across 35 libraries.

Related errors

• ind must be a positive integer; got {ind=}
• multi_dot requires at least two arrays; got len(arrays)={len
• Unsupported QR decomposition mode '{mode}'
• mode must be 'right' or 'left', got {mode!r}
• Expected assume_a to be one of {valid_assume_a}; got {assume

AI-assisted analysis of jax-ml/jax@1e1c6a8fc0 (2026-08-27). Data as JSON: /api/errors/933dd7dae11c6b1b. Report an issue: GitHub.