[MATH-326] getLInfNorm() uses wrong formula in both ArrayRealVector and OpenMapRealVector (in different ways)

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.0
• Fix Version/s: 2.1

Environment

all

• Assignee: Unassigned
• Reporter: Jake Mannix
• Created: 29/Dec/09 00:09
• Updated: 03/Apr/10 20:37
• Resolved: 29/Dec/09 12:26

Description

the L_infinity norm of a finite dimensional vector is just the max of the absolute value of its entries.

The current implementation in ArrayRealVector has a typo:

    public double getLInfNorm() {
        double max = 0;
        for (double a : data) {
            max += Math.max(max, Math.abs(a));
        }
        return max;
    }

the += should just be an =.

There is sadly a unit test assuring us that this is the correct behavior (effectively a regression-only test, not a test for correctness).

Worse, the implementation in OpenMapRealVector is not even positive semi-definite:

   
    public double getLInfNorm() {
        double max = 0;
        Iterator iter = entries.iterator();
        while (iter.hasNext()) {
            iter.advance();
            max += iter.value();
        }
        return max;
    }

I would suggest that this method be moved up to the AbstractRealVector superclass and implemented using the sparseIterator():

  public double getLInfNorm() {
    double norm = 0;
    Iterator<Entry> it = sparseIterator();
    Entry e;
    while(it.hasNext() && (e = it.next()) != null) {
      norm = Math.max(norm, Math.abs(e.getValue()));
    }
    return norm;
  }

Unit tests with negative valued vectors would be helpful to check for this kind of thing in the future.