In Euclidean geometry, the Fermat point of a triangle, also called the Torricelli point or Fermat–Torricelli point, is a point such that the sum of the three distances from each of the three vertices of the triangle to the point is the smallest possible[1] or, equivalently, the geometric median of the three vertices. It is so named because this problem was first raised by Fermat in a private letter to Evangelista Torricelli, who solved it.
The Fermat point gives a solution to the geometric median and Steiner tree problems for three points.
The Fermat point of a triangle with largest angle at most 120° is simply its first isogonic center or X(13),[2] which is constructed as follows:
An alternative method is the following:
When a triangle has an angle greater than 120°, the Fermat point is sited at the obtuse-angled vertex.
In what follows "Case 1" means the triangle has an angle exceeding 120°. "Case 2" means no angle of the triangle exceeds 120°.
Fig. 2 shows the equilateral triangles attached to the sides of the arbitrary triangle .Here is a proof using properties of concyclic points to show that the three lines in Fig 2 all intersect at the point and cut one another at angles of 60°.
The triangles are congruent because the second is a 60° rotation of the first about . Hence and . By the converse of the inscribed angle theorem applied to the segment, the points are concyclic (they lie on a circle). Similarly, the points are concyclic.
, so, using the inscribed angle theorem. Similarly, .
So . Therefore, . Using the inscribed angle theorem, this implies that the points are concyclic. So, using the inscribed angle theorem applied to the segment, . Because, the point lies on the line segment . So, the lines are concurrent (they intersect at a single point). Q.E.D.
This proof applies only in Case 2, since if, point lies inside the circumcircle of which switches the relative positions of and . However it is easily modified to cover Case 1. Then hence which means is concyclic so . Therefore, lies on .
The lines joining the centers of the circles in Fig. 2 are perpendicular to the line segments . For example, the line joining the center of the circle containing and the center of the circle containing, is perpendicular to the segment . So, the lines joining the centers of the circles also intersect at 60° angles. Therefore, the centers of the circles form an equilateral triangle. This is known as Napoleon's Theorem.
Given any Euclidean triangle and an arbitrary point let
d(P)=|PA|+|PB|+|PC|.
d(P0)<d(P)
P\neP0.
A key result that will be used is the dogleg rule, which asserts that if a triangle and a polygon have one side in common and the rest of the triangle lies inside the polygon then the triangle has a shorter perimeter than the polygon:
If is the common side, extend to cut the polygon at the point . Then the polygon's perimeter is, by the triangle inequality:
perimeter>|AB|+|AX|+|XB|=|AB|+|AC|+|CX|+|XB|\geq|AB|+|AC|+|BC|.
Let be any point outside . Associate each vertex with its remote zone; that is, the half-plane beyond the (extended) opposite side. These 3 zones cover the entire plane except for itself and clearly lies in either one or two of them. If is in two (say the and zones’ intersection) then setting
P'=A
d(P')=d(A)<d(P)
d(P')<d(P)
d(P')<d(P).
Case 1. The triangle has an angle ≥ 120°.
Without loss of generality, suppose that the angle at is ≥ 120°. Construct the equilateral triangle and for any point in (except itself) construct so that the triangle is equilateral and has the orientation shown. Then the triangle is a 60° rotation of the triangle about so these two triangles are congruent and it follows that
d(P)=|CP|+|PQ|+|QF|
|AC|+|AF|=d(A).
d(A)<d(P)
P\in\Delta,P\neA.
P'\in\Omega
d(P')<d(P)
d(A)\leqd(P')
d(A)<d(P)
d(A)<d(P)
P\neA
Case 2. The triangle has no angle ≥ 120°.
Construct the equilateral triangle, let be any point inside, and construct the equilateral triangle . Then is a 60° rotation of about so
d(P)=|PA|+|PB|+|PC|=|AP|+|PQ|+|QD|
d(P0=|AD|.
P\neP0
d(P0)=|AD|<d(P).
P'\in\Omega
d(P')<d(P)
d(P0)\leqd(P')
d(P0)<d(P)
Let be any five points in a plane. Denote the vectors
\overrightarrow{OA}, \overrightarrow{OB}, \overrightarrow{OC}, \overrightarrow{OX}
\begin{align} |a|&=a ⋅ i=(a-x) ⋅ i+x ⋅ i\leq|a-x|+x ⋅ i,\\ |b|&=b ⋅ j=(b-x) ⋅ j+x ⋅ j\leq|b-x|+x ⋅ j,\\ |c|&=c ⋅ k=(c-x) ⋅ k+x ⋅ k\leq|c-x|+x ⋅ k. \end{align}
|a|+|b|+|c|\leq|a-x|+|b-x|+|c-x|+x ⋅ (i+j+k).
If meet at at angles of 120° then, so
|a|+|b|+|c|\leq|a-x|+|b-x|+|c-x|
|OA|+|OB|+|OC|\leq|XA|+|XB|+|XC|
This argument fails when the triangle has an angle because there is no point where meet at angles of 120°. Nevertheless, it is easily fixed by redefining and placing at so that . Note that because the angle between the unit vectors is which exceeds 120°. Since
|0|\leq|0-x|+x ⋅ k,
Another approach to finding the point within a triangle, from which the sum of the distances to the vertices of the triangle is minimal, is to use one of the mathematical optimization methods; specifically, the method of Lagrange multipliers and the law of cosines.
We draw lines from the point within the triangle to its vertices and call them . Also, let the lengths of these lines be respectively. Let the angle between and be, and be . Then the angle between and is . Using the method of Lagrange multipliers we have to find the minimum of the Lagrangian, which is expressed as:
L=x+y+z+λ1(x2+y2-2xy\cos(\alpha)-a2)+λ2(y2+z2-2yz\cos(\beta)-b2)+λ3(z2+x2-2zx\cos(\alpha+\beta)-c2)
where are the lengths of the sides of the triangle.
Equating each of the five partial derivatives
\tfrac{\partialL}{\partialx},\tfrac{\partialL}{\partialy},\tfrac{\partialL}{\partialz},\tfrac{\partialL}{\partial\alpha},\tfrac{\partialL}{\partial\beta}
\begin{align} &\csc\left(A+\tfrac{\pi}{3}\right):\csc\left(B+\tfrac{\pi}{3}\right):\csc\left(C+\tfrac{\pi}{3}\right)\\ &=\sec\left(A-\tfrac{\pi}{6}\right):\sec\left(B-\tfrac{\pi}{6}\right):\sec\left(C-\tfrac{\pi}{6}\right). \end{align}
\begin{align} &\csc\left(A-\tfrac{\pi}{3}\right):\csc\left(B-\tfrac{\pi}{3}\right):\csc\left(C-\tfrac{\pi}{3}\right)\\ &=\sec\left(A+\tfrac{\pi}{6}\right):\sec\left(B+\tfrac{\pi}{6}\right):\sec\left(C+\tfrac{\pi}{6}\right). \end{align}
1-u+uvw\sec\left(A-\tfrac{\pi}{6}\right):1-v+uvw\sec\left(B-\tfrac{\pi}{6}\right):1-w+uvw\sec\left(C-\tfrac{\pi}{6}\right)
where respectively denote the Boolean variables .
\sin\left(A+\tfrac{\pi}{3}\right):\sin\left(B+\tfrac{\pi}{3}\right):\sin\left(C+\tfrac{\pi}{3}\right).
\sin\left(A-\tfrac{\pi}{3}\right):\sin\left(B-\tfrac{\pi}{3}\right):\sin\left(C-\tfrac{\pi}{3}\right).
The isogonic centers X(13) and X(14) are also known as the first Fermat point and the second Fermat point respectively. Alternatives are the positive Fermat point and the negative Fermat point. However these different names can be confusing and are perhaps best avoided. The problem is that much of the literature blurs the distinction between the Fermat point and the first Fermat point whereas it is only in Case 2 above that they are actually the same.
This question was proposed by Fermat, as a challenge to Evangelista Torricelli. He solved the problem in a similar way to Fermat's, albeit using the intersection of the circumcircles of the three regular triangles instead. His pupil, Viviani, published the solution in 1659.