Your scheme is still ambiguous at best!

You are approaching this problem incorrectly IMHO. A solution looking for a problem? It's all old hat at this stage.

- Cuchulainn
**Posts:**64438**Joined:****Location:**Drosophila melanogaster-
**Contact:**

Your scheme is still ambiguous at best!

You are approaching this problem incorrectly IMHO. A solution looking for a problem? It's all old hat at this stage.

You are approaching this problem incorrectly IMHO. A solution looking for a problem? It's all old hat at this stage.

"Compatibility means deliberately repeating other people's mistakes."

David Wheeler

http://www.datasimfinancial.com

http://www.datasim.nl

David Wheeler

http://www.datasimfinancial.com

http://www.datasim.nl

What do you mean by ambiguous ? This is a perfect example of a very simple upwind scheme, and AFAIU it is exactly the limit case of the exponential fitting.Your scheme is still ambiguous at best!

You are approaching this problem incorrectly IMHO. A solution looking for a problem? It's all old hat at this stage.

I think that this proves that one can't define a stochastic matrix using exponential fitting. Is that correct ?

Anyhow, again, without moving the mesh, it is not possible to define a stochastic matrix for the underlying PDE problem, unless massively twisting the boundary conditions.

In this thread, I thought that we were investigating properties of monotone schemes for Finance applications ?A solution looking for a problem? It's all old hat at this stage.

A quite important requirement for pricing is also that the scheme defines a stochastic transition matrix. If I summarize the picture:

1) Such schemes exist in one or two dimensions: trees based methods. They are part of PDE schemes.

2) All classical PDE schemes based on finite differences on fixed grid should fail.

3) One can define such monotone PDE schemes, defining stochastic transition matrix, at the only condition of moving the grid. The minimum being moving the grid according to the advection terms. Note also that there exists schemes moving also the mesh accordingly to the diffusive part: these are the schemes I am using.

- Cuchulainn
**Posts:**64438**Joined:****Location:**Drosophila melanogaster-
**Contact:**

It's unstable as I already said, You are mixing up forward and backward time. Please clear this up.What do you mean by ambiguous ? This is a perfect example of a very simple upwind scheme, and AFAIU it is exactly the limit case of the exponential fitting.

You are approaching this problem incorrectly IMHO. A solution looking for a problem? It's all old hat at this stage.

take initial value problem!

"Compatibility means deliberately repeating other people's mistakes."

David Wheeler

http://www.datasimfinancial.com

http://www.datasim.nl

David Wheeler

http://www.datasimfinancial.com

http://www.datasim.nl

It is not anymore, I already cleaned it up see above: [$]u_i^{n+1}=u_i^{n} +\frac{dt}{dx}(u_{i+1}^{n} - u_{i}^{n})[$] gives [$] A = (..0,1-\frac{dt}{dx},+\frac{dt}{dx},0...)[$], and of course [$]\frac{dt}{dx} \le 1[$]. It is a perfectly stable, monotone, first order, upwind scheme. Yet, you won't succeed defining a stochastic transition matrix with this scheme.It's unstable as I already said, You are mixing up forward and backward time. Please clear this up.What do you mean by ambiguous ? This is a perfect example of a very simple upwind scheme, and AFAIU it is exactly the limit case of the exponential fitting.

You are approaching this problem incorrectly IMHO. A solution looking for a problem? It's all old hat at this stage.

take initial value problem!

- Cuchulainn
**Posts:**64438**Joined:****Location:**Drosophila melanogaster-
**Contact:**

Ah, OK you had a typo.

Write it down 500 times.

Write it down 500 times.

"Compatibility means deliberately repeating other people's mistakes."

David Wheeler

http://www.datasimfinancial.com

http://www.datasim.nl

David Wheeler

http://www.datasimfinancial.com

http://www.datasim.nl

- Cuchulainn
**Posts:**64438**Joined:****Location:**Drosophila melanogaster-
**Contact:**

Now we are on same wavelength, your correct scheme still suffers from [$]CFL \lt 1[$]. The next scheme has no such restriction (and it is monotone): using your notation

[$]u_i^{n+1}=u_i^{n} +\frac{dt}{dx}(u_{i+1}^{n} - u_{i}^{n+1})[$]

or

[$]u_i^{n+1} (1 + \frac{dt}{dx}) =u_i^{n} +\frac{dt}{dx}(u_{i+1}^{n})[$]

Monotone!

QED

[$]u_i^{n+1}=u_i^{n} +\frac{dt}{dx}(u_{i+1}^{n} - u_{i}^{n+1})[$]

or

[$]u_i^{n+1} (1 + \frac{dt}{dx}) =u_i^{n} +\frac{dt}{dx}(u_{i+1}^{n})[$]

Monotone!

QED

David Wheeler

http://www.datasimfinancial.com

http://www.datasim.nl

@Cuchullain, the question is not to know if these schemes are monotone or not. They are monotone, under a CFL condition for the explicit one. The question is: does this class of schemes define a stochastic transition matrix ? The answer is no.Now we are on same wavelength, your correct scheme still suffers from [$]CFL \lt 1[$]. The next scheme has no such restriction (and it is monotone): using your notation

[$]u_i^{n+1}=u_i^{n} +\frac{dt}{dx}(u_{i+1}^{n} - u_{i}^{n+1})[$]

or

[$]u_i^{n+1} (1 + \frac{dt}{dx}) =u_i^{n} +\frac{dt}{dx}(u_{i+1}^{n})[$]

Monotone!

QED

- Cuchulainn
**Posts:**64438**Joined:****Location:**Drosophila melanogaster-
**Contact:**

The original thread title was "Monotone Methods", so what has changed? Start a new thread.

David Wheeler

http://www.datasimfinancial.com

http://www.datasim.nl

@Cuchulainn, you are slightly of bad faith here I am expecting an answer : does exponential fitting produces a stochastic transition probability matrix or not ? If you prefer the same question but with another sight view: could I use exponential fitting for my framework or not (as an alternative to the schemes that I am using) ?The original thread title was "Monotone Methods", so what has changed? Start a new thread.

- Cuchulainn
**Posts:**64438**Joined:****Location:**Drosophila melanogaster-
**Contact:**

Your question is too vague.@Cuchulainn, you are slightly of bad faith here I am expecting an answer : does exponential fitting produces a stochastic transition probability matrix or not ? If you prefer the same question but with another sight view: could I use exponential fitting for my framework or not (as an alternative to the schemes that I am using) ?The original thread title was "Monotone Methods", so what has changed? Start a new thread.

David Wheeler

http://www.datasimfinancial.com

http://www.datasim.nl

My question is very precise : does the upwind scheme defines a stochastic transition matrix ? I wrote you the scheme and the transition matrix just few lines above. We both know the answer, and I apologize for pushing this nail into.Your question is too vague.@Cuchulainn, you are slightly of bad faith here I am expecting an answer : does exponential fitting produces a stochastic transition probability matrix or not ? If you prefer the same question but with another sight view: could I use exponential fitting for my framework or not (as an alternative to the schemes that I am using) ?The original thread title was "Monotone Methods", so what has changed? Start a new thread.

What I would like to know is if we can give another answer than negative for the exponential fitting. I think that yes, with boundary conditions, but it will have a cost.

- Cuchulainn
**Posts:**64438**Joined:****Location:**Drosophila melanogaster-
**Contact:**

Do you know

1. Why exponential fitting?

2. Where can it be applied, and

3. Where not?

1. Why exponential fitting?

2. Where can it be applied, and

3. Where not?

David Wheeler

http://www.datasimfinancial.com

http://www.datasim.nl

Exponential fitting seems a good numerical scheme, hence I look forward to read your answers here.Do you know

1. Why exponential fitting?

2. Where can it be applied, and

3. Where not?

- Cuchulainn
**Posts:**64438**Joined:****Location:**Drosophila melanogaster-
**Contact:**

It's OK. But not universal. No silver bullet.

David Wheeler

http://www.datasimfinancial.com

http://www.datasim.nl