Libctl User Reference

From AbInitio

(Difference between revisions)
Jump to: navigation, search
Revision as of 22:40, 31 March 2008 (edit)
Stevenj (Talk | contribs)
(Miscellaneous utilities)
← Previous diff
Revision as of 22:41, 31 March 2008 (edit)
Stevenj (Talk | contribs)
(Miscellaneous utilities - clarification)
Next diff →
Line 132: Line 132:
; <code>(derivative2 ''function x [dx tolerance]'')</code> ; <code>(derivative2 ''function x [dx tolerance]'')</code>
; <code>(deriv2 ''function x [dx tolerance]'')</code> ; <code>(deriv2 ''function x [dx tolerance]'')</code>
-: Compute the numerical derivative of the given <code>''function''</code> at <code>''x''</code> to within ''at best'' the specified fractional <code>''tolerance''</code> (defaulting to the maximum achievable tolerance), using Ridder's method of polynomial extrapolation. <code>''dx''</code> should be a ''maximum'' displacement in <code>''x''</code> for derivative evaluation; the <code>''function''</code> should change by a significant amount (much larger than the numerical precision) over <code>''dx''</code>. <code>''dx''</code> defaults to 1% of <code>''x''</code> or <code>0.01</code>, whichever is larger. +: Compute the numerical derivative of the given <code>''function''</code> at <code>''x''</code> to within ''at best'' the specified fractional <code>''tolerance''</code> (defaulting to the best achievable tolerance), using Ridder's method of polynomial extrapolation. <code>''dx''</code> should be a ''maximum'' displacement in <code>''x''</code> for derivative evaluation; the <code>''function''</code> should change by a significant amount (much larger than the numerical precision) over <code>''dx''</code>. <code>''dx''</code> defaults to 1% of <code>''x''</code> or <code>0.01</code>, whichever is larger.
:If the return value of <code>derivative</code> is assigned to a variable <code>result</code>, then <code>(derivative-df result)</code> and <code>(derivative-df-err result)</code> give the derivative of the function and an estimate of the numerical error in the derivative, respectively. :If the return value of <code>derivative</code> is assigned to a variable <code>result</code>, then <code>(derivative-df result)</code> and <code>(derivative-df-err result)</code> give the derivative of the function and an estimate of the numerical error in the derivative, respectively.
:The <code>derivative2</code> function computes both the first and second derivatives, using minimal extra function evaluations; the second derivative and its error are then obtained by <code>(derivative-d2f result)</code> and <code>(derivative-d2f-err result)</code>. :The <code>derivative2</code> function computes both the first and second derivatives, using minimal extra function evaluations; the second derivative and its error are then obtained by <code>(derivative-d2f result)</code> and <code>(derivative-d2f-err result)</code>.

Revision as of 22:41, 31 March 2008

libctl
Manual: Introduction
Basic User Experience
Advanced User Experience
User Reference
Developer Experience
Guile and Scheme links
License and Copyright

In this section, we list all of the special functions provided for users by libctl. We do not attempt to document standard Scheme functions, with a couple of exceptions below, since there are plenty of good Scheme references [guile-links.html elsewhere].

Of course, the most important function is:

(help)
Outputs a listing of all the available classes, their properties, default values, and types. Also lists the input and output variables.

Remember, Guile lets you enter expressions and see their values interactively. This is the best way to learn how to use anything that confuses you—just try it and see how it works!

Contents

Basic Scheme functions

(set! variable value)
Change the value of variable to value.
(define variable value)
Define new variable with initial value.
(list [ element1 element2 ... ])
Returns a list consisting of zero or more elements.
(append [ list1 list2 ... ])
Concatenates zero or more lists into a single list.
(function [ arg1 arg2 ... ])
This is how you call a Scheme function in general.
(define (function [ arg1 arg2 ... ]) body)
Define a new function with zero or more arguments that returns the result of given body when it is invoked.

Command-line parameters

(define-param name default-value)
Define a variable name whose value can be set from the command line, and which assumes a value default-value if it is not set. To set the value on the command-line, include name=value on the command-line when the program is executed. In all other respects, name is an ordinary Scheme variable.
(set-param! name new-default-value)
Like set!, but does nothing if name was set on the command line.

All libctl arguments accept the command-line parameter --verbose to turn on a verbose mode. This sets the variable verbose? to true (and, depending on the program, may enable other outputs).

Complex numbers

Scheme includes full support for complex numbers and arithmetic; all of the ordinary operations (+, *, sqrt, etcetera) just work. For the same reason, you can freely use complex numbers in libctl's vector and matrix functions, below.

To specify a complex number a+bi, you simply use the syntax a+bi if a and b are constants, and (make-rectangular a b) otherwise. (You can also specify numbers in "polar" format a*eib by the syntax a@b or (make-polar a b).)

There are a few special functions provided by Scheme to manipulate complex numbers. (real-part z) and (imag-part z) return the real and imaginary parts of z, respectively. (magnitude z) returns the absolute value and (angle z) returns the phase angle. libctl also provides a (conj z) function, below, to return the complex conjugate.

3-vector functions

(vector3 x [y z])
Create a new 3-vector with the given components. If the y or z value is omitted, it is set to zero.
(vector3-x v)
(vector3-y v)
(vector3-z v)
Return the corresponding component of the vector v.
(vector3+ v1 v2)
(vector3- v1 v2)
(vector3-cross v1 v2)
Return the sum, difference, or cross product of the two vectors.
(vector3* a b)
If a and b are both vectors, returns their dot product. If one of them is a number and the other is a vector, then scales the vector by the number.
(vector3-dot v1 v2)
Returns the dot product of v1 and v2.
(vector3-cross v1 v2)
Returns the cross product of v1 and v2.
(vector3-cdot v1 v2)
Returns the conjugated dot product: v1* dot v2.
(vector3-norm v)
Returns the length (sqrt (vector3-cdot v v)) of the given vector.
(unit-vector3 x [y z])
(unit-vector3 v)
Given a vector or, alternatively, one or more components, returns a unit vector in that direction.
(vector3-close? v1 v2 tolerance)
Returns whether or not the corresponding components of the two vectors are within tolerance of each other.
(vector3= v1 v2)
Returns whether or not the two vectors are numerically equal. Beware of using this function after operations that may have some error due to the finite precision of floating-point numbers; use vector3-close? instead.
(rotate-vector3 axis theta v)
Returns the vector v rotated by an angle theta (in radians) in the right-hand direction around the axis vector (whose length is ignored). You may find the functions (deg->rad theta-deg) and (rad->deg theta-rad) useful to convert angles between degrees and radians.

3x3 matrix functions

(matrix3x3 c1 c2 c3)
Creates a 3x3 matrix with the given 3-vectors as its columns.
(matrix3x3-transpose m)
(matrix3x3-adjoint m)
(matrix3x3-determinant m)
(matrix3x3-inverse m)
Return the transpose, adjoint (conjugate transpose), determinant, or inverse of the given matrix.
(matrix3x3+ m1 m2)
(matrix3x3- m1 m2)
(matrix3x3* m1 m2)
Return the sum, difference, or product of the given matrices.
(matrix3x3* v m)
(matrix3x3* m v)
Returns the (3-vector) product of the matrix m by the vector v, with the vector multiplied on the left or the right respectively.
(matrix3x3* s m)
(matrix3x3* m s)
Scales the matrix m by the number s.
(rotation-matrix3x3 axis theta)
Like rotate-vector3, except returns the (unitary) rotation matrix that performs the given rotation. i.e., (matrix3x3* (rotation-matrix3x3 axis theta) v) produces the same result as (rotate-vector3 axis theta v).

Objects (members of classes)

(make class [ properties ... ])
Make an object of the given class. Each property is of the form (property-name property-value). A property need not be specified if it has a default value, and properties may be given in any order.
(object-property-value object property-name)
Return the value of the property whose name (symbol) is property-name in object. For example, (object-property-value a-circle-object 'radius). (Returns false if property-name is not a property of object.)

Miscellaneous utilities

(conj x)
Return the complex conjugate of a number x (for some reason, Scheme doesn't provide such a function).
(interpolate n list)
Given a list of numbers or 3-vectors, linearly interpolates between them to add n new evenly-spaced values between each pair of consecutive values in the original list.
(interpolate-uniform n list)
Similar to interpolate, but attempts to maintain a (roughly) uniform average spacing of the points in the interpolated list. In this case n is only the number of points interpolated on average between each pair of values in the original list.
(print expressions...)
Calls the Scheme display function on each of its arguments from left to right (printing them to standard output). Note that, like display, it does not append a newline to the end of the outputs; you have to do this yourself by including the "\n" string at the end of the expression list. In addition, there is a global variable print-ok?, defaulting to true, that controls whether print does anything; by setting print-ok? to false, you can disable all output.
(begin-time message-string statements...)
Like the Scheme (begin ...) construct, this executes the given sequence of statements one by one. In addition, however, it measures the elapsed time for the statements and outputs it as message-string, followed by the time, followed by a newline. The return value of begin-time is the elapsed time in seconds.
(minimize function tolerance)
Given a function of one (number) argument, finds its minimum within the specified fractional tolerance. If the return value of minimize is assigned to a variable result, then (min-arg result) and (min-val result) give the argument and value of the function at its minimum. If you can, you should use one of the variant forms of minimize, described below.
(minimize function tolerance guess)
The same as above, but you supply an initial guess for where the minimum is located.
(minimize function tolerance arg-min arg-max)
The same as above, but you supply the minimum and maximum function argument values within which to search for the minimum. This is the most preferred form of minimize, and is faster and more robust than the other two variants.
(minimize-multiple function tolerance arg1 .. argN)
Minimize a function of N numeric arguments within the specified fractional tolerance. arg1 .. argN are an initial guess for the function arguments. Returns both the arguments and value of the function at its minimum. A list of the arguments at the minimum are retrieved via min-arg, and the value via min-val.
maximize, maximize-multiple
These are the same as the minimize functions except that they maximizes the function instead of minimizing it. The functions max-arg and max-val are provided instead of min-arg and min-val.
(find-root function tolerance arg-min arg-max)
Find a root of the given function to within the specified fractional tolerance. arg-min and arg-max bracket the desired root; the function must have opposite signs at these two points!
(find-root-deriv function tolerance arg-min arg-max [arg-guess])
As find-root, but function should return a cons pair of (function-value . function-derivative); the derivative information is exploited to achieve faster convergence via Newton's method, compared to find-root. The optional argument arg-guess should be an initial guess for the root location.
(derivative function x [dx tolerance])
(deriv function x [dx tolerance])
(derivative2 function x [dx tolerance])
(deriv2 function x [dx tolerance])
Compute the numerical derivative of the given function at x to within at best the specified fractional tolerance (defaulting to the best achievable tolerance), using Ridder's method of polynomial extrapolation. dx should be a maximum displacement in x for derivative evaluation; the function should change by a significant amount (much larger than the numerical precision) over dx. dx defaults to 1% of x or 0.01, whichever is larger.
If the return value of derivative is assigned to a variable result, then (derivative-df result) and (derivative-df-err result) give the derivative of the function and an estimate of the numerical error in the derivative, respectively.
The derivative2 function computes both the first and second derivatives, using minimal extra function evaluations; the second derivative and its error are then obtained by (derivative-d2f result) and (derivative-d2f-err result).
deriv and deriv2 are identical to derivative and derivative2, except that they directly return the value of the first and second derivatives, respectively (no need to call derivative-df or derivative-d2f). (They don't provide the error estimate, however, or the ability to compute first and second derivatives simulataneously.)
There are also modified versions of these functions that compute one-sided derivatives: they only evaluate the function at arguments x that are ≥0 (or only ≤0). These functions are named derivative+, deriv+ etc. (or derivative- etc. for negative arguments) and otherwise behave identically. They are generally less accurate than the two-sided derivative, above, but are useful for functions with discontinuous derivatives.
(integrate f a b relerr [ abserr maxeval ])
Return the definite integral of the function f from a to b, to within the specified relative error relerr, using an adaptive Gaussian quadrature (in 1d) or adaptive cubature (in multiple dimensions). The optional arguments abserr and maxeval specify an absolute error tolerance (default is zero) and a maximum number of function evaluations (default is no limit). Integration stops when either the relative error or the absolute error or the maximum number of evaluations is met (note that error estimates are only approximate, though).
This function can compute multi-dimensional integrals, in which case f is a function of N variables and a and b are either lists or vectors of length N, giving the (constant) integration bounds in each dimension. (Non-constant integration bounds, i.e. non-rectilinear integration domains, can be handled by an appropriate mapping of the function f.)
(fold-left op init list)
Combine the elements of list using the binary "operator" function (op x y), with initial value init, associating from the left of the list. That is, if list consist of the elements (a b c d), then (fold-left op init list) computes (op (op (op (op init a) b) c) d). For example, if list contains numbers, then (fold-left + 0 list) returns the sum of the elements of list.
(fold-right op init list)
As fold-left, but associate from the right. For example, (op a (op b (op c (op d init)))).
(memoize func)
Return a function wrapping around the function func that "memoizes" its arguments and return values. That is, it returns the same thing as func, but if passed the same arguments as a previous call it returns a cached return value from the previous call instead of recomputing it.
Personal tools